[1/3] fscrypt: add support for inline-encryption-optimized policies
diff mbox series

Message ID 20191021230355.23136-2-ebiggers@kernel.org
State Superseded
Headers show
Series
  • fscrypt: support for inline-encryption-optimized policies
Related show

Commit Message

Eric Biggers Oct. 21, 2019, 11:03 p.m. UTC
From: Eric Biggers <ebiggers@google.com>

Some inline encryption hardware has only a small number of keyslots,
which would make it inefficient to use the traditional fscrypt per-file
keys.  The existing DIRECT_KEY encryption policy flag doesn't solve this
because it assumes that file contents and names are encrypted by the
same algorithm and that IVs are at least 24 bytes.

Therefore, add a new encryption policy flag INLINE_CRYPT_OPTIMIZED which
causes the encryption to modified as follows:

- The key for file contents encryption is derived from the values
  (master_key, mode_num, filesystem_uuid).  The per-file nonce is not
  included, so many files may share the same contents encryption key.

- The IV for encrypting each block of file contents is built as
  (inode_number << 32) | file_logical_block_num.

Including the inode number in the IVs ensures that data in different
files is encrypted differently, despite per-file keys not being used.
Limiting the inode and block numbers to 32 bits and putting the block
number in the low bits is needed to be compatible with inline encryption
hardware which only supports specifying a 64-bit data unit number which
is auto-incremented; this is what the UFS and EMMC standards support.

This IV generation method places extra constraints on filesystems: inode
numbers must be stable, which may preclude filesystem shrinking; and
inode and file logical block numbers can be at most 32-bit.  Therefore,
only allow INLINE_CRYPT_OPTIMIZED policies on filesystems that declare
that they meet these constraints.

Note that INLINE_CRYPT_OPTIMIZED is an on-disk format, not an
implementation; it doesn't require the use of inline encryption.  This
patch adds support for it using the existing filesystem layer
encryption.  A later patch will add inline encryption support.

Co-developed-by: Satya Tangirala <satyat@google.com>
Signed-off-by: Satya Tangirala <satyat@google.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 Documentation/filesystems/fscrypt.rst | 53 +++++++++++++++++++--------
 fs/crypto/crypto.c                    | 11 +++++-
 fs/crypto/fscrypt_private.h           | 20 +++++++---
 fs/crypto/keyring.c                   |  6 ++-
 fs/crypto/keysetup.c                  | 47 +++++++++++++++++++-----
 fs/crypto/policy.c                    | 42 ++++++++++++++++++++-
 include/linux/fscrypt.h               |  3 ++
 include/uapi/linux/fscrypt.h          | 15 ++++----
 8 files changed, 155 insertions(+), 42 deletions(-)

Comments

Dave Chinner Oct. 22, 2019, 5:27 a.m. UTC | #1
On Mon, Oct 21, 2019 at 04:03:53PM -0700, Eric Biggers wrote:
> From: Eric Biggers <ebiggers@google.com>
> 
> Some inline encryption hardware has only a small number of keyslots,
> which would make it inefficient to use the traditional fscrypt per-file
> keys.  The existing DIRECT_KEY encryption policy flag doesn't solve this
> because it assumes that file contents and names are encrypted by the
> same algorithm and that IVs are at least 24 bytes.
> 
> Therefore, add a new encryption policy flag INLINE_CRYPT_OPTIMIZED which
> causes the encryption to modified as follows:
> 
> - The key for file contents encryption is derived from the values
>   (master_key, mode_num, filesystem_uuid).  The per-file nonce is not
>   included, so many files may share the same contents encryption key.
> 
> - The IV for encrypting each block of file contents is built as
>   (inode_number << 32) | file_logical_block_num.
> 
> Including the inode number in the IVs ensures that data in different
> files is encrypted differently, despite per-file keys not being used.
> Limiting the inode and block numbers to 32 bits and putting the block
> number in the low bits is needed to be compatible with inline encryption
> hardware which only supports specifying a 64-bit data unit number which
> is auto-incremented; this is what the UFS and EMMC standards support.

These 32 bit size limits seem arbitrary and rules out implementing
this on larger filesystems. Why not just hash the 64 bit inode, file
offset and block numbers into a single 64 bit value? It is still
unique enough for the stated use (i.e. unique IV for each file
block) but it doesn't limit what filesystem configurations can
actually make use of this functionality....

