On Tue, Jul 14, 2026 at 08:57:24AM +0000, Jamin Lin wrote: > Hi Daniel > > > Subject: Re: [PATCH v1 09/15] crypto/cipher-gcrypt: Implement AES-GCM > > > > On Tue, Jul 14, 2026 at 07:29:14AM +0000, Jamin Lin wrote: > > > Map QCRYPTO_CIPHER_MODE_GCM to GCRY_CIPHER_MODE_GCM and > > advertise it > > > in > > > qcrypto_cipher_supports() for 128-bit block ciphers. Add a GCM driver > > > whose setiv accepts the (typically 96-bit) nonce, whose > > > encrypt/decrypt do not require block-aligned lengths, and which > > > implements setaad via > > > gcry_cipher_authenticate() and gettag via gcry_cipher_gettag(). > > > > > > Signed-off-by: Jamin Lin <[email protected]> > > > --- > > > crypto/cipher-gcrypt.c.inc | 101 > > > +++++++++++++++++++++++++++++++++++++ > > > 1 file changed, 101 insertions(+) > > > > > > diff --git a/crypto/cipher-gcrypt.c.inc b/crypto/cipher-gcrypt.c.inc > > > index 12eb9ddb5a..fce09a3c77 100644 > > > --- a/crypto/cipher-gcrypt.c.inc > > > +++ b/crypto/cipher-gcrypt.c.inc > > > @@ -65,6 +65,8 @@ static int > > qcrypto_cipher_mode_to_gcry_mode(QCryptoCipherMode mode) > > > return GCRY_CIPHER_MODE_CBC; > > > case QCRYPTO_CIPHER_MODE_CTR: > > > return GCRY_CIPHER_MODE_CTR; > > > + case QCRYPTO_CIPHER_MODE_GCM: > > > + return GCRY_CIPHER_MODE_GCM; > > > default: > > > return GCRY_CIPHER_MODE_NONE; > > > } > > > @@ -104,6 +106,10 @@ bool qcrypto_cipher_supports(QCryptoCipherAlgo > > alg, > > > case QCRYPTO_CIPHER_MODE_XTS: > > > case QCRYPTO_CIPHER_MODE_CTR: > > > return true; > > > + case QCRYPTO_CIPHER_MODE_GCM: > > > + /* GCM requires a 128-bit block cipher. */ > > > + return gcry_cipher_get_algo_blklen( > > > + qcrypto_cipher_alg_to_gcry_alg(alg)) == 16; > > > default: > > > return false; > > > } > > > @@ -228,6 +234,99 @@ static const struct QCryptoCipherDriver > > qcrypto_gcrypt_ctr_driver = { > > > .cipher_free = qcrypto_gcrypt_ctx_free, }; > > > > > > +/* > > > + * GCM is an AEAD stream mode: the IV/nonce need not match the block > > > +size, > > > + * the message length need not be a multiple of the block size, > > > +associated > > > + * data is fed with gcry_cipher_authenticate() and the authentication > > > +tag is > > > + * read back with gcry_cipher_gettag(). > > > + */ > > > +static int qcrypto_gcrypt_gcm_setiv(QCryptoCipher *cipher, > > > + const uint8_t *iv, size_t niv, > > > + Error **errp) { > > > + QCryptoCipherGcrypt *ctx = container_of(cipher, > > QCryptoCipherGcrypt, base); > > > + gcry_error_t err; > > > + > > > + gcry_cipher_reset(ctx->handle); > > > + err = gcry_cipher_setiv(ctx->handle, iv, niv); > > > + if (err != 0) { > > > + error_setg(errp, "Cannot set IV: %s", gcry_strerror(err)); > > > + return -1; > > > + } > > > + > > > + return 0; > > > +} > > > > > > +static int qcrypto_gcrypt_gcm_encrypt(QCryptoCipher *cipher, const void > > *in, > > > + void *out, size_t len, Error > > > +**errp) { > > > + QCryptoCipherGcrypt *ctx = container_of(cipher, > > QCryptoCipherGcrypt, base); > > > + gcry_error_t err; > > > + > > > + err = gcry_cipher_encrypt(ctx->handle, out, len, in, len); > > > + if (err != 0) { > > > + error_setg(errp, "Cannot encrypt data: %s", gcry_strerror(err)); > > > + return -1; > > > + } > > > + > > > + return 0; > > > +} > > > + > > > +static int qcrypto_gcrypt_gcm_decrypt(QCryptoCipher *cipher, const void > > *in, > > > + void *out, size_t len, Error > > > +**errp) { > > > + QCryptoCipherGcrypt *ctx = container_of(cipher, > > QCryptoCipherGcrypt, base); > > > + gcry_error_t err; > > > + > > > + err = gcry_cipher_decrypt(ctx->handle, out, len, in, len); > > > + if (err != 0) { > > > + error_setg(errp, "Cannot decrypt data: %s", gcry_strerror(err)); > > > + return -1; > > > + } > > > + > > > + return 0; > > > +} > > > > Is there a reason you can't use the common qcrypto_gcrypt_setiv, > > qcrypto_gcrypt_decrypt and qcrypto_gcrypt_encrypt methods ? > > > > Thanks for the review and suggestion. > > The GCM path differs from the common methods in two input checks that GCM > legitimately violates: > > - qcrypto_gcrypt_setiv() rejects niv != blocksize, but GCM uses a > 96-bit (12-byte) nonce. > - qcrypto_gcrypt_encrypt()/decrypt() reject a length that is not a > multiple of the block size, but GCM must accept arbitrary lengths > (e.g. the 60-byte NIST test vectors). > > So reusing the common methods means relaxing those two checks for GCM, > e.g.: > /* qcrypto_gcrypt_setiv() */ > if (cipher->mode != QCRYPTO_CIPHER_MODE_GCM && niv != ctx->blocksize) { > error_setg(errp, "Expected IV size %zu not %zu", ctx->blocksize, niv); > return -1; > } > /* qcrypto_gcrypt_encrypt() / _decrypt() */ > if (cipher->mode != QCRYPTO_CIPHER_MODE_GCM && > (len & (ctx->blocksize - 1))) { > error_setg(errp, "Length %zu must be a multiple of block size %zu", > len, ctx->blocksize); > return -1; > } > > I kept them separate to avoid adding GCM special-cases to the shared > block-cipher path, but I'm happy to switch the GCM driver to the common > qcrypto_gcrypt_{setiv,encrypt,decrypt} (dropping the GCM-specific ones and > keeping only the AEAD setaad/gettag hooks) > > If you prefer. I'll make that change in v2 if that's the direction you'd like.
No, leave it as it is. Reviewed-by: Daniel P. Berrangé <[email protected]> Acked-by: Daniel P. Berrangé <[email protected]> With regards, Daniel -- |: https://berrange.com ~~ https://hachyderm.io/@berrange :| |: https://libvirt.org ~~ https://entangle-photo.org :| |: https://pixelfed.art/berrange ~~ https://fstop138.berrange.com :|
