linux-stable.git/include/linux/sunrpc, branch v6.18.41

SUNRPC: pin upper rpc_clnt across the TLS connect_worker

2026-07-24T14:16:50+00:00

[ Upstream commit 46bc86c833956219bbfd246c1ffd832a479c5199 ]

The TLS connect path has a use-after-free: nothing pins the
upper rpc_clnt across the delayed connect_worker. xs_connect()
stores task->tk_client in sock_xprt::clnt as a raw pointer
and queues the worker; for TLS-secured transports that worker
is xs_tcp_tls_setup_socket(), which reads several fields out
of the saved pointer (cl_timeout, cl_program, cl_prog,
cl_vers, cl_cred, cl_stats) to construct the args for the
inner handshake rpc_clnt.

The xprt does not reference the rpc_clnt; the rpc_clnt
references the xprt. xs_destroy() does cancel the
connect_worker, but it runs only when the xprt's refcount
drops to zero, which cannot happen until the rpc_clnt
releases its cl_xprt reference in rpc_free_client_work().
When a TLS handshake fails fatally (for example, an mTLS
mount whose client cert does not match the server), the
connecting task is woken with -EACCES and exits, the mount
caller invokes rpc_shutdown_client(), and the upper rpc_clnt
is freed before the queued connect_worker fires.
xs_tcp_tls_setup_socket() then dereferences the freed clnt,
producing the refcount_t underflow Michael Nemanov reported.

Take a reference on the upper rpc_clnt in xs_connect() for
TLS transports via a new rpc_hold_client() helper, and drop
it in the connect_worker's exit path with rpc_release_client().
The xprt_lock_connect() / xprt_unlock_connect() pairing
already serialises xs_connect() with xs_tcp_tls_setup_socket(),
so the take and release are balanced one-for-one.

The non-TLS connect worker (xs_tcp_setup_socket) never reads
sock_xprt::clnt, so leave that path alone and avoid the
clnt-holds-xprt-holds-clnt cycle that would otherwise prevent
xprt destruction.

Reported-by: Michael Nemanov 
Closes: https://lore.kernel.org/linux-nfs/40e3d522-dfcf-4fc1-9c55-b5e81f1536d5@vastdata.com/
Fixes: 75eb6af7acdf ("SUNRPC: Add a TCP-with-TLS RPC transport class")
Signed-off-by: Chuck Lever 
Tested-by: Michael Nemanov 
Reviewed-by: Michael Nemanov 
Signed-off-by: Anna Schumaker 
Signed-off-by: Sasha Levin

xprtrdma: Close lost-wakeup race in xprt_rdma_alloc_slot

2026-07-24T14:16:21+00:00

[ Upstream commit 765bde47fe7f197dabeb12da76831f40d0b20377 ]

xprt_rdma_alloc_slot() and xprt_rdma_free_slot() lack serialization
between the buffer pool and the backlog queue.  A buffer freed
after rpcrdma_buffer_get() finds the pool empty but before
rpc_sleep_on() places the task on the backlog is returned to the
pool with no waiter to wake, leaving the task stuck on the backlog
indefinitely.

After joining the backlog, re-check the pool and route any
recovered buffer through xprt_wake_up_backlog(), whose queue lock
serializes with concurrent wakeups and avoids double-assignment
of slots.

Because xprt_rdma_free_slot() does not hold reserve_lock, the
XPRT_CONGESTED double-check in xprt_throttle_congested() is
ineffective: a task can join the backlog through that path after
free_slot has already found it empty and cleared the bit.  Avoid
this by using xprt_add_backlog_noncongested(), which queues the
task without setting XPRT_CONGESTED, so every allocation reaches
xprt_rdma_alloc_slot() and its post-sleep re-check.

Fixes: edb41e61a54e ("xprtrdma: Make rpc_rqst part of rpcrdma_req")
Signed-off-by: Chuck Lever 
Signed-off-by: Trond Myklebust 
Stable-dep-of: e786233d2e0b ("xprtrdma: Decouple req recycling from RPC completion")
Signed-off-by: Sasha Levin

xdrgen: Initialize data pointer for zero-length items

2026-02-26T22:59:18+00:00

[ Upstream commit 27b0fcae8f535fb882b1876227a935dcfdf576aa ]

The xdrgen decoders for strings and opaque data had an
optimization that skipped calling xdr_inline_decode() when the
item length was zero. This left the data pointer uninitialized,
which could lead to unpredictable behavior when callers access
it.

Remove the zero-length check and always call xdr_inline_decode().
When passed a length of zero, xdr_inline_decode() returns the
current buffer position, which is valid and matches the behavior
of hand-coded XDR decoders throughout the kernel.

