cppcheck/lib/checkbufferoverrun.cpp

1168 lines
50 KiB
C++

/*
* Cppcheck - A tool for static C/C++ code analysis
* Copyright (C) 2007-2023 Cppcheck team.
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
//---------------------------------------------------------------------------
// Buffer overrun..
//---------------------------------------------------------------------------
#include "checkbufferoverrun.h"
#include "astutils.h"
#include "errorlogger.h"
#include "library.h"
#include "mathlib.h"
#include "platform.h"
#include "settings.h"
#include "symboldatabase.h"
#include "token.h"
#include "tokenize.h"
#include "utils.h"
#include "valueflow.h"
#include <algorithm>
#include <cstdlib>
#include <functional>
#include <iterator>
#include <numeric> // std::accumulate
#include <sstream>
#include <tinyxml2.h>
//---------------------------------------------------------------------------
// Register this check class (by creating a static instance of it)
namespace {
CheckBufferOverrun instance;
}
//---------------------------------------------------------------------------
// CWE ids used:
static const CWE CWE131(131U); // Incorrect Calculation of Buffer Size
static const CWE CWE170(170U); // Improper Null Termination
static const CWE CWE_ARGUMENT_SIZE(398U); // Indicator of Poor Code Quality
static const CWE CWE_ARRAY_INDEX_THEN_CHECK(398U); // Indicator of Poor Code Quality
static const CWE CWE758(758U); // Reliance on Undefined, Unspecified, or Implementation-Defined Behavior
static const CWE CWE_POINTER_ARITHMETIC_OVERFLOW(758U); // Reliance on Undefined, Unspecified, or Implementation-Defined Behavior
static const CWE CWE_BUFFER_UNDERRUN(786U); // Access of Memory Location Before Start of Buffer
static const CWE CWE_BUFFER_OVERRUN(788U); // Access of Memory Location After End of Buffer
//---------------------------------------------------------------------------
static const ValueFlow::Value *getBufferSizeValue(const Token *tok)
{
const std::list<ValueFlow::Value> &tokenValues = tok->values();
const auto it = std::find_if(tokenValues.cbegin(), tokenValues.cend(), std::mem_fn(&ValueFlow::Value::isBufferSizeValue));
return it == tokenValues.cend() ? nullptr : &*it;
}
static int getMinFormatStringOutputLength(const std::vector<const Token*> &parameters, nonneg int formatStringArgNr)
{
if (formatStringArgNr <= 0 || formatStringArgNr > parameters.size())
return 0;
if (parameters[formatStringArgNr - 1]->tokType() != Token::eString)
return 0;
const std::string &formatString = parameters[formatStringArgNr - 1]->str();
bool percentCharFound = false;
int outputStringSize = 0;
bool handleNextParameter = false;
std::string digits_string;
bool i_d_x_f_found = false;
int parameterLength = 0;
int inputArgNr = formatStringArgNr;
for (int i = 1; i + 1 < formatString.length(); ++i) {
if (formatString[i] == '\\') {
if (i < formatString.length() - 1 && formatString[i + 1] == '0')
break;
++outputStringSize;
++i;
continue;
}
if (percentCharFound) {
switch (formatString[i]) {
case 'f':
case 'x':
case 'X':
case 'i':
i_d_x_f_found = true;
handleNextParameter = true;
parameterLength = 1; // TODO
break;
case 'c':
case 'e':
case 'E':
case 'g':
case 'o':
case 'u':
case 'p':
case 'n':
handleNextParameter = true;
parameterLength = 1; // TODO
break;
case 'd':
i_d_x_f_found = true;
parameterLength = 1;
if (inputArgNr < parameters.size() && parameters[inputArgNr]->hasKnownIntValue())
parameterLength = MathLib::toString(parameters[inputArgNr]->getKnownIntValue()).length();
handleNextParameter = true;
break;
case 's':
parameterLength = 0;
if (inputArgNr < parameters.size() && parameters[inputArgNr]->tokType() == Token::eString)
parameterLength = Token::getStrLength(parameters[inputArgNr]);
handleNextParameter = true;
break;
}
}
if (formatString[i] == '%')
percentCharFound = !percentCharFound;
else if (percentCharFound) {
digits_string.append(1, formatString[i]);
}
if (!percentCharFound)
outputStringSize++;
if (handleNextParameter) {
int tempDigits = std::abs(std::atoi(digits_string.c_str()));
if (i_d_x_f_found)
tempDigits = std::max(tempDigits, 1);
if (digits_string.find('.') != std::string::npos) {
const std::string endStr = digits_string.substr(digits_string.find('.') + 1);
const int maxLen = std::max(std::abs(std::atoi(endStr.c_str())), 1);
if (formatString[i] == 's') {
// For strings, the length after the dot "%.2s" will limit
// the length of the string.
