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 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
  47 #include "cluster/masklog.h"
  48
  49 /*
  50  * ocfs2 node maps are array of long int, which limits to send them freely
  51  * across the wire due to endianness issues. To workaround this, we convert
  52  * long ints to byte arrays. Following 3 routines are helper functions to
  53  * set/test/copy bits within those array of bytes
  54  */
  55 static inline void byte_set_bit(u8 nr, u8 map[])
  56 {
  57         map[nr >> 3] |= (1UL << (nr & 7));
  58 }
  59
  60 static inline int byte_test_bit(u8 nr, u8 map[])
  61 {
  62         return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
  63 }
  64
  65 static inline void byte_copymap(u8 dmap[], unsigned long smap[],
  66                         unsigned int sz)
  67 {
  68         unsigned int nn;
  69
  70         if (!sz)
  71                 return;
  72
  73         memset(dmap, 0, ((sz + 7) >> 3));
  74         for (nn = 0 ; nn < sz; nn++)
  75                 if (test_bit(nn, smap))
  76                         byte_set_bit(nn, dmap);
  77 }
  78
  79 static void dlm_free_pagevec(void **vec, int pages)
  80 {
  81         while (pages--)
  82                 free_page((unsigned long)vec[pages]);
  83         kfree(vec);
  84 }
  85
  86 static void **dlm_alloc_pagevec(int pages)
  87 {
  88         void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL);
  89         int i;
  90
  91         if (!vec)
  92                 return NULL;
  93
  94         for (i = 0; i < pages; i++)
  95                 if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
  96                         goto out_free;
  97
  98         mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n",
  99              pages, (unsigned long)DLM_HASH_PAGES,
 100              (unsigned long)DLM_BUCKETS_PER_PAGE);
 101         return vec;
 102 out_free:
 103         dlm_free_pagevec(vec, i);
 104         return NULL;
 105 }
 106
 107 /*
 108  *
 109  * spinlock lock ordering: if multiple locks are needed, obey this ordering:
 110  *    dlm_domain_lock
 111  *    struct dlm_ctxt->spinlock
 112  *    struct dlm_lock_resource->spinlock
 113  *    struct dlm_ctxt->master_lock
 114  *    struct dlm_ctxt->ast_lock
 115  *    dlm_master_list_entry->spinlock
 116  *    dlm_lock->spinlock
 117  *
 118  */
 119
 120 DEFINE_SPINLOCK(dlm_domain_lock);
 121 LIST_HEAD(dlm_domains);
 122 static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events);
 123
 124 /*
 125  * The supported protocol version for DLM communication.  Running domains
 126  * will have a negotiated version with the same major number and a minor
 127  * number equal or smaller.  The dlm_ctxt->dlm_locking_proto field should
 128  * be used to determine what a running domain is actually using.
 129  *
 130  * New in version 1.1:
 131  *      - Message DLM_QUERY_REGION added to support global heartbeat
 132  *      - Message DLM_QUERY_NODEINFO added to allow online node removes
 133  * New in version 1.2:
 134  *      - Message DLM_BEGIN_EXIT_DOMAIN_MSG added to mark start of exit domain
 135  * New in version 1.3:
 136  *      - Message DLM_DEREF_LOCKRES_DONE added to inform non-master that the
 137  *        refmap is cleared
 138  */
 139 static const struct dlm_protocol_version dlm_protocol = {
 140         .pv_major = 1,
 141         .pv_minor = 3,
 142 };
 143
 144 #define DLM_DOMAIN_BACKOFF_MS 200
 145
 146 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
 147                                   void **ret_data);
 148 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
 149                                      void **ret_data);
 150 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
 151                                    void **ret_data);
 152 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
 153                                     void *data, void **ret_data);
 154 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
 155                                    void **ret_data);
 156 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
 157                                 struct dlm_protocol_version *request);
 158
 159 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
 160
 161 void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
 162 {
 163         if (hlist_unhashed(&res->hash_node))
 164                 return;
 165
 166         mlog(0, "%s: Unhash res %.*s\n", dlm->name, res->lockname.len,
 167              res->lockname.name);
 168         hlist_del_init(&res->hash_node);
 169         dlm_lockres_put(res);
 170 }
 171
 172 void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
 173 {
 174         struct hlist_head *bucket;
 175         struct qstr *q;
 176
 177         assert_spin_locked(&dlm->spinlock);
 178
 179         q = &res->lockname;
 180         bucket = dlm_lockres_hash(dlm, q->hash);
 181
 182         /* get a reference for our hashtable */
 183         dlm_lockres_get(res);
 184
 185         hlist_add_head(&res->hash_node, bucket);
 186
 187         mlog(0, "%s: Hash res %.*s\n", dlm->name, res->lockname.len,
 188              res->lockname.name);
 189 }
 190
 191 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
 192                                                      const char *name,
 193                                                      unsigned int len,
 194                                                      unsigned int hash)
 195 {
 196         struct hlist_head *bucket;
 197         struct dlm_lock_resource *res;
 198
 199         mlog(0, "%.*s\n", len, name);
 200
 201         assert_spin_locked(&dlm->spinlock);
 202
 203         bucket = dlm_lockres_hash(dlm, hash);
 204
 205         hlist_for_each_entry(res, bucket, hash_node) {
 206                 if (res->lockname.name[0] != name[0])
 207                         continue;
 208                 if (unlikely(res->lockname.len != len))
 209                         continue;
 210                 if (memcmp(res->lockname.name + 1, name + 1, len - 1))
 211                         continue;
 212                 dlm_lockres_get(res);
 213                 return res;
 214         }
 215         return NULL;
 216 }
 217
 218 /* intended to be called by functions which do not care about lock
 219  * resources which are being purged (most net _handler functions).
 220  * this will return NULL for any lock resource which is found but
 221  * currently in the process of dropping its mastery reference.
 222  * use __dlm_lookup_lockres_full when you need the lock resource
 223  * regardless (e.g. dlm_get_lock_resource) */
 224 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
 225                                                 const char *name,
 226                                                 unsigned int len,
 227                                                 unsigned int hash)
 228 {
 229         struct dlm_lock_resource *res = NULL;
 230
 231         mlog(0, "%.*s\n", len, name);
 232
 233         assert_spin_locked(&dlm->spinlock);
 234
 235         res = __dlm_lookup_lockres_full(dlm, name, len, hash);
 236         if (res) {
 237                 spin_lock(&res->spinlock);
 238                 if (res->state & DLM_LOCK_RES_DROPPING_REF) {
 239                         spin_unlock(&res->spinlock);
 240                         dlm_lockres_put(res);
 241                         return NULL;
 242                 }
 243                 spin_unlock(&res->spinlock);
 244         }
 245
 246         return res;
 247 }
 248
 249 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
 250                                     const char *name,
 251                                     unsigned int len)
 252 {
 253         struct dlm_lock_resource *res;
 254         unsigned int hash = dlm_lockid_hash(name, len);
 255
 256         spin_lock(&dlm->spinlock);
 257         res = __dlm_lookup_lockres(dlm, name, len, hash);
 258         spin_unlock(&dlm->spinlock);
 259         return res;
 260 }
 261
 262 static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
 263 {
 264         struct dlm_ctxt *tmp;
 265
 266         assert_spin_locked(&dlm_domain_lock);
 267
 268         /* tmp->name here is always NULL terminated,
 269          * but domain may not be! */
 270         list_for_each_entry(tmp, &dlm_domains, list) {
 271                 if (strlen(tmp->name) == len &&
 272                     memcmp(tmp->name, domain, len)==0)
 273                         return tmp;
 274         }
 275
 276         return NULL;
 277 }
 278
 279 /* For null terminated domain strings ONLY */
 280 static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
 281 {
 282         assert_spin_locked(&dlm_domain_lock);
 283
 284         return __dlm_lookup_domain_full(domain, strlen(domain));
 285 }
 286
 287
 288 /* returns true on one of two conditions:
 289  * 1) the domain does not exist
 290  * 2) the domain exists and it's state is "joined" */
 291 static int dlm_wait_on_domain_helper(const char *domain)
 292 {
 293         int ret = 0;
 294         struct dlm_ctxt *tmp = NULL;
 295
 296         spin_lock(&dlm_domain_lock);
 297
 298         tmp = __dlm_lookup_domain(domain);
 299         if (!tmp)
 300                 ret = 1;
 301         else if (tmp->dlm_state == DLM_CTXT_JOINED)
 302                 ret = 1;
 303
 304         spin_unlock(&dlm_domain_lock);
 305         return ret;
 306 }
 307
 308 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
 309 {
 310         dlm_destroy_debugfs_subroot(dlm);
 311
 312         if (dlm->lockres_hash)
 313                 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
 314
 315         if (dlm->master_hash)
 316                 dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
 317
 318         kfree(dlm->name);
 319         kfree(dlm);
 320 }
 321
 322 /* A little strange - this function will be called while holding
 323  * dlm_domain_lock and is expected to be holding it on the way out. We
 324  * will however drop and reacquire it multiple times */
 325 static void dlm_ctxt_release(struct kref *kref)
 326 {
 327         struct dlm_ctxt *dlm;
 328
 329         dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
 330
 331         BUG_ON(dlm->num_joins);
 332         BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
 333
 334         /* we may still be in the list if we hit an error during join. */
 335         list_del_init(&dlm->list);
 336
 337         spin_unlock(&dlm_domain_lock);
 338
 339         mlog(0, "freeing memory from domain %s\n", dlm->name);
 340
 341         wake_up(&dlm_domain_events);
 342
 343         dlm_free_ctxt_mem(dlm);
 344
 345         spin_lock(&dlm_domain_lock);
 346 }
 347
