Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Locked thread
Computer viking
May 30, 2011
Now with less breakage.

Have you tried zeroing the last sector? GPT keeps a backup header there that might be detected.

Adbot
ADBOT LOVES YOU

Computer viking
May 30, 2011
Now with less breakage.

Only Shallow posted:

There are a few sysctl knobs that may take care of it.
The defaults tend to work fine these days, too.

Computer viking
May 30, 2011
Now with less breakage.

I'll give pulseaudio this much: It works much better than it used to. It's not going on my FreeBSD desktop, but I no longer remove it from Linux installs.

Computer viking
May 30, 2011
Now with less breakage.

So, the iconv changes in 10 made for a fun day of portmaster invocations. I should have given up, built packages for everything currently installed with poudriere, uninstalled everything, and reinstalled. Oh well, next machine.

Also: Thanks for the plex media server port ... I thought plex used one of the unimplemented linux syscalls? I've been running it on linux in a headless virtualbox instance, so being able to install it natively would remove some complexity.

Computer viking
May 30, 2011
Now with less breakage.

I know, I just had some issues with my poudriere machine and the iconv changes, so I thought "it's just a few ports, it's probably easier to mangle them into working directly instead of finding the appropriate way to patch ports for poudriere" .In the end, it ended up with exactly zero actual port changes (and merely a lot of de/re-installing and cleanup), so things have gotten much better since I tried a week or three ago.

And 9 only? Hm, I can't remember what that machine runs. It might actually be 9.1.


Today's project: zfs send/receive a 4-disk zpool onto a new 4k-blocksize pool. Swap disks between the server and the eSATA enclosure. Try to boot from the new disks. If successful, wipe old disks, and add them to the new pool with 4k alignment. Finish by adding the so-far untouched disks from the other enclosure to the pool as well. The old disks are 4k-capable but the pool they're in is 512B-aligned; I thought I should clean that up before I add more disks.

As an estimation it'll be done around midnight, I guess I'll do the rest on Monday.

And yes, I do have backups.

Computer viking fucked around with this message at 11:42 on Sep 13, 2013

Computer viking
May 30, 2011
Now with less breakage.

alyandon posted:

Will FreeBSD 10 have an installer capable of installing the OS on a root ZFS volume?

Yes and no. Going by the current (sic) releases, you can choose to set up partitions by hand in a shell, which isn't too hard ... but there is still no nice curses-based way to do ZFS.

Computer viking
May 30, 2011
Now with less breakage.

One would indeed hope so. Poudriere is capable of doing it, they have the hardware, and it would be silly to do a release without packages in the repository...

Computer viking
May 30, 2011
Now with less breakage.

Side note: God, I hate the common-in-linux thing where their stable server releases have massively outdated packages (possibly "massively outdated packages with security backports"). Samba4 on ubuntu server, you say? Nooo. You see, replacing a non-working version with something that worked would be too disruptive.

(There is a bug. It's marked as fixed. The version with the fix isn't in the LTS release, which is what the original reporter was using. Ugh.)

... what I'm trying to say is that I rather appreciate having current ports at hand.

Computer viking
May 30, 2011
Now with less breakage.

EvilMoFo posted:

Solaris hasn't been BSD based for quite a while now, not sure why this would be the place to ask. :colbert:

Well, there's probably a decent overlap of users between the two non-linux noncommercial ZFS-having OSes. I can't help, though; I haven't used anything solaris-based more than very sporadically many years ago.

Computer viking
May 30, 2011
Now with less breakage.

feld posted:

There has been a public test repo for a while

http://pkg-test.freebsd.org/pkg-test-${ABI}/latest

The 10.0-RELEASE should have zfs working in the installer. It's being tested.

The 10.0-RELEASE will also have signed packages in the repo


As of yesterday:

Bryan Dewery (as portmgr) posted:

We are pleased to announce that official binary packages are now
available for pkg, the next generation package management tool for FreeBSD.

Pkg allows you to either use ports with portmaster/portupgrade or to
have binary remote packages without ports.

We have binary packages available for i386 and amd64 on
8.3,8.4,9.1,9.2,10.0 and 11 (head).

Pkg will be the default starting in FreeBSD 10.

The pkg_install suite of tools pkg_create(1), pkg_add(1), and
pkg_info(1) (which ports also use), are deprecated and will be
discontinued in roughly 6 months. A communication regarding the
deprecation of the pkg_install suite of tools will be sent separately in
the future.

