add UTRACE_SUPPORT build option

we can now build preload-seccomp, ujail, utrace separately

Signed-off-by: Etienne CHAMPETIER <champetier.etienne@gmail.com>

move /dev/shm to /tmp/shm

Since the /dev filesystem is tiny, /dev/shm needs to live somewhere
else.

Signed-off-by: Daniel Gimpelevich <daniel@gimpelevich.san-francisco.ca.us>

inittab: always proceed to next state

If no S or K rc scripts provided we still should
be able to go on to next state, or we won't boot
properly and would not be able to reboot/shutdown later.

Signed-off-by: Sergiy Kibrik <sakib@meta.ua>

hotplug: add BUTTON to environment vars for timeout action

as done in pressed and released actions

Signed-off-by: Günther Kelleter <guenther.kelleter@devolo.de>

fix generating syscall-names.h

Sometimes the syscall number is not defined with a number but with an
offset to an other syscall and then make_syscall_h.sh created some
broken header file.

For example the bit/syscall.h from musl for i386 has this:

  #define __NR_timer_create     259
  #define __NR_timer_settime    (__NR_timer_create+1)

With this patch the resulting array looks like this:

 [259] = "timer_create",
 [(__NR_timer_create+1)] = "timer_settime",

This fixes this bug from OpenWrt:
https://dev.openwrt.org/ticket/20195

Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>

allow buttons to call delayed timeout actions

Signed-off-by: John Crispin <blogic@openwrt.org>

jail: fix jail root folder permissions

We need a+x rights on the path to the root of the jails
so we can use users other than root (like nobody)

This partly fixes jailed dnsmasq

Signed-off-by: Etienne CHAMPETIER <champetier.etienne@gmail.com>

service: if logging start app with LD_PRELOAD & lib for line buffering

Using pipe automatically switches service to block buffering which kind
of breaks our logging. We won't get anything from stdout FD until the
buffer gets filled fully or the service exits. This makes log messages
appear with an unwanted delay.
This change adds a tiny libsetlbf.so switching stdout to line buffering
and uses this lib for every logging-enabled service started by procd.
We don't need any extra change for stderr as it's unbuffered by default.

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Modified to use no buffering to align with stderr. Several cleanups.

Signed-off-by: Steven Barth <steven@midlink.org>

Attempt to deal gracefully with allocation failures.

These should probably not happen in practice, but having an explicit
error message may make debugging out-of-memory situations easier.

Revert "hotplug: support for interval commands"

This reverts commit 9562ce477476a27851ec90cfbf971b8cb41c81a6.

Revert requested by John, will be re-implemented in a different way.

fix /dev/shm permissions, this time for real

Previous patch did not account for umask, now adding that.

Signed-off-by: Daniel Gimpelevich <daniel@gimpelevich.san-francisco.ca.us>

fix /dev/shm permissions

On my Ubuntu system, the permissions are 1777. They are incorrect in
procd, leading to this:
https://forum.openwrt.org/viewtopic.php?id=57073
This in intended for both CC and DD.

Signed-off-by: Daniel Gimpelevich <daniel@gimpelevich.san-francisco.ca.us>

hotplug: support for interval commands

This allows executing code with a given interval. As every command, it
can be assign to any uevent.

Intervals may be useful for counting elapsed time since some action. It
allows e.g. indicating that button has been pressed for some time. This
is useful to let user know he can already release the button.

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>

service: reorder function to avoid forward declaration

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

service: close instance pipe fd on restart.

Otherwise we hit max number of fd limit (1024) and instances fail to
start with the following errors in syslog

    Sun Jun 14 01:27:38 2015 daemon.warn procd: pipe() failed: 24 (Too many open files)

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>

service: allow get_data of specifc instance.

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>

service: remove unused struct watch_subscribe definition.

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>

service: dump respawn params in the same order as when passed in.

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>

service: optimize relaying stdio output of daemons a bit.

Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>

trigger: make trigger_match() bool and make return value less confusing

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

allow multiple identical events to be queued

Signed-off-by: John Crispin <blogic@openwrt.org>

remove  ->
[    1.240000] init: failed to symlink /tmp -> /var

Signed-off-by: John Crispin <blogic@openwrt.org>

jail: respect byte order when setting AUDIT_ARCH

AUDIT_ARCH on ARM and MIPS differs depending on the byte order.
Thus set AUDIT_ARCH to the respective endian-specific variants.