 348 void dlm_put(struct dlm_ctxt *dlm)
 349 {
 350         spin_lock(&dlm_domain_lock);
 351         kref_put(&dlm->dlm_refs, dlm_ctxt_release);
 352         spin_unlock(&dlm_domain_lock);
 353 }
 354
 355 static void __dlm_get(struct dlm_ctxt *dlm)
 356 {
 357         kref_get(&dlm->dlm_refs);
 358 }
 359
 360 /* given a questionable reference to a dlm object, gets a reference if
 361  * it can find it in the list, otherwise returns NULL in which case
 362  * you shouldn't trust your pointer. */
 363 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
 364 {
 365         struct dlm_ctxt *target;
 366         struct dlm_ctxt *ret = NULL;
 367
 368         spin_lock(&dlm_domain_lock);
 369
 370         list_for_each_entry(target, &dlm_domains, list) {
 371                 if (target == dlm) {
 372                         __dlm_get(target);
 373                         ret = target;
 374                         break;
 375                 }
 376         }
 377
 378         spin_unlock(&dlm_domain_lock);
 379
 380         return ret;
 381 }
 382
 383 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
 384 {
 385         int ret;
 386
 387         spin_lock(&dlm_domain_lock);
 388         ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
 389                 (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
 390         spin_unlock(&dlm_domain_lock);
 391
 392         return ret;
 393 }
 394
 395 static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
 396 {
 397         if (dlm->dlm_worker) {
 398                 flush_workqueue(dlm->dlm_worker);
 399                 destroy_workqueue(dlm->dlm_worker);
 400                 dlm->dlm_worker = NULL;
 401         }
 402 }
 403
 404 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
 405 {
 406         dlm_unregister_domain_handlers(dlm);
 407         dlm_debug_shutdown(dlm);
 408         dlm_complete_thread(dlm);
 409         dlm_complete_recovery_thread(dlm);
 410         dlm_destroy_dlm_worker(dlm);
 411
 412         /* We've left the domain. Now we can take ourselves out of the
 413          * list and allow the kref stuff to help us free the
 414          * memory. */
 415         spin_lock(&dlm_domain_lock);
 416         list_del_init(&dlm->list);
 417         spin_unlock(&dlm_domain_lock);
 418
 419         /* Wake up anyone waiting for us to remove this domain */
 420         wake_up(&dlm_domain_events);
 421 }
 422
 423 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
 424 {
 425         int i, num, n, ret = 0;
 426         struct dlm_lock_resource *res;
 427         struct hlist_node *iter;
 428         struct hlist_head *bucket;
 429         int dropped;
 430
 431         mlog(0, "Migrating locks from domain %s\n", dlm->name);
 432
 433         num = 0;
 434         spin_lock(&dlm->spinlock);
 435         for (i = 0; i < DLM_HASH_BUCKETS; i++) {
 436 redo_bucket:
 437                 n = 0;
 438                 bucket = dlm_lockres_hash(dlm, i);
 439                 iter = bucket->first;
 440                 while (iter) {
 441                         n++;
 442                         res = hlist_entry(iter, struct dlm_lock_resource,
 443                                           hash_node);
 444                         dlm_lockres_get(res);
 445                         /* migrate, if necessary.  this will drop the dlm
 446                          * spinlock and retake it if it does migration. */
 447                         dropped = dlm_empty_lockres(dlm, res);
 448
 449                         spin_lock(&res->spinlock);
 450                         if (dropped)
 451                                 __dlm_lockres_calc_usage(dlm, res);
 452                         else
 453                                 iter = res->hash_node.next;
 454                         spin_unlock(&res->spinlock);
 455
 456                         dlm_lockres_put(res);
 457
 458                         if (dropped) {
 459                                 cond_resched_lock(&dlm->spinlock);
 460                                 goto redo_bucket;
 461                         }
 462                 }
 463                 cond_resched_lock(&dlm->spinlock);
 464                 num += n;
 465         }
 466         spin_unlock(&dlm->spinlock);
 467         wake_up(&dlm->dlm_thread_wq);
 468
 469         /* let the dlm thread take care of purging, keep scanning until
 470          * nothing remains in the hash */
 471         if (num) {
 472                 mlog(0, "%s: %d lock resources in hash last pass\n",
 473                      dlm->name, num);
 474                 ret = -EAGAIN;
 475         }
 476         mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
 477         return ret;
 478 }
 479
 480 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
 481 {
 482         int ret;
 483
 484         spin_lock(&dlm->spinlock);
 485         ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
 486         spin_unlock(&dlm->spinlock);
 487
 488         return ret;
 489 }
 490
 491 static int dlm_begin_exit_domain_handler(struct o2net_msg *msg, u32 len,
 492                                          void *data, void **ret_data)
 493 {
 494         struct dlm_ctxt *dlm = data;
 495         unsigned int node;
 496         struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
 497
 498         if (!dlm_grab(dlm))
 499                 return 0;
 500
 501         node = exit_msg->node_idx;
 502         mlog(0, "%s: Node %u sent a begin exit domain message\n", dlm->name, node);
 503
 504         spin_lock(&dlm->spinlock);
 505         set_bit(node, dlm->exit_domain_map);
 506         spin_unlock(&dlm->spinlock);
 507
 508         dlm_put(dlm);
 509
 510         return 0;
 511 }
 512
 513 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
 514 {
 515         /* Yikes, a double spinlock! I need domain_lock for the dlm
 516          * state and the dlm spinlock for join state... Sorry! */
 517 again:
 518         spin_lock(&dlm_domain_lock);
 519         spin_lock(&dlm->spinlock);
 520
 521         if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
 522                 mlog(0, "Node %d is joining, we wait on it.\n",
 523                           dlm->joining_node);
 524                 spin_unlock(&dlm->spinlock);
 525                 spin_unlock(&dlm_domain_lock);
 526
 527                 wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
 528                 goto again;
 529         }
 530
 531         dlm->dlm_state = DLM_CTXT_LEAVING;
 532         spin_unlock(&dlm->spinlock);
 533         spin_unlock(&dlm_domain_lock);
 534 }
 535
 536 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
 537 {
 538         int node = -1, num = 0;
 539
 540         assert_spin_locked(&dlm->spinlock);
 541
 542         printk("( ");
 543         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
 544                                      node + 1)) < O2NM_MAX_NODES) {
 545                 printk("%d ", node);
 546                 ++num;
 547         }
 548         printk(") %u nodes\n", num);
 549 }
 550
 551 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
 552                                    void **ret_data)
 553 {
 554         struct dlm_ctxt *dlm = data;
 555         unsigned int node;
 556         struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
 557
 558         mlog(0, "%p %u %p", msg, len, data);
 559
 560         if (!dlm_grab(dlm))
 561                 return 0;
 562
 563         node = exit_msg->node_idx;
 564
 565         spin_lock(&dlm->spinlock);
 566         clear_bit(node, dlm->domain_map);
 567         clear_bit(node, dlm->exit_domain_map);
 568         printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s ", node, dlm->name);
 569         __dlm_print_nodes(dlm);
 570
 571         /* notify anything attached to the heartbeat events */
 572         dlm_hb_event_notify_attached(dlm, node, 0);
 573
 574         spin_unlock(&dlm->spinlock);
 575
 576         dlm_put(dlm);
 577
 578         return 0;
 579 }
 580
 581 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, u32 msg_type,
 582                                     unsigned int node)
 583 {
 584         int status;
 585         struct dlm_exit_domain leave_msg;
 586
 587         mlog(0, "%s: Sending domain exit message %u to node %u\n", dlm->name,
 588              msg_type, node);
 589
 590         memset(&leave_msg, 0, sizeof(leave_msg));
 591         leave_msg.node_idx = dlm->node_num;
 592
 593         status = o2net_send_message(msg_type, dlm->key, &leave_msg,
 594                                     sizeof(leave_msg), node, NULL);
 595         if (status < 0)
 596                 mlog(ML_ERROR, "Error %d sending domain exit message %u "
 597                      "to node %u on domain %s\n", status, msg_type, node,
 598                      dlm->name);
 599
 600         return status;
 601 }
 602
 603 static void dlm_begin_exit_domain(struct dlm_ctxt *dlm)
 604 {
 605         int node = -1;
 606
 607         /* Support for begin exit domain was added in 1.2 */
 608         if (dlm->dlm_locking_proto.pv_major == 1 &&
 609             dlm->dlm_locking_proto.pv_minor < 2)
 610                 return;
 611
 612         /*
 613          * Unlike DLM_EXIT_DOMAIN_MSG, DLM_BEGIN_EXIT_DOMAIN_MSG is purely
 614          * informational. Meaning if a node does not receive the message,
 615          * so be it.
 616          */
 617         spin_lock(&dlm->spinlock);
 618         while (1) {
 619                 node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1);
 620                 if (node >= O2NM_MAX_NODES)
 621                         break;
 622                 if (node == dlm->node_num)
 623                         continue;
 624
 625                 spin_unlock(&dlm->spinlock);
 626                 dlm_send_one_domain_exit(dlm, DLM_BEGIN_EXIT_DOMAIN_MSG, node);
 627                 spin_lock(&dlm->spinlock);
 628         }
 629         spin_unlock(&dlm->spinlock);
 630 }
 631
 632 static void dlm_leave_domain(struct dlm_ctxt *dlm)
 633 {
 634         int node, clear_node, status;
 635
 636         /* At this point we've migrated away all our locks and won't
 637          * accept mastership of new ones. The dlm is responsible for
 638          * almost nothing now. We make sure not to confuse any joining
 639          * nodes and then commence shutdown procedure. */
 640
 641         spin_lock(&dlm->spinlock);
 642         /* Clear ourselves from the domain map */
 643         clear_bit(dlm->node_num, dlm->domain_map);
 644         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
 645                                      0)) < O2NM_MAX_NODES) {
 646                 /* Drop the dlm spinlock. This is safe wrt the domain_map.
