IRC Logs for #crux-devel Tuesday, 2015-03-17

*** ryuo has quit IRC00:33
*** Feksclaus has quit IRC01:11
*** mavrick61 has quit IRC03:50
*** mavrick61 has joined #crux-devel03:51
*** nilp has joined #crux-devel05:53
*** nilp has quit IRC06:18
*** novak_ has quit IRC06:53
*** novak has joined #crux-devel07:00
juefrinnst: sure, rpcbind was always needed for nfs09:03
juefrinnst: it's the successor for core/portmap as of CRUX 2.709:03
frinnstit "works" to mount without it09:04
frinnstsince i've obviously done it before :)09:04
frinnstanyways, sorry for the noise09:05
juenp, good to know that it works now09:10
juewill try to improve that a bit, either a readme or perhaps a check in the nfs rc-script?09:13
frinnstshouldnt be needed. a crux user should know this :-)09:29
*** Feksclaus has joined #crux-devel10:18
frinnstteK_: cryptsetup 1.6.6 has been available for ~7months or so12:30
rmullfwiw there is another issue with nfs that pidsley (from #crux) and I are experiencing that changed the required mount options and broke nfs mounts unless nfs-utils was rolled back to the previous version or mount options were changed14:35
rmullNot sure if it's a big deal because it's fixed by changing mount options, but it was unexpected and I just wanted to make sure you guys were aware of it14:35
rmullcan provide more details if asked14:36
jaegerforgive me if I missed it but what IS the problem? There are no details here14:45
rmullI was sparse just in case someone was already aware of it, sorry14:50
rmullThe assesment is that with the nfs-utils 1.3.1 -> 1.3.2 upgrade there is an additional mount option (for nfs4 mounts, at least)14:51
rmullIf no vers= option is given, it defaults to 4.2 which on my system gives "unknown mount option" or along those lines14:51
jaegerIf you mess with the 4.x versions the kernel supports does that change?14:52
rmullI gave it a vers=4.0 option, which made it work - pidsley simply rolled back to nfs-utils 1.3.1 and his resumed working14:52
rmullBoth my machines (server, client) have the latest nfs-utils, and both have built-in kernel support for nfs4 (but not nfs 4.1). I didn't see a kernel options for 4.214:52
rmullSo we're unsure where the mount without vers= gets the 4.2 from, but it can be fixed by being explicit14:53
jaegerAnything in the release notes or docs about the change?14:53
rmullI haven't checked, all information was gathered experimentally, and google didn't turn anything up. I didn't go directly to the project pages though.14:54
rmullSo there may be14:54
rmullI guess I'm not really presenting a solution other than specifying vers= mount option, just offering a "heads up"14:54
jaegerJust wondering if it's intentional behavior14:55
rmullYeah, that's reasonable. I'll take a quick glance and see if anything stands out, hold on14:55
rmullYes, looks like there is some information about it14:59
rmullAlthough I'm not sure if it's working the way it's described15:00
rmullIt says there is file called nfsmount.conf that can be used to specify minor version15:00
rmullAnd that there should be some autonegotiation logic15:00
rmullBased on this text it would seem that it should try older and older versions until one works15:01
rmullIf nfsmount.conf is not available, which on my system it is not15:01
rmullALternatively, it can be built with --enable-mountconfig=no, which would maintain the previous mount behavior and not look for the nfsmount.conf file15:01
rmullOr, sorry, that file may still be relevant, but it would act as if nfsmount.conf doesn't set the Defaultvers= for the mount15:02
rmullYou can check it out at
rmullcommit f980298853d9b12a222421342418732f65883c3015:02
jaeger;a=commit;h=f980298853d9b12a222421342418732f65883c30 <-- already there :)15:03
rmullCool. You can use your best judgment about what the best path moving forward for crux is - installing the nfsmount.conf file or something else15:04
rmullThansk for listening :)15:04
jaegerIf you create nfsmount.conf and specify 4.2 as the default does it correctly try 4.2, then 4.1, then 4.0, and then 3?15:07
rmullWill try, one moment15:08
jaegernfsmount.conf(5) is under-documented,15:08
rmullInteresting format15:10
rmullSo I have /etc/nfsmount.conf with a [ NFSMount_Global_Options ] section, and underneath that, indented by tab, I have Defaultvers=4.215:13
rmullmount -v tells me that it tries 4.2 and only 4.215:14
jaegerjue: perhaps the sample nfsmount.conf should be installed with nfs-utils. Most of it is commented by default15:14
rmullIf I change that arg to 4.0, it works15:14
jaegerthe default (commented) is Defaultvers=4, for what that's worth... but since it's commented the behavior isn't obvious, I guess15:14
rmullI'm not sure what the difference is between the value that comes after Defaultvers= and what they mean by "arg option"15:14
rmullDefaultvers=4 mounts properly15:15
rmullBut I can't test if that also works with NFS 4.1 because I don't support that in my kernel15:15
rmullAccording to the chart, it should not work with 4.115:16
jaegerI assume the "arg" part refers to the mount options15:16
jaegerso if you have Defaultvers=4.2 in nfsmount.conf and you specify -o v4 you won't get the check for v315:16
jaeger(I haven't tested it but that's how I read it)15:17
rmullIn that configuration, mount -v shows vers=4.2 and a failed mount15:17
rmullI haven't been able to verify that any "fallback logic" works yet15:18
jaegerdo you have Defaultvers= and Nfsvers= both defined?15:18
rmullHow do you recommend I set it?15:19
rmullIf I set it to 4.0, mount -v shows -o nfsvers=4.0 and there is no vers= pair, and mount works15:20
jaegerin terms of mount output or /proc/mounts I believe nfsvers= and vers= are the same, just one is deprecated15:25
rmullwell, I'll be around with my nfs server running if you want me to test anything, but I need to get some work done15:26
rmullI wish all projects had IRC channels that we could just pop into and say "whats up with this behavior"15:26
rmullalthough maybe it's better that there isn't.15:26
jaegerI'm at work and back and forth so a bit disjointed right now15:27
rmullYep, take your time, I suspect there may be something going on with upsteram15:32
rmullMaybe after working hours I'll try to see what happens in the source code15:32
jaegerif I mount with just "mount" and specify "-o nfsvers=4", I get a v3 mount, oddly15:33
rmullHaha, what fun.15:33
jaegerIf I mount with "mount.nfs4" and specify "-o nfsvers=4", I get a v4 mount15:33
jaeger(this is with no nfsmount.conf file)15:33
rmullSame with me15:34
jaegerIf I then install the example nfsmount.conf file and uncomment only the Defaultvers=4 line, it mounts with v4 succesfully15:34
jaegerI don't have 4.1 support in this kernel to test at the moment but 4.0 works15:35
rmullBut v4 only gets you 4.0, so that's not adequate15:35
rmullAssuming fallback is desired15:35
rmullBut it's a datapoint15:35
jaeger$ egrep -v '^(#|$)' /etc/nfsmount.conf15:35
jaeger[ NFSMount_Global_Options ]15:35
jaegeryeah, just gathering data now15:35
jaegerwhat kernel version are you running?15:37
rmull3.18 on the server and 3.19 on the client15:38
jaeger3.14.x here, going to add 4.1 and 4.2 support to my kernel15:39
rmullI also cant unmount the volume locally from my nfs server because the kernel nfs server is using it16:00
rmullcan't seem to shut down the nfs server daemon enough to umount the volume16:00
jaegershouldn't that be expected? If the volume is shared actively you can't unmount it?16:01
jaegeror am I misunderstanding what you're trying to do?16:01
rmullI have it unmounted on all nfs clients, I wanted to unmounted it locally on the nfs server so that I could do an xfs_repair on it16:01
rmullBut I can't because fuser says knfsd is using it16:02
rmullThe crux init scripts don't seem to be able to shut it down16:02
rmullI'm just going to reboot the machine16:02
jaegerif you kill rpc.nfsd and rpc.mountd yourself?16:02
rmullBecause the volume doesn't get mounted on boot anyway due to vgscan issues -_-16:02
rmullMy shit is so broken, haha16:02
jaegerSounds like there are some issues, yeah16:03
jaegerOne thing at a time!16:03
rmullI'll bother you guys about the vgscan thing once the nfs stull gets squared away16:03
juejaeger: yes, good idea, will add the sample file16:59
jueI have 4.1 enabled on both client and server and get that as the default mount without any configuration17:00
jueserver is 3.14.x client 3.19.117:01
