lustre/ptlrpc/rpc.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (C) 2002 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  *
  21  */
  22
  23 #define EXPORT_SYMTAB
  24 #define DEBUG_SUBSYSTEM S_RPC
  25
  26 #include <linux/module.h>
  27 #include <linux/obd_support.h>
  28 #include <linux/obd_class.h>
  29 #include <linux/lustre_lib.h>
  30 #include <linux/lustre_ha.h>
  31 #include <linux/lustre_net.h>
  32 #include <linux/init.h>
  33
  34 extern int ptlrpc_init_portals(void);
  35 extern void ptlrpc_exit_portals(void);
  36
  37 int connmgr_setup(struct obd_device *obddev, obd_count len, void *buf)
  38 {
  39         struct recovd_obd *recovd = &obddev->u.recovd;
  40         int err;
  41         ENTRY;
  42
  43         MOD_INC_USE_COUNT;
  44         memset(recovd, 0, sizeof(*recovd));
  45
  46         err = recovd_setup(recovd);
  47         if (err) {
  48                 MOD_DEC_USE_COUNT;
  49                 RETURN(err);
  50         }
  51
  52         RETURN(0);
  53 }
  54
  55 int connmgr_cleanup(struct obd_device *dev)
  56 {
  57         struct recovd_obd *recovd = &dev->u.recovd;
  58         int err;
  59
  60         err = recovd_cleanup(recovd);
  61         if (err)
  62                 LBUG();
  63
  64         MOD_DEC_USE_COUNT;
  65         RETURN(0);
  66 }
  67
  68 int connmgr_iocontrol(long cmd, struct lustre_handle *hdl, int len, void *karg,
  69                       void *uarg)
  70 {
  71         struct ptlrpc_connection *conn = NULL;
  72         struct obd_device *obd = class_conn2obd(hdl);
  73         struct recovd_obd *recovd = &obd->u.recovd;
  74         struct obd_ioctl_data *data = karg;
  75         struct list_head *tmp;
  76         int rc = 0;
  77
  78         ENTRY;
  79
  80         if (cmd != OBD_IOC_RECOVD_NEWCONN && cmd != OBD_IOC_RECOVD_FAILCONN)
  81                 RETURN(-EINVAL); /* XXX ENOSYS? */
  82
  83         /* Find the connection that's been rebuilt or has failed. */
  84         spin_lock(&recovd->recovd_lock);
  85         list_for_each(tmp, &recovd->recovd_troubled_items) {
  86                 conn = list_entry(tmp, struct ptlrpc_connection,
  87                                   c_recovd_data.rd_managed_chain);
  88
  89                 LASSERT(conn->c_recovd_data.rd_recovd == recovd); /* sanity */
  90
  91                 if (!strcmp(conn->c_remote_uuid, data->ioc_inlbuf1))
  92                         break;
  93                 conn = NULL;
  94         }
  95
  96         if (!conn) {
  97                 if (cmd == OBD_IOC_RECOVD_NEWCONN)
  98                         GOTO(out, rc = -EINVAL);
  99                 /* XXX macroize/inline and share with loop above */
 100                 list_for_each(tmp, &recovd->recovd_managed_items) {
 101                         conn = list_entry(tmp, struct ptlrpc_connection,
 102                                           c_recovd_data.rd_managed_chain);
 103
 104                         LASSERT(conn->c_recovd_data.rd_recovd == recovd);
 105
 106                         if (!strcmp(conn->c_remote_uuid, data->ioc_inlbuf1))
 107                                 break;
 108                         conn = NULL;
 109                 }
 110                 if (!conn)
 111                         GOTO(out, rc = -EINVAL);
 112         }
 113
 114         if (cmd == OBD_IOC_RECOVD_FAILCONN) {
 115                 spin_unlock(&recovd->recovd_lock);
 116                 recovd_conn_fail(conn);
 117                 spin_lock(&recovd->recovd_lock);
 118
 119                 /* Jump straight to the "failed" phase of recovery. */
 120                 conn->c_recovd_data.rd_phase = RD_FAILED;
 121                 goto out;
 122         }
 123
 124
 125         /* else (NEWCONN) */
 126         spin_lock(&conn->c_lock);
 127
 128         /* whatever happens, reset the INVALID flag */
 129         conn->c_flags &= ~CONN_INVALID;
 130
 131         /* XXX is this a good check?  should we allow readdressing of
 132          * XXX conns that aren't in recovery?
