Newer
Older
/*
* 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/
*
//
// SMPFunction.cpp
//
// This module performs the fundamental data flow analyses needed for the
// SMP project (Software Memory Protection) at the function level.
//
#include <utility>
#include <list>
#include <set>
#include <vector>
#include <algorithm>
#include <cstring>
#include <cstdlib>
#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_FRAMEFIXUP 0
#define SMP_DEBUG_DATAFLOW 0
#define SMP_DEBUG_DATAFLOW_VERBOSE 0
#define SMP_DEBUG_TYPE_INFERENCE 0
#define SMP_DEBUG_STACK_GRANULARITY 0
#define SMP_DEBUG_BUILD_RTL 1 // leave this on; serious errors reported
#define SMP_DEBUG_UNINITIALIZED_SSA_NAMES 1
clc5q
committed
#define SMP_OPTIMIZE_BLOCK_PROFILING 0
// Compute LVA/SSA or not? Turn it off for NICECAP demo on 31-JAN-2008
#define SMP_COMPUTE_LVA_SSA 1
// Compute fine-grained stack boundaries?
#define SMP_COMPUTE_STACK_GRANULARITY 1
// Insert a floating no-op instruction at top of each function to hold SSA DEFs
// of LiveIn names?
#define SMP_USE_SSA_FNOP_MARKER 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>
clc5q
committed
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;
}
// *****************************************************************
// Class SMPFunction
// *****************************************************************
// Constructor
SMPFunction::SMPFunction(func_t *Info, SMPProgram* pgm) {
clc5q
committed
this->Program = pgm;
this->FuncInfo = *Info;
clc5q
committed
this->FirstEA = this->FuncInfo.startEA;
clc5q
committed
this->FuncName[0] = '\0';
this->BlockCount = 0;
this->UseFP = false;
this->StaticFunc = false;
this->LibFunc = false;
this->IndirectCalls = false;
this->UnresolvedIndirectCalls = false;
this->IndirectJumps = false;
this->UnresolvedIndirectJumps = false;
this->DirectlyRecursive = false;
this->SharedChunks = false;
clc5q
committed
this->AnalyzedSP = false;
#if 1 // default to unsafe
this->SafeFunc = false;
#else // default to safe
this->SafeFunc = true;
this->SpecSafeFunc = true;
this->SafeCallee = true;
this->SpecSafeCallee = true;
#endif
this->WritesAboveRA = false;
this->HasIndirectWrites = false;
this->OutgoingArgsComputed = false;
clc5q
committed
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->LocalVarOffsetLimit = 0;
this->ReturnAddrStatus = FUNC_UNKNOWN;
this->SetIsSpeculative(false);
this->Blocks.clear();
this->DirectCallTargets.clear();
this->IndirectCallTargets.clear();
this->AllCallTargets.clear();
clc5q
committed
this->AllCallSources.clear();
this->InstBlockMap.clear();
this->RPOBlocks.clear();
this->IDom.clear();
this->DomTree.clear();
this->FineGrainedStackTable.clear();
clc5q
committed
this->SavedRegLoc.clear();
this->ReturnRegTypes.clear();
clc5q
committed
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);
}
} // end of SMPFunction() constructor
clc5q
committed
// Get a non-stale pointer to the func_t info for the current function.
func_t *SMPFunction::GetFuncInfo(void) {
func_t *myPtr = get_func(this->FirstEA);
assert(NULL != myPtr);
return myPtr;
}
// 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()
clc5q
committed
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
// 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();
}
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
// 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()
clc5q
committed
// 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.
func_t *FuncInfo = get_func(addr);
if (NULL == FuncInfo) {
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()
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
// Four methods to get 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()
// 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) {
bool FoundAllocInstr = false;
bool FoundDeallocInstr = false;
bool DebugFlag = false;
#if SMP_DEBUG_FRAMEFIXUP
DebugFlag |= (0 == strcmp(".init_proc", this->GetFuncName()));
#endif
// 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().
clc5q
committed
this->RetAddrSize = get_frame_retsize(this->GetFuncInfo());
// IDA Pro has trouble with functions that do not have any local
// variables. Unfortunately, the C library has plenty of these
// functions. IDA usually claims that frregs is zero and frsize
// is N, when the values should have been reversed. We can attempt
// to detect this and fix it.
bool FrameInfoFixed = this->MDFixFrameInfo();
#if SMP_DEBUG_CONTROLFLOW
msg("Returned from MDFixFrameInfo()\n");
#endif
#if SMP_DEBUG_FRAMEFIXUP
if (FrameInfoFixed) {
msg("Fixed stack frame size info: %s\n", this->FuncName);
SMPBasicBlock CurrBlock = this->Blocks.front();
msg("First basic block:\n");
for (list<list<SMPInstr>::iterator>::iterator CurrInstr = CurrBlock.GetFirstInstr();
CurrInstr != CurrBlock.GetLastInstr();
++CurrInstr) {
msg("%s\n", (*CurrInstr)->GetDisasm());
}
}
#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) msg("Searching for alloc and dealloc\n");
for (list<SMPInstr>::iterator CurrInstr = this->Instrs.begin();
CurrInstr != this->Instrs.end();
++CurrInstr) {
#if SMP_USE_SSA_FNOP_MARKER
if (this->Instrs.begin() == CurrInstr)
continue; // skip marker instruction
#endif
ea_t addr = CurrInstr->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
&& CurrInstr->MDIsFrameAllocInstr()) {
#if SMP_DEBUG_CONTROLFLOW
msg("Returned from MDIsFrameAllocInstr()\n");
#endif
this->LocalVarsAllocInstr = addr;
FoundAllocInstr = true;
if (DebugFlag) msg("Found alloc: %s\n", CurrInstr->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
msg("Returned from MDFixUseFP()\n");
#endif
#if SMP_DEBUG_FRAMEFIXUP
if (FixedUseFP) {
msg("Fixed UseFP in %s\n", this->FuncName);
}
#endif
}
else if (FoundAllocInstr) {
// We can now start searching for the DeallocInstr.
if (CurrInstr->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
msg("Returned from MDIsFrameDeallocInstr()\n");
#endif
this->LocalVarsDeallocInstr = addr;
FoundDeallocInstr = true;
}
else {
if (DebugFlag) msg("Not dealloc: %s\n", CurrInstr->GetDisasm());
}
}
} // end for (list<SMPInstr>::iterator CurrInstr ... )
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
msg("Returned from FindAllocPoint()\n");
#endif
#if SMP_DEBUG_FRAMEFIXUP
if (BADADDR == this->LocalVarsAllocInstr) {
msg("ERROR: Could not find stack frame allocation in %s\n",
FuncName);
msg("LocalVarsSize: %d SavedRegsSize: %d ArgsSize: %d\n",
LocalVarsSize, CalleeSavedRegsSize, IncomingArgsSize);
}
else {
msg("FindAllocPoint found %x for function %s\n",
this->LocalVarsAllocInstr, this->GetFuncName());
}
#endif
}
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
if (!FoundDeallocInstr) {
// Could not find the frame deallocating instruction. Bad.
// Emit diagnostic and use the last instruction in the
// function.
msg("ERROR: Could not find stack frame deallocation in %s\n",
FuncName);
}
#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;
for (list<SMPInstr>::iterator CurrInstr = this->Instrs.begin();
CurrInstr != this->Instrs.end();
++CurrInstr) {
#if SMP_USE_SSA_FNOP_MARKER
if (this->Instrs.begin() == CurrInstr)
continue; // skip marker instruction
#endif
insn_t CurrCmd = CurrInstr->GetCmd();
ea_t addr = CurrInstr->GetAddr();
if (CurrCmd.itype == NN_push)
SaveAddr = addr;
else
break;
}
this->LocalVarsAllocInstr = SaveAddr;
this->LocalVarsDeallocInstr = 0;
} // end if (LocalVarsSize > 0) ... else ...
this->CallsAlloca = this->FindAlloca();
#if SMP_COMPUTE_STACK_GRANULARITY
// Now, find the boundaries between local variables.
this->BuildLocalVarTable();
#endif
// Get callee-saved regs info for remediation use.
if (FoundAllocInstr) {
this->MDFindSavedRegs();
}
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:
// mov eax,[esp+12]
// into the statement:
// mov eax,[esp+3Ch+var_30]
// Here, 3Ch == 60 decimal is the distance between ESP and EBP, and
// var_30 is defined to have the value -30h == -48 decimal. So, the
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
// "frame size" in IDA Pro is 60 bytes, and a certain local can be
// addressed in ESP-relative manner as shown, or as [ebp+var_30] for
// EBP-relative addressing. The interactive IDA user can then edit
// the name var_30 to something mnemonic, such as "virus_size", and IDA
// will replace all occurrences with the new name, so that code references
// automatically become [ebp+virus_size]. As the user proceeds
// interactively, he eventually produces very understandable code.
// This all makes sense for producing readable assembly text. However,
// our analyses have a compiler perspective as well as a memory access
// defense perspective. SMP distinguishes between callee saved regs,
// which should not be overwritten in the function body, and local
// variables, which can be written. We view the stack frame in logical
// pieces: here are the saved regs, here are the locals, here is the
// return address, etc. We don't care which direction from EBP the
// callee-saved registers lie; we don't want to lump them in with the
// local variables. We also don't like the fact that IDA Pro will take
// the function prologue code shown above and declare frregs=4 and
// frsize=60, because frsize no longer matches the stack allocation
// statement sub esp,34h == sub esp,52. We prefer frsize=52 and frregs=12.
// So, the task of this function is to fix these stack sizes in our
// private data members for the function, while leaving the IDA database
// alone because IDA needs to maintain its own definitions of these
// variables.
