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/utsname.h>
  32 #include <linux/init.h>
  33 #include <linux/spinlock.h>
  34 #include <linux/delay.h>
  35 #include <linux/err.h>
  36 #include <linux/debugfs.h>
  37
  38 #include "cluster/heartbeat.h"
  39 #include "cluster/nodemanager.h"
  40 #include "cluster/tcp.h"
  41
  42 #include "dlmapi.h"
  43 #include "dlmcommon.h"
  44 #include "dlmdomain.h"
  45 #include "dlmdebug.h"
  46
  47 #include "dlmver.h"
  48
  49 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
  50 #include "cluster/masklog.h"
  51
  52 /*
  53  * ocfs2 node maps are array of long int, which limits to send them freely
  54  * across the wire due to endianness issues. To workaround this, we convert
  55  * long ints to byte arrays. Following 3 routines are helper functions to
  56  * set/test/copy bits within those array of bytes
  57  */
  58 static inline void byte_set_bit(u8 nr, u8 map[])
  59 {
  60         map[nr >> 3] |= (1UL << (nr & 7));
  61 }
  62
  63 static inline int byte_test_bit(u8 nr, u8 map[])
  64 {
  65         return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
  66 }
  67
  68 static inline void byte_copymap(u8 dmap[], unsigned long smap[],
  69                         unsigned int sz)
  70 {
  71         unsigned int nn;
  72
  73         if (!sz)
  74                 return;
  75
  76         memset(dmap, 0, ((sz + 7) >> 3));
  77         for (nn = 0 ; nn < sz; nn++)
  78                 if (test_bit(nn, smap))
  79                         byte_set_bit(nn, dmap);
  80 }
  81
  82 static void dlm_free_pagevec(void **vec, int pages)
  83 {
  84         while (pages--)
  85                 free_page((unsigned long)vec[pages]);
  86         kfree(vec);
  87 }
  88
  89 static void **dlm_alloc_pagevec(int pages)
  90 {
  91         void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL);
  92         int i;
  93
  94         if (!vec)
  95                 return NULL;
  96
  97         for (i = 0; i < pages; i++)
  98                 if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
  99                         goto out_free;
 100
 101         mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n",
 102              pages, (unsigned long)DLM_HASH_PAGES,
 103              (unsigned long)DLM_BUCKETS_PER_PAGE);
 104         return vec;
 105 out_free:
 106         dlm_free_pagevec(vec, i);
 107         return NULL;
 108 }
 109
 110 /*
 111  *
 112  * spinlock lock ordering: if multiple locks are needed, obey this ordering:
 113  *    dlm_domain_lock
 114  *    struct dlm_ctxt->spinlock
 115  *    struct dlm_lock_resource->spinlock
 116  *    struct dlm_ctxt->master_lock
 117  *    struct dlm_ctxt->ast_lock
 118  *    dlm_master_list_entry->spinlock
 119  *    dlm_lock->spinlock
 120  *
 121  */
 122
 123 DEFINE_SPINLOCK(dlm_domain_lock);
 124 LIST_HEAD(dlm_domains);
 125 static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events);
 126
 127 /*
 128  * The supported protocol version for DLM communication.  Running domains
 129  * will have a negotiated version with the same major number and a minor
 130  * number equal or smaller.  The dlm_ctxt->dlm_locking_proto field should
 131  * be used to determine what a running domain is actually using.
 132  */
 133 static const struct dlm_protocol_version dlm_protocol = {
 134         .pv_major = 1,
 135         .pv_minor = 0,
 136 };
 137
 138 #define DLM_DOMAIN_BACKOFF_MS 200
 139
 140 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
 141                                   void **ret_data);
 142 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
 143                                      void **ret_data);
 144 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
 145                                    void **ret_data);
 146 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
 147                                    void **ret_data);
 148 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
 149                                 struct dlm_protocol_version *request);
 150
 151 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
 152
 153 void __dlm_unhash_lockres(struct dlm_lock_resource *lockres)
 154 {
 155         if (!hlist_unhashed(&lockres->hash_node)) {
 156                 hlist_del_init(&lockres->hash_node);
 157                 dlm_lockres_put(lockres);
 158         }
 159 }
 160
 161 void __dlm_insert_lockres(struct dlm_ctxt *dlm,
 162                        struct dlm_lock_resource *res)
 163 {
 164         struct hlist_head *bucket;
 165         struct qstr *q;
 166
 167         assert_spin_locked(&dlm->spinlock);
 168
 169         q = &res->lockname;
 170         bucket = dlm_lockres_hash(dlm, q->hash);
 171
 172         /* get a reference for our hashtable */
 173         dlm_lockres_get(res);
 174
 175         hlist_add_head(&res->hash_node, bucket);
 176 }
 177
 178 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
 179                                                      const char *name,
 180                                                      unsigned int len,
 181                                                      unsigned int hash)
 182 {
 183         struct hlist_head *bucket;
 184         struct hlist_node *list;
 185
 186         mlog_entry("%.*s\n", len, name);
 187
 188         assert_spin_locked(&dlm->spinlock);
 189
 190         bucket = dlm_lockres_hash(dlm, hash);
 191
 192         hlist_for_each(list, bucket) {
 193                 struct dlm_lock_resource *res = hlist_entry(list,
 194                         struct dlm_lock_resource, hash_node);
 195                 if (res->lockname.name[0] != name[0])
 196                         continue;
 197                 if (unlikely(res->lockname.len != len))
 198                         continue;
 199                 if (memcmp(res->lockname.name + 1, name + 1, len - 1))
 200                         continue;
 201                 dlm_lockres_get(res);
 202                 return res;
 203         }
 204         return NULL;
 205 }
 206
 207 /* intended to be called by functions which do not care about lock
 208  * resources which are being purged (most net _handler functions).
