lustre/ptlrpc/recovd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  obd/rpc/recovd.c
   5  *
   6  *  Lustre High Availability Daemon
   7  *
   8  *  Copyright (C) 2001, 2002 Cluster File Systems, Inc.
   9  *
  10  *  This code is issued under the GNU General Public License.
  11  *  See the file COPYING in this distribution
  12  *
  13  *  by Peter Braam <braam@clusterfs.com>
  14  *
  15  */
  16
  17 #define DEBUG_SUBSYSTEM S_RPC
  18
  19 #include <linux/lustre_lite.h>
  20 #include <linux/lustre_ha.h>
  21 #include <linux/obd_support.h>
  22
  23 void recovd_conn_manage(struct ptlrpc_connection *conn,
  24                         struct recovd_obd *recovd, ptlrpc_recovery_cb_t recover)
  25 {
  26         struct recovd_data *rd = &conn->c_recovd_data;
  27         ENTRY;
  28
  29         rd->rd_recovd = recovd;
  30         rd->rd_recover = recover;
  31         rd->rd_phase = RD_IDLE;
  32         rd->rd_next_phase = RD_TROUBLED;
  33
  34         spin_lock(&recovd->recovd_lock);
  35         list_add(&rd->rd_managed_chain, &recovd->recovd_managed_items);
  36         spin_unlock(&recovd->recovd_lock);
  37
  38         EXIT;
  39 }
  40
  41 void recovd_conn_fail(struct ptlrpc_connection *conn)
  42 {
  43         struct recovd_data *rd = &conn->c_recovd_data;
  44         struct recovd_obd *recovd = rd->rd_recovd;
  45         ENTRY;
  46
  47         if (!recovd) {
  48                 CERROR("no recovd for connection %p\n", conn);
  49                 EXIT;
  50                 return;
  51         }
  52
  53
  54         spin_lock(&recovd->recovd_lock);
  55         if (rd->rd_phase != RD_IDLE) {
  56                 CDEBUG(D_INFO, "connection %p to %s already in recovery\n",
  57                        conn, conn->c_remote_uuid);
  58                 /* XXX need to distinguish from failure-in-recovery */
  59                 spin_unlock(&recovd->recovd_lock);
  60                 EXIT;
  61                 return;
  62         }
  63
  64         CERROR("connection %p to %s failed\n", conn, conn->c_remote_uuid);
  65         list_del(&rd->rd_managed_chain);
  66         list_add_tail(&rd->rd_managed_chain, &recovd->recovd_troubled_items);
  67         rd->rd_phase = RD_TROUBLED;
  68         spin_unlock(&recovd->recovd_lock);
  69
  70         wake_up(&recovd->recovd_waitq);
  71
  72         EXIT;
  73 }
  74
  75 /* this function must be called with recovd->recovd_lock held */
  76 void recovd_conn_fixed(struct ptlrpc_connection *conn)
  77 {
  78         struct recovd_data *rd = &conn->c_recovd_data;
  79         ENTRY;
  80
  81         spin_lock(&rd->rd_recovd->recovd_lock);
  82         list_del(&rd->rd_managed_chain);
  83         rd->rd_phase = RD_IDLE;
  84         rd->rd_next_phase = RD_TROUBLED;
  85         list_add(&rd->rd_managed_chain, &rd->rd_recovd->recovd_managed_items);
  86         spin_unlock(&rd->rd_recovd->recovd_lock);
  87
  88         EXIT;
  89 }
  90
  91
  92 static int recovd_check_event(struct recovd_obd *recovd)
  93 {
  94         int rc = 0;
  95         struct list_head *tmp;
  96
  97         ENTRY;
  98
  99         spin_lock(&recovd->recovd_lock);
 100
 101         if (recovd->recovd_state == RECOVD_STOPPING)
 102                 GOTO(out, rc = 1);
 103
 104         list_for_each(tmp, &recovd->recovd_troubled_items) {
 105
 106                 struct recovd_data *rd = list_entry(tmp, struct recovd_data,
 107                                                     rd_managed_chain);
 108
 109                 if (rd->rd_phase == rd->rd_next_phase ||
 110                     rd->rd_phase == RD_FAILED)
 111                         GOTO(out, rc = 1);
 112         }
 113
 114  out:
 115         spin_unlock(&recovd->recovd_lock);
 116         RETURN(rc);
 117 }
 118
 119 static void dump_connection_list(struct list_head *head)
 120 {
 121         struct list_head *tmp;
 122
 123         list_for_each(tmp, head) {
 124                 struct ptlrpc_connection *conn =
 125                         list_entry(tmp, struct ptlrpc_connection,
 126                                    c_recovd_data.rd_managed_chain);
 127                 CDEBUG(D_NET, "   %p = %s\n", conn, conn->c_remote_uuid);
 128         }
 129 }
 130
 131 static int recovd_handle_event(struct recovd_obd *recovd)
 132 {
 133         struct list_head *tmp, *n;
 134         int rc = 0;
 135         ENTRY;
 136
 137         spin_lock(&recovd->recovd_lock);
 138
 139         CDEBUG(D_NET, "managed: \n");
 140         dump_connection_list(&recovd->recovd_managed_items);
 141         CDEBUG(D_NET, "troubled: \n");
 142         dump_connection_list(&recovd->recovd_troubled_items);
 143
 144         /*
 145          * We use _safe here because one of the callbacks, expecially
 146          * FAILURE or PREPARED, could move list items around.
 147          */
 148         list_for_each_safe(tmp, n, &recovd->recovd_troubled_items) {
 149                 struct recovd_data *rd = list_entry(tmp, struct recovd_data,
 150                                                     rd_managed_chain);
 151
 152                 if (rd->rd_phase != RD_FAILED &&
 153                     rd->rd_phase != rd->rd_next_phase)
 154                         continue;
 155
 156                 switch (rd->rd_phase) {
 157                     case RD_FAILED:
 158                 cb_failed: /* must always reach here with recovd_lock held! */
 159                         CERROR("recovery FAILED for rd %p (conn %p): %d\n",
 160                                rd, class_rd2conn(rd), rc);
 161
 162                         spin_unlock(&recovd->recovd_lock);
 163                         (void)rd->rd_recover(rd, PTLRPC_RECOVD_PHASE_FAILURE);
 164                         spin_lock(&recovd->recovd_lock);
 165                         break;
 166
 167                     case RD_TROUBLED:
 168                         if (!rd->rd_recover) {
 169                                 CERROR("no rd_recover for rd %p (conn %p)\n",
 170                                        rd, class_rd2conn(rd));
 171                                 rc = -EINVAL;
 172                                 break;
 173                         }
 174                         CERROR("starting recovery for rd %p (conn %p)\n",
 175                                rd, class_rd2conn(rd));
 176                         rd->rd_phase = RD_PREPARING;
 177
 178                         spin_unlock(&recovd->recovd_lock);
 179                         rc = rd->rd_recover(rd, PTLRPC_RECOVD_PHASE_PREPARE);
 180                         spin_lock(&recovd->recovd_lock);
 181                         if (rc)
 182                                 goto cb_failed;
 183
 184                         rd->rd_next_phase = RD_PREPARED;
 185                         break;
 186
 187                     case RD_PREPARED:
 188                         rd->rd_phase = RD_RECOVERING;
 189
 190                         CERROR("recovery prepared for rd %p (conn %p)\n",
 191                                rd, class_rd2conn(rd));
 192
 193                         spin_unlock(&recovd->recovd_lock);
 194                         rc = rd->rd_recover(rd, PTLRPC_RECOVD_PHASE_RECOVER);
 195                         spin_lock(&recovd->recovd_lock);
 196                         if (rc)
 197                                 goto cb_failed;
 198
 199                         rd->rd_next_phase = RD_RECOVERED;
 200                         break;
 201
 202                     case RD_RECOVERED:
 203                         rd->rd_phase = RD_IDLE;
 204                         rd->rd_next_phase = RD_TROUBLED;
 205
 206                         CERROR("recovery complete for rd %p (conn %p)\n",
 207                                rd, class_rd2conn(rd));
 208                         break;
 209
 210                     default:
 211                         break;
 212                 }
 213         }
 214         spin_unlock(&recovd->recovd_lock);
 215         RETURN(0);
 216 }
 217
 218 static int recovd_main(void *arg)
 219 {
 220         struct recovd_obd *recovd = (struct recovd_obd *)arg;
 221
 222         ENTRY;
 223
 224         lock_kernel();
 225         daemonize();
 226         spin_lock_irq(&current->sigmask_lock);
 227         sigfillset(&current->blocked);
 228         recalc_sigpending(current);
 229         spin_unlock_irq(&current->sigmask_lock);
 230
 231         sprintf(current->comm, "lustre_recovd");
 232         unlock_kernel();
 233
 234         /* Signal that the thread is running. */
 235         recovd->recovd_thread = current;
 236         recovd->recovd_state = RECOVD_READY;
 237         wake_up(&recovd->recovd_ctl_waitq);
 238
 239         /* And now, loop forever on requests. */
 240         while (1) {
 241                 wait_event(recovd->recovd_waitq, recovd_check_event(recovd));
 242                 if (recovd->recovd_state == RECOVD_STOPPING)
 243                         break;
 244                 recovd_handle_event(recovd);
 245         }
 246
 247         recovd->recovd_thread = NULL;
 248         recovd->recovd_state = RECOVD_STOPPED;
 249         wake_up(&recovd->recovd_ctl_waitq);
 250         CDEBUG(D_NET, "mgr exiting process %d\n", current->pid);
 251         RETURN(0);
 252 }
 253
 254 int recovd_setup(struct recovd_obd *recovd)
 255 {
 256         int rc;
 257
 258         ENTRY;
 259
 260         INIT_LIST_HEAD(&recovd->recovd_managed_items);
 261         INIT_LIST_HEAD(&recovd->recovd_troubled_items);
 262         spin_lock_init(&recovd->recovd_lock);
 263
 264         init_waitqueue_head(&recovd->recovd_waitq);
 265         init_waitqueue_head(&recovd->recovd_recovery_waitq);
 266         init_waitqueue_head(&recovd->recovd_ctl_waitq);
 267
 268         rc = kernel_thread(recovd_main, (void *)recovd,
 269                            CLONE_VM | CLONE_FS | CLONE_FILES);
 270         if (rc < 0) {
 271                 CERROR("cannot start thread\n");
 272                 RETURN(-EINVAL);
 273         }
 274         wait_event(recovd->recovd_ctl_waitq,
 275                    recovd->recovd_state == RECOVD_READY);
 276
 277         ptlrpc_recovd = recovd;
 278
 279         RETURN(0);
 280 }
 281
 282 int recovd_cleanup(struct recovd_obd *recovd)
 283 {
 284         spin_lock(&recovd->recovd_lock);
 285         recovd->recovd_state = RECOVD_STOPPING;
 286         wake_up(&recovd->recovd_waitq);
 287         spin_unlock(&recovd->recovd_lock);
 288
 289         wait_event(recovd->recovd_ctl_waitq,
 290                    (recovd->recovd_state == RECOVD_STOPPED));
 291         RETURN(0);
 292 }
 293
 294 struct recovd_obd *ptlrpc_recovd;