Skip to content
Snippets Groups Projects
SMPFunction.cpp 237 KiB
Newer Older
jdh8d's avatar
jdh8d committed
/*
 * SMPFunction.cpp - <see below>.
 *
 * Copyright (c) 2000, 2001, 2010 - University of Virginia 
 *
 * This file is part of the Memory Error Detection System (MEDS) infrastructure.
 * This file may be used and modified for non-commercial purposes as long as 
 * all copyright, permission, and nonwarranty notices are preserved.  
 * Redistribution is prohibited without prior written consent from the University 
 * of Virginia.
 *
 * Please contact the authors for restrictions applying to commercial use.
 *
 * THIS SOURCE IS PROVIDED "AS IS" AND WITHOUT ANY EXPRESS OR IMPLIED
 * WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF
 * MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.
 *
 * Author: University of Virginia
 * e-mail: jwd@virginia.com
 * URL   : http://www.cs.virginia.edu/
 *
 * Additional copyrights 2010, 2011 by Zephyr Software LLC
 * e-mail: {clc,jwd}@zephyr-software.com
 * URL   : http://www.zephyr-software.com/
 *
jdh8d's avatar
jdh8d committed
 */

//
// SMPFunction.cpp
//
// This module performs the fundamental data flow analyses needed for the
//   SMP project (Software Memory Protection) at the function level.
//

#include <list>
#include <set>
#include <vector>
#include <algorithm>

#include <cstring>

#include <pro.h>
#include <assert.h>
#include <ida.hpp>
#include <idp.hpp>
#include <auto.hpp>
#include <bytes.hpp>
#include <funcs.hpp>
#include <allins.hpp>
#include <intel.hpp>
#include <name.hpp>
#include "SMPDataFlowAnalysis.h"
#include "SMPStaticAnalyzer.h"
#include "SMPFunction.h"
#include "SMPBasicBlock.h"
#include "SMPInstr.h"

// Set to 1 for debugging output
#define SMP_DEBUG 1
#define SMP_DEBUG2 0   // verbose
#define SMP_DEBUG3 0   // verbose
#define SMP_DEBUG_CONTROLFLOW 0  // tells what processing stage is entered
#define SMP_DEBUG_XOR 0
#define SMP_DEBUG_CHUNKS 1  // tracking down tail chunks for functions
#define SMP_DEBUG_DATAFLOW 0
#define SMP_DEBUG_DATAFLOW_VERBOSE 0
#define SMP_DEBUG_TYPE_INFERENCE 0
clc5q's avatar
clc5q committed
#define SMP_DEBUG_PROFILED_TYPE_INFERENCE 0
#define SMP_DEBUG_STACK_GRANULARITY 0
clc5q's avatar
clc5q committed
#define SMP_DEBUG_FUNC 0
#define SMP_DEBUG_FUNC_SAFETY 1
clc5q's avatar
clc5q committed
#define SMP_VERBOSE_DEBUG_FUNC 0
#define SMP_DEBUG_BUILD_RTL 1   // leave this on; serious errors reported
#define SMP_DEBUG_UNINITIALIZED_SSA_NAMES 1
#define SMP_WARN_UNUSED_DEFS 0
clc5q's avatar
clc5q committed
#define SMP_DEBUG_SWITCH_TABLE_INFO 0
#define SMP_OPTIMIZE_BLOCK_PROFILING 0
#define SMP_DECLARE_INDIRECT_TARGETS_UNSAFE 1
#define SMP_ANALYZE_STACK_POINTER 0
#define SMP_AUDIT_STACK_POINTER_DELTAS 0
#define SMP_COMPARE_IDA_STARS_STACK_POINTER_DELTAS 1

// Compute LVA/SSA or not? Turn it off for NICECAP demo on 31-JAN-2008
// Compute fine-grained stack boundaries?
#define SMP_COMPUTE_STACK_GRANULARITY 1
// Use conditional type propagation on phi functions
#define SMP_CONDITIONAL_TYPE_PROPAGATION 0

// Kludges to fix IDA Pro 5.2 errors in cc1.ncexe
#define SMP_IDAPRO52_WORKAROUND 0

// Basic block number 0 is the top of the CFG lattice.
#define SMP_TOP_BLOCK 0 

// Set SharedTailChunks to TRUE for entire printf family
//  After we restructure the parent/tail structure of the database, this
//  will go away.
#define KLUDGE_VFPRINTF_FAMILY 1

// Used for binary search by function number in SMPStaticAnalyzer.cpp
//  to trigger debugging output and find which instruction in which
//  function is causing a crash.
bool SMPBinaryDebug = false;

using namespace std;


// helper function to determine if an object is in a vector
template <class T>	
bool vector_exists(const T &item, const vector<T> &vec) {
	for (size_t i = 0; i < vec.size(); ++i) {
		if (vec[i] == item)
			return true;
	}
	return false;
}
// Comparison function for sorting.
bool LocalVarCompare(const LocalVar &LV1, const LocalVar &LV2) {
	return (LV1.offset < LV2.offset);
}

// *****************************************************************
// Class SMPFunction
// *****************************************************************

