Re: BTRFS critical (device dm-0): corrupt node: root=256 block=1035372494848 slot=364, bad key order, current (8796143471049 108 0) next (50450969 1 0)

[Qu Wenruo <wqu@suse.com>]

在 2024/10/2 02:40, Peter Volkov 写道:
> On Tue, Oct 1, 2024 at 3:09 PM David Sterba <dsterba@suse.cz> wrote:
>> On Tue, Oct 01, 2024 at 02:15:51PM +0000, Peter Volkov wrote:
>>> Hi! I've been using this system with this kernel (6.10.10) for a few
>>> months already and today out of nowhere btrfs broke with this error
>>> message:
>>>
>>> [53923.816740] page dumped because: eb page dump
>>> [53923.816743] BTRFS critical (device dm-0): corrupt node: root=256
>>> block=1035372494848 slot=364, bad key order, current (8796143471049
>>> 108 0) next (50450969 1 0)
>>
>> Quite obvious memory bitflip:
>>
>> 8796143471049 = 0x8000301c9c9
>>       50450969 = 0x301d219
>>
>> The first one should probably be 0x301c9c9, but it's impossible to tell
>> how many other data/metadata could have been hit by this or another
>> memory bitflip so check can detect the things but not fix.
> 
> Thank you David! Is my understanding correct, that btrfs catches
> memory problems,
> so this bitflip most probably means that my drive is failing?

In this particular case, it's your hardware memory, not the drive.

The error is happening at write time, so the metadata read from disk is 
fine, thus not your driver returning some weird data.

Furthermore, it's pretty hard that a simple bitflip can pass the 
internal checksums of the storage device, thus it's very unlikely it's 
your drive.

So, please do a full memtest of your system before doing anything else.


And considering your fsck result is already bad, it's no doubt that some 
bitflip has already corrupted extent tree, and I believe the csum tree 
is also corrupted.

Thanks,
Qu
> 
> --
> Peter.
>