// Fixing means we will update the data members LocalVarsSize and
// CalleeSavedRegsSize.
// NOTE: This function is both machine dependent and platform dependent.
// The prologue and epilogue code generated by gcc-linux is as discussed
// above, while on Visual Studio and other Windows x86 compilers, the
// saving of registers other than EBP happens AFTER local stack allocation.
// A Windows version of the function would expect to see the pushing
// of ESI and EDI AFTER the sub esp,34h statement.
bool SMPFunction::MDFixFrameInfo(void) {
int SavedRegsSize = 0;
int OtherPushesSize = 0; // besides callee-saved regs
int NewLocalsSize = 0;
int OldFrameTotal = this->CalleeSavedRegsSize + this->LocalVarsSize;
bool Changed = false;
bool DebugFlag = (0 == strcmp("__libc_csu_init", this->GetFuncName()));
// Iterate through the first basic block in the function. If we find
// a frame allocating Instr in it, then we have local vars. If not,
// we don't, and LocalVarsSize should have been zero. Count the callee
// register saves leading up to the local allocation. Set data members
// according to what we found if the values of the data members would
// change.
SMPBasicBlock CurrBlock = this->Blocks.front();
for (list<list<SMPInstr>::iterator>::iterator CurrIter = CurrBlock.GetFirstInstr();
CurrIter != CurrBlock.GetLastInstr();
++CurrIter) {
#if SMP_USE_SSA_FNOP_MARKER
if (CurrBlock.GetFirstInstr() == CurrIter)
continue; // skip marker instruction
#endif
list<SMPInstr>::iterator CurrInstr = *CurrIter;
if (CurrInstr->MDIsPushInstr()) {
// We will make the gcc-linux assumption that a PUSH in
// the first basic block, prior to the stack allocating
// instruction, is a callee register save. To make this
// more robust, we ensure that the register is from
// the callee saved group of registers, and that it has
// not been defined thus far in the function (else it might
// be a push of an outgoing argument to a call that happens
// in the first block when there are no locals). **!!!!**
if (CurrInstr->MDUsesCalleeSavedReg()
&& !CurrInstr->HasSourceMemoryOperand()) {
SavedRegsSize += 4; // **!!** should check the size
if (DebugFlag) msg("libc_csu_init SavedRegsSize: %d %s\n", SavedRegsSize,
CurrInstr->GetDisasm());
}
else {
// Pushes of outgoing args can be scheduled so that
// they are mixed with the pushes of callee saved regs.
OtherPushesSize += 4;
if (DebugFlag) msg("libc_csu_init OtherPushesSize: %d %s\n", OtherPushesSize,
CurrInstr->GetDisasm());
}
}
else if (CurrInstr->MDIsFrameAllocInstr()) {
if (DebugFlag) msg("libc_csu_init allocinstr: %s\n", CurrInstr->GetDisasm());
SavedRegsSize += OtherPushesSize;
// Get the size being allocated.
set<DefOrUse, LessDefUse>::iterator CurrUse;
for (CurrUse = CurrInstr->GetFirstUse(); CurrUse != CurrInstr->GetLastUse(); ++CurrUse) {
// Find the immediate operand.
if (o_imm == CurrUse->GetOp().type) {
// Get its value into LocalVarsSize.
long AllocValue = (signed long) CurrUse->GetOp().value;
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
// One compiler might have sub esp,24 and another
// might have add esp,-24. Take the absolute value.
if (0 > AllocValue)
AllocValue = -AllocValue;
if (AllocValue != (long) this->LocalVarsSize) {
Changed = true;
#if SMP_DEBUG_FRAMEFIXUP
if (AllocValue + SavedRegsSize != OldFrameTotal)
msg("Total frame size changed: %s\n", this->FuncName);
#endif
this->LocalVarsSize = (asize_t) AllocValue;
this->CalleeSavedRegsSize = (ushort) SavedRegsSize;
NewLocalsSize = this->LocalVarsSize;
}
else { // Old value was correct; no change.
NewLocalsSize = this->LocalVarsSize;
if (SavedRegsSize != this->CalleeSavedRegsSize) {
this->CalleeSavedRegsSize = (ushort) SavedRegsSize;
Changed = true;
#if SMP_DEBUG_FRAMEFIXUP
msg("Only callee regs size changed: %s\n", this->FuncName);
#endif
}
}
} // end if (o_imm == ...)
} // end for all uses
break; // After frame allocation instr, we are done
} // end if (push) .. elsif frame allocating instr
} // end for all instructions in the first basic block
// If we did not find an allocating instruction, see if it would keep
// the total size the same to set LocalVarsSize to 0 and to set
// CalleeSavedRegsSize to SavedRegsSize. If so, do it. If not, we
// might be better off to leave the numbers alone.
if (!Changed && (NewLocalsSize == 0)) {
clc5q
committed
if (DebugFlag) msg("libc_csu_init OldFrameTotal: %d \n", OldFrameTotal);
if (OldFrameTotal == SavedRegsSize) {
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
this->LocalVarsSize = 0;
Changed = true;
}
#if SMP_DEBUG_FRAMEFIXUP
else {
msg("Could not update frame sizes: %s\n", this->FuncName);
}
#endif
}
#if SMP_DEBUG_FRAMEFIXUP
if ((0 < OtherPushesSize) && (0 < NewLocalsSize))
msg("Extra pushes found of size %d in %s\n", OtherPushesSize,
this->FuncName);
#endif
return Changed;
} // end of SMPFunction::MDFixFrameInfo()
// Some functions have difficult to find stack allocations. For example, in some
// version of glibc, strpbrk() zeroes out register ECX and then pushes it more than
// 100 times in order to allocate zero-ed out local vars space for a character translation
// table. We will use the stack pointer analysis of IDA to find out if there is a point
// in the first basic block at which the stack pointer reaches the allocation total
// that IDA is expecting for the local vars region.
// If so, we return the address of the instruction at which ESP reaches its value, else
// we return BADADDR.
ea_t SMPFunction::FindAllocPoint(asize_t OriginalLocSize) {
sval_t TargetSize = - ((sval_t) OriginalLocSize); // negate; stack grows down
#if SMP_DEBUG_FRAMEFIXUP
clc5q
committed
bool DebugFlag = (0 == strcmp("_dl_runtime_resolve", this->GetFuncName()));
msg("%s OriginalLocSize: %d\n", this->GetFuncName(), OriginalLocSize);
// Limit our analysis to the first basic block in the function.
list<SMPInstr>::iterator CurrInstr;
for (CurrInstr = this->Instrs.begin(); CurrInstr != this->Instrs.end(); ++CurrInstr) {
#if SMP_USE_SSA_FNOP_MARKER
if (this->Instrs.begin() == CurrInstr)
continue; // skip marker instruction
#endif
ea_t addr = CurrInstr->GetAddr();
// get_spd() returns a cumulative delta of ESP
clc5q
committed
sval_t sp_delta = get_spd(this->GetFuncInfo(), addr);
#if SMP_DEBUG_FRAMEFIXUP
if (DebugFlag)
msg("%s delta: %d at %x\n", this->GetFuncName(), sp_delta, addr);
if (sp_delta == TargetSize) { // <= instead of == here? **!!**
// Previous instruction hit the frame size.
if (CurrInstr == this->Instrs.begin()) {
return BADADDR; // cannot back up from first instruction
}
else {
ea_t PrevAddr = (--CurrInstr)->GetAddr();
#if SMP_USE_SSA_FNOP_MARKER
if (this->Instrs.begin()->GetAddr() == PrevAddr)
return BADADDR; // don't return marker instruction
else
return PrevAddr;
#else
return PrevAddr;
#endif
if (CurrInstr->IsLastInBlock()) {
// It could be that the current instruction will cause the stack pointer
// delta to reach the TargetSize. sp_delta is not updated until after the
// current instruction, so we need to look ahead one instruction if the
// current block falls through. On the other hand, if the current block
// ends with a jump or return, we cannot hit TargetSize.
if (CurrInstr->IsBasicBlockTerminator())
return BADADDR;
list<SMPInstr>::iterator NextInstr = CurrInstr;
++NextInstr;
if (NextInstr == this->Instrs.end())
return BADADDR;
clc5q
committed
sp_delta = get_spd(this->GetFuncInfo(), NextInstr->GetAddr());
if (sp_delta == TargetSize) {
// CurrInstr will cause stack pointer delta to hit TargetSize.
return addr;
}
else {
return BADADDR;
}
} // end if LastInBlock
} // end for all instructions
#if SMP_DEBUG_FRAMEFIXUP
else {
msg("AnalyzedSP is false for %s\n", this->GetFuncName());
}
#endif
return BADADDR;
} // end of SMPFunction::FindAllocPoint()
// IDA Pro is sometimes confused by a function that uses the frame pointer
// register for other purposes. For the x86, a function that uses EBP
// as a frame pointer would begin with: push ebp; mov ebp,esp to save
// the old value of EBP and give it a new value as a frame pointer. The
// allocation of local variable space would have to come AFTER the move
// instruction. A function that begins: push ebp; push esi; sub esp,24
// is obviously not using EBP as a frame pointer. IDA is apparently
// confused by the push ebp instruction being the first instruction
// in the function. We will reset UseFP to false in this case.
// The inverse problem happens with a function that begins with instructions
// other than push ebp; mov ebp,esp; ... etc. but eventually has those
// instructions in the first basic block. For example, a C compiler generates
// for the first block of main():
// lea ecx,[esp+arg0]
// and esp, 0xfffffff0
// push dword ptr [ecx-4]
// push ebp
// mov ebp,esp
// push ecx
// sub esp,<framesize>
//
// This function is obviously using EBP as a frame pointer, but IDA Pro marks
// the function as not using a frame pointer. We will reset UseFP to true in
// this case.
