1.. SPDX-License-Identifier: GPL-2.0
2
3.. _inline_encryption:
4
5=================
6Inline Encryption
7=================
8
9Background
10==========
11
12Inline encryption hardware sits logically between memory and disk, and can
13en/decrypt data as it goes in/out of the disk.  For each I/O request, software
14can control exactly how the inline encryption hardware will en/decrypt the data
15in terms of key, algorithm, data unit size (the granularity of en/decryption),
16and data unit number (a value that determines the initialization vector(s)).
17
18Some inline encryption hardware accepts all encryption parameters including raw
19keys directly in low-level I/O requests.  However, most inline encryption
20hardware instead has a fixed number of "keyslots" and requires that the key,
21algorithm, and data unit size first be programmed into a keyslot.  Each
22low-level I/O request then just contains a keyslot index and data unit number.
23
24Note that inline encryption hardware is very different from traditional crypto
25accelerators, which are supported through the kernel crypto API.  Traditional
26crypto accelerators operate on memory regions, whereas inline encryption
27hardware operates on I/O requests.  Thus, inline encryption hardware needs to be
28managed by the block layer, not the kernel crypto API.
29
30Inline encryption hardware is also very different from "self-encrypting drives",
31such as those based on the TCG Opal or ATA Security standards.  Self-encrypting
32drives don't provide fine-grained control of encryption and provide no way to
33verify the correctness of the resulting ciphertext.  Inline encryption hardware
34provides fine-grained control of encryption, including the choice of key and
35initialization vector for each sector, and can be tested for correctness.
36
37Objective
38=========
39
40We want to support inline encryption in the kernel.  To make testing easier, we
41also want support for falling back to the kernel crypto API when actual inline
42encryption hardware is absent.  We also want inline encryption to work with
43layered devices like device-mapper and loopback (i.e. we want to be able to use
44the inline encryption hardware of the underlying devices if present, or else
45fall back to crypto API en/decryption).
46
47Constraints and notes
48=====================
49
50- We need a way for upper layers (e.g. filesystems) to specify an encryption
51  context to use for en/decrypting a bio, and device drivers (e.g. UFSHCD) need
52  to be able to use that encryption context when they process the request.
53  Encryption contexts also introduce constraints on bio merging; the block layer
54  needs to be aware of these constraints.
55
56- Different inline encryption hardware has different supported algorithms,
57  supported data unit sizes, maximum data unit numbers, etc.  We call these
58  properties the "crypto capabilities".  We need a way for device drivers to
59  advertise crypto capabilities to upper layers in a generic way.
60
61- Inline encryption hardware usually (but not always) requires that keys be
62  programmed into keyslots before being used.  Since programming keyslots may be
63  slow and there may not be very many keyslots, we shouldn't just program the
64  key for every I/O request, but rather keep track of which keys are in the
65  keyslots and reuse an already-programmed keyslot when possible.
66
67- Upper layers typically define a specific end-of-life for crypto keys, e.g.
68  when an encrypted directory is locked or when a crypto mapping is torn down.
69  At these times, keys are wiped from memory.  We must provide a way for upper
70  layers to also evict keys from any keyslots they are present in.
71
72- When possible, device-mapper devices must be able to pass through the inline
73  encryption support of their underlying devices.  However, it doesn't make
74  sense for device-mapper devices to have keyslots themselves.
75
76Basic design
77============
78
79We introduce ``struct blk_crypto_key`` to represent an inline encryption key and
80how it will be used.  This includes the actual bytes of the key; the size of the
81key; the algorithm and data unit size the key will be used with; and the number
82of bytes needed to represent the maximum data unit number the key will be used
83with.
84
85We introduce ``struct bio_crypt_ctx`` to represent an encryption context.  It
86contains a data unit number and a pointer to a blk_crypto_key.  We add pointers
87to a bio_crypt_ctx to ``struct bio`` and ``struct request``; this allows users
88of the block layer (e.g. filesystems) to provide an encryption context when
89creating a bio and have it be passed down the stack for processing by the block
90layer and device drivers.  Note that the encryption context doesn't explicitly
91say whether to encrypt or decrypt, as that is implicit from the direction of the
92bio; WRITE means encrypt, and READ means decrypt.
93
94We also introduce ``struct blk_crypto_profile`` to contain all generic inline
95encryption-related state for a particular inline encryption device.  The
96blk_crypto_profile serves as the way that drivers for inline encryption hardware
97advertise their crypto capabilities and provide certain functions (e.g.,
98functions to program and evict keys) to upper layers.  Each device driver that
99wants to support inline encryption will construct a blk_crypto_profile, then
100associate it with the disk's request_queue.
101
102The blk_crypto_profile also manages the hardware's keyslots, when applicable.
103This happens in the block layer, so that users of the block layer can just
104specify encryption contexts and don't need to know about keyslots at all, nor do
105device drivers need to care about most details of keyslot management.
