1. 概述
Valgrind 是一款用于内存调试、内存泄漏检测以及性能分析的软件开发工具。
Valgrind 发行版目前包括七个生产质量工具:一个内存错误检测器、两个线程错误检测器、一个缓存和分支预测分析器、一个调用图生成缓存和分支预测分析器,以及两个不同的堆分析器。它还包括一个实验性的 SimPoint 基本块矢量生成器。它可在以下平台上运行:X86/Linux、AMD64/Linux、ARM/Linux、ARM64/Linux、PPC32/Linux、PPC64/Linux、PPC64LE/Linux、S390X/Linux、MIPS32/Linux、MIPS64/Linux、X86/Solaris , AMD64/Solaris, ARM/Android (2.3.x 及更高版本), ARM64/Android, X86/Android (4.0 及更高版本), MIPS32/Android, X86/FreeBSD, AMD64/FreeBSD, X86/Darwin 和 AMD64/Darwin (Mac OS X 10.12)
1.1 工具
它一般包含下列工具:
1.Memcheck
最常用的工具,用来检测程序中出现的内存问题,所有对内存的读写都会被检测到,一切对 malloc()/free()/new/delete 的调用都会被捕获。所以,它能检测以下问题:
- 对未初始化内存的使用;
- 读 / 写释放后的内存块;
- 读 / 写超出 malloc 分配的内存块;
- 读 / 写不适当的栈中内存块;
- 内存泄漏,指向一块内存的指针永远丢失;
- 不正确的 malloc/free 或 new/delete 匹配;
- memcpy() 相关函数中的 dst 和 src 指针重叠。
2.Callgrind
和 gprof 类似的分析工具,但它对程序的运行观察更是入微,能给我们提供更多的信息。和 gprof 不同,它不需要在编译源代码时附加特殊选项,但加上调试选项是推荐的。Callgrind 收集程序运行时的一些数据,建立函数调用关系图,还可以有选择地进行 cache 模拟。在运行结束时,它会把分析数据写入一个文件。callgrind_annotate 可以把这个文件的内容转化成可读的形式。
3.Cachegrind
Cache 分析器,它模拟 CPU 中的一级缓存 I1,Dl 和二级缓存,能够精确地指出程序中 cache 的丢失和命中。如果需要,它还能够为我们提供 cache 丢失次数,内存引用次数,以及每行代码,每个函数,每个模块,整个程序产生的指令数。这对优化程序有很大的帮助。
4.Helgrind
它主要用来检查多线程程序中出现的竞争问题。Helgrind 寻找内存中被多个线程访问,而又没有一贯加锁的区域,这些区域往往是线程之间失去同步的地方,而且会导致难以发掘的错误。Helgrind 实现了名为 “Eraser” 的竞争检测算法,并做了进一步改进,减少了报告错误的次数。不过,Helgrind 仍然处于实验阶段。
5.Massif
堆栈分析器,它能测量程序在堆栈中使用了多少内存,告诉我们堆块,堆管理块和栈的大小。Massif 能帮助我们减少内存的使用,在带有虚拟内存的现代系统中,它还能够加速我们程序的运行,减少程序停留在交换区中的几率。 此外,lackey 和 nulgrind 也会提供。Lackey 是小型工具,很少用到;Nulgrind 只是为开发者展示如何创建一个工具。
1.2 原理
Memcheck 能够检测出内存问题,关键在于其建立了两个全局表。
Valid-Value 表
对于进程的整个地址空间中的每一个字节 (byte),都有与之对应的 8 个 bits;对于 CPU 的每个寄存器,也有一个与之对应的 bit 向量。这些 bits 负责记录该字节或者寄存器值是否具有有效的、已初始化的值。
Valid-Address 表
对于进程整个地址空间中的每一个字节 (byte),还有与之对应的 1 个 bit,负责记录该地址是否能够被读写。
检测原理
当要读写内存中某个字节时,首先检查这个字节对应的 A bit。如果该 A bit 显示该位置是无效位置,memcheck 则报告读写错误。 内核(core)类似于一个虚拟的 CPU 环境,这样当内存中的某个字节被加载到真实的 CPU 中时,该字节对应的 V bit 也被加载到虚拟的 CPU 环境中。一旦寄存器中的值,被用来产生内存地址,或者该值能够影响程序输出,则 memcheck 会检查对应的 V bits,如果该值尚未初始化,则会报告使用未初始化内存错误。
2. 安装使用
2.1 下载
centos 下载:
yum install valgrind
Ubuntu 下载
sudo apt-get install valgrind
2.2 使用
用法: valgrind[options] prog-and-args [options]:
常用选项,适用于所有 Valgrind 工具
-tool= 最常用的选项。运行 valgrind 中名为 toolname 的工具。默认 memcheck。
h –help 显示帮助信息。
-version 显示 valgrind 内核的版本,每个工具都有各自的版本。
q –quiet 安静地运行,只打印错误信息。
v –verbose 更详细的信息, 增加错误数统计。
-trace-children=no|yes 跟踪子线程? [no]
-track-fds=no|yes 跟踪打开的文件描述?[no]
-time-stamp=no|yes 增加时间戳到 LOG 信息? [no]
-log-fd= 输出 LOG 到描述符文件 [2=stderr]
-log-file= 将输出的信息写入到 filename.PID 的文件里,PID 是运行程序的进行 ID
-log-file-exactly= 输出 LOG 信息到 file
-log-file-qualifier= 取得环境变量的值来做为输出信息的文件名。 [none]
-log-socket=ipaddr:port 输出 LOG 到 socket ,ipaddr:port
LOG 信息输出:
-xml=yes 将信息以 xml 格式输出,只有 memcheck 可用
-num-callers= show callers in stack traces [12]
-error-limit=no|yes 如果太多错误,则停止显示新错误? [yes]
-error-exitcode= 如果发现错误则返回错误代码 [0=disable]
-db-attach=no|yes 当出现错误,valgrind 会自动启动调试器 gdb。[no]
-db-command= 启动调试器的命令行选项[gdb -nw %f %p]
适用于 Memcheck 工具的相关选项:
-leak-check=no|summary|full 要求对 leak 给出详细信息? [summary]
-leak-resolution=low|med|high how much bt merging in leak check [low]
-show-reachable=no|yes show reachable blocks in leak check? [no]
3. 应用例子
3.1 数组越界
malloc1.c
#include <stdio.h>
int main(int argc, char **argv) {
int *x = malloc(8 * sizeof(int));
x[9] = 0; // 数组下标越界
free(x);
return 0;
}
编译:
gcc -Wall malloc1.c -g -o malloc1
使用 Valgrind 检查程序 BUG:
# --leak-check=full 所有泄露检查
valgrind --tool=memcheck --leak-check=full ./malloc1
运行结果:
[root@hackett valgrind]# valgrind --tool=memcheck --leak-check=full ./malloc1
==550168== Memcheck, a memory error detector
==550168== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==550168== Using Valgrind-3.17.0 and LibVEX; rerun with -h for copyright info
==550168== Command: ./malloc1
==550168==
==550168== Invalid write of size 4
==550168== at 0x4005FB: main (malloc1.c:7)
==550168== Address 0x520b064 is 4 bytes after a block of size 32 alloc'd
==550168== at 0x4C360A5: malloc (vg_replace_malloc.c:380)
==550168== by 0x4005EE: main (malloc1.c:5)
==550168==
==550168==
==550168== HEAP SUMMARY:
==550168== in use at exit: 0 bytes in 0 blocks
==550168== total heap usage: 1 allocs, 1 frees, 32 bytes allocated
==550168==
==550168== All heap blocks were freed -- no leaks are possible
==550168==
==550168== For lists of detected and suppressed errors, rerun with: -s
==550168== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)
分析:
1、每一行开头的数字代表是进程 ID,这里的进程 ID 为 550168
2、提示
==550168== Invalid write of size 4
==550168== at 0x4005FB: main (malloc1.c:7)
这两行显示的是错误出现的位置,就是 x 分配了 32 byte 的空间,但是向第 36 个 byte 写数据, 所以就会显示 Invalid write 的错误,代码在 malloc1.c 的第 7 行
3.2 内存释放后进行读写
malloc2.c
#include <stdio.h>
int main(int argc, char **argv) {
char *p = malloc(1);
*p = 'a';
char c = *p;
printf("[%c]\n", c);
free(p); // 释放
c = *p; //取值 读
return 0;
}
编译:
gcc -Wall malloc2.c -g -o malloc2
使用 Valgrind 检查程序 BUG:
# --leak-check=full 所有泄露检查
valgrind --tool=memcheck --leak-check=full ./malloc2
运行结果:
[root@hackett valgrind]# valgrind --tool=memcheck --leak-check=full ./malloc2
==550063== Memcheck, a memory error detector
==550063== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==550063== Using Valgrind-3.17.0 and LibVEX; rerun with -h for copyright info
==550063== Command: ./malloc2
==550063==
[a]
==550063== Invalid read of size 1
==550063== at 0x400679: main (malloc2.c:15)
==550063== Address 0x520b040 is 0 bytes inside a block of size 1 free'd
==550063== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550063== by 0x400674: main (malloc2.c:13)
==550063== Block was alloc'd at
==550063== at 0x4C360A5: malloc (vg_replace_malloc.c:380)
==550063== by 0x40063E: main (malloc2.c:5)
==550063==
==550063==
==550063== HEAP SUMMARY:
==550063== in use at exit: 0 bytes in 0 blocks
==550063== total heap usage: 2 allocs, 2 frees, 1,025 bytes allocated
==550063==
==550063== All heap blocks were freed -- no leaks are possible
==550063==
==550063== For lists of detected and suppressed errors, rerun with: -s
==550063== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)
复制代码
分析:
1、每一行开头的数字代表是进程 ID,这里的进程 ID 为 550063