if (parameterLength > maxLen)
parameterLength = maxLen;
} else {
// For integers, the length after the dot "%.2d" can
// increase required length
if (tempDigits < maxLen)
tempDigits = maxLen;
}
}
if (tempDigits < parameterLength)
outputStringSize += parameterLength;
else
outputStringSize += tempDigits;
parameterLength = 0;
digits_string.clear();
i_d_x_f_found = false;
percentCharFound = false;
handleNextParameter = false;
++inputArgNr;
}
}
return outputStringSize;
}
//---------------------------------------------------------------------------
static bool getDimensionsEtc(const Token * const arrayToken, const Settings *settings, std::vector<Dimension> &dimensions, ErrorPath &errorPath, bool &mightBeLarger, MathLib::bigint &path)
{
const Token *array = arrayToken;
while (Token::Match(array, ".|::"))
array = array->astOperand2();
if (array->variable() && array->variable()->isArray() && !array->variable()->dimensions().empty()) {
dimensions = array->variable()->dimensions();
if (dimensions[0].num <= 1 || !dimensions[0].tok) {
visitAstNodes(arrayToken,
[&](const Token *child) {
if (child->originalName() == "->") {
mightBeLarger = true;
return ChildrenToVisit::none;
}
return ChildrenToVisit::op1_and_op2;
});
}
} else if (const Token *stringLiteral = array->getValueTokenMinStrSize(settings, &path)) {
Dimension dim;
dim.tok = nullptr;
dim.num = Token::getStrArraySize(stringLiteral);
dim.known = array->hasKnownValue();
dimensions.emplace_back(dim);
} else if (array->valueType() && array->valueType()->pointer >= 1 && (array->valueType()->isIntegral() || array->valueType()->isFloat())) {
const ValueFlow::Value *value = getBufferSizeValue(array);
if (!value)
return false;
path = value->path;
errorPath = value->errorPath;
Dimension dim;
dim.known = value->isKnown();
dim.tok = nullptr;
const int typeSize = array->valueType()->typeSize(settings->platform, array->valueType()->pointer > 1);
if (typeSize == 0)
return false;
dim.num = value->intvalue / typeSize;
dimensions.emplace_back(dim);
}
return !dimensions.empty();
}
static ValueFlow::Value makeSizeValue(MathLib::bigint size, MathLib::bigint path)
{
ValueFlow::Value v(size);
v.path = path;
return v;
}
static std::vector<ValueFlow::Value> getOverrunIndexValues(const Token* tok,
const Token* arrayToken,
const std::vector<Dimension>& dimensions,
const std::vector<const Token*>& indexTokens,
MathLib::bigint path)
{
const Token *array = arrayToken;
while (Token::Match(array, ".|::"))
array = array->astOperand2();
bool isArrayIndex = tok->str() == "[";
if (isArrayIndex) {
const Token* parent = tok;
while (Token::simpleMatch(parent, "["))
parent = parent->astParent();
if (!parent || parent->isUnaryOp("&"))
isArrayIndex = false;
}
bool overflow = false;
std::vector<ValueFlow::Value> indexValues;
for (int i = 0; i < dimensions.size() && i < indexTokens.size(); ++i) {
MathLib::bigint size = dimensions[i].num;
if (!isArrayIndex)
size++;
const bool zeroArray = array->variable() && array->variable()->isArray() && dimensions[i].num == 0;
std::vector<ValueFlow::Value> values = !zeroArray
? ValueFlow::isOutOfBounds(makeSizeValue(size, path), indexTokens[i])
: std::vector<ValueFlow::Value>{};
if (values.empty()) {
if (indexTokens[i]->hasKnownIntValue())
indexValues.push_back(indexTokens[i]->values().front());
else
indexValues.push_back(ValueFlow::Value::unknown());
continue;
}
overflow = true;
indexValues.push_back(values.front());
}
if (overflow)
return indexValues;
return {};
}
void CheckBufferOverrun::arrayIndex()
{
for (const Token *tok = mTokenizer->tokens(); tok; tok = tok->next()) {
if (tok->str() != "[")
continue;
const Token *array = tok->astOperand1();
while (Token::Match(array, ".|::"))
array = array->astOperand2();
if (!array || ((!array->variable() || array->variable()->nameToken() == array) && array->tokType() != Token::eString))
continue;
if (!array->scope()->isExecutable()) {
// LHS in non-executable scope => This is just a definition
const Token *parent = tok;
while (parent && !Token::simpleMatch(parent->astParent(), "="))
parent = parent->astParent();
if (!parent || parent == parent->astParent()->astOperand1())
continue;
}
if (astIsContainer(array))
continue;
std::vector<const Token *> indexTokens;
for (const Token *tok2 = tok; tok2 && tok2->str() == "["; tok2 = tok2->link()->next()) {
if (!tok2->astOperand2()) {
indexTokens.clear();
break;
}
indexTokens.emplace_back(tok2->astOperand2());
}
if (indexTokens.empty())
continue;
std::vector<Dimension> dimensions;
ErrorPath errorPath;
bool mightBeLarger = false;
MathLib::bigint path = 0;
if (!getDimensionsEtc(tok->astOperand1(), mSettings, dimensions, errorPath, mightBeLarger, path))
continue;
const Variable* const var = array->variable();
if (var && var->isArgument() && var->scope()) {
const Token* changeTok = var->scope()->bodyStart;
bool isChanged = false;
while ((changeTok = findVariableChanged(changeTok->next(), var->scope()->bodyEnd, /*indirect*/ 0, var->declarationId(),
/*globalvar*/ false, mSettings, mTokenizer->isCPP()))) {
if (!Token::simpleMatch(changeTok->astParent(), "[")) {
isChanged = true;
break;
}
}
if (isChanged)
continue;
}
// Positive index
if (!mightBeLarger) { // TODO check arrays with dim 1 also
const std::vector<ValueFlow::Value>& indexValues =
