OpenSSL is built with the generic linux settings for most targets,
including aarch64. These generic settings are designed for 32-bit CPU and
provide no assembler optmization: this is widely suboptimal for aarch64.
This patch simply switches to the aarch64 settings that are already
available in OpenSSL.
Here is the output of "openssl speed" before the optimization, with
"(...)" representing build flags that didn't change:
OpenSSL 1.0.2l 25 May 2017
options:bn(64,32) rc4(ptr,char) des(idx,cisc,2,int) aes(partial) blowfish(ptr)
compiler: aarch64-openwrt-linux-musl-gcc (...)
And after this patch, OpenSSL uses 64 bit mode and assembler optimizations:
OpenSSL 1.0.2l 25 May 2017
options:bn(64,64) rc4(ptr,char) des(idx,cisc,2,int) aes(partial) blowfish(ptr)
compiler: aarch64-openwrt-linux-musl-gcc (...) -DSHA1_ASM -DSHA256_ASM -DSHA512_ASM
Here are some benchmarks on a pine64+ running latest LEDE master r5142-20d363aed3:
before# openssl speed sha aes blowfish
The 'numbers' are in 1000s of bytes per second processed.
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
sha1 3918.89k 9982.43k 19148.03k 24933.03k 27325.78k
sha256 4604.51k 10240.64k 17472.51k 21355.18k 22801.07k
sha512 3662.19k 14539.41k 21443.16k 29544.11k 33177.60k
blowfish cbc 16266.63k 16940.86k 17176.92k 17237.33k 17252.35k
aes-128 cbc 19712.95k 21447.40k 22091.09k 22258.35k 22304.09k
aes-192 cbc 17680.12k 19064.47k 19572.14k 19703.13k 19737.26k
aes-256 cbc 15986.67k 17132.48k 17537.28k 17657.17k 17689.26k
after# openssl speed sha aes blowfish
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
sha1 6770.87k 26172.80k 86878.38k 205649.58k 345978.20k
sha256 20913.93k 74663.85k 184658.18k 290891.09k 351032.66k
sha512 7633.10k 30110.14k 50083.24k 71883.43k 82485.25k
blowfish cbc 16224.93k 16933.55k 17173.76k 17234.94k 17252.35k
aes-128 cbc 19425.74k 21193.31k 22065.74k 22304.77k 22380.54k
aes-192 cbc 17452.29k 18883.84k 19536.90k 19741.70k 19800.06k
aes-256 cbc 15815.89k 17003.01k 17530.03k 17695.40k 17746.60k
For some reason AES and blowfish do not benefit, but SHA performance
improves between 1.7x and 15x. SHA256 clearly benefits the most from the
optimization (4.5x on small blocks, 15x on large blocks!).
When using EVP (with "openssl speed -evp <algo>"):
# Before, EVP mode
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
sha1 3824.46k 10049.66k 19170.56k 24947.03k 27325.78k
sha256 3368.33k 8511.15k 16061.44k 20772.52k 22721.88k
sha512 2845.23k 11381.57k 19467.69k 28512.26k 33008.30k
bf-cbc 15146.74k 16623.83k 17092.01k 17211.39k 17249.62k
aes-128-cbc 17873.03k 20870.61k 21933.65k 22216.36k 22301.35k
aes-192-cbc 16184.18k 18607.15k 19447.13k 19670.02k 19737.26k
aes-256-cbc 14774.06k 16757.25k 17457.58k 17639.42k 17686.53k
# After, EVP mode
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
sha1 7056.97k 27142.10k 89515.86k 209155.41k 347419.99k
sha256 7745.70k 29750.06k 95341.48k 211001.69k 332376.75k
sha512 4550.47k 18086.06k 39997.10k 65880.75k 81431.21k
bf-cbc 15129.20k 16619.03k 17090.56k 17212.76k 17246.89k
aes-128-cbc 99619.74k 269032.34k 450214.23k 567353.00k 613933.06k
aes-192-cbc 93180.74k 231017.79k 361766.66k 433671.51k 461731.16k
aes-256-cbc 89343.23k 209858.58k 310160.04k 362234.88k 380878.85k
Blowfish does not seem to have assembler optimization at all, and SHA
still benefits (between 1.6x and 14.5x) but is generally slower than in
non-EVP mode.
