Raise the minimal GCC version to 9.3, which is required since wpewebkit-2.40.0 [1].
Similar to commit 09af6d8bfd,
we do check on >= GCC 9, because we can't check on >= GCC 9.3.
[1] f9c142d9b5
Signed-off-by: Thomas Devoogdt <thomas.devoogdt@barco.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit 603fa625b1)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Make 4.4 introduces a shuffle mode which randomizes prerequisites
in order to better flush out issues with parallel builds. On the other
hand, we use MAKE1 to build packages that are known to be broken with
parallel build. For these, passing the shuffle option would be
counter-productive and lead to spurious build failures.
The --shuffle=none option exists to turn off shuffling again. We can't
add this option unconditionally, however, because Make < 4.4 doesn't
know it. Therefore, conditionally pass --shuffle=none only if there is a
shuffle option in MAKEFLAGS.
Signed-off-by: James Hilliard <james.hilliard1@gmail.com>
Signed-off-by: Arnout Vandecappelle <arnout@mind.be>
(cherry picked from commit f664d7dc24)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The reinstall, rebuild and reconfigure commands rely on the
left-to-right order of evaluation of the dependencies to make sure that
the stamp files are removed before attempting to rebuild. However, this
order of evaluation is not guaranteed. In particular, if top-level
parallel build is enabled, they are executed in parallel and the stamp
file may not have been removed yet when it is evaluated to decide if
rebuild has to be done.
Since make 4.4, it is possible to reproduce this issue by passing
`--shuffle=reverse` to the make commandline.
To solve this, add a .WAIT directive between the clean and
install/build/configure dependencies. .WAIT was introduced in make 4.4
as well. It makes sure that the dependencies on the left are evaluated
before the dependencies on the right - exactly what we want here.
Earlier versions of make don't know about .WAIT, so we need to add a
.PHONY dependency to effectively ignore it.
Note that this doesn't fix the problem for make versions earlier than
4.4. However, the issue isn't really that important: reinstall, rebuild
and reconfigure are development tools, they're not fully reliable to
begin with, and it's anyway less likely that someone uses `make -j` when
doing a reinstall/rebuild/reconfigure.
Signed-off-by: Arnout Vandecappelle <arnout@mind.be>
Reported-by: James Hilliard <james.hilliard1@gmail.com>
(cherry picked from commit 7e3105d5c8)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
- Fix CVE-2022-48303: GNU Tar through 1.34 has a one-byte out-of-bounds
read that results in use of uninitialized memory for a conditional
jump. Exploitation to change the flow of control has not been
demonstrated. The issue occurs in from_header in list.c via a V7
archive in which mtime has approximately 11 whitespace characters.
- Update hash of COPYING (http replaced by https)
https://lists.gnu.org/archive/html/info-gnu/2023-07/msg00005.html
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit d4d483451f)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
When building for a target architecture that go does not support, the
installation fails with:
$ make host-go
[...]
ln -sf ../lib/go/bin/go /home/nyma7486/dev/work/5GCroCo/O/pouet/per-package/host-go/host/bin/
ln: failed to create symbolic link '/home/nyma7486/dev/work/5GCroCo/O/pouet/per-package/host-go/host/bin/': No such file or directory
Indeed, the HOST_DIR/bin is not guaranteed to exist when we install a
host package, so it needs to be explicitly created before we can create
entries in there.
Signed-off-by: Yann E. MORIN <yann.morin@orange.com>
Cc: Christian Stewart <christian@aperture.us>
Cc: Anisse Astier <anisse@astier.eu>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit 1db38d9282)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Without this patch, a make <pkg>_rebuild detects overwrites. Indeed, in
target_finalize steps some modifications are done on installed files (ie
strip or TARGET_FINALIZE_HOOKS for instance).
In order to avoid these modifications seen from per-package {TARGET,HOST}_DIR
and so been analyzed as some overwrites, global {TARGET,HOST}_DIR is built
using a full copy of the involved per-package files instead of hardlinks.
Signed-off-by: Herve Codina <herve.codina@bootlin.com>
Reviewed-by: Yann E. MORIN <yann.morin.1998@free.fr>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit 21d52e52d8)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Raise the minimal GCC version to 9.3, which is required since webkitgtk-2.40.0 [1].
Similar to commit ec1ff802df,
we do check on >= GCC 9, because we can't check on >= GCC 9.3.
[1] f9c142d9b5
Signed-off-by: Thomas Devoogdt <thomas.devoogdt@barco.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit de38484ec3)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
This commit fixes the S10hyperv SysV init script which expects binaries
to be locate in /sbin while they are installed in /usr/sbin. Please
note, that the systemd init scripts correctly reference them.
Furthermore, the SysV init script did not check for an actual HyperV
environment to be present, which is also corrected. In addition, this
commit also fixes check-package warnings regarding a missing DAEMON
definition.
Signed-off-by: Jens Maus <mail@jens-maus.de>
[Peter: drop from .checkpackageignore]
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit 14c4bd7bf8)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The toolchain wrapper automatically adds Position Independent
Execution and stack protector flags in the build process when selected
in the configuration. at91bootstrap being freestanding code, it
doesn't support these, so we have to disable them.
Signed-off-by: Kory Maincent <kory.maincent@bootlin.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
(cherry picked from commit 1351222486)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The "official" repository we used to reference disapeared quite some
time ago, so in commit 8c25838b53 (package/rockchip-mali: fix build
failure due to missing URL) we switched to using a mirror.
The tarballs generated on the Github side have a top-level directory
that is named "repo-name-HASH", so when we switched to a repository
named "libmali" to one named "mirrors", the content of the generated
tarball changed, even though the content of the files did not.
We can't just change the hash to the new value, or that would conflict
with the copy on s.b.o and older versions of Buildroot.
So, we drop one char from the commit hash, which eans the tarball name
changes, and thus we can calculate a new hash for that tarball, and
there will be no conflict with any existing tarball on s.b.o.
Reported-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
(cherry picked from commit f7f2de3da7)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
We want to initialize the kernel random number generator as soon as
possible, as early init scripts may also need random numbers (E.G.
syslog-ng in S01syslog-ng does).
Seedrng was presumably only using S20 because the previos urandom script
used S20, which (after som moves) dates all the way back to:
commit 8262508fc4
Author: Eric Andersen <andersen@codepoet.org>
Date: Fri Apr 26 22:01:43 2002 +0000
With this update, everything now works as expected.
-Erik
Seedrng needs persistent storage, but mount -a is run before executing the
init scripts, so S01 should be as good as S20 - Atleast with the scripts in
upstream Buildroot.
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
(cherry picked from commit 11a46002ae)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Commit b9bf1c6535 (package/opkg-utils: add opkg-utils as target pkg)
misspelled the macro to install to target, most probably as a bad
copy-paste from the host macro.
Fix that.
Reported-by: Michael Hacker <mh@superbox.one>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit 8c3c741439)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
pciutils is optional, not mandatory since version 2.1 and the addition
of the package in commit 0adc0e24ee and
98ca605e7c
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
[yann.morin.1998@free.fr:
- one item per multi-line in _DEPENDENCIES
- reorder _DEPENDENCIES
]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
(cherry picked from commit bcfcc07413)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The gnu-efi package was enabled on mips64el by commit
11b347c03a ("package/gnu-efi: add
mips64el support"). However, it has been failing to build for a long
time, and nobody bothered fixing it:
gnu-efi-3.0.15//gnuefi/crt0-efi-mips64el.S:71: Error: cannot represent BFD_RELOC_16 relocation in this object file format
Even reverting back to gnu-efi 3.0.10, which was the version used at
the time of 11b347c03a, does not fix the
issue. We tested updating to the latest gnu-efi version, 3.0.17, and
the problem still exists.
Since EFI on MIPS is extremely niche, we don't really want to invest
the time to fix this issue, so let's disable it again. If someone
cares enough, it can be fixed and re-enabled.
