NFSD: Force all NFSv4.2 COPY requests to be synchronous

We've discovered that delivering a CB_OFFLOAD operation can be
unreliable in some pretty unremarkable situations. Examples
include:

 - The server dropped the connection because it lost a forechannel
   NFSv4 request and wishes to force the client to retransmit
 - The GSS sequence number window under-flowed
 - A network partition occurred

When that happens, all pending callback operations, including
CB_OFFLOAD, are lost. NFSD does not retransmit them.

Moreover, the Linux NFS client does not yet support sending an
OFFLOAD_STATUS operation to probe whether an asynchronous COPY
operation has finished. Thus, on Linux NFS clients, when a
CB_OFFLOAD is lost, asynchronous COPY can hang until manually
interrupted.

I've tried a couple of remedies, but so far the side-effects are
worse than the disease and they have had to be reverted. So
temporarily force COPY operations to be synchronous so that the use
of CB_OFFLOAD is avoided entirely. This is a fix that can easily be
backported to LTS kernels. I am working on client patches that
introduce an implementation of OFFLOAD_STATUS.

Note that NFSD arbitrarily limits the size of a copy_file_range
to 4MB to avoid indefinitely blocking an nfsd thread. A short
COPY result is returned in that case, and the client can present
a fresh COPY request for the remainder.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
This commit is contained in:
Chuck Lever 2024-05-07 09:37:14 -04:00
parent bafa6b4d95
commit 8d915bbf39

View file

@ -1807,6 +1807,13 @@ nfsd4_copy(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
__be32 status;
struct nfsd4_copy *async_copy = NULL;
/*
* Currently, async COPY is not reliable. Force all COPY
* requests to be synchronous to avoid client application
* hangs waiting for COPY completion.
*/
nfsd4_copy_set_sync(copy, true);
copy->cp_clp = cstate->clp;
if (nfsd4_ssc_is_inter(copy)) {
trace_nfsd_copy_inter(copy);