2、提示
==550063== Invalid read of size 1
==550063== at 0x400679: main (malloc2.c:15)
==550063== Address 0x520b040 is 0 bytes inside a block of size 1 free’d
==550063== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550063== by 0x400674: main (malloc2.c:13)
这五行显示的是错误出现的位置,就是 p 指向的内存已经释放了,还对其进行读操作,所以就会显示 Invalid read 的错误,代码在 malloc2.c 的第 15 行
3.3 无效读写
malloc3.c
#include <stdio.h>
int main(int argc, char **argv) {
char *p = malloc(1);
*p = 'a';
char c = *(p+1); // 地址加1 无效读
printf("[%c]\n",c);
free(p); // 释放
return 0;
}
编译:
gcc -Wall malloc3.c -g -o malloc3
使用 Valgrind 检查程序 BUG:
# --leak-check=full 所有泄露检查
valgrind --tool=memcheck --leak-check=full ./malloc3
运行结果:
[root@iZwz97bu0gr8vx0j8l6kkzZ valgrind]# valgrind --tool=memcheck --leak-check=full ./malloc3
==550135== Memcheck, a memory error detector
==550135== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==550135== Using Valgrind-3.17.0 and LibVEX; rerun with -h for copyright info
==550135== Command: ./malloc3
==550135==
==550135== Invalid read of size 1
==550135== at 0x40064E: main (malloc3.c:9)
==550135== Address 0x520b041 is 0 bytes after a block of size 1 alloc'd
==550135== at 0x4C360A5: malloc (vg_replace_malloc.c:380)
==550135== by 0x40063E: main (malloc3.c:5)
==550135==
[]
==550135==
==550135== HEAP SUMMARY:
==550135== in use at exit: 0 bytes in 0 blocks
==550135== total heap usage: 2 allocs, 2 frees, 1,025 bytes allocated
==550135==
==550135== All heap blocks were freed -- no leaks are possible
==550135==
==550135== For lists of detected and suppressed errors, rerun with: -s
==550135== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)
分析:
1、每一行开头的数字代表是进程 ID,这里的进程 ID 为 550135
2、提示
==550135== Invalid read of size 1
==550135== at 0x40064E: main (malloc3.c:9)
这五行显示的是错误出现的位置,就是对未分配内存的空间进行读取,所以就会显示 Invalid read 的错误,代码在 malloc3.c 的第 9 行
3.4 内存泄漏
malloc4.c
#include <stdio.h>
int main(int argc, char **argv) {
char *p = malloc(1);
*p = 'a';
char c = *p;
printf("[%c]\n",c); // 申请后没有释放p 内存泄漏
return 0;
}
编译:
gcc -Wall malloc4.c -g -o malloc4
使用 Valgrind 检查程序 BUG:
# --leak-check=full 所有泄露检查
valgrind --tool=memcheck --leak-check=full ./malloc4
运行结果:
[root@hackett valgrind]# valgrind --tool=memcheck --leak-check=full ./malloc4
==550195== Memcheck, a memory error detector
==550195== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==550195== Using Valgrind-3.17.0 and LibVEX; rerun with -h for copyright info
==550195== Command: ./malloc4
==550195==
[a]
==550195==
==550195== HEAP SUMMARY:
==550195== in use at exit: 1 bytes in 1 blocks
==550195== total heap usage: 2 allocs, 1 frees, 1,025 bytes allocated
==550195==
==550195== 1 bytes in 1 blocks are definitely lost in loss record 1 of 1
==550195== at 0x4C360A5: malloc (vg_replace_malloc.c:380)
==550195== by 0x4005EE: main (malloc4.c:5)
==550195==
==550195== LEAK SUMMARY:
==550195== definitely lost: 1 bytes in 1 blocks
==550195== indirectly lost: 0 bytes in 0 blocks
==550195== possibly lost: 0 bytes in 0 blocks
==550195== still reachable: 0 bytes in 0 blocks
