Matt Caswell [Sat, 23 May 2015 20:51:21 +0000 (21:51 +0100)]
Don't send an alert if we've just received one
If the record received is for a version that we don't support, previously we
were sending an alert back. However if the incoming record already looks
like an alert then probably we shouldn't do that. So suppress an outgoing
alert if it looks like we've got one incoming.
Andy Polyakov [Wed, 20 May 2015 07:36:48 +0000 (09:36 +0200)]
bn/asm/x86_64-mont5.pl: fix valgrind error.
bn_get_bits5 was overstepping array boundary by 1 byte. It was exclusively
read overstep and data could not have been used. The only potential problem
would be if array happens to end on the very edge of last accesible page.
Andy Polyakov [Wed, 13 May 2015 20:19:59 +0000 (22:19 +0200)]
bn/asm/vis3-mont.pl: fix intermittent EC failures on SPARC T3.
BLKINIT optimization worked on T4, but for some reason appears "too
aggressive" for T3 triggering intermiitent EC failures. It's not clear
why only EC is affected...
Richard Levitte [Fri, 22 May 2015 15:54:06 +0000 (17:54 +0200)]
Fix the update target and remove duplicate file updates
We had updates of certain header files in both Makefile.org and the
Makefile in the directory the header file lived in. This is error
prone and also sometimes generates slightly different results (usually
just a comment that differs) depending on which way the update was
done.
This removes the file update targets from the top level Makefile, adds
an update: target in all Makefiles and has it depend on the depend: or
local_depend: targets, whichever is appropriate, so we don't get a
double run through the whole file tree.
Matt Caswell [Fri, 22 May 2015 12:48:49 +0000 (13:48 +0100)]
Fix typo setting up certificate masks
The certificate masks are used to select which ciphersuite we are going to
use. The variables |emask_k| and |emask_a| relate to export grade key
exchange and authentication respecitively. The variables |mask_k| and
|mask_a| are the equivalent versions for non-export grade. This fixes an
instance where the two usages of export/non-export were mixed up. In
practice it makes little difference since it still works!
Matt Caswell [Fri, 22 May 2015 12:33:19 +0000 (13:33 +0100)]
Remove export static DH ciphersuites
Remove support for the two export grade static DH ciphersuites. These two
ciphersuites were newly added (along with a number of other static DH
ciphersuites) to 1.0.2. However the two export ones have *never* worked
since they were introduced. It seems strange in any case to be adding new
export ciphersuites, and given "logjam" it also does not seem correct to
fix them.
Matt Caswell [Tue, 19 May 2015 15:03:02 +0000 (16:03 +0100)]
Fix off-by-one in BN_rand
If BN_rand is called with |bits| set to 1 and |top| set to 1 then a 1 byte
buffer overflow can occur. There are no such instances within the OpenSSL at
the moment.
Thanks to Mateusz Kocielski (LogicalTrust), Marek Kroemeke, Filip Palian for
discovering and reporting this issue.
Matt Caswell [Tue, 19 May 2015 14:19:30 +0000 (15:19 +0100)]
Reject negative shifts for BN_rshift and BN_lshift
The functions BN_rshift and BN_lshift shift their arguments to the right or
left by a specified number of bits. Unpredicatable results (including
crashes) can occur if a negative number is supplied for the shift value.
Thanks to Mateusz Kocielski (LogicalTrust), Marek Kroemeke and Filip Palian
for discovering and reporting this issue.
Emilia Kasper [Tue, 12 May 2015 14:10:05 +0000 (16:10 +0200)]
Update documentation with Diffie-Hellman best practices.
- Do not advise generation of DH parameters with dsaparam to save
computation time.
- Promote use of custom parameters more, and explicitly forbid use of
built-in parameters weaker than 2048 bits.
- Advise the callback to ignore <keylength> - it is currently called
with 1024 bits, but this value can and should be safely ignored by
servers.