// NOTE: This logic should work for both Linux and Windows x86 prologues.
bool SMPFunction::MDFixUseFP(void) {
list<SMPInstr>::iterator CurrInstr = this->Instrs.begin();
ea_t addr = CurrInstr->GetAddr();
#if SMP_USE_SSA_FNOP_MARKER
if (this->Instrs.begin() == CurrInstr)
++CurrInstr; // skip marker instruction
#endif
if (!(this->UseFP)) {
// See if we can detect the instruction "push ebp" followed by the instruction
// "mov ebp,esp" in the first basic block. The instructions do not have to be
// consecutive. If we find them, we will reset UseFP to true.
bool FirstBlockProcessed = false;
bool EBPSaved = false;
bool ESPintoEBP = false;
do {
FirstBlockProcessed = CurrInstr->IsLastInBlock();
if (!EBPSaved) { // still looking for "push ebp"
if (CurrInstr->MDIsPushInstr() && CurrInstr->GetCmd().Operands[0].is_reg(R_bp)) {
EBPSaved = true;
}
}
else if (!ESPintoEBP) { // found "push ebp", looking for "mov ebp,esp"
insn_t CurrCmd = CurrInstr->GetCmd();
if ((CurrCmd.itype == NN_mov)
&& (CurrInstr->GetFirstDef()->GetOp().is_reg(R_bp))
&& (CurrInstr->GetFirstUse()->GetOp().is_reg(R_sp))) {
ESPintoEBP = true;
FirstBlockProcessed = true; // exit loop
}
}
++CurrInstr;
addr = CurrInstr->GetAddr();
// We must get EBP set to its frame pointer value before we reach the
// local frame allocation instruction (i.e. the subtraction of locals space
// from the stack pointer).
FirstBlockProcessed |= (addr >= this->LocalVarsAllocInstr);
} while (!FirstBlockProcessed);
// If we found ESPintoEBP, we also found EBPSaved first, and we need to change
// this->UseFP to true and return true. Otherwise, return false.
this->UseFP = ESPintoEBP;
return ESPintoEBP;
} // end if (!(this->UseFP))
// At this point, this->UseFP must have been true on entry to this method and we will
// check whether it should be reset to false.
while (addr < this->LocalVarsAllocInstr) {
set<DefOrUse, LessDefUse>::iterator CurrDef = CurrInstr->GetFirstDef();
while (CurrDef != CurrInstr->GetLastDef()) {
if (CurrDef->GetOp().is_reg(R_bp))
return false; // EBP got set before locals were allocated
}
++CurrInstr;
addr = CurrInstr->GetAddr();
}
// If we found no defs of the frame pointer before the local vars
// allocation, then the frame pointer register is not being used
// as a frame pointer, just as a general callee-saved register.
this->UseFP = false;
msg("MDFixUseFP reset UseFP to false for %s\n", this->GetFuncName());
return true;
} // end of SMPFunction::MDFixUseFP()
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
// Find the callee-saved reg offsets (negative offset from return address)
// for all registers pushed onto the stack before the stack frame allocation
// instruction.
void SMPFunction::MDFindSavedRegs(void) {
list<SMPInstr>::iterator CurrInst;
int RegIndex;
func_t *CurrFunc = get_func(this->GetStartAddr());
assert(NULL != CurrFunc);
for (CurrInst = this->Instrs.begin(); CurrInst != this->Instrs.end(); ++CurrInst) {
if (CurrInst->GetAddr() > this->LocalVarsAllocInstr)
break;
if (!(CurrInst->MDIsPushInstr()))
continue;
sval_t CurrOffset = get_spd(CurrFunc, CurrInst->GetAddr());
if (CurrInst->GetCmd().itype == NN_push) {
op_t PushedReg = CurrInst->GetPushedOpnd();
if (o_reg == PushedReg.type) {
RegIndex = (int) PushedReg.reg;
if (RegIndex > R_di) {
msg("WARNING: Skipping save of register %d\n", RegIndex);
continue;
}
if (this->SavedRegLoc.at((size_t) RegIndex) == 0) {
this->SavedRegLoc[(size_t) RegIndex] = CurrOffset - 4;
}
else {
msg("WARNING: Multiple saves of register %d\n", RegIndex);
}
} // end if register push operand
} // end if PUSH instruction
else if (NN_pusha == CurrInst->GetCmd().itype) {
// **!!** Handle pushes of all regs.
this->SavedRegLoc[(size_t) R_ax] = CurrOffset - 4;
this->SavedRegLoc[(size_t) R_cx] = CurrOffset - 8;
this->SavedRegLoc[(size_t) R_dx] = CurrOffset - 12;
this->SavedRegLoc[(size_t) R_bx] = CurrOffset - 16;
this->SavedRegLoc[(size_t) R_sp] = CurrOffset - 20;
this->SavedRegLoc[(size_t) R_bp] = CurrOffset - 24;
this->SavedRegLoc[(size_t) R_si] = CurrOffset - 28;
this->SavedRegLoc[(size_t) R_di] = CurrOffset - 32;
break; // all regs accounted for
}
else if (CurrInst->MDIsEnterInstr()) {
this->SavedRegLoc[(size_t) R_bp] = CurrOffset - 4;
}
} // end for all instructions
return;
} // end of SMPFunction::MDFindSavedRegs()
// Compute the ReturnRegTypes[] as the meet over all register types
// at all return instructions.
void SMPFunction::MDFindReturnTypes(void) {
list<SMPBasicBlock>::iterator CurrBlock;
list<list<SMPInstr>::iterator>::iterator InstIter;
vector<SMPOperandType> RegTypes;
for (CurrBlock = this->Blocks.begin(); CurrBlock != this->Blocks.end(); ++CurrBlock) {
if (CurrBlock->HasReturn()) {
// Get the types of all registers at the RETURN point.
// Calculate the meet function over them.
InstIter = CurrBlock->GetLastInstr();
--InstIter;
assert(RETURN == (*InstIter)->GetDataFlowType());
set<DefOrUse, LessDefUse>::iterator CurrUse;
for (CurrUse = (*InstIter)->GetFirstUse();
CurrUse != (*InstIter)->GetLastUse();
++CurrUse) {
op_t UseOp = CurrUse->GetOp();
if ((o_reg != UseOp.type) || (R_di < UseOp.reg))
continue;
this->ReturnRegTypes[UseOp.reg]
= SMPTypeMeet(this->ReturnRegTypes.at(UseOp.reg),
CurrUse->GetType());
} // for all USEs in the RETURN instruction
} // end if current block has a RETURN
} // end for all blocks
return;
} // end of SMPFunction::MDFindReturnTypes()
// Determine local variable boundaries in the stack frame.
void SMPFunction::BuildLocalVarTable(void) {
// Currently we just use the info that IDA Pro has inferred from the direct
// addressing of stack locations.
this->SemiNaiveLocalVarID();
return;
} // end of SMPFunction::BuildLocalVarTable()
// Use the local variable offset list from IDA's stack frame structure to compute
// the table of local variable boundaries.
void SMPFunction::SemiNaiveLocalVarID(void) {
// NOTE: We use IDA Pro's offsets from this->FuncInfo (e.g. frsize) and NOT
// our own corrected values in our private data members. The offsets we
// read from the stack frame structure returned by get_frame() are consistent
// with other IDA Pro values, not with our corrected values.
bool DebugFlag = false;
#if SMP_DEBUG_STACK_GRANULARITY
DebugFlag |= (0 == strcmp("qSort3", this->GetFuncName()));
#endif
func_t *FuncPtr = get_func(this->FuncInfo.startEA);
if (NULL == FuncPtr) {
msg("ERROR in SMPFunction::SemiNaiveLocalVarID; no func ptr\n");
}
assert(NULL != FuncPtr);
struc_t *StackFrame = get_frame(FuncPtr);
if (NULL == StackFrame) {
msg("WARNING: No stack frame info from get_frame for %s\n", this->GetFuncName());
return;
}
member_t *Member = StackFrame->members;
for (size_t i = 0; i < StackFrame->memqty; ++i, ++Member) {
long offset;
if (NULL == Member) {
msg("NULL stack frame member pointer in %s\n", this->GetFuncName());
break;
}
get_member_name(Member->id, MemberName, MAXSMPVARSTR - 1);
if (MemberName == NULL) {
#if SMP_DEBUG_STACK_GRANULARITY
msg("NULL stack frame member in %s\n", this->GetFuncName());
continue;
}
offset = Member->soff;
if (MemberName[0] == ' ') {
#if SMP_DEBUG_STACK_GRANULARITY
msg("NULL stack frame name at offset %d in %s\n", offset, this->GetFuncName());
MemberName[1] = '\0';
}
if (DebugFlag) {
clc5q
committed
msg("%s local var %s at offset %ld\n", this->GetFuncName(), MemberName, offset);
}
if (offset >= (long) this->LocalVarsSize)
break; // Stop after processing locals and outgoing args
#if 0
// We want the offset from the stack pointer after local frame allocation.