 647                  * -nodes cannot be added now as the
 648                  *   query_join_handlers knows to respond with OK_NO_MAP
 649                  * -we catch the right network errors if a node is
 650                  *   removed from the map while we're sending him the
 651                  *   exit message. */
 652                 spin_unlock(&dlm->spinlock);
 653
 654                 clear_node = 1;
 655
 656                 status = dlm_send_one_domain_exit(dlm, DLM_EXIT_DOMAIN_MSG,
 657                                                   node);
 658                 if (status < 0 &&
 659                     status != -ENOPROTOOPT &&
 660                     status != -ENOTCONN) {
 661                         mlog(ML_NOTICE, "Error %d sending domain exit message "
 662                              "to node %d\n", status, node);
 663
 664                         /* Not sure what to do here but lets sleep for
 665                          * a bit in case this was a transient
 666                          * error... */
 667                         msleep(DLM_DOMAIN_BACKOFF_MS);
 668                         clear_node = 0;
 669                 }
 670
 671                 spin_lock(&dlm->spinlock);
 672                 /* If we're not clearing the node bit then we intend
 673                  * to loop back around to try again. */
 674                 if (clear_node)
 675                         clear_bit(node, dlm->domain_map);
 676         }
 677         spin_unlock(&dlm->spinlock);
 678 }
 679
 680 int dlm_shutting_down(struct dlm_ctxt *dlm)
 681 {
 682         int ret = 0;
 683
 684         spin_lock(&dlm_domain_lock);
 685
 686         if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
 687                 ret = 1;
 688
 689         spin_unlock(&dlm_domain_lock);
 690
 691         return ret;
 692 }
 693
 694 void dlm_unregister_domain(struct dlm_ctxt *dlm)
 695 {
 696         int leave = 0;
 697         struct dlm_lock_resource *res;
 698
 699         spin_lock(&dlm_domain_lock);
 700         BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
 701         BUG_ON(!dlm->num_joins);
 702
 703         dlm->num_joins--;
 704         if (!dlm->num_joins) {
 705                 /* We mark it "in shutdown" now so new register
 706                  * requests wait until we've completely left the
 707                  * domain. Don't use DLM_CTXT_LEAVING yet as we still
 708                  * want new domain joins to communicate with us at
 709                  * least until we've completed migration of our
 710                  * resources. */
 711                 dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
 712                 leave = 1;
 713         }
 714         spin_unlock(&dlm_domain_lock);
 715
 716         if (leave) {
 717                 mlog(0, "shutting down domain %s\n", dlm->name);
 718                 dlm_begin_exit_domain(dlm);
 719
 720                 /* We changed dlm state, notify the thread */
 721                 dlm_kick_thread(dlm, NULL);
 722
 723                 while (dlm_migrate_all_locks(dlm)) {
 724                         /* Give dlm_thread time to purge the lockres' */
 725                         msleep(500);
 726                         mlog(0, "%s: more migration to do\n", dlm->name);
 727                 }
 728
 729                 /* This list should be empty. If not, print remaining lockres */
 730                 if (!list_empty(&dlm->tracking_list)) {
 731                         mlog(ML_ERROR, "Following lockres' are still on the "
 732                              "tracking list:\n");
 733                         list_for_each_entry(res, &dlm->tracking_list, tracking)
 734                                 dlm_print_one_lock_resource(res);
 735                 }
 736
 737                 dlm_mark_domain_leaving(dlm);
 738                 dlm_leave_domain(dlm);
 739                 printk(KERN_NOTICE "o2dlm: Leaving domain %s\n", dlm->name);
 740                 dlm_force_free_mles(dlm);
 741                 dlm_complete_dlm_shutdown(dlm);
 742         }
 743         dlm_put(dlm);
 744 }
 745 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
 746
 747 static int dlm_query_join_proto_check(char *proto_type, int node,
 748                                       struct dlm_protocol_version *ours,
 749                                       struct dlm_protocol_version *request)
 750 {
 751         int rc;
 752         struct dlm_protocol_version proto = *request;
 753
 754         if (!dlm_protocol_compare(ours, &proto)) {
 755                 mlog(0,
 756                      "node %u wanted to join with %s locking protocol "
 757                      "%u.%u, we respond with %u.%u\n",
 758                      node, proto_type,
 759                      request->pv_major,
 760                      request->pv_minor,
 761                      proto.pv_major, proto.pv_minor);
 762                 request->pv_minor = proto.pv_minor;
 763                 rc = 0;
 764         } else {
 765                 mlog(ML_NOTICE,
 766                      "Node %u wanted to join with %s locking "
 767                      "protocol %u.%u, but we have %u.%u, disallowing\n",
 768                      node, proto_type,
 769                      request->pv_major,
 770                      request->pv_minor,
 771                      ours->pv_major,
 772                      ours->pv_minor);
 773                 rc = 1;
 774         }
 775
 776         return rc;
 777 }
 778
 779 /*
 780  * struct dlm_query_join_packet is made up of four one-byte fields.  They
 781  * are effectively in big-endian order already.  However, little-endian
 782  * machines swap them before putting the packet on the wire (because
 783  * query_join's response is a status, and that status is treated as a u32
 784  * on the wire).  Thus, a big-endian and little-endian machines will treat
 785  * this structure differently.
 786  *
 787  * The solution is to have little-endian machines swap the structure when
 788  * converting from the structure to the u32 representation.  This will
 789  * result in the structure having the correct format on the wire no matter
 790  * the host endian format.
 791  */
 792 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
 793                                           u32 *wire)
 794 {
 795         union dlm_query_join_response response;
 796
 797         response.packet = *packet;
 798         *wire = be32_to_cpu(response.intval);
 799 }
 800
 801 static void dlm_query_join_wire_to_packet(u32 wire,
 802                                           struct dlm_query_join_packet *packet)
 803 {
 804         union dlm_query_join_response response;
 805
 806         response.intval = cpu_to_be32(wire);
 807         *packet = response.packet;
 808 }
 809
 810 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
 811                                   void **ret_data)
 812 {
 813         struct dlm_query_join_request *query;
 814         struct dlm_query_join_packet packet = {
 815                 .code = JOIN_DISALLOW,
 816         };
 817         struct dlm_ctxt *dlm = NULL;
 818         u32 response;
 819         u8 nodenum;
 820
 821         query = (struct dlm_query_join_request *) msg->buf;
 822
 823         mlog(0, "node %u wants to join domain %s\n", query->node_idx,
 824                   query->domain);
 825
 826         /*
 827          * If heartbeat doesn't consider the node live, tell it
 828          * to back off and try again.  This gives heartbeat a chance
 829          * to catch up.
 830          */
 831         if (!o2hb_check_node_heartbeating_no_sem(query->node_idx)) {
 832                 mlog(0, "node %u is not in our live map yet\n",
 833                      query->node_idx);
 834
 835                 packet.code = JOIN_DISALLOW;
 836                 goto respond;
 837         }
 838
 839         packet.code = JOIN_OK_NO_MAP;
 840
 841         spin_lock(&dlm_domain_lock);
 842         dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
 843         if (!dlm)
 844                 goto unlock_respond;
 845
 846         /*
 847          * There is a small window where the joining node may not see the
 848          * node(s) that just left but still part of the cluster. DISALLOW
 849          * join request if joining node has different node map.
 850          */
 851         nodenum=0;
 852         while (nodenum < O2NM_MAX_NODES) {
 853                 if (test_bit(nodenum, dlm->domain_map)) {
 854                         if (!byte_test_bit(nodenum, query->node_map)) {
 855                                 mlog(0, "disallow join as node %u does not "
 856                                      "have node %u in its nodemap\n",
 857                                      query->node_idx, nodenum);
 858                                 packet.code = JOIN_DISALLOW;
 859                                 goto unlock_respond;
 860                         }
 861                 }
 862                 nodenum++;
 863         }
 864
 865         /* Once the dlm ctxt is marked as leaving then we don't want
 866          * to be put in someone's domain map.
