Init scripts doesn't work for Debian + OpenRC setup.

[cdluminate]

System information

Type	Version/Name
Distribution Name	Debian
Distribution Version	unstable
Linux Kernel	4.18.20
Architecture	amd64
ZFS Version	0.7.12
SPL Version	0.7.12

Describe the problem you're observing

Unlike Gentoo, Debian doesn't use /sbin/openrc-init as the pid1, but still use the init program from sysvinit-core. As a result, executing init scripts in the openrc way on Debian won't work, that because e.g. https://github.com/zfsonlinux/zfs/blob/master/etc/init.d/zfs-zed.in#L101 assumes the system pid1 is openrc once the /sbin/openrc-run is detected, and sysvinit doesn't understand the openrc way.

A dirty hack is provided in #8063 which works quite well for both Debian+sysvinit and Debian+openrc setups. A better solution requires us to detect the actual init program type. According to https://unix.stackexchange.com/questions/18209/detect-init-system-using-the-shell , a possible solution may look like

-if [ ! -e /sbin/openrc-run ]; then
-the sysvinit way
-else
-the openrc way
-fi
+if [ -e /sbin/openrc-run ] && strings $(strings /proc/1/cmdline) | grep openrc >/dev/null; then
+the openrc way
+else
+the sysvinit way
+fi

Describe how to reproduce the problem

1. setup Debian unstable base system
2. install zfsutils-linux (with debian init patch reverted)
3. install aptitude
4. aptitude install sysvinit-core
5. apt install openrc
6. reboot

Include any warning/errors/backtraces from the system logs

None, but zpools will no longer be automatically imported.

[cdluminate]

@kpande My point is that the if condition used to select sysvinit/openrc code branches is not complete. It enters a wrong code branch in Debian's case, where openrc is installed and detected, but the actuall init is still sysvinit instead of openrc. I'll test my proposed patch in a virtual machine. Should I submit a PR if the result looks good for Debian?

sysvinit/openrc is indeed not the first class init system of Debian, but to some extent such init diversity is good to have.

---

update: The proposed patch has been verified on my Debian vm, which selected the correct code branch for Debian+Openrc. If you think this patch is acceptable, I can also setup Gentoo and test it to see if I broke anything.

[gyakovlev]

nak from gentoo.

gentoo does not use /sbin/openrc-init by default. Users can configure it, but it's experimental and not feature-complete yet. So by default gentoo uses sysvinit's /sbin/init, same way as debian does according to your description (not sure how it's configured in that case, it's probably unsupported)

/sbin/openrc-run is a different beast, it's used to interpret initscripts and not related to /sbin/init or /sbin/openrc-init, don't confuse it.

again, the script checks for /sbin/openrc-run and not /sbin/openrc-init!

[gyakovlev]

your problem seems to be that the script assumes it should be run by /sbin/openrc-run if it's found and creates openrc-specific start/stop functions and omits the case loop, but in reality you expect it to be run by sysvinit with openrc installed, right?

[cdluminate]

your problem seems to be that the script assumes it should be run by /sbin/openrc-run if it's found and creates openrc-specific start/stop functions and omits the case loop, but in reality you expect it to be run by sysvinit with openrc installed, right?

@gyakovlev Yes. For Debian+sysvinit+openrc, it should enter the case code branch instead of the openrc-run code branch because in fact the openrc-run branch doesn't do anything since sysvinit don't recogonize the way openrc-run registers hooks.

[gyakovlev]

on gentoo system the shebang of the file is
#!/sbin/openrc-run, on debian it probably is #!/sbin/runscript but that condition forces openrc functions despite expecting runscript. can you check it what's your shebang?

not sure what is the correct way of detecting if the script is really run under openrc as service manager, I'll check it and report back.

we need to make sure it works in all scenarios: gentoo-sysvinit, gentoo-openrc-init, debian-sysvinit, debian-openrc, and it makes me want just create gentoo-only initscripts and don't use this fragile hackery.

[heroxbd]

if [ ! -e /sbin/openrc-run ] is not a predictor, a system can have OpenRC installed but not used.

@cdluminate openrc-run This has nothing to do with OpenRC-init.

@gyakovlev :

#!/sbin/openrc-run, on debian it probably is #!/sbin/runscript but that condition forces openrc functions despite expecting runscript.

runscript is the old name of openrc-run, they are the same thing.

can you check it what's your shebang?

This is the cause of this bug, the shebang of this script in Debian is /bin/sh, not openrc-run.

we need to make sure it works in all scenarios: gentoo-sysvinit, gentoo-openrc-init, debian-sysvinit, debian-openrc, and it makes me want just create gentoo-only initscripts and don't use this fragile hackery.

It does not matter what PID 1 you have. Therefore you are facing with only 2 cases here: OpenRC and sysv-rc.

