From 0d424049f08162f195b424fc227af3ed17380684 Mon Sep 17 00:00:00 2001 From: Noah Laptop Date: Wed, 20 Feb 2019 16:58:26 -0800 Subject: [PATCH] started integrating bearssl, wrote SSL profile that only incorporates safe TLS1.2 ciphers --- src/SSLClient.h | 7 +- src/bearssl/TLS12_only_profile.c | 459 +++++++++++++++++++++++++++++++ src/config.h | 6 +- 3 files changed, 467 insertions(+), 5 deletions(-) create mode 100644 src/bearssl/TLS12_only_profile.c diff --git a/src/SSLClient.h b/src/SSLClient.h index 36c6a5f..8b3e389 100644 --- a/src/SSLClient.h +++ b/src/SSLClient.h @@ -100,8 +100,13 @@ public: C& getClient() { return m_client; } private: - // create a copy of the class + // create a copy of the client C m_client; + // store the context values required for SSL + br_ssl_client_context m_sslctx; + br_x509_minimal_context m_x509ctx; + unsigned char m_iobuf[BR_SSL_BUFSIZE_MONO]; + br_sslio_context m_ioctx; }; #endif /** SSLClient_H_ */ \ No newline at end of file diff --git a/src/bearssl/TLS12_only_profile.c b/src/bearssl/TLS12_only_profile.c new file mode 100644 index 0000000..b58ee01 --- /dev/null +++ b/src/bearssl/TLS12_only_profile.c @@ -0,0 +1,459 @@ +/* + * Copyright (c) 2016 Thomas Pornin + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice shall be + * included in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#include "bearssl.h" + +/* + * A "profile" is an initialisation function for a SSL context, that + * configures a list of cipher suites and algorithm implementations. + * While BearSSL comes with a few predefined profiles, you might one + * to define you own, using the example below as guidance. + * + * Each individual initialisation call sets a parameter or an algorithm + * support. Setting a specific algorithm pulls in the implementation of + * that algorithm in the compiled binary, as per static linking + * behaviour. Removing some of this calls will then reduce total code + * footprint, but also mechanically prevents some features to be + * supported (protocol versions and cipher suites). + * + * The two below define profiles for the client and the server contexts, + * respectively. Of course, in a typical size-constrained application, + * you would use one or the other, not both, to avoid pulling in code + * for both. + * + * This profile has been modified to the following criteria + * * support only TLS 1.2 for security + * * Use a minimal size footprint + * * remove RSA_WITH_AES ciphers for above + */ + +void +TLS12_only_profile(br_ssl_client_context *cc + br_x509_minimal_context *xc, + const br_x509_trust_anchor *trust_anchors, size_t trust_anchors_num) +{ + /* + * The "full" profile supports all implemented cipher suites. + * + * Rationale for suite order, from most important to least + * important rule: + * + * -- Only support TLS 1.2 + * -- Don't support RSA and 3DES as they are weaker protocols + * -- Try to have Forward Secrecy (ECDHE suite) if possible. + * -- When not using Forward Secrecy, ECDH key exchange is + * better than RSA key exchange (slightly more expensive on the + * client, but much cheaper on the server, and it implies smaller + * messages). + * -- ChaCha20+Poly1305 is better than AES/GCM (faster, smaller code). + * -- GCM is better than CBC. + * -- AES-128 is preferred over AES-256 (AES-128 is already + * strong enough, and AES-256 is 40% more expensive). + */ + static const uint16_t