 867          * Also, explicitly disallow joining at certain troublesome
 868          * times (ie. during recovery). */
 869         if (dlm->dlm_state != DLM_CTXT_LEAVING) {
 870                 int bit = query->node_idx;
 871                 spin_lock(&dlm->spinlock);
 872
 873                 if (dlm->dlm_state == DLM_CTXT_NEW &&
 874                     dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
 875                         /*If this is a brand new context and we
 876                          * haven't started our join process yet, then
 877                          * the other node won the race. */
 878                         packet.code = JOIN_OK_NO_MAP;
 879                 } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
 880                         /* Disallow parallel joins. */
 881                         packet.code = JOIN_DISALLOW;
 882                 } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
 883                         mlog(0, "node %u trying to join, but recovery "
 884                              "is ongoing.\n", bit);
 885                         packet.code = JOIN_DISALLOW;
 886                 } else if (test_bit(bit, dlm->recovery_map)) {
 887                         mlog(0, "node %u trying to join, but it "
 888                              "still needs recovery.\n", bit);
 889                         packet.code = JOIN_DISALLOW;
 890                 } else if (test_bit(bit, dlm->domain_map)) {
 891                         mlog(0, "node %u trying to join, but it "
 892                              "is still in the domain! needs recovery?\n",
 893                              bit);
 894                         packet.code = JOIN_DISALLOW;
 895                 } else {
 896                         /* Alright we're fully a part of this domain
 897                          * so we keep some state as to who's joining
 898                          * and indicate to him that needs to be fixed
 899                          * up. */
 900
 901                         /* Make sure we speak compatible locking protocols.  */
 902                         if (dlm_query_join_proto_check("DLM", bit,
 903                                                        &dlm->dlm_locking_proto,
 904                                                        &query->dlm_proto)) {
 905                                 packet.code = JOIN_PROTOCOL_MISMATCH;
 906                         } else if (dlm_query_join_proto_check("fs", bit,
 907                                                               &dlm->fs_locking_proto,
 908                                                               &query->fs_proto)) {
 909                                 packet.code = JOIN_PROTOCOL_MISMATCH;
 910                         } else {
 911                                 packet.dlm_minor = query->dlm_proto.pv_minor;
 912                                 packet.fs_minor = query->fs_proto.pv_minor;
 913                                 packet.code = JOIN_OK;
 914                                 __dlm_set_joining_node(dlm, query->node_idx);
 915                         }
 916                 }
 917
 918                 spin_unlock(&dlm->spinlock);
 919         }
 920 unlock_respond:
 921         spin_unlock(&dlm_domain_lock);
 922
 923 respond:
 924         mlog(0, "We respond with %u\n", packet.code);
 925
 926         dlm_query_join_packet_to_wire(&packet, &response);
 927         return response;
 928 }
 929
 930 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
 931                                      void **ret_data)
 932 {
 933         struct dlm_assert_joined *assert;
 934         struct dlm_ctxt *dlm = NULL;
 935
 936         assert = (struct dlm_assert_joined *) msg->buf;
 937
 938         mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
 939                   assert->domain);
 940
 941         spin_lock(&dlm_domain_lock);
 942         dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
 943         /* XXX should we consider no dlm ctxt an error? */
 944         if (dlm) {
 945                 spin_lock(&dlm->spinlock);
 946
 947                 /* Alright, this node has officially joined our
 948                  * domain. Set him in the map and clean up our
 949                  * leftover join state. */
 950                 BUG_ON(dlm->joining_node != assert->node_idx);
 951
 952                 if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
 953                         mlog(0, "dlm recovery is ongoing, disallow join\n");
 954                         spin_unlock(&dlm->spinlock);
 955                         spin_unlock(&dlm_domain_lock);
 956                         return -EAGAIN;
 957                 }
 958
 959                 set_bit(assert->node_idx, dlm->domain_map);
 960                 clear_bit(assert->node_idx, dlm->exit_domain_map);
 961                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
 962
 963                 printk(KERN_NOTICE "o2dlm: Node %u joins domain %s ",
 964                        assert->node_idx, dlm->name);
 965                 __dlm_print_nodes(dlm);
 966
 967                 /* notify anything attached to the heartbeat events */
 968                 dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
 969
 970                 spin_unlock(&dlm->spinlock);
 971         }
 972         spin_unlock(&dlm_domain_lock);
 973
 974         return 0;
 975 }
 976
 977 static int dlm_match_regions(struct dlm_ctxt *dlm,
 978                              struct dlm_query_region *qr,
 979                              char *local, int locallen)
 980 {
 981         char *remote = qr->qr_regions;
 982         char *l, *r;
 983         int localnr, i, j, foundit;
 984         int status = 0;
 985
 986         if (!o2hb_global_heartbeat_active()) {
 987                 if (qr->qr_numregions) {
 988                         mlog(ML_ERROR, "Domain %s: Joining node %d has global "
 989                              "heartbeat enabled but local node %d does not\n",
 990                              qr->qr_domain, qr->qr_node, dlm->node_num);
 991                         status = -EINVAL;
 992                 }
 993                 goto bail;
 994         }
 995
 996         if (o2hb_global_heartbeat_active() && !qr->qr_numregions) {
 997                 mlog(ML_ERROR, "Domain %s: Local node %d has global "
 998                      "heartbeat enabled but joining node %d does not\n",
 999                      qr->qr_domain, dlm->node_num, qr->qr_node);
1000                 status = -EINVAL;
1001                 goto bail;
1002         }
1003
1004         r = remote;
1005         for (i = 0; i < qr->qr_numregions; ++i) {
1006                 mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r);
1007                 r += O2HB_MAX_REGION_NAME_LEN;
1008         }
1009
1010         localnr = min(O2NM_MAX_REGIONS, locallen/O2HB_MAX_REGION_NAME_LEN);
1011         localnr = o2hb_get_all_regions(local, (u8)localnr);
1012
1013         /* compare local regions with remote */
1014         l = local;
1015         for (i = 0; i < localnr; ++i) {
1016                 foundit = 0;
1017                 r = remote;
1018                 for (j = 0; j <= qr->qr_numregions; ++j) {
1019                         if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) {
1020                                 foundit = 1;
1021                                 break;
1022                         }
1023                         r += O2HB_MAX_REGION_NAME_LEN;
1024                 }
1025                 if (!foundit) {
1026                         status = -EINVAL;
1027                         mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1028                              "in local node %d but not in joining node %d\n",
1029                              qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l,
1030                              dlm->node_num, qr->qr_node);
1031                         goto bail;
1032                 }
1033                 l += O2HB_MAX_REGION_NAME_LEN;
1034         }
1035
1036         /* compare remote with local regions */
1037         r = remote;
1038         for (i = 0; i < qr->qr_numregions; ++i) {
1039                 foundit = 0;
1040                 l = local;
1041                 for (j = 0; j < localnr; ++j) {
1042                         if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) {
1043                                 foundit = 1;
1044                                 break;
1045                         }
1046                         l += O2HB_MAX_REGION_NAME_LEN;
1047                 }
1048                 if (!foundit) {
1049                         status = -EINVAL;
1050                         mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1051                              "in joining node %d but not in local node %d\n",
1052                              qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r,
1053                              qr->qr_node, dlm->node_num);
1054                         goto bail;
1055                 }
1056                 r += O2HB_MAX_REGION_NAME_LEN;
1057         }
1058
1059 bail:
1060         return status;
1061 }
1062
1063 static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map)
1064 {
1065         struct dlm_query_region *qr = NULL;
1066         int status, ret = 0, i;
1067         char *p;
1068
1069         if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1070                 goto bail;
1071
1072         qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL);
1073         if (!qr) {
1074                 ret = -ENOMEM;
1075                 mlog_errno(ret);
1076                 goto bail;
1077         }
1078
1079         qr->qr_node = dlm->node_num;
1080         qr->qr_namelen = strlen(dlm->name);
1081         memcpy(qr->qr_domain, dlm->name, qr->qr_namelen);
1082         /* if local hb, the numregions will be zero */
1083         if (o2hb_global_heartbeat_active())
1084                 qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions,
1085                                                          O2NM_MAX_REGIONS);
1086
1087         p = qr->qr_regions;
1088         for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN)
1089                 mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p);
1090
1091         i = -1;
1092         while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1093                                   i + 1)) < O2NM_MAX_NODES) {
1094                 if (i == dlm->node_num)
1095                         continue;
1096
1097                 mlog(0, "Sending regions to node %d\n", i);
1098
1099                 ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr,
1100                                          sizeof(struct dlm_query_region),
1101                                          i, &status);
1102                 if (ret >= 0)
1103                         ret = status;
1104                 if (ret) {
1105                         mlog(ML_ERROR, "Region mismatch %d, node %d\n",
1106                              ret, i);
1107                         break;
1108                 }
1109         }
1110
1111 bail:
1112         kfree(qr);
1113         return ret;
1114 }
1115
1116 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
1117                                     void *data, void **ret_data)
1118 {
1119         struct dlm_query_region *qr;
1120         struct dlm_ctxt *dlm = NULL;
1121         char *local = NULL;
1122         int status = 0;
1123
1124         qr = (struct dlm_query_region *) msg->buf;
1125
1126         mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node,
1127              qr->qr_domain);
1128
1129         /* buffer used in dlm_mast_regions() */
1130         local = kmalloc(sizeof(qr->qr_regions), GFP_KERNEL);
1131         if (!local)
1132                 return -ENOMEM;
1133
1134         status = -EINVAL;
1135
1136         spin_lock(&dlm_domain_lock);
1137         dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen);
1138         if (!dlm) {
1139                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1140                      "before join domain\n", qr->qr_node, qr->qr_domain);
1141                 goto out_domain_lock;
1142         }
1143
1144         spin_lock(&dlm->spinlock);
1145         if (dlm->joining_node != qr->qr_node) {
1146                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1147                      "but joining node is %d\n", qr->qr_node, qr->qr_domain,
1148                      dlm->joining_node);
1149                 goto out_dlm_lock;
1150         }
1151
1152         /* Support for global heartbeat was added in 1.1 */
1153         if (dlm->dlm_locking_proto.pv_major == 1 &&
1154             dlm->dlm_locking_proto.pv_minor == 0) {
1155                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1156                      "but active dlm protocol is %d.%d\n", qr->qr_node,
1157                      qr->qr_domain, dlm->dlm_locking_proto.pv_major,