 209  * this will return NULL for any lock resource which is found but
 210  * currently in the process of dropping its mastery reference.
 211  * use __dlm_lookup_lockres_full when you need the lock resource
 212  * regardless (e.g. dlm_get_lock_resource) */
 213 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
 214                                                 const char *name,
 215                                                 unsigned int len,
 216                                                 unsigned int hash)
 217 {
 218         struct dlm_lock_resource *res = NULL;
 219
 220         mlog_entry("%.*s\n", len, name);
 221
 222         assert_spin_locked(&dlm->spinlock);
 223
 224         res = __dlm_lookup_lockres_full(dlm, name, len, hash);
 225         if (res) {
 226                 spin_lock(&res->spinlock);
 227                 if (res->state & DLM_LOCK_RES_DROPPING_REF) {
 228                         spin_unlock(&res->spinlock);
 229                         dlm_lockres_put(res);
 230                         return NULL;
 231                 }
 232                 spin_unlock(&res->spinlock);
 233         }
 234
 235         return res;
 236 }
 237
 238 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
 239                                     const char *name,
 240                                     unsigned int len)
 241 {
 242         struct dlm_lock_resource *res;
 243         unsigned int hash = dlm_lockid_hash(name, len);
 244
 245         spin_lock(&dlm->spinlock);
 246         res = __dlm_lookup_lockres(dlm, name, len, hash);
 247         spin_unlock(&dlm->spinlock);
 248         return res;
 249 }
 250
 251 static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
 252 {
 253         struct dlm_ctxt *tmp = NULL;
 254         struct list_head *iter;
 255
 256         assert_spin_locked(&dlm_domain_lock);
 257
 258         /* tmp->name here is always NULL terminated,
 259          * but domain may not be! */
 260         list_for_each(iter, &dlm_domains) {
 261                 tmp = list_entry (iter, struct dlm_ctxt, list);
 262                 if (strlen(tmp->name) == len &&
 263                     memcmp(tmp->name, domain, len)==0)
 264                         break;
 265                 tmp = NULL;
 266         }
 267
 268         return tmp;
 269 }
 270
 271 /* For null terminated domain strings ONLY */
 272 static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
 273 {
 274         assert_spin_locked(&dlm_domain_lock);
 275
 276         return __dlm_lookup_domain_full(domain, strlen(domain));
 277 }
 278
 279
 280 /* returns true on one of two conditions:
 281  * 1) the domain does not exist
 282  * 2) the domain exists and it's state is "joined" */
 283 static int dlm_wait_on_domain_helper(const char *domain)
 284 {
 285         int ret = 0;
 286         struct dlm_ctxt *tmp = NULL;
 287
 288         spin_lock(&dlm_domain_lock);
 289
 290         tmp = __dlm_lookup_domain(domain);
 291         if (!tmp)
 292                 ret = 1;
 293         else if (tmp->dlm_state == DLM_CTXT_JOINED)
 294                 ret = 1;
 295
 296         spin_unlock(&dlm_domain_lock);
 297         return ret;
 298 }
 299
 300 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
 301 {
 302         dlm_destroy_debugfs_subroot(dlm);
 303
 304         if (dlm->lockres_hash)
 305                 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
 306
 307         if (dlm->master_hash)
 308                 dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
 309
 310         if (dlm->name)
 311                 kfree(dlm->name);
 312
 313         kfree(dlm);
 314 }
 315
 316 /* A little strange - this function will be called while holding
 317  * dlm_domain_lock and is expected to be holding it on the way out. We
 318  * will however drop and reacquire it multiple times */
 319 static void dlm_ctxt_release(struct kref *kref)
 320 {
 321         struct dlm_ctxt *dlm;
 322
 323         dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
 324
 325         BUG_ON(dlm->num_joins);
 326         BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
 327
 328         /* we may still be in the list if we hit an error during join. */
 329         list_del_init(&dlm->list);
 330
 331         spin_unlock(&dlm_domain_lock);
 332
 333         mlog(0, "freeing memory from domain %s\n", dlm->name);
 334
 335         wake_up(&dlm_domain_events);
 336
 337         dlm_free_ctxt_mem(dlm);
 338
 339         spin_lock(&dlm_domain_lock);
 340 }
 341
 342 void dlm_put(struct dlm_ctxt *dlm)
 343 {
 344         spin_lock(&dlm_domain_lock);
 345         kref_put(&dlm->dlm_refs, dlm_ctxt_release);
 346         spin_unlock(&dlm_domain_lock);
 347 }
 348
 349 static void __dlm_get(struct dlm_ctxt *dlm)
 350 {
 351         kref_get(&dlm->dlm_refs);
 352 }
 353
 354 /* given a questionable reference to a dlm object, gets a reference if
 355  * it can find it in the list, otherwise returns NULL in which case
 356  * you shouldn't trust your pointer. */
 357 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
 358 {
 359         struct list_head *iter;
 360         struct dlm_ctxt *target = NULL;
 361
 362         spin_lock(&dlm_domain_lock);
 363
