Nicola Tuveri [Fri, 6 Aug 2021 15:37:02 +0000 (18:37 +0300)]
[github-ci] Import run-checker daily workflow from master
The daily run-checker is scheduled to start at 6:42, instead of the
start of the hour.
The official GitHub documentation remarks the following regarding
scheduled workflows:
> Note: The schedule event can be delayed during periods of high loads
> of GitHub Actions workflow runs. High load times include the start of
> every hour. To decrease the chance of delay, schedule your workflow to
> run at a different time of the hour.
42, obviously, has been picked because it is the answer to the ultimate
question of life, the universe, and everything.
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16252)
Ingo Franzki [Wed, 11 Aug 2021 10:53:09 +0000 (12:53 +0200)]
Test EVP Cipher updating the context's IV
Ensure that an EVP_CipherUpdate operation updates the context's
IV for AES CBC, CFB, OFB, and CTR. An application can get the
updated IV via EVP_CIPHER_CTX_iv().
The s390x implementation of the CFB and OFB ciphers did not
update the IV in the context, but only within its s390x specific
context data.
Signed-off-by: Ingo Franzki <ifranzki@linux.ibm.com> Reviewed-by: Patrick Steuer <patrick.steuer@de.ibm.com> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16292)
Ingo Franzki [Wed, 11 Aug 2021 07:39:46 +0000 (09:39 +0200)]
s390x: AES OFB/CFB: Maintain running IV from cipher context
Copy the current IV from the cipher context into the kmo/kmf param before
the operation, and copy the modified IV back to the context afterwards.
Without this, an application that obtains the running IV from the context
would still get the original IV, but not the updated one.
Signed-off-by: Ingo Franzki <ifranzki@linux.ibm.com> Reviewed-by: Patrick Steuer <patrick.steuer@de.ibm.com> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16292)
Todd Short [Fri, 13 Aug 2021 13:59:59 +0000 (09:59 -0400)]
Fix potential double-free
The `sk` variable is assigned to `s->session->peer_chain`.
If `ssl3_digest_cached_records()` were to fail, then `sk` would still be
non-NULL, and subsequently freed on the error return. When the session
is freed, it will then attempt to free `s->session->peer_chain`,
resulting in a double-free (of `sk`).
Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16309)
The ASN1_STRING_get0_data(3) manual explitely cautions the reader
that the data is not necessarily NUL-terminated, and the function
X509_alias_set1(3) does not sanitize the data passed into it in any
way either, so we must assume the return value from X509_alias_get0(3)
is merely a byte array and not necessarily a string in the sense
of the C language.
I found this bug while writing manual pages for X509_print_ex(3)
and related functions. Theo Buehler <tb@openbsd.org> checked my
patch to fix the same bug in LibreSSL, see
As an aside, note that the function still produces incomplete and
misleading results when the data contains a NUL byte in the middle
and that error handling is consistently absent throughout, even
though the function provides an "int" return value obviously intended
to be 1 for success and 0 for failure, and even though this function
is called by another function that also wants to return 1 for success
and 0 for failure and even does so in many of its code paths, though
not in others. But let's stay focussed. Many things would be nice
to have in the wide wild world, but a buffer overflow must not be
allowed to remain in our backyard.
CLA: trivial
Reviewed-by: Tim Hudson <tjh@openssl.org> Reviewed-by: Paul Dale <pauli@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16108)
Matt Caswell [Thu, 15 Jul 2021 13:08:56 +0000 (14:08 +0100)]
Fix some minor record layer issues
Various comments referred to s->packet and s->packet_length instead of
s->rlayer.packet and s->rlayer.packet_length. Also fixed is a spot where
RECORD_LAYER_write_pending() should have been used. Based on the review
comments in #16077.
Matt Caswell [Tue, 13 Jul 2021 16:44:44 +0000 (17:44 +0100)]
Disallow SSL_key_update() if there are writes pending
If an application is halfway through writing application data it should
not be allowed to attempt an SSL_key_update() operation. Instead the
SSL_write() operation should be completed.
Fixes #12485
Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16098)
Matt Caswell [Tue, 13 Jul 2021 16:19:12 +0000 (17:19 +0100)]
Don't reset the packet pointer in ssl3_setup_read_buffer
Sometimes this function gets called when the buffers have already been
set up. If there is already a partial packet in the read buffer then the
packet pointer will be set to an incorrect value. The packet pointer already
gets reset to the correct value when we first read a packet anyway, so we
don't also need to do it in ssl3_setup_read_buffer.
