Matt Caswell [Mon, 12 Oct 2015 11:40:15 +0000 (12:40 +0100)]
Centralise loading default apps config file
Loading the config file after processing command line options can
cause problems, e.g. where an engine provides new ciphers/digests
these are not then recoginised on the command line. Move the
default config file loading to before the command line option
processing. Whilst we're doing this we might as well centralise
this instead of doing it individually for each application. Finally
if we do it before the OpenSSL_add_ssl_algorithms() call then
ciphersuites provided by an engine (e.g. GOST) can be available to
the apps.
RT#4085
RT#4086
Reviewed-by: Richard Levitte <levitte@openssl.org>
Emilia Kasper [Tue, 6 Oct 2015 15:20:32 +0000 (17:20 +0200)]
DTLS: remove unused cookie field
Note that this commit constifies a user callback parameter and therefore
will break compilation for applications using this callback. But unless
they are abusing write access to the buffer, the fix is trivial.
Matt Caswell [Thu, 8 Oct 2015 12:36:10 +0000 (13:36 +0100)]
Don't treat a bare OCTETSTRING as DigestInfo in int_rsa_verify
The function int_rsa_verify is an internal function used for verifying an
RSA signature. It takes an argument |dtype| which indicates the digest type
that was used. Dependant on that digest type the processing of the
signature data will vary. In particular if |dtype == NID_mdc2| and the
signature data is a bare OCTETSTRING then it is treated differently to the
default case where the signature data is treated as a DigestInfo (X509_SIG).
Due to a missing "else" keyword the logic actually correctly processes the
OCTETSTRING format signature first, and then attempts to continue and
process it as DigestInfo. This will invariably fail because we already know
that it is a bare OCTETSTRING.
This failure doesn't actualy make a real difference because it ends up at
the |err| label regardless and still returns a "success" result. This patch
just cleans things up to make it look a bit more sane.
RT#4076
Reviewed-by: Richard Levitte <levitte@openssl.org>
Matt Caswell [Wed, 7 Oct 2015 14:20:47 +0000 (15:20 +0100)]
Don't advance PACKET in ssl_check_for_safari
The function ssl_check_for_safari fingerprints the incoming extensions
to see whether it is one of the broken versions of safari. However it was
failing to reset the PACKET back to the same position it started in, hence
causing some extensions to be skipped incorrectly.
-Allow mingw debug builds to fail on Travis CI
-Fix Travis email notifications config
-Rename a variable to avoid a bogus warning with old GCC
error: declaration of ``dup'' shadows a global declaration [-Werror=shadow]
-Disable pedantic ms-format warnings with mingw
-Properly define const DH parameters
-Restore --debug flag in Travis CI builds; -d would get incorrectly passed
to ./Configure in mingw debug builds.
Signed-off-by: Rich Salz <rsalz@openssl.org> Reviewed-by: Matt Caswell <matt@openssl.org>
Move all packet parsing to the beginning of the method. This limits the
SSLv2 compatibility soup to the parsing, and makes the rest of the
processing uniform.
This is also needed for simpler EMS support: EMS servers need to do an
early scan for EMS to make resumption decisions. This'll be easier when
the entire ClientHello is parsed in the beginning.
As a side effect,
1) PACKETize ssl_get_prev_session and tls1_process_ticket; and
2) Delete dead code for SSL_OP_NETSCAPE_REUSE_CIPHER_CHANGE_BUG.
Richard Levitte [Sun, 4 Oct 2015 19:12:03 +0000 (21:12 +0200)]
Fix make depend for things being built in subdirectories
Some makedepend mechanisms remove all directory information in the
target, so a dependency can looks like this:
ssl3_record.o: record/ssl3_record.c
However, that doesn't quite suit us, our Makefile has us build
record/ssl3_record.o rather than ssl3_record.o.
To clear this up, a change to util/clean-depend.pl takes care of this
case by looking up the original file in the dependencies and restoring
the directory information from it.
Andy Polyakov [Wed, 30 Sep 2015 08:15:03 +0000 (10:15 +0200)]
Explicitly cast INVALID_SOCKET to (int) to address warnings on Windows.
Even though SOCKET is effectively declared as (void *) on Windows, it's
not actually a pointer, but an index within per-process table of
kernel objects. The table size is actually limited and its upper limit
is far below upper limit for signed 32-bit integer. This is what makes
cast in question possible.
Reviewed-by: Richard Levitte <levitte@openssl.org> Reviewed-by: Tim Hudson <tjh@openssl.org>
When a decoding error in ASN.1 occurs only free up the partial structure
at the top level. This simplifies embedded handling and fixes freeing
up of structures when presented with malformed input.
Matt Caswell [Tue, 29 Sep 2015 10:14:35 +0000 (11:14 +0100)]
Change the DEFAULT ciphersuites to exclude DES, RC4 and RC2
This patch updates the "DEFAULT" cipherstring to be
"ALL:!COMPLEMENTOFDEFAULT:!eNULL". COMPLEMENTOFDEFAULT is now defined
internally by a flag on each ciphersuite indicating whether it should be
excluded from DEFAULT or not. This gives us control at an individual
ciphersuite level as to exactly what is in DEFAULT and what is not.
Finally all DES, RC4 and RC2 ciphersuites are added to COMPLEMENTOFDEFAULT
and hence removed from DEFAULT.
Looks like someone forgot to do a "make update" since crypto/ts/Makefile
keeps changing. So include that.
Second is that the declare_dh_bn macro fools the libeay.num script.
The declarations are only needed in one file (dh_rfc5114) so remove
them from the header and put the "raw" declarations directly into that
file.
