fs/ocfs2/dlm/dlmrecovery.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
dlm_get_next_mig_cookie
dlm_set_reco_dead_node
dlm_set_reco_master
__dlm_reset_recovery
dlm_reset_recovery
dlm_dispatch_work
dlm_kick_recovery_thread
dlm_launch_recovery_thread
dlm_complete_recovery_thread
dlm_print_reco_node_status
dlm_recovery_thread
dlm_reco_master_ready
dlm_is_node_dead
dlm_is_node_recovered
dlm_wait_for_node_death
dlm_wait_for_node_recovery
dlm_in_recovery
dlm_wait_for_recovery
dlm_begin_recovery
dlm_end_recovery
dlm_print_recovery_master
dlm_do_recovery
dlm_remaster_locks
dlm_init_recovery_area
dlm_destroy_recovery_area
dlm_request_all_locks
dlm_request_all_locks_handler
dlm_request_all_locks_worker
dlm_send_all_done_msg
dlm_reco_data_done_handler
dlm_move_reco_locks_to_list
dlm_num_locks_in_lockres
dlm_send_mig_lockres_msg
dlm_init_migratable_lockres
dlm_prepare_lvb_for_migration
dlm_add_lock_to_array
dlm_add_dummy_lock
dlm_is_dummy_lock
dlm_send_one_lockres
dlm_mig_lockres_handler
dlm_mig_lockres_worker
dlm_lockres_master_requery
dlm_do_master_requery
dlm_master_requery_handler
dlm_list_num_to_pointer
dlm_process_recovery_data
dlm_move_lockres_to_recovery_list
dlm_finish_local_lockres_recovery
dlm_lvb_needs_invalidation
dlm_revalidate_lvb
dlm_free_dead_locks
dlm_do_local_recovery_cleanup
__dlm_hb_node_down
dlm_hb_node_down_cb
dlm_hb_node_up_cb
dlm_reco_ast
dlm_reco_bast
dlm_reco_unlock_ast
dlm_pick_recovery_master
dlm_send_begin_reco_message
dlm_begin_reco_handler
dlm_send_finalize_reco_message
dlm_finalize_reco_handler
   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* -*- mode: c; c-basic-offset: 8; -*-
   3  * vim: noexpandtab sw=8 ts=8 sts=0:
   4  *
   5  * dlmrecovery.c
   6  *
   7  * recovery stuff
   8  *
   9  * Copyright (C) 2004 Oracle.  All rights reserved.
  10  */
  11 
  12 
  13 #include <linux/module.h>
  14 #include <linux/fs.h>
  15 #include <linux/types.h>
  16 #include <linux/slab.h>
  17 #include <linux/highmem.h>
  18 #include <linux/init.h>
  19 #include <linux/sysctl.h>
  20 #include <linux/random.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/socket.h>
  23 #include <linux/inet.h>
  24 #include <linux/timer.h>
  25 #include <linux/kthread.h>
  26 #include <linux/delay.h>
  27 
  28 
  29 #include "../cluster/heartbeat.h"
  30 #include "../cluster/nodemanager.h"
  31 #include "../cluster/tcp.h"
  32 
  33 #include "dlmapi.h"
  34 #include "dlmcommon.h"
  35 #include "dlmdomain.h"
  36 
  37 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_RECOVERY)
  38 #include "../cluster/masklog.h"
  39 
  40 static void dlm_do_local_recovery_cleanup(struct dlm_ctxt *dlm, u8 dead_node);
  41 
  42 static int dlm_recovery_thread(void *data);
  43 static int dlm_do_recovery(struct dlm_ctxt *dlm);
  44 
  45 static int dlm_pick_recovery_master(struct dlm_ctxt *dlm);
  46 static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node);
  47 static int dlm_init_recovery_area(struct dlm_ctxt *dlm, u8 dead_node);
  48 static int dlm_request_all_locks(struct dlm_ctxt *dlm,
  49                                  u8 request_from, u8 dead_node);
  50 static void dlm_destroy_recovery_area(struct dlm_ctxt *dlm);
  51 
  52 static inline int dlm_num_locks_in_lockres(struct dlm_lock_resource *res);
  53 static void dlm_init_migratable_lockres(struct dlm_migratable_lockres *mres,
  54                                         const char *lockname, int namelen,
  55                                         int total_locks, u64 cookie,
  56                                         u8 flags, u8 master);
  57 static int dlm_send_mig_lockres_msg(struct dlm_ctxt *dlm,
  58                                     struct dlm_migratable_lockres *mres,
  59                                     u8 send_to,
  60                                     struct dlm_lock_resource *res,
  61                                     int total_locks);
  62 static int dlm_process_recovery_data(struct dlm_ctxt *dlm,
  63                                      struct dlm_lock_resource *res,
  64                                      struct dlm_migratable_lockres *mres);
  65 static int dlm_send_finalize_reco_message(struct dlm_ctxt *dlm);
  66 static int dlm_send_all_done_msg(struct dlm_ctxt *dlm,
  67                                  u8 dead_node, u8 send_to);
  68 static int dlm_send_begin_reco_message(struct dlm_ctxt *dlm, u8 dead_node);
  69 static void dlm_move_reco_locks_to_list(struct dlm_ctxt *dlm,
  70                                         struct list_head *list, u8 dead_node);
  71 static void dlm_finish_local_lockres_recovery(struct dlm_ctxt *dlm,
  72                                               u8 dead_node, u8 new_master);
  73 static void dlm_reco_ast(void *astdata);
  74 static void dlm_reco_bast(void *astdata, int blocked_type);
  75 static void dlm_reco_unlock_ast(void *astdata, enum dlm_status st);
  76 static void dlm_request_all_locks_worker(struct dlm_work_item *item,
  77                                          void *data);
  78 static void dlm_mig_lockres_worker(struct dlm_work_item *item, void *data);
  79 static int dlm_lockres_master_requery(struct dlm_ctxt *dlm,
  80                                       struct dlm_lock_resource *res,
  81                                       u8 *real_master);
  82 
  83 static u64 dlm_get_next_mig_cookie(void);
  84 
  85 static DEFINE_SPINLOCK(dlm_reco_state_lock);
  86 static DEFINE_SPINLOCK(dlm_mig_cookie_lock);
  87 static u64 dlm_mig_cookie = 1;
  88 
  89 static u64 dlm_get_next_mig_cookie(void)
  90 {
  91         u64 c;
  92         spin_lock(&dlm_mig_cookie_lock);
  93         c = dlm_mig_cookie;
  94         if (dlm_mig_cookie == (~0ULL))
  95                 dlm_mig_cookie = 1;
  96         else
  97                 dlm_mig_cookie++;
  98         spin_unlock(&dlm_mig_cookie_lock);
  99         return c;
 100 }
 101 
 102 static inline void dlm_set_reco_dead_node(struct dlm_ctxt *dlm,
 103                                           u8 dead_node)
 104 {
 105         assert_spin_locked(&dlm->spinlock);
 106         if (dlm->reco.dead_node != dead_node)
 107                 mlog(0, "%s: changing dead_node from %u to %u\n",
 108                      dlm->name, dlm->reco.dead_node, dead_node);
 109         dlm->reco.dead_node = dead_node;
 110 }
 111 
 112 static inline void dlm_set_reco_master(struct dlm_ctxt *dlm,
 113                                        u8 master)
 114 {
 115         assert_spin_locked(&dlm->spinlock);
 116         mlog(0, "%s: changing new_master from %u to %u\n",
 117              dlm->name, dlm->reco.new_master, master);
 118         dlm->reco.new_master = master;
 119 }
 120 
 121 static inline void __dlm_reset_recovery(struct dlm_ctxt *dlm)
 122 {
 123         assert_spin_locked(&dlm->spinlock);
 124         clear_bit(dlm->reco.dead_node, dlm->recovery_map);
 125         dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
 126         dlm_set_reco_master(dlm, O2NM_INVALID_NODE_NUM);
 127 }
 128 
 129 static inline void dlm_reset_recovery(struct dlm_ctxt *dlm)
 130 {
 131         spin_lock(&dlm->spinlock);
 132         __dlm_reset_recovery(dlm);
 133         spin_unlock(&dlm->spinlock);
 134 }
 135 
 136 /* Worker function used during recovery. */
 137 void dlm_dispatch_work(struct work_struct *work)
 138 {
 139         struct dlm_ctxt *dlm =
 140                 container_of(work, struct dlm_ctxt, dispatched_work);
 141         LIST_HEAD(tmp_list);
 142         struct dlm_work_item *item, *next;
 143         dlm_workfunc_t *workfunc;
 144         int tot=0;
 145 
 146         spin_lock(&dlm->work_lock);
 147         list_splice_init(&dlm->work_list, &tmp_list);
 148         spin_unlock(&dlm->work_lock);
 149 
 150         list_for_each_entry(item, &tmp_list, list) {
 151                 tot++;
 152         }
 153         mlog(0, "%s: work thread has %d work items\n", dlm->name, tot);
 154 
 155         list_for_each_entry_safe(item, next, &tmp_list, list) {
 156                 workfunc = item->func;
 157                 list_del_init(&item->list);
 158 
 159                 /* already have ref on dlm to avoid having
 160                  * it disappear.  just double-check. */
 161                 BUG_ON(item->dlm != dlm);
 162 
 163                 /* this is allowed to sleep and
 164                  * call network stuff */
 165                 workfunc(item, item->data);
 166 
 167                 dlm_put(dlm);
 168                 kfree(item);
 169         }
 170 }
 171 
 172 /*
 173  * RECOVERY THREAD
 174  */
 175 
 176 void dlm_kick_recovery_thread(struct dlm_ctxt *dlm)
 177 {
 178         /* wake the recovery thread
 179          * this will wake the reco thread in one of three places
 180          * 1) sleeping with no recovery happening
 181          * 2) sleeping with recovery mastered elsewhere
 182          * 3) recovery mastered here, waiting on reco data */
 183 
 184         wake_up(&dlm->dlm_reco_thread_wq);
 185 }
 186 
 187 /* Launch the recovery thread */
 188 int dlm_launch_recovery_thread(struct dlm_ctxt *dlm)
 189 {
 190         mlog(0, "starting dlm recovery thread...\n");
 191 
 192         dlm->dlm_reco_thread_task = kthread_run(dlm_recovery_thread, dlm,
 193                         "dlm_reco-%s", dlm->name);
 194         if (IS_ERR(dlm->dlm_reco_thread_task)) {
 195                 mlog_errno(PTR_ERR(dlm->dlm_reco_thread_task));
 196                 dlm->dlm_reco_thread_task = NULL;
 197                 return -EINVAL;
 198         }
 199 
 200         return 0;
 201 }
 202 
 203 void dlm_complete_recovery_thread(struct dlm_ctxt *dlm)
 204 {
 205         if (dlm->dlm_reco_thread_task) {
 206                 mlog(0, "waiting for dlm recovery thread to exit\n");
 207                 kthread_stop(dlm->dlm_reco_thread_task);
 208                 dlm->dlm_reco_thread_task = NULL;
 209         }
 210 }
 211 
 212 
 213 
 214 /*
 215  * this is lame, but here's how recovery works...
 216  * 1) all recovery threads cluster wide will work on recovering
 217  *    ONE node at a time
 218  * 2) negotiate who will take over all the locks for the dead node.
 219  *    thats right... ALL the locks.
 220  * 3) once a new master is chosen, everyone scans all locks
 221  *    and moves aside those mastered by the dead guy
 222  * 4) each of these locks should be locked until recovery is done
 223  * 5) the new master collects up all of secondary lock queue info
 224  *    one lock at a time, forcing each node to communicate back
 225  *    before continuing
 226  * 6) each secondary lock queue responds with the full known lock info
 227  * 7) once the new master has run all its locks, it sends a ALLDONE!
 228  *    message to everyone
 229  * 8) upon receiving this message, the secondary queue node unlocks
 230  *    and responds to the ALLDONE
 231  * 9) once the new master gets responses from everyone, he unlocks
 232  *    everything and recovery for this dead node is done
 233  *10) go back to 2) while there are still dead nodes
 234  *
 235  */
 236 
 237 static void dlm_print_reco_node_status(struct dlm_ctxt *dlm)
 238 {
 239         struct dlm_reco_node_data *ndata;
 240         struct dlm_lock_resource *res;
 241 
 242         mlog(ML_NOTICE, "%s(%d): recovery info, state=%s, dead=%u, master=%u\n",
 243              dlm->name, task_pid_nr(dlm->dlm_reco_thread_task),
 244              dlm->reco.state & DLM_RECO_STATE_ACTIVE ? "ACTIVE" : "inactive",
 245              dlm->reco.dead_node, dlm->reco.new_master);
 246 
 247         list_for_each_entry(ndata, &dlm->reco.node_data, list) {
 248                 char *st = "unknown";
 249                 switch (ndata->state) {
 250                         case DLM_RECO_NODE_DATA_INIT:
 251                                 st = "init";
 252                                 break;
 253                         case DLM_RECO_NODE_DATA_REQUESTING:
 254                                 st = "requesting";
 255                                 break;
 256                         case DLM_RECO_NODE_DATA_DEAD:
 257                                 st = "dead";
 258                                 break;
 259                         case DLM_RECO_NODE_DATA_RECEIVING:
 260                                 st = "receiving";
 261                                 break;
 262                         case DLM_RECO_NODE_DATA_REQUESTED:
 263                                 st = "requested";
 264                                 break;
 265                         case DLM_RECO_NODE_DATA_DONE:
 266                                 st = "done";
 267                                 break;
 268                         case DLM_RECO_NODE_DATA_FINALIZE_SENT:
 269                                 st = "finalize-sent";
 270                                 break;
 271                         default:
 272                                 st = "bad";
 273                                 break;
 274                 }
 275                 mlog(ML_NOTICE, "%s: reco state, node %u, state=%s\n",
 276                      dlm->name, ndata->node_num, st);
 277         }
 278         list_for_each_entry(res, &dlm->reco.resources, recovering) {
 279                 mlog(ML_NOTICE, "%s: lockres %.*s on recovering list\n",
 280                      dlm->name, res->lockname.len, res->lockname.name);
 281         }
 282 }
 283 
 284 #define DLM_RECO_THREAD_TIMEOUT_MS (5 * 1000)
 285 
 286 static int dlm_recovery_thread(void *data)
 287 {
 288         int status;
 289         struct dlm_ctxt *dlm = data;
 290         unsigned long timeout = msecs_to_jiffies(DLM_RECO_THREAD_TIMEOUT_MS);
 291 
 292         mlog(0, "dlm thread running for %s...\n", dlm->name);
 293 
 294         while (!kthread_should_stop()) {
 295                 if (dlm_domain_fully_joined(dlm)) {
 296                         status = dlm_do_recovery(dlm);
 297                         if (status == -EAGAIN) {
 298                                 /* do not sleep, recheck immediately. */
 299                                 continue;
 300                         }
 301                         if (status < 0)
 302                                 mlog_errno(status);
 303                 }
 304 
 305                 wait_event_interruptible_timeout(dlm->dlm_reco_thread_wq,
 306                                                  kthread_should_stop(),
 307                                                  timeout);
 308         }
 309 
 310         mlog(0, "quitting DLM recovery thread\n");
 311         return 0;
 312 }
 313 
 314 /* returns true when the recovery master has contacted us */
 315 static int dlm_reco_master_ready(struct dlm_ctxt *dlm)
 316 {
 317         int ready;
 318         spin_lock(&dlm->spinlock);
 319         ready = (dlm->reco.new_master != O2NM_INVALID_NODE_NUM);
 320         spin_unlock(&dlm->spinlock);
 321         return ready;
 322 }
 323 
 324 /* returns true if node is no longer in the domain
 325  * could be dead or just not joined */
 326 int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node)
 327 {
 328         int dead;
 329         spin_lock(&dlm->spinlock);
 330         dead = !test_bit(node, dlm->domain_map);
 331         spin_unlock(&dlm->spinlock);
 332         return dead;
 333 }
 334 
 335 /* returns true if node is no longer in the domain
 336  * could be dead or just not joined */
 337 static int dlm_is_node_recovered(struct dlm_ctxt *dlm, u8 node)
 338 {
 339         int recovered;
 340         spin_lock(&dlm->spinlock);
 341         recovered = !test_bit(node, dlm->recovery_map);
 342         spin_unlock(&dlm->spinlock);
 343         return recovered;
 344 }
 345 
 346 
 347 void dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout)
 348 {
 349         if (dlm_is_node_dead(dlm, node))
 350                 return;
 351 
 352         printk(KERN_NOTICE "o2dlm: Waiting on the death of node %u in "
 353                "domain %s\n", node, dlm->name);
 354 
 355         if (timeout)
 356                 wait_event_timeout(dlm->dlm_reco_thread_wq,
 357                                    dlm_is_node_dead(dlm, node),
 358                                    msecs_to_jiffies(timeout));
 359         else
 360                 wait_event(dlm->dlm_reco_thread_wq,
 361                            dlm_is_node_dead(dlm, node));
 362 }
 363 
 364 void dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout)
 365 {
 366         if (dlm_is_node_recovered(dlm, node))
 367                 return;
 368 
 369         printk(KERN_NOTICE "o2dlm: Waiting on the recovery of node %u in "
 370                "domain %s\n", node, dlm->name);
 371 
 372         if (timeout)
 373                 wait_event_timeout(dlm->dlm_reco_thread_wq,
 374                                    dlm_is_node_recovered(dlm, node),
 375                                    msecs_to_jiffies(timeout));
 376         else
 377                 wait_event(dlm->dlm_reco_thread_wq,
 378                            dlm_is_node_recovered(dlm, node));
 379 }
 380 
 381 /* callers of the top-level api calls (dlmlock/dlmunlock) should
 382  * block on the dlm->reco.event when recovery is in progress.
