Date:   Fri, 20 Jan 2017 23:24:40 +0100
From:   Damien Guibouret <damien.guibouret@...tition-saving.com>
To:     Andreas Dilger <adilger@...ger.ca>
CC:     Theodore Ts'o <tytso@....edu>,
        George Spelvin <linux@...encehorizons.net>,
        linux-ext4@...r.kernel.org
Subject: Re: kernel BUG at fs/ext4/inline.c:1943!

Andreas Dilger wrote:
> On Jan 20, 2017, at 12:14 PM, Damien Guibouret <damien.guibouret@...tition-saving.com> wrote:
> 
>>Theodore Ts'o wrote:
>>
>>>On Wed, Jan 18, 2017 at 03:21:28AM -0500, George Spelvin wrote:
>>>
>>>>I was trying to rmdir an empty directory in lost+found that accumulated
>>>>during my recent problems.
>>>>
>>>>EXT4-fs (md3): mounted filesystem with writeback data mode. Opts: data=writeback,delalloc
>>>>
>>>>$ cd /mountpoint/lost+found
>>>>$ rmdir \#1625089/
>>>>
>>>>------------[ cut here ]------------
>>>>kernel BUG at fs/ext4/inline.c:1943!
>>>>debugfs:  stat <1625089>
>>>>Inode: 1625089   Type: directory    Mode:  0775   Flags: 0x10000000
>>>>Generation: 927350643    Version: 0x00000000:00000004
>>>>User:  1000   Group:   161   Project:     0   Size: 132
>>>>File ACL: 1664090185    Directory ACL: 0
>>>>Links: 0   Blockcount: 8
>>>>Fragment:  Address: 0    Number: 0    Size: 0
>>>>ctime: 0x587f2034:472c74ec -- Wed Jan 18 02:58:44 2017
>>>>atime: 0x56b9e2f8:b68a7658 -- Tue Feb  9 08:00:40 2016
>>>>mtime: 0x56c1bc4b:a7765de8 -- Mon Feb 15 06:53:47 2016
>>>>crtime: 0x56ba9eb4:a51d90ac -- Tue Feb  9 21:21:40 2016
>>>>Size of extra inode fields: 32
>>>>Extended attributes:
>>>>system.data (72)
>>>>Inode checksum: 0xe2f12fc5
>>>>Size of inline data: 132
>>>
>>>OK, so the problem seems the inode in question has the INLINE_DATA
>>>flag set, but i_blocks is non-zero.  And it looks like the data was
>>>actually stored in an exernal data block, and the in-line xattr wasn't
>>>present.
>>>e2fsck should be checking and complaining if this is the case.  If
>>>not, it's a bug in e2fsck.
>>>And the kernel certainy shouldn't be BUG'ing when it comes across
>>>what is clearly a case of file system corruption.
>>>Cheers,
>>>					- Ted
>>>--
>>>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
>>
>>Hello,
>>
>>Perhaps I am wrong, but as the inode has a file ACL, the blockcount should be
>>different from 0?  So it seems valid on this point. Or is there something that
>>prevent inlined file to have ACL?
> 
> 
> The "File ACL" label is misleading.  ACLs are stored as xattrs, (along with other
> xattrs like SELinux labels), and also inline data.  However, unless the ACL is
> very large it would also be stored inside the inode itself.  If there isn't space
> inside the inode to store both the inline data and other xattrs, the inline data
> should be bumped to an external block first (i.e. stored as a regular file),
> since there is no benefit to doing the reverse.
> 
> 
> Cheers, Andreas
> 

I certainly do not have a complete knowledge on how it works, but the 
ext4_xattr_set_handle function does not try to move inline data to a block 
before allocating a new block for the extended attribute (or reusing a block if 
there is one with similar content). And some of its call tree does not clear 
inline data before.
A case could be some directory creation that inherits big posix ACL from its 
parent (ACL goes to a block, potentially the same than the one of its parent) 
and the directory is created with inline data.

In the ext4_getattr function there is also some comment that says it can happen 
"If there is inline data in the inode, the inode will normally not have data 
blocks allocated (it may have an external xattr block)."

Regards,

Damien
--
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