 364         list_for_each(iter, &dlm_domains) {
 365                 target = list_entry (iter, struct dlm_ctxt, list);
 366
 367                 if (target == dlm) {
 368                         __dlm_get(target);
 369                         break;
 370                 }
 371
 372                 target = NULL;
 373         }
 374
 375         spin_unlock(&dlm_domain_lock);
 376
 377         return target;
 378 }
 379
 380 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
 381 {
 382         int ret;
 383
 384         spin_lock(&dlm_domain_lock);
 385         ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
 386                 (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
 387         spin_unlock(&dlm_domain_lock);
 388
 389         return ret;
 390 }
 391
 392 static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
 393 {
 394         if (dlm->dlm_worker) {
 395                 flush_workqueue(dlm->dlm_worker);
 396                 destroy_workqueue(dlm->dlm_worker);
 397                 dlm->dlm_worker = NULL;
 398         }
 399 }
 400
 401 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
 402 {
 403         dlm_unregister_domain_handlers(dlm);
 404         dlm_debug_shutdown(dlm);
 405         dlm_complete_thread(dlm);
 406         dlm_complete_recovery_thread(dlm);
 407         dlm_destroy_dlm_worker(dlm);
 408
 409         /* We've left the domain. Now we can take ourselves out of the
 410          * list and allow the kref stuff to help us free the
 411          * memory. */
 412         spin_lock(&dlm_domain_lock);
 413         list_del_init(&dlm->list);
 414         spin_unlock(&dlm_domain_lock);
 415
 416         /* Wake up anyone waiting for us to remove this domain */
 417         wake_up(&dlm_domain_events);
 418 }
 419
 420 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
 421 {
 422         int i, num, n, ret = 0;
 423         struct dlm_lock_resource *res;
 424         struct hlist_node *iter;
 425         struct hlist_head *bucket;
 426         int dropped;
 427
 428         mlog(0, "Migrating locks from domain %s\n", dlm->name);
 429
 430         num = 0;
 431         spin_lock(&dlm->spinlock);
 432         for (i = 0; i < DLM_HASH_BUCKETS; i++) {
 433 redo_bucket:
 434                 n = 0;
 435                 bucket = dlm_lockres_hash(dlm, i);
 436                 iter = bucket->first;
 437                 while (iter) {
 438                         n++;
 439                         res = hlist_entry(iter, struct dlm_lock_resource,
 440                                           hash_node);
 441                         dlm_lockres_get(res);
 442                         /* migrate, if necessary.  this will drop the dlm
 443                          * spinlock and retake it if it does migration. */
 444                         dropped = dlm_empty_lockres(dlm, res);
 445
 446                         spin_lock(&res->spinlock);
 447                         __dlm_lockres_calc_usage(dlm, res);
 448                         iter = res->hash_node.next;
 449                         spin_unlock(&res->spinlock);
 450
 451                         dlm_lockres_put(res);
 452
 453                         if (dropped)
 454                                 goto redo_bucket;
 455                 }
 456                 cond_resched_lock(&dlm->spinlock);
 457                 num += n;
 458                 mlog(0, "%s: touched %d lockreses in bucket %d "
 459                      "(tot=%d)\n", dlm->name, n, i, num);
 460         }
 461         spin_unlock(&dlm->spinlock);
 462         wake_up(&dlm->dlm_thread_wq);
 463
 464         /* let the dlm thread take care of purging, keep scanning until
 465          * nothing remains in the hash */
 466         if (num) {
 467                 mlog(0, "%s: %d lock resources in hash last pass\n",
 468                      dlm->name, num);
 469                 ret = -EAGAIN;
 470         }
 471         mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
 472         return ret;
 473 }
 474
 475 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
 476 {
 477         int ret;
 478
 479         spin_lock(&dlm->spinlock);
 480         ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
 481         spin_unlock(&dlm->spinlock);
 482
 483         return ret;
 484 }
 485
 486 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
 487 {
 488         /* Yikes, a double spinlock! I need domain_lock for the dlm
 489          * state and the dlm spinlock for join state... Sorry! */
 490 again:
 491         spin_lock(&dlm_domain_lock);
 492         spin_lock(&dlm->spinlock);
 493
 494         if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
 495                 mlog(0, "Node %d is joining, we wait on it.\n",
 496                           dlm->joining_node);
 497                 spin_unlock(&dlm->spinlock);
 498                 spin_unlock(&dlm_domain_lock);
 499
 500                 wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
 501                 goto again;
 502         }
 503
 504         dlm->dlm_state = DLM_CTXT_LEAVING;
 505         spin_unlock(&dlm->spinlock);
 506         spin_unlock(&dlm_domain_lock);
 507 }
 508
 509 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
 510 {