1158                      dlm->dlm_locking_proto.pv_minor);
1159                 goto out_dlm_lock;
1160         }
1161
1162         status = dlm_match_regions(dlm, qr, local, sizeof(qr->qr_regions));
1163
1164 out_dlm_lock:
1165         spin_unlock(&dlm->spinlock);
1166
1167 out_domain_lock:
1168         spin_unlock(&dlm_domain_lock);
1169
1170         kfree(local);
1171
1172         return status;
1173 }
1174
1175 static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn)
1176 {
1177         struct o2nm_node *local;
1178         struct dlm_node_info *remote;
1179         int i, j;
1180         int status = 0;
1181
1182         for (j = 0; j < qn->qn_numnodes; ++j)
1183                 mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum,
1184                      &(qn->qn_nodes[j].ni_ipv4_address),
1185                      ntohs(qn->qn_nodes[j].ni_ipv4_port));
1186
1187         for (i = 0; i < O2NM_MAX_NODES && !status; ++i) {
1188                 local = o2nm_get_node_by_num(i);
1189                 remote = NULL;
1190                 for (j = 0; j < qn->qn_numnodes; ++j) {
1191                         if (qn->qn_nodes[j].ni_nodenum == i) {
1192                                 remote = &(qn->qn_nodes[j]);
1193                                 break;
1194                         }
1195                 }
1196
1197                 if (!local && !remote)
1198                         continue;
1199
1200                 if ((local && !remote) || (!local && remote))
1201                         status = -EINVAL;
1202
1203                 if (!status &&
1204                     ((remote->ni_nodenum != local->nd_num) ||
1205                      (remote->ni_ipv4_port != local->nd_ipv4_port) ||
1206                      (remote->ni_ipv4_address != local->nd_ipv4_address)))
1207                         status = -EINVAL;
1208
1209                 if (status) {
1210                         if (remote && !local)
1211                                 mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1212                                      "registered in joining node %d but not in "
1213                                      "local node %d\n", qn->qn_domain,
1214                                      remote->ni_nodenum,
1215                                      &(remote->ni_ipv4_address),
1216                                      ntohs(remote->ni_ipv4_port),
1217                                      qn->qn_nodenum, dlm->node_num);
1218                         if (local && !remote)
1219                                 mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1220                                      "registered in local node %d but not in "
1221                                      "joining node %d\n", qn->qn_domain,
1222                                      local->nd_num, &(local->nd_ipv4_address),
1223                                      ntohs(local->nd_ipv4_port),
1224                                      dlm->node_num, qn->qn_nodenum);
1225                         BUG_ON((!local && !remote));
1226                 }
1227
1228                 if (local)
1229                         o2nm_node_put(local);
1230         }
1231
1232         return status;
1233 }
1234
1235 static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map)
1236 {
1237         struct dlm_query_nodeinfo *qn = NULL;
1238         struct o2nm_node *node;
1239         int ret = 0, status, count, i;
1240
1241         if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1242                 goto bail;
1243
1244         qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL);
1245         if (!qn) {
1246                 ret = -ENOMEM;
1247                 mlog_errno(ret);
1248                 goto bail;
1249         }
1250
1251         for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) {
1252                 node = o2nm_get_node_by_num(i);
1253                 if (!node)
1254                         continue;
1255                 qn->qn_nodes[count].ni_nodenum = node->nd_num;
1256                 qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port;
1257                 qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address;
1258                 mlog(0, "Node %3d, %pI4:%u\n", node->nd_num,
1259                      &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port));
1260                 ++count;
1261                 o2nm_node_put(node);
1262         }
1263
1264         qn->qn_nodenum = dlm->node_num;
1265         qn->qn_numnodes = count;
1266         qn->qn_namelen = strlen(dlm->name);
1267         memcpy(qn->qn_domain, dlm->name, qn->qn_namelen);
1268
1269         i = -1;
1270         while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1271                                   i + 1)) < O2NM_MAX_NODES) {
1272                 if (i == dlm->node_num)
1273                         continue;
1274
1275                 mlog(0, "Sending nodeinfo to node %d\n", i);
1276
1277                 ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
1278                                          qn, sizeof(struct dlm_query_nodeinfo),
1279                                          i, &status);
1280                 if (ret >= 0)
1281                         ret = status;
1282                 if (ret) {
1283                         mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i);
1284                         break;
1285                 }
1286         }
1287
1288 bail:
1289         kfree(qn);
1290         return ret;
1291 }
1292
1293 static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len,
1294                                       void *data, void **ret_data)
1295 {
1296         struct dlm_query_nodeinfo *qn;
1297         struct dlm_ctxt *dlm = NULL;
1298         int locked = 0, status = -EINVAL;
1299
1300         qn = (struct dlm_query_nodeinfo *) msg->buf;
1301
1302         mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum,
1303              qn->qn_domain);
1304
1305         spin_lock(&dlm_domain_lock);
1306         dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen);
1307         if (!dlm) {
1308                 mlog(ML_ERROR, "Node %d queried nodes on domain %s before "
1309                      "join domain\n", qn->qn_nodenum, qn->qn_domain);
1310                 goto bail;
1311         }
1312
1313         spin_lock(&dlm->spinlock);
1314         locked = 1;
1315         if (dlm->joining_node != qn->qn_nodenum) {
1316                 mlog(ML_ERROR, "Node %d queried nodes on domain %s but "
1317                      "joining node is %d\n", qn->qn_nodenum, qn->qn_domain,
1318                      dlm->joining_node);
1319                 goto bail;
1320         }
1321
1322         /* Support for node query was added in 1.1 */
1323         if (dlm->dlm_locking_proto.pv_major == 1 &&
1324             dlm->dlm_locking_proto.pv_minor == 0) {
1325                 mlog(ML_ERROR, "Node %d queried nodes on domain %s "
1326                      "but active dlm protocol is %d.%d\n", qn->qn_nodenum,
1327                      qn->qn_domain, dlm->dlm_locking_proto.pv_major,
1328                      dlm->dlm_locking_proto.pv_minor);
1329                 goto bail;
1330         }
1331
1332         status = dlm_match_nodes(dlm, qn);
1333
1334 bail:
1335         if (locked)
1336                 spin_unlock(&dlm->spinlock);
1337         spin_unlock(&dlm_domain_lock);
1338
1339         return status;
1340 }
1341
1342 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
1343                                    void **ret_data)
1344 {
1345         struct dlm_cancel_join *cancel;
1346         struct dlm_ctxt *dlm = NULL;
1347
1348         cancel = (struct dlm_cancel_join *) msg->buf;
1349
1350         mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
1351                   cancel->domain);
1352
1353         spin_lock(&dlm_domain_lock);
1354         dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
1355
1356         if (dlm) {
1357                 spin_lock(&dlm->spinlock);
1358
1359                 /* Yikes, this guy wants to cancel his join. No
1360                  * problem, we simply cleanup our join state. */
1361                 BUG_ON(dlm->joining_node != cancel->node_idx);
1362                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1363
1364                 spin_unlock(&dlm->spinlock);
1365         }
1366         spin_unlock(&dlm_domain_lock);
1367
1368         return 0;
1369 }
1370
1371 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
1372                                     unsigned int node)
1373 {
1374         int status;
1375         struct dlm_cancel_join cancel_msg;
1376
1377         memset(&cancel_msg, 0, sizeof(cancel_msg));
1378         cancel_msg.node_idx = dlm->node_num;
1379         cancel_msg.name_len = strlen(dlm->name);
1380         memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
1381
1382         status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1383                                     &cancel_msg, sizeof(cancel_msg), node,
1384                                     NULL);
1385         if (status < 0) {
1386                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1387                      "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1388                      node);
1389                 goto bail;
1390         }
1391
1392 bail:
1393         return status;
1394 }
1395
1396 /* map_size should be in bytes. */
1397 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
1398                                  unsigned long *node_map,
1399                                  unsigned int map_size)
1400 {
1401         int status, tmpstat;
1402         int node;
1403
1404         if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
1405                          sizeof(unsigned long))) {
1406                 mlog(ML_ERROR,
1407                      "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
1408                      map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
1409                 return -EINVAL;
1410         }
1411
1412         status = 0;
1413         node = -1;
1414         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1415                                      node + 1)) < O2NM_MAX_NODES) {
1416                 if (node == dlm->node_num)
1417                         continue;
1418
1419                 tmpstat = dlm_send_one_join_cancel(dlm, node);
1420                 if (tmpstat) {
1421                         mlog(ML_ERROR, "Error return %d cancelling join on "
1422                              "node %d\n", tmpstat, node);
1423                         if (!status)
1424                                 status = tmpstat;
1425                 }
1426         }
1427
1428         if (status)
1429                 mlog_errno(status);
1430         return status;
1431 }
1432
1433 static int dlm_request_join(struct dlm_ctxt *dlm,
1434                             int node,
1435                             enum dlm_query_join_response_code *response)
1436 {
1437         int status;
1438         struct dlm_query_join_request join_msg;
1439         struct dlm_query_join_packet packet;
1440         u32 join_resp;
1441
1442         mlog(0, "querying node %d\n", node);
1443
1444         memset(&join_msg, 0, sizeof(join_msg));
1445         join_msg.node_idx = dlm->node_num;
1446         join_msg.name_len = strlen(dlm->name);
1447         memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1448         join_msg.dlm_proto = dlm->dlm_locking_proto;
1449         join_msg.fs_proto = dlm->fs_locking_proto;
1450
1451         /* copy live node map to join message */
1452         byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1453
1454         status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1455                                     sizeof(join_msg), node, &join_resp);
1456         if (status < 0 && status != -ENOPROTOOPT) {
1457                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1458                      "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1459                      node);
1460                 goto bail;
1461         }
1462         dlm_query_join_wire_to_packet(join_resp, &packet);
1463
1464         /* -ENOPROTOOPT from the net code means the other side isn't
1465             listening for our message type -- that's fine, it means
1466             his dlm isn't up, so we can consider him a 'yes' but not
1467             joined into the domain.  */
1468         if (status == -ENOPROTOOPT) {
1469                 status = 0;
1470                 *response = JOIN_OK_NO_MAP;
1471         } else {
1472                 *response = packet.code;
1473                 switch (packet.code) {
1474                 case JOIN_DISALLOW:
1475                 case JOIN_OK_NO_MAP:
1476                         break;
1477                 case JOIN_PROTOCOL_MISMATCH:
1478                         mlog(ML_NOTICE,
