Exemple #1
0
int
main(int argc, char *argv[]) {

    // This paragraph initializes the ROSE library, generates the man page for this tool, does command-line parsing for quite a
    // few switches including "--help", loads various specimen resources (ELF/PE, running process, raw memory dumps, etc),
    // disassembles, and partitions.  We could have called Engine::frontend() and done it all in one function call, but then we
    // wouldn't have a Partitioner2::Partitioner object that we need below.
    std::string purpose = "demonstrate inter-function disassembly";
    std::string description =
        "Disassembles and partitions the specimen(s), then tries to disassemble things between the functions.";
    P2::Engine engine;
    std::vector<std::string> specimens = engine.parseCommandLine(argc, argv, purpose, description).unreachedArgs();
    P2::Partitioner partitioner = engine.partition(specimens);

    // The partitioner's address usage map (AUM) describes what part of memory has been disassembled as instructions or
    // data. We're interested in the unused parts between the lowest and highest disassembled addresses, so we loop over those
    // parts.  The hull() is the entire used interval -- lowest to highest addresses used regardless of the unused areas in the
    // middle.  An AddressInterval evaluated in boolean context returns false if it's empty.
    rose_addr_t va = partitioner.aum().hull().least();
    while (AddressInterval unused = partitioner.aum().nextUnused(va)) {

        // Is the unused area beyond the last thing compiled?  We're only interested in the stuff between functions.  This
        // check also means that unused.greatest()+1 will not overflow, which simplifies later code. Overflows are easy to
        // trigger when the specimen's word size is the same as ROSE's word size.
        if (unused.least() > partitioner.aum().hull().greatest())
            break;

        // The unused address might be in the middle of some very large unmapped area of memory, or perhaps in an area that
        // doesn't have execute permission (the partitioner will only disassemble at addresses that we've marked as
        // executable). A naive implementation would just increment to the next address and try again, but that could take a
        // very long time.  This "if" statement will give us the next executable address that falls within the unused interval
        // if possible. The address is assigned to "va" if possible.
        if (!engine.memoryMap().within(unused).require(MemoryMap::EXECUTABLE).next().assignTo(va)) {
            va = unused.greatest() + 1;                 // won't overflow because of check above
            continue;
        }

        // "va" now points to an executable address that the partitioner doesn't know about yet.
        ASSERT_require(engine.memoryMap().at(va).require(MemoryMap::EXECUTABLE).exists());
        ASSERT_forbid(partitioner.aum().instructionExists(va));
        std::cout <<"unused address " <<StringUtility::addrToString(va) <<"\n";

        // Cause the partitioner to discover (disassemble) one basic block. This doesn't add the basic block to the
        // partitioner or change the partitioner in any way.  If the BB isn't something we want to keep then just forget about
        // it and garbage collection will reclaim the memory.
        P2::BasicBlock::Ptr bb = partitioner.discoverBasicBlock(va);
        if (!isGoodBasicBlock(bb)) {
            ++va;
            continue;
        }
        std::cout <<"  disassembled " <<bb->printableName() <<"\n";

        // Inform the partitioner that we wish to keep this BB.
        partitioner.attachBasicBlock(bb);

        // This BB was not reachable by any previous CFG edge, therefore it doesn't belong to any function. In order for it to
        // show up in the eventual AST we need to add it to some function (the ROSE AST has a requirement that every basic
        // block belongs to a function, although the partitioner can easily cope with the other case). The easiest way in this
        // situation is to just create a new function whose entry block is this BB.  Creating a function doesn't modify the
        // partitioner in any way, so we need to also attach the function to the partitioner.
        P2::Function::Ptr function = P2::Function::instance(va, SgAsmFunction::FUNC_USERDEF);
        function->insertBasicBlock(va);                 // allowed only before attaching function to partitioner
        partitioner.attachOrMergeFunction(function);

        // This basic block might be the first block of a whole bunch that are connected by as yet undiscovered CFG edges. We
        // can recursively discover and attach all those blocks with one Engine method.  There are also Partitioner methods to
        // do similar things, but they're lower level.
        engine.runPartitionerRecursive(partitioner);
    }

    // We've probably added a bunch more functions and basic blocks to the partitioner, but we haven't yet assigned the basic
    // blocks discovered by Engine::runPartitionerRecursive to any functions.  We might also need to assign function labels
    // from ELF/PE information, re-run some analysis, etc., so do that now.
    engine.runPartitionerFinal(partitioner);