getOverrunIndexValues(tok, tok->astOperand1(), dimensions, indexTokens, path);
if (!indexValues.empty())
arrayIndexError(tok, dimensions, indexValues);
}
// Negative index
bool neg = false;
std::vector<ValueFlow::Value> negativeIndexes;
for (const Token * indexToken : indexTokens) {
const ValueFlow::Value *negativeValue = indexToken->getValueLE(-1, mSettings);
if (negativeValue) {
negativeIndexes.emplace_back(*negativeValue);
neg = true;
} else {
negativeIndexes.emplace_back(ValueFlow::Value::unknown());
}
}
if (neg) {
negativeIndexError(tok, dimensions, negativeIndexes);
}
}
}
static std::string stringifyIndexes(const std::string& array, const std::vector<ValueFlow::Value>& indexValues)
{
if (indexValues.size() == 1)
return MathLib::toString(indexValues[0].intvalue);
std::ostringstream ret;
ret << array;
for (const ValueFlow::Value& index : indexValues) {
ret << "[";
if (index.isNonValue())
ret << "*";
else
ret << index.intvalue;
ret << "]";
}
return ret.str();
}
static std::string arrayIndexMessage(const Token* tok,
const std::vector<Dimension>& dimensions,
const std::vector<ValueFlow::Value>& indexValues,
const Token* condition)
{
auto add_dim = [](const std::string &s, const Dimension &dim) {
return s + "[" + MathLib::toString(dim.num) + "]";
};
const std::string array = std::accumulate(dimensions.cbegin(), dimensions.cend(), tok->astOperand1()->expressionString(), add_dim);
std::ostringstream errmsg;
if (condition)
errmsg << ValueFlow::eitherTheConditionIsRedundant(condition)
<< " or the array '" << array << "' is accessed at index " << stringifyIndexes(tok->astOperand1()->expressionString(), indexValues) << ", which is out of bounds.";
else
errmsg << "Array '" << array << "' accessed at index " << stringifyIndexes(tok->astOperand1()->expressionString(), indexValues) << ", which is out of bounds.";
return errmsg.str();
}
void CheckBufferOverrun::arrayIndexError(const Token* tok,
const std::vector<Dimension>& dimensions,
const std::vector<ValueFlow::Value>& indexes)
{
if (!tok) {
reportError(tok, Severity::error, "arrayIndexOutOfBounds", "Array 'arr[16]' accessed at index 16, which is out of bounds.", CWE_BUFFER_OVERRUN, Certainty::normal);
reportError(tok, Severity::warning, "arrayIndexOutOfBoundsCond", "Array 'arr[16]' accessed at index 16, which is out of bounds.", CWE_BUFFER_OVERRUN, Certainty::normal);
return;
}
const Token *condition = nullptr;
const ValueFlow::Value *index = nullptr;
for (const ValueFlow::Value& indexValue : indexes) {
if (!indexValue.errorSeverity() && !mSettings->severity.isEnabled(Severity::warning))
return;
if (indexValue.condition)
condition = indexValue.condition;
if (!index || !indexValue.errorPath.empty())
index = &indexValue;
}
reportError(getErrorPath(tok, index, "Array index out of bounds"),
index->errorSeverity() ? Severity::error : Severity::warning,
index->condition ? "arrayIndexOutOfBoundsCond" : "arrayIndexOutOfBounds",
arrayIndexMessage(tok, dimensions, indexes, condition),
CWE_BUFFER_OVERRUN,
index->isInconclusive() ? Certainty::inconclusive : Certainty::normal);
}
void CheckBufferOverrun::negativeIndexError(const Token* tok,
const std::vector<Dimension>& dimensions,
const std::vector<ValueFlow::Value>& indexes)
{
if (!tok) {
reportError(tok, Severity::error, "negativeIndex", "Negative array index", CWE_BUFFER_UNDERRUN, Certainty::normal);
return;
}
const Token *condition = nullptr;
const ValueFlow::Value *negativeValue = nullptr;
for (const ValueFlow::Value& indexValue : indexes) {
if (!indexValue.errorSeverity() && !mSettings->severity.isEnabled(Severity::warning))
return;
if (indexValue.condition)
condition = indexValue.condition;
if (!negativeValue || !indexValue.errorPath.empty())
negativeValue = &indexValue;
}
reportError(getErrorPath(tok, negativeValue, "Negative array index"),
negativeValue->errorSeverity() ? Severity::error : Severity::warning,
"negativeIndex",
arrayIndexMessage(tok, dimensions, indexes, condition),
CWE_BUFFER_UNDERRUN,
negativeValue->isInconclusive() ? Certainty::inconclusive : Certainty::normal);
}
//---------------------------------------------------------------------------
void CheckBufferOverrun::pointerArithmetic()
{
if (!mSettings->severity.isEnabled(Severity::portability))
return;
for (const Token *tok = mTokenizer->tokens(); tok; tok = tok->next()) {
if (!Token::Match(tok, "+|-"))
continue;
if (!tok->valueType() || tok->valueType()->pointer == 0)
continue;
if (!tok->isBinaryOp())
continue;
if (!tok->astOperand1()->valueType() || !tok->astOperand2()->valueType())
continue;
const Token *arrayToken, *indexToken;
if (tok->astOperand1()->valueType()->pointer > 0) {
arrayToken = tok->astOperand1();
indexToken = tok->astOperand2();
} else {
arrayToken = tok->astOperand2();
indexToken = tok->astOperand1();
}
if (!indexToken || !indexToken->valueType() || indexToken->valueType()->pointer > 0 || !indexToken->valueType()->isIntegral())
continue;
std::vector<Dimension> dimensions;
ErrorPath errorPath;
bool mightBeLarger = false;
MathLib::bigint path = 0;
if (!getDimensionsEtc(arrayToken, mSettings, dimensions, errorPath, mightBeLarger, path))
continue;
if (tok->str() == "+") {
// Positive index
if (!mightBeLarger) { // TODO check arrays with dim 1 also
const std::vector<const Token *> indexTokens{indexToken};
const std::vector<ValueFlow::Value>& indexValues =