Fixes:
http://autobuild.buildroot.net/results/d5938edcf50ebb7fdcec148d73f402845079779d/
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
(cherry picked from commit 13b0e6bbf3)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Fix the following build failure with gpsd >= 3.25 raised since commit
3c7fece853:
In file included from src/configuration.h:50,
from src/configuration.c:46:
src/gpsdclient.h:64:8: error: redefinition of 'struct fixsource_t'
64 | struct fixsource_t {
| ^~~~~~~~~~~
In file included from src/gpsdclient.h:49,
from src/configuration.h:50,
from src/configuration.c:46:
/tmp/instance-17/output-1/host/aarch64-buildroot-linux-gnu/sysroot/usr/include/gps.h:2714:8: note: originally defined here
2714 | struct fixsource_t
| ^~~~~~~~~~~
Fixes:
- http://autobuild.buildroot.org/results/47a619686bb47debd525c92aa7e14bee5c40ca9e
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit 4c32b3d9ff)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Some extensions of python-mako use python-babel, if they are used,
python-babel must be added as a runtime dependency.
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
(cherry picked from commit 411a3deb1a)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Some extensions of python-mako use python-pygments, if they are used,
python-pygments must be added as a runtime dependency.
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
(cherry picked from commit adb3f5a3e0)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
OpenBLAS RISC-V 64bit support was added in [1] and was renamed to
"RISCV64_GENERIC" in [2]. Those commits were first included in
OpenBLAS release v0.3.13. This support can now be enabled. With this
commit, we can install the library and packages such as GNU Octave on
RISC-V platforms.
This patch also adjusts the alignment for adding "RISCV64_GENERIC"
in Config.in.
[1] c167a3d6f4
[2] 265ab484c8
Signed-off-by: Yu Chien Peter Lin <peterlin@andestech.com>
Reviewed-by: Julien Olivain <ju.o@free.fr>
Tested-by: Julien Olivain <ju.o@free.fr>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit c789bcddf0)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Building go with cgo support needs to build some .c files to generate target
support code, and thus calls the cross C compiler, which is failing when the
toolchain is not built before host-go:
>>> host-go 1.21.1 Building
cd .../build/host-go-1.21.1/src && GO111MODULE=off GOCACHE=.../per-package/host-go/host/share/host-go-cache GOROOT_BOOTSTRAP=.../per-package/host-go/host/lib/go-1.19.11 GOROOT_FINAL=.../per-package/host-go/host/lib/go GOROOT=".../build/host-go-1.21.1" GOBIN=".../build/host-go-1.21.1/bin" GOOS=linux CC=/usr/bin/gcc CXX=/usr/bin/g++ CGO_ENABLED=1 CC_FOR_TARGET=".../per-package/host-go/host/bin/arm-linux-gcc" CXX_FOR_TARGET=".../per-package/host-go/host/bin/arm-linux-g++" GOOS="linux" GOARCH=arm GOARM=6 GO_ASSUME_CROSSCOMPILING=1 ./make.bash
Building Go cmd/dist using .../per-package/host-go/host/lib/go-1.19.11. (go1.19.11 linux/amd64)
go tool dist: cannot invoke C compiler [".../per-package/host-go/host/bin/arm-linux-gcc"]: fork/exec .../per-package/host-go/host/bin/arm-linux-gcc: no such file or directory
Go needs a system C compiler for use with cgo.
To set a C compiler, set CC=the-compiler.
To disable cgo, set CGO_ENABLED=0.
This happens systematically with PPD, and happens without PPD when
host-go is explicitly built (by running: "make host-go").
Since only CGO support needs to compile C files, only add the toolchain
dependency in that case.
When the target is not supported by go, then there is obviously no need
to depend on the toolchain (even if we unconditionally enable cgo
support in only-for-the-host host-go).
Signed-off-by: Christian Stewart <christian@aperture.us>
[yann.morin@orange.com:
- only add the toolchain dependency for target cgo
- reword commit log
]
Signed-off-by: Yann E. MORIN <yann.morin@orange.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Anisse Astier <anisse@astier.eu>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit 1a44f9242c)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Fixes CVE-2023-39323: Line directives ("//line") can be used to bypass the
restrictions on "//go:cgo_" directives, allowing blocked linker and compiler
flags to be passed during compilation. This can result in unexpected
execution of arbitrary code when running "go build".
go1.20.9 (released 2023-10-05) includes one security fixes to the cmd/go
package, as well as bug fixes to the go command and the linker.
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit 1da113559f)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
go1.20.8 (released 2023-09-06) includes two security fixes to the html/template
package, as well as bug fixes to the compiler, the go command, the runtime, and
the crypto/tls, go/types, net/http, and path/filepath packages.
CVE-2023-39318: html/template: improper handling of HTML-like comments within script contexts
CVE-2023-39319: html/template: improper handling of special tags within script contexts
CVE-2023-39321: crypto/tls: panic when processing post-handshake message on QUIC connections
https://go.dev/doc/devel/release#go1.20.0
Signed-off-by: Christian Stewart <christian@aperture.us>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit 2b72f1d167)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
go1.20.7 (released 2023-08-01) includes a security fix to the crypto/tls
package, as well as bug fixes to the assembler and the compiler.
Fixes CVE-2023-29409: restrict RSA keys in certificates to <= 8192 bits
Extremely large RSA keys in certificate chains can cause a client/server to
expend significant CPU time verifying signatures. Limit this by restricting the
size of RSA keys transmitted during handshakes to <= 8192 bits.
Based on a survey of publicly trusted RSA keys, there are currently only three
certificates in circulation with keys larger than this, and all three appear to
be test certificates that are not actively deployed. It is possible there are
larger keys in use in private PKIs, but we target the web PKI, so causing
breakage here in the interests of increasing the default safety of users of
crypto/tls seems reasonable.
https://go.dev/doc/devel/release#go1.20.7
Signed-off-by: Christian Stewart <christian@aperture.us>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
(cherry picked from commit ee7179497b)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
While it's not strictly necessary to update go-bootstrap-stage2, we can
continue to update the package to the latest 1.19.x until Go 1.19.x is
discontinued, at which point we can leave it at the final 1.19.x
version.
This package is only used when bootstrapping host-go.
go1.19.11 includes a security fix to the net/http package, as well as
bug fixes to cgo, the cover tool, the go command, the runtime, and the
go/printer package.
https://go.dev/doc/devel/release#go1.19.minor
Signed-off-by: Christian Stewart <christian@aperture.us>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
(cherry picked from commit df5fd7dec4)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Adjust comments in the Go package to improve clarity:
Config.in.host:
- Add comment mentioning list of supported architectures.
go.mk:
- Reword comment re: copying src/ to host/
- the previously linked issue is not relevant.
- instead: mention that src/ is needed for stdlib.
- Adjust comment re: adjusting file timestamps.
- mention this is needed to avoid rebuilding stdlib
Signed-off-by: Christian Stewart <christian@aperture.us>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
(cherry picked from commit 651835dbd7)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
CVE-2023-4911: If a tunable of the form NAME=NAME=VAL is passed in the
environment of a setuid program and NAME is valid, it may result in a
buffer overflow, which could be exploited to achieve escalated
privileges. This flaw was introduced in glibc 2.34.
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Fixes CVE_2023-5217: Heap buffer overflow in vp8 encoding in libvpx in
Google Chrome prior to 117.0.5938.132 and libvpx 1.13.1 allowed a remote
attacker to potentially exploit heap corruption via a crafted HTML page.
https://www.openwall.com/lists/oss-security/2023/09/28/5
Signed-off-by: Bernd Kuhls <bernd@kuhls.net>
[Peter: extend commit message, add _IGNORE_CVES]
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit e99999d7cb)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
- Fix CVE-2023-35852: In Suricata before 6.0.13 (when there is an
adversary who controls an external source of rules), a dataset
filename, that comes from a rule, may trigger absolute or relative
directory traversal, and lead to write access to a local filesystem.
This is addressed in 6.0.13 by requiring allow-absolute-filenames and
allow-write (in the datasets rules configuration section) if an
installation requires traversal/writing in this situation.
- Fix CVE-2023-35853: In Suricata before 6.0.13, an adversary who
controls an external source of Lua rules may be able to execute Lua
code. This is addressed in 6.0.13 by disabling Lua unless allow-rules
is true in the security lua configuration section.
