Based on Paul Fertser <fercerpav@gmail.com>'s guidance:
Change AUTORELEASE in rules.mk to:
```
AUTORELEASE = $(if $(DUMP),0,$(shell sed -i "s/\$$(AUTORELEASE)/$(call commitcount,1)/" $(CURDIR)/Makefile))
```
then update all affected packages by:
```
for i in $(git grep -l PKG_RELEASE:=.*AUTORELEASE | sed 's^.*/\([^/]*\)/Makefile^\1^';);
do
make package/$i/clean
done
```
Signed-off-by: Tianling Shen <cnsztl@immortalwrt.org>
This reduces open coding and allows to easily add a knob to enable
it treewide, where chosen packages can still opt-out via "no-lto".
Some packages used LTO, but not the linker plugin. This unifies 'em
all to attempt to produce better code.
Quoting man gcc(1):
"This improves the quality of optimization by exposing more code to the
link-time optimizer."
Also use -flto=auto instead of -flto=jobserver, as it's not guaranteed
that every buildsystem uses +$(MAKE) correctly.
Signed-off-by: Andre Heider <a.heider@gmail.com>
Inline the preinst.arm-ce script. Support for including was added in
make 4.2 and is not working with older make versions.
Fixes: https://github.com/openwrt/openwrt/issues/11866
Signed-off-by: Chen Minqiang <ptpt52@gmail.com>
Fixes denial of service attack and buffer overflow against TLS 1.3
servers using session ticket resumption. When built with
--enable-session-ticket and making use of TLS 1.3 server code in
wolfSSL, there is the possibility of a malicious client to craft a
malformed second ClientHello packet that causes the server to crash.
This issue is limited to when using both --enable-session-ticket and TLS
1.3 on the server side. Users with TLS 1.3 servers, and having
--enable-session-ticket, should update to the latest version of wolfSSL.
Thanks to Max at Trail of Bits for the report and "LORIA, INRIA, France"
for research on tlspuffin.
Complete release notes https://github.com/wolfSSL/wolfssl/releases/tag/v5.5.1-stable
Fixes: CVE-2022-39173
Fixes: https://github.com/openwrt/luci/issues/5962
References: https://github.com/wolfSSL/wolfssl/issues/5629
Tested-by: Kien Truong <duckientruong@gmail.com>
Reported-by: Kien Truong <duckientruong@gmail.com>
Signed-off-by: Petr Štetiar <ynezz@true.cz>
This reverts commit a596a8396b as I've
just discovered private email, that the issue has CVE-2022-39173
assigned so I'm going to reword the commit and push it again.
Signed-off-by: Petr Štetiar <ynezz@true.cz>
Fixes denial of service attack and buffer overflow against TLS 1.3
servers using session ticket resumption. When built with
--enable-session-ticket and making use of TLS 1.3 server code in
wolfSSL, there is the possibility of a malicious client to craft a
malformed second ClientHello packet that causes the server to crash.
This issue is limited to when using both --enable-session-ticket and TLS
1.3 on the server side. Users with TLS 1.3 servers, and having
--enable-session-ticket, should update to the latest version of wolfSSL.
Thanks to Max at Trail of Bits for the report and "LORIA, INRIA, France"
for research on tlspuffin.
Complete release notes https://github.com/wolfSSL/wolfssl/releases/tag/v5.5.1-stable
Fixes: https://github.com/openwrt/luci/issues/5962
References: https://github.com/wolfSSL/wolfssl/issues/5629
Signed-off-by: Petr Štetiar <ynezz@true.cz>
Rename libwolfssl-cpu-crypto to libwolfsslcpu-crypto so that the
regular libwolfssl version comes first when running:
opkg install libwolfssl
Normally, if the package name matches the opkg parameter, that package
is preferred. However, for libraries, the ABI version string is
appended to the package official name, and the short name won't match.
Failing a name match, the candidate packages are sorted in alphabetical
order, and a dash will come before any number. So in order to prefer
the original library, the dash should be removed from the alternative
library.
Fixes: c3e7d86d2b (wolfssl: add libwolfssl-cpu-crypto package)
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Move CONFIG_PACKAGE_libwolfssl-benchmark from the top of
PKG_CONFIG_DEPENDS to after PKG_ABI_VERSION is set.
This avoids changing the ABI version hash whether the bnechmark package
package is selected or not.
Fixes: 05df135cac (wolfssl: Rebuild when libwolfssl-benchmark gets changes)
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
libwolfssl-cpu-crypto is a variant of libwolfssl with support for
cryptographic CPU instructions on x86_64 and aarch64.
