Using the local gnulib source during autogen.sh
allows for fine-grained control over the macros
and source files for use with gettext
but part of gnulib instead of gettext,
without having to wait for a release
or deal with gnulib as a git submodule.
This is an alternative to running autoreconf.
It also removes the need to patch macros
in the case where there is a conflict
between the source and our aclocal directory.
Signed-off-by: Michael Pratt <mcpratt@pm.me>
Some users have reported that gettext builds
are attempting to link to libxml2
while it was supposed to be configured
to use it's own built-in substitute.
Configure gettext to require and link
to our local libxml2 explicitly.
Add a patch to revert upstream commit 87927a4e2
which forces libtextstyle to use the built-in libxml,
no matter what the configuration is,
making that option configurable again
after the configure script is regenerated.
Reported-by: Tianling Shen <cnsztl@immortalwrt.org>
Signed-off-by: Michael Pratt <mcpratt@pm.me>
Instead of editing the SUBDIRS variable with a patch,
it can be overriden at the end of the command line when invoking Make.
This tool has a series of recursive Makefiles in each subdirectory,
therefore SUBDIRS is set to a pattern of Make functions
so that the result is variable depending on the current subdirectory
that Make is being invoked in.
Some of the subdirectories don't have a Makefile and are just storing files
for another subdirectory Makefile target,
therefore we have to place a fake Makefile that does nothing.
Signed-off-by: Michael Pratt <mcpratt@pm.me>
This applies commit 02ac9c94 to fix this OpenSSL Security Advisory
issued on 20th April 2023[1]:
Input buffer over-read in AES-XTS implementation on 64 bit ARM
(CVE-2023-1255)
==============================================================
Severity: Low
Issue summary: The AES-XTS cipher decryption implementation for 64 bit
ARM platform contains a bug that could cause it to read past the input
buffer, leading to a crash.
Impact summary: Applications that use the AES-XTS algorithm on the 64
bit ARM platform can crash in rare circumstances. The AES-XTS algorithm
is usually used for disk encryption.
The AES-XTS cipher decryption implementation for 64 bit ARM platform
will read past the end of the ciphertext buffer if the ciphertext size
is 4 mod 5 in 16 byte blocks, e.g. 144 bytes or 1024 bytes. If the
memory after the ciphertext buffer is unmapped, this will trigger a
crash which results in a denial of service.
If an attacker can control the size and location of the ciphertext
buffer being decrypted by an application using AES-XTS on 64 bit ARM,
the application is affected. This is fairly unlikely making this issue a
Low severity one.
1. https://www.openssl.org/news/secadv/20230420.txt
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Apply two patches fixing low-severity vulnerabilities related to
certificate policies validation:
- Excessive Resource Usage Verifying X.509 Policy Constraints
(CVE-2023-0464)
Severity: Low
A security vulnerability has been identified in all supported versions
of OpenSSL related to the verification of X.509 certificate chains
that include policy constraints. Attackers may be able to exploit
this vulnerability by creating a malicious certificate chain that
triggers exponential use of computational resources, leading to a
denial-of-service (DoS) attack on affected systems.
Policy processing is disabled by default but can be enabled by passing
the `-policy' argument to the command line utilities or by calling the
`X509_VERIFY_PARAM_set1_policies()' function.
- Invalid certificate policies in leaf certificates are silently ignored
(CVE-2023-0465)
Severity: Low
Applications that use a non-default option when verifying certificates
may be vulnerable to an attack from a malicious CA to circumvent
certain checks.
Invalid certificate policies in leaf certificates are silently ignored
by OpenSSL and other certificate policy checks are skipped for that
certificate. A malicious CA could use this to deliberately assert
invalid certificate policies in order to circumvent policy checking on
the certificate altogether.
Policy processing is disabled by default but can be enabled by passing
the `-policy' argument to the command line utilities or by calling the
`X509_VERIFY_PARAM_set1_policies()' function.
Note: OpenSSL also released a fix for low-severity security advisory
CVE-2023-466. It is not included here because the fix only changes the
documentation, which is not built nor included in any OpenWrt package.
Due to the low-severity of these issues, there will be not be an
immediate new release of OpenSSL.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
This adapts the engine build infrastructure to allow building providers,
and packages the legacy provider. Providers are the successors of
engines, which have been deprecated.
The legacy provider supplies OpenSSL implementations of algorithms that
have been deemed legacy, including DES, IDEA, MDC2, SEED, and Whirlpool.
Even though these algorithms are implemented in a separate package,
their removal makes the regular library smaller by 3%, so the build
options will remain to allow lean custom builds. Their defaults will
change to 'y' if not bulding for a small flash, so that the regular
legacy package will contain a complete set of algorithms.
The engine build and configuration structure was changed to accomodate
providers, and adapt to the new style of openssl.cnf in version 3.0.
