linux-ext4 - Re: [f2fs-dev] [PATCH V2 10/13] fscrypt_encrypt_page: Loop across all blocks mapped by a page range

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20190430230840.GE48973@gmail.com>
Date:   Tue, 30 Apr 2019 16:08:41 -0700
From:   Eric Biggers <ebiggers@...nel.org>
To:     Chandan Rajendra <chandan@...ux.ibm.com>
Cc:     tytso@....edu, linux-f2fs-devel@...ts.sourceforge.net,
        hch@...radead.org, linux-fscrypt@...r.kernel.org,
        adilger.kernel@...ger.ca, linux-fsdevel@...r.kernel.org,
        jaegeuk@...nel.org, linux-ext4@...r.kernel.org
Subject: Re: [f2fs-dev] [PATCH V2 10/13] fscrypt_encrypt_page: Loop across
 all blocks mapped by a page range

On Tue, Apr 30, 2019 at 10:11:35AM -0700, Eric Biggers wrote:
> On Sun, Apr 28, 2019 at 10:01:18AM +0530, Chandan Rajendra wrote:
> > For subpage-sized blocks, this commit now encrypts all blocks mapped by
> > a page range.
> > 
> > Signed-off-by: Chandan Rajendra <chandan@...ux.ibm.com>
> > ---
> >  fs/crypto/crypto.c | 37 +++++++++++++++++++++++++------------
> >  1 file changed, 25 insertions(+), 12 deletions(-)
> > 
> > diff --git a/fs/crypto/crypto.c b/fs/crypto/crypto.c
> > index 4f0d832cae71..2d65b431563f 100644
> > --- a/fs/crypto/crypto.c
> > +++ b/fs/crypto/crypto.c
> > @@ -242,18 +242,26 @@ struct page *fscrypt_encrypt_page(const struct inode *inode,
> 
> Need to update the function comment to clearly explain what this function
> actually does now.
> 
> >  {
> >  	struct fscrypt_ctx *ctx;
> >  	struct page *ciphertext_page = page;
> > +	int i, page_nr_blks;
> >  	int err;
> >  
> >  	BUG_ON(len % FS_CRYPTO_BLOCK_SIZE != 0);
> >  
> 
> Make a 'blocksize' variable so you don't have to keep calling i_blocksize().
> 
> Also, you need to check whether 'len' and 'offs' are filesystem-block-aligned,
> since the code now assumes it.
> 
> 	const unsigned int blocksize = i_blocksize(inode);
> 
>         if (!IS_ALIGNED(len | offs, blocksize))
>                 return -EINVAL;
> 
> However, did you check whether that's always true for ubifs?  It looks like it
> may expect to encrypt a prefix of a block, that is only padded to the next
> 16-byte boundary.
> 		
> > +	page_nr_blks = len >> inode->i_blkbits;
> > +
> >  	if (inode->i_sb->s_cop->flags & FS_CFLG_OWN_PAGES) {
> >  		/* with inplace-encryption we just encrypt the page */
> > -		err = fscrypt_do_page_crypto(inode, FS_ENCRYPT, lblk_num, page,
> > -					     ciphertext_page, len, offs,
> > -					     gfp_flags);
> > -		if (err)
> > -			return ERR_PTR(err);
> > -
> > +		for (i = 0; i < page_nr_blks; i++) {
> > +			err = fscrypt_do_page_crypto(inode, FS_ENCRYPT,
> > +						lblk_num, page,
> > +						ciphertext_page,
> > +						i_blocksize(inode), offs,
> > +						gfp_flags);
> > +			if (err)
> > +				return ERR_PTR(err);

Apparently ubifs does encrypt data shorter than the filesystem block size, so
this part is wrong.

I suggest we split this into two functions, fscrypt_encrypt_block_inplace() and
fscrypt_encrypt_blocks(), so that it's conceptually simpler what each function
does.  Currently this works completely differently depending on whether the
filesystem set FS_CFLG_OWN_PAGES in its fscrypt_operations, which is weird.

I also noticed that using fscrypt_ctx for writes seems to be unnecessary.
AFAICS, page_private(bounce_page) could point directly to the pagecache page.
That would simplify things a lot, especially since then fscrypt_ctx could be
removed entirely after you convert reads to use read_callbacks_ctx.

IMO, these would be worthwhile cleanups for fscrypt by themselves, without
waiting for the read_callbacks stuff to be finalized.  Finalizing the
read_callbacks stuff will probably require reaching a consensus about how they
should work with future filesystem features like fsverity and compression.

So to move things forward, I'm considering sending out a series with the above
cleanups for fscrypt, plus the equivalent of your patches:

	"fscrypt_encrypt_page: Loop across all blocks mapped by a page range"
	"fscrypt_zeroout_range: Encrypt all zeroed out blocks of a page"
	"Add decryption support for sub-pagesized blocks" (fs/crypto/ part only)

Then hopefully we can get all that applied for 5.3 so that fs/crypto/ itself is
ready for blocksize != PAGE_SIZE; and get your changes to ext4_bio_write_page(),
__ext4_block_zero_page_range(), and ext4_block_write_begin() applied too, so
that ext4 is partially ready for encryption with blocksize != PAGE_SIZE.

Then only the read_callbacks stuff will remain, to get encryption support into
fs/mpage.c and fs/buffer.c.  Do you think that's a good plan?

Thanks!

- Eric