一 前言
对于C或C++步调员来说，面对的bug很大部分是内存利用题目，这此中比力令人头疼的就是内存走漏了，虽然我们有valgrind 和AScan等内存题目的检测工具，但是valgrind每次输出一大堆，AScan偶然候看输出结果看的是云里雾里的。再说，谁会嫌弃工具箱内里多个工具那。
二 内存走漏的一样平常查抄
2.1 根本预备
内存走漏题目的查抄步调，对于做过c或c++同砚都比力熟悉:
- 起首通过top或vmstat 、或smem（本次介绍）等工具检察内存环境，看看是否出现了内存走漏。
- 其次用pidstat 或top指定历程的方式，观察可以历程内存占用环境。
- 用memleak或gdb工具检察内存走漏。
先上测试代码：
#include <stdio.h>#include <stdlib.h>#include <pthread.h>#include <unistd.h>#define MALLOC_SIZE 256000int *fibo(int *n0, int *n1){ int *v = (int *) malloc(MALLOC_SIZE*sizeof(int)); memset(v, 0, MALLOC_SIZE*sizeof(int)); *v = *n0 + *n1; return v;}void do_test(){ int n0 = 0; int n1 = 1; int *v = NULL; int n = 2; for (n = 2; n > 0; n++) { v = fibo(&n0, &n1); n0 = n1; n1 = *v; printf("%dth => %lld\n", n, *v); //free(v) sleep(1); }}int main(void){ printf("pid=%d\n", getpid()); do_test(); return 0;}步调比力简单，编译运行起来：
gcc memtest.c ; ./a.out2.2 smem工具
这次用下新工具smem，这是一个python写的小工具，可以统计体系中全部历程占用的物理内存RSS、以及去掉共享内存的PSS、以及步调本身的独占内存USS的环境。
安装：
# centos 下yum install epel-releaseyum install smem python-matplotlib python-tk# ubuntu 下apt-get install smem常用下令：
-k 带单位表现内存
root@ubuntu-lab:/home/miao# smem -k PID User Command Swap USS PSS RSS 1009 root /usr/sbin/cron -f -P 0 304.0K 399.0K 2.9M 1137 root nginx: master process /usr/ 0 196.0K 435.0K 2.1M 931 root /usr/sbin/irqbalance --fore 0 492.0K 655.0K 4.0M ....-u -k 带单位表现每个用户的内存占用：
root@ubuntu-lab:/home/miao# smem -u -kUser Count Swap USS PSS RSS systemd-timesync 1 0 764.0K 1.1M 6.7M messagebus 1 0 924.0K 1.2M 4.9M systemd-network 1 0 1.7M 2.1M 7.4M syslog 1 0 3.0M 3.1M 6.2M www-data 4 0 2.0M 4.2M 22.4M systemd-resolve 1 0 4.8M 5.8M 12.7M miao 8 0 11.0M 16.9M 49.1M postgres 7 0 9.2M 22.0M 74.5M mysql 1 0 74.0M 74.7M 80.7M root 30 0 260.7M 284.1M 429.5M -w -k 表现体系团体内存环境雷同free
root@ubuntu-lab:/home/miao# smem -w -kArea Used Cache Noncache firmware/hardware 0 0 0 kernel image 0 0 0 kernel dynamic memory 1.5G 1.3G 268.5M userspace memory 414.0M 191.5M 222.5M free memory 2.8G 2.8G 0 -k -s uss -r 按照uss的占用从大到小排序的方式展示内存的占用环境 非常实用
root@ubuntu-lab:/home/miao# smem -k -s uss -r PID User Command Swap USS PSS RSS 1298 root /usr/bin/dockerd -H 0 74.3M 74.5M 77.9M 1068 mysql /usr/sbin/mariadbd 0 74.0M 74.8M 80.7M 939 root /usr/lib/snapd/snapd 0 44.9M 45.0M 46.7M ....好了根本下令介绍完毕，那我们来看看怎样检察内存是否走漏吧，由于内存走漏的步调占用的内存是不停再增长的（这不是废话嘛），如许我们就可以用上面的排序下令只观察上面几个历程了。
watch smem -k -s uss -r小本事，watch加在下令前面，5s实行一次下令，会高亮表现改变的部分。
2.3 memleak查抄
在ubuntu下安装memleak竟然很难安装，我用的是最新的服务器版本，反面在centos下安装后测试的：