If you are currently not using pkg and wish to, run the following as
root. Be sure not to add WITH_PKGNG=yes to your make.conf until after
pkg is installed.
code:
  # cd /usr/ports/ports-mgmt/pkg && make install clean
  # echo WITH_PKGNG=yes >> /etc/make.conf
  # pkg2ng
You can now either continue to use ports with portmaster/portupgrade, as
before or switch to using binary packages only.


To use binary packages:
  1. Ensure your pkg(8) is up-to-date. pkg -v should say at least
    1.1.4_8. If it does not, first upgrade from ports.
  2. Remove any repository-specific configuration from
    /usr/local/etc/pkg.conf, such as PACKAGESITE, MIRROR_TYPE, PUBKEY.
    If this leaves your pkg.conf empty, just remove it.
  3. mkdir -p /usr/local/etc/pkg/repos
  4. Create the file /usr/local/etc/pkg/repos/FreeBSD.conf with:
    code:
    FreeBSD: {
      url: "http://pkg.FreeBSD.org/${ABI}/latest",
      mirror_type: "srv",
      enabled: "yes"
    }

* Note that pkg.FreeBSD.org does not have a browsable web page on it and
does not have a DNS A record. This is intended as it is an SRV host.
pkg(8) knows how to properly use it. You can use 'pkg search' to browse
the available packages in the repository.

Mirrors you may use instead of the global pkg.FreeBSD.org:

pkg.eu.FreeBSD.org
pkg.us-east.FreeBSD.org
pkg.us-west.FreeBSD.org

Your system is now ready to use packages!

Refer to the handbook section on pkgng for usage at
http://www.freebsd.org/doc/en_US.ISO8859-1/books/handbook/pkgng-intro.html.
Also see 'man pkg' for examples or 'pkg help'.


Packages are built weekly from a snapshot of the Ports Collection every
Wednesday morning 01:00 UTC. They typically will be available in the
repository after a few days.

Pkg 1.2 will be released in the coming month which will bring many
improvements including officially signed packages. FreeBSD 10's pkg
bootstrap now also supports signed pkg(8) installation.


Regards,
Bryan Drewery
on behalf of portmgr@

Computer viking fucked around with this message at 15:56 on Nov 1, 2013

Computer viking
May 30, 2011
Now with less breakage.

rhag posted:

I fixed it by manually creating the a (for /) and the b (swap) partitions in the BSD partition. I don't think it was a bug but that it was just me having different expectations i guess.

For later reference: That's an MBR partition, not a DOS partition. (The latter kind of implies a FAT format.)
More to the point, yeah ... that confused me at some point as well. I tend to use GPT these days, so I hadn't seen how it works with the new installer - but it's always been A Thing.

Computer viking
May 30, 2011
Now with less breakage.

Leb posted:

Man, I still remember when FBSD 10 was just a distant pipe dream. Kudos to the devs.

4 to 5 was the first system upgrade I can remember having done, so accidentally upgrading to 11 today was ... yeah. It's been around for a while. :)

Computer viking
May 30, 2011
Now with less breakage.

Broken Machine posted:

I know that, I'm just not sure if I have to play around with the linker or whatever to get it to build and work correctly, and so I didn't really feel like messing with it and didn't.

On amd64 you'd have to do some extra work to not build with clang; it has effectively been the only compiler supported there for a while. I mean, it probably works with GCC, since some of the other archs need it - but it's GCC that is the "play around with to make it work" choice now.

Computer viking
May 30, 2011
Now with less breakage.

So, I'm trying to set up NFS 4 with a FreeBSD server and a mix of FreeBSD/Ubuntu clients, and windows 2003R2 as the kerberos server (and user/group db, through winbind).

... I don't have any questions or anything, I just want to leave some evidence in case there's an inquiry.

Computer viking
May 30, 2011
Now with less breakage.

Leb posted:

OK, so one thing I'm still a little unclear on... on 10, where exactly does the ports tree fit in with the new world order? Once built, are ports "as good as" binary packages or are there still operative differences when it comes to interactions with pkg?

They should still be treated as equal.

Computer viking
May 30, 2011
Now with less breakage.

Sir_Substance posted:

Well, here's a rookie question. I assumed that ports would be the preferred option for installing. Long installation time notwithstanding, I just kind of assumed that the BSD community would see compiling from source on the machine to be a superior option to downloading pre-compiled binaries.

Would people be recommending using pkg_add rather then ports under most circumstances?

I think the idea is to use binary packages when you can get them signed, and you're on a stable version (so not -current), and you don't need to change the options. There will probably be more subpackages and versions and such available after they drop support for the old pkg_ tools and start adding things it couldn't handle.

You might also want to compile the most CPU intensive things yourself just to get them optimized for your CPU arch, though I think the difference is smaller on amd64 than on i386 (since the lowest common denominator for amd64 is still reasonably modern).

Computer viking
May 30, 2011
Now with less breakage.

Riso posted:

Samba probably doesn't come with ADS built on BSD for the same reason Red Hat does.
Sambas ADS simply doesn't work with any ldap but the one it comes with.

Samba as an LDAP/ADS server or as a client? The former is really picky, but the latter seems to tolerate AD and whatever.

Computer viking
May 30, 2011
Now with less breakage.

evol262 posted:

It's the opposite, really. As a server, it pretends to be an NT4-style domain (except for Samba4, which peers with 2012R2). winbind will happily talk Kerberos.

Just to confirm, I use our 2003R2 - based AD domain for authenticating users on samba 3.6 . Not too bad to set up, and even winbind+pam for system users was perfectly doable.

Computer viking
May 30, 2011
Now with less breakage.

On an unrelated note, the new native iSCSI target seems to work fine for my very limited needs - I exported a zvol from the fileserver at home (a 4-disk raidz, partition of the boot SSD as ZIL), mounted it in windows, and since things seemed to work fine, installed titanfall (48GB? :catstare: ) on it, and generally didn't notice the non-local part of it at all. It was even fairly easy to set up.


On the downside, I have a sudden urge to spend money on some 10GB NICs.

Computer viking fucked around with this message at 11:53 on Apr 22, 2014

Computer viking
May 30, 2011
Now with less breakage.

That almost sounds like you've managed to accidentally mix the packaging systems ... though if you used "pkg upgrade" to upgrade it, that would be rather odd.

Computer viking
May 30, 2011
Now with less breakage.

Just for clarity, a raidz / mirror / single disk is a vdev, not a pool - the pool is the entire unified bucket of storage.

With that out of the way, you want to set up a few different vdevs (why not identical ones?) and three-way mirror over those, giving you useful capacity equal to the useful capacity of the smallest of the three mirrored vdevs? I don't think there's a simple way to do that. However, you can get reasonably close by adding the different vdevs to a pool as usual (this will give you a pool with capacity that's the sum of the vdevs in it), and then setting copies=3 . If I'm reading this right, ZFS will try to put the redundant copies on different vdevs - which ought to create a similar effect.

edit: I was bored at work. Something like this:


editagain: If it helps, you can also add more vdevs to a pool later. Any new data will preferentially go to the least full vdev in the pool (in an attempt to balance them). This means you can stuff another stack of (ideally identical) disks into the machine and expand onto them. I have no idea what copies=N will do in that case.

Computer viking fucked around with this message at 16:18 on May 9, 2014

Computer viking
May 30, 2011
Now with less breakage.

hifi posted:

Set up ssh (if you didn't enable it in the installer) and use scp

Exactly this. On the client side, WinSCP works well on Windows (though the interface is pure 1997), and I think cyberduck is fine on a Mac - or the command line tools. On Linux I use the command line tools , but you can also do fish:// or sftp:// in any KDE program (including dolphin).

Computer viking
May 30, 2011
Now with less breakage.

Ninja Rope posted:

How's the current maintainer of FreeBSD's pf doing? Is he able to keep up or does he need help?

There's a discussion in the -current mailing list about importing a more current version; so far no one has volunteered. Bapt looked at it and decided it's ... complicated, since the current version has things like multithreading that aren't in the OpenBSD version.

Computer viking
May 30, 2011
Now with less breakage.

I don't have any control over the ancient AD domain i authenticate against, so I'm using the uid mapping. As long as the username/uid mapping is stable, it should be fine - the only issue I had with differing uids between the machines was the nfs3-mounted file server.

Moving to nfs4 solved that, though that's not absolutely trivial either. Or, to be exact: it seems to work fine without being to onerous to set up now, with a recent freebsd serving ubuntu 14.04 clients. With earlier versions it was a deep well of things that should, but didn't, work. (I haven't set up sec=krb5 yet, though. That looks like it'll be "fun".)

Computer viking
May 30, 2011
Now with less breakage.

alyandon posted:

Nope, not at all unique. We have a random assortment of FreeBSD machines (v8 through v10) for some very specific tasks among a seemingly endless sea of RHEL/Centos/Windows boxes that are all tied into Windows AD for authentication and authorization.

I'll ask someone in our unixsys group about how they are pulling off the AD integration on our FreeBSD boxes and update the thread if I get a coherent answer.


I've got AD (2003R2, yay) and random FreeBSD/linux boxes. I did, however, solve it by not matching uids/gids: I use winbind for auth, and NFS4 (sending usernames, not uids, over the wire) for file shares. It would probably be an issue if I, say, made a tar file on one machine and extracted on another - but that hasn't come up so far.

Computer viking
May 30, 2011
Now with less breakage.

Huh - maybe I've been building it by hand and reflexively checking the AD box every time? Plausible enough.

(There is a check box for it, right?)

Computer viking
May 30, 2011
Now with less breakage.

You might be able to set it in /boot/loader.conf instead? (Provided you also load the Linux module there, or have it compiled in, presumably).

Computer viking
May 30, 2011
Now with less breakage.

If you can't set it in loader.conf, how about depending on the sysctl script (if it's implemented as an rc service, ofc)?

Computer viking
May 30, 2011
Now with less breakage.

In other news, I finally, finally got NFSv4 with sec=krb5 working. Apart from all the services and config setup that has to be right, it helps to compile a kernel with GSSAPI. Without it, everything appears to work*, but reading/writing files always fails with I/O errors.


* I did finally notice an error in the messages log that sent me in the right direction.

Computer viking
May 30, 2011
Now with less breakage.

Oh, neat. Nothing I remotely need, but that's sort of secondary.

In other news, kerberized nfs4 (with a windows 2003r2 server I have no control over as the kerberos server) is completely mystifying. It stopped working overnight and nothing I've done has worked, short of removing sec=krb5 from the mount options.

Oh well, there's always next week.

Computer viking
May 30, 2011
Now with less breakage.

Is net.inet.ip.forwarding=1 ?
I have something similar (that works) set up at work, so I can take a look tomorrow.

In other news I've almost finished writing an nss module that forwards passwd and group queries over TCP , so you can get identical username:uid and group:gid mappings. I'd love it if the AD domain at work upgraded to a functional level that includes working uid/uid fields, but in the meantime etc.

Computer viking
May 30, 2011
Now with less breakage.

Try "route add default 10.0.0.x" in the jail, where x is the IP of the host. If that helps, you can set defaultrouter="10.0.0.x" in the jail's rc.conf.

Also, read this for more details: http://www.freebsd.no/doc/handbook/network-routing.html

Computer viking fucked around with this message at 10:37 on May 6, 2016

Computer viking
May 30, 2011
Now with less breakage.

Oh right. Maybe you need to pass port 80 from any to your external IP?

Computer viking
May 30, 2011
Now with less breakage.

Nystral posted:

antidotal evidence
You know, that makes at least as much sense as "anecdotal".


Unrelated, I have an NFSv4 problem. Granted, solving a problem with NFSv4+krb5 seems to be like regex ("now you have two problems"), but I think I've got it very nearly sorted.

I have a FreeBSD 10.3 fileserver, a FreeBSD 10.3 client, base kerberos (so heimdal), and a windows server 2003 R2 domain as the kerberos server. Following this it's actually working fine ... for some hours, and then the mounts completely die on the client: All IO to the NFS mount including umount gets stuck permanently in rpccon, and the only real way out is to reboot the client. It works fine for a while again after rebooting. Of course, there isn't a single word about this in the logs on either machine.

At a random guess, there is a kerberos ticket timing out in there somewhere ... but which one, and how do I renew it?

Computer viking
May 30, 2011
Now with less breakage.

To follow up on that: Updating the file server to 11 seems to have fixed it. In hindsight I suspect it's because mountd was changed to use the -S flag by default ("don't break horribly when updating the exports list, but possibly hang for a bit instead") , but I'm not about to roll back to 10.3 to test that theory.

Adbot
ADBOT LOVES YOU

Computer viking
May 30, 2011
Now with less breakage.

Generally speaking you could just put them in the host fstab, mounting "into" the jail directory. It'll be fun making the user ids match, but you'd kind of have that problem anyway.

If FreeNAS has implemented it on top of ezjail, I think it will read and mount from /etc/fstab.jailname (or something like it) when starting the jail, but there isn't anything like that for plain jails.

Of course, there might also be a separate FreeNAS solution; I wouldn't know.

  • Locked thread