On aarch64, wolfSSL does not perform run-time detection, so the library
will crash when the AES functions are called. A preinst script attempts
to check for support by querying /proc/cpuinfo, if installed in a
running system. When building an image, the script will check the
DISTRIB_TARGET value in /etc/openwrt_release, and will abort
installation if target is bcm27xx.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
This forces a rebuild of the wolfssl package when the
libwolfssl-benchmark OpenWrt package gets activated or deactivated.
Without this change the wolfssl build will fail when it compiled without
libwolfssl-benchmark before and it gets activated for the next build.
Fixes: 18fd12edb8 ("wolfssl: add benchmark utility")
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Apply upstream patch[1] to fix breakage around math libraries.
This can likely be removed when 5.5.0-stable is tagged and released.
Build system: x86_64
Build-tested: bcm2711/RPi4B
Run-tested: bcm2711/RPi4B
1. https://github.com/wolfSSL/wolfssl/pull/5390
Signed-off-by: John Audia <therealgraysky@proton.me>
Disable the usage of target specific CPU crypto instructions by default
to allow the package being shared again. Since WolfSSL does not offer
a stable ABI or a long term support version suitable for OpenWrt release
timeframes, we're forced to frequently update it which is greatly
complicated by the package being nonshared.
People who want or need CPU crypto instruction support can enable it in
menuconfig while building custom images for the few platforms that support
them.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
The armvirt target is also used to run OpenWrt in lxc on other targets
like a Raspberry Pi. If we set WOLFSSL_HAS_CPU_CRYPTO by default the
wolfssl binray is only working when the CPU supports the hardware crypto
extension.
Some targets like the Raspberry Pi do not support the ARM CPU crypto
extension, compile wolfssl without it by default. It is still possible
to activate it in custom builds.
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
This version fixes two vulnerabilities:
-CVE-2022-34293[high]: Potential for DTLS DoS attack
-[medium]: Ciphertext side channel attack on ECC and DH operations.
The patch fixing x86 aesni build has been merged upstream.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Apply an upstream patch that removes unnecessary CFLAGs, avoiding
generation of incompatible code.
Commit 0bd5367233 is reverted so the
accelerated version builds by default on x86_64.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Without this, WOLFSSL_HAS_DH can be disabled even if WOLFSSL_HAS_WPAS is
enabled, resulting in an "Anonymous suite requires DH" error when trying
to compile wolfssl.
Signed-off-by: Pascal Ernster <git@hardfalcon.net>
Reviewed-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
This enables building WolfSSL with Curve448, which can be used by
Strongswan. This has been tested on a Linksys E8450, running OpenWrt
22.03-rc4.
This allows parity with OpenSSL, which already supports Curve448 in
OpenWrt 21.02.
Fixesopenwrt/packages#18812.
Signed-off-by: Joel Low <joel@joelsplace.sg>
WolfSSL is crashing with an illegal opcode in some x86_64 CPUs that have
AES instructions but lack other extensions that are used by WolfSSL
when AES-NI is enabled.
Disable the option by default for now until the issue is properly fixed.
People can enable them in a custom build if they are sure it will work
for them.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Openvpn forces CONFIG_WOLFSSL_HAS_OPENVPN=y. When the phase1 bots build
the now non-shared package, openvpn will not be selected, and WolfSSL
will be built without it. Then phase2 bots have CONFIG_ALL=y, which
will select openvpn and force CONFIG_WOLFSSL_HAS_OPENVPN=y. This
changes the version hash, causing dependency failures, as shared
packages expect the phase2 hash.
Fixes: #9738
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
libwolfssl-benchmark should NOT be compiled as nonshared but
currently there is a bug where, on buildbot stage2, the package
is recompiled to build libwolfssl-benchmark and the dependency
change to the new libwolfssl version.
Each dependant package will now depend on the new wolfssl package
instead of the one previously on stage1 that has a different package
HASH.
Set the nonshared PKGFLAGS global while this gets investigated
and eventually fixed.
Fixes: 0a2edc2714 ("wolfssl: enable CPU crypto instructions")
Signed-off-by: Christian 'Ansuel' Marangi <ansuelsmth@gmail.com>
This enables AES & SHA CPU instructions for compatible armv8, and x86_64
architectures. Add this to the hardware acceleration choice, since they
can't be enabled at the same time.
The package was marked non-shared, since the arm CPUs may or may not
have crypto extensions enabled based on licensing; bcm27xx does not
enable them. There is no run-time detection of this for arm.
NOTE:
Should this be backported to a release branch, it must be done shortly
before a new minor release, because the change to nonshared will remove
libwolfssl from the shared packages, but the nonshared are only built in
a subsequent release!