getOverrunIndexValues(tok, arrayToken, dimensions, indexTokens, path);
if (!indexValues.empty())
pointerArithmeticError(tok, indexToken, &indexValues.front());
}
if (const ValueFlow::Value *neg = indexToken->getValueLE(-1, mSettings))
pointerArithmeticError(tok, indexToken, neg);
} else if (tok->str() == "-") {
if (arrayToken->variable() && arrayToken->variable()->isArgument())
continue;
const Token *array = arrayToken;
while (Token::Match(array, ".|::"))
array = array->astOperand2();
if (array->variable() && array->variable()->isArray()) {
const ValueFlow::Value *v = indexToken->getValueGE(1, mSettings);
if (v)
pointerArithmeticError(tok, indexToken, v);
}
}
}
}
void CheckBufferOverrun::pointerArithmeticError(const Token *tok, const Token *indexToken, const ValueFlow::Value *indexValue)
{
if (!tok) {
reportError(tok, Severity::portability, "pointerOutOfBounds", "Pointer arithmetic overflow.", CWE_POINTER_ARITHMETIC_OVERFLOW, Certainty::normal);
reportError(tok, Severity::portability, "pointerOutOfBoundsCond", "Pointer arithmetic overflow.", CWE_POINTER_ARITHMETIC_OVERFLOW, Certainty::normal);
return;
}
std::string errmsg;
if (indexValue->condition)
errmsg = "Undefined behaviour, when '" + indexToken->expressionString() + "' is " + MathLib::toString(indexValue->intvalue) + " the pointer arithmetic '" + tok->expressionString() + "' is out of bounds.";
else
errmsg = "Undefined behaviour, pointer arithmetic '" + tok->expressionString() + "' is out of bounds.";
reportError(getErrorPath(tok, indexValue, "Pointer arithmetic overflow"),
Severity::portability,
indexValue->condition ? "pointerOutOfBoundsCond" : "pointerOutOfBounds",
errmsg,
CWE_POINTER_ARITHMETIC_OVERFLOW,
indexValue->isInconclusive() ? Certainty::inconclusive : Certainty::normal);
}
//---------------------------------------------------------------------------
ValueFlow::Value CheckBufferOverrun::getBufferSize(const Token *bufTok) const
{
if (!bufTok->valueType())
return ValueFlow::Value(-1);
const Variable *var = bufTok->variable();
if (!var || var->dimensions().empty()) {
const ValueFlow::Value *value = getBufferSizeValue(bufTok);
if (value)
return *value;
}
if (!var)
return ValueFlow::Value(-1);
const MathLib::bigint dim = std::accumulate(var->dimensions().cbegin(), var->dimensions().cend(), 1LL, [](MathLib::bigint i1, const Dimension &dim) {
return i1 * dim.num;
});
ValueFlow::Value v;
v.setKnown();
v.valueType = ValueFlow::Value::ValueType::BUFFER_SIZE;
if (var->isPointerArray())
v.intvalue = dim * mSettings->platform.sizeof_pointer;
else if (var->isPointer())
return ValueFlow::Value(-1);
else {
const MathLib::bigint typeSize = bufTok->valueType()->typeSize(mSettings->platform);
v.intvalue = dim * typeSize;
}
return v;
}
//---------------------------------------------------------------------------
static bool checkBufferSize(const Token *ftok, const Library::ArgumentChecks::MinSize &minsize, const std::vector<const Token *> &args, const MathLib::bigint bufferSize, const Settings *settings, const Tokenizer* tokenizer)
{
const Token * const arg = (minsize.arg > 0 && minsize.arg - 1 < args.size()) ? args[minsize.arg - 1] : nullptr;
const Token * const arg2 = (minsize.arg2 > 0 && minsize.arg2 - 1 < args.size()) ? args[minsize.arg2 - 1] : nullptr;
switch (minsize.type) {
case Library::ArgumentChecks::MinSize::Type::STRLEN:
if (settings->library.isargformatstr(ftok, minsize.arg)) {
return getMinFormatStringOutputLength(args, minsize.arg) < bufferSize;
} else if (arg) {
const Token *strtoken = arg->getValueTokenMaxStrLength();
if (strtoken)
return Token::getStrLength(strtoken) < bufferSize;
}
break;
case Library::ArgumentChecks::MinSize::Type::ARGVALUE: {
if (arg && arg->hasKnownIntValue()) {
MathLib::bigint myMinsize = arg->getKnownIntValue();
const unsigned int baseSize = tokenizer->sizeOfType(minsize.baseType);
if (baseSize != 0)
myMinsize *= baseSize;
return myMinsize <= bufferSize;
}
break;
}
case Library::ArgumentChecks::MinSize::Type::SIZEOF:
// TODO
break;
case Library::ArgumentChecks::MinSize::Type::MUL:
if (arg && arg2 && arg->hasKnownIntValue() && arg2->hasKnownIntValue())
return (arg->getKnownIntValue() * arg2->getKnownIntValue()) <= bufferSize;
break;
case Library::ArgumentChecks::MinSize::Type::VALUE: {
MathLib::bigint myMinsize = minsize.value;
const unsigned int baseSize = tokenizer->sizeOfType(minsize.baseType);
if (baseSize != 0)
myMinsize *= baseSize;
return myMinsize <= bufferSize;
}
case Library::ArgumentChecks::MinSize::Type::NONE:
break;
}
return true;
}
void CheckBufferOverrun::bufferOverflow()
{
const SymbolDatabase *symbolDatabase = mTokenizer->getSymbolDatabase();
for (const Scope * scope : symbolDatabase->functionScopes) {
for (const Token *tok = scope->bodyStart; tok != scope->bodyEnd; tok = tok->next()) {
if (!Token::Match(tok, "%name% (") || Token::simpleMatch(tok, ") {"))
continue;
if (!mSettings->library.hasminsize(tok))
continue;
const std::vector<const Token *> args = getArguments(tok);
for (int argnr = 0; argnr < args.size(); ++argnr) {
if (!args[argnr]->valueType() || args[argnr]->valueType()->pointer == 0)
continue;
const std::vector<Library::ArgumentChecks::MinSize> *minsizes = mSettings->library.argminsizes(tok, argnr + 1);
if (!minsizes || minsizes->empty())
continue;
// Get buffer size..
