Quota not being enforced

[ghostrider]

Hello,

I've been looking for an answer to this and I haven't found anything, so I'm raising this post.
I have a dataset with no children that I want to set some quotas on, so for this purpose I have set the quota to 3GB. However, to my surprise when I try to extrapolate the quota by creating a file I see that the quota is not enforced and I can create files sizes with several times beyond what's defined, just as if having a quota configured or not makes no difference.

This is the quota definition of my dataset:

This is what is displayed as free space for this dataset after the quota is defined, which looks as expected:

But after the quota is set, when I try to fill a file with the contents of /dev/zero, I can extrapolate it without any problems:

root@cloud:/mnt/data # su -m www -c 'cat /dev/zero > A' ^C root@cloud:/mnt/data # ls -lah total 189 drwxrwx--- 8 www www 14B Nov 9 22:41 . drwxr-xr-x 3 root wheel 3B Oct 15 14:10 .. [B]-rw-r--r-- 1 www www 482G Nov 9 23:01 A[/B] (...)

I'm mounting the dataset inside a jail, and from there I'm trying to verify that the quota is working, although I don't think it makes any difference.


Is this supposed to happen, can anyone explain this behavior?

 

[Ericloewe]

This seems a bit extreme, but it's true that quotas are not strictly enforced to the byte. For performance reasons, quotas are a little bit fuzzy, but my understanding is that we're talking about a few MB or so. Maybe this goes up to a whole TXG's worth of data?

 

[ghostrider]

This seems a bit extreme, but it's true that quotas are not strictly enforced to the byte. For performance reasons, quotas are a little bit fuzzy, but my understanding is that we're talking about a few MB or so. Maybe this goes up to a whole TXG's worth of data?

This is not a difference in a few MB, but in the GB/TB range.
The quota, in the example above, is set to 3GB, but I'm able to create files in this dataset of 482GB simply because I stop the command and didn't let running for longer, if it kept going it would reach differences of TB.

If the quota is not enforced, what's the point?

 

[Ericloewe]

Well, the 400+ GB are all zeros, right? So they're compressing down substantially. What is the actual usage reported by ZFS?

 

[ghostrider]

Well spotted! The file is stored with zeros and that was why I was experiencing the problem that neither the quota was enforced nor the available space of the dataset was decreasing as the file grew.
If I use /dev/random instead, the quota is now enforced as the available space decreases:
root@cloud:/mnt/data # su -m www -c 'cat /dev/random > A' cat: stdout: Disc quota exceeded

I wasn't aware that this was due to compression, thanks for pointing that out!

 

[ghostrider]

Btw, while setting a quota in another mounted dataset, I get the following error:

This dataset is currently mounted in a Linux VM, could this be the cause of the error?

 

[Ericloewe]

I'm not sure why it's saying that it's invalid, is that dataset for an SMB share?

 

[ghostrider]

It's currently used as an NFS share.