Cheers,

Dave.
Eric Biggers Oct. 22, 2019, 6 a.m. UTC | #2
On Tue, Oct 22, 2019 at 04:27:12PM +1100, Dave Chinner wrote:
> On Mon, Oct 21, 2019 at 04:03:53PM -0700, Eric Biggers wrote:
> > From: Eric Biggers <ebiggers@google.com>
> > 
> > Some inline encryption hardware has only a small number of keyslots,
> > which would make it inefficient to use the traditional fscrypt per-file
> > keys.  The existing DIRECT_KEY encryption policy flag doesn't solve this
> > because it assumes that file contents and names are encrypted by the
> > same algorithm and that IVs are at least 24 bytes.
> > 
> > Therefore, add a new encryption policy flag INLINE_CRYPT_OPTIMIZED which
> > causes the encryption to modified as follows:
> > 
> > - The key for file contents encryption is derived from the values
> >   (master_key, mode_num, filesystem_uuid).  The per-file nonce is not
> >   included, so many files may share the same contents encryption key.
> > 
> > - The IV for encrypting each block of file contents is built as
> >   (inode_number << 32) | file_logical_block_num.
> > 
> > Including the inode number in the IVs ensures that data in different
> > files is encrypted differently, despite per-file keys not being used.
> > Limiting the inode and block numbers to 32 bits and putting the block
> > number in the low bits is needed to be compatible with inline encryption
> > hardware which only supports specifying a 64-bit data unit number which
> > is auto-incremented; this is what the UFS and EMMC standards support.
> 
> These 32 bit size limits seem arbitrary and rules out implementing
> this on larger filesystems. Why not just hash the 64 bit inode, file
> offset and block numbers into a single 64 bit value? It is still
> unique enough for the stated use (i.e. unique IV for each file
> block) but it doesn't limit what filesystem configurations can
> actually make use of this functionality....
> 

That won't work because we need consecutive file blocks to have consecutive IVs
as often as possible.  The crypto support in the UFS and EMMC standards takes
only a single 64-bit "data unit number" (DUN) per request, which the hardware
uses as the first 64 bits of the IV and automatically increments for each data
unit (i.e. for each filesystem block, in this case).

If every block had some random DUN, we'd have to submit a separate bio for every
single block.  And they wouldn't be mergable, so each one would cause a separate
disk request.  That would be really terrible for performance.

Also, a 64 bit hash value isn't sufficiently safe against hash collisions.

An alternative which would work nicely on ext4 and xfs (if xfs supported
fscrypt) would be to pass the physical block number as the DUN.  However, that
wouldn't work at all on f2fs because f2fs moves data blocks around.  And since
most people who want to use this are using f2fs, f2fs support is essential.

Also keep in mind that the proposed format can still be used on a specific
filesystem instance with fewer than 2^32 inodes and blocks, even if that type of
filesystem can support more inodes and blocks in general.

- Eric
Theodore Y. Ts'o Oct. 22, 2019, 1:30 p.m. UTC | #3
On Mon, Oct 21, 2019 at 11:00:04PM -0700, Eric Biggers wrote:
> That won't work because we need consecutive file blocks to have consecutive IVs
> as often as possible.  The crypto support in the UFS and EMMC standards takes
> only a single 64-bit "data unit number" (DUN) per request, which the hardware
> uses as the first 64 bits of the IV and automatically increments for each data
> unit (i.e. for each filesystem block, in this case).

It seems very likely that for systems that are using UFS and eMMC
(which are overwhelming lower-end devices --- e.g., embedded and
mobile handsets) 32-bit inode and logical block numbers will be just
fine.

If and when we actually get inline crypto support for server-class
systems, hopefully they will support 128-bit DUN's, and/or they will
have sufficiently fast key load times such that we can use per-file
keying.

> An alternative which would work nicely on ext4 and xfs (if xfs supported
> fscrypt) would be to pass the physical block number as the DUN.  However, that
> wouldn't work at all on f2fs because f2fs moves data blocks around.  And since
> most people who want to use this are using f2fs, f2fs support is essential.

And that is something fscrypt supports already, so if people really
did want to use 64-bit logical block numbers, they could do that, at
the cost of giving up the ability to shrink the file system (which XFS
doesn't support anyway....)

							- Ted
Eric Biggers Oct. 22, 2019, 4:15 p.m. UTC | #4
On Tue, Oct 22, 2019 at 09:30:01AM -0400, Theodore Y. Ts'o wrote:
> > An alternative which would work nicely on ext4 and xfs (if xfs supported
> > fscrypt) would be to pass the physical block number as the DUN.  However, that
> > wouldn't work at all on f2fs because f2fs moves data blocks around.  And since
> > most people who want to use this are using f2fs, f2fs support is essential.
> 
> And that is something fscrypt supports already, so if people really
> did want to use 64-bit logical block numbers, they could do that, at
> the cost of giving up the ability to shrink the file system (which XFS
> doesn't support anyway....)