const Token *argtok = args[argnr];
while (argtok && argtok->isCast())
argtok = argtok->astOperand2() ? argtok->astOperand2() : argtok->astOperand1();
while (Token::Match(argtok, ".|::"))
argtok = argtok->astOperand2();
if (!argtok || !argtok->variable())
continue;
if (argtok->valueType() && argtok->valueType()->pointer == 0)
continue;
// TODO: strcpy(buf+10, "hello");
const ValueFlow::Value bufferSize = getBufferSize(argtok);
if (bufferSize.intvalue <= 0)
continue;
// buffer size == 1 => do not warn for dynamic memory
if (bufferSize.intvalue == 1 && Token::simpleMatch(argtok->astParent(), ".")) { // TODO: check if parent was allocated dynamically
const Token *tok2 = argtok;
while (Token::simpleMatch(tok2->astParent(), "."))
tok2 = tok2->astParent();
while (Token::Match(tok2, "[|."))
tok2 = tok2->astOperand1();
const Variable *var = tok2 ? tok2->variable() : nullptr;
if (var) {
if (var->isPointer())
continue;
if (var->isArgument() && var->isReference())
continue;
}
}
const bool error = std::none_of(minsizes->begin(), minsizes->end(), [=](const Library::ArgumentChecks::MinSize &minsize) {
return checkBufferSize(tok, minsize, args, bufferSize.intvalue, mSettings, mTokenizer);
});
if (error)
bufferOverflowError(args[argnr], &bufferSize, Certainty::normal);
}
}
}
}
void CheckBufferOverrun::bufferOverflowError(const Token *tok, const ValueFlow::Value *value, Certainty certainty)
{
reportError(getErrorPath(tok, value, "Buffer overrun"), Severity::error, "bufferAccessOutOfBounds", "Buffer is accessed out of bounds: " + (tok ? tok->expressionString() : "buf"), CWE_BUFFER_OVERRUN, certainty);
}
//---------------------------------------------------------------------------
void CheckBufferOverrun::arrayIndexThenCheck()
{
if (!mSettings->severity.isEnabled(Severity::portability))
return;
const SymbolDatabase *symbolDatabase = mTokenizer->getSymbolDatabase();
for (const Scope * const scope : symbolDatabase->functionScopes) {
for (const Token *tok = scope->bodyStart; tok && tok != scope->bodyEnd; tok = tok->next()) {
if (Token::simpleMatch(tok, "sizeof (")) {
tok = tok->linkAt(1);
continue;
}
if (Token::Match(tok, "%name% [ %var% ]")) {
tok = tok->next();
const int indexID = tok->next()->varId();
const std::string& indexName(tok->strAt(1));
// Iterate AST upwards
const Token* tok2 = tok;
const Token* tok3 = tok2;
while (tok2->astParent() && tok2->tokType() != Token::eLogicalOp && tok2->str() != "?") {
tok3 = tok2;
tok2 = tok2->astParent();
}
// Ensure that we ended at a logical operator and that we came from its left side
if (tok2->tokType() != Token::eLogicalOp || tok2->astOperand1() != tok3)
continue;
// check if array index is ok
// statement can be closed in parentheses, so "(| " is using
if (Token::Match(tok2, "&& (| %varid% <|<=", indexID))
arrayIndexThenCheckError(tok, indexName);
else if (Token::Match(tok2, "&& (| %any% >|>= %varid% !!+", indexID))
arrayIndexThenCheckError(tok, indexName);
}
}
}
}
void CheckBufferOverrun::arrayIndexThenCheckError(const Token *tok, const std::string &indexName)
{
reportError(tok, Severity::style, "arrayIndexThenCheck",
"$symbol:" + indexName + "\n"
"Array index '$symbol' is used before limits check.\n"
"Defensive programming: The variable '$symbol' is used as an array index before it "
"is checked that is within limits. This can mean that the array might be accessed out of bounds. "
"Reorder conditions such as '(a[i] && i < 10)' to '(i < 10 && a[i])'. That way the array will "
"not be accessed if the index is out of limits.", CWE_ARRAY_INDEX_THEN_CHECK, Certainty::normal);
}
//---------------------------------------------------------------------------
void CheckBufferOverrun::stringNotZeroTerminated()
{
// this is currently 'inconclusive'. See TestBufferOverrun::terminateStrncpy3
if (!mSettings->severity.isEnabled(Severity::warning) || !mSettings->certainty.isEnabled(Certainty::inconclusive))
return;
const SymbolDatabase *symbolDatabase = mTokenizer->getSymbolDatabase();
for (const Scope * const scope : symbolDatabase->functionScopes) {
for (const Token *tok = scope->bodyStart; tok && tok != scope->bodyEnd; tok = tok->next()) {
if (!Token::simpleMatch(tok, "strncpy ("))
continue;
const std::vector<const Token *> args = getArguments(tok);
if (args.size() != 3)
continue;
const Token *sizeToken = args[2];
if (!sizeToken->hasKnownIntValue())
continue;
const ValueFlow::Value &bufferSize = getBufferSize(args[0]);
if (bufferSize.intvalue < 0 || sizeToken->getKnownIntValue() < bufferSize.intvalue)
continue;
if (Token::simpleMatch(args[1], "(") && Token::simpleMatch(args[1]->astOperand1(), ". c_str") && args[1]->astOperand1()->astOperand1()) {
const std::list<ValueFlow::Value>& contValues = args[1]->astOperand1()->astOperand1()->values();
auto it = std::find_if(contValues.cbegin(), contValues.cend(), [](const ValueFlow::Value& value) {
return value.isContainerSizeValue() && !value.isImpossible();
});
if (it != contValues.end() && it->intvalue < sizeToken->getKnownIntValue())
continue;
} else {
const Token* srcValue = args[1]->getValueTokenMaxStrLength();
if (srcValue && Token::getStrLength(srcValue) < sizeToken->getKnownIntValue())
continue;
}
// Is the buffer zero terminated after the call?
