1=====================================
2Filesystem-level encryption (fscrypt)
3=====================================
4
5Introduction
6============
7
8fscrypt is a library which filesystems can hook into to support
9transparent encryption of files and directories.
10
11Note: "fscrypt" in this document refers to the kernel-level portion,
12implemented in ``fs/crypto/``, as opposed to the userspace tool
13`fscrypt <https://github.com/google/fscrypt>`_.  This document only
14covers the kernel-level portion.  For command-line examples of how to
15use encryption, see the documentation for the userspace tool `fscrypt
16<https://github.com/google/fscrypt>`_.  Also, it is recommended to use
17the fscrypt userspace tool, or other existing userspace tools such as
18`fscryptctl <https://github.com/google/fscryptctl>`_ or `Android's key
19management system
20<https://source.android.com/security/encryption/file-based>`_, over
21using the kernel's API directly.  Using existing tools reduces the
22chance of introducing your own security bugs.  (Nevertheless, for
23completeness this documentation covers the kernel's API anyway.)
24
25Unlike dm-crypt, fscrypt operates at the filesystem level rather than
26at the block device level.  This allows it to encrypt different files
27with different keys and to have unencrypted files on the same
28filesystem.  This is useful for multi-user systems where each user's
29data-at-rest needs to be cryptographically isolated from the others.
30However, except for filenames, fscrypt does not encrypt filesystem
31metadata.
32
33Unlike eCryptfs, which is a stacked filesystem, fscrypt is integrated
34directly into supported filesystems --- currently ext4, F2FS, and
35UBIFS.  This allows encrypted files to be read and written without
36caching both the decrypted and encrypted pages in the pagecache,
37thereby nearly halving the memory used and bringing it in line with
38unencrypted files.  Similarly, half as many dentries and inodes are
39needed.  eCryptfs also limits encrypted filenames to 143 bytes,
40causing application compatibility issues; fscrypt allows the full 255
41bytes (NAME_MAX).  Finally, unlike eCryptfs, the fscrypt API can be
42used by unprivileged users, with no need to mount anything.
43
44fscrypt does not support encrypting files in-place.  Instead, it
45supports marking an empty directory as encrypted.  Then, after
46userspace provides the key, all regular files, directories, and
47symbolic links created in that directory tree are transparently
48encrypted.
49
50Threat model
51============
52
53Offline attacks
54---------------
55
56Provided that userspace chooses a strong encryption key, fscrypt
57protects the confidentiality of file contents and filenames in the
58event of a single point-in-time permanent offline compromise of the
59block device content.  fscrypt does not protect the confidentiality of
60non-filename metadata, e.g. file sizes, file permissions, file
61timestamps, and extended attributes.  Also, the existence and location
62of holes (unallocated blocks which logically contain all zeroes) in
63files is not protected.
64
65fscrypt is not guaranteed to protect confidentiality or authenticity
66if an attacker is able to manipulate the filesystem offline prior to
67an authorized user later accessing the filesystem.
68
69Online attacks
70--------------
71
72fscrypt (and storage encryption in general) can only provide limited
73protection, if any at all, against online attacks.  In detail:
74
75Side-channel attacks
76~~~~~~~~~~~~~~~~~~~~
77
78fscrypt is only resistant to side-channel attacks, such as timing or
79electromagnetic attacks, to the extent that the underlying Linux
80Cryptographic API algorithms or inline encryption hardware are.  If a
81vulnerable algorithm is used, such as a table-based implementation of
82AES, it may be possible for an attacker to mount a side channel attack
83against the online system.  Side channel attacks may also be mounted
84against applications consuming decrypted data.
85
86Unauthorized file access
87~~~~~~~~~~~~~~~~~~~~~~~~
88
89After an encryption key has been added, fscrypt does not hide the
90plaintext file contents or filenames from other users on the same
91system.  Instead, existing access control mechanisms such as file mode
92bits, POSIX ACLs, LSMs, or namespaces should be used for this purpose.
93
94(For the reasoning behind this, understand that while the key is
95added, the confidentiality of the data, from the perspective of the
96system itself, is *not* protected by the mathematical properties of
97encryption but rather only by the correctness of the kernel.
98Therefore, any encryption-specific access control checks would merely
99be enforced by kernel *code* and therefore would be largely redundant
100with the wide variety of access control mechanisms already available.)
101
102Kernel memory compromise
103~~~~~~~~~~~~~~~~~~~~~~~~
104
105An attacker who compromises the system enough to read from arbitrary
106memory, e.g. by mounting a physical attack or by exploiting a kernel
107security vulnerability, can compromise all encryption keys that are
108currently in use.
109
110However, fscrypt allows encryption keys to be removed from the kernel,
111which may protect them from later compromise.
112
113In more detail, the FS_IOC_REMOVE_ENCRYPTION_KEY ioctl (or the
114FS_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS ioctl) can wipe a master
115encryption key from kernel memory.  If it does so, it will also try to
116evict all cached inodes which had been "unlocked" using the key,
117thereby wiping their per-file keys and making them once again appear
118"locked", i.e. in ciphertext or encrypted form.
119
120However, these ioctls have some limitations:
121
122- Per-file keys for in-use files will *not* be removed or wiped.
123  Therefore, for maximum effect, userspace should close the relevant
124  encrypted files and directories before removing a master key, as
125  well as kill any processes whose working directory is in an affected
126  encrypted directory.
127
128- The kernel cannot magically wipe copies of the master key(s) that
129  userspace might have as well.  Therefore, userspace must wipe all
130  copies of the master key(s) it makes as well; normally this should
131  be done immediately after FS_IOC_ADD_ENCRYPTION_KEY, without waiting
132  for FS_IOC_REMOVE_ENCRYPTION_KEY.  Naturally, the same also applies
133  to all higher levels in the key hierarchy.  Userspace should also
134  follow other security precautions such as mlock()ing memory
135  containing keys to prevent it from being swapped out.
136
137- In general, decrypted contents and filenames in the kernel VFS
138  caches are freed but not wiped.  Therefore, portions thereof may be
139  recoverable from freed memory, even after the corresponding key(s)
140  were wiped.  To partially solve this, you can set
141  CONFIG_PAGE_POISONING=y in your kernel config and add page_poison=1
142  to your kernel command line.  However, this has a performance cost.
143
144- Secret keys might still exist in CPU registers, in crypto
145  accelerator hardware (if used by the crypto API to implement any of
146  the algorithms), or in other places not explicitly considered here.
147
148Limitations of v1 policies
149~~~~~~~~~~~~~~~~~~~~~~~~~~
150
151v1 encryption policies have some weaknesses with respect to online
152attacks:
153
154- There is no verification that the provided master key is correct.
155  Therefore, a malicious user can temporarily associate the wrong key
156  with another user's encrypted files to which they have read-only
157  access.  Because of filesystem caching, the wrong key will then be
158  used by the other user's accesses to those files, even if the other
159  user has the correct key in their own keyring.  This violates the
160  meaning of "read-only access".
161
162- A compromise of a per-file key also compromises the master key from
163  which it was derived.
164
165- Non-root users cannot securely remove encryption keys.
166
167All the above problems are fixed with v2 encryption policies.  For
168this reason among others, it is recommended to use v2 encryption
169policies on all new encrypted directories.
170
171Key hierarchy
172=============
173
174Master Keys
175-----------
176
177Each encrypted directory tree is protected by a *master key*.  Master
178keys can be up to 64 bytes long, and must be at least as long as the
179greater of the security strength of the contents and filenames
180encryption modes being used.  For example, if any AES-256 mode is
181used, the master key must be at least 256 bits, i.e. 32 bytes.  A
182stricter requirement applies if the key is used by a v1 encryption
183policy and AES-256-XTS is used; such keys must be 64 bytes.
184
185To "unlock" an encrypted directory tree, userspace must provide the
186appropriate master key.  There can be any number of master keys, each
187of which protects any number of directory trees on any number of
188filesystems.
