fs/ocfs2/dlm/dlmdomain.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * dlmdomain.c
   5  *
   6  * defines domain join / leave apis
   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 #include <linux/module.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/init.h>
  32 #include <linux/spinlock.h>
  33 #include <linux/delay.h>
  34 #include <linux/err.h>
  35 #include <linux/debugfs.h>
  36
  37 #include "cluster/heartbeat.h"
  38 #include "cluster/nodemanager.h"
  39 #include "cluster/tcp.h"
  40
  41 #include "dlmapi.h"
  42 #include "dlmcommon.h"
  43 #include "dlmdomain.h"
  44 #include "dlmdebug.h"
  45
  46 #include "dlmver.h"
  47
  48 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
  49 #include "cluster/masklog.h"
  50
  51 /*
  52  * ocfs2 node maps are array of long int, which limits to send them freely
  53  * across the wire due to endianness issues. To workaround this, we convert
  54  * long ints to byte arrays. Following 3 routines are helper functions to
  55  * set/test/copy bits within those array of bytes
  56  */
  57 static inline void byte_set_bit(u8 nr, u8 map[])
  58 {
  59         map[nr >> 3] |= (1UL << (nr & 7));
  60 }
  61
  62 static inline int byte_test_bit(u8 nr, u8 map[])
  63 {
  64         return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
  65 }
  66
  67 static inline void byte_copymap(u8 dmap[], unsigned long smap[],
  68                         unsigned int sz)
  69 {
  70         unsigned int nn;
  71
  72         if (!sz)
  73                 return;
  74
  75         memset(dmap, 0, ((sz + 7) >> 3));
  76         for (nn = 0 ; nn < sz; nn++)
  77                 if (test_bit(nn, smap))
  78                         byte_set_bit(nn, dmap);
  79 }
  80
  81 static void dlm_free_pagevec(void **vec, int pages)
  82 {
  83         while (pages--)
  84                 free_page((unsigned long)vec[pages]);
  85         kfree(vec);
  86 }
  87
  88 static void **dlm_alloc_pagevec(int pages)
  89 {
  90         void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL);
  91         int i;
  92
  93         if (!vec)
  94                 return NULL;
  95
  96         for (i = 0; i < pages; i++)
  97                 if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
  98                         goto out_free;
  99
 100         mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n",
 101              pages, (unsigned long)DLM_HASH_PAGES,
 102              (unsigned long)DLM_BUCKETS_PER_PAGE);
 103         return vec;
 104 out_free:
 105         dlm_free_pagevec(vec, i);
 106         return NULL;
 107 }
 108
 109 /*
 110  *
 111  * spinlock lock ordering: if multiple locks are needed, obey this ordering:
 112  *    dlm_domain_lock
 113  *    struct dlm_ctxt->spinlock
 114  *    struct dlm_lock_resource->spinlock
 115  *    struct dlm_ctxt->master_lock
 116  *    struct dlm_ctxt->ast_lock
 117  *    dlm_master_list_entry->spinlock
 118  *    dlm_lock->spinlock
 119  *
 120  */
 121
 122 DEFINE_SPINLOCK(dlm_domain_lock);
 123 LIST_HEAD(dlm_domains);
 124 static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events);
 125
 126 /*
 127  * The supported protocol version for DLM communication.  Running domains
 128  * will have a negotiated version with the same major number and a minor
 129  * number equal or smaller.  The dlm_ctxt->dlm_locking_proto field should
 130  * be used to determine what a running domain is actually using.
 131  *
 132  * New in version 1.1:
 133  *      - Message DLM_QUERY_REGION added to support global heartbeat
 134  *      - Message DLM_QUERY_NODEINFO added to allow online node removes
 135  */
 136 static const struct dlm_protocol_version dlm_protocol = {
 137         .pv_major = 1,
 138         .pv_minor = 1,
 139 };
 140
 141 #define DLM_DOMAIN_BACKOFF_MS 200
 142
 143 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
 144                                   void **ret_data);
 145 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
 146                                      void **ret_data);
 147 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
 148                                    void **ret_data);
 149 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
 150                                     void *data, void **ret_data);
 151 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
 152                                    void **ret_data);
 153 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
 154                                 struct dlm_protocol_version *request);
 155
 156 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
 157
 158 void __dlm_unhash_lockres(struct dlm_lock_resource *lockres)
 159 {
 160         if (!hlist_unhashed(&lockres->hash_node)) {
 161                 hlist_del_init(&lockres->hash_node);
 162                 dlm_lockres_put(lockres);
 163         }
 164 }
 165
 166 void __dlm_insert_lockres(struct dlm_ctxt *dlm,
 167                        struct dlm_lock_resource *res)
 168 {
 169         struct hlist_head *bucket;
 170         struct qstr *q;
 171
 172         assert_spin_locked(&dlm->spinlock);
 173
 174         q = &res->lockname;
 175         bucket = dlm_lockres_hash(dlm, q->hash);
 176
 177         /* get a reference for our hashtable */
 178         dlm_lockres_get(res);
 179
 180         hlist_add_head(&res->hash_node, bucket);
 181 }
 182
 183 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
 184                                                      const char *name,
 185                                                      unsigned int len,
 186                                                      unsigned int hash)
 187 {
 188         struct hlist_head *bucket;
 189         struct hlist_node *list;
 190
 191         mlog_entry("%.*s\n", len, name);
 192
 193         assert_spin_locked(&dlm->spinlock);
 194
 195         bucket = dlm_lockres_hash(dlm, hash);
 196
 197         hlist_for_each(list, bucket) {
 198                 struct dlm_lock_resource *res = hlist_entry(list,
 199                         struct dlm_lock_resource, hash_node);
 200                 if (res->lockname.name[0] != name[0])
 201                         continue;
 202                 if (unlikely(res->lockname.len != len))
 203                         continue;
 204                 if (memcmp(res->lockname.name + 1, name + 1, len - 1))
 205                         continue;
 206                 dlm_lockres_get(res);
 207                 return res;
 208         }
 209         return NULL;
 210 }
 211
 212 /* intended to be called by functions which do not care about lock
 213  * resources which are being purged (most net _handler functions).
 214  * this will return NULL for any lock resource which is found but
 215  * currently in the process of dropping its mastery reference.
 216  * use __dlm_lookup_lockres_full when you need the lock resource
 217  * regardless (e.g. dlm_get_lock_resource) */
 218 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
 219                                                 const char *name,
 220                                                 unsigned int len,
 221                                                 unsigned int hash)
 222 {
 223         struct dlm_lock_resource *res = NULL;
 224
 225         mlog_entry("%.*s\n", len, name);
 226
 227         assert_spin_locked(&dlm->spinlock);
 228
 229         res = __dlm_lookup_lockres_full(dlm, name, len, hash);
 230         if (res) {
 231                 spin_lock(&res->spinlock);
 232                 if (res->state & DLM_LOCK_RES_DROPPING_REF) {
 233                         spin_unlock(&res->spinlock);
 234                         dlm_lockres_put(res);
 235                         return NULL;
 236                 }
 237                 spin_unlock(&res->spinlock);
 238         }
 239
 240         return res;
 241 }
 242
 243 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
 244                                     const char *name,
 245                                     unsigned int len)
 246 {
 247         struct dlm_lock_resource *res;
 248         unsigned int hash = dlm_lockid_hash(name, len);
 249
 250         spin_lock(&dlm->spinlock);
 251         res = __dlm_lookup_lockres(dlm, name, len, hash);
 252         spin_unlock(&dlm->spinlock);
 253         return res;
 254 }
 255
 256 static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
 257 {
 258         struct dlm_ctxt *tmp = NULL;
 259         struct list_head *iter;
 260
 261         assert_spin_locked(&dlm_domain_lock);
 262
 263         /* tmp->name here is always NULL terminated,
 264          * but domain may not be! */
 265         list_for_each(iter, &dlm_domains) {
 266                 tmp = list_entry (iter, struct dlm_ctxt, list);
 267                 if (strlen(tmp->name) == len &&
 268                     memcmp(tmp->name, domain, len)==0)
 269                         break;
 270                 tmp = NULL;
 271         }
 272
 273         return tmp;
 274 }
 275
 276 /* For null terminated domain strings ONLY */
 277 static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
 278 {
 279         assert_spin_locked(&dlm_domain_lock);
 280
 281         return __dlm_lookup_domain_full(domain, strlen(domain));
 282 }
 283
 284
 285 /* returns true on one of two conditions:
 286  * 1) the domain does not exist
 287  * 2) the domain exists and it's state is "joined" */
 288 static int dlm_wait_on_domain_helper(const char *domain)
 289 {
 290         int ret = 0;
 291         struct dlm_ctxt *tmp = NULL;
 292
 293         spin_lock(&dlm_domain_lock);
 294
 295         tmp = __dlm_lookup_domain(domain);
 296         if (!tmp)
 297                 ret = 1;
 298         else if (tmp->dlm_state == DLM_CTXT_JOINED)
 299                 ret = 1;
 300
 301         spin_unlock(&dlm_domain_lock);
 302         return ret;
 303 }
 304
 305 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
 306 {
 307         dlm_destroy_debugfs_subroot(dlm);
 308
 309         if (dlm->lockres_hash)
 310                 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
 311
 312         if (dlm->master_hash)
 313                 dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
 314
 315         if (dlm->name)
 316                 kfree(dlm->name);
 317
 318         kfree(dlm);
 319 }
 320
 321 /* A little strange - this function will be called while holding
 322  * dlm_domain_lock and is expected to be holding it on the way out. We
 323  * will however drop and reacquire it multiple times */
 324 static void dlm_ctxt_release(struct kref *kref)
 325 {
 326         struct dlm_ctxt *dlm;
 327
 328         dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
 329
 330         BUG_ON(dlm->num_joins);
 331         BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
 332
 333         /* we may still be in the list if we hit an error during join. */
 334         list_del_init(&dlm->list);
 335
 336         spin_unlock(&dlm_domain_lock);
 337
 338         mlog(0, "freeing memory from domain %s\n", dlm->name);
 339
 340         wake_up(&dlm_domain_events);
 341
 342         dlm_free_ctxt_mem(dlm);
 343
 344         spin_lock(&dlm_domain_lock);
 345 }
 346
 347 void dlm_put(struct dlm_ctxt *dlm)
 348 {
 349         spin_lock(&dlm_domain_lock);
 350         kref_put(&dlm->dlm_refs, dlm_ctxt_release);
 351         spin_unlock(&dlm_domain_lock);
 352 }
 353
 354 static void __dlm_get(struct dlm_ctxt *dlm)
 355 {
 356         kref_get(&dlm->dlm_refs);
 357 }
 358
 359 /* given a questionable reference to a dlm object, gets a reference if
 360  * it can find it in the list, otherwise returns NULL in which case
 361  * you shouldn't trust your pointer. */
 362 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
 363 {
 364         struct list_head *iter;
 365         struct dlm_ctxt *target = NULL;
 366
 367         spin_lock(&dlm_domain_lock);
 368
 369         list_for_each(iter, &dlm_domains) {
 370                 target = list_entry (iter, struct dlm_ctxt, list);
 371
 372                 if (target == dlm) {
 373                         __dlm_get(target);
 374                         break;
 375                 }
 376
 377                 target = NULL;
 378         }
 379
 380         spin_unlock(&dlm_domain_lock);
 381
 382         return target;
 383 }
 384
 385 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
 386 {
 387         int ret;
 388
 389         spin_lock(&dlm_domain_lock);
 390         ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
 391                 (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
 392         spin_unlock(&dlm_domain_lock);
 393
 394         return ret;
 395 }
 396
 397 static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
 398 {
 399         if (dlm->dlm_worker) {
 400                 flush_workqueue(dlm->dlm_worker);
 401                 destroy_workqueue(dlm->dlm_worker);
 402                 dlm->dlm_worker = NULL;
 403         }
 404 }
 405
 406 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
 407 {
 408         dlm_unregister_domain_handlers(dlm);
 409         dlm_debug_shutdown(dlm);
 410         dlm_complete_thread(dlm);
 411         dlm_complete_recovery_thread(dlm);
 412         dlm_destroy_dlm_worker(dlm);
 413
 414         /* We've left the domain. Now we can take ourselves out of the
 415          * list and allow the kref stuff to help us free the
 416          * memory. */
 417         spin_lock(&dlm_domain_lock);
 418         list_del_init(&dlm->list);
 419         spin_unlock(&dlm_domain_lock);
 420
 421         /* Wake up anyone waiting for us to remove this domain */
 422         wake_up(&dlm_domain_events);
 423 }
 424
 425 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
 426 {
 427         int i, num, n, ret = 0;
 428         struct dlm_lock_resource *res;
 429         struct hlist_node *iter;
 430         struct hlist_head *bucket;
 431         int dropped;
 432
 433         mlog(0, "Migrating locks from domain %s\n", dlm->name);
 434
 435         num = 0;
 436         spin_lock(&dlm->spinlock);
 437         for (i = 0; i < DLM_HASH_BUCKETS; i++) {
 438 redo_bucket:
 439                 n = 0;
 440                 bucket = dlm_lockres_hash(dlm, i);
 441                 iter = bucket->first;
 442                 while (iter) {
 443                         n++;
 444                         res = hlist_entry(iter, struct dlm_lock_resource,
 445                                           hash_node);
 446                         dlm_lockres_get(res);
 447                         /* migrate, if necessary.  this will drop the dlm
 448                          * spinlock and retake it if it does migration. */
 449                         dropped = dlm_empty_lockres(dlm, res);
 450
 451                         spin_lock(&res->spinlock);
 452                         __dlm_lockres_calc_usage(dlm, res);
 453                         iter = res->hash_node.next;
 454                         spin_unlock(&res->spinlock);
 455
 456                         dlm_lockres_put(res);
 457
 458                         if (dropped)
 459                                 goto redo_bucket;
 460                 }
 461                 cond_resched_lock(&dlm->spinlock);
 462                 num += n;
 463         }
 464         spin_unlock(&dlm->spinlock);
 465         wake_up(&dlm->dlm_thread_wq);
 466
 467         /* let the dlm thread take care of purging, keep scanning until
 468          * nothing remains in the hash */
 469         if (num) {
 470                 mlog(0, "%s: %d lock resources in hash last pass\n",
 471                      dlm->name, num);
 472                 ret = -EAGAIN;
 473         }
 474         mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
 475         return ret;
 476 }
 477
 478 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
 479 {
 480         int ret;
 481
 482         spin_lock(&dlm->spinlock);
 483         ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
 484         spin_unlock(&dlm->spinlock);
 485
 486         return ret;
 487 }
 488
 489 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
 490 {
 491         /* Yikes, a double spinlock! I need domain_lock for the dlm
 492          * state and the dlm spinlock for join state... Sorry! */
 493 again:
 494         spin_lock(&dlm_domain_lock);
 495         spin_lock(&dlm->spinlock);
 496
 497         if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
 498                 mlog(0, "Node %d is joining, we wait on it.\n",
 499                           dlm->joining_node);
 500                 spin_unlock(&dlm->spinlock);
 501                 spin_unlock(&dlm_domain_lock);
 502
 503                 wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
 504                 goto again;
 505         }
 506
 507         dlm->dlm_state = DLM_CTXT_LEAVING;
 508         spin_unlock(&dlm->spinlock);
 509         spin_unlock(&dlm_domain_lock);
 510 }
 511
 512 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
 513 {
 514         int node = -1;
 515
 516         assert_spin_locked(&dlm->spinlock);
 517
 518         printk(KERN_NOTICE "o2dlm: Nodes in domain %s: ", dlm->name);
 519
 520         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
 521                                      node + 1)) < O2NM_MAX_NODES) {
 522                 printk("%d ", node);
 523         }
 524         printk("\n");
 525 }
 526
 527 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
 528                                    void **ret_data)
 529 {
 530         struct dlm_ctxt *dlm = data;
 531         unsigned int node;
 532         struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
 533
 534         mlog_entry("%p %u %p", msg, len, data);
 535
 536         if (!dlm_grab(dlm))
 537                 return 0;
 538
 539         node = exit_msg->node_idx;
 540
 541         printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s\n", node, dlm->name);
 542
 543         spin_lock(&dlm->spinlock);
 544         clear_bit(node, dlm->domain_map);
 545         __dlm_print_nodes(dlm);
 546
 547         /* notify anything attached to the heartbeat events */
 548         dlm_hb_event_notify_attached(dlm, node, 0);
 549
 550         spin_unlock(&dlm->spinlock);
 551
 552         dlm_put(dlm);
 553
 554         return 0;
 555 }
 556
 557 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm,
 558                                     unsigned int node)
 559 {
 560         int status;
 561         struct dlm_exit_domain leave_msg;
 562
 563         mlog(0, "Asking node %u if we can leave the domain %s me = %u\n",
 564                   node, dlm->name, dlm->node_num);
 565
 566         memset(&leave_msg, 0, sizeof(leave_msg));
 567         leave_msg.node_idx = dlm->node_num;
 568
 569         status = o2net_send_message(DLM_EXIT_DOMAIN_MSG, dlm->key,
 570                                     &leave_msg, sizeof(leave_msg), node,
 571                                     NULL);
 572         if (status < 0)
 573                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
 574                      "node %u\n", status, DLM_EXIT_DOMAIN_MSG, dlm->key, node);
 575         mlog(0, "status return %d from o2net_send_message\n", status);
 576
 577         return status;
 578 }
 579
 580
 581 static void dlm_leave_domain(struct dlm_ctxt *dlm)
 582 {
 583         int node, clear_node, status;
 584
 585         /* At this point we've migrated away all our locks and won't
 586          * accept mastership of new ones. The dlm is responsible for
 587          * almost nothing now. We make sure not to confuse any joining
 588          * nodes and then commence shutdown procedure. */
 589
 590         spin_lock(&dlm->spinlock);
 591         /* Clear ourselves from the domain map */
 592         clear_bit(dlm->node_num, dlm->domain_map);
 593         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
 594                                      0)) < O2NM_MAX_NODES) {
 595                 /* Drop the dlm spinlock. This is safe wrt the domain_map.
 596                  * -nodes cannot be added now as the
 597                  *   query_join_handlers knows to respond with OK_NO_MAP
 598                  * -we catch the right network errors if a node is
 599                  *   removed from the map while we're sending him the
 600                  *   exit message. */
 601                 spin_unlock(&dlm->spinlock);
 602
 603                 clear_node = 1;
 604
 605                 status = dlm_send_one_domain_exit(dlm, node);
 606                 if (status < 0 &&
 607                     status != -ENOPROTOOPT &&
 608                     status != -ENOTCONN) {
 609                         mlog(ML_NOTICE, "Error %d sending domain exit message "
 610                              "to node %d\n", status, node);
 611
 612                         /* Not sure what to do here but lets sleep for
 613                          * a bit in case this was a transient
 614                          * error... */
 615                         msleep(DLM_DOMAIN_BACKOFF_MS);
 616                         clear_node = 0;
 617                 }
 618
 619                 spin_lock(&dlm->spinlock);
 620                 /* If we're not clearing the node bit then we intend
 621                  * to loop back around to try again. */
 622                 if (clear_node)
 623                         clear_bit(node, dlm->domain_map);
 624         }
 625         spin_unlock(&dlm->spinlock);
 626 }
 627
 628 int dlm_joined(struct dlm_ctxt *dlm)
 629 {
 630         int ret = 0;
 631
 632         spin_lock(&dlm_domain_lock);
 633
 634         if (dlm->dlm_state == DLM_CTXT_JOINED)
 635                 ret = 1;
 636
 637         spin_unlock(&dlm_domain_lock);
 638
 639         return ret;
 640 }
 641
 642 int dlm_shutting_down(struct dlm_ctxt *dlm)
 643 {
 644         int ret = 0;
 645
 646         spin_lock(&dlm_domain_lock);
 647
 648         if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
 649                 ret = 1;
 650
 651         spin_unlock(&dlm_domain_lock);
 652
 653         return ret;
 654 }
 655
 656 void dlm_unregister_domain(struct dlm_ctxt *dlm)
 657 {
 658         int leave = 0;
 659         struct dlm_lock_resource *res;
 660
 661         spin_lock(&dlm_domain_lock);
 662         BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
 663         BUG_ON(!dlm->num_joins);
 664
 665         dlm->num_joins--;
 666         if (!dlm->num_joins) {
 667                 /* We mark it "in shutdown" now so new register
 668                  * requests wait until we've completely left the
 669                  * domain. Don't use DLM_CTXT_LEAVING yet as we still
 670                  * want new domain joins to communicate with us at
 671                  * least until we've completed migration of our
 672                  * resources. */
 673                 dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
 674                 leave = 1;
 675         }
 676         spin_unlock(&dlm_domain_lock);
 677
 678         if (leave) {
 679                 mlog(0, "shutting down domain %s\n", dlm->name);
 680
 681                 /* We changed dlm state, notify the thread */
 682                 dlm_kick_thread(dlm, NULL);
 683
 684                 while (dlm_migrate_all_locks(dlm)) {
 685                         /* Give dlm_thread time to purge the lockres' */
 686                         msleep(500);
 687                         mlog(0, "%s: more migration to do\n", dlm->name);
 688                 }
 689
 690                 /* This list should be empty. If not, print remaining lockres */
 691                 if (!list_empty(&dlm->tracking_list)) {
 692                         mlog(ML_ERROR, "Following lockres' are still on the "
 693                              "tracking list:\n");
 694                         list_for_each_entry(res, &dlm->tracking_list, tracking)
 695                                 dlm_print_one_lock_resource(res);
 696                 }
 697
 698                 dlm_mark_domain_leaving(dlm);
 699                 dlm_leave_domain(dlm);
 700                 dlm_force_free_mles(dlm);
 701                 dlm_complete_dlm_shutdown(dlm);
 702         }
 703         dlm_put(dlm);
 704 }
 705 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
 706
 707 static int dlm_query_join_proto_check(char *proto_type, int node,
 708                                       struct dlm_protocol_version *ours,
 709                                       struct dlm_protocol_version *request)
 710 {
 711         int rc;
 712         struct dlm_protocol_version proto = *request;
 713
 714         if (!dlm_protocol_compare(ours, &proto)) {
 715                 mlog(0,
 716                      "node %u wanted to join with %s locking protocol "
 717                      "%u.%u, we respond with %u.%u\n",
 718                      node, proto_type,
 719                      request->pv_major,
 720                      request->pv_minor,
 721                      proto.pv_major, proto.pv_minor);
 722                 request->pv_minor = proto.pv_minor;
 723                 rc = 0;
 724         } else {
 725                 mlog(ML_NOTICE,
 726                      "Node %u wanted to join with %s locking "
 727                      "protocol %u.%u, but we have %u.%u, disallowing\n",
 728                      node, proto_type,
 729                      request->pv_major,
 730                      request->pv_minor,
 731                      ours->pv_major,
 732                      ours->pv_minor);
 733                 rc = 1;
 734         }
 735
 736         return rc;
 737 }
 738
 739 /*
 740  * struct dlm_query_join_packet is made up of four one-byte fields.  They
 741  * are effectively in big-endian order already.  However, little-endian
 742  * machines swap them before putting the packet on the wire (because
 743  * query_join's response is a status, and that status is treated as a u32
 744  * on the wire).  Thus, a big-endian and little-endian machines will treat
 745  * this structure differently.
 746  *
 747  * The solution is to have little-endian machines swap the structure when
 748  * converting from the structure to the u32 representation.  This will
 749  * result in the structure having the correct format on the wire no matter
 750  * the host endian format.
 751  */
 752 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
 753                                           u32 *wire)
 754 {
 755         union dlm_query_join_response response;
 756
 757         response.packet = *packet;
 758         *wire = cpu_to_be32(response.intval);
 759 }
 760
 761 static void dlm_query_join_wire_to_packet(u32 wire,
 762                                           struct dlm_query_join_packet *packet)
 763 {
 764         union dlm_query_join_response response;
 765
 766         response.intval = cpu_to_be32(wire);
 767         *packet = response.packet;
 768 }
 769
 770 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
 771                                   void **ret_data)
 772 {
 773         struct dlm_query_join_request *query;
 774         struct dlm_query_join_packet packet = {
 775                 .code = JOIN_DISALLOW,
 776         };
 777         struct dlm_ctxt *dlm = NULL;
 778         u32 response;
 779         u8 nodenum;
 780
 781         query = (struct dlm_query_join_request *) msg->buf;
 782
 783         mlog(0, "node %u wants to join domain %s\n", query->node_idx,
 784                   query->domain);
 785
 786         /*
 787          * If heartbeat doesn't consider the node live, tell it
 788          * to back off and try again.  This gives heartbeat a chance
 789          * to catch up.
 790          */
 791         if (!o2hb_check_node_heartbeating(query->node_idx)) {
 792                 mlog(0, "node %u is not in our live map yet\n",
 793                      query->node_idx);
 794
 795                 packet.code = JOIN_DISALLOW;
 796                 goto respond;
 797         }
 798
 799         packet.code = JOIN_OK_NO_MAP;
 800
 801         spin_lock(&dlm_domain_lock);
 802         dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
 803         if (!dlm)
 804                 goto unlock_respond;
 805
 806         /*
 807          * There is a small window where the joining node may not see the
 808          * node(s) that just left but still part of the cluster. DISALLOW
 809          * join request if joining node has different node map.
 810          */
 811         nodenum=0;
 812         while (nodenum < O2NM_MAX_NODES) {
 813                 if (test_bit(nodenum, dlm->domain_map)) {
 814                         if (!byte_test_bit(nodenum, query->node_map)) {
 815                                 mlog(0, "disallow join as node %u does not "
 816                                      "have node %u in its nodemap\n",
 817                                      query->node_idx, nodenum);
 818                                 packet.code = JOIN_DISALLOW;
 819                                 goto unlock_respond;
 820                         }
 821                 }
 822                 nodenum++;
 823         }
 824
 825         /* Once the dlm ctxt is marked as leaving then we don't want
 826          * to be put in someone's domain map.
 827          * Also, explicitly disallow joining at certain troublesome
 828          * times (ie. during recovery). */
 829         if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) {
 830                 int bit = query->node_idx;
 831                 spin_lock(&dlm->spinlock);
 832
 833                 if (dlm->dlm_state == DLM_CTXT_NEW &&
 834                     dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
 835                         /*If this is a brand new context and we
 836                          * haven't started our join process yet, then
 837                          * the other node won the race. */
 838                         packet.code = JOIN_OK_NO_MAP;
 839                 } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
 840                         /* Disallow parallel joins. */
 841                         packet.code = JOIN_DISALLOW;
 842                 } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
 843                         mlog(0, "node %u trying to join, but recovery "
 844                              "is ongoing.\n", bit);
 845                         packet.code = JOIN_DISALLOW;
 846                 } else if (test_bit(bit, dlm->recovery_map)) {
 847                         mlog(0, "node %u trying to join, but it "
 848                              "still needs recovery.\n", bit);
 849                         packet.code = JOIN_DISALLOW;
 850                 } else if (test_bit(bit, dlm->domain_map)) {
 851                         mlog(0, "node %u trying to join, but it "
 852                              "is still in the domain! needs recovery?\n",
 853                              bit);
 854                         packet.code = JOIN_DISALLOW;
 855                 } else {
 856                         /* Alright we're fully a part of this domain
 857                          * so we keep some state as to who's joining
 858                          * and indicate to him that needs to be fixed
 859                          * up. */
 860
 861                         /* Make sure we speak compatible locking protocols.  */
 862                         if (dlm_query_join_proto_check("DLM", bit,
 863                                                        &dlm->dlm_locking_proto,
 864                                                        &query->dlm_proto)) {
 865                                 packet.code = JOIN_PROTOCOL_MISMATCH;
 866                         } else if (dlm_query_join_proto_check("fs", bit,
 867                                                               &dlm->fs_locking_proto,
 868                                                               &query->fs_proto)) {
 869                                 packet.code = JOIN_PROTOCOL_MISMATCH;
 870                         } else {
 871                                 packet.dlm_minor = query->dlm_proto.pv_minor;
 872                                 packet.fs_minor = query->fs_proto.pv_minor;
 873                                 packet.code = JOIN_OK;
 874                                 __dlm_set_joining_node(dlm, query->node_idx);
 875                         }
 876                 }
 877
 878                 spin_unlock(&dlm->spinlock);
 879         }
 880 unlock_respond:
 881         spin_unlock(&dlm_domain_lock);
 882
 883 respond:
 884         mlog(0, "We respond with %u\n", packet.code);
 885
 886         dlm_query_join_packet_to_wire(&packet, &response);
 887         return response;
 888 }
 889
 890 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
 891                                      void **ret_data)
 892 {
 893         struct dlm_assert_joined *assert;
 894         struct dlm_ctxt *dlm = NULL;
 895
 896         assert = (struct dlm_assert_joined *) msg->buf;
 897
 898         mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
 899                   assert->domain);
 900
 901         spin_lock(&dlm_domain_lock);
 902         dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
 903         /* XXX should we consider no dlm ctxt an error? */
 904         if (dlm) {
 905                 spin_lock(&dlm->spinlock);
 906
 907                 /* Alright, this node has officially joined our
 908                  * domain. Set him in the map and clean up our
 909                  * leftover join state. */
 910                 BUG_ON(dlm->joining_node != assert->node_idx);
 911                 set_bit(assert->node_idx, dlm->domain_map);
 912                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
 913
 914                 printk(KERN_NOTICE "o2dlm: Node %u joins domain %s\n",
 915                        assert->node_idx, dlm->name);
 916                 __dlm_print_nodes(dlm);
 917
 918                 /* notify anything attached to the heartbeat events */
 919                 dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
 920
 921                 spin_unlock(&dlm->spinlock);
 922         }
 923         spin_unlock(&dlm_domain_lock);
 924
 925         return 0;
 926 }
 927
 928 static int dlm_match_regions(struct dlm_ctxt *dlm,
 929                              struct dlm_query_region *qr)
 930 {
 931         char *local = NULL, *remote = qr->qr_regions;
 932         char *l, *r;
 933         int localnr, i, j, foundit;
 934         int status = 0;
 935
 936         if (!o2hb_global_heartbeat_active()) {
 937                 if (qr->qr_numregions) {
 938                         mlog(ML_ERROR, "Domain %s: Joining node %d has global "
 939                              "heartbeat enabled but local node %d does not\n",
 940                              qr->qr_domain, qr->qr_node, dlm->node_num);
 941                         status = -EINVAL;
 942                 }
 943                 goto bail;
 944         }
 945
 946         if (o2hb_global_heartbeat_active() && !qr->qr_numregions) {
 947                 mlog(ML_ERROR, "Domain %s: Local node %d has global "
 948                      "heartbeat enabled but joining node %d does not\n",
 949                      qr->qr_domain, dlm->node_num, qr->qr_node);
 950                 status = -EINVAL;
 951                 goto bail;
 952         }
 953
 954         r = remote;
 955         for (i = 0; i < qr->qr_numregions; ++i) {
 956                 mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r);
 957                 r += O2HB_MAX_REGION_NAME_LEN;
 958         }
 959
 960         local = kmalloc(sizeof(qr->qr_regions), GFP_ATOMIC);
 961         if (!local) {
 962                 status = -ENOMEM;
 963                 goto bail;
 964         }
 965
 966         localnr = o2hb_get_all_regions(local, O2NM_MAX_REGIONS);
 967
 968         /* compare local regions with remote */
 969         l = local;
 970         for (i = 0; i < localnr; ++i) {
 971                 foundit = 0;
 972                 r = remote;
 973                 for (j = 0; j <= qr->qr_numregions; ++j) {
 974                         if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) {
 975                                 foundit = 1;
 976                                 break;
 977                         }
 978                         r += O2HB_MAX_REGION_NAME_LEN;
 979                 }
 980                 if (!foundit) {
 981                         status = -EINVAL;
 982                         mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
 983                              "in local node %d but not in joining node %d\n",
 984                              qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l,
 985                              dlm->node_num, qr->qr_node);
 986                         goto bail;
 987                 }
 988                 l += O2HB_MAX_REGION_NAME_LEN;
 989         }
 990
 991         /* compare remote with local regions */
 992         r = remote;
 993         for (i = 0; i < qr->qr_numregions; ++i) {
 994                 foundit = 0;
 995                 l = local;
 996                 for (j = 0; j < localnr; ++j) {
 997                         if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) {
 998                                 foundit = 1;
 999                                 break;
1000                         }
1001                         l += O2HB_MAX_REGION_NAME_LEN;
1002                 }
1003                 if (!foundit) {
1004                         status = -EINVAL;
1005                         mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1006                              "in joining node %d but not in local node %d\n",
1007                              qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r,
1008                              qr->qr_node, dlm->node_num);
1009                         goto bail;
1010                 }
1011                 r += O2HB_MAX_REGION_NAME_LEN;
1012         }
1013
1014 bail:
1015         kfree(local);
1016
1017         return status;
1018 }
1019
1020 static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map)
1021 {
1022         struct dlm_query_region *qr = NULL;
1023         int status, ret = 0, i;
1024         char *p;
1025
1026         if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1027                 goto bail;
1028
1029         qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL);
1030         if (!qr) {
1031                 ret = -ENOMEM;
1032                 mlog_errno(ret);
1033                 goto bail;
1034         }
1035
1036         qr->qr_node = dlm->node_num;
1037         qr->qr_namelen = strlen(dlm->name);
1038         memcpy(qr->qr_domain, dlm->name, qr->qr_namelen);
1039         /* if local hb, the numregions will be zero */
1040         if (o2hb_global_heartbeat_active())
1041                 qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions,
1042                                                          O2NM_MAX_REGIONS);
1043
1044         p = qr->qr_regions;
1045         for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN)
1046                 mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p);
1047
1048         i = -1;
1049         while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1050                                   i + 1)) < O2NM_MAX_NODES) {
1051                 if (i == dlm->node_num)
1052                         continue;
1053
1054                 mlog(0, "Sending regions to node %d\n", i);
1055
1056                 ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr,
1057                                          sizeof(struct dlm_query_region),
1058                                          i, &status);
1059                 if (ret >= 0)
1060                         ret = status;
1061                 if (ret) {
1062                         mlog(ML_ERROR, "Region mismatch %d, node %d\n",
1063                              ret, i);
1064                         break;
1065                 }
1066         }
1067
1068 bail:
1069         kfree(qr);
1070         return ret;
1071 }
1072
1073 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
1074                                     void *data, void **ret_data)
1075 {
1076         struct dlm_query_region *qr;
1077         struct dlm_ctxt *dlm = NULL;
1078         int status = 0;
1079         int locked = 0;
1080
1081         qr = (struct dlm_query_region *) msg->buf;
1082
1083         mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node,
1084              qr->qr_domain);
1085
1086         status = -EINVAL;
1087
1088         spin_lock(&dlm_domain_lock);
1089         dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen);
1090         if (!dlm) {
1091                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1092                      "before join domain\n", qr->qr_node, qr->qr_domain);
1093                 goto bail;
1094         }
1095
1096         spin_lock(&dlm->spinlock);
1097         locked = 1;
1098         if (dlm->joining_node != qr->qr_node) {
1099                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1100                      "but joining node is %d\n", qr->qr_node, qr->qr_domain,
1101                      dlm->joining_node);
1102                 goto bail;
1103         }
1104
1105         /* Support for global heartbeat was added in 1.1 */
1106         if (dlm->dlm_locking_proto.pv_major == 1 &&
1107             dlm->dlm_locking_proto.pv_minor == 0) {
1108                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1109                      "but active dlm protocol is %d.%d\n", qr->qr_node,
1110                      qr->qr_domain, dlm->dlm_locking_proto.pv_major,
1111                      dlm->dlm_locking_proto.pv_minor);
1112                 goto bail;
1113         }
1114
1115         status = dlm_match_regions(dlm, qr);
1116
1117 bail:
1118         if (locked)
1119                 spin_unlock(&dlm->spinlock);
1120         spin_unlock(&dlm_domain_lock);
1121
1122         return status;
1123 }
1124
1125 static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn)
1126 {
1127         struct o2nm_node *local;
1128         struct dlm_node_info *remote;
1129         int i, j;
1130         int status = 0;
1131
1132         for (j = 0; j < qn->qn_numnodes; ++j)
1133                 mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum,
1134                      &(qn->qn_nodes[j].ni_ipv4_address),
1135                      ntohs(qn->qn_nodes[j].ni_ipv4_port));
1136
1137         for (i = 0; i < O2NM_MAX_NODES && !status; ++i) {
1138                 local = o2nm_get_node_by_num(i);
1139                 remote = NULL;
1140                 for (j = 0; j < qn->qn_numnodes; ++j) {
1141                         if (qn->qn_nodes[j].ni_nodenum == i) {
1142                                 remote = &(qn->qn_nodes[j]);
1143                                 break;
1144                         }
1145                 }
1146
1147                 if (!local && !remote)
1148                         continue;
1149
1150                 if ((local && !remote) || (!local && remote))
1151                         status = -EINVAL;
1152
1153                 if (!status &&
1154                     ((remote->ni_nodenum != local->nd_num) ||
1155                      (remote->ni_ipv4_port != local->nd_ipv4_port) ||
1156                      (remote->ni_ipv4_address != local->nd_ipv4_address)))
1157                         status = -EINVAL;
1158
1159                 if (status) {
1160                         if (remote && !local)
1161                                 mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1162                                      "registered in joining node %d but not in "
1163                                      "local node %d\n", qn->qn_domain,
1164                                      remote->ni_nodenum,
1165                                      &(remote->ni_ipv4_address),
1166                                      ntohs(remote->ni_ipv4_port),
1167                                      qn->qn_nodenum, dlm->node_num);
1168                         if (local && !remote)
1169                                 mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1170                                      "registered in local node %d but not in "
1171                                      "joining node %d\n", qn->qn_domain,
1172                                      local->nd_num, &(local->nd_ipv4_address),
1173                                      ntohs(local->nd_ipv4_port),
1174                                      dlm->node_num, qn->qn_nodenum);
1175                         BUG_ON((!local && !remote));
1176                 }
1177
1178                 if (local)
1179                         o2nm_node_put(local);
1180         }
1181
1182         return status;
1183 }
1184
1185 static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map)
1186 {
1187         struct dlm_query_nodeinfo *qn = NULL;
1188         struct o2nm_node *node;
1189         int ret = 0, status, count, i;
1190
1191         if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1192                 goto bail;
1193
1194         qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL);
1195         if (!qn) {
1196                 ret = -ENOMEM;
1197                 mlog_errno(ret);
1198                 goto bail;
1199         }
1200
1201         for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) {
1202                 node = o2nm_get_node_by_num(i);
1203                 if (!node)
1204                         continue;
1205                 qn->qn_nodes[count].ni_nodenum = node->nd_num;
1206                 qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port;
1207                 qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address;
1208                 mlog(0, "Node %3d, %pI4:%u\n", node->nd_num,
1209                      &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port));
1210                 ++count;
1211                 o2nm_node_put(node);
1212         }
1213
1214         qn->qn_nodenum = dlm->node_num;
1215         qn->qn_numnodes = count;
1216         qn->qn_namelen = strlen(dlm->name);
1217         memcpy(qn->qn_domain, dlm->name, qn->qn_namelen);
1218
1219         i = -1;
1220         while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1221                                   i + 1)) < O2NM_MAX_NODES) {
1222                 if (i == dlm->node_num)
1223                         continue;
1224
1225                 mlog(0, "Sending nodeinfo to node %d\n", i);
1226
1227                 ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
1228                                          qn, sizeof(struct dlm_query_nodeinfo),
1229                                          i, &status);
1230                 if (ret >= 0)
1231                         ret = status;
1232                 if (ret) {
1233                         mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i);
1234                         break;
1235                 }
1236         }
1237
1238 bail:
1239         kfree(qn);
1240         return ret;
1241 }
1242
1243 static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len,
1244                                       void *data, void **ret_data)
1245 {
1246         struct dlm_query_nodeinfo *qn;
1247         struct dlm_ctxt *dlm = NULL;
1248         int locked = 0, status = -EINVAL;
1249
1250         qn = (struct dlm_query_nodeinfo *) msg->buf;
1251
1252         mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum,
1253              qn->qn_domain);
1254
1255         spin_lock(&dlm_domain_lock);
1256         dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen);
1257         if (!dlm) {
1258                 mlog(ML_ERROR, "Node %d queried nodes on domain %s before "
1259                      "join domain\n", qn->qn_nodenum, qn->qn_domain);
1260                 goto bail;
1261         }
1262
1263         spin_lock(&dlm->spinlock);
1264         locked = 1;
1265         if (dlm->joining_node != qn->qn_nodenum) {
1266                 mlog(ML_ERROR, "Node %d queried nodes on domain %s but "
1267                      "joining node is %d\n", qn->qn_nodenum, qn->qn_domain,
1268                      dlm->joining_node);
1269                 goto bail;
1270         }
1271
1272         /* Support for node query was added in 1.1 */
1273         if (dlm->dlm_locking_proto.pv_major == 1 &&
1274             dlm->dlm_locking_proto.pv_minor == 0) {
1275                 mlog(ML_ERROR, "Node %d queried nodes on domain %s "
1276                      "but active dlm protocol is %d.%d\n", qn->qn_nodenum,
1277                      qn->qn_domain, dlm->dlm_locking_proto.pv_major,
1278                      dlm->dlm_locking_proto.pv_minor);
1279                 goto bail;
1280         }
1281
1282         status = dlm_match_nodes(dlm, qn);
1283
1284 bail:
1285         if (locked)
1286                 spin_unlock(&dlm->spinlock);
1287         spin_unlock(&dlm_domain_lock);
1288
1289         return status;
1290 }
1291
1292 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
1293                                    void **ret_data)
1294 {
1295         struct dlm_cancel_join *cancel;
1296         struct dlm_ctxt *dlm = NULL;
1297
1298         cancel = (struct dlm_cancel_join *) msg->buf;
1299
1300         mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
1301                   cancel->domain);
1302
1303         spin_lock(&dlm_domain_lock);
1304         dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
1305
1306         if (dlm) {
1307                 spin_lock(&dlm->spinlock);
1308
1309                 /* Yikes, this guy wants to cancel his join. No
1310                  * problem, we simply cleanup our join state. */
1311                 BUG_ON(dlm->joining_node != cancel->node_idx);
1312                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1313
1314                 spin_unlock(&dlm->spinlock);
1315         }
1316         spin_unlock(&dlm_domain_lock);
1317
1318         return 0;
1319 }
1320
1321 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
1322                                     unsigned int node)
1323 {
1324         int status;
1325         struct dlm_cancel_join cancel_msg;
1326
1327         memset(&cancel_msg, 0, sizeof(cancel_msg));
1328         cancel_msg.node_idx = dlm->node_num;
1329         cancel_msg.name_len = strlen(dlm->name);
1330         memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
1331
1332         status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1333                                     &cancel_msg, sizeof(cancel_msg), node,
1334                                     NULL);
1335         if (status < 0) {
1336                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1337                      "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1338                      node);
1339                 goto bail;
1340         }
1341
1342 bail:
1343         return status;
1344 }
1345
1346 /* map_size should be in bytes. */
1347 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
1348                                  unsigned long *node_map,
1349                                  unsigned int map_size)
1350 {
1351         int status, tmpstat;
1352         unsigned int node;
1353
1354         if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
1355                          sizeof(unsigned long))) {
1356                 mlog(ML_ERROR,
1357                      "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
1358                      map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
1359                 return -EINVAL;
1360         }
1361
1362         status = 0;
1363         node = -1;
1364         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1365                                      node + 1)) < O2NM_MAX_NODES) {
1366                 if (node == dlm->node_num)
1367                         continue;
1368
1369                 tmpstat = dlm_send_one_join_cancel(dlm, node);
1370                 if (tmpstat) {
1371                         mlog(ML_ERROR, "Error return %d cancelling join on "
1372                              "node %d\n", tmpstat, node);
1373                         if (!status)
1374                                 status = tmpstat;
1375                 }
1376         }
1377
1378         if (status)
1379                 mlog_errno(status);
1380         return status;
1381 }
1382
1383 static int dlm_request_join(struct dlm_ctxt *dlm,
1384                             int node,
1385                             enum dlm_query_join_response_code *response)
1386 {
1387         int status;
1388         struct dlm_query_join_request join_msg;
1389         struct dlm_query_join_packet packet;
1390         u32 join_resp;
1391
1392         mlog(0, "querying node %d\n", node);
1393
1394         memset(&join_msg, 0, sizeof(join_msg));
1395         join_msg.node_idx = dlm->node_num;
1396         join_msg.name_len = strlen(dlm->name);
1397         memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1398         join_msg.dlm_proto = dlm->dlm_locking_proto;
1399         join_msg.fs_proto = dlm->fs_locking_proto;
1400
1401         /* copy live node map to join message */
1402         byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1403
1404         status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1405                                     sizeof(join_msg), node, &join_resp);
1406         if (status < 0 && status != -ENOPROTOOPT) {
1407                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1408                      "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1409                      node);
1410                 goto bail;
1411         }
1412         dlm_query_join_wire_to_packet(join_resp, &packet);
1413
1414         /* -ENOPROTOOPT from the net code means the other side isn't
1415             listening for our message type -- that's fine, it means
1416             his dlm isn't up, so we can consider him a 'yes' but not
1417             joined into the domain.  */
1418         if (status == -ENOPROTOOPT) {
1419                 status = 0;
1420                 *response = JOIN_OK_NO_MAP;
1421         } else if (packet.code == JOIN_DISALLOW ||
1422                    packet.code == JOIN_OK_NO_MAP) {
1423                 *response = packet.code;
1424         } else if (packet.code == JOIN_PROTOCOL_MISMATCH) {
1425                 mlog(ML_NOTICE,
1426                      "This node requested DLM locking protocol %u.%u and "
1427                      "filesystem locking protocol %u.%u.  At least one of "
1428                      "the protocol versions on node %d is not compatible, "
1429                      "disconnecting\n",
1430                      dlm->dlm_locking_proto.pv_major,
1431                      dlm->dlm_locking_proto.pv_minor,
1432                      dlm->fs_locking_proto.pv_major,
1433                      dlm->fs_locking_proto.pv_minor,
1434                      node);
1435                 status = -EPROTO;
1436                 *response = packet.code;
1437         } else if (packet.code == JOIN_OK) {
1438                 *response = packet.code;
1439                 /* Use the same locking protocol as the remote node */
1440                 dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1441                 dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1442                 mlog(0,
1443                      "Node %d responds JOIN_OK with DLM locking protocol "
1444                      "%u.%u and fs locking protocol %u.%u\n",
1445                      node,
1446                      dlm->dlm_locking_proto.pv_major,
1447                      dlm->dlm_locking_proto.pv_minor,
1448                      dlm->fs_locking_proto.pv_major,
1449                      dlm->fs_locking_proto.pv_minor);
1450         } else {
1451                 status = -EINVAL;
1452                 mlog(ML_ERROR, "invalid response %d from node %u\n",
1453                      packet.code, node);
1454         }
1455
1456         mlog(0, "status %d, node %d response is %d\n", status, node,
1457              *response);
1458
1459 bail:
1460         return status;
1461 }
1462
1463 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1464                                     unsigned int node)
1465 {
1466         int status;
1467         struct dlm_assert_joined assert_msg;
1468
1469         mlog(0, "Sending join assert to node %u\n", node);
1470
1471         memset(&assert_msg, 0, sizeof(assert_msg));
1472         assert_msg.node_idx = dlm->node_num;
1473         assert_msg.name_len = strlen(dlm->name);
1474         memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1475
1476         status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1477                                     &assert_msg, sizeof(assert_msg), node,
1478                                     NULL);
1479         if (status < 0)
1480                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1481                      "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1482                      node);
1483
1484         return status;
1485 }
1486
1487 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1488                                   unsigned long *node_map)
1489 {
1490         int status, node, live;
1491
1492         status = 0;
1493         node = -1;
1494         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1495                                      node + 1)) < O2NM_MAX_NODES) {
1496                 if (node == dlm->node_num)
1497                         continue;
1498
1499                 do {
1500                         /* It is very important that this message be
1501                          * received so we spin until either the node
1502                          * has died or it gets the message. */
1503                         status = dlm_send_one_join_assert(dlm, node);
1504
1505                         spin_lock(&dlm->spinlock);
1506                         live = test_bit(node, dlm->live_nodes_map);
1507                         spin_unlock(&dlm->spinlock);
1508
1509                         if (status) {
1510                                 mlog(ML_ERROR, "Error return %d asserting "
1511                                      "join on node %d\n", status, node);
1512
1513                                 /* give us some time between errors... */
1514                                 if (live)
1515                                         msleep(DLM_DOMAIN_BACKOFF_MS);
1516                         }
1517                 } while (status && live);
1518         }
1519 }
1520
1521 struct domain_join_ctxt {
1522         unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1523         unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1524 };
1525
1526 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1527                                    struct domain_join_ctxt *ctxt,
1528                                    enum dlm_query_join_response_code response)
1529 {
1530         int ret;
1531
1532         if (response == JOIN_DISALLOW) {
1533                 mlog(0, "Latest response of disallow -- should restart\n");
1534                 return 1;
1535         }
1536
1537         spin_lock(&dlm->spinlock);
1538         /* For now, we restart the process if the node maps have
1539          * changed at all */
1540         ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1541                      sizeof(dlm->live_nodes_map));
1542         spin_unlock(&dlm->spinlock);
1543
1544         if (ret)
1545                 mlog(0, "Node maps changed -- should restart\n");
1546
1547         return ret;
1548 }
1549
1550 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1551 {
1552         int status = 0, tmpstat, node;
1553         struct domain_join_ctxt *ctxt;
1554         enum dlm_query_join_response_code response = JOIN_DISALLOW;
1555
1556         mlog_entry("%p", dlm);
1557
1558         ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1559         if (!ctxt) {
1560                 status = -ENOMEM;
1561                 mlog_errno(status);
1562                 goto bail;
1563         }
1564
1565         /* group sem locking should work for us here -- we're already
1566          * registered for heartbeat events so filling this should be
1567          * atomic wrt getting those handlers called. */
1568         o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1569
1570         spin_lock(&dlm->spinlock);
1571         memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1572
1573         __dlm_set_joining_node(dlm, dlm->node_num);
1574
1575         spin_unlock(&dlm->spinlock);
1576
1577         node = -1;
1578         while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1579                                      node + 1)) < O2NM_MAX_NODES) {
1580                 if (node == dlm->node_num)
1581                         continue;
1582
1583                 status = dlm_request_join(dlm, node, &response);
1584                 if (status < 0) {
1585                         mlog_errno(status);
1586                         goto bail;
1587                 }
1588
1589                 /* Ok, either we got a response or the node doesn't have a
1590                  * dlm up. */
1591                 if (response == JOIN_OK)
1592                         set_bit(node, ctxt->yes_resp_map);
1593
1594                 if (dlm_should_restart_join(dlm, ctxt, response)) {
1595                         status = -EAGAIN;
1596                         goto bail;
1597                 }
1598         }
1599
1600         mlog(0, "Yay, done querying nodes!\n");
1601
1602         /* Yay, everyone agree's we can join the domain. My domain is
1603          * comprised of all nodes who were put in the
1604          * yes_resp_map. Copy that into our domain map and send a join
1605          * assert message to clean up everyone elses state. */
1606         spin_lock(&dlm->spinlock);
1607         memcpy(dlm->domain_map, ctxt->yes_resp_map,
1608                sizeof(ctxt->yes_resp_map));
1609         set_bit(dlm->node_num, dlm->domain_map);
1610         spin_unlock(&dlm->spinlock);
1611
1612         /* Support for global heartbeat and node info was added in 1.1 */
1613         if (dlm_protocol.pv_major > 1 || dlm_protocol.pv_minor > 0) {
1614                 status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map);
1615                 if (status) {
1616                         mlog_errno(status);
1617                         goto bail;
1618                 }
1619                 status = dlm_send_regions(dlm, ctxt->yes_resp_map);
1620                 if (status) {
1621                         mlog_errno(status);
1622                         goto bail;
1623                 }
1624         }
1625
1626         dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1627
1628         /* Joined state *must* be set before the joining node
1629          * information, otherwise the query_join handler may read no
1630          * current joiner but a state of NEW and tell joining nodes
1631          * we're not in the domain. */
1632         spin_lock(&dlm_domain_lock);
1633         dlm->dlm_state = DLM_CTXT_JOINED;
1634         dlm->num_joins++;
1635         spin_unlock(&dlm_domain_lock);
1636
1637 bail:
1638         spin_lock(&dlm->spinlock);
1639         __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1640         if (!status)
1641                 __dlm_print_nodes(dlm);
1642         spin_unlock(&dlm->spinlock);
1643
1644         if (ctxt) {
1645                 /* Do we need to send a cancel message to any nodes? */
1646                 if (status < 0) {
1647                         tmpstat = dlm_send_join_cancels(dlm,
1648                                                         ctxt->yes_resp_map,
1649                                                         sizeof(ctxt->yes_resp_map));
1650                         if (tmpstat < 0)
1651                                 mlog_errno(tmpstat);
1652                 }
1653                 kfree(ctxt);
1654         }
1655
1656         mlog(0, "returning %d\n", status);
1657         return status;
1658 }
1659
1660 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1661 {
1662         o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up);
1663         o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down);
1664         o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1665 }
1666
1667 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1668 {
1669         int status;
1670
1671         mlog(0, "registering handlers.\n");
1672
1673         o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1674                             dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1675         status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down);
1676         if (status)
1677                 goto bail;
1678
1679         o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1680                             dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1681         status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up);
1682         if (status)
1683                 goto bail;
1684
1685         status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1686                                         sizeof(struct dlm_master_request),
1687                                         dlm_master_request_handler,
1688                                         dlm, NULL, &dlm->dlm_domain_handlers);
1689         if (status)
1690                 goto bail;
1691
1692         status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1693                                         sizeof(struct dlm_assert_master),
1694                                         dlm_assert_master_handler,
1695                                         dlm, dlm_assert_master_post_handler,
1696                                         &dlm->dlm_domain_handlers);
1697         if (status)
1698                 goto bail;
1699
1700         status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1701                                         sizeof(struct dlm_create_lock),
1702                                         dlm_create_lock_handler,
1703                                         dlm, NULL, &dlm->dlm_domain_handlers);
1704         if (status)
1705                 goto bail;
1706
1707         status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1708                                         DLM_CONVERT_LOCK_MAX_LEN,
1709                                         dlm_convert_lock_handler,
1710                                         dlm, NULL, &dlm->dlm_domain_handlers);
1711         if (status)
1712                 goto bail;
1713
1714         status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1715                                         DLM_UNLOCK_LOCK_MAX_LEN,
1716                                         dlm_unlock_lock_handler,
1717                                         dlm, NULL, &dlm->dlm_domain_handlers);
1718         if (status)
1719                 goto bail;
1720
1721         status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1722                                         DLM_PROXY_AST_MAX_LEN,
1723                                         dlm_proxy_ast_handler,
1724                                         dlm, NULL, &dlm->dlm_domain_handlers);
1725         if (status)
1726                 goto bail;
1727
1728         status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1729                                         sizeof(struct dlm_exit_domain),
1730                                         dlm_exit_domain_handler,
1731                                         dlm, NULL, &dlm->dlm_domain_handlers);
1732         if (status)
1733                 goto bail;
1734
1735         status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1736                                         sizeof(struct dlm_deref_lockres),
1737                                         dlm_deref_lockres_handler,
1738                                         dlm, NULL, &dlm->dlm_domain_handlers);
1739         if (status)
1740                 goto bail;
1741
1742         status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1743                                         sizeof(struct dlm_migrate_request),
1744                                         dlm_migrate_request_handler,
1745                                         dlm, NULL, &dlm->dlm_domain_handlers);
1746         if (status)
1747                 goto bail;
1748
1749         status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1750                                         DLM_MIG_LOCKRES_MAX_LEN,
1751                                         dlm_mig_lockres_handler,
1752                                         dlm, NULL, &dlm->dlm_domain_handlers);
1753         if (status)
1754                 goto bail;
1755
1756         status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1757                                         sizeof(struct dlm_master_requery),
1758                                         dlm_master_requery_handler,
1759                                         dlm, NULL, &dlm->dlm_domain_handlers);
1760         if (status)
1761                 goto bail;
1762
1763         status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1764                                         sizeof(struct dlm_lock_request),
1765                                         dlm_request_all_locks_handler,
1766                                         dlm, NULL, &dlm->dlm_domain_handlers);
1767         if (status)
1768                 goto bail;
1769
1770         status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1771                                         sizeof(struct dlm_reco_data_done),
1772                                         dlm_reco_data_done_handler,
1773                                         dlm, NULL, &dlm->dlm_domain_handlers);
1774         if (status)
1775                 goto bail;
1776
1777         status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1778                                         sizeof(struct dlm_begin_reco),
1779                                         dlm_begin_reco_handler,
1780                                         dlm, NULL, &dlm->dlm_domain_handlers);
1781         if (status)
1782                 goto bail;
1783
1784         status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1785                                         sizeof(struct dlm_finalize_reco),
1786                                         dlm_finalize_reco_handler,
1787                                         dlm, NULL, &dlm->dlm_domain_handlers);
1788         if (status)
1789                 goto bail;
1790
1791 bail:
1792         if (status)
1793                 dlm_unregister_domain_handlers(dlm);
1794
1795         return status;
1796 }
1797
1798 static int dlm_join_domain(struct dlm_ctxt *dlm)
1799 {
1800         int status;
1801         unsigned int backoff;
1802         unsigned int total_backoff = 0;
1803
1804         BUG_ON(!dlm);
1805
1806         mlog(0, "Join domain %s\n", dlm->name);
1807
1808         status = dlm_register_domain_handlers(dlm);
1809         if (status) {
1810                 mlog_errno(status);
1811                 goto bail;
1812         }
1813
1814         status = dlm_debug_init(dlm);
1815         if (status < 0) {
1816                 mlog_errno(status);
1817                 goto bail;
1818         }
1819
1820         status = dlm_launch_thread(dlm);
1821         if (status < 0) {
1822                 mlog_errno(status);
1823                 goto bail;
1824         }
1825
1826         status = dlm_launch_recovery_thread(dlm);
1827         if (status < 0) {
1828                 mlog_errno(status);
1829                 goto bail;
1830         }
1831
1832         dlm->dlm_worker = create_singlethread_workqueue("dlm_wq");
1833         if (!dlm->dlm_worker) {
1834                 status = -ENOMEM;
1835                 mlog_errno(status);
1836                 goto bail;
1837         }
1838
1839         do {
1840                 status = dlm_try_to_join_domain(dlm);
1841
1842                 /* If we're racing another node to the join, then we
1843                  * need to back off temporarily and let them
1844                  * complete. */
1845 #define DLM_JOIN_TIMEOUT_MSECS  90000
1846                 if (status == -EAGAIN) {
1847                         if (signal_pending(current)) {
1848                                 status = -ERESTARTSYS;
1849                                 goto bail;
1850                         }
1851
1852                         if (total_backoff >
1853                             msecs_to_jiffies(DLM_JOIN_TIMEOUT_MSECS)) {
1854                                 status = -ERESTARTSYS;
1855                                 mlog(ML_NOTICE, "Timed out joining dlm domain "
1856                                      "%s after %u msecs\n", dlm->name,
1857                                      jiffies_to_msecs(total_backoff));
1858                                 goto bail;
1859                         }
1860
1861                         /*
1862                          * <chip> After you!
1863                          * <dale> No, after you!
1864                          * <chip> I insist!
1865                          * <dale> But you first!
1866                          * ...
1867                          */
1868                         backoff = (unsigned int)(jiffies & 0x3);
1869                         backoff *= DLM_DOMAIN_BACKOFF_MS;
1870                         total_backoff += backoff;
1871                         mlog(0, "backoff %d\n", backoff);
1872                         msleep(backoff);
1873                 }
1874         } while (status == -EAGAIN);
1875
1876         if (status < 0) {
1877                 mlog_errno(status);
1878                 goto bail;
1879         }
1880
1881         status = 0;
1882 bail:
1883         wake_up(&dlm_domain_events);
1884
1885         if (status) {
1886                 dlm_unregister_domain_handlers(dlm);
1887                 dlm_debug_shutdown(dlm);
1888                 dlm_complete_thread(dlm);
1889                 dlm_complete_recovery_thread(dlm);
1890                 dlm_destroy_dlm_worker(dlm);
1891         }
1892
1893         return status;
1894 }
1895
1896 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1897                                 u32 key)
1898 {
1899         int i;
1900         int ret;
1901         struct dlm_ctxt *dlm = NULL;
1902
1903         dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1904         if (!dlm) {
1905                 mlog_errno(-ENOMEM);
1906                 goto leave;
1907         }
1908
1909         dlm->name = kstrdup(domain, GFP_KERNEL);
1910         if (dlm->name == NULL) {
1911                 mlog_errno(-ENOMEM);
1912                 kfree(dlm);
1913                 dlm = NULL;
1914                 goto leave;
1915         }
1916
1917         dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1918         if (!dlm->lockres_hash) {
1919                 mlog_errno(-ENOMEM);
1920                 kfree(dlm->name);
1921                 kfree(dlm);
1922                 dlm = NULL;
1923                 goto leave;
1924         }
1925
1926         for (i = 0; i < DLM_HASH_BUCKETS; i++)
1927                 INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
1928
1929         dlm->master_hash = (struct hlist_head **)
1930                                 dlm_alloc_pagevec(DLM_HASH_PAGES);
1931         if (!dlm->master_hash) {
1932                 mlog_errno(-ENOMEM);
1933                 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
1934                 kfree(dlm->name);
1935                 kfree(dlm);
1936                 dlm = NULL;
1937                 goto leave;
1938         }
1939
1940         for (i = 0; i < DLM_HASH_BUCKETS; i++)
1941                 INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
1942
1943         dlm->key = key;
1944         dlm->node_num = o2nm_this_node();
1945
1946         ret = dlm_create_debugfs_subroot(dlm);
1947         if (ret < 0) {
1948                 dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
1949                 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
1950                 kfree(dlm->name);
1951                 kfree(dlm);
1952                 dlm = NULL;
1953                 goto leave;
1954         }
1955
1956         spin_lock_init(&dlm->spinlock);
1957         spin_lock_init(&dlm->master_lock);
1958         spin_lock_init(&dlm->ast_lock);
1959         spin_lock_init(&dlm->track_lock);
1960         INIT_LIST_HEAD(&dlm->list);
1961         INIT_LIST_HEAD(&dlm->dirty_list);
1962         INIT_LIST_HEAD(&dlm->reco.resources);
1963         INIT_LIST_HEAD(&dlm->reco.received);
1964         INIT_LIST_HEAD(&dlm->reco.node_data);
1965         INIT_LIST_HEAD(&dlm->purge_list);
1966         INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
1967         INIT_LIST_HEAD(&dlm->tracking_list);
1968         dlm->reco.state = 0;
1969
1970         INIT_LIST_HEAD(&dlm->pending_asts);
1971         INIT_LIST_HEAD(&dlm->pending_basts);
1972
1973         mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
1974                   dlm->recovery_map, &(dlm->recovery_map[0]));
1975
1976         memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
1977         memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
1978         memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
1979
1980         dlm->dlm_thread_task = NULL;
1981         dlm->dlm_reco_thread_task = NULL;
1982         dlm->dlm_worker = NULL;
1983         init_waitqueue_head(&dlm->dlm_thread_wq);
1984         init_waitqueue_head(&dlm->dlm_reco_thread_wq);
1985         init_waitqueue_head(&dlm->reco.event);
1986         init_waitqueue_head(&dlm->ast_wq);
1987         init_waitqueue_head(&dlm->migration_wq);
1988         INIT_LIST_HEAD(&dlm->mle_hb_events);
1989
1990         dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
1991         init_waitqueue_head(&dlm->dlm_join_events);
1992
1993         dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
1994         dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
1995
1996         atomic_set(&dlm->res_tot_count, 0);
1997         atomic_set(&dlm->res_cur_count, 0);
1998         for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
1999                 atomic_set(&dlm->mle_tot_count[i], 0);
2000                 atomic_set(&dlm->mle_cur_count[i], 0);
2001         }
2002
2003         spin_lock_init(&dlm->work_lock);
2004         INIT_LIST_HEAD(&dlm->work_list);
2005         INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
2006
2007         kref_init(&dlm->dlm_refs);
2008         dlm->dlm_state = DLM_CTXT_NEW;
2009
2010         INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
2011
2012         mlog(0, "context init: refcount %u\n",
2013                   atomic_read(&dlm->dlm_refs.refcount));
2014
2015 leave:
2016         return dlm;
2017 }
2018
2019 /*
2020  * Compare a requested locking protocol version against the current one.
2021  *
2022  * If the major numbers are different, they are incompatible.
2023  * If the current minor is greater than the request, they are incompatible.
2024  * If the current minor is less than or equal to the request, they are
2025  * compatible, and the requester should run at the current minor version.
2026  */
2027 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
2028                                 struct dlm_protocol_version *request)
2029 {
2030         if (existing->pv_major != request->pv_major)
2031                 return 1;
2032
2033         if (existing->pv_minor > request->pv_minor)
2034                 return 1;
2035
2036         if (existing->pv_minor < request->pv_minor)
2037                 request->pv_minor = existing->pv_minor;
2038
2039         return 0;
2040 }
2041
2042 /*
2043  * dlm_register_domain: one-time setup per "domain".
2044  *
2045  * The filesystem passes in the requested locking version via proto.
2046  * If registration was successful, proto will contain the negotiated
2047  * locking protocol.
2048  */
2049 struct dlm_ctxt * dlm_register_domain(const char *domain,
2050                                u32 key,
2051                                struct dlm_protocol_version *fs_proto)
2052 {
2053         int ret;
2054         struct dlm_ctxt *dlm = NULL;
2055         struct dlm_ctxt *new_ctxt = NULL;
2056
2057         if (strlen(domain) >= O2NM_MAX_NAME_LEN) {
2058                 ret = -ENAMETOOLONG;
2059                 mlog(ML_ERROR, "domain name length too long\n");
2060                 goto leave;
2061         }
2062
2063         if (!o2hb_check_local_node_heartbeating()) {
2064                 mlog(ML_ERROR, "the local node has not been configured, or is "
2065                      "not heartbeating\n");
2066                 ret = -EPROTO;
2067                 goto leave;
2068         }
2069
2070         mlog(0, "register called for domain \"%s\"\n", domain);
2071
2072 retry:
2073         dlm = NULL;
2074         if (signal_pending(current)) {
2075                 ret = -ERESTARTSYS;
2076                 mlog_errno(ret);
2077                 goto leave;
2078         }
2079
2080         spin_lock(&dlm_domain_lock);
2081
2082         dlm = __dlm_lookup_domain(domain);
2083         if (dlm) {
2084                 if (dlm->dlm_state != DLM_CTXT_JOINED) {
2085                         spin_unlock(&dlm_domain_lock);
2086
2087                         mlog(0, "This ctxt is not joined yet!\n");
2088                         wait_event_interruptible(dlm_domain_events,
2089                                                  dlm_wait_on_domain_helper(
2090                                                          domain));
2091                         goto retry;
2092                 }
2093
2094                 if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
2095                         spin_unlock(&dlm_domain_lock);
2096                         mlog(ML_ERROR,
2097                              "Requested locking protocol version is not "
2098                              "compatible with already registered domain "
2099                              "\"%s\"\n", domain);
2100                         ret = -EPROTO;
2101                         goto leave;
2102                 }
2103
2104                 __dlm_get(dlm);
2105                 dlm->num_joins++;
2106
2107                 spin_unlock(&dlm_domain_lock);
2108
2109                 ret = 0;
2110                 goto leave;
2111         }
2112
2113         /* doesn't exist */
2114         if (!new_ctxt) {
2115                 spin_unlock(&dlm_domain_lock);
2116
2117                 new_ctxt = dlm_alloc_ctxt(domain, key);
2118                 if (new_ctxt)
2119                         goto retry;
2120
2121                 ret = -ENOMEM;
2122                 mlog_errno(ret);
2123                 goto leave;
2124         }
2125
2126         /* a little variable switch-a-roo here... */
2127         dlm = new_ctxt;
2128         new_ctxt = NULL;
2129
2130         /* add the new domain */
2131         list_add_tail(&dlm->list, &dlm_domains);
2132         spin_unlock(&dlm_domain_lock);
2133
2134         /*
2135          * Pass the locking protocol version into the join.  If the join
2136          * succeeds, it will have the negotiated protocol set.
2137          */
2138         dlm->dlm_locking_proto = dlm_protocol;
2139         dlm->fs_locking_proto = *fs_proto;
2140
2141         ret = dlm_join_domain(dlm);
2142         if (ret) {
2143                 mlog_errno(ret);
2144                 dlm_put(dlm);
2145                 goto leave;
2146         }
2147
2148         /* Tell the caller what locking protocol we negotiated */
2149         *fs_proto = dlm->fs_locking_proto;
2150
2151         ret = 0;
2152 leave:
2153         if (new_ctxt)
2154                 dlm_free_ctxt_mem(new_ctxt);
2155
2156         if (ret < 0)
2157                 dlm = ERR_PTR(ret);
2158
2159         return dlm;
2160 }
2161 EXPORT_SYMBOL_GPL(dlm_register_domain);
2162
2163 static LIST_HEAD(dlm_join_handlers);
2164
2165 static void dlm_unregister_net_handlers(void)
2166 {
2167         o2net_unregister_handler_list(&dlm_join_handlers);
2168 }
2169
2170 static int dlm_register_net_handlers(void)
2171 {
2172         int status = 0;
2173
2174         status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
2175                                         sizeof(struct dlm_query_join_request),
2176                                         dlm_query_join_handler,
2177                                         NULL, NULL, &dlm_join_handlers);
2178         if (status)
2179                 goto bail;
2180
2181         status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
2182                                         sizeof(struct dlm_assert_joined),
2183                                         dlm_assert_joined_handler,
2184                                         NULL, NULL, &dlm_join_handlers);
2185         if (status)
2186                 goto bail;
2187
2188         status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
2189                                         sizeof(struct dlm_cancel_join),
2190                                         dlm_cancel_join_handler,
2191                                         NULL, NULL, &dlm_join_handlers);
2192         if (status)
2193                 goto bail;
2194
2195         status = o2net_register_handler(DLM_QUERY_REGION, DLM_MOD_KEY,
2196                                         sizeof(struct dlm_query_region),
2197                                         dlm_query_region_handler,
2198                                         NULL, NULL, &dlm_join_handlers);
2199
2200         if (status)
2201                 goto bail;
2202
2203         status = o2net_register_handler(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
2204                                         sizeof(struct dlm_query_nodeinfo),
2205                                         dlm_query_nodeinfo_handler,
2206                                         NULL, NULL, &dlm_join_handlers);
2207 bail:
2208         if (status < 0)
2209                 dlm_unregister_net_handlers();
2210
2211         return status;
2212 }
2213
2214 /* Domain eviction callback handling.
2215  *
2216  * The file system requires notification of node death *before* the
2217  * dlm completes it's recovery work, otherwise it may be able to
2218  * acquire locks on resources requiring recovery. Since the dlm can
2219  * evict a node from it's domain *before* heartbeat fires, a similar
2220  * mechanism is required. */
2221
2222 /* Eviction is not expected to happen often, so a per-domain lock is
2223  * not necessary. Eviction callbacks are allowed to sleep for short
2224  * periods of time. */
2225 static DECLARE_RWSEM(dlm_callback_sem);
2226
2227 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
2228                                         int node_num)
2229 {
2230         struct list_head *iter;
2231         struct dlm_eviction_cb *cb;
2232
2233         down_read(&dlm_callback_sem);
2234         list_for_each(iter, &dlm->dlm_eviction_callbacks) {
2235                 cb = list_entry(iter, struct dlm_eviction_cb, ec_item);
2236
2237                 cb->ec_func(node_num, cb->ec_data);
2238         }
2239         up_read(&dlm_callback_sem);
2240 }
2241
2242 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
2243                            dlm_eviction_func *f,
2244                            void *data)
2245 {
2246         INIT_LIST_HEAD(&cb->ec_item);
2247         cb->ec_func = f;
2248         cb->ec_data = data;
2249 }
2250 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
2251
2252 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
2253                               struct dlm_eviction_cb *cb)
2254 {
2255         down_write(&dlm_callback_sem);
2256         list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
2257         up_write(&dlm_callback_sem);
2258 }
2259 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
2260
2261 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
2262 {
2263         down_write(&dlm_callback_sem);
2264         list_del_init(&cb->ec_item);
2265         up_write(&dlm_callback_sem);
2266 }
2267 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
2268
2269 static int __init dlm_init(void)
2270 {
2271         int status;
2272
2273         dlm_print_version();
2274
2275         status = dlm_init_mle_cache();
2276         if (status) {
2277                 mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
2278                 goto error;
2279         }
2280
2281         status = dlm_init_master_caches();
2282         if (status) {
2283                 mlog(ML_ERROR, "Could not create o2dlm_lockres and "
2284                      "o2dlm_lockname slabcaches\n");
2285                 goto error;
2286         }
2287
2288         status = dlm_init_lock_cache();
2289         if (status) {
2290                 mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
2291                 goto error;
2292         }
2293
2294         status = dlm_register_net_handlers();
2295         if (status) {
2296                 mlog(ML_ERROR, "Unable to register network handlers\n");
2297                 goto error;
2298         }
2299
2300         status = dlm_create_debugfs_root();
2301         if (status)
2302                 goto error;
2303
2304         return 0;
2305 error:
2306         dlm_unregister_net_handlers();
2307         dlm_destroy_lock_cache();
2308         dlm_destroy_master_caches();
2309         dlm_destroy_mle_cache();
2310         return -1;
2311 }
2312
2313 static void __exit dlm_exit (void)
2314 {
2315         dlm_destroy_debugfs_root();
2316         dlm_unregister_net_handlers();
2317         dlm_destroy_lock_cache();
2318         dlm_destroy_master_caches();
2319         dlm_destroy_mle_cache();
2320 }
2321
2322 MODULE_AUTHOR("Oracle");
2323 MODULE_LICENSE("GPL");
2324
2325 module_init(dlm_init);
2326 module_exit(dlm_exit);