I was talking about the physical block number (offset from the start of the
filesystem -- ext4_fsblk_t on ext4), not the file logical block number (offset
in the file data -- ext4_lblk_t on ext4).  fscrypt doesn't currently support
using the physical block number.

- Eric
Christoph Hellwig Oct. 23, 2019, 9:27 a.m. UTC | #5
On Tue, Oct 22, 2019 at 09:30:01AM -0400, Theodore Y. Ts'o wrote:
> If and when we actually get inline crypto support for server-class
> systems, hopefully they will support 128-bit DUN's, and/or they will
> have sufficiently fast key load times such that we can use per-file
> keying.

NVMe is working on a key per I/O feature.  So at very least the naming
of this option should be "crappy_underwhelming_embedded_inline_crypto"
Christoph Hellwig Oct. 23, 2019, 9:28 a.m. UTC | #6
On Mon, Oct 21, 2019 at 11:00:04PM -0700, Eric Biggers wrote:
> An alternative which would work nicely on ext4 and xfs (if xfs supported
> fscrypt) would be to pass the physical block number as the DUN.  However, that
> wouldn't work at all on f2fs because f2fs moves data blocks around.

XFS can also move data blocks around.  Even ext4 can do that for limited
cases (defrag).
Theodore Y. Ts'o Oct. 23, 2019, 12:57 p.m. UTC | #7
On Wed, Oct 23, 2019 at 02:27:18AM -0700, Christoph Hellwig wrote:
> On Tue, Oct 22, 2019 at 09:30:01AM -0400, Theodore Y. Ts'o wrote:
> > If and when we actually get inline crypto support for server-class
> > systems, hopefully they will support 128-bit DUN's, and/or they will
> > have sufficiently fast key load times such that we can use per-file
> > keying.
> 
> NVMe is working on a key per I/O feature.  So at very least the naming
> of this option should be "crappy_underwhelming_embedded_inline_crypto"

If and when the vaporware shows up in real hardware, and assuming that
fscrypt is useful for this hardware, we can name it
"super_duper_fancy_inline_crypto".  :-)

Remember that fscrypt only encrypts the data and the file name.  It
doesn't encrypt the metadata.  It has very specific use cases for
Android and ChromeOS where you have multiple users that need to use
different keys, and in the case of ChromeOS, we want to be able to
efficiently use the space so that while user A is logged in, we can
delete files in user B's cache directory without user B's keys being
present.  (This is why we can't use fixed per-user partitions with
dm-crypt; that solution was considered and rejected before we started
work on fscrypt.)

If you aren't working under tight space and cost constraints, it's
actually better to encrypt the whole partition, so that all of the
metadata can be protected.  fscrypt is deployed in millions and
millions of devices, and is solving real world problems.  However, it
never claimed to be the only way to address encryption in the storage
stack --- and it's not at all clear fscrypt is the way that makes the
most amount of sense for NVMe devices.  So let's cross that bridge
when we get to it.

Cheers,

	       	   	      	       	      - Ted
Christoph Hellwig Oct. 24, 2019, 1:27 a.m. UTC | #8
> If and when the vaporware shows up in real hardware, and assuming that
> fscrypt is useful for this hardware, we can name it
> "super_duper_fancy_inline_crypto".  :-)

I think you are entirely missing the point.  The point is that naming
the option someting related to inline encryption is fundamentally
wrong.  It is related to a limitation of existing inline crypto
engines, not related to the fudamental model.  And all the other
rambling below don't matter either.
Eric Biggers Oct. 24, 2019, 2:44 a.m. UTC | #9
On Wed, Oct 23, 2019 at 06:27:59PM -0700, Christoph Hellwig wrote:
> > If and when the vaporware shows up in real hardware, and assuming that
> > fscrypt is useful for this hardware, we can name it
> > "super_duper_fancy_inline_crypto".  :-)
> 
> I think you are entirely missing the point.  The point is that naming
> the option someting related to inline encryption is fundamentally
> wrong.  It is related to a limitation of existing inline crypto
> engines, not related to the fudamental model.  And all the other
> rambling below don't matter either.
> 

Would you be happy with something that more directly describes the change the
flag makes, like FSCRYPT_POLICY_FLAG_CONTENTS_IV_INO_LBLK_64?  I.e., the IVs for
contents encryption are 64-bit and contain the inode and logical block numbers.

