How Can PVS-Studio Help in the Detection of Vulnerabilities?

• PVS-Studio is a Tool that Prevents not Only Bugs, but also Vulnerabilities
• Terminology
• PVS-Studio: A Different Point of View
  • Background
  • The relation between PVS-Studio and CWE
• CVE Bases
• Some CVE that Could Have Been Found Using PVS-Studio
  • illumos-gate
  • Network Audio System
  • Ytnef (Yerase's TNEF Stream Reader)
  • MySQL
  • iOS
• Effective Use of Static Analysis
• Conclusion
• Additional Links

A vulnerability in terms of computer security, is a flaw in the system allowing someone to violate the integrity, or deliberately cause a malfunction, of the program. Practice shows that even a seemingly insignificant bug can be a serious vulnerability. Vulnerabilities can be avoided by using different methods of validation and verification of software, including static analysis. This article will cover the topic of how PVS-Studio copes with the task of vulnerability search.

PVS-Studio is a Tool that Prevents not Only Bugs, but also Vulnerabilities

Later in the article I will tell how we came to this conclusion. But first, I would like to say a few words about PVS-Studio itself.

PVS-Studio is a static code analyzer that searches for bugs (and vulnerabilities) in programs written in C, C++, and C#. It works under Windows and Linux, and can be integrated into Visual Studio IDE as a plugin. At this point the analyzer has more than 450 diagnostic rules, each of them is described in the documentation.

By the time this article was posted, we had checked more than 280 open source projects, where we found more than 11 000 errors. It's quite interesting, the number of these bugs which are real vulnerabilities...

You can download PVS-Studio on the official site, and try it yourself.

By the way, we offer PVS-Studio licenses to security experts. If you are an expert in the field of security, and search for vulnerabilities, you may contact us to get a license. More details about this offer can be found in the article "Handing out PVS-Studio Analyzer Licenses to Security Experts".

Terminology

In the case that you are well aware of the terminology, and know the differences between CVE and CWE as well as their similarities, you may skip this section. Still, I suggest that everybody else to take a look at it, so it will be easier to understand the topic in the future.

CWE (Common Weakness Enumeration) - a combined list of security defects. Targeted at both the development community and the community of security practitioners, Common Weakness Enumeration (CWE) is a formal list or dictionary of common software weaknesses that can occur in software's architecture, design, code, or implementation that can lead to exploitable security vulnerabilities. CWE was created to serve as a common language for describing software security weaknesses; as a standard measuring stick for software security tools targeting these weaknesses; and to provide a common baseline standard for weakness identification, mitigation, and prevention efforts.

CVE (Common Vulnerabilities and Exposures) - program errors that can be directly used by hackers.

MITRE corporation started working on the classification of software vulnerabilities in 1999, when the list of common vulnerabilities and the software liabilities (CVE) came into being. In 2005 within the framework of further development of the CVE system, a team of authors started the work on the preparatory classification of vulnerabilities, attacks, crashes and other kinds of security issues with a view to define common software security defects. However, despite the self-sufficiency of the classification created in the scope of CVE, it appeared to be too rough for the definition and classification of methods of code security assessment, used by the analyzers. Thus, CWE list was created to resolve this problem.

PVS-Studio: A Different Point of View

Background

Historically, we have positioned PVS-Studio as a tool to search for errors. In the articles about our project analyses, we have always used corresponding terminology: a bug, an error, a typo. It's clear that different errors have different levels of severity: there may be some code fragments that contain redundant or misleading code, but there are some errors that cause the whole system to crash at 5 in the morning every third day. Everything was clear, this concept didn't go any further for a long time - errors were just errors.

However, over time, it turned out that some of the errors detected by PVS-Studio can be more serious. For example, incorrectly used printf function can cause many more negative consequences than the output of a wrong message in stdout. When it became clear that quite a number of diagnostic rules can detect not only errors, but weaknesses (CWE), we decided to investigate this question in more detail and see how the diagnostic rules of PVS-Studio can be related to CWE.

The relation between PVS-Studio and CWE

Note. The table in this article is outdated. The up-to-date table that classifies PVS-Studio warnings according to CWE is available here.