// Constructor
SMPFunction::SMPFunction(func_t *Info, SMPProgram* pgm) {
	this->FuncInfo = *Info;
#if 0
#endif
	this->UseFP = false;
	this->StaticFunc = false;
	this->LibFunc = false;
	this->IndirectCalls = false;
	this->UnresolvedIndirectCalls = false;
	this->IndirectJumps = false;
	this->UnresolvedIndirectJumps = false;
	this->SharedChunks = false;
	this->UnsharedChunks = false;
	this->CallsAlloca = false;
	this->STARSStackPtrAnalysisPerformed = false;
	this->BuiltRTLs = false;
	this->SpecSafeFunc = false;
	this->SafeCallee = false;
	this->SpecSafeCallee = false;
	this->SafeFunc = true;
	this->SpecSafeFunc = true;
	this->SafeCallee = true;
	this->SpecSafeCallee = true;
#endif
	this->NeedsStackReferent = true;
clc5q's avatar
clc5q committed
	this->SpecNeedsStackReferent = true;
	this->HasIndirectWrites = false;
	this->PossibleIndirectCallTarget = false;
	this->PossibleTailCallTarget = false;
	this->OutgoingArgsComputed = false;
	this->GoodLocalVarTable = false;
	this->OutgoingArgsSize = 0;
	this->TypedDefs = 0;
	this->UntypedDefs = 0;
	this->TypedPhiDefs = 0;
	this->UntypedPhiDefs = 0;
	this->SafeBlocks = 0;
	this->UnsafeBlocks = 0;
	this->Size = 0;
	this->LocalVarsSize = 0;
	this->CalleeSavedRegsSize = 0;
	this->RetAddrSize = 0;
	this->IncomingArgsSize = 0;
	this->OutgoingArgsSize = 0;
	this->LocalVarsAllocInstr = BADADDR;
	this->LocalVarsDeallocInstr = BADADDR;
	this->AllocPointDelta = 0;
	this->MinStackDelta = 0;
	this->NetStackDelta = CALLING_CONVENTION_DEFAULT_FUNCTION_STACK_DELTA;
	this->PreAllocStackDelta = CALLING_CONVENTION_DEFAULT_PREFRAMEALLOC_STACK_DELTA;
	this->FramePointerStackDelta = 0;
	this->ReturnAddrStatus = FUNC_UNKNOWN;
	this->SetIsSpeculative(false);

	this->Instrs.clear();
	this->Blocks.clear();
	this->DirectCallTargets.clear();
	this->IndirectCallTargets.clear();
	this->AllCallTargets.clear();
	this->InstBlockMap.clear();
	this->RPOBlocks.clear();
	this->IDom.clear();
	this->DomTree.clear();
	this->GlobalNames.clear();
	this->BlocksDefinedIn.clear();
	this->SSACounter.clear();
	this->SSAStack.clear();
	this->LocalVarTable.clear();
	this->StackFrameMap.clear();
	this->FineGrainedStackTable.clear();
	this->SavedRegLoc.clear();
	this->ReturnRegTypes.clear();
	this->LiveInSet.clear();
	this->LiveOutSet.clear();
	this->KillSet.clear();
	this->GlobalDefAddrBySSA.clear();

	for (int RegIndex = R_ax; RegIndex <= R_di; ++RegIndex) {
		this->SavedRegLoc.push_back(0); // zero offset means reg not saved
		this->ReturnRegTypes.push_back(UNINIT);
	}

	// The sizes of the three regions of the stack frame other than the
	//  return address are stored in the function structure.
	this->LocalVarsSize = this->FuncInfo.frsize;
	this->CalleeSavedRegsSize = this->FuncInfo.frregs;
	this->IncomingArgsSize = this->FuncInfo.argsize;

	// The return address size can be obtained in a machine independent
	//  way by calling get_frame_retsize(). 
	this->RetAddrSize = get_frame_retsize(this->GetFuncInfo());

} // end of SMPFunction() constructor
SMPFunction::~SMPFunction() {
	list<SMPInstr *>::iterator InstIter;
	for (InstIter = this->Instrs.begin(); InstIter != this->Instrs.end(); ++InstIter) {
		SMPInstr *CurrInst = (*InstIter);
		delete CurrInst;
	}

	list<SMPBasicBlock *>::iterator BlockIter;
	for (BlockIter = this->Blocks.begin(); BlockIter != this->Blocks.end(); ++BlockIter) {
		SMPBasicBlock *CurrBlock = (*BlockIter);
		delete CurrBlock;
	}
// Get a non-stale pointer to the func_t info for the current function.
func_t *SMPFunction::GetFuncInfo(void) {
	func_t *myPtr = SMP_get_func(this->FirstEA);
// Reset the Processed flags in all blocks to false.
void SMPFunction::ResetProcessedBlocks(void) {
	list<SMPBasicBlock *>::iterator CurrBlock;
	for (CurrBlock = this->Blocks.begin(); CurrBlock != this->Blocks.end(); ++CurrBlock) {
		(*CurrBlock)->SetProcessed(false);
	}
	return;
} // end of SMPFunction::ResetProcessedBlocks()

// Return an iterator for the beginning of the LiveInSet. 
set<op_t, LessOp>::iterator SMPFunction::GetFirstLiveIn(void) {
	return this->LiveInSet.begin();
} // end of SMPBasicBlock::GetFirstLiveIn()

