git.oblomov.eu Git - linux-2.6/blob - fs/ocfs2/dlm/dlmmaster.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * dlmmod.c
   5  *
   6  * standalone DLM module
   7  *
   8  * Copyright (C) 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  *
  25  */
  26
  27
  28 #include <linux/module.h>
  29 #include <linux/fs.h>
  30 #include <linux/types.h>
  31 #include <linux/slab.h>
  32 #include <linux/highmem.h>
  33 #include <linux/utsname.h>
  34 #include <linux/init.h>
  35 #include <linux/sysctl.h>
  36 #include <linux/random.h>
  37 #include <linux/blkdev.h>
  38 #include <linux/socket.h>
  39 #include <linux/inet.h>
  40 #include <linux/spinlock.h>
  41 #include <linux/delay.h>
  42
  43
  44 #include "cluster/heartbeat.h"
  45 #include "cluster/nodemanager.h"
  46 #include "cluster/tcp.h"
  47
  48 #include "dlmapi.h"
  49 #include "dlmcommon.h"
  50 #include "dlmdomain.h"
  51
  52 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_MASTER)
  53 #include "cluster/masklog.h"
  54
  55 enum dlm_mle_type {
  56         DLM_MLE_BLOCK,
  57         DLM_MLE_MASTER,
  58         DLM_MLE_MIGRATION
  59 };
  60
  61 struct dlm_lock_name
  62 {
  63         u8 len;
  64         u8 name[DLM_LOCKID_NAME_MAX];
  65 };
  66
  67 struct dlm_master_list_entry
  68 {
  69         struct list_head list;
  70         struct list_head hb_events;
  71         struct dlm_ctxt *dlm;
  72         spinlock_t spinlock;
  73         wait_queue_head_t wq;
  74         atomic_t woken;
  75         struct kref mle_refs;
  76         int inuse;
  77         unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
  78         unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
  79         unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
  80         unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
  81         u8 master;
  82         u8 new_master;
  83         enum dlm_mle_type type;
  84         struct o2hb_callback_func mle_hb_up;
  85         struct o2hb_callback_func mle_hb_down;
  86         union {
  87                 struct dlm_lock_resource *res;
  88                 struct dlm_lock_name name;
  89         } u;
  90 };
  91
  92 static void dlm_mle_node_down(struct dlm_ctxt *dlm,
  93                               struct dlm_master_list_entry *mle,
  94                               struct o2nm_node *node,
  95                               int idx);
  96 static void dlm_mle_node_up(struct dlm_ctxt *dlm,
  97                             struct dlm_master_list_entry *mle,
  98                             struct o2nm_node *node,
  99                             int idx);
 100
 101 static void dlm_assert_master_worker(struct dlm_work_item *item, void *data);
 102 static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname,
 103                                 unsigned int namelen, void *nodemap,
 104                                 u32 flags);
 105
 106 static inline int dlm_mle_equal(struct dlm_ctxt *dlm,
 107                                 struct dlm_master_list_entry *mle,
 108                                 const char *name,
 109                                 unsigned int namelen)
 110 {
 111         struct dlm_lock_resource *res;
 112
 113         if (dlm != mle->dlm)
 114                 return 0;
 115
 116         if (mle->type == DLM_MLE_BLOCK ||
 117             mle->type == DLM_MLE_MIGRATION) {
 118                 if (namelen != mle->u.name.len ||
 119                     memcmp(name, mle->u.name.name, namelen)!=0)
 120                         return 0;
 121         } else {
 122                 res = mle->u.res;
 123                 if (namelen != res->lockname.len ||
 124                     memcmp(res->lockname.name, name, namelen) != 0)
 125                         return 0;
 126         }
 127         return 1;
 128 }
 129
 130 #define dlm_print_nodemap(m)  _dlm_print_nodemap(m,#m)
 131 static void _dlm_print_nodemap(unsigned long *map, const char *mapname)
 132 {
 133         int i;
 134         printk("%s=[ ", mapname);
 135         for (i=0; i<O2NM_MAX_NODES; i++)
 136                 if (test_bit(i, map))
 137                         printk("%d ", i);
 138         printk("]");
 139 }
 140
 141 static void dlm_print_one_mle(struct dlm_master_list_entry *mle)
 142 {
 143         int refs;
 144         char *type;
 145         char attached;
 146         u8 master;
 147         unsigned int namelen;
 148         const char *name;
 149         struct kref *k;
 150         unsigned long *maybe = mle->maybe_map,
 151                       *vote = mle->vote_map,
 152                       *resp = mle->response_map,
 153                       *node = mle->node_map;
 154
 155         k = &mle->mle_refs;
 156         if (mle->type == DLM_MLE_BLOCK)
 157                 type = "BLK";
 158         else if (mle->type == DLM_MLE_MASTER)
 159                 type = "MAS";
 160         else
 161                 type = "MIG";
 162         refs = atomic_read(&k->refcount);
 163         master = mle->master;
 164         attached = (list_empty(&mle->hb_events) ? 'N' : 'Y');
 165
 166         if (mle->type != DLM_MLE_MASTER) {
 167                 namelen = mle->u.name.len;
 168                 name = mle->u.name.name;
 169         } else {
 170                 namelen = mle->u.res->lockname.len;
 171                 name = mle->u.res->lockname.name;
 172         }
 173
 174         mlog(ML_NOTICE, "%.*s: %3s refs=%3d mas=%3u new=%3u evt=%c inuse=%d ",
 175                   namelen, name, type, refs, master, mle->new_master, attached,
 176                   mle->inuse);
 177         dlm_print_nodemap(maybe);
 178         printk(", ");
 179         dlm_print_nodemap(vote);
 180         printk(", ");
 181         dlm_print_nodemap(resp);
 182         printk(", ");
 183         dlm_print_nodemap(node);
 184         printk(", ");
 185         printk("\n");
 186 }
 187
 188 #if 0
 189 /* Code here is included but defined out as it aids debugging */
 190
 191 static void dlm_dump_mles(struct dlm_ctxt *dlm)
 192 {
 193         struct dlm_master_list_entry *mle;
 194         struct list_head *iter;
 195
 196         mlog(ML_NOTICE, "dumping all mles for domain %s:\n", dlm->name);
 197         spin_lock(&dlm->master_lock);
 198         list_for_each(iter, &dlm->master_list) {
 199                 mle = list_entry(iter, struct dlm_master_list_entry, list);
 200                 dlm_print_one_mle(mle);
 201         }
 202         spin_unlock(&dlm->master_lock);
 203 }
 204
 205 int dlm_dump_all_mles(const char __user *data, unsigned int len)
 206 {
 207         struct list_head *iter;
 208         struct dlm_ctxt *dlm;
 209
 210         spin_lock(&dlm_domain_lock);
 211         list_for_each(iter, &dlm_domains) {
 212                 dlm = list_entry (iter, struct dlm_ctxt, list);
 213                 mlog(ML_NOTICE, "found dlm: %p, name=%s\n", dlm, dlm->name);
 214                 dlm_dump_mles(dlm);
 215         }
 216         spin_unlock(&dlm_domain_lock);
 217         return len;
 218 }
 219 EXPORT_SYMBOL_GPL(dlm_dump_all_mles);
 220
 221 #endif  /*  0  */
 222
 223
 224 static kmem_cache_t *dlm_mle_cache = NULL;
 225
 226
 227 static void dlm_mle_release(struct kref *kref);
 228 static void dlm_init_mle(struct dlm_master_list_entry *mle,
 229                         enum dlm_mle_type type,
 230                         struct dlm_ctxt *dlm,
 231                         struct dlm_lock_resource *res,
 232                         const char *name,
 233                         unsigned int namelen);
 234 static void dlm_put_mle(struct dlm_master_list_entry *mle);
 235 static void __dlm_put_mle(struct dlm_master_list_entry *mle);
 236 static int dlm_find_mle(struct dlm_ctxt *dlm,
 237                         struct dlm_master_list_entry **mle,
 238                         char *name, unsigned int namelen);
 239
 240 static int dlm_do_master_request(struct dlm_master_list_entry *mle, int to);
 241
 242
 243 static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm,
 244                                      struct dlm_lock_resource *res,
 245                                      struct dlm_master_list_entry *mle,
 246                                      int *blocked);
 247 static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm,
 248                                     struct dlm_lock_resource *res,
 249                                     struct dlm_master_list_entry *mle,
 250                                     int blocked);
 251 static int dlm_add_migration_mle(struct dlm_ctxt *dlm,
 252                                  struct dlm_lock_resource *res,
 253                                  struct dlm_master_list_entry *mle,
 254                                  struct dlm_master_list_entry **oldmle,
 255                                  const char *name, unsigned int namelen,
 256                                  u8 new_master, u8 master);
 257
 258 static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
 259                                     struct dlm_lock_resource *res);
 260 static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
 261                                       struct dlm_lock_resource *res);
 262 static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm,
 263                                        struct dlm_lock_resource *res,
 264                                        u8 target);
 265 static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm,
 266                                        struct dlm_lock_resource *res);
 267
 268
 269 int dlm_is_host_down(int errno)
 270 {
 271         switch (errno) {
 272                 case -EBADF:
 273                 case -ECONNREFUSED:
 274                 case -ENOTCONN:
 275                 case -ECONNRESET:
 276                 case -EPIPE:
 277                 case -EHOSTDOWN:
 278                 case -EHOSTUNREACH:
 279                 case -ETIMEDOUT:
 280                 case -ECONNABORTED:
 281                 case -ENETDOWN:
 282                 case -ENETUNREACH:
 283                 case -ENETRESET:
 284                 case -ESHUTDOWN:
 285                 case -ENOPROTOOPT:
 286                 case -EINVAL:   /* if returned from our tcp code,
 287                                    this means there is no socket */
 288                         return 1;
 289         }
 290         return 0;
 291 }
 292
 293
 294 /*
 295  * MASTER LIST FUNCTIONS
 296  */
 297
 298
 299 /*
 300  * regarding master list entries and heartbeat callbacks:
 301  *
 302  * in order to avoid sleeping and allocation that occurs in
 303  * heartbeat, master list entries are simply attached to the
 304  * dlm's established heartbeat callbacks.  the mle is attached
 305  * when it is created, and since the dlm->spinlock is held at
 306  * that time, any heartbeat event will be properly discovered
 307  * by the mle.  the mle needs to be detached from the
 308  * dlm->mle_hb_events list as soon as heartbeat events are no
 309  * longer useful to the mle, and before the mle is freed.
 310  *
 311  * as a general rule, heartbeat events are no longer needed by
 312  * the mle once an "answer" regarding the lock master has been
 313  * received.
 314  */
 315 static inline void __dlm_mle_attach_hb_events(struct dlm_ctxt *dlm,
 316                                               struct dlm_master_list_entry *mle)
 317 {
 318         assert_spin_locked(&dlm->spinlock);
 319
 320         list_add_tail(&mle->hb_events, &dlm->mle_hb_events);
 321 }
 322
 323
 324 static inline void __dlm_mle_detach_hb_events(struct dlm_ctxt *dlm,
 325                                               struct dlm_master_list_entry *mle)
 326 {
 327         if (!list_empty(&mle->hb_events))
 328                 list_del_init(&mle->hb_events);
 329 }
 330
 331
 332 static inline void dlm_mle_detach_hb_events(struct dlm_ctxt *dlm,
 333                                             struct dlm_master_list_entry *mle)
 334 {
 335         spin_lock(&dlm->spinlock);
 336         __dlm_mle_detach_hb_events(dlm, mle);
 337         spin_unlock(&dlm->spinlock);
 338 }
 339
 340 static void dlm_get_mle_inuse(struct dlm_master_list_entry *mle)
 341 {
 342         struct dlm_ctxt *dlm;
 343         dlm = mle->dlm;
 344
 345         assert_spin_locked(&dlm->spinlock);
 346         assert_spin_locked(&dlm->master_lock);
 347         mle->inuse++;
 348         kref_get(&mle->mle_refs);
 349 }
 350
 351 static void dlm_put_mle_inuse(struct dlm_master_list_entry *mle)
 352 {
 353         struct dlm_ctxt *dlm;
 354         dlm = mle->dlm;
 355
 356         spin_lock(&dlm->spinlock);
 357         spin_lock(&dlm->master_lock);
 358         mle->inuse--;
 359         __dlm_put_mle(mle);
 360         spin_unlock(&dlm->master_lock);
 361         spin_unlock(&dlm->spinlock);
 362
 363 }
 364
 365 /* remove from list and free */
 366 static void __dlm_put_mle(struct dlm_master_list_entry *mle)
 367 {
 368         struct dlm_ctxt *dlm;
 369         dlm = mle->dlm;
 370
 371         assert_spin_locked(&dlm->spinlock);
 372         assert_spin_locked(&dlm->master_lock);
 373         if (!atomic_read(&mle->mle_refs.refcount)) {
 374                 /* this may or may not crash, but who cares.