However, AES performance is improved between 5.5x and 27.5x, which is
really impressive! For aes-128-cbc on large blocks, a core i7-6600U
@2.60GHz is only twice as fast...
Signed-off-by: Baptiste Jonglez <git@bitsofnetworks.org>
This fixes the following security problems:
CVE-2017-3731: Truncated packet could crash via OOB read
CVE-2017-3732: BN_mod_exp may produce incorrect results on x86_64
CVE-2016-7055: Montgomery multiplication may produce incorrect results
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
CVE-2016-0704
s2_srvr.c overwrite the wrong bytes in the master-key when applying
Bleichenbacher protection for export cipher suites. This provides a
Bleichenbacher oracle, and could potentially allow more efficient variants of
the DROWN attack.
CVE-2016-0703
s2_srvr.c did not enforce that clear-key-length is 0 for non-export ciphers.
If clear-key bytes are present for these ciphers, they *displace* encrypted-key
bytes. This leads to an efficient divide-and-conquer key recovery attack: if
an eavesdropper has intercepted an SSLv2 handshake, they can use the server as
an oracle to determine the SSLv2 master-key, using only 16 connections to the
server and negligible computation. More importantly, this leads to a more
efficient version of DROWN that is effective against non-export ciphersuites,
and requires no significant computation.
CVE-2016-0702
A side-channel attack was found which makes use of cache-bank conflicts on
the Intel Sandy-Bridge microarchitecture which could lead to the recovery of
RSA keys. The ability to exploit this issue is limited as it relies on an
attacker who has control of code in a thread running on the same hyper-
threaded core as the victim thread which is performing decryptions.
CVE-2016-0799
The internal |fmtstr| function used in processing a "%s" format string in
the BIO_*printf functions could overflow while calculating the length of a
string and cause an OOB read when printing very long strings. Additionally
the internal |doapr_outch| function can attempt to write to an OOB memory
location (at an offset from the NULL pointer) in the event of a memory
allocation failure. In 1.0.2 and below this could be caused where the size
of a buffer to be allocated is greater than INT_MAX. E.g. this could be in
processing a very long "%s" format string. Memory leaks can also occur.
The first issue may mask the second issue dependent on compiler behaviour.
These problems could enable attacks where large amounts of untrusted data is
passed to the BIO_*printf functions. If applications use these functions in
this way then they could be vulnerable. OpenSSL itself uses these functions
when printing out human-readable dumps of ASN.1 data. Therefore applications
that print this data could be vulnerable if the data is from untrusted sources.
OpenSSL command line applications could also be vulnerable where they print out
ASN.1 data, or if untrusted data is passed as command line arguments. Libssl is
not considered directly vulnerable. Additionally certificates etc received via
remote connections via libssl are also unlikely to be able to trigger these
issues because of message size limits enforced within libssl.
CVE-2016-0797
In the BN_hex2bn function the number of hex digits is calculated using an int
value |i|. Later |bn_expand| is called with a value of |i * 4|. For large
values of |i| this can result in |bn_expand| not allocating any memory because
|i * 4| is negative. This can leave the internal BIGNUM data field as NULL
leading to a subsequent NULL ptr deref. For very large values of |i|, the
calculation |i * 4| could be a positive value smaller than |i|. In this case
memory is allocated to the internal BIGNUM data field, but it is insufficiently
sized leading to heap corruption. A similar issue exists in BN_dec2bn. This
could have security consequences if BN_hex2bn/BN_dec2bn is ever called by user
applications with very large untrusted hex/dec data. This is anticipated to be
a rare occurrence. All OpenSSL internal usage of these functions use data that
is not expected to be untrusted, e.g. config file data or application command
line arguments. If user developed applications generate config file data based
on untrusted data then it is possible that this could also lead to security
consequences. This is also anticipated to be rare.
