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, 2003 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 #ifdef __KERNEL__
  27 # include <linux/module.h>
  28 # include <linux/init.h>
  29 #else
  30 # include <liblustre.h>
  31 #endif
  32 #include <linux/obd.h>
  33 #include <linux/obd_support.h>
  34 #include <linux/obd_class.h>
  35 #include <linux/lustre_lib.h>
  36 #include <linux/lustre_ha.h>
  37 #include <linux/lustre_net.h>
  38 #include <linux/lprocfs_status.h>
  39
  40 extern int ptlrpc_init_portals(void);
  41 extern void ptlrpc_exit_portals(void);
  42
  43 static __u32 ptlrpc_last_xid = 0;
  44 static spinlock_t ptlrpc_last_xid_lock = SPIN_LOCK_UNLOCKED;
  45
  46 __u32 ptlrpc_next_xid(void)
  47 {
  48         __u32 tmp;
  49         spin_lock(&ptlrpc_last_xid_lock);
  50         tmp = ++ptlrpc_last_xid;
  51         spin_unlock(&ptlrpc_last_xid_lock);
  52         return tmp;
  53 }
  54
  55 int connmgr_setup(struct obd_device *obddev, obd_count len, void *buf)
  56 {
  57         struct recovd_obd *recovd = &obddev->u.recovd;
  58         int err;
  59         ENTRY;
  60
  61         memset(recovd, 0, sizeof(*recovd));
  62
  63         err = recovd_setup(recovd);
  64         RETURN(err);
  65 }
  66
  67 int connmgr_cleanup(struct obd_device *dev)
  68 {
  69         struct recovd_obd *recovd = &dev->u.recovd;
  70         int err;
  71
  72         err = recovd_cleanup(recovd);
  73         RETURN(err);
  74 }
  75
  76 int connmgr_iocontrol(unsigned int cmd, struct lustre_handle *hdl, int len,
  77                       void *karg, void *uarg)
  78 {
  79         struct ptlrpc_connection *conn = NULL;
  80         struct obd_device *obd = class_conn2obd(hdl);
  81         struct recovd_obd *recovd = &obd->u.recovd;
  82         struct obd_ioctl_data *data = karg;
  83         struct list_head *tmp;
  84         int rc = 0;
  85
  86         ENTRY;
  87
  88         if (cmd != OBD_IOC_RECOVD_NEWCONN && cmd != OBD_IOC_RECOVD_FAILCONN)
  89                 RETURN(-EINVAL); /* XXX ENOSYS? */
  90
  91         /* Find the connection that's been rebuilt or has failed. */
  92         spin_lock(&recovd->recovd_lock);
  93         list_for_each(tmp, &recovd->recovd_troubled_items) {
  94                 conn = list_entry(tmp, struct ptlrpc_connection,
  95                                   c_recovd_data.rd_managed_chain);
  96
  97                 LASSERT(conn->c_recovd_data.rd_recovd == recovd); /* sanity */
  98 #warning check buffer overflow in next line
  99                 if (!strcmp(conn->c_remote_uuid.uuid, data->ioc_inlbuf1))
 100                         break;
 101                 conn = NULL;
 102         }
 103
 104         if (!conn) {
 105                 if (cmd == OBD_IOC_RECOVD_NEWCONN)
 106                         GOTO(out, rc = -EINVAL);
 107                 /* XXX macroize/inline and share with loop above */
 108                 list_for_each(tmp, &recovd->recovd_managed_items) {
 109                         conn = list_entry(tmp, struct ptlrpc_connection,
 110                                           c_recovd_data.rd_managed_chain);
 111
 112                         LASSERT(conn->c_recovd_data.rd_recovd == recovd);
 113
 114 #warning check buffer overflow in next line
 115                         if (!strcmp(conn->c_remote_uuid.uuid,
 116                                     data->ioc_inlbuf1))
 117                                 break;
 118                         conn = NULL;
 119                 }
 120                 if (!conn)
 121                         GOTO(out, rc = -EINVAL);
 122         }
 123
 124         if (cmd == OBD_IOC_RECOVD_FAILCONN) {
 125                 spin_unlock(&recovd->recovd_lock);
 126                 recovd_conn_fail(conn);
 127                 spin_lock(&recovd->recovd_lock);
 128                 goto out;
 129         }
 130
 131
 132         /* else (NEWCONN) */
 133         spin_lock(&conn->c_lock);
 134
 135         /* whatever happens, reset the INVALID flag */
 136         conn->c_flags &= ~CONN_INVALID;
 137
 138         /* XXX is this a good check?  should we allow readdressing of
 139          * XXX conns that aren't in recovery?