// Get termination iterator marker for the LiveIn set, for use by predecessors.
set<op_t, LessOp>::iterator SMPFunction::GetLastLiveIn(void) {
	return this->LiveInSet.end();
}

// Get iterator for the start of the LiveOut set.
set<op_t, LessOp>::iterator SMPFunction::GetFirstLiveOut(void) {
	return this->LiveOutSet.begin();
}

// Get termination iterator marker for the LiveOut set.
set<op_t, LessOp>::iterator SMPFunction::GetLastLiveOut(void) {
	return this->LiveOutSet.end();
}

// Get iterator for the start of the VarKill set.
set<op_t, LessOp>::iterator SMPFunction::GetFirstVarKill(void) {
	return this->KillSet.begin();
}

// Get termination iterator marker for the VarKill set.
set<op_t, LessOp>::iterator SMPFunction::GetLastVarKill(void) {
	return this->KillSet.end();
}

// Four methods to get values from the maps of global reg/SSA to FG info.
//  For local names, see corresponding methods in SMPBasicBlock.
unsigned short SMPFunction::GetDefSignMiscInfo(int DefHashValue) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalDefFGInfoBySSA.find(DefHashValue);
	if (MapIter != this->GlobalDefFGInfoBySSA.end())
		return MapIter->second.SignMiscInfo;
	else
		return 0;
} // end of SMPFunction::GetDefSignMiscInfo()

unsigned short SMPFunction::GetUseSignMiscInfo(int UseHashValue) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalUseFGInfoBySSA.find(UseHashValue);
	if (MapIter != this->GlobalUseFGInfoBySSA.end())
		return MapIter->second.SignMiscInfo;
	else
		return 0;
} // end of SMPFunction::GetUseSignMiscInfo()

unsigned short SMPFunction::GetDefWidthTypeInfo(int DefHashValue) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalDefFGInfoBySSA.find(DefHashValue);
	if (MapIter != this->GlobalDefFGInfoBySSA.end())
		return MapIter->second.SizeInfo;
	else
		return 0;
} // end of SMPFunction::GetDefWidthTypeInfo()

unsigned short SMPFunction::GetUseWidthTypeInfo(int UseHashValue) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalUseFGInfoBySSA.find(UseHashValue);
	if (MapIter != this->GlobalUseFGInfoBySSA.end())
		return MapIter->second.SizeInfo;
	else
		return 0;
} // end of SMPFunction::GetUseWidthTypeInfo()

struct FineGrainedInfo SMPFunction::GetDefFGInfo(int DefHashValue) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalDefFGInfoBySSA.find(DefHashValue);
	if (MapIter != this->GlobalDefFGInfoBySSA.end())
		return MapIter->second;
	else {
		struct FineGrainedInfo EmptyFG;
		EmptyFG.SignMiscInfo = 0;
		EmptyFG.SizeInfo = 0;
		return EmptyFG;
	}
} // end of SMPFunction::GetDefFGInfo()

struct FineGrainedInfo SMPFunction::GetUseFGInfo(int UseHashValue) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalUseFGInfoBySSA.find(UseHashValue);
	if (MapIter != this->GlobalUseFGInfoBySSA.end())
		return MapIter->second;
	else {
		struct FineGrainedInfo EmptyFG;
		EmptyFG.SignMiscInfo = 0;
		EmptyFG.SizeInfo = 0;
		return EmptyFG;
	}
} // end of SMPFunction::GetUseFGInfo()

// Add a caller to the list of all callers of this function.
void SMPFunction::AddCallSource(ea_t addr) {
	// Convert call instruction address to beginning address of the caller.
		SMP_msg("SERIOUS WARNING: Call location %x not in a function.\n", addr);
		return;
	}
	ea_t FirstAddr = FuncInfo->startEA;
	assert(BADADDR != FirstAddr);
	this->AllCallSources.insert(FirstAddr);
	return;
} // end of SMPFunction::AddCallSource()

// Six methods to set values into the maps of global reg/SSA to FG info.
//  For local names, see corresponding methods in SMPBasicBlock.
void SMPFunction::UpdateDefSignMiscInfo(int DefHashValue, unsigned short NewInfo) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalDefFGInfoBySSA.find(DefHashValue);
	if (MapIter == this->GlobalDefFGInfoBySSA.end()) {
		// Not found; insert first.
		struct FineGrainedInfo NewFGInfo;
		NewFGInfo.SignMiscInfo = NewInfo;
		NewFGInfo.SizeInfo = 0;
		pair<int, struct FineGrainedInfo> MapItem(DefHashValue, NewFGInfo);
		MapResult = this->GlobalDefFGInfoBySSA.insert(MapItem);
		assert(MapResult.second); // Was not previously found, insertion must work.
	}
	else { // found; just OR in the new bits.
		MapIter->second.SignMiscInfo |= NewInfo;
	}

	return;
} // end of SMPFunction::UpdateDefSignMiscInfo()

void SMPFunction::UpdateUseSignMiscInfo(int UseHashValue, unsigned short NewInfo) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalUseFGInfoBySSA.find(UseHashValue);
	if (MapIter == this->GlobalUseFGInfoBySSA.end()) {
		// Not found; insert first.
		struct FineGrainedInfo NewFGInfo;
		NewFGInfo.SignMiscInfo = NewInfo;
		NewFGInfo.SizeInfo = 0;
		pair<int, struct FineGrainedInfo> MapItem(UseHashValue, NewFGInfo);
		MapResult = this->GlobalUseFGInfoBySSA.insert(MapItem);
		assert(MapResult.second); // Was not previously found, insertion must work.
	}
	else { // found; just OR in the new bits.
		MapIter->second.SignMiscInfo |= NewInfo;
	}

