On Fri, Feb 12, 2016 at 09:26:28AM -0800, Marc MERLIN wrote:

On Fri, Feb 12, 2016 at 08:33:11AM +0800, Qu Wenruo wrote:

quoted

There is still a last chance.

If btrfsck still report original error about "bad file extent" in root: 
 45851/45852/...
Btrfs-debug-tree may provide useful info by dumping only that root.

# btrfs-debug-tree -t 45851

But the problem is, there is no filename fuzz option.
You need to mask all the filenames in INODE_REF/DIR_ITEM/DIR_INDEX by 
script, or just grep the affected inode info following the pattern "key 
(<INODE_NUM>".

At least this should tell us what's the problem and we can check 
manually to determine if it's fixable.

Mmmh, so the fsck is looking a bit worse now, here is the output:
http://marc.merlins.org/tmp/ggm-broken-ds1-fsck.txt

I'm not super sure what inode I should use for debug tree. Can you suggest
one?

I ran the dump 
gargamel:~# btrfs-image -s -c 9 /dev/mapper/dshelf1old
/mnt/dshelf1/ds1old.dump
Error adding space cache blocks -5
Error flushing pending -5
create failed (Success)

and a du every so often showed the file go to 9.3GB before btrfs-image
deleted it:
9.3G    /mnt/dshelf1/ds1old.dump

So I'm at a point where I need to delete this filesystem. I believe it'll be
monday morning soon on your side of the world :) 
Can you let me know if there is anything else I can/should get before I wipe
it?

Thanks,
Marc
-- 
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
                                      .... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/