ThePassionate opened a new pull request, #21619:
URL: https://github.com/apache/nuttx/pull/21619

   ## Summary
   
   Two framework steps that the rpmsg / virtio / vhost crypto transports (to 
follow in separate PRs) build on, plus one style-only cleanup.
   
   * crypto: split backend registration from the /dev/crypto interface
     `devcrypto_register()` created the /dev/crypto character device and also 
registered the software and hardware crypto backends into the driver table. The 
two serve different users: the driver table is what in-kernel callers of 
`crypto_newsession()` / `crypto_invoke()` use, while /dev/crypto is the BSD 
compatible ioctl interface for user space. The backend registration 
(`swcr_init()`, `swkey_init()`, `hwcr_init()`) moves to a new 
`crypto_backends_init()` in crypto/backends.c, and `drivers_initialize()` calls 
the two independently, so a build can have the backends without compiling 
cryptodev.c. The new hidden `CRYPTO_BACKENDS` option is selected by 
`CRYPTO_CRYPTODEV`, and the software cipher / software key management / 
hardware options now depend on it instead of `CRYPTO_CRYPTODEV`.
   
   * crypto/cryptodev: pass operation, key, padding and hash type to kprocess
     `struct crypt_kop` only carried the operation code and the bignum 
parameters, so a backend that receives a key operation could not tell whether 
an RSA request is an encryption or a signature, whether the key is public or 
private, or which padding and digest the caller wants. Drivers that forward the 
request to another processor or to a hardware engine need that information to 
describe the operation to the other side. Add `crk_optype`, `crk_keytype`, 
`crk_padding` and `crk_hash` to `struct crypt_kop`, the matching `krp_*` fields 
to `struct cryptkop`, and the value definitions (`CRYPTO_OP_SIGN` / `VERIFY` / 
`GENKEY`, `CRYPTO_KEY_TYPE_*`, `CRYPTO_RSA_*_PADDING`, `CRYPTO_RSA_<hash>`). 
All new fields are zero for callers that clear the struct before use, and zero 
means "decrypt, no key type, raw padding, no hash", so backends that do not 
look at them behave exactly as before.
   
   * include/crypto: fix an nxstyle error in cryptodev.h
     Style-only: one long-standing declaration on two lines is split so the 
file passes `tools/checkpatch.sh` as-is.
   
   ## Impact
   
   * No configuration break: `CRYPTO_CRYPTODEV` selects `CRYPTO_BACKENDS`, so 
every existing defconfig (e.g. sim:crypto) builds unchanged; no defconfig 
changes are needed.
   * New optional fields in `struct crypt_kop` / `struct cryptkop`, zero-filled 
meaning "as before"; no behavioral change for existing callers or backends.
   * In-kernel crypto users (rpmsg crypto, virtio-crypto, vhost-crypto) will 
select `CRYPTO_BACKENDS` and consume the kprocess type fields; those drivers 
are not in tree yet and follow in separate PRs.
   
   ## Testing
   
   Host: Ubuntu 22.04.5 LTS, x86_64, gcc 13.1.0
   Board: sim:crypto
   nuttx: this branch on top of master 066314c6373
   apps: master 8c81529153 + apache/nuttx-apps#3996 
(https://github.com/apache/nuttx-apps/pull/3996)
   
   Build steps:
   
   ```
   ./tools/configure.sh -l sim:crypto
   # extra options so that the new backend registration and key management 
paths compile
   CONFIG_CRYPTO_CRYPTODEV_SOFTWARE_KEYMGMT=y
   CONFIG_MTD=y
   CONFIG_MTD_CONFIG=y
   CONFIG_MTD_CONFIG_NAMED=y
   CONFIG_NSH_BUILTIN_APPS=y
   make olddefconfig
   make
   ```
   
   Result: build completes with no errors and no new warnings; `.config` shows 
`CRYPTO_BACKENDS=y` selected by `CRYPTO_CRYPTODEV=y`.
   
   Run `./nuttx` and the tests from apps/testing/drivers/crypto:
   
   ```
   nsh> aescbc
   aescbc test ok
   nsh> aesctr
   OK test vector 0..8
   nsh> aesxts
   OK encrypt/decrypt test vector 0..13
   nsh> 3descbc
   ok, encrypt with /dev/crypto, decrypt with /dev/crypto
   nsh> chacha20
   chacha20: all tests passed
   nsh> chachapoly
   chachapoly: all tests passed
   nsh> hmac
   hmac md5/sha1/sha224/sha256 success (20 vectors, includes 131-byte key cases)
   nsh> hash
   (hash vectors pass)
   ```
   
   The symmetric suite and hmac/hash pass. rsa / ecdsa / dhm, which exercise 
the kprocess path through the new fields, are covered by zero-filled structs 
and behave as before; the dedicated asymmetric test runs are appended below.
   
   `tools/checkpatch.sh -c -u -m -g $(git merge-base origin/master HEAD)..HEAD` 
passes.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to