189
190Master keys must be real cryptographic keys, i.e. indistinguishable
191from random bytestrings of the same length.  This implies that users
192**must not** directly use a password as a master key, zero-pad a
193shorter key, or repeat a shorter key.  Security cannot be guaranteed
194if userspace makes any such error, as the cryptographic proofs and
195analysis would no longer apply.
196
197Instead, users should generate master keys either using a
198cryptographically secure random number generator, or by using a KDF
199(Key Derivation Function).  The kernel does not do any key stretching;
200therefore, if userspace derives the key from a low-entropy secret such
201as a passphrase, it is critical that a KDF designed for this purpose
202be used, such as scrypt, PBKDF2, or Argon2.
203
204Key derivation function
205-----------------------
206
207With one exception, fscrypt never uses the master key(s) for
208encryption directly.  Instead, they are only used as input to a KDF
209(Key Derivation Function) to derive the actual keys.
210
211The KDF used for a particular master key differs depending on whether
212the key is used for v1 encryption policies or for v2 encryption
213policies.  Users **must not** use the same key for both v1 and v2
214encryption policies.  (No real-world attack is currently known on this
215specific case of key reuse, but its security cannot be guaranteed
216since the cryptographic proofs and analysis would no longer apply.)
217
218For v1 encryption policies, the KDF only supports deriving per-file
219encryption keys.  It works by encrypting the master key with
220AES-128-ECB, using the file's 16-byte nonce as the AES key.  The
221resulting ciphertext is used as the derived key.  If the ciphertext is
222longer than needed, then it is truncated to the needed length.
223
224For v2 encryption policies, the KDF is HKDF-SHA512.  The master key is
225passed as the "input keying material", no salt is used, and a distinct
226"application-specific information string" is used for each distinct
227key to be derived.  For example, when a per-file encryption key is
228derived, the application-specific information string is the file's
229nonce prefixed with "fscrypt\\0" and a context byte.  Different
230context bytes are used for other types of derived keys.
231
232HKDF-SHA512 is preferred to the original AES-128-ECB based KDF because
233HKDF is more flexible, is nonreversible, and evenly distributes
234entropy from the master key.  HKDF is also standardized and widely
235used by other software, whereas the AES-128-ECB based KDF is ad-hoc.
236
237Per-file encryption keys
238------------------------
239
240Since each master key can protect many files, it is necessary to
241"tweak" the encryption of each file so that the same plaintext in two
242files doesn't map to the same ciphertext, or vice versa.  In most
243cases, fscrypt does this by deriving per-file keys.  When a new
244encrypted inode (regular file, directory, or symlink) is created,
245fscrypt randomly generates a 16-byte nonce and stores it in the
246inode's encryption xattr.  Then, it uses a KDF (as described in `Key
247derivation function`_) to derive the file's key from the master key
248and nonce.
249
250Key derivation was chosen over key wrapping because wrapped keys would
251require larger xattrs which would be less likely to fit in-line in the
252filesystem's inode table, and there didn't appear to be any
253significant advantages to key wrapping.  In particular, currently
254there is no requirement to support unlocking a file with multiple
255alternative master keys or to support rotating master keys.  Instead,
256the master keys may be wrapped in userspace, e.g. as is done by the
257`fscrypt <https://github.com/google/fscrypt>`_ tool.
258
259DIRECT_KEY policies
260-------------------
261
262The Adiantum encryption mode (see `Encryption modes and usage`_) is
263suitable for both contents and filenames encryption, and it accepts
264long IVs --- long enough to hold both an 8-byte logical block number
265and a 16-byte per-file nonce.  Also, the overhead of each Adiantum key
266is greater than that of an AES-256-XTS key.
267
268Therefore, to improve performance and save memory, for Adiantum a
269"direct key" configuration is supported.  When the user has enabled
270this by setting FSCRYPT_POLICY_FLAG_DIRECT_KEY in the fscrypt policy,
271per-file encryption keys are not used.  Instead, whenever any data
272(contents or filenames) is encrypted, the file's 16-byte nonce is
273included in the IV.  Moreover:
274
275- For v1 encryption policies, the encryption is done directly with the
276  master key.  Because of this, users **must not** use the same master
277  key for any other purpose, even for other v1 policies.
278
279- For v2 encryption policies, the encryption is done with a per-mode
280  key derived using the KDF.  Users may use the same master key for
281  other v2 encryption policies.
282
283IV_INO_LBLK_64 policies
284-----------------------
285
286When FSCRYPT_POLICY_FLAG_IV_INO_LBLK_64 is set in the fscrypt policy,
287the encryption keys are derived from the master key, encryption mode
288number, and filesystem UUID.  This normally results in all files
289protected by the same master key sharing a single contents encryption
290key and a single filenames encryption key.  To still encrypt different
291files' data differently, inode numbers are included in the IVs.
292Consequently, shrinking the filesystem may not be allowed.
293
294This format is optimized for use with inline encryption hardware
295compliant with the UFS standard, which supports only 64 IV bits per
296I/O request and may have only a small number of keyslots.
297
298IV_INO_LBLK_32 policies
299-----------------------
300
301IV_INO_LBLK_32 policies work like IV_INO_LBLK_64, except that for
302IV_INO_LBLK_32, the inode number is hashed with SipHash-2-4 (where the
303SipHash key is derived from the master key) and added to the file
304logical block number mod 2^32 to produce a 32-bit IV.
305
306This format is optimized for use with inline encryption hardware
307compliant with the eMMC v5.2 standard, which supports only 32 IV bits
308per I/O request and may have only a small number of keyslots.  This
309format results in some level of IV reuse, so it should only be used
310when necessary due to hardware limitations.
311
312Key identifiers
313---------------
314
315For master keys used for v2 encryption policies, a unique 16-byte "key
316identifier" is also derived using the KDF.  This value is stored in
317the clear, since it is needed to reliably identify the key itself.
318
319Dirhash keys
320------------
321
322For directories that are indexed using a secret-keyed dirhash over the
323plaintext filenames, the KDF is also used to derive a 128-bit
324SipHash-2-4 key per directory in order to hash filenames.  This works
325just like deriving a per-file encryption key, except that a different
326KDF context is used.  Currently, only casefolded ("case-insensitive")
327encrypted directories use this style of hashing.
328
329Encryption modes and usage
330==========================
331
332fscrypt allows one encryption mode to be specified for file contents
333and one encryption mode to be specified for filenames.  Different
334directory trees are permitted to use different encryption modes.
335Currently, the following pairs of encryption modes are supported:
336
337- AES-256-XTS for contents and AES-256-CTS-CBC for filenames
338- AES-128-CBC for contents and AES-128-CTS-CBC for filenames
339- Adiantum for both contents and filenames
340- AES-256-XTS for contents and AES-256-HCTR2 for filenames (v2 policies only)
341- SM4-XTS for contents and SM4-CTS-CBC for filenames (v2 policies only)
342
343If unsure, you should use the (AES-256-XTS, AES-256-CTS-CBC) pair.
344
345AES-128-CBC was added only for low-powered embedded devices with
346crypto accelerators such as CAAM or CESA that do not support XTS.  To
347use AES-128-CBC, CONFIG_CRYPTO_ESSIV and CONFIG_CRYPTO_SHA256 (or
348another SHA-256 implementation) must be enabled so that ESSIV can be
349used.
350
351Adiantum is a (primarily) stream cipher-based mode that is fast even
352on CPUs without dedicated crypto instructions.  It's also a true
353wide-block mode, unlike XTS.  It can also eliminate the need to derive
354per-file encryption keys.  However, it depends on the security of two
355primitives, XChaCha12 and AES-256, rather than just one.  See the
356paper "Adiantum: length-preserving encryption for entry-level
357processors" (https://eprint.iacr.org/2018/720.pdf) for more details.
358To use Adiantum, CONFIG_CRYPTO_ADIANTUM must be enabled.  Also, fast
359implementations of ChaCha and NHPoly1305 should be enabled, e.g.
360CONFIG_CRYPTO_CHACHA20_NEON and CONFIG_CRYPTO_NHPOLY1305_NEON for ARM.
