2020-10-27 01:06:59 by Michael Forney | Files touched by this commit (1) |
Log message:
mbedtls: avoid implementation-defined find(1) usage
It is implementation-defined whether find(1) replaces {} in arguments
that are not exactly equal to {}, so use a for-loop instead.
|
2020-09-03 22:30:56 by Nia Alarie | Files touched by this commit (3) |
Log message:
mbedtls: Update to 2.24.0
= mbed TLS 2.24.0 branch released 2020-09-01
API changes
* In the PSA API, rename the types of elliptic curve and Diffie-Hellman
group families to psa_ecc_family_t and psa_dh_family_t, in line with the
PSA Crypto API specification version 1.0.0.
Rename associated macros as well:
PSA_ECC_CURVE_xxx renamed to PSA_ECC_FAMILY_xxx
PSA_DH_GROUP_xxx renamed to PSA_DH_FAMILY_xxx
PSA_KEY_TYPE_GET_CURVE renamed to to PSA_KEY_TYPE_ECC_GET_FAMILY
PSA_KEY_TYPE_GET_GROUP renamed to PSA_KEY_TYPE_DH_GET_FAMILY
Default behavior changes
* Stop storing persistent information about externally stored keys created
through PSA Crypto with a volatile lifetime. Reported in #3288 and
contributed by Steven Cooreman in #3382.
Features
* The new function mbedtls_ecp_write_key() exports private ECC keys back to
a byte buffer. It is the inverse of the existing mbedtls_ecp_read_key().
* Support building on e2k (Elbrus) architecture: correctly enable
-Wformat-signedness, and fix the code that causes signed-one-bit-field
and sign-compare warnings. Contributed by makise-homura (Igor Molchanov)
<akemi_homura@kurisa.ch>.
Security
* Fix a vulnerability in the verification of X.509 certificates when
matching the expected common name (the cn argument of
mbedtls_x509_crt_verify()) with the actual certificate name: when the
subjecAltName extension is present, the expected name was compared to any
name in that extension regardless of its type. This means that an
attacker could for example impersonate a 4-bytes or 16-byte domain by
getting a certificate for the corresponding IPv4 or IPv6 (this would
require the attacker to control that IP address, though). Similar attacks
using other subjectAltName name types might be possible. Found and
reported by kFYatek in #3498.
* When checking X.509 CRLs, a certificate was only considered as revoked if
its revocationDate was in the past according to the local clock if
available. In particular, on builds without MBEDTLS_HAVE_TIME_DATE,
certificates were never considered as revoked. On builds with
MBEDTLS_HAVE_TIME_DATE, an attacker able to control the local clock (for
example, an untrusted OS attacking a secure enclave) could prevent
revocation of certificates via CRLs. Fixed by no longer checking the
revocationDate field, in accordance with RFC 5280. Reported by
yuemonangong in #3340. Reported independently and fixed by
Raoul Strackx and Jethro Beekman in #3433.
* In (D)TLS record decryption, when using a CBC ciphersuites without the
Encrypt-then-Mac extension, use constant code flow memory access patterns
to extract and check the MAC. This is an improvement to the existing
countermeasure against Lucky 13 attacks. The previous countermeasure was
effective against network-based attackers, but less so against local
attackers. The new countermeasure defends against local attackers, even
if they have access to fine-grained measurements. In particular, this
fixes a local Lucky 13 cache attack found and reported by Tuba Yavuz,
Farhaan Fowze, Ken (Yihan) Bai, Grant Hernandez, and Kevin Butler
(University of Florida) and Dave Tian (Purdue University).
* Fix side channel in RSA private key operations and static (finite-field)
Diffie-Hellman. An adversary with precise enough timing and memory access
information (typically an untrusted operating system attacking a secure
enclave) could bypass an existing counter-measure (base blinding) and
potentially fully recover the private key.
* Fix a 1-byte buffer overread in mbedtls_x509_crl_parse_der().
Credit to OSS-Fuzz for detecting the problem and to Philippe Antoine
for pinpointing the problematic code.
