Subject: Re: [vserver] ext4 inode tagging
From: Herbert Poetzl <herbert@13thfloor.at>
Date: Sun, 15 Jan 2012 02:45:48 +0100

On Sat, Jan 14, 2012 at 09:47:25PM +0100, Roberto Puzzanghera wrote:
> Hi Corey,

> [...]
>> roberto, are you wanting to use disk limits
>> (http://linux-vserver.org/Disk_Limits_and_Quota) or user/group quotas
>> (http://linux-vserver.org/Quotas) on ext4?

>> you said you were "following the howto on the wiki" and that included
>> mounting with "-o tag".  the disk limits wiki page details mounting with 
>> "-o
>> tag" (and appears to be the only wiki page that includes that:
>> http://www.google.com/search?q=site:linux-vserver.org+%22-o+tag%22) and 
>> disk
>> limits haven't worked on ext4 for some time
>> (http://list.linux-vserver.org/archive?mss:4959:201104:cnkmighjiiigipolfgel
>> and attached log).

>> but in your mount options you are using jqfmt, usrjquota,&  grpjquota, 
>> which
>> is for (journalling) user/group quotas, not "disk limits".  if you want
>> user/group quotas, then i don't believe tagging is necessary (or at least 
>> the
>> linux-vserver quotas wiki page doesn't specify such).

>> if you can clarify what you are trying to accomplish (ie "user/group 
>> quotas"
>> or "disk limits"), then your problem can probably be addressed more quickly
>> (eg if you want to do disk limits on ext4, then i could have told you it
>> probably doesn't work as testfs.sh says it "fails" from at least 2.6.35
>> through the latest 3.0.16 and following the disk limit instructions
>> probably won't enable user/group quotas).

> I'm successfully using users/groups quota on Kernel:
> 3.1.4-vs2.3.2.1 according to the wiki page you have metioned
> above and now, for the first time, I'm going to use disk
> limits. I can get disk limits working on ext3 of course.

> The testfs script seems to show that ext4 inode tagging is
> working fine, but despite of the fact that I mount the partion
> in this way inside fstab:

> /dev/sda3        /usr/local       ext4 
> data=ordered,rw,usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0,tag 
> 1 2

> as Herbert pointed out, the tagging is not recognized during
> the mounting.

> I don't have ideas on the purpose, and google searching
> produced anything, anyway tonight, during users' sleep time, 
> I will try again.

> Also I would like to know if the failed tag option recognition
> is an issue which just comes with ext4, or is an issue related
> to linux-vserver and ext4.

well, as the 'tag' option is Linux-VServer specific, it
naturally won't happen on a vanilla kernel (as there
simply is no such option in vanilla)

but I don't think the issue is ext4 related (from the
kernel PoV), otherwise the testfs script would have 
failed, as far as my guessing goes there are a bunch
of options:

 - the mount utility (for ext4) does mangle the options
   somehow, and the 'tag' is not passed to the kernel
   (most likely, would explain why ext3 works)

 - the scripts doing the mount clean up the arguments
   (on accident or purpose) and thus remove the 'tag'
   option before it can be passed to the kernel

 - the mount is done in several steps and/or involves
   several remounts, which somehow 'lose' the tag on
   the way ...

not sure how to pinpoint this though, but if it cannot
be tracked down by analyzing the scripts/sources, we
could add some kernel debug messages to print the
mount/remount/unmount requests ...

HTH,
Herbert

> Thank you, best regards
> Roberto Puzzanghera

>> corey
>> --
>> undefined@pobox.com