1479                              "This node requested DLM locking protocol %u.%u and "
1480                              "filesystem locking protocol %u.%u.  At least one of "
1481                              "the protocol versions on node %d is not compatible, "
1482                              "disconnecting\n",
1483                              dlm->dlm_locking_proto.pv_major,
1484                              dlm->dlm_locking_proto.pv_minor,
1485                              dlm->fs_locking_proto.pv_major,
1486                              dlm->fs_locking_proto.pv_minor,
1487                              node);
1488                         status = -EPROTO;
1489                         break;
1490                 case JOIN_OK:
1491                         /* Use the same locking protocol as the remote node */
1492                         dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1493                         dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1494                         mlog(0,
1495                              "Node %d responds JOIN_OK with DLM locking protocol "
1496                              "%u.%u and fs locking protocol %u.%u\n",
1497                              node,
1498                              dlm->dlm_locking_proto.pv_major,
1499                              dlm->dlm_locking_proto.pv_minor,
1500                              dlm->fs_locking_proto.pv_major,
1501                              dlm->fs_locking_proto.pv_minor);
1502                         break;
1503                 default:
1504                         status = -EINVAL;
1505                         mlog(ML_ERROR, "invalid response %d from node %u\n",
1506                              packet.code, node);
1507                         /* Reset response to JOIN_DISALLOW */
1508                         *response = JOIN_DISALLOW;
1509                         break;
1510                 }
1511         }
1512
1513         mlog(0, "status %d, node %d response is %d\n", status, node,
1514              *response);
1515
1516 bail:
1517         return status;
1518 }
1519
1520 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1521                                     unsigned int node)
1522 {
1523         int status;
1524         int ret;
1525         struct dlm_assert_joined assert_msg;
1526
1527         mlog(0, "Sending join assert to node %u\n", node);
1528
1529         memset(&assert_msg, 0, sizeof(assert_msg));
1530         assert_msg.node_idx = dlm->node_num;
1531         assert_msg.name_len = strlen(dlm->name);
1532         memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1533
1534         status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1535                                     &assert_msg, sizeof(assert_msg), node,
1536                                     &ret);
1537         if (status < 0)
1538                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1539                      "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1540                      node);
1541         else
1542                 status = ret;
1543
1544         return status;
1545 }
1546
1547 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1548                                   unsigned long *node_map)
1549 {
1550         int status, node, live;
1551
1552         status = 0;
1553         node = -1;
1554         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1555                                      node + 1)) < O2NM_MAX_NODES) {
1556                 if (node == dlm->node_num)
1557                         continue;
1558
1559                 do {
1560                         /* It is very important that this message be
1561                          * received so we spin until either the node
1562                          * has died or it gets the message. */
1563                         status = dlm_send_one_join_assert(dlm, node);
1564
1565                         spin_lock(&dlm->spinlock);
1566                         live = test_bit(node, dlm->live_nodes_map);
1567                         spin_unlock(&dlm->spinlock);
1568
1569                         if (status) {
1570                                 mlog(ML_ERROR, "Error return %d asserting "
1571                                      "join on node %d\n", status, node);
1572
1573                                 /* give us some time between errors... */
1574                                 if (live)
1575                                         msleep(DLM_DOMAIN_BACKOFF_MS);
1576                         }
1577                 } while (status && live);
1578         }
1579 }
1580
1581 struct domain_join_ctxt {
1582         unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1583         unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1584 };
1585
1586 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1587                                    struct domain_join_ctxt *ctxt,
1588                                    enum dlm_query_join_response_code response)
1589 {
1590         int ret;
1591
1592         if (response == JOIN_DISALLOW) {
1593                 mlog(0, "Latest response of disallow -- should restart\n");
1594                 return 1;
1595         }
1596
1597         spin_lock(&dlm->spinlock);
1598         /* For now, we restart the process if the node maps have
1599          * changed at all */
1600         ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1601                      sizeof(dlm->live_nodes_map));
1602         spin_unlock(&dlm->spinlock);
1603
1604         if (ret)
1605                 mlog(0, "Node maps changed -- should restart\n");
1606
1607         return ret;
1608 }
1609
1610 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1611 {
1612         int status = 0, tmpstat, node;
1613         struct domain_join_ctxt *ctxt;
1614         enum dlm_query_join_response_code response = JOIN_DISALLOW;
1615
1616         mlog(0, "%p", dlm);
1617
1618         ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1619         if (!ctxt) {
1620                 status = -ENOMEM;
1621                 mlog_errno(status);
1622                 goto bail;
1623         }
1624
1625         /* group sem locking should work for us here -- we're already
1626          * registered for heartbeat events so filling this should be
1627          * atomic wrt getting those handlers called. */
1628         o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1629
1630         spin_lock(&dlm->spinlock);
1631         memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1632
1633         __dlm_set_joining_node(dlm, dlm->node_num);
1634
1635         spin_unlock(&dlm->spinlock);
1636
1637         node = -1;
1638         while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1639                                      node + 1)) < O2NM_MAX_NODES) {
1640                 if (node == dlm->node_num)
1641                         continue;
1642
1643                 status = dlm_request_join(dlm, node, &response);
1644                 if (status < 0) {
1645                         mlog_errno(status);
1646                         goto bail;
1647                 }
1648
1649                 /* Ok, either we got a response or the node doesn't have a
1650                  * dlm up. */
1651                 if (response == JOIN_OK)
1652                         set_bit(node, ctxt->yes_resp_map);
1653
1654                 if (dlm_should_restart_join(dlm, ctxt, response)) {
1655                         status = -EAGAIN;
1656                         goto bail;
1657                 }
1658         }
1659
1660         mlog(0, "Yay, done querying nodes!\n");
1661
1662         /* Yay, everyone agree's we can join the domain. My domain is
1663          * comprised of all nodes who were put in the
1664          * yes_resp_map. Copy that into our domain map and send a join
1665          * assert message to clean up everyone elses state. */
1666         spin_lock(&dlm->spinlock);
1667         memcpy(dlm->domain_map, ctxt->yes_resp_map,
1668                sizeof(ctxt->yes_resp_map));
1669         set_bit(dlm->node_num, dlm->domain_map);
1670         spin_unlock(&dlm->spinlock);
1671
1672         /* Support for global heartbeat and node info was added in 1.1 */
1673         if (dlm->dlm_locking_proto.pv_major > 1 ||
1674             dlm->dlm_locking_proto.pv_minor > 0) {
1675                 status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map);
1676                 if (status) {
1677                         mlog_errno(status);
1678                         goto bail;
1679                 }
1680                 status = dlm_send_regions(dlm, ctxt->yes_resp_map);
1681                 if (status) {
1682                         mlog_errno(status);
1683                         goto bail;
1684                 }
1685         }
1686
1687         dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1688
1689         /* Joined state *must* be set before the joining node
1690          * information, otherwise the query_join handler may read no
1691          * current joiner but a state of NEW and tell joining nodes
1692          * we're not in the domain. */
1693         spin_lock(&dlm_domain_lock);
1694         dlm->dlm_state = DLM_CTXT_JOINED;
1695         dlm->num_joins++;
1696         spin_unlock(&dlm_domain_lock);
1697
1698 bail:
1699         spin_lock(&dlm->spinlock);
1700         __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1701         if (!status) {
1702                 printk(KERN_NOTICE "o2dlm: Joining domain %s ", dlm->name);
1703                 __dlm_print_nodes(dlm);
1704         }
1705         spin_unlock(&dlm->spinlock);
1706
1707         if (ctxt) {
1708                 /* Do we need to send a cancel message to any nodes? */
1709                 if (status < 0) {
1710                         tmpstat = dlm_send_join_cancels(dlm,
1711                                                         ctxt->yes_resp_map,
1712                                                         sizeof(ctxt->yes_resp_map));
1713                         if (tmpstat < 0)
1714                                 mlog_errno(tmpstat);
1715                 }
1716                 kfree(ctxt);
1717         }
1718
1719         mlog(0, "returning %d\n", status);
1720         return status;
1721 }
1722
1723 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1724 {
1725         o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up);
1726         o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down);
1727         o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1728 }
1729
1730 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1731 {
1732         int status;
1733
1734         mlog(0, "registering handlers.\n");
1735
1736         o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1737                             dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1738         o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1739                             dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1740
1741         status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down);
1742         if (status)
1743                 goto bail;
1744
1745         status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up);
1746         if (status)
1747                 goto bail;
1748
1749         status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1750                                         sizeof(struct dlm_master_request),
1751                                         dlm_master_request_handler,
1752                                         dlm, NULL, &dlm->dlm_domain_handlers);
1753         if (status)
1754                 goto bail;
1755
1756         status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1757                                         sizeof(struct dlm_assert_master),
1758                                         dlm_assert_master_handler,
1759                                         dlm, dlm_assert_master_post_handler,
1760                                         &dlm->dlm_domain_handlers);
1761         if (status)
1762                 goto bail;
1763
1764         status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1765                                         sizeof(struct dlm_create_lock),
1766                                         dlm_create_lock_handler,
1767                                         dlm, NULL, &dlm->dlm_domain_handlers);
1768         if (status)
1769                 goto bail;
1770
1771         status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1772                                         DLM_CONVERT_LOCK_MAX_LEN,
1773                                         dlm_convert_lock_handler,
1774                                         dlm, NULL, &dlm->dlm_domain_handlers);
1775         if (status)
1776                 goto bail;
1777
1778         status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1779                                         DLM_UNLOCK_LOCK_MAX_LEN,