 133          */
 134         if (conn->c_recovd_data.rd_phase != RD_PREPARING) {
 135                 spin_unlock(&conn->c_lock);
 136                 GOTO(out, rc = -EALREADY);
 137         }
 138
 139         if (data->ioc_inllen2) {
 140                 CERROR("conn %p UUID change %s -> %s\n",
 141                        conn, conn->c_remote_uuid, data->ioc_inlbuf2);
 142                 strcpy(conn->c_remote_uuid, data->ioc_inlbuf2);
 143         } else {
 144                 CERROR("conn %p UUID %s reconnected\n", conn,
 145                        conn->c_remote_uuid);
 146         }
 147         ptlrpc_readdress_connection(conn, conn->c_remote_uuid);
 148         spin_unlock(&conn->c_lock);
 149
 150         conn->c_recovd_data.rd_phase = RD_PREPARED;
 151         wake_up(&recovd->recovd_waitq);
 152  out:
 153         spin_unlock(&recovd->recovd_lock);
 154         RETURN(rc);
 155 }
 156
 157 static int connmgr_connect(struct lustre_handle *conn, struct obd_device *src,
 158                            obd_uuid_t cluuid, struct recovd_obd *recovd,
 159                            ptlrpc_recovery_cb_t recover)
 160 {
 161         return class_connect(conn, src, cluuid);
 162 }
 163
 164 /* use obd ops to offer management infrastructure */
 165 static struct obd_ops recovd_obd_ops = {
 166         o_setup:       connmgr_setup,
 167         o_cleanup:     connmgr_cleanup,
 168         o_iocontrol:   connmgr_iocontrol,
 169         o_connect:     connmgr_connect,
 170         o_disconnect:  class_disconnect
 171 };
 172
 173 static int __init ptlrpc_init(void)
 174 {
 175         int rc;
 176         rc = ptlrpc_init_portals();
 177         if (rc)
 178                 RETURN(rc);
 179         ptlrpc_init_connection();
 180         class_register_type(&recovd_obd_ops, LUSTRE_HA_NAME);
 181         ptlrpc_put_connection_superhack = ptlrpc_put_connection;
 182         return 0;
 183 }
 184
 185 static void __exit ptlrpc_exit(void)
 186 {
 187         class_unregister_type(LUSTRE_HA_NAME);
 188         ptlrpc_exit_portals();
 189         ptlrpc_cleanup_connection();
 190 }
 191
 192 /* recovd.c */
 193 EXPORT_SYMBOL(ptlrpc_recovd);
 194 EXPORT_SYMBOL(recovd_conn_fail);
 195 EXPORT_SYMBOL(recovd_conn_manage);
 196 EXPORT_SYMBOL(recovd_conn_fixed);
 197 EXPORT_SYMBOL(recovd_setup);
 198 EXPORT_SYMBOL(recovd_cleanup);
 199
 200 /* connection.c */
 201 EXPORT_SYMBOL(ptlrpc_readdress_connection);
 202 EXPORT_SYMBOL(ptlrpc_get_connection);
 203 EXPORT_SYMBOL(ptlrpc_put_connection);
 204 EXPORT_SYMBOL(ptlrpc_connection_addref);
 205 EXPORT_SYMBOL(ptlrpc_init_connection);
 206 EXPORT_SYMBOL(ptlrpc_cleanup_connection);
 207
 208 /* niobuf.c */
 209 EXPORT_SYMBOL(ptlrpc_send_bulk);
 210 EXPORT_SYMBOL(ptlrpc_register_bulk);
 211 EXPORT_SYMBOL(ptlrpc_abort_bulk);
 212 EXPORT_SYMBOL(ptlrpc_reply);
 213 EXPORT_SYMBOL(ptlrpc_error);
 214 EXPORT_SYMBOL(ptlrpc_resend_req);
 215 EXPORT_SYMBOL(ptl_send_rpc);
 216 EXPORT_SYMBOL(ptlrpc_link_svc_me);
 217
 218 /* client.c */
 219 EXPORT_SYMBOL(ptlrpc_init_client);
 220 EXPORT_SYMBOL(ptlrpc_cleanup_client);
 221 EXPORT_SYMBOL(ptlrpc_req_to_uuid);
 222 EXPORT_SYMBOL(ptlrpc_uuid_to_connection);
 223 EXPORT_SYMBOL(ptlrpc_queue_wait);
 224 EXPORT_SYMBOL(ptlrpc_continue_req);
 225 EXPORT_SYMBOL(ptlrpc_replay_req);
 226 EXPORT_SYMBOL(ptlrpc_restart_req);
 227 EXPORT_SYMBOL(ptlrpc_prep_req);
 228 EXPORT_SYMBOL(ptlrpc_free_req);
 229 EXPORT_SYMBOL(ptlrpc_req_finished);
 230 EXPORT_SYMBOL(ptlrpc_prep_bulk);
 231 EXPORT_SYMBOL(ptlrpc_free_bulk);
 232 EXPORT_SYMBOL(ptlrpc_prep_bulk_page);
 233 EXPORT_SYMBOL(ptlrpc_free_bulk_page);
 234 EXPORT_SYMBOL(ptlrpc_check_status);
 235
 236 /* service.c */
 237 EXPORT_SYMBOL(ptlrpc_init_svc);
 238 EXPORT_SYMBOL(ptlrpc_stop_all_threads);
 239 EXPORT_SYMBOL(ptlrpc_start_thread);
 240 EXPORT_SYMBOL(ptlrpc_unregister_service);
 241
 242 /* pack_generic.c */
 243 EXPORT_SYMBOL(lustre_pack_msg);
 244 EXPORT_SYMBOL(lustre_msg_size);
 245 EXPORT_SYMBOL(lustre_unpack_msg);
 246 EXPORT_SYMBOL(lustre_msg_buf);
 247
 248 /* recover.c */
 249 EXPORT_SYMBOL(ptlrpc_run_recovery_upcall);
 250 EXPORT_SYMBOL(ptlrpc_reconnect_import);
 251 EXPORT_SYMBOL(ptlrpc_replay);
 252
 253 MODULE_AUTHOR("Cluster File Systems, Inc <info@clusterfs.com>");
 254 MODULE_DESCRIPTION("Lustre Request Processor v1.0");
 255 MODULE_LICENSE("GPL");
 256
 257 module_init(ptlrpc_init);
 258 module_exit(ptlrpc_exit);