 140          */
 141         if (conn->c_recovd_data.rd_phase != RD_PREPARING) {
 142                 spin_unlock(&conn->c_lock);
 143                 GOTO(out, rc = -EALREADY);
 144         }
 145
 146         if (data->ioc_inllen2) {
 147                 CERROR("conn %p UUID change %s -> %s\n",
 148                        conn, conn->c_remote_uuid.uuid, data->ioc_inlbuf2);
 149                 obd_str2uuid(&conn->c_remote_uuid, data->ioc_inlbuf2);
 150         } else {
 151                 CERROR("conn %p UUID %s reconnected\n", conn,
 152                        conn->c_remote_uuid.uuid);
 153         }
 154         ptlrpc_readdress_connection(conn, &conn->c_remote_uuid);
 155         spin_unlock(&conn->c_lock);
 156
 157         conn->c_recovd_data.rd_phase = RD_PREPARED;
 158         wake_up(&recovd->recovd_waitq);
 159  out:
 160         spin_unlock(&recovd->recovd_lock);
 161         RETURN(rc);
 162 }
 163
 164 static int connmgr_connect(struct lustre_handle *conn, struct obd_device *src,
 165                            struct obd_uuid *cluuid, struct recovd_obd *recovd,
 166                            ptlrpc_recovery_cb_t recover)
 167 {
 168         return class_connect(conn, src, cluuid);
 169 }
 170
 171 int connmgr_attach(struct obd_device *dev, obd_count len, void *data)
 172 {
 173         struct lprocfs_static_vars lvars;
 174         int rc = 0;
 175
 176         lprocfs_init_vars(&lvars);
 177         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
 178         return rc;
 179 }
 180
 181 int conmgr_detach(struct obd_device *dev)
 182 {
 183         return lprocfs_obd_detach(dev);
 184 }
 185
 186 /* use obd ops to offer management infrastructure */
 187 static struct obd_ops recovd_obd_ops = {
 188         o_owner:        THIS_MODULE,
 189         o_attach:       connmgr_attach,
 190         o_detach:       conmgr_detach,
 191         o_setup:        connmgr_setup,
 192         o_cleanup:      connmgr_cleanup,
 193         o_iocontrol:    connmgr_iocontrol,
 194         o_connect:      connmgr_connect,
 195         o_disconnect:   class_disconnect
 196 };
 197
 198
 199
 200 __init int ptlrpc_init(void)
 201 {
 202         struct lprocfs_static_vars lvars;
 203         int rc;
 204         ENTRY;
 205
 206         rc = ptlrpc_init_portals();
 207         if (rc)
 208                 RETURN(rc);
 209         ptlrpc_init_connection();
 210
 211         lprocfs_init_vars(&lvars);
 212         rc = class_register_type(&recovd_obd_ops, lvars.module_vars,
 213                                  LUSTRE_HA_NAME);
 214         if (rc)
 215                 RETURN(rc);
 216         ptlrpc_put_connection_superhack = ptlrpc_put_connection;
 217         ptlrpc_abort_inflight_superhack = ptlrpc_abort_inflight;
 218         RETURN(0);
 219 }
 220
 221 static void __exit ptlrpc_exit(void)
 222 {
 223         class_unregister_type(LUSTRE_HA_NAME);
 224         ptlrpc_exit_portals();
