From: "Darrick J. Wong" <djwong@us.ibm.com>
Subject: Re: [PATCH 01/54] libext2fs: Read and write full size inodes
Date: Mon, 12 Mar 2012 15:23:25 -0700
Message-ID: <20120312222325.GM15164@tux1.beaverton.ibm.com>
References: <20120306235720.11945.30629.stgit@elm3b70.beaverton.ibm.com> <20120306235727.11945.11066.stgit@elm3b70.beaverton.ibm.com> <20120309233631.GE5635@thunk.org>
Reply-To: djwong@us.ibm.com
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: Andreas Dilger <adilger.kernel@dilger.ca>,
	Sunil Mushran <sunil.mushran@oracle.com>,
	Amir Goldstein <amir73il@gmail.com>,
	Andi Kleen <andi@firstfloor.org>,
	Mingming Cao <cmm@us.ibm.com>,
	Joel Becker <jlbec@evilplan.org>, linux-ext4@vger.kernel.org,
	Coly Li <colyli@gmail.com>
To: "Ted Ts'o" <tytso@mit.edu>
Content-Disposition: inline
In-Reply-To: <20120309233631.GE5635@thunk.org>
Sender: linux-ext4-owner@vger.kernel.org

On Fri, Mar 09, 2012 at 06:36:31PM -0500, Ted Ts'o wrote:
> On Tue, Mar 06, 2012 at 03:57:27PM -0800, Darrick J. Wong wrote:
> > Change libext2fs to read and write full-size inodes in preparation for the
> > metadata checksumming patchset, which will require this.  Due to ABI
> > compatibility requirements, this change must be hidden from client programs.
> > 
> > Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
> 
> After applying this first patch, the e2fsprogs regression test suite
> blew up spectacularly caused by the malloc-managed free lists pointers
> getting corrupted:
> 
> *** glibc detected *** /usr/projects/e2fsprogs/e2fsprogs/build/debugfs/debugfs: free(): invalid next size (fast): 0x000000000063f9d0 ***
> ======= Backtrace: =========
> /lib/libc.so.6(+0x77806)[0x7ffff6e4a806]
> /lib/libc.so.6(cfree+0x73)[0x7ffff6e510d3]
> /usr/projects/e2fsprogs/e2fsprogs/build/lib/libext2fs.so.2(ext2fs_free_mem+0x30)[0x7ffff7bb562f]
> 
> 
> Interestingly, valgrind was *not* useful in finding the problem;
> apparently it was getting confused by the ext2fs_get_mem()
> abstraction, which is unfortunate.  I'll have to look into that at
> some point.
> 
> Anyway, the problem was in ext2fs_write_inode_full(), and it could be
> replicated by simply writing to an inode, i.e.
> 
>    mke2fs  -F -O resize_inode -o Linux -b 1024 /tmp/image  16384
>    debugfs -w -R "set_inode_field <7> mtime now" /tmp/image
> 
> is enough to trigger it.  The problem is with a 128 byte ext2 file
> system, ext2fs_write_inode_full() is passed a large inode and so
> bufsize is 156, but EXT2_INODE_SIZE(fs->super) is 128.  So at
> lib/ext2fs/inode.c:698:
> 
> 	memcpy(w_inode, inode, bufsize);
> 
> you end up writing 156 bytes into a memory buffer that was allocated
> to a size of 128 bytes.  Hilarity ensues.
> 
> The fix is relatively simple:
> 
> 	memcpy(w_inode, inode, (bufsize > length) ? length : bufsize);
> 
> Anyway, this is *why* running the regression tests are important.
> (And why projects which don't have regression test suites are just
> asking for trouble.)
> 
> They catch all sorts of interesting oversights like this....

Hmm, good catch.  To be honest I wasn't aware that there /was/ a make check.
I'll contribute my checksum tests when I figure out how to integrate them.

I'll send out patches to fix the other 45 failures this afternoon; I traced
most of them to a mistake I made in e2fsck/rehash.c.

--D