There is not a clean upgrade path for the /etc/ssl/openssl.cnf file,
installed by the openssl-conf package. It is recommended to rename or
remove the old config file when flashing an image with the updated
openssl-conf package, then apply the changes manually.
An old openssl.cnf file will silently work, but new engine or provider
packages will not be enabled. Any remaining engine config files under
/etc/ssl/engines.cnf.d can be removed.
On the build side, the include file used by engine packages was renamed
to openssl-module.mk, so the engine packages in other feeds need to
adapt.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Engines that are built into the main libcrypto OpenSSL library can't be
disabled through UCI. Add a 'builtin' setting to signal that the engine
can't be disabled through UCI, and show a message explaining this in
case buitin=1 and enabled=0.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Building openssl with OPENSSL_SMALL_FOOTPRINT yelds only from 1% to 3%
decrease in size, dropping performance from 2% to 91%, depending on the
target and algorithm.
For example, using AES256-GCM with 1456-bytes operations, X86_64 appears
to be the least affected with 2% performance penalty and 1% reduction in
size; mips drops performance by 13%, size by 3%; Arm drops 29% in
performance, 2% in size.
On aarch64, it slows down ghash so much that I consider it broken
(-91%). SMALL_FOOTPRINT will reduce AES256-GCM performance by 88%, and
size by only 1%. It makes an AES-capable CPU run AES128-GCM at 35% of
the speed of Chacha20-Poly1305:
Block-size=1456 bytes AES256-GCM AES128-GCM ChaCha20-Poly1305
SMALL_FOOTPRINT 62014.44 65063.23 177090.50
regular 504220.08 565630.28 182706.16
OpenSSL 1.1.1 numbers are about the same, so this should have been
noticed a long time ago.
This creates an option to use OPENSSL_SMALL_FOOTPRINT, but it is turned
off by default unless SMALL_FLASH or LOW_MEMORY_FOOTPRINT is used.
Compiling with -O3 instead of -Os, for comparison, will increase size by
about 14-15%, with no measureable effect on AES256-GCM performance, and
about 2% increase in Chacha20-Poly1305 performance on Aarch64.
There are no Arm devices with the small flash feature, so drop the
conditional default. The package is built on phase2, so even if we
include an Arm device with small flash later, a no-asm library would
have to be built from source anyway.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Changes:
1c6f0f3 libtraceevent: version 1.7.2
73f6a8a libtraceevent: Fix some missing commas in big endian blocks
da2ea6b libtraceevent: Rename "ok" to "token_has_paren" in process_sizeof()
e6f7cfa libtraceevent: No need for testing ok in else if (!ok) in process_sizeof()
a4b1ba5 libtraceevent: Fix double free in parsing sizeof()
Signed-off-by: Nick Hainke <vincent@systemli.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>
This reduces open coding and allows to easily add a knob to
enable it treewide, where chosen packages can still opt-out via
"no-gc-sections".
Note: libnl, mbedtls and opkg only used the CFLAGS part without the
LDFLAGS counterpart. That doesn't help at all if the goal is to produce
smaller binaries. I consider that an accident, and this fixes it.
Note: there are also packages using only the LDFLAGS part. I didn't
touch those, as gc might have been disabled via CFLAGS intentionally.
Signed-off-by: Andre Heider <a.heider@gmail.com>
Fix the trivial abscence of $() when assigning engine config files to
the main libopenssl-config package even if the corresponding engines
were not built into the main library.
This is mostly cosmetic, since scripts/ipkg-build tests the file's
presence before it is actually included in the package's conffiles.
Fixes: 30b0351039 "openssl: configure engine packages during install"
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
The bump to 3.0.8 inadvertently removed patches that are needed here,
but were not adopted upstream. The most important one changes the
default value of the DIGESTS setting from ALL to NONE. The absence of
this patch causes a sysupgrade failure while the engine is in use with
digests enabled. When this happens, the system fails to boot with a
kernel panic.
Also, explicitly set DIGESTS to NONE in the provided config file, and
change the default ciphers setting to disable ECB, which has been
recommended for a long time and may cause trouble with some apps.
The config file change by itself is not enough because the config file
may be preserved during sysupgrade.
For people affected by this bug:
You can either:
1. remove, the libopenssl-devcrypto package
2. disable the engine in /etc/config/openssl;
3. change /etc/ssl/engines.cnf.d/devcrypto.cnf to set DIGESTS=NONE;
4. update libopenssl-devcrypto to >=3.0.8-3
However, after doing any of the above, **you must reboot the device
before running sysupgrade** to ensure no running application is using
the engine. Running `/etc/init.d/openssl restart` is not enough.
Fixes: 7e7e76afca "openssl: bump to 3.0.8"
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
PowerPC CONFIG_ARCH is defined as powerpc, not ppc. Fix that in the
DEPENDS condition.
Arc needs to be built with libatomic. Change the OpenSSL configuration
file, and add it to the libatomic DEPENDS condition.
Fixes: 7e7e76afca "openssl: bump to 3.0.8"
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
