linux.git/fs/nfsd, branch master

nfsd: export NFSv4 callback op stats via netlink

2026-08-10T13:54:35+00:00

Add a proc4cb-ops nested attribute to the server-stats netlink dump,
reusing the existing server-proc-entry (op/count) layout. The dump
gains a callback section that emits one entry per callback opcode
(OP_CB_GETATTR..OP_CB_OFFLOAD) from the per-netns callback counters,
paged across messages like the other per-operation sections.

This lets nfsstat report NFSv4 backchannel operation counts over
netlink, including CB_GETATTR which corresponds to the procfs
wdeleg_getattr line.

Assisted-by: LLM
Signed-off-by: Jeff Layton 
Link: https://patch.msgid.link/20260717-exportd-netlink-v7-6-b7ce17b83b60@kernel.org
Signed-off-by: Chuck Lever

nfsd: count NFSv4 callback operations per netns

2026-08-10T13:54:35+00:00

The NFS server tracks per-operation call counts for the forward channel
(proc4ops) but keeps no statistics for the NFSv4 backchannel (callback)
operations it sends to clients.

Add a per-netns array of percpu counters for callback operations, indexed
by RFC 8881 callback opcode (OP_CB_GETATTR..OP_CB_OFFLOAD), and bump the
relevant counter in nfsd4_run_cb(), which is hit exactly once per callback
that is actually queued.

CB_GETATTR is sent when a GETATTR conflicts with an outstanding write
delegation, which is roughly what the dedicated wdeleg_getattr counter
tracked. The two are not identical: the old counter incremented on every
such conflict, whereas the CB_GETATTR counter only counts callbacks that
are actually queued, so concurrent conflicts that coalesce onto an
already in-flight CB_GETATTR are now counted once rather than once per
conflict. Report the procfs "wdeleg_getattr" line from the CB_GETATTR
counter and drop the now-redundant NFSD_STATS_WDELEG_GETATTR counter, its
helper, and its increment site.

Assisted-by: LLM
Signed-off-by: Jeff Layton 
Link: https://patch.msgid.link/20260717-exportd-netlink-v7-5-b7ce17b83b60@kernel.org
Signed-off-by: Chuck Lever

sunrpc: remove unused svc_version vs_count field

2026-08-10T13:54:35+00:00

Now that svc_seq_show() and the nfsd netlink stats handler both use
the per-netns svc_stat vs_count arrays, the global per-version
vs_count percpu counters are no longer read by anything. Remove the
vs_count field from struct svc_version and all the associated
DEFINE_PER_CPU_ALIGNED arrays and initializers across nfsd, lockd,
and the NFS client callback service.

Assisted-by: LLM
Signed-off-by: Jeff Layton 
Link: https://patch.msgid.link/20260717-exportd-netlink-v7-4-b7ce17b83b60@kernel.org
Signed-off-by: Chuck Lever

nfsd: implement server-stats-get netlink handler

2026-08-10T13:54:35+00:00

Implement nfsd_nl_server_stats_get_dumpit() which exposes the
NFS server statistics currently available via /proc/net/rpc/nfsd
through the nfsd generic netlink family.

The handler uses a dump operation to stream statistics across one
or more netlink messages. The reply is divided into sections that
are emitted in order:

  - scalar stats (reply cache, filehandle, IO, network, RPC),
    emitted once in the first message, then

  - per-version procedure counts (proc2/3/4-ops) and the NFSv4
    per-operation counts (proc4ops-ops), using the per-netns
    vs_count arrays.

cb->args[0] tracks the current section and cb->args[1] the entry
index within it, so a section that does not fit in the current
message is closed and resumed in the next one. This matters because
the first dump message is allocated at NLMSG_GOODSIZE (a single page
on most architectures) regardless of the client's receive buffer;
packing every counter into one message would overflow it and fail
the dump with -EMSGSIZE. Userspace merges the attributes from every
message.

This allows nfsstat to retrieve server statistics via netlink
with a procfs fallback for older kernels.

Assisted-by: LLM
Signed-off-by: Jeff Layton 
Link: https://patch.msgid.link/20260717-exportd-netlink-v7-3-b7ce17b83b60@kernel.org
Signed-off-by: Chuck Lever

sunrpc: add per-netns per-procedure call counts to svc_stat

2026-08-10T13:54:35+00:00

The existing per-procedure call counts live in global
svc_version->vs_count[] arrays which are not network-namespace-aware.
Add per-netns equivalents in struct svc_stat so the upcoming netlink
stats interface can return namespace-scoped statistics.

Add a vs_count pointer array to struct svc_stat, along with
svc_stat_alloc_counts() and svc_stat_free_counts() helpers to manage
per-version percpu call count arrays.

Increment the per-net counter alongside the global one in
svc_generic_init_request(). Call the alloc/free helpers from
nfsd_net_init() and nfsd_net_exit().

