ext4: don't allow journal inode to have encrypt flag

From: Eric Biggers <ebiggers@google.com>

Mounting a filesystem whose journal inode has the encrypt flag causes a
NULL dereference in fscrypt_limit_io_blocks() when the 'inlinecrypt'
mount option is used.

The problem is that when jbd2_journal_init_inode() calls bmap(), it
eventually finds its way into ext4_iomap_begin(), which calls
fscrypt_limit_io_blocks().  fscrypt_limit_io_blocks() requires that if
the inode is encrypted, then its encryption key must already be set up.
That's not the case here, since the journal inode is never "opened" like
a normal file would be.  Hence the crash.

A reproducer is:

    mkfs.ext4 -F /dev/vdb
    debugfs -w /dev/vdb -R "set_inode_field <8> flags 0x80808"
    mount /dev/vdb /mnt -o inlinecrypt

To fix this, make ext4 consider journal inodes with the encrypt flag to
be invalid.  (Note, maybe other flags should be rejected on the journal
inode too.  For now, this is just the minimal fix for the above issue.)

I've marked this as fixing the commit that introduced the call to
fscrypt_limit_io_blocks(), since that's what made an actual crash start
being possible.  But this fix could be applied to any version of ext4
that supports the encrypt feature.

Reported-by: syzbot+ba9dac45bc76c490b7c3@syzkaller.appspotmail.com
Fixes: 38ea50daa7a4 ("ext4: support direct I/O with fscrypt using blk-crypto")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 fs/ext4/super.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)


base-commit: 8f71a2b3f435f29b787537d1abedaa7d8ebe6647

On Tue, Nov 01, 2022 at 10:33:12PM -0700, Eric Biggers wrote:
> From: Eric Biggers <ebiggers@google.com>
> 
> Mounting a filesystem whose journal inode has the encrypt flag causes a
> NULL dereference in fscrypt_limit_io_blocks() when the 'inlinecrypt'
> mount option is used.
> 
> The problem is that when jbd2_journal_init_inode() calls bmap(), it
> eventually finds its way into ext4_iomap_begin(), which calls
> fscrypt_limit_io_blocks().  fscrypt_limit_io_blocks() requires that if
> the inode is encrypted, then its encryption key must already be set up.
> That's not the case here, since the journal inode is never "opened" like
> a normal file would be.  Hence the crash.
> 
> A reproducer is:
> 
>     mkfs.ext4 -F /dev/vdb
>     debugfs -w /dev/vdb -R "set_inode_field <8> flags 0x80808"
>     mount /dev/vdb /mnt -o inlinecrypt
> 
> To fix this, make ext4 consider journal inodes with the encrypt flag to
> be invalid.  (Note, maybe other flags should be rejected on the journal
> inode too.  For now, this is just the minimal fix for the above issue.)
> 
> I've marked this as fixing the commit that introduced the call to
> fscrypt_limit_io_blocks(), since that's what made an actual crash start
> being possible.  But this fix could be applied to any version of ext4
> that supports the encrypt feature.
> 
> Reported-by: syzbot+ba9dac45bc76c490b7c3@syzkaller.appspotmail.com
> Fixes: 38ea50daa7a4 ("ext4: support direct I/O with fscrypt using blk-crypto")
> Cc: stable@vger.kernel.org
> Signed-off-by: Eric Biggers <ebiggers@google.com>
> ---
>  fs/ext4/super.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/ext4/super.c b/fs/ext4/super.c
> index 7950904fbf04f..2274f730b87e5 100644
> --- a/fs/ext4/super.c
> +++ b/fs/ext4/super.c
> @@ -5723,7 +5723,7 @@ static struct inode *ext4_get_journal_inode(struct super_block *sb,
>  
>  	ext4_debug("Journal inode found at %p: %lld bytes\n",
>  		  journal_inode, journal_inode->i_size);
> -	if (!S_ISREG(journal_inode->i_mode)) {
> +	if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) {
>  		ext4_msg(sb, KERN_ERR, "invalid journal inode");
>  		iput(journal_inode);
>  		return NULL;
> 
> base-commit: 8f71a2b3f435f29b787537d1abedaa7d8ebe6647
> -- 

Ping.

- Eric

On Tue, 1 Nov 2022 22:33:12 -0700, Eric Biggers wrote:
> From: Eric Biggers <ebiggers@google.com>
> 
> Mounting a filesystem whose journal inode has the encrypt flag causes a
> NULL dereference in fscrypt_limit_io_blocks() when the 'inlinecrypt'
> mount option is used.
> 
> The problem is that when jbd2_journal_init_inode() calls bmap(), it
> eventually finds its way into ext4_iomap_begin(), which calls
> fscrypt_limit_io_blocks().  fscrypt_limit_io_blocks() requires that if
> the inode is encrypted, then its encryption key must already be set up.
> That's not the case here, since the journal inode is never "opened" like
> a normal file would be.  Hence the crash.
> 
> [...]

Applied, thanks!

[1/1] ext4: don't allow journal inode to have encrypt flag
      commit: 29cef51d8522c4d8953856afaffcaf1b754e4f6c

Best regards,

On Tue, Dec 06, 2022 at 04:01:48PM -0500, Theodore Ts'o wrote:
> On Tue, 1 Nov 2022 22:33:12 -0700, Eric Biggers wrote:
> > From: Eric Biggers <ebiggers@google.com>
> > 
> > Mounting a filesystem whose journal inode has the encrypt flag causes a
> > NULL dereference in fscrypt_limit_io_blocks() when the 'inlinecrypt'
> > mount option is used.
> > 
> > The problem is that when jbd2_journal_init_inode() calls bmap(), it
> > eventually finds its way into ext4_iomap_begin(), which calls
> > fscrypt_limit_io_blocks().  fscrypt_limit_io_blocks() requires that if
> > the inode is encrypted, then its encryption key must already be set up.
> > That's not the case here, since the journal inode is never "opened" like
> > a normal file would be.  Hence the crash.
> > 
> > [...]
> 
> Applied, thanks!
> 
> [1/1] ext4: don't allow journal inode to have encrypt flag
>       commit: 29cef51d8522c4d8953856afaffcaf1b754e4f6c
> 
> Best regards,
> -- 
> Theodore Ts'o <tytso@mit.edu>

Thanks Ted.  Note that I also sent an e2fsprogs patch to make e2fsck fix this
situation: https://lore.kernel.org/r/20221102220551.3940-1-ebiggers@kernel.org

- Eric