Some platforms like Raspberry Pi require patching some backup files like
cmdline.txt in order to set the correct root PARTUUID.
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
Due to previous refactoring in sysupgrade, writing backup archives to
stdout became impossible since the hardcoded gzip output redirection
did not account for the `-` special case filename.
Fix this issue by substituting `-` with `/proc/self/fd/1` in the tar
archive output path variable.
Also remove a redundant `rm -f` of the target file path that occurs
before the file could've possibly been written.
Fixes: #14773
Fixes: https://github.com/openwrt/luci/issues/6961
Fixes: e36cc53092 ("base-files: sysupgrade: use tar helper to include installed_packages.txt")
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Disabled services should be kept disabled after sysupgrade. This can be
easily handled using a proper uci-defaults script.
Extend sysupgrade to check for disabled services, generate uci-defaults
script disabling them and include it in backup.
Cc: Christian Marangi <ansuelsmth@gmail.com>
Cc: Jo-Philipp Wich <jo@mein.io>
Cc: Jonas Gorski <jonas.gorski@gmail.com>
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: Jo-Philipp Wich <jo@mein.io>
Replace mount + overlay with manually built tar archive that gets
prepended to the actual config files backup. This allows more
flexibility with including extra backup files. They can be included at
any paths and don't require writing to flash or mounting an overlay
which has its own limitations (mount points).
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: Jo-Philipp Wich <jo@mein.io>
This allows building uncompressed tar archives from shell scripts (and
compressing them later if needed)
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
[rmilecki: adapt to sysupgrade needs]
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
After commit ad62247800 ("base-files: improve lib/upgrade/common.sh")
behavior of export_bootdevice has been made consistent in such way that
always the whole disk device is exported (as that was the case already
when matching via UUID) rather than the partition device.
Do the same for the device holding the fitblk backing partition.
Fixes: 5992f976b3 ("base-files: recognize bootdevice on devices using fitblk")
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Boards using the fitblk driver need special treatment when it comes to
detecting the actual block device used to store the image used to boot
from. Transparently handle this in 'export_bootdevice' and provide new
'fitblk_get_bootdev' function to replace implementations in
/lib/upgrade/platform.sh.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Usage of word "add" was somehow misleading in those functions:
1. They don't really add (as in: append) anything. Result files are
created from scratch.
2. It wasn't clear what adding files means. It could be understood as
adding actual files somewhere (to existing archive?).
Also the word "add" was also a bit ambiguous.
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Rename function to more accurate and self-explanatory name:
1. Use "archive" in name as this functions creates tar archive
2. Avoid "conffiles" as this function may archive more than that
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
This reverts commit 4fa9aaf0be.
That seemed like a good idea allowing us to include any runtime
generated file in archive. Unfortuantely it broke backups with files
from mounted directories.
When mounting overlay with / as lowerdir its mounts don't propagete in
the mountpoint. That resulted in empty directories:
/tmp/overlay.XXXXXX/backup/tmp/
/tmp/overlay.XXXXXX/backup/var/
/tmp/overlay.XXXXXX/backup/dev/
/tmp/overlay.XXXXXX/backup/proc/
etc.
As some platforms / users try to backup files like /var/dhcp.leases or
/boot/cmdline.txt it means we can't use that solution.
Link: http://lists.openwrt.org/pipermail/openwrt-devel/2024-February/042320.html
Link: https://lore.kernel.org/linux-fsdevel/67bb0571-a6e0-44ea-9ab6-91c267d0642f@gmail.com/T/#u
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
This reverts commit bf304d10e9.
That uci-defaults script worked great but generating it required
mounting root dir as overlay lowerdir that needs to be reverted.
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Disabled services should be kept disabled after sysupgrade. This can be
easily handled using a proper uci-defaults script.
Extend sysupgrade to check for disabled services, generate uci-defaults
script disabling them and include it in backup.
Cc: Christian Marangi <ansuelsmth@gmail.com>
Cc: Jo-Philipp Wich <jo@mein.io>
Cc: Jonas Gorski <jonas.gorski@gmail.com>
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: Jo-Philipp Wich <jo@mein.io>
Setting overlay while creating backup allows including extra files in
archive without actually writing them to flash. Right now this feature
is limited to /etc/backup/ directory and is used only for including
installed_packages.txt.