 511         int node = -1;
 512
 513         assert_spin_locked(&dlm->spinlock);
 514
 515         printk(KERN_INFO "ocfs2_dlm: Nodes in domain (\"%s\"): ", dlm->name);
 516
 517         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
 518                                      node + 1)) < O2NM_MAX_NODES) {
 519                 printk("%d ", node);
 520         }
 521         printk("\n");
 522 }
 523
 524 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
 525                                    void **ret_data)
 526 {
 527         struct dlm_ctxt *dlm = data;
 528         unsigned int node;
 529         struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
 530
 531         mlog_entry("%p %u %p", msg, len, data);
 532
 533         if (!dlm_grab(dlm))
 534                 return 0;
 535
 536         node = exit_msg->node_idx;
 537
 538         printk(KERN_INFO "ocfs2_dlm: Node %u leaves domain %s\n", node, dlm->name);
 539
 540         spin_lock(&dlm->spinlock);
 541         clear_bit(node, dlm->domain_map);
 542         __dlm_print_nodes(dlm);
 543
 544         /* notify anything attached to the heartbeat events */
 545         dlm_hb_event_notify_attached(dlm, node, 0);
 546
 547         spin_unlock(&dlm->spinlock);
 548
 549         dlm_put(dlm);
 550
 551         return 0;
 552 }
 553
 554 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm,
 555                                     unsigned int node)
 556 {
 557         int status;
 558         struct dlm_exit_domain leave_msg;
 559
 560         mlog(0, "Asking node %u if we can leave the domain %s me = %u\n",
 561                   node, dlm->name, dlm->node_num);
 562
 563         memset(&leave_msg, 0, sizeof(leave_msg));
 564         leave_msg.node_idx = dlm->node_num;
 565
 566         status = o2net_send_message(DLM_EXIT_DOMAIN_MSG, dlm->key,
 567                                     &leave_msg, sizeof(leave_msg), node,
 568                                     NULL);
 569
 570         mlog(0, "status return %d from o2net_send_message\n", status);
 571
 572         return status;
 573 }
 574
 575
 576 static void dlm_leave_domain(struct dlm_ctxt *dlm)
 577 {
 578         int node, clear_node, status;
 579
 580         /* At this point we've migrated away all our locks and won't
 581          * accept mastership of new ones. The dlm is responsible for
 582          * almost nothing now. We make sure not to confuse any joining
 583          * nodes and then commence shutdown procedure. */
 584
 585         spin_lock(&dlm->spinlock);
 586         /* Clear ourselves from the domain map */
 587         clear_bit(dlm->node_num, dlm->domain_map);
 588         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
 589                                      0)) < O2NM_MAX_NODES) {
 590                 /* Drop the dlm spinlock. This is safe wrt the domain_map.
 591                  * -nodes cannot be added now as the
 592                  *   query_join_handlers knows to respond with OK_NO_MAP
 593                  * -we catch the right network errors if a node is
 594                  *   removed from the map while we're sending him the
 595                  *   exit message. */
 596                 spin_unlock(&dlm->spinlock);
 597
 598                 clear_node = 1;
 599
 600                 status = dlm_send_one_domain_exit(dlm, node);
 601                 if (status < 0 &&
 602                     status != -ENOPROTOOPT &&
 603                     status != -ENOTCONN) {
 604                         mlog(ML_NOTICE, "Error %d sending domain exit message "
 605                              "to node %d\n", status, node);
 606
 607                         /* Not sure what to do here but lets sleep for
 608                          * a bit in case this was a transient
 609                          * error... */
 610                         msleep(DLM_DOMAIN_BACKOFF_MS);
 611                         clear_node = 0;
 612                 }
 613
 614                 spin_lock(&dlm->spinlock);
 615                 /* If we're not clearing the node bit then we intend
 616                  * to loop back around to try again. */
 617                 if (clear_node)
 618                         clear_bit(node, dlm->domain_map);
 619         }
 620         spin_unlock(&dlm->spinlock);
 621 }
 622
 623 int dlm_joined(struct dlm_ctxt *dlm)
 624 {
 625         int ret = 0;
 626
 627         spin_lock(&dlm_domain_lock);
 628
 629         if (dlm->dlm_state == DLM_CTXT_JOINED)
 630                 ret = 1;
 631
 632         spin_unlock(&dlm_domain_lock);
 633
 634         return ret;
 635 }
 636
 637 int dlm_shutting_down(struct dlm_ctxt *dlm)
 638 {
 639         int ret = 0;
 640
 641         spin_lock(&dlm_domain_lock);
 642
 643         if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
 644                 ret = 1;
 645
 646         spin_unlock(&dlm_domain_lock);
 647
 648         return ret;
 649 }
 650
 651 void dlm_unregister_domain(struct dlm_ctxt *dlm)
 652 {
 653         int leave = 0;
 654         struct dlm_lock_resource *res;
 655
 656         spin_lock(&dlm_domain_lock);
 657         BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
 658         BUG_ON(!dlm->num_joins);
 659
 660         dlm->num_joins--;
 661         if (!dlm->num_joins) {
 662                 /* We mark it "in shutdown" now so new register
 663                  * requests wait until we've completely left the
 664                  * domain. Don't use DLM_CTXT_LEAVING yet as we still
 665                  * want new domain joins to communicate with us at
 666                  * least until we've completed migration of our
 667                  * resources. */
 668                 dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
 669                 leave = 1;
 670         }
 671         spin_unlock(&dlm_domain_lock);
 672
 673         if (leave) {
 674                 mlog(0, "shutting down domain %s\n", dlm->name);
 675
 676                 /* We changed dlm state, notify the thread */
 677                 dlm_kick_thread(dlm, NULL);
 678
 679                 while (dlm_migrate_all_locks(dlm)) {
 680                         /* Give dlm_thread time to purge the lockres' */
 681                         msleep(500);
 682                         mlog(0, "%s: more migration to do\n", dlm->name);
 683                 }
 684
 685                 /* This list should be empty. If not, print remaining lockres */
 686                 if (!list_empty(&dlm->tracking_list)) {
 687                         mlog(ML_ERROR, "Following lockres' are still on the "
 688                              "tracking list:\n");
 689                         list_for_each_entry(res, &dlm->tracking_list, tracking)
 690                                 dlm_print_one_lock_resource(res);
 691                 }
 692
 693                 dlm_mark_domain_leaving(dlm);
 694                 dlm_leave_domain(dlm);
 695                 dlm_complete_dlm_shutdown(dlm);
 696         }
 697         dlm_put(dlm);
 698 }
 699 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
 700
 701 static int dlm_query_join_proto_check(char *proto_type, int node,
 702                                       struct dlm_protocol_version *ours,
 703                                       struct dlm_protocol_version *request)
 704 {
 705         int rc;
 706         struct dlm_protocol_version proto = *request;
 707
 708         if (!dlm_protocol_compare(ours, &proto)) {
 709                 mlog(0,
 710                      "node %u wanted to join with %s locking protocol "
 711                      "%u.%u, we respond with %u.%u\n",
 712                      node, proto_type,
 713                      request->pv_major,
 714                      request->pv_minor,
 715                      proto.pv_major, proto.pv_minor);
 716                 request->pv_minor = proto.pv_minor;
 717                 rc = 0;
 718         } else {
 719                 mlog(ML_NOTICE,
 720                      "Node %u wanted to join with %s locking "
 721                      "protocol %u.%u, but we have %u.%u, disallowing\n",
 722                      node, proto_type,
 723                      request->pv_major,
 724                      request->pv_minor,
 725                      ours->pv_major,
 726                      ours->pv_minor);
 727                 rc = 1;
 728         }
 729
 730         return rc;
 731 }
 732
 733 /*
 734  * struct dlm_query_join_packet is made up of four one-byte fields.  They
 735  * are effectively in big-endian order already.  However, little-endian
 736  * machines swap them before putting the packet on the wire (because
 737  * query_join's response is a status, and that status is treated as a u32
 738  * on the wire).  Thus, a big-endian and little-endian machines will treat
 739  * this structure differently.
 740  *
 741  * The solution is to have little-endian machines swap the structure when
 742  * converting from the structure to the u32 representation.  This will
 743  * result in the structure having the correct format on the wire no matter
 744  * the host endian format.
 745  */
 746 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
 747                                           u32 *wire)
 748 {
 749         union dlm_query_join_response response;
 750
 751         response.packet = *packet;
 752         *wire = cpu_to_be32(response.intval);
 753 }
 754
 755 static void dlm_query_join_wire_to_packet(u32 wire,
 756                                           struct dlm_query_join_packet *packet)
 757 {
 758         union dlm_query_join_response response;
 759
 760         response.intval = cpu_to_be32(wire);
 761         *packet = response.packet;
 762 }
 763
 764 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
 765                                   void **ret_data)
 766 {
 767         struct dlm_query_join_request *query;
 768         struct dlm_query_join_packet packet = {
 769                 .code = JOIN_DISALLOW,
 770         };
 771         struct dlm_ctxt *dlm = NULL;
 772         u32 response;
 773         u8 nodenum;
 774
 775         query = (struct dlm_query_join_request *) msg->buf;
 776
 777         mlog(0, "node %u wants to join domain %s\n", query->node_idx,
 778                   query->domain);
 779
 780         /*
 781          * If heartbeat doesn't consider the node live, tell it
 782          * to back off and try again.  This gives heartbeat a chance
 783          * to catch up.
 784          */
 785         if (!o2hb_check_node_heartbeating(query->node_idx)) {
 786                 mlog(0, "node %u is not in our live map yet\n",
 787                      query->node_idx);
 788
 789                 packet.code = JOIN_DISALLOW;
 790                 goto respond;
 791         }
 792
 793         packet.code = JOIN_OK_NO_MAP;
 794
 795         spin_lock(&dlm_domain_lock);
 796         dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
 797         if (!dlm)
 798                 goto unlock_respond;
 799
 800         /*
 801          * There is a small window where the joining node may not see the
 802          * node(s) that just left but still part of the cluster. DISALLOW
 803          * join request if joining node has different node map.
 804          */
 805         nodenum=0;
 806         while (nodenum < O2NM_MAX_NODES) {
 807                 if (test_bit(nodenum, dlm->domain_map)) {
 808                         if (!byte_test_bit(nodenum, query->node_map)) {
 809                                 mlog(0, "disallow join as node %u does not "
 810                                      "have node %u in its nodemap\n",
 811                                      query->node_idx, nodenum);
 812                                 packet.code = JOIN_DISALLOW;
 813                                 goto unlock_respond;
 814                         }
 815                 }
 816                 nodenum++;
 817         }
 818
 819         /* Once the dlm ctxt is marked as leaving then we don't want
 820          * to be put in someone's domain map.
 821          * Also, explicitly disallow joining at certain troublesome
 822          * times (ie. during recovery). */
 823         if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) {
 824                 int bit = query->node_idx;
 825                 spin_lock(&dlm->spinlock);
 826
 827                 if (dlm->dlm_state == DLM_CTXT_NEW &&
 828                     dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
 829                         /*If this is a brand new context and we
 830                          * haven't started our join process yet, then
 831                          * the other node won the race. */
 832                         packet.code = JOIN_OK_NO_MAP;
 833                 } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
 834                         /* Disallow parallel joins. */
 835                         packet.code = JOIN_DISALLOW;
 836                 } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
 837                         mlog(0, "node %u trying to join, but recovery "
 838                              "is ongoing.\n", bit);
 839                         packet.code = JOIN_DISALLOW;
 840                 } else if (test_bit(bit, dlm->recovery_map)) {
 841                         mlog(0, "node %u trying to join, but it "
 842                              "still needs recovery.\n", bit);
 843                         packet.code = JOIN_DISALLOW;
 844                 } else if (test_bit(bit, dlm->domain_map)) {
 845                         mlog(0, "node %u trying to join, but it "
 846                              "is still in the domain! needs recovery?\n",
 847                              bit);
 848                         packet.code = JOIN_DISALLOW;
 849                 } else {
 850                         /* Alright we're fully a part of this domain
 851                          * so we keep some state as to who's joining
 852                          * and indicate to him that needs to be fixed
 853                          * up. */
 854
 855                         /* Make sure we speak compatible locking protocols.  */
 856                         if (dlm_query_join_proto_check("DLM", bit,
 857                                                        &dlm->dlm_locking_proto,
 858                                                        &query->dlm_proto)) {
 859                                 packet.code = JOIN_PROTOCOL_MISMATCH;
 860                         } else if (dlm_query_join_proto_check("fs", bit,
 861                                                               &dlm->fs_locking_proto,
 862                                                               &query->fs_proto)) {
 863                                 packet.code = JOIN_PROTOCOL_MISMATCH;
 864                         } else {
 865                                 packet.dlm_minor = query->dlm_proto.pv_minor;
 866                                 packet.fs_minor = query->fs_proto.pv_minor;
 867                                 packet.code = JOIN_OK;
 868                                 __dlm_set_joining_node(dlm, query->node_idx);
 869                         }
 870                 }
 871
 872                 spin_unlock(&dlm->spinlock);
 873         }
 874 unlock_respond:
 875         spin_unlock(&dlm_domain_lock);
 876
 877 respond:
 878         mlog(0, "We respond with %u\n", packet.code);
 879
 880         dlm_query_join_packet_to_wire(&packet, &response);
 881         return response;
 882 }
 883
 884 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
 885                                      void **ret_data)
 886 {
 887         struct dlm_assert_joined *assert;
 888         struct dlm_ctxt *dlm = NULL;
 889
 890         assert = (struct dlm_assert_joined *) msg->buf;
 891
 892         mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
 893                   assert->domain);
 894
 895         spin_lock(&dlm_domain_lock);
 896         dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
 897         /* XXX should we consider no dlm ctxt an error? */
 898         if (dlm) {
 899                 spin_lock(&dlm->spinlock);
 900
 901                 /* Alright, this node has officially joined our
 902                  * domain. Set him in the map and clean up our
 903                  * leftover join state. */
 904                 BUG_ON(dlm->joining_node != assert->node_idx);
 905                 set_bit(assert->node_idx, dlm->domain_map);
 906                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
