Reject elliptic curve lists of odd lengths.

The Supported Elliptic Curves extension contains a vector of NamedCurves
of 2 bytes each, so the total length must be even. Accepting odd-length
lists was observed to lead to a non-exploitable one-byte out-of-bounds
read in the latest development branches (1.0.2 and master). Released
versions of OpenSSL are not affected.

Thanks to Felix Groebert of the Google Security Team for reporting this issue.

Reviewed-by: Matt Caswell <matt@openssl.org>

Fix broken build

Add includes missing from commit 33eab3f6af51c6d7fe68ee4960930df4f1616bd5

Reviewed-by: Geoff Thorpe <geoff@openssl.org>

Replace GOST_R_MALLOC_FAILURE and GOST_R_NO_MEMORY with ERR_R_MALLOC_FAILURE

Reviewed-by: Richard Levitte <levitte@openssl.org>

capi_get_provname: Check return values

Reviewed-by: Richard Levitte <levitte@openssl.org>

ssl_create_cipher_list: check whether push onto cipherstack succeeds

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Richard Levitte <levitte@openssl.org>

ssl_cert_dup: Fix memory leak

Always use goto err on failure and call ssl_cert_free() on the error path so all
fields and "ret" itself are freed

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Richard Levitte <levitte@openssl.org>

dtls1_new: free s on error path

Reviewed-by: Richard Levitte <levitte@openssl.org>

dtls1_heartbeat: check for NULL after allocating s->cert->ctypes

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Richard Levitte <levitte@openssl.org>

dtls1_process_heartbeat: check for NULL after allocating buffer

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Richard Levitte <levitte@openssl.org>

capi_get_key: check for NULL after allocating key

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Richard Levitte <levitte@openssl.org>

capi_cert_get_fname: check for NULL after allocating wfname

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Richard Levitte <levitte@openssl.org>

capi_get_provname: free name on error if it was malloc'ed

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Richard Levitte <levitte@openssl.org>

pkey_gost_mac_keygen: check for NULL after allocating keydata

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Richard Levitte <levitte@openssl.org>

pkey_gost_ctrl: check for NULL after allocating pctx->shared_ukm

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Richard Levitte <levitte@openssl.org>

Update changes to indicate that SSLv2 support has been removed

Reviewed-by: Matt Caswell <matt@openssl.org>

Remove incorrect code inadvertently introduced through commit 59669b6ab.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Remove SSLv2 support

The only support for SSLv2 left is receiving a SSLv2 compatible client hello.

Reviewed-by: Richard Levitte <levitte@openssl.org>

New location on website for binaries.

Reviewed-by: Bodo Moeller <bodo@openssl.org>

Remove "#if 0" code

Reviewed-by: Tim Hudson <tjh@openssl.org>

Only use the fallback mtu after 2 unsuccessful retransmissions if it is less
than the mtu we are already using

Reviewed-by: Tim Hudson <tjh@openssl.org>

Updates to s_client and s_server to remove the constant 28 (for IPv4 header
and UDP header) when setting an mtu. This constant is not always correct (e.g.
if using IPv6). Use the new DTLS_CTRL functions instead.

Reviewed-by: Tim Hudson <tjh@openssl.org>

If we really get a situation where the underlying mtu is less than the minimum
we will support then dtls1_do_write can go into an infinite loop. This commit
fixes that.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Fix dtls_query_mtu so that it will always either complete with an mtu that is
at least the minimum or it will fail.
There were some instances in dtls1_query_mtu where the final mtu can end up
being less than the minimum, i.e. where the user has set an mtu manually. This
shouldn't be allowed. Also remove dtls1_guess_mtu that, despite having
logic for guessing an mtu, was actually only ever used to work out the minimum
mtu to use.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Remove instances in libssl of the constant 28 (for size of IPv4 header + UDP)
and instead use the value provided by the underlying BIO. Also provide some
new DTLS_CTRLs so that the library user can set the mtu without needing to
know this constant. These new DTLS_CTRLs provide the capability to set the
link level mtu to be used (i.e. including this IP/UDP overhead). The previous
DTLS_CTRLs required the library user to subtract this overhead first.

Reviewed-by: Tim Hudson <tjh@openssl.org>

There are a number of instances throughout the code where the constant 28 is
used with no explanation. Some of this was introduced as part of RT#1929. The
value 28 is the length of the IP header (20 bytes) plus the UDP header (8
bytes). However use of this constant is incorrect because there may be
instances where a different value is needed, e.g. an IPv4 header is 20 bytes
but an IPv6 header is 40. Similarly you may not be using UDP (e.g. SCTP).
This commit introduces a new BIO_CTRL that provides the value to be used for
this mtu "overhead". It will be used by subsequent commits.

Reviewed-by: Tim Hudson <tjh@openssl.org>

The first call to query the mtu in dtls1_do_write correctly checks that the
mtu that we have received is not less than the minimum. If its less it uses the
minimum instead. The second call to query the mtu does not do that, but
instead uses whatever comes back. We have seen an instance in RT#3592 where we
have got an unreasonably small mtu come back. This commit makes both query
checks consistent.

Reviewed-by: Tim Hudson <tjh@openssl.org>

The SSL_OP_NO_QUERY_MTU option is supposed to stop the mtu from being
automatically updated, and we should use the one provided instead.
Unfortunately there are a couple of locations where this is not respected.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Verify that we have a sensible message len and fail if not
RT#3592 provides an instance where the OPENSSL_assert that this commit
replaces can be hit. I was able to recreate this issue by forcing the
underlying BIO to misbehave and come back with very small mtu values. This
happens the second time around the while loop after we have detected that the
MTU has been exceeded following the call to dtls1_write_bytes.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Use the SSLv23 method by default

If SSLv2 and SSLv3 are both disabled we still support SSL/TLS.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Remove all .cvsignore files

Reviewed-by: Tim Hudson <tjh@openssl.org>

Check for FindNextFile when defining it rather than FindFirstFile

Reviewed-by: Matt Caswell <matt@openssl.org>

[PR3597] Advance to the next state variant when reusing messages.

Previously, state variant was not advanced, which resulted in state
being stuck in the st1 variant (usually "_A").

This broke certificate callback retry logic when accepting connections
that were using SSLv2 ClientHello (hence reusing the message), because
their state never advanced to SSL3_ST_SR_CLNT_HELLO_C variant required
for the retry code path.

Reported by Yichun Zhang (agentzh).

Signed-off-by: Piotr Sikora <piotr@cloudflare.com>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>

Correct some layout issues, convert all remaining tabs to appropriate amounts of spaces.

Reviewed-by: Matt Caswell <matt@openssl.org>

Improves the proxy certificates howto doc.

The current documentation contains a bunch of spelling and grammar mistakes. I also
found it hard to understand some paragraphs, so here is my attempt to improve its
readability.

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>

Fix warning in ssl2_enc

Reviewed-by: Richard Levitte <levitte@openssl.org>

Remove more references to dtls1_enc

Reviewed-by: Richard Levitte <levitte@openssl.org>

Check EVP_Cipher return values for SSL2

Reviewed-by: Richard Levitte <levitte@openssl.org>

Delete unused file

Reviewed-by: Richard Levitte <levitte@openssl.org>

Add checks to the return value of EVP_Cipher to prevent silent encryption failure.

PR#1767

Reviewed-by: Richard Levitte <levitte@openssl.org>

Remove redundant checks in ssl_cert_dup. This was causing spurious error messages when using GOST

PR#3613

Reviewed-by: Richard Levitte <levitte@openssl.org>

Tidy up ocsp help output

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Add documentation on -timeout option in the ocsp utility

PR#3612

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

NetWare compilation fix.

Workaround for NetWare CodeWarrior compiler which doesn't properly lookup
includes when in same directory as the C file which includes it.

PR#3569
Reviewed-by: Stephen Henson <steve@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

Updates to X509_NAME_get_index_by_NID.pod submitted by user Bernardh via the wiki
Minor changes made by Matt Caswell

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Updates to X509_NAME_add_entry_by_txt.pod submitted by user Bernardh via the wiki
Minor changes made by Matt Caswell.

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Updates to EVP_PKEY_encrypt.pod submitted by user Bernardh via the wiki
Minor changes made by Matt Caswell.

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Add include of ssl.h which is required by srtp.h

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Fixed memory leak due to incorrect freeing of DTLS reassembly bit mask

PR#3608

Reviewed-by: Tim Hudson <tjh@openssl.org>

Corrected comments in ssl.h about SSLv23_method and friends

PR#3574

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Print out Suite B status.

When using the -xcert option to test certificate validity print out
if we pass Suite B compliance. We print out "not tested" if we aren't
in Suite B mode.
Reviewed-by: Matt Caswell <matt@openssl.org>

Fix SuiteB chain checking logic.

Reviewed-by: Matt Caswell <matt@openssl.org>

RT2679: Fix error if keysize too short

In keygen, return KEY_SIZE_TOO_SMALL not INVALID_KEYBITS.

** I also increased the minimum from 256 to 512, which is now
documented in CHANGES file. **

Reviewed-by: Matt Caswell <matt@openssl.org>

RT3488: Update doc for OPENSSL_config

Fix CONF_load_modules to CONF_modules_load.
Document that it calls exit.
Advise against using it now.
Add an error print to stderr.

Reviewed-by: Matt Caswell <matt@openssl.org>

Do not resume a session if the negotiated protocol version does not match
the session's version (server).

See also BoringSSL's commit bdf5e72f50e25f0e45e825c156168766d8442dde.

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

When using EVP_PKEY_derive with a KDF set, a negative error from
ECDH_compute_key is silently ignored and the KDF is run on duff data

Thanks to github user tomykaira for the suggested fix.

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Clean up CHANGES

Reviewed-by: Matt Caswell <matt@openssl.org>

Ensure SSL3_FLAGS_CCS_OK (or d1->change_cipher_spec_ok for DTLS) is reset
once the ChangeCipherSpec message is received. Previously, the server would
set the flag once at SSL3_ST_SR_CERT_VRFY and again at SSL3_ST_SR_FINISHED.
This would allow a second CCS to arrive and would corrupt the server state.

(Because the first CCS would latch the correct keys and subsequent CCS
messages would have to be encrypted, a MitM attacker cannot exploit this,
though.)

Thanks to Joeri de Ruiter for reporting this issue.

Reviewed-by: Matt Caswell <matt@openssl.org>

Always require an advertised NewSessionTicket message.

The server must send a NewSessionTicket message if it advertised one
in the ServerHello, so make a missing ticket message an alert
in the client.

An equivalent change was independently made in BoringSSL, see commit
6444287806d801b9a45baf1f6f02a0e3a16e144c.

Reviewed-by: Matt Caswell <matt@openssl.org>

Remove ssl3_check_finished.

The client sends a session ID with the session ticket, and uses
the returned ID to detect resumption, so we do not need to peek
at handshake messages: s->hit tells us explicitly if we're resuming.

An equivalent change was independently made in BoringSSL, see commit
407886f589cf2dbaed82db0a44173036c3bc3317.

Reviewed-by: Matt Caswell <matt@openssl.org>

Set s->hit when resuming from external pre-shared secret.

The same change was independently made in BoringSSL, see commit
9eaeef81fa2d4fd6246dc02b6203fa936a5eaf67

Reviewed-by: Matt Caswell <matt@openssl.org>

Reset s->tlsext_ticket_expected in ssl_scan_serverhello_tlsext.
This ensures that it's zeroed even if the SSL object is reused
(as in ssltest.c). It also ensures that it applies to DTLS, too.

Reviewed-by: Matt Caswell <matt@openssl.org>

Fix s_server -ssl2. Previously this reported "Error setting EC curve"

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Fix excert logic.

If no keyfile has been specified use the certificate file instead.

Fix typo: we need to check the chain is not NULL, not the chain file.
Reviewed-by: Matt Caswell <matt@openssl.org>
(cherry picked from commit 786370b1b09b919d9306f27336e13783e4fe3fd0)

New option no-ssl3-method which removes SSLv3_*method

When no-ssl3 is set only make SSLv3 disabled by default. Retain -ssl3
options for s_client/s_server/ssltest.

When no-ssl3-method is set SSLv3_*method() is removed and all -ssl3
options.

We should document this somewhere, e.g. wiki, FAQ or manual page.
Reviewed-by: Emilia Käsper <emilia@openssl.org>

Only handle RI extension for SSLv3

Don't send or parse any extensions other than RI (which is needed
to handle secure renegotation) for SSLv3.
Reviewed-by: Matt Caswell <matt@openssl.org>

Process signature algorithms before deciding on certificate.

