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