Rich Salz [Tue, 12 May 2015 15:49:32 +0000 (11:49 -0400)]
Add NULL checks from master
The big "don't check for NULL" cleanup requires backporting some
of the lowest-level functions to actually do nothing if NULL is
given. This will make it easier to backport fixes to release
branches, where master assumes those lower-level functions are "safe"
This commit addresses those tickets: 3798 3799 3801.
Hanno Böck [Mon, 11 May 2015 10:33:37 +0000 (11:33 +0100)]
Call of memcmp with null pointers in obj_cmp()
The function obj_cmp() (file crypto/objects/obj_dat.c) can in some
situations call memcmp() with a null pointer and a zero length.
This is invalid behaviour. When compiling openssl with undefined
behaviour sanitizer (add -fsanitize=undefined to compile flags) this
can be seen. One example that triggers this behaviour is the pkcs7
command (but there are others, e.g. I've seen it with the timestamp
function):
apps/openssl pkcs7 -in test/testp7.pem
What happens is that obj_cmp takes objects of the type ASN1_OBJECT and
passes their ->data pointer to memcmp. Zero-sized ASN1_OBJECT
structures can have a null pointer as data.
Matt Caswell [Wed, 6 May 2015 20:31:16 +0000 (21:31 +0100)]
Don't allow a CCS when expecting a CertificateVerify
Currently we set change_cipher_spec_ok to 1 before calling
ssl3_get_cert_verify(). This is because this message is optional and if it
is not sent then the next thing we would expect to get is the CCS. However,
although it is optional, we do actually know whether we should be receiving
one in advance. If we have received a client cert then we should expect
a CertificateVerify message. By the time we get to this point we will
already have bombed out if we didn't get a Certificate when we should have
done, so it is safe just to check whether |peer| is NULL or not. If it is
we won't get a CertificateVerify, otherwise we will. Therefore we should
change the logic so that we only attempt to get the CertificateVerify if
we are expecting one, and not allow a CCS in this scenario.
Whilst this is good practice for TLS it is even more important for DTLS.
In DTLS messages can be lost. Therefore we may be in a situation where a
CertificateVerify message does not arrive even though one was sent. In that
case the next message the server will receive will be the CCS. This could
also happen if messages get re-ordered in-flight. In DTLS if
|change_cipher_spec_ok| is not set and a CCS is received it is ignored.
However if |change_cipher_spec_ok| *is* set then a CCS arrival will
immediately move the server into the next epoch. Any messages arriving for
the previous epoch will be ignored. This means that, in this scenario, the
handshake can never complete. The client will attempt to retransmit
missing messages, but the server will ignore them because they are the wrong
epoch. The server meanwhile will still be waiting for the CertificateVerify
which is never going to arrive.
Matt Caswell [Thu, 23 Apr 2015 19:01:33 +0000 (20:01 +0100)]
Add Error state
Reusing an SSL object when it has encountered a fatal error can
have bad consequences. This is a bug in application code not libssl
but libssl should be more forgiving and not crash.
Matt Caswell [Mon, 4 May 2015 22:15:46 +0000 (23:15 +0100)]
Remove libcrypto to libssl dependency
Remove dependency on ssl_locl.h from v3_scts.c, and incidentally fix a build problem with
kerberos (the dependency meant v3_scts.c was trying to include krb5.h, but without having been
passed the relevanant -I flags to the compiler)
Richard Levitte [Mon, 4 May 2015 15:34:40 +0000 (17:34 +0200)]
RT2943: Check sizes if -iv and -K arguments
RT2943 only complains about the incorrect check of -K argument size,
we might as well do the same thing with the -iv argument.
Before this, we only checked that the given argument wouldn't give a
bitstring larger than EVP_MAX_KEY_LENGTH. we can be more precise and
check against the size of the actual cipher used.
Matt Caswell [Wed, 29 Apr 2015 12:22:18 +0000 (13:22 +0100)]
Fix buffer overrun in RSA signing
The problem occurs in EVP_PKEY_sign() when using RSA with X931 padding.
It is only triggered if the RSA key size is smaller than the digest length.
So with SHA512 you can trigger the overflow with anything less than an RSA
512 bit key. I managed to trigger a 62 byte overflow when using a 16 bit RSA
key. This wasn't sufficient to cause a crash, although your mileage may
vary.
