lkml.org 
[lkml]   [2012]   [Jun]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: Punching hole using fallocate is not removing the uninit extent from extent tree
From
Add in mail loop.

2012/6/18, Namjae Jeon <linkinjeon@gmail.com>:
> 2012/6/18, Lukáš Czerner <lczerner@redhat.com>:
>> On Mon, 28 May 2012, Ashish Sangwan wrote:
>>
>>> Date: Mon, 28 May 2012 19:38:05 +0530
>>> From: Ashish Sangwan <ashishsangwan2@gmail.com>
>>> To: linux-ext4@vger.kernel.org
>>> Subject: Punching hole using fallocate is not removing the uninit extent
>>> from
>>> extent tree
>>
>> Hi Ashish,
>>
>> I am looking at you patch, however I am not able to reproduce this.
>> Can you please send more information (script preferably) on how to
>> reproduce this problem ?
>>
> Hi. Lukas.
> If you use the below script, you can easily reproduce this problem.
> And I can not attach script file, so I write script code in this mail.
> you can paste it to file.
>
> fragmentation.sh
> -----------------------------------------------------------------------
> dd if=/dev/zero of=12kb bs=4096 count=6
> index=0
> while [ $? == 0 ]
> do
> index=$(($index + 1))
> cp 12kb $1/file1.$index
> done
> echo "Partition filled"
> sync
> df -h
> index=0
> while [ $? == 0 ]
> do
> index=$(($index + 2))
> sync
> rm $1/file1.$index
> done
> sync
> echo "fragmented partition $1 with 4KB files"
> -------------------------------------------------------------------------
>
> 1. create smaill partition(500MB) to reproduce rapidly.
> 2. plz run this script to make fragmentation partition after making
> dummy directory like this.
> mkdir test_dir
> ./fragmentation.sh test_dir
>
> 3. dd if=/dev/zero of=d_file bs=1024 count=10240
> 4. you can see file depth is not zero.(dump_extents d_file
> 5. and try to fallocate offset : 4096 lengh 4096.
>
>> Also what kernel version did try this on ?
> We are using 3.0.20 kernel patched your punch hole patches.
>
> Please let us know in case of any queries.
>
> Thanks.
>
>>
>> Thanks!
>> -Lukas
>>
>>>
>>> I have created a formatted EXT4 partition such that every single
>>> extent is exactly 6blocks (24KB) of length.
>>> I used hole punch on 2 different files.
>>>
>>> CASE 1: In first situation, file size is 72KB. There are total 3
>>> extents each 24KB length. Using fallocate to punch hole starting at
>>> offset 4096 and length 4096,
>>> dump_extents gives the following expected output :
>>>
>>> Before punching hole :
>>> Level Entries Logical Physical Length Flags
>>> 0/ 0 1/ 2 0 - 5 1856 - 1861 6
>>> 0/ 0 2/ 2 6 - 11 1868 - 1873 6
>>>
>>> After punching hole :
>>> Level Entries Logical Physical Length Flags
>>> 0/ 0 1/ 3 0 - 0 1856 - 1856 1
>>> 0/ 0 2/ 3 2 - 5 1858 - 1861 4
>>> 0/ 0 3/ 3 6 - 11 1868 - 1873 6
>>>
>>> The 1st extent: 0-5, is splitted into 3 extents, "0-0", "1-1", "2-5"
>>> Extent 1-1 is first marked as uninitialized in function
>>> ext4_ext_map_blocks() and later removed from the extent tree by
>>> ext4_ext_remove_space().
>>>
>>> CASE 2: File size is 9.4MB. There are total 400 extents each 24KB
>>> length, depth of extent tree at root header is 1 and there are 2 index
>>> entries.
>>>
>>> dump_extents output before punching hole:
>>> Level Entries Logical Physical Length Flags
>>> 0/ 1 1/ 2 0 - 2039 1922 2040
>>> 1/ 1 1/340 0 - 5 1856 - 1861 6
>>> 1/ 1 2/340 6 - 11 1868 - 1873 6
>>> < Continued likewise till 340/340 >
>>> 1/ 1 340/340 2034 - 2039 5942 - 5947 6
>>> 0/ 1 2/ 2 2040 - 2399 1923 360
>>> 1/ 1 1/ 60 2040 - 2045 5954 - 5959 6
>>> 1/ 1 2/ 60 2046 - 2051 5966 - 5971 6
>>> < Continued likewise till 60/60 >
>>> 1/ 1 60/ 60 2394 - 2399 6662 - 6667 6
>>>
>>> dump_extents output after punching hole :
>>> 0/ 1 1/ 3 0 - 5 1922 6
>>> 1/ 1 1/ 3 0 - 0 1856 - 1856 1
>>> 1/ 1 2/ 3 1 - 1 1857 - 1857 1 Uninit
>>> 1/ 1 3/ 3 2 - 5 1858 - 1861 4
>>> 0/ 1 2/ 3 6 - 2039 6674 2034
>>> 1/ 1 1/339 6 - 11 1868 - 1873 6
>>> 1/ 1 2/339 12 - 17 1880 - 1885 6
>>> < Continued like wise...>
>>>
>>> Comparing CASE2 with CASE1, still uninit extent "1-1" is present
>>> within the extent tree.
>>>
>>> In function ext4_ext_remove_space(), there is call to function
>>> ext4_ext_rm_leaf which is responsible for removal of this extent.
>>> But this function is not getting called in CASE 2 :
>>> if (i == depth) {
>>> /* this is leaf block */
>>> err = ext4_ext_rm_leaf(handle, inode, path,
>>> start, end);
>>> /* root level has p_bh == NULL, brelse() eats this */
>>> brelse(path[i].p_bh);
>>> path[i].p_bh = NULL;
>>> i--;
>>> continue;
>>> }
>>>
>>> Varibale "i" does not become equals to "depth" because
>>> ext4_ext_more_to_rm is returning "0" hence the following if condition
>>> is turning out to be false for 1st extent index:
>>> if (ext4_ext_more_to_rm(path + i)) {
>>>
>>> Looking at the defination of ext4_ext_more_to_rm :
>>> /*
>>> * ext4_ext_more_to_rm:
>>> * returns 1 if current index has to be freed (even partial)
>>> */
>>> static int
>>> ext4_ext_more_to_rm(struct ext4_ext_path *path)
>>> {
>>> BUG_ON(path->p_idx == NULL);
>>> if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
>>> return 0;
>>>
>>> /*
>>> * if truncate on deeper level happened, it wasn't partial,
>>> * so we have to consider current index for truncation
>>> */
>>> if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) <=
>>> This condition is turning out to be true
>>> return 0; <= The function is returning zero from
>>> here.
>>> return 1;
>>> }
>>>
>>> I could not understand the significance of the above mentioned if
>>> condition check, if anyone could explain a little, it will be help.
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>>>
>
--
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/

\
 
 \ /
  Last update: 2012-06-18 20:41    [W:0.075 / U:0.020 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site