Commit 9ef9088c15 switched the SSL/SSL_CTX
statistics counters to using Thread-Sanitizer-friendly primitives.
However, it erroneously converted an addition of -1
(for s->session_ctx->stats.sess_accept) to an addition of +1, since that
is the only counter API provided by the internal tsan_assist.h header
until the previous commit. This means that for each accepted (initial)
connection, the session_ctx's counter would get doubly incremented, and the
(switched) ctx's counter would also get incremented.
Restore the counter decrement so that each accepted connection increments
exactly one counter exactly once (in net effect).
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7464)
(cherry picked from commit 2aaa0b146b)
The existing tsan_counter() API increments a reference counter.
Provide a new API, tsan_decr(), to decrement such a reference counter.
This can be used, for example, when a reference is added to the session_ctx's
sess_accept stats but should more properly be tracked in the regular ctx's
statistics.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7464)
(cherry picked from commit 95658c3243)
If engine building fails for some reason, we must make sure to close
the /dev/crypto handle.
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/7506)
(cherry picked from commit 681e8cacdb)
We opened /dev/crypto once for each session, which is quite unnecessary.
With this change, we open /dev/crypto once at engine init, and close
it on unload.
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/7506)
(cherry picked from commit 458c7dad9e)
Copying an EVP_MD_CTX, including the implementation local bits, is a
necessary operation. In this case, though, it's the same as
initializing the local bits to be "copied to".
Fixes#7495
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/7506)
(cherry picked from commit 36af124bfb)
The documentation says some commands care, but the code says differently.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7440)
(cherry picked from commit d91d443f0d)
This makes sure that any resulting directory target in the build files
also depend on object files meant for shared libraries.
As a side effect, we move the production of the dirinfo structure from
common.tmpl to Configure, to make it easier to check the result.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7452)
(cherry picked from commit b6e660754c)
This excluses user from additional PATH adjustments in case NDK has
llvm-ar.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7443)
(cherry picked from commit ddf2e8c8cc)
03ad7c009e failed if one didn't pass
explicit -D__ANDROID_API__=N :-(
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7443)
(cherry picked from commit 71f2b3171e)
In TLSv1.2 and below a CertificateRequest is sent after the Certificate
from the server. This means that by the time the client_cert_cb is called
on receipt of the CertificateRequest a call to SSL_get_peer_certificate()
will return the server certificate as expected. In TLSv1.3 a
CertificateRequest is sent before a Certificate message so calling
SSL_get_peer_certificate() returns NULL.
To workaround this we delay calling the client_cert_cb until after we
have processed the CertificateVerify message, when we are doing TLSv1.3.
Fixes#7384
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/7413)
(cherry picked from commit e45620140f)
The RAND_DRBG_TYPE preprocessor define depends on a NID, so we have
to include obj_mac.h to make the header selfcontained.
Fixes#7521
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7524)
(cherry picked from commit cf3d6ef7af)
pkey_test_ctrl() was designed for parsing values, not for using in
test runs. Relying on its returned value when it returned 1 even for
control errors made it particularly useless for mac_test_run().
Here, it gets replaced with a MAC specific control function, that
parses values the same way but is designed for use in a _run() rather
than a _parse() function.
This uncovers a SipHash test with an invalid control that wasn't
caught properly. After all, that stanza is supposed to test that
invalid control values do generate an error. Now we catch that.
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7500)
(cherry picked from commit ce5d64c79c)
We made the build of foo.obj depend on foo.d, meaning the latter gets
built first. Unfortunately, the way the compiler works, we are forced
to redirect all output to foo.d, meaning that if the source contains
an error, the build fails without showing those errors.
We therefore remove the dependency and force the build of foo.d to
always happen after build of foo.obj.
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7469)
(cherry picked from commit ecc347f5f5)
Needed to clear a clash between ERR_load_CRYPTO_strings and
ERR_load_crypto_strings
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7194)
(cherry picked from commit cc3a2e4f51)
Preallocate an extra limb for some of the big numbers to avoid a reallocation
that can potentially provide a side channel.
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/7486)
(cherry picked from commit 99540ec794)
Avoid a timing attack that leaks information via a side channel that
triggers when a BN is resized. Increasing the size of the BNs
prior to doing anything with them suppresses the attack.
Thanks due to Samuel Weiser for finding and locating this.
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/7486)
(cherry picked from commit a9cfb8c2aa)
There is a side channel attack against the division used to calculate one of
the modulo inverses in the DSA algorithm. This change takes advantage of the
primality of the modulo and Fermat's little theorem to calculate the inverse
without leaking information.
Thanks to Samuel Weiser for finding and reporting this.
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/7487)
(cherry picked from commit 415c335635)
Found by Coverity Scan
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7511)
(cherry picked from commit 59f90557dd)
This also gave enough reason to collect the stuff that's common for
all iOS config targets into the template "ios-common".
Fixes#7318
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7403)
(cherry picked from commit 04e3f9a114)
The new Configure summary box (41349b5e6d) now hides the warning
about the missing seed source (2805ee1e09) too much. To make it
more visible again, add warning markers.
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7509)
(cherry picked from commit caa8595276)
Commit 5b4cb385c1 (#7382) introduced a bug which had the effect
that RAND_add()/RAND_seed() failed for buffer sizes less than
32 bytes. The reason was that now the added random data was used
exlusively as entropy source for reseeding. When the random input
was too short or contained not enough entropy, the DRBG failed
without querying the available entropy sources.