[root@xxx]# python2 /usr/share/bcc/tools/memleak -p 160399Attaching to pid 160399, Ctrl+C to quit.[17:27:25] Top 10 stacks with outstanding allocations: 5120000 bytes in 5 allocations from stack fibo+0x1a [a.out] do_test+0x41 [a.out] main+0x24 [a.out] __libc_start_main+0xf5 [libc-2.17.so][17:27:30] Top 10 stacks with outstanding allocations: 10240000 bytes in 10 allocations from stack fibo+0x1a [a.out] do_test+0x41 [a.out] main+0x24 [a.out] __libc_start_main+0xf5 [libc-2.17.so][17:27:35] Top 10 stacks with outstanding allocations: 15360000 bytes in 15 allocations from stack fibo+0x1a [a.out] do_test+0x41 [a.out] main+0x24 [a.out] __libc_start_main+0xf5 [libc-2.17.so][17:27:40] Top 10 stacks with outstanding allocations: 19456000 bytes in 19 allocations from stackfibo 函数出现内存走漏，把走漏的字节数都打印了出来，我们改了下代码把free的解释去掉，再用memleak检察等了一会照旧没有走漏信息，分析已经修复了，如下：
[root@xxx]# python2 /usr/share/bcc/tools/memleak -p 165349Attaching to pid 165349, Ctrl+C to quit.[17:35:21] Top 10 stacks with outstanding allocations:[17:35:26] Top 10 stacks with outstanding allocations:[17:35:31] Top 10 stacks with outstanding allocations:[17:35:36] Top 10 stacks with outstanding allocations:三 gdb 检察内存走漏
大概你对memleak已经很熟悉了，那来看看gdb检察函数的内存走漏方法吧，这个方法只是检察具体的一个函数是否存在内存走漏，肯定的场景下照旧蛮实用的。
把代码中的for (n = 2; n > 0; n++) 改成for (n = 2; n > 0&& n <10; n++)
(gdb) b mainBreakpoint 1 at 0x400739: file memleaktest.c, line 34.(gdb) rStarting program: /home/miaohq/testcode/./a.out Breakpoint 1, main () at memleaktest.c:3434 printf("pid=%d\n", getpid());Missing separate debuginfos, use: debuginfo-install glibc-2.17-325.el7_9.x86_64(gdb) call malloc_stats()Arena 0:system bytes = 0in use bytes = 0Total (incl. mmap):system bytes = 0in use bytes = 0max mmap regions = 0max mmap bytes = 0$1 = -136490560(gdb) npid=18197735 do_test();(gdb) call malloc_stats()Arena 0:system bytes = 0in use bytes = 0Total (incl. mmap):system bytes = 0in use bytes = 0max mmap regions = 0max mmap bytes = 0$2 = -136490560(gdb) n2th => 13th => 24th => 35th => 56th => 87th => 138th => 219th => 3436 return 0;(gdb) call malloc_stats()Arena 0:system bytes = 0in use bytes = 0Total (incl. mmap):system bytes = 8224768in use bytes = 8224768max mmap regions = 8max mmap bytes = 8224768$3 = -136490560(gdb) p 256000*4*8$4 = 8192000(gdb) Total (incl. mmap):即本步调占用的总内存，看到显着的增长部分即为未开释的内存，步调使用的内存增长：8224768 稍大于256000*4*8 分配的内存，内存分配须要存储链表另有一些对齐缘故起因以是会多分配些。
free之后的场景：
(gdb) call malloc_stats()Arena 0:system bytes = 0in use bytes = 0Total (incl. mmap):system bytes = 0in use bytes = 0max mmap regions = 0max mmap bytes = 0$1 = -136490560(gdb) npid=18340635 do_test();(gdb) n2th => 13th => 24th => 35th => 56th => 87th => 138th => 219th => 3436 return 0;(gdb) call malloc_stats()Arena 0:system bytes = 1159168in use bytes = 0Total (incl. mmap):system bytes = 1159168in use bytes = 0max mmap regions = 1max mmap bytes = 1028096$2 = -136490560(gdb) in use bytes 为0了。 |