Assisted-by: LLM
Signed-off-by: Jeff Layton 
Link: https://patch.msgid.link/20260717-exportd-netlink-v7-1-b7ce17b83b60@kernel.org
Signed-off-by: Chuck Lever

NFSD: Eliminate percpu counter contention in IO byte accounting

2026-08-10T13:54:35+00:00

nfsd_stats_io_write_add() and nfsd_stats_io_read_add() accumulate
byte counts in per-net-namespace and per-export percpu_counters
using percpu_counter_add(), which applies the default batch
threshold of max(32, 2*nr_cpus).

For a 4 KB NFS WRITE, the amount (4096) always exceeds this
threshold, so percpu_counter_add_batch() acquires the counter's
global spinlock on every update. Each WRITE RPC updates two
counters (per-net and per-export), producing two global lock
acquisitions per operation. Profiling on a 10-CPU RDMA NFS
server shows 0.44% of total CPU cycles spent contending on
these locks during a small random write workload.

Switch to percpu_counter_add_local(), which batches with
INT_MAX so that updates always remain on the per-CPU fast
path regardless of the amount. All readers of these counters
already use percpu_counter_sum_positive(), which sums the
per-CPU deltas under the global lock, so read accuracy is
unaffected.

Reviewed-by: Jeff Layton 
Reviewed-by: NeilBrown 
Link: https://patch.msgid.link/20260717001232.438792-5-cel@kernel.org
Signed-off-by: Chuck Lever

NFSD: Eliminate percpu counter contention in reply cache statistics

2026-08-10T13:54:35+00:00

Each RPC passes through nfsd_cache_lookup(), which increments one
of nfsd_stats_rc_hits_inc(), nfsd_stats_rc_misses_inc(), or
nfsd_stats_rc_nocache_inc(). These helpers update
per-net-namespace percpu_counters with percpu_counter_inc(),
which applies the default batch threshold of max(32, 2*nr_cpus).

Once a CPU's local delta reaches that threshold, the update folds
into the shared counter under its global spinlock. On a busy
multi-CPU server this produces lock traffic on a counter cacheline
shared across all CPUs, growing with the request rate.

Switch to percpu_counter_add_local(fbc, 1), which batches with
INT_MAX so that increments always remain on the per-CPU fast path.
This matches the treatment already applied to the IO byte and DRC
memory counters. All readers of these counters use
percpu_counter_sum_positive(), which sums the per-CPU deltas under
the global lock, so read accuracy is unaffected.

Reviewed-by: Jeff Layton 
Reviewed-by: NeilBrown 
Link: https://patch.msgid.link/20260717001232.438792-4-cel@kernel.org
Signed-off-by: Chuck Lever

NFSD: Eliminate percpu counter contention in DRC memory accounting

2026-08-10T13:54:35+00:00

The DRC memory usage counter (NFSD_STATS_DRC_MEM_USAGE) tracks
bytes, but percpu_counter_add() uses the global percpu_counter_batch
threshold of max(32, 2*nr_cpus). Each DRC entry add or removal
updates the counter by sizeof(struct nfsd_cacherep) (~144 bytes),
which always exceeds the batch threshold. percpu_counter_add()
then acquires the counter's global spinlock on every update,
serializing all nfsd threads.

On a 10-CPU NFS server handling a high rate of non-idempotent
NFSv3 operations, this lock accounts for a measurable fraction
of total spin lock overhead because nfsd_cache_lookup() both
inserts a new entry and prunes up to three old entries per RPC,
producing 4-7 global lock acquisitions per operation.

Switch to percpu_counter_add_local() and percpu_counter_sub_local(),
which batch with INT_MAX so that updates always remain on the per-CPU
fast path regardless of the amount. The only reader of this counter uses
percpu_counter_sum_positive(), which sums the per-CPU deltas under the
global lock, so read accuracy is unaffected.

Reviewed-by: Jeff Layton 
Reviewed-by: NeilBrown 
Link: https://patch.msgid.link/20260717001232.438792-3-cel@kernel.org
Signed-off-by: Chuck Lever

NFSD: Fix off-by-one in DRC bucket pruning limit

2026-08-10T13:54:35+00:00

nfsd_prune_bucket_locked() evicts an entry before checking
the freed count against @max. The check uses "++freed > max",
which does not break until freed exceeds max, resulting in
max + 1 evictions. Use ">=" so the limit stated in the
function comment is honored.

Fixes: a9507f6af145 ("NFSD: Replace nfsd_prune_bucket()")
Cc: stable@vger.kernel.org
Reviewed-by: Jeff Layton 
Reviewed-by: NeilBrown 
Link: https://patch.msgid.link/20260717001232.438792-2-cel@kernel.org
Signed-off-by: Chuck Lever

NFSD: Relocate NFSv4 "supported attributes" to new header

2026-08-10T13:54:35+00:00

These NFSv4 attribute bitmask definitions live in nfsd.h, which
nearly every nfsd source file includes, yet only nfs4proc.c and
nfs4xdr.c reference them. Move them to a dedicated header so only
those two consumers pull them in.

While moving the block, correct the stale QUOTA_* annotation: the
promised support never materialized, so these attributes are
unlikely to be supported any time soon rather than forthcoming.

Link: https://patch.msgid.link/20260712204554.125308-10-cel@kernel.org
Reviewed-by: Jeff Layton 
Signed-off-by: Chuck Lever