Re: [PATCH v5 7/9] fscrypt: add inline encryption support

From: Christoph Hellwig <hch@infradead.org>
To: Christoph Hellwig <hch@infradead.org>,
	Satya Tangirala <satyat@google.com>,
	linux-scsi@vger.kernel.org, Kim Boojin <boojin.kim@samsung.com>,
	Kuohong Wang <kuohong.wang@mediatek.com>,
	Barani Muthukumaran <bmuthuku@qti.qualcomm.com>,
	linux-f2fs-devel@lists.sourceforge.net,
	linux-block@vger.kernel.org, linux-fscrypt@vger.kernel.org,
	linux-fsdevel@vger.kernel.org
Subject: Re: [PATCH v5 7/9] fscrypt: add inline encryption support
Date: Mon, 4 Nov 2019 16:15:54 -0800	[thread overview]
Message-ID: <20191105001554.GA24056@infradead.org> (raw)
In-Reply-To: <20191031222500.GB111219@gmail.com>

On Thu, Oct 31, 2019 at 03:25:03PM -0700, Eric Biggers wrote:
> It's more important to clean up the IS_ENCRYPTED(inode) &&
> S_ISREG(inode->i_mode) checks that are duplicated in fs/{ext4,f2fs}/, so I've
> been thinking of adding a helper:
> 
> static inline bool fscrypt_needs_contents_encryption(const struct inode *inode)
> {
>         return IS_ENABLED(CONFIG_FS_ENCRYPTION) && IS_ENCRYPTED(inode) &&
>                S_ISREG(inode->i_mode);
> }

Sounds fine.

> I don't think combining these things is a good idea because it would restrict
> the use of inline encryption to filesystems that allow IV_INO_LBLK_64 encryption
> policies, i.e. filesystems that have stable inode numbers, 32-bit inodes, and
> 32-bit file logical block numbers.
> 
> The on-disk format (i.e. the type of encryption policy chosen) and the
> implementation (inline or filesystem-layer crypto) are really two separate
> things.  This was one of the changes in v4 => v5 of this patchset; these two
> things used to be conflated but now they are separate.  Now you can use inline
> encryption with the existing fscrypt policies too.
> 
> We could use two separate SB_* flags, like SB_INLINE_CRYPT and
> SB_IV_INO_LBLK_64_SUPPORT.

Yes, I think that is a good idea.

> However, the ->has_stable_inodes() and
> ->get_ino_and_lblk_bits() methods are nice because they separate the filesystem
> properties from the question of "is this encryption policy supported".
> Declaring the filesystem properties is easier to do because it doesn't require
> any fscrypt-specific knowledge.  Also, fs/crypto/ could use these properties in
> different ways in the future, e.g. if another IV generation scheme is added.

I don't really like writing up method boilerplates for something that
is a simple boolean flag.