* Zeroising of plaintext buffers in mbedtls_ssl_read() to erase unused
application data from memory. Reported in #689 by
Johan Uppman Bruce of Sectra.
Bugfix
* Library files installed after a CMake build no longer have execute
permission.
* Use local labels in mbedtls_padlock_has_support() to fix an invalid symbol
redefinition if the function is inlined.
Reported in #3451 and fix contributed in #3452 by okhowang.
* Fix the endianness of Curve25519 keys imported/exported through the PSA
APIs. psa_import_key and psa_export_key will now correctly expect/output
Montgomery keys in little-endian as defined by RFC7748. Contributed by
Steven Cooreman in #3425.
* Fix build errors when the only enabled elliptic curves are Montgomery
curves. Raised by signpainter in #941 and by Taiki-San in #1412. This
also fixes missing declarations reported by Steven Cooreman in #1147.
* Fix self-test failure when the only enabled short Weierstrass elliptic
curve is secp192k1. Fixes #2017.
* PSA key import will now correctly import a Curve25519/Curve448 public key
instead of erroring out. Contributed by Steven Cooreman in #3492.
* Use arc4random_buf on NetBSD instead of rand implementation with cyclical
lower bits. Fix contributed in #3540.
* Fix a memory leak in mbedtls_md_setup() when using HMAC under low memory
conditions. Reported and fix suggested by Guido Vranken in #3486.
* Fix bug in redirection of unit test outputs on platforms where stdout is
defined as a macro. First reported in #2311 and fix contributed in #3528.
Changes
* Only pass -Wformat-signedness to versions of GCC that support it. Reported
in #3478 and fix contributed in #3479 by okhowang.
* Reduce the stack consumption of mbedtls_x509write_csr_der() which
previously could lead to stack overflow on constrained devices.
Contributed by Doru Gucea and Simon Leet in #3464.
* Undefine the ASSERT macro before defining it locally, in case it is defined
in a platform header. Contributed by Abdelatif Guettouche in #3557.
* Update copyright notices to use Linux Foundation guidance. As a result,
the copyright of contributors other than Arm is now acknowledged, and the
years of publishing are no longer tracked in the source files. This also
eliminates the need for the lines declaring the files to be part of
MbedTLS. Fixes #3457.
* Add the command line parameter key_pwd to the ssl_client2 and ssl_server2
example applications which allows to provide a password for the key file
specified through the existing key_file argument. This allows the use of
these applications with password-protected key files. Analogously but for
ssl_server2 only, add the command line parameter key_pwd2 which allows to
set a password for the key file provided through the existing key_file2
argument.
|
2020-07-07 13:21:39 by Nia Alarie | Files touched by this commit (1) |
Log message:
mbedtls: Set BUILDLINK_ABI_DEPENDS
|
2020-07-07 13:16:38 by Nia Alarie | Files touched by this commit (1) |
Log message:
mbedtls: force python3
|
2020-07-07 13:16:10 by Nia Alarie | Files touched by this commit (9) | |
Log message:
mbedtls: Update to 2.23.0
= mbed TLS 2.23.0 branch released 2020-07-01
Default behavior changes
* In the experimental PSA secure element interface, change the encoding of
key lifetimes to encode a persistence level and the location. Although C
prototypes do not effectively change, code calling
psa_register_se_driver() must be modified to pass the driver's location
instead of the keys' lifetime. If the library is upgraded on an existing
device, keys created with the old lifetime value will not be readable or
removable through Mbed TLS after the upgrade.
Features
* New functions in the error module return constant strings for
high- and low-level error codes, complementing mbedtls_strerror()
which constructs a string for any error code, including compound
ones, but requires a writable buffer. Contributed by Gaurav Aggarwal
in #3176.
* The new utility programs/ssl/ssl_context_info prints a human-readable
dump of an SSL context saved with mbedtls_ssl_context_save().
* Add support for midipix, a POSIX layer for Microsoft Windows.
* Add new mbedtls_x509_crt_parse_der_with_ext_cb() routine which allows
parsing unsupported certificate extensions via user provided callback.