Extend this solution to make it more generic:
1. Always mount overlay while creating backup
2. Overlay whole / to don't limit it to /etc/backup/
This allows including any additional files in backups and adding more
sysupgrade features.
Cc: Luiz Angelo Daros de Luca <luizluca@gmail.com>
Cc: Christian Marangi <ansuelsmth@gmail.com>
Cc: Jo-Philipp Wich <jo@mein.io>
Cc: Jonas Gorski <jonas.gorski@gmail.com>
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
When tar was failing, it was exiting immediately. Some files and the
tmpfs mount (-k) would remain breaking the next backup attempt.
Also remove redundant $? from exit builtin call as exit already returns
the last command exit code when called.
Signed-off-by: Luiz Angelo Daros de Luca <luizluca@gmail.com>
tar stderr was probably discarded only to remove this message:
tar: removing leading '/' from member names
However, together with that, any other error would also be discarded.
It is easier to fix that allowing the error message to be printed.
In sysupgrade, the backup file list only uses absolute paths. That way,
the solution is to remove the leading '/' from all files (sed) and chdir
to / (option -C /)
Signed-off-by: Luiz Angelo Daros de Luca <luizluca@gmail.com>
PoE devices in the realtek target have the possibility to add PSE info
to the board description via 02_network. Make this available for all
targets, by moving the uci_set_poe() function to the globally available
uci-default.sh script.
Signed-off-by: Sander Vanheule <sander@svanheule.net>
Add the possibility that colored LEDs can also be configured via the uci.
config led 'led1'
option name '<name>'
option sysfs '<path>'
option trigger 'default-on'
option default '1'
--> option color_{$color} '<0-255>'
The supported names of the variable "${color}" for the selected LED can be
queried in the file with the name 'multi_index'.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
Setting the trigger and checking whether the trigger can be set belong
together and should not be interrupted by other lines of code.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
There are monochrome LEDs that can only display one color. However, there
are also LEDs that can display multiple colors. This can be tested in the
led subsystem of the kernel if the files 'multi_index' and 'multi_intensity'
are present in the folder '/sys/class/leds/<ledname>'.
Until now it was not possible to reset the default color. This commit adds
the missing information in the file '/var/run/led.state' so that the bootup
color can be seen on the LED again when the LED configuration has been changed.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
Send error output of umount to /dev/null to mute error in case
ubiblock device has already been unmounted (which is usually the
case).
Gets rid of bogus error message:
umount: can't unmount /dev/ubiblock0_4: Invalid argument
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
It shouldn't gate on the value, since the value will ostensibly
always be set; instead it should depend on the variable being
prepended to being non-empty.
Fixes#14403
Signed-off-by: Philip Prindeville <philipp@redfish-solutions.com>
Add support for configuring rootfs mount options from cmdline.
Rootfs mount options can be passed by declaring in the kernel
cmdline as much options as needed prefixed with "rootfs_mount_options."
An example usage is with rootfs with F2FS filesystem to enable
compress_algorithm to reduce flash wear by compressing the files before
writing to flash.
Example usage:
"... rootfs_mount_options.compress_algorithm=zstd ..."
To pass multiple options:
"... rootfs_mount_options.compress_algorithm=zstd rootfs_mount_options.noinline_data ..."
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
Allow "postinst" scripts to perform extra actions after applying all
kind of fixups implemented using uci-defaults.
This is needed e.g. by uhttpd-mod-ubus which after installation in a
running systems needs to:
1. Update uhttpd config using its uci-defaults script
2. Reload uhttpd
While this approach makes sense there is a risk it'll blow up some
corner case postinst usages. There is only 1 way to find out.
Cc: Hauke Mehrtens <hauke@hauke-m.de>
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Explain some of the more obscure logic, or where we deviate from
what the original awk code did. Also, give a count of the usable
addresses on the subnet.