 907
 908                 printk(KERN_INFO "ocfs2_dlm: Node %u joins domain %s\n",
 909                        assert->node_idx, dlm->name);
 910                 __dlm_print_nodes(dlm);
 911
 912                 /* notify anything attached to the heartbeat events */
 913                 dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
 914
 915                 spin_unlock(&dlm->spinlock);
 916         }
 917         spin_unlock(&dlm_domain_lock);
 918
 919         return 0;
 920 }
 921
 922 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
 923                                    void **ret_data)
 924 {
 925         struct dlm_cancel_join *cancel;
 926         struct dlm_ctxt *dlm = NULL;
 927
 928         cancel = (struct dlm_cancel_join *) msg->buf;
 929
 930         mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
 931                   cancel->domain);
 932
 933         spin_lock(&dlm_domain_lock);
 934         dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
 935
 936         if (dlm) {
 937                 spin_lock(&dlm->spinlock);
 938
 939                 /* Yikes, this guy wants to cancel his join. No
 940                  * problem, we simply cleanup our join state. */
 941                 BUG_ON(dlm->joining_node != cancel->node_idx);
 942                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
 943
 944                 spin_unlock(&dlm->spinlock);
 945         }
 946         spin_unlock(&dlm_domain_lock);
 947
 948         return 0;
 949 }
 950
 951 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
 952                                     unsigned int node)
 953 {
 954         int status;
 955         struct dlm_cancel_join cancel_msg;
 956
 957         memset(&cancel_msg, 0, sizeof(cancel_msg));
 958         cancel_msg.node_idx = dlm->node_num;
 959         cancel_msg.name_len = strlen(dlm->name);
 960         memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
 961
 962         status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
 963                                     &cancel_msg, sizeof(cancel_msg), node,
 964                                     NULL);
 965         if (status < 0) {
 966                 mlog_errno(status);
 967                 goto bail;
 968         }
 969
 970 bail:
 971         return status;
 972 }
 973
 974 /* map_size should be in bytes. */
 975 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
 976                                  unsigned long *node_map,
 977                                  unsigned int map_size)
 978 {
 979         int status, tmpstat;
 980         unsigned int node;
 981
 982         if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
 983                          sizeof(unsigned long))) {
 984                 mlog(ML_ERROR,
 985                      "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
 986                      map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
 987                 return -EINVAL;
 988         }
 989
 990         status = 0;
 991         node = -1;
 992         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
 993                                      node + 1)) < O2NM_MAX_NODES) {
 994                 if (node == dlm->node_num)
 995                         continue;
 996
 997                 tmpstat = dlm_send_one_join_cancel(dlm, node);
 998                 if (tmpstat) {
 999                         mlog(ML_ERROR, "Error return %d cancelling join on "
1000                              "node %d\n", tmpstat, node);
1001                         if (!status)
1002                                 status = tmpstat;
1003                 }
1004         }
1005
1006         if (status)
1007                 mlog_errno(status);
1008         return status;
1009 }
1010
1011 static int dlm_request_join(struct dlm_ctxt *dlm,
1012                             int node,
1013                             enum dlm_query_join_response_code *response)
1014 {
1015         int status;
1016         struct dlm_query_join_request join_msg;
1017         struct dlm_query_join_packet packet;
1018         u32 join_resp;
1019
1020         mlog(0, "querying node %d\n", node);
1021
1022         memset(&join_msg, 0, sizeof(join_msg));
1023         join_msg.node_idx = dlm->node_num;
1024         join_msg.name_len = strlen(dlm->name);
1025         memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1026         join_msg.dlm_proto = dlm->dlm_locking_proto;
1027         join_msg.fs_proto = dlm->fs_locking_proto;
1028
1029         /* copy live node map to join message */
1030         byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1031
1032         status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1033                                     sizeof(join_msg), node,
1034                                     &join_resp);
1035         if (status < 0 && status != -ENOPROTOOPT) {
1036                 mlog_errno(status);
1037                 goto bail;
1038         }
1039         dlm_query_join_wire_to_packet(join_resp, &packet);
1040
1041         /* -ENOPROTOOPT from the net code means the other side isn't
1042             listening for our message type -- that's fine, it means
1043             his dlm isn't up, so we can consider him a 'yes' but not
1044             joined into the domain.  */
1045         if (status == -ENOPROTOOPT) {
1046                 status = 0;
1047                 *response = JOIN_OK_NO_MAP;
1048         } else if (packet.code == JOIN_DISALLOW ||
1049                    packet.code == JOIN_OK_NO_MAP) {
1050                 *response = packet.code;
1051         } else if (packet.code == JOIN_PROTOCOL_MISMATCH) {
1052                 mlog(ML_NOTICE,
1053                      "This node requested DLM locking protocol %u.%u and "
1054                      "filesystem locking protocol %u.%u.  At least one of "
1055                      "the protocol versions on node %d is not compatible, "
1056                      "disconnecting\n",
1057                      dlm->dlm_locking_proto.pv_major,