Fixes #13729
Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16098)
apps: Use the first detected address family if IPv6 is not available
This is a follow up of 15729bef385211bc2a0497e2d53a45c45d677d2c. Even
when the host does not support IPv6 at all, BIO_lookup_ex may now
return IN6ADDR_ANY in addition to INADDR_ANY, as the second element of
the ai_next field.
After eee8a40aa5e06841eed6fa8eb4f6109238d59aea, the do_server function
prefers the IPv6 address and fails on the BIO_socket call. This adds
a fallback code to retry with the IPv4 address returned as the first
element to avoid the error.
The failure had been partially avoided in the previous code with
AI_ADDRCONFIG, because getaddrinfo returns only IPv4 address if no
IPv6 address is associated with external interface. However, it would
be still a problem if the external interface has an IPv6 address
assigned, while the loopback interface doesn't.
Signed-off-by: Daiki Ueno <dueno@redhat.com> Reviewed-by: Paul Dale <pauli@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16078)
BIO_lookup_ex: use AI_ADDRCONFIG only if explicit host name is given
The flag only affects which record types are queried via DNS (A or
AAAA, or both). When node is NULL and AF_UNSPEC is used, it prevents
getaddrinfo returning the right address associated with the loopback
interface.
Signed-off-by: Daiki Ueno <dueno@redhat.com> Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com> Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16039)
Matt Caswell [Wed, 23 Jun 2021 07:54:12 +0000 (08:54 +0100)]
Avoid "excessive message size" for session tickets
We received a report of an "excessive message size" for a received
session ticket. Our maximum size was significantly less than the theoretical
maximum. The server may put any data it likes in the session ticket
including (for example) the full certificate chain so we should be able to
handle longer tickets. Update the value to the maximum allowed by the spec.
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15877)
Matt Caswell [Tue, 6 Jul 2021 15:24:07 +0000 (16:24 +0100)]
Fix s_server PSK handling
Issue #15951 describes a scenario which causes s_server to fail when using
a PSK. In the originally described issue this only impacted master and not
1.1.1. However, in fact this issue does also impact 1.1.1 - but only if you
additionally supply the option "-no_ticket" to the s_server command line.
The difference between the behaviour in master and 1.1.1 is due to 9c13b49,
which changed PSK_MAX_IDENTITY_LEN from 128 to 256. It just so happens that
a default OpenSSL TLSv1.3 ticket length happens to fall between those 2
values. Tickets are presented in TLSv1.3 as a PSK "identity". Passing
"no_ticket" doesn't actually stop TLSv1.3 tickets completely, it just
forces the use of "session ids as a ticket" instead. This significantly
reduces the ticket size to below 128 in 1.1.1.
The problem was due to s_server setting a TLSv1.2 PSK callback and a
TLSv1.3 PSK callback. For backwards compat reasons the TLSv1.2 PSK
callbacks also work in TLSv1.3 but are not preferred. In the described
scenario we use a PSK to create the initial connection. Subsequent to that
we attempt a resumption using a TLSv1.3 ticket (psk). If the psk length is
below PSK_MAX_IDENTITY_LEN then we first call the TLSv1.2 PSK callback.
Subsequently we call the TLSv1.3 PSK callback. Unfortunately s_server's
TLSv1.2 PSK callback accepts the identity regardless, even though it is an
unexpected value, and hence the binder subsequently fails to verify.
The fix is to bail early in the TLSv1.2 callback if we detect we are being
called from a TLSv1.3 connection.
Fixes #15951
Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/16008)
Oliver Mihatsch [Mon, 5 Jul 2021 14:23:03 +0000 (16:23 +0200)]
Fix memory leak in i2d_ASN1_bio_stream
When creating a signed S/MIME message using SMIME_write_CMS()
if the reading from the bio fails, the state is therefore
still ASN1_STATE_START when BIO_flush() is called by i2d_ASN1_bio_stream().
This results in calling asn1_bio_flush_ex cleanup but will only
reset retry flags as the state is not ASN1_STATE_POST_COPY.
Therefore 48 bytes (Linux x86_64) leaked since the
ndef_prefix_free / ndef_suffix_free callbacks are not executed
and the ndef_aux structure is not freed.
