On Wed, Apr 09, 2014 at 11:04:49AM -0700, Darrick J. Wong wrote:

On Wed, Apr 09, 2014 at 12:56:41PM +0800, Zheng Liu wrote:

quoted

On Tue, Apr 08, 2014 at 11:40:03AM -0700, Darrick J. Wong wrote:

quoted

On Tue, Apr 08, 2014 at 07:54:25PM +0800, Zheng Liu wrote:

quoted

On Mon, Mar 10, 2014 at 08:06:04PM -0700, Darrick J. Wong wrote:

quoted

Hi all,

A couple of errors to complain about before I send out the e2fsprogs patchbomb.
:)

Zheng: I've been running the metadata checksum test with inline_data set.
flat_dir_test is a stress test which copies /usr/share/doc into a filesystem
and then "enlarges" the directories by recursively renaming "$foo" to
"$foo.longer", with the results below.  e2fsck complained that the directories
involved (4156, 4251) have multiple links to the subdir inode.  I'm not sure
what this is all about; the only (circumstantial) evidence I have is that it
goes away if I don't turn on inline_data.

Hi Darrick,

I guess that you were talking about this bug at LSF/MM submmit this
year, right?  Thanks for reporting this bug.  I am trying to take a
closer look at it.  It would be great if you could tell me where I can
find the your test program ('flat_dir_test').

It's buried in the metadata checksum test suite, alas.  The test copies an
arbitrary directory (/usr/share/doc) to the filesystem, then does a depth-first
traversal of the copied tree, renaming everything it finds from "oldname" to
"oldname.longer" to force the kernel to shuffle directory entries all over the
place, and likely causing spill-out from formerly inlinedata directories.

I write a simple script to try to reproduce the problem.  But,
unfortunately, I failed.  The kernel is based on ext4/dev branch, and
e2fsprogs is based on e2fsprogs/next branch.  I paste the script below
and the output of 'dumpe2fs -h'.  Please let me know if I miss
something.

It is necessary to umount and then mount the FS in between populating the FS
and renaming the files.  flat_dir_test also renames *everything*, which is why
you need the "-depth" so that the renames don't break find:

find $mnt/testdir -depth | while read f; do
        mv $f $f.longer
done

This doesn't affect the script's ability to reproduce errors; it just makes it
run faster.  The errors I see look like this:

[   51.207952] EXT4-fs error (device vda): ext4_generic_delete_entry:2118: inode #4332: block 687: comm mv: bad entry in directory: rec_len % 4 != 0 - offset=0(0), inode=33188, rec_len=1201, name_len=0
[   51.221472] EXT4-fs error (device vda) in ext4_delete_entry:2178: IO failure
[   51.230300] EXT4-fs warning (device vda): ext4_rename:3151: Deleting old file (4332), 2, error=-5
[   51.512388] EXT4-fs error (device vda): ext4_generic_delete_entry:2118: inode #4427: block 693: comm mv: bad entry in directory: directory entry across range - offset=2056(2056), inode=1397064974, rec_len=34064, name_len=69
[   51.524525] EXT4-fs error (device vda) in ext4_delete_entry:2178: IO failure

Welp, I figured out why this was happening -- if renaming within the same
directory causes inline data conversion during a add-then-delete-dirent
operation, the delete operation incorrectly tries to delete out of i_block
(which is now an extent tree), causing the FS error.

Patch soon.

--D

--D

quoted

Thanks,
                                                - Zheng


#!/bin/bash

dev='/dev/sda1'
mnt='/mnt/sda1'
e2fsprogs_base="$HOME/projects/ext4-dev/e2fsprogs"
mkfs="$e2fsprogs_base/misc/mke2fs"
fsck="$e2fsprogs_base/e2fsck/e2fsck"

sudo umount $mnt
sudo $mkfs -t ext4 -O inline_data,metadata_csum,64bit $dev
sudo mount -t ext4 $dev $mnt
sudo chown wenqing:wenqing $mnt

cd $mnt
mkdir testdir
cp -rf /usr/share/doc/* testdir/

function do_rename()
{
	tmpfilename=$(mktemp --dry-run)
	postfix=$(echo $tmpfilename | awk -F'/' '{print $3}')
	mv "$1" "$1-$postfix"
}

function do_dir()
{
	for ent in $1/*; do
		if [ -d $ent ]; then
			do_dir $ent
		else
			do_rename $ent
		fi
	done
}

do_dir $mnt/testdir

cd $e2fsprogs_base
sudo umount $mnt
sudo $fsck -f $dev

exit 0


dumpe2fs 1.43-WIP (4-Feb-2014)
Filesystem volume name:   <none>
Last mounted on:          /mnt/sda1
Filesystem UUID:          5643cf62-dd3b-4883-9991-4d7f319e3e1e
Filesystem magic number:  0xEF53
Filesystem revision #:    1 (dynamic)
Filesystem features:      has_journal ext_attr resize_inode dir_index filetype extent 64bit flex_bg inline_data sparse_super large_file huge_file dir_nlink extra_isize metadata_csum
Filesystem flags:         signed_directory_hash 
Default mount options:    user_xattr acl
Filesystem state:         clean
Errors behavior:          Continue
Filesystem OS type:       Linux
Inode count:              5021696
Block count:              20081242
Reserved block count:     1004062
Free blocks:              19663835
Free inodes:              5007473
First block:              0
Block size:               4096
Fragment size:            4096
Group descriptor size:    64
Reserved GDT blocks:      1024
Blocks per group:         32768
Fragments per group:      32768
Inodes per group:         8192
Inode blocks per group:   512
Flex block group size:    16
Filesystem created:       Wed Apr  9 11:41:36 2014
Last mount time:          Wed Apr  9 11:41:42 2014
Last write time:          Wed Apr  9 11:43:28 2014
Mount count:              0
Maximum mount count:      -1
Last checked:             Wed Apr  9 11:43:28 2014
Check interval:           0 (<none>)
Lifetime writes:          1601 MB
Reserved blocks uid:      0 (user root)
Reserved blocks gid:      0 (group root)
First inode:              11
Inode size:	          256
Required extra isize:     28
Desired extra isize:      28
Journal inode:            8
Default directory hash:   half_md4
Directory Hash Seed:      698d5a93-bfa8-4b55-b4ab-ae94c4e9e5e0
Journal backup:           inode blocks
Checksum type:            crc32c
Checksum:                 0x7c4d3477
Journal features:         journal_64bit
Journal size:             128M
Journal length:           32768
Journal sequence:         0x00000017
Journal start:            0

--
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