Recommended way to track down array out-of-bound access/write in C program

Question

Consider writing implementation for some not-so-obvious algorithm in C. For example let it be recursive quicksort, that I have found in K. N. King's "C Programming: A Modern Approach, 2nd Edition" book, that it's available from here. The most interesting part consist of two following definitions:

void quicksort(int a[], int low, int high)
{
    int middle;

    if (low >= high)
        return;

    middle = split(a, low, high);
    quicksort(a, low, middle - 1);
    quicksort(a, middle + 1, high);
}

int split(int a[], int low, int high)
{
    int part_element = a[low];

    for (;;) {
       while (low < high && part_element <= a[high])
           high--;
       if (low >= high)
           break;
       a[low++] = a[high];

       while (low < high && a[low] <= part_element)
           low++;
       if (low >= high)
           break;
       a[high--] = a[low];
    }

    a[high] = part_element;
    return high;
}

Both while loops can be optimized by removing low < high tests:

for (;;) {
    while (part_element < a[high])
        high--;
    if (low >= high)
        break;
    a[low++] = a[high];
    a[high] = part_element;

    while (a[low] <= part_element)
        low++;
    if (low >= high)
        break;
    a[high--] = a[low];
    a[low] = part_element;
}

What is the recommended way to make sure that every access or write to array (allocated on stack) is actually valid (i.e. not provoking undefined behaviour) ? What I already tried is to:

• manually debug with gdb on some actual data
• pass source code to static analysis tools like split or cppcheck
• valgrind with --tool=exp-sgcheck switch

For example having five elements array {8, 1, 2, 3, 4}:

#define N 5

int main(void)
{
    int a[N] = {8, 1, 2, 3, 4}, i;

    quicksort(a, 0, N - 1);

    printf("After sort:");
    for (i = 0; i < N; i++)
        printf(" %d", a[i]);
    putchar('\n');

    return 0;
}

Result is (most certainly it's implemention dependent):

After sort: 1 1 2 4 8

1. GDB

(gdb) p low
$1 = 3
(gdb) p high
$2 = 4
(gdb) p a[low]
$3 = 1
(gdb) p part_element
$4 = 8
(gdb) s
47              low++;
(gdb) s
46          while (a[low] <= part_element)
(gdb) s
47              low++;
(gdb) s
46          while (a[low] <= part_element)
(gdb) p low
$5 = 5
(gdb) p high
$6 = 4
(gdb) bt full
#0  split (a=0x7fffffffe140, low=5, high=4) at qsort.c:46
        part_element = 8
#1  0x00000000004005df in quicksort (a=0x7fffffffe140, low=0, high=4) at qsort.c:30
        middle = <value optimized out>
#2  0x0000000000400656 in main () at qsort.c:14
        a = {4, 1, 2, 1, 8}
        i = <value optimized out>

As you see low variable went outside boundary:

(gdb) p low
$5 = 5

2. Static analysis tools

$ splint -retvalint -exportlocal qsort.c 
Splint 3.1.2 --- 07 Feb 2011

Finished checking --- no warnings

$ cppcheck qsort.c 
Checking qsort.c...

3. Valgrind with --tool=exp-sgcheck

$ valgrind --tool=exp-sgcheck ./a.out 
==5480== exp-sgcheck, a stack and global array overrun detector
==5480== NOTE: This is an Experimental-Class Valgrind Tool
==5480== Copyright (C) 2003-2012, and GNU GPL'd, by OpenWorks Ltd et al.
==5480== Using Valgrind-3.8.1 and LibVEX; rerun with -h for copyright info
==5480== Command: ./a.out
==5480== 
==5480== Invalid read of size 4
==5480==    at 0x4005A0: split (qsort.c:46)
==5480==    by 0x4005DE: quicksort (qsort.c:30)
==5480==    by 0x400655: main (qsort.c:14)
==5480==  Address 0x7ff000114 expected vs actual:
==5480==  Expected: stack array "a" of size 20 in frame 2 back from here
==5480==  Actual:   unknown
==5480==  Actual:   is 0 after Expected
==5480== 
After sort: 1 1 2 4 8
==5480== 
==5480== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)

The location at 0x4005A0: split (qsort.c:46) is matching to the same place as I found by gdb manually.