 375                  * it's a BUG. */
 376                 mlog(ML_ERROR, "bad mle: %p\n", mle);
 377                 dlm_print_one_mle(mle);
 378                 BUG();
 379         } else
 380                 kref_put(&mle->mle_refs, dlm_mle_release);
 381 }
 382
 383
 384 /* must not have any spinlocks coming in */
 385 static void dlm_put_mle(struct dlm_master_list_entry *mle)
 386 {
 387         struct dlm_ctxt *dlm;
 388         dlm = mle->dlm;
 389
 390         spin_lock(&dlm->spinlock);
 391         spin_lock(&dlm->master_lock);
 392         __dlm_put_mle(mle);
 393         spin_unlock(&dlm->master_lock);
 394         spin_unlock(&dlm->spinlock);
 395 }
 396
 397 static inline void dlm_get_mle(struct dlm_master_list_entry *mle)
 398 {
 399         kref_get(&mle->mle_refs);
 400 }
 401
 402 static void dlm_init_mle(struct dlm_master_list_entry *mle,
 403                         enum dlm_mle_type type,
 404                         struct dlm_ctxt *dlm,
 405                         struct dlm_lock_resource *res,
 406                         const char *name,
 407                         unsigned int namelen)
 408 {
 409         assert_spin_locked(&dlm->spinlock);
 410
 411         mle->dlm = dlm;
 412         mle->type = type;
 413         INIT_LIST_HEAD(&mle->list);
 414         INIT_LIST_HEAD(&mle->hb_events);
 415         memset(mle->maybe_map, 0, sizeof(mle->maybe_map));
 416         spin_lock_init(&mle->spinlock);
 417         init_waitqueue_head(&mle->wq);
 418         atomic_set(&mle->woken, 0);
 419         kref_init(&mle->mle_refs);
 420         memset(mle->response_map, 0, sizeof(mle->response_map));
 421         mle->master = O2NM_MAX_NODES;
 422         mle->new_master = O2NM_MAX_NODES;
 423         mle->inuse = 0;
 424
 425         if (mle->type == DLM_MLE_MASTER) {
 426                 BUG_ON(!res);
 427                 mle->u.res = res;
 428         } else if (mle->type == DLM_MLE_BLOCK) {
 429                 BUG_ON(!name);
 430                 memcpy(mle->u.name.name, name, namelen);
 431                 mle->u.name.len = namelen;
 432         } else /* DLM_MLE_MIGRATION */ {
 433                 BUG_ON(!name);
 434                 memcpy(mle->u.name.name, name, namelen);
 435                 mle->u.name.len = namelen;
 436         }
 437
 438         /* copy off the node_map and register hb callbacks on our copy */
 439         memcpy(mle->node_map, dlm->domain_map, sizeof(mle->node_map));
 440         memcpy(mle->vote_map, dlm->domain_map, sizeof(mle->vote_map));
 441         clear_bit(dlm->node_num, mle->vote_map);
 442         clear_bit(dlm->node_num, mle->node_map);
 443
 444         /* attach the mle to the domain node up/down events */
 445         __dlm_mle_attach_hb_events(dlm, mle);
 446 }
 447
 448
 449 /* returns 1 if found, 0 if not */
 450 static int dlm_find_mle(struct dlm_ctxt *dlm,
 451                         struct dlm_master_list_entry **mle,
 452                         char *name, unsigned int namelen)
 453 {
 454         struct dlm_master_list_entry *tmpmle;
 455         struct list_head *iter;
 456
 457         assert_spin_locked(&dlm->master_lock);
 458
 459         list_for_each(iter, &dlm->master_list) {
 460                 tmpmle = list_entry(iter, struct dlm_master_list_entry, list);
 461                 if (!dlm_mle_equal(dlm, tmpmle, name, namelen))
 462                         continue;
 463                 dlm_get_mle(tmpmle);
 464                 *mle = tmpmle;
 465                 return 1;
 466         }
 467         return 0;
 468 }
 469
 470 void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up)
 471 {
 472         struct dlm_master_list_entry *mle;
 473         struct list_head *iter;
 474
 475         assert_spin_locked(&dlm->spinlock);
 476
 477         list_for_each(iter, &dlm->mle_hb_events) {
 478                 mle = list_entry(iter, struct dlm_master_list_entry,
 479                                  hb_events);
 480                 if (node_up)
 481                         dlm_mle_node_up(dlm, mle, NULL, idx);
 482                 else
 483                         dlm_mle_node_down(dlm, mle, NULL, idx);
 484         }
 485 }
 486
 487 static void dlm_mle_node_down(struct dlm_ctxt *dlm,
 488                               struct dlm_master_list_entry *mle,
 489                               struct o2nm_node *node, int idx)
 490 {
 491         spin_lock(&mle->spinlock);
 492
 493         if (!test_bit(idx, mle->node_map))
 494                 mlog(0, "node %u already removed from nodemap!\n", idx);
 495         else
 496                 clear_bit(idx, mle->node_map);
 497
 498         spin_unlock(&mle->spinlock);
 499 }
 500
 501 static void dlm_mle_node_up(struct dlm_ctxt *dlm,
 502                             struct dlm_master_list_entry *mle,
 503                             struct o2nm_node *node, int idx)
 504 {
 505         spin_lock(&mle->spinlock);
 506
 507         if (test_bit(idx, mle->node_map))
 508                 mlog(0, "node %u already in node map!\n", idx);
 509         else
 510                 set_bit(idx, mle->node_map);
 511
 512         spin_unlock(&mle->spinlock);
 513 }
 514
 515
 516 int dlm_init_mle_cache(void)
 517 {
 518         dlm_mle_cache = kmem_cache_create("dlm_mle_cache",
 519                                           sizeof(struct dlm_master_list_entry),
 520                                           0, SLAB_HWCACHE_ALIGN,
 521                                           NULL, NULL);
 522         if (dlm_mle_cache == NULL)
 523                 return -ENOMEM;
 524         return 0;
 525 }
 526
 527 void dlm_destroy_mle_cache(void)
 528 {
 529         if (dlm_mle_cache)
 530                 kmem_cache_destroy(dlm_mle_cache);
 531 }
 532
 533 static void dlm_mle_release(struct kref *kref)
 534 {
 535         struct dlm_master_list_entry *mle;
 536         struct dlm_ctxt *dlm;
 537
 538         mlog_entry_void();
 539
 540         mle = container_of(kref, struct dlm_master_list_entry, mle_refs);
 541         dlm = mle->dlm;
 542
 543         if (mle->type != DLM_MLE_MASTER) {
 544                 mlog(0, "calling mle_release for %.*s, type %d\n",
 545                      mle->u.name.len, mle->u.name.name, mle->type);
 546         } else {
 547                 mlog(0, "calling mle_release for %.*s, type %d\n",
 548                      mle->u.res->lockname.len,
 549                      mle->u.res->lockname.name, mle->type);
 550         }
 551         assert_spin_locked(&dlm->spinlock);
 552         assert_spin_locked(&dlm->master_lock);
 553
 554         /* remove from list if not already */
 555         if (!list_empty(&mle->list))
 556                 list_del_init(&mle->list);
 557
 558         /* detach the mle from the domain node up/down events */
 559         __dlm_mle_detach_hb_events(dlm, mle);
 560
 561         /* NOTE: kfree under spinlock here.
 562          * if this is bad, we can move this to a freelist. */
 563         kmem_cache_free(dlm_mle_cache, mle);
 564 }
 565
 566
 567 /*
 568  * LOCK RESOURCE FUNCTIONS
 569  */
 570
 571 static void dlm_set_lockres_owner(struct dlm_ctxt *dlm,
 572                                   struct dlm_lock_resource *res,
 573                                   u8 owner)
 574 {
 575         assert_spin_locked(&res->spinlock);
 576
 577         mlog_entry("%.*s, %u\n", res->lockname.len, res->lockname.name, owner);
 578
 579         if (owner == dlm->node_num)
 580                 atomic_inc(&dlm->local_resources);
 581         else if (owner == DLM_LOCK_RES_OWNER_UNKNOWN)
 582                 atomic_inc(&dlm->unknown_resources);
 583         else
 584                 atomic_inc(&dlm->remote_resources);
 585
 586         res->owner = owner;
 587 }
 588
 589 void dlm_change_lockres_owner(struct dlm_ctxt *dlm,
 590                               struct dlm_lock_resource *res, u8 owner)
 591 {
 592         assert_spin_locked(&res->spinlock);
 593
 594         if (owner == res->owner)
 595                 return;
 596
 597         if (res->owner == dlm->node_num)
 598                 atomic_dec(&dlm->local_resources);
 599         else if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN)
 600                 atomic_dec(&dlm->unknown_resources);
 601         else
 602                 atomic_dec(&dlm->remote_resources);
 603
 604         dlm_set_lockres_owner(dlm, res, owner);
 605 }
 606
 607
 608 static void dlm_lockres_release(struct kref *kref)
 609 {
 610         struct dlm_lock_resource *res;
 611
 612         res = container_of(kref, struct dlm_lock_resource, refs);
 613
 614         /* This should not happen -- all lockres' have a name
 615          * associated with them at init time. */
 616         BUG_ON(!res->lockname.name);
 617
 618         mlog(0, "destroying lockres %.*s\n", res->lockname.len,
 619              res->lockname.name);
 620
 621         if (!hlist_unhashed(&res->hash_node) ||
 622             !list_empty(&res->granted) ||
 623             !list_empty(&res->converting) ||
 624             !list_empty(&res->blocked) ||
 625             !list_empty(&res->dirty) ||
 626             !list_empty(&res->recovering) ||
 627             !list_empty(&res->purge)) {
 628                 mlog(ML_ERROR,
 629                      "Going to BUG for resource %.*s."
 630                      "  We're on a list! [%c%c%c%c%c%c%c]\n",
 631                      res->lockname.len, res->lockname.name,
 632                      !hlist_unhashed(&res->hash_node) ? 'H' : ' ',
 633                      !list_empty(&res->granted) ? 'G' : ' ',
 634                      !list_empty(&res->converting) ? 'C' : ' ',
 635                      !list_empty(&res->blocked) ? 'B' : ' ',
 636                      !list_empty(&res->dirty) ? 'D' : ' ',
 637                      !list_empty(&res->recovering) ? 'R' : ' ',
 638                      !list_empty(&res->purge) ? 'P' : ' ');
 639
 640                 dlm_print_one_lock_resource(res);
 641         }
 642
 643         /* By the time we're ready to blow this guy away, we shouldn't
 644          * be on any lists. */
 645         BUG_ON(!hlist_unhashed(&res->hash_node));
 646         BUG_ON(!list_empty(&res->granted));
 647         BUG_ON(!list_empty(&res->converting));
 648         BUG_ON(!list_empty(&res->blocked));
 649         BUG_ON(!list_empty(&res->dirty));
 650         BUG_ON(!list_empty(&res->recovering));
 651         BUG_ON(!list_empty(&res->purge));
 652
 653         kfree(res->lockname.name);
 654
 655         kfree(res);
 656 }
 657
 658 void dlm_lockres_put(struct dlm_lock_resource *res)
 659 {
 660         kref_put(&res->refs, dlm_lockres_release);
 661 }
 662
 663 static void dlm_init_lockres(struct dlm_ctxt *dlm,
 664                              struct dlm_lock_resource *res,
 665                              const char *name, unsigned int namelen)
 666 {
 667         char *qname;
 668
 669         /* If we memset here, we lose our reference to the kmalloc'd
 670          * res->lockname.name, so be sure to init every field
 671          * correctly! */
 672
 673         qname = (char *) res->lockname.name;
 674         memcpy(qname, name, namelen);
 675
 676         res->lockname.len = namelen;
 677         res->lockname.hash = dlm_lockid_hash(name, namelen);
 678
 679         init_waitqueue_head(&res->wq);
 680         spin_lock_init(&res->spinlock);
 681         INIT_HLIST_NODE(&res->hash_node);
 682         INIT_LIST_HEAD(&res->granted);
 683         INIT_LIST_HEAD(&res->converting);
 684         INIT_LIST_HEAD(&res->blocked);
 685         INIT_LIST_HEAD(&res->dirty);
 686         INIT_LIST_HEAD(&res->recovering);
 687         INIT_LIST_HEAD(&res->purge);
 688         atomic_set(&res->asts_reserved, 0);
 689         res->migration_pending = 0;
 690
 691         kref_init(&res->refs);
 692
 693         /* just for consistency */
 694         spin_lock(&res->spinlock);
 695         dlm_set_lockres_owner(dlm, res, DLM_LOCK_RES_OWNER_UNKNOWN);
 696         spin_unlock(&res->spinlock);
 697
 698         res->state = DLM_LOCK_RES_IN_PROGRESS;
 699
 700         res->last_used = 0;
 701
 702         memset(res->lvb, 0, DLM_LVB_LEN);
 703 }
 704
 705 struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
 706                                    const char *name,
 707                                    unsigned int namelen)
 708 {
 709         struct dlm_lock_resource *res;
 710
 711         res = kmalloc(sizeof(struct dlm_lock_resource), GFP_NOFS);
 712         if (!res)
 713                 return NULL;
 714
 715         res->lockname.name = kmalloc(namelen, GFP_NOFS);
 716         if (!res->lockname.name) {
 717                 kfree(res);
 718                 return NULL;
 719         }
 720
 721         dlm_init_lockres(dlm, res, name, namelen);
 722         return res;
 723 }
 724
 725 /*
 726  * lookup a lock resource by name.
 727  * may already exist in the hashtable.
 728  * lockid is null terminated
 729  *
 730  * if not, allocate enough for the lockres and for
 731  * the temporary structure used in doing the mastering.
 732  *
 733  * also, do a lookup in the dlm->master_list to see
 734  * if another node has begun mastering the same lock.
 735  * if so, there should be a block entry in there
 736  * for this name, and we should *not* attempt to master
 737  * the lock here.   need to wait around for that node
 738  * to assert_master (or die).
 739  *
 740  */
 741 struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
 742                                           const char *lockid,
 743                                           int namelen,
 744                                           int flags)
 745 {
 746         struct dlm_lock_resource *tmpres=NULL, *res=NULL;
 747         struct dlm_master_list_entry *mle = NULL;
 748         struct dlm_master_list_entry *alloc_mle = NULL;
 749         int blocked = 0;
 750         int ret, nodenum;
 751         struct dlm_node_iter iter;
 752         unsigned int hash;
 753         int tries = 0;
 754         int bit, wait_on_recovery = 0;
 755
 756         BUG_ON(!lockid);
 757
 758         hash = dlm_lockid_hash(lockid, namelen);
 759
 760         mlog(0, "get lockres %s (len %d)\n", lockid, namelen);
 761
 762 lookup:
 763         spin_lock(&dlm->spinlock);
 764         tmpres = __dlm_lookup_lockres(dlm, lockid, namelen, hash);
 765         if (tmpres) {
 766                 spin_unlock(&dlm->spinlock);
 767                 mlog(0, "found in hash!\n");
 768                 if (res)
 769                         dlm_lockres_put(res);
 770                 res = tmpres;
 771                 goto leave;
 772         }
 773
 774         if (!res) {
 775                 spin_unlock(&dlm->spinlock);
 776                 mlog(0, "allocating a new resource\n");
 777                 /* nothing found and we need to allocate one. */
 778                 alloc_mle = (struct dlm_master_list_entry *)
 779                         kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
 780                 if (!alloc_mle)
 781                         goto leave;
 782                 res = dlm_new_lockres(dlm, lockid, namelen);
 783                 if (!res)
 784                         goto leave;
 785                 goto lookup;
 786         }
 787
 788         mlog(0, "no lockres found, allocated our own: %p\n", res);
 789
 790         if (flags & LKM_LOCAL) {
 791                 /* caller knows it's safe to assume it's not mastered elsewhere
 792                  * DONE!  return right away */
 793                 spin_lock(&res->spinlock);
 794                 dlm_change_lockres_owner(dlm, res, dlm->node_num);
 795                 __dlm_insert_lockres(dlm, res);
 796                 spin_unlock(&res->spinlock);
 797                 spin_unlock(&dlm->spinlock);
 798                 /* lockres still marked IN_PROGRESS */
 799                 goto wake_waiters;
 800         }
 801
 802         /* check master list to see if another node has started mastering it */
 803         spin_lock(&dlm->master_lock);
 804
 805         /* if we found a block, wait for lock to be mastered by another node */
 806         blocked = dlm_find_mle(dlm, &mle, (char *)lockid, namelen);
 807         if (blocked) {
 808                 if (mle->type == DLM_MLE_MASTER) {
 809                         mlog(ML_ERROR, "master entry for nonexistent lock!\n");
 810                         BUG();
 811                 } else if (mle->type == DLM_MLE_MIGRATION) {
 812                         /* migration is in progress! */
 813                         /* the good news is that we now know the
 814                          * "current" master (mle->master). */
 815
 816                         spin_unlock(&dlm->master_lock);
 817                         assert_spin_locked(&dlm->spinlock);
 818
 819                         /* set the lockres owner and hash it */
 820                         spin_lock(&res->spinlock);
 821                         dlm_set_lockres_owner(dlm, res, mle->master);
 822                         __dlm_insert_lockres(dlm, res);
 823                         spin_unlock(&res->spinlock);
 824                         spin_unlock(&dlm->spinlock);
 825
 826                         /* master is known, detach */
 827                         dlm_mle_detach_hb_events(dlm, mle);
 828                         dlm_put_mle(mle);
 829                         mle = NULL;
 830                         goto wake_waiters;
 831                 }
 832         } else {
 833                 /* go ahead and try to master lock on this node */
 834                 mle = alloc_mle;
 835                 /* make sure this does not get freed below */
 836                 alloc_mle = NULL;
 837                 dlm_init_mle(mle, DLM_MLE_MASTER, dlm, res, NULL, 0);
 838                 set_bit(dlm->node_num, mle->maybe_map);
 839                 list_add(&mle->list, &dlm->master_list);
 840
 841                 /* still holding the dlm spinlock, check the recovery map
 842                  * to see if there are any nodes that still need to be
 843                  * considered.  these will not appear in the mle nodemap
 844                  * but they might own this lockres.  wait on them. */
 845                 bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
 846                 if (bit < O2NM_MAX_NODES) {
 847                         mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to"
 848                              "recover before lock mastery can begin\n",
 849                              dlm->name, namelen, (char *)lockid, bit);
 850                         wait_on_recovery = 1;
 851                 }
 852         }
 853
 854         /* at this point there is either a DLM_MLE_BLOCK or a
 855          * DLM_MLE_MASTER on the master list, so it's safe to add the
 856          * lockres to the hashtable.  anyone who finds the lock will
 857          * still have to wait on the IN_PROGRESS. */
 858
 859         /* finally add the lockres to its hash bucket */
 860         __dlm_insert_lockres(dlm, res);
 861         /* get an extra ref on the mle in case this is a BLOCK
 862          * if so, the creator of the BLOCK may try to put the last
 863          * ref at this time in the assert master handler, so we
 864          * need an extra one to keep from a bad ptr deref. */
 865         dlm_get_mle_inuse(mle);
 866         spin_unlock(&dlm->master_lock);
 867         spin_unlock(&dlm->spinlock);
 868
 869 redo_request:
 870         while (wait_on_recovery) {
 871                 /* any cluster changes that occurred after dropping the
 872                  * dlm spinlock would be detectable be a change on the mle,
 873                  * so we only need to clear out the recovery map once. */
 874                 if (dlm_is_recovery_lock(lockid, namelen)) {
 875                         mlog(ML_NOTICE, "%s: recovery map is not empty, but "
 876                              "must master $RECOVERY lock now\n", dlm->name);
 877                         if (!dlm_pre_master_reco_lockres(dlm, res))
 878                                 wait_on_recovery = 0;
 879                         else {
 880                                 mlog(0, "%s: waiting 500ms for heartbeat state "
 881                                     "change\n", dlm->name);
 882                                 msleep(500);
 883                         }
 884                         continue;
 885                 }
 886
 887                 dlm_kick_recovery_thread(dlm);
 888                 msleep(1000);
 889                 dlm_wait_for_recovery(dlm);
 890
 891                 spin_lock(&dlm->spinlock);
 892                 bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
 893                 if (bit < O2NM_MAX_NODES) {
 894                         mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to"
 895                              "recover before lock mastery can begin\n",
 896                              dlm->name, namelen, (char *)lockid, bit);
 897                         wait_on_recovery = 1;
 898                 } else
 899                         wait_on_recovery = 0;
 900                 spin_unlock(&dlm->spinlock);
 901
 902                 if (wait_on_recovery)
 903                         dlm_wait_for_node_recovery(dlm, bit, 10000);
 904         }
 905
 906         /* must wait for lock to be mastered elsewhere */
 907         if (blocked)
 908                 goto wait;
 909
 910         ret = -EINVAL;
 911         dlm_node_iter_init(mle->vote_map, &iter);
 912         while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
 913                 ret = dlm_do_master_request(mle, nodenum);
 914                 if (ret < 0)
 915                         mlog_errno(ret);
 916                 if (mle->master != O2NM_MAX_NODES) {
 917                         /* found a master ! */
 918                         if (mle->master <= nodenum)
 919                                 break;
 920                         /* if our master request has not reached the master
 921                          * yet, keep going until it does.  this is how the
 922                          * master will know that asserts are needed back to
 923                          * the lower nodes. */
 924                         mlog(0, "%s:%.*s: requests only up to %u but master "
 925                              "is %u, keep going\n", dlm->name, namelen,
 926                              lockid, nodenum, mle->master);
 927                 }
 928         }
 929
 930 wait:
 931         /* keep going until the response map includes all nodes */
 932         ret = dlm_wait_for_lock_mastery(dlm, res, mle, &blocked);
 933         if (ret < 0) {
 934                 wait_on_recovery = 1;
 935                 mlog(0, "%s:%.*s: node map changed, redo the "
 936                      "master request now, blocked=%d\n",
 937                      dlm->name, res->lockname.len,
 938                      res->lockname.name, blocked);
 939                 if (++tries > 20) {
 940                         mlog(ML_ERROR, "%s:%.*s: spinning on "
 941                              "dlm_wait_for_lock_mastery, blocked=%d\n",
 942                              dlm->name, res->lockname.len,
 943                              res->lockname.name, blocked);
 944                         dlm_print_one_lock_resource(res);
 945                         dlm_print_one_mle(mle);
 946                         tries = 0;
 947                 }
 948                 goto redo_request;
 949         }
 950
 951         mlog(0, "lockres mastered by %u\n", res->owner);
 952         /* make sure we never continue without this */
 953         BUG_ON(res->owner == O2NM_MAX_NODES);
 954
 955         /* master is known, detach if not already detached */
 956         dlm_mle_detach_hb_events(dlm, mle);
 957         dlm_put_mle(mle);
 958         /* put the extra ref */
 959         dlm_put_mle_inuse(mle);
 960
 961 wake_waiters:
 962         spin_lock(&res->spinlock);
 963         res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
 964         spin_unlock(&res->spinlock);
 965         wake_up(&res->wq);
 966
 967 leave:
 968         /* need to free the unused mle */
 969         if (alloc_mle)
 970                 kmem_cache_free(dlm_mle_cache, alloc_mle);
 971
 972         return res;
 973 }
 974
 975
 976 #define DLM_MASTERY_TIMEOUT_MS   5000
 977
 978 static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm,
 979                                      struct dlm_lock_resource *res,
 980                                      struct dlm_master_list_entry *mle,
 981                                      int *blocked)
 982 {
 983         u8 m;
 984         int ret, bit;
 985         int map_changed, voting_done;
 986         int assert, sleep;
 987
 988 recheck:
 989         ret = 0;
 990         assert = 0;
 991
 992         /* check if another node has already become the owner */
 993         spin_lock(&res->spinlock);
 994         if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
 995                 mlog(0, "%s:%.*s: owner is suddenly %u\n", dlm->name,
 996                      res->lockname.len, res->lockname.name, res->owner);
 997                 spin_unlock(&res->spinlock);
 998                 /* this will cause the master to re-assert across
 999                  * the whole cluster, freeing up mles */
1000                 if (res->owner != dlm->node_num) {
1001                         ret = dlm_do_master_request(mle, res->owner);
1002                         if (ret < 0) {
1003                                 /* give recovery a chance to run */
1004                                 mlog(ML_ERROR, "link to %u went down?: %d\n", res->owner, ret);
1005                                 msleep(500);
1006                                 goto recheck;
1007                         }
1008                 }
1009                 ret = 0;
1010                 goto leave;
1011         }
1012         spin_unlock(&res->spinlock);
1013
1014         spin_lock(&mle->spinlock);
1015         m = mle->master;
1016         map_changed = (memcmp(mle->vote_map, mle->node_map,
1017                               sizeof(mle->vote_map)) != 0);
1018         voting_done = (memcmp(mle->vote_map, mle->response_map,
1019                              sizeof(mle->vote_map)) == 0);
1020
1021         /* restart if we hit any errors */
1022         if (map_changed) {
1023                 int b;
1024                 mlog(0, "%s: %.*s: node map changed, restarting\n",
1025                      dlm->name, res->lockname.len, res->lockname.name);
1026                 ret = dlm_restart_lock_mastery(dlm, res, mle, *blocked);
1027                 b = (mle->type == DLM_MLE_BLOCK);
1028                 if ((*blocked && !b) || (!*blocked && b)) {
1029                         mlog(0, "%s:%.*s: status change: old=%d new=%d\n",
1030                              dlm->name, res->lockname.len, res->lockname.name,
1031                              *blocked, b);
1032                         *blocked = b;
1033                 }
1034                 spin_unlock(&mle->spinlock);
1035                 if (ret < 0) {
1036                         mlog_errno(ret);
1037                         goto leave;
1038                 }
1039                 mlog(0, "%s:%.*s: restart lock mastery succeeded, "
1040                      "rechecking now\n", dlm->name, res->lockname.len,
1041                      res->lockname.name);
1042                 goto recheck;
1043         } else {
1044                 if (!voting_done) {
1045                         mlog(0, "map not changed and voting not done "
1046                              "for %s:%.*s\n", dlm->name, res->lockname.len,
1047                              res->lockname.name);
1048                 }
1049         }
1050
1051         if (m != O2NM_MAX_NODES) {
1052                 /* another node has done an assert!
1053                  * all done! */
1054                 sleep = 0;
1055         } else {
1056                 sleep = 1;
1057                 /* have all nodes responded? */
1058                 if (voting_done && !*blocked) {
1059                         bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
1060                         if (dlm->node_num <= bit) {
1061                                 /* my node number is lowest.
1062                                  * now tell other nodes that I am
1063                                  * mastering this. */
1064                                 mle->master = dlm->node_num;
1065                                 assert = 1;
1066                                 sleep = 0;
1067                         }
1068                         /* if voting is done, but we have not received
1069                          * an assert master yet, we must sleep */
1070                 }
1071         }
1072
1073         spin_unlock(&mle->spinlock);
1074
1075         /* sleep if we haven't finished voting yet */
1076         if (sleep) {
1077                 unsigned long timeo = msecs_to_jiffies(DLM_MASTERY_TIMEOUT_MS);
1078
1079                 /*
1080                 if (atomic_read(&mle->mle_refs.refcount) < 2)
1081                         mlog(ML_ERROR, "mle (%p) refs=%d, name=%.*s\n", mle,
1082                         atomic_read(&mle->mle_refs.refcount),
1083                         res->lockname.len, res->lockname.name);
1084                 */
1085                 atomic_set(&mle->woken, 0);
1086                 (void)wait_event_timeout(mle->wq,
1087                                          (atomic_read(&mle->woken) == 1),
1088                                          timeo);
1089                 if (res->owner == O2NM_MAX_NODES) {
1090                         mlog(0, "waiting again\n");
1091                         goto recheck;
1092                 }
1093                 mlog(0, "done waiting, master is %u\n", res->owner);
1094                 ret = 0;
1095                 goto leave;
1096         }
1097
1098         ret = 0;   /* done */
1099         if (assert) {
1100                 m = dlm->node_num;
1101                 mlog(0, "about to master %.*s here, this=%u\n",
1102                      res->lockname.len, res->lockname.name, m);
1103                 ret = dlm_do_assert_master(dlm, res->lockname.name,
1104                                            res->lockname.len, mle->vote_map, 0);
1105                 if (ret) {
1106                         /* This is a failure in the network path,
1107                          * not in the response to the assert_master
1108                          * (any nonzero response is a BUG on this node).
1109                          * Most likely a socket just got disconnected
1110                          * due to node death. */
1111                         mlog_errno(ret);
1112                 }
1113                 /* no longer need to restart lock mastery.
1114                  * all living nodes have been contacted. */
1115                 ret = 0;
1116         }
1117
1118         /* set the lockres owner */
1119         spin_lock(&res->spinlock);
1120         dlm_change_lockres_owner(dlm, res, m);
1121         spin_unlock(&res->spinlock);
1122
1123 leave:
1124         return ret;
1125 }
1126
1127 struct dlm_bitmap_diff_iter
1128 {
1129         int curnode;
1130         unsigned long *orig_bm;
1131         unsigned long *cur_bm;
1132         unsigned long diff_bm[BITS_TO_LONGS(O2NM_MAX_NODES)];
1133 };
1134
1135 enum dlm_node_state_change
1136 {
1137         NODE_DOWN = -1,
1138         NODE_NO_CHANGE = 0,
1139         NODE_UP
1140 };
1141
1142 static void dlm_bitmap_diff_iter_init(struct dlm_bitmap_diff_iter *iter,
1143                                       unsigned long *orig_bm,
1144                                       unsigned long *cur_bm)
1145 {
1146         unsigned long p1, p2;
1147         int i;
1148
1149         iter->curnode = -1;
1150         iter->orig_bm = orig_bm;
1151         iter->cur_bm = cur_bm;
1152
1153         for (i = 0; i < BITS_TO_LONGS(O2NM_MAX_NODES); i++) {
1154                 p1 = *(iter->orig_bm + i);
1155                 p2 = *(iter->cur_bm + i);
1156                 iter->diff_bm[i] = (p1 & ~p2) | (p2 & ~p1);
1157         }
1158 }
1159
1160 static int dlm_bitmap_diff_iter_next(struct dlm_bitmap_diff_iter *iter,
1161                                      enum dlm_node_state_change *state)
1162 {
1163         int bit;
1164
1165         if (iter->curnode >= O2NM_MAX_NODES)
1166                 return -ENOENT;
1167
1168         bit = find_next_bit(iter->diff_bm, O2NM_MAX_NODES,
1169                             iter->curnode+1);
1170         if (bit >= O2NM_MAX_NODES) {
1171                 iter->curnode = O2NM_MAX_NODES;
1172                 return -ENOENT;
1173         }
1174
1175         /* if it was there in the original then this node died */
1176         if (test_bit(bit, iter->orig_bm))
1177                 *state = NODE_DOWN;
1178         else
1179                 *state = NODE_UP;
1180
1181         iter->curnode = bit;
1182         return bit;
1183 }
1184
1185
1186 static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm,
1187                                     struct dlm_lock_resource *res,
1188                                     struct dlm_master_list_entry *mle,
1189                                     int blocked)
1190 {
1191         struct dlm_bitmap_diff_iter bdi;
1192         enum dlm_node_state_change sc;
1193         int node;
1194         int ret = 0;
1195
1196         mlog(0, "something happened such that the "
1197              "master process may need to be restarted!\n");
1198
1199         assert_spin_locked(&mle->spinlock);
1200
1201         dlm_bitmap_diff_iter_init(&bdi, mle->vote_map, mle->node_map);
1202         node = dlm_bitmap_diff_iter_next(&bdi, &sc);
1203         while (node >= 0) {
1204                 if (sc == NODE_UP) {
1205                         /* a node came up.  clear any old vote from
1206                          * the response map and set it in the vote map
1207                          * then restart the mastery. */
1208                         mlog(ML_NOTICE, "node %d up while restarting\n", node);
1209
1210                         /* redo the master request, but only for the new node */
1211                         mlog(0, "sending request to new node\n");
1212                         clear_bit(node, mle->response_map);
1213                         set_bit(node, mle->vote_map);
1214                 } else {
1215                         mlog(ML_ERROR, "node down! %d\n", node);
1216                         if (blocked) {
1217                                 int lowest = find_next_bit(mle->maybe_map,
1218                                                        O2NM_MAX_NODES, 0);
1219
1220                                 /* act like it was never there */
1221                                 clear_bit(node, mle->maybe_map);
1222
1223                                 if (node == lowest) {
1224                                         mlog(0, "expected master %u died"
1225                                             " while this node was blocked "
1226                                             "waiting on it!\n", node);
1227                                         lowest = find_next_bit(mle->maybe_map,
1228                                                         O2NM_MAX_NODES,
1229                                                         lowest+1);
1230                                         if (lowest < O2NM_MAX_NODES) {
1231                                                 mlog(0, "%s:%.*s:still "
1232                                                      "blocked. waiting on %u "
1233                                                      "now\n", dlm->name,
1234                                                      res->lockname.len,
1235                                                      res->lockname.name,
1236                                                      lowest);
1237                                         } else {
1238                                                 /* mle is an MLE_BLOCK, but
1239                                                  * there is now nothing left to
1240                                                  * block on.  we need to return
1241                                                  * all the way back out and try
1242                                                  * again with an MLE_MASTER.
1243                                                  * dlm_do_local_recovery_cleanup
1244                                                  * has already run, so the mle
1245                                                  * refcount is ok */
1246                                                 mlog(0, "%s:%.*s: no "
1247                                                      "longer blocking. try to "
1248                                                      "master this here\n",
1249                                                      dlm->name,
1250                                                      res->lockname.len,
1251                                                      res->lockname.name);
1252                                                 mle->type = DLM_MLE_MASTER;
1253                                                 mle->u.res = res;
1254                                         }
1255                                 }
1256                         }
1257
1258                         /* now blank out everything, as if we had never
1259                          * contacted anyone */
1260                         memset(mle->maybe_map, 0, sizeof(mle->maybe_map));
1261                         memset(mle->response_map, 0, sizeof(mle->response_map));
1262                         /* reset the vote_map to the current node_map */
1263                         memcpy(mle->vote_map, mle->node_map,
1264                                sizeof(mle->node_map));
1265                         /* put myself into the maybe map */
1266                         if (mle->type != DLM_MLE_BLOCK)
1267                                 set_bit(dlm->node_num, mle->maybe_map);
1268                 }
1269                 ret = -EAGAIN;
1270                 node = dlm_bitmap_diff_iter_next(&bdi, &sc);
1271         }
1272         return ret;
1273 }
1274
1275
1276 /*
1277  * DLM_MASTER_REQUEST_MSG
1278  *
1279  * returns: 0 on success,
1280  *          -errno on a network error
1281  *
1282  * on error, the caller should assume the target node is "dead"
1283  *
1284  */
1285
1286 static int dlm_do_master_request(struct dlm_master_list_entry *mle, int to)
1287 {
1288         struct dlm_ctxt *dlm = mle->dlm;
1289         struct dlm_master_request request;
1290         int ret, response=0, resend;
1291
1292         memset(&request, 0, sizeof(request));
1293         request.node_idx = dlm->node_num;
1294
1295         BUG_ON(mle->type == DLM_MLE_MIGRATION);
1296
1297         if (mle->type != DLM_MLE_MASTER) {
1298                 request.namelen = mle->u.name.len;
1299                 memcpy(request.name, mle->u.name.name, request.namelen);
1300         } else {
1301                 request.namelen = mle->u.res->lockname.len;
1302                 memcpy(request.name, mle->u.res->lockname.name,
1303                         request.namelen);
1304         }
1305
1306 again:
1307         ret = o2net_send_message(DLM_MASTER_REQUEST_MSG, dlm->key, &request,
1308                                  sizeof(request), to, &response);
1309         if (ret < 0)  {
1310                 if (ret == -ESRCH) {
1311                         /* should never happen */
1312                         mlog(ML_ERROR, "TCP stack not ready!\n");
1313                         BUG();
1314                 } else if (ret == -EINVAL) {
1315                         mlog(ML_ERROR, "bad args passed to o2net!\n");
1316                         BUG();
1317                 } else if (ret == -ENOMEM) {
1318                         mlog(ML_ERROR, "out of memory while trying to send "
1319                              "network message!  retrying\n");
1320                         /* this is totally crude */
1321                         msleep(50);
1322                         goto again;
1323                 } else if (!dlm_is_host_down(ret)) {
1324                         /* not a network error. bad. */
1325                         mlog_errno(ret);
1326                         mlog(ML_ERROR, "unhandled error!");
1327                         BUG();
1328                 }
1329                 /* all other errors should be network errors,
1330                  * and likely indicate node death */
1331                 mlog(ML_ERROR, "link to %d went down!\n", to);
1332                 goto out;
1333         }
1334
1335         ret = 0;
1336         resend = 0;
1337         spin_lock(&mle->spinlock);
1338         switch (response) {
1339                 case DLM_MASTER_RESP_YES:
1340                         set_bit(to, mle->response_map);
1341                         mlog(0, "node %u is the master, response=YES\n", to);
1342                         mle->master = to;
1343                         break;
1344                 case DLM_MASTER_RESP_NO:
1345                         mlog(0, "node %u not master, response=NO\n", to);
1346                         set_bit(to, mle->response_map);
1347                         break;
1348                 case DLM_MASTER_RESP_MAYBE:
1349                         mlog(0, "node %u not master, response=MAYBE\n", to);
1350                         set_bit(to, mle->response_map);
1351                         set_bit(to, mle->maybe_map);
1352                         break;
1353                 case DLM_MASTER_RESP_ERROR:
1354                         mlog(0, "node %u hit an error, resending\n", to);
1355                         resend = 1;
1356                         response = 0;
1357                         break;
1358                 default:
1359                         mlog(ML_ERROR, "bad response! %u\n", response);
1360                         BUG();
1361         }
1362         spin_unlock(&mle->spinlock);
1363         if (resend) {
1364                 /* this is also totally crude */
1365                 msleep(50);
1366                 goto again;
1367         }
1368
1369 out:
1370         return ret;
1371 }
1372
1373 /*
1374  * locks that can be taken here:
1375  * dlm->spinlock
1376  * res->spinlock
1377  * mle->spinlock
1378  * dlm->master_list
1379  *
1380  * if possible, TRIM THIS DOWN!!!
1381  */
1382 int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data)
1383 {
1384         u8 response = DLM_MASTER_RESP_MAYBE;
1385         struct dlm_ctxt *dlm = data;
1386         struct dlm_lock_resource *res = NULL;
1387         struct dlm_master_request *request = (struct dlm_master_request *) msg->buf;
1388         struct dlm_master_list_entry *mle = NULL, *tmpmle = NULL;
1389         char *name;
1390         unsigned int namelen, hash;
1391         int found, ret;
1392         int set_maybe;
1393         int dispatch_assert = 0;
1394
1395         if (!dlm_grab(dlm))
1396                 return DLM_MASTER_RESP_NO;
1397
1398         if (!dlm_domain_fully_joined(dlm)) {
1399                 response = DLM_MASTER_RESP_NO;
1400                 goto send_response;
1401         }
1402
1403         name = request->name;
1404         namelen = request->namelen;
1405         hash = dlm_lockid_hash(name, namelen);
1406
1407         if (namelen > DLM_LOCKID_NAME_MAX) {
1408                 response = DLM_IVBUFLEN;
1409                 goto send_response;
1410         }
1411
1412 way_up_top:
1413         spin_lock(&dlm->spinlock);
1414         res = __dlm_lookup_lockres(dlm, name, namelen, hash);
1415         if (res) {
1416                 spin_unlock(&dlm->spinlock);
1417
1418                 /* take care of the easy cases up front */
1419                 spin_lock(&res->spinlock);
1420                 if (res->state & DLM_LOCK_RES_RECOVERING) {
1421                         spin_unlock(&res->spinlock);
1422                         mlog(0, "returning DLM_MASTER_RESP_ERROR since res is "
1423                              "being recovered\n");
1424                         response = DLM_MASTER_RESP_ERROR;
1425                         if (mle)
1426                                 kmem_cache_free(dlm_mle_cache, mle);
1427                         goto send_response;
1428                 }
1429
1430                 if (res->owner == dlm->node_num) {
1431                         spin_unlock(&res->spinlock);
1432                         // mlog(0, "this node is the master\n");
1433                         response = DLM_MASTER_RESP_YES;
1434                         if (mle)
1435                                 kmem_cache_free(dlm_mle_cache, mle);
1436
1437                         /* this node is the owner.
1438                          * there is some extra work that needs to
1439                          * happen now.  the requesting node has
1440                          * caused all nodes up to this one to
1441                          * create mles.  this node now needs to
1442                          * go back and clean those up. */
1443                         dispatch_assert = 1;
1444                         goto send_response;
1445                 } else if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
1446                         spin_unlock(&res->spinlock);
1447                         // mlog(0, "node %u is the master\n", res->owner);
1448                         response = DLM_MASTER_RESP_NO;
1449                         if (mle)
1450                                 kmem_cache_free(dlm_mle_cache, mle);
1451                         goto send_response;
1452                 }
1453
1454                 /* ok, there is no owner.  either this node is
1455                  * being blocked, or it is actively trying to
1456                  * master this lock. */
1457                 if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) {
1458                         mlog(ML_ERROR, "lock with no owner should be "
1459                              "in-progress!\n");
1460                         BUG();
1461                 }
1462
1463                 // mlog(0, "lockres is in progress...\n");
1464                 spin_lock(&dlm->master_lock);
1465                 found = dlm_find_mle(dlm, &tmpmle, name, namelen);
1466                 if (!found) {
1467                         mlog(ML_ERROR, "no mle found for this lock!\n");
1468                         BUG();
1469                 }
1470                 set_maybe = 1;
1471                 spin_lock(&tmpmle->spinlock);
1472                 if (tmpmle->type == DLM_MLE_BLOCK) {
1473                         // mlog(0, "this node is waiting for "
1474                         // "lockres to be mastered\n");
1475                         response = DLM_MASTER_RESP_NO;
1476                 } else if (tmpmle->type == DLM_MLE_MIGRATION) {
1477                         mlog(0, "node %u is master, but trying to migrate to "
1478                              "node %u.\n", tmpmle->master, tmpmle->new_master);
1479                         if (tmpmle->master == dlm->node_num) {
1480                                 response = DLM_MASTER_RESP_YES;
1481                                 mlog(ML_ERROR, "no owner on lockres, but this "
1482                                      "node is trying to migrate it to %u?!\n",
1483                                      tmpmle->new_master);
1484                                 BUG();
1485                         } else {
1486                                 /* the real master can respond on its own */
1487                                 response = DLM_MASTER_RESP_NO;
1488                         }
1489                 } else if (tmpmle->master != DLM_LOCK_RES_OWNER_UNKNOWN) {
1490                         set_maybe = 0;
1491                         if (tmpmle->master == dlm->node_num) {
1492                                 response = DLM_MASTER_RESP_YES;
1493                                 /* this node will be the owner.
1494                                  * go back and clean the mles on any
1495                                  * other nodes */
1496                                 dispatch_assert = 1;
1497                         } else
1498                                 response = DLM_MASTER_RESP_NO;
1499                 } else {
1500                         // mlog(0, "this node is attempting to "
1501                         // "master lockres\n");
1502                         response = DLM_MASTER_RESP_MAYBE;
1503                 }
1504                 if (set_maybe)
1505                         set_bit(request->node_idx, tmpmle->maybe_map);
1506                 spin_unlock(&tmpmle->spinlock);
1507
1508                 spin_unlock(&dlm->master_lock);
1509                 spin_unlock(&res->spinlock);
1510
1511                 /* keep the mle attached to heartbeat events */
1512                 dlm_put_mle(tmpmle);
1513                 if (mle)
1514                         kmem_cache_free(dlm_mle_cache, mle);
1515                 goto send_response;
1516         }
1517
1518         /*
1519          * lockres doesn't exist on this node
1520          * if there is an MLE_BLOCK, return NO
1521          * if there is an MLE_MASTER, return MAYBE
1522          * otherwise, add an MLE_BLOCK, return NO
1523          */
1524         spin_lock(&dlm->master_lock);
1525         found = dlm_find_mle(dlm, &tmpmle, name, namelen);
1526         if (!found) {
1527                 /* this lockid has never been seen on this node yet */
1528                 // mlog(0, "no mle found\n");
1529                 if (!mle) {
1530                         spin_unlock(&dlm->master_lock);
1531                         spin_unlock(&dlm->spinlock);
1532
1533                         mle = (struct dlm_master_list_entry *)
1534                                 kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
1535                         if (!mle) {
1536                                 response = DLM_MASTER_RESP_ERROR;
1537                                 mlog_errno(-ENOMEM);
1538                                 goto send_response;
1539                         }
1540                         goto way_up_top;
1541                 }
1542
1543                 // mlog(0, "this is second time thru, already allocated, "
1544                 // "add the block.\n");
1545                 dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL, name, namelen);
1546                 set_bit(request->node_idx, mle->maybe_map);
1547                 list_add(&mle->list, &dlm->master_list);
1548                 response = DLM_MASTER_RESP_NO;
1549         } else {
1550                 // mlog(0, "mle was found\n");
1551                 set_maybe = 1;
1552                 spin_lock(&tmpmle->spinlock);
1553                 if (tmpmle->master == dlm->node_num) {
1554                         mlog(ML_ERROR, "no lockres, but an mle with this node as master!\n");
1555                         BUG();
1556                 }
1557                 if (tmpmle->type == DLM_MLE_BLOCK)
1558                         response = DLM_MASTER_RESP_NO;
1559                 else if (tmpmle->type == DLM_MLE_MIGRATION) {
1560                         mlog(0, "migration mle was found (%u->%u)\n",
1561                              tmpmle->master, tmpmle->new_master);
1562                         /* real master can respond on its own */
1563                         response = DLM_MASTER_RESP_NO;
1564                 } else
1565                         response = DLM_MASTER_RESP_MAYBE;
1566                 if (set_maybe)
1567                         set_bit(request->node_idx, tmpmle->maybe_map);
1568                 spin_unlock(&tmpmle->spinlock);
1569         }
1570         spin_unlock(&dlm->master_lock);
1571         spin_unlock(&dlm->spinlock);
1572
1573         if (found) {
1574                 /* keep the mle attached to heartbeat events */
1575                 dlm_put_mle(tmpmle);
1576         }
1577 send_response:
1578
1579         if (dispatch_assert) {
1580                 if (response != DLM_MASTER_RESP_YES)
1581                         mlog(ML_ERROR, "invalid response %d\n", response);
1582                 if (!res) {
1583                         mlog(ML_ERROR, "bad lockres while trying to assert!\n");
1584                         BUG();
1585                 }
1586                 mlog(0, "%u is the owner of %.*s, cleaning everyone else\n",
1587                              dlm->node_num, res->lockname.len, res->lockname.name);
1588                 ret = dlm_dispatch_assert_master(dlm, res, 0, request->node_idx,
1589                                                  DLM_ASSERT_MASTER_MLE_CLEANUP);
1590                 if (ret < 0) {
1591                         mlog(ML_ERROR, "failed to dispatch assert master work\n");
1592                         response = DLM_MASTER_RESP_ERROR;
1593                 }
1594         }
1595
1596         dlm_put(dlm);
1597         return response;
1598 }
1599
1600 /*
1601  * DLM_ASSERT_MASTER_MSG
1602  */
1603
1604
1605 /*
1606  * NOTE: this can be used for debugging
1607  * can periodically run all locks owned by this node
1608  * and re-assert across the cluster...
1609  */
1610 static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname,
1611                                 unsigned int namelen, void *nodemap,
1612                                 u32 flags)
1613 {
1614         struct dlm_assert_master assert;
1615         int to, tmpret;
1616         struct dlm_node_iter iter;
1617         int ret = 0;
1618         int reassert;
1619
1620         BUG_ON(namelen > O2NM_MAX_NAME_LEN);
1621 again:
1622         reassert = 0;
1623
1624         /* note that if this nodemap is empty, it returns 0 */
1625         dlm_node_iter_init(nodemap, &iter);
1626         while ((to = dlm_node_iter_next(&iter)) >= 0) {
1627                 int r = 0;
1628                 struct dlm_master_list_entry *mle = NULL;
1629
1630                 mlog(0, "sending assert master to %d (%.*s)\n", to,
1631                      namelen, lockname);
1632                 memset(&assert, 0, sizeof(assert));
1633                 assert.node_idx = dlm->node_num;
1634                 assert.namelen = namelen;
1635                 memcpy(assert.name, lockname, namelen);
1636                 assert.flags = cpu_to_be32(flags);
1637
1638                 tmpret = o2net_send_message(DLM_ASSERT_MASTER_MSG, dlm->key,
1639                                             &assert, sizeof(assert), to, &r);
1640                 if (tmpret < 0) {
1641                         mlog(0, "assert_master returned %d!\n", tmpret);
1642                         if (!dlm_is_host_down(tmpret)) {
1643                                 mlog(ML_ERROR, "unhandled error=%d!\n", tmpret);
1644                                 BUG();
1645                         }
1646                         /* a node died.  finish out the rest of the nodes. */
1647                         mlog(0, "link to %d went down!\n", to);
1648                         /* any nonzero status return will do */
1649                         ret = tmpret;
1650                 } else if (r < 0) {
1651                         /* ok, something horribly messed.  kill thyself. */
1652                         mlog(ML_ERROR,"during assert master of %.*s to %u, "
1653                              "got %d.\n", namelen, lockname, to, r);
1654                         spin_lock(&dlm->spinlock);
1655                         spin_lock(&dlm->master_lock);
1656                         if (dlm_find_mle(dlm, &mle, (char *)lockname,
1657                                          namelen)) {
1658                                 dlm_print_one_mle(mle);
1659                                 __dlm_put_mle(mle);
1660                         }
1661                         spin_unlock(&dlm->master_lock);
1662                         spin_unlock(&dlm->spinlock);
1663                         BUG();
1664                 } else if (r == EAGAIN) {
1665                         mlog(0, "%.*s: node %u create mles on other "
1666                              "nodes and requests a re-assert\n",
1667                              namelen, lockname, to);
1668                         reassert = 1;
1669                 }
1670         }
1671
1672         if (reassert)
1673                 goto again;
1674
1675         return ret;
1676 }
1677
1678 /*
1679  * locks that can be taken here:
1680  * dlm->spinlock
1681  * res->spinlock
1682  * mle->spinlock
1683  * dlm->master_list
1684  *
1685  * if possible, TRIM THIS DOWN!!!
1686  */
1687 int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data)
1688 {
1689         struct dlm_ctxt *dlm = data;
1690         struct dlm_master_list_entry *mle = NULL;
1691         struct dlm_assert_master *assert = (struct dlm_assert_master *)msg->buf;
1692         struct dlm_lock_resource *res = NULL;
1693         char *name;
1694         unsigned int namelen, hash;
1695         u32 flags;
1696         int master_request = 0;
1697         int ret = 0;
1698
1699         if (!dlm_grab(dlm))
1700                 return 0;
1701
1702         name = assert->name;
1703         namelen = assert->namelen;
1704         hash = dlm_lockid_hash(name, namelen);
1705         flags = be32_to_cpu(assert->flags);
1706
1707         if (namelen > DLM_LOCKID_NAME_MAX) {
1708                 mlog(ML_ERROR, "Invalid name length!");
1709                 goto done;
1710         }
1711
1712         spin_lock(&dlm->spinlock);
1713
1714         if (flags)
1715                 mlog(0, "assert_master with flags: %u\n", flags);
1716
1717         /* find the MLE */
1718         spin_lock(&dlm->master_lock);
1719         if (!dlm_find_mle(dlm, &mle, name, namelen)) {
1720                 /* not an error, could be master just re-asserting */
1721                 mlog(0, "just got an assert_master from %u, but no "
1722                      "MLE for it! (%.*s)\n", assert->node_idx,
1723                      namelen, name);
1724         } else {
1725                 int bit = find_next_bit (mle->maybe_map, O2NM_MAX_NODES, 0);
1726                 if (bit >= O2NM_MAX_NODES) {
1727                         /* not necessarily an error, though less likely.
1728                          * could be master just re-asserting. */
1729                         mlog(0, "no bits set in the maybe_map, but %u "
1730                              "is asserting! (%.*s)\n", assert->node_idx,
1731                              namelen, name);
1732                 } else if (bit != assert->node_idx) {
1733                         if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) {
1734                                 mlog(0, "master %u was found, %u should "
1735                                      "back off\n", assert->node_idx, bit);
1736                         } else {
1737                                 /* with the fix for bug 569, a higher node
1738                                  * number winning the mastery will respond
1739                                  * YES to mastery requests, but this node
1740                                  * had no way of knowing.  let it pass. */
1741                                 mlog(0, "%u is the lowest node, "
1742                                      "%u is asserting. (%.*s)  %u must "
1743                                      "have begun after %u won.\n", bit,
1744                                      assert->node_idx, namelen, name, bit,
1745                                      assert->node_idx);
1746                         }
1747                 }
1748                 if (mle->type == DLM_MLE_MIGRATION) {
1749                         if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) {
1750                                 mlog(0, "%s:%.*s: got cleanup assert"
1751                                      " from %u for migration\n",
1752                                      dlm->name, namelen, name,
1753                                      assert->node_idx);
1754                         } else if (!(flags & DLM_ASSERT_MASTER_FINISH_MIGRATION)) {
1755                                 mlog(0, "%s:%.*s: got unrelated assert"
1756                                      " from %u for migration, ignoring\n",
1757                                      dlm->name, namelen, name,
1758                                      assert->node_idx);
1759                                 __dlm_put_mle(mle);
1760                                 spin_unlock(&dlm->master_lock);
1761                                 spin_unlock(&dlm->spinlock);
1762                                 goto done;
1763                         }
1764                 }
1765         }
1766         spin_unlock(&dlm->master_lock);
1767
1768         /* ok everything checks out with the MLE
1769          * now check to see if there is a lockres */
1770         res = __dlm_lookup_lockres(dlm, name, namelen, hash);
1771         if (res) {
1772                 spin_lock(&res->spinlock);
1773                 if (res->state & DLM_LOCK_RES_RECOVERING)  {
1774                         mlog(ML_ERROR, "%u asserting but %.*s is "
1775                              "RECOVERING!\n", assert->node_idx, namelen, name);
1776                         goto kill;
1777                 }
1778                 if (!mle) {
1779                         if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN &&
1780                             res->owner != assert->node_idx) {
1781                                 mlog(ML_ERROR, "assert_master from "
1782                                           "%u, but current owner is "
1783                                           "%u! (%.*s)\n",
1784                                        assert->node_idx, res->owner,
1785                                        namelen, name);
1786                                 goto kill;
1787                         }
1788                 } else if (mle->type != DLM_MLE_MIGRATION) {
1789                         if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
1790                                 /* owner is just re-asserting */
1791                                 if (res->owner == assert->node_idx) {
1792                                         mlog(0, "owner %u re-asserting on "
1793                                              "lock %.*s\n", assert->node_idx,
1794                                              namelen, name);
1795                                         goto ok;
1796                                 }
1797                                 mlog(ML_ERROR, "got assert_master from "
1798                                      "node %u, but %u is the owner! "
1799                                      "(%.*s)\n", assert->node_idx,
1800                                      res->owner, namelen, name);
1801                                 goto kill;
1802                         }
1803                         if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) {
1804                                 mlog(ML_ERROR, "got assert from %u, but lock "
1805                                      "with no owner should be "
1806                                      "in-progress! (%.*s)\n",
1807                                      assert->node_idx,
1808                                      namelen, name);
1809                                 goto kill;
1810                         }
1811                 } else /* mle->type == DLM_MLE_MIGRATION */ {
1812                         /* should only be getting an assert from new master */
1813                         if (assert->node_idx != mle->new_master) {
1814                                 mlog(ML_ERROR, "got assert from %u, but "
1815                                      "new master is %u, and old master "
1816                                      "was %u (%.*s)\n",
1817                                      assert->node_idx, mle->new_master,
1818                                      mle->master, namelen, name);
1819                                 goto kill;
1820                         }
1821
1822                 }
1823 ok:
1824                 spin_unlock(&res->spinlock);
1825         }
1826         spin_unlock(&dlm->spinlock);
1827
1828         // mlog(0, "woo!  got an assert_master from node %u!\n",
1829         //           assert->node_idx);
1830         if (mle) {
1831                 int extra_ref = 0;
1832                 int nn = -1;
1833                 int rr, err = 0;
1834
1835                 spin_lock(&mle->spinlock);
1836                 if (mle->type == DLM_MLE_BLOCK || mle->type == DLM_MLE_MIGRATION)
1837                         extra_ref = 1;
1838                 else {
1839                         /* MASTER mle: if any bits set in the response map
1840                          * then the calling node needs to re-assert to clear
1841                          * up nodes that this node contacted */
1842                         while ((nn = find_next_bit (mle->response_map, O2NM_MAX_NODES,
1843                                                     nn+1)) < O2NM_MAX_NODES) {
1844                                 if (nn != dlm->node_num && nn != assert->node_idx)
1845                                         master_request = 1;
1846                         }
1847                 }
1848                 mle->master = assert->node_idx;
1849                 atomic_set(&mle->woken, 1);
1850                 wake_up(&mle->wq);
1851                 spin_unlock(&mle->spinlock);
1852
1853                 if (res) {
1854                         spin_lock(&res->spinlock);
1855                         if (mle->type == DLM_MLE_MIGRATION) {
1856                                 mlog(0, "finishing off migration of lockres %.*s, "
1857                                         "from %u to %u\n",
1858                                         res->lockname.len, res->lockname.name,
1859                                         dlm->node_num, mle->new_master);
1860                                 res->state &= ~DLM_LOCK_RES_MIGRATING;
1861                                 dlm_change_lockres_owner(dlm, res, mle->new_master);
1862                                 BUG_ON(res->state & DLM_LOCK_RES_DIRTY);
1863                         } else {
1864                                 dlm_change_lockres_owner(dlm, res, mle->master);
1865                         }
1866                         spin_unlock(&res->spinlock);
1867                 }
1868
1869                 /* master is known, detach if not already detached.
1870                  * ensures that only one assert_master call will happen
1871                  * on this mle. */
1872                 spin_lock(&dlm->spinlock);
1873                 spin_lock(&dlm->master_lock);
1874
1875                 rr = atomic_read(&mle->mle_refs.refcount);
1876                 if (mle->inuse > 0) {
1877                         if (extra_ref && rr < 3)
1878                                 err = 1;
1879                         else if (!extra_ref && rr < 2)
1880                                 err = 1;
1881                 } else {
1882                         if (extra_ref && rr < 2)
1883                                 err = 1;
1884                         else if (!extra_ref && rr < 1)
1885                                 err = 1;
1886                 }
1887                 if (err) {
1888                         mlog(ML_ERROR, "%s:%.*s: got assert master from %u "
1889                              "that will mess up this node, refs=%d, extra=%d, "
1890                              "inuse=%d\n", dlm->name, namelen, name,
1891                              assert->node_idx, rr, extra_ref, mle->inuse);
1892                         dlm_print_one_mle(mle);
1893                 }
1894                 list_del_init(&mle->list);
1895                 __dlm_mle_detach_hb_events(dlm, mle);
1896                 __dlm_put_mle(mle);
1897                 if (extra_ref) {
1898                         /* the assert master message now balances the extra
1899                          * ref given by the master / migration request message.
1900                          * if this is the last put, it will be removed
1901                          * from the list. */
1902                         __dlm_put_mle(mle);
1903                 }
1904                 spin_unlock(&dlm->master_lock);
1905                 spin_unlock(&dlm->spinlock);
1906         } else if (res) {
1907                 if (res->owner != assert->node_idx) {
1908                         mlog(0, "assert_master from %u, but current "
1909                              "owner is %u (%.*s), no mle\n", assert->node_idx,
1910                              res->owner, namelen, name);
1911                 }
1912         }
1913
1914 done:
1915         ret = 0;
1916         if (res)
1917                 dlm_lockres_put(res);
1918         dlm_put(dlm);
1919         if (master_request) {
1920                 mlog(0, "need to tell master to reassert\n");
1921                 ret = EAGAIN;  // positive. negative would shoot down the node.
1922         }
1923         return ret;
1924
1925 kill:
1926         /* kill the caller! */
1927         mlog(ML_ERROR, "Bad message received from another node.  Dumping state "
1928              "and killing the other node now!  This node is OK and can continue.\n");
1929         __dlm_print_one_lock_resource(res);
1930         spin_unlock(&res->spinlock);
1931         spin_unlock(&dlm->spinlock);
1932         dlm_lockres_put(res);
1933         dlm_put(dlm);
1934         return -EINVAL;
1935 }
1936
1937 int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
1938                                struct dlm_lock_resource *res,
1939                                int ignore_higher, u8 request_from, u32 flags)
1940 {
1941         struct dlm_work_item *item;
1942         item = kcalloc(1, sizeof(*item), GFP_NOFS);
1943         if (!item)
1944                 return -ENOMEM;
1945
1946
1947         /* queue up work for dlm_assert_master_worker */
1948         dlm_grab(dlm);  /* get an extra ref for the work item */
1949         dlm_init_work_item(dlm, item, dlm_assert_master_worker, NULL);
1950         item->u.am.lockres = res; /* already have a ref */
1951         /* can optionally ignore node numbers higher than this node */
1952         item->u.am.ignore_higher = ignore_higher;
1953         item->u.am.request_from = request_from;
1954         item->u.am.flags = flags;
1955
1956         if (ignore_higher)
1957                 mlog(0, "IGNORE HIGHER: %.*s\n", res->lockname.len,
1958                      res->lockname.name);
1959
1960         spin_lock(&dlm->work_lock);
1961         list_add_tail(&item->list, &dlm->work_list);
1962         spin_unlock(&dlm->work_lock);
1963
1964         queue_work(dlm->dlm_worker, &dlm->dispatched_work);
1965         return 0;
1966 }
1967
1968 static void dlm_assert_master_worker(struct dlm_work_item *item, void *data)
1969 {
1970         struct dlm_ctxt *dlm = data;
1971         int ret = 0;
1972         struct dlm_lock_resource *res;
1973         unsigned long nodemap[BITS_TO_LONGS(O2NM_MAX_NODES)];
1974         int ignore_higher;
1975         int bit;
1976         u8 request_from;
1977         u32 flags;
1978
1979         dlm = item->dlm;
1980         res = item->u.am.lockres;
1981         ignore_higher = item->u.am.ignore_higher;
1982         request_from = item->u.am.request_from;
1983         flags = item->u.am.flags;
1984
1985         spin_lock(&dlm->spinlock);
1986         memcpy(nodemap, dlm->domain_map, sizeof(nodemap));
1987         spin_unlock(&dlm->spinlock);
1988
1989         clear_bit(dlm->node_num, nodemap);
1990         if (ignore_higher) {
1991                 /* if is this just to clear up mles for nodes below
1992                  * this node, do not send the message to the original
1993                  * caller or any node number higher than this */
1994                 clear_bit(request_from, nodemap);
1995                 bit = dlm->node_num;
1996                 while (1) {
1997                         bit = find_next_bit(nodemap, O2NM_MAX_NODES,
1998                                             bit+1);
1999                         if (bit >= O2NM_MAX_NODES)
2000                                 break;
2001                         clear_bit(bit, nodemap);
2002                 }
2003         }
2004
2005         /*
2006          * If we're migrating this lock to someone else, we are no
2007          * longer allowed to assert out own mastery.  OTOH, we need to
2008          * prevent migration from starting while we're still asserting
2009          * our dominance.  The reserved ast delays migration.
2010          */
2011         spin_lock(&res->spinlock);
2012         if (res->state & DLM_LOCK_RES_MIGRATING) {
2013                 mlog(0, "Someone asked us to assert mastery, but we're "
2014                      "in the middle of migration.  Skipping assert, "
2015                      "the new master will handle that.\n");
2016                 spin_unlock(&res->spinlock);
2017                 goto put;
2018         } else
2019                 __dlm_lockres_reserve_ast(res);
2020         spin_unlock(&res->spinlock);
2021
2022         /* this call now finishes out the nodemap
2023          * even if one or more nodes die */
2024         mlog(0, "worker about to master %.*s here, this=%u\n",
2025                      res->lockname.len, res->lockname.name, dlm->node_num);
2026         ret = dlm_do_assert_master(dlm, res->lockname.name,
2027                                    res->lockname.len,
2028                                    nodemap, flags);
2029         if (ret < 0) {
2030                 /* no need to restart, we are done */
2031                 if (!dlm_is_host_down(ret))
2032                         mlog_errno(ret);
2033         }
2034
2035         /* Ok, we've asserted ourselves.  Let's let migration start. */
2036         dlm_lockres_release_ast(dlm, res);
2037
2038 put:
2039         dlm_lockres_put(res);
2040
2041         mlog(0, "finished with dlm_assert_master_worker\n");
2042 }
2043
2044 /* SPECIAL CASE for the $RECOVERY lock used by the recovery thread.
2045  * We cannot wait for node recovery to complete to begin mastering this
2046  * lockres because this lockres is used to kick off recovery! ;-)
2047  * So, do a pre-check on all living nodes to see if any of those nodes
2048  * think that $RECOVERY is currently mastered by a dead node.  If so,
2049  * we wait a short time to allow that node to get notified by its own
2050  * heartbeat stack, then check again.  All $RECOVERY lock resources
2051  * mastered by dead nodes are purged when the hearbeat callback is
2052  * fired, so we can know for sure that it is safe to continue once
2053  * the node returns a live node or no node.  */
2054 static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm,
2055                                        struct dlm_lock_resource *res)
2056 {
2057         struct dlm_node_iter iter;
2058         int nodenum;
2059         int ret = 0;
2060         u8 master = DLM_LOCK_RES_OWNER_UNKNOWN;
2061
2062         spin_lock(&dlm->spinlock);
2063         dlm_node_iter_init(dlm->domain_map, &iter);
2064         spin_unlock(&dlm->spinlock);
2065
2066         while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
2067                 /* do not send to self */
2068                 if (nodenum == dlm->node_num)
2069                         continue;
2070                 ret = dlm_do_master_requery(dlm, res, nodenum, &master);
2071                 if (ret < 0) {
2072                         mlog_errno(ret);
2073                         if (!dlm_is_host_down(ret))
2074                                 BUG();
2075                         /* host is down, so answer for that node would be
2076                          * DLM_LOCK_RES_OWNER_UNKNOWN.  continue. */
2077                         ret = 0;
2078                 }
2079
2080                 if (master != DLM_LOCK_RES_OWNER_UNKNOWN) {
2081                         /* check to see if this master is in the recovery map */
2082                         spin_lock(&dlm->spinlock);
2083                         if (test_bit(master, dlm->recovery_map)) {
2084                                 mlog(ML_NOTICE, "%s: node %u has not seen "
2085                                      "node %u go down yet, and thinks the "
2086                                      "dead node is mastering the recovery "
2087                                      "lock.  must wait.\n", dlm->name,
2088                                      nodenum, master);
2089                                 ret = -EAGAIN;
2090                         }
2091                         spin_unlock(&dlm->spinlock);
2092                         mlog(0, "%s: reco lock master is %u\n", dlm->name,
2093                              master);
2094                         break;
2095                 }
2096         }
2097         return ret;
2098 }
2099
2100
2101 /*
2102  * DLM_MIGRATE_LOCKRES
2103  */
2104
2105
2106 int dlm_migrate_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
2107                         u8 target)
2108 {
2109         struct dlm_master_list_entry *mle = NULL;
2110         struct dlm_master_list_entry *oldmle = NULL;
2111         struct dlm_migratable_lockres *mres = NULL;
2112         int ret = -EINVAL;
2113         const char *name;
2114         unsigned int namelen;
2115         int mle_added = 0;
2116         struct list_head *queue, *iter;
2117         int i;
2118         struct dlm_lock *lock;
2119         int empty = 1;
2120
2121         if (!dlm_grab(dlm))
2122                 return -EINVAL;
2123
2124         name = res->lockname.name;
2125         namelen = res->lockname.len;
2126
2127         mlog(0, "migrating %.*s to %u\n", namelen, name, target);
2128
2129         /*
2130          * ensure this lockres is a proper candidate for migration
2131          */
2132         spin_lock(&res->spinlock);
2133         if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
2134                 mlog(0, "cannot migrate lockres with unknown owner!\n");
2135                 spin_unlock(&res->spinlock);
2136                 goto leave;
2137         }
2138         if (res->owner != dlm->node_num) {
2139                 mlog(0, "cannot migrate lockres this node doesn't own!\n");
2140                 spin_unlock(&res->spinlock);
2141                 goto leave;
2142         }
2143         mlog(0, "checking queues...\n");
2144         queue = &res->granted;
2145         for (i=0; i<3; i++) {
2146                 list_for_each(iter, queue) {
2147                         lock = list_entry (iter, struct dlm_lock, list);
2148                         empty = 0;
2149                         if (lock->ml.node == dlm->node_num) {
2150                                 mlog(0, "found a lock owned by this node "
2151                                      "still on the %s queue!  will not "
2152                                      "migrate this lockres\n",
2153                                      i==0 ? "granted" :
2154                                      (i==1 ? "converting" : "blocked"));
2155                                 spin_unlock(&res->spinlock);
2156                                 ret = -ENOTEMPTY;
2157                                 goto leave;
2158                         }
2159                 }
2160                 queue++;
2161         }
2162         mlog(0, "all locks on this lockres are nonlocal.  continuing\n");
2163         spin_unlock(&res->spinlock);
2164
2165         /* no work to do */
2166         if (empty) {
2167                 mlog(0, "no locks were found on this lockres! done!\n");
2168                 ret = 0;
2169                 goto leave;
2170         }
2171
2172         /*
2173          * preallocate up front
2174          * if this fails, abort
2175          */
2176
2177         ret = -ENOMEM;
2178         mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_NOFS);
2179         if (!mres) {
2180                 mlog_errno(ret);
2181                 goto leave;
2182         }
2183
2184         mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache,
2185                                                                 GFP_NOFS);
2186         if (!mle) {
2187                 mlog_errno(ret);
2188                 goto leave;
2189         }
2190         ret = 0;
2191
2192         /*
2193          * find a node to migrate the lockres to
2194          */
2195
2196         mlog(0, "picking a migration node\n");
2197         spin_lock(&dlm->spinlock);
2198         /* pick a new node */
2199         if (!test_bit(target, dlm->domain_map) ||
2200             target >= O2NM_MAX_NODES) {
2201                 target = dlm_pick_migration_target(dlm, res);
2202         }
2203         mlog(0, "node %u chosen for migration\n", target);
2204
2205         if (target >= O2NM_MAX_NODES ||
2206             !test_bit(target, dlm->domain_map)) {
2207                 /* target chosen is not alive */
2208                 ret = -EINVAL;
2209         }
2210
2211         if (ret) {
2212                 spin_unlock(&dlm->spinlock);
2213                 goto fail;
2214         }
2215
2216         mlog(0, "continuing with target = %u\n", target);
2217
2218         /*
2219          * clear any existing master requests and
2220          * add the migration mle to the list
2221          */
2222         spin_lock(&dlm->master_lock);
2223         ret = dlm_add_migration_mle(dlm, res, mle, &oldmle, name,
2224                                     namelen, target, dlm->node_num);
2225         spin_unlock(&dlm->master_lock);
2226         spin_unlock(&dlm->spinlock);
2227
2228         if (ret == -EEXIST) {
2229                 mlog(0, "another process is already migrating it\n");
2230                 goto fail;
2231         }
2232         mle_added = 1;
2233
2234         /*
2235          * set the MIGRATING flag and flush asts
2236          * if we fail after this we need to re-dirty the lockres
2237          */
2238         if (dlm_mark_lockres_migrating(dlm, res, target) < 0) {
2239                 mlog(ML_ERROR, "tried to migrate %.*s to %u, but "
2240                      "the target went down.\n", res->lockname.len,
2241                      res->lockname.name, target);
2242                 spin_lock(&res->spinlock);
2243                 res->state &= ~DLM_LOCK_RES_MIGRATING;
2244                 spin_unlock(&res->spinlock);
2245                 ret = -EINVAL;
2246         }
2247
2248 fail:
2249         if (oldmle) {
2250                 /* master is known, detach if not already detached */
2251                 dlm_mle_detach_hb_events(dlm, oldmle);
2252                 dlm_put_mle(oldmle);
2253         }
2254
2255         if (ret < 0) {
2256                 if (mle_added) {
2257                         dlm_mle_detach_hb_events(dlm, mle);
2258                         dlm_put_mle(mle);
2259                 } else if (mle) {
2260                         kmem_cache_free(dlm_mle_cache, mle);
2261                 }
2262                 goto leave;
2263         }
2264
2265         /*
2266          * at this point, we have a migration target, an mle
2267          * in the master list, and the MIGRATING flag set on
2268          * the lockres
2269          */
2270
2271
2272         /* get an extra reference on the mle.
2273          * otherwise the assert_master from the new
2274          * master will destroy this.
2275          * also, make sure that all callers of dlm_get_mle
2276          * take both dlm->spinlock and dlm->master_lock */
2277         spin_lock(&dlm->spinlock);
2278         spin_lock(&dlm->master_lock);
2279         dlm_get_mle_inuse(mle);
2280         spin_unlock(&dlm->master_lock);
2281         spin_unlock(&dlm->spinlock);
2282
2283         /* notify new node and send all lock state */
2284         /* call send_one_lockres with migration flag.
2285          * this serves as notice to the target node that a
2286          * migration is starting. */
2287         ret = dlm_send_one_lockres(dlm, res, mres, target,
2288                                    DLM_MRES_MIGRATION);
2289
2290         if (ret < 0) {
2291                 mlog(0, "migration to node %u failed with %d\n",
2292                      target, ret);
2293                 /* migration failed, detach and clean up mle */
2294                 dlm_mle_detach_hb_events(dlm, mle);
2295                 dlm_put_mle(mle);
2296                 dlm_put_mle_inuse(mle);
2297                 spin_lock(&res->spinlock);
2298                 res->state &= ~DLM_LOCK_RES_MIGRATING;
2299                 spin_unlock(&res->spinlock);
2300                 goto leave;
2301         }
2302
2303         /* at this point, the target sends a message to all nodes,
2304          * (using dlm_do_migrate_request).  this node is skipped since
2305          * we had to put an mle in the list to begin the process.  this
2306          * node now waits for target to do an assert master.  this node
2307          * will be the last one notified, ensuring that the migration
2308          * is complete everywhere.  if the target dies while this is
2309          * going on, some nodes could potentially see the target as the
2310          * master, so it is important that my recovery finds the migration
2311          * mle and sets the master to UNKNONWN. */
2312
2313
2314         /* wait for new node to assert master */
2315         while (1) {
2316                 ret = wait_event_interruptible_timeout(mle->wq,
2317                                         (atomic_read(&mle->woken) == 1),
2318                                         msecs_to_jiffies(5000));
2319
2320                 if (ret >= 0) {
2321                         if (atomic_read(&mle->woken) == 1 ||
2322                             res->owner == target)
2323                                 break;
2324
2325                         mlog(0, "timed out during migration\n");
2326                         /* avoid hang during shutdown when migrating lockres
2327                          * to a node which also goes down */
2328                         if (dlm_is_node_dead(dlm, target)) {
2329                                 mlog(0, "%s:%.*s: expected migration "
2330                                      "target %u is no longer up, restarting\n",
2331                                      dlm->name, res->lockname.len,
2332                                      res->lockname.name, target);
2333                                 ret = -ERESTARTSYS;
2334                         }
2335                 }
2336                 if (ret == -ERESTARTSYS) {
2337                         /* migration failed, detach and clean up mle */
2338                         dlm_mle_detach_hb_events(dlm, mle);
2339                         dlm_put_mle(mle);
2340                         dlm_put_mle_inuse(mle);
2341                         spin_lock(&res->spinlock);
2342                         res->state &= ~DLM_LOCK_RES_MIGRATING;
2343                         spin_unlock(&res->spinlock);
2344                         goto leave;
2345                 }
2346                 /* TODO: if node died: stop, clean up, return error */
2347         }
2348
2349         /* all done, set the owner, clear the flag */
2350         spin_lock(&res->spinlock);
2351         dlm_set_lockres_owner(dlm, res, target);
2352         res->state &= ~DLM_LOCK_RES_MIGRATING;
2353         dlm_remove_nonlocal_locks(dlm, res);
2354         spin_unlock(&res->spinlock);
2355         wake_up(&res->wq);
2356
2357         /* master is known, detach if not already detached */
2358         dlm_mle_detach_hb_events(dlm, mle);
2359         dlm_put_mle_inuse(mle);
2360         ret = 0;
2361
2362         dlm_lockres_calc_usage(dlm, res);
2363
2364 leave:
2365         /* re-dirty the lockres if we failed */
2366         if (ret < 0)
2367                 dlm_kick_thread(dlm, res);
2368
2369         /* TODO: cleanup */
2370         if (mres)
2371                 free_page((unsigned long)mres);
2372
2373         dlm_put(dlm);
2374
2375         mlog(0, "returning %d\n", ret);
2376         return ret;
2377 }
2378
2379 int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock)
2380 {
2381         int ret;
2382         spin_lock(&dlm->ast_lock);
2383         spin_lock(&lock->spinlock);
2384         ret = (list_empty(&lock->bast_list) && !lock->bast_pending);
2385         spin_unlock(&lock->spinlock);
2386         spin_unlock(&dlm->ast_lock);
2387         return ret;
2388 }
2389
2390 static int dlm_migration_can_proceed(struct dlm_ctxt *dlm,
2391                                      struct dlm_lock_resource *res,
2392                                      u8 mig_target)
2393 {
2394         int can_proceed;
2395         spin_lock(&res->spinlock);
2396         can_proceed = !!(res->state & DLM_LOCK_RES_MIGRATING);
2397         spin_unlock(&res->spinlock);
2398
2399         /* target has died, so make the caller break out of the
2400          * wait_event, but caller must recheck the domain_map */
2401         spin_lock(&dlm->spinlock);
2402         if (!test_bit(mig_target, dlm->domain_map))
2403                 can_proceed = 1;
2404         spin_unlock(&dlm->spinlock);
2405         return can_proceed;
2406 }
2407
2408 int dlm_lockres_is_dirty(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
2409 {
2410         int ret;
2411         spin_lock(&res->spinlock);
2412         ret = !!(res->state & DLM_LOCK_RES_DIRTY);
2413         spin_unlock(&res->spinlock);
2414         return ret;
2415 }
2416
2417
2418 static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm,
2419                                        struct dlm_lock_resource *res,
2420                                        u8 target)
2421 {
2422         int ret = 0;
2423
2424         mlog(0, "dlm_mark_lockres_migrating: %.*s, from %u to %u\n",
2425                res->lockname.len, res->lockname.name, dlm->node_num,
2426                target);
2427         /* need to set MIGRATING flag on lockres.  this is done by
2428          * ensuring that all asts have been flushed for this lockres. */
2429         spin_lock(&res->spinlock);
2430         BUG_ON(res->migration_pending);
2431         res->migration_pending = 1;
2432         /* strategy is to reserve an extra ast then release
2433          * it below, letting the release do all of the work */
2434         __dlm_lockres_reserve_ast(res);
2435         spin_unlock(&res->spinlock);
2436
2437         /* now flush all the pending asts.. hang out for a bit */
2438         dlm_kick_thread(dlm, res);
2439         wait_event(dlm->ast_wq, !dlm_lockres_is_dirty(dlm, res));
2440         dlm_lockres_release_ast(dlm, res);
2441
2442         mlog(0, "about to wait on migration_wq, dirty=%s\n",
2443                res->state & DLM_LOCK_RES_DIRTY ? "yes" : "no");
2444         /* if the extra ref we just put was the final one, this
2445          * will pass thru immediately.  otherwise, we need to wait
2446          * for the last ast to finish. */
2447 again:
2448         ret = wait_event_interruptible_timeout(dlm->migration_wq,
2449                    dlm_migration_can_proceed(dlm, res, target),
2450                    msecs_to_jiffies(1000));
2451         if (ret < 0) {
2452                 mlog(0, "woken again: migrating? %s, dead? %s\n",
2453                        res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no",
2454                        test_bit(target, dlm->domain_map) ? "no":"yes");
2455         } else {
2456                 mlog(0, "all is well: migrating? %s, dead? %s\n",
2457                        res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no",
2458                        test_bit(target, dlm->domain_map) ? "no":"yes");
2459         }
2460         if (!dlm_migration_can_proceed(dlm, res, target)) {
2461                 mlog(0, "trying again...\n");
2462                 goto again;
2463         }
2464
2465         /* did the target go down or die? */
2466         spin_lock(&dlm->spinlock);
2467         if (!test_bit(target, dlm->domain_map)) {
2468                 mlog(ML_ERROR, "aha. migration target %u just went down\n",
2469                      target);
2470                 ret = -EHOSTDOWN;
2471         }
2472         spin_unlock(&dlm->spinlock);
2473
2474         /*
2475          * at this point:
2476          *
2477          *   o the DLM_LOCK_RES_MIGRATING flag is set
2478          *   o there are no pending asts on this lockres
2479          *   o all processes trying to reserve an ast on this
2480          *     lockres must wait for the MIGRATING flag to clear
2481          */
2482         return ret;
2483 }
2484
2485 /* last step in the migration process.
2486  * original master calls this to free all of the dlm_lock
2487  * structures that used to be for other nodes. */
2488 static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
2489                                       struct dlm_lock_resource *res)
2490 {
2491         struct list_head *iter, *iter2;
2492         struct list_head *queue = &res->granted;
2493         int i;
2494         struct dlm_lock *lock;
2495
2496         assert_spin_locked(&res->spinlock);
2497
2498         BUG_ON(res->owner == dlm->node_num);
2499
2500         for (i=0; i<3; i++) {
2501                 list_for_each_safe(iter, iter2, queue) {
2502                         lock = list_entry (iter, struct dlm_lock, list);
2503                         if (lock->ml.node != dlm->node_num) {
2504                                 mlog(0, "putting lock for node %u\n",
2505                                      lock->ml.node);
2506                                 /* be extra careful */
2507                                 BUG_ON(!list_empty(&lock->ast_list));
2508                                 BUG_ON(!list_empty(&lock->bast_list));
2509                                 BUG_ON(lock->ast_pending);
2510                                 BUG_ON(lock->bast_pending);
2511                                 list_del_init(&lock->list);
2512                                 dlm_lock_put(lock);
2513                         }
2514                 }
2515                 queue++;
2516         }
2517 }
2518
2519 /* for now this is not too intelligent.  we will
2520  * need stats to make this do the right thing.
2521  * this just finds the first lock on one of the
2522  * queues and uses that node as the target. */
2523 static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
2524                                     struct dlm_lock_resource *res)
2525 {
2526         int i;
2527         struct list_head *queue = &res->granted;
2528         struct list_head *iter;
2529         struct dlm_lock *lock;
2530         int nodenum;
2531
2532         assert_spin_locked(&dlm->spinlock);
2533
2534         spin_lock(&res->spinlock);
2535         for (i=0; i<3; i++) {
2536                 list_for_each(iter, queue) {
2537                         /* up to the caller to make sure this node
2538                          * is alive */
2539                         lock = list_entry (iter, struct dlm_lock, list);
2540                         if (lock->ml.node != dlm->node_num) {
2541                                 spin_unlock(&res->spinlock);
2542                                 return lock->ml.node;
2543                         }
2544                 }
2545                 queue++;
2546         }
2547         spin_unlock(&res->spinlock);
2548         mlog(0, "have not found a suitable target yet! checking domain map\n");
2549
2550         /* ok now we're getting desperate.  pick anyone alive. */
2551         nodenum = -1;
2552         while (1) {
2553                 nodenum = find_next_bit(dlm->domain_map,
2554                                         O2NM_MAX_NODES, nodenum+1);
2555                 mlog(0, "found %d in domain map\n", nodenum);
2556                 if (nodenum >= O2NM_MAX_NODES)
2557                         break;
2558                 if (nodenum != dlm->node_num) {
2559                         mlog(0, "picking %d\n", nodenum);
2560                         return nodenum;
2561                 }
2562         }
2563
2564         mlog(0, "giving up.  no master to migrate to\n");
2565         return DLM_LOCK_RES_OWNER_UNKNOWN;
2566 }
2567
2568
2569
2570 /* this is called by the new master once all lockres
2571  * data has been received */
2572 static int dlm_do_migrate_request(struct dlm_ctxt *dlm,
2573                                   struct dlm_lock_resource *res,
2574                                   u8 master, u8 new_master,
2575                                   struct dlm_node_iter *iter)
2576 {
2577         struct dlm_migrate_request migrate;
2578         int ret, status = 0;
2579         int nodenum;
2580
2581         memset(&migrate, 0, sizeof(migrate));
2582         migrate.namelen = res->lockname.len;
2583         memcpy(migrate.name, res->lockname.name, migrate.namelen);
2584         migrate.new_master = new_master;
2585         migrate.master = master;
2586
2587         ret = 0;
2588
2589         /* send message to all nodes, except the master and myself */
2590         while ((nodenum = dlm_node_iter_next(iter)) >= 0) {
2591                 if (nodenum == master ||
2592                     nodenum == new_master)
2593                         continue;
2594
2595                 ret = o2net_send_message(DLM_MIGRATE_REQUEST_MSG, dlm->key,
2596                                          &migrate, sizeof(migrate), nodenum,
2597                                          &status);
2598                 if (ret < 0)
2599                         mlog_errno(ret);
2600                 else if (status < 0) {
2601                         mlog(0, "migrate request (node %u) returned %d!\n",
2602                              nodenum, status);
2603                         ret = status;
2604                 }
2605         }
2606
2607         if (ret < 0)
2608                 mlog_errno(ret);
2609
2610         mlog(0, "returning ret=%d\n", ret);
2611         return ret;
2612 }
2613
2614
2615 /* if there is an existing mle for this lockres, we now know who the master is.
2616  * (the one who sent us *this* message) we can clear it up right away.
2617  * since the process that put the mle on the list still has a reference to it,
2618  * we can unhash it now, set the master and wake the process.  as a result,
2619  * we will have no mle in the list to start with.  now we can add an mle for
2620  * the migration and this should be the only one found for those scanning the
2621  * list.  */
2622 int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data)
2623 {
2624         struct dlm_ctxt *dlm = data;
2625         struct dlm_lock_resource *res = NULL;
2626         struct dlm_migrate_request *migrate = (struct dlm_migrate_request *) msg->buf;
2627         struct dlm_master_list_entry *mle = NULL, *oldmle = NULL;
2628         const char *name;
2629         unsigned int namelen, hash;
2630         int ret = 0;
2631
2632         if (!dlm_grab(dlm))
2633                 return -EINVAL;
2634
2635         name = migrate->name;
2636         namelen = migrate->namelen;
2637         hash = dlm_lockid_hash(name, namelen);
2638
2639         /* preallocate.. if this fails, abort */
2640         mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache,
2641                                                          GFP_NOFS);
2642
2643         if (!mle) {
2644                 ret = -ENOMEM;
2645                 goto leave;
2646         }
2647
2648         /* check for pre-existing lock */
2649         spin_lock(&dlm->spinlock);
2650         res = __dlm_lookup_lockres(dlm, name, namelen, hash);
2651         spin_lock(&dlm->master_lock);
2652
2653         if (res) {
2654                 spin_lock(&res->spinlock);
2655                 if (res->state & DLM_LOCK_RES_RECOVERING) {
2656                         /* if all is working ok, this can only mean that we got
2657                         * a migrate request from a node that we now see as
2658                         * dead.  what can we do here?  drop it to the floor? */
2659                         spin_unlock(&res->spinlock);
2660                         mlog(ML_ERROR, "Got a migrate request, but the "
2661                              "lockres is marked as recovering!");
2662                         kmem_cache_free(dlm_mle_cache, mle);
2663                         ret = -EINVAL; /* need a better solution */
2664                         goto unlock;
2665                 }
2666                 res->state |= DLM_LOCK_RES_MIGRATING;
2667                 spin_unlock(&res->spinlock);
2668         }
2669
2670         /* ignore status.  only nonzero status would BUG. */
2671         ret = dlm_add_migration_mle(dlm, res, mle, &oldmle,
2672                                     name, namelen,
2673                                     migrate->new_master,
2674                                     migrate->master);
2675
2676 unlock:
2677         spin_unlock(&dlm->master_lock);
2678         spin_unlock(&dlm->spinlock);
2679
2680         if (oldmle) {
2681                 /* master is known, detach if not already detached */
2682                 dlm_mle_detach_hb_events(dlm, oldmle);
2683                 dlm_put_mle(oldmle);
2684         }
2685
2686         if (res)
2687                 dlm_lockres_put(res);
2688 leave:
2689         dlm_put(dlm);
2690         return ret;
2691 }
2692
2693 /* must be holding dlm->spinlock and dlm->master_lock
2694  * when adding a migration mle, we can clear any other mles
2695  * in the master list because we know with certainty that
2696  * the master is "master".  so we remove any old mle from
2697  * the list after setting it's master field, and then add
2698  * the new migration mle.  this way we can hold with the rule
2699  * of having only one mle for a given lock name at all times. */
2700 static int dlm_add_migration_mle(struct dlm_ctxt *dlm,
2701                                  struct dlm_lock_resource *res,
2702                                  struct dlm_master_list_entry *mle,
2703                                  struct dlm_master_list_entry **oldmle,
2704                                  const char *name, unsigned int namelen,
2705                                  u8 new_master, u8 master)
2706 {
2707         int found;
2708         int ret = 0;
2709
2710         *oldmle = NULL;
2711
2712         mlog_entry_void();
2713
2714         assert_spin_locked(&dlm->spinlock);
2715         assert_spin_locked(&dlm->master_lock);
2716
2717         /* caller is responsible for any ref taken here on oldmle */
2718         found = dlm_find_mle(dlm, oldmle, (char *)name, namelen);
2719         if (found) {
2720                 struct dlm_master_list_entry *tmp = *oldmle;
2721                 spin_lock(&tmp->spinlock);
2722                 if (tmp->type == DLM_MLE_MIGRATION) {
2723                         if (master == dlm->node_num) {
2724                                 /* ah another process raced me to it */
2725                                 mlog(0, "tried to migrate %.*s, but some "
2726                                      "process beat me to it\n",
2727                                      namelen, name);
2728                                 ret = -EEXIST;
2729                         } else {
2730                                 /* bad.  2 NODES are trying to migrate! */
2731                                 mlog(ML_ERROR, "migration error  mle: "
2732                                      "master=%u new_master=%u // request: "
2733                                      "master=%u new_master=%u // "
2734                                      "lockres=%.*s\n",
2735                                      tmp->master, tmp->new_master,
2736                                      master, new_master,
2737                                      namelen, name);
2738                                 BUG();
2739                         }
2740                 } else {
2741                         /* this is essentially what assert_master does */
2742                         tmp->master = master;
2743                         atomic_set(&tmp->woken, 1);
2744                         wake_up(&tmp->wq);
2745                         /* remove it from the list so that only one
2746                          * mle will be found */
2747                         list_del_init(&tmp->list);
2748                         __dlm_mle_detach_hb_events(dlm, mle);
2749                 }
2750                 spin_unlock(&tmp->spinlock);
2751         }
2752
2753         /* now add a migration mle to the tail of the list */
2754         dlm_init_mle(mle, DLM_MLE_MIGRATION, dlm, res, name, namelen);
2755         mle->new_master = new_master;
2756         mle->master = master;
2757         /* do this for consistency with other mle types */
2758         set_bit(new_master, mle->maybe_map);
2759         list_add(&mle->list, &dlm->master_list);
2760
2761         return ret;
2762 }
2763
2764
2765 void dlm_clean_master_list(struct dlm_ctxt *dlm, u8 dead_node)
2766 {
2767         struct list_head *iter, *iter2;
2768         struct dlm_master_list_entry *mle;
2769         struct dlm_lock_resource *res;
2770         unsigned int hash;
2771
2772         mlog_entry("dlm=%s, dead node=%u\n", dlm->name, dead_node);
2773 top:
2774         assert_spin_locked(&dlm->spinlock);
2775
2776         /* clean the master list */
2777         spin_lock(&dlm->master_lock);
2778         list_for_each_safe(iter, iter2, &dlm->master_list) {
2779                 mle = list_entry(iter, struct dlm_master_list_entry, list);
2780
2781                 BUG_ON(mle->type != DLM_MLE_BLOCK &&
2782                        mle->type != DLM_MLE_MASTER &&
2783                        mle->type != DLM_MLE_MIGRATION);
2784
2785                 /* MASTER mles are initiated locally.  the waiting
2786                  * process will notice the node map change
2787                  * shortly.  let that happen as normal. */
2788                 if (mle->type == DLM_MLE_MASTER)
2789                         continue;
2790
2791
2792                 /* BLOCK mles are initiated by other nodes.
2793                  * need to clean up if the dead node would have
2794                  * been the master. */
2795                 if (mle->type == DLM_MLE_BLOCK) {
2796                         int bit;
2797
2798                         spin_lock(&mle->spinlock);
2799                         bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
2800                         if (bit != dead_node) {
2801                                 mlog(0, "mle found, but dead node %u would "
2802                                      "not have been master\n", dead_node);
2803                                 spin_unlock(&mle->spinlock);
2804                         } else {
2805                                 /* must drop the refcount by one since the
2806                                  * assert_master will never arrive.  this
2807                                  * may result in the mle being unlinked and
2808                                  * freed, but there may still be a process
2809                                  * waiting in the dlmlock path which is fine. */
2810                                 mlog(0, "node %u was expected master\n",
2811                                      dead_node);
2812                                 atomic_set(&mle->woken, 1);
2813                                 spin_unlock(&mle->spinlock);
2814                                 wake_up(&mle->wq);
2815                                 /* do not need events any longer, so detach
2816                                  * from heartbeat */
2817                                 __dlm_mle_detach_hb_events(dlm, mle);
2818                                 __dlm_put_mle(mle);
2819                         }
2820                         continue;
2821                 }
2822
2823                 /* everything else is a MIGRATION mle */
2824
2825                 /* the rule for MIGRATION mles is that the master
2826                  * becomes UNKNOWN if *either* the original or
2827                  * the new master dies.  all UNKNOWN lockreses
2828                  * are sent to whichever node becomes the recovery
2829                  * master.  the new master is responsible for
2830                  * determining if there is still a master for
2831                  * this lockres, or if he needs to take over
2832                  * mastery.  either way, this node should expect
2833                  * another message to resolve this. */
2834                 if (mle->master != dead_node &&
2835                     mle->new_master != dead_node)
2836                         continue;
2837
2838                 /* if we have reached this point, this mle needs to
2839                  * be removed from the list and freed. */
2840
2841                 /* remove from the list early.  NOTE: unlinking
2842                  * list_head while in list_for_each_safe */
2843                 __dlm_mle_detach_hb_events(dlm, mle);
2844                 spin_lock(&mle->spinlock);
2845                 list_del_init(&mle->list);
2846                 atomic_set(&mle->woken, 1);
2847                 spin_unlock(&mle->spinlock);
2848                 wake_up(&mle->wq);
2849
2850                 mlog(0, "%s: node %u died during migration from "
2851                      "%u to %u!\n", dlm->name, dead_node,
2852                      mle->master, mle->new_master);
2853                 /* if there is a lockres associated with this
2854                  * mle, find it and set its owner to UNKNOWN */
2855                 hash = dlm_lockid_hash(mle->u.name.name, mle->u.name.len);
2856                 res = __dlm_lookup_lockres(dlm, mle->u.name.name,
2857                                            mle->u.name.len, hash);
2858                 if (res) {
2859                         /* unfortunately if we hit this rare case, our
2860                          * lock ordering is messed.  we need to drop
2861                          * the master lock so that we can take the
2862                          * lockres lock, meaning that we will have to
2863                          * restart from the head of list. */
2864                         spin_unlock(&dlm->master_lock);
2865
2866                         /* move lockres onto recovery list */
2867                         spin_lock(&res->spinlock);
2868                         dlm_set_lockres_owner(dlm, res,
2869                                         DLM_LOCK_RES_OWNER_UNKNOWN);
2870                         dlm_move_lockres_to_recovery_list(dlm, res);
2871                         spin_unlock(&res->spinlock);
2872                         dlm_lockres_put(res);
2873
2874                         /* about to get rid of mle, detach from heartbeat */
2875                         __dlm_mle_detach_hb_events(dlm, mle);
2876
2877                         /* dump the mle */
2878                         spin_lock(&dlm->master_lock);
2879                         __dlm_put_mle(mle);
2880                         spin_unlock(&dlm->master_lock);
2881
2882                         /* restart */
2883                         goto top;
2884                 }
2885
2886                 /* this may be the last reference */
2887                 __dlm_put_mle(mle);
2888         }
2889         spin_unlock(&dlm->master_lock);
2890 }
2891
2892
2893 int dlm_finish_migration(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
2894                          u8 old_master)
2895 {
2896         struct dlm_node_iter iter;
2897         int ret = 0;
2898
2899         spin_lock(&dlm->spinlock);
2900         dlm_node_iter_init(dlm->domain_map, &iter);
2901         clear_bit(old_master, iter.node_map);
2902         clear_bit(dlm->node_num, iter.node_map);
2903         spin_unlock(&dlm->spinlock);
2904
2905         mlog(0, "now time to do a migrate request to other nodes\n");
2906         ret = dlm_do_migrate_request(dlm, res, old_master,
2907                                      dlm->node_num, &iter);
2908         if (ret < 0) {
2909                 mlog_errno(ret);
2910                 goto leave;
2911         }
2912
2913         mlog(0, "doing assert master of %.*s to all except the original node\n",
2914              res->lockname.len, res->lockname.name);
2915         /* this call now finishes out the nodemap
2916          * even if one or more nodes die */
2917         ret = dlm_do_assert_master(dlm, res->lockname.name,
2918                                    res->lockname.len, iter.node_map,
2919                                    DLM_ASSERT_MASTER_FINISH_MIGRATION);
2920         if (ret < 0) {
2921                 /* no longer need to retry.  all living nodes contacted. */
2922                 mlog_errno(ret);
2923                 ret = 0;
2924         }
2925
2926         memset(iter.node_map, 0, sizeof(iter.node_map));
2927         set_bit(old_master, iter.node_map);
2928         mlog(0, "doing assert master of %.*s back to %u\n",
2929              res->lockname.len, res->lockname.name, old_master);
2930         ret = dlm_do_assert_master(dlm, res->lockname.name,
2931                                    res->lockname.len, iter.node_map,
2932                                    DLM_ASSERT_MASTER_FINISH_MIGRATION);
2933         if (ret < 0) {
2934                 mlog(0, "assert master to original master failed "
2935                      "with %d.\n", ret);
2936                 /* the only nonzero status here would be because of
2937                  * a dead original node.  we're done. */
2938                 ret = 0;
2939         }
2940
2941         /* all done, set the owner, clear the flag */
2942         spin_lock(&res->spinlock);
2943         dlm_set_lockres_owner(dlm, res, dlm->node_num);
2944         res->state &= ~DLM_LOCK_RES_MIGRATING;
2945         spin_unlock(&res->spinlock);
2946         /* re-dirty it on the new master */
2947         dlm_kick_thread(dlm, res);
2948         wake_up(&res->wq);
2949 leave:
2950         return ret;
2951 }
2952
2953 /*
2954  * LOCKRES AST REFCOUNT
2955  * this is integral to migration
2956  */
2957
2958 /* for future intent to call an ast, reserve one ahead of time.
2959  * this should be called only after waiting on the lockres
2960  * with dlm_wait_on_lockres, and while still holding the
2961  * spinlock after the call. */
2962 void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res)
2963 {
2964         assert_spin_locked(&res->spinlock);
2965         if (res->state & DLM_LOCK_RES_MIGRATING) {
2966                 __dlm_print_one_lock_resource(res);
2967         }
2968         BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
2969
2970         atomic_inc(&res->asts_reserved);
2971 }
2972
2973 /*
2974  * used to drop the reserved ast, either because it went unused,
2975  * or because the ast/bast was actually called.
2976  *
2977  * also, if there is a pending migration on this lockres,
2978  * and this was the last pending ast on the lockres,
2979  * atomically set the MIGRATING flag before we drop the lock.
2980  * this is how we ensure that migration can proceed with no
2981  * asts in progress.  note that it is ok if the state of the
2982  * queues is such that a lock should be granted in the future
2983  * or that a bast should be fired, because the new master will
2984  * shuffle the lists on this lockres as soon as it is migrated.
2985  */
2986 void dlm_lockres_release_ast(struct dlm_ctxt *dlm,
2987                              struct dlm_lock_resource *res)
2988 {
2989         if (!atomic_dec_and_lock(&res->asts_reserved, &res->spinlock))
2990                 return;
2991
2992         if (!res->migration_pending) {
2993                 spin_unlock(&res->spinlock);
2994                 return;
2995         }
2996
2997         BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
2998         res->migration_pending = 0;
2999         res->state |= DLM_LOCK_RES_MIGRATING;
3000         spin_unlock(&res->spinlock);
3001         wake_up(&res->wq);
3002         wake_up(&dlm->migration_wq);
3003 }