We are often asked if our static analyzer PVS-Studio can detect memory leaks. To avoid emailing the same text again and again, we decided to post a detailed answer in our blog. Yes, PVS-Studio can detect memory leaks and leaks of other resources. This is achieved by means of several special diagnostics, the use of which will be demonstrated in this article. The examples are taken from real projects.
A memory leak is a type of resource leak that occurs when a computer program incorrectly manages memory allocations in such a way that memory which is no longer needed is not released. In object-oriented programming, a memory leak may happen when an object is stored in memory but cannot be accessed by the running code. In CWE classification, memory leaks are known as CWE-401 weaknesses.
Memory leaks are one of the types of resource leaks. An example of another type of a leak is a file handle leak: it occurs when the program opens a file but doesn't close it and fails to return the file handle to the operating system. In CWE classification, these defects are given the code CWE-404.
Memory and resource leaks may cause Denial of Service errors.
Memory and resource leaks are detected by dynamic and static code analyzers. Our static analyzer PVS-Studio is one of such tools.
PVS-Studio uses the following diagnostics to detect these types of errors:
Let's take a look at several examples of memory leaks detected by PVS-Studio in the source code of open-source projects.
Example 1.
Project NetDefender. PVS-Studio diagnostic message: V773 The 'm_pColumns' pointer was not released in destructor. A memory leak is possible. fireview.cpp 95
Note that two objects are created in the constructor:
CFireView::CFireView() : CFormView(CFireView::IDD)
{
m_pBrush = new CBrush;
ASSERT(m_pBrush);
m_clrBk = RGB(148, 210, 252);
m_clrText = RGB(0, 0, 0);
m_pBrush->CreateSolidBrush(m_clrBk);
m_pColumns = new CStringList;
ASSERT(m_pColumns);
_rows = 1;
start = TRUE;
block = TRUE;
allow = TRUE;
ping = TRUE;
m_style=StyleTile;
}However, only one object, which address is stored in the m_pBrush variable, is destroyed in the destructor:
CFireView::~CFireView()
{
if(m_pBrush)
{
delete m_pBrush;
}
}The programmer must have simply forgotten about the m_pColumns variable, which results in a memory leak.
Example 2.
Project Far2l (Linux port of FAR v2). The interesting thing about this bug is that it triggers two different PVS-Studio diagnostics at once:
BOOL WINAPI _export SEVENZ_OpenArchive(const char *Name,
int *Type)
{
Traverser *t = new Traverser(Name);
if (!t->Valid())
{
return FALSE;
delete t;
}
delete s_selected_traverser;
s_selected_traverser = t;
return TRUE;
}The return and delete operator are swapped. As a result, delete will never be executed. The analyzer issues two messages: one about unreachable code, the other about a memory leak.
Example 3.
Project Firebird. PVS-Studio diagnostic message: V701 realloc() possible leak: when realloc() fails in allocating memory, original pointer 's->base' is lost. Consider assigning realloc() to a temporary pointer. mstring.c 42
int mputchar(struct mstring *s, int ch)
{
if (!s || !s->base) return ch;
if (s->ptr == s->end) {
int len = s->end - s->base;
if ((s->base = realloc(s->base, len+len+TAIL))) {
s->ptr = s->base + len;
s->end = s->base + len+len+TAIL; }
else {
s->ptr = s->end = 0;
return ch;
}
}
*s->ptr++ = ch;
return ch;
}The function in question adds a character to a string. The buffer storing the string is extended by calling function realloc. The problem here is that if realloc fails to increase the buffer's size, a memory leak will occur. This happens because when there is no available memory block large enough, the realloc function returns NULL without releasing the previous block. Since the calling function's return result is immediately written to variable s->base, it is simply impossible to free the previously allocated storage.
The bug can be fixed by adding a temporary variable and a call to function free:
int mputchar(struct mstring *s, int ch)
{
if (!s || !s->base) return ch;
if (s->ptr == s->end) {
void *old = s->base;
int len = s->end - s->base;
if ((s->base = realloc(s->base, len+len+TAIL))) {
s->ptr = s->base + len;
s->end = s->base + len+len+TAIL; }
else {
free(old);
s->ptr = s->end = 0;
return ch;
}
}
*s->ptr++ = ch;
return ch;
}PVS-Studio's diagnostics are a good example of how static analyzers can detect various types of resource leaks. It should be noted, however, that static analyzers perform worse than dynamic analyzers in this field.
Static analyzers find bugs by examining the source code and keeping track of how pointers are used, which is a very difficult task. Pointers may be passed between functions in tricky ways, so the analyzer is not always able to follow them and notice potential memory leaks. In some cases, it's simply impossible as the analyzer doesn't know what input data will be fed to the program.
Detecting memory or resource leaks is much easier for dynamic analyzers because they don't have to keep track of the data. They just need to remember the location inside the program where a resource is allocated, and check if the program has released it before exiting. If it hasn't, then it's a bug. So, dynamic analyzers are more careful and reliable in detecting the various types of leaks.
This doesn't mean that dynamic analysis is more powerful than static analysis. Both methodologies have strong and weak points of their own. Detecting resource leaks is something that dynamic analyzers are better at. But in other fields such as search of typos and unreachable code, they are ineffective or useless at all.
Don't view it as "static analysis vs dynamic analysis". They don't compete; they complement each other. If you want to improve your code's quality and reliability, you should use both types of analysis. I've written a lot about it, and I don't feel like repeating myself. For further reading, see the following articles:
PVS-Studio static analyzer can detect a wide range of problems caused by memory and resource leaks. Use it regularly to have bugs eliminated as soon as they appear in your code or during night builds:
The PVS-Studio team wishes you bugless code!