你的位置:首页 > 操作系统

[操作系统]使用valgrind检查内存


工欲善其事,必先利其器。Valgrind作为一个免费且优秀的工具包,平时大部分人可能都是使用valgrind检测内存问题,如内存泄露,越界等。其实Valgrind的用途远不止于此,其实际上为一个工具包,除了检查内存问题以外,还有其它多项用途。我准备将其大致介绍一下。本不想再介绍Valgrind检测内存问题的用法的,但是又一想,毕竟这是Valgrind的一个最有名的用途,如果少了它,不免有些遗憾,所以还是把检查内存问题作为第一篇吧。
 
请看一下代码:
  1. #include <stdlib.h>
  2. #include <stdio.h>
  3. #include <string.h>
  4. static void mem_leak1(void)
  5. {
  6.     char *p = malloc(1);
  7. }
  8. static void mem_leak2(void)
  9. {
  10.     FILE *fp = fopen("test.txt", "w");
  11. }
  12. static void mem_overrun1(void)
  13. {
  14.     char *p = malloc(1);
  15.     *(short*)p = 2;
  16.     free(p);
  17. }
  18. static void mem_overrun2(void)
  19. {
  20.     char array[5];
  21.     strcpy(array, "hello");
  22. }
  23. static void mem_double_free(void)
  24. {
  25.     char *p = malloc(1);
  26.     free(p);
  27.     free(p);
  28. }
  29. static void mem_use_wild_pointer(void)
  30. {
  31.     char *p = (void*)0x80184800;
  32.     *p = 1;
  33. }
  34. static void mem_free_wild_pointer(void)
  35. {
  36.     char *p;
  37.     free(p);
  38. }
  39. int main()
  40. {
  41.     mem_leak1();
  42.     mem_leak2();
  43.     mem_overrun1();
  44.     mem_overrun2();
  45.     mem_double_free();
  46.     //mem_use_wild_pointer();
  47.     mem_free_wild_pointer();
  48.     return 0;
  49. }


这里一共列出了七种常见的内存问题:
1. 动态内存泄露;
2. 资源泄露,这里以文件描述符为例;
3. 动态内存越界;
4.数组内存越界;
5.动态内存double free;
6.使用野指针,即未初始化的指针;
7.释放野指针,即未初始化的指针;
 
  其中由于本示例代码过于简单,第6中情况,使用野指针会直接导致crash,所以在main中,并没有真正的调用那个示例代码。由于valgrind只能检测执行到的代码,所以在后面的报告中,不会报告第6种错误情况。但是,在大型的项目中,有可能使用野指针并不会导致程序crash。另外上面的7中情况,有些情况严格的说,实际上可以归为一类。
 
下面看怎样执行valgrind来检测内存错误:
  1. valgrind --track-fds=yes --leak-check=full --undef-value-errors=yes ./a.out