This commit makes drbg_add() act smarter: it checks the entropy
requirements explicitely. If the random input fails this check,
it won't be added as entropy input, but only as additional data.
More precisely, the behaviour depends on whether an os entropy
source was configured (which is the default on most os):
- If an os entropy source is avaible then we declare the buffer
content as additional data by setting randomness to zero and
trigger a regular reseeding.
- If no os entropy source is available, a reseeding will fail
inevitably. So drbg_add() uses a trick to mix the buffer contents
into the DRBG state without forcing a reseeding: it generates a
dummy random byte, using the buffer content as additional data.
Related-to: #7449
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7456)
(cherry picked from commit 8817215d5c)
In commit 8bf3665196 some renamings andd typo fixes were made
while adding back the DRBG-HMAC and DRBG-HASH implementation.
Since the commit could not be backported, a lot of unnecessary
differences between master and 1.1.1 were introduced.
These differences result in tiresome merge conflicts when
cherry-picking. To minimize these merge-conflicts, this patch
ports all 'non-feature' changes of commit 8bf3665196
(e.g., renamings of private variables, fixes of typographical
errors, comment changes) manually back to 1.1.1.
The commits a83dc59afa (#7399) and 8817215d5c (#7456)
failed to cherry-pick previously to 1.1.1, with this patch
they both cherry-pick without conflicts.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/7505)
Since 1fb9fdc30 we may attempt to buffer a record from the next epoch
that has already been buffered. Prior to that this never occurred.
We simply ignore a failure to buffer a duplicated record.
Fixes#6902
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7414)
(cherry picked from commit 840facc3cc)
Reviewed-by: Paul Yang <yang.yang@baishancloud.com>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7474)
(cherry picked from commit 21311777ad)
In commit 820e414d28 (pr #5247) the summary output of the
Configure command was optimized towards instructing people
how to create issue reports.
It turned out that the wording of this message can confuse new
OpenSSL users and make them think that they are seeing an error
message. This commit makes the summary output start with a success
to prevent a misunderstanding. Also it gives more hints to new
OpenSSL users.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7499)
(cherry picked from commit 41349b5e6d)
Increase the load buffer size such that it exceeds the chunk
size by a comfortable amount. This is done to avoid calling
RAND_add() with a small final chunk. Instead, such a small
final chunk will be added together with the previous chunk
(unless it's the only one).
Related-to: #7449
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7456)
The failure of RAND_load_file was only noticed because of the
heap corruption which was reported in #7499 and fixed in commit
5b4cb385c1. To prevent this in the future, RAND_load_file()
now explicitly checks RAND_status() and reports an error if it
fails.
Related-to: #7449
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7456)
When building shared libraries on Windows, we had a clash between
'libcrypto.lib' the static routine library and 'libcrypto.lib' the
import library.
We now change it so the static versions of our libraries get '_static'
appended to their names. These will never get installed, but can
still be used for our internal purposes, such as internal tests.
When building non-shared, the renaming mechanism doesn't come into
play. In that case, the static libraries 'libcrypto.lib' and
'libssl.lib' are installed, just as always.
Fixes#7492
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7496)
(cherry picked from commit b3023ced6b)
This bug was introduced by #7382 which enhanced RAND_add() to
accept large buffer sizes. As a consequence, RAND_add() now fails
for buffer sizes less than 32 bytes (i.e. less than 256 bits).
In addition, rand_drbg_get_entropy() forgets to reset the attached
drbg->pool in the case of an error, which leads to the heap corruption.
The problem occurred with RAND_load_file(), which reads the file in
chunks of 1024 bytes each. If the size of the final chunk is less than
32 bytes, then RAND_add() fails, whence RAND_load_file() fails
silently for buffer sizes n = k * 1024 + r with r = 1,...,31.
This commit fixes the heap corruption only. The other issues will
be addressed in a separate pull request.
Thanks to Gisle Vanem for reporting this issue.
Fixes#7449
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7455)
(cherry picked from commit 5b4cb385c1)
We only had targets for the "simple" shared library names (libfoo.so
and not libfoo.so.x.y on Unix, import library libfoo.lib but not
libfoo.dll on Windows). This has created some confusion why it wasn't
possible to rebuild the less "simple" name directly (just as an
example, someone who mistook the import library on Windows for a
static library, removed the DLL and then found it was difficult to
rebuild directly), so we change the target to include all possible
names.
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7451)
(cherry picked from commit d8cac50b02)
An unknown PSK identity could be because its actually a session resumption
attempt. Sessions resumptions and external PSKs are indistinguishable so
the callbacks need to fail gracefully if they don't recognise the identity.
Fixes#7433
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7434)
(cherry picked from commit 2d015189b9)
Previously when a ClientHello arrives with a valid cookie using
DTLSv1_listen() we only "peeked" at the message and left it on the
underlying fd. This works fine for single threaded applications but for
multi-threaded apps this does not work since the fd is typically reused for
the server thread, while a new fd is created and connected for the client.
By "peeking" we leave the message on the server fd, and consequently we
think we've received another valid ClientHello and so we create yet another
fd for the client, and so on until we run out of fds.
In this new approach we remove the ClientHello and buffer it in the SSL
object.
Fixes#6934
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/7375)
(cherry picked from commit 079ef6bd53)
Rather than using init_buf we use the record layer read and write buffers
in DTLSv1_listen(). These seem more appropriate anyway and will help with
the next commit.
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/7375)
(cherry picked from commit 2fc4c77c3f)