Reviewed-by: Richard Levitte <levitte@openssl.org>
David Woodhouse [Fri, 11 Sep 2015 18:56:32 +0000 (14:56 -0400)]
Fix no-stdio build
Much related/similar work also done by
Ivan Nestlerode <ivan.nestlerode@sonos.com>
+Replace FILE BIO's with dummy ops that fail.
+Include <stdio.h> for sscanf() even with no-stdio (since the declaration
is there). We rely on sscanf() to parse the OPENSSL_ia32cap environment
variable, since it can be larger than a 'long'. And we don't rely on the
availability of strtoull().
+Remove OPENSSL_stderr(); not used.
+Make OPENSSL_showfatal() do nothing (currently without stdio there's
nothing we can do).
+Remove file-based functionality from ssl/. The function
prototypes were already gone, but not the functions themselves.
+Remove unviable conf functionality via SYS_UEFI
+Add fallback definition of BUFSIZ.
+Remove functions taking FILE * from header files.
+Add missing DECLARE_PEM_write_fp_const
+Disable X509_LOOKUP_hash_dir(). X509_LOOKUP_file() was already compiled out,
so remove its prototype.
+Use OPENSSL_showfatal() in CRYPTO_destroy_dynlockid().
+Eliminate SRP_VBASE_init() and supporting functions. Users will need to
build the verifier manually instead.
+Eliminate compiler warning for unused do_pk8pkey_fp().
+Disable TEST_ENG_OPENSSL_PKEY.
+Disable GOST engine as is uses [f]printf all over the place.
+Eliminate compiler warning for unused send_fp_chars().
Signed-off-by: Rich Salz <rsalz@akamai.com> Reviewed-by: Tim Hudson <tjh@openssl.org>
RFC 5077 section 3.3 says:
If the server determines that it does not want to include a
ticket after it has included the SessionTicket extension in the
ServerHello, then it sends a zero-length ticket in the
NewSessionTicket handshake message.
Previously the client would fail upon attempting to allocate a
zero-length buffer. Now, we have the client ignore the empty ticket and
keep the existing session.
Matt Caswell [Tue, 22 Sep 2015 07:54:43 +0000 (08:54 +0100)]
Change ossltest engine to manually allocate cipher_data
The ossltest engine wraps the built-in implementation of aes128-cbc.
Normally in an engine the cipher_data structure is automatically allocated
by the EVP layer. However this relies on the engine specifying up front
the size of that cipher_data structure. In the case of ossltest this value
isn't available at compile time. This change makes the ossltest engine
allocate its own cipher_data structure instead of leaving it to the EVP
layer.
Matt Caswell [Tue, 22 Sep 2015 15:02:50 +0000 (16:02 +0100)]
Fix the OCSP test on Windows
The windows test uses the pseudo file "nul" to indicate no file for the
-CApath option. This does not work on all versions of Windows. Instead use
the new -no-CApath option.
Matt Caswell [Tue, 22 Sep 2015 15:00:52 +0000 (16:00 +0100)]
Add support for -no-CApath and -no-CAfile options
For those command line options that take the verification options
-CApath and -CAfile, if those options are absent then the default path or
file is used instead. It is not currently possible to specify *no* path or
file at all. This change adds the options -no-CApath and -no-CAfile to
specify that the default locations should not be used to all relevant
applications.
Matt Caswell [Wed, 23 Sep 2015 09:02:18 +0000 (10:02 +0100)]
Fix s_server DTLSv1_listen issues
Use sockaddr_storage not sockaddr for the client IP address to allow for
IPv6.
Also fixed a section of code which was conditional on OPENSSL_NO_DTLS1
which should not have been.
Matt Caswell [Fri, 10 Apr 2015 12:10:05 +0000 (13:10 +0100)]
Add -listen documentation
This commit adds documentation for the new -listen option to s_server. Along
the way it also adds documentation for -dtls, -dtls1 and -dtls1_2 which was
missing.
Matt Caswell [Thu, 9 Apr 2015 09:01:05 +0000 (10:01 +0100)]
Add support for DTLSv1_listen in s_server
DTLSv1_listen is a commonly used function within DTLS solutions for
listening for new incoming connections. This commit adds support to s_server
for using it.
Matt Caswell [Mon, 14 Sep 2015 21:49:35 +0000 (22:49 +0100)]
DTLSv1_listen rewrite
The existing implementation of DTLSv1_listen() is fundamentally flawed. This
function is used in DTLS solutions to listen for new incoming connections
from DTLS clients. A client will send an initial ClientHello. The server
will respond with a HelloVerifyRequest containing a unique cookie. The
client the responds with a second ClientHello - which this time contains the
cookie.
Once the cookie has been verified then DTLSv1_listen() returns to user code,
which is typically expected to continue the handshake with a call to (for
example) SSL_accept().
Whilst listening for incoming ClientHellos, the underlying BIO is usually in
an unconnected state. Therefore ClientHellos can come in from *any* peer.
The arrival of the first ClientHello without the cookie, and the second one
with it, could be interspersed with other intervening messages from
different clients.
The whole purpose of this mechanism is as a defence against DoS attacks. The
idea is to avoid allocating state on the server until the client has
verified that it is capable of receiving messages at the address it claims
to come from. However the existing DTLSv1_listen() implementation completely
fails to do this. It attempts to super-impose itself on the standard state
machine and reuses all of this code. However the standard state machine
expects to operate in a stateful manner with a single client, and this can
cause various problems.
A second more minor issue is that the return codes from this function are
quite confused, with no distinction made between fatal and non-fatal errors.
Most user code treats all errors as non-fatal, and simply retries the call
to DTLSv1_listen().
This commit completely rewrites the implementation of DTLSv1_listen() and
provides a stand alone implementation that does not rely on the existing
state machine. It also provides more consistent return codes.