上面那些option的具体含义,可以参加valgrind --help,其中有些option默认就是打开的,不过我习惯于明确的使用option,以示清晰。
 
看执行后的报告:
  1. ==2326== Memcheck, a memory error detector
  2. ==2326== Copyright (C) 2002-2009, and GNU GPL'd, by Julian Seward et al.
  3. ==2326== Using Valgrind-3.5.0 and LibVEX; rerun with -h for copyright info
  4. ==2326== Command: ./a.out
  5. ==2326==
  6. /* 这里检测到了动态内存的越界,提示Invalid write。*/
  7. ==2326== Invalid write of size 2
  8. ==2326== at 0x80484B4: mem_overrun1 (in /home/fgao/works/test/a.out)
  9. ==2326== by 0x8048553: main (in /home/fgao/works/test/a.out)
  10. ==2326== Address 0x40211f0 is 0 bytes inside a block of size 1 alloc'd
  11. ==2326== at 0x4005BDC: malloc (vg_replace_malloc.c:195)
  12. ==2326== by 0x80484AD: mem_overrun1 (in /home/fgao/works/test/a.out)
  13. ==2326== by 0x8048553: main (in /home/fgao/works/test/a.out)
  14. ==2326==
 /* 这里检测到了double free问题,提示Invalid Free */
  1. ==2326== Invalid free() / delete / delete[]
  2. ==2326== at 0x40057F6: free (vg_replace_malloc.c:325)
  3. ==2326== by 0x8048514: mem_double_free (in /home/fgao/works/test/a.out)
  4. ==2326== by 0x804855D: main (in /home/fgao/works/test/a.out)
  5. ==2326== Address 0x4021228 is 0 bytes inside a block of size 1 free'd
  6. ==2326== at 0x40057F6: free (vg_replace_malloc.c:325)
  7. ==2326== by 0x8048509: mem_double_free (in /home/fgao/works/test/a.out)
  8. ==2326== by 0x804855D: main (in /home/fgao/works/test/a.out)
  9. ==2326==
  10. /* 这里检测到了未初始化变量 */
  11. ==2326== Conditional jump or move depends on uninitialised value(s)
  12. ==2326== at 0x40057B6: free (vg_replace_malloc.c:325)
  13. ==2326== by 0x804853C: mem_free_wild_pointer (in /home/fgao/works/test/a.out)
  14. ==2326== by 0x8048562: main (in /home/fgao/works/test/a.out)
  15. ==2326==
 /* 这里检测到了非法是否野指针 */
  1. ==2326== Invalid free() / delete / delete[]
  2. ==2326== at 0x40057F6: free (vg_replace_malloc.c:325)
  3. ==2326== by 0x804853C: mem_free_wild_pointer (in /home/fgao/works/test/a.out)
  4. ==2326== by 0x8048562: main (in /home/fgao/works/test/a.out)
  5. ==2326== Address 0x4021228 is 0 bytes inside a block of size 1 free'd
  6. ==2326== at 0x40057F6: free (vg_replace_malloc.c:325)
  7. ==2326== by 0x8048509: mem_double_free (in /home/fgao/works/test/a.out)
  8. ==2326== by 0x804855D: main (in /home/fgao/works/test/a.out)
  9. ==2326==
  10. ==2326==
  11. /* 
  12. 这里检测到了文件指针资源的泄露,下面提示说有4个文件描述符在退出时仍是打开的。
  13. 描述符0,1,2无需关心,通过报告,可以发现程序中自己明确打开的文件描述符没有关闭。
  14. */
  15. ==2326== FILE DESCRIPTORS: 4 open at exit.
  16. ==2326== Open file descriptor 3: test.txt
  17. ==2326== at 0x68D613: __open_nocancel (in /lib/libc-2.12.so)
  18. ==2326== by 0x61F8EC: __fopen_internal (in /lib/libc-2.12.so)
  19. ==2326== by 0x61F94B: fopen@@GLIBC_2.1 (in /lib/libc-2.12.so)
  20. ==2326== by 0x8048496: mem_leak2 (in /home/fgao/works/test/a.out)
  21. ==2326== by 0x804854E: main (in /home/fgao/works/test/a.out)
  22. ==2326==
  23. ==2326== Open file descriptor 2: /dev/pts/4
  24. ==2326==
  25. ==2326==
  26. ==2326== Open file descriptor 1: /dev/pts/4
  27. ==2326==
  28. ==2326==
  29. ==2326== Open file descriptor 0: /dev/pts/4
  30. ==2326==
  31. ==2326==
  32. ==2326==
  33. /* 堆信息的总结:一共调用4次alloc,4次free。之所以正好相等,因为上面有一函数少了free,有一个函数多了一个free */
  34. ==2326== HEAP SUMMARY:
  35. ==2326== in use at exit: 353 bytes in 2 blocks
  36. ==2326== total heap usage: 4 allocs, 4 frees, 355 bytes allocated
  37. ==2326==
  38. /* 检测到一个字节的内存泄露 */
  39. ==2326== 1 bytes in 1 blocks are definitely lost in loss record 1 of 2
  40. ==2326== at 0x4005BDC: malloc (vg_replace_malloc.c:195)
  41. ==2326== by 0x8048475: mem_leak1 (in /home/fgao/works/test/a.out)
  42. ==2326== by 0x8048549: main (in /home/fgao/works/test/a.out)
  43. ==2326==
  44. /* 内存泄露的总结 */
  45. ==2326== LEAK SUMMARY:
  46. ==2326== definitely lost: 1 bytes in 1 blocks
  47. ==2326== indirectly lost: 0 bytes in 0 blocks
  48. ==2326== possibly lost: 0 bytes in 0 blocks
  49. ==2326== still reachable: 352 bytes in 1 blocks
  50. ==2326== suppressed: 0 bytes in 0 blocks
  51. ==2326== Reachable blocks (those to which a pointer was found) are not shown.
  52. ==2326== To see them, rerun with: --leak-check=full --show-reachable=yes
  53. ==2326==
  54. ==2326== For counts of detected and suppressed errors, rerun with: -v
  55. ==2326== Use --track-origins=yes to see where uninitialised values come from
  56. ==2326== ERROR SUMMARY: 5 errors from 5 contexts (suppressed: 12 from 8)


 
这个只是一个简单的示例程序,即使没有Valgrind,我们也可以很轻易的发现问题。但是在真实的项目中,当代码量达到万行,十万行,甚至百万行时。由于申请的内存可能不是在一个地方使用,不可避免的被传来传去。这时,如果光是看review代码来检查问题,可能很难找到根本原因。这时,使用Valgrind则可以很容易的发现问题所在。
 
当然,Valgrind也不是万能的。我也遇到过Valgrind无法找到问题,反而我通过不断的review代码找到了症结。发现问题,解决问题,毕竟是末流。最好的方法,就是不引入内存问题。这可以通过良好的代码风格和设计来实现的。