361
362AES-256-HCTR2 is another true wide-block encryption mode that is intended for
363use on CPUs with dedicated crypto instructions.  AES-256-HCTR2 has the property
364that a bitflip in the plaintext changes the entire ciphertext.  This property
365makes it desirable for filename encryption since initialization vectors are
366reused within a directory.  For more details on AES-256-HCTR2, see the paper
367"Length-preserving encryption with HCTR2"
368(https://eprint.iacr.org/2021/1441.pdf).  To use AES-256-HCTR2,
369CONFIG_CRYPTO_HCTR2 must be enabled.  Also, fast implementations of XCTR and
370POLYVAL should be enabled, e.g. CRYPTO_POLYVAL_ARM64_CE and
371CRYPTO_AES_ARM64_CE_BLK for ARM64.
372
373SM4 is a Chinese block cipher that is an alternative to AES.  It has
374not seen as much security review as AES, and it only has a 128-bit key
375size.  It may be useful in cases where its use is mandated.
376Otherwise, it should not be used.  For SM4 support to be available, it
377also needs to be enabled in the kernel crypto API.
378
379New encryption modes can be added relatively easily, without changes
380to individual filesystems.  However, authenticated encryption (AE)
381modes are not currently supported because of the difficulty of dealing
382with ciphertext expansion.
383
384Contents encryption
385-------------------
386
387For file contents, each filesystem block is encrypted independently.
388Starting from Linux kernel 5.5, encryption of filesystems with block
389size less than system's page size is supported.
390
391Each block's IV is set to the logical block number within the file as
392a little endian number, except that:
393
394- With CBC mode encryption, ESSIV is also used.  Specifically, each IV
395  is encrypted with AES-256 where the AES-256 key is the SHA-256 hash
396  of the file's data encryption key.
397
398- With `DIRECT_KEY policies`_, the file's nonce is appended to the IV.
399  Currently this is only allowed with the Adiantum encryption mode.
400
401- With `IV_INO_LBLK_64 policies`_, the logical block number is limited
402  to 32 bits and is placed in bits 0-31 of the IV.  The inode number
403  (which is also limited to 32 bits) is placed in bits 32-63.
404
405- With `IV_INO_LBLK_32 policies`_, the logical block number is limited
406  to 32 bits and is placed in bits 0-31 of the IV.  The inode number
407  is then hashed and added mod 2^32.
408
409Note that because file logical block numbers are included in the IVs,
410filesystems must enforce that blocks are never shifted around within
411encrypted files, e.g. via "collapse range" or "insert range".
412
413Filenames encryption
414--------------------
415
416For filenames, each full filename is encrypted at once.  Because of
417the requirements to retain support for efficient directory lookups and
418filenames of up to 255 bytes, the same IV is used for every filename
419in a directory.
420
421However, each encrypted directory still uses a unique key, or
422alternatively has the file's nonce (for `DIRECT_KEY policies`_) or
423inode number (for `IV_INO_LBLK_64 policies`_) included in the IVs.
424Thus, IV reuse is limited to within a single directory.
425
426With CTS-CBC, the IV reuse means that when the plaintext filenames share a
427common prefix at least as long as the cipher block size (16 bytes for AES), the
428corresponding encrypted filenames will also share a common prefix.  This is
429undesirable.  Adiantum and HCTR2 do not have this weakness, as they are
430wide-block encryption modes.
431
432All supported filenames encryption modes accept any plaintext length
433>= 16 bytes; cipher block alignment is not required.  However,
434filenames shorter than 16 bytes are NUL-padded to 16 bytes before
435being encrypted.  In addition, to reduce leakage of filename lengths
436via their ciphertexts, all filenames are NUL-padded to the next 4, 8,
43716, or 32-byte boundary (configurable).  32 is recommended since this
438provides the best confidentiality, at the cost of making directory
439entries consume slightly more space.  Note that since NUL (``\0``) is
440not otherwise a valid character in filenames, the padding will never
441produce duplicate plaintexts.
442
443Symbolic link targets are considered a type of filename and are
444encrypted in the same way as filenames in directory entries, except
445that IV reuse is not a problem as each symlink has its own inode.
446
447User API
448========
449
450Setting an encryption policy
451----------------------------
452
453FS_IOC_SET_ENCRYPTION_POLICY
454~~~~~~~~~~~~~~~~~~~~~~~~~~~~
455
456The FS_IOC_SET_ENCRYPTION_POLICY ioctl sets an encryption policy on an
457empty directory or verifies that a directory or regular file already
458has the specified encryption policy.  It takes in a pointer to
459struct fscrypt_policy_v1 or struct fscrypt_policy_v2, defined as
460follows::
461
462    #define FSCRYPT_POLICY_V1               0
463    #define FSCRYPT_KEY_DESCRIPTOR_SIZE     8
464    struct fscrypt_policy_v1 {
465            __u8 version;
466            __u8 contents_encryption_mode;
467            __u8 filenames_encryption_mode;
468            __u8 flags;
469            __u8 master_key_descriptor[FSCRYPT_KEY_DESCRIPTOR_SIZE];
470    };
471    #define fscrypt_policy  fscrypt_policy_v1
472
473    #define FSCRYPT_POLICY_V2               2
474    #define FSCRYPT_KEY_IDENTIFIER_SIZE     16
475    struct fscrypt_policy_v2 {
476            __u8 version;
477            __u8 contents_encryption_mode;
478            __u8 filenames_encryption_mode;
479            __u8 flags;
480            __u8 __reserved[4];
481            __u8 master_key_identifier[FSCRYPT_KEY_IDENTIFIER_SIZE];
482    };
483
484This structure must be initialized as follows:
485
486- ``version`` must be FSCRYPT_POLICY_V1 (0) if
487  struct fscrypt_policy_v1 is used or FSCRYPT_POLICY_V2 (2) if
488  struct fscrypt_policy_v2 is used. (Note: we refer to the original
489  policy version as "v1", though its version code is really 0.)
490  For new encrypted directories, use v2 policies.
491
492- ``contents_encryption_mode`` and ``filenames_encryption_mode`` must
493  be set to constants from ``<linux/fscrypt.h>`` which identify the
494  encryption modes to use.  If unsure, use FSCRYPT_MODE_AES_256_XTS
495  (1) for ``contents_encryption_mode`` and FSCRYPT_MODE_AES_256_CTS
496  (4) for ``filenames_encryption_mode``.
497
498- ``flags`` contains optional flags from ``<linux/fscrypt.h>``:
499
500  - FSCRYPT_POLICY_FLAGS_PAD_*: The amount of NUL padding to use when
501    encrypting filenames.  If unsure, use FSCRYPT_POLICY_FLAGS_PAD_32
502    (0x3).
503  - FSCRYPT_POLICY_FLAG_DIRECT_KEY: See `DIRECT_KEY policies`_.
504  - FSCRYPT_POLICY_FLAG_IV_INO_LBLK_64: See `IV_INO_LBLK_64
505    policies`_.
506  - FSCRYPT_POLICY_FLAG_IV_INO_LBLK_32: See `IV_INO_LBLK_32
507    policies`_.
508
509  v1 encryption policies only support the PAD_* and DIRECT_KEY flags.
510  The other flags are only supported by v2 encryption policies.
511
512  The DIRECT_KEY, IV_INO_LBLK_64, and IV_INO_LBLK_32 flags are
513  mutually exclusive.
514
515- For v2 encryption policies, ``__reserved`` must be zeroed.
516
517- For v1 encryption policies, ``master_key_descriptor`` specifies how
518  to find the master key in a keyring; see `Adding keys`_.  It is up
519  to userspace to choose a unique ``master_key_descriptor`` for each
520  master key.  The e4crypt and fscrypt tools use the first 8 bytes of
521  ``SHA-512(SHA-512(master_key))``, but this particular scheme is not
522  required.  Also, the master key need not be in the keyring yet when
523  FS_IOC_SET_ENCRYPTION_POLICY is executed.  However, it must be added
524  before any files can be created in the encrypted directory.
