Program Listing for File taintAnalysis.h

Program Listing for File taintAnalysis.h#

↰ Return to documentation for file (src/midend/programAnalysis/genericDataflow/simpleAnalyses/taintAnalysis.h)

#include <featureTests.h>
#ifdef ROSE_ENABLE_SOURCE_ANALYSIS

#ifndef ROSE_TaintAnalysis_H
#define ROSE_TaintAnalysis_H

// Tainted flow analysis.
//
// The original version of this tainted flow analysis was written 2012-09 by someone other than the author of the
// genericDataflow framework.  It is based on the sign analysis (sgnAnalysis.[Ch]) in this same directory since documentation
// for the genericDataflow framework is fairly sparse: 5 pages in the tutorial, not counting the code listings) and no doxygen
// documentation.
//
// This file contains two types of comments:
//    1. Comments that try to document some of the things I've discovered through playing with the genericDataflow framework.
//    2. Comments and suggestions about usability, consistency, applicability to binary analysis, etc.
//
// [RPM 2012-09]


// USABILITY: Names of header files aren't consistent across the genericDataflow files. E.g., in the "lattice" directory we
// have "lattice.h" that defines the Lattice class, but "ConstrGraph.h" that defines "ConstrGraph" (and apparently no
// documentation as to what "Constr" means).
#include "lattice.h"
#include "dataflow.h"
#include "liveDeadVarAnalysis.h"        // misspelled? Shouldn't it be liveDeadVarsAnalysis or LiveDeadVarsAnalysis?


// USABILITY: The abundant use of dynamic_cast makes it seem like something's wrong with the whole dataflow design.  And I
//            couldn't find any documentation about when it's okay to cast from Lattice to one of its subclasses, so I've made
//            the assumption throughout that when the dynamic_cast returns null, the node in question points to a variable or
//            expression that the live/dead analysis has determined to be dead.
//
// USABILITY: No doxygen comments throughout genericDataflow framework?!?  But it looks like there's some doxygen-like stuff
//            describing a few function parameters, so is it using some other documenting system?  I at least added the headers
//            to the docs/Rose/rose.cfg file so doxygen picks up the structure.
//
// USABILITY: The genericDataflow framework always produces files named "index.html", "summary.html", and "detail.html" and
//            a directory named "dbg_imgs" regardless of any debug settings.  These are apprently the result of the Dbg::init()
//            call made from the main program, and this call is required (omitting it results in segmentation faults).  The
//            file names are constant and therefore one should expect tests to clobber each other's outputs when run in
//            parallel.  The contents of the files cannot be trusted.  Furthermore, since these are HTML files, an aborted test
//            will generate only a partial HTML file which some browsers will choke on.


/******************************************************************************************************************************
 * Taint Lattice
 ******************************************************************************************************************************/

class TaintLattice: public FiniteLattice {
public:

    enum Vertex {
        VERTEX_BOTTOM,
        VERTEX_UNTAINTED,
        VERTEX_TAINTED,
        // no need for a top since that would imply that the value is tainted. I.e., VERTEX_TAINTED *is* our top.
    };

protected:
    Vertex vertex;
public:

    TaintLattice(): vertex(VERTEX_BOTTOM) {}

    virtual void initialize() override {
        *this = TaintLattice();
    }

    Vertex get_vertex() const { return vertex; }
    bool set_vertex(Vertex v);
    virtual Lattice *copy() const override {
        return new TaintLattice(*this);
    }

    // USABILITY: The base class defines copy() without a const argument, so we must do the same here.
    virtual void copy(/*const*/ Lattice *other_) override;


    // USABILITY: The base class defines '==' with non-const argument and "this", so we must do the same here.
    // USABILITY: This is not a real equality predicate since it's not reflexive.  In other words, (A==B) does not imply (B==A)
    //            for all values of A and B.
    virtual bool operator==(/*const*/ Lattice *other_) /*const*/ override;

    // USABILITY: The base class defines str() with non-const "this", so we must do the same here.  That means that if we want
    //            to use this functionality from our own methods (that have const "this") we have to distill it out to some
    //            other place.
    // USABILITY: The "prefix" argument is pointless. Why not just use StringUtility::prefixLines() in the base class rather
    //            than replicate this functionality all over the place?
    virtual std::string str(/*const*/ std::string /*&*/prefix) /*const*/ override {
        return prefix + to_string();
    }

    // USABILITY: We define this only because of deficiencies with the "str" signature in the base class.  Otherwise our
    //            printing method (operator<<) could just use str().  We're trying to avoid evil const_cast.
    std::string to_string() const;

    // USABILITY: The base class defines meetUpdate() with a non-const argument, so we must do the same here.
    virtual bool meetUpdate(/*const*/ Lattice *other_) override;

    friend std::ostream& operator<<(std::ostream &o, const TaintLattice &lattice);
};

/******************************************************************************************************************************
 * Taint Flow Analysis
 ******************************************************************************************************************************/

class TaintAnalysis: public IntraFWDataflow {
protected:
    LiveDeadVarsAnalysis* ldv_analysis;
    std::ostream *debug;

public:
    // USABILITY: Documentation as to why a live/dead analysis is used in SgnAnalysis would be nice. I tried doing it without
    //            originally to make things simpler, but it seems that the FiniteVarsExprProductLattice depends on it even
    //            though I saw commented out code and comments somewhere(?) that indicated otherwise.
    TaintAnalysis(LiveDeadVarsAnalysis *ldv_analysis)
        : ldv_analysis(ldv_analysis), debug(NULL) {}

    std::ostream *get_debug() const { return debug; }
    void set_debug(std::ostream *os) { debug = os; }
    // BINARIES:  The "Function" type is a wrapper around SgFunctionDeclaration and the data flow traversals depend on this
    //            fact.  Binaries don't have SgFunctionDeclaration nodes (they have SgAsmFunction, which is a bit different).
    //
    // NOTE:      The "DataflowNode" is just a VirtualCFG::DataflowNode that contains a VirtualCFG::CFGNode pointer and a
    //            "filter". I didn't find any documentation for how "filter" is used.
    //
    // USABILITY: The "initLattices" and "initFacts" are not documented. They're apparently only outputs for this function
    //            since they seem to be empty on every call and are not const.  They're apparently not parallel arrays since
    //            the examples I was looking at don't push the same number of items into each vector.
    //
    // USABILITY: Copied from src/midend/programAnalysis/genericDataflow/simpleAnalyses/sgnAnalysis.C.  I'm not sure what
    //            it's doing yet since there's no doxygen documentation for FiniteVarsExprsProductLattice or any of its
    //            members.
    //
    // BINARIES:  This might not work for binaries because FiniteVarsExprsProductLattice seems to do things in terms of
    //            variables.  Variables are typically lacking from binary specimens and most existing binary analysis
    //            describes things in terms of static register names or dynamic memory locations.
    void genInitState(const Function& func, const DataflowNode& node, const NodeState& state,
                      std::vector<Lattice*>& initLattices, std::vector<NodeFact*>& initFacts);

    // USABILITY: Not documented in doxygen, so I'm more or less copying from the SgnAnalysis::transfer() method defined in
    //            src/midend/programAnalysis/genericDataflow/sgnAnalysis.C.
    bool transfer(const Function& func, const DataflowNode& node_, NodeState& state, const std::vector<Lattice*>& dfInfo);

protected:
    static std::string lattice_info(const TaintLattice *lattice) {
        return lattice ? lattice->to_string() : "dead";
    }

    bool magic_tainted(SgNode *node, FiniteVarsExprsProductLattice *prodLat);
};

#endif
#endif