By always calling free function callback in asn1_bio_free() the
memory leak is fixed.
Pauli [Mon, 5 Jul 2021 01:01:59 +0000 (11:01 +1000)]
bn: procduce correct sign for result of BN_mod()
There is a problem that appears when calling BN_div(a, c, a, b) with negative b.
In this case, the sign of the remainder c is incorrect. The problem only
occurs if the dividend and the quotient are the same BIGNUM.
Fixes #15982
Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/15991)
Tomas Mraz [Wed, 30 Jun 2021 09:17:09 +0000 (11:17 +0200)]
doc: Mention the update of der data pointers in d2i/i2d
Fixes #15958
Reviewed-by: Richard Levitte <levitte@openssl.org> Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15973)
David Benjamin [Tue, 29 Jun 2021 18:41:12 +0000 (14:41 -0400)]
Fix use of uninitialized memory in test_rsa_oaep
48f1739600f33c92387debce2002acec6e365f1d did not convert the RSA OAEP
tests correctly. The corrupted ciphertext and truncation tests were
really decrypting uninitialized memory, rather than the sample
ciphertext. This results in an error in tools like MSan.
The test is somewhat roundabout. In the original version, before the
conversion, ctext_ex was an OAEP test vector from key1(), etc.,
functions. The test would:
1. Encrypt ptext_ex as ctext.
2. Decrypt ctext and check it gives ptext_ex.
3. Decrypt ctext_ex and check it gives ptext_ex.
4. Try corrupted and truncated versions of ctext.
48f1739600f33c92387debce2002acec6e365f1d then moved steps 1 and 2 into
test_rsa_simple, which meant ctext is no longer available for step 4. It
then mistakenly left the variable around, but uninitialized, so the test
wasn't testing anything. (Confusingly, test_rsa_simple outputs ctext_ex
to the caller, but doesn't do anything with it. The ctext_ex output is
also only usable for OAEP, not PKCS#1 v1.5.)
It doesn't really matter whether we use ctext or ctext_ex for step 4, so
this PR fixes it by using ctext_ex instead.
Reviewed-by: Shane Lontis <shane.lontis@oracle.com> Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15950)
Lars Immisch [Thu, 5 Mar 2020 10:26:06 +0000 (11:26 +0100)]
Use getauxval on Android with API level > 18
We received analytics that devices of the device family Oppo A37x
are crashing with SIGILL when trying to load libcrypto.so.
These crashes were fixed by using the system-supplied getauxval function.
Reviewed-by: Paul Dale <pauli@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15763)
Patrick Steuer [Tue, 8 Jun 2021 08:22:53 +0000 (10:22 +0200)]
Test EVP_CipherInit sequences and resets
Various EVP_CipherInit sequences including partial inits and initializations
with different "enc" flags caused problems on s390x. Similarly, cipher
reinitialization and especially GCM reinitialization with different tag length
led to wrong results. Add some unit tests to cover these rather exotic use
cases.
Signed-off-by: Patrick Steuer <patrick.steuer@de.ibm.com> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14900)
Patrick Steuer [Fri, 16 Apr 2021 15:09:46 +0000 (15:09 +0000)]
s390x: cipher must set EVP_CIPH_ALWAYS_CALL_INIT flag
The s390x cipher implementations must call their init function
even if the key argument is NULL to allow initializing the
cipher operation's context in any order.
Signed-off-by: Patrick Steuer <patrick.steuer@de.ibm.com> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14900)
Matt Caswell [Wed, 2 Jun 2021 16:19:23 +0000 (17:19 +0100)]
Only call dtls1_start_timer() once
The function dtls1_handle_timeout() calls dtls1_double_timeout() which
was calling dtls1_start_timer(). However dtls1_start_timer() is also
called directly by dtls1_handle_timeout(). We only need to start the timer
once.
Fixes #15561
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15595)
bonniegong [Wed, 2 Jun 2021 07:35:18 +0000 (15:35 +0800)]
Check the return value of ASN1_STRING_length
ASN1_STRING_length gets the field 'length' of msg, which
can be manipulated through a crafted input.
Add a check to avoid error execution of OPENSSL_malloc().
