OpenWrt/LEDE Project

Attached to Project: OpenWrt/LEDE Project
Opened by Baptiste Jonglez - 13.06.2020
Last edited by Baptiste Jonglez - 06.07.2020

I’ve been trying to debug this regression in 18.06.8: in some circumstances, rpcd fails to start.

This is mostly visible as it breaks LuCI, see e.g. https://github.com/openwrt/luci/issues/3773 or https://forum.openwrt.org/t/luci-error-after-upgrade-to-r10949-or-r10951-etc-config-luci-seems-to-be-corrupt/56880

To reproduce on 18.06.8:

• remove the rpcd section in /etc/config/rpcd
• reboot (this is important)
• result: rpcd is not started, and the following log message is printed in logread:

Thu Feb 27 21:26:37 2020 daemon.info procd: Not starting instance rpcd::instance1, command not set

More details:

When this issue happens, it becomes impossible to start rpcd with procd, even when adding back the rpcd section:

root@OpenWrt:~# PROCD_DEBUG=1 /etc/init.d/rpcd start
{ "name": "rpcd", "script": "\/etc\/init.d\/rpcd", "instances": { "instance1": { "command": [ "\/sbin\/rpcd" ] } }, "triggers": [ ], "data": { } }
root@OpenWrt:~# ps | grep rpc
 1614 root      1200 S    grep rpc

root@OpenWrt:~# uci add rpcd rpcd
cfg027c4e
root@OpenWrt:~# uci set rpcd.@rpcd[-1].timeout=30
root@OpenWrt:~# uci commit
root@OpenWrt:~# PROCD_DEBUG=1 /etc/init.d/rpcd start
{ "name": "rpcd", "script": "\/etc\/init.d\/rpcd", "instances": { "instance1": { "command": [ "\/sbin\/rpcd", "-t", "30" ] } }, "triggers": [ ], "data": { } }
root@OpenWrt:~# ps | grep rpc
 1636 root      1200 S    grep rpc

root@OpenWrt:~# uci set rpcd.@rpcd[-1].socket=/var/run/ubus.sock
root@OpenWrt:~# uci commit
root@OpenWrt:~# PROCD_DEBUG=1 /etc/init.d/rpcd start
{ "name": "rpcd", "script": "\/etc\/init.d\/rpcd", "instances": { "instance1": { "command": [ "\/sbin\/rpcd", "-s", "\/var\/run\/ubus.sock", "-t", "30" ] } }, "triggers": [ ], "data": { } }
root@OpenWrt:~# ps | grep rpc
 1680 root      1200 S    grep rpc

However, running rpcd manually works perfectly well (and fixes LuCI):

root@OpenWrt:~# rpcd

Workaround:

To workaround the issue, it is necessary to:

• add a rpcd section with either a socket or timeout option
• reboot

At this point, rpcd is started correctly, and everything works fine. It is even possible to delete the rpcd section and restart rpcd, it will still start correctly.

Finding the root cause:

There are very few commits between 18.06.7 and 18.06.8. None of these commits is touching procd or rpcd.

However, there has been a libubox fix in 82fbd857471292ca71dc06b05f11089962f33a4f. This is currently the prime suspect: I will try to revert this commit, and also try with the further libubox fixes that have not yet been backported.