In practice RSA keys of this length are never used and X931 padding is very
rare. Even if someone did use an excessively short RSA key, the chances of
them combining that with a longer digest and X931 padding is very
small. For these reasons I do not believe there is a security implication to
this. Thanks to Kevin Wojtysiak (Int3 Solutions) and Paramjot Oberoi (Int3
Solutions) for reporting this issue.
Matt Caswell [Wed, 29 Apr 2015 08:58:10 +0000 (09:58 +0100)]
Add sanity check to print_bin function
Add a sanity check to the print_bin function to ensure that the |off|
argument is positive. Thanks to Kevin Wojtysiak (Int3 Solutions) and
Paramjot Oberoi (Int3 Solutions) for reporting this issue.
Matt Caswell [Tue, 28 Apr 2015 14:28:23 +0000 (15:28 +0100)]
Add sanity check to ssl_get_prev_session
Sanity check the |len| parameter to ensure it is positive. Thanks to Kevin
Wojtysiak (Int3 Solutions) and Paramjot Oberoi (Int3 Solutions) for
reporting this issue.
Matt Caswell [Tue, 28 Apr 2015 14:19:50 +0000 (15:19 +0100)]
Sanity check the return from final_finish_mac
The return value is checked for 0. This is currently safe but we should
really check for <= 0 since -1 is frequently used for error conditions.
Thanks to Kevin Wojtysiak (Int3 Solutions) and Paramjot Oberoi (Int3
Solutions) for reporting this issue.
Matt Caswell [Mon, 27 Apr 2015 14:41:42 +0000 (15:41 +0100)]
Add sanity check in ssl3_cbc_digest_record
For SSLv3 the code assumes that |header_length| > |md_block_size|. Whilst
this is true for all SSLv3 ciphersuites, this fact is far from obvious by
looking at the code. If this were not the case then an integer overflow
would occur, leading to a subsequent buffer overflow. Therefore I have
added an explicit sanity check to ensure header_length is always valid.
Thanks to Kevin Wojtysiak (Int3 Solutions) and Paramjot Oberoi (Int3
Solutions) for reporting this issue.
Matt Caswell [Mon, 27 Apr 2015 14:41:03 +0000 (15:41 +0100)]
Clarify logic in BIO_*printf functions
The static function dynamically allocates an output buffer if the output
grows larger than the static buffer that is normally used. The original
logic implied that |currlen| could be greater than |maxlen| which is
incorrect (and if so would cause a buffer overrun). Also the original
logic would call OPENSSL_malloc to create a dynamic buffer equal to the
size of the static buffer, and then immediately call OPENSSL_realloc to
make it bigger, rather than just creating a buffer than was big enough in
the first place. Thanks to Kevin Wojtysiak (Int3 Solutions) and Paramjot
Oberoi (Int3 Solutions) for reporting this issue.
Matt Caswell [Mon, 27 Apr 2015 10:13:56 +0000 (11:13 +0100)]
Sanity check EVP_EncodeUpdate buffer len
There was already a sanity check to ensure the passed buffer length is not
zero. Extend this to ensure that it also not negative. Thanks to Kevin
Wojtysiak (Int3 Solutions) and Paramjot Oberoi (Int3 Solutions) for
reporting this issue.
Matt Caswell [Mon, 27 Apr 2015 10:07:06 +0000 (11:07 +0100)]
Sanity check EVP_CTRL_AEAD_TLS_AAD
The various implementations of EVP_CTRL_AEAD_TLS_AAD expect a buffer of at
least 13 bytes long. Add sanity checks to ensure that the length is at
least that. Also add a new constant (EVP_AEAD_TLS1_AAD_LEN) to evp.h to
represent this length. Thanks to Kevin Wojtysiak (Int3 Solutions) and
Paramjot Oberoi (Int3 Solutions) for reporting this issue.