Fixes: 4b132aacb076 ("tools: Add xdrgen")
Reviewed-by: Jeff Layton 
Reviewed-by: NeilBrown 
Signed-off-by: Chuck Lever 
Signed-off-by: Sasha Levin

Merge tag 'nfsd-6.18' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux

2025-10-06T20:22:21+00:00

Pull nfsd updates from Chuck Lever:
 "Mike Snitzer has prototyped a mechanism for disabling I/O caching in
  NFSD. This is introduced in v6.18 as an experimental feature. This
  enables scaling NFSD in /both/ directions:

   - NFS service can be supported on systems with small memory
     footprints, such as low-cost cloud instances

   - Large NFS workloads will be less likely to force the eviction of
     server-local activity, helping it avoid thrashing

  Jeff Layton contributed a number of fixes to the new attribute
  delegation implementation (based on a pending Internet RFC) that we
  hope will make attribute delegation reliable enough to enable by
  default, as it is on the Linux NFS client.

  The remaining patches in this pull request are clean-ups and minor
  optimizations. Many thanks to the contributors, reviewers, testers,
  and bug reporters who participated during the v6.18 NFSD development
  cycle"

* tag 'nfsd-6.18' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux: (42 commits)
  nfsd: discard nfserr_dropit
  SUNRPC: Make RPCSEC_GSS_KRB5 select CRYPTO instead of depending on it
  NFSD: Add io_cache_{read,write} controls to debugfs
  NFSD: Do the grace period check in ->proc_layoutget
  nfsd: delete unnecessary NULL check in __fh_verify()
  NFSD: Allow layoutcommit during grace period
  NFSD: Disallow layoutget during grace period
  sunrpc: fix "occurence"->"occurrence"
  nfsd: Don't force CRYPTO_LIB_SHA256 to be built-in
  nfsd: nfserr_jukebox in nlm_fopen should lead to a retry
  NFSD: Reduce DRC bucket size
  NFSD: Delay adding new entries to LRU
  SUNRPC: Move the svc_rpcb_cleanup() call sites
  NFS: Remove rpcbind cleanup for NFSv4.0 callback
  nfsd: unregister with rpcbind when deleting a transport
  NFSD: Drop redundant conversion to bool
  sunrpc: eliminate return pointer in svc_tcp_sendmsg()
  sunrpc: fix pr_notice in svc_tcp_sendto() to show correct length
  nfsd: decouple the xprtsec policy check from check_nfsd_access()
  NFSD: Fix destination buffer size in nfsd4_ssc_setup_dul()
  ...

SUNRPC: Update gssx_accept_sec_context() to use xdr_set_scratch_folio()

2025-09-23T17:29:50+00:00

This was the last caller of xdr_set_scratch_page(), so I remove this
function while I'm at it.

Signed-off-by: Anna Schumaker

SUNRPC: Update svcxdr_init_decode() to call xdr_set_scratch_folio()

2025-09-23T17:29:50+00:00

The only snag here is that __folio_alloc_node() doesn't handle
NUMA_NO_NODE, so I also need to update svc_pool_map_get_node() to return
numa_mem_id() instead. I arrived at this approach by  looking at what
other users of __folio_alloc_node() do for this case.

Signed-off-by: Anna Schumaker

SUNRPC: Introduce xdr_set_scratch_folio()

2025-09-23T17:29:50+00:00

This will replace xdr_set_scratch_page() when we switch pages to folios.

Signed-off-by: Anna Schumaker

SUNRPC: Move the svc_rpcb_cleanup() call sites

2025-09-23T17:28:19+00:00

Clean up: because svc_rpcb_cleanup() and svc_xprt_destroy_all()
are always invoked in pairs, we can deduplicate code by moving
the svc_rpcb_cleanup() call sites into svc_xprt_destroy_all().

Signed-off-by: Chuck Lever 
Tested-by: Olga Kornievskaia 
Signed-off-by: Anna Schumaker

sunrpc: add a Kconfig option to redirect dfprintk() output to trace buffer

2025-09-23T17:28:19+00:00

We have a lot of old dprintk() call sites that aren't going anywhere
anytime soon. At the same time, turning them up is a serious burden on
the host due to the console locking overhead.

Add a new Kconfig option that redirects dfprintk() output to the trace
buffer. This is more efficient than logging to the console and allows
for proper interleaving of dprintk and static tracepoint events.

Since using trace_printk() causes scary warnings to pop at boot time,
this new option defaults to "n".

Signed-off-by: Jeff Layton 
Reviewed-by: Chuck Lever 
Reviewed-by: Simon Horman 
Signed-off-by: Anna Schumaker

sunrpc: remove dfprintk_cont() and dfprintk_rcu_cont()

2025-09-23T17:28:19+00:00

KERN_CONT hails from a simpler time, when SMP wasn't the norm. These
days, it doesn't quite work right since another printk() can always race
in between the first one and the one being "continued".

Nothing calls dprintk_rcu_cont(), so just remove it. The only caller of
dprintk_cont() is in nfs_commit_release_pages(). Just use a normal
dprintk() there instead, since this is not SMP-safe anyway.

Signed-off-by: Jeff Layton 
Reviewed-by: Chuck Lever 
Reviewed-by: Simon Horman 
Signed-off-by: Anna Schumaker