Signed-off-by: Daniel Golle <daniel@makrotopia.org>

jail: add support for ARM architecture

SECCOMP_FILTER isn't supported on ARM OABI.
Thus enable seccomp support in jail only for EABI.
thumb might work as well as it apparently implies EABI, but
yet doesn't set __ARM_EABI__.

The REG_SYSCALL macro seems to be an unused left-over.
However, it's defined for other architectures as well.

Signed-off-by: Daniel Golle <daniel@makrotopia.org>

trace: add support for ARM architecture

Signed-off-by: Daniel Golle <daniel@makrotopia.org>

instance: handle setgid() before setuid()

When attempting to run a service with an unprivileged user and group
id procd, the following error might occur:

procd: failed to set uid:1000, gid:1000

This is due to the fact that procd first performs the setuid(), then
the setgid() call.

Usually there no sufficient permissions after a setuid() anymore to
change the effective group id of the process.

Refactor the code to:

  * Swap the invocations (first gid, then uid)
  * Don't set user or group id if it is 0
  * Handle errors independently and make them more verbose

Signed-off-by: Jo-Philipp Wich <jow@openwrt.org>

instance: avoid dumping invalid service instances - prevents a potential crash

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

fix include order

this broke x86_64 builds on uclibc

Signed-off-by: John Crispin <blogic@openwrt.org>

properly handle return codes

Signed-off-by: John Crispin <blogic@openwrt.org>

make it optional to remount the tmpfs as ro

Signed-off-by: John Crispin <blogic@openwrt.org>

make jail build optional

Signed-off-by: John Crispin <blogic@openwrt.org>

procd can now start jailed processes

Signed-off-by: John Crispin <blogic@openwrt.org>

add initial version of ujail and utrace

Signed-off-by: John Crispin <blogic@openwrt.org>

check for empty parameters in askfirst

Signed-off-by: Zefir Kurtisi <zefir.kurtisi@neratec.com>

switch to _DEFAULT_SOURCE for modern glibc compat

Signed-off-by: Jeff Waugh <jdub@bethesignal.org>

json 0.12 fixes
Signed-off-by: John Crispin <blogic@openwrt.org>

service: rename variables / struct members called stdout, stderr to avoid conflicts with system defines

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

service: don't use stdio log channel

Signed-off-by: Jo-Philipp Wich <jow@openwrt.org>

procd: support relayoing daemon stdout/stderr to syslog

This commit adds support to procd for relaying stdout and stderr streams to
the system log. That is mainly useful for services not using syslog, e.g.
uhttpd.

Signed-off-by: Jo-Philipp Wich <jow@openwrt.org>

Convert log calls to ulog() api

Signed-off-by: Jo-Philipp Wich <jow@openwrt.org>

procd: increase memory allocated for tmpfs on zram

Devices with <= 32MB of ram get half of memory allocated to zram (up to 16MB).
Devices with > 32MB of ram get just 8MB of memory allocated to zram.

Increase memory allocated to devices with > 32MB ram to 16MB.

Signed-off-by: Nathan Hintz <nlhintz@hotmail.com>

Make build of upgraded optional

Signed-off-by: Jo-Philipp Wich <jow@openwrt.org>

system: increase memory field sizes to 64bit

On an Alix APU board with 4GB of available ram, the total memory is reported
as "-179417088" bytes. Increase the ubus field sizes to 64bit integers in
order to avoid overflows.

Signed-off-by: Jo-Philipp Wich <jow@openwrt.org>

system: attempt to parse /proc/device-tree/model

Fallback to /proc/device-tree/model if /tmp/sysinfo/model is not available.

Signed-off-by: Jo-Philipp Wich <jow@openwrt.org>

system: improve system name detection

Skip entries like "Processor: 0" which are common on x86, otherwise
an "ubus call system board" will just return "system: 0".