525
526  For v2 encryption policies, ``master_key_descriptor`` has been
527  replaced with ``master_key_identifier``, which is longer and cannot
528  be arbitrarily chosen.  Instead, the key must first be added using
529  `FS_IOC_ADD_ENCRYPTION_KEY`_.  Then, the ``key_spec.u.identifier``
530  the kernel returned in the struct fscrypt_add_key_arg must
531  be used as the ``master_key_identifier`` in
532  struct fscrypt_policy_v2.
533
534If the file is not yet encrypted, then FS_IOC_SET_ENCRYPTION_POLICY
535verifies that the file is an empty directory.  If so, the specified
536encryption policy is assigned to the directory, turning it into an
537encrypted directory.  After that, and after providing the
538corresponding master key as described in `Adding keys`_, all regular
539files, directories (recursively), and symlinks created in the
540directory will be encrypted, inheriting the same encryption policy.
541The filenames in the directory's entries will be encrypted as well.
542
543Alternatively, if the file is already encrypted, then
544FS_IOC_SET_ENCRYPTION_POLICY validates that the specified encryption
545policy exactly matches the actual one.  If they match, then the ioctl
546returns 0.  Otherwise, it fails with EEXIST.  This works on both
547regular files and directories, including nonempty directories.
548
549When a v2 encryption policy is assigned to a directory, it is also
550required that either the specified key has been added by the current
551user or that the caller has CAP_FOWNER in the initial user namespace.
552(This is needed to prevent a user from encrypting their data with
553another user's key.)  The key must remain added while
554FS_IOC_SET_ENCRYPTION_POLICY is executing.  However, if the new
555encrypted directory does not need to be accessed immediately, then the
556key can be removed right away afterwards.
557
558Note that the ext4 filesystem does not allow the root directory to be
559encrypted, even if it is empty.  Users who want to encrypt an entire
560filesystem with one key should consider using dm-crypt instead.
561
562FS_IOC_SET_ENCRYPTION_POLICY can fail with the following errors:
563
564- ``EACCES``: the file is not owned by the process's uid, nor does the
565  process have the CAP_FOWNER capability in a namespace with the file
566  owner's uid mapped
567- ``EEXIST``: the file is already encrypted with an encryption policy
568  different from the one specified
569- ``EINVAL``: an invalid encryption policy was specified (invalid
570  version, mode(s), or flags; or reserved bits were set); or a v1
571  encryption policy was specified but the directory has the casefold
572  flag enabled (casefolding is incompatible with v1 policies).
573- ``ENOKEY``: a v2 encryption policy was specified, but the key with
574  the specified ``master_key_identifier`` has not been added, nor does
575  the process have the CAP_FOWNER capability in the initial user
576  namespace
577- ``ENOTDIR``: the file is unencrypted and is a regular file, not a
578  directory
579- ``ENOTEMPTY``: the file is unencrypted and is a nonempty directory
580- ``ENOTTY``: this type of filesystem does not implement encryption
581- ``EOPNOTSUPP``: the kernel was not configured with encryption
582  support for filesystems, or the filesystem superblock has not
583  had encryption enabled on it.  (For example, to use encryption on an
584  ext4 filesystem, CONFIG_FS_ENCRYPTION must be enabled in the
585  kernel config, and the superblock must have had the "encrypt"
586  feature flag enabled using ``tune2fs -O encrypt`` or ``mkfs.ext4 -O
587  encrypt``.)
588- ``EPERM``: this directory may not be encrypted, e.g. because it is
589  the root directory of an ext4 filesystem
590- ``EROFS``: the filesystem is readonly
591
592Getting an encryption policy
593----------------------------
594
595Two ioctls are available to get a file's encryption policy:
596
597- `FS_IOC_GET_ENCRYPTION_POLICY_EX`_
598- `FS_IOC_GET_ENCRYPTION_POLICY`_
599
600The extended (_EX) version of the ioctl is more general and is
601recommended to use when possible.  However, on older kernels only the
602original ioctl is available.  Applications should try the extended
603version, and if it fails with ENOTTY fall back to the original
604version.
605
606FS_IOC_GET_ENCRYPTION_POLICY_EX
607~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
608
609The FS_IOC_GET_ENCRYPTION_POLICY_EX ioctl retrieves the encryption
610policy, if any, for a directory or regular file.  No additional
611permissions are required beyond the ability to open the file.  It
612takes in a pointer to struct fscrypt_get_policy_ex_arg,
613defined as follows::
614
615    struct fscrypt_get_policy_ex_arg {
616            __u64 policy_size; /* input/output */
617            union {
618                    __u8 version;
619                    struct fscrypt_policy_v1 v1;
620                    struct fscrypt_policy_v2 v2;
621            } policy; /* output */
622    };
623
624The caller must initialize ``policy_size`` to the size available for
625the policy struct, i.e. ``sizeof(arg.policy)``.
626
627On success, the policy struct is returned in ``policy``, and its
628actual size is returned in ``policy_size``.  ``policy.version`` should
629be checked to determine the version of policy returned.  Note that the
630version code for the "v1" policy is actually 0 (FSCRYPT_POLICY_V1).
631
632FS_IOC_GET_ENCRYPTION_POLICY_EX can fail with the following errors:
633
634- ``EINVAL``: the file is encrypted, but it uses an unrecognized
635  encryption policy version
636- ``ENODATA``: the file is not encrypted
637- ``ENOTTY``: this type of filesystem does not implement encryption,
638  or this kernel is too old to support FS_IOC_GET_ENCRYPTION_POLICY_EX
639  (try FS_IOC_GET_ENCRYPTION_POLICY instead)
640- ``EOPNOTSUPP``: the kernel was not configured with encryption
641  support for this filesystem, or the filesystem superblock has not
642  had encryption enabled on it
643- ``EOVERFLOW``: the file is encrypted and uses a recognized
644  encryption policy version, but the policy struct does not fit into
645  the provided buffer
646
647Note: if you only need to know whether a file is encrypted or not, on
648most filesystems it is also possible to use the FS_IOC_GETFLAGS ioctl
649and check for FS_ENCRYPT_FL, or to use the statx() system call and
650check for STATX_ATTR_ENCRYPTED in stx_attributes.
651
652FS_IOC_GET_ENCRYPTION_POLICY
653~~~~~~~~~~~~~~~~~~~~~~~~~~~~
654
655The FS_IOC_GET_ENCRYPTION_POLICY ioctl can also retrieve the
656encryption policy, if any, for a directory or regular file.  However,
657unlike `FS_IOC_GET_ENCRYPTION_POLICY_EX`_,
658FS_IOC_GET_ENCRYPTION_POLICY only supports the original policy
659version.  It takes in a pointer directly to struct fscrypt_policy_v1
660rather than struct fscrypt_get_policy_ex_arg.
661
662The error codes for FS_IOC_GET_ENCRYPTION_POLICY are the same as those
663for FS_IOC_GET_ENCRYPTION_POLICY_EX, except that
664FS_IOC_GET_ENCRYPTION_POLICY also returns ``EINVAL`` if the file is
665encrypted using a newer encryption policy version.
666
667Getting the per-filesystem salt
668-------------------------------
669
670Some filesystems, such as ext4 and F2FS, also support the deprecated
671ioctl FS_IOC_GET_ENCRYPTION_PWSALT.  This ioctl retrieves a randomly
672generated 16-byte value stored in the filesystem superblock.  This
673value is intended to used as a salt when deriving an encryption key
674from a passphrase or other low-entropy user credential.
675
676FS_IOC_GET_ENCRYPTION_PWSALT is deprecated.  Instead, prefer to
677generate and manage any needed salt(s) in userspace.
678
679Getting a file's encryption nonce
680---------------------------------
681
682Since Linux v5.7, the ioctl FS_IOC_GET_ENCRYPTION_NONCE is supported.
683On encrypted files and directories it gets the inode's 16-byte nonce.
684On unencrypted files and directories, it fails with ENODATA.
685
686This ioctl can be useful for automated tests which verify that the
687encryption is being done correctly.  It is not needed for normal use
688of fscrypt.
689
690Adding keys
691-----------
692
693FS_IOC_ADD_ENCRYPTION_KEY
694~~~~~~~~~~~~~~~~~~~~~~~~~
695
696The FS_IOC_ADD_ENCRYPTION_KEY ioctl adds a master encryption key to
697the filesystem, making all files on the filesystem which were
698encrypted using that key appear "unlocked", i.e. in plaintext form.