CLA: trivial
Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Paul Dale <pauli@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15583)
Trev Larock [Fri, 28 May 2021 02:26:41 +0000 (02:26 +0000)]
Modify ssl_handshake_hash to call SSLfatal
When EVP_MD_CTX_new fails call SSLfatal before the goto err.
This resolves a state machine issue on the out of memory condition:
ssl/statem/statem.c:643: OpenSSL internal error: Assertion failed:
(s)->statem.in_init && (s)->statem.state == MSG_FLOW_ERROR
Fixes #15491.
CLA: trivial
Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Paul Dale <pauli@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15511)
Christian Heimes [Tue, 30 Mar 2021 10:02:42 +0000 (12:02 +0200)]
Inherit hostflags verify params even without hosts
X509_VERIFY_PARAM_inherit() now copies hostflags independently of hosts.
Previously hostflags were only copied when at least one host was set.
Typically applications don't configure hosts on SSL_CTX. The change
enables applications to configure hostflags on SSL_CTX and have OpenSSL
copy the flags from SSL_CTX to SSL.
Fixes: https://github.com/openssl/openssl/issues/14579 Signed-off-by: Christian Heimes <christian@python.org> Reviewed-by: Paul Dale <pauli@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14856)
Theo Buehler [Sat, 1 May 2021 11:09:10 +0000 (13:09 +0200)]
Test oct2point for hybrid point encoding of (0, y)
Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com> Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15112)
Theo Buehler [Sat, 1 May 2021 10:25:50 +0000 (12:25 +0200)]
Avoid division by zero in hybrid point encoding
In hybrid and compressed point encodings, the form octet contains a bit
of information allowing to calculate y from x. For a point on a binary
curve, this bit is zero if x is zero, otherwise it must match the
rightmost bit of of the field element y / x. The existing code only
considers the second possibility. It could thus incorrecly fail with a
division by zero error as found by Guido Vranken's cryptofuzz.
This commit adds a few explanatory comments to oct2point. The only
actual code change is in the last hunk which adds a BN_is_zero(x)
check to avoid the division by zero.
Fixes #15021
Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com> Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15112)
Benjamin Kaduk [Tue, 30 Mar 2021 06:05:22 +0000 (23:05 -0700)]
Update expected results for tls13kexmodes tests
One of the scenarios constructed in these tests was erroneously
producing successful handshakes until the previous commits, but should
have been failing. Update our expected behavior to match the
specification requirements, and adjust the commentary slightly for
a test case relevant for the other preceding commit.
Benjamin Kaduk [Tue, 30 Mar 2021 04:27:49 +0000 (21:27 -0700)]
Don't send key_share for PSK-only key exchange
TLS 1.3 allows for the "psk_ke" and "psk_dhe_ke" key-exchange modes.
Only the latter mode introduces a new ephemeral (Diffie-Hellman)
key exchange, with the PSK being the only key material used in the
former case.
It's a compliance requirement of RFC 8446 that the server MUST NOT
send a KeyShareEntry when using the "psk_ke" mode, but prior to
this commit we would send a key-share based solely on whether the
client sent one. This bug goes unnoticed in our internal test suite
since openssl communicating with openssl can never negotiate the
PSK-only key-exchange mode. However, we should still be compliant
with the spec, so check whether the DHE mode was offered and don't
send a key-share if it wasn't.
Benjamin Kaduk [Tue, 30 Mar 2021 06:03:49 +0000 (23:03 -0700)]
Improve RFC 8446 PSK key exchange mode compliance
It's a MUST-level requirement that if the client sends a pre_shared_key
extension not accompanied by a psk_key_exchange_modes extension, the
server must abort the handshake. Prior to this commit the server
would continue on.
Fred Hornsey [Wed, 18 Nov 2020 04:20:43 +0000 (22:20 -0600)]
Support for Android NDK r22
This is a backport of #13434, Fixes #13685.
I think builds using standalone toolchain are fine so I left them alone,
but `Configure` will fail if using the NDK directly because the
`platforms` and `sysroot` directories were removed.
If `sysroot` is missing, omit the `--sysroot` and `-gcc-toolchain`
arguments and use the triplet form clang command.
Also since `platforms` was being used for the default API level, use
`meta/platforms.json` instead if needed.