Actually, we could use the same key derivation and IV generation for directories
and symlinks too, which would result in just FSCRYPT_POLICY_FLAG_IV_INO_LBLK_64.
(lblk is 0 when encrypting a filename.)

Although, in general it would be nice to name the settings in ways that are
easier for people not intimately familiar with the crypto to understand...

- Eric
Christoph Hellwig Oct. 24, 2019, 7:04 a.m. UTC | #10
On Wed, Oct 23, 2019 at 07:44:59PM -0700, Eric Biggers wrote:
> Would you be happy with something that more directly describes the change the
> flag makes

Yes.

> , like FSCRYPT_POLICY_FLAG_CONTENTS_IV_INO_LBLK_64?  I.e., the IVs for
> contents encryption are 64-bit and contain the inode and logical block numbers.
> 
> Actually, we could use the same key derivation and IV generation for directories
> and symlinks too, which would result in just FSCRYPT_POLICY_FLAG_IV_INO_LBLK_64.
> (lblk is 0 when encrypting a filename.)

I think not making it crazy verbose is a helpful, but at the same time
it should be somewhat descriptive.

> Although, in general it would be nice to name the settings in ways that are
> easier for people not intimately familiar with the crypto to understand...

For the andoid case the actual users won't ever really see it, and if
you set up the thing yourself it probably helps a lot to try to
understand what your are doing.
Paul Crowley Oct. 24, 2019, 9:54 a.m. UTC | #11
On Thu, 24 Oct 2019 at 00:04, Christoph Hellwig <hch@infradead.org> wrote:
> I think not making it crazy verbose is a helpful, but at the same time
> it should be somewhat descriptive.

What would your suggested name be?

Patch
diff mbox series

diff --git a/Documentation/filesystems/fscrypt.rst b/Documentation/filesystems/fscrypt.rst
index 6ec459be3de16..d059185f79ba9 100644
--- a/Documentation/filesystems/fscrypt.rst
+++ b/Documentation/filesystems/fscrypt.rst
@@ -257,12 +257,13 @@  the master keys may be wrapped in userspace, e.g. as is done by the
 `fscrypt <https://github.com/google/fscrypt>`_ tool.
 
 Including the inode number in the IVs was considered.  However, it was
-rejected as it would have prevented ext4 filesystems from being
-resized, and by itself still wouldn't have been sufficient to prevent
-the same key from being directly reused for both XTS and CTS-CBC.
+rejected (except for INLINE_CRYPT_OPTIMIZED policies) as it would have
+prevented ext4 filesystems from being resized, and by itself still
+wouldn't have been sufficient to prevent the same key from being
+directly reused for both XTS and CTS-CBC.
 
-DIRECT_KEY and per-mode keys
-----------------------------
+DIRECT_KEY policies
+-------------------
 
 The Adiantum encryption mode (see `Encryption modes and usage`_) is
 suitable for both contents and filenames encryption, and it accepts
@@ -285,6 +286,22 @@  IV.  Moreover:
   key derived using the KDF.  Users may use the same master key for
   other v2 encryption policies.
 
+INLINE_CRYPT_OPTIMIZED policies
+-------------------------------
+
+When FSCRYPT_POLICY_FLAG_INLINE_CRYPT_OPTIMIZED is set in an
+encryption policy, file contents are encrypted using a key derived
+from the master key, encryption mode number, and filesystem UUID.  In
+this case, many files may share the same key.  To still encrypt
+different files' data differently, inode numbers are included in the
+IVs.  Consequently, shrinking the filesystem may not be allowed.
+
+This format is optimized for use with inline encryption hardware that
+supports only a small number of keyslots and allows specifying only 8
+IV bytes, e.g. the crypto capabilities in the UFSHCI v2.1+ standard.
+
+This flag has no effect on filenames encryption.
+
 Key identifiers
 ---------------
 
@@ -342,11 +359,14 @@  a little endian number, except that:
   is encrypted with AES-256 where the AES-256 key is the SHA-256 hash
   of the file's data encryption key.
 
-- In the "direct key" configuration (FSCRYPT_POLICY_FLAG_DIRECT_KEY
-  set in the fscrypt_policy), the file's nonce is also appended to the
-  IV.  Currently this is only allowed with the Adiantum encryption
+- With `DIRECT_KEY policies`_, the file's nonce is also appended to
+  the IV.  Currently this is only allowed with the Adiantum encryption
   mode.
 
