Request backport of bf5a54bc0e3d to 6.18.y

Request backport of bf5a54bc0e3d to 6.18.y

[Rachel Menge]

Hi stable team,

Please consider applying the following upstream commit to 6.18.y:

bf5a54bc0e3d8962474fcd611f1266eba036232f
doc/netlink: netlink-raw: Add max check

The 6.18.y tree contains these related changes:

2fd0880f0272 ("net: reject oversized tx_queue_len at netlink parse time")
982c7f66c041 ("neighbour: Enforce min/max to NDTPA_INTERVAL_PROBE_TIME_MS.")

Those changes add checks.max to the rt-link.yaml and rt-neigh.yaml
Netlink specifications. However, the 6.18.y version of
Documentation/netlink/netlink-raw.yaml neither defines len-or-limit nor
permits checks.max.

I found this while building YNL from v6.18.55. I reproduced the failure
against the unmodified v6.18.55 tag with:

   make -C tools/net/ynl/generated rt-link-user.h

YNL code generation fails JSON schema validation because the rt-link and
rt-neigh specifications use checks.max while netlink-raw.yaml rejects that
property:

   jsonschema.exceptions.ValidationError:
   Additional properties are not allowed ('max' was unexpected)

The requested upstream commit updates only the build-time YAML schema. It
does not change kernel runtime behavior.

I also checked 6.1.y, 6.6.y, 6.12.y, and 6.19.y. Those branches do not
contain the rt-link.yaml and rt-neigh.yaml checks.max additions that create
this mismatch, so this request is limited to 6.18.y.

Thanks,
Rachel Menge

Re: Request backport of bf5a54bc0e3d to 6.18.y

[Anton Danilov]

On Fri, Oct 09, 2026 at 03:36:42PM -0700, Rachel Menge wrote:
> Please consider applying the following upstream commit to 6.18.y:
>
> bf5a54bc0e3d8962474fcd611f1266eba036232f
> doc/netlink: netlink-raw: Add max check

I see the same on v6.18.55: rt-link-user.h fails since 6.18.53 and
rt-neigh-user.h since 6.18.54, and the in-tree cli.py fails the same
way for both families. With bf5a54bc0e3d on top, make -C tools/net/ynl
passes, and the generated rt-link and rt-neigh code is byte-identical
to a build of the same tree with the two checks hunks dropped.

> I also checked 6.1.y, 6.6.y, 6.12.y, and 6.19.y. Those branches do not
> contain the rt-link.yaml and rt-neigh.yaml checks.max additions that create
> this mismatch, so this request is limited to 6.18.y.

6.12.y is affected too. Since 6.12.111 it carries 45ca9f59b6c7 ("net:
reject oversized tx_queue_len at netlink parse time"), which adds
checks.max to rt_link.yaml (the file name has an underscore there).
6.12.y does not generate C code for netlink-raw specs, so the build is
not affected by this one, but cli.py --spec rt_link.yaml fails with
the same ValidationError.

bf5a54bc0e3d applies to 6.12.y as well. The 6.12.y YNL build has been
broken since 6.12.93 by handshake.yaml, so I sent a 6.12.y series that
fixes that and includes bf5a54bc0e3d:

https://lore.kernel.org/netdev/20261010035116.578385-1-littlesmilingcloud@gmail.com/

One more thing for 6.18.y, not a regression: conntrack.yaml and
nftables.yaml fail the schema check since v6.18 itself ('labels mask'
and 'set id' do not match the attribute name pattern), so cli.py fails
for these two families. Upstream fixed that in 1adc241f3940 ("ynl: fix
schema check errors"), which renames the two attributes to labels-mask
and set-id. It applies cleanly to 6.18.55, and with it and bf5a54bc0e3d
all specs in 6.18.55 pass the schema check. The C build is not affected,
these two families are not generated there.

Tested-by: Anton Danilov <littlesmilingcloud@gmail.com> # v6.18.55

---

Anton Danilov

Re: Request backport of bf5a54bc0e3d to 6.18.y

[Sasha Levin]

> Please consider applying the following upstream commit to 6.18.y:
>
> bf5a54bc0e3d8962474fcd611f1266eba036232f
> doc/netlink: netlink-raw: Add max check

Queued for 6.18, along with the schema check fix Anton suggested,
thanks.

-- 
Thanks,
Sasha