	return;
} // end of SMPFunction::UpdateUseSignMiscInfo()

void SMPFunction::UpdateDefWidthTypeInfo(int DefHashValue, unsigned short NewInfo) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalDefFGInfoBySSA.find(DefHashValue);
	if (MapIter == this->GlobalDefFGInfoBySSA.end()) {
		// Not found; insert first.
		struct FineGrainedInfo NewFGInfo;
		NewFGInfo.SignMiscInfo = 0;
		NewFGInfo.SizeInfo = NewInfo;
		pair<int, struct FineGrainedInfo> MapItem(DefHashValue, NewFGInfo);
		MapResult = this->GlobalDefFGInfoBySSA.insert(MapItem);
		assert(MapResult.second); // Was not previously found, insertion must work.
	}
	else { // found; just OR in the new bits.
		MapIter->second.SizeInfo |= NewInfo;
	}

	return;
} // end of SMPFunction::UpdateDefWidthTypeInfo()

void SMPFunction::UpdateUseWidthTypeInfo(int UseHashValue, unsigned short NewInfo) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalUseFGInfoBySSA.find(UseHashValue);
	if (MapIter == this->GlobalUseFGInfoBySSA.end()) {
		// Not found; insert first.
		struct FineGrainedInfo NewFGInfo;
		NewFGInfo.SignMiscInfo = 0;
		NewFGInfo.SizeInfo = NewInfo;
		pair<int, struct FineGrainedInfo> MapItem(UseHashValue, NewFGInfo);
		MapResult = this->GlobalUseFGInfoBySSA.insert(MapItem);
		assert(MapResult.second); // Was not previously found, insertion must work.
	}
	else { // found; just OR in the new bits.
		MapIter->second.SizeInfo |= NewInfo;
	}

	return;
} // end of SMPFunction::UpdateUseWidthTypeInfo()

void SMPFunction::UpdateDefFGInfo(int DefHashValue, struct FineGrainedInfo NewFG) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalDefFGInfoBySSA.find(DefHashValue);
	if (MapIter == this->GlobalDefFGInfoBySSA.end()) {
		// Not found; insert it.
		pair<int, struct FineGrainedInfo> MapItem(DefHashValue, NewFG);
		MapResult = this->GlobalDefFGInfoBySSA.insert(MapItem);
		assert(MapResult.second); // Was not previously found, insertion must work.
	}
	else { // found; just put in the new bits.
		MapIter->second.SignMiscInfo |= NewFG.SignMiscInfo;
		MapIter->second.SizeInfo |= NewFG.SizeInfo;
	}

	return;
} // end of SMPFunction::UpdateDefFGInfo()

void SMPFunction::UpdateUseFGInfo(int UseHashValue, struct FineGrainedInfo NewFG) {
	map<int, struct FineGrainedInfo>::iterator MapIter;
	pair<map<int, struct FineGrainedInfo>::iterator, bool> MapResult;

	MapIter = this->GlobalUseFGInfoBySSA.find(UseHashValue);
	if (MapIter == this->GlobalUseFGInfoBySSA.end()) {
		// Not found; insert it.
		pair<int, struct FineGrainedInfo> MapItem(UseHashValue, NewFG);
		MapResult = this->GlobalUseFGInfoBySSA.insert(MapItem);
		assert(MapResult.second); // Was not previously found, insertion must work.
	}
	else { // found; just put in the new bits.
		MapIter->second.SignMiscInfo |= NewFG.SignMiscInfo;
		MapIter->second.SizeInfo |= NewFG.SizeInfo;
	}

	return;
} // end of SMPFunction::UpdateUseFGInfo()

// Erase a range of instructions from the Instrs list, usually corresponding
//  the the range of a basic block.
void SMPFunction::EraseInstRange(ea_t FirstAddr, ea_t LastAddr) {
	list<SMPInstr *>::iterator InstIter = this->Instrs.begin();
	SMPInstr *CurrInst;
	ea_t InstAddr;

	while (InstIter != this->Instrs.end()) {
		CurrInst = (*InstIter);
		InstAddr = CurrInst->GetAddr();
		if ((InstAddr >= FirstAddr) && (InstAddr <= LastAddr)) {
			InstIter = this->Instrs.erase(InstIter);
		}
		else {
			++InstIter;
		}
	}
} // end of SMPFunction::EraseInstRange()

// Get stack delta from a callee function that is unable to provide the info from
//  its own analyses (e.g. analyses failed or have not been performed yet, due to
//  a mutually recursive clique in the call graph). We have three approaches in
//  this case: Use a default value, consult IDA Pro's analyses, or see if we can
//  detect a stack adjustment after the call instruction, from which we could infer
//  the stack delta of the callee.
sval_t SMPFunction::GetStackDeltaForCallee(ea_t CallAddr) {
	bool success = false;
	sval_t CalleeDelta = CALLING_CONVENTION_DEFAULT_FUNCTION_STACK_DELTA;

