fs/dlm/recoverd.c

   1 /******************************************************************************
   2 *******************************************************************************
   3 **
   4 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   5 **  Copyright (C) 2004-2011 Red Hat, Inc.  All rights reserved.
   6 **
   7 **  This copyrighted material is made available to anyone wishing to use,
   8 **  modify, copy, or redistribute it subject to the terms and conditions
   9 **  of the GNU General Public License v.2.
  10 **
  11 *******************************************************************************
  12 ******************************************************************************/
  13
  14 #include "dlm_internal.h"
  15 #include "lockspace.h"
  16 #include "member.h"
  17 #include "dir.h"
  18 #include "ast.h"
  19 #include "recover.h"
  20 #include "lowcomms.h"
  21 #include "lock.h"
  22 #include "requestqueue.h"
  23 #include "recoverd.h"
  24
  25
  26 /* If the start for which we're re-enabling locking (seq) has been superseded
  27    by a newer stop (ls_recover_seq), we need to leave locking disabled.
  28
  29    We suspend dlm_recv threads here to avoid the race where dlm_recv a) sees
  30    locking stopped and b) adds a message to the requestqueue, but dlm_recoverd
  31    enables locking and clears the requestqueue between a and b. */
  32
  33 static int enable_locking(struct dlm_ls *ls, uint64_t seq)
  34 {
  35         int error = -EINTR;
  36
  37         down_write(&ls->ls_recv_active);
  38
  39         spin_lock(&ls->ls_recover_lock);
  40         if (ls->ls_recover_seq == seq) {
  41                 set_bit(LSFL_RUNNING, &ls->ls_flags);
  42                 /* unblocks processes waiting to enter the dlm */
  43                 up_write(&ls->ls_in_recovery);
  44                 clear_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
  45                 error = 0;
  46         }
  47         spin_unlock(&ls->ls_recover_lock);
  48
  49         up_write(&ls->ls_recv_active);
  50         return error;
  51 }
  52
  53 static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
  54 {
  55         unsigned long start;
  56         int error, neg = 0;
  57
  58         log_rinfo(ls, "dlm_recover %llu", (unsigned long long)rv->seq);
  59
  60         mutex_lock(&ls->ls_recoverd_active);
  61
  62         dlm_callback_suspend(ls);
  63
  64         dlm_clear_toss(ls);
  65
  66         /*
  67          * This list of root rsb's will be the basis of most of the recovery
  68          * routines.
  69          */
  70
  71         dlm_create_root_list(ls);
  72
  73         /*
  74          * Add or remove nodes from the lockspace's ls_nodes list.
  75          */
  76
  77         error = dlm_recover_members(ls, rv, &neg);
  78         if (error) {
  79                 log_rinfo(ls, "dlm_recover_members error %d", error);
  80                 goto fail;
  81         }
  82
  83         dlm_recover_dir_nodeid(ls);
  84
  85         ls->ls_recover_dir_sent_res = 0;
  86         ls->ls_recover_dir_sent_msg = 0;
  87         ls->ls_recover_locks_in = 0;
  88
  89         dlm_set_recover_status(ls, DLM_RS_NODES);
  90
  91         error = dlm_recover_members_wait(ls);
  92         if (error) {
  93                 log_rinfo(ls, "dlm_recover_members_wait error %d", error);
  94                 goto fail;
  95         }
  96
  97         start = jiffies;
  98
  99         /*
 100          * Rebuild our own share of the directory by collecting from all other
 101          * nodes their master rsb names that hash to us.
 102          */
 103
 104         error = dlm_recover_directory(ls);
 105         if (error) {
 106                 log_rinfo(ls, "dlm_recover_directory error %d", error);
 107                 goto fail;
 108         }
 109
 110         dlm_set_recover_status(ls, DLM_RS_DIR);
 111
 112         error = dlm_recover_directory_wait(ls);
 113         if (error) {
 114                 log_rinfo(ls, "dlm_recover_directory_wait error %d", error);
 115                 goto fail;
 116         }
 117
 118         log_rinfo(ls, "dlm_recover_directory %u out %u messages",
 119                   ls->ls_recover_dir_sent_res, ls->ls_recover_dir_sent_msg);
 120
 121         /*
 122          * We may have outstanding operations that are waiting for a reply from
 123          * a failed node.  Mark these to be resent after recovery.  Unlock and
 124          * cancel ops can just be completed.
