Message-ID: <CADtmLM_erJXuuq6gMt_YvRip1P4b+DQ0j1P3ETV_gY=G1+-aCw@mail.gmail.com>
Date:	Tue, 23 Feb 2016 08:09:00 +0100
From:	Alexander Peganz <a.peganz@...il.com>
To:	Andreas Dilger <adilger@...ger.ca>
Cc:	linux-ext4@...r.kernel.org
Subject: Re: support request: how to fix corruption after offline-shrink?

I guess I should have ordered more disks than I did so I could do a
full dd... I'm currently reshaping two other arrays to free up
additional disks. They should be done by the end of the week.

Meanwhile I have started copying the most critical data from a ro
mount to another fs with some free space. It'll be interesting to see
what differences there will be between those files and the files on a
repaired fs. I should really get into the habit of storing checksums
for files.

Thanks for the info on the feature bits. That one has been nagging me
for quite a while.

2016-02-22 22:51 GMT+01:00 Andreas Dilger <adilger@...ger.ca>:
> On Feb 22, 2016, at 2:58 AM, Alexander Peganz <a.peganz@...il.com> wrote:
>> Shrinking an ext4 filesystem with resize2fs 1.42.5 from Debian
>> Wheezy's e2fsprogs corrupted the filesystem. I have found out from
>> mailing list archives and blog and forum posts that offline resizing
>> with such old versions of resize2fs is prone to corrupt ext4
>> filesystems. So I probably have run into one of those bugs. If I
>> understand the older messages I found correctly the data is actually
>> still complete and undamaged, but some of the metadata was somewhat
>> scrambled during the resize. Now I am looking for the most reliable
>> way to safe the most data.
>>
>>
>> I have since updated e2fsprogs to Stretch's 1.42.13. Checking with
>> e2fsck -fn the fs gives me a few hundred error messages each of:
>> Inode X, end of extent exceeds allowed value
>> Logical start X does not match logical start Y at next level.
>> Inode X, i_blocks is Y, should be Z.
>> Plus a long list of Block bitmap differences.
>>
>> tune2fs -l states the fs is clean with errors with the following
>> features: has_journal ext_attr resize_inode dir_index filetype extent
>> flex_bg sparse_super large_file huge_file uninit_bg dir_nlink
>> extra_isize
>>
>> My first instinct was to e2fsck -fp the fs, but -p tells me it cannot
>> safely fix the fs. I dabbled a bit with debugfs (admittedly not really
>> knowing what exactly I'm doing) and the fs seems to be largely intact,
>> with little more than a hundred files of the 6TB (around 4 in use)
>> being affected - although I moved around 2TB worth of files to another
>> fs earlier before noticing the corruption, so a few dozen of those are
>> probably damaged.
>>
>>
>> What I'd like to know is how to proceed from here. If I run e2fsck -fy
>> and hope for the best - can this only make things better or do I risk
>> causing further damage?
>>
>
>> I am currently waiting for a few additional disks, once they get here
>> I could try mounting the fs (I'm guessing mount can be convinced to
>> mount the fs without checking it first when the interval- and mount
>> count checks are disabled beforehand with tune2fs?) and just copying
>> files over to the new disks, but I guess that I would loose the chance
>> to repair any files that are currently damaged?
>
> If you have the capacity to do so, it is recommended to make a full "dd"
> backup of the original filesystem device, and then run "e2fsck -fy" on
> the backup, so that you can always make _another_ copy from the original
> should this go badly.  If the "e2fsck -fy" on the backup goes well, you
> can run e2fsck on the primary copy, or just use the new copy and reformat
> the original (after possibly keeping it around for some time for safety).
>
>
>> Any assistance that can be provided is greatly appreciated!
>>
>>
>> PS:
>> In case it helps here is the brief history of the fs as far as I remember it:
>> The fs was created unter Ubuntu 10.04LTS, so probably with a really
>> old version of mke2fs. It was online-grown with 10.04's resize2fs when
>> more disks were added to the RAID array. The array was later moved to
>> a Debian Wheezy server were it was in use for a few years before the
>> fateful offline shrink was performed.
>>
>>
>> PPS:
>> Not related at all to the problem but something that has always
>> confused me and I never found definite info on: if features that seem
>> to be supersets of other features (e.g. huge_file > large_file,
>> sparse_super2 > sparse_super) are both enabled on a fs I'm guessing
>> the more powerful one "wins"? Or are both flags required?
>
> In some cases the new feature supersedes the older one, but often they
> are complimentary.  For example "large_file" allows storing the high 32
> bits of the file size (i.e. files > 2^32 bytes in size = 4GB), while
> "huge_file" allows storing the high 16 bits of the block count (i.e.
> files > 2^32 sectors in size = 2TB), so they both need to be enabled.
>
> Cheers, Andreas
>
>
>
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists