Credit to OSS-Fuzz for finding this.
CVE-2017-3738
(Imported from upstream's 5630661aecbea5fe3c4740f5fea744a1f07a6253 and
77d75993651b63e872244a3256e37967bb3c3e9e.)
Confirmed with Intel SDE that the fix makes the test vector pass and
that, without the fix, the test vector does not. (Well, we knew the
latter already, since it was our test vector.)
(cherry-picked from 296a61d600)
Change-Id: I167aa3407ddab3b434bacbd18e099c55aa40ac4c
Reviewed-on: https://boringssl-review.googlesource.com/23884
Reviewed-by: Adam Langley <agl@google.com>
Reviewed-on: https://boringssl-review.googlesource.com/23924
Due to a copy-paste error, the call to |left_shift_3| is missing after
reducing the password scalar in SPAKE2. This means that three bits of
the password leak in Alice's message. (Two in Bob's message as the point
N happens to have order 4l, not 8l.)
The “correct” fix is to put in the missing call to |left_shift_3|, but
that would be a breaking change. In order to fix this in a unilateral
way, we add points of small order to the masking point to bring it into
prime-order subgroup.
BUG=chromium:778101
Change-Id: I440931a3df7f009b324d2a3e3af2d893a101804f
Reviewed-on: https://boringssl-review.googlesource.com/22445
Reviewed-by: Adam Langley <agl@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: Adam Langley <agl@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
The only difference is whether there's an alert to send back, but we'll
need to allow an "error without alert" in several cases anyway:
1. If the server sees an HTTP request or garbage instead of a
ClientHello, it shouldn't send an alert.
2. Resurfaced errors.
Just make zero signal no alert for now. Later on, I'm thinking we might
just want to put the alert into the outgoing buffer and make it further
uniform.
This also gives us only one error state to keep track of rather than
two.
Bug: 206
Change-Id: Ia821d9f89abd2ca6010e8851220d4e070bc42fa1
Reviewed-on: https://boringssl-review.googlesource.com/21286
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Reviewed-by: Steven Valdez <svaldez@google.com>
This is analogous to the Go stack's handshakeErr field. Since it's quite
common for callers to run two I/O operations in parallel[*] like
SSL_read and SSL_write (or SSL_read and SSL_do_handshake for client
0-RTT). Accordingly, the new handshake state machine jams itself up on
handshake error, but to fully work with such callers, we should also
replay the error state.
This doesn't yet catch all cases (there are some parts of the read flow
which need to be fixed). Those will be resolved in later changes.
[*] Not actually in parallel, of course, but logically in parallel on a
non-blocking socket.
Bug: 206
Change-Id: I5a4d37a258b9e3fc555b732938b0528b839650f8
Reviewed-on: https://boringssl-review.googlesource.com/21285
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Reviewed-by: Steven Valdez <svaldez@google.com>
This will be useful for the SSL stack to properly resurface handshake
failures. Leave this in a private header and, along the way, hide the
various types.
(ERR_NUM_ERRORS didn't change in meaning. The old documentation was
wrong.)
Bug: 206
Change-Id: I4c6ca98d162d11ad5e17e4baf439a18fbe371018
Reviewed-on: https://boringssl-review.googlesource.com/21284
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Reviewed-by: Steven Valdez <svaldez@google.com>
crypto/bio/bio_test.cc - I'm not sure where this was added for, but none
of the functions used there appear to have feature macros documented.
crypto/bio/printf.c - -std=c99 provides (v)snprintf.
crypto/lhash/lhash_test.cc - we no longer call rand_r.
crypto/mem.c - we no longer call strdup and -std=c99 provides (v)snprintf.
Apple messed up their headers and, if _POSIX_C_SOURCE is defined but
_DARWIN_C_SOURCE isn't, pthread.h no longer defines mach_port_t. They
then shipped a version of libc++ headers that is missing this fix, so
the build breaks:
https://github.com/llvm-mirror/libcxx/commit/bcc92d75df0274b9593ebd097fcae60494e3bffc
If one uses XCode, they've hacked their pthread.h to provide mach_port_t
if defined(__cplusplus), but the standalone tools appear to be old and
missing this.
We can work around this by also defining _DARWIN_C_SOURCE in C++ files
that need _POSIX_C_SOURCE, but it appears none of these files actually
need it.
Change-Id: I5df9453730696100eb22b809febeb65053701322
Reviewed-on: https://boringssl-review.googlesource.com/20964
Reviewed-by: Adam Langley <agl@google.com>
The exponent is secret, so we should be using the consttime variant. See
also upstream's f9cbf470180841966338db1f4c28d99ec4debec4.
Change-Id: I233d4223ded5b80711d7c8f906e3579c36b24cd0
Reviewed-on: https://boringssl-review.googlesource.com/20924
Reviewed-by: Adam Langley <agl@google.com>
Although we are derived from 1.0.2, we mimic 1.1.0 in some ways around
our FOO_up_ref functions and opaque libssl types. This causes some
difficulties when porting third-party code as any OPENSSL_VERSION_NUMBER
checks for 1.1.0 APIs we have will be wrong.
Moreover, adding accessors without changing OPENSSL_VERSION_NUMBER can
break external projects. It is common to implement a compatibility
version of an accessor under #ifdef as a static function. This then
conflicts with our headers if we, unlike OpenSSL 1.0.2, have this
function.
This change switches OPENSSL_VERSION_NUMBER to 1.1.0 and atomically adds
enough accessors for software with 1.1.0 support already. The hope is
this will unblock hiding SSL_CTX and SSL_SESSION, which will be
especially useful with C++-ficiation. The cost is we will hit some
growing pains as more 1.1.0 consumers enter the ecosystem and we
converge on the right set of APIs to import from upstream.
It does not remove any 1.0.2 APIs, so we will not require that all
projects support 1.1.0. The exception is APIs which changed in 1.1.0 but
did not change the function signature. Those are breaking changes.
Specifically:
- SSL_CTX_sess_set_get_cb is now const-correct.
- X509_get0_signature is now const-correct.
For C++ consumers only, this change temporarily includes an overload
hack for SSL_CTX_sess_set_get_cb that keeps the old callback working.
This is a workaround for Node not yet supporting OpenSSL 1.1.0.
The version number is set at (the as yet unreleased) 1.1.0g to denote
that this change includes https://github.com/openssl/openssl/pull/4384.
Bug: 91
Change-Id: I5eeb27448a6db4c25c244afac37f9604d9608a76
Reviewed-on: https://boringssl-review.googlesource.com/10340
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
SSL_CTX_sessions is the only think making us expose LHASH as public API
and nothing uses it. Nothing can use it anyway as it's not thread-safe.
I haven't actually removed it yet since SSL_CTX is public, but once the
types are opaque, we could trim the number of symbols ssl.h pulls in
with some work.
Relatedly, fix thread safety of SSL_CTX_sess_number.
Change-Id: I75a6c93509d462cd5ed3ce76c587f0d1e7cd0797
Reviewed-on: https://boringssl-review.googlesource.com/20804
Commit-Queue: David Benjamin <davidben@google.com>
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
David Benjamin