Fix a bug in handling of 128 byte long PSK identity in
psk_client_callback.
OpenSSL supports PSK identities of up to (and including) 128 bytes in
length. PSK identity is obtained via the psk_client_callback,
implementors of which are expected to provide a NULL-terminated
identity. However, the callback is invoked with only 128 bytes of
storage thus making it impossible to return a 128 byte long identity and
the required additional NULL byte.
This CL fixes the issue by passing in a 129 byte long buffer into the
psk_client_callback. As a safety precaution, this CL also zeroes out the
buffer before passing it into the callback, uses strnlen for obtaining
the length of the identity returned by the callback, and aborts the
handshake if the identity (without the NULL terminator) is longer than
128 bytes.
https://rt.openssl.org/Ticket/Display.html?id=2608
Previously, this input to the base64 code:
================================================================================-
Would cause the output length of EVP_DecodeUpdate to be negative. When
that happened in the base64 BIO, it would crash. In PEM decoding, the
ASN.1 code actually maintains signed lengths and manages to simply error
out!
Move ECC SSL extensions to the end.
WebSphere Application Server 7.0 appears to be intolerant of an empty
extension at the end. To that end, also ensure we never send an empty
padding extension.
PSK identity hint can be stored in SSL_CTX and in SSL/SSL_SESSION,
similar to other TLS parameters, with the value in SSL/SSL_SESSION
taking precedence over the one in SSL_CTX. The value in SSL_CTX is
shared (used as the default) between all SSL instances associated
with that SSL_CTX, whereas the value in SSL/SSL_SESSION is confined
to that particular TLS/SSL connection/session.
The existing implementation of TLS-PSK does not correctly distinguish
between PSK identity hint in SSL_CTX and in SSL/SSL_SESSION. This
change fixes these issues:
1. SSL_use_psk_identity_hint does nothing and returns "success" when
the SSL object does not have an associated SSL_SESSION.
2. On the client, the hint in SSL_CTX (which is shared between
multiple SSL instances) is overwritten with the hint received from
server or reset to NULL if no hint was received.
3. On the client, psk_client_callback is invoked with the hint from
SSL_CTX rather than from current SSL/SSL_SESSION (i.e., the one
received from the server). Issue #2 above masks this issue.
4. On the server, the hint in SSL/SSL_SESSION is ignored and the hint
from SSL_CTX is sent to the client.
5. On the server, the hint in SSL/SSL_SESSION is reset to the one in
SSL_CTX after the ClientKeyExchange message step.
This change fixes the issues by:
* Adding storage for the hint in the SSL object. The idea being that
the hint in the associated SSL_SESSION takes precedence.
* Reading the hint during the handshake only from the associated
SSL_SESSION object.
* Initializing the hint in SSL object with the one from the SSL_CTX
object.
* Initializing the hint in SSL_SESSION object with the one from the
SSL object.
* Making SSL_use_psk_identity_hint and SSL_get_psk_identity_hint
set/get the hint to/from SSL_SESSION associated with the provided
SSL object, or, if no SSL_SESSION is available, set/get the hint
to/from the provided SSL object.
* Removing code which resets the hint during handshake.
Fix handling of points at infinity in ec_GFp_simple_points_make_affine.
When inverting an array of Z coordinates, the algorithm is supposed to
treat any 0 essentially like a 1 to remain in the multiplicative group;
however, for one of the cases, we incorrectly multiplied by 0 and thus
ended up with garbage.
Fix limit checks in ssl_add_clienthello_tlsext and
ssl_add_serverhello_tlsext.
Some of the limit checks reference p rather than ret. p is the original
buffer position, not the current one. Fix those and rename p to orig so
it's clearer.
This change implements equal-preference groups of cipher suites. This
allows, for example, a server to prefer one of AES-GCM or ChaCha20
ciphers, but to allow the client to pick which one. When coupled with
clients that will boost AES-GCM in their preferences when AES-NI is
present, this allows us to use AES-GCM when the hardware exists and
ChaCha20 otherwise.
This patch removes support for empty records (which is almost
universally disabled via SSL_OP_ALL) and adds optional support for 1/n-1
record splitting.
The latter is not enabled by default, since it's not typically used on
servers, but it should be enabled in web browsers since there are known
attacks in that case (see BEAST).
This change adjusts the stack pointer during CBC decryption. The code
was previously using the red zone across function calls and valgrind
thinks that the "unused" stack is undefined after a function call.
Some RSA private keys are specified with only n, e and d. Although we
can use these keys directly, it's nice to have a uniform representation
that includes the precomputed CRT values. This change adds a function
that can recover the primes from a minimal private key of that form.
This patch adds support for a different cipher list when the connection
is using TLS 1.1. This is intended to support the case where we want to
use AES with >= TLS 1.1 clients but RC4 otherwise because of the BEAST
attack.
This change saves several EC routines from crashing when an EC_KEY is
missing a public key. The public key is optional in the EC private key
format and, without this patch, running the following through `openssl
ec` causes a crash:
-----BEGIN EC PRIVATE KEY-----
MBkCAQEECAECAwQFBgcIoAoGCCqGSM49AwEH
-----END EC PRIVATE KEY-----
Ensure that, when generating small primes, the result is actually of the
requested size. Fixes OpenSSL #2701.
This change does not address the cases of generating safe primes, or
where the |add| parameter is non-NULL.
This change adds functions to check membership of various cipher
families. Clients and servers need this in order to optimise the size of
records because different families have different amounts of prefix and
postfix overhead.
Limit the number of empty records that will be processed consecutively
in order to prevent ssl3_get_record from never returning.
Reported by "oftc_must_be_destroyed" and George Kadianakis.
(The issue was reported by Shay Gueron.)
The final reduction in Montgomery multiplication computes if (X >= m) then X =
X - m else X = X
In OpenSSL, this was done by computing T = X - m, doing a constant-time
selection of the *addresses* of X and T, and loading from the resulting
address. But this is not cache-neutral.
This patch changes the behaviour by loading both X and T into registers, and
doing a constant-time selection of the *values*.
TODO(fork): only some of the fixes from the original patch still apply to
the 1.0.2 code.
