Merge tag 'nfsd-6.5' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux

Pull nfsd updates from Chuck Lever:

 - Clean-ups in the READ path in anticipation of MSG_SPLICE_PAGES

 - Better NUMA awareness when allocating pages and other objects

 - A number of minor clean-ups to XDR encoding

 - Elimination of a race when accepting a TCP socket

 - Numerous observability enhancements

* tag 'nfsd-6.5' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux: (46 commits)
  nfsd: remove redundant assignments to variable len
  svcrdma: Fix stale comment
  NFSD: Distinguish per-net namespace initialization
  nfsd: move init of percpu reply_cache_stats counters back to nfsd_init_net
  SUNRPC: Address RCU warning in net/sunrpc/svc.c
  SUNRPC: Use sysfs_emit in place of strlcpy/sprintf
  SUNRPC: Remove transport class dprintk call sites
  SUNRPC: Fix comments for transport class registration
  svcrdma: Remove an unused argument from __svc_rdma_put_rw_ctxt()
  svcrdma: trace cc_release calls
  svcrdma: Convert "might sleep" comment into a code annotation
  NFSD: Add an nfsd4_encode_nfstime4() helper
  SUNRPC: Move initialization of rq_stime
  SUNRPC: Optimize page release in svc_rdma_sendto()
  svcrdma: Prevent page release when nothing was received
  svcrdma: Revert 2a1e4f21d841 ("svcrdma: Normalize Send page handling")
  SUNRPC: Revert 579900670ac7 ("svcrdma: Remove unused sc_pages field")
  SUNRPC: Revert cc93ce9529a6 ("svcrdma: Retain the page backing rq_res.head[0].iov_base")
  NFSD: add encoding of op_recall flag for write delegation
  NFSD: Add "official" reviewers for this subsystem
  ...

1 files changed, 14 insertions, 10 deletions

diff --git a/net/sunrpc/xprtrdma/svc_rdma_rw.c b/net/sunrpc/xprtrdma/svc_rdma_rw.c
index 11cf7c646644..e460e25a1d6d 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_rw.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_rw.c
@@ -62,8 +62,8 @@ svc_rdma_get_rw_ctxt(struct svcxprt_rdma *rdma, unsigned int sges)
 	if (node) {
 		ctxt = llist_entry(node, struct svc_rdma_rw_ctxt, rw_node);
 	} else {
-		ctxt = kmalloc(struct_size(ctxt, rw_first_sgl, SG_CHUNK_SIZE),
-			       GFP_KERNEL);
+		ctxt = kmalloc_node(struct_size(ctxt, rw_first_sgl, SG_CHUNK_SIZE),
+				    GFP_KERNEL, ibdev_to_node(rdma->sc_cm_id->device));
 		if (!ctxt)
 			goto out_noctx;
 
@@ -84,8 +84,7 @@ out_noctx:
 	return NULL;
 }
 
-static void __svc_rdma_put_rw_ctxt(struct svcxprt_rdma *rdma,
-				   struct svc_rdma_rw_ctxt *ctxt,
+static void __svc_rdma_put_rw_ctxt(struct svc_rdma_rw_ctxt *ctxt,
 				   struct llist_head *list)
 {
 	sg_free_table_chained(&ctxt->rw_sg_table, SG_CHUNK_SIZE);
@@ -95,7 +94,7 @@ static void __svc_rdma_put_rw_ctxt(struct svcxprt_rdma *rdma,
 static void svc_rdma_put_rw_ctxt(struct svcxprt_rdma *rdma,
 				 struct svc_rdma_rw_ctxt *ctxt)
 {
-	__svc_rdma_put_rw_ctxt(rdma, ctxt, &rdma->sc_rw_ctxts);
+	__svc_rdma_put_rw_ctxt(ctxt, &rdma->sc_rw_ctxts);
 }
 
 /**
@@ -191,6 +190,8 @@ static void svc_rdma_cc_release(struct svc_rdma_chunk_ctxt *cc,
 	struct svc_rdma_rw_ctxt *ctxt;
 	LLIST_HEAD(free);
 
+	trace_svcrdma_cc_release(&cc->cc_cid, cc->cc_sqecount);
+
 	first = last = NULL;
 	while ((ctxt = svc_rdma_next_ctxt(&cc->cc_rwctxts)) != NULL) {
 		list_del(&ctxt->rw_list);
@@ -198,7 +199,7 @@ static void svc_rdma_cc_release(struct svc_rdma_chunk_ctxt *cc,
 		rdma_rw_ctx_destroy(&ctxt->rw_ctx, rdma->sc_qp,
 				    rdma->sc_port_num, ctxt->rw_sg_table.sgl,
 				    ctxt->rw_nents, dir);
-		__svc_rdma_put_rw_ctxt(rdma, ctxt, &free);
+		__svc_rdma_put_rw_ctxt(ctxt, &free);
 
 		ctxt->rw_node.next = first;
 		first = &ctxt->rw_node;
@@ -234,7 +235,8 @@ svc_rdma_write_info_alloc(struct svcxprt_rdma *rdma,
 {
 	struct svc_rdma_write_info *info;
 
-	info = kmalloc(sizeof(*info), GFP_KERNEL);
+	info = kmalloc_node(sizeof(*info), GFP_KERNEL,
+			    ibdev_to_node(rdma->sc_cm_id->device));
 	if (!info)
 		return info;
 
@@ -304,7 +306,8 @@ svc_rdma_read_info_alloc(struct svcxprt_rdma *rdma)
 {
 	struct svc_rdma_read_info *info;
 
-	info = kmalloc(sizeof(*info), GFP_KERNEL);
+	info = kmalloc_node(sizeof(*info), GFP_KERNEL,
+			    ibdev_to_node(rdma->sc_cm_id->device));
 	if (!info)
 		return info;
 
@@ -351,8 +354,7 @@ static void svc_rdma_wc_read_done(struct ib_cq *cq, struct ib_wc *wc)
 	return;
 }
 
-/* This function sleeps when the transport's Send Queue is congested.
- *
+/*
  * Assumptions:
  * - If ib_post_send() succeeds, only one completion is expected,
  *   even if one or more WRs are flushed. This is true when posting
@@ -367,6 +369,8 @@ static int svc_rdma_post_chunk_ctxt(struct svc_rdma_chunk_ctxt *cc)
 	struct ib_cqe *cqe;
 	int ret;
 
+	might_sleep();
+
 	if (cc->cc_sqecount > rdma->sc_sq_depth)
 		return -EINVAL;