On Tue, 2019-08-27 at 06:25 -0400, Jan Stancek wrote: > That theory is probably not correct for this case, since EIO I see > appears > to originate from write and nfs_writeback_result(). This function > also > produces message we saw in logs from Naresh. > > I can't find where/how is resp->count updated on WRITE reply in > NFSv2. > Issue also goes away with patch below, though I can't speak about its > correctness: > > NFS version Type Test Return code > nfsvers=2 tcp -b:base 0 > nfsvers=2 tcp -g:general 0 > nfsvers=2 tcp -s:special 0 > nfsvers=2 tcp -l:lock 0 > Total time: 141 > > diff --git a/fs/nfs/nfs2xdr.c b/fs/nfs/nfs2xdr.c > index cbc17a203248..4913c6da270b 100644 > --- a/fs/nfs/nfs2xdr.c > +++ b/fs/nfs/nfs2xdr.c > @@ -897,6 +897,16 @@ static int nfs2_xdr_dec_writeres(struct rpc_rqst > *req, struct xdr_stream *xdr, > void *data) > { > struct nfs_pgio_res *result = data; > + struct rpc_task *rq_task = req->rq_task; > + > + if (rq_task) { > + struct nfs_pgio_args *args = rq_task- > >tk_msg.rpc_argp; > + > + if (args) { > + result->count = args->count; > + } > + } > > /* All NFSv2 writes are "file sync" writes */ > result->verf->committed = NFS_FILE_SYNC; Thanks! I've moved the above to nfs_write_done() so that we do it only on success (see http://git.linux-nfs.org/?p=trondmy/linux-nfs.git;a=commitdiff;h=3ba5688da709dd0f7d917029c206bc1848a6ae74 ) -- Trond Myklebust Linux NFS client maintainer, Hammerspace trond.myklebust@xxxxxxxxxxxxxxx