serialization: obfuscate the serialized search paths

We noticed some Swift clients rely on the serialized search paths in the module to
find dependencies and droping these paths altogether can lead to build failures like
rdar://85840921.

This change teaches the serialization to obfuscate the search paths and the deserialization
to recover them. This allows clients to keep accessing these paths without exposing
them when shipping the module to other users.
This commit is contained in:
Xi Ge
2021-11-30 18:52:21 -08:00
parent f00f2841ab
commit 0047d81f9a
21 changed files with 139 additions and 27 deletions

View File

@@ -361,6 +361,7 @@ ModuleFile::getModuleName(ASTContext &Ctx, StringRef modulePath,
serialization::ValidationInfo loadInfo = ModuleFileSharedCore::load(
modulePath.str(), std::move(newBuf), nullptr, nullptr,
/*isFramework*/ isFramework, Ctx.SILOpts.EnableOSSAModules,
Ctx.SearchPathOpts.DeserializedPathRecoverer,
loadedModuleFile);
Name = loadedModuleFile->Name.str();
return std::move(moduleBuf.get());
@@ -999,6 +1000,13 @@ Optional<CommentInfo> ModuleFile::getCommentForDecl(const Decl *D) const {
return getCommentForDeclByUSR(USRBuffer.str());
}
void ModuleFile::collectSerializedSearchPath(
llvm::function_ref<void(StringRef)> callback) const {
for (auto path: Core->SearchPaths) {
callback(path.Path);
}
}
void ModuleFile::collectBasicSourceFileInfo(
llvm::function_ref<void(const BasicSourceFileInfo &)> callback) const {
if (Core->SourceFileListData.empty())

View File

@@ -743,7 +743,8 @@ public:
Optional<Fingerprint> loadFingerprint(const IterableDeclContext *IDC) const;
void collectBasicSourceFileInfo(
llvm::function_ref<void(const BasicSourceFileInfo &)> callback) const;
void collectSerializedSearchPath(
llvm::function_ref<void(StringRef)> callback) const;
// MARK: Deserialization interface

View File

@@ -84,7 +84,8 @@ static bool enterTopLevelModuleBlock(llvm::BitstreamCursor &cursor,
/// Returns true on success.
static bool readOptionsBlock(llvm::BitstreamCursor &cursor,
SmallVectorImpl<uint64_t> &scratch,
ExtendedValidationInfo &extendedInfo) {
ExtendedValidationInfo &extendedInfo,
PathObfuscator &pathRecoverer) {
while (!cursor.AtEndOfStream()) {
Expected<llvm::BitstreamEntry> maybeEntry = cursor.advance();
if (!maybeEntry) {
@@ -119,7 +120,7 @@ static bool readOptionsBlock(llvm::BitstreamCursor &cursor,
unsigned kind = maybeKind.get();
switch (kind) {
case options_block::SDK_PATH:
extendedInfo.setSDKPath(blobData);
extendedInfo.setSDKPath(pathRecoverer.recover(blobData));
break;
case options_block::XCC:
extendedInfo.addExtraClangImporterOption(blobData);
@@ -171,7 +172,8 @@ static bool readOptionsBlock(llvm::BitstreamCursor &cursor,
static ValidationInfo validateControlBlock(
llvm::BitstreamCursor &cursor, SmallVectorImpl<uint64_t> &scratch,
std::pair<uint16_t, uint16_t> expectedVersion, bool requiresOSSAModules,
ExtendedValidationInfo *extendedInfo) {
ExtendedValidationInfo *extendedInfo,
PathObfuscator &pathRecoverer) {
// The control block is malformed until we've at least read a major version
// number.
ValidationInfo result;
@@ -202,7 +204,7 @@ static ValidationInfo validateControlBlock(
result.status = Status::Malformed;
return result;
}
if (!readOptionsBlock(cursor, scratch, *extendedInfo)) {
if (!readOptionsBlock(cursor, scratch, *extendedInfo, pathRecoverer)) {
result.status = Status::Malformed;
return result;
}
@@ -465,10 +467,11 @@ ValidationInfo serialization::validateSerializedAST(
result.status = Status::Malformed;
return result;
}
PathObfuscator localObfuscator;
result = validateControlBlock(
cursor, scratch,
{SWIFTMODULE_VERSION_MAJOR, SWIFTMODULE_VERSION_MINOR},
requiresOSSAModules, extendedInfo);
requiresOSSAModules, extendedInfo, localObfuscator);
if (result.status == Status::Malformed)
return result;
} else if (dependencies &&
@@ -938,7 +941,7 @@ getActualImportControl(unsigned rawValue) {
}
}
bool ModuleFileSharedCore::readModuleDocIfPresent() {
bool ModuleFileSharedCore::readModuleDocIfPresent(PathObfuscator &pathRecoverer) {
if (!this->ModuleDocInputBuffer)
return true;
@@ -977,7 +980,7 @@ bool ModuleFileSharedCore::readModuleDocIfPresent() {
info = validateControlBlock(
docCursor, scratch, {SWIFTDOC_VERSION_MAJOR, SWIFTDOC_VERSION_MINOR},
RequiresOSSAModules,
/*extendedInfo*/ nullptr);
/*extendedInfo*/ nullptr, pathRecoverer);
if (info.status != Status::Valid)
return false;
// Check that the swiftdoc is actually for this module.
@@ -1083,7 +1086,7 @@ bool ModuleFileSharedCore::readDeclLocsBlock(llvm::BitstreamCursor &cursor) {
return false;
}
bool ModuleFileSharedCore::readModuleSourceInfoIfPresent() {
bool ModuleFileSharedCore::readModuleSourceInfoIfPresent(PathObfuscator &pathRecoverer) {
if (!this->ModuleSourceInfoInputBuffer)
return true;
@@ -1121,7 +1124,8 @@ bool ModuleFileSharedCore::readModuleSourceInfoIfPresent() {
infoCursor, scratch,
{SWIFTSOURCEINFO_VERSION_MAJOR, SWIFTSOURCEINFO_VERSION_MINOR},
RequiresOSSAModules,
/*extendedInfo*/ nullptr);
/*extendedInfo*/ nullptr,
pathRecoverer);
if (info.status != Status::Valid)
return false;
// Check that the swiftsourceinfo is actually for this module.
@@ -1196,7 +1200,7 @@ ModuleFileSharedCore::ModuleFileSharedCore(
std::unique_ptr<llvm::MemoryBuffer> moduleDocInputBuffer,
std::unique_ptr<llvm::MemoryBuffer> moduleSourceInfoInputBuffer,
bool isFramework, bool requiresOSSAModules,
serialization::ValidationInfo &info)
serialization::ValidationInfo &info, PathObfuscator &pathRecoverer)
: ModuleInputBuffer(std::move(moduleInputBuffer)),
ModuleDocInputBuffer(std::move(moduleDocInputBuffer)),
ModuleSourceInfoInputBuffer(std::move(moduleSourceInfoInputBuffer)),
@@ -1247,7 +1251,7 @@ ModuleFileSharedCore::ModuleFileSharedCore(
info = validateControlBlock(
cursor, scratch,
{SWIFTMODULE_VERSION_MAJOR, SWIFTMODULE_VERSION_MINOR},
RequiresOSSAModules, &extInfo);
RequiresOSSAModules, &extInfo, pathRecoverer);
if (info.status != Status::Valid) {
error(info.status);
return;
@@ -1372,7 +1376,8 @@ ModuleFileSharedCore::ModuleFileSharedCore(
bool isSystem;
input_block::SearchPathLayout::readRecord(scratch, isFramework,
isSystem);
SearchPaths.push_back({blobData, isFramework, isSystem});
SearchPaths.push_back({pathRecoverer.recover(blobData), isFramework,
isSystem});
break;
}
case input_block::MODULE_INTERFACE_PATH: {
@@ -1572,8 +1577,8 @@ ModuleFileSharedCore::ModuleFileSharedCore(
return;
}
// Read source info file.
readModuleSourceInfoIfPresent();
if (!readModuleDocIfPresent()) {
readModuleSourceInfoIfPresent(pathRecoverer);
if (!readModuleDocIfPresent(pathRecoverer)) {
info.status = error(Status::MalformedDocumentation);
return;
}

View File

@@ -149,7 +149,7 @@ private:
SmallVector<Dependency, 8> Dependencies;
struct SearchPath {
StringRef Path;
std::string Path;
bool IsFramework;
bool IsSystem;
};
@@ -374,7 +374,7 @@ private:
std::unique_ptr<llvm::MemoryBuffer> moduleDocInputBuffer,
std::unique_ptr<llvm::MemoryBuffer> moduleSourceInfoInputBuffer,
bool isFramework, bool requiresOSSAModules,
serialization::ValidationInfo &info);
serialization::ValidationInfo &info, PathObfuscator &pathRecoverer);
/// Change the status of the current module.
Status error(Status issue) {
@@ -464,7 +464,7 @@ private:
/// Loads data from #ModuleDocInputBuffer.
///
/// Returns false if there was an error.
bool readModuleDocIfPresent();
bool readModuleDocIfPresent(PathObfuscator &pathRecoverer);
/// Reads the source loc block, which contains USR to decl location mapping.
///
@@ -474,7 +474,7 @@ private:
/// Loads data from #ModuleSourceInfoInputBuffer.
///
/// Returns false if there was an error.
bool readModuleSourceInfoIfPresent();
bool readModuleSourceInfoIfPresent(PathObfuscator &pathRecoverer);
/// Read an on-disk decl hash table stored in
/// \c sourceinfo_block::DeclUSRSLayout format.
@@ -510,12 +510,13 @@ public:
std::unique_ptr<llvm::MemoryBuffer> moduleDocInputBuffer,
std::unique_ptr<llvm::MemoryBuffer> moduleSourceInfoInputBuffer,
bool isFramework, bool requiresOSSAModules,
PathObfuscator &pathRecoverer,
std::shared_ptr<const ModuleFileSharedCore> &theModule) {
serialization::ValidationInfo info;
auto *core = new ModuleFileSharedCore(
std::move(moduleInputBuffer), std::move(moduleDocInputBuffer),
std::move(moduleSourceInfoInputBuffer), isFramework,
requiresOSSAModules, info);
requiresOSSAModules, info, pathRecoverer);
if (!moduleInterfacePath.empty()) {
ArrayRef<char> path;
core->allocateBuffer(path, moduleInterfacePath);

View File

@@ -1077,9 +1077,11 @@ void Serializer::writeHeader(const SerializationOptions &options) {
options_block::XCCLayout XCC(Out);
const auto &PathRemapper = options.DebuggingOptionsPrefixMap;
const auto &PathObfuscator = options.PathObfuscator;
auto sdkPath = M->getASTContext().SearchPathOpts.SDKPath;
SDKPath.emit(
ScratchRecord,
PathRemapper.remapPath(M->getASTContext().SearchPathOpts.SDKPath));
PathObfuscator.obfuscate(PathRemapper.remapPath(sdkPath)));
auto &Opts = options.ExtraClangOptions;
for (auto Arg = Opts.begin(), E = Opts.end(); Arg != E; ++Arg) {
StringRef arg(*Arg);
@@ -1157,16 +1159,17 @@ void Serializer::writeInputBlock(const SerializationOptions &options) {
input_block::ModuleInterfaceLayout ModuleInterface(Out);
if (options.SerializeOptionsForDebugging) {
const auto &PathObfuscator = options.PathObfuscator;
const auto &PathMapper = options.DebuggingOptionsPrefixMap;
const SearchPathOptions &searchPathOpts = M->getASTContext().SearchPathOpts;
// Put the framework search paths first so that they'll be preferred upon
// deserialization.
for (auto &framepath : searchPathOpts.FrameworkSearchPaths)
SearchPath.emit(ScratchRecord, /*framework=*/true, framepath.IsSystem,
PathMapper.remapPath(framepath.Path));
PathObfuscator.obfuscate(PathMapper.remapPath(framepath.Path)));
for (auto &path : searchPathOpts.ImportSearchPaths)
SearchPath.emit(ScratchRecord, /*framework=*/false, /*system=*/false,
PathMapper.remapPath(path));
PathObfuscator.obfuscate(PathMapper.remapPath(path)));
}
// Note: We're not using StringMap here because we don't need to own the

View File

@@ -404,7 +404,9 @@ llvm::ErrorOr<ModuleDependencies> SerializedModuleLoaderBase::scanModuleFile(
bool isFramework = false;
serialization::ValidationInfo loadInfo = ModuleFileSharedCore::load(
modulePath.str(), std::move(moduleBuf.get()), nullptr, nullptr,
isFramework, isRequiredOSSAModules(), loadedModuleFile);
isFramework, isRequiredOSSAModules(),
Ctx.SearchPathOpts.DeserializedPathRecoverer,
loadedModuleFile);
const std::string moduleDocPath;
const std::string sourceInfoPath;
@@ -730,7 +732,9 @@ LoadedFile *SerializedModuleLoaderBase::loadAST(
serialization::ValidationInfo loadInfo = ModuleFileSharedCore::load(
moduleInterfacePath, std::move(moduleInputBuffer),
std::move(moduleDocInputBuffer), std::move(moduleSourceInfoInputBuffer),
isFramework, isRequiredOSSAModules(), loadedModuleFileCore);
isFramework, isRequiredOSSAModules(),
Ctx.SearchPathOpts.DeserializedPathRecoverer,
loadedModuleFileCore);
SerializedASTFile *fileUnit = nullptr;
if (loadInfo.status == serialization::Status::Valid) {
@@ -1555,3 +1559,8 @@ void SerializedASTFile::collectBasicSourceFileInfo(
llvm::function_ref<void(const BasicSourceFileInfo &)> callback) const {
File.collectBasicSourceFileInfo(callback);
}
void SerializedASTFile::collectSerializedSearchPath(
llvm::function_ref<void(StringRef)> callback) const {
File.collectSerializedSearchPath(callback);
}