1780                                         dlm_unlock_lock_handler,
1781                                         dlm, NULL, &dlm->dlm_domain_handlers);
1782         if (status)
1783                 goto bail;
1784
1785         status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1786                                         DLM_PROXY_AST_MAX_LEN,
1787                                         dlm_proxy_ast_handler,
1788                                         dlm, NULL, &dlm->dlm_domain_handlers);
1789         if (status)
1790                 goto bail;
1791
1792         status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1793                                         sizeof(struct dlm_exit_domain),
1794                                         dlm_exit_domain_handler,
1795                                         dlm, NULL, &dlm->dlm_domain_handlers);
1796         if (status)
1797                 goto bail;
1798
1799         status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1800                                         sizeof(struct dlm_deref_lockres),
1801                                         dlm_deref_lockres_handler,
1802                                         dlm, NULL, &dlm->dlm_domain_handlers);
1803         if (status)
1804                 goto bail;
1805
1806         status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1807                                         sizeof(struct dlm_migrate_request),
1808                                         dlm_migrate_request_handler,
1809                                         dlm, NULL, &dlm->dlm_domain_handlers);
1810         if (status)
1811                 goto bail;
1812
1813         status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1814                                         DLM_MIG_LOCKRES_MAX_LEN,
1815                                         dlm_mig_lockres_handler,
1816                                         dlm, NULL, &dlm->dlm_domain_handlers);
1817         if (status)
1818                 goto bail;
1819
1820         status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1821                                         sizeof(struct dlm_master_requery),
1822                                         dlm_master_requery_handler,
1823                                         dlm, NULL, &dlm->dlm_domain_handlers);
1824         if (status)
1825                 goto bail;
1826
1827         status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1828                                         sizeof(struct dlm_lock_request),
1829                                         dlm_request_all_locks_handler,
1830                                         dlm, NULL, &dlm->dlm_domain_handlers);
1831         if (status)
1832                 goto bail;
1833
1834         status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1835                                         sizeof(struct dlm_reco_data_done),
1836                                         dlm_reco_data_done_handler,
1837                                         dlm, NULL, &dlm->dlm_domain_handlers);
1838         if (status)
1839                 goto bail;
1840
1841         status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1842                                         sizeof(struct dlm_begin_reco),
1843                                         dlm_begin_reco_handler,
1844                                         dlm, NULL, &dlm->dlm_domain_handlers);
1845         if (status)
1846                 goto bail;
1847
1848         status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1849                                         sizeof(struct dlm_finalize_reco),
1850                                         dlm_finalize_reco_handler,
1851                                         dlm, NULL, &dlm->dlm_domain_handlers);
1852         if (status)
1853                 goto bail;
1854
1855         status = o2net_register_handler(DLM_BEGIN_EXIT_DOMAIN_MSG, dlm->key,
1856                                         sizeof(struct dlm_exit_domain),
1857                                         dlm_begin_exit_domain_handler,
1858                                         dlm, NULL, &dlm->dlm_domain_handlers);
1859         if (status)
1860                 goto bail;
1861
1862         status = o2net_register_handler(DLM_DEREF_LOCKRES_DONE, dlm->key,
1863                                         sizeof(struct dlm_deref_lockres_done),
1864                                         dlm_deref_lockres_done_handler,
1865                                         dlm, NULL, &dlm->dlm_domain_handlers);
1866 bail:
1867         if (status)
1868                 dlm_unregister_domain_handlers(dlm);
1869
1870         return status;
1871 }
1872
1873 static int dlm_join_domain(struct dlm_ctxt *dlm)
1874 {
1875         int status;
1876         unsigned int backoff;
1877         unsigned int total_backoff = 0;
1878         char wq_name[O2NM_MAX_NAME_LEN];
1879
1880         BUG_ON(!dlm);
1881
1882         mlog(0, "Join domain %s\n", dlm->name);
1883
1884         status = dlm_register_domain_handlers(dlm);
1885         if (status) {
1886                 mlog_errno(status);
1887                 goto bail;
1888         }
1889
1890         status = dlm_launch_thread(dlm);
1891         if (status < 0) {
1892                 mlog_errno(status);
1893                 goto bail;
1894         }
1895
1896         status = dlm_launch_recovery_thread(dlm);
1897         if (status < 0) {
1898                 mlog_errno(status);
1899                 goto bail;
1900         }
1901
1902         status = dlm_debug_init(dlm);
1903         if (status < 0) {
1904                 mlog_errno(status);
1905                 goto bail;
1906         }
1907
1908         snprintf(wq_name, O2NM_MAX_NAME_LEN, "dlm_wq-%s", dlm->name);
1909         dlm->dlm_worker = create_singlethread_workqueue(wq_name);
1910         if (!dlm->dlm_worker) {
1911                 status = -ENOMEM;
1912                 mlog_errno(status);
1913                 goto bail;
1914         }
1915
1916         do {
1917                 status = dlm_try_to_join_domain(dlm);
1918
1919                 /* If we're racing another node to the join, then we
1920                  * need to back off temporarily and let them
1921                  * complete. */
1922 #define DLM_JOIN_TIMEOUT_MSECS  90000
1923                 if (status == -EAGAIN) {
1924                         if (signal_pending(current)) {
1925                                 status = -ERESTARTSYS;
1926                                 goto bail;
1927                         }
1928
1929                         if (total_backoff > DLM_JOIN_TIMEOUT_MSECS) {
1930                                 status = -ERESTARTSYS;
1931                                 mlog(ML_NOTICE, "Timed out joining dlm domain "
1932                                      "%s after %u msecs\n", dlm->name,
1933                                      total_backoff);
1934                                 goto bail;
1935                         }
1936
1937                         /*
1938                          * <chip> After you!
1939                          * <dale> No, after you!
1940                          * <chip> I insist!
1941                          * <dale> But you first!
1942                          * ...
1943                          */
1944                         backoff = (unsigned int)(jiffies & 0x3);
1945                         backoff *= DLM_DOMAIN_BACKOFF_MS;
1946                         total_backoff += backoff;
1947                         mlog(0, "backoff %d\n", backoff);
1948                         msleep(backoff);
1949                 }
1950         } while (status == -EAGAIN);
1951
1952         if (status < 0) {
1953                 mlog_errno(status);
1954                 goto bail;
1955         }
1956
1957         status = 0;
1958 bail:
1959         wake_up(&dlm_domain_events);
1960
1961         if (status) {
1962                 dlm_unregister_domain_handlers(dlm);
1963                 dlm_debug_shutdown(dlm);
1964                 dlm_complete_thread(dlm);
1965                 dlm_complete_recovery_thread(dlm);
1966                 dlm_destroy_dlm_worker(dlm);
1967         }
1968
1969         return status;
1970 }
1971
1972 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1973                                 u32 key)
1974 {
1975         int i;
1976         int ret;
1977         struct dlm_ctxt *dlm = NULL;
1978
1979         dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1980         if (!dlm) {
1981                 ret = -ENOMEM;
1982                 mlog_errno(ret);
1983                 goto leave;
1984         }
1985
1986         dlm->name = kstrdup(domain, GFP_KERNEL);
1987         if (dlm->name == NULL) {
1988                 ret = -ENOMEM;
1989                 mlog_errno(ret);
1990                 goto leave;
1991         }
1992
1993         dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1994         if (!dlm->lockres_hash) {
1995                 ret = -ENOMEM;
1996                 mlog_errno(ret);
1997                 goto leave;
1998         }
1999
2000         for (i = 0; i < DLM_HASH_BUCKETS; i++)
2001                 INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
2002
2003         dlm->master_hash = (struct hlist_head **)
2004                                 dlm_alloc_pagevec(DLM_HASH_PAGES);
2005         if (!dlm->master_hash) {
2006                 ret = -ENOMEM;
2007                 mlog_errno(ret);
2008                 goto leave;
2009         }
2010
2011         for (i = 0; i < DLM_HASH_BUCKETS; i++)
2012                 INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
2013
2014         dlm->key = key;
2015         dlm->node_num = o2nm_this_node();
2016
2017         ret = dlm_create_debugfs_subroot(dlm);
2018         if (ret < 0)
2019                 goto leave;
2020
2021         spin_lock_init(&dlm->spinlock);
2022         spin_lock_init(&dlm->master_lock);
2023         spin_lock_init(&dlm->ast_lock);
2024         spin_lock_init(&dlm->track_lock);
2025         INIT_LIST_HEAD(&dlm->list);
2026         INIT_LIST_HEAD(&dlm->dirty_list);
2027         INIT_LIST_HEAD(&dlm->reco.resources);
2028         INIT_LIST_HEAD(&dlm->reco.node_data);
2029         INIT_LIST_HEAD(&dlm->purge_list);
2030         INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
2031         INIT_LIST_HEAD(&dlm->tracking_list);
2032         dlm->reco.state = 0;
2033
2034         INIT_LIST_HEAD(&dlm->pending_asts);
2035         INIT_LIST_HEAD(&dlm->pending_basts);
2036
2037         mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
2038                   dlm->recovery_map, &(dlm->recovery_map[0]));
2039
2040         memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
2041         memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
2042         memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
2043
2044         dlm->dlm_thread_task = NULL;
2045         dlm->dlm_reco_thread_task = NULL;
2046         dlm->dlm_worker = NULL;
2047         init_waitqueue_head(&dlm->dlm_thread_wq);
2048         init_waitqueue_head(&dlm->dlm_reco_thread_wq);
2049         init_waitqueue_head(&dlm->reco.event);
2050         init_waitqueue_head(&dlm->ast_wq);
2051         init_waitqueue_head(&dlm->migration_wq);
2052         INIT_LIST_HEAD(&dlm->mle_hb_events);
2053
2054         dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
2055         init_waitqueue_head(&dlm->dlm_join_events);
2056
2057         dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
2058         dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
2059
2060         atomic_set(&dlm->res_tot_count, 0);
2061         atomic_set(&dlm->res_cur_count, 0);
2062         for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
2063                 atomic_set(&dlm->mle_tot_count[i], 0);
2064                 atomic_set(&dlm->mle_cur_count[i], 0);
2065         }
2066
2067         spin_lock_init(&dlm->work_lock);
2068         INIT_LIST_HEAD(&dlm->work_list);
2069         INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
2070
2071         kref_init(&dlm->dlm_refs);
2072         dlm->dlm_state = DLM_CTXT_NEW;
2073
2074         INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
2075
2076         mlog(0, "context init: refcount %u\n",
2077                   atomic_read(&dlm->dlm_refs.refcount));
2078
2079 leave:
2080         if (ret < 0 && dlm) {
2081                 if (dlm->master_hash)
2082                         dlm_free_pagevec((void **)dlm->master_hash,
2083                                         DLM_HASH_PAGES);
2084
2085                 if (dlm->lockres_hash)
2086                         dlm_free_pagevec((void **)dlm->lockres_hash,
2087                                         DLM_HASH_PAGES);
2088
2089                 kfree(dlm->name);
2090                 kfree(dlm);
2091                 dlm = NULL;
2092         }
2093         return dlm;
2094 }
2095
2096 /*
2097  * Compare a requested locking protocol version against the current one.