Reviewed-by: Tim Hudson <tjh@openssl.org> Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/13694)
Richard Levitte [Tue, 20 Apr 2021 06:43:30 +0000 (08:43 +0200)]
ASN1: Ensure that d2i_ASN1_OBJECT() frees the strings on ASN1_OBJECT reuse
The 'sn' and 'ln' strings may be dynamically allocated, and the
ASN1_OBJECT flags have a bit set to say this. If an ASN1_OBJECT with
such strings is passed to d2i_ASN1_OBJECT() for reuse, the strings
must be freed, or there is a memory leak.
Fixes #14667
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/14938)
Richard Levitte [Thu, 22 Apr 2021 12:37:40 +0000 (14:37 +0200)]
Don't remove $(TARFILE) when cleaning
This file is outside the source tree, so we have no business removing
it. This is especially concerning if that was the tarball the user
had to create the source tree.
Fixes #14981
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14985)
Some compilers(g++ on Solaris/Illumos) define __STDC__VERSION__ in c++ .
This causes c++ code that uses openssl to break on these compilers since
_Noreturn is not a keyword in c++ .
CLA: trivial
Reviewed-by: Richard Levitte <levitte@openssl.org> Reviewed-by: Paul Dale <pauli@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14944)
Pauli [Sun, 18 Apr 2021 22:57:18 +0000 (08:57 +1000)]
engine: fix double free on error path.
In function try_decode_PKCS8Encrypted, p8 is freed via X509_SIG_free() at line 481.
If function new_EMBEDDED() returns a null pointer at line 483, the execution will goto nop8.
In the nop8 branch, p8 is freed again at line 491.
Bug reported by @Yunlongs
Fixes #14915
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14921)
Pauli [Sun, 18 Apr 2021 22:55:37 +0000 (08:55 +1000)]
ts: fix double free on error path.
In function int_ts_RESP_verify_token, if (flags & TS_VFY_DATA) is true, function ts_compute_imprint() will be called at line 299.
In the implementation of ts_compute_imprint, it allocates md_alg at line 406.
But after the allocation, if the execution goto err, then md_alg will be freed in the first time by X509_ALGOR_free at line 439.
After that, ts_compute_imprint returns 0 and the execution goto err branch of int_ts_RESP_verify_token.
In the err branch, md_alg will be freed in the second time at line 320.
Bug reported by @Yunlongs
Fixes #14914
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14921)
Pauli [Sun, 18 Apr 2021 22:51:38 +0000 (08:51 +1000)]
srp: fix double free,
In function SRP_create_verifier_ex, it calls SRP_create_verifier_BN_ex(..., &v, ..) at line 653.
In the implementation of SRP_create_verifier_BN_ex(), *verify (which is the paremeter of v) is allocated a pointer via BN_new() at line 738.
And *verify is freed via BN_clear_free() at line 743, and return 0.
Then the execution continues up to goto err at line 655, and the freed v is freed again at line 687.
Bug reported by @Yunlongs
Fixes #14913
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14921)
Dave Coombs [Tue, 6 Apr 2021 16:49:21 +0000 (12:49 -0400)]
crl2pkcs7 shouldn't include empty optional sets
If using crl2pkcs7 -nocrl and with no -certfiles, we shouldn't include
the implicitly tagged [0] certs and [1] crls sets as they are marked
optional and would be empty.
Reviewed-by: Paul Dale <pauli@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14781)
If utils/mkdir-p.pl fails to create a target dir because of insufficient file system
permissions, the subsequent test for dir existence always fails and overwrites
the system error. As a result, a user is presented with a misleading error message.
E.g. if a user tries to create a dir under /usr/local and does not have permissions
for it, the reported error message is "Cannot create directory /usr/local/lib: No such file or directory",
whereas the expected error message is "Cannot create directory /usr/local/lib: Permission denied".
This commit introduces a fix by declaring an additional local variable to cache
the original error message from mkdir. If -d check fails and overwrites the system
error, the user is still presented with the original error from mkdir.
CLA: Trivial
Reviewed-by: Shane Lontis <shane.lontis@oracle.com> Reviewed-by: Richard Levitte <levitte@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14487)
Matt Caswell [Thu, 18 Mar 2021 16:52:10 +0000 (16:52 +0000)]
Ensure buffer/length pairs are always in sync
Following on from CVE-2021-3449 which was caused by a non-zero length
associated with a NULL buffer, other buffer/length pairs are updated to
ensure that they too are always in sync.
