银河麒麟V10 SP2 auditd服务内存泄露问题
这几天遇到基于海光服务器的银河麒麟V10 SP2版本操作系统出现内存无故增长问题。
排查发现auditd服务,占用了大量内存。
我的环境是银河麒麟V10 SP2 524,audit版本audit-3.0-5.se.06
==5037== HEAP SUMMARY: ==5037== in use at exit: 3,022 bytes in 210 blocks ==5037== total heap usage: 415 allocs, 205 frees, 159,455 bytes allocated ==5037== ==5037== 122 bytes in 35 blocks are definitely lost in loss record 1 of 6 ==5037== at 0x483479B: malloc (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so) ==5037== by 0x488435C: xmalloc (in /usr/lib64/libaudit.so.1.0.0) ==5037== by 0x4884393: xmemdup (in /usr/lib64/libaudit.so.1.0.0) ==5037== by 0x48848FF: ??? (in /usr/lib64/libaudit.so.1.0.0) ==5037== by 0x4884B4C: get_file_sys_info (in /usr/lib64/libaudit.so.1.0.0) ==5037== by 0x11217B: ??? (in /usr/sbin/auditd) ==5037== by 0x112D83: ??? (in /usr/sbin/auditd) ==5037== by 0x10ECA7: ??? (in /usr/sbin/auditd) ==5037== by 0x4BD0B26: (below main) (libc-start.c:308) ==5037== ==5037== 122 bytes in 35 blocks are definitely lost in loss record 2 of 6 ==5037== at 0x483479B: malloc (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so) ==5037== by 0x488435C: xmalloc (in /usr/lib64/libaudit.so.1.0.0) ==5037== by 0x4884393: xmemdup (in /usr/lib64/libaudit.so.1.0.0) ==5037== by 0x488494E: ??? (in /usr/lib64/libaudit.so.1.0.0) ==5037== by 0x4884B4C: get_file_sys_info (in /usr/lib64/libaudit.so.1.0.0) ==5037== by 0x11217B: ??? (in /usr/sbin/auditd) ==5037== by 0x112D83: ??? (in /usr/sbin/auditd) ==5037== by 0x10ECA7: ??? (in /usr/sbin/auditd) ==5037== by 0x4BD0B26: (below main) (libc-start.c:308)
分析auditd服务存在内存泄露问题,位置在/usr/lib64/libaudit.so.1.0.0的get_file_sys_info处,malloc以后未释放。
auditd是审计服务,会对ssh的会话做日志记录,该问题出现在写日志前判断分区时的一部分代码。
经过循环ssh登录退出,也复现了这个现象。
yum update audit
升级到audit-3.0-5.se.07.ky10及以后版本即可解决
我也找到了官方发布的补丁:https://www.kylinos.cn/index.php/support/update/6.html
另外,关注我公众号,了解更多与云计算信创安全相关信息
另外还有一个问题,如果服务器安装的时候包含了图形界面,还会出现死机情况
由于mate-indicators进程也存在内存泄露问题
观察发现mate-indicators也会慢慢增大内存占用,并且会让服务器死机
影响版本
V10-SP1-0711及以后版本
V10-SP2-0524版本
解决方案
升级mate-indicators至mate-indicators-20150918kord0ukui58-10.p07.ky10及以后版本。
或将mate-indicators直接kill掉。