Contributed by Nicola Di Lieto <nicola.dilieto@gmail.com> in #3243 as
a solution to #3241.
* Pass the "certificate policies" extension to the callback supplied to
mbedtls_x509_crt_parse_der_with_ext_cb() if it contains unsupported
policies (#3419).
* Added support to entropy_poll for the kern.arandom syscall supported on
some BSD systems. Contributed by Nia Alarie in #3423.
* Add support for Windows 2000 in net_sockets. Contributed by opatomic. #3239
Security
* Fix a side channel vulnerability in modular exponentiation that could
reveal an RSA private key used in a secure enclave. Noticed by Sangho Lee,
Ming-Wei Shih, Prasun Gera, Taesoo Kim and Hyesoon Kim (Georgia Institute
of Technology); and Marcus Peinado (Microsoft Research). Reported by Raoul
Strackx (Fortanix) in #3394.
* Fix side channel in mbedtls_ecp_check_pub_priv() and
mbedtls_pk_parse_key() / mbedtls_pk_parse_keyfile() (when loading a
private key that didn't include the uncompressed public key), as well as
mbedtls_ecp_mul() / mbedtls_ecp_mul_restartable() when called with a NULL
f_rng argument. An attacker with access to precise enough timing and
memory access information (typically an untrusted operating system
attacking a secure enclave) could fully recover the ECC private key.
Found and reported by Alejandro Cabrera Aldaya and Billy Brumley.
* Fix issue in Lucky 13 counter-measure that could make it ineffective when
hardware accelerators were used (using one of the MBEDTLS_SHAxxx_ALT
macros). This would cause the original Lucky 13 attack to be possible in
those configurations, allowing an active network attacker to recover
plaintext after repeated timing measurements under some conditions.
Reported and fix suggested by Luc Perneel in #3246.
Bugfix
* Fix the Visual Studio Release x64 build configuration for mbedtls itself.
Completes a previous fix in Mbed TLS 2.19 that only fixed the build for
the example programs. Reported in #1430 and fix contributed by irwir.
* Fix undefined behavior in X.509 certificate parsing if the
pathLenConstraint basic constraint value is equal to INT_MAX.
The actual effect with almost every compiler is the intended
behavior, so this is unlikely to be exploitable anywhere. #3192
* Fix issue with a detected HW accelerated record error not being exposed
due to shadowed variable. Contributed by Sander Visser in #3310.
* Avoid NULL pointer dereferencing if mbedtls_ssl_free() is called with a
NULL pointer argument. Contributed by Sander Visser in #3312.
* Fix potential linker errors on dual world platforms by inlining
mbedtls_gcc_group_to_psa(). This allows the pk.c module to link separately
from psa_crypto.c. Fixes #3300.
* Remove dead code in X.509 certificate parsing. Contributed by irwir in
#2855.
* Include asn1.h in error.c. Fixes #3328 reported by David Hu.
* Fix potential memory leaks in ecp_randomize_jac() and ecp_randomize_mxz()
when PRNG function fails. Contributed by Jonas Lejeune in #3318.
* Remove unused macros from MSVC projects. Reported in #3297 and fix
submitted in #3333 by irwir.
* Add additional bounds checks in ssl_write_client_hello() preventing
output buffer overflow if the configuration declared a buffer that was
too small.
* Set _POSIX_C_SOURCE to at least 200112L in C99 code. Reported in #3420 and
fix submitted in #3421 by Nia Alarie.
* Fix building library/net_sockets.c and the ssl_mail_client program on
NetBSD. Contributed by Nia Alarie in #3422.
* Fix false positive uninitialised variable reported by cpp-check.
Contributed by Sander Visser in #3311.
* Update iv and len context pointers manually when reallocating buffers
using the MBEDTLS_SSL_VARIABLE_BUFFER_LENGTH feature. This caused issues
when receiving a connection with CID, when these fields were shifted
in ssl_parse_record_header().
