Re: [PATCH v3 04/22] fscrypt: add extent-based encryption

[Eric Biggers <ebiggers@kernel.org>]

On Thu, Oct 20, 2022 at 06:55:04PM -0400, Sweet Tea Dorminy wrote:
> 
> 
> On 10/20/22 17:45, Eric Biggers wrote:
> > On Thu, Oct 20, 2022 at 12:58:23PM -0400, Sweet Tea Dorminy wrote:
> > > Some filesystems need to encrypt data based on extents, rather than on
> > > inodes, due to features incompatible with inode-based encryption. For
> > > instance, btrfs can have multiple inodes referencing a single block of
> > > data, and moves logical data blocks to different physical locations on
> > > disk in the background; these two features mean traditional inode-based
> > > file contents encryption will not work for btrfs.
> > > 
> > > This change introduces fscrypt_extent_context objects, in analogy to
> > > existing context objects based on inodes. For a filesystem which opts to
> > > use extent-based encryption, a new hook provides a new
> > > fscrypt_extent_context, generated in close analogy to the IVs generated
> > > with existing policies. During file content encryption/decryption, the
> > > existing fscrypt_context object provides key information, while the new
> > > fscrypt_extent_context provides IV information. For filename encryption,
> > > the existing IV generation methods are still used, since filenames are
> > > not stored in extents.
> > > 
> > > Signed-off-by: Sweet Tea Dorminy <sweettea-kernel@dorminy.me>
> > > ---
> > >   fs/crypto/crypto.c          | 20 ++++++++--
> > >   fs/crypto/fscrypt_private.h | 25 +++++++++++-
> > >   fs/crypto/inline_crypt.c    | 28 ++++++++++---
> > >   fs/crypto/policy.c          | 79 +++++++++++++++++++++++++++++++++++++
> > >   include/linux/fscrypt.h     | 47 ++++++++++++++++++++++
> > >   5 files changed, 189 insertions(+), 10 deletions(-)
> > > 
> > > diff --git a/fs/crypto/crypto.c b/fs/crypto/crypto.c
> > > index 7fe5979fbea2..08b495dc5c0c 100644
> > > --- a/fs/crypto/crypto.c
> > > +++ b/fs/crypto/crypto.c
> > > @@ -81,8 +81,22 @@ void fscrypt_generate_iv(union fscrypt_iv *iv, u64 lblk_num,
> > >   			 const struct fscrypt_info *ci)
> > >   {
> > >   	u8 flags = fscrypt_policy_flags(&ci->ci_policy);
> > > +	struct inode *inode = ci->ci_inode;
> > > +	const struct fscrypt_operations *s_cop = inode->i_sb->s_cop;
> > > -	memset(iv, 0, ci->ci_mode->ivsize);
> > > +	memset(iv, 0, sizeof(*iv));
> > > +	if (s_cop->get_extent_context && lblk_num != U64_MAX) {
> > > +		size_t extent_offset;
> > > +		union fscrypt_extent_context ctx;
> > > +		int ret;
> > > +
> > > +		ret = fscrypt_get_extent_context(inode, lblk_num, &ctx,
> > > +						 &extent_offset, NULL);
> > > +		WARN_ON_ONCE(ret);
> > > +		memcpy(iv->raw, ctx.v1.iv.raw, sizeof(*iv));
> > > +		iv->lblk_num += cpu_to_le64(extent_offset);
> > > +		return;
> > > +	}
> > 
> > Please read through my review comment
> > https://lore.kernel.org/linux-fscrypt/Yx6MnaUqUTdjCmX+@quark/ again, as it
> > doesn't seem that you've addressed it.
> > 
> > - Eric
> 
> I probably didn't understand it correctly. I think there were three points
> in it:
> 
> 1) reconsider per-extent keys
> 2) make IV generation work for non-directkey policies as similarly as
> possible to how they work in inode-based filesystems
> 3) never use 'file-based' except in contrast to dm-crypt and other
> block-layer encryption.
> 
> For point 2, I changed the initial extent context generation to match up
> with fscrypt_generate_iv() (and probably didn't call that out enough in the
> description). (Looking at it again, I could literally call
> fscrypt_generate_iv() to generate the initial extent context; I didn't
> realize that before).
> 
> Then adding lblk_num to the existing lblk_num in the iv from the start of
> the extent should be the same as the iv->lblk_num setting in the inode-based
> case: for lblk 12, for instance, the same IV should result from inode-based
> with lblk 12, as with extent-based with an initial lblk_num of 9 and an
> extent_offset of 3. For shared extents, they'll be different, but for
> singly-referenced extents, the IVs should be exactly the same in theory.
> 
> I'm not sure whether I misunderstood the points or didn't address them
> fully, I apologize. Would you be up for elaborating where I missed, either
> by email or by videochat whenever works for you?

It seems you misunderstood point (2).  See what I said below:

	So if you do want to implement the DIRECT_KEY method, the natural thing
	to do would be to store a 16-byte nonce along with each extent, and use
	the DIRECT_KEY IV generation method as-is.  It seems that you've done it
	a bit differently; you store a 32-byte nonce and generate the IV as
	'nonce + lblk_num', instead of 'nonce || lblk_num'.  I think that's a
	mistake -- it should be exactly the same.

	If the issue is that the 'nonce || lblk_num' method doesn't allow for
	AES-XTS support, we could extend DIRECT_KEY to do 'nonce + lblk_num'
	*if* the algorithm has a 16-byte IV size and thus has to tolerate some
	chance of IV reuse.  Note that this change would be unrelated to
	extent-based encryption, and could be applied regardless of it.

So:

1.) Provided that you've decided against per-extent keys, and are not trying to
    support UFS and eMMC inline encryption hardware, then you should *only*
    support DIRECT_KEY -- not other settings that don't make sense.

2.) There should be a preparatory patch that makes DIRECT_KEY be allowed when
    the IV length is 16 bytes, using the method 'nonce + lblk_num' -- assuming
    that you need AES-XTS support and aren't planning on supporting Adiantum or
    AES-HCTR2 only.  (The small chance for IV reuse that it results in is not
    ideal, but it's probably tolerable.  Maybe the nonce should also be hashed
    with a secret key, like what IV_INO_LBLK_32 does with the inode number; I'll
    have to think about it.)  If you plan to just support AES-HCTR2 instead of
    AES-XTS, then you'd need a patch to allow AES-HCTR2 for contents encryption,
    as currently it is only allowed for filenames.

3.) Each extent context should contain a 16-byte random nonce, that is newly
    generated just for that extent -- not copied from anywhere.

4.) IVs should be generated using the DIRECT_KEY method.  That is,
    'nonce || lblk_num' if the IV length allows it, otherwise 'nonce + lblk_num'
    as mentioned in (2).  For inode-based encryption, nonce means the inode's
    nonce, and lblk_num means the index of the block in the inode.  For
    extent-based encryption, nonce will mean the extent's nonce, and lblk_num
    will mean the index of the block in the extent.

- Eric