Matt Caswell [Mon, 27 Apr 2015 10:04:56 +0000 (11:04 +0100)]
Sanity check DES_enc_write buffer length
Add a sanity check to DES_enc_write to ensure the buffer length provided
is not negative. Thanks to Kevin Wojtysiak (Int3 Solutions) and Paramjot
Oberoi (Int3 Solutions) for reporting this issue.
Matt Caswell [Wed, 29 Apr 2015 15:15:40 +0000 (16:15 +0100)]
Add length sanity check in SSLv2 n_do_ssl_write()
Fortify flagged up a problem in n_do_ssl_write() in SSLv2. Analysing the
code I do not believe there is a real problem here. However the logic flows
are complicated enough that a sanity check of |len| is probably worthwhile.
Thanks to Kevin Wojtysiak (Int3 Solutions) and Paramjot Oberoi (Int3
Solutions) for reporting this issue.
The above commit was backported to the 1.0.2 branch as part of backporting
the alternative chain verify algorithm changes. However it has been pointed
out (credit to Shigeki Ohtsu) that this is unnecessary in 1.0.2 as this
commit is a work around for loop checking that only exists in master.
Reviewed-by: Richard Levitte <levitte@openssl.org>
The function CRYPTO_strdup (aka OPENSSL_strdup) fails to check the return
value from CRYPTO_malloc to see if it is NULL before attempting to use it.
This patch adds a NULL check.
This change partially reverts the commits and brings the lookahead back
in reduced capacity for TLS + EAP-FAST only. Since EAP-FAST does not
support regular session tickets, the lookahead now only checks for a
Finished message.
Regular handshakes are unaffected by this change.
Reviewed-by: David Benjamin <davidben@chromium.org> Reviewed-by: Matt Caswell <matt@openssl.org>
(cherry picked from commit 6e3d015363ed09c4eff5c02ad41153387ffdf5af)
Andy Polyakov [Fri, 23 Jan 2015 16:27:10 +0000 (17:27 +0100)]
aes/asm/aesni-x86[_64].pl update.
This addresses
- request for improvement for faster key setup in RT#3576;
- clearing registers and stack in RT#3554 (this is more of a gesture to
see if there will be some traction from compiler side);
- more commentary around input parameters handling and stack layout
(desired when RT#3553 was reviewed);
- minor size and single block performance optimization (was lying around);
Disable loop checking when we retry verification with an alternative path.
This fixes the case where an intermediate CA is explicitly trusted and part
of the untrusted certificate list. By disabling loop checking for this case
the untrusted CA can be replaced by the explicitly trusted case and
verification will succeed.
Matt Caswell [Tue, 27 Jan 2015 10:50:38 +0000 (10:50 +0000)]
Add -no_alt_chains option to apps to implement the new
X509_V_FLAG_NO_ALT_CHAINS flag. Using this option means that when building
certificate chains, the first chain found will be the one used. Without this
flag, if the first chain found is not trusted then we will keep looking to
see if we can build an alternative chain instead.
Matt Caswell [Tue, 27 Jan 2015 10:03:29 +0000 (10:03 +0000)]
In certain situations the server provided certificate chain may no longer be
valid. However the issuer of the leaf, or some intermediate cert is in fact
in the trust store.
When building a trust chain if the first attempt fails, then try to see if
alternate chains could be constructed that are trusted.
A 0-length ciphers list is never permitted. The old code only used to
reject an empty ciphers list for connections with a session ID. It
would later error out on a NULL structure, so this change just moves
the alert closer to the problem source.
The disabled set of -Weverything is hard to maintain across versions.
Use -Wall -Wextra but also document other useful warnings that currently trigger.
Matt Caswell [Fri, 10 Apr 2015 15:49:33 +0000 (16:49 +0100)]
Fix ssl_get_prev_session overrun
If OpenSSL is configured with no-tlsext then ssl_get_prev_session can read
past the end of the ClientHello message if the session_id length in the
ClientHello is invalid. This should not cause any security issues since the
underlying buffer is 16k in size. It should never be possible to overrun by
that many bytes.
This is probably made redundant by the previous commit - but you can never be
too careful.
With thanks to Qinghao Tang for reporting this issue.
