A stray comma was being appended to the last package version dependency,
causing it to be missed for ABI version checks
Signed-off-by: Felix Fietkau <nbd@nbd.name>
The version string generated for ARM Trusted-Firmware-A was stated as
"OpenWRT". Fix that by changing it to the exact spelling "OpenWrt"
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Multiple sources are hosted on OpenWrts source server only. The source
URLs to point to the server vary based on different epochs in OpenWrts
history.
Replace all by @OPENWRT which is an "empty" mirror, therefore using the
fallback servers sources.cdn.openwrt.org and sources.openwrt.org.
Signed-off-by: Paul Spooren <mail@aparcar.org>
The license folder is a core part of OpenWrt and all GPL-2.0 licensed.
Use SPDX license tags to allow machines to check licenses.
Signed-off-by: Paul Spooren <mail@aparcar.org>
[rebase, keep some Copyright lines, sharpen commit message]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
Ran update_kernel.sh in a fresh clone without any existing toolchains.
Removed upstreamed patches:
imx6: 303-ARM-dts-imx6qdl-gw52xx-fix-duplicate-regulator-namin.patch
Build system: x86_64
Build-tested: ipq806x/R7800, bcm27xx/bcm2711
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
All modification made by update_kernel.sh in a fresh clone without
existing toolchains.
Build-tested: bcm27xx/bcm2711, ipq806x/R7800,
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
Upon boot it now prints:
NOTICE: BL1: v2.4(release):OpenWRT v2.4-1 (espressobin-v3-v5-1gb-2cs) (Marvell-devel-18.12.0)
Signed-off-by: Andre Heider <a.heider@gmail.com>
A header used in ELECOM WRC-300GHBK2-I and WRC-1750GHBK2-I/C is also
used in ELECOM WRC-2533GHBK-I, so split the code to generate the header
and move it to image-commands.mk to use from ramips target.
Signed-off-by: INAGAKI Hiroshi <musashino.open@gmail.com>
Reviewed-by: Sungbo Eo <mans0n@gorani.run>
/lib/functions.sh was executable for no obvious reason and its
execute property was even checked in package-ipkg.mk just to
source it afterwards.
Remove the execute bit and shebang as this is clearly a library.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
Reviewed-by: Philip Prindeville <philipp@redfish-solutions.com>
The majority of our targets provide a default value for the variable
SUPPORTED_DEVICES, which is used in images to check against the
compatible on a running device:
SUPPORTED_DEVICES := $(subst _,$(comma),$(1))
At the moment, this is implemented in the Device/Default block of
the individual targets or even subtargets. However, since we
standardized device names and compatible in the recent past, almost
all targets are following the same scheme now:
device/image name: vendor_model
compatible: vendor,model
The equal redundant definitions are a symptom of this process.
Consequently, this patch moves the definition to image.mk making it
a global default. For the few targets not using the scheme above,
SUPPORTED_DEVICES will be defined to a different value in
Device/Default anyway, overwriting the default. In other words:
This change is supposed to be cosmetic.
This can be used as a global measure to get the current compatible
with: $(firstword $(SUPPORTED_DEVICES))
(Though this is not precisely an achievement of this commit.)
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
All modification made by update_kernel.sh in a fresh clone without
existing toolchains.
Build system: x86_64
Build-tested: ipq806x/R7800, bcm27xx/bcm2711
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
Tested-by: Curtis Deptuck <curtdept@me.com> [x86/64]
All modification made by update_kernel.sh in a fresh clone without
existing toolchains.
Build system: x86_64
Build-tested: ipq806x/R7800, bcm27xx/bcm2711
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
Tested-by: Curtis Deptuck <curtdept@me.com> [x86/64]
Kerning seems to be very off-putting for some people so the logo
designer thankfully updated guidelines to something which is now
considered final.
Signed-off-by: Paul Spooren <mail@aparcar.org>
All modification made by update_kernel.sh in a fresh clone without
existing toolchains.
Build system: x86_64
Build-tested: ipq806x/R7800, bcm27xx/bcm2711
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
Tested-by: Curtis Deptuck <curtdept@me.com> [x86/64]
Currently `qemu-img` is used to convert raw x86 images to VDI and VMDK
images, used for virtual machines.
Having `qemu-img` in tree requires us to maintain an ancient version of
`qemu-utils`, which recently required extra work to compile with newer
compiler version.
This commit prints a warning message in case `qemu-img` is missing.
As a next step the in-tree version of `qemu-img` can be removed.
Signed-off-by: Paul Spooren <mail@aparcar.org>
Linux 5.9 introduces support for ZSTD ramdisk and initrd compression,
make sure we enable/disable the relevant options when building an
initramfs enabled kernel.
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
When we use an external kernel tree which may not have been fully
cleaned, there may be user_headers left which do not match the target
architecture, leading to build failures for packages that do an explicit
inclusion of user_headers (such as iproute2 or iptables). Make sure we
delete them while preparing the directory.
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
This keeps the configuration, like the size of the cache, and the
statistics intact. Move the removal of the cache directory to the
distclean target, but only delete the .ccache directory inside of our
build tree, as we should not mess with a user-configured external ccache
directory this destructively.
Signed-off-by: Sven Wegener <sven.wegener@stealer.net>
clang's gcc emulation does the right thing with -print-file-name now,
drop the wrapper
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
`which` utility is not shipped by default for example on recent Arch
Linux and then any steps relying on its presence fails, like for example
following Python3 prereq build check:
$ python3 --version
Python 3.9.1
$ make
/bin/sh: line 1: which: command not found
/bin/sh: line 1: which: command not found
/bin/sh: line 1: which: command not found
...
Checking 'python3'... failed.
...
Fix this by switching to Bash builtin `command` which should provide
same functionality.
Fixes: FS#3525
Signed-off-by: Petr Štetiar <ynezz@true.cz>
With commit 2ca084cc ("build: improve ccache support") these variables
are being set globally and we don't need them for specific targets.
Signed-off-by: Sven Wegener <sven.wegener@stealer.net>
While parsing the nm output, we need to account for the fact that 64-bit kernels
have 64-bit wide addresses. While at it, replace the grep | sed combo with a
single awk invocation and a stronger regex.
Fixes: 2ef0acc5fc "kernel-build: fix
STRIP_KERNEL_EXPORTS for recent kernels"
Signed-off-by: Rui Salvaterra <rsalvaterra@gmail.com>
Multiple prereq checks are only required within the build system but not
for the ImageBuilder. These checks are excluded by using ifndef IB.
This commit merges the three ifndef IB blocks together.
Signed-off-by: Paul Spooren <mail@aparcar.org>
All modifications made by update_kernel.sh run in a fresh clone
without any existing toolchains.
Build system: x86_64
Build-tested: ipq806x/R7800, ath79/generic, bcm27xx/bcm2711
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
Some images are created using different filesystems, most popular
squashfs and ext4. To allow downstream projects to distinguesh between
those, add the `filesystem` information to created json files.
Signed-off-by: Paul Spooren <mail@aparcar.org>
Removed since included upstream and could be reverse-applied by quilt:
backport-5.4/315-v5.10-usbnet-ipeth-fix-connectivity-with-ios-14.patch
Remaining modifications made by update_kernel.sh
Build system: x86_64
Build-tested: ipq806x/R7800, ath79/generic, bcm27xx/bcm2711
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
Tested-by: Curtis Deptuck <curtdept@me.com> [build/run x86_64]
It was observed that the MD5 would not change after source files had been
modified, looking deeper into the build process it was discovered that
find_md5 build function makes a list of the files being built and then
passes the list to a summing utility on stdin. The resultant MD5 is of
the file list, not the contents of the files.
The MD5 would change if the ordering of the list changed, or items were
removed or deleted.
The proposed fix is to add the modification time after the filename and
then sort the list to prevent find returning files in a different order
falsely re-triggering a rebuild. The MD5 will now change when a file is
modified or files are added/removed from the list.
Using 'T@' to show time in epoch for timezone independent behaviour.
Signed-off-by: John Beckett <john.beckett@net2edge.com>
This is a neat project, but offers no benefit to OpenWrt. The initial
reason for it was to be a replacement for libstdcpp as it is smaller
and lacks compatibility for C++98. Unfortunately, compiling several
packages with it results in larger ipk sizes.
While not a member of the packages feed, this will be moved to
packages-abandoned to keep it somewhere.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
On non-GNU systems, zcat often does not handle gzip decompression.
Use gzip -dc like the regular unpack command
Signed-off-by: Felix Fietkau <nbd@nbd.name>
All modifications made by update_kernel.sh/no human intervention needed
Build system: x86_64
Build-tested: ipq806x/R7800, ath79/generic, bcm27xx/bcm2711
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
[another refresh]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
The include/trusted-firmware-a.mk file is based on the
include/u-boot.mk file and should be used to build a Trusted Firmware-A
(TFA) which was previously named Arm trusted firmware.
This is useful for targets where the TFA is board specific like for
Marvell SoCs and probably also NXP Layerscape SoCs.
This also makes use of this abstraction in the
arm-trusted-firmware-mvebu package to build board specific ATF binaries.
The ATF binaries will be automatically activated and build when the
board is selected in the normal build or all boards are selected. This
should also activate the build when build bot creates images.
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Some Git hoster (e.g. sr.ht) disable hosting of svg images (xml) to
avoid XSS attacks. To show the logo correctly on all code hosters use a
"safe" PNG image.
Also move logo(s) to include/ folder to lower autocomplete churn with
the `logs/` folder. While at it, replace absolute logo path and make it
relative, as this may break other code hosters as well.
Signed-off-by: Paul Spooren <mail@aparcar.org>
Commit f98878e4c1 ("cmake.mk: set C/CXX compiler for host builds as
well") has introduced regression as it didn't taken usage of ccache into
the account so fix it by handling ccache use cases as well.
In order to get this working we need to export HOSTCXX_NOCACHE in
rules.mk as well.
Fixes: f98878e4c1 ("cmake.mk: set C/CXX compiler for host builds as well")
Reported-by: Ansuel Smith <ansuelsmth@gmail.com>
Tested-by: Ansuel Smith <ansuelsmth@gmail.com>
Signed-off-by: Petr Štetiar <ynezz@true.cz>
It is impossible to locate package that failed the build just from log
once more build is run in parallel (that is more than one make job). The
only way is to scout log files for failed package going back trough log.
This change makes it so error is printed for package that failed every
time.
Signed-off-by: Karel Kočí <karel.koci@nic.cz>
Without this, cmake will use whatever CC/CXX is set to, which could be
clang. In that case, at least libjson-c/host will fail to compile.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Currently it's assumed, that already downloaded tarballs are always
fine, so no checksum checking is performed and the tarball is used even
if it might be corrupted.
From now on, we're going to always check the downloaded tarballs before
considering them valid.
