[solved] Root on nilfs - garbage collection doesn't start

[szatox]

I decided to give nilfs a shot. The trick is, this filesystem requires a garbage collection process (which runs as a userspace process) to stop it from choking.
Usually this process is started during mount, but it's not strictly required (can be disabled with options) and - apparently - it does not start if kernel can't read it because root is not mounted yet.
I can either start it later on manually, or invoke 'mount -o remount /', which also starts garbage collection at this point, even though mount options haven't changed.

Question is: is there a smarter way to do that than adding my own service that would remount root? I might go with custom initramfs again one day (and solve it there), but for now I'm building stuff with genkernel and using openrc, for the simple reason it's faster than doing everything myself.

[NeddySeagoon]

szatox,

Add it to /etc/local.d/

You need a file called nilfs.start containing the line mount -o remount /
chmod +x the file.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[szatox]

Yes, I am aware of rc.local. I don't like it because it's ugly and hackish. However, the vast majority of systems has no use of runlevels, so I guess it's good enough.
I can't really think of an actually good way to do that in userspace. It would be nice to have kernel module handle it internally, but it would certainly turn out much more hackish. And it probably is a userspace process for a reason. Even if the reason is "userspace stuff is easier to write".

Thanks

[russK]

Is it OK to periodically remount? How about a cron job?

[szatox]

Cron is not needed here, since garbage collector is a deamon process. It can be started manually, or automatically when you mount a filesystem.
Starting it automatically is a better option, because this way kernel is aware of this deamon making things more managable. E.g. GC will die when you unmount that FS so you don't have to think about it.
I suppose the problem here originates from FS not being readable _yet_ when it's being mounted for the first time (so kernel can't read and start the GC) and once root is mounted, kernel has no reason to mount it again so the event spawning GC deamon never occurs when it could be served.
With secondary mountpoints it "just works"™

Basically the best thing I can think of would be a /etc/fstab entry forcing mount -a to take it into consideration. However, mount is smart enough to only mount those filesystems that are not already mounted.

[NeddySeagoon]

szatox,

Doesn't the kernel do a remount anyway?

First mount read only, so rootfsck can run, then a remount rw?
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[szatox]

Bingo!
You gave me a good idea.
I changed a few things in openrc scripts (to print mounts at various points during boot), and entered debug mode in initramfs and I found out this little bit in /etc/init.d/root: