Lines of Code and related lines-oriented-statistics with Perl

6 02 2011

A project I’m involved right now is making a Static code analyzer. The main goal is to produce a RoR front-end webapp with the capability to submit code ad analyze it statically. Our main goal languages are C/C++, but we will attach other tools to do some work for other languages. For Static I mean without the need to run the program. One thing we will never be able to answer with Static analysis is the behavior of the program, but we can answer some things that with dynamic analysis we can not, so there is not such a think like one is better or more complete than the other.

The RoR front-end is almost done and quick contribution I gave was a Perl script that receives a folder as input and analyzes all the source code available inside the folder (recursive). You can find the README.markdown under the same folder.
This analysis is just oriented to quantity of lines of code.

How to use

Well, if you don’t want to read the README.markdown file and experiment the script I will show you the output and the command you need to produce this images.

So, by default you only need to say the input folder and the output prefix name for the images, like so we an say:

[ulissesaraujocosta@maclisses:trab1]-$ perl count.pl -open ../../../Static-Code-Analyzer/ -out work

This will produce the following 3 images (click on the images to enlarge):



By default the script always produce this 3 images: number of files per language, number of lines per language and the ratio between them (the average of lines per file, per language).

We can also see this values in percentage, related to overall project (folder).
Image number 3 will be the same, because does not make sense ratio percentage.

[ulissesaraujocosta@maclisses:trab1]-$ perl count.pl -open ../../../Static-Code-Analyzer/ -out work_percent -percent



And my favorite, produce an image with a overall picture of the percentage use of each language on the project.

[ulissesaraujocosta@maclisses:trab1]-$ perl count.pl -open ../../../Static-Code-Analyzer/ -out work_All -all

With this script you can also generate pie charts, by default it uses bars charts, please read the README.markdown for more information.

How to improve and support more languages?

If you want to improve the script feel free to fork on github and maybe we can discuss more about the script.
To support more languages you just have to add one more entrance in the hashtable and write:

extension => {"nrFiles" => 0, "nrLines" => 0, "comments" => function_to_catch_comments, "nrComments" => 0, "percentageNrFiles" => 0, "percentageNrLines" => 0, "percentageNrComments" => 0}

This is an example of the entrace for C++:

"cpp"  => {"nrFiles" => 0, "nrLines" => 0, "comments" => sub { return shift =~ m/(\*(.|\n|\r)*?\*)|(^[ \t\n]*\/\/.*)/; },    "nrComments" => 0,
                         "percentageNrFiles" => 0, "percentageNrLines" => 0, "percentageNrComments" => 0
                        },

This is a simple presentation I gave about the module I used: GD::Graph





Splint the static C code checker

3 05 2009

Splint is this great tool to statically analyze C code. Splint is an evolution of Lint. Lint makes analysis of C code and verifies unused declarations, type inconsistencies, use before definitions, unreachable code, ignored return values, execution path with no return and infinite loops. However, Lint was not a sufficiently powerful tool. Splint was created as a descendant of Lint allowing annotations in C code, so, it do more checks than its predecessor. In this post I will present some verifications that Splint does, and show the philosophy and how to use it. Splint allows annotations on functions, variables, parameters and types.

Undefined Variables

Splint detects instances where the value of a variable is used before it is defined. Anyway, many static analyzers can also do this, it is a very basic check. However, thanks to the splint annotations supports, they can be used to describe what storage must be defined and what storage must be undefined at interface points, this means that all storage reachable from a global variable, parameter to a function, or function return value is defined before and after the function call.

A special case of undefined variables is undefined function parameters. Sometimes we do a certain C function that return values, or change parameters values. The out annotation denotes a pointer to storage that may be undefined. Splint have a great storage model, so does not report an error when a pointer to allocated but undefined storage is passed as an out parameter. If in the body of a function an outparameter is allocated but not bounded to a value Splint reports an error. You can see out as an parameter that will be defined inside a function. The opposite happens to the in annotation, this can be used to denote a parameter that must be completely defined.
Here is a small example of it:

extern void setVal (/*@out@*/ int *x);
extern int getVal (/*@in@*/ int *x);
extern int mysteryVal(int *x);

int dumbfunc(/*@out@*/ int *x, int i) {
    if (i > 3)
11         return *x;
    else if (i > 1)
13         return getVal (x);
    else if (i == 0)
15         return mysteryVal (x);
    else {
18         setVal (x);
19         return *x;
    }
}

And here it is the output of Splint:

> splint usedef.c
usedef.c:11: Value *x used before definition
usedef.c:13: Passed storage x not completely defined
    (*x is undefined): getVal (x)
usedef.c:15: Passed storage x not completely defined
    (*x is undefined): mysteryVal (x)
Finished checking --- 3 code warnings

As you can see, no error is reported for line 18, because the incomplete defined storage x is passed to a function that will define it, so we can return *x with no problems.

Types

Strong type checking often reveals programming errors. Splint can check primitive C types more strictly and flexibly than typical compilers.

C, does not make types verification. Because for C all basic non numerical types are just int’s (Enum, Pointer, Char). Splint supports stricter checking of built-in C types. The char and enum types can be checked as distinct types, and the different numeric types can be type-checked strictly. Also the primitive char type can be type-checked as a distinct type. If char is used as a distinct type, common errors involving assigning ints to chars are detected.

In Splint, you can always turn off some default verifications, and Splint will assume the C compiler strategy to verify the code. If you run Splint with +charint is on, char types are indistinguishable from ints. So you can make cast’s from int to char or from cahr to int. But never forgot that this can lead to errors, as you can imagine if you do (char) 1000 you will not get an expected result because the characters only go up to 256.

Splint reports errors in the use of Enums, if a value that is not an enumerator member is assigned to the enum type or if an enum type is used as an operand to an arithmetic operator. Anyway you can turn this off, activating the enumint flag and then you can use enums and ints types interchangeably.

Memory Management

About half the bugs in typical C programs can be attributed to memory management problems.

Any C programmer who has already made a reasonable size program, has been confronted with unexpected errors related with memory management problems. This happens basically because C is a low level language so, does not have any system like Java’s garbage collector. All the memory management have to be done by the programmer, so errors may append. Some only appear sporadically, and some may only be apparent when compiled on a different platform. Splint is able to detect many memory management errors at compile time, like: using storage that may have been deallocated, memory leaks or returning a pointer to stack-allocated storage. This is possible because Splint makes a memory model when is working on the code, anyway I will not talk about this subject.

Splint can see and report deallocating storage when there are other live references to the same storage and failing to deallocate storage before the last reference to it is lost. The solution here is simple, Splint makes an obligation to release storage and attach this obligation to the reference to which the storage is assigned.

For references not shared Splint uses the only annotation to indicate that a reference is the only pointer to the object it points to, here we also use the null annotation to say that the output of malloc could be NULL:

/* @only@ */ /* @null@ */ void * malloc ( size_t size );

Here is a simple example of C annotated code with memory leaks and use after free variables:

1 extern /*@only@*/ int *glob;
/*@only@*/ int *f (/*@only@*/ int *x, int *y, int *z) /*@globals glob;@*/ {
8    int *m = (int *)
9    malloc (sizeof (int));
11   glob = y;   //Memory leak
12   free (x);
13   *m = *x;   //Use after free
14    return z;   //Memory leak detected
}

Here is the Splint output for the file above:

> splint only.c
only.c:11: Only storage glob (type int *) not released before assignment: glob = y
only.c:1: Storage glob becomes only
only.c:11: Implicitly temp storage y assigned to only: glob = y
only.c:13: Dereference of possibly null pointer m: *m
only.c:8: Storage m may become null
only.c:13: Variable x used after being released
only.c:12: Storage x released
only.c:14: Implicitly temp storage z returned as only: z
only.c:14: Fresh storage m not released before return
only.c:9: Fresh storage m allocated

Splint errors and warnings are very human readable, so you just have to read them to understand.
The first warning says that variable glob was not released (we say, in line 1, that it is only) before the assignment. In line 13 we have a dereference of possibly null pointer, because in line 12 we have done free to x variable, an now we want to use it’s value. As you can see, is very easy to understand the output of Splint.

Splint can also make verifications in stack based references. A memory error occurs if a pointer into stack is live after the function returns. Splint detects errors involving stack references exported from a function through return values or assignments to references reachable from global variables or actual parameters.
No annotations are needed to detect stack reference errors. It is clear from declarations if storage is allocated on the function stack.

Here is an example of stack-allocated storage:

int *glob;

int *f (int **x) {
    int sa[2] = { 0, 1 };
    int loc = 3;
9   glob = &loc;
10  *x = &sa[0];
12  return &loc;
}

And here is the Splint output:

> splint stack.c
stack.c:12: Stack-allocated storage &loc reachable from return value: &loc
stack.c:12: Stack-allocated storage *x reachable from parameter x
stack.c:10: Storage *x becomes stack
stack.c:12: Stack-allocated storage glob reachable from global glob
stack.c:9: Storage glob becomes stack

Control Flow

In order to avoid certain errors, Splint have to understand the control flow of the program, so Splint do some checks related to control flow. Many of these checks are possible because of the extra information that is known in annotations. Without this additional information Splint assumes that all functions return and execution continues normally.

noreturn annotation is used to denote a function that never returns.

extern /* @noreturn@ */ void fatalerror ( char *s);

We also have maynoreturn and alwaysreturns annotations, but Splint must assume that a function returns normally when checking the code and doesn’t verify if a function really returns.