Based on the results of detecting the correlation between the warnings of PVS-Studio and CWE we created the following table:

CWE	PVS-Studio	CWE Description
CWE-14	V597	Compiler Removal of Code to Clear Buffers
CWE-36	V631, V3039	Absolute Path Traversal
CWE-121	V755	Stack-based Buffer Overflow
CWE-122	V755	Heap-based Buffer Overflow
CWE-123	V575	Write-what-where Condition
CWE-129	V557, V781, V3106	Improper Validation of Array Index
CWE-190	V636	Integer Overflow or Wraparound
CWE-193	V645	Off-by-one Error
CWE-252	V522, V575	Unchecked Return Value
CWE-253	V544, V545, V676, V716, V721, V724	Incorrect Check of Function Return Value
CWE-390	V565	Detection of Error Condition Without Action
CWE-476	V522, V595, V664, V757, V769, V3019, V3042, V3080, V3095, V3105, V3125	NULL Pointer Dereference
CWE-481	V559, V3055	Assigning instead of comparing
CWE-482	V607	Comparing instead of Assigning
CWE-587	V566	Assignment of a Fixed Address to a Pointer
CWE-369	V609, V3064	Divide By Zero
CWE-416	V723, V774	Use after free
CWE-467	V511, V512, V568	Use of sizeof() on a Pointer Type
CWE-805	V512, V594, V3106	Buffer Access with Incorrect Length Value
CWE-806	V512	Buffer Access Using Size of Source Buffer
CWE-483	V640, V3043	Incorrect Block Delimitation
CWE-134	V576, V618, V3025	Use of Externally-Controlled Format String
CWE-135	V518, V635	Incorrect Calculation of Multi-Byte String Length
CWE-462	V766, V3058	Duplicate Key in Associative List (Alist)
CWE-401	V701, V773	Improper Release of Memory Before Removing Last Reference ('Memory Leak')
CWE-468	V613, V620, V643	Incorrect Pointer Scaling
CWE-588	V641	Attempt to Access Child of a Non-structure Pointer
CWE-843	V641	Access of Resource Using Incompatible Type ('Type Confusion')
CWE-131	V512, V514, V531, V568, V620, V627, V635, V641, V645, V651, V687, V706, V727	Incorrect Calculation of Buffer Size
CWE-195	V569	Signed to Unsigned Conversion Error
CWE-197	V642	Numeric Truncation Error
CWE-762	V611, V780	Mismatched Memory Management Routines
CWE-478	V577, V719, V622, V3002	Missing Default Case in Switch Statement
CWE-415	V586	Double Free
CWE-188	V557, V3106	Reliance on Data/Memory Layout
CWE-562	V558	Return of Stack Variable Address
CWE-690	V522, V3080	Unchecked Return Value to NULL Pointer Dereference
CWE-457	V573, V614, V730, V670, V3070, V3128	Use of Uninitialized Variable
CWE-404	V611, V773	Improper Resource Shutdown or Release
CWE-563	V519, V603, V751, V763, V3061, V3065, V3077, V3117	Assignment to Variable without Use ('Unused Variable')
CWE-561	V551, V695, V734, V776, V779, V3021	Dead Code
CWE-570	V501, V547, V517, V560, V625, V654, V3022, V3063	Expression is Always False
CWE-571	V501, V547, V560, V617, V654, V694, V768, V3022, V3063	Expression is Always True
CWE-670	V696	Always-Incorrect Control Flow Implementation
CWE-674	V3110	Uncontrolled Recursion
CWE-681	V601	Incorrect Conversion between Numeric Types
CWE-688	V549	Function Call With Incorrect Variable or Reference as Argument
CWE-697	V556, V668	Insufficient Comparison

Table N1 - The first test variant of the correspondence between CWE and PVS-Studio diagnostics

The above is not the final variant of the table, but it gives some idea of how some of the PVS-Studio warnings are related to CWE. Now it is clear that PVS-Studio successfully detects (and has always detected) not only bugs in the code of the program, but also potential vulnerabilities, i.e. CWE. There were several articles written on this topic, they are listed in the end of this article.

CVE Bases

A potential vulnerability (CWE) is not yet an actual vulnerability (CVE). Real vulnerabilities, found both in open source, and in proprietary projects, are collected on the http://cve.mitre.org site. There you may find a description of a particular vulnerability, additional links (discussions, a bulletin of vulnerability fixes, links to the commits, remediate vulnerabilities and so on.) Optionally, the database can be downloaded in the necessary format. At the time of writing this article, .txt file of the base of vulnerabilities was about 100MB and more than 2.7 million of lines. Quite impressive, yes?

While doing some research for this article, I found quite an interesting resource that could be helpful to those who are interested - http://www.cvedetails.com/. It is convenient due to such features as:

• Search of CVE by the CWE identifier;
• Search of CVE in a certain product;
• Viewing statistics of appearance/fixes of vulnerabilities;
• Viewing various data tables, in one or another way related to CVE (for example, rating of companies, in whose products was the largest number of vulnerabilities found);
• And with more besides.