==550195== suppressed: 0 bytes in 0 blocks
==550195==
==550195== For lists of detected and suppressed errors, rerun with: -s
==550195== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)
分析:
1、每一行开头的数字代表是进程 ID,这里的进程 ID 为 550195
2、提示
==550195== 1 bytes in 1 blocks are definitely lost in loss record 1 of 1
==550195== at 0x4C360A5: malloc (vg_replace_malloc.c:380)
==550195== by 0x4005EE: main (malloc4.c:5)
这三行显示的是错误出现的位置,是 stack trace 内存泄漏,代码在 malloc1.c 的第 5 行,申请了内存没有进行对应的 free
3.5 内存多次释放
malloc5.c
#include <stdio.h>
int main(int argc, char **argv) {
char *p = malloc(1);
*p = 'a';
char c = *p; // 地址加1
printf("[%c]\n",c);
free(p);
free(p);// 内存多次释放
free(p);// 内存多次释放
return 0;
}
编译:
gcc -Wall malloc5.c -g -o malloc5
使用 Valgrind 检查程序 BUG:
# --leak-check=full 所有泄露检查
valgrind --tool=memcheck --leak-check=full ./malloc5
运行结果:
[root@hackett valgrind]# valgrind --tool=memcheck --leak-check=full ./malloc5
==550227== Memcheck, a memory error detector
==550227== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==550227== Using Valgrind-3.17.0 and LibVEX; rerun with -h for copyright info
==550227== Command: ./malloc5
==550227==
[a]
==550227== Invalid free() / delete / delete[] / realloc()
==550227== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550227== by 0x400680: main (malloc5.c:14)
==550227== Address 0x520b040 is 0 bytes inside a block of size 1 free'd
==550227== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550227== by 0x400674: main (malloc5.c:13)
==550227== Block was alloc'd at
==550227== at 0x4C360A5: malloc (vg_replace_malloc.c:380)
==550227== by 0x40063E: main (malloc5.c:5)
==550227==
==550227== Invalid free() / delete / delete[] / realloc()
==550227== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550227== by 0x40068C: main (malloc5.c:15)
==550227== Address 0x520b040 is 0 bytes inside a block of size 1 free'd
==550227== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550227== by 0x400674: main (malloc5.c:13)
==550227== Block was alloc'd at
==550227== at 0x4C360A5: malloc (vg_replace_malloc.c:380)
==550227== by 0x40063E: main (malloc5.c:5)
==550227==
==550227==
==550227== HEAP SUMMARY:
==550227== in use at exit: 0 bytes in 0 blocks
==550227== total heap usage: 2 allocs, 4 frees, 1,025 bytes allocated
==550227==
==550227== All heap blocks were freed -- no leaks are possible
==550227==
==550227== For lists of detected and suppressed errors, rerun with: -s
==550227== ERROR SUMMARY: 2 errors from 2 contexts (suppressed: 0 from 0)
分析:
1、每一行开头的数字代表是进程 ID,这里的进程 ID 为 550227
2、
==550227== Invalid free() / delete / delete[] / realloc()
==550227== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550227== by 0x400680: main (malloc5.c:14)
这三行显示的是第一个错误出现的位置,就是多次释放内存问题, 所以就会显示 Invalid free() / delete / delete[] / realloc() 的错误,代码在 malloc5.c 的第 14 行
3、
==550227== Invalid free() / delete / delete[] / realloc()
==550227== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550227== by 0x40068C: main (malloc5.c:15)
