Skip to contentSkip to conversation

system/skeleton: make /var/run a symlink to /run

PATCH New Norbert Lange 2 participants 10 messages Jun 29, 2020 – Jul 16, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 system/skeleton: make /var/run a symlink to /run

both systemd and FHS3 denote /var/run as a compatibility directory, new systems should use /run.

There seems to be little reason to not switch to the new FHS layout with buildroot, and adding the compatibility symlink.

[1] - https://refspecs.linuxfoundation.org/FHS_3.0/fhs/ch05s13.html

---
 package/skeleton-init-systemd/skeleton-init-systemd.mk | 2 --
 system/skeleton/var/run                                | 1 +
 2 files changed, 1 insertion(+), 2 deletions(-)
 create mode 120000 system/skeleton/var/run
package/skeleton-init-systemd/skeleton-init-systemd.mk +0 -2
index 5d6b716f7a..528cf43e61 100644
@@ -60,8 +60,6 @@ endif
 define SKELETON_INIT_SYSTEMD_INSTALL_TARGET_CMDS
 	mkdir -p $(TARGET_DIR)/home
 	mkdir -p $(TARGET_DIR)/srv
-	mkdir -p $(TARGET_DIR)/var
-	ln -s ../run $(TARGET_DIR)/var/run
 	$(SKELETON_INIT_SYSTEMD_ROOT_RO_OR_RW)
 endef
 
system/skeleton/var/run new file +1 -0
new file mode 120000
index 0000000000..84ba55b912
@@ -0,0 +1 @@
+../run
\ No newline at end of file
Signature · 1 line
2.27.0
#1 lore raw

PATCH 1/1 system/skeleton: make /var/run a symlink to /run

Am Mo., 29. Juni 2020 um 12:47 Uhr schrieb Norbert Lange <[email protected]>:

Quoted 38 lines

both systemd and FHS3 denote /var/run as a compatibility directory, new systems should use /run.

There seems to be little reason to not switch to the new FHS layout with buildroot, and adding the compatibility symlink.

[1] - https://refspecs.linuxfoundation.org/FHS_3.0/fhs/ch05s13.html

---
 package/skeleton-init-systemd/skeleton-init-systemd.mk | 2 --
 system/skeleton/var/run                                | 1 +
 2 files changed, 1 insertion(+), 2 deletions(-)
 create mode 120000 system/skeleton/var/run
package/skeleton-init-systemd/skeleton-init-systemd.mk +0 -2
index 5d6b716f7a..528cf43e61 100644
@@ -60,8 +60,6 @@ endif
 define SKELETON_INIT_SYSTEMD_INSTALL_TARGET_CMDS
        mkdir -p $(TARGET_DIR)/home
        mkdir -p $(TARGET_DIR)/srv
-       mkdir -p $(TARGET_DIR)/var
-       ln -s ../run $(TARGET_DIR)/var/run
        $(SKELETON_INIT_SYSTEMD_ROOT_RO_OR_RW)
 endef
 
system/skeleton/var/run new file +1 -1
new file mode 120000
index 0000000000..84ba55b912
@@ -0,0 +1 @@
+../run
\ No newline at end of file
--
Quoted 2 lines

2.27.0

Some future change would be to change relative symlinks touching the target root to absolutes.
potentially as last step after the fakeroot scripts. Reasoning can be found at the debian manual (allowing symlinks in root).

Norbert

[1] - https://www.debian.org/doc/debian-policy/ch-files.html#symbolic-links

#2 lore raw

PATCH 1/1 system/skeleton: make /var/run a symlink to /run

In reply to Norbert Lange

Hello Norbert,

Quoted 7 lines · On Mon, 29 Jun 2020 12:47:49 +0200 Norbert Lange <[email protected]> wrote:

both systemd and FHS3 denote /var/run as a compatibility directory, new systems should use /run.

There seems to be little reason to not switch to the new FHS layout with buildroot, and adding the compatibility symlink.

[1] - https://refspecs.linuxfoundation.org/FHS_3.0/fhs/ch05s13.html

The commit log is a bit confusing here, especially when looking at the patch. Indeed, we see you are *removing* code that creates /var/run as a symlink to /run, but you're saying that you're making /var/run a symlink to /run.

However, what really happens is that /var/run was already a symlink to /run when the selected init system is systemd, but not with other init systems. What your patch does is that it moves all init system to this convention.