+- With `INLINE_CRYPT_OPTIMIZED policies`_, the block number is limited
+  to 32 bits and is placed in bits 0-31 of the IV.  The inode number
+  (which is also limited to 32 bits) is placed in bits 32-63.
+
 Filenames encryption
 --------------------
 
@@ -432,12 +452,15 @@  This structure must be initialized as follows:
   (1) for ``contents_encryption_mode`` and FSCRYPT_MODE_AES_256_CTS
   (4) for ``filenames_encryption_mode``.
 
-- ``flags`` must contain a value from ``<linux/fscrypt.h>`` which
-  identifies the amount of NUL-padding to use when encrypting
-  filenames.  If unsure, use FSCRYPT_POLICY_FLAGS_PAD_32 (0x3).
-  Additionally, if the encryption modes are both
-  FSCRYPT_MODE_ADIANTUM, this can contain
-  FSCRYPT_POLICY_FLAG_DIRECT_KEY; see `DIRECT_KEY and per-mode keys`_.
+- ``flags`` contains optional flags from ``<linux/fscrypt.h>``:
+
+  - FSCRYPT_POLICY_FLAGS_PAD_*: The amount of NUL padding to use when
+    encrypting filenames.  If unsure, use FSCRYPT_POLICY_FLAGS_PAD_32
+    (0x3).
+  - FSCRYPT_POLICY_FLAG_DIRECT_KEY: See `DIRECT_KEY policies`_.
+  - FSCRYPT_POLICY_FLAG_INLINE_CRYPT_OPTIMIZED: See
+    `INLINE_CRYPT_OPTIMIZED policies`_.  This is mutually exclusive
+    with DIRECT_KEY and is not supported on v1 policies.
 
 - For v2 encryption policies, ``__reserved`` must be zeroed.
 
@@ -1090,7 +1113,7 @@  policy structs (see `Setting an encryption policy`_), except that the
 context structs also contain a nonce.  The nonce is randomly generated
 by the kernel and is used as KDF input or as a tweak to cause
 different files to be encrypted differently; see `Per-file keys`_ and
-`DIRECT_KEY and per-mode keys`_.
+`DIRECT_KEY policies`_.
 
 Data path changes
 -----------------
diff --git a/fs/crypto/crypto.c b/fs/crypto/crypto.c
index ced8ad9f2d019..6be16837accdd 100644
--- a/fs/crypto/crypto.c
+++ b/fs/crypto/crypto.c
@@ -73,11 +73,18 @@  EXPORT_SYMBOL(fscrypt_free_bounce_page);
 void fscrypt_generate_iv(union fscrypt_iv *iv, u64 lblk_num,
 			 const struct fscrypt_info *ci)
 {
+	u8 flags = fscrypt_policy_flags(&ci->ci_policy);
+
 	memset(iv, 0, ci->ci_mode->ivsize);
-	iv->lblk_num = cpu_to_le64(lblk_num);
 
-	if (fscrypt_is_direct_key_policy(&ci->ci_policy))
+	if ((flags & FSCRYPT_POLICY_FLAG_INLINE_CRYPT_OPTIMIZED) &&
+	    S_ISREG(ci->ci_inode->i_mode)) {
+		WARN_ON_ONCE((u32)lblk_num != lblk_num);
+		lblk_num |= (u64)ci->ci_inode->i_ino << 32;
+	} else if (flags & FSCRYPT_POLICY_FLAG_DIRECT_KEY) {
 		memcpy(iv->nonce, ci->ci_nonce, FS_KEY_DERIVATION_NONCE_SIZE);
+	}
+	iv->lblk_num = cpu_to_le64(lblk_num);
 }
 
 /* Encrypt or decrypt a single filesystem block of file contents */
diff --git a/fs/crypto/fscrypt_private.h b/fs/crypto/fscrypt_private.h
index dacf8fcbac3be..3428488612b84 100644
--- a/fs/crypto/fscrypt_private.h
+++ b/fs/crypto/fscrypt_private.h
@@ -163,6 +163,9 @@  struct fscrypt_info {
 	/* The actual crypto transform used for encryption and decryption */
 	struct crypto_skcipher *ci_ctfm;
 
