fs/dlm/recover.c

   1 /******************************************************************************
   2 *******************************************************************************
   3 **
   4 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   5 **  Copyright (C) 2004-2005 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 "dir.h"
  17 #include "config.h"
  18 #include "ast.h"
  19 #include "memory.h"
  20 #include "rcom.h"
  21 #include "lock.h"
  22 #include "lowcomms.h"
  23 #include "member.h"
  24 #include "recover.h"
  25
  26
  27 /*
  28  * Recovery waiting routines: these functions wait for a particular reply from
  29  * a remote node, or for the remote node to report a certain status.  They need
  30  * to abort if the lockspace is stopped indicating a node has failed (perhaps
  31  * the one being waited for).
  32  */
  33
  34 /*
  35  * Wait until given function returns non-zero or lockspace is stopped
  36  * (LS_RECOVERY_STOP set due to failure of a node in ls_nodes).  When another
  37  * function thinks it could have completed the waited-on task, they should wake
  38  * up ls_wait_general to get an immediate response rather than waiting for the
  39  * timer to detect the result.  A timer wakes us up periodically while waiting
  40  * to see if we should abort due to a node failure.  This should only be called
  41  * by the dlm_recoverd thread.
  42  */
  43
  44 static void dlm_wait_timer_fn(unsigned long data)
  45 {
  46         struct dlm_ls *ls = (struct dlm_ls *) data;
  47         mod_timer(&ls->ls_timer, jiffies + (dlm_config.recover_timer * HZ));
  48         wake_up(&ls->ls_wait_general);
  49 }
  50
  51 int dlm_wait_function(struct dlm_ls *ls, int (*testfn) (struct dlm_ls *ls))
  52 {
  53         int error = 0;
  54
  55         init_timer(&ls->ls_timer);
  56         ls->ls_timer.function = dlm_wait_timer_fn;
  57         ls->ls_timer.data = (long) ls;
  58         ls->ls_timer.expires = jiffies + (dlm_config.recover_timer * HZ);
  59         add_timer(&ls->ls_timer);
  60
  61         wait_event(ls->ls_wait_general, testfn(ls) || dlm_recovery_stopped(ls));
  62         del_timer_sync(&ls->ls_timer);
  63
  64         if (dlm_recovery_stopped(ls)) {
  65                 log_debug(ls, "dlm_wait_function aborted");
  66                 error = -EINTR;
  67         }
  68         return error;
  69 }
  70
  71 /*
  72  * An efficient way for all nodes to wait for all others to have a certain
  73  * status.  The node with the lowest nodeid polls all the others for their
  74  * status (wait_status_all) and all the others poll the node with the low id
  75  * for its accumulated result (wait_status_low).  When all nodes have set
  76  * status flag X, then status flag X_ALL will be set on the low nodeid.
  77  */
  78
  79 uint32_t dlm_recover_status(struct dlm_ls *ls)
  80 {
  81         uint32_t status;
  82         spin_lock(&ls->ls_recover_lock);
  83         status = ls->ls_recover_status;
  84         spin_unlock(&ls->ls_recover_lock);
  85         return status;
  86 }
  87
  88 void dlm_set_recover_status(struct dlm_ls *ls, uint32_t status)
  89 {
  90         spin_lock(&ls->ls_recover_lock);
  91         ls->ls_recover_status |= status;
  92         spin_unlock(&ls->ls_recover_lock);
  93 }
  94
  95 static int wait_status_all(struct dlm_ls *ls, uint32_t wait_status)
  96 {
  97         struct dlm_rcom *rc = (struct dlm_rcom *) ls->ls_recover_buf;
  98         struct dlm_member *memb;
  99         int error = 0, delay;
 100
 101         list_for_each_entry(memb, &ls->ls_nodes, list) {
 102                 delay = 0;
 103                 for (;;) {
 104                         if (dlm_recovery_stopped(ls)) {
 105                                 error = -EINTR;
 106                                 goto out;
 107                         }
 108
 109                         error = dlm_rcom_status(ls, memb->nodeid);
 110                         if (error)
 111                                 goto out;
 112
 113                         if (rc->rc_result & wait_status)
 114                                 break;
 115                         if (delay < 1000)
 116                                 delay += 20;
 117                         msleep(delay);
 118                 }
 119         }
 120  out:
 121         return error;
 122 }
 123
 124 static int wait_status_low(struct dlm_ls *ls, uint32_t wait_status)
 125 {
 126         struct dlm_rcom *rc = (struct dlm_rcom *) ls->ls_recover_buf;
 127         int error = 0, delay = 0, nodeid = ls->ls_low_nodeid;
 128
 129         for (;;) {
 130                 if (dlm_recovery_stopped(ls)) {
 131                         error = -EINTR;
 132                         goto out;
 133                 }
 134
 135                 error = dlm_rcom_status(ls, nodeid);
 136                 if (error)
 137                         break;
 138
 139                 if (rc->rc_result & wait_status)
 140                         break;
 141                 if (delay < 1000)
 142                         delay += 20;
 143                 msleep(delay);
 144         }
 145  out:
 146         return error;
 147 }
 148
 149 static int wait_status(struct dlm_ls *ls, uint32_t status)
 150 {
 151         uint32_t status_all = status << 1;
 152         int error;
 153
 154         if (ls->ls_low_nodeid == dlm_our_nodeid()) {
 155                 error = wait_status_all(ls, status);
 156                 if (!error)
 157                         dlm_set_recover_status(ls, status_all);
 158         } else
 159                 error = wait_status_low(ls, status_all);
 160
 161         return error;
 162 }
 163
 164 int dlm_recover_members_wait(struct dlm_ls *ls)
 165 {
 166         return wait_status(ls, DLM_RS_NODES);
 167 }
 168
 169 int dlm_recover_directory_wait(struct dlm_ls *ls)
 170 {
 171         return wait_status(ls, DLM_RS_DIR);
 172 }
 173
 174 int dlm_recover_locks_wait(struct dlm_ls *ls)
 175 {
 176         return wait_status(ls, DLM_RS_LOCKS);
 177 }
 178
 179 int dlm_recover_done_wait(struct dlm_ls *ls)
 180 {
 181         return wait_status(ls, DLM_RS_DONE);
 182 }
 183
 184 /*
 185  * The recover_list contains all the rsb's for which we've requested the new
 186  * master nodeid.  As replies are returned from the resource directories the
 187  * rsb's are removed from the list.  When the list is empty we're done.
 188  *
 189  * The recover_list is later similarly used for all rsb's for which we've sent
 190  * new lkb's and need to receive new corresponding lkid's.
 191  *
 192  * We use the address of the rsb struct as a simple local identifier for the
 193  * rsb so we can match an rcom reply with the rsb it was sent for.
 194  */
 195
 196 static int recover_list_empty(struct dlm_ls *ls)
 197 {
 198         int empty;
 199
 200         spin_lock(&ls->ls_recover_list_lock);
 201         empty = list_empty(&ls->ls_recover_list);
 202         spin_unlock(&ls->ls_recover_list_lock);
 203
 204         return empty;
 205 }
 206
 207 static void recover_list_add(struct dlm_rsb *r)
 208 {
 209         struct dlm_ls *ls = r->res_ls;
 210
 211         spin_lock(&ls->ls_recover_list_lock);
 212         if (list_empty(&r->res_recover_list)) {
 213                 list_add_tail(&r->res_recover_list, &ls->ls_recover_list);
 214                 ls->ls_recover_list_count++;
 215                 dlm_hold_rsb(r);
 216         }
 217         spin_unlock(&ls->ls_recover_list_lock);
 218 }
 219
 220 static void recover_list_del(struct dlm_rsb *r)
 221 {
 222         struct dlm_ls *ls = r->res_ls;
 223
 224         spin_lock(&ls->ls_recover_list_lock);
 225         list_del_init(&r->res_recover_list);
 226         ls->ls_recover_list_count--;
 227         spin_unlock(&ls->ls_recover_list_lock);
 228
 229         dlm_put_rsb(r);
 230 }
 231
 232 static struct dlm_rsb *recover_list_find(struct dlm_ls *ls, uint64_t id)
 233 {
 234         struct dlm_rsb *r = NULL;
 235
 236         spin_lock(&ls->ls_recover_list_lock);
 237
 238         list_for_each_entry(r, &ls->ls_recover_list, res_recover_list) {
 239                 if (id == (unsigned long) r)
 240                         goto out;
 241         }
 242         r = NULL;
 243  out:
 244         spin_unlock(&ls->ls_recover_list_lock);
 245         return r;
 246 }
 247
 248 static void recover_list_clear(struct dlm_ls *ls)
 249 {
 250         struct dlm_rsb *r, *s;
 251
 252         spin_lock(&ls->ls_recover_list_lock);
 253         list_for_each_entry_safe(r, s, &ls->ls_recover_list, res_recover_list) {
 254                 list_del_init(&r->res_recover_list);
 255                 r->res_recover_locks_count = 0;
 256                 dlm_put_rsb(r);
 257                 ls->ls_recover_list_count--;
 258         }
 259
 260         if (ls->ls_recover_list_count != 0) {
 261                 log_error(ls, "warning: recover_list_count %d",
 262                           ls->ls_recover_list_count);
 263                 ls->ls_recover_list_count = 0;
 264         }
 265         spin_unlock(&ls->ls_recover_list_lock);
 266 }
 267
 268
 269 /* Master recovery: find new master node for rsb's that were
 270    mastered on nodes that have been removed.
 271
 272    dlm_recover_masters
 273    recover_master
 274    dlm_send_rcom_lookup            ->  receive_rcom_lookup
 275                                        dlm_dir_lookup
 276    receive_rcom_lookup_reply       <-
 277    dlm_recover_master_reply
 278    set_new_master
 279    set_master_lkbs
 280    set_lock_master
 281 */
 282
 283 /*
 284  * Set the lock master for all LKBs in a lock queue
 285  * If we are the new master of the rsb, we may have received new
 286  * MSTCPY locks from other nodes already which we need to ignore
 287  * when setting the new nodeid.
 288  */
 289
 290 static void set_lock_master(struct list_head *queue, int nodeid)
 291 {
 292         struct dlm_lkb *lkb;
 293
 294         list_for_each_entry(lkb, queue, lkb_statequeue)
 295                 if (!(lkb->lkb_flags & DLM_IFL_MSTCPY))
 296                         lkb->lkb_nodeid = nodeid;
 297 }
 298
 299 static void set_master_lkbs(struct dlm_rsb *r)
 300 {
 301         set_lock_master(&r->res_grantqueue, r->res_nodeid);
 302         set_lock_master(&r->res_convertqueue, r->res_nodeid);
 303         set_lock_master(&r->res_waitqueue, r->res_nodeid);
 304 }
 305
 306 /*
 307  * Propogate the new master nodeid to locks
 308  * The NEW_MASTER flag tells dlm_recover_locks() which rsb's to consider.
 309  * The NEW_MASTER2 flag tells recover_lvb() and set_locks_purged() which
 310  * rsb's to consider.
 311  */
 312
 313 static void set_new_master(struct dlm_rsb *r, int nodeid)
 314 {
 315         lock_rsb(r);
 316         r->res_nodeid = nodeid;
 317         set_master_lkbs(r);
 318         rsb_set_flag(r, RSB_NEW_MASTER);
 319         rsb_set_flag(r, RSB_NEW_MASTER2);
 320         unlock_rsb(r);
 321 }
 322
 323 /*
 324  * We do async lookups on rsb's that need new masters.  The rsb's
 325  * waiting for a lookup reply are kept on the recover_list.
 326  */
 327
 328 static int recover_master(struct dlm_rsb *r)
 329 {
 330         struct dlm_ls *ls = r->res_ls;
 331         int error, dir_nodeid, ret_nodeid, our_nodeid = dlm_our_nodeid();
 332
 333         dir_nodeid = dlm_dir_nodeid(r);
 334
 335         if (dir_nodeid == our_nodeid) {
 336                 error = dlm_dir_lookup(ls, our_nodeid, r->res_name,
 337                                        r->res_length, &ret_nodeid);
 338                 if (error)
 339                         log_error(ls, "recover dir lookup error %d", error);
 340
 341                 if (ret_nodeid == our_nodeid)
 342                         ret_nodeid = 0;
 343                 set_new_master(r, ret_nodeid);
 344         } else {
 345                 recover_list_add(r);
 346                 error = dlm_send_rcom_lookup(r, dir_nodeid);
 347         }
 348
 349         return error;
 350 }
 351
 352 /*
 353  * When not using a directory, most resource names will hash to a new static
 354  * master nodeid and the resource will need to be remastered.
 355  */
 356
 357 static int recover_master_static(struct dlm_rsb *r)
 358 {
 359         int master = dlm_dir_nodeid(r);
 360
 361         if (master == dlm_our_nodeid())
 362                 master = 0;
 363
 364         if (r->res_nodeid != master) {
 365                 if (is_master(r))
 366                         dlm_purge_mstcpy_locks(r);
 367                 set_new_master(r, master);
 368                 return 1;
 369         }
 370         return 0;
 371 }
 372
 373 /*
 374  * Go through local root resources and for each rsb which has a master which
 375  * has departed, get the new master nodeid from the directory.  The dir will
 376  * assign mastery to the first node to look up the new master.  That means
 377  * we'll discover in this lookup if we're the new master of any rsb's.
 378  *
 379  * We fire off all the dir lookup requests individually and asynchronously to
 380  * the correct dir node.
 381  */
 382
 383 int dlm_recover_masters(struct dlm_ls *ls)
 384 {
 385         struct dlm_rsb *r;
 386         int error = 0, count = 0;
 387
 388         log_debug(ls, "dlm_recover_masters");
 389
 390         down_read(&ls->ls_root_sem);
 391         list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
 392                 if (dlm_recovery_stopped(ls)) {
 393                         up_read(&ls->ls_root_sem);
 394                         error = -EINTR;
 395                         goto out;
 396                 }
 397
 398                 if (dlm_no_directory(ls))
 399                         count += recover_master_static(r);
 400                 else if (!is_master(r) && dlm_is_removed(ls, r->res_nodeid)) {
 401                         recover_master(r);
 402                         count++;
 403                 }
 404
 405                 schedule();
 406         }
 407         up_read(&ls->ls_root_sem);
 408
 409         log_debug(ls, "dlm_recover_masters %d resources", count);
 410
 411         error = dlm_wait_function(ls, &recover_list_empty);
 412  out:
 413         if (error)
 414                 recover_list_clear(ls);
 415         return error;
 416 }
 417
 418 int dlm_recover_master_reply(struct dlm_ls *ls, struct dlm_rcom *rc)
 419 {
 420         struct dlm_rsb *r;
 421         int nodeid;
 422
 423         r = recover_list_find(ls, rc->rc_id);
 424         if (!r) {
 425                 log_error(ls, "dlm_recover_master_reply no id %llx",
 426                           (unsigned long long)rc->rc_id);
 427                 goto out;
 428         }
 429
 430         nodeid = rc->rc_result;
 431         if (nodeid == dlm_our_nodeid())
 432                 nodeid = 0;
 433
 434         set_new_master(r, nodeid);
 435         recover_list_del(r);
 436
 437         if (recover_list_empty(ls))
 438                 wake_up(&ls->ls_wait_general);
 439  out:
 440         return 0;
 441 }
 442
 443
 444 /* Lock recovery: rebuild the process-copy locks we hold on a
 445    remastered rsb on the new rsb master.
 446
 447    dlm_recover_locks
 448    recover_locks
 449    recover_locks_queue
 450    dlm_send_rcom_lock              ->  receive_rcom_lock
 451                                        dlm_recover_master_copy
 452    receive_rcom_lock_reply         <-
 453    dlm_recover_process_copy
 454 */
 455
 456
 457 /*
 458  * keep a count of the number of lkb's we send to the new master; when we get
 459  * an equal number of replies then recovery for the rsb is done
 460  */
 461
 462 static int recover_locks_queue(struct dlm_rsb *r, struct list_head *head)
 463 {
 464         struct dlm_lkb *lkb;
 465         int error = 0;
 466
 467         list_for_each_entry(lkb, head, lkb_statequeue) {
 468                 error = dlm_send_rcom_lock(r, lkb);
 469                 if (error)
 470                         break;
 471                 r->res_recover_locks_count++;
 472         }
 473
 474         return error;
 475 }
 476
 477 static int recover_locks(struct dlm_rsb *r)
 478 {
 479         int error = 0;
 480
 481         lock_rsb(r);
 482
 483         DLM_ASSERT(!r->res_recover_locks_count, dlm_dump_rsb(r););
 484
 485         error = recover_locks_queue(r, &r->res_grantqueue);
 486         if (error)
 487                 goto out;
 488         error = recover_locks_queue(r, &r->res_convertqueue);
 489         if (error)
 490                 goto out;
 491         error = recover_locks_queue(r, &r->res_waitqueue);
 492         if (error)
 493                 goto out;
 494
 495         if (r->res_recover_locks_count)
 496                 recover_list_add(r);
 497         else
 498                 rsb_clear_flag(r, RSB_NEW_MASTER);
 499  out:
 500         unlock_rsb(r);
 501         return error;
 502 }
 503
 504 int dlm_recover_locks(struct dlm_ls *ls)
 505 {
 506         struct dlm_rsb *r;
 507         int error, count = 0;
 508
 509         log_debug(ls, "dlm_recover_locks");
 510
 511         down_read(&ls->ls_root_sem);
 512         list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
 513                 if (is_master(r)) {
 514                         rsb_clear_flag(r, RSB_NEW_MASTER);
 515                         continue;
 516                 }
 517
 518                 if (!rsb_flag(r, RSB_NEW_MASTER))
 519                         continue;
 520
 521                 if (dlm_recovery_stopped(ls)) {
 522                         error = -EINTR;
 523                         up_read(&ls->ls_root_sem);
 524                         goto out;
 525                 }
 526
 527                 error = recover_locks(r);
 528                 if (error) {
 529                         up_read(&ls->ls_root_sem);
 530                         goto out;
 531                 }
 532
 533                 count += r->res_recover_locks_count;
 534         }
 535         up_read(&ls->ls_root_sem);
 536
 537         log_debug(ls, "dlm_recover_locks %d locks", count);
 538
 539         error = dlm_wait_function(ls, &recover_list_empty);
 540  out:
 541         if (error)
 542                 recover_list_clear(ls);
 543         else
 544                 dlm_set_recover_status(ls, DLM_RS_LOCKS);
 545         return error;
 546 }
 547
 548 void dlm_recovered_lock(struct dlm_rsb *r)
 549 {
 550         DLM_ASSERT(rsb_flag(r, RSB_NEW_MASTER), dlm_dump_rsb(r););
 551
 552         r->res_recover_locks_count--;
 553         if (!r->res_recover_locks_count) {
 554                 rsb_clear_flag(r, RSB_NEW_MASTER);
 555                 recover_list_del(r);
 556         }
 557
 558         if (recover_list_empty(r->res_ls))
 559                 wake_up(&r->res_ls->ls_wait_general);
 560 }
 561
 562 /*
 563  * The lvb needs to be recovered on all master rsb's.  This includes setting
 564  * the VALNOTVALID flag if necessary, and determining the correct lvb contents
 565  * based on the lvb's of the locks held on the rsb.
 566  *
 567  * RSB_VALNOTVALID is set if there are only NL/CR locks on the rsb.  If it
 568  * was already set prior to recovery, it's not cleared, regardless of locks.
 569  *
 570  * The LVB contents are only considered for changing when this is a new master
 571  * of the rsb (NEW_MASTER2).  Then, the rsb's lvb is taken from any lkb with
 572  * mode > CR.  If no lkb's exist with mode above CR, the lvb contents are taken
 573  * from the lkb with the largest lvb sequence number.
 574  */
 575
 576 static void recover_lvb(struct dlm_rsb *r)
 577 {
 578         struct dlm_lkb *lkb, *high_lkb = NULL;
 579         uint32_t high_seq = 0;
 580         int lock_lvb_exists = 0;
 581         int big_lock_exists = 0;
 582         int lvblen = r->res_ls->ls_lvblen;
 583
 584         list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
 585                 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
 586                         continue;
 587
 588                 lock_lvb_exists = 1;
 589
 590                 if (lkb->lkb_grmode > DLM_LOCK_CR) {
 591                         big_lock_exists = 1;
 592                         goto setflag;
 593                 }
 594
 595                 if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) {
 596                         high_lkb = lkb;
 597                         high_seq = lkb->lkb_lvbseq;
 598                 }
 599         }
 600
 601         list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
 602                 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
 603                         continue;
 604
 605                 lock_lvb_exists = 1;
 606
 607                 if (lkb->lkb_grmode > DLM_LOCK_CR) {
 608                         big_lock_exists = 1;
 609                         goto setflag;
 610                 }
 611
 612                 if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) {
 613                         high_lkb = lkb;
 614                         high_seq = lkb->lkb_lvbseq;
 615                 }
 616         }
 617
 618  setflag:
 619         if (!lock_lvb_exists)
 620                 goto out;
 621
 622         if (!big_lock_exists)
 623                 rsb_set_flag(r, RSB_VALNOTVALID);
 624
 625         /* don't mess with the lvb unless we're the new master */
 626         if (!rsb_flag(r, RSB_NEW_MASTER2))
 627                 goto out;
 628
 629         if (!r->res_lvbptr) {
 630                 r->res_lvbptr = allocate_lvb(r->res_ls);
 631                 if (!r->res_lvbptr)
 632                         goto out;
 633         }
 634
 635         if (big_lock_exists) {
 636                 r->res_lvbseq = lkb->lkb_lvbseq;
 637                 memcpy(r->res_lvbptr, lkb->lkb_lvbptr, lvblen);
 638         } else if (high_lkb) {
 639                 r->res_lvbseq = high_lkb->lkb_lvbseq;
 640                 memcpy(r->res_lvbptr, high_lkb->lkb_lvbptr, lvblen);
 641         } else {
 642                 r->res_lvbseq = 0;
 643                 memset(r->res_lvbptr, 0, lvblen);
 644         }
 645  out:
 646         return;
 647 }
 648
 649 /* All master rsb's flagged RECOVER_CONVERT need to be looked at.  The locks
 650    converting PR->CW or CW->PR need to have their lkb_grmode set. */
 651
 652 static void recover_conversion(struct dlm_rsb *r)
 653 {
 654         struct dlm_lkb *lkb;
 655         int grmode = -1;
 656
 657         list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
 658                 if (lkb->lkb_grmode == DLM_LOCK_PR ||
 659                     lkb->lkb_grmode == DLM_LOCK_CW) {
 660                         grmode = lkb->lkb_grmode;
 661                         break;
 662                 }
 663         }
 664
 665         list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
 666                 if (lkb->lkb_grmode != DLM_LOCK_IV)
 667                         continue;
 668                 if (grmode == -1)
 669                         lkb->lkb_grmode = lkb->lkb_rqmode;
 670                 else
 671                         lkb->lkb_grmode = grmode;
 672         }
 673 }
 674
 675 /* We've become the new master for this rsb and waiting/converting locks may
 676    need to be granted in dlm_grant_after_purge() due to locks that may have
 677    existed from a removed node. */
 678
 679 static void set_locks_purged(struct dlm_rsb *r)
 680 {
 681         if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue))
 682                 rsb_set_flag(r, RSB_LOCKS_PURGED);
 683 }
 684
 685 void dlm_recover_rsbs(struct dlm_ls *ls)
 686 {
 687         struct dlm_rsb *r;
 688         int count = 0;
 689
 690         log_debug(ls, "dlm_recover_rsbs");
 691
 692         down_read(&ls->ls_root_sem);
 693         list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
 694                 lock_rsb(r);
 695                 if (is_master(r)) {
 696                         if (rsb_flag(r, RSB_RECOVER_CONVERT))
 697                                 recover_conversion(r);
 698                         if (rsb_flag(r, RSB_NEW_MASTER2))
 699                                 set_locks_purged(r);
 700                         recover_lvb(r);
 701                         count++;
 702                 }
 703                 rsb_clear_flag(r, RSB_RECOVER_CONVERT);
 704                 rsb_clear_flag(r, RSB_NEW_MASTER2);
 705                 unlock_rsb(r);
 706         }
 707         up_read(&ls->ls_root_sem);
 708
 709         log_debug(ls, "dlm_recover_rsbs %d rsbs", count);
 710 }
 711
 712 /* Create a single list of all root rsb's to be used during recovery */
 713
 714 int dlm_create_root_list(struct dlm_ls *ls)
 715 {
 716         struct dlm_rsb *r;
 717         int i, error = 0;
 718
 719         down_write(&ls->ls_root_sem);
 720         if (!list_empty(&ls->ls_root_list)) {
 721                 log_error(ls, "root list not empty");
 722                 error = -EINVAL;
 723                 goto out;
 724         }
 725
 726         for (i = 0; i < ls->ls_rsbtbl_size; i++) {
 727                 read_lock(&ls->ls_rsbtbl[i].lock);
 728                 list_for_each_entry(r, &ls->ls_rsbtbl[i].list, res_hashchain) {
 729                         list_add(&r->res_root_list, &ls->ls_root_list);
 730                         dlm_hold_rsb(r);
 731                 }
 732                 read_unlock(&ls->ls_rsbtbl[i].lock);
 733         }
 734  out:
 735         up_write(&ls->ls_root_sem);
 736         return error;
 737 }
 738
 739 void dlm_release_root_list(struct dlm_ls *ls)
 740 {
 741         struct dlm_rsb *r, *safe;
 742
 743         down_write(&ls->ls_root_sem);
 744         list_for_each_entry_safe(r, safe, &ls->ls_root_list, res_root_list) {
 745                 list_del_init(&r->res_root_list);
 746                 dlm_put_rsb(r);
 747         }
 748         up_write(&ls->ls_root_sem);
 749 }
 750
 751 void dlm_clear_toss_list(struct dlm_ls *ls)
 752 {
 753         struct dlm_rsb *r, *safe;
 754         int i;
 755
 756         for (i = 0; i < ls->ls_rsbtbl_size; i++) {
 757                 write_lock(&ls->ls_rsbtbl[i].lock);
 758                 list_for_each_entry_safe(r, safe, &ls->ls_rsbtbl[i].toss,
 759                                          res_hashchain) {
 760                         list_del(&r->res_hashchain);
 761                         free_rsb(r);
 762                 }
 763                 write_unlock(&ls->ls_rsbtbl[i].lock);
 764         }
 765 }
 766