699It can be executed on any file or directory on the target filesystem,
700but using the filesystem's root directory is recommended.  It takes in
701a pointer to struct fscrypt_add_key_arg, defined as follows::
702
703    struct fscrypt_add_key_arg {
704            struct fscrypt_key_specifier key_spec;
705            __u32 raw_size;
706            __u32 key_id;
707            __u32 __reserved[8];
708            __u8 raw[];
709    };
710
711    #define FSCRYPT_KEY_SPEC_TYPE_DESCRIPTOR        1
712    #define FSCRYPT_KEY_SPEC_TYPE_IDENTIFIER        2
713
714    struct fscrypt_key_specifier {
715            __u32 type;     /* one of FSCRYPT_KEY_SPEC_TYPE_* */
716            __u32 __reserved;
717            union {
718                    __u8 __reserved[32]; /* reserve some extra space */
719                    __u8 descriptor[FSCRYPT_KEY_DESCRIPTOR_SIZE];
720                    __u8 identifier[FSCRYPT_KEY_IDENTIFIER_SIZE];
721            } u;
722    };
723
724    struct fscrypt_provisioning_key_payload {
725            __u32 type;
726            __u32 __reserved;
727            __u8 raw[];
728    };
729
730struct fscrypt_add_key_arg must be zeroed, then initialized
731as follows:
732
733- If the key is being added for use by v1 encryption policies, then
734  ``key_spec.type`` must contain FSCRYPT_KEY_SPEC_TYPE_DESCRIPTOR, and
735  ``key_spec.u.descriptor`` must contain the descriptor of the key
736  being added, corresponding to the value in the
737  ``master_key_descriptor`` field of struct fscrypt_policy_v1.
738  To add this type of key, the calling process must have the
739  CAP_SYS_ADMIN capability in the initial user namespace.
740
741  Alternatively, if the key is being added for use by v2 encryption
742  policies, then ``key_spec.type`` must contain
743  FSCRYPT_KEY_SPEC_TYPE_IDENTIFIER, and ``key_spec.u.identifier`` is
744  an *output* field which the kernel fills in with a cryptographic
745  hash of the key.  To add this type of key, the calling process does
746  not need any privileges.  However, the number of keys that can be
747  added is limited by the user's quota for the keyrings service (see
748  ``Documentation/security/keys/core.rst``).
749
750- ``raw_size`` must be the size of the ``raw`` key provided, in bytes.
751  Alternatively, if ``key_id`` is nonzero, this field must be 0, since
752  in that case the size is implied by the specified Linux keyring key.
753
754- ``key_id`` is 0 if the raw key is given directly in the ``raw``
755  field.  Otherwise ``key_id`` is the ID of a Linux keyring key of
756  type "fscrypt-provisioning" whose payload is
757  struct fscrypt_provisioning_key_payload whose ``raw`` field contains
758  the raw key and whose ``type`` field matches ``key_spec.type``.
759  Since ``raw`` is variable-length, the total size of this key's
760  payload must be ``sizeof(struct fscrypt_provisioning_key_payload)``
761  plus the raw key size.  The process must have Search permission on
762  this key.
763
764  Most users should leave this 0 and specify the raw key directly.
765  The support for specifying a Linux keyring key is intended mainly to
766  allow re-adding keys after a filesystem is unmounted and re-mounted,
767  without having to store the raw keys in userspace memory.
768
769- ``raw`` is a variable-length field which must contain the actual
770  key, ``raw_size`` bytes long.  Alternatively, if ``key_id`` is
771  nonzero, then this field is unused.
772
773For v2 policy keys, the kernel keeps track of which user (identified
774by effective user ID) added the key, and only allows the key to be
775removed by that user --- or by "root", if they use
776`FS_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS`_.
777
778However, if another user has added the key, it may be desirable to
779prevent that other user from unexpectedly removing it.  Therefore,
780FS_IOC_ADD_ENCRYPTION_KEY may also be used to add a v2 policy key
781*again*, even if it's already added by other user(s).  In this case,
782FS_IOC_ADD_ENCRYPTION_KEY will just install a claim to the key for the
783current user, rather than actually add the key again (but the raw key
784must still be provided, as a proof of knowledge).
785
786FS_IOC_ADD_ENCRYPTION_KEY returns 0 if either the key or a claim to
787the key was either added or already exists.
788
789FS_IOC_ADD_ENCRYPTION_KEY can fail with the following errors:
790
791- ``EACCES``: FSCRYPT_KEY_SPEC_TYPE_DESCRIPTOR was specified, but the
792  caller does not have the CAP_SYS_ADMIN capability in the initial
793  user namespace; or the raw key was specified by Linux key ID but the
794  process lacks Search permission on the key.
795- ``EDQUOT``: the key quota for this user would be exceeded by adding
796  the key
797- ``EINVAL``: invalid key size or key specifier type, or reserved bits
798  were set
799- ``EKEYREJECTED``: the raw key was specified by Linux key ID, but the
800  key has the wrong type
801- ``ENOKEY``: the raw key was specified by Linux key ID, but no key
802  exists with that ID
803- ``ENOTTY``: this type of filesystem does not implement encryption
804- ``EOPNOTSUPP``: the kernel was not configured with encryption
805  support for this filesystem, or the filesystem superblock has not
806  had encryption enabled on it
807
808Legacy method
809~~~~~~~~~~~~~
810
811For v1 encryption policies, a master encryption key can also be
812provided by adding it to a process-subscribed keyring, e.g. to a
813session keyring, or to a user keyring if the user keyring is linked
814into the session keyring.
815
816This method is deprecated (and not supported for v2 encryption
817policies) for several reasons.  First, it cannot be used in
818combination with FS_IOC_REMOVE_ENCRYPTION_KEY (see `Removing keys`_),
819so for removing a key a workaround such as keyctl_unlink() in
820combination with ``sync; echo 2 > /proc/sys/vm/drop_caches`` would
821have to be used.  Second, it doesn't match the fact that the
822locked/unlocked status of encrypted files (i.e. whether they appear to
823be in plaintext form or in ciphertext form) is global.  This mismatch
824has caused much confusion as well as real problems when processes
825running under different UIDs, such as a ``sudo`` command, need to
826access encrypted files.
827
828Nevertheless, to add a key to one of the process-subscribed keyrings,
829the add_key() system call can be used (see:
830``Documentation/security/keys/core.rst``).  The key type must be
831"logon"; keys of this type are kept in kernel memory and cannot be
832read back by userspace.  The key description must be "fscrypt:"
833followed by the 16-character lower case hex representation of the
834``master_key_descriptor`` that was set in the encryption policy.  The
835key payload must conform to the following structure::
836
837    #define FSCRYPT_MAX_KEY_SIZE            64
838
839    struct fscrypt_key {
840            __u32 mode;
841            __u8 raw[FSCRYPT_MAX_KEY_SIZE];
842            __u32 size;
843    };
844
845``mode`` is ignored; just set it to 0.  The actual key is provided in
846``raw`` with ``size`` indicating its size in bytes.  That is, the
847bytes ``raw[0..size-1]`` (inclusive) are the actual key.
848
849The key description prefix "fscrypt:" may alternatively be replaced
850with a filesystem-specific prefix such as "ext4:".  However, the
851filesystem-specific prefixes are deprecated and should not be used in
852new programs.
853
854Removing keys
855-------------
856
857Two ioctls are available for removing a key that was added by
858`FS_IOC_ADD_ENCRYPTION_KEY`_:
859
860- `FS_IOC_REMOVE_ENCRYPTION_KEY`_
861- `FS_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS`_
862
863These two ioctls differ only in cases where v2 policy keys are added
864or removed by non-root users.
865
866These ioctls don't work on keys that were added via the legacy
867process-subscribed keyrings mechanism.
868
869Before using these ioctls, read the `Kernel memory compromise`_
870section for a discussion of the security goals and limitations of
871these ioctls.