Signed-off-by: Philip Prindeville <philipp@redfish-solutions.com>
This is useful if you later need to perform numeric range-checking
on addresses, i.e. to see if an address falls inside a CIDR range,
etc. and what interface it corresponds to.
Signed-off-by: Philip Prindeville <philipp@redfish-solutions.com>
Similar to the *_get_mac_binary function, also split the common parts
off mtd_get_mac_ascii into new get_mac_ascii function and introduce
mmc_get_mac_ascii which uses it.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
The 'label' property in led node has been deprecated and we'd better
to avoid using it. This patch allows us to extract DT OF LED name
from the newly introduced LED properties "color", "function" and
"function-enumerator".
Signed-off-by: Shiji Yang <yangshiji66@qq.com>
dnmasq.init now invokes ipcalc.sh as either:
ipcalc.sh address/netmask ...
or:
ipcalc.sh address/prefix
but the existing version doesn't accept the 2nd notation. We're
trying to rationalize the usage of ipcalc.sh, and here we add
support for the 2nd format.
Signed-off-by: Philip Prindeville <philipp@redfish-solutions.com>
Add additional uci-defaults function for configuring GRO settings and
conduit for network devices.
Tweaking the GRO values might increase performance on some low spec
device that lack some offload feature on gmac.
Tweaking conduit interface is specific to DSA based devices and is
useful for multi-CPU scenario where one CPU is dedicated to one single
port.
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
Generalize ucidef_set_network_device functions to use a more generic
_ucidef_set_network_device_common that takes as args the option and the
value to apply instead of hardcoding.
This is to reduce duplicated code in preparation for addition of
additional option for board.d usage.
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
Sometimes it's useful to be able to prepend to a variable as
well, such as when dealing with domain names, e.g.
prepend fdqn "$subdomain" "."
will result in:
fqdn="$subdomain.$fqdn"
Signed-off-by: Philip Prindeville <philipp@redfish-solutions.com>
The ucidef_set_network_device_* functions in uci-defaults.sh disagree
on whether to use "network-device" or "network_device" in board.json.
With the additional caveat that jshn will translate hyphens (-) into
underscores (_). This casues problems in netifd which expected
"network_device" causing boards which depend on assigning MACs in
board.json via uci-defaults.sh (or jshn in general) to fail.
This commit addresses the issue by using network_device in
uci-defaults.sh.
The bug was uncovered in the forums here:
https://forum.openwrt.org/t/support-for-rtl838x-based-managed-switches/57875/2596
This was exposed by commit 4ebba8a05d ("realtek: add support for HPE
1920-8g-poe+") where the board_config_load call from 03_gpio introduced
the key normalization by jshn.
Fixes: 9290539ca9 ("base-files: allow setting device and bridge macs")
Tested-by: Stijn Segers <foss@volatilesystems.org>
Signed-off-by: Michael 'ASAP' Weinrich <michael@a5ap.net>
[ improve commit title, description and fix wrong Tested-by tag ]
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
The previous code handling the equal-condition might be removed or
altered in the future and the case might be overlooked.
Signed-off-by: Leon M. Busch-George <leon@georgemail.eu>
Some users have their routers configured to supply a DHCP range that
includes the local interface address.
That worked with dnsmasq because it automatically skips the local
address.
Re-enable those existing configurations for the release and hint at
possible future problems.
Signed-off-by: Leon M. Busch-George <leon@georgemail.eu>
[ wrap commit description and remove unecessary text ]
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
To avoid confusion when working with ipcalc.sh, clarify that the last two
parameters belong to the range calculation and rename 'num' to the slightly
less ambiguous 'size'.
Signed-off-by: Leon M. Busch-George <leon@georgemail.eu>
Add a function 'ipcalc' to /lib/functions.sh that sets variables more
safely using export.
With this new function, dnsmasq also handles the return value of ipcalc
correctly.
Fixes: e4bd3de1be ("dnsmasq: refuse to add empty DHCP range")
Co-Authored-By: Philip Prindeville <philipp@redfish-solutions.com>
Signed-off-by: Leon M. Busch-George <leon@georgemail.eu>
Printing a broadcast address doesn't make any sense for /31 and /32
prefixes.