106
107Specifically, for each keyslot, the block layer (via the blk_crypto_profile)
108keeps track of which blk_crypto_key that keyslot contains (if any), and how many
109in-flight I/O requests are using it.  When the block layer creates a
110``struct request`` for a bio that has an encryption context, it grabs a keyslot
111that already contains the key if possible.  Otherwise it waits for an idle
112keyslot (a keyslot that isn't in-use by any I/O), then programs the key into the
113least-recently-used idle keyslot using the function the device driver provided.
114In both cases, the resulting keyslot is stored in the ``crypt_keyslot`` field of
115the request, where it is then accessible to device drivers and is released after
116the request completes.
117
118``struct request`` also contains a pointer to the original bio_crypt_ctx.
119Requests can be built from multiple bios, and the block layer must take the
120encryption context into account when trying to merge bios and requests.  For two
121bios/requests to be merged, they must have compatible encryption contexts: both
122unencrypted, or both encrypted with the same key and contiguous data unit
123numbers.  Only the encryption context for the first bio in a request is
124retained, since the remaining bios have been verified to be merge-compatible
125with the first bio.
126
127To make it possible for inline encryption to work with request_queue based
128layered devices, when a request is cloned, its encryption context is cloned as
129well.  When the cloned request is submitted, it is then processed as usual; this
130includes getting a keyslot from the clone's target device if needed.
131
132blk-crypto-fallback
133===================
134
135It is desirable for the inline encryption support of upper layers (e.g.
136filesystems) to be testable without real inline encryption hardware, and
137likewise for the block layer's keyslot management logic.  It is also desirable
138to allow upper layers to just always use inline encryption rather than have to
139implement encryption in multiple ways.
140
141Therefore, we also introduce *blk-crypto-fallback*, which is an implementation
142of inline encryption using the kernel crypto API.  blk-crypto-fallback is built
143into the block layer, so it works on any block device without any special setup.
144Essentially, when a bio with an encryption context is submitted to a
145block_device that doesn't support that encryption context, the block layer will
146handle en/decryption of the bio using blk-crypto-fallback.
147
148For encryption, the data cannot be encrypted in-place, as callers usually rely
149on it being unmodified.  Instead, blk-crypto-fallback allocates bounce pages,
150fills a new bio with those bounce pages, encrypts the data into those bounce
151pages, and submits that "bounce" bio.  When the bounce bio completes,
152blk-crypto-fallback completes the original bio.  If the original bio is too
153large, multiple bounce bios may be required; see the code for details.
154
155For decryption, blk-crypto-fallback "wraps" the bio's completion callback
156(``bi_complete``) and private data (``bi_private``) with its own, unsets the
157bio's encryption context, then submits the bio.  If the read completes
158successfully, blk-crypto-fallback restores the bio's original completion
159callback and private data, then decrypts the bio's data in-place using the
160kernel crypto API.  Decryption happens from a workqueue, as it may sleep.
161Afterwards, blk-crypto-fallback completes the bio.
162
163In both cases, the bios that blk-crypto-fallback submits no longer have an
164encryption context.  Therefore, lower layers only see standard unencrypted I/O.
165
166blk-crypto-fallback also defines its own blk_crypto_profile and has its own
167"keyslots"; its keyslots contain ``struct crypto_skcipher`` objects.  The reason
168for this is twofold.  First, it allows the keyslot management logic to be tested
169without actual inline encryption hardware.  Second, similar to actual inline
170encryption hardware, the crypto API doesn't accept keys directly in requests but
171rather requires that keys be set ahead of time, and setting keys can be
172expensive; moreover, allocating a crypto_skcipher can't happen on the I/O path
173at all due to the locks it takes.  Therefore, the concept of keyslots still
174makes sense for blk-crypto-fallback.
175
176Note that regardless of whether real inline encryption hardware or
177blk-crypto-fallback is used, the ciphertext written to disk (and hence the
178on-disk format of data) will be the same (assuming that both the inline
179encryption hardware's implementation and the kernel crypto API's implementation
180of the algorithm being used adhere to spec and function correctly).
181
182blk-crypto-fallback is optional and is controlled by the
183``CONFIG_BLK_INLINE_ENCRYPTION_FALLBACK`` kernel configuration option.
184
185API presented to users of the block layer
186=========================================
187
188``blk_crypto_config_supported()`` allows users to check ahead of time whether
189inline encryption with particular crypto settings will work on a particular
190block_device -- either via hardware or via blk-crypto-fallback.  This function
191takes in a ``struct blk_crypto_config`` which is like blk_crypto_key, but omits
192the actual bytes of the key and instead just contains the algorithm, data unit
193size, etc.  This function can be useful if blk-crypto-fallback is disabled.
194
195``blk_crypto_init_key()`` allows users to initialize a blk_crypto_key.
196
197Users must call ``blk_crypto_start_using_key()`` before actually starting to use
198a blk_crypto_key on a block_device (even if ``blk_crypto_config_supported()``
199was called earlier).  This is needed to initialize blk-crypto-fallback if it
200will be needed.  This must not be called from the data path, as this may have to
201allocate resources, which may deadlock in that case.