suites[] = { + BR_TLS_ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256, + BR_TLS_ECDHE_RSA_WITH_CHACHA20_POLY1305_SHA256, + + BR_TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256, + BR_TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256, + BR_TLS_ECDHE_ECDSA_WITH_AES_256_GCM_SHA384, + BR_TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384, + BR_TLS_ECDHE_ECDSA_WITH_AES_128_CBC_SHA256, + BR_TLS_ECDHE_RSA_WITH_AES_128_CBC_SHA256, + BR_TLS_ECDHE_ECDSA_WITH_AES_256_CBC_SHA384, + BR_TLS_ECDHE_RSA_WITH_AES_256_CBC_SHA384, + + BR_TLS_ECDH_ECDSA_WITH_AES_128_GCM_SHA256, + BR_TLS_ECDH_RSA_WITH_AES_128_GCM_SHA256, + BR_TLS_ECDH_ECDSA_WITH_AES_256_GCM_SHA384, + BR_TLS_ECDH_RSA_WITH_AES_256_GCM_SHA384, + BR_TLS_ECDH_ECDSA_WITH_AES_128_CBC_SHA256, + BR_TLS_ECDH_RSA_WITH_AES_128_CBC_SHA256, + BR_TLS_ECDH_ECDSA_WITH_AES_256_CBC_SHA384, + BR_TLS_ECDH_RSA_WITH_AES_256_CBC_SHA384, + }; + + /* + * Client context must be cleared at some point. This sets + * every value and pointer to 0 or NULL. + */ + br_ssl_client_zero(cc); + + /* + * Define minimum and maximum protocol versions. Supported + * versions are: + * BR_TLS10 TLS 1.0 + * BR_TLS11 TLS 1.1 + * BR_TLS12 TLS 1.2 + */ + br_ssl_engine_set_versions(&cc->eng, BR_TLS12, BR_TLS12); + + /* + * Set the PRF implementation(s). + * For TLS 1.0 and 1.1, the "prf10" is needed. + * For TLS 1.2, this depends on the cipher suite: + * -- cipher suites with a name ending in "SHA384" need "prf_sha384"; + * -- all others need "prf_sha256". + * + * Note that a cipher suite like TLS_RSA_WITH_AES_128_CBC_SHA will + * use SHA-1 for the per-record MAC (that's what the final "SHA" + * means), but still SHA-256 for the PRF when selected along with + * the TLS-1.2 protocol version. + */ + // br_ssl_engine_set_prf10(&cc->eng, &br_tls10_prf); + br_ssl_engine_set_prf_sha256(&cc->eng, &br_tls12_sha256_prf); + br_ssl_engine_set_prf_sha384(&cc->eng, &br_tls12_sha384_prf); + + /* + * Set hash functions for the engine. Required hash functions + * depend on the protocol and cipher suite: + * + * -- TLS 1.0 and 1.1 require both MD5 and SHA-1. + * -- With TLS 1.2, cipher suites with a name ending in "SHA384" + * require SHA-384. + * -- With TLS 1.2, cipher suites with a name ending in "SHA256" + * require SHA-256. + * -- With TLS 1.2, cipher suites with a name ending in "SHA" + * require both SHA-256 and SHA-1. + * + * Moreover, these hash functions are also used to compute + * hashes supporting signatures on the server side (for ECDHE_* + * cipher suites), and on the client side (for client + * certificates, except in the case of full static ECDH). In TLS + * 1.0 and 1.1, SHA-1 (and also MD5) will be used, but with TLS + * 1.2 these hash functions are negotiated between client and + * server; SHA-256 and/or SHA-384 should be sufficient in + * practice. + * + * Note that with current implementations, SHA-224 and SHA-256 + * share the same file, so if you use one, you may have the other + * one with no additional overhead. Similarly, SHA-384 and SHA-512 + * share the same implementation code. + */ + // br_ssl_engine_set_hash(&cc->eng, br_md5_ID, &br_md5_vtable); + // br_ssl_engine_set_hash(&cc->eng, br_sha1_ID, &br_sha1_vtable); + br_ssl_engine_set_hash(&cc->eng, br_sha224_ID, &br_sha224_vtable); + br_ssl_engine_set_hash(&cc->eng, br_sha256_ID, &br_sha256_vtable); + br_ssl_engine_set_hash(&cc->eng, br_sha384_ID, &br_sha384_vtable); + // br_ssl_engine_set_hash(&cc->eng, br_sha512_ID, &br_sha512_vtable); + + /* + * Set the cipher suites. All specified cipher suite MUST be + * supported, and the relevant algorithms MUST have been + * configured (failure to provide needed implementations may + * trigger unwanted behaviours like segfaults or overflows). + */ + br_ssl_engine_set_suites(&cc->eng, suites, + (sizeof suites) / (sizeof suites[0])); + + /* + * Public-key algorithm implementations. + * + * -- RSA public core ("rsapub") is needed for "RSA" key exchange + * (cipher suites whose name starts with TLS_RSA). + * + * -- RSA signature verification ("rsavrfy") is needed for + * "ECDHE_RSA" cipher suites (not ECDH_RSA). + * + * -- Elliptic curve implementation ("ec") is needed for cipher + * suites that use elliptic curves (both "ECDH" and "ECDHE" + * cipher suites). + * + * -- ECDSA signature verification is needed for "ECDHE_ECDSA" + * cipher suites (but not for ECDHE_RSA, ECDH_ECDSA or ECDH_RSA). + * + * Normally, you use the "default" implementations, obtained + * through relevant function calls. These functions return + * implementations that are deemed "best" for the current + * platform, where "best" means "fastest within constant-time + * implementations". Selecting the default implementation is a + * mixture of compile-time and runtime checks. + * + * Nevertheless, specific implementations may be selected + * explicitly, e.g. to use code which is slower but with a + * smaller footprint. + * + * The RSA code comes in three variants, called "i15", "i31" and + * "i32". The "i31" code is somewhat faster than the "i32" code. + * Usually, "i31" is faster than "i15", except on some specific + * architectures (ARM Cortex M0, M0+, M1 and M3) where the "i15" + * should be preferred (the "i15" code is constant-time, while + * the "i31" is not, and the "i15" code is faster anyway). + * + * ECDSA code also comes in "i15" and "i31" variants. As in the + * case of RSA, the "i31" code is faster, except on the small + * ARM Cortex M, where the "i15" code is faster and safer. + * + * There are no less than 10 elliptic curve implementations: + * + * - ec_c25519_i15, ec_c25519_i31, ec_c25519_m15 and ec_c25519_m31 + * implement Curve25519. + * + * - ec_p256_m15 and ec_p256_m31 implement NIST curve P-256. + * + * - ec_prime_i15 and ec_prime_i31 implement NIST curves P-256, + * P-384 and P-521. + * + * - ec_all_m15 is an aggregate implementation that uses + * ec_c25519_m15, ec_p256_m15 and ec_prime_i15. + * + * - ec_all_m31 is an aggregate implementation that uses + * ec_c25519_m31, ec_p256_m31 and ec_prime_i31. + * + * For a given curve, "m15" is faster than "i15" (but possibly + * with a larger code footprint) and "m31" is faster than "i31" + * (there again with a larger code footprint). For best + * performance, use ec_all_m31, except on the small ARM Cortex M + * where ec_all_m15 should be used. Referencing the other + * implementations directly will result in smaller code, but + * support for fewer curves and possibly lower performance. + */ + // br_ssl_client_set_default_rsapub(cc); + // br_ssl_engine_set_default_rsavrfy(&cc->eng); + // br_ssl_engine_set_default_ecdsa(&cc->eng); + //* Alternate: set implementations explicitly. + // br_ssl_client_set_rsapub(cc, &br_rsa_i31_public); + br_ssl_client_set_rsavrfy(cc, &br_rsa_i15_pkcs1_vrfy); + br_ssl_engine_set_ec(&cc->eng, &br_ec_all_m15); + br_ssl_engine_set_ecdsa(&cc->eng, &br_ecdsa_i15_vrfy_asn1); + //*/ + + /* + * Record handler: + * -- Cipher suites in AES_128_CBC, AES_256_CBC and 3DES_EDE_CBC + * need the CBC record handler ("set_cbc"). + * -- Cipher suites in AES_128_GCM and AES_256_GCM need the GCM + * record handler ("set_gcm"). + * -- Cipher suites in CHACHA20_POLY1305 need the ChaCha20+Poly1305 + * record handler ("set_chapol"). + */ + br_ssl_engine_set_cbc(&cc->eng, + &br_sslrec_in_cbc_vtable, + &br_sslrec_out_cbc_vtable); + br_ssl_engine_set_gcm(&cc->eng, + &br_sslrec_in_gcm_vtable, + &br_sslrec_out_gcm_vtable); + // br_ssl_engine_set_chapol(&cc->eng, + // &br_sslrec_in_chapol_vtable, + // &br_sslrec_out_chapol_vtable); + + /* + * Set the ChaCha20 and Poly1305 implementations + * Not included in this file orignally for some reason + */ + br_ssl_engine_set_default_chapol(&cc->eng); + + /* + * Symmetric encryption: + * -- AES_128_CBC and AES_256_CBC require an "aes_cbc" implementation + * (actually two implementations, for encryption and decryption). + * -- 3DES_EDE_CBC requires a "des_cbc" implementation + * (actually two implementations, for encryption and decryption). + * -- AES_128_GCM and AES_256_GCM require an "aes_ctr" imeplementation + * and also a GHASH implementation. + * + * Two 3DES implementations are provided: + * + * des_tab Classical table-based implementation; it is + * not constant-time. + * + * dest_ct Constant-time DES/3DES implementation. It is + * slower than des_tab. + * + * Four AES implementations are provided: + * + * aes_ct Constant-time AES implementation, for 32-bit + * systems. + * + * aes_ct64 Constant-time AES implementation, for 64-bit + * systems. It actually also runs on 32-bit systems, + * but, on such systems, it yields larger code and + * slightly worse performance. On 64-bit systems, + * aes_ct64 is about twice faster than aes_ct for + * CTR processing (GCM encryption and decryption), + * and for CBC (decryption only). + * + * aes_small Smallest implementation provided, but also the + * slowest, and it is not constant-time. Use it + * only if desperate for code size. + * + * aes_big Classical table-based AES implementation. This + * is decently fast and still resonably compact, + * but it is not constant-time. + * + * aes_x86ni Very fast implementation that uses the AES-NI + * opcodes on recent x86 CPU. But it may not be + * compiled in the library if the compiler or + * architecture is not supported; and the CPU + * may also not support the opcodes. Selection + * functions are provided to test for availability + * of the code and the opcodes. + * + * Whether having constant-time implementations is absolutely + * required for security depends on the context (in particular + * whether the target architecture actually has cache memory), + * and while side-channel analysis for non-constant-time AES + * code has been demonstrated in lab conditions, it certainly + * does not apply to all actual usages, and it has never been + * spotted in the wild. It is still considered cautious to use + * constant-time code by default, and to consider the other + * implementations only if duly measured performance issues make + * it mandatory. + */ + br_ssl_engine_set_aes_cbc(&cc->eng, + &br_aes_ct_cbcenc_vtable, + &br_aes_ct_cbcdec_vtable); + br_ssl_engine_set_aes_ctr(&cc->eng, + &br_aes_ct_ctr_vtable); + /* Alternate: aes_ct64 + br_ssl_engine_set_aes_cbc(&cc->eng, + &br_aes_ct64_cbcenc_vtable, + &br_aes_ct64_cbcdec_vtable); + br_ssl_engine_set_aes_ctr(&cc->eng, + &br_aes_ct64_ctr_vtable); + */ + /* Alternate: aes_small + br_ssl_engine_set_aes_cbc(&cc->eng, + &br_aes_small_cbcenc_vtable, + &br_aes_small_cbcdec_vtable); + br_ssl_engine_set_aes_ctr(&cc->eng, + &br_aes_small_ctr_vtable); + */ + /* Alternate: aes_big + br_ssl_engine_set_aes_cbc(&cc->eng, + &br_aes_big_cbcenc_vtable, + &br_aes_big_cbcdec_vtable); + br_ssl_engine_set_aes_ctr(&cc->eng, + &br_aes_big_ctr_vtable); + */ + /* 3DES Disabled + br_ssl_engine_set_des_cbc(&cc->eng, + &br_des_ct_cbcenc_vtable, + &br_des_ct_cbcdec_vtable); + */ + /* Alternate: des_tab + br_ssl_engine_set_des_cbc(&cc->eng, + &br_des_tab_cbcenc_vtable, + &br_des_tab_cbcdec_vtable); + */ + + /* + * GHASH is needed for AES_128_GCM and AES_256_GCM. Three + * implementations are provided: + * + * ctmul Uses 32-bit multiplications with a 64-bit result. + * + * ctmul32 Uses 32-bit multiplications with a 32-bit result. + * + * ctmul64 Uses 64-bit multiplications with a 64-bit result. + * + * On 64-bit platforms, ctmul64 is the smallest and fastest of + * the three. On 32-bit systems, ctmul should be preferred. The + * ctmul32 implementation is meant to be used for the specific + * 32-bit systems that do not have a 32x32->64 multiplier (i.e. + * the ARM Cortex-M0 and Cortex-M0+). + * + * These implementations are all constant-time as long as the + * underlying