Some CVE that Could Have Been Found Using PVS-Studio

I am writing this article to demonstrate that the PVS-Studio analyzer can protect an application from vulnerabilities (at least, from some of them).

We have never investigated whether a certain defect, detected by PVS-Studio, can be exploited as a vulnerability. This is quite complicated and we have never set such a task. Therefore, I will do otherwise: I'll take several vulnerabilities that were have already detected and described, and show that they could have been avoided, if the code had been regularly checked by PVS-Studio.

Note. The vulnerabilities described in the article weren't found in synthetic examples, but in real source files, taken from old project revisions.

illumos-gate

The first vulnerability that we are going to talk about was detected in the source code of the illumos-gate project. illumos-gate is an open source project (available at the repository of GitHub), forming the core of an operating system, rooted in Unix in BSD.

The vulnerability has a name CVE-2014-9491.

Description of CVE-2014-9491: The devzvol_readdir function in illumos does not check the return value of a strchr call, which allows remote attackers to cause a denial of service (NULL pointer dereference and panic) via unspecified vectors.

The problem code was in the function devzvol_readdir:

static int devzvol_readdir(....)
{
  ....
  char *ptr;
  ....
  ptr = strchr(ptr + 1, '/') + 1;
  rw_exit(&sdvp->sdev_contents);
  sdev_iter_datasets(dvp, ZFS_IOC_DATASET_LIST_NEXT, ptr);
  ....
}

The function strchr returns a pointer to the first symbol occurrence, passed as a second argument. However, the function can return a null pointer in case the symbol wasn't found in the source string. But this fact was forgotten, or not taken into account. As a result, the return value is just added 1, the result is written to the ptr variable, and then the pointer is handled "as is". If the obtained pointer was null, then by adding 1 to it, we will get an invalid pointer, whose verification against NULL won't mean its validity. Under certain conditions this code can lead to a kernel panic.

PVS-Studio detects this vulnerability with the diagnostic rule V769, saying that the pointer returned by the strchr function can be null, and at the same time it gets damaged (due to adding 1):

V769 The 'strchr(ptr + 1, '/')' pointer in the 'strchr(ptr + 1, '/') + 1' expression could be nullptr. In such case, the resulting value will be senseless and it should not be used.

Network Audio System

Network Audio System (NAS) - network-transparent, client-server audio transport system, whose source code is available on SourceForge. NAS works on Unix and Microsoft Windows.

The vulnerability detected in this project has the code name CVE-2013-4258.

Description of CVE-2013-4258: Format string vulnerability in the osLogMsg function in server/os/aulog.c in Network Audio System (NAS) 1.9.3 allows remote attackers to cause a denial of service (crash) and possibly execute arbitrary code via format string specifiers in unspecified vectors, related to syslog.

The code was the following:

....
if (NasConfig.DoDaemon) {   /* daemons use syslog */
  openlog("nas", LOG_PID, LOG_DAEMON);
  syslog(LOG_DEBUG, buf);
  closelog();
} else {
  errfd = stderr;
....

In this fragment a syslog function is used incorrectly. Function declaration looks as follows:

void syslog(int priority, const char *format, ...);

The second parameter should be a format string, and all the others - data required for this string. Here the format string is missing, and a target message is passed directly as an argument (variable buf). This was the cause of the vulnerability which may lead to execution of arbitrary code.

If we believe the records in the SecurityFocus base, the vulnerability showed up in Debian and Gentoo.

What about PVS-Studio then? PVS-Studio detects this error with the diagnostic rule V618 and issues a warning:

V618 It's dangerous to call the 'syslog' function in such a manner, as the line being passed could contain format specification. The example of the safe code: printf("%s", str);

The mechanism of function annotation, built in the analyzer, helps to detect errors of this kind; the amount of annotated functions is more than 6500 for C and C++, and more than 900 for C#.

Here is the correct call of this function, remediating this vulnerability:

syslog(LOG_DEBUG, "%s", buf);

It uses a format string of "%s", which makes the call of the syslog function safe.

Ytnef (Yerase's TNEF Stream Reader)

Ytnef - an open source program available on GitHub. It is designed to decode the TNEF streams, created in Outlook, for example.

Over the last several months, there were quite a number of vulnerabilities detected that are described here. Lets' consider one of the CVE given in this list - CVE-2017-6298.

Description of CVE-2017-6298: An issue was discovered in ytnef before 1.9.1. This is related to a patch described as "1 of 9. Null Pointer Deref / calloc return value not checked."

