6.3 Cryptographic Functions

Some functions operating on S-expressions support ‘flags’ to influence the operation. These flags have to be listed in a sub-S-expression named ‘flags’. Flag names are case-sensitive. The following flags are known:

comp

nocomp

If supported by the algorithm and curve the comp flag requests that points are returned in compact (compressed) representation. The nocomp flag requests that points are returned with full coordinates. The default depends on the the algorithm and curve. The compact representation requires a small overhead before a point can be used but halves the size of a to be conveyed public key. If comp is used with the “EdDSA” algorithm the key generation prefix the public key with a 0x40 byte.

pkcs1

Use PKCS#1 block type 2 padding for encryption, block type 1 padding for signing.

oaep

Use RSA-OAEP padding for encryption.

pss

Use RSA-PSS padding for signing.

eddsa

Use the EdDSA scheme signing instead of the default ECDSA algorithm. Note that the EdDSA uses a special form of the public key.

rfc6979

For DSA and ECDSA use a deterministic scheme for the k parameter.

no-blinding

Do not use a technique called ‘blinding’, which is used by default in order to prevent leaking of secret information. Blinding is only implemented by RSA, but it might be implemented by other algorithms in the future as well, when necessary.

param

For ECC key generation also return the domain parameters. For ECC signing and verification override default parameters by provided domain parameters of the public or private key.

transient-key

This flag is only meaningful for RSA, DSA, and ECC key generation. If given the key is created using a faster and a somewhat less secure random number generator. This flag may be used for keys which are only used for a short time or per-message and do not require full cryptographic strength.

no-keytest

This flag skips internal failsafe tests to assert that a generated key is properly working. It currently has an effect only for standard ECC key generation. It is mostly useful along with transient-key to achieve fastest ECC key generation.

use-x931

Force the use of the ANSI X9.31 key generation algorithm instead of the default algorithm. This flag is only meaningful for RSA key generation and usually not required. Note that this algorithm is implicitly used if either derive-parms is given or Libgcrypt is in FIPS mode.

use-fips186

Force the use of the FIPS 186 key generation algorithm instead of the default algorithm. This flag is only meaningful for DSA and usually not required. Note that this algorithm is implicitly used if either derive-parms is given or Libgcrypt is in FIPS mode. As of now FIPS 186-2 is implemented; after the approval of FIPS 186-3 the code will be changed to implement 186-3.

use-fips186-2

Force the use of the FIPS 186-2 key generation algorithm instead of the default algorithm. This algorithm is slightly different from FIPS 186-3 and allows only 1024 bit keys. This flag is only meaningful for DSA and only required for FIPS testing backward compatibility.

Now that we know the key basics, we can carry on and explain how to encrypt and decrypt data. In almost all cases the data is a random session key which is in turn used for the actual encryption of the real data. There are 2 functions to do this:

Function: gcry_error_t gcry_pk_encrypt (gcry_sexp_t *r_ciph, gcry_sexp_t data, gcry_sexp_t pkey)

Obviously a public key must be provided for encryption. It is expected as an appropriate S-expression (see above) in pkey. The data to be encrypted can either be in the simple old format, which is a very simple S-expression consisting only of one MPI, or it may be a more complex S-expression which also allows to specify flags for operation, like e.g. padding rules.

If you don’t want to let Libgcrypt handle the padding, you must pass an appropriate MPI using this expression for data:

(data
  (flags raw)
  (value mpi))

This has the same semantics as the old style MPI only way. MPI is the actual data, already padded appropriate for your protocol. Most RSA based systems however use PKCS#1 padding and so you can use this S-expression for data:

(data
  (flags pkcs1)
  (value block))

Here, the "flags" list has the "pkcs1" flag which let the function know that it should provide PKCS#1 block type 2 padding. The actual data to be encrypted is passed as a string of octets in block. The function checks that this data actually can be used with the given key, does the padding and encrypts it.

If the function could successfully perform the encryption, the return value will be 0 and a new S-expression with the encrypted result is allocated and assigned to the variable at the address of r_ciph. The caller is responsible to release this value using gcry_sexp_release. In case of an error, an error code is returned and r_ciph will be set to NULL.

The returned S-expression has this format when used with RSA:

(enc-val
  (rsa
    (a a-mpi)))

Where a-mpi is an MPI with the result of the RSA operation. When using the Elgamal algorithm, the return value will have this format:

(enc-val
  (elg
    (a a-mpi)
    (b b-mpi)))

Where a-mpi and b-mpi are MPIs with the result of the Elgamal encryption operation.