+	/* True if the key should be freed when this fscrypt_info is freed */
+	bool ci_owns_key;
+
 	/*
 	 * Encryption mode used for this inode.  It corresponds to either the
 	 * contents or filenames encryption mode, depending on the inode type.
@@ -279,9 +282,10 @@  extern int fscrypt_init_hkdf(struct fscrypt_hkdf *hkdf, const u8 *master_key,
  * outputs are unique and cryptographically isolated, i.e. knowledge of one
  * output doesn't reveal another.
  */
-#define HKDF_CONTEXT_KEY_IDENTIFIER	1
-#define HKDF_CONTEXT_PER_FILE_KEY	2
-#define HKDF_CONTEXT_PER_MODE_KEY	3
+#define HKDF_CONTEXT_KEY_IDENTIFIER		1
+#define HKDF_CONTEXT_PER_FILE_KEY		2
+#define HKDF_CONTEXT_DIRECT_KEY			3
+#define HKDF_CONTEXT_INLINE_CRYPT_OPTIMIZED_KEY	4
 
 extern int fscrypt_hkdf_expand(struct fscrypt_hkdf *hkdf, u8 context,
 			       const u8 *info, unsigned int infolen,
@@ -378,8 +382,14 @@  struct fscrypt_master_key {
 	struct list_head	mk_decrypted_inodes;
 	spinlock_t		mk_decrypted_inodes_lock;
 
-	/* Per-mode tfms for DIRECT_KEY policies, allocated on-demand */
-	struct crypto_skcipher	*mk_mode_keys[__FSCRYPT_MODE_MAX + 1];
+	/* Crypto API transforms for DIRECT_KEY policies, allocated on-demand */
+	struct crypto_skcipher	*mk_direct_tfms[__FSCRYPT_MODE_MAX + 1];
+
+	/*
+	 * Crypto API transforms for filesystem-layer implementation of
+	 * INLINE_CRYPT_OPTIMIZED policies, allocated on-demand.
+	 */
+	struct crypto_skcipher	*mk_inline_crypt_optimized_tfms[__FSCRYPT_MODE_MAX + 1];
 
 } __randomize_layout;
 
diff --git a/fs/crypto/keyring.c b/fs/crypto/keyring.c
index c34fa7c61b43b..6ca3364b2c968 100644
--- a/fs/crypto/keyring.c
+++ b/fs/crypto/keyring.c
@@ -43,8 +43,10 @@  static void free_master_key(struct fscrypt_master_key *mk)
 
 	wipe_master_key_secret(&mk->mk_secret);
 
-	for (i = 0; i < ARRAY_SIZE(mk->mk_mode_keys); i++)
-		crypto_free_skcipher(mk->mk_mode_keys[i]);
+	for (i = 0; i <= __FSCRYPT_MODE_MAX; i++) {
+		crypto_free_skcipher(mk->mk_direct_tfms[i]);
+		crypto_free_skcipher(mk->mk_inline_crypt_optimized_tfms[i]);
+	}
 
 	key_put(mk->mk_users);
 	kzfree(mk);
diff --git a/fs/crypto/keysetup.c b/fs/crypto/keysetup.c
index b03b33643e4b2..79af9c2713057 100644
--- a/fs/crypto/keysetup.c
+++ b/fs/crypto/keysetup.c
@@ -116,40 +116,54 @@  int fscrypt_set_derived_key(struct fscrypt_info *ci, const u8 *derived_key)
 		return PTR_ERR(tfm);
 
 	ci->ci_ctfm = tfm;
+	ci->ci_owns_key = true;
 	return 0;
 }
 
 static int setup_per_mode_key(struct fscrypt_info *ci,
-			      struct fscrypt_master_key *mk)
+			      struct fscrypt_master_key *mk,
+			      struct crypto_skcipher **tfms,
+			      u8 hkdf_context, bool include_fs_uuid)
 {
+	const struct inode *inode = ci->ci_inode;
+	const struct super_block *sb = inode->i_sb;
 	struct fscrypt_mode *mode = ci->ci_mode;
 	u8 mode_num = mode - available_modes;
 	struct crypto_skcipher *tfm, *prev_tfm;
 	u8 mode_key[FSCRYPT_MAX_KEY_SIZE];
+	u8 hkdf_info[sizeof(mode_num) + sizeof(sb->s_uuid)];
+	unsigned int hkdf_infolen = 0;
 	int err;
 
-	if (WARN_ON(mode_num >= ARRAY_SIZE(mk->mk_mode_keys)))
+	if (WARN_ON(mode_num > __FSCRYPT_MODE_MAX))
 		return -EINVAL;
 
 	/* pairs with cmpxchg() below */
-	tfm = READ_ONCE(mk->mk_mode_keys[mode_num]);
+	tfm = READ_ONCE(tfms[mode_num]);
 	if (likely(tfm != NULL))
 		goto done;
 