Steps to reproduce:
1. Remove cached tarball
rm dl/libubox-2020-08-06-9e52171d.tar.xz
2. Download valid tarball again
make package/libubox/download
3. Invalidate the tarball
sed -i 's/PKG_MIRROR_HASH:=../PKG_MIRROR_HASH:=ff/' package/libs/libubox/Makefile
4. Now compile with corrupt tarball source
make package/libubox/{clean,compile}
Signed-off-by: Petr Štetiar <ynezz@true.cz>
When setting the option IPK_FILES_CHECKSUMS the build system stores
checksums of all package file as metadata. In combination with pkg_check
this allows to see if a package is broken, e.g. caused by bad flash.
To create those checksums the tool `sha256sum` were used while the rest
of OpenWrt uses `mkhash`, a small & fast implementation of sha256. As
the build system does not check the existence of `sha256sum` and the
stderr output is moved to /dev/null, a situation where the option is
enabled but no actual checksum are created may occur.
Instead of adding `sha256sum` as a requirement, this replaces it with
`mkhash sha256` and adapts the `sed` pipe command to fit spacing.
CC: Xu Wang <xwang1498@gmx.com>
CC: Michal Hrusecky <Michal@Hrusecky.net>
Signed-off-by: Paul Spooren <mail@aparcar.org>
Removed since could be reverse-applied by quilt and found to be
included upstream:
backport-5.4/789-net-usb-qmi_wwan-Set-DTR-quirk-for-MR400.patch
All modifications made by update_kernel.sh
Build system: x86_64
Build-tested: ipq806x/R7800, bcm27xx/bcm2711, ath79/generic
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
Tested-by: Curtis Deptuck <curtdept@me.com> [x86_64 build/run]
Allow a device recipe to specify a custom UIMAGE_MAGIC value, as used by
OpenWrt's -M flag for mkimage. This allows to automatically customize
the magic bytes in all calls to Build/uImage for this device, similar to
the behaviour of UIMAGE_NAME. Since the -M argument is inserted before
the user arguments, it can be overriden.
The following example would use 0x87654321 for the KERNEL image, but
0x12345678 for the KERNEL_INITRAMFS image:
define Device/MyDevice
UIMAGE_MAGIC := 0x87654321
KERNEL := ... | uImage lzma
KERNEL_INITRAMFS := ... | uImage lzma -M 0x12345678
...
endef
Fixes: df8e6be59a ("rtl838x: add new architecture")
[UIMAGE_MAGIC was not declared as a device variable]
Signed-off-by: Sander Vanheule <sander@svanheule.net>
[rebase, improve formatting of "Fixes"]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
For some build recipes, the argument to Build/uImage is used to sneak in
extra arguments for mkimage, whereas this appears to have been intended
to specificy the compression method only.
Use the first provided word for -C to be backwards compatible with
current calls to Build/uImage. Use the rest of the call arguments to
override the provided defaults. Only the input file name (-d) and the
output file name cannot overriden.
Signed-off-by: Sander Vanheule <sander@svanheule.net>
Manually rebased patches:
ath79/patches-5.4/910-unaligned_access_hacks.patch
bcm27xx/patches-5.4/950-0135-spi-spi-bcm2835-Disable-forced-software-CS.patch
bcm27xx/patches-5.4/950-0414-SQUASH-Fix-spi-driver-compiler-warnings.patch
ipq806x/patches-5.4/093-4-v5.8-ipq806x-PCI-qcom-Use-bulk-clk-api-and-assert-on-error.patch
Removed since could be reverse-applied by quilt and found to be included upstream:
ipq806x/patches-5.4/096-PCI-qcom-Make-sure-PCIe-is-reset-before-init-for-rev.patch
All modifications made by update_kernel.sh
Build system: x86_64
Build-tested: ipq806x/R7800, ath79/generic, bcm27xx/bcm2711
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
[refresh altered targets after rebase]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
Commit 5d76065 moved the creation of the symvers directory to
include/kernel-build.mk. This is fine when building from scratch. But
when unpacking an SDK the directory doesn't exist and because the kernel
won't be built (again) this directory will not be created by the build
system, causing build failure if make tries to copy files into it.
This moves the creation of the symvers directory back into
include/kernel.mk so that the directory is created in any case.
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Apple ships a broken make version with the Xcode command line tools.
Homebrew installs make as gmake by default in order to not collide with
Apple's version.
Exit with an error if the broken one is used accidentally
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Define wildcard patterns for filtering in target/linux/generic/config-filter
Preparation for supporting newer kernels
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This replaces the previous (deprecated) method of collecting symvers data
in $(PKG_BUILD_DIR)/Module.symvers, which does not work on newer kernels
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This change adds the configuration option to build and include
the nft_queue kernel module, which allows traffic to be queued up
to userspace from an nftables rule
Tested-by: Sébastien Delafond sdelafond@gmail.com
Signed-off-by: Brett Mastbergen <bmastbergen@untangle.com>
The source date epoch is the only reproducible date close to the actual
build date. It can be used for tooling like the firmware wizard to show
the image age.
Signed-off-by: Paul Spooren <mail@aparcar.org>
Manually rebased patches:
bcm27xx:
patches-5.4/950-0267-xhci-add-quirk-for-host-controllers-that-don-t-updat.patch
bcm53xx:
patches-5.4/180-usb-xhci-add-support-for-performing-fake-doorbell.patch
layerscape:
patches-5.4/802-can-0025-can-flexcan-add-LPSR-mode-support-for-i.MX7D.patch
patches-5.4/808-i2c-0002-MLK-10893-i2c-imx-add-irqf_no_suspend.patch
patches-5.4/820-usb-0016-MLK-16735-usb-host-add-XHCI_CDNS_HOST-flag.patch
Removed since could be reverse-applied by quilt:
mediatek:
patches-5.4/0700-arm-dts-mt7623-add-missing-pause-for-switchport.patch
All modifications made by update_kernel.sh
Build system: x86_64
Build-tested: ipq806x/R7800, ath79/generic, bcm27xx/bcm2711, x86_64
Run-tested: ipq806x/R7800, x86_64
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
Tested-by: Curtis Deptuck <curtdept@me.com> [x86_64]
Rebase of 802-can-0025-can-flexcan-add-LPSR-mode-support-for-i.MX7D.patch
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
This removes switches dependent on kernel version 4.19 as well as
several packages/modules selected only for that version.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
We use 5.4 on all targets by default, and 4.19 has never been released
in a stable version. There is no reason to keep it.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
LegacyDevice is not used anymore in our tree, so let's drop it
together with the relevant definitions and recipes.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
The definitions in image-commands.mk seem to have no particular
order. Sort them alphabetically to make it easier to actually
find anything there. No other changes made beyond moving entire
blocks.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
Build/combined-image is only used in ath25 target, and that defines
its own version. Thus, drop the unused definition in image-commands.mk.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
This drops unused legacy recipes Image/Build/SysupgradeNAND and
Image/Build/UbinizeImage.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
This reverts commit 7f94e2afcf.
Package kmod-nft-core is missing dependencies for the following libraries:
nft_reject.ko
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Nightly snapshot builds of OpenWrt change their kernels versions
frequently and lose thereby compatibility to kmods from the upstream
target specific packages feed.
To allow opkg to install packages over multiple days a kmod archive is
offered at $target/$subtarget/kmods/$kernelversion and added as a feed
to created snapshot images via a buildbot step[1].
Instead of using a buildstep add the kmod feed directly via
FeedSourcesAppend to be included in the ImageBuilder repositories.conf
as well. This is conditionally only done for SNAPSHOT builds and when
running as BUILDBOT. Releases are unaffected as they don't include
kernel version changes and local builds may use different kernel
versions or magics than available upstream.
This commit allows in a future step to ship ImageBuilders without a
locally stored kmod archive.
[1]: https://git.openwrt.org/?p=buildbot.git;a=blob;f=phase1/master.cfg;h=3ba7a1606e89b095b10555e703ea96e93295deec;hb=HEAD#l1025
Signed-off-by: Paul Spooren <mail@aparcar.org>
The mips kernel vmlinux image supports adding an empty ELF section
for DTB to be later inserted into with MIPS_ELF_APPENDED_DTB.
This ELF + inserted DTB image can then be directly booted on some
devices.
Example usage:
image/subtarget.mk:
KERNEL_NAME := vmlinux.elf
KERNEL_INITRAMFS_NAME := vmlinux-initramfs.elf
KERNEL := kernel-bin | append-dtb-elf
On mt7621 memory size needed to be manually specified.
Signed-off-by: John Thomson <git@johnthomson.fastmail.com.au>
Kernel commit 1ac89d20150e ("netfilter: nat: merge nf_nat_redirect into
nf_nat") made the redirect module part of the nat core and changed the
CONFIG_NF_NAT_REDIRECT option to a boolean, without prompt, affecting
kernel 4.18 onwards. CONFIG_NF_NAT_REDIRECT now can only be selected by
CONFIG_NFT_REDIR or NETFILTER_XT_TARGET_REDIRECT
Fixes: FS#2476
Ref: https://bugs.openwrt.org/index.php?do=details&task_id=2476
Fixes: FS#2990 (partial)
Ref: https://bugs.openwrt.org/index.php?do=details&task_id=2990
Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
[note that the option has no prompt and can only be selected by other
kconfig options]
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
It was removed in upstream linux commit faec18db ("netfilter: nat:
remove l4proto->manip_pkt"). This happened since linux 5.0
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
Kernel commit 22fc4c4c9fd6 ("netfilter: conntrack: gre: switch module to
be built-in") moved the CT GRE code into the core nf_conntrack.ko module
and changed the CONFIG_NF_CT_PROTO_GRE option to boolean for kernel 5.1
and onwards.
CONFIG_NF_CT_PROTO_GRE at the moment has no prompt and can only be
selected by NF_CONNTRACK_PPTP
Fixes: FS#2990 (partial)
Ref: https://bugs.openwrt.org/index.php?do=details&task_id=2990
Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
[note that the option now can not be enabled on its own]
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
The upstream linux commit is 3bf195ae ("netfilter: nat: merge
nf_nat_ipv4,6 into nat core"). It was included since linux 5.1
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
CONFIG_IP_NF_TARGET_REDIRECT is a compat option since upstream commit
2cbc78a2 ("netfilter: combine ipt_REDIRECT and ip6t_REDIRECT"). That
happened since linux 3.10
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
CONFIG_IP_NF_TARGET_MASQUERADE and its counterpart
CONFIG_IP6_NF_TARGET_MASQUERADE are "backwards-compat option for the
user's convenience"
Related commit d22c1755 ("netfilter: fix NAT packaging with kernels
5.2+")
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
Add procd-ujail and procd-seccomp to DEFAULT_PACKAGES if not building
for space-constraint (FEATURES:=small_flash) targets.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Rather than unconditionally adding busybox and procd to the set of
default packages, add busybox-selinux and procd-selinux in case
CONFIG_SELINUX is set.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
All modifications made by update_kernel.sh
Build system: x86_64
Build-tested: ipq806x/R7800, ath79/generic, bcm27xx/bcm2711
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
* add d-link_dgs-1210-10p support
* make sure mips16 is disabled
* add a generic sub target
* add proper cflags
Signed-off-by: John Crispin <john@phrozen.org>
This reverts commit ef7c34c1d1.
The commit seems to break all buildbots with messages like:
/builder/shared-workdir/build/include/toplevel.mk:15:
/builder/shared-workdir/build/include/toplevel-vars.mk: No such file or directory
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
version_abbrev uses $(shell) and the ?= is causing make to run the command
over and over again, causing a significant build slowdown
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Usage of current R1 ISA is inconsistent with the MIPS32 subtarget, little
used and has limited utility for testing.
Many distros target a minimum R2 ISA. Debian MIPS 32-bit/64-bit ports all
use MIPS R2 ISA since Stretch, for example. Fedora's MIPS arch also targets
the R2 ISA for 32-bit/64-bit.
Widely used MIPS64 platforms like Octeon are based on the MIPS R2 ISA or
later, and benefit from having a compatible test platform in OpenWRT.
While Linux does support MIPS64 R1 targets, its usefulness for development
and testing is limited. As an example, the modern Linux eBPF JIT requires
a MIPS R2 ISA or later.
Signed-off-by: Tony Ambardar <itugrok@yahoo.com>
[Refresh config and fix README]
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
All modifications made by update_kernel.sh
Build system: x86_64
Build-tested: ipq806x/R7800, ath79/generic, bcm27xx/bcm2711
Run-tested: ipq806x/R7800
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
The buildroot and SDK both require the compilers (gcc, g++) to be
installed on the host system, however the ImageBuilder uses precompiled
binaries.
This patch changes the prerequirements checks to skip the checking for
the compilers if running as ImageBuilder. A similar change has been
made for libncurses-dev in 4a1a58a3e2.
Signed-off-by: Sven Roederer <devel-sven@geroedel.de>
Acked-by: Paul Spooren <mail@aparcar.org>
Seemingly unneeded based on new upstream code so manually deleted:
layerscape:
820-usb-0007-usb-dwc3-gadget-increase-timeout-value-for-send-ep-c.patch
Manually merged:
generic-hack:
251-sound_kconfig.patch
All other modifications made by update_kernel.sh
Build system: x86_64
Build-tested: ipq806x/R7800, ath79/generic, bcm27xx/bcm2711
Run-tested: ipq806x/R7800, lantiq/Easybox 904 xDSL
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
[add lantiq test report, minor commit message clarification]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
The former nft_chain_nat_ipv4 and nft_chain_nat_ipv6 modules have been merged
into a common nft_chain_nat module starting with Linux 5.1.
Ensure that this common module is shipped along with kmod-nft-nat on recent
kernels.
While we're at it, also apply version constraints to other nft modules that
have been merged into the core with newer kernels.
Ref: https://bugs.openwrt.org/index.php?do=details&task_id=2815#comment8016
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Instead of hardcoding 'targeted' policy, evaluate /etc/selinux/config
in rootfs to choose according to which policy files in the rootfs got
to be labeled.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
By installing policycoreutils to host/bin it is also available within
the ImageBuilder and SDK, allowing to correctly label both filesystems
and packages.
Signed-off-by: Paul Spooren <mail@aparcar.org>
All modifications made by update_kernel.sh
Build system: x86_64
Build-tested: ipq806x, ath79/generic, bcm72xx/bcm2711
Run-tested: ipq806x (R7800)
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
There are currently several variants of 'wpad' package but the 'iwinfo'
is included by default only if 'wpad', 'wpad-{basic*,mini}' or 'nas'
packages are included in {DEVICE,DEFAULT}_PACKAGES. Use 'wpad-*'
pattern to include 'iwinfo' with any 'wpad' variant.
Signed-off-by: Piotr Dymacz <pepe2k@gmail.com>
'setfiles' and others should be installed to $(STAGING_DIR_HOSTPKG)/bin
rather than $(...)/sbin which isn't in PATH.
Also using -Wl,-rpath to set library search location instead of setting
LD_LIBRARY_PATH when calling setfiles in image.mk.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Some bootloaders are really keen on just one special
fdt in a multi-image fit image. This is a problem, because
currently this is fixed to "fdt@1".
This patch introduces a new device variable:
DEVICE_FDT_NUM that allows to specify the right
fdt number.
If the value is absent "1" will be chosen.
Signed-off-by: Christian Lamparter <chunkeey@gmail.com>
All modifications made by update_kernel.sh
Build system: x86_64
Build-tested: ipq806x, lantiq/xrx200 and ath79/generic
Run-tested: ipq806x (R7800), lantiq (Easybox 904 xDSL)
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
[add test on lantiq]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
All modifications made by update_kernel.sh/no manual intervention needed
Run-tested: ipq806x (R7800), ath79 (Archer C7v5), x86/64
No dmesg regressions, everything appears functional
Signed-off-by: John Audia <graysky@archlinux.us>
[add run test from PR]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
This reverts commit 34cc2c9a99.
The reverted shell code is a very poor reimplementation of the existing
package-metadata.pl usergroup subcommand and the resulting file is not
used anymore, so drop this code.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
This file can be subsequently used to resolve symbolic user or group names
to their numeric IDs when packing ipk archives.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Multiple packages contain a USERID variable defining required user and
group for the package to run. With the recent addition of
"PKG_FILE_MODES" it is possible to define user and group of specific
files, replacing (possibly insecure) post-inst scripts. These modes are
set during build time and put directly into the packages.