 125          */
 126
 127         dlm_recover_waiters_pre(ls);
 128
 129         error = dlm_recovery_stopped(ls);
 130         if (error)
 131                 goto fail;
 132
 133         if (neg || dlm_no_directory(ls)) {
 134                 /*
 135                  * Clear lkb's for departed nodes.
 136                  */
 137
 138                 dlm_recover_purge(ls);
 139
 140                 /*
 141                  * Get new master nodeid's for rsb's that were mastered on
 142                  * departed nodes.
 143                  */
 144
 145                 error = dlm_recover_masters(ls);
 146                 if (error) {
 147                         log_rinfo(ls, "dlm_recover_masters error %d", error);
 148                         goto fail;
 149                 }
 150
 151                 /*
 152                  * Send our locks on remastered rsb's to the new masters.
 153                  */
 154
 155                 error = dlm_recover_locks(ls);
 156                 if (error) {
 157                         log_rinfo(ls, "dlm_recover_locks error %d", error);
 158                         goto fail;
 159                 }
 160
 161                 dlm_set_recover_status(ls, DLM_RS_LOCKS);
 162
 163                 error = dlm_recover_locks_wait(ls);
 164                 if (error) {
 165                         log_rinfo(ls, "dlm_recover_locks_wait error %d", error);
 166                         goto fail;
 167                 }
 168
 169                 log_rinfo(ls, "dlm_recover_locks %u in",
 170                           ls->ls_recover_locks_in);
 171
 172                 /*
 173                  * Finalize state in master rsb's now that all locks can be
 174                  * checked.  This includes conversion resolution and lvb
 175                  * settings.
 176                  */
 177
 178                 dlm_recover_rsbs(ls);
 179         } else {
 180                 /*
 181                  * Other lockspace members may be going through the "neg" steps
 182                  * while also adding us to the lockspace, in which case they'll
 183                  * be doing the recover_locks (RS_LOCKS) barrier.
 184                  */
 185                 dlm_set_recover_status(ls, DLM_RS_LOCKS);
 186
 187                 error = dlm_recover_locks_wait(ls);
 188                 if (error) {
 189                         log_rinfo(ls, "dlm_recover_locks_wait error %d", error);
 190                         goto fail;
 191                 }
 192         }
 193
 194         dlm_release_root_list(ls);
 195
 196         /*
 197          * Purge directory-related requests that are saved in requestqueue.
 198          * All dir requests from before recovery are invalid now due to the dir
 199          * rebuild and will be resent by the requesting nodes.
 200          */
 201
 202         dlm_purge_requestqueue(ls);
 203
 204         dlm_set_recover_status(ls, DLM_RS_DONE);
 205
 206         error = dlm_recover_done_wait(ls);
 207         if (error) {
 208                 log_rinfo(ls, "dlm_recover_done_wait error %d", error);
 209                 goto fail;
 210         }
 211
 212         dlm_clear_members_gone(ls);
 213
 214         dlm_adjust_timeouts(ls);
 215
 216         dlm_callback_resume(ls);
 217
 218         error = enable_locking(ls, rv->seq);
 219         if (error) {
 220                 log_rinfo(ls, "enable_locking error %d", error);
 221                 goto fail;
 222         }
 223
 224         error = dlm_process_requestqueue(ls);
 225         if (error) {
 226                 log_rinfo(ls, "dlm_process_requestqueue error %d", error);
 227                 goto fail;
 228         }
 229
 230         error = dlm_recover_waiters_post(ls);
 231         if (error) {
 232                 log_rinfo(ls, "dlm_recover_waiters_post error %d", error);
 233                 goto fail;
 234         }
 235
 236         dlm_recover_grant(ls);
 237
 238         log_rinfo(ls, "dlm_recover %llu generation %u done: %u ms",
 239                   (unsigned long long)rv->seq, ls->ls_generation,
 240                   jiffies_to_msecs(jiffies - start));
 241         mutex_unlock(&ls->ls_recoverd_active);
 242
 243         dlm_lsop_recover_done(ls);
 244         return 0;
 245
 246  fail:
 247         dlm_release_root_list(ls);
 248         log_rinfo(ls, "dlm_recover %llu error %d",
 249                   (unsigned long long)rv->seq, error);
 250         mutex_unlock(&ls->ls_recoverd_active);
 251         return error;
 252 }
 253
 254 /* The dlm_ls_start() that created the rv we take here may already have been
 255    stopped via dlm_ls_stop(); in that case we need to leave the RECOVERY_STOP
 256    flag set. */
 257
 258 static void do_ls_recovery(struct dlm_ls *ls)
 259 {
 260         struct dlm_recover *rv = NULL;
 261
 262         spin_lock(&ls->ls_recover_lock);
 263         rv = ls->ls_recover_args;
 264         ls->ls_recover_args = NULL;
 265         if (rv && ls->ls_recover_seq == rv->seq)
 266                 clear_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
 267         spin_unlock(&ls->ls_recover_lock);
 268
 269         if (rv) {
 270                 ls_recover(ls, rv);
 271                 kfree(rv->nodes);
 272                 kfree(rv);
 273         }
 274 }
 275
 276 static int dlm_recoverd(void *arg)
 277 {
 278         struct dlm_ls *ls;
 279
 280         ls = dlm_find_lockspace_local(arg);
 281         if (!ls) {
 282                 log_print("dlm_recoverd: no lockspace %p", arg);
 283                 return -1;
 284         }
 285
 286         down_write(&ls->ls_in_recovery);
 287         set_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
 288         wake_up(&ls->ls_recover_lock_wait);
 289
 290         while (1) {
 291                 /*
 292                  * We call kthread_should_stop() after set_current_state().
 293                  * This is because it works correctly if kthread_stop() is
 294                  * called just before set_current_state().
 295                  */
 296                 set_current_state(TASK_INTERRUPTIBLE);
 297                 if (kthread_should_stop()) {
 298                         set_current_state(TASK_RUNNING);
 299                         break;
 300                 }
 301                 if (!test_bit(LSFL_RECOVER_WORK, &ls->ls_flags) &&
 302                     !test_bit(LSFL_RECOVER_DOWN, &ls->ls_flags)) {
 303                         if (kthread_should_stop())
 304                                 break;
 305                         schedule();
 306                 }
 307                 set_current_state(TASK_RUNNING);
 308
 309                 if (test_and_clear_bit(LSFL_RECOVER_DOWN, &ls->ls_flags)) {
 310                         down_write(&ls->ls_in_recovery);
 311                         set_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
 312                         wake_up(&ls->ls_recover_lock_wait);
 313                 }
 314
 315                 if (test_and_clear_bit(LSFL_RECOVER_WORK, &ls->ls_flags))
 316                         do_ls_recovery(ls);
 317         }
 318
 319         if (test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags))
 320                 up_write(&ls->ls_in_recovery);
 321
 322         dlm_put_lockspace(ls);
 323         return 0;
 324 }
 325
 326 int dlm_recoverd_start(struct dlm_ls *ls)
 327 {
 328         struct task_struct *p;
 329         int error = 0;
 330
 331         p = kthread_run(dlm_recoverd, ls, "dlm_recoverd");
 332         if (IS_ERR(p))
 333                 error = PTR_ERR(p);
 334         else
 335                 ls->ls_recoverd_task = p;
 336         return error;
 337 }
 338
 339 void dlm_recoverd_stop(struct dlm_ls *ls)
 340 {
 341         kthread_stop(ls->ls_recoverd_task);
 342 }
 343
 344 void dlm_recoverd_suspend(struct dlm_ls *ls)
 345 {
 346         wake_up(&ls->ls_wait_general);
 347         mutex_lock(&ls->ls_recoverd_active);
 348 }
 349
 350 void dlm_recoverd_resume(struct dlm_ls *ls)
 351 {
 352         mutex_unlock(&ls->ls_recoverd_active);
 353 }
 354