Implement defaultuserquota/defaultgroupquota #16283
Conversation
|
If anyone can point me at some docs on how to correctly incorporate these changes into a FreeBSD source tree, I'll happily work on making this work on that as well. |
|
@seidelma sorry no one has had a chance to take a look at this yet. I'll try to take a look at it early next week. In the meantime you may want to rebase it on master since there have been some important ZTS fixes that have gone in recently. |
|
@seidelma would you mind re-basing your commits on top of master? We've had some test suite fixes go in recently that should fix some of the failures on Fedora 39-40. |
seidelma
force-pushed
the
seidelma/feature/defaultuserquota
branch
from
b4c6c6e
to
6f20f8c
Compare
|
Hmm... checkstyle isn't happy with something ABI related. Can you try removing your ABI changes, let checkstyle run in the PR runner, and then download the new ABI files, and then include them in your commits? |
Actually, looks like the newly generated ABI files are here: Would you mind including them in your commits? |
|
To apply these sames changes to FreeBSD, from reviewing the diff, it looks like all you'll need to do is replicate the changes you made to |
This change adds the 'defaultuserquota' and 'defaultgroupquota'
properties to ZFS datasets to apply a quota to users and groups that do
not have a specific quota assigned. The default quota checking
mechanism works alongside the existing 'userquota' and 'groupquota'
checks, only taking effect if no quota is assigned for a particular
user/group. This means that it's possible to exceed a default quota
by quite a lot before the user/groupused property is updated and
further writes are denied, which was already the case for
user/groupquota.
Default quotas are implemented similarly to 'normal' user quotas,
but instead of being user properties that are preserved across
snapshots, they default back to none. NB: this is different from the
observed Solaris behavior, which is to preserve default quotas across
snapshot/clone/promote.
For instance, Solaris has:
# zfs set defaultuserquota=100M tank/fs
# zfs snap tank/fs@snap
# zfs clone tank/fs@snap tank/fs-clone
# zfs get -H defaultuserquota tank/fs-clone
tank/fs-clone defaultuserquota 100M -
Whereas this commit does:
# zfs set defaultuserquota=100M tank/fs
# zfs snap tank/fs@snap
# zfs clone tank/fs@snap tank/fs-clone
# zfs get -H defaultuserquota tank/fs-clone
tank/fs-clone defaultuserquota none default
It should also be possible to implement a default project quota using
an analogous process, if doing so makes sense.
Signed-off-by: Todd Seidelmann <seidelma@wharton.upenn.edu>
These are based on the existing userquota tests. Note that because the
default{user|group}quota behavior differs from Solaris during
snapshot/clone/promote, some of the tests fail by design.
Signed-off-by: Todd Seidelmann <seidelma@wharton.upenn.edu>
Add defaultuserquota and defaultgroupquota dataset properties to the zfsprops.7 manpage. Signed-off-by: Todd Seidelmann <seidelma@wharton.upenn.edu>
Signed-off-by: Todd Seidelmann <seidelma@wharton.upenn.edu>
seidelma
force-pushed
the
seidelma/feature/defaultuserquota
branch
from
6f20f8c
to
91b5c44
Compare
|
I've tested the FreeBSD changes against 15-CURRENT (specifically, freebsd/freebsd-src@2b4aa2816). It appears to function with the manual testing I've done, so I'm working on getting the new test cases working with the automated testing framework. |
Signed-off-by: Todd Seidelmann <seidelma@wharton.upenn.edu>
seidelma
force-pushed
the
seidelma/feature/defaultuserquota
branch
from
0426a40
to
8889165
Compare
|
Would it be useful to implement defaults for userobjquota and groupobjquota as well? |
Motivation and Context
This change implements default user and group quotas on Linux, an oft-requested feature present in Solaris 11.3.
#5431
Description
This change adds the 'defaultuserquota' and 'defaultgroupquota' properties to ZFS datasets, which apply a quota to users or groups that don't have a specific quota assigned:
It includes tests (some of which fail by design since the behavior is slightly different from Oracle's) and manpage changes.
How Has This Been Tested?
The code changes were tested on Rocky Linux 9 (x86_64 and aarch64) and Ubuntu 24.04 (aarch64) using the zfs-test suite, as well as a comparative behavior analysis between OpenZFS and ZFS on Oracle Solaris 11.4 (x86_64).
In addition, upgrading OpenZFS from 2.1.15 as well as current master/HEAD to my branch was also tested on Rocky 9.
Types of changes
Checklist:
Signed-off-by.