Whamcloud - gitweb
LU-13004 ptlrpc: Allow BULK_BUF_KIOV to accept a kvec
[fs/lustre-release.git] / lustre / ptlrpc / connection.c
index e643013..369eace 100644 (file)
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
  *
- *  Copyright (C) 2002 Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   This file is part of Lustre, http://www.lustre.org.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   Lustre is free software; you can redistribute it and/or
- *   modify it under the terms of version 2 of the GNU General Public
- *   License as published by the Free Software Foundation.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   Lustre is distributed in the hope that it will be useful,
- *   but WITHOUT ANY WARRANTY; without even the implied warranty of
- *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   GNU General Public License for more details.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
- *   You should have received a copy of the GNU General Public License
- *   along with Lustre; if not, write to the Free Software
- *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
  *
+ * Copyright (c) 2011, Intel Corporation.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
 #define DEBUG_SUBSYSTEM S_RPC
+#include <obd_support.h>
+#include <obd_class.h>
+#include <lustre_net.h>
 
-#include <linux/obd_support.h>
-#include <linux/lustre_net.h>
+#include "ptlrpc_internal.h"
 
-static spinlock_t conn_lock;
-static struct list_head conn_list;
-static struct list_head conn_unused_list;
+static struct cfs_hash *conn_hash;
+static struct cfs_hash_ops conn_hash_ops;
 
-/* If UUID is NULL, c->c_remote_uuid must be all zeroes
- * If UUID is non-NULL, c->c_remote_uuid must match. */
-static int match_connection_uuid(struct ptlrpc_connection *c, obd_uuid_t uuid)
+struct ptlrpc_connection *
+ptlrpc_connection_get(struct lnet_process_id peer, lnet_nid_t self,
+                     struct obd_uuid *uuid)
 {
-        obd_uuid_t zero_uuid = {0};
+       struct ptlrpc_connection *conn, *conn2;
+       ENTRY;
+
+       peer.nid = LNetPrimaryNID(peer.nid);
+       conn = cfs_hash_lookup(conn_hash, &peer);
+       if (conn)
+               GOTO(out, conn);
+
+       OBD_ALLOC_PTR(conn);
+       if (!conn)
+               RETURN(NULL);
+
+       conn->c_peer = peer;
+       conn->c_self = self;
+       INIT_HLIST_NODE(&conn->c_hash);
+       atomic_set(&conn->c_refcount, 1);
+       if (uuid)
+               obd_str2uuid(&conn->c_remote_uuid, uuid->uuid);
+
+       /*
+        * Add the newly created conn to the hash, on key collision we
+        * lost a racing addition and must destroy our newly allocated
+        * connection.  The object which exists in the hash will be
+        * returned and may be compared against out object.
+        */
+       /* In the function below, .hs_keycmp resolves to
+        * conn_keycmp() */
+       /* coverity[overrun-buffer-val] */
+       conn2 = cfs_hash_findadd_unique(conn_hash, &peer, &conn->c_hash);
+       if (conn != conn2) {
+               OBD_FREE_PTR(conn);
+               conn = conn2;
+       }
+       EXIT;
+out:
+       CDEBUG(D_INFO, "conn=%p refcount %d to %s\n",
+              conn, atomic_read(&conn->c_refcount),
+              libcfs_nid2str(conn->c_peer.nid));
+       return conn;
+}
 
-        if (uuid)
-                return memcmp(c->c_remote_uuid, uuid, sizeof(uuid));
+int ptlrpc_connection_put(struct ptlrpc_connection *conn)
+{
+       int rc = 0;
+       ENTRY;
+
+       if (!conn)
+               RETURN(rc);
+
+       LASSERT(atomic_read(&conn->c_refcount) > 1);
+
+       /*
+        * We do not remove connection from hashtable and
+        * do not free it even if last caller released ref,
+        * as we want to have it cached for the case it is
+        * needed again.
+        *
+        * Deallocating it and later creating new connection
+        * again would be wastful. This way we also avoid
+        * expensive locking to protect things from get/put
+        * race when found cached connection is freed by
+        * ptlrpc_connection_put().
+        *
+        * It will be freed later in module unload time,
+        * when ptlrpc_connection_fini()->lh_exit->conn_exit()
+        * path is called.
+        */
+       if (atomic_dec_return(&conn->c_refcount) == 1)
+               rc = 1;
+
+       CDEBUG(D_INFO, "PUT conn=%p refcount %d to %s\n",
+              conn, atomic_read(&conn->c_refcount),
+              libcfs_nid2str(conn->c_peer.nid));
+
+       RETURN(rc);
+}
+
+struct ptlrpc_connection *
+ptlrpc_connection_addref(struct ptlrpc_connection *conn)
+{
+       ENTRY;
 
-        return memcmp(c->c_remote_uuid, zero_uuid, sizeof(zero_uuid));
+       atomic_inc(&conn->c_refcount);
+       CDEBUG(D_INFO, "conn=%p refcount %d to %s\n",
+              conn, atomic_read(&conn->c_refcount),
+              libcfs_nid2str(conn->c_peer.nid));
+
+       RETURN(conn);
 }
 
-struct ptlrpc_connection *ptlrpc_get_connection(struct lustre_peer *peer,
-                                                obd_uuid_t uuid)
+int ptlrpc_connection_init(void)
 {
-        struct list_head *tmp, *pos;
-        struct ptlrpc_connection *c;
         ENTRY;
 
-        CDEBUG(D_INFO, "peer is %08x %08lx %08lx\n",
-               peer->peer_nid, peer->peer_ni.nal_idx, peer->peer_ni.handle_idx);
-
-        spin_lock(&conn_lock);
-        list_for_each(tmp, &conn_list) {
-                c = list_entry(tmp, struct ptlrpc_connection, c_link);
-                if (memcmp(peer, &c->c_peer, sizeof(*peer)) == 0 &&
-                    !match_connection_uuid(c, uuid)) {
-                        ptlrpc_connection_addref(c);
-                        GOTO(out, c);
-                }
-        }
-
-        list_for_each_safe(tmp, pos, &conn_unused_list) {
-                c = list_entry(tmp, struct ptlrpc_connection, c_link);
-                if (memcmp(peer, &c->c_peer, sizeof(*peer)) == 0 &&
-                    !match_connection_uuid(c, uuid)) {
-                        ptlrpc_connection_addref(c);
-                        list_del(&c->c_link);
-                        list_add(&c->c_link, &conn_list);
-                        GOTO(out, c);
-                }
-        }
-
-        /* FIXME: this should be a slab once we can validate slab addresses
-         * without OOPSing */
-        OBD_ALLOC(c, sizeof(*c));
-        if (c == NULL)
-                GOTO(out, c);
-
-        c->c_level = LUSTRE_CONN_NEW;
-        c->c_xid_in = 1;
-        c->c_xid_out = 1;
-        c->c_generation = 1;
-        c->c_epoch = 1;
-        c->c_bootcount = 0;
-        c->c_flags = 0;
-        if (uuid)
-                strcpy(c->c_remote_uuid, uuid);
-        INIT_LIST_HEAD(&c->c_delayed_head);
-        INIT_LIST_HEAD(&c->c_sending_head);
-        INIT_LIST_HEAD(&c->c_dying_head);
-        INIT_LIST_HEAD(&c->c_imports);
-        INIT_LIST_HEAD(&c->c_exports);
-        INIT_LIST_HEAD(&c->c_sb_chain);
-        atomic_set(&c->c_refcount, 0);
-        ptlrpc_connection_addref(c);
-        spin_lock_init(&c->c_lock);
-
-        memcpy(&c->c_peer, peer, sizeof(c->c_peer));
-        list_add(&c->c_link, &conn_list);
+        conn_hash = cfs_hash_create("CONN_HASH",
+                                    HASH_CONN_CUR_BITS,
+                                    HASH_CONN_MAX_BITS,
+                                    HASH_CONN_BKT_BITS, 0,
+                                    CFS_HASH_MIN_THETA,
+                                    CFS_HASH_MAX_THETA,
+                                    &conn_hash_ops, CFS_HASH_DEFAULT);
+        if (!conn_hash)
+                RETURN(-ENOMEM);
+
+        RETURN(0);
+}
 
+void ptlrpc_connection_fini(void) {
+        ENTRY;
+        cfs_hash_putref(conn_hash);
         EXIT;
- out:
-        spin_unlock(&conn_lock);
-        return c;
 }
 
-int ptlrpc_put_connection(struct ptlrpc_connection *c)
+/*
+ * Hash operations for net_peer<->connection
+ */
+static unsigned
+conn_hashfn(struct cfs_hash *hs, const void *key, unsigned mask)
 {
-        int rc = 0;
-        ENTRY;
+       return cfs_hash_djb2_hash(key, sizeof(struct lnet_process_id), mask);
+}
 
-        if (c == NULL) {
-                CERROR("NULL connection\n");
-                RETURN(0);
-        }
-
-        CDEBUG(D_INFO, "connection=%p refcount %d\n",
-               c, atomic_read(&c->c_refcount) - 1);
-        if (atomic_dec_and_test(&c->c_refcount)) {
-                spin_lock(&conn_lock);
-                list_del(&c->c_link);
-                list_add(&c->c_link, &conn_unused_list);
-                spin_unlock(&conn_lock);
-                rc = 1;
-        }
-        if (atomic_read(&c->c_refcount) < 0)
-                CERROR("connection %p refcount %d!\n",
-                       c, atomic_read(&c->c_refcount));
-
-        RETURN(rc);
+static int
+conn_keycmp(const void *key, struct hlist_node *hnode)
+{
+       struct ptlrpc_connection *conn;
+       const struct lnet_process_id *conn_key;
+
+       LASSERT(key != NULL);
+       conn_key = (struct lnet_process_id *)key;
+       conn = hlist_entry(hnode, struct ptlrpc_connection, c_hash);
+
+       return conn_key->nid == conn->c_peer.nid &&
+               conn_key->pid == conn->c_peer.pid;
 }
 
-struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *c)
+static void *
+conn_key(struct hlist_node *hnode)
 {
-        ENTRY;
-        CDEBUG(D_INFO, "connection=%p refcount %d\n",
-               c, atomic_read(&c->c_refcount) + 1);
-        atomic_inc(&c->c_refcount);
-        RETURN(c);
+       struct ptlrpc_connection *conn;
+       conn = hlist_entry(hnode, struct ptlrpc_connection, c_hash);
+       return &conn->c_peer;
 }
 
-void ptlrpc_init_connection(void)
+static void *
+conn_object(struct hlist_node *hnode)
 {
-        INIT_LIST_HEAD(&conn_list);
-        INIT_LIST_HEAD(&conn_unused_list);
-        conn_lock = SPIN_LOCK_UNLOCKED;
+       return hlist_entry(hnode, struct ptlrpc_connection, c_hash);
 }
 
-void ptlrpc_cleanup_connection(void)
+static void
+conn_get(struct cfs_hash *hs, struct hlist_node *hnode)
 {
-        struct list_head *tmp, *pos;
-        struct ptlrpc_connection *c;
-
-        spin_lock(&conn_lock);
-        list_for_each_safe(tmp, pos, &conn_unused_list) {
-                c = list_entry(tmp, struct ptlrpc_connection, c_link);
-                list_del(&c->c_link);
-                OBD_FREE(c, sizeof(*c));
-        }
-        list_for_each_safe(tmp, pos, &conn_list) {
-                c = list_entry(tmp, struct ptlrpc_connection, c_link);
-                CERROR("Connection %p has refcount %d at cleanup (nid=%lu)!\n",
-                       c, atomic_read(&c->c_refcount),
-                       (unsigned long)c->c_peer.peer_nid);
-                list_del(&c->c_link);
-                OBD_FREE(c, sizeof(*c));
-        }
-        spin_unlock(&conn_lock);
+       struct ptlrpc_connection *conn;
+
+       conn = hlist_entry(hnode, struct ptlrpc_connection, c_hash);
+       atomic_inc(&conn->c_refcount);
 }
+
+static void
+conn_put_locked(struct cfs_hash *hs, struct hlist_node *hnode)
+{
+       struct ptlrpc_connection *conn;
+
+       conn = hlist_entry(hnode, struct ptlrpc_connection, c_hash);
+       atomic_dec(&conn->c_refcount);
+}
+
+static void
+conn_exit(struct cfs_hash *hs, struct hlist_node *hnode)
+{
+       struct ptlrpc_connection *conn;
+
+       conn = hlist_entry(hnode, struct ptlrpc_connection, c_hash);
+       /*
+        * Nothing should be left. Connection user put it and
+        * connection also was deleted from table by this time
+        * so we should have 0 refs.
+        */
+       LASSERTF(atomic_read(&conn->c_refcount) == 0,
+                "Busy connection with %d refs\n",
+                atomic_read(&conn->c_refcount));
+       OBD_FREE_PTR(conn);
+}
+
+static struct cfs_hash_ops conn_hash_ops = {
+       .hs_hash        = conn_hashfn,
+       .hs_keycmp      = conn_keycmp,
+       .hs_key         = conn_key,
+       .hs_object      = conn_object,
+       .hs_get         = conn_get,
+       .hs_put_locked  = conn_put_locked,
+       .hs_exit        = conn_exit,
+};