linux-ext4 - Re: [BUG+PATCH] mke2fs: Inode checksum does not match inode while creating orphan file

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Date:   Thu, 24 Aug 2023 22:05:55 +0200
From:   Jan Kara <jack@...e.cz>
To:     Sven Zühlsdorf <sven.zuehlsdorf@...em.de>
Cc:     Theodore Ts'o <tytso@....edu>, Jan Kara <jack@...e.cz>,
        linux-ext4@...r.kernel.org
Subject: Re: [BUG+PATCH] mke2fs: Inode checksum does not match inode while
 creating orphan file

On Thu 24-08-23 14:29:50, Sven Zühlsdorf wrote:
> Hi,
> 
> using version 1.7.0 (commit 25ad8a43) I encountered a bug in mke2fs, as
> creating new filesystems now fails on some devices:
> > # mke2fs -t ext4 /dev/nvme0n1p2
> > mke2fs 1.47.0 (5-Feb-2023)
> > Discarding device blocks: done
> > Creating filesystem with 4194304 4k blocks and 1048576 inodes
> > Filesystem UUID: d379784e-c92d-431f-b527-c4088299a914
> > Superblock backups stored on blocks:
> > 	32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208,
> > 	4096000
> > 
> > Allocating group tables: done
> > Writing inode tables: done
> > Creating journal (32768 blocks): done
> > mke2fs: Inode checksum does not match inode while creating orphan file
> 
> As a workaround, disabling the new orphan_file feature (i.e. `mke2fs -O
> ^orphan_file ...') allows mke2fs to succeed.
> 
> I could trace this to ext2fs_create_orphan_file's call to ext2fs_read_inode
> which reads the inode from disk, disregarding that the inode may have just
> been created and not written to disk yet.
> On devices where discarding produces NULs this isn't an issue, since
> ext2fs_read_inode will read a zeroed-out struct inode and the checksum
> verification function has a special case for the checksum still being zero.
> I assume enabling orphan_file after a file system has been in use for some
> time could run into this bug as well, but I haven't verified that.
> On some of the devices we use, however, discard results in random looking
> data, leading to above checksum failure.
> 
> From other places creating inodes I cobbled together the attached patch,
> allowing mke2fs to succceed; a subsequent forced fsck succeeds with no
> issues as well.

Thanks for the analysis and the fix! It looks good, I've just somewhat
fixed up whitespace (end of lines got somehow garbled for me) and also
removed the pointless inode reading in ext2fs_create_orphan_file(). The
result is attached.

								Honza

-- 
Jan Kara <jack@...e.com>
SUSE Labs, CR

View attachment "0001-ext2fs-Fix-initialization-of-orphan-file.patch" of type "text/x-patch" (2978 bytes)