    // Most ROSE analysis is performed on an abstract syntax tree, so generate one.  If the specime is an ELF or PE container
    // then the returned global block will also be attached somewhere below a SgProject node, otherwise the returned global
    // block is the root of the AST and there is no project (e.g., like when the specimen is a raw memory dump).
    SgAsmBlock *gblock = P2::Modules::buildAst(partitioner, engine.interpretation());

    // Generate an assembly listing. These unparser properties are all optional, but they result in more informative assembly
    // listings.
    AsmUnparser unparser;
    unparser.set_registers(partitioner.instructionProvider().registerDictionary());
    unparser.add_control_flow_graph(ControlFlow().build_block_cfg_from_ast<ControlFlow::BlockGraph>(gblock));
    unparser.staticDataDisassembler.init(engine.disassembler());
    unparser.unparse(std::cout, gblock);
}
Exemple #2
0
int
main(int argc, char *argv[]) {
    ROSE_INITIALIZE;
    Diagnostics::initAndRegister(&mlog, "tool");

    // Parse the command-line to configure the partitioner engine, obtain the executable and its arguments, and generate a man
    // page, adjust global settings, etc. This demo tool has no switches of its own, which makes this even easier. For a
    // production tool, it's probably better to obtain the parser and register only those switches we need (e.g., no need for
    // AST generation switches since we skip that step), to set it up to use our own diagnostic stream instead of exceptions,
    // and to adjust this tool's synopsis in the documentation.  Examples of all of these can be found in other demos.
    P2::Engine engine;
    engine.doingPostAnalysis(false);                    // no need for any post-analysis phases (user can override on cmdline)
    std::vector<std::string> command;
    try {
        command = engine.parseCommandLine(argc, argv, purpose, description).unreachedArgs();
    } catch (const std::runtime_error &e) {
        mlog[FATAL] <<"invalid command-line: " <<e.what() <<"\n";
        exit(1);
    }
    if (command.empty()) {
        mlog[FATAL] <<"no executable specified\n";
        exit(1);
    }

    // Since we'll be tracing this program's execution, we might as well disassemble the process's memory directly. That way we
    // don't have to worry about ROSE mapping the specimen to the same virtual address as the kernel (which might be using
    // address randomization). We can stop short of generating the AST because we won't need it.
    BinaryAnalysis::BinaryDebugger debugger(command);
    std::string specimenResourceName = "proc:noattach:" + StringUtility::numberToString(debugger.isAttached());
    P2::Partitioner partitioner = engine.partition(specimenResourceName);
    partitioner.memoryMap()->dump(std::cerr);           // show the memory map as a debugging aid

    // Create a global control flow graph whose vertices are instructions from a global CFG whose verts are mostly basic
    // blocks.
    InsnCfg insnCfg;
    const P2::ControlFlowGraph &bbCfg = partitioner.cfg();
    BOOST_FOREACH (const P2::ControlFlowGraph::Vertex &bbVert, bbCfg.vertices()) {
        if (P2::BasicBlock::Ptr bb = isBasicBlock(bbVert)) {
            const std::vector<SgAsmInstruction*> &insns = bb->instructions();

            // Each basic block has one or more instructions that need to be inserted into our instruction control flow graph
            // with edges from each instruction to the next.  The insertEdgeWithVertices automatically inserts missing
            // vertices, and doesn't insert vertices that already exist, making it convenient for this type of construction.
            for (size_t i=1; i<insns.size(); ++i)
                insnCfg.insertEdgeWithVertices(insns[i-1], insns[i]);

            // The final instruction of this block needs to flow into each of the initial instructions of the successor basic
            // blocks. Be careful that the successors are actually existing basic blocks.  Note that in ROSE's global CFG, a
            // function call has at least two successors: the function being called (normal edges), and the address to which
            // the function returns ("callret" edges). There are other types of edges too, but we want only the normal edges.
            BOOST_FOREACH (const P2::ControlFlowGraph::Edge &bbEdge, bbVert.outEdges()) {
                if (bbEdge.value().type() == P2::E_NORMAL) {
                    if (P2::BasicBlock::Ptr target = isBasicBlock(*bbEdge.target()))
                        insnCfg.insertEdgeWithVertices(insns.back(), target->instructions()[0]);
                }
            }
        }
    }
    mlog[INFO] <<"block CFG: "
               <<StringUtility::plural(bbCfg.nVertices(), "vertices", "vertex") <<", "
               <<StringUtility::plural(bbCfg.nEdges(), "edges") <<"\n";
    mlog[INFO] <<"insn CFG:  "
               <<StringUtility::plural(insnCfg.nVertices(), "vertices", "vertex") <<", "
               <<StringUtility::plural(insnCfg.nEdges(), "edges") <<"\n";
    