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Enabling different hardware crypto acceleration should not change the
library ABI. Add them to PKG_CONFIG_DEPENDS after the ABI version hash
has been computed.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
This is mostly a bug fix release, including two that were already
patched here:
- 300-fix-SSL_get_verify_result-regression.patch
- 400-wolfcrypt-src-port-devcrypto-devcrypto_aes.c-remove-.patch
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Fixes two high-severity vulnerabilities:
- CVE-2022-25640: A TLS v1.3 server who requires mutual authentication
can be bypassed. If a malicious client does not send the
certificate_verify message a client can connect without presenting a
certificate even if the server requires one.
- CVE-2022-25638: A TLS v1.3 client attempting to authenticate a TLS
v1.3 server can have its certificate heck bypassed. If the sig_algo in
the certificate_verify message is different than the certificate
message checking may be bypassed.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Backport fix for API breakage of SSL_get_verify_result() introduced in
v5.1.1-stable. In v4.8.1-stable SSL_get_verify_result() used to return
X509_V_OK when used on LE powered sites or other sites utilizing
relaxed/alternative cert chain validation feature. After an update to
v5.1.1-stable that API calls started returning X509_V_ERR_INVALID_CA
error and thus rendered all such connection attempts imposible:
$ docker run -it openwrt/rootfs:x86_64-21.02.2 sh -c "wget https://letsencrypt.org"
Downloading 'https://letsencrypt.org'
Connecting to 18.159.128.50:443
Connection error: Invalid SSL certificate
Fixes: #9283
References: https://github.com/wolfSSL/wolfssl/issues/4879
Signed-off-by: Petr Štetiar <ynezz@true.cz>
x509v3 SAN extension is required to generate a certificate compatible with
chromium-based web browsers (version >58)
It can be disabled via unsetting CONFIG_WOLFSSL_ALT_NAMES
Signed-off-by: Sergey V. Lobanov <sergey@lobanov.in>
It's the default anyway and this just looks confusing, as if it wasn't.
Switch to AUTORELEASE while at it.
The binary size is unchanged.
Signed-off-by: Andre Heider <a.heider@gmail.com>
This gates out anything that might introduce semantically frivolous jitter,
maximizing chance of identical object files.
The binary size shrinks by 8kb:
1244352 staging_dir/target-mipsel_24kc_musl/usr/lib/libwolfssl.so.4.8.1.39c36f2f
1236160 staging_dir/target-mipsel_24kc_musl/usr/lib/libwolfssl.so.4.8.1.39c36f2f
Signed-off-by: Andre Heider <a.heider@gmail.com>
"Alternate certification chains, as oppossed to requiring full chain
validataion. Certificate validation behavior is relaxed, similar to
openssl and browsers. Only the peer certificate must validate to a trusted
certificate. Without this, all certificates sent by a peer must be
used in the trust chain or the connection will be rejected."
This fixes e.g. uclient-fetch and curl connecting to servers using a Let's
Encrypt certificate which are cross-signed by the now expired
DST Root CA X3, see [0].
This is the recommended solution from upstream [1].
The binary size increases by ~12.3kb:
1236160 staging_dir/target-mipsel_24kc_musl/usr/lib/libwolfssl.so.4.8.1.39c36f2f
1248704 staging_dir/target-mipsel_24kc_musl/usr/lib/libwolfssl.so.4.8.1.39c36f2f
[0] https://github.com/openwrt/packages/issues/16674
[1] https://github.com/wolfSSL/wolfssl/issues/4443#issuecomment-934926793
Signed-off-by: Andre Heider <a.heider@gmail.com>
[bump PKG_RELEASE]
Signed-off-by: David Bauer <mail@david-bauer.net>
Changes from 4.7.0:
Fix one high (OCSP verification issue) and two low vulnerabilities
Improve compatibility layer
Other improvements and fixes
For detailed changes refer to https://github.com/wolfSSL/wolfssl/releases
Signed-off-by: Ivan Pavlov <AuthorReflex@gmail.com>
Support for wolfSSL has been upstreamed to the master OpenVPN branch
in f6dca235ae560597a0763f0c98fcc9130b80ccf4, so we can use wolfSSL
directly in OpenVPN. So no more needed differnt SSL engine for OpenVPN
in systems based on wolfSSL library
Compiled && tested on ramips/mt7620, ramips/mt7621
Signed-off-by: Ivan Pavlov <AuthorReflex@gmail.com>
Since commit 6467de5a8840 ("Randomize z ordinates in scalar
mult when timing resistant") wolfssl requires a RNG for an EC
key when the hardened built option is selected.
wc_ecc_set_rng is only available when built hardened, so there
is no safe way to install the RNG to the key regardless whether
or not wolfssl is compiled hardened.
Always export wc_ecc_set_rng so tools such as hostapd can install
RNG regardless of the built settings for wolfssl.
Signed-off-by: David Bauer <mail@david-bauer.net>