872
873FS_IOC_REMOVE_ENCRYPTION_KEY
874~~~~~~~~~~~~~~~~~~~~~~~~~~~~
875
876The FS_IOC_REMOVE_ENCRYPTION_KEY ioctl removes a claim to a master
877encryption key from the filesystem, and possibly removes the key
878itself.  It can be executed on any file or directory on the target
879filesystem, but using the filesystem's root directory is recommended.
880It takes in a pointer to struct fscrypt_remove_key_arg, defined
881as follows::
882
883    struct fscrypt_remove_key_arg {
884            struct fscrypt_key_specifier key_spec;
885    #define FSCRYPT_KEY_REMOVAL_STATUS_FLAG_FILES_BUSY      0x00000001
886    #define FSCRYPT_KEY_REMOVAL_STATUS_FLAG_OTHER_USERS     0x00000002
887            __u32 removal_status_flags;     /* output */
888            __u32 __reserved[5];
889    };
890
891This structure must be zeroed, then initialized as follows:
892
893- The key to remove is specified by ``key_spec``:
894
895    - To remove a key used by v1 encryption policies, set
896      ``key_spec.type`` to FSCRYPT_KEY_SPEC_TYPE_DESCRIPTOR and fill
897      in ``key_spec.u.descriptor``.  To remove this type of key, the
898      calling process must have the CAP_SYS_ADMIN capability in the
899      initial user namespace.
900
901    - To remove a key used by v2 encryption policies, set
902      ``key_spec.type`` to FSCRYPT_KEY_SPEC_TYPE_IDENTIFIER and fill
903      in ``key_spec.u.identifier``.
904
905For v2 policy keys, this ioctl is usable by non-root users.  However,
906to make this possible, it actually just removes the current user's
907claim to the key, undoing a single call to FS_IOC_ADD_ENCRYPTION_KEY.
908Only after all claims are removed is the key really removed.
909
910For example, if FS_IOC_ADD_ENCRYPTION_KEY was called with uid 1000,
911then the key will be "claimed" by uid 1000, and
912FS_IOC_REMOVE_ENCRYPTION_KEY will only succeed as uid 1000.  Or, if
913both uids 1000 and 2000 added the key, then for each uid
914FS_IOC_REMOVE_ENCRYPTION_KEY will only remove their own claim.  Only
915once *both* are removed is the key really removed.  (Think of it like
916unlinking a file that may have hard links.)
917
918If FS_IOC_REMOVE_ENCRYPTION_KEY really removes the key, it will also
919try to "lock" all files that had been unlocked with the key.  It won't
920lock files that are still in-use, so this ioctl is expected to be used
921in cooperation with userspace ensuring that none of the files are
922still open.  However, if necessary, this ioctl can be executed again
923later to retry locking any remaining files.
924
925FS_IOC_REMOVE_ENCRYPTION_KEY returns 0 if either the key was removed
926(but may still have files remaining to be locked), the user's claim to
927the key was removed, or the key was already removed but had files
928remaining to be the locked so the ioctl retried locking them.  In any
929of these cases, ``removal_status_flags`` is filled in with the
930following informational status flags:
931
932- ``FSCRYPT_KEY_REMOVAL_STATUS_FLAG_FILES_BUSY``: set if some file(s)
933  are still in-use.  Not guaranteed to be set in the case where only
934  the user's claim to the key was removed.
935- ``FSCRYPT_KEY_REMOVAL_STATUS_FLAG_OTHER_USERS``: set if only the
936  user's claim to the key was removed, not the key itself
937
938FS_IOC_REMOVE_ENCRYPTION_KEY can fail with the following errors:
939
940- ``EACCES``: The FSCRYPT_KEY_SPEC_TYPE_DESCRIPTOR key specifier type
941  was specified, but the caller does not have the CAP_SYS_ADMIN
942  capability in the initial user namespace
943- ``EINVAL``: invalid key specifier type, or reserved bits were set
944- ``ENOKEY``: the key object was not found at all, i.e. it was never
945  added in the first place or was already fully removed including all
946  files locked; or, the user does not have a claim to the key (but
947  someone else does).
948- ``ENOTTY``: this type of filesystem does not implement encryption
949- ``EOPNOTSUPP``: the kernel was not configured with encryption
950  support for this filesystem, or the filesystem superblock has not
951  had encryption enabled on it
952
953FS_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS
954~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
955
956FS_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS is exactly the same as
957`FS_IOC_REMOVE_ENCRYPTION_KEY`_, except that for v2 policy keys, the
958ALL_USERS version of the ioctl will remove all users' claims to the
959key, not just the current user's.  I.e., the key itself will always be
960removed, no matter how many users have added it.  This difference is
961only meaningful if non-root users are adding and removing keys.
962
963Because of this, FS_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS also requires
964"root", namely the CAP_SYS_ADMIN capability in the initial user
965namespace.  Otherwise it will fail with EACCES.
966
967Getting key status
968------------------
969
970FS_IOC_GET_ENCRYPTION_KEY_STATUS
971~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
972
973The FS_IOC_GET_ENCRYPTION_KEY_STATUS ioctl retrieves the status of a
974master encryption key.  It can be executed on any file or directory on
975the target filesystem, but using the filesystem's root directory is
976recommended.  It takes in a pointer to
977struct fscrypt_get_key_status_arg, defined as follows::
978
979    struct fscrypt_get_key_status_arg {
980            /* input */
981            struct fscrypt_key_specifier key_spec;
982            __u32 __reserved[6];
983
984            /* output */
985    #define FSCRYPT_KEY_STATUS_ABSENT               1
986    #define FSCRYPT_KEY_STATUS_PRESENT              2
987    #define FSCRYPT_KEY_STATUS_INCOMPLETELY_REMOVED 3
988            __u32 status;
989    #define FSCRYPT_KEY_STATUS_FLAG_ADDED_BY_SELF   0x00000001
990            __u32 status_flags;
991            __u32 user_count;
992            __u32 __out_reserved[13];
993    };
994
995The caller must zero all input fields, then fill in ``key_spec``:
996
997    - To get the status of a key for v1 encryption policies, set
998      ``key_spec.type`` to FSCRYPT_KEY_SPEC_TYPE_DESCRIPTOR and fill
999      in ``key_spec.u.descriptor``.
1000
1001    - To get the status of a key for v2 encryption policies, set
1002      ``key_spec.type`` to FSCRYPT_KEY_SPEC_TYPE_IDENTIFIER and fill
1003      in ``key_spec.u.identifier``.
1004
1005On success, 0 is returned and the kernel fills in the output fields:
1006
1007- ``status`` indicates whether the key is absent, present, or
1008  incompletely removed.  Incompletely removed means that the master
1009  secret has been removed, but some files are still in use; i.e.,
1010  `FS_IOC_REMOVE_ENCRYPTION_KEY`_ returned 0 but set the informational
1011  status flag FSCRYPT_KEY_REMOVAL_STATUS_FLAG_FILES_BUSY.
1012
1013- ``status_flags`` can contain the following flags:
1014
1015    - ``FSCRYPT_KEY_STATUS_FLAG_ADDED_BY_SELF`` indicates that the key
1016      has added by the current user.  This is only set for keys
1017      identified by ``identifier`` rather than by ``descriptor``.
1018
1019- ``user_count`` specifies the number of users who have added the key.
1020  This is only set for keys identified by ``identifier`` rather than
1021  by ``descriptor``.
1022
1023FS_IOC_GET_ENCRYPTION_KEY_STATUS can fail with the following errors:
1024
1025- ``EINVAL``: invalid key specifier type, or reserved bits were set
1026- ``ENOTTY``: this type of filesystem does not implement encryption
1027- ``EOPNOTSUPP``: the kernel was not configured with encryption
1028  support for this filesystem, or the filesystem superblock has not
1029  had encryption enabled on it
1030
1031Among other use cases, FS_IOC_GET_ENCRYPTION_KEY_STATUS can be useful
1032for determining whether the key for a given encrypted directory needs
1033to be added before prompting the user for the passphrase needed to
1034derive the key.