这三行显示的是第二个错误出现的位置,就是多次释放内存问题, 所以就会显示 Invalid free() / delete / delete[] / realloc() 的错误,代码在 malloc5.c 的第 15 行
3.6 动态内存管理
常见的内存分配方式分三种:静态存储,栈上分配,堆上分配。全局变量属于静态存储,它们是在编译时就被分配了存储空间,函数内的局部变量属于栈上分配,而最灵活的内存使用方式当属堆上分配,也叫做内存动态分配了。常用的内存动态分配函数包括:malloc, alloc, realloc, new 等,动态释放函数包括 free, delete。
一旦成功申请了动态内存,我们就需要自己对其进行内存管理,而这又是最容易犯错误的。
malloc6.c
#include <stdio.h>
// 内存动态管理
int main(int argc, char **argv) {
int i;
char *p = (char *)malloc(10);
char *pt = p;
for(i = 0; i < 10; i++) {
p[i] = 'A'+i;
}
free(p);
pt[1] = 'x';
free(pt);
return 0;
}
编译:
gcc -Wall malloc6.c -g -o malloc6
使用 Valgrind 检查程序 BUG:
# --leak-check=full 所有泄露检查
valgrind --tool=memcheck --leak-check=full ./malloc6
运行结果:
[root@hackett valgrind]# valgrind --tool=memcheck --leak-check=full ./malloc6
==550239== Memcheck, a memory error detector
==550239== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==550239== Using Valgrind-3.17.0 and LibVEX; rerun with -h for copyright info
==550239== Command: ./malloc6
==550239==
==550239== Invalid write of size 1
==550239== at 0x400639: main (malloc6.c:16)
==550239== Address 0x520b041 is 1 bytes inside a block of size 10 free'd
==550239== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550239== by 0x400630: main (malloc6.c:14)
==550239== Block was alloc'd at
==550239== at 0x4C360A5: malloc (vg_replace_malloc.c:380)
==550239== by 0x4005EE: main (malloc6.c:7)
==550239==
==550239== Invalid free() / delete / delete[] / realloc()
==550239== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550239== by 0x400647: main (malloc6.c:18)
==550239== Address 0x520b040 is 0 bytes inside a block of size 10 free'd
==550239== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550239== by 0x400630: main (malloc6.c:14)
==550239== Block was alloc'd at
==550239== at 0x4C360A5: malloc (vg_replace_malloc.c:380)
==550239== by 0x4005EE: main (malloc6.c:7)
==550239==
==550239==
==550239== HEAP SUMMARY:
==550239== in use at exit: 0 bytes in 0 blocks
==550239== total heap usage: 1 allocs, 2 frees, 10 bytes allocated
==550239==
==550239== All heap blocks were freed -- no leaks are possible
==550239==
==550239== For lists of detected and suppressed errors, rerun with: -s
==550239== ERROR SUMMARY: 2 errors from 2 contexts (suppressed: 0 from 0)
分析:
1、每一行开头的数字代表是进程 ID,这里的进程 ID 为 550239
2、
==550239== Invalid write of size 1
==550239== at 0x400639: main (malloc6.c:16)
这一行显示这里有一个错误,就是发生非法写操作, 所以就会显示 Invalid write 的错误,代码在 malloc6.c 的第 16 行
3、
==550239== Invalid free() / delete / delete[] / realloc()
==550239== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550239== by 0x400647: main (malloc6.c:18)
==550239== Address 0x520b040 is 0 bytes inside a block of size 10 free’d
==550239== at 0x4C38A03: free (vg_replace_malloc.c:755)
==550239== by 0x400630: main (malloc6.c:14)
这五行显示的是第二个错误出现的位置,指针 p 和 pt 指向的是同一块内存,却被先后释放两次。系统会在堆上维护一个动态内存链表,如果被释放,就意味着该块内存可以继续被分配给其他部分,如果内存被释放后再访问,就可能覆盖其他部分的信息,这是一种严重的错误,上述程序第 16 行中就在释放后仍然写这块内存。
总结:
- 申请内存在使用完成后就要释放。如果没有释放,或少释放了就是内存泄露;多释放也会产生问题。
- 注意数组的大小,别越界读写访问非法内存
- malloc 和 new 要对应 free 和 delete 使用。