SUNRPC: receive buffer size estimation values almost never change

Avoid unnecessary cache sloshing by placing the buffer size
estimation update logic behind an atomic bit flag.

The size of GSS information included in each wrapped Reply does
not change during the lifetime of a GSS context. Therefore, the
au_rslack and au_ralign fields need to be updated only once after
establishing a fresh GSS credential.

Thus a slack size update must occur after a cred is created,
duplicated, renewed, or expires. I'm not sure I have this exactly
right. A trace point is introduced to track updates to these
variables to enable troubleshooting the problem if I missed a spot.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

include/linux/sunrpc/auth.h

@@ -76,7 +76,7 @@ struct rpc_auth {
 	unsigned int		au_verfsize;	/* size of reply verifier */
 	unsigned int		au_ralign;	/* words before UL header */
 
-	unsigned int		au_flags;
+	unsigned long		au_flags;
 	const struct rpc_authops *au_ops;
 	rpc_authflavor_t	au_flavor;	/* pseudoflavor (note may
 						 * differ from the flavor in
@@ -89,7 +89,8 @@ struct rpc_auth {
 };
 
 /* rpc_auth au_flags */
-#define RPCAUTH_AUTH_DATATOUCH	0x00000002
+#define RPCAUTH_AUTH_DATATOUCH		(1)
+#define RPCAUTH_AUTH_UPDATE_SLACK	(2)
 
 struct rpc_auth_create_args {
 	rpc_authflavor_t pseudoflavor;

include/trace/events/rpcgss.h

@@ -291,6 +291,40 @@ TRACE_EVENT(rpcgss_need_reencode,
 		__entry->ret ? "" : "un")
 );
 
+TRACE_EVENT(rpcgss_update_slack,
+	TP_PROTO(
+		const struct rpc_task *task,
+		const struct rpc_auth *auth
+	),
+
+	TP_ARGS(task, auth),
+
+	TP_STRUCT__entry(
+		__field(unsigned int, task_id)
+		__field(unsigned int, client_id)
+		__field(u32, xid)
+		__field(const void *, auth)
+		__field(unsigned int, rslack)
+		__field(unsigned int, ralign)
+		__field(unsigned int, verfsize)
+	),
+
+	TP_fast_assign(
+		__entry->task_id = task->tk_pid;
+		__entry->client_id = task->tk_client->cl_clid;
+		__entry->xid = be32_to_cpu(task->tk_rqstp->rq_xid);
+		__entry->auth = auth;
+		__entry->rslack = auth->au_rslack;
+		__entry->ralign = auth->au_ralign;
+		__entry->verfsize = auth->au_verfsize;
+	),
+
+	TP_printk("task:%u@%u xid=0x%08x auth=%p rslack=%u ralign=%u verfsize=%u\n",
+		__entry->task_id, __entry->client_id, __entry->xid,
+		__entry->auth, __entry->rslack, __entry->ralign,
+		__entry->verfsize)
+);
+
 DECLARE_EVENT_CLASS(rpcgss_svc_seqno_class,
 	TP_PROTO(
 		__be32 xid,

net/sunrpc/auth_gss/auth_gss.c

@@ -1054,11 +1054,11 @@ gss_create_new(const struct rpc_auth_create_args *args, struct rpc_clnt *clnt)
 	auth->au_rslack = GSS_KRB5_MAX_SLACK_NEEDED >> 2;
 	auth->au_verfsize = GSS_VERF_SLACK >> 2;
 	auth->au_ralign = GSS_VERF_SLACK >> 2;
-	auth->au_flags = 0;
+	__set_bit(RPCAUTH_AUTH_UPDATE_SLACK, &auth->au_flags);
 	auth->au_ops = &authgss_ops;
 	auth->au_flavor = flavor;
 	if (gss_pseudoflavor_to_datatouch(gss_auth->mech, flavor))
-		auth->au_flags |= RPCAUTH_AUTH_DATATOUCH;
+		__set_bit(RPCAUTH_AUTH_DATATOUCH, &auth->au_flags);
 	refcount_set(&auth->au_count, 1);
 	kref_init(&gss_auth->kref);
 
@@ -1613,6 +1613,7 @@ static int gss_renew_cred(struct rpc_task *task)
 	new = gss_lookup_cred(auth, &acred, RPCAUTH_LOOKUP_NEW);
 	if (IS_ERR(new))
 		return PTR_ERR(new);
+
 	task->tk_rqstp->rq_cred = new;
 	put_rpccred(oldcred);
 	return 0;
@@ -1709,7 +1710,8 @@ gss_validate(struct rpc_task *task, struct xdr_stream *xdr)
 
 	/* We leave it to unwrap to calculate au_rslack. For now we just
 	 * calculate the length of the verifier: */
-	cred->cr_auth->au_verfsize = XDR_QUADLEN(len) + 2;
+	if (test_bit(RPCAUTH_AUTH_UPDATE_SLACK, &cred->cr_auth->au_flags))
+		cred->cr_auth->au_verfsize = XDR_QUADLEN(len) + 2;
 	status = 0;
 out:
 	gss_put_ctx(ctx);
@@ -1927,13 +1929,30 @@ static int gss_wrap_req(struct rpc_task *task, struct xdr_stream *xdr)
 	return status;
 }
 
-static int
-gss_unwrap_resp_auth(struct rpc_cred *cred)
+/**
+ * gss_update_rslack - Possibly update RPC receive buffer size estimates
+ * @task: rpc_task for incoming RPC Reply being unwrapped
+ * @cred: controlling rpc_cred for @task
+ * @before: XDR words needed before each RPC Reply message
+ * @after: XDR words needed following each RPC Reply message
+ *
+ */
+static void gss_update_rslack(struct rpc_task *task, struct rpc_cred *cred,
+			      unsigned int before, unsigned int after)
 {
 	struct rpc_auth *auth = cred->cr_auth;
 
-	auth->au_rslack = auth->au_verfsize;
-	auth->au_ralign = auth->au_verfsize;
+	if (test_and_clear_bit(RPCAUTH_AUTH_UPDATE_SLACK, &auth->au_flags)) {
+		auth->au_ralign = auth->au_verfsize + before;
+		auth->au_rslack = auth->au_verfsize + after;
+		trace_rpcgss_update_slack(task, auth);
+	}
+}
+
+static int
+gss_unwrap_resp_auth(struct rpc_task *task, struct rpc_cred *cred)
+{
+	gss_update_rslack(task, cred, 0, 0);
 	return 0;
 }
 
@@ -1956,7 +1975,6 @@ gss_unwrap_resp_integ(struct rpc_task *task, struct rpc_cred *cred,
 		      struct xdr_stream *xdr)
 {
 	struct xdr_buf gss_data, *rcv_buf = &rqstp->rq_rcv_buf;
-	struct rpc_auth *auth = cred->cr_auth;
 	u32 len, offset, seqno, maj_stat;
 	struct xdr_netobj mic;
 	int ret;
@@ -2005,8 +2023,7 @@ gss_unwrap_resp_integ(struct rpc_task *task, struct rpc_cred *cred,
 	if (maj_stat != GSS_S_COMPLETE)
 		goto bad_mic;
 
-	auth->au_rslack = auth->au_verfsize + 2 + 1 + XDR_QUADLEN(mic.len);
-	auth->au_ralign = auth->au_verfsize + 2;
+	gss_update_rslack(task, cred, 2, 2 + 1 + XDR_QUADLEN(mic.len));
 	ret = 0;
 
 out:
@@ -2031,7 +2048,6 @@ gss_unwrap_resp_priv(struct rpc_task *task, struct rpc_cred *cred,
 {
 	struct xdr_buf *rcv_buf = &rqstp->rq_rcv_buf;
 	struct kvec *head = rqstp->rq_rcv_buf.head;
-	struct rpc_auth *auth = cred->cr_auth;
 	u32 offset, opaque_len, maj_stat;
 	__be32 *p;
 
@@ -2058,8 +2074,8 @@ gss_unwrap_resp_priv(struct rpc_task *task, struct rpc_cred *cred,
 	 */
 	xdr_init_decode(xdr, rcv_buf, p, rqstp);
 
-	auth->au_rslack = auth->au_verfsize + 2 + ctx->gc_gss_ctx->slack;
-	auth->au_ralign = auth->au_verfsize + 2 + ctx->gc_gss_ctx->align;
+	gss_update_rslack(task, cred, 2 + ctx->gc_gss_ctx->align,
+			  2 + ctx->gc_gss_ctx->slack);
 
 	return 0;
 unwrap_failed:
@@ -2130,7 +2146,7 @@ gss_unwrap_resp(struct rpc_task *task, struct xdr_stream *xdr)
 		goto out_decode;
 	switch (gss_cred->gc_service) {
 	case RPC_GSS_SVC_NONE:
-		status = gss_unwrap_resp_auth(cred);
+		status = gss_unwrap_resp_auth(task, cred);
 		break;
 	case RPC_GSS_SVC_INTEGRITY:
 		status = gss_unwrap_resp_integ(task, cred, ctx, rqstp, xdr);

net/sunrpc/xprtrdma/rpc_rdma.c

@@ -892,8 +892,8 @@ rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst)
 	 * or privacy, direct data placement of individual data items
 	 * is not allowed.
 	 */
-	ddp_allowed = !(rqst->rq_cred->cr_auth->au_flags &
-						RPCAUTH_AUTH_DATATOUCH);
+	ddp_allowed = !test_bit(RPCAUTH_AUTH_DATATOUCH,
+				&rqst->rq_cred->cr_auth->au_flags);
 
 	/*
 	 * Chunks needed for results?