 	BUILD_BUG_ON(sizeof(mode_num) != 1);
+	BUILD_BUG_ON(sizeof(sb->s_uuid) != 16);
+	BUILD_BUG_ON(sizeof(hkdf_info) != 17);
+	hkdf_info[hkdf_infolen++] = mode_num;
+	if (include_fs_uuid) {
+		memcpy(&hkdf_info[hkdf_infolen], &sb->s_uuid,
+		       sizeof(sb->s_uuid));
+		hkdf_infolen += sizeof(sb->s_uuid);
+	}
 	err = fscrypt_hkdf_expand(&mk->mk_secret.hkdf,
-				  HKDF_CONTEXT_PER_MODE_KEY,
-				  &mode_num, sizeof(mode_num),
+				  hkdf_context, hkdf_info, hkdf_infolen,
 				  mode_key, mode->keysize);
 	if (err)
 		return err;
-	tfm = fscrypt_allocate_skcipher(mode, mode_key, ci->ci_inode);
+	tfm = fscrypt_allocate_skcipher(mode, mode_key, inode);
 	memzero_explicit(mode_key, mode->keysize);
 	if (IS_ERR(tfm))
 		return PTR_ERR(tfm);
 
 	/* pairs with READ_ONCE() above */
-	prev_tfm = cmpxchg(&mk->mk_mode_keys[mode_num], NULL, tfm);
+	prev_tfm = cmpxchg(&tfms[mode_num], NULL, tfm);
 	if (prev_tfm != NULL) {
 		crypto_free_skcipher(tfm);
 		tfm = prev_tfm;
@@ -180,7 +194,21 @@  static int fscrypt_setup_v2_file_key(struct fscrypt_info *ci,
 				     ci->ci_mode->friendly_name);
 			return -EINVAL;
 		}
-		return setup_per_mode_key(ci, mk);
+		return setup_per_mode_key(ci, mk, mk->mk_direct_tfms,
+					  HKDF_CONTEXT_DIRECT_KEY, false);
+	} else if ((ci->ci_policy.v2.flags &
+		    FSCRYPT_POLICY_FLAG_INLINE_CRYPT_OPTIMIZED) &&
+		   S_ISREG(ci->ci_inode->i_mode)) {
+		/*
+		 * INLINE_CRYPT_OPTIMIZED: file contents keys are derived from
+		 * (master_key, mode_num, filesystem_uuid), and inode number is
+		 * included in the IVs.  This format is optimized for use with
+		 * inline encryption hardware.
+		 */
+		return setup_per_mode_key(ci, mk,
+					  mk->mk_inline_crypt_optimized_tfms,
+					  HKDF_CONTEXT_INLINE_CRYPT_OPTIMIZED_KEY,
+					  true);
 	}
 
 	err = fscrypt_hkdf_expand(&mk->mk_secret.hkdf,
@@ -304,8 +332,7 @@  static void put_crypt_info(struct fscrypt_info *ci)
 
 	if (ci->ci_direct_key)
 		fscrypt_put_direct_key(ci->ci_direct_key);
-	else if (ci->ci_ctfm != NULL &&
-		 !fscrypt_is_direct_key_policy(&ci->ci_policy))
+	else if (ci->ci_owns_key)
 		crypto_free_skcipher(ci->ci_ctfm);
 
 	key = ci->ci_master_key;
diff --git a/fs/crypto/policy.c b/fs/crypto/policy.c
index 4072ba644595b..fddd90db2d787 100644
--- a/fs/crypto/policy.c
+++ b/fs/crypto/policy.c
@@ -29,6 +29,41 @@  bool fscrypt_policies_equal(const union fscrypt_policy *policy1,
 	return !memcmp(policy1, policy2, fscrypt_policy_size(policy1));
 }
 