CVE-2016-0798
The SRP user database lookup method SRP_VBASE_get_by_user had confusing memory
management semantics; the returned pointer was sometimes newly allocated, and
sometimes owned by the callee. The calling code has no way of distinguishing
these two cases. Specifically, SRP servers that configure a secret seed to hide
valid login information are vulnerable to a memory leak: an attacker connecting
with an invalid username can cause a memory leak of around 300 bytes per
connection. Servers that do not configure SRP, or configure SRP but do not
configure a seed are not vulnerable. In Apache, the seed directive is known as
SSLSRPUnknownUserSeed. To mitigate the memory leak, the seed handling in
SRP_VBASE_get_by_user is now disabled even if the user has configured a seed.
Applications are advised to migrate to SRP_VBASE_get1_by_user. However, note
that OpenSSL makes no strong guarantees about the indistinguishability of valid
and invalid logins. In particular, computations are currently not carried out
in constant time.
CVE-2016-0705
A double free bug was discovered when OpenSSL parses malformed DSA private keys
and could lead to a DoS attack or memory corruption for applications that
receive DSA private keys from untrusted sources. This scenario is considered
rare.
CVE-2016-0800
A cross-protocol attack was discovered that could lead to decryption of TLS
sessions by using a server supporting SSLv2 and EXPORT cipher suites as a
Bleichenbacher RSA padding oracle. Note that traffic between clients and non-
vulnerable servers can be decrypted provided another server supporting SSLv2
and EXPORT ciphers (even with a different protocol such as SMTP, IMAP or POP)
shares the RSA keys of the non-vulnerable server. This vulnerability is known
as DROWN (CVE-2016-0800). Recovering one session key requires the attacker to
perform approximately 2^50 computation, as well as thousands of connections to
the affected server. A more efficient variant of the DROWN attack exists
against unpatched OpenSSL servers using versions that predate 1.0.2a, 1.0.1m,
1.0.0r and 0.9.8zf released on 19/Mar/2015 (see CVE-2016-0703 below). Users can
avoid this issue by disabling the SSLv2 protocol in all their SSL/TLS servers,
if they've not done so already. Disabling all SSLv2 ciphers is also sufficient,
provided the patches for CVE-2015-3197 (fixed in OpenSSL 1.0.1r and 1.0.2f)
have been deployed. Servers that have not disabled the SSLv2 protocol, and are
not patched for CVE-2015-3197 are vulnerable to DROWN even if all SSLv2
ciphers are nominally disabled, because malicious clients can force the use of
SSLv2 with EXPORT ciphers. OpenSSL 1.0.2g and 1.0.1s deploy the following
mitigation against DROWN: SSLv2 is now by default disabled at build-time.
Builds that are not configured with "enable-ssl2" will not support SSLv2.
Even if "enable-ssl2" is used, users who want to negotiate SSLv2 via the
version-flexible SSLv23_method() will need to explicitly call either of:
SSL_CTX_clear_options(ctx, SSL_OP_NO_SSLv2); or SSL_clear_options(ssl,
SSL_OP_NO_SSLv2); as appropriate. Even if either of those is used, or the
application explicitly uses the version-specific SSLv2_method() or its client
or server variants, SSLv2 ciphers vulnerable to exhaustive search key recovery
have been removed. Specifically, the SSLv2 40-bit EXPORT ciphers, and SSLv2
56-bit DES are no longer available. In addition, weak ciphers in SSLv3 and up
are now disabled in default builds of OpenSSL. Builds that are not configured
with "enable-weak-ssl-ciphers" will not provide any "EXPORT" or "LOW" strength
ciphers.
Signed-off-by: Jo-Philipp Wich <jow@openwrt.org>
SVN-Revision: 48868
Tested myself on ixp4xx and mvebu, and (originally)
by Daniel on i.MX6. Also tested on a MIPS target,
to make sure the change to ASFLAGS does not break things.
Based on a patch submitted by Daniel Drown:
https://lists.openwrt.org/pipermail/openwrt-devel/2014-July/026639.html
Signed-off-by: Claudio Leite <leitec@staticky.com>
Signed-off-by: Daniel Drown <dan-openwrt@drown.org>
SVN-Revision: 44618
This version includes this changes:
Don't include gmt_unix_time in TLS server and client random values
Fix for TLS record tampering bug CVE-2013-4353
Fix for TLS version checking bug CVE-2013-6449
Fix for DTLS retransmission bug CVE-2013-6450
Signed-off-by: Peter Wagner <tripolar@gmx.at>
SVN-Revision: 39853
