| Message ID | 1524752664-27946-1-git-send-email-selva.nair@gmail.com |
|---|---|
| State | Accepted |
| Headers | show |
| Series | [Openvpn-devel] Pass the hash without the DigestInfo header to NCryptSignHash() | expand |
Hi, On 26-04-18 16:24, selva.nair@gmail.com wrote: > From: Selva Nair <selva.nair@gmail.com> > > In case of TLS 1.2 signatures, the callback rsa_priv_enc() gets > the hash with the DigestInfo prepended. Signing this using > NCryptSignHash() with hash algorithm id set to NULL works in most cases. > But when using some hardware tokens, the data gets interpreted as the pre > TLS 1.2 MD5+SHA1 hash and is silently truncated to 36 bytes. > Avoid this by passing the raw hash to NCryptSignHash() and let it > add the DigestInfo. > > To get the raw hash we set the RSA_sign() method in the rsa_method > structure. This callback bypasses rsa_priv_enc() and gets called with > the hash type and the hash. > > Fixes Trac #1050 > --- > > Tested by JJK and gizi as described in Trac #1050 > > src/openvpn/cryptoapi.c | 84 ++++++++++++++++++++++++++++++++++++++++++++++--- > 1 file changed, 79 insertions(+), 5 deletions(-) > > diff --git a/src/openvpn/cryptoapi.c b/src/openvpn/cryptoapi.c > index 89d253c..d765954 100644 > --- a/src/openvpn/cryptoapi.c > +++ b/src/openvpn/cryptoapi.c > @@ -217,14 +217,16 @@ rsa_pub_dec(int flen, const unsigned char *from, unsigned char *to, RSA *rsa, in > * Sign the hash in 'from' using NCryptSignHash(). This requires an NCRYPT > * key handle in cd->crypt_prov. On return the signature is in 'to'. Returns > * the length of the signature or 0 on error. > + * If the hash_algo is not NULL, PKCS #1 DigestInfo header gets added > + * to 'from', else it is signed as is. > * For now we support only RSA and the padding is assumed to be PKCS1 v1.5 > */ > static int > -priv_enc_CNG(const CAPI_DATA *cd, const unsigned char *from, int flen, > - unsigned char *to, int tlen, int padding) > +priv_enc_CNG(const CAPI_DATA *cd, const wchar_t *hash_algo, const unsigned char *from, > + int flen, unsigned char *to, int tlen, int padding) > { > NCRYPT_KEY_HANDLE hkey = cd->crypt_prov; > - DWORD len; > + DWORD len = 0; > ASSERT(cd->key_spec == CERT_NCRYPT_KEY_SPEC); > > msg(D_LOW, "Signing hash using CNG: data size = %d", flen); > @@ -232,7 +234,7 @@ priv_enc_CNG(const CAPI_DATA *cd, const unsigned char *from, int flen, > /* The hash OID is already in 'from'. So set the hash algorithm > * in the padding info struct to NULL. > */ > - BCRYPT_PKCS1_PADDING_INFO padinfo = {NULL}; > + BCRYPT_PKCS1_PADDING_INFO padinfo = {hash_algo}; > DWORD status; > > status = NCryptSignHash(hkey, padding? &padinfo : NULL, (BYTE*) from, flen, > @@ -270,7 +272,7 @@ rsa_priv_enc(int flen, const unsigned char *from, unsigned char *to, RSA *rsa, i > } > if (cd->key_spec == CERT_NCRYPT_KEY_SPEC) > { > - return priv_enc_CNG(cd, from, flen, to, RSA_size(rsa), padding); > + return priv_enc_CNG(cd, NULL, from, flen, to, RSA_size(rsa), padding); > } > > /* Unfortunately, there is no "CryptSign()" function in CryptoAPI, that would > @@ -334,6 +336,69 @@ rsa_priv_enc(int flen, const unsigned char *from, unsigned char *to, RSA *rsa, i > return len; > } > > +/* > + * Sign the hash in |m| and return the signature in |sig|. > + * Returns 1 on success, 0 on error. > + * NCryptSignHash() is used to sign and it is instructed to add the > + * the PKCS #1 DigestInfo header to |m| unless the hash algorithm is > + * the MD5/SHA1 combination used in TLS 1.1 and earlier versions. > + */ > +static int > +rsa_sign_CNG(int type, const unsigned char *m, unsigned int m_len, > + unsigned char *sig, unsigned int *siglen, const RSA *rsa) > +{ > + CAPI_DATA *cd = (CAPI_DATA *) RSA_meth_get0_app_data(RSA_get_method(rsa)); > + const wchar_t *alg = NULL; > + int padding = RSA_PKCS1_PADDING; > + > + *siglen = 0; > + if (cd == NULL) > + { > + RSAerr(RSA_F_RSA_OSSL_PRIVATE_ENCRYPT, ERR_R_PASSED_NULL_PARAMETER); > + return 0; > + } > + > + switch (type) > + { > + case NID_md5: > + alg = BCRYPT_MD5_ALGORITHM; > + break; > + > + case NID_sha1: > + alg = BCRYPT_SHA1_ALGORITHM; > + break; > + > + case NID_sha256: > + alg = BCRYPT_SHA256_ALGORITHM; > + break; > + > + case NID_sha384: > + alg = BCRYPT_SHA384_ALGORITHM; > + break; > + > + case NID_sha512: > + alg = BCRYPT_SHA512_ALGORITHM; > + break; > + > + case NID_md5_sha1: > + if (m_len != SSL_SIG_LENGTH) > + { > + RSAerr(RSA_F_RSA_SIGN, RSA_R_INVALID_MESSAGE_LENGTH); > + return 0; > + } > + /* No DigestInfo header is required -- set alg-name to NULL */ > + alg = NULL; > + break; > + default: > + msg(M_WARN, "cryptoapicert: Unknown hash type NID=0x%x", type); > + RSAerr(RSA_F_RSA_SIGN, RSA_R_UNKNOWN_ALGORITHM_TYPE); > + return 0; > + } > + > + *siglen = priv_enc_CNG(cd, alg, m, (int)m_len, sig, RSA_size(rsa), padding); > + return (siglen == 0) ? 0 : 1; > +} > + > /* decrypt */ > static int > rsa_priv_dec(int flen, const unsigned char *from, unsigned char *to, RSA *rsa, int padding) > @@ -555,6 +620,15 @@ SSL_CTX_use_CryptoAPI_certificate(SSL_CTX *ssl_ctx, const char *cert_prop) > RSA_meth_set_finish(my_rsa_method, finish); > RSA_meth_set0_app_data(my_rsa_method, cd); > > + /* For CNG, set the RSA_sign method which gets priority over priv_enc(). > + * This method is called with the raw hash without the digestinfo > + * header and works better when using NCryptSignHash() with some tokens. > + */ > + if (cd->key_spec == CERT_NCRYPT_KEY_SPEC) > + { > + RSA_meth_set_sign(my_rsa_method, rsa_sign_CNG); > + } > + > rsa = RSA_new(); > if (rsa == NULL) > { > Change makes sense and code looks good. Since this is already tested by Selva, JJK and gizi, I just did the stare-at-code part of review. Acked-by: Steffan Karger <steffan@karger.me> -Steffan
Your patch has been applied to the master branch. Steffan says "it is
a bugfix so it should go to 2.4", but the underlying infrastructure
does not seem to be there yet (git cherry-pick tries to bring in lots
of extra stuff). So we can do that, but I need to be told which master
commits I need to bring to 2.4 first.
I have not even stared-at-code very much here, trusting the testers and
Steffan's code-staring abilities.
As a side note (rambling about comments today), this hunk makes the
comment above it slightly misleading:
/* The hash OID is already in 'from'. So set the hash algorithm
* in the padding info struct to NULL.
*/
- BCRYPT_PKCS1_PADDING_INFO padinfo = {NULL};
+ BCRYPT_PKCS1_PADDING_INFO padinfo = {hash_algo};
.. since we do not set it to NULL anymore...