The supported signature algorithms extension needs to be processed before
the certificate to use is decided and before a cipher is selected (as the
set of shared signature algorithms supported may impact the choice).
Reviewed-by: Matt Caswell <matt@openssl.org>
(cherry picked from commit 56e8dc542bd693b2dccea8828b3d8e5fc6932d0c)

Conflicts:
ssl/ssl.h
ssl/ssl_err.c

Added references to RFC 7027

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Prevent use of binary curves when OPENSSL_NO_EC2M is defined

Reviewed-by: Matt Caswell <matt@openssl.org>

Fixed cms-test.pl for no-ec2m

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Updated comment references to draft-ietf-tls-ecc-12 to refer to RFC4492 instead

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Don't use msg on error.

Don't attempt to access msg structure if recvmsg returns an error.

PR#3483
Reviewed-by: Stephen Henson <steve@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>

Add whrlpool and camellia .s files to perlasm list

Change-Id: I626d751f19f24df6b967c17498d6189cc0acb96c
Signed-off-by: Mike Bland <mbland@acm.org>
Signed-off-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

Add missing SRC variable

This is the only Makefile without SRC defined. This change enables a
standard Makefile include directive to cover crypto/jpake/*.d files.

This was automatically applied by AddSrcVarIfNeeded() in:
https://code.google.com/p/mike-bland/source/browse/openssl/update_makefiles.py

Change-Id: I030204a1bc873b5de5b06c8ddc0b94bb224c6650
Signed-off-by: Mike Bland <mbland@acm.org>
Signed-off-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

Remove redundant test targets outside of test/

These correspond to targets of the same name in test/Makefile that clash when
using the single-makefile build method using GitConfigure and GitMake.

Change-Id: If7e900c75f4341b446608b6916a3d76f202026ea
Signed-off-by: Mike Bland <mbland@acm.org>
Signed-off-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

Improve variable parsing when generating MINFO

Before this change, variables for which a '=' appeared in the assignment would
be parsed as the entire string up until the final '='. For example:

  BUILD_CMD=shlib_target=; if [ -n "$(SHARED_LIBS)" ]; then \

would result in the variable name "BUILD_CMD=shlib_target". This doesn't
appear to harm the current generation of MINFO, but creates problems for other
Makefile-related work I'm attempting.

Change-Id: I1f3a606d67fd5464bb459e8f36c23b3e967b77e1
Signed-off-by: Mike Bland <mbland@acm.org>
Signed-off-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

{,darwin64-}debug-test-64-clang Configure targets

These are based on debug-ben-debug-64-clang and is intended to produce
consistent settings for folks involved in the unit testing effort detailed at:

http://wiki.openssl.org/index.php/Unit_Testing

-fsanitize has been removed from the set of clang flags for now. Apparently
clang 3.1, which ships with FreeBSD 9.1, completely ignores -fsanitize. Clang
3.3, which ships with FreeBSD 9.2, compiles with it, but fails to link due to
the absence of libasan:

http://lists.freebsd.org/pipermail/freebsd-hackers/2013-December/043995.html
https://www.mail-archive.com/cfe-commits@cs.uiuc.edu/msg92260.html
http://reviews.llvm.org/D2644

We need -Wno-error=unused-const-variable because of this error:
.../crypto/ec/ec_lib.c:74:19: error: unused variable 'EC_version' [-Werror,-Wunused-const-variable]
static const char EC_version[] = "EC" OPENSSL_VERSION_PTEXT;

Change-Id: I2cba53537137186114c083049ea1233550a741f9
Signed-off-by: Mike Bland <mbland@acm.org>
Signed-off-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

Emit PERLASM_SCHEME to fix GitMake on OS X

This fixes the errors when trying to assemble .s files using GitMake on OS X.

Change-Id: I2221f558619302d22e0c57d7203173d634155678
Signed-off-by: Mike Bland <mbland@acm.org>
Signed-off-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

Add cscope.out and .d files to .gitignore

cscope.out is generated by cscope as described in:
http://wiki.openssl.org/index.php/Testing_and_Development_Tools_and_Tips

.d files are compiler-generated Makefile dependency files (e.g. using
'gcc -MMD -MP foo.c').

Change-Id: I2338858a6b6ee0527837d10a8b55cff1689023fd
Signed-off-by: Mike Bland <mbland@acm.org>
Signed-off-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

Additional output for ssltest.