    // Run the executable to obtain a trace.  We use the instruction pointer to look up a SgAsmInstruction in the insnCfg and
    // thus map the trace onto the instruction CFG.
    mlog[INFO] <<"running subordinate to obtain trace: " <<boost::join(command, " ") <<"\n";
    std::set<rose_addr_t> missingAddresses;
    Trace trace;
    while (!debugger.isTerminated()) {
        // Find the instruction CFG vertex corresponding to the current execution address. It could be that the execution
        // address doesn't exist in the CFG, and this can be caused by a number of things including failure of ROSE to
        // statically find the address, dynamic libraries that weren't loaded statically, etc.
        rose_addr_t va = debugger.executionAddress();
        InsnCfg::ConstVertexIterator vertex = insnCfg.findVertexKey(va);
        if (!insnCfg.isValidVertex(vertex)) {
            missingAddresses.insert(va);
        } else {
            trace.append(vertex->id());
        }
        debugger.singleStep();
    }
    mlog[INFO] <<"subordinate " <<debugger.howTerminated() <<"\n";
    mlog[INFO] <<"trace length: " <<StringUtility::plural(trace.size(), "instructions") <<"\n";
    Diagnostics::mfprintf(mlog[INFO])("overall burstiness: %6.2f%%\n", 100.0 * trace.burstiness());
    mlog[INFO] <<"distinct executed addresses missing from CFG: " <<missingAddresses.size() <<"\n";

    // Print a list of CFG vertices that were never reached.  We use std::cout rather than diagnostics because this is one of
    // the main outputs of this demo. The "if" condition is constant time.
    BOOST_FOREACH (const InsnCfg::Vertex &vertex, insnCfg.vertices()) {
        if (!trace.exists(vertex.id()))
            std::cout <<"not executed: " <<unparseInstructionWithAddress(vertex.value()) <<"\n";
    }

    // Print list of addresses that were executed but did not appear in the CFG
    BOOST_FOREACH (rose_addr_t va, missingAddresses)
        std::cout <<"missing address: " <<StringUtility::addrToString(va) <<"\n";

    // Print those branch instructions that were executed by the trace but always took the same branch.  Just to mix things up,
    // I'll iterate over the trace labels this time instead of the CFG vertices.  Remember, the labels are the integer IDs of
    // the CFG vertices. The "if" condition executes in constant time, as does the next line.
    for (size_t i = 0; i < trace.nLabels(); ++i) {
        if (insnCfg.findVertex(i)->nOutEdges() > 1 && trace.successors(i).size() == 1) {
            SgAsmInstruction *successor = insnCfg.findVertex(*trace.successorSet(i).begin())->value();
            std::cout <<"single flow: " <<unparseInstructionWithAddress(insnCfg.findVertex(i)->value())
                      <<" --> " <<unparseInstructionWithAddress(successor) <<"\n";
        }
    }

    // Get a list of executed instructions that are branch points and sort them by their burstiness.  The "if" condition is
    // constant time.
    std::vector<InsnTraceInfo> info;
    BOOST_FOREACH (const InsnCfg::Vertex &vertex, insnCfg.vertices()) {
        if (vertex.nOutEdges() > 1 && trace.exists(vertex.id()))
            info.push_back(InsnTraceInfo(vertex.value(), trace.burstiness(vertex.id()), trace.size(vertex.id())));
    }
    std::sort(info.begin(), info.end());
    std::reverse(info.begin(), info.end());
    BOOST_FOREACH (const InsnTraceInfo &record, info) {
        Diagnostics::mfprintf(std::cout)("burstiness %6.2f%% %5zu hits at %s\n",
                                         100.0*record.burstiness, record.nHits,
                                         unparseInstructionWithAddress(record.insn).c_str());
    }