commit 6b495dc4c5cfc118091ddc9c19330b3c9e3e3dff (master)
Author: Selva Nair
Date: Thu Apr 26 10:24:24 2018 -0400
Pass the hash without the DigestInfo header to NCryptSignHash()
Acked-by: Steffan Karger <steffan.karger@fox-it.com>
Message-Id: <1524752664-27946-1-git-send-email-selva.nair@gmail.com>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg16840.html
Signed-off-by: Gert Doering <gert@greenie.muc.de>
--
kind regards,
Gert Doering
Hi, > As a side note (rambling about comments today), this hunk makes the > comment above it slightly misleading: > > /* The hash OID is already in 'from'. So set the hash algorithm > * in the padding info struct to NULL. > */ > - BCRYPT_PKCS1_PADDING_INFO padinfo = {NULL}; > + BCRYPT_PKCS1_PADDING_INFO padinfo = {hash_algo}; > > .. since we do not set it to NULL anymore... > A perfect case of how fast comments and code diverge -- I added that comment only the other day.. My bad. Will try to remember to fix next time this is touched. Knuth's literate programming never caught on, did it.. Thanks for the ack and merge.. For 2.4, have to look through and figure out --- it has been a while.. Selva <div dir="ltr">Hi,<br><div><br><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> <br> As a side note (rambling about comments today), this hunk makes the<br> comment above it slightly misleading:<br> <br> /* The hash OID is already in 'from'. So set the hash algorithm<br> * in the padding info struct to NULL.<br> */<br> - BCRYPT_PKCS1_PADDING_INFO padinfo = {NULL};<br> + BCRYPT_PKCS1_PADDING_INFO padinfo = {hash_algo};<br> <br> .. since we do not set it to NULL anymore...<br></blockquote><div><br></div><div>A perfect case of how fast comments and code diverge -- I added that comment only<br>the other day.. My bad. Will try to remember to fix next time this is touched.<br><br>Knuth's literate programming never caught on, did it..<br><br></div><div>Thanks for the ack and merge..<br><br></div><div>For 2.4, have to look through and figure out --- it has been a while..<br><br></div><div>Selva</div></div></div></div>
Hi On Fri, Oct 5, 2018 at 6:51 AM Gert Doering <gert@greenie.muc.de> wrote: > Your patch has been applied to the master branch. Steffan says "it is > a bugfix so it should go to 2.4", but the underlying infrastructure > does not seem to be there yet (git cherry-pick tries to bring in lots > of extra stuff). So we can do that, but I need to be told which master > commits I need to bring to 2.4 first. > cherry-picking this needs some manual conflict resolution but no pulling in anything extra from master (one function got refactored and renamed in master and hence the conflict). A patch for 2.4 follows -- same as for master except for the context of one hunk. Thanks, Selva <div dir="ltr"><div dir="ltr">Hi<br><div><br><div class="gmail_quote"><div dir="ltr">On Fri, Oct 5, 2018 at 6:51 AM Gert Doering <<a href="mailto:gert@greenie.muc.de" target="_blank">gert@greenie.muc.de</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Your patch has been applied to the master branch. Steffan says "it is<br> a bugfix so it should go to 2.4", but the underlying infrastructure <br> does not seem to be there yet (git cherry-pick tries to bring in lots<br> of extra stuff). So we can do that, but I need to be told which master<br> commits I need to bring to 2.4 first.<br></blockquote><div><br></div><div>cherry-picking this needs some manual conflict resolution but no pulling in<br></div><div>anything extra from master (one function got refactored and renamed in master</div><div>and hence the conflict).</div><div><br></div><div>A patch for 2.4 follows -- same as for master except for the context of one hunk.<br><br></div><div>Thanks,<br><br></div><div>Selva<br></div></div></div></div></div>