Function: gcry_error_t gcry_pk_decrypt (gcry_sexp_t *r_plain, gcry_sexp_t data, gcry_sexp_t skey)

Obviously a private key must be provided for decryption. It is expected as an appropriate S-expression (see above) in skey. The data to be decrypted must match the format of the result as returned by gcry_pk_encrypt, but should be enlarged with a flags element:

(enc-val
  (flags)
  (elg
    (a a-mpi)
    (b b-mpi)))

This function does not remove padding from the data by default. To let Libgcrypt remove padding, give a hint in ‘flags’ telling which padding method was used when encrypting:

(flags padding-method)

Currently padding-method is either pkcs1 for PKCS#1 block type 2 padding, or oaep for RSA-OAEP padding.

The function returns 0 on success or an error code. The variable at the address of r_plain will be set to NULL on error or receive the decrypted value on success. The format of r_plain is a simple S-expression part (i.e. not a valid one) with just one MPI if there was no flags element in data; if at least an empty flags is passed in data, the format is:

(value plaintext)

Another operation commonly performed using public key cryptography is signing data. In some sense this is even more important than encryption because digital signatures are an important instrument for key management. Libgcrypt supports digital signatures using 2 functions, similar to the encryption functions:

Function: gcry_error_t gcry_pk_sign (gcry_sexp_t *r_sig, gcry_sexp_t data, gcry_sexp_t skey)

This function creates a digital signature for data using the private key skey and place it into the variable at the address of r_sig. data may either be the simple old style S-expression with just one MPI or a modern and more versatile S-expression which allows to let Libgcrypt handle padding:

 (data
  (flags pkcs1)
  (hash hash-algo block))

This example requests to sign the data in block after applying PKCS#1 block type 1 style padding. hash-algo is a string with the hash algorithm to be encoded into the signature, this may be any hash algorithm name as supported by Libgcrypt. Most likely, this will be "sha256" or "sha1". It is obvious that the length of block must match the size of that message digests; the function checks that this and other constraints are valid.

If PKCS#1 padding is not required (because the caller does already provide a padded value), either the old format or better the following format should be used:

(data
  (flags raw)
  (value mpi))

Here, the data to be signed is directly given as an MPI.

For DSA the input data is expected in this format:

(data
  (flags raw)
  (value mpi))

Here, the data to be signed is directly given as an MPI. It is expect that this MPI is the the hash value. For the standard DSA using a MPI is not a problem in regard to leading zeroes because the hash value is directly used as an MPI. For better standard conformance it would be better to explicit use a memory string (like with pkcs1) but that is currently not supported. However, for deterministic DSA as specified in RFC6979 this can’t be used. Instead the following input is expected.

(data
  (flags rfc6979)
  (hash hash-algo block))

Note that the provided hash-algo is used for the internal HMAC; it should match the hash-algo used to create block.

The signature is returned as a newly allocated S-expression in r_sig using this format for RSA:

(sig-val
  (rsa
    (s s-mpi)))

Where s-mpi is the result of the RSA sign operation. For DSA the S-expression returned is:

(sig-val
  (dsa
    (r r-mpi)
    (s s-mpi)))

Where r-mpi and s-mpi are the result of the DSA sign operation.

For Elgamal signing (which is slow, yields large numbers and probably is not as secure as the other algorithms), the same format is used with "elg" replacing "dsa"; for ECDSA signing, the same format is used with "ecdsa" replacing "dsa".

For the EdDSA algorithm (cf. Ed25515) the required input parameters are:

(data
  (flags eddsa)
  (hash-algo sha512)
  (value message))

Note that the message may be of any length; hashing is part of the algorithm. Using a large data block for message is not suggested; in that case the used protocol should better require that a hash of the message is used as input to the EdDSA algorithm.

The operation most commonly used is definitely the verification of a signature. Libgcrypt provides this function:

Function: gcry_error_t gcry_pk_verify (gcry_sexp_t sig, gcry_sexp_t data, gcry_sexp_t pkey)

This is used to check whether the signature sig matches the data. The public key pkey must be provided to perform this verification. This function is similar in its parameters to gcry_pk_sign with the exceptions that the public key is used instead of the private key and that no signature is created but a signature, in a format as created by gcry_pk_sign, is passed to the function in sig.

The result is 0 for success (i.e. the data matches the signature), or an error code where the most relevant code is GCRY_ERR_BAD_SIGNATURE to indicate that the signature does not match the provided data.