1035
1036FS_IOC_GET_ENCRYPTION_KEY_STATUS can only get the status of keys in
1037the filesystem-level keyring, i.e. the keyring managed by
1038`FS_IOC_ADD_ENCRYPTION_KEY`_ and `FS_IOC_REMOVE_ENCRYPTION_KEY`_.  It
1039cannot get the status of a key that has only been added for use by v1
1040encryption policies using the legacy mechanism involving
1041process-subscribed keyrings.
1042
1043Access semantics
1044================
1045
1046With the key
1047------------
1048
1049With the encryption key, encrypted regular files, directories, and
1050symlinks behave very similarly to their unencrypted counterparts ---
1051after all, the encryption is intended to be transparent.  However,
1052astute users may notice some differences in behavior:
1053
1054- Unencrypted files, or files encrypted with a different encryption
1055  policy (i.e. different key, modes, or flags), cannot be renamed or
1056  linked into an encrypted directory; see `Encryption policy
1057  enforcement`_.  Attempts to do so will fail with EXDEV.  However,
1058  encrypted files can be renamed within an encrypted directory, or
1059  into an unencrypted directory.
1060
1061  Note: "moving" an unencrypted file into an encrypted directory, e.g.
1062  with the `mv` program, is implemented in userspace by a copy
1063  followed by a delete.  Be aware that the original unencrypted data
1064  may remain recoverable from free space on the disk; prefer to keep
1065  all files encrypted from the very beginning.  The `shred` program
1066  may be used to overwrite the source files but isn't guaranteed to be
1067  effective on all filesystems and storage devices.
1068
1069- Direct I/O is supported on encrypted files only under some
1070  circumstances.  For details, see `Direct I/O support`_.
1071
1072- The fallocate operations FALLOC_FL_COLLAPSE_RANGE and
1073  FALLOC_FL_INSERT_RANGE are not supported on encrypted files and will
1074  fail with EOPNOTSUPP.
1075
1076- Online defragmentation of encrypted files is not supported.  The
1077  EXT4_IOC_MOVE_EXT and F2FS_IOC_MOVE_RANGE ioctls will fail with
1078  EOPNOTSUPP.
1079
1080- The ext4 filesystem does not support data journaling with encrypted
1081  regular files.  It will fall back to ordered data mode instead.
1082
1083- DAX (Direct Access) is not supported on encrypted files.
1084
1085- The maximum length of an encrypted symlink is 2 bytes shorter than
1086  the maximum length of an unencrypted symlink.  For example, on an
1087  EXT4 filesystem with a 4K block size, unencrypted symlinks can be up
1088  to 4095 bytes long, while encrypted symlinks can only be up to 4093
1089  bytes long (both lengths excluding the terminating null).
1090
1091Note that mmap *is* supported.  This is possible because the pagecache
1092for an encrypted file contains the plaintext, not the ciphertext.
1093
1094Without the key
1095---------------
1096
1097Some filesystem operations may be performed on encrypted regular
1098files, directories, and symlinks even before their encryption key has
1099been added, or after their encryption key has been removed:
1100
1101- File metadata may be read, e.g. using stat().
1102
1103- Directories may be listed, in which case the filenames will be
1104  listed in an encoded form derived from their ciphertext.  The
1105  current encoding algorithm is described in `Filename hashing and
1106  encoding`_.  The algorithm is subject to change, but it is
1107  guaranteed that the presented filenames will be no longer than
1108  NAME_MAX bytes, will not contain the ``/`` or ``\0`` characters, and
1109  will uniquely identify directory entries.
1110
1111  The ``.`` and ``..`` directory entries are special.  They are always
1112  present and are not encrypted or encoded.
1113
1114- Files may be deleted.  That is, nondirectory files may be deleted
1115  with unlink() as usual, and empty directories may be deleted with
1116  rmdir() as usual.  Therefore, ``rm`` and ``rm -r`` will work as
1117  expected.
1118
1119- Symlink targets may be read and followed, but they will be presented
1120  in encrypted form, similar to filenames in directories.  Hence, they
1121  are unlikely to point to anywhere useful.
1122
1123Without the key, regular files cannot be opened or truncated.
1124Attempts to do so will fail with ENOKEY.  This implies that any
1125regular file operations that require a file descriptor, such as
1126read(), write(), mmap(), fallocate(), and ioctl(), are also forbidden.
1127
1128Also without the key, files of any type (including directories) cannot
1129be created or linked into an encrypted directory, nor can a name in an
1130encrypted directory be the source or target of a rename, nor can an
1131O_TMPFILE temporary file be created in an encrypted directory.  All
1132such operations will fail with ENOKEY.
1133
1134It is not currently possible to backup and restore encrypted files
1135without the encryption key.  This would require special APIs which
1136have not yet been implemented.
1137
1138Encryption policy enforcement
1139=============================
1140
1141After an encryption policy has been set on a directory, all regular
1142files, directories, and symbolic links created in that directory
1143(recursively) will inherit that encryption policy.  Special files ---
1144that is, named pipes, device nodes, and UNIX domain sockets --- will
1145not be encrypted.
1146
1147Except for those special files, it is forbidden to have unencrypted
1148files, or files encrypted with a different encryption policy, in an
1149encrypted directory tree.  Attempts to link or rename such a file into
1150an encrypted directory will fail with EXDEV.  This is also enforced
1151during ->lookup() to provide limited protection against offline
1152attacks that try to disable or downgrade encryption in known locations
1153where applications may later write sensitive data.  It is recommended
1154that systems implementing a form of "verified boot" take advantage of
1155this by validating all top-level encryption policies prior to access.
1156
1157Inline encryption support
1158=========================
1159
1160By default, fscrypt uses the kernel crypto API for all cryptographic
1161operations (other than HKDF, which fscrypt partially implements
1162itself).  The kernel crypto API supports hardware crypto accelerators,
1163but only ones that work in the traditional way where all inputs and
1164outputs (e.g. plaintexts and ciphertexts) are in memory.  fscrypt can
1165take advantage of such hardware, but the traditional acceleration
1166model isn't particularly efficient and fscrypt hasn't been optimized
1167for it.
1168
1169Instead, many newer systems (especially mobile SoCs) have *inline
1170encryption hardware* that can encrypt/decrypt data while it is on its
1171way to/from the storage device.  Linux supports inline encryption
1172through a set of extensions to the block layer called *blk-crypto*.
1173blk-crypto allows filesystems to attach encryption contexts to bios
1174(I/O requests) to specify how the data will be encrypted or decrypted
1175in-line.  For more information about blk-crypto, see
1176:ref:`Documentation/block/inline-encryption.rst <inline_encryption>`.
1177
1178On supported filesystems (currently ext4 and f2fs), fscrypt can use
1179blk-crypto instead of the kernel crypto API to encrypt/decrypt file
1180contents.  To enable this, set CONFIG_FS_ENCRYPTION_INLINE_CRYPT=y in
1181the kernel configuration, and specify the "inlinecrypt" mount option
1182when mounting the filesystem.
1183
1184Note that the "inlinecrypt" mount option just specifies to use inline
1185encryption when possible; it doesn't force its use.  fscrypt will
1186still fall back to using the kernel crypto API on files where the
1187inline encryption hardware doesn't have the needed crypto capabilities
1188(e.g. support for the needed encryption algorithm and data unit size)
1189and where blk-crypto-fallback is unusable.  (For blk-crypto-fallback
1190to be usable, it must be enabled in the kernel configuration with
1191CONFIG_BLK_INLINE_ENCRYPTION_FALLBACK=y.)
1192
1193Currently fscrypt always uses the filesystem block size (which is
1194usually 4096 bytes) as the data unit size.  Therefore, it can only use
1195inline encryption hardware that supports that data unit size.
1196
1197Inline encryption doesn't affect the ciphertext or other aspects of
1198the on-disk format, so users may freely switch back and forth between
1199using "inlinecrypt" and not using "inlinecrypt".