#if 0
#ifdef STARS_IDA_INTERFACE
	if (this->FuncInfo.flags & (FUNC_SP_READY | FUNC_PURGED_OK)) {
		// FUNC_SP_READY => AnalyzedSP; FUNC_PURGED_OK => analysis of effect on stack succeeded.
		if (this->FuncInfo.argsize != 0) {
		}
	}
#endif // STARS_IDA_INTERFACE
#endif

	SMPBasicBlock *CallBlock = this->GetBlockFromInstAddr(CallAddr);
	sval_t BlockAnalysisDelta = CallBlock->ComputeStackAdjustmentAfterCall(CallAddr);
	if (0 != BlockAnalysisDelta) {
		CalleeDelta -= BlockAnalysisDelta;
		SMP_msg("INFO: Block analysis produced callee delta of %d bytes after %x\n", CalleeDelta, CallAddr);
	}

	return CalleeDelta;
} // end of SMPFunction::GetStackDeltaForCallee()

// Analyze changes to the stack pointer over all instructions.
bool SMPFunction::AnalyzeStackPointerDeltas(void) {
	list<pair<SMPBasicBlock *, sval_t> > WorkList;
	list<SMPInstr *>::iterator InstIter;
	SMPInstr *CurrInst;
	sval_t CurrentDelta = 0;
	bool ConsistentNetDelta = true; // Net change to stack pointer is consistent at all RETURN locations
	bool ReturnSeen = false;
	bool IDAProSucceeded = this->AnalyzedSP;
	bool FirstBlockProcessed = false;
	bool FPSaved = false; // found save of frame pointer; prelude to initializing it
	bool SPintoFP = false; // found move of stack pointer into frame pointer (FP init)

#if SMP_COMPARE_IDA_STARS_STACK_POINTER_DELTAS
	bool DebugFlag = (0 == strcmp("find_derivation", this->GetFuncName()));
	bool TraceFlag = (0 == strcmp("_dl_start_profile", this->GetFuncName()));
	bool IDATraceFlag = (0 == strcmp("do_length", this->GetFuncName()));
#endif

#if 1
	// Temporarily pull the functions that call alloca out of the stack pointer delta computations, so
	//  that we can focus on solving other problems.
	if (this->CallsAlloca) {
		if (!this->AnalyzedSP) {
			SMP_msg("INFO: Leaving alloca-calling function %s stack pointer deltas unanalyzed.\n", this->GetFuncName());
			return false; // leave it unsolved
		}
		else {
			SMP_msg("INFO: Using IDA Pro stack pointer deltas for alloca-calling function %s .\n", this->GetFuncName());
			InstIter = this->Instrs.begin();
#if SMP_USE_SSA_FNOP_MARKER
			++InstIter; // skip marker pseudo-instruction
#endif
			while (InstIter != this->Instrs.end()) {
				CurrInst = *InstIter;
				sval_t IDAProDelta = get_spd(this->GetFuncInfo(), CurrInst->GetAddr());
				CurrInst->SetStackPtrOffset(IDAProDelta);
				++InstIter;
				if (IDATraceFlag) {
					SMP_msg("INFO: IDA Pro stack delta trace: %d at %x\n", IDAProDelta, CurrInst->GetAddr());
				}
			}
			return true; // leave it solved
		}
	}
#endif

	// Mark all blocks as unprocessed
	this->ResetProcessedBlocks();

	this->AnalyzedSP = true;

	// Put the entry block on the work list.
	assert(0 < this->Blocks.size());
	pair<SMPBasicBlock *, sval_t> WorkingPair (this->Blocks.front(), CurrentDelta);
	WorkList.push_back(WorkingPair);

	// While blocks exist on the work list
	//  if block already processed, confirm that we are re-entering
	//    the block with the same stack pointer delta as previously,
	//    and pop it off the work list
	//    otherwise declare the stack pointer to be un-analyzeable;
	//  else
	//     iterate through all instructions in the block, analyzing
	//     the stack pointer delta of each inst and accumulating current delta
	//     At the end of the block, put the successor blocks on the work list.
	do {
		SMPBasicBlock *CurrBlock = WorkList.front().first;
		sval_t IncomingDelta = WorkList.front().second;

		if (CurrBlock->IsProcessed()) { // already processed
			InstIter = CurrBlock->GetFirstInstr();
			sval_t PrevIncomingDelta = (*InstIter)->GetStackPtrOffset();
			if (IncomingDelta == PrevIncomingDelta) {
				// No error, already processed.
				WorkList.pop_front(); // discard already processed block.
			}
			else if (this->CallsAlloca) {
				// Calls to alloca() become additional stack allocations, which can produce
				//  multiple possible stack deltas for an instruction if different paths
				//  to the instruction do not hit the same alloca() calls, so it is not
				//  an error to have conflicting deltas in the functions that call alloca().
				//  We want to converge to the smallest magnitude deltas, which are the greatest
				//  values because the deltas are negative. This is the opposite of IDA Pro, which
				//  seems to use the largest stack deltas it has seen.
				if (PrevIncomingDelta >= IncomingDelta) {
					// Old incoming delta should be retained.
					WorkList.pop_front(); // discard already processed block.
				}
				else {
					CurrBlock->SetProcessed(false);
					continue;  // Make the loop come around and process this block again, using
							   //  the new incoming delta. Because we do this only when it decreases
							   //  the stack size as seen by this block, no infinite loop is possible.
				}
			}
			else {
				this->AnalyzedSP = false;
				SMP_msg("ERROR: Stack delta: PrevIncoming is %d NewIncoming is %d at %x\n",
					PrevIncomingDelta, IncomingDelta, (*InstIter)->GetAddr());
				WorkList.clear();
			}
		}
		else { // not already processed
			CurrBlock->SetProcessed(true);
			WorkList.pop_front();
			for (InstIter = CurrBlock->GetFirstInstr(); InstIter != CurrBlock->GetLastInstr(); ++InstIter) {
				CurrInst = (*InstIter);
				if (CurrInst->GetAddr() == this->GetFirstFrameAllocInstAddr()) {
					// Record the reset point for frame deallocations
					this->PreAllocStackDelta = IncomingDelta;
				}
#if 0
				if (this->CallsAlloca) { // keep deltas in a set; paths can lead to multiple deltas per inst
					bool DeltaAlreadyFound = CurrInst->FindStackPtrDelta(IncomingDelta);
					CurrInst->SetStackPtrOffset(IncomingDelta);
#if 0
				}
#endif
#if SMP_COMPARE_IDA_STARS_STACK_POINTER_DELTAS
				if (DebugFlag && IDAProSucceeded && !this->CallsAlloca) {
					sval_t IDAProDelta = get_spd(this->GetFuncInfo(), CurrInst->GetAddr());
					if ((IDAProDelta != IncomingDelta) && (!CurrInst->MDIsHaltInstr())) {
						// IDA Pro special-cases the HALT instruction to make it appear that the
						//  incoming stack delta is zero. We do no such special case delta adjudstment,
						//  so we suppress error messages, as our delta will be non-zero.
						SMP_msg("ERROR: At %x IDA Pro has stack pointer delta of %d and we compute %d\n", CurrInst->GetAddr(),
							IDAProDelta, IncomingDelta);
					}
				}
				if (TraceFlag) {
					SMP_msg("INFO: Stack delta trace: %d at %x\n", IncomingDelta, CurrInst->GetAddr());
				}
#endif
				if (!FirstBlockProcessed) { // Look for initialization of frame pointer, record its stack delta
					FirstBlockProcessed = CurrInst->IsLastInBlock();
					if (!FPSaved) { // still looking for "push <framepointerreg>"
						if (CurrInst->MDIsPushInstr() && CurrInst->GetCmd().Operands[0].is_reg(MD_FRAME_POINTER_REG)) {
							FPSaved = true;
						}
					}
					else if (!SPintoFP) { // found "push <framepointerreg>", looking for "fp := sp"
						insn_t CurrCmd = CurrInst->GetCmd();
						if ((CurrCmd.itype == MD_MOVE_INSTRUCTION) 
							&& (CurrInst->GetFirstDef()->GetOp().is_reg(MD_FRAME_POINTER_REG))
							&& (CurrInst->GetFirstUse()->GetOp().is_reg(MD_STACK_POINTER_REG))) {
							SPintoFP = true;
							this->FramePointerStackDelta = IncomingDelta;
							FirstBlockProcessed = true; // stop looking
							assert(this->UsesFramePointer());
						}
					}
				}

				CurrentDelta = CurrInst->AnalyzeStackPointerDelta(IncomingDelta, this->GetFramePtrStackDelta());
				if (SMP_STACK_POINTER_BITWISE_AND_CODE == CurrentDelta) {
					// For now, we ignore instructions that AND a constant into the stack pointer.
					CurrentDelta = 0;
					SMP_msg("WARNING: Stack pointer bitwise AND ignored at %x\n", CurrInst->GetAddr());
				}
				else if (SMP_STACK_DELTA_ERROR_CODE == CurrentDelta) {
					this->AnalyzedSP = false;
					SMP_msg("ERROR: Stack delta unanalyzeable at %x\n", CurrInst->GetAddr());
					WorkList.clear();
					break;
				}
				IncomingDelta += CurrentDelta;
				if ((RETURN == CurrInst->GetDataFlowType()) && (!CurrInst->IsCondTailCall())) {
					// We hope to see a consistent outgoing delta from all RETURN points.
					//  We special-case the conditional jump used as tail call, because it must be followed
					//  by a real return instruction later. If the jump is taken, it acts as a return, but
					//  it has not yet popped the stack.
					if (ReturnSeen) { // This is not the first RETURN seen.
						if (IncomingDelta != this->NetStackDelta) { // Inconsistent
							SMP_msg("ERROR: Inconsistent stack deltas at return instruction at %x\n", CurrInst->GetAddr());
							ConsistentNetDelta = false;
							this->AnalyzedSP = false;
							WorkList.clear();
							break;
						}
					}
					else { // First RETURN statement seen.
						ReturnSeen = true;
						this->NetStackDelta = IncomingDelta;
#if SMP_AUDIT_STACK_POINTER_DELTAS
						if (0 != IncomingDelta) {
							SMP_msg("WARNING: Stack delta not zero after return instruction at %x\n", CurrInst->GetAddr());
						}
			}
			// Push the successor blocks onto the work list
			if (this->AnalyzedSP) { // if we do not have an error already
				list<SMPBasicBlock *>::iterator SuccIter;
				for (SuccIter = CurrBlock->GetFirstSucc(); SuccIter != CurrBlock->GetLastSucc(); ++SuccIter) {
					pair<SMPBasicBlock *, sval_t> SuccPair (*SuccIter, IncomingDelta);
					WorkList.push_back(SuccPair);
				}
			}
		}
	} while (!WorkList.empty());

