I was running a webserver in a jail, following directions in the FreeBSD Handbook. I did a 'zfs umount' of the dataset thinking that I wanted to change the mount point. I changed my mind and kept the same mountpoint. I remounted the dataset, but my data is not there. There should be a jail located at /usr/local/jails/containers/lighty but there is no such directory anymore. What am I missing?

well besides the obvious thing of missing the data :-)

mns: Hidden mount. Does it show as mounted?

mason: not sure about hidden mount. How do I look for that?

what does "mount" say about it?

mason: doing 'mount' gives me this: zroot/jails on /usr/local/jails (zfs, local, noatime, nfsv4acls)

oh I did a 'zfs mount -a' and that brought everything back. What's the different between 'zfs mount zroot/jails' and 'zfs mount -a' ?

mns: So, you can have a mountpoint that sits on another mountpoint, so if the first one is missing the second one can't mount.

Alternately, if you get mounts in the wrong order, you can hide something that's already mounted.

after doing 'zfs mount zroot/jails', I did a 'ls /usr/local/jails' and it would show me the subdirectories that should have been there, but nothing underneath those directories.

mns: Sounds like you've got one dataset per jail perhaps, which is fine.

mason: I followed the Handbook when I set this up last year. docs.freebsd.org/en/books/handbook/jails/#creating-thin-jail-nullfs

Ah, bit more complex than I like.

mns: Sorry, so, in this case, zfs only controls some of the mounts, but not the nullfs mounts.

Try "mount -a" and see if that helps.

"zfs mount -a" worked. "mount -a" I don't think makes a difference in this case since I'm using ZFS.

mason: how do you manage your jails if the handbook is complex?

mns: You're doing nullfs with ZFS?

yes

zfs to create the template jails, nullfs to mount ports and ssl certs inside the jails

Anyway, to answer your question, I just have a dataset per jail. The duplication isn't huge.

so what they call thick jails ?

Relative to thin, yes.

might have to look at that. you have to update each dataset separately when updating your jails though, yes?

yes

Plus side to that, I can schedule things more easily.

I think mns is thinking that mount is recursive down the tree of possible mounts that might possibly be mounted below. But that would cause problems so it isn't done that way. One might have two datasets such that they only want one of them mounted at a time but both on the same mount point for example. So mounting is not automatically recursive down the tree.

Seems reasonable.

I'll also note that if one wants to change the mount point of a dataset that I never umount it first. Not necessary. Instead just change the mount point "zfs set mountpoint=/jails/foo zroot/jails/foo" or whatever and zfs will handle moving/umounting/mounting the mount point from a previous location to the new location. Pretty sure that works even if the mount point is busy as it moves it rather than umount'ing it.

rwp: I would just have to make sure that the mount point exists prior to setting the mountpoint yes?

mason: yeah that's a fair point about scheduling the upgrades

out of curiosity, how many jails do you run on your system, mason?

ooh, I'd have to check. Multiple systems.

Five on one box here.

Three on another.

ok so multiple jails on multiple systems

That said, more before long as I migrate some LXC containers.

you use anything to manage them? appjail, bastille, cbsd, etc? or roll your own scripts?

Just the base tools.

I have a dataset with the base set in it and minimal config, and I zfs send/receive that to new jails.

interesting. I may have to look into that sort of thing.

If you're happy with what you've got, stick with it. But I do like this. It's quite simple.

I'm just doing this because I was following the handbook. I just have one jail right now. Will have more down the pipeline. Main goal for now is to learn about jails and vnets

thanks for the help and explanations mason and rwp.

mns: Sure. Sorry my explanations were a little incoherent at first. Distracted this evening.

mns, zfs will create the mount point for you automatically. I don't think it will create parent directories so the parent directory must exist. But zfs will manage the mount point completely. Happy to help! I am also incoherent often too.

how do you set multiple 'sharenfs' lines in zfs?

rtprio, Set all of the options for a dataset in one sharenfs property.

The sharenfs property is documented in the zfsprops(7) man page.

IIUIC you can create a custom property. Maybe sharenfs within that? just guessing though. e&oe.

I just got my WayPonDEV youyeetoo X1 and can confirm that FreeBSD runs (unsurprisingly) perfectly fine on it. :)

well

Fuck.

cut(1) has a fatal flaw.

what's that

It drops the double quotes from double-quoted fields.

whatever you do with cut, you could do with awk

CrtxReavr: it doesn't: echo '"foo","bar"' | cut -d, -f1 prints "foo" with the quotes

echo '"foo,bar","fubar"' | cut -d, -f1

why would cut know to group that together

it's returning the value of the first ','-separated field, which is '"foo'

No shit?

that's how cut works, it's not a CSV parser and doesn't understand quoting

not a flaw in the command, it's just not intended for that use...

CrtxReavr: cut does what I expect in your example, it gives me "foo

if you're trying to work it that way, you'll need to use another delimiter

`tr -d '"'` would remove the quotes

(and stay posix)

CrtxReavr: So, realize that the shell eats quotes most of the time.

Your programs never see quotes or redirection.