- Drop first patch (not needed since
c8a3aa608e)
https://github.com/OISF/suricata/blob/suricata-6.0.14/ChangeLog
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
(cherry picked from commit ce17f93e82)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Fix CVE-2023-38633: A directory traversal problem in the URL decoder of
librsvg before 2.56.3 could be used by local or remote attackers to
disclose files (on the local filesystem outside of the expected area),
as demonstrated by href=".?../../../../../../../../../../etc/passwd" in
an xi:include element.
https://gitlab.gnome.org/GNOME/librsvg/-/blob/2.50.9/NEWS
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
(cherry picked from commit e7988c7060)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The COPYING also contains a BSD-3-Clause license. The BSD-3-Clause
applies to "manual page unifdef.1 and the portability support code in
the FreeBSD subdirectory". The BSD-2-Clause applies to everything else.
Signed-off-by: Brandon Maier <brandon.maier@collins.com>
Reviewed-by: Thomas Devoogdt <thomas@devoogdt.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
(cherry picked from commit d8cea23ce4)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Fixes the following security issues:
CVE-2023-4527: If the system is configured in no-aaaa mode via
/etc/resolv.conf, getaddrinfo is called for the AF_UNSPEC address
family, and a DNS response is received over TCP that is larger than
2048 bytes, getaddrinfo may potentially disclose stack contents via
the returned address data, or crash.
CVE-2023-4806: When an NSS plugin only implements the
_gethostbyname2_r and _getcanonname_r callbacks, getaddrinfo could use
memory that was freed during buffer resizing, potentially causing a
crash or read or write to arbitrary memory.
CVE-2023-5156: The fix for CVE-2023-4806 introduced a memory leak when
an application calls getaddrinfo for AF_INET6 with AI_CANONNAME,
AI_ALL and AI_V4MAPPED flags set.
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Reviewed-by: Romain Naour <romain.naour@smile.fr>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Commit 84c24ab1b5 (package/nodejs: fix parallel build) made use of
BR2_JLEVEL to set the number of jobs nodejs should use instead of using
the number of CPUs (+2).
However, BR2_JLEVEL can be set to 0 by the user, to let Buildroot detect
the number of CPUs (+1), and stores it in PARALLEL_JOBS, and leaves
BR2_JLEVEL untouched, so 0.
Thus, we can end up spawning a build by passing -j0 to ninja, which it
interprets as "no -limit yolo" and does not limit the number oj jobs it
spawns, which usually ends up in an OOM somewhere...
Fix this by using PARALLEL_JOBS.
Reported-by: Cédric & Co
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
(cherry picked from commit d574e2a4f4)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Fix CVE-2023-3341: The code that processes control channel messages sent
to `named` calls certain functions recursively during packet parsing.
Recursion depth is only limited by the maximum accepted packet size;
depending on the environment, this may cause the packet-parsing code to
run out of available stack memory, causing `named` to terminate
unexpectedly. Since each incoming control channel message is fully
parsed before its contents are authenticated, exploiting this flaw does
not require the attacker to hold a valid RNDC key; only network access
to the control channel's configured TCP port is necessary. This issue
affects BIND 9 versions 9.2.0 through 9.16.43, 9.18.0 through 9.18.18,
9.19.0 through 9.19.16, 9.9.3-S1 through 9.16.43-S1, and 9.18.0-S1
through 9.18.18-S1.
https://ftp.isc.org/isc/bind9/9.16.44/CHANGES
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
(cherry picked from commit 54f6e1f81f)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
When nodejs is build, a qemu wrapper script is used to execute some
programs built for the target in user-mode emulation. However, when the
target and build machines are similar (e.g. x86_74), running those
programs fails, with errors such as:
cd ../../tools/v8_gypfiles; python ../../deps/v8/tools/run.py ../../out/Release/v8-qemu-wrapper ../../out/Release/bytecode_builtins_list_generator ../../out/Release/obj.host/gen/generate-bytecode-output-root/builtins-generated/bytecodes-builtins-list.h
../../out/Release/bytecode_builtins_list_generator: /lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.29' not found (required by ../../out/Release/bytecode_builtins_list_generator)
../../out/Release/bytecode_builtins_list_generator: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.33' not found (required by ../../out/Release/bytecode_builtins_list_generator)
../../out/Release/bytecode_builtins_list_generator: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.32' not found (required by ../../out/Release/bytecode_builtins_list_generator)
../../out/Release/bytecode_builtins_list_generator: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.34' not found (required by ../../out/Release/bytecode_builtins_list_generator)
Return code is 1
So the question is: why the heck does Qemu use the host C library?
To answer this question, we first have to look at how the -L option of
Qemu is implemented. This option is documented as such:
-L path QEMU_LD_PREFIX set the elf interpreter prefix to 'path'
The v8-qemu-wrapper script makes this option point to $(STAGING_DIR),
so that the ELF interpreter used is the one in $(STAGING_DIR).
However, contrary to what the option documentation says, this option
does much more than setting the ELF interpreter prefix: it is going to
affect how *all* system calls manipulating files (open, etc.) are
going to work.
When this option is passed, the function init_paths() in
https://git.qemu.org/?p=qemu.git;a=blob;f=util/path.c is called at
initialization time, and essentially its sets the global "base"
variable to point to the directory passed as -L argument.
Then, for every single syscall that manipulates a path, this path will
be passed through the path() function in the same file. This function
will first attempt to resolve the path with "base" as a prefix, and if
not, return the unprefixed path.
After adding some traces into this function, I was able to understand
what happens:
(1) -L$(STAGING_DIR) is passed, causing "base" to point to
$(STAGING_DIR)
(2) The target ELF interpreter from $(STAGING_DIR) is properly invoked
(3) When this ELF interpreter then resolves the libc.so.6 library, it
first looks for /etc/ld.so.cache.
(4) Qemu first looks for /etc/ld.so.cache with the -L prefix, i.e
$(STAGING_DIR)/etc/ld.so.cache, but it does not exist. So, the Qemu
system call emulation falls back to /etc/ld.so.cache, which means
the target ELF interpreter reads the /etc/ld.so.cache of the host
system.
(5) This /etc/ld.so.cache of the host system says that libc.so.6 is in
/lib/x86_64-linux-gnu/
(6) The target ELF interpreter therefore tries to use
/lib/x86_64-linux-gnu/libc.so.6. The Qemu system call emulation
first tries $(STAGING_DIR)/lib/x86_64-linux-gnu/libc.so.6, but
this library does not exist (it is in
$(STAGING_DIR)/lib/libc.so.6), so the Qemu system call emulation
falls back to /lib/x86_64-linux-gnu/libc.so.6 of the host system,
which exist... but is too old compared to the target C library.
Indeed, results from ld.so.cache take precedence over the simple
resolution of library paths in /usr/lib and /lib.
We see 3 possible ideas to resolve this problem:
(A) Change the behavior of Qemu to not fallback to unprefixed paths:
when -L is passed, all path-related system calls should see the
paths prefixed by the -L option.
Issue with this is that this change is unlikely to get accepted by
Qemu upstream. And there might be some side effects we have not
really identified.
(B) Create an empty $(STAGING_DIR)/etc/ld.so.cache. We have tested
this solution and it works: it gets used instead of the host
/etc/ld.so.cache. Because $(STAGING_DIR)/etc/ld.so.cache is empty,
there's no libc.so.6 match, so the target ELF interpreter goes
through its normal library location resolution logic, which falls
back to trying in /usr/lib and /lib, which works as those paths
ends up being prefixed with $(STAGING_DIR) by Qemu.
(C) Pass LD_LIBRARY_PATH pointing to $(STAGING_DIR)/lib and
$(STAGING_DIR)/usr/lib in the Qemu wrapper. This works because
LD_LIBRARY_PATH paths have precedence over paths given by
ld.so.cache.
This is the solution already used by the GOI qemu wrapper in
package/gobject-introspection/g-ir-scanner-qemuwrapper.in.
We chose to go with the third option, because it has been proven to work
for the GOI wrapper, and has been reported to solve #14366. Even though
the first option would be the best, it is also the one that has the
least chances to land any time soon (if ever); the second has not been
exercised, and the impact is not fully understood either (e.g what about
non-glibc toolchains?).
Fixes: #14366
Signed-off-by: Jens Maus <mail@jens-maus.de>
[yann.morin.1998@free.fr:
- add whole analsys done by Thomas in:
https://lore.kernel.org/buildroot/20221031213926.50d3c778@windsurf/
]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
(cherry picked from commit 278d1db56b)
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
