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 #include <linux/lprocfs_status.h>
  34
  35
  36
  37 extern int ptlrpc_init_portals(void);
  38 extern void ptlrpc_exit_portals(void);
  39
  40 extern lprocfs_vars_t status_var_nm_1[];
  41 extern lprocfs_vars_t status_class_var[];
  42
  43 int connmgr_setup(struct obd_device *obddev, obd_count len, void *buf)
  44 {
  45         struct recovd_obd *recovd = &obddev->u.recovd;
  46         int err;
  47         ENTRY;
  48
  49         MOD_INC_USE_COUNT;
  50         memset(recovd, 0, sizeof(*recovd));
  51
  52         err = recovd_setup(recovd);
  53         if (err) {
  54                 MOD_DEC_USE_COUNT;
  55                 RETURN(err);
  56         }
  57
  58         RETURN(0);
  59 }
  60
  61 int connmgr_cleanup(struct obd_device *dev)
  62 {
  63         struct recovd_obd *recovd = &dev->u.recovd;
  64         int err;
  65
  66         err = recovd_cleanup(recovd);
  67         if (err)
  68                 LBUG();
  69
  70         MOD_DEC_USE_COUNT;
  71         RETURN(0);
  72 }
  73
  74 int connmgr_iocontrol(long cmd, struct lustre_handle *hdl, int len, void *karg,
  75                       void *uarg)
  76 {
  77         struct ptlrpc_connection *conn = NULL;
  78         struct obd_device *obd = class_conn2obd(hdl);
  79         struct recovd_obd *recovd = &obd->u.recovd;
  80         struct obd_ioctl_data *data = karg;
  81         struct list_head *tmp;
  82         int rc = 0;
  83
  84         ENTRY;
  85
  86         if (cmd != OBD_IOC_RECOVD_NEWCONN && cmd != OBD_IOC_RECOVD_FAILCONN)
  87                 RETURN(-EINVAL); /* XXX ENOSYS? */
  88
  89         /* Find the connection that's been rebuilt or has failed. */
  90         spin_lock(&recovd->recovd_lock);
  91         list_for_each(tmp, &recovd->recovd_troubled_items) {
  92                 conn = list_entry(tmp, struct ptlrpc_connection,
  93                                   c_recovd_data.rd_managed_chain);
  94
  95                 LASSERT(conn->c_recovd_data.rd_recovd == recovd); /* sanity */
  96
  97                 if (!strcmp(conn->c_remote_uuid, data->ioc_inlbuf1))
  98                         break;
  99                 conn = NULL;
 100         }
 101
 102         if (!conn) {
 103                 if (cmd == OBD_IOC_RECOVD_NEWCONN)
 104                         GOTO(out, rc = -EINVAL);
 105                 /* XXX macroize/inline and share with loop above */
 106                 list_for_each(tmp, &recovd->recovd_managed_items) {
 107                         conn = list_entry(tmp, struct ptlrpc_connection,
 108                                           c_recovd_data.rd_managed_chain);
 109
 110                         LASSERT(conn->c_recovd_data.rd_recovd == recovd);
 111
 112                         if (!strcmp(conn->c_remote_uuid, data->ioc_inlbuf1))
 113                                 break;
 114                         conn = NULL;
 115                 }
 116                 if (!conn)
 117                         GOTO(out, rc = -EINVAL);
 118         }
 119
 120         if (cmd == OBD_IOC_RECOVD_FAILCONN) {
 121                 spin_unlock(&recovd->recovd_lock);
 122                 recovd_conn_fail(conn);
 123                 spin_lock(&recovd->recovd_lock);
 124
 125                 /* Jump straight to the "failed" phase of recovery. */
 126                 conn->c_recovd_data.rd_phase = RD_FAILED;
 127                 goto out;
 128         }
 129
 130
 131         /* else (NEWCONN) */
 132         spin_lock(&conn->c_lock);
 133
 134         /* whatever happens, reset the INVALID flag */
 135         conn->c_flags &= ~CONN_INVALID;
 136
 137         /* XXX is this a good check?  should we allow readdressing of
 138          * XXX conns that aren't in recovery?
 139          */
 140         if (conn->c_recovd_data.rd_phase != RD_PREPARING) {
 141                 spin_unlock(&conn->c_lock);
 142                 GOTO(out, rc = -EALREADY);
 143         }
 144
 145         if (data->ioc_inllen2) {
 146                 CERROR("conn %p UUID change %s -> %s\n",
 147                        conn, conn->c_remote_uuid, data->ioc_inlbuf2);
 148                 strcpy(conn->c_remote_uuid, data->ioc_inlbuf2);
 149         } else {
 150                 CERROR("conn %p UUID %s reconnected\n", conn,
 151                        conn->c_remote_uuid);
 152         }
 153         ptlrpc_readdress_connection(conn, conn->c_remote_uuid);
 154         spin_unlock(&conn->c_lock);
 155
 156         conn->c_recovd_data.rd_phase = RD_PREPARED;
 157         wake_up(&recovd->recovd_waitq);
 158  out:
 159         spin_unlock(&recovd->recovd_lock);
 160         RETURN(rc);
 161 }
 162
 163 static int connmgr_connect(struct lustre_handle *conn, struct obd_device *src,
 164                            obd_uuid_t cluuid, struct recovd_obd *recovd,
 165                            ptlrpc_recovery_cb_t recover)
 166 {
 167         return class_connect(conn, src, cluuid);
 168 }
 169 int connmgr_attach(struct obd_device *dev,
 170                    obd_count len, void *data)
 171 {
 172         int rc;
 173         rc = lprocfs_reg_obd(dev, (lprocfs_vars_t*)status_var_nm_1, (void*)dev);
 174         return rc;
 175 }
 176
 177 int conmgr_detach(struct obd_device *dev)
 178 {
 179         int rc;
 180         rc = lprocfs_dereg_obd(dev);
 181         return rc;
 182
 183 }
 184 /* use obd ops to offer management infrastructure */
 185 static struct obd_ops recovd_obd_ops = {
 186         o_attach:      connmgr_attach,
 187         o_detach:      conmgr_detach,
 188         o_setup:       connmgr_setup,
 189         o_cleanup:     connmgr_cleanup,
 190         o_iocontrol:   connmgr_iocontrol,
 191         o_connect:     connmgr_connect,
 192         o_disconnect:  class_disconnect
 193 };
 194
 195 static int __init ptlrpc_init(void)
 196 {
 197         int rc;
 198         rc = ptlrpc_init_portals();
 199         if (rc)
 200                 RETURN(rc);
 201         ptlrpc_init_connection();
 202         rc = class_register_type(&recovd_obd_ops,
 203                                  (lprocfs_vars_t*)status_class_var,
 204                                  LUSTRE_HA_NAME);
 205         if (rc)
 206                 RETURN(rc);
 207         ptlrpc_put_connection_superhack = ptlrpc_put_connection;
 208
 209         if (rc)
 210                 RETURN(rc);
 211
 212         return 0;
 213 }
 214
 215 static void __exit ptlrpc_exit(void)
 216 {
 217         class_unregister_type(LUSTRE_HA_NAME);
 218         ptlrpc_exit_portals();
 219         ptlrpc_cleanup_connection();
 220 }
 221
 222 /* recovd.c */
 223 EXPORT_SYMBOL(ptlrpc_recovd);
 224 EXPORT_SYMBOL(recovd_conn_fail);
 225 EXPORT_SYMBOL(recovd_conn_manage);
 226 EXPORT_SYMBOL(recovd_conn_fixed);
 227 EXPORT_SYMBOL(recovd_setup);
 228 EXPORT_SYMBOL(recovd_cleanup);
 229
 230 /* connection.c */
 231 EXPORT_SYMBOL(ptlrpc_readdress_connection);
 232 EXPORT_SYMBOL(ptlrpc_get_connection);
 233 EXPORT_SYMBOL(ptlrpc_put_connection);
 234 EXPORT_SYMBOL(ptlrpc_connection_addref);
 235 EXPORT_SYMBOL(ptlrpc_init_connection);
 236 EXPORT_SYMBOL(ptlrpc_cleanup_connection);
 237
 238 /* niobuf.c */
 239 EXPORT_SYMBOL(ptlrpc_send_bulk);
 240 EXPORT_SYMBOL(ptlrpc_register_bulk);
 241 EXPORT_SYMBOL(ptlrpc_abort_bulk);
 242 EXPORT_SYMBOL(ptlrpc_reply);
 243 EXPORT_SYMBOL(ptlrpc_error);
 244 EXPORT_SYMBOL(ptlrpc_resend_req);
 245 EXPORT_SYMBOL(ptl_send_rpc);
 246 EXPORT_SYMBOL(ptlrpc_link_svc_me);
 247
 248 /* client.c */
 249 EXPORT_SYMBOL(ptlrpc_init_client);
 250 EXPORT_SYMBOL(ptlrpc_cleanup_client);
 251 EXPORT_SYMBOL(ptlrpc_req_to_uuid);
 252 EXPORT_SYMBOL(ptlrpc_uuid_to_connection);
 253 EXPORT_SYMBOL(ptlrpc_queue_wait);
 254 EXPORT_SYMBOL(ptlrpc_continue_req);
 255 EXPORT_SYMBOL(ptlrpc_replay_req);
 256 EXPORT_SYMBOL(ptlrpc_restart_req);
 257 EXPORT_SYMBOL(ptlrpc_prep_req);
 258 EXPORT_SYMBOL(ptlrpc_free_req);
 259 EXPORT_SYMBOL(ptlrpc_req_finished);
 260 EXPORT_SYMBOL(ptlrpc_prep_bulk);
 261 EXPORT_SYMBOL(ptlrpc_free_bulk);
 262 EXPORT_SYMBOL(ptlrpc_prep_bulk_page);
 263 EXPORT_SYMBOL(ptlrpc_free_bulk_page);
 264 EXPORT_SYMBOL(ptlrpc_check_status);
 265
 266 /* service.c */
 267 EXPORT_SYMBOL(ptlrpc_init_svc);
 268 EXPORT_SYMBOL(ptlrpc_stop_all_threads);
 269 EXPORT_SYMBOL(ptlrpc_start_thread);
 270 EXPORT_SYMBOL(ptlrpc_unregister_service);
 271
 272 /* pack_generic.c */
 273 EXPORT_SYMBOL(lustre_pack_msg);
 274 EXPORT_SYMBOL(lustre_msg_size);
 275 EXPORT_SYMBOL(lustre_unpack_msg);
 276 EXPORT_SYMBOL(lustre_msg_buf);
 277
 278 /* recover.c */
 279 EXPORT_SYMBOL(ptlrpc_run_recovery_upcall);
 280 EXPORT_SYMBOL(ptlrpc_reconnect_import);
 281 EXPORT_SYMBOL(ptlrpc_replay);
 282
 283 MODULE_AUTHOR("Cluster File Systems, Inc <info@clusterfs.com>");
 284 MODULE_DESCRIPTION("Lustre Request Processor v1.0");
 285 MODULE_LICENSE("GPL");
 286
 287 module_init(ptlrpc_init);
 288 module_exit(ptlrpc_exit);