Changes
* Fix warnings about signedness issues in format strings. The build is now
clean of -Wformat-signedness warnings. Contributed by Kenneth Soerensen
in #3153.
* Fix minor performance issue in operations on Curve25519 caused by using a
suboptimal modular reduction in one place. Found and fix contributed by
Aurelien Jarno in #3209.
* Combine identical cases in switch statements in md.c. Contributed
by irwir in #3208.
* Simplify a bounds check in ssl_write_certificate_request(). Contributed
by irwir in #3150.
* Unify the example programs termination to call mbedtls_exit() instead of
using a return command. This has been done to enable customization of the
behavior in bare metal environments.
* Fix mbedtls_x509_dn_gets to escape non-ASCII characters as "?".
Contributed by Koh M. Nakagawa in #3326.
* Use FindPython3 when cmake version >= 3.15.0
* Abort the ClientHello writing function as soon as some extension doesn't
fit into the record buffer. Previously, such extensions were silently
dropped. As a consequence, the TLS handshake now fails when the output
buffer is not large enough to hold the ClientHello.
* The unit tests now rely on header files in tests/include/test and source
files in tests/src. When building with make or cmake, the files in
tests/src are compiled and the resulting object linked into each test
executable.
* The ECP module, enabled by `MBEDTLS_ECP_C`, now depends on
`MBEDTLS_CTR_DRBG_C` or `MBEDTLS_HMAC_DRBG_C` for some side-channel
coutermeasures. If side channels are not a concern, this dependency can
be avoided by enabling the new option `MBEDTLS_ECP_NO_INTERNAL_RNG`.
* Align MSVC error flag with GCC and Clang. Contributed by Carlos Gomes
Martinho. #3147
* Remove superfluous assignment in mbedtls_ssl_parse_certificate(). Reported
in #3182 and fix submitted by irwir. #3217
* Fix typo in XTS tests. Reported and fix submitted by Kxuan. #3319
|
2020-06-29 14:39:36 by Nia Alarie | Files touched by this commit (4) |
Log message:
mbedtls: Add KERN_ARND support.
Motivation: the default behaviour of reopening /dev/urandom repeatedly
for every 128 bytes of entropy required is _exceedingly_ slow on NetBSD.
Not helped is using fread(), which assumes a long-lived file and buffers
excessively. This change makes the standard gen_entropy tool run in
milliseconds instead of seconds when it generates 48K of randomness.
Not only that, but sysctl is a lot more robust in e.g. chroots, resource
limited processes, etc.
Risk: On NetBSD, the security properties of the previous and current
behaviour are identical.
Upstreamed: https://github.com/ARMmbed/mbedtls/pull/3423
Bump PKGREVISION.
|
2020-06-11 13:43:50 by Nia Alarie | Files touched by this commit (6) |
Log message:
mbedtls: Update patches. Add links to PRs.
|
2020-06-02 17:07:59 by Nia Alarie | Files touched by this commit (3) |
Log message:
mbedtls: attempt to fix build on SunOS
|
2020-04-18 16:21:56 by Nia Alarie | Files touched by this commit (2) |
Log message:
mbedtls: Update to 2.16.6
= mbed TLS 2.16.6 branch released 2020-04-14
Security
* Fix side channel in ECC code that allowed an adversary with access to
precise enough timing and memory access information (typically an
untrusted operating system attacking a secure enclave) to fully recover
an ECDSA private key. Found and reported by Alejandro Cabrera Aldaya,
Billy Brumley and Cesar Pereida Garcia. CVE-2020-10932
* Fix a potentially remotely exploitable buffer overread in a
DTLS client when parsing the Hello Verify Request message.
Bugfix
* Fix compilation failure when both MBEDTLS_SSL_PROTO_DTLS and
MBEDTLS_SSL_HW_RECORD_ACCEL are enabled.
* Fix a function name in a debug message. Contributed by Ercan Ozturk in
#3013.
|
2020-03-22 08:47:00 by Nia Alarie | Files touched by this commit (1) |
Log message:
mbedtls: doesn't need gmake
|