请问dirtytime_expire_seconds的意义

u4linux

请大牛解释一下sysctl参数vm.dirtytime_expire_seconds的意义，谢谢！
Document/sysctl/vm.txt的版本好像停留在2.6.29，之后加入的参数都找不到，在网上也没有查到相关资料。

Buddy_Zhang1

我在 vm.txt 只有找到 dirty_expire_centisecs,把您找到的 vm.dirtytime_expire_seconds 内容贴出来一下

u4linux

回复 2# Buddy_Zhang1


    我在整理vm的参数，使用sysctl命令看到的vm.dirtytime_expire_seconds，我只找到了引入这个参数的patch，但是这个patch里也没有对这个参数的工作原理解释太多。

nswcfd

http://lxr.oss.org.cn/source/fs/fs-writeback.c#L87

77 /*
78  * If an inode is constantly having its pages dirtied, but then the
79  * updates stop dirtytime_expire_interval seconds in the past, it's
80  * possible for the worst case time between when an inode has its
81  * timestamps updated and when they finally get written out to be two
82  * dirtytime_expire_intervals.  We set the default to 12 hours (in
83  * seconds), which means most of the time inodes will have their
84  * timestamps written to disk after 12 hours, but in the worst case a
85  * few inodes might not their timestamps updated for 24 hours.
86  */
87 unsigned int dirtytime_expire_interval = 12 * 60 * 60;

1249         {
1250                 .procname       = "dirtytime_expire_seconds",
1251                 .data           = &dirtytime_expire_interval,
1252                 .maxlen         = sizeof(dirty_expire_interval),
1253                 .mode           = 0644,
1254                 .proc_handler   = dirtytime_interval_handler,
1255                 .extra1         = &zero,
1256         },

1900 /*
1901  * Wake up bdi's periodically to make sure dirtytime inodes gets
1902  * written back periodically.  We deliberately do *not* check the
1903  * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1904  * kernel to be constantly waking up once there are any dirtytime
1905  * inodes on the system.  So instead we define a separate delayed work
1906  * function which gets called much more rarely.  (By default, only
1907  * once every 12 hours.)
1908  *
1909  * If there is any other write activity going on in the file system,
1910  * this function won't be necessary.  But if the only thing that has
1911  * happened on the file system is a dirtytime inode caused by an atime
1912  * update, we need this infrastructure below to make sure that inode
1913  * eventually gets pushed out to disk.
1914  */
1915 static void wakeup_dirtytime_writeback(struct work_struct *w);
1916 static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback);

u4linux

回复 4# nswcfd


    我也找到这段代码了，但是没看太明白，能给解释一下不？

nswcfd

就是一个定时器控制参数，控制什么时候执行wakeup_dirtytime_writeback

nswcfd

控制dirty页面什么时候写回的。