Superseded. Reposting as a reply to the original regression report:
https://lore.kernel.org/all/177349021750.3039212.10211295677877269201@eldamar.lan/
On Fri, Apr 3, 2026 at 11:30 PM Tushar Sariya <tushar.sariya77@gmail.com> wrote:
>
> From: Tushar Sariya <tushar.97@hotmail.com>
>
> The NFS automount clone path (nfs4_clone_server) is missing the
> session-size clamping that top-level mounts get in
> nfs4_server_common_setup(). This was exposed by 2b092175f5e3 which
> changed submounts to no longer unconditionally inherit the parent's
> already-clamped rsize/wsize. On servers that enforce tight
> max_request_size budgets (reported on Dell EMC Isilon/OneFS), the
> child mount ends up with raw unclamped values that exceed the session
> channel limits, resulting in NFS4ERR_REQ_TOO_BIG and user-visible EIO.
>
> Note: I was unable to reproduce the exact failure locally as it appears
> to require a server that enforces tight max_request_size budgets. The
> fix is based on code analysis — the clone path is missing the same
> session-limit clamping that top-level mounts apply in
> nfs4_server_common_setup(). Tested that the kernel builds and boots
> successfully.
>
> Tushar Sariya (1):
> NFSv4.1: Apply session size limits on clone path
>
> fs/nfs/internal.h | 2 ++
> fs/nfs/nfs4client.c | 4 ++--
> fs/nfs/nfs4proc.c | 3 +++
> 3 files changed, 7 insertions(+), 2 deletions(-)
>
> --
> 2.43.0
>