To allow user and group names rather than the numeric values, a mapping
like `/etc/passwd` is required by the `ipkg-build` script, mapping names
defined in "PKG_FILE_MODES" to a numeric value, as the build system does
not create any users during build.
This commit adds a single line to the `prepare-tmpinfo` target, so that
everytime the feeds are updated the *passwd like* content of
`./tmp/userids` is updated.
Signed-off-by: Paul Spooren <mail@aparcar.org>
All modifications made by update_kernel.sh/no manual intervention needed
Build-tested: x86_64
Run-tested: ipq806x (R7800)
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
Currently the global variable PKG_FILE_MODES is used for all ipkg
creations. This works for Makefiles which output a single package, or
variants of a single package.
But if a Makefile outputs multiple packages that each contain different
files, setting PKG_FILE_MODES causes build failure when any of the files
in the variable do not exist in the folder that is currently being
packaged.
Example:
/openwrt/staging_dir/host/bin/fakeroot -l /openwrt/staging_dir/host/lib/libfakeroot.so -f /openwrt/staging_dir/host/bin/faked /openwrt/scripts/ipkg-build -m "/usr/lib/mariadb/plugin/auth_pam_tool_dir:root:376:0750" /openwrt/build_dir/target-mips_24kc_musl/mariadb-10.4.13/ipkg-mips_24kc/mariadb-server-plugin-disks /openwrt/bin/packages/mips_24kc/packages
+chown: cannot access '/openwrt/build_dir/target-mips_24kc_musl/mariadb-10.4.13/ipkg-mips_24kc/mariadb-server-plugin-disks//usr/lib/mariadb/plugin/auth_pam_tool_dir': No such file or directory
This commit changes the file mode handling a bit. The file mode can now
be set either globally via PKG_FILE_MODES (no behavior change) or on a
per-package basis via FILE_MODES. This way specific file modes can be
used for any particular package.
This behavior is already used for other OpenWrt variables, hence it is
familiar:
PKG_MAINTAINER vs MAINTAINER
PKG_SOURCE_SUBDIR vs SUBDIR
PKG_LICENSE vs LICENSE
...
Signed-off-by: Sebastian Kemper <sebastian_ml@gmx.net>
Manually merged:
hack-5.4
230-openwrt_lzma_options.patch
bcm27xx
950-0283-hid-usb-Add-device-quirks-for-Freeway-Airmouse-T3-an.patch
x86
011-tune_lzma_options.patch
Remove upstreamed patches in collaboration with Ansuel Smith:
ipq806x
093-1-v5.8-ipq806x-PCI-qcom-Add-missing-ipq806x-clocks-in-PCIe-driver.patch
093-2-v5.8-ipq806x-PCI-qcom-Change-duplicate-PCI-reset-to-phy-reset.patch
093-3-v5.8-ipq806x-PCI-qcom-Add-missing-reset-for-ipq806x.patch
All other modifications made by update_kernel.sh
Build-tested: bcm27xx/bcm2708, ipq806x, x86/64
Run-tested: ipq806x (R7800), x86/64
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
[update commit message/tested]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
Adding inline shell invocations in per-target variables causes them to be
executed over and over again, which causes a significant slowdown.
Fix this by evaluating it only once per package directory
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This removes switches dependent on kernel version 4.14 as well as
several packages/modules selected only for that version.
This also removes sched-cake-virtual, which is not required anymore
now that we have only one variant of cake.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
Using fakeroot without passing the paths to libfakeroot.sh and faked
causes havoc. Use the $(FAKEROOT) Make variable which includes them.
Fixes: 353ce2e521 ("build: ipkg-build use fakeroot with PKG_FILE_MODES")
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
The variable VERSION_REPO is used by opkg to download package(list)s.
Now that the default installation support encrypted HTTP opkg should
make use of it.
Suggested-by: Petr Štetiar <ynezz@true.cz>
Suggested-by: Baptiste Jonglez <baptiste@bitsofnetworks.org>
Signed-off-by: Paul Spooren <mail@aparcar.org>
Acked-by: Baptiste Jonglez <baptiste@bitsofnetworks.org>
The line of default packages became very long and it is easier to read
one package per line, therefore split it by newlines and sort it
alphabetically.
Signed-off-by: Paul Spooren <mail@aparcar.org>
To allow HTTPS usage on a router it requires both certificates
(ca-bundle) and a fitting libustream library (libustream-wolfssl)
By adding both, uclient-fetch and wget can connect to encrypted HTTP.
This allows opkg to update package lists in a more secure fashion.
Suggested-by: Petr Štetiar <ynezz@true.cz>
Suggested-by: Baptiste Jonglez <baptiste@bitsofnetworks.org>
Signed-off-by: Paul Spooren <mail@aparcar.org>
The usage of granular `SOURCE_DATE_EPOCH` for packages is an
incrementing integer which could be useful for downstream tooling,
therefore add it to the packages manifest.
Signed-off-by: Paul Spooren <mail@aparcar.org>
With the new `SOURCE` argument of `get_source_date_epoch` it is possible
to set package timestamps based on actual package changes rather thane
$TOPDIR changes.
This commit adds a new variable PKG_SOURCE_DATE_EPOCH which is used by
the `ipkg` build script. As a fallback the existing SOURCE_DATE_EPOCH is
used or as last resort the current time.
The redundant checks for `.git/` and `.svn/` are removed.
