nfs4 + portage(distifiles) problem

zhushazang · Tux's lil' helper Joined: 01 Dec 2005 Posts: 81

I have four machines (desktop) tha use gentoo in my network.

After first problems with nfs4's implementation two machines (nfs4 "server" and a first nfs4 "client") working correctly. But when i tried put oher "nfs4 clients" to use the same /usr/portage/distfiles occour an error. I can crete, delete, write and read inside nfs4 partition, but the error appear in the other 2 machines that will use nfs4 mounted directory.

depontius · Advocate Joined: 05 May 2004 Posts: 3509

By default, nfs is set up with the root "squashed", essentially turned into noboty. So root has less capability on a mounted filesystem than even an ordinary user. Normally this is OK - even desirable because it gives your users extra privacy, but putting distfiles on nfs is one situation where it isn't.

Look for the "root_squash"/"no_root_squash" options. Normally root_squash is the default. I believe you have to have "no_root_squash" in both /etc/exports on the server, and on the mount definition line in /etc/fstab on the client. I've never done this, so I don't know exactly. But I believe this is your problem.

I do have slight root squash problems on my systems, still. My /home is nfs4 mounted, and every now an then some ebuild or another wants to touch something in /home, and fails. (IIRC, it might be mythtv, because /home/mythtv does exist.) I have to unmount /home and repeat the emerge. For me this works, because my local /home is really mounted on /local. On the native /home are per-user symlinks to /local/(user). For me, the mythtv user is a bit of a special case, in that it's box-local. So on my nfs server home/mythtv is yet another symlink to /local/mythtv. So for me, unmounting /home gets the right files in /home/mythtv updates, because it's really updating /local/mythtv, and that's what's normally used. It's just that when /home is nfs4-mounted, root can't traverse the symlink.
_________________
.sigs waste space and bandwidth

zhushazang · Tux's lil' helper Joined: 01 Dec 2005 Posts: 81

Now working.

I needed only configure correctly /etc/idmapd.conf

The line Domain need be the same in every machine that'll use the nfs exported directory.

Thanks for all

mattst88 · Developer Joined: 28 Oct 2004 Posts: 422

For reference, the relevant bug report is https://bugs.gentoo.org/show_bug.cgi?id=318847
_________________
My Wiki page

clytle374 · Apprentice Joined: 01 Aug 2006 Posts: 221

I just developed this problem with a portage share.

I set the domain line in /etc/idmapd.conf the same on both machines, and it still doesn't work

Is there any other config for idmap? I can't find anything on it, I did find mention that it needs to be running, like in a service? If so, I can't find one.

thanks
Cory
_________________
NooB since RedHat 6.1

titanofold · Posted: Sat May 07, 2011 8:06 pm Post subject:

clytle374 · Apprentice Joined: 01 Aug 2006 Posts: 221

Thanks, I had solved this and totally forgot about this post.

Problem lies with something changing and I had to use mountvers=3 in fstab to get proper permissions.

Unfortunately I don't remember what I did to fix rpc.idmapd I think revdep-rebuild fixed it.

Thanks
Cory
_________________
NooB since RedHat 6.1

genterminl · Posted: Sun Jan 22, 2012 11:11 pm Post subject:

I'm facing a similar problem. I originally posted here. I even filed a bug, that was closed as a duplicate of the one mentioned above. I simply gave up for over a year, but now I need a larger tmpdir, which I only have on the machine I'm using as the nfs server.

In my case, on a completely unpredictable basis, an emerge, after saying "Completed installing ...." fails with a python stacktrace ending with

genterminl · Posted: Tue Jan 24, 2012 4:41 pm Post subject:

Short update. I just paid more attention to the exact error "OSError: [Errno 22] Invalid argument..." and the invalid argument is the file name ..../CHOST.nnnn and it is trying to do a chmod. By the time I look, the file is CHOST, without the numeric extension. At that point, emerge --resume has about 50% chance of completing that emerge, and doing "ebuild path/to/ebuild merge" seems to always complete successfully.

In addition, however, EVERY emerge or ebuild ends with

Hu · Administrator Joined: 06 Mar 2007 Posts: 21842

How space constrained is your build system? It seems weird that it would be more efficient to endure the performance of building in an NFS mount than to build locally.

genterminl · Posted: Wed Jan 25, 2012 3:25 am Post subject:

libreoffice. Everything else I could do locally. That's the only one I really don't have room for. However, at this point, I'm on a mission just to figure out what the problem is. I'm starting to play with nfs mount options, as I really do suspect caching of some sort. Setting noac has made performance go to &*)(&) but I have had no failures of the chown, but still get the failures to delete folders that are apparently actually empty.

elvis_ · n00b Joined: 15 Feb 2004 Posts: 24 Location: dogonfire.com

I had a similar problem and solved it by commenting out the ldap lines in

/etc/idmapd.conf on the nfs server machine. I restarted the service and it all started working again.

vostorga · Posted: Mon Mar 19, 2012 4:55 pm Post subject:

genterminl · Posted: Wed Mar 21, 2012 7:33 pm Post subject:

I never did have any LDAP lines in my idmapd.conf files. I don't recall if there were any in the template, but if so, I know I would have immediately commented them out, as I have no use for LDAP.