	this->STARSStackPtrAnalysisPerformed = true;
	if (this->AnalyzedSP && (CALLING_CONVENTION_DEFAULT_FUNCTION_STACK_DELTA != this->NetStackDelta)) {
		SMP_msg("WARNING: Non-default stack ptr delta %d for function: %s\n", this->NetStackDelta, this->GetFuncName());
	}

	if (IDAProSucceeded) {
		if (!this->AnalyzedSP) {
			SMP_msg("ERROR: Stack Ptr Delta Analysis succeeded in IDA, failed in STARS for %x : %s\n", this->FirstEA,
				this->GetFuncName());
		}
	}
	else {
		if (this->AnalyzedSP) {
			SMP_msg("SUCCESS: Stack Ptr Delta Analysis failed in IDA, succeeded in STARS for %x : %s\n", this->FirstEA,
				this->GetFuncName());
		}
	}
	return this->AnalyzedSP;
} // end of SMPFunction::AnalyzeStackPointerDeltas()

void SMPFunction::FindAllAllocsAndDeallocs(void) {
	bool FoundAllocInstr = false;
	bool FoundDeallocInstr = false;
clc5q's avatar
clc5q committed
	bool DebugFlag = false;
#if SMP_DEBUG_FRAMEFIXUP
	DebugFlag |= (0 == strcmp("frame_dummy", this->GetFuncName()));
#endif

	// Now, if LocalVarsSize is not zero, we need to find the instruction
	//  in the function prologue that allocates space on the stack for
	//  local vars. This code could be made more robust in the future
	//  by matching LocalVarsSize to the immediate value in the allocation
	//  instruction. However, IDA Pro is sometimes a little off on this
	//  number. **!!**
	if (0 < this->LocalVarsSize) {
		if (DebugFlag) SMP_msg("Searching for alloc and dealloc\n");
		list<SMPInstr *>::iterator InstIter = this->Instrs.begin();
#if SMP_USE_SSA_FNOP_MARKER
		++InstIter;  // skip marker instruction
		for ( ; InstIter != this->Instrs.end(); ++InstIter) {
			SMPInstr *CurrInst = (*InstIter);
			ea_t addr = CurrInst->GetAddr();

			// Keep the most recent instruction in the DeallocInstr
			//  in case we reach the return without seeing a dealloc.
			if (!FoundDeallocInstr) {
				this->LocalVarsDeallocInstr = addr;
			}

			if (!FoundAllocInstr
				&& CurrInst->MDIsFrameAllocInstr()) {
#if SMP_DEBUG_CONTROLFLOW
				SMP_msg("Returned from MDIsFrameAllocInstr()\n");
				this->LocalVarsAllocInstr = addr;
				FoundAllocInstr = true;
				if (DebugFlag) SMP_msg("Found alloc: %s\n", CurrInst->GetDisasm());
				// As soon as we have found the local vars allocation,
				//  we can try to fix incorrect sets of UseFP by IDA.
				// NOTE: We might want to extend this in the future to
				//  handle functions that have no locals.  **!!**
				bool FixedUseFP = MDFixUseFP();
#if SMP_DEBUG_CONTROLFLOW
#if SMP_DEBUG_FRAMEFIXUP
				if (FixedUseFP) {
					SMP_msg("Fixed UseFP in %s\n", this->GetFuncName());
				}
#endif
			}
			else if (FoundAllocInstr) {
				// We can now start searching for the DeallocInstr.
				if (CurrInst->MDIsFrameDeallocInstr(UseFP, this->LocalVarsSize)) {
					// Keep saving the most recent addr that looks
					//  like the DeallocInstr until we reach the
					//  end of the function. Last one to look like
					//  it is used as the DeallocInstr.
#if SMP_DEBUG_CONTROLFLOW
					SMP_msg("Returned from MDIsFrameDeallocInstr()\n");
					this->LocalVarsDeallocInstr = addr;
					FoundDeallocInstr = true;
				}
					if (DebugFlag) SMP_msg("Not dealloc: %s\n", CurrInst->GetDisasm());
		} // end for (list<SMPInstr *>::iterator InstIter ... )
		if (!FoundAllocInstr) {
			// Could not find the frame allocating instruction.  Bad.
			// See if we can find the point at which the stack allocation reaches
			//  a total of FuncInfo.frsize+frregs, regardless of whether it happened by push
			//  instructions or some other means.
			this->LocalVarsAllocInstr = this->FindAllocPoint(this->FuncInfo.frsize + this->FuncInfo.frregs);
#if SMP_DEBUG_CONTROLFLOW
			SMP_msg("Returned from FindAllocPoint()\n");
#if SMP_DEBUG_FRAMEFIXUP
			if (BADADDR == this->LocalVarsAllocInstr) {
				SMP_msg("ERROR: Could not find stack frame allocation in %s\n",
					this->GetFuncName());
				SMP_msg("LocalVarsSize: %d  SavedRegsSize: %d ArgsSize: %d\n",
					LocalVarsSize, CalleeSavedRegsSize, IncomingArgsSize);
			}
			else {
				SMP_msg("FindAllocPoint found %x for function %s\n",
					this->LocalVarsAllocInstr, this->GetFuncName());
			}
#endif
		}
#if SMP_DEBUG_FRAMEFIXUP
		if (!FoundDeallocInstr) {
			// Could not find the frame deallocating instruction.  Bad.
			// Emit diagnostic and use the last instruction in the
			// function.
			SMP_msg("ERROR: Could not find stack frame deallocation in %s\n",
				this->GetFuncName());
		}
#endif
	}
	// else LocalVarsSize was zero, meaning that we need to search 
	//  for the end of the function prologue code and emit stack frame
	//  annotations from that address (i.e. this method returns that
	//  address). We will approximate this by finding the end of the
	//  sequence of PUSH instructions at the beginning of the function.
	//  The last PUSH instruction should be the last callee-save-reg
	//  instruction. We can make this more robust in the future by
	//  making sure that we do not count a PUSH of anything other than
	//  a register. **!!**
	// NOTE: 2nd prologue instr is usually mov ebp,esp
	// THE ASSUMPTION THAT WE HAVE ONLY PUSH INSTRUCTIONS BEFORE
	// THE ALLOCATING INSTR IS ONLY TRUE WHEN LOCALVARSSIZE == 0;
	else {
		ea_t SaveAddr = this->FuncInfo.startEA;
		list<SMPInstr *>::iterator InstIter = this->Instrs.begin();
#if SMP_USE_SSA_FNOP_MARKER
		++InstIter;  // skip marker instruction
		for ( ;	InstIter != this->Instrs.end(); ++InstIter) {
			SMPInstr *CurrInst = (*InstIter);
			insn_t CurrCmd = CurrInst->GetCmd();
			ea_t addr = CurrInst->GetAddr();
			if (CurrCmd.itype == NN_push)
				SaveAddr = addr;
			else
				break;
		}
		this->LocalVarsAllocInstr = SaveAddr;
		this->LocalVarsDeallocInstr = 0;
	} // end if (LocalVarsSize > 0) ... else ...