202
203Next, to attach an encryption context to a bio, users should call
204``bio_crypt_set_ctx()``.  This function allocates a bio_crypt_ctx and attaches
205it to a bio, given the blk_crypto_key and the data unit number that will be used
206for en/decryption.  Users don't need to worry about freeing the bio_crypt_ctx
207later, as that happens automatically when the bio is freed or reset.
208
209Finally, when done using inline encryption with a blk_crypto_key on a
210block_device, users must call ``blk_crypto_evict_key()``.  This ensures that
211the key is evicted from all keyslots it may be programmed into and unlinked from
212any kernel data structures it may be linked into.
213
214In summary, for users of the block layer, the lifecycle of a blk_crypto_key is
215as follows:
216
2171. ``blk_crypto_config_supported()`` (optional)
2182. ``blk_crypto_init_key()``
2193. ``blk_crypto_start_using_key()``
2204. ``bio_crypt_set_ctx()`` (potentially many times)
2215. ``blk_crypto_evict_key()`` (after all I/O has completed)
2226. Zeroize the blk_crypto_key (this has no dedicated function)
223
224If a blk_crypto_key is being used on multiple block_devices, then
225``blk_crypto_config_supported()`` (if used), ``blk_crypto_start_using_key()``,
226and ``blk_crypto_evict_key()`` must be called on each block_device.
227
228API presented to device drivers
229===============================
230
231A device driver that wants to support inline encryption must set up a
232blk_crypto_profile in the request_queue of its device.  To do this, it first
233must call ``blk_crypto_profile_init()`` (or its resource-managed variant
234``devm_blk_crypto_profile_init()``), providing the number of keyslots.
235
236Next, it must advertise its crypto capabilities by setting fields in the
237blk_crypto_profile, e.g. ``modes_supported`` and ``max_dun_bytes_supported``.
238
239It then must set function pointers in the ``ll_ops`` field of the
240blk_crypto_profile to tell upper layers how to control the inline encryption
241hardware, e.g. how to program and evict keyslots.  Most drivers will need to
242implement ``keyslot_program`` and ``keyslot_evict``.  For details, see the
243comments for ``struct blk_crypto_ll_ops``.
244
245Once the driver registers a blk_crypto_profile with a request_queue, I/O
246requests the driver receives via that queue may have an encryption context.  All
247encryption contexts will be compatible with the crypto capabilities declared in
248the blk_crypto_profile, so drivers don't need to worry about handling
249unsupported requests.  Also, if a nonzero number of keyslots was declared in the
250blk_crypto_profile, then all I/O requests that have an encryption context will
251also have a keyslot which was already programmed with the appropriate key.
252
253If the driver implements runtime suspend and its blk_crypto_ll_ops don't work
254while the device is runtime-suspended, then the driver must also set the ``dev``
255field of the blk_crypto_profile to point to the ``struct device`` that will be
256resumed before any of the low-level operations are called.
257
258If there are situations where the inline encryption hardware loses the contents
259of its keyslots, e.g. device resets, the driver must handle reprogramming the
260keyslots.  To do this, the driver may call ``blk_crypto_reprogram_all_keys()``.
261
262Finally, if the driver used ``blk_crypto_profile_init()`` instead of
263``devm_blk_crypto_profile_init()``, then it is responsible for calling
264``blk_crypto_profile_destroy()`` when the crypto profile is no longer needed.
265
266Layered Devices
267===============
268
269Request queue based layered devices like dm-rq that wish to support inline
270encryption need to create their own blk_crypto_profile for their request_queue,
271and expose whatever functionality they choose. When a layered device wants to
272pass a clone of that request to another request_queue, blk-crypto will
273initialize and prepare the clone as necessary; see
274``blk_crypto_insert_cloned_request()``.
275
276Interaction between inline encryption and blk integrity
277=======================================================
278
279At the time of this patch, there is no real hardware that supports both these
280features. However, these features do interact with each other, and it's not
281completely trivial to make them both work together properly. In particular,
282when a WRITE bio wants to use inline encryption on a device that supports both
283features, the bio will have an encryption context specified, after which
284its integrity information is calculated (using the plaintext data, since
285the encryption will happen while data is being written), and the data and
286integrity info is sent to the device. Obviously, the integrity info must be
287verified before the data is encrypted. After the data is encrypted, the device
288must not store the integrity info that it received with the plaintext data
289since that might reveal information about the plaintext data. As such, it must
290re-generate the integrity info from the ciphertext data and store that on disk
291instead. Another issue with storing the integrity info of the plaintext data is
292that it changes the on disk format depending on whether hardware inline
293encryption support is present or the kernel crypto API fallback is used (since
294if the fallback is used, the device will receive the integrity info of the
295ciphertext, not that of the plaintext).
296
297Because there isn't any real hardware yet, it seems prudent to assume that
298hardware implementations might not implement both features together correctly,
299and disallow the combination for now. Whenever a device supports integrity, the
300kernel will pretend that the device does not support hardware inline encryption
301(by setting the blk_crypto_profile in the request_queue of the device to NULL).
302When the crypto API fallback is enabled, this means that all bios with and
303encryption context will use the fallback, and IO will complete as usual.  When
304the fallback is disabled, a bio with an encryption context will be failed.
305