Re: linux-next: manual merge of the block tree with the drbd tree

[Christoph Böhmwalder]

Am 25.03.26 um 16:41 schrieb Mark Brown:
> Hi all,
> 
> Today's linux-next merge of the block tree got a conflict in:
> 
>   drivers/block/drbd/drbd_nl.c
> 
> between commit:
> 
>   ffa847f8ff11d ("drbd: rework netlink interface for DRBD 9 multi-peer config")
> 
> from the drbd tree and commit:
> 
>   630bbba45cfd3 ("drbd: use genl pre_doit/post_doit")
> 
> from the block tree.
> 
> These changes are both quite large and the conflicts substantial,
> ffa847f8ff11d ("drbd: rework netlink interface for DRBD 9 multi-peer
> config") in particular has a diffstat of:
> 
>  drivers/block/drbd/drbd_nl.c | 7260 ++++++++++++++++++++++++++++++------------
>  1 file changed, 5167 insertions(+), 2093 deletions(-)
> 
> which would be enormous even for a non-mechanical change, based on the
> size and a glance over the changelog it seems clear to me that it should
> be split up into a series of patches.  This would make resolving the
> conflicts much more tractable as it would be clearer what each change is
> trying to accomplish.
> 
> 630bbba45cfd3 ("drbd: use genl pre_doit/post_doit") is much less of a
> concern as while it's fairly large it is mostly mechanical.
> 
> Since I've got no confidence in my ability to do so rather than
> resolving this I have marked the DRBD driver as BROKEN for today
> (there's a half completed merge in the tree which shows how far I got)
> and I will reorder the drbd tree after the block tree going forward
> since it seems to feed into there.

Ah, sorry, that's on me, I should have called ahead.

It's a long story: the DRBD changes in drbd-next (and thus linux-next)
replay about 15 years of development history. The out-of-tree DRBD
module got de-synced from the in-tree version, and never truly caught
up. The gap just grew over the years, and we're trying to fix that now.
That's how we ended up with the enormous patch series that is now in
linux-next.

We submitted the preliminary series to linux-next to "test the waters"
in regards to integration with the mainline kernel, CI checks, etc.
This preliminary series still breaks some old userspace tooling, though,
so we can't truly submit it "officially" yet.

And that is precisely what we are working on now: we intend on starting
a new genetlink family that enables compatibility with both the old and
new userspace tools. This will take some more time, we are taking the
first preparing steps now (such as "drbd: use genl pre_doit/post_doit").
Since this is useful for the current in-tree module as well, we have
decided to submit those patches "normally" via the block tree.
And obviously that caused conflicts with the next tree, since that has
not been rebased (yet).

Jens, how do you want to handle this?
Should I send the (technically working, but maybe
old-userspace-breaking) DRBD 9 patch series to you so you can carry it
in block/for-next? So far I was under the impression that these patches
would be too large and unfinished for the block/for-next branch.

Mark, thanks for your efforts and please keep the drbd-next branch
marked as broken for now. I will rebase the tree in the following days,
and we'll see via which path it gets resubmitted.

Thanks,
Christoph

-- 
Christoph Böhmwalder
LINBIT | Keeping the Digital World Running
DRBD HA —  Disaster Recovery — Software defined Storage