So perhaps a better commit title would be:

===

system/skeleton: always make /var/run a symlink to /run

While Buildroot was already making /var/run a symlink to /run when systemd was selected as the init system, it was not the case with other init systems.

However, FHS3 denotes /var/run as a compatibility directory, and new systems should use /run. So let's switch to this convention regardless of the init system being used.

===

Also, was there any specific motivation/issue for this change, or just make things "nicer" ?

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#3 lore raw

PATCH 1/1 system/skeleton: make /var/run a symlink to /run

Am Di., 14. Juli 2020 um 15:15 Uhr schrieb Thomas Petazzoni <[email protected]>:

Quoted 34 lines

Hello Norbert,

Quoted 7 lines · On Mon, 29 Jun 2020 12:47:49 +0200 Norbert Lange <[email protected]> wrote:

both systemd and FHS3 denote /var/run as a compatibility directory, new systems should use /run.

There seems to be little reason to not switch to the new FHS layout with buildroot, and adding the compatibility symlink.

[1] - https://refspecs.linuxfoundation.org/FHS_3.0/fhs/ch05s13.html

The commit log is a bit confusing here, especially when looking at the patch. Indeed, we see you are *removing* code that creates /var/run as a symlink to /run, but you're saying that you're making /var/run a symlink to /run.

However, what really happens is that /var/run was already a symlink to /run when the selected init system is systemd, but not with other init systems. What your patch does is that it moves all init system to this convention.

So perhaps a better commit title would be:

===

system/skeleton: always make /var/run a symlink to /run

While Buildroot was already making /var/run a symlink to /run when systemd was selected as the init system, it was not the case with other init systems.

Quoted 5 lines

However, FHS3 denotes /var/run as a compatibility directory, and new systems should use /run. So let's switch to this convention regardless of the init system being used.

===

Fine with that, can you change that pls?

Quoted 3 lines

Also, was there any specific motivation/issue for this change, or just make things "nicer" ?

Removal of stuff like this:
ifeq ($(BR2_INIT_SYSTEMD),y)
LTTNG_LIBUST_CONF_OPTS += --with-lttng-system-rundir=/run/lttng endif

You get warnings everytime you start systemd with units using a var/run directory
(eg. dbus). So cleaning that up would be the next step once this change is accepted.

Norbert

#4 lore raw

PATCH 1/1 system/skeleton: make /var/run a symlink to /run

Quoted 7 lines · On Tue, 14 Jul 2020 15:35:22 +0200 Norbert Lange <[email protected]> wrote:
Quoted 2 lines

Also, was there any specific motivation/issue for this change, or just make things "nicer" ?

Removal of stuff like this:
ifeq ($(BR2_INIT_SYSTEMD),y)
LTTNG_LIBUST_CONF_OPTS += --with-lttng-system-rundir=/run/lttng endif

I'm confused, this snippet of code is for systemd configurations only, and your patch doesn't change anything for systemd-based systems, where /var/run was already a symlink to /run.

Could you clarify ?

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#5 lore raw

PATCH 1/1 system/skeleton: make /var/run a symlink to /run

Am Di., 14. Juli 2020 um 15:52 Uhr schrieb Thomas Petazzoni <[email protected]>:

Quoted 12 lines
Quoted 7 lines · On Tue, 14 Jul 2020 15:35:22 +0200 Norbert Lange <[email protected]> wrote:
Quoted 2 lines

Also, was there any specific motivation/issue for this change, or just make things "nicer" ?

Removal of stuff like this:
ifeq ($(BR2_INIT_SYSTEMD),y)
LTTNG_LIBUST_CONF_OPTS += --with-lttng-system-rundir=/run/lttng endif

Quoted 5 lines

I'm confused, this snippet of code is for systemd configurations only, and your patch doesn't change anything for systemd-based systems, where /var/run was already a symlink to /run.

Could you clarify ?

chronological:

-   Buildroot uses /var/run as RUNTIME directory for SysV, and /run for systemd.
-   I add the lttng patch above to use either /var/run/lttng or /run/lttng
-   I consider doing the same for dbus as I get warnings otherwise

(actually have a post_install script replacing /var/run with /run)
- I consider it would be alot easier to just use /run everywhere and
add this patch to see if that's accepted, before doing more if-butt-ing like with lttng

So the end result should be that /var/run is treated as legacy unconditionally, and packages would always use /run,
and don't depend on a symlink (potentially on a network drive). the symlink schoul just be a safety net.