bool isZeroTerminated = false;
for (const Token *tok2 = tok->next()->link(); tok2 != scope->bodyEnd; tok2 = tok2->next()) {
if (!Token::simpleMatch(tok2, "] ="))
continue;
const Token *rhs = tok2->next()->astOperand2();
if (!rhs || !rhs->hasKnownIntValue() || rhs->getKnownIntValue() != 0)
continue;
if (isSameExpression(mTokenizer->isCPP(), false, args[0], tok2->link()->astOperand1(), mSettings->library, false, false))
isZeroTerminated = true;
}
if (isZeroTerminated)
continue;
// TODO: Locate unsafe string usage..
terminateStrncpyError(tok, args[0]->expressionString());
}
}
}
void CheckBufferOverrun::terminateStrncpyError(const Token *tok, const std::string &varname)
{
const std::string shortMessage = "The buffer '$symbol' may not be null-terminated after the call to strncpy().";
reportError(tok, Severity::warning, "terminateStrncpy",
"$symbol:" + varname + '\n' +
shortMessage + '\n' +
shortMessage + ' ' +
"If the source string's size fits or exceeds the given size, strncpy() does not add a "
"zero at the end of the buffer. This causes bugs later in the code if the code "
"assumes buffer is null-terminated.", CWE170, Certainty::inconclusive);
}
//---------------------------------------------------------------------------
void CheckBufferOverrun::argumentSize()
{
// Check '%type% x[10]' arguments
if (!mSettings->severity.isEnabled(Severity::warning))
return;
const SymbolDatabase *symbolDatabase = mTokenizer->getSymbolDatabase();
for (const Scope * const scope : symbolDatabase->functionScopes) {
for (const Token *tok = scope->bodyStart; tok != scope->bodyEnd; tok = tok->next()) {
if (!tok->function() || !Token::Match(tok, "%name% ("))
continue;
// If argument is '%type% a[num]' then check bounds against num
const Function *callfunc = tok->function();
const std::vector<const Token *> callargs = getArguments(tok);
for (nonneg int paramIndex = 0; paramIndex < callargs.size() && paramIndex < callfunc->argCount(); ++paramIndex) {
const Variable* const argument = callfunc->getArgumentVar(paramIndex);
if (!argument || !argument->nameToken() || !argument->isArray())
continue;
if (!argument->valueType() || !callargs[paramIndex]->valueType())
continue;
if (argument->valueType()->type != callargs[paramIndex]->valueType()->type)
continue;
const Token * calldata = callargs[paramIndex];
while (Token::Match(calldata, "::|."))
calldata = calldata->astOperand2();
if (!calldata->variable() || !calldata->variable()->isArray())
continue;
if (calldata->variable()->dimensions().size() != argument->dimensions().size())
continue;
bool err = false;
for (int d = 0; d < argument->dimensions().size(); ++d) {
const auto& dim1 = calldata->variable()->dimensions()[d];
const auto& dim2 = argument->dimensions()[d];
if (!dim1.known || !dim2.known)
break;
if (dim1.num < dim2.num)
err = true;
}
if (err)
argumentSizeError(tok, tok->str(), paramIndex, callargs[paramIndex]->expressionString(), calldata->variable(), argument);
}
}
}
}
void CheckBufferOverrun::argumentSizeError(const Token *tok, const std::string &functionName, nonneg int paramIndex, const std::string &paramExpression, const Variable *paramVar, const Variable *functionArg)
{
const std::string strParamNum = std::to_string(paramIndex + 1) + getOrdinalText(paramIndex + 1);
ErrorPath errorPath;
errorPath.emplace_back(tok, "Function '" + functionName + "' is called");
if (functionArg)
errorPath.emplace_back(functionArg->nameToken(), "Declaration of " + strParamNum + " function argument.");
if (paramVar)
errorPath.emplace_back(paramVar->nameToken(), "Passing buffer '" + paramVar->name() + "' to function that is declared here");
errorPath.emplace_back(tok, "");
reportError(errorPath, Severity::warning, "argumentSize",
"$symbol:" + functionName + '\n' +
"Buffer '" + paramExpression + "' is too small, the function '" + functionName + "' expects a bigger buffer in " + strParamNum + " argument", CWE_ARGUMENT_SIZE, Certainty::normal);
}
//---------------------------------------------------------------------------
// CTU..
//---------------------------------------------------------------------------
std::string CheckBufferOverrun::MyFileInfo::toString() const
{
std::string xml;
if (!unsafeArrayIndex.empty())
xml = " <array-index>\n" + CTU::toString(unsafeArrayIndex) + " </array-index>\n";
if (!unsafePointerArith.empty())
xml += " <pointer-arith>\n" + CTU::toString(unsafePointerArith) + " </pointer-arith>\n";
return xml;
}
bool CheckBufferOverrun::isCtuUnsafeBufferUsage(const Check *check, const Token *argtok, MathLib::bigint *offset, int type)
{
if (!offset)
return false;
const CheckBufferOverrun *c = dynamic_cast<const CheckBufferOverrun *>(check);
if (!c)
return false;
if (!argtok->valueType() || argtok->valueType()->typeSize(c->mSettings->platform) == 0)
return false;
const Token *indexTok = nullptr;
if (type == 1 && Token::Match(argtok, "%name% [") && argtok->astParent() == argtok->next() && !Token::simpleMatch(argtok->linkAt(1), "] ["))
indexTok = argtok->next()->astOperand2();
else if (type == 2 && Token::simpleMatch(argtok->astParent(), "+"))
indexTok = (argtok == argtok->astParent()->astOperand1()) ?