1200
1201Direct I/O support
1202==================
1203
1204For direct I/O on an encrypted file to work, the following conditions
1205must be met (in addition to the conditions for direct I/O on an
1206unencrypted file):
1207
1208* The file must be using inline encryption.  Usually this means that
1209  the filesystem must be mounted with ``-o inlinecrypt`` and inline
1210  encryption hardware must be present.  However, a software fallback
1211  is also available.  For details, see `Inline encryption support`_.
1212
1213* The I/O request must be fully aligned to the filesystem block size.
1214  This means that the file position the I/O is targeting, the lengths
1215  of all I/O segments, and the memory addresses of all I/O buffers
1216  must be multiples of this value.  Note that the filesystem block
1217  size may be greater than the logical block size of the block device.
1218
1219If either of the above conditions is not met, then direct I/O on the
1220encrypted file will fall back to buffered I/O.
1221
1222Implementation details
1223======================
1224
1225Encryption context
1226------------------
1227
1228An encryption policy is represented on-disk by
1229struct fscrypt_context_v1 or struct fscrypt_context_v2.  It is up to
1230individual filesystems to decide where to store it, but normally it
1231would be stored in a hidden extended attribute.  It should *not* be
1232exposed by the xattr-related system calls such as getxattr() and
1233setxattr() because of the special semantics of the encryption xattr.
1234(In particular, there would be much confusion if an encryption policy
1235were to be added to or removed from anything other than an empty
1236directory.)  These structs are defined as follows::
1237
1238    #define FSCRYPT_FILE_NONCE_SIZE 16
1239
1240    #define FSCRYPT_KEY_DESCRIPTOR_SIZE  8
1241    struct fscrypt_context_v1 {
1242            u8 version;
1243            u8 contents_encryption_mode;
1244            u8 filenames_encryption_mode;
1245            u8 flags;
1246            u8 master_key_descriptor[FSCRYPT_KEY_DESCRIPTOR_SIZE];
1247            u8 nonce[FSCRYPT_FILE_NONCE_SIZE];
1248    };
1249
1250    #define FSCRYPT_KEY_IDENTIFIER_SIZE  16
1251    struct fscrypt_context_v2 {
1252            u8 version;
1253            u8 contents_encryption_mode;
1254            u8 filenames_encryption_mode;
1255            u8 flags;
1256            u8 __reserved[4];
1257            u8 master_key_identifier[FSCRYPT_KEY_IDENTIFIER_SIZE];
1258            u8 nonce[FSCRYPT_FILE_NONCE_SIZE];
1259    };
1260
1261The context structs contain the same information as the corresponding
1262policy structs (see `Setting an encryption policy`_), except that the
1263context structs also contain a nonce.  The nonce is randomly generated
1264by the kernel and is used as KDF input or as a tweak to cause
1265different files to be encrypted differently; see `Per-file encryption
1266keys`_ and `DIRECT_KEY policies`_.
1267
1268Data path changes
1269-----------------
1270
1271When inline encryption is used, filesystems just need to associate
1272encryption contexts with bios to specify how the block layer or the
1273inline encryption hardware will encrypt/decrypt the file contents.
1274
1275When inline encryption isn't used, filesystems must encrypt/decrypt
1276the file contents themselves, as described below:
1277
1278For the read path (->read_folio()) of regular files, filesystems can
1279read the ciphertext into the page cache and decrypt it in-place.  The
1280folio lock must be held until decryption has finished, to prevent the
1281folio from becoming visible to userspace prematurely.
1282
1283For the write path (->writepage()) of regular files, filesystems
1284cannot encrypt data in-place in the page cache, since the cached
1285plaintext must be preserved.  Instead, filesystems must encrypt into a
1286temporary buffer or "bounce page", then write out the temporary
1287buffer.  Some filesystems, such as UBIFS, already use temporary
1288buffers regardless of encryption.  Other filesystems, such as ext4 and
1289F2FS, have to allocate bounce pages specially for encryption.
1290
1291Filename hashing and encoding
1292-----------------------------
1293
1294Modern filesystems accelerate directory lookups by using indexed
1295directories.  An indexed directory is organized as a tree keyed by
1296filename hashes.  When a ->lookup() is requested, the filesystem
1297normally hashes the filename being looked up so that it can quickly
1298find the corresponding directory entry, if any.
1299
1300With encryption, lookups must be supported and efficient both with and
1301without the encryption key.  Clearly, it would not work to hash the
1302plaintext filenames, since the plaintext filenames are unavailable
1303without the key.  (Hashing the plaintext filenames would also make it
1304impossible for the filesystem's fsck tool to optimize encrypted
1305directories.)  Instead, filesystems hash the ciphertext filenames,
1306i.e. the bytes actually stored on-disk in the directory entries.  When
1307asked to do a ->lookup() with the key, the filesystem just encrypts
1308the user-supplied name to get the ciphertext.
1309
1310Lookups without the key are more complicated.  The raw ciphertext may
1311contain the ``\0`` and ``/`` characters, which are illegal in
1312filenames.  Therefore, readdir() must base64url-encode the ciphertext
1313for presentation.  For most filenames, this works fine; on ->lookup(),
1314the filesystem just base64url-decodes the user-supplied name to get
1315back to the raw ciphertext.
1316
1317However, for very long filenames, base64url encoding would cause the
1318filename length to exceed NAME_MAX.  To prevent this, readdir()
1319actually presents long filenames in an abbreviated form which encodes
1320a strong "hash" of the ciphertext filename, along with the optional
1321filesystem-specific hash(es) needed for directory lookups.  This
1322allows the filesystem to still, with a high degree of confidence, map
1323the filename given in ->lookup() back to a particular directory entry
1324that was previously listed by readdir().  See
1325struct fscrypt_nokey_name in the source for more details.
1326
1327Note that the precise way that filenames are presented to userspace
1328without the key is subject to change in the future.  It is only meant
1329as a way to temporarily present valid filenames so that commands like
1330``rm -r`` work as expected on encrypted directories.
1331
1332Tests
1333=====
1334
1335To test fscrypt, use xfstests, which is Linux's de facto standard
1336filesystem test suite.  First, run all the tests in the "encrypt"
1337group on the relevant filesystem(s).  One can also run the tests
1338with the 'inlinecrypt' mount option to test the implementation for
1339inline encryption support.  For example, to test ext4 and
1340f2fs encryption using `kvm-xfstests
1341<https://github.com/tytso/xfstests-bld/blob/master/Documentation/kvm-quickstart.md>`_::
1342
1343    kvm-xfstests -c ext4,f2fs -g encrypt
1344    kvm-xfstests -c ext4,f2fs -g encrypt -m inlinecrypt
1345
1346UBIFS encryption can also be tested this way, but it should be done in
1347a separate command, and it takes some time for kvm-xfstests to set up
1348emulated UBI volumes::
1349
1350    kvm-xfstests -c ubifs -g encrypt
1351
1352No tests should fail.  However, tests that use non-default encryption
1353modes (e.g. generic/549 and generic/550) will be skipped if the needed
1354algorithms were not built into the kernel's crypto API.  Also, tests
1355that access the raw block device (e.g. generic/399, generic/548,
1356generic/549, generic/550) will be skipped on UBIFS.
1357
1358Besides running the "encrypt" group tests, for ext4 and f2fs it's also
1359possible to run most xfstests with the "test_dummy_encryption" mount
1360option.  This option causes all new files to be automatically
1361encrypted with a dummy key, without having to make any API calls.
1362This tests the encrypted I/O paths more thoroughly.  To do this with
1363kvm-xfstests, use the "encrypt" filesystem configuration::
1364
1365    kvm-xfstests -c ext4/encrypt,f2fs/encrypt -g auto
1366    kvm-xfstests -c ext4/encrypt,f2fs/encrypt -g auto -m inlinecrypt
1367
1368Because this runs many more tests than "-g encrypt" does, it takes
1369much longer to run; so also consider using `gce-xfstests
1370<https://github.com/tytso/xfstests-bld/blob/master/Documentation/gce-xfstests.md>`_
1371instead of kvm-xfstests::
1372
1373    gce-xfstests -c ext4/encrypt,f2fs/encrypt -g auto
1374    gce-xfstests -c ext4/encrypt,f2fs/encrypt -g auto -m inlinecrypt
1375