Reviewed-by: Tomas Mraz <tomas@openssl.org> Reviewed-by: Paul Dale <pauli@openssl.org>
As the variable peer_sigalgslen is not cleared on ssl rehandshake, it's
possible to crash an openssl tls secured server remotely by sending a
manipulated hello message in a rehandshake.
On such a manipulated rehandshake, tls1_set_shared_sigalgs() calls
tls12_shared_sigalgs() with the peer_sigalgslen of the previous
handshake, while the peer_sigalgs has been freed.
As a result tls12_shared_sigalgs() walks over the available
peer_sigalgs and tries to access data of a NULL pointer.
This issue was introduced by c589c34e61 (Add support for the TLS 1.3
signature_algorithms_cert extension, 2018-01-11).
Signed-off-by: Peter Kästle <peter.kaestle@nokia.com> Signed-off-by: Samuel Sapalski <samuel.sapalski@nokia.com>
CVE-2021-3449
CLA: trivial
Reviewed-by: Tomas Mraz <tomas@openssl.org> Reviewed-by: Paul Dale <pauli@openssl.org> Reviewed-by: Matt Caswell <matt@openssl.org>
Matt Caswell [Thu, 18 Mar 2021 15:25:42 +0000 (15:25 +0000)]
Teach TLSProxy how to encrypt <= TLSv1.2 ETM records
Previously TLSProxy only knew how to "repack" messages for TLSv1.3.
Most of the handshake in <= TLSv1.2 is unencrypted so this hasn't been
too much of restriction. However we now want to modify reneg handshakes
which are encrypted so we need to add that capability.
Reviewed-by: Tomas Mraz <tomas@openssl.org> Reviewed-by: Paul Dale <pauli@openssl.org>
Tomas Mraz [Mon, 22 Mar 2021 08:51:52 +0000 (08:51 +0000)]
check_chain_extensions: Do not override error return value by check_curve
The X509_V_FLAG_X509_STRICT flag enables additional security checks of the
certificates present in a certificate chain. It is not set by default.
Starting from OpenSSL version 1.1.1h a check to disallow certificates with
explicitly encoded elliptic curve parameters in the chain was added to the
strict checks.
An error in the implementation of this check meant that the result of a
previous check to confirm that certificates in the chain are valid CA
certificates was overwritten. This effectively bypasses the check
that non-CA certificates must not be able to issue other certificates.
If a "purpose" has been configured then a subsequent check that the
certificate is consistent with that purpose also checks that it is a
valid CA. Therefore where a purpose is set the certificate chain will
still be rejected even when the strict flag has been used. A purpose is
set by default in libssl client and server certificate verification
routines, but it can be overriden by an application.
Affected applications explicitly set the X509_V_FLAG_X509_STRICT
verification flag and either do not set a purpose for the certificate
verification or, in the case of TLS client or server applications,
override the default purpose to make it not set.
CVE-2021-3450
Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Paul Dale <pauli@openssl.org>
Chenglong Zhang [Mon, 22 Mar 2021 07:29:28 +0000 (15:29 +0800)]
Fix missing INVALID_EXTENSION
Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Ben Kaduk <kaduk@mit.edu> Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14639)
Richard Levitte [Sat, 20 Mar 2021 08:09:40 +0000 (09:09 +0100)]
ASN1: Reset the content dump flag after dumping
When encountering a badly coded item, the DER printer (ASN1_print_dump())
sets a flag to ensure that an additional hex dump of the offending content
is printed as part of the output. Unfortunately, this flag is never reset,
which means that all following items are printed with the extra hex dump,
whether they are faulty or not.
Resetting the flag after hex dumping ensures that only the faulty contents
are printed with the additional hex dump.
Fixes #14626
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14627)
Matt Caswell [Thu, 18 Mar 2021 10:22:51 +0000 (10:22 +0000)]
Add a missing RUN_ONCE in rand_lib.c
Some of the callbacks in rand_lib.c were being invoked without the
RUN_ONCE for that file being called. We add it during rand_pool_new
which should cover all cases.
Fixes #7870
Fixes #11144
Reviewed-by: Tomas Mraz <tomas@openssl.org> Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14603)
Pauli [Wed, 17 Mar 2021 02:23:52 +0000 (12:23 +1000)]
apps: fix coverity 966560: division by zero
Reviewed-by: Tomas Mraz <tomas@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14586)