Re: [PATCH storage v2 0/7] add native ZFS over NVMe/TCP backend

[Joaquin Varela <joaquinvarela@neatech.ar>]

Hi Max,

thank you very much for the detailed review, and sorry that it took a
few days. I have just sent v3 ("[PATCH storage v3 0/4] add ZFS over
NVMe/TCP storage plugin"), together with the docs and manager patches.
It addresses all three points and the helper suggestions: the plugin
inherits from PVE::Storage::Plugin, LunCmd::NVMET is gone, ZFS/configfs
parsing and operation planning are local. In patch 1/4 the target runs
quoted simple commands,
sometimes joined with '&&', plus one pipe that writes the key, and the
PVE helpers replace the builtins.
Instead of dir_glob_regex, which returns only the first match, the
sysfs walks nest dir_glob_foreach.

Two parts are questions for you, explained in the cover letter: an
optional patch that guards transaction commands on the target against a
delayed command of an abandoned operation, which adds target-side
validation and a small sh guard with the only shell control flow on the
target, and a longer default
wait for the shared storage lock. Both are separate patches, so either
can be dropped without touching the other storage patches (dropping
the guard also drops one sentence and one paragraph of the docs
patch). If the size of the series gets in the way, I can split the
backend patch and split or trim the tests.

The contribution is limited to the storage backend and its integration
with PVE. It does not implement target high availability, pool takeover
or target fencing. The optional guard protects control-plane operations
on a single target too; it is not host or guest-I/O fencing. Multi-head
target compatibility remains conditional and has not been qualified.

The v3 cover Message-ID is:
<20261005002609.571-1-joaquinvarela@neatech.ar>

v3 also refuses storages that share an NVMe/TCP listener on the
target, because after a target restart they could lose their
controllers. The cover letter has the details.

Kind regards,
Joaquin Varela