// This subtraction would make it relative to the original stack pointer.
offset -= this->FuncInfo.frsize;
#endif
struct LocalVar TempLocal;
TempLocal.offset = offset;
qstrncpy(TempLocal.VarName, MemberName, sizeof(TempLocal.VarName) - 1);
this->LocalVarTable.push_back(TempLocal);
} // end for all stack frame members
if (this->LocalVarTable.empty())
return;
#if SMP_DEBUG_STACK_GRANULARITY
msg("Computing %d local var sizes\n", this->LocalVarTable.size());
// Now we want to fill in the size field for each local
size_t VarLimit = this->LocalVarTable.size() - 1;
assert(this->LocalVarTable.size() > 0);
for (size_t VarIndex = 0; VarIndex < VarLimit; ++VarIndex) {
this->LocalVarTable[VarIndex].size = this->LocalVarTable[VarIndex + 1].offset
- this->LocalVarTable[VarIndex].offset;
}
#if SMP_DEBUG_STACK_GRANULARITY
msg("Computing last local var size for frsize %d\n", this->FuncInfo.frsize);
#endif
// Size of last local is total frsize minus savedregs in frame minus offset of last local
size_t SavedRegsSpace = 0; // portion of frsize that is saved regs, not locals.
if (this->CalleeSavedRegsSize > this->FuncInfo.frregs) {
// IDA Pro counts the save of EBP in frregs, but then EBP gets its new
// value and callee saved regs other than the old EBP push get counted
// in frsize rather than frregs. CalleeSavedRegsSize includes all saved
// regs on the stack, both above and below the current EBP offset.
// NOTE: For windows, this has to be done differently, as callee saved regs
// happen at the bottom of the local frame, not the top.
#if 0
SavedRegsSpace = this->CalleeSavedRegsSize - this->FuncInfo.frregs;
#else
SavedRegsSpace = this->FuncInfo.frsize - this->LocalVarsSize;
#endif
this->LocalVarTable.back().size = this->FuncInfo.frsize
- SavedRegsSpace - this->LocalVarTable.back().offset;
this->LocalVarOffsetLimit = this->LocalVarTable.back().offset
+ (adiff_t) this->LocalVarTable.back().size;
// IDA Pro can have difficulty with some irregular functions such as are found
// in the C startup code. The frsize value might be bogus. Just punt on the
// local variable ID if that is the case.
if (this->LocalVarOffsetLimit > (adiff_t) this->FuncInfo.frsize) {
this->LocalVarTable.clear();
msg("WARNING: Bad frsize for %s ; abandoning SemiNaiveLocalVarID.\n", this->FuncName);
return;
}
assert(this->LocalVarOffsetLimit <= (adiff_t) this->FuncInfo.frsize);
// Find out how many of the locals are really outgoing args.
if (this->AnalyzedSP && !this->CallsAlloca && (BADADDR != this->LocalVarsAllocInstr)) {
this->FindOutgoingArgsSize();
}
else {
msg("FindOutgoingArgsSize not called for %s ", this->GetFuncName());
msg("AnalyzedSP: %d CallsAlloca: %d LocalVarsAllocInstr: %x \n",
this->AnalyzedSP, this->CallsAlloca, this->LocalVarsAllocInstr);
}
return;
} // end of SMPFunction::SemiNaiveLocalVarID()
// Determine how many bytes at the bottom of the stack frame (i.e. at bottom of
// this->LocalVarsSize) are used for outgoing args. This is the case when the cdecl
// calling convention is used, e.g. gcc/linux allocates local var space + out args space
// in a single allocation and then writes outarg values directly to ESP+0, ESP+4, etc.
void SMPFunction::FindOutgoingArgsSize(void) {
// Compute the lowest value reached by the stack pointer.
list<SMPInstr>::iterator CurrInst;
this->MinStackDelta = 20000; // Final value should be negative
unsigned int BitWidthMask;
bool DebugFlag = false;
#if SMP_DEBUG_STACK_GRANULARITY
DebugFlag = (0 == strcmp("error_for_asm", this->GetFuncName()));
#endif
this->OutgoingArgsComputed = true;
if (DebugFlag) {
msg("DEBUG: Entered FindOutgoingArgsSize for %s\n", this->GetFuncName());
#if SMP_IDAPRO52_WORKAROUND
this->OutgoingArgsSize = 16;
return;
for (CurrInst = this->Instrs.begin(); CurrInst != this->Instrs.end(); ++CurrInst) {
#if SMP_USE_SSA_FNOP_MARKER
if (this->Instrs.begin() == CurrInst)
continue; // skip marker instruction
#endif
ea_t addr = CurrInst->GetAddr();
clc5q
committed
sval_t sp_delta = get_spd(this->GetFuncInfo(), addr);
if (sp_delta < this->MinStackDelta)
this->MinStackDelta = sp_delta;
if (addr == this->LocalVarsAllocInstr) {
// Total stack pointer delta is sp_delta for the next instruction,
// because IDA updates the sp delta AFTER each instruction.
list<SMPInstr>::iterator NextInst = CurrInst;
++NextInst;
clc5q
committed
sp_delta = get_spd(this->GetFuncInfo(), NextInst->GetAddr());
this->AllocPointDelta = sp_delta;
}
}
#if SMP_DEBUG_STACK_GRANULARITY
msg("AllocPointDelta: %d MinStackDelta: %d\n", this->AllocPointDelta, this->MinStackDelta);
#endif
assert(0 > this->MinStackDelta);
// Allocate a vector of stack frame entries, one for each byte of the stack frame.
// This will be our memory map for analyzing stack usage.
int limit = 0;
#if 1
if (this->LocalVarOffsetLimit > 0)
limit = this->LocalVarOffsetLimit;
#endif
for (int i = this->MinStackDelta; i < limit; ++i) {
struct StackFrameEntry TempEntry;
struct FineGrainedInfo TempFineGrained;
TempEntry.VarPtr = NULL;
TempEntry.offset = (long) i;
TempEntry.Read = false;
TempEntry.Written = false;
TempEntry.AddressTaken = false;
TempEntry.ESPRelativeAccess = false;
TempEntry.EBPRelativeAccess = false;
this->StackFrameMap.push_back(TempEntry);
TempFineGrained.SignMiscInfo = 0;
TempFineGrained.SizeInfo = 0;
this->FineGrainedStackTable.push_back(TempFineGrained);
}
// Fill in the VarPtr fields for each StackFrameMap entry.
if (0 <= this->AllocPointDelta) {
msg("FATAL ERROR: AllocPointDelta = %d in %s\n", this->AllocPointDelta, this->GetFuncName());
}
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
assert(0 > this->AllocPointDelta);
for (size_t i = 0; i < this->LocalVarTable.size(); ++i) {
assert(this->LocalVarTable.at(i).offset >= 0);
// Picture that AllocPointDelta is -200, MinStackDelta is -210, and
// the LocalVarTable[i].offset is +8 (i.e. 8 bytes above alloc point).
// Then base = 8 + (-200 - -210) = 8 + 10 = 18, the proper offset into
// the StackFrameMap.
size_t base = (size_t) (this->LocalVarTable.at(i).offset
+ (this->AllocPointDelta - this->MinStackDelta));
size_t limit = base + this->LocalVarTable.at(i).size;
if (limit > this->StackFrameMap.size()) {
msg("ERROR: base = %d limit = %d StackFrameMap size = %d\n", base, limit,
this->StackFrameMap.size());
}
assert(limit <= this->StackFrameMap.size());
for (size_t MapIndex = base; MapIndex < limit; ++MapIndex) {
this->StackFrameMap[MapIndex].VarPtr = &(this->LocalVarTable.at(i));
}
}
// Iterate through all instructions and record stack frame accesses in the StackFrameMap.
for (CurrInst = this->Instrs.begin(); CurrInst != this->Instrs.end(); ++CurrInst) {
#if SMP_USE_SSA_FNOP_MARKER
if (this->Instrs.begin() == CurrInst)
continue; // skip marker instruction
#endif
clc5q
committed
sval_t sp_delta = get_spd(this->GetFuncInfo(), CurrInst->GetAddr());
if (0 < sp_delta) {
// Stack underflow; about to assert
msg("Stack underflow at %x %s sp_delta: %d\n", CurrInst->GetAddr(),
CurrInst->GetDisasm(), sp_delta);
}
assert(0 >= sp_delta);
ea_t offset;
size_t DataSize;
bool UsedFramePointer;
bool SignedMove;
bool UnsignedMove;
if (CurrInst->HasDestMemoryOperand()) {
set<DefOrUse, LessDefUse>::iterator CurrDef;
for (CurrDef = CurrInst->GetFirstDef(); CurrDef != CurrInst->GetLastDef(); ++CurrDef) {
op_t TempOp = CurrDef->GetOp();
if (TempOp.type != o_phrase && TempOp.type != o_displ)
continue;
if (this->MDGetStackOffsetAndSize(CurrInst, TempOp, sp_delta, offset, DataSize, UsedFramePointer,
SignedMove, UnsignedMove)) {
assert(0 <= offset);
if (offset >= this->FuncInfo.frsize)
continue; // limit processing to outgoing args and locals
if ((offset + DataSize) > this->StackFrameMap.size()) {
msg("ERROR: offset = %d DataSize = %d FrameMapSize = %d\n",
offset, DataSize, this->StackFrameMap.size());
}
assert((offset + DataSize) <= this->StackFrameMap.size());
for (int j = 0; j < (int) DataSize; ++j) {
this->StackFrameMap[offset + j].Written = true;
if (!UsedFramePointer) {
this->StackFrameMap[offset + j].ESPRelativeAccess = true;
}
else {
this->StackFrameMap[offset + j].EBPRelativeAccess = true;
BitWidthMask = ComputeOperandBitWidthMask(TempOp, DataSize);
this->FineGrainedStackTable.at(offset).SizeInfo |= BitWidthMask;
this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_WRITTEN;
if (!UsedFramePointer) {
this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_SP_RELATIVE;
}
else {
this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_FP_RELATIVE;
}
// We will process the signedness of stores later, so that loads can take precedence
// over stores in determining signedness.
} // end if MDGetStackOffsetAndSize()
} // end for all DEFs
} // end if DestMemoryOperand
if (CurrInst->HasSourceMemoryOperand()) {
set<DefOrUse, LessDefUse>::iterator CurrUse;
for (CurrUse = CurrInst->GetFirstUse(); CurrUse != CurrInst->GetLastUse(); ++CurrUse) {
op_t TempOp = CurrUse->GetOp();
if (TempOp.type != o_phrase && TempOp.type != o_displ)
continue;
if (this->MDGetStackOffsetAndSize(CurrInst, TempOp, sp_delta, offset, DataSize, UsedFramePointer,
SignedMove, UnsignedMove)) {
assert(0 <= offset);
if (offset >= this->FuncInfo.frsize)
continue; // limit processing to outgoing args and locals
if ((offset + DataSize) > this->StackFrameMap.size()) {
msg("ERROR: offset = %d DataSize = %d FrameMapSize = %d\n",
offset, DataSize, this->StackFrameMap.size());
}
assert((offset + DataSize) <= this->StackFrameMap.size());
for (int j = 0; j < (int) DataSize; ++j) {
this->StackFrameMap[offset + j].Read = true;
if (!UsedFramePointer)
this->StackFrameMap[offset + j].ESPRelativeAccess = true;
else
this->StackFrameMap[offset + j].EBPRelativeAccess = true;
}
BitWidthMask = ComputeOperandBitWidthMask(TempOp, DataSize);
this->FineGrainedStackTable.at(offset).SizeInfo |= BitWidthMask;
this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_READ;
if (!UsedFramePointer) {
this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_SP_RELATIVE;
}
else {
this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_FP_RELATIVE;
}
if (SignedMove) {
this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_SIGNED;
}
else if (UnsignedMove) {
this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_UNSIGNED;
}
} // end if MDGetStackOffsetAndSize()
} // end if SourceMemoryOperand
// NOTE: Detect taking the address of stack locations. **!!**
} // end for all instructions
// If function is a leaf function, set OutgoingArgsSize to zero and return.
if (this->IsLeaf()) {
this->OutgoingArgsSize = 0;
return;
}
// For non-leaf functions, set the OutgoingArgsSize to the write-only, ESP-relative
// region of the bottom of the StackFrameMap.
for (size_t MapIndex = 0; MapIndex < this->StackFrameMap.size(); ++MapIndex) {
// Some of the bottom of the stack frame might be below the local frame allocation.
// These are pushes that happened after allocation, etc. We skip over these
// locations and define the outgoing args region to start strictly at the bottom
// of the local frame allocation.
struct StackFrameEntry TempEntry = this->StackFrameMap.at(MapIndex);
if (DebugFlag) {
clc5q
committed
msg("StackFrameMap entry %d: offset: %ld Read: %d Written: %d ESP: %d EBP: %d\n",
MapIndex, TempEntry.offset, TempEntry.Read, TempEntry.Written,
TempEntry.ESPRelativeAccess, TempEntry.EBPRelativeAccess);
}
if (TempEntry.offset < this->AllocPointDelta)
continue;
if (TempEntry.Read || TempEntry.EBPRelativeAccess || !TempEntry.Written
|| !TempEntry.ESPRelativeAccess)
break;
this->OutgoingArgsSize++;
}
// Sometimes we encounter unused stack space above the outgoing args. Lump this space
// in with the outgoing args. We detect this by noting when the outgoing args space
// has only partially used the space assigned to a local var.
if ((0 < this->OutgoingArgsSize) && (this->OutgoingArgsSize < this->FuncInfo.frsize)) {
long MapIndex = (this->AllocPointDelta - this->MinStackDelta);
assert(0 <= MapIndex);
MapIndex += (((long) this->OutgoingArgsSize) - 1);
struct StackFrameEntry TempEntry = this->StackFrameMap.at((size_t) MapIndex);
clc5q
committed
if (NULL == TempEntry.VarPtr) { // Gap in stack frame; IDA 6.0
msg("Gap in stack frame: %s\n", this->FuncName);
}
else if (this->OutgoingArgsSize < (TempEntry.VarPtr->offset + TempEntry.VarPtr->size)) {
clc5q
committed
#if SMP_DEBUG_FRAMEFIXUP
msg("OutGoingArgsSize = %d", this->OutgoingArgsSize);
clc5q
committed
#endif
this->OutgoingArgsSize = TempEntry.VarPtr->offset + TempEntry.VarPtr->size;
clc5q
committed
#if SMP_DEBUG_FRAMEFIXUP
msg(" adjusted to %d\n", this->OutgoingArgsSize);
clc5q
committed
#endif
}
}
return;
} // end of SMPFunction::FindOutgoingArgsSize()
// If TempOp reads or writes to a stack location, return the offset (relative to the initial
// stack pointer value) and the size in bytes of the data access. Also return whether the
// access was frame-pointer-relative, and whether signedness can be inferred due to a load
// from the stack being zero-extended or sign-extended.
// NOTE: TempOp must be of type o_displ or o_phrase, as no other operand type could be a
// stack memory access.
// sp_delta is the stack pointer delta of the current instruction, relative to the initial
// stack pointer value for the function.
// Return true if a stack memory access was found in TempOp, false otherwise.
bool SMPFunction::MDGetStackOffsetAndSize(list<SMPInstr>::iterator Instr, op_t TempOp, sval_t sp_delta, ea_t &offset, size_t &DataSize, bool &FP,
bool &Signed, bool &Unsigned) {
clc5q
committed
int BaseReg;
int IndexReg;
ushort ScaleFactor;
assert((o_displ == TempOp.type) || (o_phrase == TempOp.type));
clc5q
committed
MDExtractAddressFields(TempOp, BaseReg, IndexReg, ScaleFactor, offset);
clc5q
committed
if (TempOp.type == o_phrase) {
assert(offset == 0); // implicit zero, as in [esp] ==> [esp+0]
}
if ((BaseReg == R_sp) || (IndexReg == R_sp)) {
// ESP-relative constant offset
offset += sp_delta; // base offsets from entry ESP value
offset -= this->MinStackDelta; // convert to StackFrameMap index
// Get size of data written
DataSize = GetOpDataSize(TempOp);
FP = false;
unsigned short opcode = Instr->GetCmd().itype;
Unsigned = (opcode == NN_movzx);
Signed = (opcode == NN_movsx);
return true;
}
else if (this->UseFP && ((BaseReg == R_bp) || (IndexReg == R_bp))) {
offset -= this->FuncInfo.frregs; // base offsets from entry ESP value
offset -= this->MinStackDelta; // convert to StackFrameMap index
DataSize = GetOpDataSize(TempOp);
FP = true;
unsigned short opcode = Instr->GetCmd().itype;
Unsigned = (opcode == NN_movzx);
Signed = (opcode == NN_movsx);
return true;
}
else {
return false;
}
} // end of SMPFunction::MDGetStackOffsetAndSize()
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
// Return fine grained stack entry for stack op TempOp from instruction at InstAddr
bool SMPFunction::MDGetFGStackLocInfo(ea_t InstAddr, op_t TempOp, struct FineGrainedInfo &FGEntry) {
int BaseReg;
int IndexReg;
ushort ScaleFactor;
ea_t offset;
int SignedOffset;
assert((o_displ == TempOp.type) || (o_phrase == TempOp.type));
MDExtractAddressFields(TempOp, BaseReg, IndexReg, ScaleFactor, offset);
sval_t sp_delta = get_spd(this->GetFuncInfo(), InstAddr);
SignedOffset = (int) offset;
if (TempOp.type == o_phrase) {
assert(offset == 0); // implicit zero, as in [esp] ==> [esp+0]
}
if ((BaseReg == R_sp) || (IndexReg == R_sp)) {
// ESP-relative constant offset
offset += sp_delta; // base offsets from entry ESP value
offset -= this->MinStackDelta; // convert to StackFrameMap index
}
else if (this->UseFP && ((BaseReg == R_bp) || (IndexReg == R_bp))) {
offset -= this->FuncInfo.frregs; // base offsets from entry ESP value
offset -= this->MinStackDelta; // convert to StackFrameMap index
}
else {
return false;
}
// We did not return false, so we should have a good offset. Use it to
// pass back the fine grained stack table entry for that offset.
if ((0 > offset) || (offset >= this->FineGrainedStackTable.size())) {
msg("ERROR: FG stack table index out of range in MDGetFGStackLocInfo at %x\n", InstAddr);
FGEntry.SignMiscInfo = 0;
FGEntry.SizeInfo = 0;
}
else {
FGEntry = this->FineGrainedStackTable.at(offset);
}
return true;
} // end of SMPFunction::MDGetFGStackLocInfo()
// retrieve DEF addr from GlobalDefAddrBySSA or return BADADDR
ea_t SMPFunction::GetGlobalDefAddr(op_t DefOp, int SSANum) {
map<int, ea_t>::iterator DefAddrMapIter;
map<int, ea_t>::iterator MapResult;
ea_t DefAddr = BADADDR; // BADADDR means we did not find it
int HashedName = HashGlobalNameAndSSA(DefOp, SSANum);
MapResult = this->GlobalDefAddrBySSA.find(HashedName);
if (MapResult != this->GlobalDefAddrBySSA.end()) { // Found it.
