void
RiscOperators::dumpState() {
Sawyer::Message::Stream out(thread_->tracing(TRACE_STATE));
out.enable();
out <<"Semantic state for thread " <<thread_->get_tid() <<":\n";
if (currentInstruction()) {
out <<" instruction #" <<nInsns() <<" at " <<unparseInstructionWithAddress(currentInstruction()) <<"\n";
} else {
out <<" processed " <<StringUtility::plural(nInsns(), "instructions") <<"\n";
}
out <<" registers:\n";
BaseSemantics::Formatter format;
format.set_line_prefix(" ");
out <<(*currentState()->registerState()+format);
out <<" memory:\n";
thread_->get_process()->mem_showmap(out, "memory:", " ");
if (ARCH_X86 == architecture_) {
out <<" segments:\n";
RegisterNames regNames(currentState()->registerState()->get_register_dictionary());
BOOST_FOREACH (const SegmentInfoMap::Node &node, segmentInfo_.nodes()) {
RegisterDescriptor segreg(x86_regclass_segment, node.key(), 0, 16);
out <<" " <<regNames(segreg) <<": base=" <<StringUtility::addrToString(node.value().base)
<<" limit=" <<StringUtility::addrToString(node.value().limit)
<<" present=" <<(node.value().present?"yes":"no") <<"\n";
}
}
std::string
NoOperation::StateNormalizer::toString(const BaseSemantics::DispatcherPtr &cpu, const BaseSemantics::StatePtr &state_) {
BaseSemantics::StatePtr state = state_;
BaseSemantics::RiscOperatorsPtr ops = cpu->get_operators();
if (!state)
return "";
bool isCloned = false; // do we have our own copy of the state?
// If possible and appropriate, remove the instruction pointer register
const RegisterDescriptor regIp = cpu->instructionPointerRegister();
BaseSemantics::RegisterStateGenericPtr rstate = BaseSemantics::RegisterStateGeneric::promote(state->registerState());
if (rstate && rstate->is_partly_stored(regIp)) {
BaseSemantics::SValuePtr ip = ops->readRegister(cpu->instructionPointerRegister());
if (ip->is_number()) {
state = state->clone();
isCloned = true;
rstate = BaseSemantics::RegisterStateGeneric::promote(state->registerState());
rstate->erase_register(regIp, ops.get());
}
}
// Get the memory state, cloning the state if not done so above.
BaseSemantics::MemoryCellStatePtr mem =
boost::dynamic_pointer_cast<BaseSemantics::MemoryCellState>(state->memoryState());
if (mem && !isCloned) {
state = state->clone();
isCloned = true;
mem = BaseSemantics::MemoryCellState::promote(state->memoryState());
}
// Erase memory that has never been written (i.e., cells that sprang into existence by reading an address) of which appears
// to have been recently popped from the stack.
CellErasurePredicate predicate(ops, ops->readRegister(cpu->stackPointerRegister()), ignorePoppedMemory_);
if (mem)
mem->eraseMatchingCells(predicate);
BaseSemantics::Formatter fmt;
fmt.set_show_latest_writers(false);
fmt.set_show_properties(false);
std::ostringstream ss;
ss <<(*state+fmt);
return ss.str();
}
/* 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
}
}
// Show the register state for BaseSemantics::RegisterStateGeneric in the same format as for RegisterStateX86. This is
// for comparison of the two register states when verifying results. It's also close to the format used by the old binary
// semantics API.
void
show_state(const BaseSemantics::RiscOperatorsPtr &ops)
{
#if SEMANTIC_DOMAIN == MULTI_DOMAIN
std::cout <<*ops;
return;
#endif
struct ShowReg {
BaseSemantics::RiscOperatorsPtr ops;
std::ostream &o;
std::string prefix;
ShowReg(const BaseSemantics::RiscOperatorsPtr &ops, std::ostream &o, const std::string &prefix)
: ops(ops), o(o), prefix(prefix) {}
void operator()(const char *name, const char *abbr=NULL) {
const RegisterDictionary *regdict = ops->get_state()->get_register_state()->get_register_dictionary();
const RegisterDescriptor *desc = regdict->lookup(name);
assert(desc);
(*this)(*desc, abbr?abbr:name);
}
void operator()(const RegisterDescriptor &desc, const char *abbr) {
BaseSemantics::RegisterStatePtr regstate = ops->get_state()->get_register_state();
FormatRestorer fmt(o);
o <<prefix <<std::setw(8) <<std::left <<abbr <<"= { ";
fmt.restore();
BaseSemantics::SValuePtr val = regstate->readRegister(desc, ops.get());
o <<*val <<" }\n";
}
void operator()(unsigned majr, unsigned minr, unsigned offset, unsigned nbits, const char *abbr) {
(*this)(RegisterDescriptor(majr, minr, offset, nbits), abbr);
}
} show(ops, std::cout, " ");
std::cout <<"registers:\n";
show("eax", "ax");
show("ecx", "cx");
show("edx", "dx");
show("ebx", "bx");
show("esp", "sp");
show("ebp", "bp");
show("esi", "si");
show("edi", "di");
show("es");
show("cs");
show("ss");
show("ds");
show("fs");
show("gs");
show("cf");
show(x86_regclass_flags, 0, 1, 1, "?1");
show("pf");
show(x86_regclass_flags, 0, 3, 1, "?3");
show("af");
show(x86_regclass_flags, 0, 5, 1, "?5");
show("zf");
show("sf");
show("tf");
show("if");
show("df");
show("of");
show(x86_regclass_flags, 0, 12, 1, "iopl0");
show(x86_regclass_flags, 0, 13, 1, "iopl1");
show("nt");
show(x86_regclass_flags, 0, 15, 1, "?15");
show("rf");
show("vm");
show(x86_regclass_flags, 0, 18, 1, "ac");
show(x86_regclass_flags, 0, 19, 1, "vif");
show(x86_regclass_flags, 0, 20, 1, "vip");
show(x86_regclass_flags, 0, 21, 1, "id");
show(x86_regclass_flags, 0, 22, 1, "?22");
show(x86_regclass_flags, 0, 23, 1, "?23");
show(x86_regclass_flags, 0, 24, 1, "?24");
show(x86_regclass_flags, 0, 25, 1, "?25");
show(x86_regclass_flags, 0, 26, 1, "?26");
show(x86_regclass_flags, 0, 27, 1, "?27");
show(x86_regclass_flags, 0, 28, 1, "?28");
show(x86_regclass_flags, 0, 29, 1, "?29");
show(x86_regclass_flags, 0, 30, 1, "?30");
show(x86_regclass_flags, 0, 31, 1, "?31");
show("eip", "ip");
BaseSemantics::Formatter memfmt;
memfmt.set_line_prefix(" ");
std::cout <<"memory:\n";
ops->get_state()->print_memory(std::cout, memfmt);
}
/* 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
}
}