+static bool supported_inline_crypt_optimized_policy(
+					const struct fscrypt_policy_v2 *policy,
+					const struct inode *inode)
+{
+	struct super_block *sb = inode->i_sb;
+	int ino_bits = 64, lblk_bits = 64;
+
+	if (policy->flags & FSCRYPT_POLICY_FLAG_DIRECT_KEY) {
+		fscrypt_warn(inode,
+			     "The DIRECT_KEY and INLINE_CRYPT_OPTIMIZED flags are mutually exclusive");
+		return false;
+	}
+	/*
+	 * Since INLINE_CRYPT_OPTIMIZED policies include the inode number in the
+	 * IVs, it must not be allowed for the filesystem to renumber the
+	 * inodes, e.g. via filesystem shrinking.
+	 */
+	if (!sb->s_cop->has_stable_inodes ||
+	    !sb->s_cop->has_stable_inodes(sb)) {
+		fscrypt_warn(inode,
+			     "Can't use INLINE_CRYPT_OPTIMIZED policy on filesystem '%s' because it doesn't have stable inode numbers",
+			     sb->s_id);
+		return false;
+	}
+	if (sb->s_cop->get_ino_and_lblk_bits)
+		sb->s_cop->get_ino_and_lblk_bits(sb, &ino_bits, &lblk_bits);
+	if (ino_bits > 32 || lblk_bits > 32) {
+		fscrypt_warn(inode,
+			     "Can't use INLINE_CRYPT_OPTIMIZED policy on filesystem '%s' because it doesn't use 32-bit inode and block numbers",
+			     sb->s_id);
+		return false;
+	}
+	return true;
+}
+
 /**
  * fscrypt_supported_policy - check whether an encryption policy is supported
  *
@@ -55,7 +90,8 @@  bool fscrypt_supported_policy(const union fscrypt_policy *policy_u,
 			return false;
 		}
 
-		if (policy->flags & ~FSCRYPT_POLICY_FLAGS_VALID) {
+		if (policy->flags & ~(FSCRYPT_POLICY_FLAGS_PAD_MASK |
+				      FSCRYPT_POLICY_FLAG_DIRECT_KEY)) {
 			fscrypt_warn(inode,
 				     "Unsupported encryption flags (0x%02x)",
 				     policy->flags);
@@ -83,6 +119,10 @@  bool fscrypt_supported_policy(const union fscrypt_policy *policy_u,
 			return false;
 		}
 
+		if ((policy->flags & FSCRYPT_POLICY_FLAG_INLINE_CRYPT_OPTIMIZED)
+		    && !supported_inline_crypt_optimized_policy(policy, inode))
+			return false;
+
 		if (memchr_inv(policy->__reserved, 0,
 			       sizeof(policy->__reserved))) {
 			fscrypt_warn(inode,
diff --git a/include/linux/fscrypt.h b/include/linux/fscrypt.h
index 04f5ed6284454..1a7bffe78ed56 100644
--- a/include/linux/fscrypt.h
+++ b/include/linux/fscrypt.h
@@ -61,6 +61,9 @@  struct fscrypt_operations {
 	bool (*dummy_context)(struct inode *);
 	bool (*empty_dir)(struct inode *);
 	unsigned int max_namelen;
+	bool (*has_stable_inodes)(struct super_block *sb);
+	void (*get_ino_and_lblk_bits)(struct super_block *sb,
+				      int *ino_bits_ret, int *lblk_bits_ret);
 };
 
 static inline bool fscrypt_has_encryption_key(const struct inode *inode)
diff --git a/include/uapi/linux/fscrypt.h b/include/uapi/linux/fscrypt.h
index 39ccfe9311c38..6a8fd8ff2d2ac 100644
--- a/include/uapi/linux/fscrypt.h
+++ b/include/uapi/linux/fscrypt.h
@@ -11,13 +11,14 @@ 
 #include <linux/types.h>
 
 /* Encryption policy flags */
-#define FSCRYPT_POLICY_FLAGS_PAD_4		0x00
-#define FSCRYPT_POLICY_FLAGS_PAD_8		0x01
-#define FSCRYPT_POLICY_FLAGS_PAD_16		0x02
-#define FSCRYPT_POLICY_FLAGS_PAD_32		0x03
-#define FSCRYPT_POLICY_FLAGS_PAD_MASK		0x03
-#define FSCRYPT_POLICY_FLAG_DIRECT_KEY		0x04
-#define FSCRYPT_POLICY_FLAGS_VALID		0x07
+#define FSCRYPT_POLICY_FLAGS_PAD_4			0x00
+#define FSCRYPT_POLICY_FLAGS_PAD_8			0x01
+#define FSCRYPT_POLICY_FLAGS_PAD_16			0x02
+#define FSCRYPT_POLICY_FLAGS_PAD_32			0x03
+#define FSCRYPT_POLICY_FLAGS_PAD_MASK			0x03
+#define FSCRYPT_POLICY_FLAG_DIRECT_KEY			0x04
+#define FSCRYPT_POLICY_FLAG_INLINE_CRYPT_OPTIMIZED	0x08
+#define FSCRYPT_POLICY_FLAGS_VALID			0x0F
 
 /* Encryption algorithms */
 #define FSCRYPT_MODE_AES_256_XTS		1