All the fixed fragments which could contain null pointer dereference were approximately the same:

vl->data = calloc(vl->size, sizeof(WORD));
temp_word = SwapWord((BYTE*)d, sizeof(WORD));
memcpy(vl->data, &temp_word, vl->size);

In all these cases the vulnerabilities are caused by incorrect use of the calloc function. This function can return a null pointer in case the program failed to allocate the requested memory block. But the resulting pointer is not tested for NULL, and is used on account that calloc will always return a non-null pointer. This is slightly unreasonable.

How does PVS-Studio detect vulnerabilities? Quite easily: the analyzer has a lot of diagnostic rules, which detect the work with null pointers.

In particular, the vulnerabilities described above would be detected by V575 diagnostic. Here is what the warning looks like:

V575 The potential null pointer is passed into 'memcpy' function. Inspect the first argument.

The analyzer detected that a potentially null pointer, resulting from the call of the calloc function, is passed to the memcpy function without the verification against NULL.

That's how PVS-Studio detected this vulnerability. If the analyzer was used regularly while writing code, this problem could be avoided before it got to the version control system.

MySQL

MySQL is an open-source relational database management system. Usually MySQL is used as a server accessed by local or remote clients; however, the distribution kit includes a library of internal server, allowing the building of MySQL into standalone programs.

Let's consider one of the vulnerabilities, detected in this project - CVE-2012-2122.

The description of CVE-2012-2122: sql/password.c in Oracle MySQL 5.1.x before 5.1.63, 5.5.x before 5.5.24, and 5.6.x before 5.6.6, and MariaDB 5.1.x before 5.1.62, 5.2.x before 5.2.12, 5.3.x before 5.3.6, and 5.5.x before 5.5.23, when running in certain environments with certain implementations of the memcmp function, allows remote attackers to bypass authentication by repeatedly authenticating with the same incorrect password, which eventually causes a token comparison to succeed due to an improperly-checked return value.

Here is the code, having a vulnerability:

typedef char my_bool;
my_bool
check_scramble(const char *scramble_arg, const char *message,
               const uint8 *hash_stage2)
{
  ....
  return memcmp(hash_stage2, hash_stage2_reassured, SHA1_HASH_SIZE);
}

The type of the return value of the memcmp function is int, and the type of the return value of the check_scramble is my_bool, but actually - char. As a result, there is implicit conversion of int to char, during which the significant bits are lost. This resulted in the fact that in 1 out of 256 cases, it was possible to login with any password, knowing the user's name. In view of the fact that 300 attempts of connection took less than a second, this protection is as good as no protection. You may find more details about this vulnerability via the links listed on the following page: CVE-2012-2122.

PVS-Studio detects this issue with the help of the diagnostic rule V642. The warning is the following:

V642 Saving the 'memcmp' function result inside the 'char' type variable is inappropriate. The significant bits could be lost breaking the program's logic. password.c

As you can see, it was possible to detect this vulnerability using PVS-Studio.

iOS

iOS - a mobile operating system for smartphones, tablets and portable players, developed and manufactured by Apple.

Let's consider one of the vulnerabilities that was detected in this operating system; CVE-2014-1266. Fortunately, the code fragment where we may see what the issue is about, is publicly available.

Description of the CVE-2014-1266 vulnerability: The SSLVerifySignedServerKeyExchange function in libsecurity_ssl/lib/sslKeyExchange.c in the Secure Transport feature in the Data Security component in Apple iOS 6.x before 6.1.6 and 7.x before 7.0.6, Apple TV 6.x before 6.0.2, and Apple OS X 10.9.x before 10.9.2 does not check the signature in a TLS Server Key Exchange message, which allows man-in-the-middle attackers to spoof SSL servers by (1) using an arbitrary private key for the signing step or (2) omitting the signing step.

The code fragment causing the vulnerability was as follows:

static OSStatus
SSLVerifySignedServerKeyExchange(SSLContext *ctx, 
                                 bool isRsa, 
                                 SSLBuffer signedParams,
                                 uint8_t *signature, 
                                 UInt16 signatureLen)
{
  OSStatus err;
  ....

  if ((err = SSLHashSHA1.update(&hashCtx, &serverRandom)) != 0)
    goto fail;
  if ((err = SSLHashSHA1.update(&hashCtx, &signedParams)) != 0)
    goto fail;
    goto fail;
  if ((err = SSLHashSHA1.final(&hashCtx, &hashOut)) != 0)
    goto fail;
  ....

fail:
  SSLFreeBuffer(&signedHashes);
  SSLFreeBuffer(&hashCtx);
  return err;
}

The problem is in two goto operators, written close to each other. The first refers to the if statement, while the second - doesn't. Thus, regardless of the values of previous conditions, the control flow will jump to the "fail" label. Because of the second goto operator, the value err will be successful. This allowed man-in-the-middle attackers to spoof SSL servers.