Norbert

#6 lore raw

PATCH 1/1 system/skeleton: make /var/run a symlink to /run

Quoted 9 lines · On Tue, 14 Jul 2020 16:22:25 +0200 Norbert Lange <[email protected]> wrote:

chronological:

(quoted diff) +0 -3
-   Buildroot uses /var/run as RUNTIME directory for SysV, and /run for systemd.
-   I add the lttng patch above to use either /var/run/lttng or /run/lttng
-   I consider doing the same for dbus as I get warnings otherwise

(actually have a post_install script replacing /var/run with /run)
- I consider it would be alot easier to just use /run everywhere and
add this patch to see if that's accepted, before doing more if-butt-ing like with lttng

But then, you have to anyway tweak lttng to pass:

LTTNG_LIBUST_CONF_OPTS += --with-lttng-system-rundir=/run/lttng

but unconditionally, and no longer depending on whether systemd is used or not. Correct ?

But... /run is already a tmpfs in sysv-based systems, so what prevents today from using /run/lttng even for sysv based systems ?

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#7 lore raw

PATCH 1/1 system/skeleton: make /var/run a symlink to /run

Am Mi., 15. Juli 2020 um 15:43 Uhr schrieb Thomas Petazzoni <[email protected]>:

Quoted 14 lines
Quoted 9 lines · On Tue, 14 Jul 2020 16:22:25 +0200 Norbert Lange <[email protected]> wrote:

chronological:

(quoted diff) +0 -3
-   Buildroot uses /var/run as RUNTIME directory for SysV, and /run for systemd.
-   I add the lttng patch above to use either /var/run/lttng or /run/lttng
-   I consider doing the same for dbus as I get warnings otherwise

(actually have a post_install script replacing /var/run with /run)
- I consider it would be alot easier to just use /run everywhere and
add this patch to see if that's accepted, before doing more if-butt-ing like with lttng

Quoted 6 lines

But then, you have to anyway tweak lttng to pass:

LTTNG_LIBUST_CONF_OPTS += --with-lttng-system-rundir=/run/lttng

but unconditionally, and no longer depending on whether systemd is used or not. Correct ?

That's the idea, yes.

Quoted 2 lines

But... /run is already a tmpfs in sysv-based systems, so what prevents today from using /run/lttng even for sysv based systems ?

Ok, I missed that bit, thought it would have unrelated /run and /var/run directories.
So this patch is just a cleanup then, and buildroot already *always* uses a tmpfs /run (with compatibilty symlink /var/run),
and I can use /run with lttng and dbus whether this cleanup is merged or not?

I'd redo the patch with those removed, if you agree:

package/skeleton-init-openrc/skeleton/var/run
package/skeleton-init-sysv/skeleton/var/run

Or do nothing if you dont ;)
My new argument would be that this would prevent false assumptions (like mine).

Norbert

#8 lore raw

PATCH 1/1 system/skeleton: make /var/run a symlink to /run

Quoted 9 lines · On Wed, 15 Jul 2020 15:53:12 +0200 Norbert Lange <[email protected]> wrote:
Quoted 2 lines

But... /run is already a tmpfs in sysv-based systems, so what prevents today from using /run/lttng even for sysv based systems ?

Ok, I missed that bit, thought it would have unrelated /run and /var/run directories.
So this patch is just a cleanup then, and buildroot already *always* uses a tmpfs /run (with compatibilty symlink /var/run),
and I can use /run with lttng and dbus whether this cleanup is merged or not?

Quoted 7 lines

I'd redo the patch with those removed, if you agree:

package/skeleton-init-openrc/skeleton/var/run
package/skeleton-init-sysv/skeleton/var/run

Or do nothing if you dont ;)
My new argument would be that this would prevent false assumptions (like mine).

OK, the more I look into this, the more I'm confused. To me, it seems like in both systemd *and* non-systemd cases, /var/run is already a symlink to /run, but just implemented in a different way.

For the systemd case, we have:

        ln -s ../run $(TARGET_DIR)/var/run

in package/skeleton-init-systemd/skeleton-init-systemd.dmk.