argtok->astParent()->astOperand2() :
argtok->astParent()->astOperand1();
if (!indexTok)
return false;
if (!indexTok->hasKnownIntValue())
return false;
*offset = indexTok->getKnownIntValue() * argtok->valueType()->typeSize(c->mSettings->platform);
return true;
}
bool CheckBufferOverrun::isCtuUnsafeArrayIndex(const Check *check, const Token *argtok, MathLib::bigint *offset)
{
return CheckBufferOverrun::isCtuUnsafeBufferUsage(check, argtok, offset, 1);
}
bool CheckBufferOverrun::isCtuUnsafePointerArith(const Check *check, const Token *argtok, MathLib::bigint *offset)
{
return CheckBufferOverrun::isCtuUnsafeBufferUsage(check, argtok, offset, 2);
}
/** @brief Parse current TU and extract file info */
Check::FileInfo *CheckBufferOverrun::getFileInfo(const Tokenizer *tokenizer, const Settings *settings) const
{
CheckBufferOverrun checkBufferOverrun(tokenizer, settings, nullptr);
MyFileInfo *fileInfo = new MyFileInfo;
fileInfo->unsafeArrayIndex = CTU::getUnsafeUsage(tokenizer, settings, &checkBufferOverrun, isCtuUnsafeArrayIndex);
fileInfo->unsafePointerArith = CTU::getUnsafeUsage(tokenizer, settings, &checkBufferOverrun, isCtuUnsafePointerArith);
if (fileInfo->unsafeArrayIndex.empty() && fileInfo->unsafePointerArith.empty()) {
delete fileInfo;
return nullptr;
}
return fileInfo;
}
Check::FileInfo * CheckBufferOverrun::loadFileInfoFromXml(const tinyxml2::XMLElement *xmlElement) const
{
const std::string arrayIndex("array-index");
const std::string pointerArith("pointer-arith");
MyFileInfo *fileInfo = new MyFileInfo;
for (const tinyxml2::XMLElement *e = xmlElement->FirstChildElement(); e; e = e->NextSiblingElement()) {
if (e->Name() == arrayIndex)
fileInfo->unsafeArrayIndex = CTU::loadUnsafeUsageListFromXml(e);
else if (e->Name() == pointerArith)
fileInfo->unsafePointerArith = CTU::loadUnsafeUsageListFromXml(e);
}
if (fileInfo->unsafeArrayIndex.empty() && fileInfo->unsafePointerArith.empty()) {
delete fileInfo;
return nullptr;
}
return fileInfo;
}
/** @brief Analyse all file infos for all TU */
bool CheckBufferOverrun::analyseWholeProgram(const CTU::FileInfo *ctu, const std::list<Check::FileInfo*> &fileInfo, const Settings& settings, ErrorLogger &errorLogger)
{
if (!ctu)
return false;
bool foundErrors = false;
(void)settings; // This argument is unused
const std::map<std::string, std::list<const CTU::FileInfo::CallBase *>> callsMap = ctu->getCallsMap();
for (Check::FileInfo *fi1 : fileInfo) {
const MyFileInfo *fi = dynamic_cast<const MyFileInfo*>(fi1);
if (!fi)
continue;
for (const CTU::FileInfo::UnsafeUsage &unsafeUsage : fi->unsafeArrayIndex)
foundErrors |= analyseWholeProgram1(callsMap, unsafeUsage, 1, errorLogger);
for (const CTU::FileInfo::UnsafeUsage &unsafeUsage : fi->unsafePointerArith)
foundErrors |= analyseWholeProgram1(callsMap, unsafeUsage, 2, errorLogger);
}
return foundErrors;
}
bool CheckBufferOverrun::analyseWholeProgram1(const std::map<std::string, std::list<const CTU::FileInfo::CallBase *>> &callsMap, const CTU::FileInfo::UnsafeUsage &unsafeUsage, int type, ErrorLogger &errorLogger)
{
const CTU::FileInfo::FunctionCall *functionCall = nullptr;
const std::list<ErrorMessage::FileLocation> &locationList =
CTU::FileInfo::getErrorPath(CTU::FileInfo::InvalidValueType::bufferOverflow,
unsafeUsage,
callsMap,
"Using argument ARG",
&functionCall,
false);
if (locationList.empty())
return false;
const char *errorId = nullptr;
std::string errmsg;
CWE cwe(0);
if (type == 1) {
errorId = "ctuArrayIndex";
if (unsafeUsage.value > 0)
errmsg = "Array index out of bounds; '" + unsafeUsage.myArgumentName + "' buffer size is " + MathLib::toString(functionCall->callArgValue) + " and it is accessed at offset " + MathLib::toString(unsafeUsage.value) + ".";
else
errmsg = "Array index out of bounds; buffer '" + unsafeUsage.myArgumentName + "' is accessed at offset " + MathLib::toString(unsafeUsage.value) + ".";
cwe = (unsafeUsage.value > 0) ? CWE_BUFFER_OVERRUN : CWE_BUFFER_UNDERRUN;
} else {
errorId = "ctuPointerArith";
errmsg = "Pointer arithmetic overflow; '" + unsafeUsage.myArgumentName + "' buffer size is " + MathLib::toString(functionCall->callArgValue);
cwe = CWE_POINTER_ARITHMETIC_OVERFLOW;
}
const ErrorMessage errorMessage(locationList,
emptyString,
Severity::error,
errmsg,
errorId,
cwe, Certainty::normal);
errorLogger.reportErr(errorMessage);
return true;
}
void CheckBufferOverrun::objectIndex()
{
const SymbolDatabase *symbolDatabase = mTokenizer->getSymbolDatabase();
for (const Scope *functionScope : symbolDatabase->functionScopes) {
for (const Token *tok = functionScope->bodyStart; tok != functionScope->bodyEnd; tok = tok->next()) {
if (!Token::simpleMatch(tok, "["))
continue;
const Token *obj = tok->astOperand1();