multiplication opcode is constant-time (which is + * true for all modern systems, but not for older architectures + * such that ARM9 or 80486). + */ + /* + br_ssl_engine_set_ghash(&cc->eng, + &br_ghash_ctmul); */ + //* Alternate: ghash_ctmul32 + br_ssl_engine_set_ghash(&cc->eng, + &br_ghash_ctmul32); + //*/ + /* Alternate: ghash_ctmul64 + br_ssl_engine_set_ghash(&cc->eng, + &br_ghash_ctmul64); + */ + + /* + * For a client, the normal case is to validate the server + * certificate with regards to a set of trust anchors. This + * entails using a br_x509_minimal_context structure, configured + * with the relevant algorithms, as shown below. + * + * Alternatively, the client could "know" the intended server + * public key through an out-of-band mechanism, in which case + * a br_x509_knownkey_context is appropriate, for a much reduced + * code footprint. + * + * We assume here that the following extra parameters have been + * provided: + * + * xc engine context (br_x509_minimal_context *) + * trust_anchors trust anchors (br_x509_trust_anchor *) + * trust_anchors_num number of trust anchors (size_t) + */ + + /* + * The X.509 engine needs a hash function for processing the + * subject and issuer DN of certificates and trust anchors. Any + * supported hash function is appropriate; here we use SHA-256. + * The trust an + */ + br_x509_minimal_init(xc, &br_sha256_vtable, + trust_anchors, trust_anchors_num); + + /* + * Set suites and asymmetric crypto implementations. We use the + * "i31" code for RSA (it is somewhat faster than the "i32" + * implementation). These implementations are used for + * signature verification on certificates, but not for the + * SSL-specific usage of the server's public key. For instance, + * if the server has an EC public key but the rest of the chain + * (intermediate CA, root...) use RSA, then you would need only + * the RSA verification function below. + */ + // br_x509_minimal_set_rsa(xc, &br_rsa_i31_pkcs1_vrfy); + br_x509_minimal_set_rsa(xc, &br_rsa_i15_pkcs1_vrfy); + // br_x509_minimal_set_ecdsa(xc, + // &br_ec_prime_i31, &br_ecdsa_i31_vrfy_asn1); + br_x509_minimal_set_ecdsa(xc, + &br_ec_all_m15, &br_ecdsa_i15_vrfy_asn1); + + /* + * Set supported hash functions. These are for signatures on + * certificates. There again, you only need the hash functions + * that are actually used in certificates, but if a given + * function was included for the SSL engine, you may as well + * add it here. + * + * Note: the engine explicitly rejects signatures that use MD5. + * Thus, there is no need for MD5 here. + */ + // br_ssl_engine_set_hash(xc, br_sha1_ID, &br_sha1_vtable); + br_ssl_engine_set_hash(xc, br_sha224_ID, &br_sha224_vtable); + br_ssl_engine_set_hash(xc, br_sha256_ID, &br_sha256_vtable); + br_ssl_engine_set_hash(xc, br_sha384_ID, &br_sha384_vtable); + // br_ssl_engine_set_hash(xc, br_sha512_ID, &br_sha512_vtable); + + /* + * Link the X.509 engine in the SSL engine. + */ + br_ssl_engine_set_x509(&cc->eng, &xc->vtable); +} diff --git a/src/config.h b/src/config.h index 8ea4d8a..001b352 100644 --- a/src/config.h +++ b/src/config.h @@ -47,9 +47,8 @@ * substantially more efficient than 32-bit multiplications that yield * 64-bit results. This is typically the case on low-end ARM Cortex M * systems (M0, M0+, M1, and arguably M3 and M4 as well). - * -#define BR_LOMUL 1 */ +#define BR_LOMUL 1 /* * When BR_SLOW_MUL is enabled, multiplications are assumed to be @@ -159,9 +158,8 @@ * This is meant for the low-end cores (Cortex M0, M0+, M1, M3). * Note: if BR_LOMUL is not explicitly enabled or disabled, then * enabling BR_ARMEL_CORTEXM_GCC also enables BR_LOMUL. - * -#define BR_ARMEL_CORTEXM_GCC 1 */ +#define BR_ARMEL_CORTEXM_GCC 1 /* * When BR_AES_X86NI is enabled, the AES implementation using the x86 "NI"