[gyakovlev]

runscript is the old name of openrc-run, they are the same thing.

yeah, I haven't used debian for a while, thanks for clarification.

It does not matter what PID 1 you have. Therefore you are facing with only 2 cases here: OpenRC and sysv-rc.

true, I meant is the script should not make assumptions about init, but pick up correct service manager and not guess by checking if file exists. Like you said it can exists but be unused.

This is the cause of this bug, the shebang of this script in Debian is /bin/sh, not openrc-run.

it may be a symptom, not the cause, it depends how debian wants to run the script.
seems they want to run the script under sysvinit despite having openrc installed or active.

looks like Makefile makes (no pun intended) some assumptions about openrc the same way as initscript
https://github.com/zfsonlinux/zfs/blob/a8577bdb32e091645df901d8501e44ef50748389/etc/init.d/Makefile.am#L24-L28

I guess the if zfs package is built on a system without openrc shebang is set to #!/bin/sh, but the script still misses the case loop and tries to set openrc-specific functions.

@cdluminate @heroxbd can you help me understand how exactly openrc is used on debian in your case?
does it replace service manager completely or it works with it in conjunction with sysvinit and manages it's own set of services?
I checked this page https://wiki.debian.org/OpenRC and it's still not quite clear.

my current thoughts are that the script should:
if shebang is #!/sbin/openrc-run - try to use openrc-run unconditionally
if shebang is #!/bin/sh - never check if openrc-run is executable and create sysvinit functions unconditionally.

any ideas how to correctly check the caller of the script in POSIX sh or dash, without bashisms?
bash has $PPID, caller builtin, $BASH_SOURCE but it seems those things can't be used in that case.

[cdluminate]

@cdluminate @heroxbd can you help me understand how exactly openrc is used on debian in your case?
does it replace service manager completely or it works with it in conjunction with sysvinit and manages it's own set of services?

It only replaces sysv-rc and works in conjunction with other parts of sysvinit. And I learned that this script has nothing to do with init.

my current thoughts are that the script should:
if shebang is #!/sbin/openrc-run - try to use openrc-run unconditionally
if shebang is #!/bin/sh - never check if openrc-run is executable and create sysvinit functions unconditionally.

This looks good to me and should be able to solve this issue.

any ideas how to correctly check the caller of the script in POSIX sh or dash, without bashisms?
bash has $PPID, caller builtin, $BASH_SOURCE but it seems those things can't be used in that case.

If we stick to the above assumption, we can simply check the @SHELL@ string.

[cdluminate]

How about this patch?

Index: zfs/etc/init.d/zfs-import.in
===================================================================
--- zfs.orig/etc/init.d/zfs-import.in
+++ zfs/etc/init.d/zfs-import.in
@@ -308,8 +308,7 @@ do_start()
 
 # ----------------------------------------------------
 
-if [ ! -e /sbin/openrc-run ]
-then
+if ! (echo @SHELL@ | grep openrc 1>/dev/null 2>/dev/null); then
 	case "$1" in
 		start)
 			do_start
Index: zfs/etc/init.d/zfs-mount.in
===================================================================
--- zfs.orig/etc/init.d/zfs-mount.in
+++ zfs/etc/init.d/zfs-mount.in
@@ -199,8 +199,7 @@ do_stop()
 
 # ----------------------------------------------------
 
-if [ ! -e /sbin/openrc-run ]
-then
+if ! (echo @SHELL@ | grep openrc 1>/dev/null 2>/dev/null); then
 	case "$1" in
 		start)
 			do_start
Index: zfs/etc/init.d/zfs-share.in
===================================================================
--- zfs.orig/etc/init.d/zfs-share.in
+++ zfs/etc/init.d/zfs-share.in
@@ -58,7 +58,7 @@ do_stop()
 
 # ----------------------------------------------------
 
-if [ ! -e /sbin/openrc-run ]; then
+if ! (echo @SHELL@ | grep openrc 1>/dev/null 2>/dev/null); then
 	case "$1" in
 		start)
 			do_start
Index: zfs/etc/init.d/zfs-zed.in
===================================================================
--- zfs.orig/etc/init.d/zfs-zed.in
+++ zfs/etc/init.d/zfs-zed.in
@@ -98,7 +98,7 @@ do_reload()
 
 # ----------------------------------------------------
 
-if [ ! -e /sbin/openrc-run ]; then
+if ! (echo @SHELL@ | grep openrc 1>/dev/null 2>/dev/null); then
 	case "$1" in
 		start)
 			do_start

[cdluminate]

#8359

[stale]

This issue has been automatically marked as "stale" because it has not had any activity for a while. It will be closed in 90 days if no further activity occurs. Thank you for your contributions.