Signed-off-by: Paul Spooren <mail@aparcar.org>
The `ipkg-build` script converts a folder into a `opkg` installable
package. Until now it would use root:root for all packages and try to
preserve file modes.
This has the two drawbacks of packages want to add non-root files or add
SUID files, like the `sudo` package does.
To give more flexibility regarding file modes and avoid init script
hacks, a new variable called `PKG_FILE_MODES`. The variable contains a
list of files modes in the format `path:owner:group:mode`.
An example for the `sudo` package below:
```
PKG_FILE_MODES:=\
/usr/bin/sudo:root:root:4755 \
/etc/sudoers:root:root:0440
```
The `ipkg-build` now runs within a fakeroot environment to set any mode
and directly store it in the resulting `ipk` package archive.
Both options `-o` and `-g` are no longer required due to the introduction
of the more flexible `-m` options, which takes the `PKG_FILE_MODES` as
input.
Lastly the option `-c` is removed as it's unused within the script.
Signed-off-by: Paul Spooren <mail@aparcar.org>
This allows the build process to prepare a squashfs filesystem for use
with SELinux.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
[rebase, add commit message]
Signed-off-by: W. Michael Petullo <mike@flyn.org>
CMake provides a user package registry (stored in ~/.cmake/packages) and
a system package registry (not available on non-Windows platforms).
The "export(PACKAGE)" command may store information in the user package
registry, and the "find_package()" command may search both user and
system package registries for information.
This sets various variables to disable the use of these package
registries (both saving and retrieval of package information).
This also sets deprecated variables that perform similar functions, in
case external toolchains include older versions of CMake.
Signed-off-by: Jeffery To <jeffery.to@gmail.com>
This PR is a blend of several kernel bumps authored by ldir taken from his
staging tree w/ some further adjustments made by me and update_kernel.sh
Summary:
Deleted upstreamed patches:
generic:
742-v5.5-net-sfp-add-support-for-module-quirks.patch
743-v5.5-net-sfp-add-some-quirks-for-GPON-modules.patch
bcm63xx:
022-v5.8-mtd-rawnand-brcmnand-correctly-verify-erased-pages.patch
024-v5.8-mtd-rawnand-brcmnand-fix-CS0-layout.patch
mediatek:
0402-net-ethernet-mtk_eth_soc-Always-call-mtk_gmac0_rgmii.patch
Deleted patches applied differently upstream:
generic:
641-sch_cake-fix-IP-protocol-handling-in-the-presence-of.patch
Manually merged patches:
generic:
395-v5.8-net-sch_cake-Take-advantage-of-skb-hash-where-appropriate.patch
bcm27xx:
950-0132-lan78xx-Debounce-link-events-to-minimize-poll-storm.patch
layerscape:
701-net-0231-enetc-Use-DT-protocol-information-to-set-up-the-port.patch
Build system: x86_64
Build-tested: ath79/generic, bcm27xx/bcm2708, bcm27xx/bcm2711,
imx6, mvebu/cortexa9, sunxi/a53
Run-tested: Netgear R7800 (ipq806x)
No dmesg regressions, everything functional
Signed-off-by: John Audia <graysky@archlinux.us>
Tested-By: Lucian Cristian <Lucian.cristian@gmail.com> [mvebu]
Tested-By: Curtis Deptuck <curtdept@me.com> [x86/64]
[do not remove 395-v5.8-net-sch_cake-Take-advantage-... patch,
adjust and refresh patches, adjust commit message]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
Tested-By: John Audia <graysky@archlinux.us> [ipq806x]
The README file is displayed on "make help", but updating the
command has been overlooked during the README -> README.md
rename.
Fix it.
Fixes: d0113711a3 ("README: port to 21st century")
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
Fix shellcheck SC2230
> which is non-standard. Use builtin 'command -v' instead.
Using `command -v` is POSIX compliant while `which` is not. Also to
mention, `command -v` is a shell builtin whereas `which` is a separate
busybox applet.
Once applied to everything concerning OpenWrt we can disable the busybox
feature `which` and save 3.8kB.
Acked-by: Stijn Tintel <stijn@linux-ipv6.be>
Signed-off-by: Paul Spooren <mail@aparcar.org>
[also replace cases in zram-swap]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
It has been reported that the current message displayed during
upgrade with compat_version change is misleading for "legacy"
devices, i.e. those without the "new" fwtool. This is partially
caused by the fact that we need to exploit the supported_devices
string to get some message text displayed for these devices.
This patch modifies the message to make it more helpful and
include additional information, e.g.
Device linksys,wrt3200acm not supported by this image
Supported devices: linksys,wrt3200acm linksys-whateverelse - Image
version mismatch: image 1.1, device 1.0. Please wipe config during
upgrade (force required) or reinstall. Reason: Config cannot be
migrated from swconfig to DSA
Note that the line breaks (except the one before Supported devices)
are added manually here, I hesitate to hack \n into the
supported_devices as well. The "Reason:" will only be displayed if
DEVICE_COMPAT_MESSAGE is set for the device, otherwise
"Please check documentation ..." will be shown instead.
While at it, also rearrange the code in image-commands.mk to
make lines shorter and remove the double filter-out command.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
Add package which provides size optimized wpad with support for just
WPA-PSK, SAE (WPA3-Personal), 802.11r and 802.11w.