 225         ptlrpc_cleanup_connection();
 226 }
 227
 228 /* rpc.c */
 229 EXPORT_SYMBOL(ptlrpc_next_xid);
 230
 231 /* recovd.c */
 232 EXPORT_SYMBOL(ptlrpc_recovd);
 233 EXPORT_SYMBOL(recovd_conn_fail);
 234 EXPORT_SYMBOL(recovd_conn_manage);
 235 EXPORT_SYMBOL(recovd_conn_fixed);
 236 EXPORT_SYMBOL(recovd_setup);
 237 EXPORT_SYMBOL(recovd_cleanup);
 238
 239 /* connection.c */
 240 EXPORT_SYMBOL(ptlrpc_readdress_connection);
 241 EXPORT_SYMBOL(ptlrpc_get_connection);
 242 EXPORT_SYMBOL(ptlrpc_put_connection);
 243 EXPORT_SYMBOL(ptlrpc_connection_addref);
 244 EXPORT_SYMBOL(ptlrpc_init_connection);
 245 EXPORT_SYMBOL(ptlrpc_cleanup_connection);
 246
 247 /* niobuf.c */
 248 EXPORT_SYMBOL(ptlrpc_bulk_put);
 249 EXPORT_SYMBOL(ptlrpc_bulk_get);
 250 EXPORT_SYMBOL(ptlrpc_register_bulk_put);
 251 EXPORT_SYMBOL(ptlrpc_register_bulk_get);
 252 EXPORT_SYMBOL(ptlrpc_abort_bulk);
 253 EXPORT_SYMBOL(ptlrpc_reply);
 254 EXPORT_SYMBOL(ptlrpc_error);
 255 EXPORT_SYMBOL(ptlrpc_resend_req);
 256 EXPORT_SYMBOL(ptl_send_rpc);
 257 EXPORT_SYMBOL(ptlrpc_link_svc_me);
 258 EXPORT_SYMBOL(obd_brw_set_new);
 259 EXPORT_SYMBOL(obd_brw_set_add);
 260 EXPORT_SYMBOL(obd_brw_set_del);
 261 EXPORT_SYMBOL(obd_brw_set_decref);
 262 EXPORT_SYMBOL(obd_brw_set_addref);
 263
 264 /* client.c */
 265 EXPORT_SYMBOL(ptlrpc_init_client);
 266 EXPORT_SYMBOL(ptlrpc_cleanup_client);
 267 EXPORT_SYMBOL(ptlrpc_req_to_uuid);
 268 EXPORT_SYMBOL(ptlrpc_uuid_to_connection);
 269 EXPORT_SYMBOL(ptlrpc_queue_wait);
 270 EXPORT_SYMBOL(ptlrpc_continue_req);
 271 EXPORT_SYMBOL(ptlrpc_replay_req);
 272 EXPORT_SYMBOL(ptlrpc_restart_req);
 273 EXPORT_SYMBOL(ptlrpc_prep_req);
 274 EXPORT_SYMBOL(ptlrpc_free_req);
 275 EXPORT_SYMBOL(ptlrpc_abort);
 276 EXPORT_SYMBOL(ptlrpc_req_finished);
 277 EXPORT_SYMBOL(ptlrpc_request_addref);
 278 EXPORT_SYMBOL(ptlrpc_prep_bulk);
 279 EXPORT_SYMBOL(ptlrpc_free_bulk);
 280 EXPORT_SYMBOL(ptlrpc_prep_bulk_page);
 281 EXPORT_SYMBOL(ptlrpc_free_bulk_page);
 282 EXPORT_SYMBOL(ll_brw_sync_wait);
 283 EXPORT_SYMBOL(ptlrpc_abort_inflight);
 284 EXPORT_SYMBOL(ptlrpc_retain_replayable_request);
 285
 286 /* service.c */
 287 EXPORT_SYMBOL(ptlrpc_init_svc);
 288 EXPORT_SYMBOL(ptlrpc_stop_all_threads);
 289 EXPORT_SYMBOL(ptlrpc_start_thread);
 290 EXPORT_SYMBOL(ptlrpc_unregister_service);
 291
 292 /* pack_generic.c */
 293 EXPORT_SYMBOL(lustre_pack_msg);
 294 EXPORT_SYMBOL(lustre_msg_size);
 295 EXPORT_SYMBOL(lustre_unpack_msg);
 296 EXPORT_SYMBOL(lustre_msg_buf);
 297
 298 /* recover.c */
 299 EXPORT_SYMBOL(ptlrpc_run_recovery_upcall);
 300 EXPORT_SYMBOL(ptlrpc_reconnect_import);
 301 EXPORT_SYMBOL(ptlrpc_replay);
 302 EXPORT_SYMBOL(ptlrpc_resend);
 303 EXPORT_SYMBOL(ptlrpc_wake_delayed);
 304
 305 #ifdef __KERNEL__
 306 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 307 MODULE_DESCRIPTION("Lustre Request Processor");
 308 MODULE_LICENSE("GPL");
 309
 310 module_init(ptlrpc_init);
 311 module_exit(ptlrpc_exit);
 312 #endif