Signed-off-by: Jo-Philipp Wich <jow@openwrt.org>

inittab: don't close fds unconditionally since this breaks UML

Signed-off-by: Steven Barth <steven@midlink.org>

only write to the watchdog if the fd is valid

Signed-off-by: John Crispin <blogic@openwrt.org>

previous commit accidentiall dropped the cgroup mount

Signed-off-by: John Crispin <blogic@openwrt.org>

add support for zram compressed tmpfs

Signed-off-by: John Crispin <blogic@openwrt.org>

service: fix ubus list command

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

initd: mount cgroup

If kernel is compiled with cgroup support it should be mounted. This change
does not effect kernels without cgroup support.

Signed-off-by: Luka Perkov <luka@openwrt.org>

inittab: clean up tty opening code, fix console shell job control issues

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

call setsid during startup.

this fixes a bug where pgrp was not set up properly leading to a trail of carnage;

Signed-off-by: John Crispin <blogic@openwrt.org>

kmodloader takes longer than wdt timeout

on systms with slow flash the watchdog will trigger when a lot of modules are
included in the image.

Signed-off-by: John Crispin <blogic@openwrt.org>

get_cmdline_val: search for entire name, not just suffix

also fix writing of 0-byte to buffer

Signed-off-by: Steven Barth <steven@midlink.org>

procd: Make askconsole work again when no tty is specified in inittab

Consider:
::askconsole:/bin/ash --login

askconsole( ) checks for the existance of the tty, but if none is
specified it will skip the remainder of the function. This means
fork_worker( ) is never called and no process is spawned. This
would leave routers without an initial console.

Signed-off-by: Michel Stam <m.stam@fugro.nl>

Honour tty field in /etc/inittab

The problem was caused by procd not opening /dev/tty* (whichever was
specified for the ID field /etc/inittab), causing /proc/PID/fd to
point to /dev/console instead.

This is a rework of e63051d9, which did not initialise the console
pointer and did not check the tty pointer in askconsole. askfirst
was not completely fixed as it expected the console parameter on
the commandline, which is no longer necessary because procd opens
the console prior to fork()-ing.

Signed-off-by: Michel Stam <m.stam@fugro.nl>

Revert "Honour tty field in /etc/inittab"

This reverts commit e63051d9843ddbafb1fabfd97d60e853bdeac129.
This unbreaks the initial console on UML and possibly other platforms.

Signed-off-by: Steven Barth <cyrus@openwrt.org>

Fix regression in command line parsing

79872ea6 reduced the command line buffer breaking various platforms.

Signed-off-by: Steven Barth <steven@midlink.org>

Log startup/shutdown to console

procd has the habit of logging startup/shutdown via
rcS to syslog, which is pointless in case of a
shutdown, and unlikely to be complete on a startup
(as syslog is not running). Write to the console
instead.

Signed-off-by: Michel Stam <m.stam@fugro.nl>
Signed-off-by: John Crispin <blogic@openwrt.org>

minor fixes to michels patches

Signed-off-by: John Crispin <blogic@openwrt.org>

Honour tty field in /etc/inittab

The problem was caused by procd not opening /dev/tty* (whichever was
specified for the ID field /etc/inittab), causing /proc/PID/fd to
point to /dev/console instead.

Signed-off-by: Michel Stam <m.stam@fugro.nl>

Fix ctrl+alt+del support

The previous patch did not catch SIGINT, which is used by the
kernel to indicate to the init process that the system should
reboot.

Signed-off-by: Michel Stam <m.stam@fugro.nl>

Show the shutdown sequence on the active virtual terminal

procd by default writes to /dev/console. When rebooting, this means that the
terminal on which the reboot sequence was started will not see what is going
on. This patch fixes that by reopening stdin, stdout and stderr to the console
device specified on the commandline, /dev/tty0 or /dev/console upon reboot.

Also, due to (probably) pivot-root, /proc/1/fd shows 1-3 pointing to
/console. This patch also fixes that.

Signed-off-by: Michel Stam <m.stam@fugro.nl>

Use one generic routine to access /proc/cmdline

Signed-off-by: Michel Stam <m.stam@fugro.nl>

make procd wait for ubus to come up

Signed-off-by: John Crispin <blogic@openwrt.org>

Fix scroll lock blocking procd

On x86, pressing the scrolllock button may effectively prevent procd from rebooting
the system. This happens because procd tries to write to /dev/console, which in that
situation is blocked, effectively blocking procd from rebooting the system.