2098  *
2099  * If the major numbers are different, they are incompatible.
2100  * If the current minor is greater than the request, they are incompatible.
2101  * If the current minor is less than or equal to the request, they are
2102  * compatible, and the requester should run at the current minor version.
2103  */
2104 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
2105                                 struct dlm_protocol_version *request)
2106 {
2107         if (existing->pv_major != request->pv_major)
2108                 return 1;
2109
2110         if (existing->pv_minor > request->pv_minor)
2111                 return 1;
2112
2113         if (existing->pv_minor < request->pv_minor)
2114                 request->pv_minor = existing->pv_minor;
2115
2116         return 0;
2117 }
2118
2119 /*
2120  * dlm_register_domain: one-time setup per "domain".
2121  *
2122  * The filesystem passes in the requested locking version via proto.
2123  * If registration was successful, proto will contain the negotiated
2124  * locking protocol.
2125  */
2126 struct dlm_ctxt * dlm_register_domain(const char *domain,
2127                                u32 key,
2128                                struct dlm_protocol_version *fs_proto)
2129 {
2130         int ret;
2131         struct dlm_ctxt *dlm = NULL;
2132         struct dlm_ctxt *new_ctxt = NULL;
2133
2134         if (strlen(domain) >= O2NM_MAX_NAME_LEN) {
2135                 ret = -ENAMETOOLONG;
2136                 mlog(ML_ERROR, "domain name length too long\n");
2137                 goto leave;
2138         }
2139
2140         mlog(0, "register called for domain \"%s\"\n", domain);
2141
2142 retry:
2143         dlm = NULL;
2144         if (signal_pending(current)) {
2145                 ret = -ERESTARTSYS;
2146                 mlog_errno(ret);
2147                 goto leave;
2148         }
2149
2150         spin_lock(&dlm_domain_lock);
2151
2152         dlm = __dlm_lookup_domain(domain);
2153         if (dlm) {
2154                 if (dlm->dlm_state != DLM_CTXT_JOINED) {
2155                         spin_unlock(&dlm_domain_lock);
2156
2157                         mlog(0, "This ctxt is not joined yet!\n");
2158                         wait_event_interruptible(dlm_domain_events,
2159                                                  dlm_wait_on_domain_helper(
2160                                                          domain));
2161                         goto retry;
2162                 }
2163
2164                 if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
2165                         spin_unlock(&dlm_domain_lock);
2166                         mlog(ML_ERROR,
2167                              "Requested locking protocol version is not "
2168                              "compatible with already registered domain "
2169                              "\"%s\"\n", domain);
2170                         ret = -EPROTO;
2171                         goto leave;
2172                 }
2173
2174                 __dlm_get(dlm);
2175                 dlm->num_joins++;
2176
2177                 spin_unlock(&dlm_domain_lock);
2178
2179                 ret = 0;
2180                 goto leave;
2181         }
2182
2183         /* doesn't exist */
2184         if (!new_ctxt) {
2185                 spin_unlock(&dlm_domain_lock);
2186
2187                 new_ctxt = dlm_alloc_ctxt(domain, key);
2188                 if (new_ctxt)
2189                         goto retry;
2190
2191                 ret = -ENOMEM;
2192                 mlog_errno(ret);
2193                 goto leave;
2194         }
2195
2196         /* a little variable switch-a-roo here... */
2197         dlm = new_ctxt;
2198         new_ctxt = NULL;
2199
2200         /* add the new domain */
2201         list_add_tail(&dlm->list, &dlm_domains);
2202         spin_unlock(&dlm_domain_lock);
2203
2204         /*
2205          * Pass the locking protocol version into the join.  If the join
2206          * succeeds, it will have the negotiated protocol set.
2207          */
2208         dlm->dlm_locking_proto = dlm_protocol;
2209         dlm->fs_locking_proto = *fs_proto;
2210
2211         ret = dlm_join_domain(dlm);
2212         if (ret) {
2213                 mlog_errno(ret);
2214                 dlm_put(dlm);
2215                 goto leave;
2216         }
2217
2218         /* Tell the caller what locking protocol we negotiated */
2219         *fs_proto = dlm->fs_locking_proto;
2220
2221         ret = 0;
2222 leave:
2223         if (new_ctxt)
2224                 dlm_free_ctxt_mem(new_ctxt);
2225
2226         if (ret < 0)
2227                 dlm = ERR_PTR(ret);
2228
2229         return dlm;
2230 }
2231 EXPORT_SYMBOL_GPL(dlm_register_domain);
2232
2233 static LIST_HEAD(dlm_join_handlers);
2234
2235 static void dlm_unregister_net_handlers(void)
2236 {
2237         o2net_unregister_handler_list(&dlm_join_handlers);
2238 }
2239
2240 static int dlm_register_net_handlers(void)
2241 {
2242         int status = 0;
2243
2244         status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
2245                                         sizeof(struct dlm_query_join_request),
2246                                         dlm_query_join_handler,
2247                                         NULL, NULL, &dlm_join_handlers);
2248         if (status)
2249                 goto bail;
2250
2251         status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
2252                                         sizeof(struct dlm_assert_joined),
2253                                         dlm_assert_joined_handler,
2254                                         NULL, NULL, &dlm_join_handlers);
2255         if (status)
2256                 goto bail;
2257
2258         status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
2259                                         sizeof(struct dlm_cancel_join),
2260                                         dlm_cancel_join_handler,
2261                                         NULL, NULL, &dlm_join_handlers);
2262         if (status)
2263                 goto bail;
2264
2265         status = o2net_register_handler(DLM_QUERY_REGION, DLM_MOD_KEY,
2266                                         sizeof(struct dlm_query_region),
2267                                         dlm_query_region_handler,
2268                                         NULL, NULL, &dlm_join_handlers);
2269
2270         if (status)
2271                 goto bail;
2272
2273         status = o2net_register_handler(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
2274                                         sizeof(struct dlm_query_nodeinfo),
2275                                         dlm_query_nodeinfo_handler,
2276                                         NULL, NULL, &dlm_join_handlers);
2277 bail:
2278         if (status < 0)
2279                 dlm_unregister_net_handlers();
2280
2281         return status;
2282 }
2283
2284 /* Domain eviction callback handling.
2285  *
2286  * The file system requires notification of node death *before* the
2287  * dlm completes it's recovery work, otherwise it may be able to
2288  * acquire locks on resources requiring recovery. Since the dlm can
2289  * evict a node from it's domain *before* heartbeat fires, a similar
2290  * mechanism is required. */
2291
2292 /* Eviction is not expected to happen often, so a per-domain lock is
2293  * not necessary. Eviction callbacks are allowed to sleep for short
2294  * periods of time. */
2295 static DECLARE_RWSEM(dlm_callback_sem);
2296
2297 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
2298                                         int node_num)
2299 {
2300         struct dlm_eviction_cb *cb;
2301
2302         down_read(&dlm_callback_sem);
2303         list_for_each_entry(cb, &dlm->dlm_eviction_callbacks, ec_item) {
2304                 cb->ec_func(node_num, cb->ec_data);
2305         }
2306         up_read(&dlm_callback_sem);
2307 }
2308
2309 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
2310                            dlm_eviction_func *f,
2311                            void *data)
2312 {
2313         INIT_LIST_HEAD(&cb->ec_item);
2314         cb->ec_func = f;
2315         cb->ec_data = data;
2316 }
2317 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
2318
2319 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
2320                               struct dlm_eviction_cb *cb)
2321 {
2322         down_write(&dlm_callback_sem);
2323         list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
2324         up_write(&dlm_callback_sem);
2325 }
2326 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
2327
2328 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
2329 {
2330         down_write(&dlm_callback_sem);
2331         list_del_init(&cb->ec_item);
2332         up_write(&dlm_callback_sem);
2333 }
2334 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
2335
2336 static int __init dlm_init(void)
2337 {
2338         int status;
2339
2340         status = dlm_init_mle_cache();
2341         if (status) {
2342                 mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
2343                 goto error;
2344         }
2345
2346         status = dlm_init_master_caches();
2347         if (status) {
2348                 mlog(ML_ERROR, "Could not create o2dlm_lockres and "
2349                      "o2dlm_lockname slabcaches\n");
2350                 goto error;
2351         }
2352
2353         status = dlm_init_lock_cache();
2354         if (status) {
2355                 mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
2356                 goto error;
2357         }
2358
2359         status = dlm_register_net_handlers();
2360         if (status) {
2361                 mlog(ML_ERROR, "Unable to register network handlers\n");
2362                 goto error;
2363         }
2364
2365         status = dlm_create_debugfs_root();
2366         if (status)
2367                 goto error;
2368
2369         return 0;
2370 error:
2371         dlm_unregister_net_handlers();
2372         dlm_destroy_lock_cache();
2373         dlm_destroy_master_caches();
2374         dlm_destroy_mle_cache();
2375         return -1;
2376 }
2377
2378 static void __exit dlm_exit (void)
2379 {
2380         dlm_destroy_debugfs_root();
2381         dlm_unregister_net_handlers();
2382         dlm_destroy_lock_cache();
2383         dlm_destroy_master_caches();
2384         dlm_destroy_mle_cache();
2385 }
2386
2387 MODULE_AUTHOR("Oracle");
2388 MODULE_LICENSE("GPL");
2389 MODULE_DESCRIPTION("OCFS2 Distributed Lock Management");
2390
2391 module_init(dlm_init);
2392 module_exit(dlm_exit);