Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 17 Dec 2020 15:44:11 +0000
From:   Satya Tangirala <satyat@google.com>
To:     Chao Yu <yuchao0@huawei.com>
Cc:     "Theodore Y . Ts'o" <tytso@mit.edu>,
        Jaegeuk Kim <jaegeuk@kernel.org>,
        Eric Biggers <ebiggers@kernel.org>, Chao Yu <chao@kernel.org>,
        linux-kernel@vger.kernel.org, linux-fscrypt@vger.kernel.org,
        linux-f2fs-devel@lists.sourceforge.net
Subject: Re: [PATCH 0/3] add support for metadata encryption to F2FS
Message-ID: <X9t8y3rElyAPCLoD@google.com>
References: <20201005073606.1949772-1-satyat@google.com>
 <471e0eb7-b035-03da-3ee3-35d5880a6748@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <471e0eb7-b035-03da-3ee3-35d5880a6748@huawei.com>
Precedence: bulk

On Sat, Oct 10, 2020 at 05:53:06PM +0800, Chao Yu wrote:
> On 2020/10/5 15:36, Satya Tangirala wrote:
> > This patch series adds support for metadata encryption to F2FS using
> > blk-crypto.
> 
> It looks this implementation is based on hardware crypto engine, could you
> please add this info into f2fs.rst as well like inlinecrypt...
> 
To be precise, the implementation requires either a hardware crypto
engine *or* blk-crypto-fallback. I tried to clarify this a little better
in the commit messages and in fscrypt.rst, but thinking about it again
now, I think I should add a section about metadata encryption at the end
of f2fs.rst. I'll do that when I send out v3 of this patch series (I
just sent out v2).
> > 
> > Patch 3 wires up F2FS with the functions introduced in Patch 2. F2FS
> > will encrypt every block (that's not being encrypted by some other
> > encryption key, e.g. a per-file key) with the metadata encryption key
> > except the superblock (and the redundant copy of the superblock). The DUN
> > of a block is the offset of the block from the start of the F2FS
> > filesystem.
> 
> Why not using nid as DUN, then GC could migrate encrypted node block directly via
> meta inode's address space like we do for encrypted data block, rather than
> decrypting node block to node page and then encrypting node page with DUN of new
> blkaddr it migrates to.
> 
The issue is, the bi_crypt_context in a bio holds a single DUN value,
which is the DUN for the first data unit in the bio. blk-crypto assumes
that the DUN of each subsequent data unit can be computed by simply
incrementing the DUN. So physically contiguous data units can only be put
into the same bio if they also have contiguous DUNs. I don't know much
about nids, but if the nid is invariant w.r.t the physical block location,
then there might be more fragmentation of bios in regular read/writes
(because physical contiguity may have no relation to DUN contiguity). So I
think we should continue using the fsblk number as the DUN.