Re: change of zfs mount options?

Simon Burge <[email protected]> writes:

> Greg Troxel wrote:
>
>> I've had a netbsd-10 amdt64 machine for a long time, most data on zfs,
>> xen dom0.  I recently upgraded it to 11RC4, and found that
>> pkgsrc/devel/gettext-tools wouldn't build, because pkgsrc was mounted
>> noexec.
>>
>> A "zfs set exec=yes" on the dataset resolved this.
>>
>> I have no memory of this happening before, so I wonder if defaults
>> changed.
>>
>> Perhaps this is "fixes for property overrides", part of a big pullup of
>> zfs fixes (yay!) and it always should have been like that.
>
> If I had to guess, I'd suspect
> https://mail-index.NetBSD.org/source-changes/2026/03/17/msg161042.html
> or possibly
> https://mail-index.NetBSD.org/source-changes/2026/03/01/msg160827.html
>
> Are you able to easily test with either of these changes reversed?

Thanks for the pointers.

I could semi-easily test, but tl;dr is that I now think there is no bug.



I have 3 machines with zfs.  All are up to date with their branches.

which		version	initialized	exec/setuid  
desktop		10	2023-08-15	on/on
xen		11	2021-02-11	off/off
n11-test	11      2024-10-30	on/on


zpool history is a great feature.  It tells me that I set exec/setuid to
off 3 days after creation on xen, on all datasets.

I just created a test dataset on netbsd-10, and it was off/off and
mounted without noexec/nosuid.  I then set the properties to on/on and
the mount options changed.

  tank1/test-suid on /tank1/test-suid type zfs (noexec, nosuid, local)


Now, I think:

  I had intentionally set exec/setuid off long ago, as general paranoia,
  perhaps not wise.

  I am really unclear on whether this was effective or not, because most
  packages don't run binaries from pkgsrc.  I want to say it was not
  effective, simply because that seems more likely than me not having a
  problem from 2021 to last week.

  The xen machine is now correctly propagating exec/setuid flags to
  mount properties.

  My netbsd-10 machine propagates flags if I set them.

so overall I am now thinking

  I had done this on purpose and it just never caused a problem.

  I have no reason to think our zfs code in 10 or 11 is wrong about
  exec/setuid.