Project Quotas in ext4

Question

This is regarding setting of Project Quotas/ Directory Quotas on ext4 filesystems. The patch for allowing Project Quotas in ext4 filesystems was submitted via patch https://lore.kernel.org/patchwork/patch/541895/ .

I have tried the below steps (followed by relevant commands): 1. Create file system on block device mkfs.ext4 /dev/nvme0n1

2. Enable Project quota tune2fs -O project -Q prjquota /dev/nvme0n1

3. Mount the block device: mount -o prjquota /dev/nvme0n1 /test

4. Enable Quotas on mount path: quotaon -Pv -F vfsv1 /test

5. Create the Project ID / Project Name for the Project Quota: echo 51:/test/first >> /etc/projects echo testproj:51 >> /etc/projid

6. Edit the Quota related to Project user: testproj edquota -P testproj

set the soft block unit to 10, hard unit to 20

7. Conform the Quota is set

repquota -avugP

Report for project quotas on device /dev/nvme0n1 Block grace time: 7days; Inode grace time: 7days Block limits File limits

Project used soft hard grace used soft hard grace

testproj -- 0 10 20 0 0 0

8. Start writing/ performing IO on said directory under /test/first

fallocate -l 10G ten.txt

9. I can see that a file with size 10G is created at said location, and running the above command in #7 I can see no change in inodes consumed.

Are there any other steps to enable quotas on ext4 file systems?

Kernel Version:4.15.0-36-generic

Answer 1

I believe you are missing a

chattr +P -p 51 /test/first

/etc/projects seems only used by XFS tools. Also /etc/projid is only there for pretty printing.

fyi this is the procedure I came up with:

(step 0 to actually create the block device:

dd if=/dev/zero of=/tmp/fs bs=1024 count=80000
losetup -f /tmp/fs
losetup -l

)

1. create a file system with big enough inodes:

mkfs.ext4 -I 256 /dev/loop0

2. enable project quotas and make sure the file system is mounted with it by default (here using extended options with -E, avoiding the mount option in your step3, but also quite sneaky as you don't see it in /proc/mounts as mounted as such)

tune2fs -Q prjquota  /dev/loop0
tune2fs -E mount_opts=prjquota /dev/loop0

3. mount it

mount /dev/loop0 /mnt/loop/

4. the quota on command does not seem useful, so skipping this one

5. set a proj id, but as a pure courtesy for the next sys admin logging in to your box. Not actually required

echo testproj:51 >> /etc/projid

6. actually make your folder part of the project (which was missing from your list)

mkdir abc
chattr +P -p 51 abc

7. edit the quota. Let's use the setquota tool, that could be used later in some ansible playbook someday, unlike edquota which runs an interactive editor:

setquota -P testproj 0 1234 0 0 /mnt/loop/

8. confirm the quota is set

repquota  -P   /mnt/loop/
# in some parsable format, assuming you wrote some simple enough strings in projid, since the xml formatter is pretty basic
repquota -P /mnt/loop/ -O xml

9. verify it works:

as a normal user:

dd if=/dev/zero of=someoutput oflag=append
loop0: write failed, project block limit reached.
dd: writing to 'someoutput': Disk quota exceeded
2471+0 records in
2470+0 records out
1264640 bytes (1.3 MB, 1.2 MiB) copied, 0.00985608 s, 128 MB/s

10. verify you can trivially escape it though, as a normal user:

chattr  -p 43 someoutput
dd if=/dev/zero of=someoutput oflag=append
dd: writing to 'someoutput': No space left on device
127427+0 records in
127426+0 records out
65242112 bytes (65 MB, 62 MiB) copied, 0.561987 s, 116 MB/s

here completely filling the file system.

EDIT: more info on the limitations of project quotas Re: Project Quota file owner could change its project ID?, Re: ext4 and project quotas bugs (/ features)