 383  * the dlm recovery thread will set this state when it begins
 384  * recovering a dead node (as the new master or not) and clear
 385  * the state and wake as soon as all affected lock resources have
 386  * been marked with the RECOVERY flag */
 387 static int dlm_in_recovery(struct dlm_ctxt *dlm)
 388 {
 389         int in_recovery;
 390         spin_lock(&dlm->spinlock);
 391         in_recovery = !!(dlm->reco.state & DLM_RECO_STATE_ACTIVE);
 392         spin_unlock(&dlm->spinlock);
 393         return in_recovery;
 394 }
 395 
 396 
 397 void dlm_wait_for_recovery(struct dlm_ctxt *dlm)
 398 {
 399         if (dlm_in_recovery(dlm)) {
 400                 mlog(0, "%s: reco thread %d in recovery: "
 401                      "state=%d, master=%u, dead=%u\n",
 402                      dlm->name, task_pid_nr(dlm->dlm_reco_thread_task),
 403                      dlm->reco.state, dlm->reco.new_master,
 404                      dlm->reco.dead_node);
 405         }
 406         wait_event(dlm->reco.event, !dlm_in_recovery(dlm));
 407 }
 408 
 409 static void dlm_begin_recovery(struct dlm_ctxt *dlm)
 410 {
 411         assert_spin_locked(&dlm->spinlock);
 412         BUG_ON(dlm->reco.state & DLM_RECO_STATE_ACTIVE);
 413         printk(KERN_NOTICE "o2dlm: Begin recovery on domain %s for node %u\n",
 414                dlm->name, dlm->reco.dead_node);
 415         dlm->reco.state |= DLM_RECO_STATE_ACTIVE;
 416 }
 417 
 418 static void dlm_end_recovery(struct dlm_ctxt *dlm)
 419 {
 420         spin_lock(&dlm->spinlock);
 421         BUG_ON(!(dlm->reco.state & DLM_RECO_STATE_ACTIVE));
 422         dlm->reco.state &= ~DLM_RECO_STATE_ACTIVE;
 423         spin_unlock(&dlm->spinlock);
 424         printk(KERN_NOTICE "o2dlm: End recovery on domain %s\n", dlm->name);
 425         wake_up(&dlm->reco.event);
 426 }
 427 
 428 static void dlm_print_recovery_master(struct dlm_ctxt *dlm)
 429 {
 430         printk(KERN_NOTICE "o2dlm: Node %u (%s) is the Recovery Master for the "
 431                "dead node %u in domain %s\n", dlm->reco.new_master,
 432                (dlm->node_num == dlm->reco.new_master ? "me" : "he"),
 433                dlm->reco.dead_node, dlm->name);
 434 }
 435 
 436 static int dlm_do_recovery(struct dlm_ctxt *dlm)
 437 {
 438         int status = 0;
 439         int ret;
 440 
 441         spin_lock(&dlm->spinlock);
 442 
 443         if (dlm->migrate_done) {
 444                 mlog(0, "%s: no need do recovery after migrating all "
 445                      "lock resources\n", dlm->name);
 446                 spin_unlock(&dlm->spinlock);
 447                 return 0;
 448         }
 449 
 450         /* check to see if the new master has died */
 451         if (dlm->reco.new_master != O2NM_INVALID_NODE_NUM &&
 452             test_bit(dlm->reco.new_master, dlm->recovery_map)) {
 453                 mlog(0, "new master %u died while recovering %u!\n",
 454                      dlm->reco.new_master, dlm->reco.dead_node);
 455                 /* unset the new_master, leave dead_node */
 456                 dlm_set_reco_master(dlm, O2NM_INVALID_NODE_NUM);
 457         }
 458 
 459         /* select a target to recover */
 460         if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) {
 461                 int bit;
 462 
 463                 bit = find_next_bit (dlm->recovery_map, O2NM_MAX_NODES, 0);
 464                 if (bit >= O2NM_MAX_NODES || bit < 0)
 465                         dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
 466                 else
 467                         dlm_set_reco_dead_node(dlm, bit);
 468         } else if (!test_bit(dlm->reco.dead_node, dlm->recovery_map)) {
 469                 /* BUG? */
 470                 mlog(ML_ERROR, "dead_node %u no longer in recovery map!\n",
 471                      dlm->reco.dead_node);
 472                 dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
 473         }
 474 
 475         if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) {
 476                 // mlog(0, "nothing to recover!  sleeping now!\n");
 477                 spin_unlock(&dlm->spinlock);
 478                 /* return to main thread loop and sleep. */
 479                 return 0;
 480         }
 481         mlog(0, "%s(%d):recovery thread found node %u in the recovery map!\n",
 482              dlm->name, task_pid_nr(dlm->dlm_reco_thread_task),
 483              dlm->reco.dead_node);
 484 
 485         /* take write barrier */
 486         /* (stops the list reshuffling thread, proxy ast handling) */
 487         dlm_begin_recovery(dlm);
 488 
 489         spin_unlock(&dlm->spinlock);
 490 
 491         if (dlm->reco.new_master == dlm->node_num)
 492                 goto master_here;
 493 
 494         if (dlm->reco.new_master == O2NM_INVALID_NODE_NUM) {
 495                 /* choose a new master, returns 0 if this node
 496                  * is the master, -EEXIST if it's another node.
 497                  * this does not return until a new master is chosen
 498                  * or recovery completes entirely. */
 499                 ret = dlm_pick_recovery_master(dlm);
 500                 if (!ret) {
 501                         /* already notified everyone.  go. */
 502                         goto master_here;
 503                 }
 504                 mlog(0, "another node will master this recovery session.\n");
 505         }
 506 
 507         dlm_print_recovery_master(dlm);
 508 
 509         /* it is safe to start everything back up here
 510          * because all of the dead node's lock resources
 511          * have been marked as in-recovery */
 512         dlm_end_recovery(dlm);
 513 
 514         /* sleep out in main dlm_recovery_thread loop. */
 515         return 0;
 516 
 517 master_here:
 518         dlm_print_recovery_master(dlm);
 519 
 520         status = dlm_remaster_locks(dlm, dlm->reco.dead_node);
 521         if (status < 0) {
 522                 /* we should never hit this anymore */
 523                 mlog(ML_ERROR, "%s: Error %d remastering locks for node %u, "
 524                      "retrying.\n", dlm->name, status, dlm->reco.dead_node);
 525                 /* yield a bit to allow any final network messages
 526                  * to get handled on remaining nodes */
 527                 msleep(100);
 528         } else {
 529                 /* success!  see if any other nodes need recovery */
 530                 mlog(0, "DONE mastering recovery of %s:%u here(this=%u)!\n",
 531                      dlm->name, dlm->reco.dead_node, dlm->node_num);
 532                 spin_lock(&dlm->spinlock);
 533                 __dlm_reset_recovery(dlm);
 534                 dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE;
 535                 spin_unlock(&dlm->spinlock);
 536         }
 537         dlm_end_recovery(dlm);
 538 
 539         /* continue and look for another dead node */
 540         return -EAGAIN;
 541 }
 542 
 543 static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node)
 544 {
 545         int status = 0;
 546         struct dlm_reco_node_data *ndata;
 547         int all_nodes_done;
 548         int destroy = 0;
 549         int pass = 0;
 550 
 551         do {
 552                 /* we have become recovery master.  there is no escaping
 553                  * this, so just keep trying until we get it. */
 554                 status = dlm_init_recovery_area(dlm, dead_node);
 555                 if (status < 0) {
 556                         mlog(ML_ERROR, "%s: failed to alloc recovery area, "
 557                              "retrying\n", dlm->name);
 558                         msleep(1000);
 559                 }
 560         } while (status != 0);
 561 
 562         /* safe to access the node data list without a lock, since this
 563          * process is the only one to change the list */
 564         list_for_each_entry(ndata, &dlm->reco.node_data, list) {
 565                 BUG_ON(ndata->state != DLM_RECO_NODE_DATA_INIT);
 566                 ndata->state = DLM_RECO_NODE_DATA_REQUESTING;
 567 
 568                 mlog(0, "%s: Requesting lock info from node %u\n", dlm->name,
 569                      ndata->node_num);
 570 
 571                 if (ndata->node_num == dlm->node_num) {
 572                         ndata->state = DLM_RECO_NODE_DATA_DONE;
 573                         continue;
 574                 }
 575 
 576                 do {
 577                         status = dlm_request_all_locks(dlm, ndata->node_num,
 578                                                        dead_node);
 579                         if (status < 0) {
 580                                 mlog_errno(status);
 581                                 if (dlm_is_host_down(status)) {
 582                                         /* node died, ignore it for recovery */
 583                                         status = 0;
 584                                         ndata->state = DLM_RECO_NODE_DATA_DEAD;
 585                                         /* wait for the domain map to catch up
 586                                          * with the network state. */
 587                                         wait_event_timeout(dlm->dlm_reco_thread_wq,
 588                                                            dlm_is_node_dead(dlm,
 589                                                                 ndata->node_num),
 590                                                            msecs_to_jiffies(1000));
 591                                         mlog(0, "waited 1 sec for %u, "
 592                                              "dead? %s\n", ndata->node_num,
 593                                              dlm_is_node_dead(dlm, ndata->node_num) ?
 594                                              "yes" : "no");
 595                                 } else {
 596                                         /* -ENOMEM on the other node */
 597                                         mlog(0, "%s: node %u returned "
 598                                              "%d during recovery, retrying "
 599                                              "after a short wait\n",
 600                                              dlm->name, ndata->node_num,
 601                                              status);
 602                                         msleep(100);
 603                                 }
 604                         }
 605                 } while (status != 0);
 606 
 607                 spin_lock(&dlm_reco_state_lock);
 608                 switch (ndata->state) {
 609                         case DLM_RECO_NODE_DATA_INIT:
 610                         case DLM_RECO_NODE_DATA_FINALIZE_SENT:
 611                         case DLM_RECO_NODE_DATA_REQUESTED:
 612                                 BUG();
 613                                 break;
 614                         case DLM_RECO_NODE_DATA_DEAD:
 615                                 mlog(0, "node %u died after requesting "
 616                                      "recovery info for node %u\n",
 617                                      ndata->node_num, dead_node);
 618                                 /* fine.  don't need this node's info.
 619                                  * continue without it. */
 620                                 break;
 621                         case DLM_RECO_NODE_DATA_REQUESTING:
 622                                 ndata->state = DLM_RECO_NODE_DATA_REQUESTED;
 623                                 mlog(0, "now receiving recovery data from "
 624                                      "node %u for dead node %u\n",
 625                                      ndata->node_num, dead_node);
 626                                 break;
 627                         case DLM_RECO_NODE_DATA_RECEIVING:
 628                                 mlog(0, "already receiving recovery data from "
 629                                      "node %u for dead node %u\n",
 630                                      ndata->node_num, dead_node);
 631                                 break;
 632                         case DLM_RECO_NODE_DATA_DONE:
 633                                 mlog(0, "already DONE receiving recovery data "
 634                                      "from node %u for dead node %u\n",
 635                                      ndata->node_num, dead_node);
 636                                 break;
 637                 }
 638                 spin_unlock(&dlm_reco_state_lock);
 639         }
 640 
 641         mlog(0, "%s: Done requesting all lock info\n", dlm->name);
 642 
 643         /* nodes should be sending reco data now
 644          * just need to wait */
 645 
 646         while (1) {
 647                 /* check all the nodes now to see if we are
 648                  * done, or if anyone died */
 649                 all_nodes_done = 1;
 650                 spin_lock(&dlm_reco_state_lock);
 651                 list_for_each_entry(ndata, &dlm->reco.node_data, list) {
 652                         mlog(0, "checking recovery state of node %u\n",
 653                              ndata->node_num);
 654                         switch (ndata->state) {
 655                                 case DLM_RECO_NODE_DATA_INIT:
 656                                 case DLM_RECO_NODE_DATA_REQUESTING:
 657                                         mlog(ML_ERROR, "bad ndata state for "
 658                                              "node %u: state=%d\n",
 659                                              ndata->node_num, ndata->state);
 660                                         BUG();
 661                                         break;
 662                                 case DLM_RECO_NODE_DATA_DEAD:
 663                                         mlog(0, "node %u died after "
 664                                              "requesting recovery info for "
 665                                              "node %u\n", ndata->node_num,
 666                                              dead_node);
 667                                         break;
 668                                 case DLM_RECO_NODE_DATA_RECEIVING:
 669                                 case DLM_RECO_NODE_DATA_REQUESTED:
 670                                         mlog(0, "%s: node %u still in state %s\n",
 671                                              dlm->name, ndata->node_num,
 672                                              ndata->state==DLM_RECO_NODE_DATA_RECEIVING ?
 673                                              "receiving" : "requested");
 674                                         all_nodes_done = 0;
 675                                         break;
 676                                 case DLM_RECO_NODE_DATA_DONE:
 677                                         mlog(0, "%s: node %u state is done\n",
 678                                              dlm->name, ndata->node_num);
 679                                         break;
 680                                 case DLM_RECO_NODE_DATA_FINALIZE_SENT:
 681                                         mlog(0, "%s: node %u state is finalize\n",
 682                                              dlm->name, ndata->node_num);
 683                                         break;
 684                         }
 685                 }
 686                 spin_unlock(&dlm_reco_state_lock);
 687 
 688                 mlog(0, "pass #%d, all_nodes_done?: %s\n", ++pass,
 689                      all_nodes_done?"yes":"no");
 690                 if (all_nodes_done) {
 691                         int ret;
 692 
 693                         /* Set this flag on recovery master to avoid
 694                          * a new recovery for another dead node start
 695                          * before the recovery is not done. That may
 696                          * cause recovery hung.*/
 697                         spin_lock(&dlm->spinlock);
 698                         dlm->reco.state |= DLM_RECO_STATE_FINALIZE;
 699                         spin_unlock(&dlm->spinlock);
 700 
 701                         /* all nodes are now in DLM_RECO_NODE_DATA_DONE state
 702                          * just send a finalize message to everyone and
 703                          * clean up */
 704                         mlog(0, "all nodes are done! send finalize\n");
 705                         ret = dlm_send_finalize_reco_message(dlm);
 706                         if (ret < 0)
 707                                 mlog_errno(ret);
 708 
 709                         spin_lock(&dlm->spinlock);
 710                         dlm_finish_local_lockres_recovery(dlm, dead_node,
 711                                                           dlm->node_num);
 712                         spin_unlock(&dlm->spinlock);
 713                         mlog(0, "should be done with recovery!\n");
 714 
 715                         mlog(0, "finishing recovery of %s at %lu, "
 716                              "dead=%u, this=%u, new=%u\n", dlm->name,
 717                              jiffies, dlm->reco.dead_node,
 718                              dlm->node_num, dlm->reco.new_master);
 719                         destroy = 1;
 720                         status = 0;
 721                         /* rescan everything marked dirty along the way */
 722                         dlm_kick_thread(dlm, NULL);
 723                         break;
 724                 }
 725                 /* wait to be signalled, with periodic timeout
 726                  * to check for node death */
 727                 wait_event_interruptible_timeout(dlm->dlm_reco_thread_wq,
 728                                          kthread_should_stop(),
 729                                          msecs_to_jiffies(DLM_RECO_THREAD_TIMEOUT_MS));
 730 
 731         }
 732 
 733         if (destroy)
 734                 dlm_destroy_recovery_area(dlm);
 735 
 736         return status;
 737 }
 738 
 739 static int dlm_init_recovery_area(struct dlm_ctxt *dlm, u8 dead_node)
 740 {
 741         int num=0;
 742         struct dlm_reco_node_data *ndata;
 743 
 744         spin_lock(&dlm->spinlock);
 745         memcpy(dlm->reco.node_map, dlm->domain_map, sizeof(dlm->domain_map));
 746         /* nodes can only be removed (by dying) after dropping
 747          * this lock, and death will be trapped later, so this should do */
 748         spin_unlock(&dlm->spinlock);
 749 
 750         while (1) {
 751                 num = find_next_bit (dlm->reco.node_map, O2NM_MAX_NODES, num);
 752                 if (num >= O2NM_MAX_NODES) {
 753                         break;
 754                 }
 755                 BUG_ON(num == dead_node);
 756 
 757                 ndata = kzalloc(sizeof(*ndata), GFP_NOFS);
 758                 if (!ndata) {
 759                         dlm_destroy_recovery_area(dlm);
 760                         return -ENOMEM;
 761                 }
 762                 ndata->node_num = num;
 763                 ndata->state = DLM_RECO_NODE_DATA_INIT;
 764                 spin_lock(&dlm_reco_state_lock);
 765                 list_add_tail(&ndata->list, &dlm->reco.node_data);
 766                 spin_unlock(&dlm_reco_state_lock);
 767                 num++;
 768         }
 769 
 770         return 0;
 771 }
 772 
 773 static void dlm_destroy_recovery_area(struct dlm_ctxt *dlm)
 774 {
 775         struct dlm_reco_node_data *ndata, *next;
 776         LIST_HEAD(tmplist);
 777 
 778         spin_lock(&dlm_reco_state_lock);
 779         list_splice_init(&dlm->reco.node_data, &tmplist);
 780         spin_unlock(&dlm_reco_state_lock);
 781 
 782         list_for_each_entry_safe(ndata, next, &tmplist, list) {
 783                 list_del_init(&ndata->list);
 784                 kfree(ndata);
 785         }
 786 }
 787 
 788 static int dlm_request_all_locks(struct dlm_ctxt *dlm, u8 request_from,
 789                                  u8 dead_node)
 790 {
 791         struct dlm_lock_request lr;
 792         int ret;
 793         int status;
 794 
 795         mlog(0, "\n");
 796 
 797 
 798         mlog(0, "dlm_request_all_locks: dead node is %u, sending request "
 799                   "to %u\n", dead_node, request_from);
 800 
 801         memset(&lr, 0, sizeof(lr));
 802         lr.node_idx = dlm->node_num;
 803         lr.dead_node = dead_node;
 804 
 805         // send message
 806         ret = o2net_send_message(DLM_LOCK_REQUEST_MSG, dlm->key,
 807                                  &lr, sizeof(lr), request_from, &status);
 808 
 809         /* negative status is handled by caller */
 810         if (ret < 0)
 811                 mlog(ML_ERROR, "%s: Error %d send LOCK_REQUEST to node %u "
 812                      "to recover dead node %u\n", dlm->name, ret,
 813                      request_from, dead_node);
 814         else
 815                 ret = status;
 816         // return from here, then
 817         // sleep until all received or error
 818         return ret;
 819 
 820 }
 821 
 822 int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data,
 823                                   void **ret_data)
 824 {
 825         struct dlm_ctxt *dlm = data;
 826         struct dlm_lock_request *lr = (struct dlm_lock_request *)msg->buf;
 827         char *buf = NULL;
 828         struct dlm_work_item *item = NULL;
 829 
 830         if (!dlm_grab(dlm))
 831                 return -EINVAL;
 832 
 833         if (lr->dead_node != dlm->reco.dead_node) {
 834                 mlog(ML_ERROR, "%s: node %u sent dead_node=%u, but local "
 835                      "dead_node is %u\n", dlm->name, lr->node_idx,
 836                      lr->dead_node, dlm->reco.dead_node);
 837                 dlm_print_reco_node_status(dlm);
 838                 /* this is a hack */
 839                 dlm_put(dlm);
 840                 return -ENOMEM;
 841         }
 842         BUG_ON(lr->dead_node != dlm->reco.dead_node);
 843 
 844         item = kzalloc(sizeof(*item), GFP_NOFS);
 845         if (!item) {
 846                 dlm_put(dlm);
 847                 return -ENOMEM;
 848         }
 849 
 850         /* this will get freed by dlm_request_all_locks_worker */
 851         buf = (char *) __get_free_page(GFP_NOFS);
 852         if (!buf) {
 853                 kfree(item);
 854                 dlm_put(dlm);
 855                 return -ENOMEM;
 856         }
 857 
 858         /* queue up work for dlm_request_all_locks_worker */
 859         dlm_grab(dlm);  /* get an extra ref for the work item */
 860         dlm_init_work_item(dlm, item, dlm_request_all_locks_worker, buf);
 861         item->u.ral.reco_master = lr->node_idx;
 862         item->u.ral.dead_node = lr->dead_node;
 863         spin_lock(&dlm->work_lock);
 864         list_add_tail(&item->list, &dlm->work_list);
 865         spin_unlock(&dlm->work_lock);
 866         queue_work(dlm->dlm_worker, &dlm->dispatched_work);
 867 
 868         dlm_put(dlm);
 869         return 0;
 870 }
 871 
 872 static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data)
 873 {
 874         struct dlm_migratable_lockres *mres;
 875         struct dlm_lock_resource *res;
 876         struct dlm_ctxt *dlm;
 877         LIST_HEAD(resources);
 878         int ret;
 879         u8 dead_node, reco_master;
 880         int skip_all_done = 0;
 881 
 882         dlm = item->dlm;
 883         dead_node = item->u.ral.dead_node;
 884         reco_master = item->u.ral.reco_master;
 885         mres = (struct dlm_migratable_lockres *)data;
 886 
 887         mlog(0, "%s: recovery worker started, dead=%u, master=%u\n",
 888              dlm->name, dead_node, reco_master);
 889 
 890         if (dead_node != dlm->reco.dead_node ||
 891             reco_master != dlm->reco.new_master) {
 892                 /* worker could have been created before the recovery master
 893                  * died.  if so, do not continue, but do not error. */
 894                 if (dlm->reco.new_master == O2NM_INVALID_NODE_NUM) {
 895                         mlog(ML_NOTICE, "%s: will not send recovery state, "
 896                              "recovery master %u died, thread=(dead=%u,mas=%u)"
 897                              " current=(dead=%u,mas=%u)\n", dlm->name,
 898                              reco_master, dead_node, reco_master,
 899                              dlm->reco.dead_node, dlm->reco.new_master);
 900                 } else {
 901                         mlog(ML_NOTICE, "%s: reco state invalid: reco(dead=%u, "
 902                              "master=%u), request(dead=%u, master=%u)\n",
 903                              dlm->name, dlm->reco.dead_node,
 904                              dlm->reco.new_master, dead_node, reco_master);
 905                 }
 906                 goto leave;
 907         }
 908 
 909         /* lock resources should have already been moved to the
 910          * dlm->reco.resources list.  now move items from that list
 911          * to a temp list if the dead owner matches.  note that the
 912          * whole cluster recovers only one node at a time, so we
 913          * can safely move UNKNOWN lock resources for each recovery
 914          * session. */
 915         dlm_move_reco_locks_to_list(dlm, &resources, dead_node);
 916 
 917         /* now we can begin blasting lockreses without the dlm lock */
 918 
 919         /* any errors returned will be due to the new_master dying,
 920          * the dlm_reco_thread should detect this */
 921         list_for_each_entry(res, &resources, recovering) {
 922                 ret = dlm_send_one_lockres(dlm, res, mres, reco_master,
 923                                         DLM_MRES_RECOVERY);
 924                 if (ret < 0) {
 925                         mlog(ML_ERROR, "%s: node %u went down while sending "
 926                              "recovery state for dead node %u, ret=%d\n", dlm->name,
 927                              reco_master, dead_node, ret);
 928                         skip_all_done = 1;
 929                         break;
 930                 }
 931         }
 932 
 933         /* move the resources back to the list */
 934         spin_lock(&dlm->spinlock);
 935         list_splice_init(&resources, &dlm->reco.resources);
 936         spin_unlock(&dlm->spinlock);
 937 
 938         if (!skip_all_done) {
 939                 ret = dlm_send_all_done_msg(dlm, dead_node, reco_master);
 940                 if (ret < 0) {
 941                         mlog(ML_ERROR, "%s: node %u went down while sending "
 942                              "recovery all-done for dead node %u, ret=%d\n",
 943                              dlm->name, reco_master, dead_node, ret);
 944                 }
 945         }
 946 leave:
 947         free_page((unsigned long)data);
 948 }
 949 
 950 
 951 static int dlm_send_all_done_msg(struct dlm_ctxt *dlm, u8 dead_node, u8 send_to)
 952 {
 953         int ret, tmpret;
 954         struct dlm_reco_data_done done_msg;
 955 
 956         memset(&done_msg, 0, sizeof(done_msg));
 957         done_msg.node_idx = dlm->node_num;
 958         done_msg.dead_node = dead_node;
 959         mlog(0, "sending DATA DONE message to %u, "
 960              "my node=%u, dead node=%u\n", send_to, done_msg.node_idx,
 961              done_msg.dead_node);
 962 
 963         ret = o2net_send_message(DLM_RECO_DATA_DONE_MSG, dlm->key, &done_msg,
 964                                  sizeof(done_msg), send_to, &tmpret);
 965         if (ret < 0) {
 966                 mlog(ML_ERROR, "%s: Error %d send RECO_DATA_DONE to node %u "
 967                      "to recover dead node %u\n", dlm->name, ret, send_to,
 968                      dead_node);
 969                 if (!dlm_is_host_down(ret)) {
 970                         BUG();
 971                 }
 972         } else
 973                 ret = tmpret;
 974         return ret;
 975 }
 976 
 977 
 978 int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data,
 979                                void **ret_data)
 980 {
 981         struct dlm_ctxt *dlm = data;
 982         struct dlm_reco_data_done *done = (struct dlm_reco_data_done *)msg->buf;
 983         struct dlm_reco_node_data *ndata = NULL;
 984         int ret = -EINVAL;
 985 
 986         if (!dlm_grab(dlm))
 987                 return -EINVAL;
 988 
 989         mlog(0, "got DATA DONE: dead_node=%u, reco.dead_node=%u, "
 990              "node_idx=%u, this node=%u\n", done->dead_node,
 991              dlm->reco.dead_node, done->node_idx, dlm->node_num);
 992 
 993         mlog_bug_on_msg((done->dead_node != dlm->reco.dead_node),
 994                         "Got DATA DONE: dead_node=%u, reco.dead_node=%u, "
 995                         "node_idx=%u, this node=%u\n", done->dead_node,
 996                         dlm->reco.dead_node, done->node_idx, dlm->node_num);
 997 
 998         spin_lock(&dlm_reco_state_lock);
 999         list_for_each_entry(ndata, &dlm->reco.node_data, list) {
1000                 if (ndata->node_num != done->node_idx)
1001                         continue;
1002 
1003                 switch (ndata->state) {
1004                         /* should have moved beyond INIT but not to FINALIZE yet */
1005                         case DLM_RECO_NODE_DATA_INIT:
1006                         case DLM_RECO_NODE_DATA_DEAD:
1007                         case DLM_RECO_NODE_DATA_FINALIZE_SENT:
1008                                 mlog(ML_ERROR, "bad ndata state for node %u:"
1009                                      " state=%d\n", ndata->node_num,
1010                                      ndata->state);
1011                                 BUG();
1012                                 break;
1013                         /* these states are possible at this point, anywhere along
1014                          * the line of recovery */
1015                         case DLM_RECO_NODE_DATA_DONE:
1016                         case DLM_RECO_NODE_DATA_RECEIVING:
1017                         case DLM_RECO_NODE_DATA_REQUESTED:
1018                         case DLM_RECO_NODE_DATA_REQUESTING:
1019                                 mlog(0, "node %u is DONE sending "
1020                                           "recovery data!\n",
1021                                           ndata->node_num);
1022 
1023                                 ndata->state = DLM_RECO_NODE_DATA_DONE;
1024                                 ret = 0;
1025                                 break;
1026                 }
1027         }
1028         spin_unlock(&dlm_reco_state_lock);
1029 
1030         /* wake the recovery thread, some node is done */
1031         if (!ret)
1032                 dlm_kick_recovery_thread(dlm);
1033 
1034         if (ret < 0)
1035                 mlog(ML_ERROR, "failed to find recovery node data for node "
1036                      "%u\n", done->node_idx);
1037         dlm_put(dlm);
1038 
1039         mlog(0, "leaving reco data done handler, ret=%d\n", ret);
1040         return ret;
1041 }
1042 
1043 static void dlm_move_reco_locks_to_list(struct dlm_ctxt *dlm,
1044                                         struct list_head *list,
1045                                         u8 dead_node)
1046 {
1047         struct dlm_lock_resource *res, *next;
1048         struct dlm_lock *lock;
1049 
1050         spin_lock(&dlm->spinlock);
1051         list_for_each_entry_safe(res, next, &dlm->reco.resources, recovering) {
1052                 /* always prune any $RECOVERY entries for dead nodes,
1053                  * otherwise hangs can occur during later recovery */
1054                 if (dlm_is_recovery_lock(res->lockname.name,
1055                                          res->lockname.len)) {
1056                         spin_lock(&res->spinlock);
1057                         list_for_each_entry(lock, &res->granted, list) {
1058                                 if (lock->ml.node == dead_node) {
1059                                         mlog(0, "AHA! there was "
1060                                              "a $RECOVERY lock for dead "
1061                                              "node %u (%s)!\n",
1062                                              dead_node, dlm->name);
1063                                         list_del_init(&lock->list);
1064                                         dlm_lock_put(lock);
1065                                         /* Can't schedule DLM_UNLOCK_FREE_LOCK
1066                                          * - do manually */
1067                                         dlm_lock_put(lock);
1068                                         break;
1069                                 }
1070                         }
1071                         spin_unlock(&res->spinlock);
1072                         continue;
1073                 }
1074 
1075                 if (res->owner == dead_node) {
1076                         mlog(0, "found lockres owned by dead node while "
1077                                   "doing recovery for node %u. sending it.\n",
1078                                   dead_node);
1079                         list_move_tail(&res->recovering, list);
1080                 } else if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
1081                         mlog(0, "found UNKNOWN owner while doing recovery "
1082                                   "for node %u. sending it.\n", dead_node);
1083                         list_move_tail(&res->recovering, list);
1084                 }
1085         }
1086         spin_unlock(&dlm->spinlock);
1087 }
1088 
1089 static inline int dlm_num_locks_in_lockres(struct dlm_lock_resource *res)
1090 {
1091         int total_locks = 0;
1092         struct list_head *iter, *queue = &res->granted;
1093         int i;
1094 
1095         for (i=0; i<3; i++) {
1096                 list_for_each(iter, queue)
1097                         total_locks++;
1098                 queue++;
1099         }
1100         return total_locks;
1101 }
1102 
1103 
1104 static int dlm_send_mig_lockres_msg(struct dlm_ctxt *dlm,
1105                                       struct dlm_migratable_lockres *mres,
1106                                       u8 send_to,
1107                                       struct dlm_lock_resource *res,
1108                                       int total_locks)
1109 {
1110         u64 mig_cookie = be64_to_cpu(mres->mig_cookie);
1111         int mres_total_locks = be32_to_cpu(mres->total_locks);
1112         int ret = 0, status = 0;
1113         u8 orig_flags = mres->flags,
1114            orig_master = mres->master;
1115 
1116         BUG_ON(mres->num_locks > DLM_MAX_MIGRATABLE_LOCKS);
1117         if (!mres->num_locks)
1118                 return 0;
1119 
1120         /* add an all-done flag if we reached the last lock */
1121         orig_flags = mres->flags;
1122         BUG_ON(total_locks > mres_total_locks);
1123         if (total_locks == mres_total_locks)
1124                 mres->flags |= DLM_MRES_ALL_DONE;
1125 
1126         mlog(0, "%s:%.*s: sending mig lockres (%s) to %u\n",
1127              dlm->name, res->lockname.len, res->lockname.name,
1128              orig_flags & DLM_MRES_MIGRATION ? "migration" : "recovery",
1129              send_to);
1130 
1131         /* send it */
1132         ret = o2net_send_message(DLM_MIG_LOCKRES_MSG, dlm->key, mres,
1133                                  struct_size(mres, ml, mres->num_locks),
1134                                  send_to, &status);
1135         if (ret < 0) {
1136                 /* XXX: negative status is not handled.
1137                  * this will end up killing this node. */
1138                 mlog(ML_ERROR, "%s: res %.*s, Error %d send MIG_LOCKRES to "
1139                      "node %u (%s)\n", dlm->name, mres->lockname_len,
1140                      mres->lockname, ret, send_to,
1141                      (orig_flags & DLM_MRES_MIGRATION ?
1142                       "migration" : "recovery"));
1143         } else {
1144                 /* might get an -ENOMEM back here */
1145                 ret = status;
1146                 if (ret < 0) {
1147                         mlog_errno(ret);
1148 
1149                         if (ret == -EFAULT) {
1150                                 mlog(ML_ERROR, "node %u told me to kill "
1151                                      "myself!\n", send_to);
1152                                 BUG();
1153                         }
1154                 }
1155         }
1156 
1157         /* zero and reinit the message buffer */
1158         dlm_init_migratable_lockres(mres, res->lockname.name,
1159                                     res->lockname.len, mres_total_locks,
1160                                     mig_cookie, orig_flags, orig_master);
1161         return ret;
1162 }
1163 
1164 static void dlm_init_migratable_lockres(struct dlm_migratable_lockres *mres,
1165                                         const char *lockname, int namelen,
1166                                         int total_locks, u64 cookie,
1167                                         u8 flags, u8 master)
1168 {
1169         /* mres here is one full page */
1170         clear_page(mres);
1171         mres->lockname_len = namelen;
1172         memcpy(mres->lockname, lockname, namelen);
1173         mres->num_locks = 0;
1174         mres->total_locks = cpu_to_be32(total_locks);
1175         mres->mig_cookie = cpu_to_be64(cookie);
1176         mres->flags = flags;
1177         mres->master = master;
1178 }
1179 
1180 static void dlm_prepare_lvb_for_migration(struct dlm_lock *lock,
1181                                           struct dlm_migratable_lockres *mres,
1182                                           int queue)
1183 {
1184         if (!lock->lksb)
1185                return;
1186 
1187         /* Ignore lvb in all locks in the blocked list */
1188         if (queue == DLM_BLOCKED_LIST)
1189                 return;
1190 
1191         /* Only consider lvbs in locks with granted EX or PR lock levels */
1192         if (lock->ml.type != LKM_EXMODE && lock->ml.type != LKM_PRMODE)
1193                 return;
1194 
1195         if (dlm_lvb_is_empty(mres->lvb)) {
1196                 memcpy(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN);
1197                 return;
1198         }
1199 
1200         /* Ensure the lvb copied for migration matches in other valid locks */
1201         if (!memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))
1202                 return;
1203 
1204         mlog(ML_ERROR, "Mismatched lvb in lock cookie=%u:%llu, name=%.*s, "
1205              "node=%u\n",
1206              dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
1207              dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
1208              lock->lockres->lockname.len, lock->lockres->lockname.name,
1209              lock->ml.node);
1210         dlm_print_one_lock_resource(lock->lockres);
1211         BUG();
1212 }
1213 
1214 /* returns 1 if this lock fills the network structure,
1215  * 0 otherwise */
1216 static int dlm_add_lock_to_array(struct dlm_lock *lock,
1217                                  struct dlm_migratable_lockres *mres, int queue)
1218 {
1219         struct dlm_migratable_lock *ml;
1220         int lock_num = mres->num_locks;
1221 
1222         ml = &(mres->ml[lock_num]);
1223         ml->cookie = lock->ml.cookie;
1224         ml->type = lock->ml.type;
1225         ml->convert_type = lock->ml.convert_type;
1226         ml->highest_blocked = lock->ml.highest_blocked;
1227         ml->list = queue;
1228         if (lock->lksb) {
1229                 ml->flags = lock->lksb->flags;
1230                 dlm_prepare_lvb_for_migration(lock, mres, queue);
1231         }
1232         ml->node = lock->ml.node;
1233         mres->num_locks++;
1234         /* we reached the max, send this network message */
1235         if (mres->num_locks == DLM_MAX_MIGRATABLE_LOCKS)
1236                 return 1;
1237         return 0;
1238 }
1239 
1240 static void dlm_add_dummy_lock(struct dlm_ctxt *dlm,
1241                                struct dlm_migratable_lockres *mres)
1242 {
1243         struct dlm_lock dummy;
1244         memset(&dummy, 0, sizeof(dummy));
1245         dummy.ml.cookie = 0;
1246         dummy.ml.type = LKM_IVMODE;
1247         dummy.ml.convert_type = LKM_IVMODE;
1248         dummy.ml.highest_blocked = LKM_IVMODE;
1249         dummy.lksb = NULL;
1250         dummy.ml.node = dlm->node_num;
1251         dlm_add_lock_to_array(&dummy, mres, DLM_BLOCKED_LIST);
1252 }
1253 
1254 static inline int dlm_is_dummy_lock(struct dlm_ctxt *dlm,
1255                                     struct dlm_migratable_lock *ml,
1256                                     u8 *nodenum)
1257 {
1258         if (unlikely(ml->cookie == 0 &&
1259             ml->type == LKM_IVMODE &&
1260             ml->convert_type == LKM_IVMODE &&
1261             ml->highest_blocked == LKM_IVMODE &&
1262             ml->list == DLM_BLOCKED_LIST)) {
1263                 *nodenum = ml->node;
1264                 return 1;
1265         }
1266         return 0;
1267 }
1268 
1269 int dlm_send_one_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
1270                          struct dlm_migratable_lockres *mres,
1271                          u8 send_to, u8 flags)
1272 {
1273         struct list_head *queue;
1274         int total_locks, i;
1275         u64 mig_cookie = 0;
1276         struct dlm_lock *lock;
1277         int ret = 0;
1278 
1279         BUG_ON(!(flags & (DLM_MRES_RECOVERY|DLM_MRES_MIGRATION)));
1280 
1281         mlog(0, "sending to %u\n", send_to);
1282 
1283         total_locks = dlm_num_locks_in_lockres(res);
1284         if (total_locks > DLM_MAX_MIGRATABLE_LOCKS) {
1285                 /* rare, but possible */
1286                 mlog(0, "argh.  lockres has %d locks.  this will "
1287                           "require more than one network packet to "
1288                           "migrate\n", total_locks);
1289                 mig_cookie = dlm_get_next_mig_cookie();
1290         }
1291 
1292         dlm_init_migratable_lockres(mres, res->lockname.name,
1293                                     res->lockname.len, total_locks,
1294                                     mig_cookie, flags, res->owner);
1295 
1296         total_locks = 0;
1297         for (i=DLM_GRANTED_LIST; i<=DLM_BLOCKED_LIST; i++) {
1298                 queue = dlm_list_idx_to_ptr(res, i);
1299                 list_for_each_entry(lock, queue, list) {
1300                         /* add another lock. */
1301                         total_locks++;
1302                         if (!dlm_add_lock_to_array(lock, mres, i))
1303                                 continue;
1304 
1305                         /* this filled the lock message,
1306                          * we must send it immediately. */
1307                         ret = dlm_send_mig_lockres_msg(dlm, mres, send_to,
1308                                                        res, total_locks);
1309                         if (ret < 0)
1310                                 goto error;
1311                 }
1312         }
1313         if (total_locks == 0) {
1314                 /* send a dummy lock to indicate a mastery reference only */
1315                 mlog(0, "%s:%.*s: sending dummy lock to %u, %s\n",
1316                      dlm->name, res->lockname.len, res->lockname.name,
1317                      send_to, flags & DLM_MRES_RECOVERY ? "recovery" :
1318                      "migration");
1319                 dlm_add_dummy_lock(dlm, mres);
1320         }
1321         /* flush any remaining locks */
1322         ret = dlm_send_mig_lockres_msg(dlm, mres, send_to, res, total_locks);
1323         if (ret < 0)
1324                 goto error;
1325         return ret;
1326 
1327 error:
1328         mlog(ML_ERROR, "%s: dlm_send_mig_lockres_msg returned %d\n",
1329              dlm->name, ret);
1330         if (!dlm_is_host_down(ret))
1331                 BUG();
1332         mlog(0, "%s: node %u went down while sending %s "
1333              "lockres %.*s\n", dlm->name, send_to,
1334              flags & DLM_MRES_RECOVERY ?  "recovery" : "migration",
1335              res->lockname.len, res->lockname.name);
1336         return ret;
1337 }
1338 
1339 
1340 
1341 /*
1342  * this message will contain no more than one page worth of
1343  * recovery data, and it will work on only one lockres.
1344  * there may be many locks in this page, and we may need to wait
1345  * for additional packets to complete all the locks (rare, but
1346  * possible).
1347  */
1348 /*
1349  * NOTE: the allocation error cases here are scary
1350  * we really cannot afford to fail an alloc in recovery
1351  * do we spin?  returning an error only delays the problem really
1352  */
1353 
1354 int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
1355                             void **ret_data)
1356 {
1357         struct dlm_ctxt *dlm = data;
1358         struct dlm_migratable_lockres *mres =
1359                 (struct dlm_migratable_lockres *)msg->buf;
1360         int ret = 0;
1361         u8 real_master;
1362         u8 extra_refs = 0;
1363         char *buf = NULL;
1364         struct dlm_work_item *item = NULL;
1365         struct dlm_lock_resource *res = NULL;
1366         unsigned int hash;
1367 
1368         if (!dlm_grab(dlm))
1369                 return -EINVAL;
1370 
1371         if (!dlm_joined(dlm)) {
1372                 mlog(ML_ERROR, "Domain %s not joined! "
1373                           "lockres %.*s, master %u\n",
1374                           dlm->name, mres->lockname_len,
1375                           mres->lockname, mres->master);
1376                 dlm_put(dlm);
1377                 return -EINVAL;
1378         }
1379 
1380         BUG_ON(!(mres->flags & (DLM_MRES_RECOVERY|DLM_MRES_MIGRATION)));
1381 
1382         real_master = mres->master;
1383         if (real_master == DLM_LOCK_RES_OWNER_UNKNOWN) {
1384                 /* cannot migrate a lockres with no master */
1385                 BUG_ON(!(mres->flags & DLM_MRES_RECOVERY));
1386         }
1387 
1388         mlog(0, "%s message received from node %u\n",
1389                   (mres->flags & DLM_MRES_RECOVERY) ?
1390                   "recovery" : "migration", mres->master);
1391         if (mres->flags & DLM_MRES_ALL_DONE)
1392                 mlog(0, "all done flag.  all lockres data received!\n");
1393 
1394         ret = -ENOMEM;
1395         buf = kmalloc(be16_to_cpu(msg->data_len), GFP_NOFS);
1396         item = kzalloc(sizeof(*item), GFP_NOFS);
1397         if (!buf || !item)
1398                 goto leave;
1399 
1400         /* lookup the lock to see if we have a secondary queue for this
1401          * already...  just add the locks in and this will have its owner
1402          * and RECOVERY flag changed when it completes. */
1403         hash = dlm_lockid_hash(mres->lockname, mres->lockname_len);
1404         spin_lock(&dlm->spinlock);
1405         res = __dlm_lookup_lockres_full(dlm, mres->lockname, mres->lockname_len,
1406                         hash);
1407         if (res) {
1408                 /* this will get a ref on res */
1409                 /* mark it as recovering/migrating and hash it */
1410                 spin_lock(&res->spinlock);
1411                 if (res->state & DLM_LOCK_RES_DROPPING_REF) {
1412                         mlog(0, "%s: node is attempting to migrate "
1413                                 "lockres %.*s, but marked as dropping "
1414                                 " ref!\n", dlm->name,
1415                                 mres->lockname_len, mres->lockname);
1416                         ret = -EINVAL;
1417                         spin_unlock(&res->spinlock);
1418                         spin_unlock(&dlm->spinlock);
1419                         dlm_lockres_put(res);
1420                         goto leave;
1421                 }
1422 
1423                 if (mres->flags & DLM_MRES_RECOVERY) {
1424                         res->state |= DLM_LOCK_RES_RECOVERING;
1425                 } else {
1426                         if (res->state & DLM_LOCK_RES_MIGRATING) {
1427                                 /* this is at least the second
1428                                  * lockres message */
1429                                 mlog(0, "lock %.*s is already migrating\n",
1430                                           mres->lockname_len,
1431                                           mres->lockname);
1432                         } else if (res->state & DLM_LOCK_RES_RECOVERING) {
1433                                 /* caller should BUG */
1434                                 mlog(ML_ERROR, "node is attempting to migrate "
1435                                      "lock %.*s, but marked as recovering!\n",
1436                                      mres->lockname_len, mres->lockname);
1437                                 ret = -EFAULT;
1438                                 spin_unlock(&res->spinlock);
1439                                 spin_unlock(&dlm->spinlock);
1440                                 dlm_lockres_put(res);
1441                                 goto leave;
1442                         }
1443                         res->state |= DLM_LOCK_RES_MIGRATING;
1444                 }
1445                 spin_unlock(&res->spinlock);
1446                 spin_unlock(&dlm->spinlock);
1447         } else {
1448                 spin_unlock(&dlm->spinlock);
1449                 /* need to allocate, just like if it was
1450                  * mastered here normally  */
1451                 res = dlm_new_lockres(dlm, mres->lockname, mres->lockname_len);
1452                 if (!res)
1453                         goto leave;
1454 
1455                 /* to match the ref that we would have gotten if
1456                  * dlm_lookup_lockres had succeeded */
1457                 dlm_lockres_get(res);
1458 
1459                 /* mark it as recovering/migrating and hash it */
1460                 if (mres->flags & DLM_MRES_RECOVERY)
1461                         res->state |= DLM_LOCK_RES_RECOVERING;
1462                 else
1463                         res->state |= DLM_LOCK_RES_MIGRATING;
1464 
1465                 spin_lock(&dlm->spinlock);
1466                 __dlm_insert_lockres(dlm, res);
1467                 spin_unlock(&dlm->spinlock);
1468 
1469                 /* Add an extra ref for this lock-less lockres lest the
1470                  * dlm_thread purges it before we get the chance to add
1471                  * locks to it */
1472                 dlm_lockres_get(res);
1473 
1474                 /* There are three refs that need to be put.
1475                  * 1. Taken above.
1476                  * 2. kref_init in dlm_new_lockres()->dlm_init_lockres().
1477                  * 3. dlm_lookup_lockres()
1478                  * The first one is handled at the end of this function. The
1479                  * other two are handled in the worker thread after locks have
1480                  * been attached. Yes, we don't wait for purge time to match
1481                  * kref_init. The lockres will still have atleast one ref
1482                  * added because it is in the hash __dlm_insert_lockres() */
1483                 extra_refs++;
1484 
1485                 /* now that the new lockres is inserted,
1486                  * make it usable by other processes */
1487                 spin_lock(&res->spinlock);
1488                 res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
1489                 spin_unlock(&res->spinlock);
1490                 wake_up(&res->wq);
1491         }
1492 
1493         /* at this point we have allocated everything we need,
1494          * and we have a hashed lockres with an extra ref and
1495          * the proper res->state flags. */
1496         ret = 0;
1497         spin_lock(&res->spinlock);
1498         /* drop this either when master requery finds a different master
1499          * or when a lock is added by the recovery worker */
1500         dlm_lockres_grab_inflight_ref(dlm, res);
1501         if (mres->master == DLM_LOCK_RES_OWNER_UNKNOWN) {
1502                 /* migration cannot have an unknown master */
1503                 BUG_ON(!(mres->flags & DLM_MRES_RECOVERY));
1504                 mlog(0, "recovery has passed me a lockres with an "
1505                           "unknown owner.. will need to requery: "
1506                           "%.*s\n", mres->lockname_len, mres->lockname);
1507         } else {
1508                 /* take a reference now to pin the lockres, drop it
1509                  * when locks are added in the worker */
1510                 dlm_change_lockres_owner(dlm, res, dlm->node_num);
1511         }
1512         spin_unlock(&res->spinlock);
1513 
1514         /* queue up work for dlm_mig_lockres_worker */
1515         dlm_grab(dlm);  /* get an extra ref for the work item */
1516         memcpy(buf, msg->buf, be16_to_cpu(msg->data_len));  /* copy the whole message */
1517         dlm_init_work_item(dlm, item, dlm_mig_lockres_worker, buf);
1518         item->u.ml.lockres = res; /* already have a ref */
1519         item->u.ml.real_master = real_master;
1520         item->u.ml.extra_ref = extra_refs;
1521         spin_lock(&dlm->work_lock);
1522         list_add_tail(&item->list, &dlm->work_list);
1523         spin_unlock(&dlm->work_lock);
1524         queue_work(dlm->dlm_worker, &dlm->dispatched_work);
1525 
1526 leave:
1527         /* One extra ref taken needs to be put here */
1528         if (extra_refs)
1529                 dlm_lockres_put(res);
1530 
1531         dlm_put(dlm);
1532         if (ret < 0) {
1533                 kfree(buf);
1534                 kfree(item);
1535                 mlog_errno(ret);
1536         }
1537 
1538         return ret;
1539 }
1540 
1541 
1542 static void dlm_mig_lockres_worker(struct dlm_work_item *item, void *data)
1543 {
1544         struct dlm_ctxt *dlm;
1545         struct dlm_migratable_lockres *mres;
1546         int ret = 0;
1547         struct dlm_lock_resource *res;
1548         u8 real_master;
1549         u8 extra_ref;
1550 
1551         dlm = item->dlm;
1552         mres = (struct dlm_migratable_lockres *)data;
1553 
1554         res = item->u.ml.lockres;
1555         real_master = item->u.ml.real_master;
1556         extra_ref = item->u.ml.extra_ref;
1557 
1558         if (real_master == DLM_LOCK_RES_OWNER_UNKNOWN) {
1559                 /* this case is super-rare. only occurs if
1560                  * node death happens during migration. */
1561 again:
1562                 ret = dlm_lockres_master_requery(dlm, res, &real_master);
1563                 if (ret < 0) {
1564                         mlog(0, "dlm_lockres_master_requery ret=%d\n",
1565                                   ret);
1566                         goto again;
1567                 }
1568                 if (real_master == DLM_LOCK_RES_OWNER_UNKNOWN) {
1569                         mlog(0, "lockres %.*s not claimed.  "
1570                                    "this node will take it.\n",
1571                                    res->lockname.len, res->lockname.name);
1572                 } else {
1573                         spin_lock(&res->spinlock);
1574                         dlm_lockres_drop_inflight_ref(dlm, res);
1575                         spin_unlock(&res->spinlock);
1576                         mlog(0, "master needs to respond to sender "
1577                                   "that node %u still owns %.*s\n",
1578                                   real_master, res->lockname.len,
1579                                   res->lockname.name);
1580                         /* cannot touch this lockres */
1581                         goto leave;
1582                 }
1583         }
1584 
1585         ret = dlm_process_recovery_data(dlm, res, mres);
1586         if (ret < 0)
1587                 mlog(0, "dlm_process_recovery_data returned  %d\n", ret);
1588         else
1589                 mlog(0, "dlm_process_recovery_data succeeded\n");
1590 
1591         if ((mres->flags & (DLM_MRES_MIGRATION|DLM_MRES_ALL_DONE)) ==
1592                            (DLM_MRES_MIGRATION|DLM_MRES_ALL_DONE)) {
1593                 ret = dlm_finish_migration(dlm, res, mres->master);
1594                 if (ret < 0)
1595                         mlog_errno(ret);
1596         }
1597 
1598 leave:
1599         /* See comment in dlm_mig_lockres_handler() */
1600         if (res) {
1601                 if (extra_ref)
1602                         dlm_lockres_put(res);
1603                 dlm_lockres_put(res);
1604         }
1605         kfree(data);
1606 }
1607 
1608 
1609 
1610 static int dlm_lockres_master_requery(struct dlm_ctxt *dlm,
1611                                       struct dlm_lock_resource *res,
1612                                       u8 *real_master)
1613 {
1614         struct dlm_node_iter iter;
1615         int nodenum;
1616         int ret = 0;
1617 
1618         *real_master = DLM_LOCK_RES_OWNER_UNKNOWN;
1619 
1620         /* we only reach here if one of the two nodes in a
1621          * migration died while the migration was in progress.
1622          * at this point we need to requery the master.  we
1623          * know that the new_master got as far as creating
1624          * an mle on at least one node, but we do not know
1625          * if any nodes had actually cleared the mle and set
1626          * the master to the new_master.  the old master
1627          * is supposed to set the owner to UNKNOWN in the
1628          * event of a new_master death, so the only possible
1629          * responses that we can get from nodes here are
1630          * that the master is new_master, or that the master
1631          * is UNKNOWN.
1632          * if all nodes come back with UNKNOWN then we know
1633          * the lock needs remastering here.
1634          * if any node comes back with a valid master, check
1635          * to see if that master is the one that we are
1636          * recovering.  if so, then the new_master died and
1637          * we need to remaster this lock.  if not, then the
1638          * new_master survived and that node will respond to
1639          * other nodes about the owner.
1640          * if there is an owner, this node needs to dump this
1641          * lockres and alert the sender that this lockres
1642          * was rejected. */
1643         spin_lock(&dlm->spinlock);
1644         dlm_node_iter_init(dlm->domain_map, &iter);
1645         spin_unlock(&dlm->spinlock);
1646 
1647         while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
1648                 /* do not send to self */
1649                 if (nodenum == dlm->node_num)
1650                         continue;
1651                 ret = dlm_do_master_requery(dlm, res, nodenum, real_master);
1652                 if (ret < 0) {
1653                         mlog_errno(ret);
1654                         if (!dlm_is_host_down(ret))
1655                                 BUG();
1656                         /* host is down, so answer for that node would be
1657                          * DLM_LOCK_RES_OWNER_UNKNOWN.  continue. */
1658                 }
1659                 if (*real_master != DLM_LOCK_RES_OWNER_UNKNOWN) {
1660                         mlog(0, "lock master is %u\n", *real_master);
1661                         break;
1662                 }
1663         }
1664         return ret;
1665 }
1666 
1667 
1668 int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
1669                           u8 nodenum, u8 *real_master)
1670 {
1671         int ret = -EINVAL;
1672         struct dlm_master_requery req;
1673         int status = DLM_LOCK_RES_OWNER_UNKNOWN;
1674 
1675         memset(&req, 0, sizeof(req));
1676         req.node_idx = dlm->node_num;
1677         req.namelen = res->lockname.len;
1678         memcpy(req.name, res->lockname.name, res->lockname.len);
1679 
1680 resend:
1681         ret = o2net_send_message(DLM_MASTER_REQUERY_MSG, dlm->key,
1682                                  &req, sizeof(req), nodenum, &status);
1683         if (ret < 0)
1684                 mlog(ML_ERROR, "Error %d when sending message %u (key "
1685                      "0x%x) to node %u\n", ret, DLM_MASTER_REQUERY_MSG,
1686                      dlm->key, nodenum);
1687         else if (status == -ENOMEM) {
1688                 mlog_errno(status);
1689                 msleep(50);
1690                 goto resend;
1691         } else {
1692                 BUG_ON(status < 0);
1693                 BUG_ON(status > DLM_LOCK_RES_OWNER_UNKNOWN);
1694                 *real_master = (u8) (status & 0xff);
1695                 mlog(0, "node %u responded to master requery with %u\n",
1696                           nodenum, *real_master);
1697                 ret = 0;
1698         }
1699         return ret;
1700 }
1701 
1702 
1703 /* this function cannot error, so unless the sending
1704  * or receiving of the message failed, the owner can
1705  * be trusted */
1706 int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data,
1707                                void **ret_data)
1708 {
1709         struct dlm_ctxt *dlm = data;
1710         struct dlm_master_requery *req = (struct dlm_master_requery *)msg->buf;
1711         struct dlm_lock_resource *res = NULL;
1712         unsigned int hash;
1713         int master = DLM_LOCK_RES_OWNER_UNKNOWN;
1714         u32 flags = DLM_ASSERT_MASTER_REQUERY;
1715         int dispatched = 0;
1716 
1717         if (!dlm_grab(dlm)) {
1718                 /* since the domain has gone away on this
1719                  * node, the proper response is UNKNOWN */
1720                 return master;
1721         }
1722 
1723         hash = dlm_lockid_hash(req->name, req->namelen);
1724 
1725         spin_lock(&dlm->spinlock);
1726         res = __dlm_lookup_lockres(dlm, req->name, req->namelen, hash);
1727         if (res) {
1728                 spin_lock(&res->spinlock);
1729                 master = res->owner;
1730                 if (master == dlm->node_num) {
1731                         int ret = dlm_dispatch_assert_master(dlm, res,
1732                                                              0, 0, flags);
1733                         if (ret < 0) {
1734                                 mlog_errno(ret);
1735                                 spin_unlock(&res->spinlock);
1736                                 dlm_lockres_put(res);
1737                                 spin_unlock(&dlm->spinlock);
1738                                 dlm_put(dlm);
1739                                 /* sender will take care of this and retry */
1740                                 return ret;
1741                         } else {
1742                                 dispatched = 1;
1743                                 __dlm_lockres_grab_inflight_worker(dlm, res);
1744                                 spin_unlock(&res->spinlock);
1745                         }
1746                 } else {
1747                         /* put.. incase we are not the master */
1748                         spin_unlock(&res->spinlock);
1749                         dlm_lockres_put(res);
1750                 }
1751         }
1752         spin_unlock(&dlm->spinlock);
1753 
1754         if (!dispatched)
1755                 dlm_put(dlm);
1756         return master;
1757 }
1758 
1759 static inline struct list_head *
1760 dlm_list_num_to_pointer(struct dlm_lock_resource *res, int list_num)
1761 {
1762         struct list_head *ret;
1763         BUG_ON(list_num < 0);
1764         BUG_ON(list_num > 2);
1765         ret = &(res->granted);
1766         ret += list_num;
1767         return ret;
1768 }
1769 /* TODO: do ast flush business
1770  * TODO: do MIGRATING and RECOVERING spinning
1771  */
1772 
1773 /*
1774 * NOTE about in-flight requests during migration:
1775 *
1776 * Before attempting the migrate, the master has marked the lockres as
1777 * MIGRATING and then flushed all of its pending ASTS.  So any in-flight
1778 * requests either got queued before the MIGRATING flag got set, in which
1779 * case the lock data will reflect the change and a return message is on
1780 * the way, or the request failed to get in before MIGRATING got set.  In
1781 * this case, the caller will be told to spin and wait for the MIGRATING
1782 * flag to be dropped, then recheck the master.
1783 * This holds true for the convert, cancel and unlock cases, and since lvb
1784 * updates are tied to these same messages, it applies to lvb updates as
1785 * well.  For the lock case, there is no way a lock can be on the master
1786 * queue and not be on the secondary queue since the lock is always added
1787 * locally first.  This means that the new target node will never be sent
1788 * a lock that he doesn't already have on the list.
1789 * In total, this means that the local lock is correct and should not be
1790 * updated to match the one sent by the master.  Any messages sent back
1791 * from the master before the MIGRATING flag will bring the lock properly
1792 * up-to-date, and the change will be ordered properly for the waiter.
1793 * We will *not* attempt to modify the lock underneath the waiter.
1794 */
1795 
1796 static int dlm_process_recovery_data(struct dlm_ctxt *dlm,
1797                                      struct dlm_lock_resource *res,
1798                                      struct dlm_migratable_lockres *mres)
1799 {
1800         struct dlm_migratable_lock *ml;
1801         struct list_head *queue, *iter;
1802         struct list_head *tmpq = NULL;
1803         struct dlm_lock *newlock = NULL;
1804         struct dlm_lockstatus *lksb = NULL;
1805         int ret = 0;
1806         int i, j, bad;
1807         struct dlm_lock *lock;
1808         u8 from = O2NM_MAX_NODES;
1809         __be64 c;
1810 
1811         mlog(0, "running %d locks for this lockres\n", mres->num_locks);
1812         for (i=0; i<mres->num_locks; i++) {
1813                 ml = &(mres->ml[i]);
1814 
1815                 if (dlm_is_dummy_lock(dlm, ml, &from)) {
1816                         /* placeholder, just need to set the refmap bit */
1817                         BUG_ON(mres->num_locks != 1);
1818                         mlog(0, "%s:%.*s: dummy lock for %u\n",
1819                              dlm->name, mres->lockname_len, mres->lockname,
1820                              from);
1821                         spin_lock(&res->spinlock);
1822                         dlm_lockres_set_refmap_bit(dlm, res, from);
1823                         spin_unlock(&res->spinlock);
1824                         break;
1825                 }
1826                 BUG_ON(ml->highest_blocked != LKM_IVMODE);
1827                 newlock = NULL;
1828                 lksb = NULL;
1829 
1830                 queue = dlm_list_num_to_pointer(res, ml->list);
1831                 tmpq = NULL;
1832 
1833                 /* if the lock is for the local node it needs to
1834                  * be moved to the proper location within the queue.
1835                  * do not allocate a new lock structure. */
1836                 if (ml->node == dlm->node_num) {
1837                         /* MIGRATION ONLY! */
1838                         BUG_ON(!(mres->flags & DLM_MRES_MIGRATION));
1839 
1840                         lock = NULL;
1841                         spin_lock(&res->spinlock);
1842                         for (j = DLM_GRANTED_LIST; j <= DLM_BLOCKED_LIST; j++) {
1843                                 tmpq = dlm_list_idx_to_ptr(res, j);
1844                                 list_for_each(iter, tmpq) {
1845                                         lock = list_entry(iter,
1846                                                   struct dlm_lock, list);
1847                                         if (lock->ml.cookie == ml->cookie)
1848                                                 break;
1849                                         lock = NULL;
1850                                 }
1851                                 if (lock)
1852                                         break;
1853                         }
1854 
1855                         /* lock is always created locally first, and
1856                          * destroyed locally last.  it must be on the list */
1857                         if (!lock) {
1858                                 c = ml->cookie;
1859                                 mlog(ML_ERROR, "Could not find local lock "
1860                                                "with cookie %u:%llu, node %u, "
1861                                                "list %u, flags 0x%x, type %d, "
1862                                                "conv %d, highest blocked %d\n",
1863                                      dlm_get_lock_cookie_node(be64_to_cpu(c)),
1864                                      dlm_get_lock_cookie_seq(be64_to_cpu(c)),
1865                                      ml->node, ml->list, ml->flags, ml->type,
1866                                      ml->convert_type, ml->highest_blocked);
1867                                 __dlm_print_one_lock_resource(res);
1868                                 BUG();
1869                         }
1870 
1871                         if (lock->ml.node != ml->node) {
1872                                 c = lock->ml.cookie;
1873                                 mlog(ML_ERROR, "Mismatched node# in lock "
1874                                      "cookie %u:%llu, name %.*s, node %u\n",
1875                                      dlm_get_lock_cookie_node(be64_to_cpu(c)),
1876                                      dlm_get_lock_cookie_seq(be64_to_cpu(c)),
1877                                      res->lockname.len, res->lockname.name,
1878                                      lock->ml.node);
1879                                 c = ml->cookie;
1880                                 mlog(ML_ERROR, "Migrate lock cookie %u:%llu, "
1881                                      "node %u, list %u, flags 0x%x, type %d, "
1882                                      "conv %d, highest blocked %d\n",
1883                                      dlm_get_lock_cookie_node(be64_to_cpu(c)),
1884                                      dlm_get_lock_cookie_seq(be64_to_cpu(c)),
1885                                      ml->node, ml->list, ml->flags, ml->type,
1886                                      ml->convert_type, ml->highest_blocked);
1887                                 __dlm_print_one_lock_resource(res);
1888                                 BUG();
1889                         }
1890 
1891                         if (tmpq != queue) {
1892                                 c = ml->cookie;
1893                                 mlog(0, "Lock cookie %u:%llu was on list %u "
1894                                      "instead of list %u for %.*s\n",
1895                                      dlm_get_lock_cookie_node(be64_to_cpu(c)),
1896                                      dlm_get_lock_cookie_seq(be64_to_cpu(c)),
1897                                      j, ml->list, res->lockname.len,
1898                                      res->lockname.name);
1899                                 __dlm_print_one_lock_resource(res);
1900                                 spin_unlock(&res->spinlock);
1901                                 continue;
1902                         }
1903 
1904                         /* see NOTE above about why we do not update
1905                          * to match the master here */
1906 
1907                         /* move the lock to its proper place */
1908                         /* do not alter lock refcount.  switching lists. */
1909                         list_move_tail(&lock->list, queue);
1910                         spin_unlock(&res->spinlock);
1911 
1912                         mlog(0, "just reordered a local lock!\n");
1913                         continue;
1914                 }
1915 
1916                 /* lock is for another node. */
1917                 newlock = dlm_new_lock(ml->type, ml->node,
1918                                        be64_to_cpu(ml->cookie), NULL);
1919                 if (!newlock) {
1920                         ret = -ENOMEM;
1921                         goto leave;
1922                 }
1923                 lksb = newlock->lksb;
1924                 dlm_lock_attach_lockres(newlock, res);
1925 
1926                 if (ml->convert_type != LKM_IVMODE) {
1927                         BUG_ON(queue != &res->converting);
1928                         newlock->ml.convert_type = ml->convert_type;
1929                 }
1930                 lksb->flags |= (ml->flags &
1931                                 (DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB));
1932 
1933                 if (ml->type == LKM_NLMODE)
1934                         goto skip_lvb;
1935 
1936                 /*
1937                  * If the lock is in the blocked list it can't have a valid lvb,
1938                  * so skip it
1939                  */
1940                 if (ml->list == DLM_BLOCKED_LIST)
1941                         goto skip_lvb;
1942 
1943                 if (!dlm_lvb_is_empty(mres->lvb)) {
1944                         if (lksb->flags & DLM_LKSB_PUT_LVB) {
1945                                 /* other node was trying to update
1946                                  * lvb when node died.  recreate the
1947                                  * lksb with the updated lvb. */
1948                                 memcpy(lksb->lvb, mres->lvb, DLM_LVB_LEN);
1949                                 /* the lock resource lvb update must happen
1950                                  * NOW, before the spinlock is dropped.
1951                                  * we no longer wait for the AST to update
1952                                  * the lvb. */
1953                                 memcpy(res->lvb, mres->lvb, DLM_LVB_LEN);
1954                         } else {
1955                                 /* otherwise, the node is sending its
1956                                  * most recent valid lvb info */
1957                                 BUG_ON(ml->type != LKM_EXMODE &&
1958                                        ml->type != LKM_PRMODE);
1959                                 if (!dlm_lvb_is_empty(res->lvb) &&
1960                                     (ml->type == LKM_EXMODE ||
1961                                      memcmp(res->lvb, mres->lvb, DLM_LVB_LEN))) {
1962                                         int i;
1963                                         mlog(ML_ERROR, "%s:%.*s: received bad "
1964                                              "lvb! type=%d\n", dlm->name,
1965                                              res->lockname.len,
1966                                              res->lockname.name, ml->type);
1967                                         printk("lockres lvb=[");
1968                                         for (i=0; i<DLM_LVB_LEN; i++)
1969                                                 printk("%02x", res->lvb[i]);
1970                                         printk("]\nmigrated lvb=[");
1971                                         for (i=0; i<DLM_LVB_LEN; i++)
1972                                                 printk("%02x", mres->lvb[i]);
1973                                         printk("]\n");
1974                                         dlm_print_one_lock_resource(res);
1975                                         BUG();
1976                                 }
1977                                 memcpy(res->lvb, mres->lvb, DLM_LVB_LEN);
1978                         }
1979                 }
1980 skip_lvb:
1981 
1982                 /* NOTE:
1983                  * wrt lock queue ordering and recovery:
1984                  *    1. order of locks on granted queue is
1985                  *       meaningless.
1986                  *    2. order of locks on converting queue is
1987                  *       LOST with the node death.  sorry charlie.
1988                  *    3. order of locks on the blocked queue is
1989                  *       also LOST.
1990                  * order of locks does not affect integrity, it
1991                  * just means that a lock request may get pushed
1992                  * back in line as a result of the node death.
1993                  * also note that for a given node the lock order
1994                  * for its secondary queue locks is preserved
1995                  * relative to each other, but clearly *not*
1996                  * preserved relative to locks from other nodes.
1997                  */
1998                 bad = 0;
1999                 spin_lock(&res->spinlock);
2000                 list_for_each_entry(lock, queue, list) {
2001                         if (lock->ml.cookie == ml->cookie) {
2002                                 c = lock->ml.cookie;
2003                                 mlog(ML_ERROR, "%s:%.*s: %u:%llu: lock already "
2004                                      "exists on this lockres!\n", dlm->name,
2005                                      res->lockname.len, res->lockname.name,
2006                                      dlm_get_lock_cookie_node(be64_to_cpu(c)),
2007                                      dlm_get_lock_cookie_seq(be64_to_cpu(c)));
2008 
2009                                 mlog(ML_NOTICE, "sent lock: type=%d, conv=%d, "
2010                                      "node=%u, cookie=%u:%llu, queue=%d\n",
2011                                      ml->type, ml->convert_type, ml->node,
2012                                      dlm_get_lock_cookie_node(be64_to_cpu(ml->cookie)),
2013                                      dlm_get_lock_cookie_seq(be64_to_cpu(ml->cookie)),
2014                                      ml->list);
2015 
2016                                 __dlm_print_one_lock_resource(res);
2017                                 bad = 1;
2018                                 break;
2019                         }
2020                 }
2021                 if (!bad) {
2022                         dlm_lock_get(newlock);
2023                         if (mres->flags & DLM_MRES_RECOVERY &&
2024                                         ml->list == DLM_CONVERTING_LIST &&
2025                                         newlock->ml.type >
2026                                         newlock->ml.convert_type) {
2027                                 /* newlock is doing downconvert, add it to the
2028                                  * head of converting list */
2029                                 list_add(&newlock->list, queue);
2030                         } else
2031                                 list_add_tail(&newlock->list, queue);
2032                         mlog(0, "%s:%.*s: added lock for node %u, "
2033                              "setting refmap bit\n", dlm->name,
2034                              res->lockname.len, res->lockname.name, ml->node);
2035                         dlm_lockres_set_refmap_bit(dlm, res, ml->node);
2036                 }
2037                 spin_unlock(&res->spinlock);
2038         }
2039         mlog(0, "done running all the locks\n");
2040 
2041 leave:
2042         /* balance the ref taken when the work was queued */
2043         spin_lock(&res->spinlock);
2044         dlm_lockres_drop_inflight_ref(dlm, res);
2045         spin_unlock(&res->spinlock);
2046 
2047         if (ret < 0)
2048                 mlog_errno(ret);
2049 
2050         return ret;
2051 }
2052 
2053 void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
2054                                        struct dlm_lock_resource *res)
2055 {
2056         int i;
2057         struct list_head *queue;
2058         struct dlm_lock *lock, *next;
2059 
2060         assert_spin_locked(&dlm->spinlock);
2061         assert_spin_locked(&res->spinlock);
2062         res->state |= DLM_LOCK_RES_RECOVERING;
2063         if (!list_empty(&res->recovering)) {
2064                 mlog(0,
2065                      "Recovering res %s:%.*s, is already on recovery list!\n",
2066                      dlm->name, res->lockname.len, res->lockname.name);
2067                 list_del_init(&res->recovering);
2068                 dlm_lockres_put(res);
2069         }
2070         /* We need to hold a reference while on the recovery list */
2071         dlm_lockres_get(res);
2072         list_add_tail(&res->recovering, &dlm->reco.resources);
2073 
2074         /* find any pending locks and put them back on proper list */
2075         for (i=DLM_BLOCKED_LIST; i>=DLM_GRANTED_LIST; i--) {
2076                 queue = dlm_list_idx_to_ptr(res, i);
2077                 list_for_each_entry_safe(lock, next, queue, list) {
2078                         dlm_lock_get(lock);
2079                         if (lock->convert_pending) {
2080                                 /* move converting lock back to granted */
2081                                 mlog(0, "node died with convert pending "
2082                                      "on %.*s. move back to granted list.\n",
2083                                      res->lockname.len, res->lockname.name);
2084                                 dlm_revert_pending_convert(res, lock);
2085                                 lock->convert_pending = 0;
2086                         } else if (lock->lock_pending) {
2087                                 /* remove pending lock requests completely */
2088                                 BUG_ON(i != DLM_BLOCKED_LIST);
2089                                 mlog(0, "node died with lock pending "
2090                                      "on %.*s. remove from blocked list and skip.\n",
2091                                      res->lockname.len, res->lockname.name);
2092                                 /* lock will be floating until ref in
2093                                  * dlmlock_remote is freed after the network
2094                                  * call returns.  ok for it to not be on any
2095                                  * list since no ast can be called
2096                                  * (the master is dead). */
2097                                 dlm_revert_pending_lock(res, lock);
2098                                 lock->lock_pending = 0;
2099                         } else if (lock->unlock_pending) {
2100                                 /* if an unlock was in progress, treat as
2101                                  * if this had completed successfully
2102                                  * before sending this lock state to the
2103                                  * new master.  note that the dlm_unlock
2104                                  * call is still responsible for calling
2105                                  * the unlockast.  that will happen after
2106                                  * the network call times out.  for now,
2107                                  * just move lists to prepare the new
2108                                  * recovery master.  */
2109                                 BUG_ON(i != DLM_GRANTED_LIST);
2110                                 mlog(0, "node died with unlock pending "
2111                                      "on %.*s. remove from blocked list and skip.\n",
2112                                      res->lockname.len, res->lockname.name);
2113                                 dlm_commit_pending_unlock(res, lock);
2114                                 lock->unlock_pending = 0;
2115                         } else if (lock->cancel_pending) {
2116                                 /* if a cancel was in progress, treat as
2117                                  * if this had completed successfully
2118                                  * before sending this lock state to the
2119                                  * new master */
2120                                 BUG_ON(i != DLM_CONVERTING_LIST);
2121                                 mlog(0, "node died with cancel pending "
2122                                      "on %.*s. move back to granted list.\n",
2123                                      res->lockname.len, res->lockname.name);
2124                                 dlm_commit_pending_cancel(res, lock);
2125                                 lock->cancel_pending = 0;
2126                         }
2127                         dlm_lock_put(lock);
2128                 }
2129         }
2130 }
2131 
2132 
2133 
2134 /* removes all recovered locks from the recovery list.
2135  * sets the res->owner to the new master.
2136  * unsets the RECOVERY flag and wakes waiters. */
2137 static void dlm_finish_local_lockres_recovery(struct dlm_ctxt *dlm,
2138                                               u8 dead_node, u8 new_master)
2139 {
2140         int i;
2141         struct hlist_head *bucket;
2142         struct dlm_lock_resource *res, *next;
2143 
2144         assert_spin_locked(&dlm->spinlock);
2145 
2146         list_for_each_entry_safe(res, next, &dlm->reco.resources, recovering) {
2147                 if (res->owner == dead_node) {
2148                         mlog(0, "%s: res %.*s, Changing owner from %u to %u\n",
2149                              dlm->name, res->lockname.len, res->lockname.name,
2150                              res->owner, new_master);
2151                         list_del_init(&res->recovering);
2152                         spin_lock(&res->spinlock);
2153                         /* new_master has our reference from
2154                          * the lock state sent during recovery */
2155                         dlm_change_lockres_owner(dlm, res, new_master);
2156                         res->state &= ~DLM_LOCK_RES_RECOVERING;
2157                         if (__dlm_lockres_has_locks(res))
2158                                 __dlm_dirty_lockres(dlm, res);
2159                         spin_unlock(&res->spinlock);
2160                         wake_up(&res->wq);
2161                         dlm_lockres_put(res);
2162                 }
2163         }
2164 
2165         /* this will become unnecessary eventually, but
2166          * for now we need to run the whole hash, clear
2167          * the RECOVERING state and set the owner
2168          * if necessary */
2169         for (i = 0; i < DLM_HASH_BUCKETS; i++) {
2170                 bucket = dlm_lockres_hash(dlm, i);
2171                 hlist_for_each_entry(res, bucket, hash_node) {
2172                         if (res->state & DLM_LOCK_RES_RECOVERY_WAITING) {
2173                                 spin_lock(&res->spinlock);
2174                                 res->state &= ~DLM_LOCK_RES_RECOVERY_WAITING;
2175                                 spin_unlock(&res->spinlock);
2176                                 wake_up(&res->wq);
2177                         }
2178 
2179                         if (!(res->state & DLM_LOCK_RES_RECOVERING))
2180                                 continue;
2181 
2182                         if (res->owner != dead_node &&
2183                             res->owner != dlm->node_num)
2184                                 continue;
2185 
2186                         if (!list_empty(&res->recovering)) {
2187                                 list_del_init(&res->recovering);
2188                                 dlm_lockres_put(res);
2189                         }
2190 
2191                         /* new_master has our reference from
2192                          * the lock state sent during recovery */
2193                         mlog(0, "%s: res %.*s, Changing owner from %u to %u\n",
2194                              dlm->name, res->lockname.len, res->lockname.name,
2195                              res->owner, new_master);
2196                         spin_lock(&res->spinlock);
2197                         dlm_change_lockres_owner(dlm, res, new_master);
2198                         res->state &= ~DLM_LOCK_RES_RECOVERING;
2199                         if (__dlm_lockres_has_locks(res))
2200                                 __dlm_dirty_lockres(dlm, res);
2201                         spin_unlock(&res->spinlock);
2202                         wake_up(&res->wq);
2203                 }
2204         }
2205 }
2206 
2207 static inline int dlm_lvb_needs_invalidation(struct dlm_lock *lock, int local)
2208 {
2209         if (local) {
2210                 if (lock->ml.type != LKM_EXMODE &&
2211                     lock->ml.type != LKM_PRMODE)
2212                         return 1;
2213         } else if (lock->ml.type == LKM_EXMODE)
2214                 return 1;
2215         return 0;
2216 }
2217 
2218 static void dlm_revalidate_lvb(struct dlm_ctxt *dlm,
2219                                struct dlm_lock_resource *res, u8 dead_node)
2220 {
2221         struct list_head *queue;
2222         struct dlm_lock *lock;
2223         int blank_lvb = 0, local = 0;
2224         int i;
2225         u8 search_node;
2226 
2227         assert_spin_locked(&dlm->spinlock);
2228         assert_spin_locked(&res->spinlock);
2229 
2230         if (res->owner == dlm->node_num)
2231                 /* if this node owned the lockres, and if the dead node
2232                  * had an EX when he died, blank out the lvb */
2233                 search_node = dead_node;
2234         else {
2235                 /* if this is a secondary lockres, and we had no EX or PR
2236                  * locks granted, we can no longer trust the lvb */
2237                 search_node = dlm->node_num;
2238                 local = 1;  /* check local state for valid lvb */
2239         }
2240 
2241         for (i=DLM_GRANTED_LIST; i<=DLM_CONVERTING_LIST; i++) {
2242                 queue = dlm_list_idx_to_ptr(res, i);
2243                 list_for_each_entry(lock, queue, list) {
2244                         if (lock->ml.node == search_node) {
2245                                 if (dlm_lvb_needs_invalidation(lock, local)) {
2246                                         /* zero the lksb lvb and lockres lvb */
2247                                         blank_lvb = 1;
2248                                         memset(lock->lksb->lvb, 0, DLM_LVB_LEN);
2249                                 }
2250                         }
2251                 }
2252         }
2253 
2254         if (blank_lvb) {
2255                 mlog(0, "clearing %.*s lvb, dead node %u had EX\n",
2256                      res->lockname.len, res->lockname.name, dead_node);
2257                 memset(res->lvb, 0, DLM_LVB_LEN);
2258         }
2259 }
2260 
2261 static void dlm_free_dead_locks(struct dlm_ctxt *dlm,
2262                                 struct dlm_lock_resource *res, u8 dead_node)
2263 {
2264         struct dlm_lock *lock, *next;
2265         unsigned int freed = 0;
2266 
2267         /* this node is the lockres master:
2268          * 1) remove any stale locks for the dead node
2269          * 2) if the dead node had an EX when he died, blank out the lvb
2270          */
2271         assert_spin_locked(&dlm->spinlock);
2272         assert_spin_locked(&res->spinlock);
2273 
2274         /* We do two dlm_lock_put(). One for removing from list and the other is
2275          * to force the DLM_UNLOCK_FREE_LOCK action so as to free the locks */
2276 
2277         /* TODO: check pending_asts, pending_basts here */
2278         list_for_each_entry_safe(lock, next, &res->granted, list) {
2279                 if (lock->ml.node == dead_node) {
2280                         list_del_init(&lock->list);
2281                         dlm_lock_put(lock);
2282                         /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
2283                         dlm_lock_put(lock);
2284                         freed++;
2285                 }
2286         }
2287         list_for_each_entry_safe(lock, next, &res->converting, list) {
2288                 if (lock->ml.node == dead_node) {
2289                         list_del_init(&lock->list);
2290                         dlm_lock_put(lock);
2291                         /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
2292                         dlm_lock_put(lock);
2293                         freed++;
2294                 }
2295         }
2296         list_for_each_entry_safe(lock, next, &res->blocked, list) {
2297                 if (lock->ml.node == dead_node) {
2298                         list_del_init(&lock->list);
2299                         dlm_lock_put(lock);
2300                         /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
2301                         dlm_lock_put(lock);
2302                         freed++;
2303                 }
2304         }
2305 
2306         if (freed) {
2307                 mlog(0, "%s:%.*s: freed %u locks for dead node %u, "
2308                      "dropping ref from lockres\n", dlm->name,
2309                      res->lockname.len, res->lockname.name, freed, dead_node);
2310                 if(!test_bit(dead_node, res->refmap)) {
2311                         mlog(ML_ERROR, "%s:%.*s: freed %u locks for dead node %u, "
2312                              "but ref was not set\n", dlm->name,
2313                              res->lockname.len, res->lockname.name, freed, dead_node);
2314                         __dlm_print_one_lock_resource(res);
2315                 }
2316                 res->state |= DLM_LOCK_RES_RECOVERY_WAITING;
2317                 dlm_lockres_clear_refmap_bit(dlm, res, dead_node);
2318         } else if (test_bit(dead_node, res->refmap)) {
2319                 mlog(0, "%s:%.*s: dead node %u had a ref, but had "
2320                      "no locks and had not purged before dying\n", dlm->name,
2321                      res->lockname.len, res->lockname.name, dead_node);
2322                 dlm_lockres_clear_refmap_bit(dlm, res, dead_node);
2323         }
2324 
2325         /* do not kick thread yet */
2326         __dlm_dirty_lockres(dlm, res);
2327 }
2328 
2329 static void dlm_do_local_recovery_cleanup(struct dlm_ctxt *dlm, u8 dead_node)
2330 {
2331         struct dlm_lock_resource *res;
2332         int i;
2333         struct hlist_head *bucket;
2334         struct hlist_node *tmp;
2335         struct dlm_lock *lock;
2336 
2337 
2338         /* purge any stale mles */
2339         dlm_clean_master_list(dlm, dead_node);
2340 
2341         /*
2342          * now clean up all lock resources.  there are two rules:
2343          *
2344          * 1) if the dead node was the master, move the lockres
2345          *    to the recovering list.  set the RECOVERING flag.
2346          *    this lockres needs to be cleaned up before it can
2347          *    be used further.
2348          *
2349          * 2) if this node was the master, remove all locks from
2350          *    each of the lockres queues that were owned by the
2351          *    dead node.  once recovery finishes, the dlm thread
2352          *    can be kicked again to see if any ASTs or BASTs
2353          *    need to be fired as a result.
2354          */
2355         for (i = 0; i < DLM_HASH_BUCKETS; i++) {
2356                 bucket = dlm_lockres_hash(dlm, i);
2357                 hlist_for_each_entry_safe(res, tmp, bucket, hash_node) {
2358                         /* always prune any $RECOVERY entries for dead nodes,
2359                          * otherwise hangs can occur during later recovery */
2360                         if (dlm_is_recovery_lock(res->lockname.name,
2361                                                  res->lockname.len)) {
2362                                 spin_lock(&res->spinlock);
2363                                 list_for_each_entry(lock, &res->granted, list) {
2364                                         if (lock->ml.node == dead_node) {
2365                                                 mlog(0, "AHA! there was "
2366                                                      "a $RECOVERY lock for dead "
2367                                                      "node %u (%s)!\n",
2368                                                      dead_node, dlm->name);
2369                                                 list_del_init(&lock->list);
2370                                                 dlm_lock_put(lock);
2371                                                 /* Can't schedule
2372                                                  * DLM_UNLOCK_FREE_LOCK
2373                                                  * - do manually */
2374                                                 dlm_lock_put(lock);
2375                                                 break;
2376                                         }
2377                                 }
2378 
2379                                 if ((res->owner == dead_node) &&
2380                                                         (res->state & DLM_LOCK_RES_DROPPING_REF)) {
2381                                         dlm_lockres_get(res);
2382                                         __dlm_do_purge_lockres(dlm, res);
2383                                         spin_unlock(&res->spinlock);
2384                                         wake_up(&res->wq);
2385                                         dlm_lockres_put(res);
2386                                         continue;
2387                                 } else if (res->owner == dlm->node_num)
2388                                         dlm_lockres_clear_refmap_bit(dlm, res, dead_node);
2389                                 spin_unlock(&res->spinlock);
2390                                 continue;
2391                         }
2392                         spin_lock(&res->spinlock);
2393                         /* zero the lvb if necessary */
2394                         dlm_revalidate_lvb(dlm, res, dead_node);
2395                         if (res->owner == dead_node) {
2396                                 if (res->state & DLM_LOCK_RES_DROPPING_REF) {
2397                                         mlog(0, "%s:%.*s: owned by "
2398                                                 "dead node %u, this node was "
2399                                                 "dropping its ref when master died. "
2400                                                 "continue, purging the lockres.\n",
2401                                                 dlm->name, res->lockname.len,
2402                                                 res->lockname.name, dead_node);
2403                                         dlm_lockres_get(res);
2404                                         __dlm_do_purge_lockres(dlm, res);
2405                                         spin_unlock(&res->spinlock);
2406                                         wake_up(&res->wq);
2407                                         dlm_lockres_put(res);
2408                                         continue;
2409                                 }
2410                                 dlm_move_lockres_to_recovery_list(dlm, res);
2411                         } else if (res->owner == dlm->node_num) {
2412                                 dlm_free_dead_locks(dlm, res, dead_node);
2413                                 __dlm_lockres_calc_usage(dlm, res);
2414                         } else if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
2415                                 if (test_bit(dead_node, res->refmap)) {
2416                                         mlog(0, "%s:%.*s: dead node %u had a ref, but had "
2417                                                 "no locks and had not purged before dying\n",
2418                                                 dlm->name, res->lockname.len,
2419                                                 res->lockname.name, dead_node);
2420                                         dlm_lockres_clear_refmap_bit(dlm, res, dead_node);
2421                                 }
2422                         }
2423                         spin_unlock(&res->spinlock);
2424                 }
2425         }
2426 
2427 }
2428 
2429 static void __dlm_hb_node_down(struct dlm_ctxt *dlm, int idx)
2430 {
2431         assert_spin_locked(&dlm->spinlock);
2432 
2433         if (dlm->reco.new_master == idx) {
2434                 mlog(0, "%s: recovery master %d just died\n",
2435                      dlm->name, idx);
2436                 if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
2437                         /* finalize1 was reached, so it is safe to clear
2438                          * the new_master and dead_node.  that recovery
2439                          * is complete. */
2440                         mlog(0, "%s: dead master %d had reached "
2441                              "finalize1 state, clearing\n", dlm->name, idx);
2442                         dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE;
2443                         __dlm_reset_recovery(dlm);
2444                 }
2445         }
2446 
2447         /* Clean up join state on node death. */
2448         if (dlm->joining_node == idx) {
2449                 mlog(0, "Clearing join state for node %u\n", idx);
2450                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
2451         }
2452 
2453         /* check to see if the node is already considered dead */
2454         if (!test_bit(idx, dlm->live_nodes_map)) {
2455                 mlog(0, "for domain %s, node %d is already dead. "
2456                      "another node likely did recovery already.\n",
2457                      dlm->name, idx);
2458                 return;
2459         }
2460 
2461         /* check to see if we do not care about this node */
2462         if (!test_bit(idx, dlm->domain_map)) {
2463                 /* This also catches the case that we get a node down
2464                  * but haven't joined the domain yet. */
2465                 mlog(0, "node %u already removed from domain!\n", idx);
2466                 return;
2467         }
2468 
2469         clear_bit(idx, dlm->live_nodes_map);
2470 
2471         /* make sure local cleanup occurs before the heartbeat events */
2472         if (!test_bit(idx, dlm->recovery_map))
2473                 dlm_do_local_recovery_cleanup(dlm, idx);
2474 
2475         /* notify anything attached to the heartbeat events */
2476         dlm_hb_event_notify_attached(dlm, idx, 0);
2477 
2478         mlog(0, "node %u being removed from domain map!\n", idx);
2479         clear_bit(idx, dlm->domain_map);
2480         clear_bit(idx, dlm->exit_domain_map);
2481         /* wake up migration waiters if a node goes down.
2482          * perhaps later we can genericize this for other waiters. */
2483         wake_up(&dlm->migration_wq);
2484 
2485         set_bit(idx, dlm->recovery_map);
2486 }
2487 
2488 void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data)
2489 {
2490         struct dlm_ctxt *dlm = data;
2491 
2492         if (!dlm_grab(dlm))
2493                 return;
2494 
2495         /*
2496          * This will notify any dlm users that a node in our domain
2497          * went away without notifying us first.
2498          */
2499         if (test_bit(idx, dlm->domain_map))
2500                 dlm_fire_domain_eviction_callbacks(dlm, idx);
2501 
2502         spin_lock(&dlm->spinlock);
2503         __dlm_hb_node_down(dlm, idx);
2504         spin_unlock(&dlm->spinlock);
2505 
2506         dlm_put(dlm);
2507 }
2508 
2509 void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data)
2510 {
2511         struct dlm_ctxt *dlm = data;
2512 
2513         if (!dlm_grab(dlm))
2514                 return;
2515 
2516         spin_lock(&dlm->spinlock);
2517         set_bit(idx, dlm->live_nodes_map);
2518         /* do NOT notify mle attached to the heartbeat events.
2519          * new nodes are not interesting in mastery until joined. */
2520         spin_unlock(&dlm->spinlock);
2521 
2522         dlm_put(dlm);
2523 }
2524 
2525 static void dlm_reco_ast(void *astdata)
2526 {
2527         struct dlm_ctxt *dlm = astdata;
2528         mlog(0, "ast for recovery lock fired!, this=%u, dlm=%s\n",
2529              dlm->node_num, dlm->name);
2530 }
2531 static void dlm_reco_bast(void *astdata, int blocked_type)
2532 {
2533         struct dlm_ctxt *dlm = astdata;
2534         mlog(0, "bast for recovery lock fired!, this=%u, dlm=%s\n",
2535              dlm->node_num, dlm->name);
2536 }
2537 static void dlm_reco_unlock_ast(void *astdata, enum dlm_status st)
2538 {
2539         mlog(0, "unlockast for recovery lock fired!\n");
2540 }
2541 
2542 /*
2543  * dlm_pick_recovery_master will continually attempt to use
2544  * dlmlock() on the special "$RECOVERY" lockres with the
2545  * LKM_NOQUEUE flag to get an EX.  every thread that enters
2546  * this function on each node racing to become the recovery
2547  * master will not stop attempting this until either:
2548  * a) this node gets the EX (and becomes the recovery master),
2549  * or b) dlm->reco.new_master gets set to some nodenum
2550  * != O2NM_INVALID_NODE_NUM (another node will do the reco).
2551  * so each time a recovery master is needed, the entire cluster
2552  * will sync at this point.  if the new master dies, that will
2553  * be detected in dlm_do_recovery */
2554 static int dlm_pick_recovery_master(struct dlm_ctxt *dlm)
2555 {
2556         enum dlm_status ret;
2557         struct dlm_lockstatus lksb;
2558         int status = -EINVAL;
2559 
2560         mlog(0, "starting recovery of %s at %lu, dead=%u, this=%u\n",
2561              dlm->name, jiffies, dlm->reco.dead_node, dlm->node_num);
2562 again:
2563         memset(&lksb, 0, sizeof(lksb));
2564 
2565         ret = dlmlock(dlm, LKM_EXMODE, &lksb, LKM_NOQUEUE|LKM_RECOVERY,
2566                       DLM_RECOVERY_LOCK_NAME, DLM_RECOVERY_LOCK_NAME_LEN,
2567                       dlm_reco_ast, dlm, dlm_reco_bast);
2568 
2569         mlog(0, "%s: dlmlock($RECOVERY) returned %d, lksb=%d\n",
2570              dlm->name, ret, lksb.status);
2571 
2572         if (ret == DLM_NORMAL) {
2573                 mlog(0, "dlm=%s dlmlock says I got it (this=%u)\n",
2574                      dlm->name, dlm->node_num);
2575 
2576                 /* got the EX lock.  check to see if another node
2577                  * just became the reco master */
2578                 if (dlm_reco_master_ready(dlm)) {
2579                         mlog(0, "%s: got reco EX lock, but %u will "
2580                              "do the recovery\n", dlm->name,
2581                              dlm->reco.new_master);
2582                         status = -EEXIST;
2583                 } else {
2584                         status = 0;
2585 
2586                         /* see if recovery was already finished elsewhere */
2587                         spin_lock(&dlm->spinlock);
2588                         if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) {
2589                                 status = -EINVAL;
2590                                 mlog(0, "%s: got reco EX lock, but "
2591                                      "node got recovered already\n", dlm->name);
2592                                 if (dlm->reco.new_master != O2NM_INVALID_NODE_NUM) {
2593                                         mlog(ML_ERROR, "%s: new master is %u "
2594                                              "but no dead node!\n",
2595                                              dlm->name, dlm->reco.new_master);
2596                                         BUG();
2597                                 }
2598                         }
2599                         spin_unlock(&dlm->spinlock);
2600                 }
2601 
2602                 /* if this node has actually become the recovery master,
2603                  * set the master and send the messages to begin recovery */
2604                 if (!status) {
2605                         mlog(0, "%s: dead=%u, this=%u, sending "
2606                              "begin_reco now\n", dlm->name,
2607                              dlm->reco.dead_node, dlm->node_num);
2608                         status = dlm_send_begin_reco_message(dlm,
2609                                       dlm->reco.dead_node);
2610                         /* this always succeeds */
2611                         BUG_ON(status);
2612 
2613                         /* set the new_master to this node */
2614                         spin_lock(&dlm->spinlock);
2615                         dlm_set_reco_master(dlm, dlm->node_num);
2616                         spin_unlock(&dlm->spinlock);
2617                 }
2618 
2619                 /* recovery lock is a special case.  ast will not get fired,
2620                  * so just go ahead and unlock it. */
2621                 ret = dlmunlock(dlm, &lksb, 0, dlm_reco_unlock_ast, dlm);
2622                 if (ret == DLM_DENIED) {
2623                         mlog(0, "got DLM_DENIED, trying LKM_CANCEL\n");
2624                         ret = dlmunlock(dlm, &lksb, LKM_CANCEL, dlm_reco_unlock_ast, dlm);
2625                 }
2626                 if (ret != DLM_NORMAL) {
2627                         /* this would really suck. this could only happen
2628                          * if there was a network error during the unlock
2629                          * because of node death.  this means the unlock
2630                          * is actually "done" and the lock structure is
2631                          * even freed.  we can continue, but only
2632                          * because this specific lock name is special. */
2633                         mlog(ML_ERROR, "dlmunlock returned %d\n", ret);
2634                 }
2635         } else if (ret == DLM_NOTQUEUED) {
2636                 mlog(0, "dlm=%s dlmlock says another node got it (this=%u)\n",
2637                      dlm->name, dlm->node_num);
2638                 /* another node is master. wait on
2639                  * reco.new_master != O2NM_INVALID_NODE_NUM
2640                  * for at most one second */
2641                 wait_event_timeout(dlm->dlm_reco_thread_wq,
2642                                          dlm_reco_master_ready(dlm),
2643                                          msecs_to_jiffies(1000));
2644                 if (!dlm_reco_master_ready(dlm)) {
2645                         mlog(0, "%s: reco master taking awhile\n",
2646                              dlm->name);
2647                         goto again;
2648                 }
2649                 /* another node has informed this one that it is reco master */
2650                 mlog(0, "%s: reco master %u is ready to recover %u\n",
2651                      dlm->name, dlm->reco.new_master, dlm->reco.dead_node);
2652                 status = -EEXIST;
2653         } else if (ret == DLM_RECOVERING) {
2654                 mlog(0, "dlm=%s dlmlock says master node died (this=%u)\n",
2655                      dlm->name, dlm->node_num);
2656                 goto again;
2657         } else {
2658                 struct dlm_lock_resource *res;
2659 
2660                 /* dlmlock returned something other than NOTQUEUED or NORMAL */
2661                 mlog(ML_ERROR, "%s: got %s from dlmlock($RECOVERY), "
2662                      "lksb.status=%s\n", dlm->name, dlm_errname(ret),
2663                      dlm_errname(lksb.status));
2664                 res = dlm_lookup_lockres(dlm, DLM_RECOVERY_LOCK_NAME,
2665                                          DLM_RECOVERY_LOCK_NAME_LEN);
2666                 if (res) {
2667                         dlm_print_one_lock_resource(res);
2668                         dlm_lockres_put(res);
2669                 } else {
2670                         mlog(ML_ERROR, "recovery lock not found\n");
2671                 }
2672                 BUG();
2673         }
2674 
2675         return status;
2676 }
2677 
2678 static int dlm_send_begin_reco_message(struct dlm_ctxt *dlm, u8 dead_node)
2679 {
2680         struct dlm_begin_reco br;
2681         int ret = 0;
2682         struct dlm_node_iter iter;
2683         int nodenum;
2684         int status;
2685 
2686         mlog(0, "%s: dead node is %u\n", dlm->name, dead_node);
2687 
2688         spin_lock(&dlm->spinlock);
2689         dlm_node_iter_init(dlm->domain_map, &iter);
2690         spin_unlock(&dlm->spinlock);
2691 
2692         clear_bit(dead_node, iter.node_map);
2693 
2694         memset(&br, 0, sizeof(br));
2695         br.node_idx = dlm->node_num;
2696         br.dead_node = dead_node;
2697 
2698         while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
2699                 ret = 0;
2700                 if (nodenum == dead_node) {
2701                         mlog(0, "not sending begin reco to dead node "
2702                                   "%u\n", dead_node);
2703                         continue;
2704                 }
2705                 if (nodenum == dlm->node_num) {
2706                         mlog(0, "not sending begin reco to self\n");
2707                         continue;
2708                 }
2709 retry:
2710                 ret = -EINVAL;
2711                 mlog(0, "attempting to send begin reco msg to %d\n",
2712                           nodenum);
2713                 ret = o2net_send_message(DLM_BEGIN_RECO_MSG, dlm->key,
2714                                          &br, sizeof(br), nodenum, &status);
2715                 /* negative status is handled ok by caller here */
2716                 if (ret >= 0)
2717                         ret = status;
2718                 if (dlm_is_host_down(ret)) {
2719                         /* node is down.  not involved in recovery
2720                          * so just keep going */
2721                         mlog(ML_NOTICE, "%s: node %u was down when sending "
2722                              "begin reco msg (%d)\n", dlm->name, nodenum, ret);
2723                         ret = 0;
2724                 }
2725 
2726                 /*
2727                  * Prior to commit aad1b15310b9bcd59fa81ab8f2b1513b59553ea8,
2728                  * dlm_begin_reco_handler() returned EAGAIN and not -EAGAIN.
2729                  * We are handling both for compatibility reasons.
2730                  */
2731                 if (ret == -EAGAIN || ret == EAGAIN) {
2732                         mlog(0, "%s: trying to start recovery of node "
2733                              "%u, but node %u is waiting for last recovery "
2734                              "to complete, backoff for a bit\n", dlm->name,
2735                              dead_node, nodenum);
2736                         msleep(100);
2737                         goto retry;
2738                 }
2739                 if (ret < 0) {
2740                         struct dlm_lock_resource *res;
2741 
2742                         /* this is now a serious problem, possibly ENOMEM
2743                          * in the network stack.  must retry */
2744                         mlog_errno(ret);
2745                         mlog(ML_ERROR, "begin reco of dlm %s to node %u "
2746                              "returned %d\n", dlm->name, nodenum, ret);
2747                         res = dlm_lookup_lockres(dlm, DLM_RECOVERY_LOCK_NAME,
2748                                                  DLM_RECOVERY_LOCK_NAME_LEN);
2749                         if (res) {
2750                                 dlm_print_one_lock_resource(res);
2751                                 dlm_lockres_put(res);
2752                         } else {
2753                                 mlog(ML_ERROR, "recovery lock not found\n");
2754                         }
2755                         /* sleep for a bit in hopes that we can avoid
2756                          * another ENOMEM */
2757                         msleep(100);
2758                         goto retry;
2759                 }
2760         }
2761 
2762         return ret;
2763 }
2764 
2765 int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data,
2766                            void **ret_data)
2767 {
2768         struct dlm_ctxt *dlm = data;
2769         struct dlm_begin_reco *br = (struct dlm_begin_reco *)msg->buf;
2770 
2771         /* ok to return 0, domain has gone away */
2772         if (!dlm_grab(dlm))
2773                 return 0;
2774 
2775         spin_lock(&dlm->spinlock);
2776         if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
2777                 mlog(0, "%s: node %u wants to recover node %u (%u:%u) "
2778                      "but this node is in finalize state, waiting on finalize2\n",
2779                      dlm->name, br->node_idx, br->dead_node,
2780                      dlm->reco.dead_node, dlm->reco.new_master);
2781                 spin_unlock(&dlm->spinlock);
2782                 dlm_put(dlm);
2783                 return -EAGAIN;
2784         }
2785         spin_unlock(&dlm->spinlock);
2786 
2787         mlog(0, "%s: node %u wants to recover node %u (%u:%u)\n",
2788              dlm->name, br->node_idx, br->dead_node,
2789              dlm->reco.dead_node, dlm->reco.new_master);
2790 
2791         dlm_fire_domain_eviction_callbacks(dlm, br->dead_node);
2792 
2793         spin_lock(&dlm->spinlock);
2794         if (dlm->reco.new_master != O2NM_INVALID_NODE_NUM) {
2795                 if (test_bit(dlm->reco.new_master, dlm->recovery_map)) {
2796                         mlog(0, "%s: new_master %u died, changing "
2797                              "to %u\n", dlm->name, dlm->reco.new_master,
2798                              br->node_idx);
2799                 } else {
2800                         mlog(0, "%s: new_master %u NOT DEAD, changing "
2801                              "to %u\n", dlm->name, dlm->reco.new_master,
2802                              br->node_idx);
2803                         /* may not have seen the new master as dead yet */
2804                 }
2805         }
2806         if (dlm->reco.dead_node != O2NM_INVALID_NODE_NUM) {
2807                 mlog(ML_NOTICE, "%s: dead_node previously set to %u, "
2808                      "node %u changing it to %u\n", dlm->name,
2809                      dlm->reco.dead_node, br->node_idx, br->dead_node);
2810         }
2811         dlm_set_reco_master(dlm, br->node_idx);
2812         dlm_set_reco_dead_node(dlm, br->dead_node);
2813         if (!test_bit(br->dead_node, dlm->recovery_map)) {
2814                 mlog(0, "recovery master %u sees %u as dead, but this "
2815                      "node has not yet.  marking %u as dead\n",
2816                      br->node_idx, br->dead_node, br->dead_node);
2817                 if (!test_bit(br->dead_node, dlm->domain_map) ||
2818                     !test_bit(br->dead_node, dlm->live_nodes_map))
2819                         mlog(0, "%u not in domain/live_nodes map "
2820                              "so setting it in reco map manually\n",
2821                              br->dead_node);
2822                 /* force the recovery cleanup in __dlm_hb_node_down
2823                  * both of these will be cleared in a moment */
2824                 set_bit(br->dead_node, dlm->domain_map);
2825                 set_bit(br->dead_node, dlm->live_nodes_map);
2826                 __dlm_hb_node_down(dlm, br->dead_node);
2827         }
2828         spin_unlock(&dlm->spinlock);
2829 
2830         dlm_kick_recovery_thread(dlm);
2831 
2832         mlog(0, "%s: recovery started by node %u, for %u (%u:%u)\n",
2833              dlm->name, br->node_idx, br->dead_node,
2834              dlm->reco.dead_node, dlm->reco.new_master);
2835 
2836         dlm_put(dlm);
2837         return 0;
2838 }
2839 
2840 #define DLM_FINALIZE_STAGE2  0x01
2841 static int dlm_send_finalize_reco_message(struct dlm_ctxt *dlm)
2842 {
2843         int ret = 0;
2844         struct dlm_finalize_reco fr;
2845         struct dlm_node_iter iter;
2846         int nodenum;
2847         int status;
2848         int stage = 1;
2849 
2850         mlog(0, "finishing recovery for node %s:%u, "
2851              "stage %d\n", dlm->name, dlm->reco.dead_node, stage);
2852 
2853         spin_lock(&dlm->spinlock);
2854         dlm_node_iter_init(dlm->domain_map, &iter);
2855         spin_unlock(&dlm->spinlock);
2856 
2857 stage2:
2858         memset(&fr, 0, sizeof(fr));
2859         fr.node_idx = dlm->node_num;
2860         fr.dead_node = dlm->reco.dead_node;
2861         if (stage == 2)
2862                 fr.flags |= DLM_FINALIZE_STAGE2;
2863 
2864         while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
2865                 if (nodenum == dlm->node_num)
2866                         continue;
2867                 ret = o2net_send_message(DLM_FINALIZE_RECO_MSG, dlm->key,
2868                                          &fr, sizeof(fr), nodenum, &status);
2869                 if (ret >= 0)
2870                         ret = status;
2871                 if (ret < 0) {
2872                         mlog(ML_ERROR, "Error %d when sending message %u (key "
2873                              "0x%x) to node %u\n", ret, DLM_FINALIZE_RECO_MSG,
2874                              dlm->key, nodenum);
2875                         if (dlm_is_host_down(ret)) {
2876                                 /* this has no effect on this recovery
2877                                  * session, so set the status to zero to
2878                                  * finish out the last recovery */
2879                                 mlog(ML_ERROR, "node %u went down after this "
2880                                      "node finished recovery.\n", nodenum);
2881                                 ret = 0;
2882                                 continue;
2883                         }
2884                         break;
2885                 }
2886         }
2887         if (stage == 1) {
2888                 /* reset the node_iter back to the top and send finalize2 */
2889                 iter.curnode = -1;
2890                 stage = 2;
2891                 goto stage2;
2892         }
2893 
2894         return ret;
2895 }
2896 
2897 int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data,
2898                               void **ret_data)
2899 {
2900         struct dlm_ctxt *dlm = data;
2901         struct dlm_finalize_reco *fr = (struct dlm_finalize_reco *)msg->buf;
2902         int stage = 1;
2903 
2904         /* ok to return 0, domain has gone away */
2905         if (!dlm_grab(dlm))
2906                 return 0;
2907 
2908         if (fr->flags & DLM_FINALIZE_STAGE2)
2909                 stage = 2;
2910 
2911         mlog(0, "%s: node %u finalizing recovery stage%d of "
2912              "node %u (%u:%u)\n", dlm->name, fr->node_idx, stage,
2913              fr->dead_node, dlm->reco.dead_node, dlm->reco.new_master);
2914 
2915         spin_lock(&dlm->spinlock);
2916 
2917         if (dlm->reco.new_master != fr->node_idx) {
2918                 mlog(ML_ERROR, "node %u sent recovery finalize msg, but node "
2919                      "%u is supposed to be the new master, dead=%u\n",
2920                      fr->node_idx, dlm->reco.new_master, fr->dead_node);
2921                 BUG();
2922         }
2923         if (dlm->reco.dead_node != fr->dead_node) {
2924                 mlog(ML_ERROR, "node %u sent recovery finalize msg for dead "
2925                      "node %u, but node %u is supposed to be dead\n",
2926                      fr->node_idx, fr->dead_node, dlm->reco.dead_node);
2927                 BUG();
2928         }
2929 
2930         switch (stage) {
2931                 case 1:
2932                         dlm_finish_local_lockres_recovery(dlm, fr->dead_node, fr->node_idx);
2933                         if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
2934                                 mlog(ML_ERROR, "%s: received finalize1 from "
2935                                      "new master %u for dead node %u, but "
2936                                      "this node has already received it!\n",
2937                                      dlm->name, fr->node_idx, fr->dead_node);
2938                                 dlm_print_reco_node_status(dlm);
2939                                 BUG();
2940                         }
2941                         dlm->reco.state |= DLM_RECO_STATE_FINALIZE;
2942                         spin_unlock(&dlm->spinlock);
2943                         break;
2944                 case 2:
2945                         if (!(dlm->reco.state & DLM_RECO_STATE_FINALIZE)) {
2946                                 mlog(ML_ERROR, "%s: received finalize2 from "
2947                                      "new master %u for dead node %u, but "
2948                                      "this node did not have finalize1!\n",
2949                                      dlm->name, fr->node_idx, fr->dead_node);
2950                                 dlm_print_reco_node_status(dlm);
2951                                 BUG();
2952                         }
2953                         dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE;
2954                         __dlm_reset_recovery(dlm);
2955                         spin_unlock(&dlm->spinlock);
2956                         dlm_kick_recovery_thread(dlm);
2957                         break;
2958         }
2959 
2960         mlog(0, "%s: recovery done, reco master was %u, dead now %u, master now %u\n",
2961              dlm->name, fr->node_idx, dlm->reco.dead_node, dlm->reco.new_master);
2962 
2963         dlm_put(dlm);
2964         return 0;
2965 }
/* [<][>][^][v][top][bottom][index][help] */
root/fs/ocfs2/dlm/dlmrecovery.c

DEFINITIONS