Strictly speaking, the same goes for the network address but it is useful
to get the first address in the prefix, e.g. to create a canonical
CIDR notation "$NETWORK/$PREFIX".
Signed-off-by: Leon M. Busch-George <leon@georgemail.eu>
For /31 and /32 prefixes, there are only host addresses - no network and
broadcast address with all-zero and all-one bits.
Reflect this when setting the limit.
Signed-off-by: Leon M. Busch-George <leon@georgemail.eu>
The start and end addresses are inclusive.
Thus, adding num without substracting one results in num + 1 addresses.
Add the substraction and to implement the documented behaviour.
Signed-off-by: Leon M. Busch-George <leon@georgemail.eu>
Add 2 seconds sleep after each forcibly killed/tried-to-kill process
in the final process termination loop in sysupgrade stage2.
This is needed especially for qualcommax/ipq807x, where ath11k
wireless driver may have a long 10-20 seconds delay after termination
before actually getting killed. This often breaks sysupgrade.
The current KILL loop in kill_remaining does all 10 kill attempts
consecutively without any delay, as evidenced here in a failing sysupgrade.
It does not allow any time for the process to finalize its internal
termination.
Sat Sep 2 19:05:56 EEST 2023 upgrade: Sending TERM to remaining processes ...
Sat Sep 2 19:05:56 EEST 2023 upgrade: Sending signal TERM to hostapd (2122)
Sat Sep 2 19:05:56 EEST 2023 upgrade: Sending signal TERM to hostapd (2138)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending KILL to remaining processes ...
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending signal KILL to hostapd (2122)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending signal KILL to hostapd (2138)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending signal KILL to hostapd (2138)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending signal KILL to hostapd (2138)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending signal KILL to hostapd (2138)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending signal KILL to hostapd (2138)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending signal KILL to hostapd (2138)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending signal KILL to hostapd (2138)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending signal KILL to hostapd (2138)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending signal KILL to hostapd (2138)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Sending signal KILL to hostapd (2138)
Sat Sep 2 19:06:00 EEST 2023 upgrade: Failed to kill all processes.
sysupgrade aborted with return code: 256
The change in this commit adds a 2 seconds delay after each kill attempt
in order to allow some processes to more gracefully handle their
internal termination.
The result is like this:
Sun Sep 3 11:15:10 EEST 2023 upgrade: Sending TERM to remaining processes ...
Sun Sep 3 11:15:10 EEST 2023 upgrade: Sending signal TERM to hostapd (2309)
Sun Sep 3 11:15:10 EEST 2023 upgrade: Sending signal TERM to hostapd (2324)
Sun Sep 3 11:15:14 EEST 2023 upgrade: Sending KILL to remaining processes ...
Sun Sep 3 11:15:14 EEST 2023 upgrade: Sending signal KILL to hostapd (2309)
[ 699.827521] br-lan: port 7(hn5wpa2r) entered disabled state
[ 699.908673] device hn5wpa2r left promiscuous mode
[ 699.908721] br-lan: port 7(hn5wpa2r) entered disabled state
[ 701.038029] br-lan: port 6(hn5wpa3) entered disabled state
Sun Sep 3 11:15:16 EEST 2023 upgrade: Sending signal KILL to hostapd (2324)
[ 702.058256] br-lan: port 5(hn2wlan) entered disabled state
[ 709.250063] stage2 (8237): drop_caches: 3
Sun Sep 3 11:15:25 EEST 2023 upgrade: Switching to ramdisk...
The delay introduced here only kicks in if there is some process that
does not get terminated by the first TERM call. Then there is at least
one 2 sec wait after the first KILL loop round.
This commit is related to discussion in PRs #12235 and #12632
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Reviewed-by: Robert Marko <robimarko@gmail.com>
The duplicate sections are caused by a race condition at boot, when board.json
is not available. In that case, the final phy name cannot be resolved, and extra
sections referring to the path are created.
Fix this by making sure that wifi config is not being run before board.json
is created.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
