Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH 4.14 024/110] btrfs: use proper endianness accessors for
 super_copy
To:     Christoph Biedl <linux-kernel.bfrz@manchmal.in-ulm.de>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc:     linux-kernel@vger.kernel.org, stable@vger.kernel.org,
        Liu Bo <bo.li.liu@oracle.com>, David Sterba <dsterba@suse.com>
References: <20180307191039.748351103@linuxfoundation.org>
 <20180307191042.810088712@linuxfoundation.org>
 <1521139304@msgid.manchmal.in-ulm.de>
From:   Anand Jain <anand.jain@oracle.com>
Message-ID: <f5dc1577-356a-5a0f-193b-d421d61ce45c@oracle.com>
Date:   Sat, 17 Mar 2018 00:21:04 +0800
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.6.0
MIME-Version: 1.0
In-Reply-To: <1521139304@msgid.manchmal.in-ulm.de>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 03/16/2018 02:55 AM, Christoph Biedl wrote:
> Greg Kroah-Hartman wrote...
> 
>> 4.14-stable review patch.  If anyone has any objections, please let me know.
> 
>> commit 3c181c12c431fe33b669410d663beb9cceefcd1b upstream.
> (...)
> 
>> If the filesystem is always used on a same endian host, this will not
>> be a problem.
> 
>  From my observations I cannot quite subscribe to that.
> 
> On big-endian systems, this change intruduces severe corruption,
> resulting in complete loss of the data on the used block device.

Thanks for the report.

That's really bad, my mistake. I am digging to know how it happened. 
Our on-disk root bytenr are little-endian compatible.  So using the 
cpu_to_le for write on a big-endian arch is a correct thing to do. If it 
fails,  certainly there is something which I have overlooked. I am 
digging to know. Thanks for the report again.

Fsck won't be able to figure out the correct on-disk btyenr either.

If there isn't any backup we could try to find out the correct pointers 
manually.  However, restore from the backup approach is much better.

-Anand


> Steps to reproduce (tested on ppc/powerpc and parisc/hppa):
> 
> # mkfs.btrfs $DEV
> # mount $DEV /mnt/tmp/
> # umount /mnt/tmp/
> 
> This simple umount corrupts the file system:
> 
> # mount $DEV /mnt/tmp/
> mount: /mnt/tmp: wrong fs type, bad option, bad superblock on $DEV, missing codepage or helper program, or other error.
> 
> # dmesg:
> BTRFS critical (device <dev>): unable to find logical 4294967296 length 4096
> BTRFS critical (device <dev>): unable to find logical 4294967296 length 4096
> BTRFS critical (device <dev>): unable to find logical 18102363734671360 length 16384
> BTRFS error (device <dev>): failed to read chunk root
> BTRFS error (device <dev>): open_ctree failed
> 
> Also fsck is of no help:
> 
> # btrfsck $DEV
> Couldn't map the block 18102363734671360
> No mapping for 18102363734671360-18102363734687744
> Couldn't map the block 18102363734671360
> bytenr mismatch, want=18102363734671360, have=0
> ERROR: cannot read chunk root
> ERROR: cannot open file system
> 
> 
> Trying mount or fsck on a little-endian system does not help either. So
> I consider the data on that device lost - luckily I use btrfs only for
> files where a backup exists all the time.
> 
> 
> Reverting that change restored the previous error-free behaviour. I
> didn't check HEAD, i.e. v4.16-rc5, since the upstream commt was the last
> that affected these files. Still I could give this a try if anybody
> wishes so.
> 
> Cheers,
> 
>      Christoph
>