Answer 1

What is the recommended way to make sure that every access or write to array (allocated on stack) is actually valid (i.e. not provoking undefined behaviour) ?

What if use clang on Linux with the options -fsanitize=addressand -fsanitize=undefined? It is also available in gcc: http://gcc.gnu.org/gcc-4.8/changes.html.

---

clang with the option -fsanitize=undefined

This is an example:

#include <stdlib.h>

#define N 5

int main(int argc, char *argv[])
{
  int a[N] = {8, 1, 2, 3, 4}, i;

  int r =0;
  int end = atoi(argv[1]);
  for (int i = 0; i != end; ++i)
    r += a[i];

  return r;
}

Then

clang -fno-omit-frame-pointer -fsanitize=undefined -g out_boundary.c -o out_boundary_clang

$ ./out_boundary_clang 5
$ ./out_boundary_clang 6
out_boundary.c:12:10: runtime error: index 5 out of bounds for type 'int [5]'
Illegal instruction (core dumped)

And then analyze a core file

Program terminated with signal 4, Illegal instruction.
#0  main (argc=2, argv=0x7fff3a1c28c8) at out_boundary.c:12
12          r += a[i];
(gdb) p i
$1 = 5

---

clang with the option -fsanitize=address

This is a quote:

The tool can detect the following types of bugs:

* Out-of-bounds accesses to heap, stack and globals
* Use-after-free
* Use-after-return (to some extent)
* Double-free, invalid free
* Memory leaks (experimental)

clang -fno-omit-frame-pointer -fsanitize=address -g out_boundary.c -o out_boundary_clang

And then:

$ ./out_boundary_clang 6 2>&1 | asan_symbolize.py
=================================================================
==9634==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7fff91bb2ad4 at pc 0x459c67 bp 0x7fff91bb2910 sp 0x7fff91bb2908
READ of size 4 at 0x7fff91bb2ad4 thread T0
    #0 0x459c66 in main out_boundary.c:12
    #1 0x3a1d81ed1c in __libc_start_main ??:0
    #2 0x4594ac in _start ??:0
Address 0x7fff91bb2ad4 is located in stack of thread T0 at offset 244 in frame
    #0 0x45957f in main out_boundary.c:6
  This frame has 8 object(s):
    [32, 36) ''
    [96, 100) ''
    [160, 168) ''
    [224, 244) 'a'
    [288, 292) 'i'
    [352, 356) 'r'
    [416, 420) 'end'
    [480, 484) 'i1'
HINT: this may be a false positive if your program uses some custom stack unwind mechanism or swapcontext
      (longjmp and C++ exceptions *are* supported)
Shadow bytes around the buggy address:
  0x10007236e500: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x10007236e510: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x10007236e520: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x10007236e530: 00 00 00 00 00 00 00 00 00 00 00 00 f1 f1 f1 f1
  0x10007236e540: 04 f4 f4 f4 f2 f2 f2 f2 04 f4 f4 f4 f2 f2 f2 f2
=>0x10007236e550: 00 f4 f4 f4 f2 f2 f2 f2 00 00[04]f4 f2 f2 f2 f2
  0x10007236e560: 04 f4 f4 f4 f2 f2 f2 f2 04 f4 f4 f4 f2 f2 f2 f2
  0x10007236e570: 04 f4 f4 f4 f2 f2 f2 f2 04 f4 f4 f4 f3 f3 f3 f3
  0x10007236e580: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x10007236e590: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x10007236e5a0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Shadow byte legend (one shadow byte represents 8 application bytes):
  Addressable:           00
  Partially addressable: 01 02 03 04 05 06 07
  Heap left redzone:     fa
  Heap right redzone:    fb
  Freed heap region:     fd
  Stack left redzone:    f1
  Stack mid redzone:     f2
  Stack right redzone:   f3
  Stack partial redzone: f4
  Stack after return:    f5
  Stack use after scope: f8
  Global redzone:        f9
  Global init order:     f6
  Poisoned by user:      f7
  ASan internal:         fe
==9634==ABORTING

Or you can use both this options. Useful links:

• http://clang.llvm.org/docs/UsersManual.html#controlling-code-generation
• http://developer.mozilla.org/en-US/docs/Mozilla/Testing/Firefox_and_Address_Sanitizer
• http://clang.llvm.org/docs/AddressSanitizer.html