Print out more details of the conection in ssltest specifically:
server certificate curve name for EC, server temporary key (if any)
and peer signing digest.
Reviewed-by: Matt Caswell <matt@openssl.org>

Add SSL_CONF support to ssltest.

Add command line support for SSL_CONF: server side arguments are
prefixed by -s_ (e.g. -s_no_ssl3) and client side with -c_.
Reviewed-by: Matt Caswell <matt@openssl.org>

Fix cross reference table generator.

If the hash or public key algorithm is "undef" the signature type
will receive special handling and shouldn't be included in the
cross reference table.
Reviewed-by: Tim Hudson <tjh@openssl.org>

Fixes a minor typo in the EVP docs.

Out is the buffer which needs to contain at least inl + cipher_block_size - 1 bytes. Outl
is just an int*.

Reviewed-by: Emilia Käsper <emilia@openssl.org>

Correct timestamp output when clock_precision_digits > 0

PR#3535

Reviewed-by: Stephen Henson <steve@openssl.org>

Fix free of garbage pointer. PR#3595

Reviewed-by: Emilia Käsper <emilia@openssl.org>

Fix warning about negative unsigned intergers

Reviewed-by: Richard Levitte <levitte@openssl.org>

Fix datarace reported by valgrind/helgrind

This doesn't really fix the datarace but changes it so it can only happens
once. This isn't really a problem since we always just set it to the same
value. We now just stop writing it after the first time.

PR3584, https://bugs.debian.org/534534

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Fix spelling of EECDH

Reviewed-by: Matt Caswell <matt@openssl.org>

armv4cpuid.S: fix compilation error in pre-ARMv7 build.

PR: 3474
Reviewed-by: Kurt Roeckx <kurt@openssl.org>

Fix WIN32 build by disabling bn* calls.

The trial division and probable prime with coprime tests are disabled
on WIN32 builds because they use internal functions not exported from
the WIN32 DLLs.
Reviewed-by: Emilia Käsper <emilia@openssl.org>

ec/asm/ecp_nistz256-x86_64.pl: fix inconsistency in path handling.

Reviewed-by: Tim Hudson <tjh@openssl.org>

md32_common.h: address compiler warning in HOST_c2l.

Reviewed-by: Stephen Henson <steve@openssl.org>

Use only unsigned arithmetic in constant-time operations

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Emilia Käsper <emilia@openssl.org>

Tighten session ticket handling

Tighten client-side session ticket handling during renegotiation:
ensure that the client only accepts a session ticket if the server sends
the extension anew in the ServerHello. Previously, a TLS client would
reuse the old extension state and thus accept a session ticket if one was
announced in the initial ServerHello.

Reviewed-by: Bodo Moeller <bodo@openssl.org>

Add missing CHANGES interval [1.0.1h, 1.0.1i]

Reviewed-by: Rich Salz <rsalz@openssl.org>

Sync CHANGES

Reviewed-by: Rich Salz <rsalz@openssl.org>

Fix ssltest logic when some protocols are compiled out.

Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Geoff Thorpe <geoff@openssl.org>

Copy negotiated parameters in SSL_set_SSL_CTX.

SSL_set_SSL_CTX is used to change the SSL_CTX for SNI, keep the
supported signature algorithms and raw cipherlist.
Reviewed-by: Tim Hudson <tjh@openssl.org>

Process signature algorithms in ClientHello late.

Reviewed-by: Tim Hudson <tjh@openssl.org>

ecp_nistz256 update.

Facilitate switch to custom scatter-gather routines. This modification
does not change algorithms, only makes it possible to implement
alternative. This is achieved by a) moving precompute table to assembly
(perlasm parses ecp_nistz256_table.c and is free to rearrange data to
match gathering algorithm); b) adhering to explicit scatter subroutine
(which for now is simply a memcpy). First implementations that will use
this option are 32-bit assembly implementations, ARMv4 and x86, where
equivalent of current read-whole-table-select-single-value algorithm
is too time-consuming. [On side note, switching to scatter-gather on
x86_64 would allow to improve server-side ECDSA performance by ~5%].

Reviewed-by: Bodo Moeller <bodo@openssl.org>

Configure: add ios64 target.

Reviewed-by: Steve Marquess <marquess@openssl.org>