DefAddr = (ea_t) MapResult->second;
}
return DefAddr;
} // end of SMPFunction::GetGlobalDefAddr()
// Retrieve block iterator for InstAddr from InstBlockMap; assert if failure
list<SMPBasicBlock>::iterator SMPFunction::GetBlockFromInstAddr(ea_t InstAddr) {
map<ea_t, list<SMPBasicBlock>::iterator>::iterator MapEntry;
MapEntry = this->InstBlockMap.find(InstAddr);
assert(MapEntry != this->InstBlockMap.end());
return MapEntry->second;
}
// Given block # and PhiDef op_t and SSANum, return the Phi iterator or assert.
set<SMPPhiFunction, LessPhi>::iterator SMPFunction::GetPhiIterForPhiDef(size_t BlockNumber, op_t DefOp, int SSANum) {
list<SMPBasicBlock>::iterator DefBlock = this->RPOBlocks.at(BlockNumber);
set<SMPPhiFunction, LessPhi>::iterator PhiIter = DefBlock->FindPhi(DefOp);
assert(PhiIter != DefBlock->GetLastPhi());
return PhiIter;
}
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
// Is DestOp within the outgoing args area? Assume it must be an ESP-relative
// DEF operand in order to be a write to the outgoing args area.
bool SMPFunction::WritesToOutgoingArgs(op_t DestOp) {
bool OutArgWrite = false;
int BaseReg, IndexReg;
ushort ScaleFactor;
ea_t offset;
if (this->IsLeaf())
return false;
MDExtractAddressFields(DestOp, BaseReg, IndexReg, ScaleFactor, offset);
if ((BaseReg != R_sp) && (IndexReg != R_sp))
return false;
if (((BaseReg == R_sp) && (IndexReg != R_none))
|| ((IndexReg == R_sp) && (BaseReg != R_none))
|| (0 < ScaleFactor)) {
msg("WARNING: WritesToOutgoingArgs called with indexed write.");
PrintOperand(DestOp);
return false;
}
if (!this->OutgoingArgsComputed) {
OutArgWrite = true; // be conservative
}
else {
OutArgWrite = (offset < this->OutgoingArgsSize);
}
return OutArgWrite;
} // end of SMPFunction::WritesToOutgoingArgs()
// Is DestOp a direct memory access above the local vars frame?
bool SMPFunction::WritesAboveLocalFrame(op_t DestOp) {
bool InArgWrite = false;
int BaseReg, IndexReg;
ushort ScaleFactor;
ea_t offset;
MDExtractAddressFields(DestOp, BaseReg, IndexReg, ScaleFactor, offset);
bool ESPrelative = (BaseReg == R_sp) || (IndexReg == R_sp);
bool EBPrelative = this->UseFP && ((BaseReg == R_bp) || (IndexReg == R_bp));
if (!(ESPrelative || EBPrelative))
return false;
if (((IndexReg != R_none) && (BaseReg != R_none))
|| (0 < ScaleFactor)) {
msg("WARNING: WritesAboveLocalFrame called with indexed write.");
PrintOperand(DestOp);
return false;
}
InArgWrite = (ESPrelative && (SignedOffset > ((long) this->LocalVarsSize)))
|| (EBPrelative && (SignedOffset > 0));
return InArgWrite;
}// end of SMPFunction::WritesAboveLocalFrame()
// Is DestOp an indexed write above the local vars frame?
bool SMPFunction::IndexedWritesAboveLocalFrame(op_t DestOp)
{
bool InArgWrite = false;
int BaseReg, IndexReg;
ushort ScaleFactor;
ea_t offset;
MDExtractAddressFields(DestOp, BaseReg, IndexReg, ScaleFactor, offset);
bool ESPrelative = (BaseReg == R_sp) || (IndexReg == R_sp);
bool EBPrelative = this->UseFP && ((BaseReg == R_bp) || (IndexReg == R_bp));
if (!(ESPrelative || EBPrelative))
return false;
InArgWrite = (ESPrelative && (offset > this->LocalVarsSize))
|| (EBPrelative && (offset > 0));
return InArgWrite;
} // end of SMPFunction::IndexedWritesAboveLocalFrame
// Find evidence of calls to alloca(), which appear as stack space allocations (i.e.
// subtractions from the stack pointer) AFTER the local frame allocation instruction
// for this function.
// Return true if such an allocation is found and false otherwise.
bool SMPFunction::FindAlloca(void) {
list<SMPInstr>::iterator CurrInst;
for (CurrInst = this->Instrs.begin(); CurrInst != this->Instrs.end(); ++CurrInst) {
#if SMP_USE_SSA_FNOP_MARKER
if (this->Instrs.begin() == CurrInst)
continue; // skip marker instruction
#endif
if ((CurrInst->GetAddr() > this->LocalVarsAllocInstr) && CurrInst->MDIsFrameAllocInstr()) {
return true;
}
}
return false;
} // end of SMPFunction::FindAlloca()
// Emit the annotations describing the regions of the stack frame.
void SMPFunction::EmitStackFrameAnnotations(FILE *AnnotFile, list<SMPInstr>::iterator Instr) {
ea_t addr = Instr->GetAddr();
#if 0
if (0 < IncomingArgsSize) {
qfprintf(AnnotFile, "%10x %6d INARGS STACK esp + %d %s \n",
addr, IncomingArgsSize,
(LocalVarsSize + CalleeSavedRegsSize + RetAddrSize),
Instr->GetDisasm());
}
#endif
if (0 < RetAddrSize) {
qfprintf(AnnotFile, "%10x %6d MEMORYHOLE STACK esp + %d ReturnAddress \n",
addr, RetAddrSize, (LocalVarsSize + CalleeSavedRegsSize));
}
if (0 < CalleeSavedRegsSize) {
qfprintf(AnnotFile, "%10x %6d MEMORYHOLE STACK esp + %d CalleeSavedRegs \n",
addr, CalleeSavedRegsSize, LocalVarsSize);
if (0 < LocalVarsSize) {
unsigned long ParentReferentID = DataReferentID++;
clc5q
committed
qfprintf(AnnotFile, "%10x %6d DATAREF STACK %ld esp + %d PARENT LocalFrame LOCALFRAME\n",
addr, LocalVarsSize, ParentReferentID, 0);
#if SMP_COMPUTE_STACK_GRANULARITY
if (this->AnalyzedSP && !this->CallsAlloca && (BADADDR != this->LocalVarsAllocInstr)) {
// We can only fine-grain the stack frame if we were able to analyze the stack
if (this->OutgoingArgsSize > 0) {
clc5q
committed
qfprintf(AnnotFile, "%10x %6d DATAREF STACK %ld esp + %d CHILDOF %ld OFFSET %d OutArgsRegion OUTARGS\n",
addr, this->OutgoingArgsSize, DataReferentID, 0, ParentReferentID, 0);
++DataReferentID;
#if SMP_DEBUG_STACK_GRANULARITY
msg("LocalVarTable of size %d for function %s\n", this->LocalVarTable.size(),
this->GetFuncName());
for (size_t i = 0; i < this->LocalVarTable.size(); ++i) {
#if SMP_DEBUG_STACK_GRANULARITY
msg("Entry %d offset %d size %d name %s\n", i, this->LocalVarTable[i].offset,
this->LocalVarTable[i].size, this->LocalVarTable[i].VarName);
// Don't emit annotations for incoming or outgoing args or anything else
// above or below the current local frame.
if ((this->LocalVarTable[i].offset >= (long) this->FuncInfo.frsize)
|| (this->LocalVarTable[i].offset < (long) this->OutgoingArgsSize))
continue;
clc5q
committed
qfprintf(AnnotFile, "%10x %6d DATAREF STACK %ld esp + %ld CHILDOF %ld OFFSET %ld LOCALVAR %s \n",
addr, this->LocalVarTable[i].size, DataReferentID,
this->LocalVarTable[i].offset, ParentReferentID,
this->LocalVarTable[i].offset, this->LocalVarTable[i].VarName);
++DataReferentID;
} // end if (this->AnalyzedSP and not Alloca .... )
} // end if (0 < LocalVarsSize)
return;
} // end of SMPFunction::EmitStackFrameAnnotations()
// Main data flow analysis driver. Goes through the function and
// fills all objects for instructions, basic blocks, and the function
// itself.
void SMPFunction::Analyze(void) {
clc5q
committed
bool FoundAllCallers = false;
list<SMPInstr>::iterator FirstInBlock = this->Instrs.end();
// For starting a basic block
list<SMPInstr>::iterator LastInBlock = this->Instrs.end();
// Terminating a basic block
#if SMP_DEBUG_CONTROLFLOW
msg("Entering SMPFunction::Analyze.\n");
#endif
// Get some basic info from the FuncInfo structure.
this->Size = this->FuncInfo.endEA - this->FuncInfo.startEA;
this->UseFP = (0 != (this->FuncInfo.flags & (FUNC_FRAME | FUNC_BOTTOMBP)));
this->StaticFunc = (0 != (this->FuncInfo.flags & FUNC_STATIC));
this->LibFunc = (0 != (this->FuncInfo.flags & FUNC_LIB));
get_func_name(this->FuncInfo.startEA, this->FuncName,
sizeof(this->FuncName) - 1);
this->BlockCount = 0;
this->AnalyzedSP = this->FuncInfo.analyzed_sp();
#if SMP_DEBUG_CONTROLFLOW
msg("SMPFunction::Analyze: got basic info.\n");
#endif
// Cycle through all chunks that belong to the function.
clc5q
committed
func_tail_iterator_t FuncTail(this->GetFuncInfo());
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
size_t ChunkCounter = 0;
for (bool ChunkOK = FuncTail.main(); ChunkOK; ChunkOK = FuncTail.next()) {
const area_t &CurrChunk = FuncTail.chunk();
++ChunkCounter;
if (1 < ChunkCounter) {
this->SharedChunks = true;
#if SMP_DEBUG_CHUNKS
msg("Found tail chunk for %s at %x\n", this->FuncName, CurrChunk.startEA);
#endif
}
// Build the instruction and block lists for the function.
for (ea_t addr = CurrChunk.startEA; addr < CurrChunk.endEA;
addr = get_item_end(addr)) {
flags_t InstrFlags = getFlags(addr);
if (isHead(InstrFlags) && isCode(InstrFlags)) {
SMPInstr CurrInst = SMPInstr(addr);
// Fill in the instruction data members.
#if SMP_DEBUG_CONTROLFLOW
msg("SMPFunction::Analyze: calling CurrInst::Analyze.\n");
#endif
CurrInst.Analyze();
if (SMPBinaryDebug) {
msg("Disasm: %s \n", CurrInst.GetDisasm());
}
#if SMP_USE_SSA_FNOP_MARKER
if (this->Instrs.empty()) {
// First instruction in function. We want to create a pseudo-instruction
// at the top of the function that can hold SSA DEFs for LiveIn names
// to the function. We use a floating point no-op as the pseudo-inst.
// The code address is one less than the start address of the function.
SMPInstr MarkerInst = SMPInstr(addr - 1);
MarkerInst.AnalyzeMarker();
assert(FirstInBlock == this->Instrs.end());
this->Instrs.push_back(MarkerInst);
}
#endif
if (this->AnalyzedSP) {
// Audit the IDA SP analysis.
clc5q
committed
sval_t sp_delta = get_spd(this->GetFuncInfo(), addr);
// sp_delta is difference between current value of stack pointer
// and value of the stack pointer coming into the function. It
// is updated AFTER each instruction. Thus, it should not get back
// above zero (e.g. to +4) until after a return instruction.
if (sp_delta > 0) {
// Stack pointer has underflowed, according to IDA's analysis,
// which is probably incorrect.
this->AnalyzedSP = false;
msg("Resetting AnalyzedSP to false for %s\n", this->GetFuncName());
msg("Underflowing instruction: %s sp_delta: %d\n", CurrInst.GetDisasm(),
sp_delta);
}
else if (sp_delta == 0) {
// Search for tail calls.
if (CurrInst.IsBranchToFarChunk()) {
// After the stack has been restored to the point at which
// we are ready to return, we instead find a jump to a
// far chunk. This is the classic tail call optimization:
// the return statement has been replaced with a jump to
// another function, which will return not to this function,
// but to the caller of this function.
CurrInst.SetTailCall();
msg("Found tail call at %x from %s: %s\n", addr, this->GetFuncName(),
CurrInst.GetDisasm());
// Just like a return instruction, we must make
// DEF-USE chains reach the tail call.
CurrInst.MDAddRegUse(R_ax, false);
CurrInst.MDAddRegUse(R_bx, false);
CurrInst.MDAddRegUse(R_cx, false);
CurrInst.MDAddRegUse(R_dx, false);
CurrInst.MDAddRegUse(R_bp, false);
CurrInst.MDAddRegUse(R_si, false);
CurrInst.MDAddRegUse(R_di, false);
clc5q
committed
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
// Find all functions that call the current function.
xrefblk_t CurrXrefs;
if (!FoundAllCallers) {
for (bool ok = CurrXrefs.first_to(CurrInst.GetAddr(), XREF_ALL);
ok;
ok = CurrXrefs.next_to()) {
if ((CurrXrefs.from != 0) && (CurrXrefs.iscode)) {
// Make sure it is not a fall-through. Must be a
// control-flow instruction of some sort, including
// direct or indirect calls or tail calls.
SMPInstr CallInst(CurrXrefs.from);
CallInst.Analyze();
SMPitype CallType = CallInst.GetDataFlowType();
if ((COND_BRANCH <= CallType) && (RETURN >= CallType)) {
// Found a caller, with its call address in CurrXrefs.from
this->AddCallSource(CurrXrefs.from);
}
}
}
FoundAllCallers = true; // only do this for first inst
}
SMPitype DataFlowType = CurrInst.GetDataFlowType();
if ((DataFlowType == INDIR_CALL)|| (DataFlowType == CALL)) {
// See if IDA has determined the target of the call.
ea_t TargetAddr = CurrInst.GetCallTarget();
bool LinkedToTarget = (BADADDR != TargetAddr);
if (LinkedToTarget) {
if (0 == TargetAddr) {
msg("WARNING: Ignoring NULL call target (unreachable) at %x\n", CurrInst.GetAddr());
}
else {
this->AllCallTargets.push_back(TargetAddr);
if (INDIR_CALL == DataFlowType) {
this->IndirectCallTargets.push_back(TargetAddr);
}
else {
this->DirectCallTargets.push_back(TargetAddr);
}
if (DataFlowType == INDIR_CALL) {
this->IndirectCalls = true;
this->UnresolvedIndirectCalls = (!LinkedToTarget);
}
} // end if INDIR_CALL or CALL
else if (DataFlowType == INDIR_JUMP)
this->IndirectJumps = true;
// Before we insert the instruction into the instruction
// list, determine if it is a jump target that does not
// follow a basic block terminator. This is the special case
// of a CASE in a SWITCH that falls through into another
// CASE, for example. The first sequence of statements
// was not terminated by a C "break;" statement, so it
// looks like straight line code, but there is an entry
// point at the beginning of the second CASE sequence and
// we have to split basic blocks at the entry point.
if ((FirstInBlock != this->Instrs.end())
&& CurrInst.IsJumpTarget()) {
#if SMP_DEBUG_CONTROLFLOW
msg("SMPFunction::Analyze: hit special jump target case.\n");
#endif
LastInBlock = --(this->Instrs.end());
SMPBasicBlock CurrBlock = SMPBasicBlock(this, FirstInBlock,
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
LastInBlock);
CurrBlock.Analyze();
// If not the first chunk in the function, it is a shared
// tail chunk.
if (ChunkCounter > 1) {
CurrBlock.SetShared();
}
FirstInBlock = this->Instrs.end();
LastInBlock = this->Instrs.end();
this->Blocks.push_back(CurrBlock);
this->BlockCount += 1;
}
#if SMP_DEBUG_CONTROLFLOW
msg("SMPFunction::Analyze: putting CurrInst on list.\n");
#endif
// Insert instruction at end of list.
this->Instrs.push_back(CurrInst);
// Find basic block leaders and terminators.
if (FirstInBlock == this->Instrs.end()) {
#if SMP_DEBUG_CONTROLFLOW
msg("SMPFunction::Analyze: setting FirstInBlock.\n");
#if SMP_USE_SSA_FNOP_MARKER
if (2 == this->Instrs.size()) {
// Just pushed first real instruction, after the fnop marker.
FirstInBlock = this->Instrs.begin();
}
else {
FirstInBlock = --(this->Instrs.end());
}
#else
FirstInBlock = --(this->Instrs.end());
}
if (CurrInst.IsBasicBlockTerminator()) {
#if SMP_DEBUG_CONTROLFLOW
msg("SMPFunction::Analyze: found block terminator.\n");
#endif
LastInBlock = --(this->Instrs.end());
SMPBasicBlock CurrBlock = SMPBasicBlock(this, FirstInBlock, LastInBlock);
CurrBlock.Analyze();
// If not the first chunk in the function, it is a shared
// tail chunk.
if (ChunkCounter > 1) {
CurrBlock.SetShared();
}
FirstInBlock = this->Instrs.end();
LastInBlock = this->Instrs.end();
this->Blocks.push_back(CurrBlock);
this->BlockCount += 1;
// Is the instruction a branch to a target outside the function? If
// so, this function has shared tail chunks.
if (CurrInst.IsBranchToFarChunk() && (!CurrInst.IsTailCall())) {
this->SharedChunks = true;
}
}
} // end if (isHead(InstrFlags) && isCode(InstrFlags)
} // end for (ea_t addr = CurrChunk.startEA; ... )
// Handle the special case in which a function does not terminate
// with a return instruction or any other basic block terminator.
// Sometimes IDA Pro sees a call to a NORET function and decides
// to not include the dead code after it in the function. That
// dead code includes the return instruction, so the function no
// longer includes a return instruction and terminates with a CALL.
if (FirstInBlock != this->Instrs.end()) {
LastInBlock = --(this->Instrs.end());
SMPBasicBlock CurrBlock = SMPBasicBlock(this, FirstInBlock, LastInBlock);
CurrBlock.Analyze();
// If not the first chunk in the function, it is a shared
// tail chunk.
if (ChunkCounter > 1) {
CurrBlock.SetShared();
}
FirstInBlock = this->Instrs.end();
LastInBlock = this->Instrs.end();
this->Blocks.push_back(CurrBlock);
this->BlockCount += 1;
}
} // end for (bool ChunkOK = ...)
// Now that we have all instructions and basic blocks, link each instruction
// to its basic block. Note that the instruction has to be linked to the copy
// of the basic block in this->Blocks(), not to the original SMPBasicBlock
// object that was constructed and destructed on the stack above. (Ouch!
// Very painful memory corruption debugging lesson.)
list<SMPBasicBlock>::iterator CurrBlock;
list<list<SMPInstr>::iterator>::iterator InstIter;
for (CurrBlock = this->Blocks.begin(); CurrBlock != this->Blocks.end(); ++CurrBlock) {
for (InstIter = CurrBlock->GetFirstInstr(); InstIter != CurrBlock->GetLastInstr(); ++InstIter) {
(*InstIter)->SetBlock(CurrBlock->GetThisBlock());
}
}
#if KLUDGE_VFPRINTF_FAMILY
if (0 != strstr(this->GetFuncName(), "printf")) {
this->SharedChunks = true;
msg("Kludging function %s\n", this->GetFuncName());
}
#endif
#if SMP_IDAPRO52_WORKAROUND
if (0 == strcmp(this->GetFuncName(), "error_for_asm")) {
this->SharedChunks = true;
msg("Kludging function %s\n", this->GetFuncName());
}
#endif
// Set up basic block links and map of instructions to blocks.
if (!(this->HasSharedChunks())) {
this->SetLinks();
this->RPONumberBlocks();
// Figure out the stack frame and related info.
this->SetStackFrameInfo();
clc5q
committed
list<SMPInstr>::iterator CurrInst;
bool GoodRTL;
this->BuiltRTLs = true;
for (CurrInst = this->Instrs.begin(); CurrInst != this->Instrs.end(); ++CurrInst) {
// Build tree RTLs for the instruction.
GoodRTL = CurrInst->BuildRTL();
this->BuiltRTLs = (this->BuiltRTLs && GoodRTL);
clc5q
committed
if (!GoodRTL) {
msg("ERROR: Cannot build RTL at %x for %s\n", CurrInst->GetAddr(),
CurrInst->GetDisasm());
}
clc5q
committed
if (GoodRTL)
CurrInst->SyncAllRTs();
// Detect indirect memory references.
CurrInst->AnalyzeIndirectRefs(this->UseFP);
clc5q
committed
} // end for all instructions
} // end if not shared chunks
else { // has shared chunks; still want to compute stack frame info
#if SMP_DEBUG_CONTROLFLOW
msg("SMPFunction::Analyze: set stack frame info.\n");
#ifdef SMP_DEBUG_FUNC
msg(" %s has shared chunks \n", this->GetFuncName());
#endif
// Figure out the stack frame and related info.
this->SetStackFrameInfo();
}
// We can finally search for stack loads now that UseFP has been fixed by
// GetStackFrameInfo(). Otherwise, we would do this in SMPInstr::Analyze(),
// but the UseFP flag is not ready that early.
list<SMPInstr>::iterator StLoadInstIter = this->Instrs.begin();
while (StLoadInstIter != this->Instrs.end()) {
StLoadInstIter->MDFindLoadFromStack(this->UseFP);
++StLoadInstIter;
}
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
// Audit the call instructions and call targets.
if ((!this->AllCallTargets.empty()) || this->UnresolvedIndirectCalls) {
bool FoundBadCallTarget = false;
vector<ea_t>::iterator CurrTarget = this->AllCallTargets.begin();
while (CurrTarget != this->AllCallTargets.end()) {
if ((this->FirstEA <= *CurrTarget) && (this->FuncInfo.endEA >= *CurrTarget)) {
// Found a call target that is within the function.
FoundBadCallTarget = true;
if (this->FirstEA == *CurrTarget) { // Direct recursion, not a pseudo-jump
this->DirectlyRecursive = true;
}
CurrTarget = this->AllCallTargets.erase(CurrTarget);
}
else {
++CurrTarget;
}
}
if (FoundBadCallTarget) {
// We have to mark the pseudo-call instructions and audit the direct and
// indirect call target vectors.
// Audit direct call targets.
CurrTarget = this->DirectCallTargets.begin();
while (CurrTarget != this->DirectCallTargets.end()) {
if ((this->FirstEA <= *CurrTarget) && (this->FuncInfo.endEA >= *CurrTarget)) {
// Found a call target that is within the function.
CurrTarget = this->DirectCallTargets.erase(CurrTarget);
}
else {
++CurrTarget;
}
}
// Audit indirect call targets.
CurrTarget = this->IndirectCallTargets.begin();
while (CurrTarget != this->IndirectCallTargets.end()) {
if ((this->FirstEA <= *CurrTarget) && (this->FuncInfo.endEA >= *CurrTarget)) {
// Found a call target that is within the function.
CurrTarget = this->IndirectCallTargets.erase(CurrTarget);
}
else {
++CurrTarget;
}
}
// Find calls used as jumps.
list<SMPInstr>::iterator InstIter = this->Instrs.begin();
while (InstIter != this->Instrs.end()) {
SMPitype InstFlow = InstIter->GetDataFlowType();
if ((CALL == InstFlow) || (INDIR_CALL == InstFlow)) {
InstIter->AnalyzeCallInst(this->FirstEA, this->FuncInfo.endEA);
}
++InstIter;
}
} // end if (FoundBadCallTarget)
}
} // end of SMPFunction::Analyze()
// For each instruction, mark the non-flags-reg DEFs as having live
// metadata (mmStrata needs to fetch and track this metadata for this
// instruction) or dead metadata (won't be used as addressing reg, won't
// be stored to memory, won't be returned to caller).
void SMPFunction::AnalyzeMetadataLiveness(void) {
bool changed;
int BaseReg;
int IndexReg;
ushort ScaleFactor;
ea_t offset;
op_t BaseOp, IndexOp, ReturnOp, DefOp, UseOp;
BaseOp.type = o_reg;
IndexOp.type = o_reg;
ReturnOp.type = o_reg;
list<SMPInstr>::iterator CurrInst;
set<DefOrUse, LessDefUse>::iterator CurrDef;
set<DefOrUse, LessDefUse>::iterator CurrUse;
set<DefOrUse, LessDefUse>::iterator NextUse;
int IterationCount = 0;
#if SMP_DEBUG_DATAFLOW
if (0 == strcmp("uw_frame_state_for", this->GetFuncName())) {
#endif
++IterationCount;
bool SafeMemDest;
if (DebugFlag) {
msg("AnalyzeMetadataLiveness iteration count: %d \n", IterationCount);
}
for (CurrInst = this->Instrs.begin(); CurrInst != this->Instrs.end(); ++CurrInst) {
SafeMemDest = false; // true for some SafeFunc instructions
// Skip the SSA marker instruction.
if (NN_fnop == CurrInst->GetCmd().itype)
continue;
if (DebugFlag) {
msg("Inst addr: %x \n", CurrInst->GetAddr());
}
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
CurrDef = CurrInst->GetFirstDef();
while (CurrDef != CurrInst->GetLastDef()) {
if (DEF_METADATA_UNANALYZED == CurrDef->GetMetadataStatus()) {
DefOp = CurrDef->GetOp();
// Handle special registers never used as address regs.
if (DefOp.is_reg(X86_FLAGS_REG)
|| ((o_trreg <= DefOp.type) && (o_xmmreg >= DefOp.type))) {
CurrDef = CurrInst->SetDefMetadata(DefOp,
DEF_METADATA_UNUSED);
changed = true;
}
else if (DefOp.is_reg(R_sp)
|| (this->UseFP && DefOp.is_reg(R_bp))) {
// Stack pointer register DEFs always have live
// metadata, but we don't need to propagate back
// through particular DEF-USE chains.
CurrDef = CurrInst->SetDefMetadata(DefOp, DEF_METADATA_USED);
changed = true;
}
else if ((o_mem <= DefOp.type) && (o_displ >= DefOp.type)) {
// DEF is a memory operand. The addressing registers
// therefore have live metadata, and the memory metadata is live.
// EXCEPTION: If the function is Safe, then direct stack writes
// to local variables (above the outgoing args area of the frame)
// are not live metadata, and there will be no indirect local frame
// writes, by definition of "safe." So, for safe funcs, only
// the o_mem (globals) and indirect writes are live metadata.
if (this->SafeFunc && MDIsStackAccessOpnd(DefOp, this->UseFP)
&& (!this->WritesAboveLocalFrame(DefOp))
&& (!this->WritesToOutgoingArgs(DefOp))) {
++CurrDef;
SafeMemDest = true;
continue;
}
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
CurrDef = CurrInst->SetDefMetadata(DefOp, DEF_METADATA_USED);
changed = true;
MDExtractAddressFields(DefOp, BaseReg, IndexReg,
ScaleFactor, offset);
if (R_none != BaseReg) {
BaseOp.reg = MDCanonicalizeSubReg((ushort) BaseReg);
if (BaseOp.is_reg(R_sp)
|| (this->UseFP && BaseOp.is_reg(R_bp))) {
; // do nothing; DEF handled by case above
}
else {
CurrUse = CurrInst->FindUse(BaseOp);
if (CurrUse == CurrInst->GetLastUse()) {
msg("ERROR: BaseReg %d not in USE list at %x for %s\n",
BaseOp.reg, CurrInst->GetAddr(),
CurrInst->GetDisasm());
}
assert(CurrUse != CurrInst->GetLastUse());
if (this->IsGlobalName(BaseOp)) {
changed |= this->PropagateGlobalMetadata(BaseOp,
DEF_METADATA_USED, CurrUse->GetSSANum());
}
else {
changed |= CurrInst->GetBlock()->PropagateLocalMetadata(BaseOp,
DEF_METADATA_USED, CurrUse->GetSSANum());
}
}
} // end if R_none != BaseReg
if (R_none != IndexReg) {
IndexOp.reg = MDCanonicalizeSubReg((ushort) IndexReg);
if (IndexOp.is_reg(R_sp)
|| (this->UseFP && IndexOp.is_reg(R_bp))) {
; // do nothing; DEF handled by case above
}
else {
CurrUse = CurrInst->FindUse(IndexOp);
if (CurrUse == CurrInst->GetLastUse()) {
msg("ERROR: IndexReg %d not in USE list at %x for %s\n",
IndexOp.reg, CurrInst->GetAddr(),
CurrInst->GetDisasm());
}
assert(CurrUse != CurrInst->GetLastUse());
if (0 != ScaleFactor) {
; // mmStrata knows scaled reg is NUMERIC
// ... its metadata is not fetched
}
else if (this->IsGlobalName(IndexOp)) {
changed |= this->PropagateGlobalMetadata(IndexOp,
DEF_METADATA_USED, CurrUse->GetSSANum());
Loading
Loading full blame...