	return;
} // end of SMPFunction::FindAllAllocsAndDeallocs()

// Figure out the different regions of the stack frame, and find the
//  instructions that allocate and deallocate the local variables space
//  on the stack frame.
// The stack frame info will be used to emit stack
//  annotations when Analyze() reaches the stack allocation
//  instruction that sets aside space for local vars.
// Set the address of the instruction at which these
//  annotations should be emitted. This should normally
//  be an instruction such as:  sub esp,48
//  However, for a function with no local variables at all,
//  we will need to determine which instruction should be
//  considered to be the final instruction of the function
//  prologue and return its address.
// Likewise, we find the stack deallocating instruction in
//  the function epilogue.
void SMPFunction::SetStackFrameInfo(void) {
	for (list<SMPInstr *>::iterator InstIter = this->Instrs.begin(); InstIter != this->Instrs.end(); ++InstIter) {
		// We can finally search for stack loads now that UseFP has been fixed by
		//  MDFixUseFP(). Otherwise, we would do this in SMPInstr::Analyze(),
		//  but the UseFP flag is not ready that early.
		(*InstIter)->MDFindLoadFromStack(this->UseFP);

		// Fix up machine dependent quirks in the def and use lists.
		//  This used to be called from within SMPInstr.Analyze(), but info such as UseFP
		//  is not available that early.
		(*InstIter)->MDFixupDefUseLists();
#if SMP_COMPUTE_STACK_GRANULARITY
	// Now, find the boundaries between local variables.
	this->BuildLocalVarTable();
#endif

	// Get callee-saved regs info for remediation use.
	if (BADADDR != this->GetFirstFrameAllocInstAddr()) {
	return;
} // end of SMPFunction::SetStackFrameInfo()

// IDA Pro defines the sizes of regions in the stack frame in a way
//  that suits its purposes but not ours. The frsize field of the func_info_t
//  structure measures the distance between the stack pointer and the
//  frame pointer (ESP and EBP in the x86). This region includes some
//  of the callee-saved registers. So, the frregs field only includes
//  the callee-saved registers that are above the frame pointer.
//  x86 standard prologue on gcc/linux:
//    push ebp      ; save old frame pointer
//    mov ebp,esp   ; new frame pointer = current stack pointer
//    push esi      ; callee save reg
//    push edi      ; callee save reg
//    sub esp,34h   ; allocate 52 bytes for local variables
//
//  Notice that EBP acquires its final frame pointer value AFTER the
//  old EBP has been pushed. This means that, of the three callee saved
//  registers, one is above where EBP points and two are below.
//  IDA Pro is concerned with generating readable addressing expressions
//  for items on the stack. None of the callee-saved regs will ever
//  be addressed in the function; they will be dormant until they are popped
//  off the stack in the function epilogue. In order to create readable
//  disassembled code, IDA defines named constant offsets for locals. These
//  offsets are negative values (x86 stack grows downward from EBP toward
//  ESP). When ESP_relative addressing occurs, IDA converts a statement: