//! [basicReadTest] static void basicReadTest(const P2::Partitioner &partitioner) { std::cout <<"\n" <<std::string(40, '=') <<"\nbasicReadTest\n" <<std::string(40, '=') <<"\n"; SymbolicSemantics::Formatter fmt; fmt.set_line_prefix(" "); // Create the RiscOperators and the initial state. const RegisterDictionary *regdict = partitioner.instructionProvider().registerDictionary(); const RegisterDescriptor REG = partitioner.instructionProvider().stackPointerRegister(); const std::string REG_NAME = RegisterNames(regdict)(REG); BaseSemantics::RiscOperatorsPtr ops = SymbolicSemantics::RiscOperators::instance(regdict); ops->currentState()->memoryState()->set_byteOrder(partitioner.instructionProvider().defaultByteOrder()); BaseSemantics::StatePtr initialState = ops->currentState()->clone(); ops->initialState(initialState); // lazily evaluated initial state std::cout <<"Initial state before reading:\n" <<(*initialState+fmt); // Read some memory and a register, which should cause them to spring into existence in both the current state and the // initial state. BaseSemantics::SValuePtr addr1 = ops->number_(32, 0); BaseSemantics::SValuePtr dflt1m = ops->number_(32, 0x11223344); BaseSemantics::SValuePtr read1m = ops->readMemory(RegisterDescriptor(), addr1, dflt1m, ops->boolean_(true)); BaseSemantics::SValuePtr dflt1r = ops->undefined_(REG.get_nbits()); BaseSemantics::SValuePtr read1r = ops->readRegister(REG, dflt1r); std::cout <<"Initial state after reading " <<*read1m <<" from address " <<*addr1 <<"\n" <<"and " <<*read1r <<" from " <<REG_NAME <<"\n" <<(*initialState+fmt); ASSERT_always_require(read1m->must_equal(dflt1m)); ASSERT_always_require(read1r->must_equal(dflt1r)); // Create a new current state and read again. We should get the same value even though the current state is empty. BaseSemantics::StatePtr curState = ops->currentState()->clone(); curState->clear(); ops->currentState(curState); BaseSemantics::SValuePtr dflt2m = ops->number_(32, 0x55667788); BaseSemantics::SValuePtr read2m = ops->readMemory(RegisterDescriptor(), addr1, dflt2m, ops->boolean_(true)); BaseSemantics::SValuePtr dflt2r = ops->undefined_(REG.get_nbits()); BaseSemantics::SValuePtr read2r = ops->readRegister(REG, dflt2r); std::cout <<"Initial state after reading " <<*read2m <<" from address " <<*addr1 <<"\n" <<"and " <<*read2r <<" from " <<REG_NAME <<"\n" <<(*initialState+fmt); ASSERT_always_require(read1m->must_equal(read2m)); ASSERT_always_require(read1r->must_equal(read2r)); // Disable the initial state. If we re-read the same address we'll still get the same result because it's now present in // the current state also. ops->initialState(BaseSemantics::StatePtr()); BaseSemantics::SValuePtr dflt3m = ops->number_(32, 0x99aabbcc); BaseSemantics::SValuePtr read3m = ops->readMemory(RegisterDescriptor(), addr1, dflt3m, ops->boolean_(true)); BaseSemantics::SValuePtr dflt3r = ops->undefined_(REG.get_nbits()); BaseSemantics::SValuePtr read3r = ops->readRegister(REG, dflt3r); ASSERT_always_require(read1m->must_equal(read3m)); ASSERT_always_require(read1r->must_equal(read3r)); }
/* Analyze a single interpretation a block at a time */ static void analyze_interp(SgAsmInterpretation *interp) { /* Get the set of all instructions except instructions that are part of left-over blocks. */ struct AllInstructions: public SgSimpleProcessing, public std::map<rose_addr_t, SgAsmX86Instruction*> { void visit(SgNode *node) { SgAsmX86Instruction *insn = isSgAsmX86Instruction(node); SgAsmFunction *func = SageInterface::getEnclosingNode<SgAsmFunction>(insn); if (func && 0==(func->get_reason() & SgAsmFunction::FUNC_LEFTOVERS)) insert(std::make_pair(insn->get_address(), insn)); } } insns; insns.traverse(interp, postorder); while (!insns.empty()) { std::cout <<"=====================================================================================\n" <<"=== Starting a new basic block ===\n" <<"=====================================================================================\n"; AllInstructions::iterator si = insns.begin(); SgAsmX86Instruction *insn = si->second; insns.erase(si); BaseSemantics::RiscOperatorsPtr operators = make_ops(); BaseSemantics::Formatter formatter; formatter.set_suppress_initial_values(); formatter.set_show_latest_writers(do_usedef); BaseSemantics::DispatcherPtr dispatcher; if (do_trace) { // Enable RiscOperators tracing, but turn off a bunch of info that makes comparisons with a known good answer // difficult. Sawyer::Message::PrefixPtr prefix = Sawyer::Message::Prefix::instance(); prefix->showProgramName(false); prefix->showThreadId(false); prefix->showElapsedTime(false); prefix->showFacilityName(Sawyer::Message::Prefix::NEVER); prefix->showImportance(false); Sawyer::Message::UnformattedSinkPtr sink = Sawyer::Message::StreamSink::instance(std::cout); sink->prefix(prefix); sink->defaultPropertiesNS().useColor = false; TraceSemantics::RiscOperatorsPtr trace = TraceSemantics::RiscOperators::instance(operators); trace->stream().destination(sink); trace->stream().enable(); dispatcher = DispatcherX86::instance(trace, 32); } else { dispatcher = DispatcherX86::instance(operators, 32); } operators->set_solver(make_solver()); // The fpstatus_top register must have a concrete value if we'll use the x86 floating-point stack (e.g., st(0)) if (const RegisterDescriptor *REG_FPSTATUS_TOP = regdict->lookup("fpstatus_top")) { BaseSemantics::SValuePtr st_top = operators->number_(REG_FPSTATUS_TOP->get_nbits(), 0); operators->writeRegister(*REG_FPSTATUS_TOP, st_top); } #if SEMANTIC_DOMAIN == SYMBOLIC_DOMAIN BaseSemantics::SValuePtr orig_esp; if (do_test_subst) { // Only request the orig_esp if we're going to use it later because it causes an esp value to be instantiated // in the state, which is printed in the output, and thus changes the answer. BaseSemantics::RegisterStateGeneric::promote(operators->get_state()->get_register_state())->initialize_large(); orig_esp = operators->readRegister(*regdict->lookup("esp")); std::cout <<"Original state:\n" <<*operators; } #endif /* Perform semantic analysis for each instruction in this block. The block ends when we no longer know the value of * the instruction pointer or the instruction pointer refers to an instruction that doesn't exist or which has already * been processed. */ while (1) { /* Analyze current instruction */ std::cout <<"\n" <<unparseInstructionWithAddress(insn) <<"\n"; try { dispatcher->processInstruction(insn); # if 0 /*DEBUGGING [Robb P. Matzke 2013-05-01]*/ show_state(operators); // for comparing RegisterStateGeneric with the old RegisterStateX86 output # else std::cout <<(*operators + formatter); # endif } catch (const BaseSemantics::Exception &e) { std::cout <<e <<"\n"; } /* Never follow CALL instructions */ if (insn->get_kind()==x86_call || insn->get_kind()==x86_farcall) break; /* Get next instruction of this block */ BaseSemantics::SValuePtr ip = operators->readRegister(dispatcher->findRegister("eip")); if (!ip->is_number()) break; rose_addr_t next_addr = ip->get_number(); si = insns.find(next_addr); if (si==insns.end()) break; insn = si->second; insns.erase(si); } // Test substitution on the symbolic state. #if SEMANTIC_DOMAIN == SYMBOLIC_DOMAIN if (do_test_subst) { SymbolicSemantics::SValuePtr from = SymbolicSemantics::SValue::promote(orig_esp); BaseSemantics::SValuePtr newvar = operators->undefined_(32); newvar->set_comment("frame_pointer"); SymbolicSemantics::SValuePtr to = SymbolicSemantics::SValue::promote(operators->add(newvar, operators->number_(32, 4))); std::cout <<"Substituting from " <<*from <<" to " <<*to <<"\n"; SymbolicSemantics::RiscOperators::promote(operators)->substitute(from, to); std::cout <<"Substituted state:\n" <<(*operators+formatter); } #endif } }
/* Analyze a single interpretation a block at a time */ static void analyze_interp(SgAsmInterpretation *interp) { /* Get the set of all instructions except instructions that are part of left-over blocks. */ struct AllInstructions: public SgSimpleProcessing, public std::map<rose_addr_t, SgAsmx86Instruction*> { void visit(SgNode *node) { SgAsmx86Instruction *insn = isSgAsmx86Instruction(node); SgAsmFunction *func = SageInterface::getEnclosingNode<SgAsmFunction>(insn); if (func && 0==(func->get_reason() & SgAsmFunction::FUNC_LEFTOVERS)) insert(std::make_pair(insn->get_address(), insn)); } } insns; insns.traverse(interp, postorder); while (!insns.empty()) { std::cout <<"=====================================================================================\n" <<"=== Starting a new basic block ===\n" <<"=====================================================================================\n"; AllInstructions::iterator si = insns.begin(); SgAsmx86Instruction *insn = si->second; insns.erase(si); #if SEMANTIC_API == NEW_API BaseSemantics::RiscOperatorsPtr operators = make_ops(); BaseSemantics::Formatter formatter; formatter.set_suppress_initial_values(); BaseSemantics::DispatcherPtr dispatcher; if (do_trace) { TraceSemantics::RiscOperatorsPtr trace = TraceSemantics::RiscOperators::instance(operators); trace->set_stream(stdout); dispatcher = DispatcherX86::instance(trace); } else { dispatcher = DispatcherX86::instance(operators); } operators->set_solver(make_solver()); #else // OLD_API typedef X86InstructionSemantics<MyPolicy, MyValueType> MyDispatcher; MyPolicy operators; MyDispatcher dispatcher(operators); # if SEMANTIC_DOMAIN == SYMBOLIC_DOMAIN operators.set_solver(make_solver()); SymbolicSemantics::Formatter formatter; formatter.expr_formatter.do_rename = true; formatter.expr_formatter.add_renames = true; # elif SEMANTIC_DOMAIN != FINDCONST_DOMAIN && SEMANTIC_DOMAIN != FINDCONSTABI_DOMAIN BaseSemantics::Formatter formatter; # endif #endif #if SEMANTIC_DOMAIN == SYMBOLIC_DOMAIN && SEMANTIC_API == NEW_API BaseSemantics::SValuePtr orig_esp; if (do_test_subst) { // Only request the orig_esp if we're going to use it later because it causes an esp value to be instantiated // in the state, which is printed in the output, and thus changes the answer. BaseSemantics::RegisterStateGeneric::promote(operators->get_state()->get_register_state())->initialize_large(); orig_esp = operators->readRegister(*regdict->lookup("esp")); std::cout <<"Original state:\n" <<*operators; } #endif /* Perform semantic analysis for each instruction in this block. The block ends when we no longer know the value of * the instruction pointer or the instruction pointer refers to an instruction that doesn't exist or which has already * been processed. */ while (1) { /* Analyze current instruction */ std::cout <<"\n" <<unparseInstructionWithAddress(insn) <<"\n"; #if SEMANTIC_API == NEW_API try { dispatcher->processInstruction(insn); # if 0 /*DEBUGGING [Robb P. Matzke 2013-05-01]*/ show_state(operators); // for comparing RegisterStateGeneric with the old RegisterStateX86 output # else std::cout <<(*operators + formatter); # endif } catch (const BaseSemantics::Exception &e) { std::cout <<e <<"\n"; } #else // OLD API try { dispatcher.processInstruction(insn); # if SEMANTIC_DOMAIN == FINDCONST_DOMAIN || SEMANTIC_DOMAIN == FINDCONSTABI_DOMAIN operators.print(std::cout); # else operators.print(std::cout, formatter); # endif } catch (const MyDispatcher::Exception &e) { std::cout <<e <<"\n"; break; # if SEMANTIC_DOMAIN == PARTSYM_DOMAIN } catch (const MyPolicy::Exception &e) { std::cout <<e <<"\n"; break; # endif } catch (const SMTSolver::Exception &e) { std::cout <<e <<" [ "<<unparseInstructionWithAddress(insn) <<"]\n"; break; } #endif /* Never follow CALL instructions */ if (insn->get_kind()==x86_call || insn->get_kind()==x86_farcall) break; /* Get next instruction of this block */ #if SEMANTIC_API == NEW_API BaseSemantics::SValuePtr ip = operators->readRegister(dispatcher->findRegister("eip")); if (!ip->is_number()) break; rose_addr_t next_addr = ip->get_number(); #else // OLD_API # if SEMANTIC_DOMAIN == PARTSYM_DOMAIN || SEMANTIC_DOMAIN == SYMBOLIC_DOMAIN MyValueType<32> ip = operators.get_ip(); if (!ip.is_known()) break; rose_addr_t next_addr = ip.known_value(); # elif SEMANTIC_DOMAIN == NULL_DOMAIN || SEMANTIC_DOMAIN == INTERVAL_DOMAIN MyValueType<32> ip = operators.readRegister<32>(dispatcher.REG_EIP); if (!ip.is_known()) break; rose_addr_t next_addr = ip.known_value(); # elif SEMANTIC_DOMAIN == MULTI_DOMAIN PartialSymbolicSemantics::ValueType<32> ip = operators.readRegister<32>(dispatcher.REG_EIP) .get_subvalue(MyMultiSemanticsClass::SP0()); if (!ip.is_known()) break; rose_addr_t next_addr = ip.known_value(); # else if (operators.newIp->get().name) break; rose_addr_t next_addr = operators.newIp->get().offset; # endif #endif si = insns.find(next_addr); if (si==insns.end()) break; insn = si->second; insns.erase(si); } // Test substitution on the symbolic state. #if SEMANTIC_DOMAIN == SYMBOLIC_DOMAIN && SEMANTIC_API == NEW_API if (do_test_subst) { SymbolicSemantics::SValuePtr from = SymbolicSemantics::SValue::promote(orig_esp); BaseSemantics::SValuePtr newvar = operators->undefined_(32); newvar->set_comment("frame_pointer"); SymbolicSemantics::SValuePtr to = SymbolicSemantics::SValue::promote(operators->add(newvar, operators->number_(32, 4))); std::cout <<"Substituting from " <<*from <<" to " <<*to <<"\n"; SymbolicSemantics::RiscOperators::promote(operators)->substitute(from, to); std::cout <<"Substituted state:\n" <<(*operators+formatter); } #endif } }
int main(int argc, char *argv[]) { ROSE_INITIALIZE; Diagnostics::initAndRegister(&::mlog, "tool"); parseCommandLine(argc, argv); // Create the machine state const RegisterDictionary *registers = RegisterDictionary::dictionary_amd64(); const RegisterDescriptor EAX = *registers->lookup("eax"); SmtSolverPtr solver; BaseSemantics::RiscOperatorsPtr ops = SymbolicSemantics::RiscOperators::instance(registers, solver); ops->currentState()->memoryState()->set_byteOrder(ByteOrder::ORDER_LSB); // Initialize the machine state with some writes and get the string representation. BaseSemantics::SValuePtr eax = ops->number_(32, 1234); ops->writeRegister(EAX, eax); BaseSemantics::SValuePtr addr0 = ops->number_(32, 0x1000); BaseSemantics::SValuePtr mem0 = ops->number_(8, 123); ops->writeMemory(RegisterDescriptor(), addr0, mem0, ops->boolean_(true)); std::ostringstream s0; s0 <<*ops; // Peek at parts of the state that exist BaseSemantics::SValuePtr v1 = ops->peekRegister(EAX, ops->undefined_(32)); ASSERT_always_not_null(v1); ASSERT_always_require(v1->must_equal(eax, solver)); BaseSemantics::SValuePtr mem1 = ops->peekMemory(RegisterDescriptor(), addr0, ops->undefined_(8)); ASSERT_always_not_null(mem1); ASSERT_always_require(mem1->must_equal(mem0, solver)); std::ostringstream s1; s1 <<*ops; ASSERT_always_require2(s0.str() == s1.str(), s1.str()); // Peek at parts of the state that don't exist const RegisterDescriptor EBX = *registers->lookup("ebx"); BaseSemantics::SValuePtr ebx = ops->undefined_(32); BaseSemantics::SValuePtr v2 = ops->peekRegister(EBX, ebx); ASSERT_always_not_null(v2); ASSERT_always_require(v2->must_equal(ebx, solver)); BaseSemantics::SValuePtr addr2 = ops->number_(32, 0x2000); BaseSemantics::SValuePtr mem2init = ops->undefined_(8); BaseSemantics::SValuePtr mem2 = ops->peekMemory(RegisterDescriptor(), addr2, mem2init); ASSERT_always_not_null(mem2); ASSERT_always_require(mem2->must_equal(mem2init, solver)); std::ostringstream s2; s2 <<*ops; ASSERT_always_require2(s0.str() == s2.str(), s2.str()); // Peek at parts of the state that partly exist. const RegisterDescriptor RAX = *registers->lookup("rax"); BaseSemantics::SValuePtr zero64 = ops->number_(64, 0); BaseSemantics::SValuePtr v3 = ops->peekRegister(RAX, zero64); ASSERT_always_not_null(v3); ASSERT_always_require(v3->must_equal(ops->number_(64, 1234), solver)); BaseSemantics::SValuePtr zero32 = ops->number_(32, 0); BaseSemantics::SValuePtr mem3ans = ops->number_(32, 123); BaseSemantics::SValuePtr mem3 = ops->peekMemory(RegisterDescriptor(), addr0, zero32); ASSERT_always_not_null(mem3); ASSERT_always_require(mem3->must_equal(mem3ans, solver)); std::ostringstream s3; s3 <<*ops; ASSERT_always_require2(s0.str() == s3.str(), s3.str()); std::cout <<s3.str(); }