jaegerok, with 3, 4, 4.1, and 4.2 support all in my kernel I can mount the test export with or without nfsmount.conf18:06
jaegerthe server in this case only supports 3 and 4, I'll have to test 4.1 and 4.2 elsewhere, I guess18:06
rmullSo the client needs to support everything up to and including 4.2 to be able to do the fallback properly?18:27
jaegerpossibly. I need to test again on a kernel that doesn't support 4.1 and 4.2 to a server that does18:31
jaegerhaving problems with my client right now, though18:31
jaegerrpc.statd won't start on my client for some reason18:31
juerpcbind running?18:33
jaegeryeah, though I'm going to restart it because rpcinfo -p doesn't work either18:34
jaegerok, looking better now. Not sure what happened18:34
jaegerok, so some more data with CRUX as both server and client18:36
jaegerboth machines have 4.1 and 4.2 support18:36
jaegerI can mount correctly versions 3, 4.0, and 4.1. I've no idea how to mount 4.2, that doesn't seem to be an option. Maybe just a different behavior of the 4.1 support?18:37
jaegerspecifying 4.2 directly results in "Protocol not supported"18:37
jaegeralso, if I have nfsmount on the client and I specify Defaultvers=3, I get version 318:37
jaegerif I have nfsmount.conf on the client and specify Defaultvers=4, I get version 4.018:37
jaegerif I remove nfsmount.conf I get version 4.1 by default18:37
jaegernow I'll remove 4.1/4.2 support from the server and test again18:38
juejaeger: same here with 4.2 support18:43
juedunno why not, but a 'cat /proc/fs/nfsd/versions'18:46
jue-2 +3 +4 +4.1 -4.218:46
jaegerthis is also interesting:18:50
jaeger# CONFIG_NFS_V4_1 is not set18:51
jaeger-2 +3 +4 +4.1 -4.218:51
jaeger4.1 seems to work even without 4.1 support in the kernel18:51
jaegerok, it seems the auto-fallback does work for me with nfsmount.conf19:03
jaegerI disabled 4.1 on the server by adding "-N 4.1" to the nfsd command options19:03
jaeger$ sudo cat /proc/fs/nfsd/versions19:03
jaeger-2 +3 +4 -4.1 -4.219:03
jaegerDefaultvers=4.2 in nfsmount.conf19:04
jaegertype nfs4 (rw,relatime,vers=4.0,19:04
jaegerif I remove nfsmount.conf at this point it still mounts without trouble, version 4.019:05
jaegerrmull: not sure what to say at this point, it seems to work properly for me19:14
juesame for me, never had any problems to mount shares from a v4 enabled server19:15
juebut I have v4 enabaled since a long time and never tested with a v3-only server19:16
jaegerthat was between two CRUX machines... falling back now to trying network storage here at work which is oracle/sun ZFS19:17
jaegerslightly different behavior there.19:18
jaegerif nfsmount.conf exists and has either Defaultvers=3 or Defaultvers=4 in it the mount works19:19
jaegerif nfsmount.conf does not exist I get the "an incorrect mount option was specified" error19:19
jaegerif nfsmount.conf has 4.1 or 4.2 in it the same error occurs. So it seems the auto-fallback does NOT work when mounting an export from my oracle storage. It DOES work when mounting an export from CRUX19:20
juejaeger: guess it works too if you use -o vers=4.0 as mount option without having a nfsmount.conf?20:26
jueso a save default for us would be to have a nfsmount.conf with Defaultvers set to 420:31
jaegerThat seems reasonable to me20:31
rmulljaeger: Thanks very much for testing20:42
rmullyou da man20:42
rmullSo crux will be shipping the nfsmount.conf file as the only change? I will retest once the new port is out20:43
rmullThough it would probably be more useful to test prior20:43
rmullI will test.20:43
rmullWhat's the install line I should add to the Pkgfile (if you have it handy?)20:44
jaegerinstall -m 0644 utils/mount/nfsmount.conf ${PKG}/etc/ <-- I think that should do it20:45
rmullPerfect, thanks20:45
jaegerwell, that will install it, you'd still need to uncomment the defaultvers line20:45
rmullLooks good, automagically inserts vers=4.0 as desired20:49
rmullAny other information needed from me?20:49
jaegerIf that solves the problem, I don't think so20:49
rmullGreat, thanks jaeger / jue20:49
*** nilp has joined #crux-devel22:01
jaegerglad to help22:27
*** nilp has quit IRC22:51
jaegerhrmm... might put crux on my surface pro 223:41

Generated by 2.11.0 by Marius Gedminas - find it at!