From: Namjae Jeon Subject: Re: Punching hole using fallocate is not removing the uninit extent from extent tree Date: Fri, 22 Jun 2012 07:44:35 +0900 Message-ID: References: Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Ashish Sangwan , linux-ext4@vger.kernel.org, Amit Sahrawat , "linux-kernel@vger.kernel.org" , "Ted Ts'o" To: =?UTF-8?B?THVrw6HFoSBDemVybmVy?= Return-path: Received: from mail-wg0-f44.google.com ([74.125.82.44]:49495 "EHLO mail-wg0-f44.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1759106Ab2FUWoh convert rfc822-to-8bit (ORCPT ); Thu, 21 Jun 2012 18:44:37 -0400 In-Reply-To: Sender: linux-ext4-owner@vger.kernel.org List-ID: Hi Lukas. Have you reproduced this problem ? Plz reivew our patch. Thanks. 2012/6/18, Namjae Jeon : > Add in mail loop. > > 2012/6/18, Namjae Jeon : >> 2012/6/18, Luk=C3=A1=C5=A1 Czerner : >>> On Mon, 28 May 2012, Ashish Sangwan wrote: >>> >>>> Date: Mon, 28 May 2012 19:38:05 +0530 >>>> From: Ashish Sangwan >>>> 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 mai= l. >> you can paste it to file. >> >> fragmentation.sh >> --------------------------------------------------------------------= --- >> dd if=3D/dev/zero of=3D12kb bs=3D4096 count=3D6 >> index=3D0 >> while [ $? =3D=3D 0 ] >> do >> index=3D$(($index + 1)) >> cp 12kb $1/file1.$index >> done >> echo "Partition filled" >> sync >> df -h >> index=3D0 >> while [ $? =3D=3D 0 ] >> do >> index=3D$(($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=3D/dev/zero of=3Dd_file bs=3D1024 count=3D10240 >> 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 a= t >>>> 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 i= ndex >>>> 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 =3D=3D depth) { >>>> /* this is leaf block */ >>>> err =3D ext4_ext_rm_leaf(handle, inode, pat= h, >>>> start, end); >>>> /* root level has p_bh =3D=3D NULL, brelse() eats this */ >>>> brelse(path[i].p_bh); >>>> path[i].p_bh =3D NULL; >>>> i--; >>>> continue; >>>> } >>>> >>>> Varibale "i" does not become equals to "depth" because >>>> ext4_ext_more_to_rm is returning "0" hence the following if condit= ion >>>> 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 =3D=3D 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) =3D=3D path->p_blo= ck) <=3D >>>> This condition is turning out to be true >>>> return 0; <=3D 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= =2E >>>> -- >>>> To unsubscribe from this list: send the line "unsubscribe linux-ex= t4" >>>> 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-ext4" i= n the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html