ZFS Quota proxy module

The project aims at providing a proxy between UNIX quota and ZFS On Linux quota. The user is able to use standard utilities such as repquota and edquota to show or edit zfs quotas.

It was originally developed for the OpenVZ6-enabled kernel here at FastVPS but includes proxy filesystem named ZQFS for the CentOS7 vanilla kernels as well.

Requirements

SPL and ZFS installed with the appopriate developer packages are required. Autoconf and gang are required to generate ./configure file.

Vanilla CentOS 6 kernel lacks necessary infrastructure and the module cannot be compiled here. Vanilla CentOS 7 kernel works well.

"Development Tools" group has to be installed as well:

yum groupinstall -y "Development Tools"

Installation

After the requirements are met and the kernel-devel along with spl and zfs are installed it is as easy as the following:

git clone https://github.com/paboldin/zfs-quota
cd zfs-quota
automake --add-missing --force-missing &>/dev/null
autoreconf -fisv
./configure
make -j
make install

By default this builds and installs kernel modules zfs-quota and zqfs for the currently active kernel. To specify another kernel use --with-linux-obj configuration option with the kernel directory as the argument.

Note that inode quotas are not yet merged upstream so one will have to pull the code from the appropriate pull request.

Using with OpenVZ

First, the regular vzdquota module must be disabled and DISK_QUOTA set to no:

sed -e '/^\s\+MODULES=.*vzdquota/ {
   s/#*/#/; p; s/vzdquota/zfs-quota/; s/#*// }' -i /etc/init.d/vz
sed -e '/DISK_QUOTA=/ { s/yes/no/; }' -i /etc/vz/vz.conf

The machine must be reboot afterwards as the vzdquota module does not support the module unload.

This will disable OpenVZ per-container (level1) quotas as well. They will be replaced with ZFS diskspace quota.

Next just deploy a VE above a ZFS instance as described in this manual.

The OpenVZ uses a simfs filesystem and appropriate device as the target for quotactl syscalls performed by quota utils executed from inside the VE. To enable container access to this interface it is necessary to allow VE to use user/group disk quota. To do this execute:

vzctl set $CTID --quotaugidlimit 1 --save

Now (re)start the VE and execute the repquota from inside it:

vzctl exec $CTID repquota -a

It should provide the output with the current disk usage. NOTE that the maximum output quota id by default equals 131072. This can be changed with zfs-quota module parameter vz_qid_limit either during module load or via file /sys/module/zfs_quota/parameters/vz_qid_limit and remounting the simfs filesystem.

Usage with ZQFS

Vanilla kernel users can use the module too though this usage requires proxy file system that has to exploit kernel in a tricky-hacky manner. In particular the ZQFS module creates an alias (almost hardlink) for a directory inode and overrides a stat call to the directory root so it matches the specified device.

That being said here is how you can use it. Just mount the zqfs type file system above the ZFS:

mount -t zqfs /dev/zqfs/mydev /mnt/zqfs -o fsroot=/zfs/data[,limit=qid_limit]

This will mount the ZQFS from the given ZFS root and create appropriate device node. The optional limit is used to specify maximum QID that will be shown for user.

TODO

• Implement vzdquota interface as a vzquota executable to set ugidlimit.