const Token *idx = tok->astOperand2();
if (!idx || !obj)
continue;
if (idx->hasKnownIntValue()) {
if (idx->getKnownIntValue() == 0)
continue;
}
if (idx->hasKnownIntValue() && idx->getKnownIntValue() == 0)
continue;
std::vector<ValueFlow::Value> values = ValueFlow::getLifetimeObjValues(obj, false, -1);
for (const ValueFlow::Value& v:values) {
if (v.lifetimeKind != ValueFlow::Value::LifetimeKind::Address)
continue;
const Variable *var = v.tokvalue->variable();
if (!var)
continue;
if (var->isReference())
continue;
if (var->isRValueReference())
continue;
if (var->isArray())
continue;
if (var->isPointer()) {
if (!var->valueType())
continue;
if (!obj->valueType())
continue;
if (var->valueType()->pointer > obj->valueType()->pointer)
continue;
}
if (obj->valueType() && var->valueType() && (obj->isCast() || (mTokenizer->isCPP() && isCPPCast(obj)) || obj->valueType()->pointer)) { // allow cast to a different type
const auto varSize = var->valueType()->typeSize(mSettings->platform);
if (varSize == 0)
continue;
if (obj->valueType()->type != var->valueType()->type) {
if (ValueFlow::isOutOfBounds(makeSizeValue(varSize, v.path), idx).empty())
continue;
}
}
if (v.path != 0) {
std::vector<ValueFlow::Value> idxValues;
std::copy_if(idx->values().cbegin(),
idx->values().cend(),
std::back_inserter(idxValues),
[&](const ValueFlow::Value& vidx) {
if (!vidx.isIntValue())
return false;
return vidx.path == v.path || vidx.path == 0;
});
if (std::any_of(idxValues.cbegin(), idxValues.cend(), [&](const ValueFlow::Value& vidx) {
if (vidx.isImpossible())
return (vidx.intvalue == 0);
else
return (vidx.intvalue != 0);
})) {
objectIndexError(tok, &v, idx->hasKnownIntValue());
}
} else {
objectIndexError(tok, &v, idx->hasKnownIntValue());
}
}
}
}
}
void CheckBufferOverrun::objectIndexError(const Token *tok, const ValueFlow::Value *v, bool known)
{
ErrorPath errorPath;
std::string name;
if (v) {
name = v->tokvalue->variable()->name();
errorPath = v->errorPath;
}
errorPath.emplace_back(tok, "");
std::string verb = known ? "is" : "might be";
reportError(errorPath,
known ? Severity::error : Severity::warning,
"objectIndex",
"The address of local variable '" + name + "' " + verb + " accessed at non-zero index.",
CWE758,
Certainty::normal);
}
static bool isVLAIndex(const Token* tok)
{
if (!tok)
return false;
if (tok->varId() != 0U)
return true;
if (tok->str() == "?") {
// this is a VLA index if both expressions around the ":" is VLA index
return tok->astOperand2() &&
tok->astOperand2()->str() == ":" &&
isVLAIndex(tok->astOperand2()->astOperand1()) &&
isVLAIndex(tok->astOperand2()->astOperand2());
}
return isVLAIndex(tok->astOperand1()) || isVLAIndex(tok->astOperand2());
}
void CheckBufferOverrun::negativeArraySize()
{
const SymbolDatabase* symbolDatabase = mTokenizer->getSymbolDatabase();
for (const Variable* var : symbolDatabase->variableList()) {
if (!var || !var->isArray())
continue;
const Token* const nameToken = var->nameToken();
if (!Token::Match(nameToken, "%var% [") || !nameToken->next()->astOperand2())
continue;
const ValueFlow::Value* sz = nameToken->next()->astOperand2()->getValueLE(-1, mSettings);
// don't warn about constant negative index because that is a compiler error
if (sz && isVLAIndex(nameToken->next()->astOperand2()))
negativeArraySizeError(nameToken);
}
for (const Scope* functionScope : symbolDatabase->functionScopes) {
for (const Token* tok = functionScope->bodyStart; tok != functionScope->bodyEnd; tok = tok->next()) {
if (!tok->isKeyword() || tok->str() != "new" || !tok->astOperand1() || tok->astOperand1()->str() != "[")
continue;
const Token* valOperand = tok->astOperand1()->astOperand2();
if (!valOperand)
continue;
const ValueFlow::Value* sz = valOperand->getValueLE(-1, mSettings);
if (sz)
negativeMemoryAllocationSizeError(tok, sz);
}
}
}
void CheckBufferOverrun::negativeArraySizeError(const Token* tok)
{
const std::string arrayName = tok ? tok->expressionString() : std::string();
const std::string line1 = arrayName.empty() ? std::string() : ("$symbol:" + arrayName + '\n');
reportError(tok, Severity::error, "negativeArraySize",
line1 +
"Declaration of array '" + arrayName + "' with negative size is undefined behaviour", CWE758, Certainty::normal);
}
void CheckBufferOverrun::negativeMemoryAllocationSizeError(const Token* tok, const ValueFlow::Value* value)
{
const std::string msg = "Memory allocation size is negative.";
const ErrorPath errorPath = getErrorPath(tok, value, msg);
const bool inconclusive = value != nullptr && !value->isKnown();
reportError(errorPath, inconclusive ? Severity::warning : Severity::error, "negativeMemoryAllocationSize",
msg, CWE131, inconclusive ? Certainty::inconclusive : Certainty::normal);
}