To describe non-returning functions the noreturnwhentrue and noreturnwhenfalse mean that a function never returns if the first argument is true or false.

/* @noreturnwhenfalse@ */ void assert (/* @sef@ */ bool /* @alt int@ */ pred );

The sef annotation denotes a parameter as side effect free, and the alt int indicate that it may be either a Boolean or an integer.

Undefined Behavior

The order which side effect take place in C is not entirely defined by the code.

A sequence point is some point in the C code where it is guaranteed that all side effects of previous evaluations have been performed.
An example of sequence points is:

  • a function call (after the arguments have been evaluated)
  • at the end of a if, while, for or do statement
  • a &&, || and ?

Here is a simple C program that have undefined behavior.

extern int glob;
extern int mystery (void);
extern int modglob (void) /*@globals glob@*/ /*@modifies glob@*/;

int f (int x, int y[]) {
11    int i = x++ * x;
13    y[i] = i++;
14    i += modglob() * glob;
15    i += mystery() * glob;
16    return i;
}

And here is the Splint output:

> splint order.c
order.c:11: Expression has undefined behavior (value of right operand modified by left operand): x++ * x
order.c:13: Expression has undefined behavior (left operand uses i, modified by right operand): y[i] = i++
order.c:14: Expression has undefined behavior (value of right operand modified by left operand): modglob() * glob
order.c:15: Expression has undefined behavior (unconstrained function mystery used in left operand may set global variable glob used in right operand): mystery() * glob

Regard control flow, Splint has more options to check C code. I will not talk about all of them here.

Buffer Sizes

Buffer overflow errors are a particularly dangerous type of bug in C, they are responsible for half of all security attacks. This is happens because C does not perform runtime bound checking (for performance reasons), and so attackers can exploit program bugs to, for example, gain full access to a machine.

Splint models blocks of memory using two properties:
maxSet, maxSet(b) denotes the highest address beyond b that can be safely used as lvalue, for instance:

char buffer[MAXSIZE] we have maxSet(buffer) = MAXSIZE - 1

and maxRead, maxRead(b) denotes the highest index of a buffer that can be safely used as rvalue.

When a buffer is accessed as an lvalue, Splint generates a precondition constraint involving the maxSet property and when a buffer is accessed as an rvalue, Splint generates a precondition constraint involving the maxRead property.

We can use this two properties in buffer sizes annotations, this annotations uses the clause requires and ensures just like JML, VDM and Frama-C. When a function with requires clause is called, the call site must be checked to satisfy the constraints implied by requires.

Here is an example of its use:

void /* @alt char * @*/ strcpy
(/* @out@ */ /* @returned@ */ char *s1 , char *s2)
/* @modifies * s1@ */
/* @requires maxSet (s1) >= maxRead (s2) @*/
/* @ensures maxRead (s1) == maxRead (s2) @*/;

This is a possible annotation for strcpy standard library C function. We say that s1 is the return value of the function (returned), and that the pointer to s1 is the only one (only). We also say that this function modifies s1 and we specify the pre and post conditions.

Splint can also do bound checking. This is more complex than other checks done by Splint, so, memory bound warnings contain extensive information about the unresolved constraint as you can see in this example:

int buf [10];
buf [10] = 3;
setChar .c :5:4: Likely out -of - bounds store : buf [10]
Unable to resolve constraint : requires 9 >= 10
needed to satisfy precondition : requires maxSet ( buf @ setChar .c :5:4) >= 10

The ultimate test: wu-ftpd

wu-ftpd version 2.5.0 was about 20.000 lines of code and took less than four seconds to check all of wu-ftpd on a 1.2-GHz Athlon machine. Splint detected the known flaws as well as finding some previously unknown flaws (!)

One of the problems of these automatic static analysis tools is that it can produce false problems.
Running Splint on wu-ftpd without adding annotations produced 166 warnings for potential out-of-bounds writes. After adding 66 annotations, it produced 101 warnings: 25 of these indicated real problems and 76 were false.

Pros and Cons

Pros:

  • Lightweight static analysis detects software vulnerabilities
  • Splint definitely improves code quality
  • Suitable for real programs…

Cons:

  • …although it produces more warning messages that lead to confusion
  • It won’t eliminate all security risks
  • Hasn’t been developed since 2007, they need new volunteers

Conclusions

No tool will eliminate all security risks, lightweight static analysis tools (Splint) play an important role in identifying security vulnerabilities.

Here is the presentation about that subject:

Outroduction

All the examples provide here is from the Splint manual.








Follow

Get every new post delivered to your Inbox.

Join 185 other followers