Signed-off-by: Petr Štetiar <ynezz@true.cz>
[adapt to recent changes, add dependency for WPA_WOLFSSL config]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
Instead of just printing a warning that an image is too big, also
print both actual size and limit in the string:
WARNING: Image file somename.bin is too big: 2096101 > 1048576
Since the kernel size is checked via the same function (if
KERNEL_SIZE is specified), this will also apply to the kernel
image size check.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
The recipe check-kernel-size is not used in the entire tree. Instead,
we already check the size of the kernel image in Device/Build/kernel
in image.mk via check-size function if KERNEL_SIZE is defined.
Therefore, drop the function. Using it would be redundant anyway.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
So far, the compatibility mechanism only works if both device and
image are already updated to the new routines. This patch extends
the sysupgrade metadata and fwtool_check_image() to account for
"older" images as well:
The basic mechanism for older devices to check for image compatibility
is the supported_devices entry. This can be exploited by putting
a custom message into this variable of the metadata, so older FW
will produce a mismatch and print the message as it thinks it's the
list of supported devices. So, we have two cases:
device 1.0, image 1.0:
The metadata will just contain supported_devices as before.
device 1.0, image 1.1:
The metadata will contain:
"new_supported_devices":["device_string1", "device_string2", ...],
"supported_devices":["Image version 1.1 incompatible to device: ..."]
If the device is "legacy", i.e. does not have the updated fwtool.sh,
it will just fail with image check and print the content of
supported_devices. If DEVICE_COMPAT_MESSAGE is set, this will be
printed on old devices as well through the same mechanism. Otherwise
a generic "Please check documentation ..." is appended.
Upgrade can still be performed with -F like when
SUPPORTED_DEVICES has been removed to prevent bricking.
If the device has updated fwtool.sh (but is 1.0), it will just use
the new_supported_devices instead, and work as intended (flashing
with -n will work, flashing without will print the appropriate
warning).
This mechanism should provide a fair tradeoff between simplicity
and functionality.
Since we touched a lot of fields in metadata, this also bumps
metadata_version to 1.1.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
We regularly encounter the situation that devices are subject to
changes that will make them incompatible to previous versions.
Removing SUPPORTED_DEVICES will not really be helpful in most of these
cases, as this only helps after a rename.
To solve this situation, this patchset introduces a compatibility
version for devices. It will be implemented via a per-device
Make variable DEVICE_COMPAT_VERSION, which will be set to 1.0
globally by default and then can be overwritten as needed.
Furthermore, a variable DEVICE_COMPAT_MESSAGE is added, where
a message to be displayed during sysupgrade may be specified
optionally.
This patch only implements the build variables and adds them
to the sysupgrade metadata, the evaluation will be addressed
in a subsequent patch.
To set it, one would just need to add the following to a device node:
define Device/somedevice
...
DEVICE_COMPAT_VERSION := 1.1
DEVICE_COMPAT_MESSAGE := Config cannot be migrated from swconfig to DSA
endef
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
By specifying "BROKEN := 1" or "BROKEN := y" for a device, it will be
hidden (and deselected) by default. By that, it provides a stronger
option to "disable" a device beyond just using DEFAULT := n.
To make these devices visible, just enable the BROKEN option in
developer settings as already implemented for targets and packages.
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
update_kernel.sh refreshed all patches, no human interaction was needed
Build system: x86_64
Run-tested: Netgear R7800 (ipq806x)
Signed-off-by: John Audia <graysky@archlinux.us>
Set CCACHE_DIR to $(TOPDIR)/.ccache and CCACHE_BASEDIR to $(TOPDIR).
This allows to do clean and dirclean. Cache hit rate for test build
after dirclean is ~65%.
If CCACHE is enabled stats are printed out at the end of building process.
CCACHE_DIR config variable allows to override default, which could be useful
when sharing cache with many builds.
cacheclean make target allows to clean the cache.
Changes from v1:
- remove ccache directory using CCACHE_DIR variable
- remove ccache leftovers from sdk and toolchain make files
- introduce CONFIG_CCACHE_DIR variable
- introduce cacheclean make target
Signed-off-by: Roman Yeryomin <roman@advem.lv>
Instead of using xargs to pass a huge number of files to
script/ipkg-remove, which will usually pick only one, use a more
restrictive wildcard so that, currently, at the most 325 files are
examined, instead of up to over 2,300. The 325-file package is python,
which is picking up python3* ipks. It is about to be removed.
Runner-up is ddns-scripts with 7 files.
This makes a second run of make package/luci/compile go from
real 16.40s; user 17.42s; sys 2.73s
to
real 10.71s; user 9.51s; sys 1.27s
There is a caveat though: if one were to remove the ABI_VERSION of a
package that ends in a digit [0-9], then the old package ipk will not be
removed from the bin directory by make package/abc2/clean.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
The wildcard call to clean up luci package (luci*) can pick up over
2,300 files when the full tree is built. Running make package/luci/clean
or a second run of make package/luci/compile would fail with an
'Argument list too long' error.
To avoid that, a maybe_use_xargs function was created that runs the
command straight as usual if the number of arguments is < 512, or saves
the list in a temporary file and feeds it to xargs otherwise.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Instead of calling $(wildcard) to check if the removal list is empty,
then calling it again to actually remove the files, define a function so
that the arguments are expanded only once when it gets called.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
f4f8f4a180 broke ffmpeg compilation with x86
The reason is that ffmpeg's x86 assembly requires at least MMX, which the
pentium CPU_TYPE was preventing.
Fixes ffmpeg compilation on x86_legacy and x86_geode.
Ref: https://github.com/openwrt/openwrt/pull/3061
Signed-off-by: Rosen Penev <rosenp@gmail.com>
