PATCH] Perhaps invalid goal for file block #1 in ext3(2)_find_goal

stone_wang_at_sohu.com
Date: 11/30/04

  • Next message: stanojr_at_blackhole.websupport.sk: "Re: quota deadlock" 
    Date:	Tue, 30 Nov 2004 20:40:23 +0800 (CST)
To: <linux-kernel@vger.kernel.org>

    Hi:

    I am learning how Ext2/3 works and how we could better use it for our servers.

    In the process,i found something strange in ext3_find_goal:
    when we create a file,write initial 4k bytes to it,then we reboot the system,and after that,we again append another 4k bytes to the file,thus the file becomes 8k bytes now;but under this condition,the file will be in 2 chunks,not continous.

    I did simple tests(i reboot the system to make sure that the inode cache get outdated between the 2 write/append operations,and i am sure that under some VM pressure result will be the same--ext3_read_inode clear ei->i_next_alloc_block to zero):

    #dd if=/dev/zero of=/root/4kfile bs=1024 count=4
    #cp /root/4kfile tmpfile
    # ll tmpfile
    -rw-r--r-- 1 root root 4096 Dec 17 13:05 tmpfile
    # filefrag tmpfile
    tmpfile: 1 extent found
    ------THEN I REBOOT THE LINUX SYSTEM----
    #cat /root/4kfile >>tmpfile
    # ll tmpfile
    -rw-r--r-- 1 root root 8192 Dec 17 13:14 tmpfile
    # filefrag tmpfile
    tmpfile: 2 extents found, perfection would be 1 extent

    debugfs shows that there are free blocks just next to the block#0 of tmpfile.

    Generally,continous creating/appending(without rebooting the system or VM pressure)a file get 1 extent.

    Here are my simple patch for kernel 2.6.9,it works for me.

    ---2.6.9/fs/ext3/inode.c 2004-12-17
    +++2.6.9-ext3-find-goal-fixed/fs/ext3/inode.c 2004-12-17
    @@ -525,7 +525,7 @@
     {
            struct ext3_inode_info *ei = EXT3_I(inode);
            /* Writer: ->i_next_alloc* */
    - if (block == ei->i_next_alloc_block + 1) {
    + if ((block == ei->i_next_alloc_block + 1)&&(ei->i_next_alloc_goal!=0)) {
                    ei->i_next_alloc_block++;
                    ei->i_next_alloc_goal++;
            }

    Regards.

    Peter Wang
    2004.11.29
    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

  • Next message: stanojr_at_blackhole.websupport.sk: "Re: quota deadlock"