PVS-Studio detects this issue using two diagnostic rules - V640 and V779. These are the warnings:

• V640 The code's operational logic does not correspond with its formatting. The statement is indented to the right, but it is always executed. It is possible that curly brackets are missing.
• V779 Unreachable code detected. It is possible that an error is present

Thus, the analyzer warns about several things that seemed suspicious to it.

• the logic of the program does not comply with the code formatting: judging by the alignment, we get the impression that both goto statements refer to the if statement, but it isn't so. The first goto is really in the condition, but the second - not.
• unreachable code: as the second goto runs without a condition, the code following it won't get executed.

It turns out that here PVS-Studio also coped with the work successfully.

Effective Use of Static Analysis

The aim of this article, as I mentioned earlier, is to show that the PVS-Studio analyzer successfully detects vulnerabilities. The approach chosen to achieve this objective is the demonstration that the analyzer finds some well-known vulnerabilities. The material was necessary to confirm the fact that it is possible to search for vulnerabilities using static analysis.

Now I would like to speak about the ways to do it more effectively. Ideally, vulnerabilities should be detected before they turn into vulnerabilities (i.e. when someone finds them and understands how they can be exploited); the earlier they are found, the better. By using static analysis in the proper way, the vulnerabilities can be detected at the coding stage. Below is the description of how this can be achieved.

Note. In this section I am going to use the word "error" for consistency. But, as we have already seen, simple bugs can be potential - and then real - vulnerabilities. Please do not forget this.

In general, the earlier the error is found and fixed, the lower the cost of fixing it. The figure provides data from the book by Capers Jones "Applied Software Measurement".

As you can see on the graphs, approximately 85% of errors are made at the coding stage, when the cost of the fix is minimal. As the error continues living in the code, the cost of its fix is constantly rising; if it costs only 25$ to fix the error at the coding stage, then after the release of the software, this figure increases up to tens of thousands dollars. Not to mention the cost of the vulnerabilities, found after the release.

It follows a simple conclusion - the sooner the error is detected and fixed, the better. The aim of static analysis is the earliest possible detection of errors in the code. Static analysis is not the replacement of other validation and verification tools, but a great addition to them.

How to get most of the benefit from a static analyzer? The first rule - the code must be checked regularly. Ideally, the error should be fixed at the coding stage, before it is committed to the version control system.

Nevertheless, it can be quite inconvenient to run continuous checks on the developer's machine. Besides that, the analysis of the code can be quite long, which won't let you recheck the code after the fixes. PVS-Studio has a special incremental analysis mode implemented, which allows analysis of only the code which was modified/edited since the last build. Moreover, this feature allows the running of the analysis automatically after the build, which means the developer doesn't have to think about manually starting it. After the analysis is completed, the programmer will be notified if there were errors detected in the modified files.

But even using the analyzer in such a way, there is a chance of an error getting into the version control system. That's why it's important to have a 'second level of protection' - to use a static analyzer on the build server. For example, to integrate the code analysis to the process of night builds. This will allow the checking of projects at night, and in the morning collecting information on the errors that got into the version control system. An important thing here is to immediately fix errors detected this way - preferably the next day. Otherwise, over time, nobody will pay attention to the new errors and there will be little use in such checks.

Implementation of static analysis into the development process may seem a non-trivial task, if the project is not being developed from scratch. The article, "What is a quick way to integrate static analysis in a big project?" gives a clear explanation of how to start using static analysis correctly.

Conclusion

I hope I was able to show that:

• even a seemingly simple bug may be a serious vulnerability;
• PVS-Studio successfully copes not only with the detection of errors in the code, but with CWE and CVE as well.

And if the cost of a simple bug increases over the time, the cost of a vulnerability can be enormous. At the same time, with the help of static analysis, a lot of vulnerabilities can be fixed even before they get into the version control system; not to mention before someone finds them and starts exploiting them.

Lastly, I would like to recommend trying PVS-Studio on your project - what if you find something that would save your project from getting to the CVE base?

Additional Links

• Handing out PVS-Studio Analyzer Licenses to Security Experts;
• PVS-Studio: searching software weaknesses.
• How to find 56 potential vulnerabilities in FreeBSD code in one evening.
• PVS-Studio Team Willing to Work on Improving Tizen Project (open letter).
• "What is a quick way to integrate static analysis in a big project?"