NFSv4.1: Ensure smooth handover of slots from one task to the next waiting

Currently, we see a lot of bouncing for the value of highest_used_slotid
due to the fact that slots are getting freed, instead of getting instantly
transmitted to the next waiting task.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

4 files changed, 69 insertions, 12 deletions

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 99d99a5a3f6..992233561db 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -401,14 +401,15 @@ static void nfs41_sequence_free_slot(struct nfs4_sequence_res *res)
 	if (tbl->highest_used_slotid > tbl->target_highest_slotid)
 		send_new_highest_used_slotid = true;
 
+	if (nfs41_wake_and_assign_slot(tbl, res->sr_slot)) {
+		send_new_highest_used_slotid = false;
+		goto out_unlock;
+	}
 	nfs4_free_slot(tbl, res->sr_slot);
 
 	if (tbl->highest_used_slotid != NFS4_NO_SLOT)
 		send_new_highest_used_slotid = false;
-	if (!nfs4_session_draining(session)) {
-		if (rpc_wake_up_next(&tbl->slot_tbl_waitq) != NULL)
-			send_new_highest_used_slotid = false;
-	}
+out_unlock:
 	spin_unlock(&tbl->slot_tbl_lock);
 	res->sr_slot = NULL;
 	if (send_new_highest_used_slotid)
@@ -1465,6 +1466,7 @@ unlock_no_action:
 	rcu_read_unlock();
 out_no_action:
 	task->tk_action = NULL;
+	nfs4_sequence_done(task, &data->o_res.seq_res);
 }
 
 static void nfs4_open_done(struct rpc_task *task, void *calldata)
@@ -2135,6 +2137,7 @@ static void nfs4_close_prepare(struct rpc_task *task, void *data)
 	if (!call_close) {
 		/* Note: exit _without_ calling nfs4_close_done */
 		task->tk_action = NULL;
+		nfs4_sequence_done(task, &calldata->res.seq_res);
 		goto out;
 	}
 
@@ -4384,6 +4387,7 @@ static void nfs4_locku_prepare(struct rpc_task *task, void *data)
 	if (test_bit(NFS_LOCK_INITIALIZED, &calldata->lsp->ls_flags) == 0) {
 		/* Note: exit _without_ running nfs4_locku_done */
 		task->tk_action = NULL;
+		nfs4_sequence_done(task, &calldata->res.seq_res);
 		return;
 	}
 	calldata->timestamp = jiffies;
diff --git a/fs/nfs/nfs4session.c b/fs/nfs/nfs4session.c
index 701170293ce..066cfa101b4 100644
--- a/fs/nfs/nfs4session.c
+++ b/fs/nfs/nfs4session.c
@@ -217,11 +217,65 @@ static void nfs4_destroy_slot_tables(struct nfs4_session *session)
 	nfs4_shrink_slot_table(&session->bc_slot_table, 0);
 }
 
+static bool nfs41_assign_slot(struct rpc_task *task, void *pslot)
+{
+	struct nfs4_sequence_args *args = task->tk_msg.rpc_argp;
+	struct nfs4_sequence_res *res = task->tk_msg.rpc_resp;
+	struct nfs4_slot *slot = pslot;
+	struct nfs4_slot_table *tbl = slot->table;
+
+	if (nfs4_session_draining(tbl->session) && !args->sa_privileged)
+		return false;
+	slot->renewal_time = jiffies;
+	slot->generation = tbl->generation;
+	args->sa_slot = slot;
+	res->sr_slot = slot;
+	res->sr_status_flags = 0;
+	res->sr_status = 1;
+	return true;
+}
+
+static bool __nfs41_wake_and_assign_slot(struct nfs4_slot_table *tbl,
+		struct nfs4_slot *slot)
+{
+	if (rpc_wake_up_first(&tbl->slot_tbl_waitq, nfs41_assign_slot, slot))
+		return true;
+	return false;
+}
+
+bool nfs41_wake_and_assign_slot(struct nfs4_slot_table *tbl,
+		struct nfs4_slot *slot)
+{
+	if (slot->slot_nr > tbl->max_slotid)
+		return false;
+	return __nfs41_wake_and_assign_slot(tbl, slot);
+}
+
+static bool nfs41_try_wake_next_slot_table_entry(struct nfs4_slot_table *tbl)
+{
+	struct nfs4_slot *slot = nfs4_alloc_slot(tbl);
+	if (!IS_ERR(slot)) {
+		bool ret = __nfs41_wake_and_assign_slot(tbl, slot);
+		if (ret)
+			return ret;
+		nfs4_free_slot(tbl, slot);
+	}
+	return false;
+}
+
+void nfs41_wake_slot_table(struct nfs4_slot_table *tbl)
+{
+	for (;;) {
+		if (!nfs41_try_wake_next_slot_table_entry(tbl))
+			break;
+	}
+}
+
 /* Update the client's idea of target_highest_slotid */
 static void nfs41_set_target_slotid_locked(struct nfs4_slot_table *tbl,
 		u32 target_highest_slotid)
 {
-	unsigned int max_slotid, i;
+	unsigned int max_slotid;
 
 	if (tbl->target_highest_slotid == target_highest_slotid)
 		return;
@@ -229,9 +283,8 @@ static void nfs41_set_target_slotid_locked(struct nfs4_slot_table *tbl,
 	tbl->generation++;
 
 	max_slotid = min(NFS4_MAX_SLOT_TABLE - 1, tbl->target_highest_slotid);
-	for (i = tbl->max_slotid + 1; i <= max_slotid; i++)
-		rpc_wake_up_next(&tbl->slot_tbl_waitq);
 	tbl->max_slotid = max_slotid;
+	nfs41_wake_slot_table(tbl);
 }
 
 void nfs41_set_target_slotid(struct nfs4_slot_table *tbl,
diff --git a/fs/nfs/nfs4session.h b/fs/nfs/nfs4session.h
index bdd14a60722..7db73937016 100644
--- a/fs/nfs/nfs4session.h
+++ b/fs/nfs/nfs4session.h
@@ -94,6 +94,10 @@ static inline bool nfs4_session_draining(struct nfs4_session *session)
 	return !!test_bit(NFS4_SESSION_DRAINING, &session->session_state);
 }
 
+bool nfs41_wake_and_assign_slot(struct nfs4_slot_table *tbl,
+		struct nfs4_slot *slot);
+void nfs41_wake_slot_table(struct nfs4_slot_table *tbl);
+
 /*
  * Determine if sessions are in use.
  */
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index 7d73df5a05d..78e90a80fc3 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -255,17 +255,13 @@ static void nfs4_end_drain_session(struct nfs_client *clp)
 {
 	struct nfs4_session *ses = clp->cl_session;
 	struct nfs4_slot_table *tbl;
-	unsigned int i;
 
 	if (ses == NULL)
 		return;
 	tbl = &ses->fc_slot_table;
 	if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 		spin_lock(&tbl->slot_tbl_lock);
-		for (i = 0; i <= tbl->max_slotid; i++) {
-			if (rpc_wake_up_next(&tbl->slot_tbl_waitq) == NULL)
-				break;
-		}
+		nfs41_wake_slot_table(tbl);
 		spin_unlock(&tbl->slot_tbl_lock);
 	}
 }