diff --git a/src/openvpn/cryptoapi.c b/src/openvpn/cryptoapi.c index 89d253c..d765954 100644 --- a/src/openvpn/cryptoapi.c +++ b/src/openvpn/cryptoapi.c @@ -217,14 +217,16 @@ rsa_pub_dec(int flen, const unsigned char *from, unsigned char *to, RSA *rsa, in * Sign the hash in 'from' using NCryptSignHash(). This requires an NCRYPT * key handle in cd->crypt_prov. On return the signature is in 'to'. Returns * the length of the signature or 0 on error. + * If the hash_algo is not NULL, PKCS #1 DigestInfo header gets added + * to 'from', else it is signed as is. * For now we support only RSA and the padding is assumed to be PKCS1 v1.5 */ static int -priv_enc_CNG(const CAPI_DATA *cd, const unsigned char *from, int flen, - unsigned char *to, int tlen, int padding) +priv_enc_CNG(const CAPI_DATA *cd, const wchar_t *hash_algo, const unsigned char *from, + int flen, unsigned char *to, int tlen, int padding) { NCRYPT_KEY_HANDLE hkey = cd->crypt_prov; - DWORD len; + DWORD len = 0; ASSERT(cd->key_spec == CERT_NCRYPT_KEY_SPEC); msg(D_LOW, "Signing hash using CNG: data size = %d", flen); @@ -232,7 +234,7 @@ priv_enc_CNG(const CAPI_DATA *cd, const unsigned char *from, int flen, /* The hash OID is already in 'from'. So set the hash algorithm * in the padding info struct to NULL. */ - BCRYPT_PKCS1_PADDING_INFO padinfo = {NULL}; + BCRYPT_PKCS1_PADDING_INFO padinfo = {hash_algo}; DWORD status; status = NCryptSignHash(hkey, padding? &padinfo : NULL, (BYTE*) from, flen, @@ -270,7 +272,7 @@ rsa_priv_enc(int flen, const unsigned char *from, unsigned char *to, RSA *rsa, i } if (cd->key_spec == CERT_NCRYPT_KEY_SPEC) { - return priv_enc_CNG(cd, from, flen, to, RSA_size(rsa), padding); + return priv_enc_CNG(cd, NULL, from, flen, to, RSA_size(rsa), padding); } /* Unfortunately, there is no "CryptSign()" function in CryptoAPI, that would @@ -334,6 +336,69 @@ rsa_priv_enc(int flen, const unsigned char *from, unsigned char *to, RSA *rsa, i return len; } +/* + * Sign the hash in |m| and return the signature in |sig|. + * Returns 1 on success, 0 on error. + * NCryptSignHash() is used to sign and it is instructed to add the + * the PKCS #1 DigestInfo header to |m| unless the hash algorithm is + * the MD5/SHA1 combination used in TLS 1.1 and earlier versions. + */ +static int +rsa_sign_CNG(int type, const unsigned char *m, unsigned int m_len, + unsigned char *sig, unsigned int *siglen, const RSA *rsa) +{ + CAPI_DATA *cd = (CAPI_DATA *) RSA_meth_get0_app_data(RSA_get_method(rsa)); + const wchar_t *alg = NULL; + int padding = RSA_PKCS1_PADDING; + + *siglen = 0; + if (cd == NULL) + { + RSAerr(RSA_F_RSA_OSSL_PRIVATE_ENCRYPT, ERR_R_PASSED_NULL_PARAMETER); + return 0; + } + + switch (type) + { + case NID_md5: + alg = BCRYPT_MD5_ALGORITHM; + break; + + case NID_sha1: + alg = BCRYPT_SHA1_ALGORITHM; + break; + + case NID_sha256: + alg = BCRYPT_SHA256_ALGORITHM; + break; + + case NID_sha384: + alg = BCRYPT_SHA384_ALGORITHM; + break; + + case NID_sha512: + alg = BCRYPT_SHA512_ALGORITHM; + break; + + case NID_md5_sha1: + if (m_len != SSL_SIG_LENGTH) + { + RSAerr(RSA_F_RSA_SIGN, RSA_R_INVALID_MESSAGE_LENGTH); + return 0; + } + /* No DigestInfo header is required -- set alg-name to NULL */ + alg = NULL; + break; + default: + msg(M_WARN, "cryptoapicert: Unknown hash type NID=0x%x", type); + RSAerr(RSA_F_RSA_SIGN, RSA_R_UNKNOWN_ALGORITHM_TYPE); + return 0; + } + + *siglen = priv_enc_CNG(cd, alg, m, (int)m_len, sig, RSA_size(rsa), padding); + return (siglen == 0) ? 0 : 1; +} + /* decrypt */ static int rsa_priv_dec(int flen, const unsigned char *from, unsigned char *to, RSA *rsa, int padding) @@ -555,6 +620,15 @@ SSL_CTX_use_CryptoAPI_certificate(SSL_CTX *ssl_ctx, const char *cert_prop) RSA_meth_set_finish(my_rsa_method, finish); RSA_meth_set0_app_data(my_rsa_method, cd); + /* For CNG, set the RSA_sign method which gets priority over priv_enc(). + * This method is called with the raw hash without the digestinfo + * header and works better when using NCryptSignHash() with some tokens. + */ + if (cd->key_spec == CERT_NCRYPT_KEY_SPEC) + { + RSA_meth_set_sign(my_rsa_method, rsa_sign_CNG); + } + rsa = RSA_new(); if (rsa == NULL) {
From: Selva Nair <selva.nair@gmail.com> In case of TLS 1.2 signatures, the callback rsa_priv_enc() gets the hash with the DigestInfo prepended. Signing this using NCryptSignHash() with hash algorithm id set to NULL works in most cases. But when using some hardware tokens, the data gets interpreted as the pre TLS 1.2 MD5+SHA1 hash and is silently truncated to 36 bytes. Avoid this by passing the raw hash to NCryptSignHash() and let it add the DigestInfo. To get the raw hash we set the RSA_sign() method in the rsa_method structure. This callback bypasses rsa_priv_enc() and gets called with the hash type and the hash. Fixes Trac #1050 --- Tested by JJK and gizi as described in Trac #1050 src/openvpn/cryptoapi.c | 84 ++++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 79 insertions(+), 5 deletions(-)