This patch puts procd's stderr access into non-blocking, which will prevent the
boot from being stuck.

Signed-off-by: Michel Stam <m.stam@fugro.nl>

make procd handle ctrl+alt+del

Signed-off-by: John Crispin <blogic@openwrt.org>

Fix 'reboot' message when the system is powering down

While executing a system halt (via the powerbutton or otherwise),
the system displays rebooting, whereas it should display that
it is going to turn off.

Signed-off-by: Michel Stam <m.stam@fugro.nl>

Reboot busybox style via procd

It has been observed that, very rarely, the system does not reboot
when procd enters the shutdown state. Busybox seems to have an
obscure fix in its shutdown, where the parent (procd) process
is put in an infinite loop, and a special child is executed to
execute reboot(). This may very well be the cause.

Signed-off-by: Michel Stam <m.stam@fugro.nl>

Kill processes on shutdown

Procd as it currently is does not kill the running processes during shutdown.
As this might cause undesired behaviour, this should be implemented.
Implementation was taken from busybox 1.19.4

Signed-off-by: Michel Stam <m.stam@fugro.nl>

Fix procd not handling rapid reboot signals very well

When keeping ctrl-alt-del pressed, or when running;
while true; do
    /sbin/reboot
done
effectively the procd shutdown cycle gets into an undefined state,

Signed-off-by: Michel Stam <m.stam@fugro.nl>

fix a copy/paste bug leading to a bad loop limitation

https://dev.openwrt.org/ticket/17864

Signed-off-by: John Crispin <blogic@openwrt.org>

fix timeout on shutdown

Signed-off-by: John Crispin <blogic@openwrt.org>

instances can now start as a !root user

Signed-off-by: John Crispin <blogic@openwrt.org>

add .data to instance_dump()

Signed-off-by: John Crispin <blogic@openwrt.org>

add an event for "service.update" triggers

Signed-off-by: John Crispin <blogic@openwrt.org>

hotplug: avoid using the deprecated is_error() call, check for NULL pointer instead

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

explicitly include json/json.h where needed

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

system.c: parse quoted version info from /etc/openwrt_release

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

Make failure to resize hotplug buffer non-fatal

Some architectures seem to have broken SO_RCVBUFFORCE.

Signed-off-by: Steven Barth <steven@midlink.org>

hotplug: fix firmware loading

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

service: fix trigger related double-free

The memdup is unnecessary, because the data is contained inside
in->config.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

service: implement get_data to dump the data fields for every instance

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

instance: allow filling "data" with fields of arbitrary type

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

make static linking work for upgraded

Signed-off-by: John Crispin <blogic@openwrt.org>

only build upgraded on mips

Signed-off-by: John Crispin <blogic@openwrt.org>

move upgrade to its own subfolder

Signed-off-by: John Crispin <blogic@openwrt.org>

rename the folder option to path

Signed-off-by: John Crispin <blogic@openwrt.org>

procd: force -lc -lgcc to handle cross-dependencies between those two

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

Increase buffers for hotplug and cmdline for UML

Thanks to Markus Stenberg for debugging and testing.

Signed-off-by: Steven Barth <steven@midlink.org>

only add the nandupgrade method when /sbin/upgraded exists

Signed-off-by: John Crispin <blogic@openwrt.org>

add upgraded support for nand flashing

Signed-off-by: John Crispin <blogic@openwrt.org>

allow instances to report errors. if an error is set, the instance wont be started

Signed-off-by: John Crispin <blogic@openwrt.org>

service: do not emit any service events before the ubus connection is up

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

service: register the subscriber object before the event handler

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

service: fix a debug message typo

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

service: fix a leftover debug hack

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

0467aebb79f64d1c9d05a84367fda7c02c4ea6b8 included too much

Signed-off-by: John Crispin <blogic@openwrt.org>

wait for kmod loader to complete on first boot

Signed-off-by: John Crispin <blogic@openwrt.org>