Matt Caswell [Fri, 10 Apr 2015 16:25:27 +0000 (17:25 +0100)]
Check for ClientHello message overruns
The ClientHello processing is insufficiently rigorous in its checks to make
sure that we don't read past the end of the message. This does not have
security implications due to the size of the underlying buffer - but still
needs to be fixed.
With thanks to Qinghao Tang for reporting this issue.
While *pval is usually a pointer in rare circumstances it can be a long
value. One some platforms (e.g. WIN64) where
sizeof(long) < sizeof(ASN1_VALUE *) this will write past the field.
*pval is initialised correctly in the rest of ASN1_item_ex_new so setting it
to NULL is unecessary anyway.
Thanks to Julien Kauffmann for reporting this issue.
Richard Levitte [Wed, 8 Apr 2015 17:26:11 +0000 (19:26 +0200)]
Have mkerr.pl treat already existing multiline string defs properly
Since source reformat, we ended up with some error reason string
definitions that spanned two lines. That in itself is fine, but we
sometimes edited them to provide better strings than what could be
automatically determined from the reason macro, for example:
{ERR_REASON(SSL_R_NO_GOST_CERTIFICATE_SENT_BY_PEER),
"Peer haven't sent GOST certificate, required for selected ciphersuite"},
However, mkerr.pl didn't treat those two-line definitions right, and
they ended up being retranslated to whatever the macro name would
indicate, for example:
{ERR_REASON(SSL_R_NO_GOST_CERTIFICATE_SENT_BY_PEER),
"No gost certificate sent by peer"},
Clearly not what we wanted. This change fixes this problem.
Richard Levitte [Sat, 4 Apr 2015 14:53:44 +0000 (16:53 +0200)]
Appease clang -Wshadow
The macros BSWAP4 and BSWAP8 have statetemnt expressions
implementations that use local variable names that shadow variables
outside the macro call, generating warnings like this
e_aes_cbc_hmac_sha1.c:263:14: warning: declaration shadows a local variable
[-Wshadow]
seqnum = BSWAP8(blocks[0].q[0]);
^
../modes/modes_lcl.h:41:29: note: expanded from macro 'BSWAP8'
^
e_aes_cbc_hmac_sha1.c:223:12: note: previous declaration is here
size_t ret = 0;
^
Have clang be quiet by modifying the macro variable names slightly
(suffixing them with an underscore).
Richard Levitte [Sat, 4 Apr 2015 14:33:20 +0000 (16:33 +0200)]
Appease clang -Wgnu-statement-expression
We use GNU statement expressions in crypto/md32_common.h, surrounded
by checks that GNU C is indeed used to compile. It seems that clang,
at least on Linux, pretends to be GNU C, therefore finds the statement
expressions and then warns about them.
Harden SSLv2-supporting servers against Bleichenbacher's attack.
There is no indication that the timing differences are exploitable in
OpenSSL, and indeed there is some indication (Usenix '14) that they
are too small to be exploitable. Nevertheless, be careful and apply
the same countermeasures as in s3_srvr.c
Thanks to Nimrod Aviram, Sebastian Schinzel and Yuval Shavitt for
reporting this issue.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Andy Polyakov [Tue, 3 Mar 2015 21:05:25 +0000 (22:05 +0100)]
aes/asm/aesv8-armx.pl: optimize for Cortex-A5x.
ARM has optimized Cortex-A5x pipeline to favour pairs of complementary
AES instructions. While modified code improves performance of post-r0p0
Cortex-A53 performance by >40% (for CBC decrypt and CTR), it hurts
original r0p0. We favour later revisions, because one can't prevent
future from coming. Improvement on post-r0p0 Cortex-A57 exceeds 50%,
while new code is not slower on r0p0, or Apple A7 for that matter.
Douglas E Engert [Wed, 25 Mar 2015 23:52:28 +0000 (23:52 +0000)]
Ensure EC private keys retain leading zeros
RFC5915 requires the use of the I2OSP primitive as defined in RFC3447
for storing an EC Private Key. This converts the private key into an
OCTETSTRING and retains any leading zeros. This commit ensures that those
leading zeros are present if required.