For the sysv/openrc case, we have the skeleton itself that contains:

$ ls -l package/skeleton-init-sysv/skeleton/var/run 
lrwxrwxrwx 1 thomas thomas 6 26 mars  17:10 package/skeleton-init-sysv/skeleton/var/run -> ../run
$ ls -l package/skeleton-init-openrc/skeleton/var/run 
lrwxrwxrwx 1 thomas thomas 6 18 mai   17:32 package/skeleton-init-openrc/skeleton/var/run -> ../run

So, it all cases, /var/run is already a symlink to /run. Am I missing something here ?

So, if anything needs to be done, it's just cleaning up that and doing it in the same way for all init systems, perhaps in a common place. But we need to be careful, skeleton-init-systemd.mk has some special/complex logic to handle /var and generate a /etc/tmpfiles.d/var-factory.conf to generate the contents of /var at boot time when the rootfs is read only and therefore the entire /var is in tmpfs.

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#9 lore raw

PATCH 1/1 system/skeleton: make /var/run a symlink to /run

Am Mi., 15. Juli 2020 um 17:38 Uhr schrieb Thomas Petazzoni <[email protected]>:

Quoted 38 lines
Quoted 9 lines · On Wed, 15 Jul 2020 15:53:12 +0200 Norbert Lange <[email protected]> wrote:
Quoted 2 lines

But... /run is already a tmpfs in sysv-based systems, so what prevents today from using /run/lttng even for sysv based systems ?

Ok, I missed that bit, thought it would have unrelated /run and /var/run directories.
So this patch is just a cleanup then, and buildroot already *always* uses a tmpfs /run (with compatibilty symlink /var/run),
and I can use /run with lttng and dbus whether this cleanup is merged or not?

Quoted 7 lines

I'd redo the patch with those removed, if you agree:

package/skeleton-init-openrc/skeleton/var/run
package/skeleton-init-sysv/skeleton/var/run

Or do nothing if you dont ;)
My new argument would be that this would prevent false assumptions (like mine).

OK, the more I look into this, the more I'm confused. To me, it seems like in both systemd *and* non-systemd cases, /var/run is already a symlink to /run, but just implemented in a different way.

For the systemd case, we have:

        ln -s ../run $(TARGET_DIR)/var/run

in package/skeleton-init-systemd/skeleton-init-systemd.dmk.

For the sysv/openrc case, we have the skeleton itself that contains:

$ ls -l package/skeleton-init-sysv/skeleton/var/run
lrwxrwxrwx 1 thomas thomas 6 26 mars  17:10 package/skeleton-init-sysv/skeleton/var/run -> ../run
$ ls -l package/skeleton-init-openrc/skeleton/var/run
lrwxrwxrwx 1 thomas thomas 6 18 mai   17:32 package/skeleton-init-openrc/skeleton/var/run -> ../run
Quoted 5 lines

So, it all cases, /var/run is already a symlink to /run. Am I missing something here ?

So, if anything needs to be done, it's just cleaning up that and doing it in the same way for all init systems, perhaps in a common place.

I thought about it some more, I think it's fine the way it is for now. Sometime later I'd remove the logic in systemd as tmpfilesd already will create the link - but only if sysv compatibility is enabled. (Still working on a systemd patch to allow tmpfilesd be executed safely on the host, then this can be cleaned up).

Quoted 6 lines

But
we need to be careful, skeleton-init-systemd.mk has some special/complex logic to handle /var and generate a /etc/tmpfiles.d/var-factory.conf to generate the contents of /var at boot time when the rootfs is read only and therefore the entire /var is in tmpfs.

Yeah.. about that...

First it's weird that a generic option (REMOUNT_ROOTFS_RW) does more than just remounting the rootfs. This should be its own option.

Then it is already broken or atleast insanely fragile.

- it kills previous files in /usr/share/factory/var
- it doesn't handle symlinks (just think of /var already containing
a symlink into that factory),
especially relative ones.
- it has sideeffects with tmpfile .confs that are ordered before and
touch /var
- it has sideeffects with other PRE_CMD_HOOKS touching /var

Whatever the reason was, that this got included, it seems to be a very narrow and specialised usecase,
and hurting everything else.

With systemd 246 it's possible to boot into a complete ro rootfs, basic services will work (course you can't write to /var). A tmpfs overlay or a script + service that creates a bind mount seems a way better solution to me.

Norbert.

PS.
I tried to ping you in regards to the fakeroot 1.24 upgrade:
http://lists.busybox.net/pipermail/buildroot/2020-May/282161.html