1058                      dlm->dlm_locking_proto.pv_minor,
1059                      dlm->fs_locking_proto.pv_major,
1060                      dlm->fs_locking_proto.pv_minor,
1061                      node);
1062                 status = -EPROTO;
1063                 *response = packet.code;
1064         } else if (packet.code == JOIN_OK) {
1065                 *response = packet.code;
1066                 /* Use the same locking protocol as the remote node */
1067                 dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1068                 dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1069                 mlog(0,
1070                      "Node %d responds JOIN_OK with DLM locking protocol "
1071                      "%u.%u and fs locking protocol %u.%u\n",
1072                      node,
1073                      dlm->dlm_locking_proto.pv_major,
1074                      dlm->dlm_locking_proto.pv_minor,
1075                      dlm->fs_locking_proto.pv_major,
1076                      dlm->fs_locking_proto.pv_minor);
1077         } else {
1078                 status = -EINVAL;
1079                 mlog(ML_ERROR, "invalid response %d from node %u\n",
1080                      packet.code, node);
1081         }
1082
1083         mlog(0, "status %d, node %d response is %d\n", status, node,
1084              *response);
1085
1086 bail:
1087         return status;
1088 }
1089
1090 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1091                                     unsigned int node)
1092 {
1093         int status;
1094         struct dlm_assert_joined assert_msg;
1095
1096         mlog(0, "Sending join assert to node %u\n", node);
1097
1098         memset(&assert_msg, 0, sizeof(assert_msg));
1099         assert_msg.node_idx = dlm->node_num;
1100         assert_msg.name_len = strlen(dlm->name);
1101         memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1102
1103         status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1104                                     &assert_msg, sizeof(assert_msg), node,
1105                                     NULL);
1106         if (status < 0)
1107                 mlog_errno(status);
1108
1109         return status;
1110 }
1111
1112 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1113                                   unsigned long *node_map)
1114 {
1115         int status, node, live;
1116
1117         status = 0;
1118         node = -1;
1119         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1120                                      node + 1)) < O2NM_MAX_NODES) {
1121                 if (node == dlm->node_num)
1122                         continue;
1123
1124                 do {
1125                         /* It is very important that this message be
1126                          * received so we spin until either the node
1127                          * has died or it gets the message. */
1128                         status = dlm_send_one_join_assert(dlm, node);
1129
1130                         spin_lock(&dlm->spinlock);
1131                         live = test_bit(node, dlm->live_nodes_map);
1132                         spin_unlock(&dlm->spinlock);
1133
1134                         if (status) {
1135                                 mlog(ML_ERROR, "Error return %d asserting "
1136                                      "join on node %d\n", status, node);
1137
1138                                 /* give us some time between errors... */
1139                                 if (live)
1140                                         msleep(DLM_DOMAIN_BACKOFF_MS);
1141                         }
1142                 } while (status && live);
1143         }
1144 }
1145
1146 struct domain_join_ctxt {
1147         unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1148         unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1149 };
1150
1151 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1152                                    struct domain_join_ctxt *ctxt,
1153                                    enum dlm_query_join_response_code response)
1154 {
1155         int ret;
1156
1157         if (response == JOIN_DISALLOW) {
1158                 mlog(0, "Latest response of disallow -- should restart\n");
1159                 return 1;
1160         }
1161
1162         spin_lock(&dlm->spinlock);
1163         /* For now, we restart the process if the node maps have
1164          * changed at all */
1165         ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1166                      sizeof(dlm->live_nodes_map));
1167         spin_unlock(&dlm->spinlock);
1168
1169         if (ret)
1170                 mlog(0, "Node maps changed -- should restart\n");
1171
1172         return ret;
1173 }
1174
1175 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1176 {
1177         int status = 0, tmpstat, node;
1178         struct domain_join_ctxt *ctxt;
1179         enum dlm_query_join_response_code response = JOIN_DISALLOW;
1180
1181         mlog_entry("%p", dlm);
1182
1183         ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1184         if (!ctxt) {
1185                 status = -ENOMEM;
1186                 mlog_errno(status);
1187                 goto bail;
1188         }
1189
1190         /* group sem locking should work for us here -- we're already
1191          * registered for heartbeat events so filling this should be
1192          * atomic wrt getting those handlers called. */
1193         o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1194
1195         spin_lock(&dlm->spinlock);
1196         memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1197
1198         __dlm_set_joining_node(dlm, dlm->node_num);
1199
1200         spin_unlock(&dlm->spinlock);
1201
1202         node = -1;
1203         while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1204                                      node + 1)) < O2NM_MAX_NODES) {
1205                 if (node == dlm->node_num)
1206                         continue;
1207
1208                 status = dlm_request_join(dlm, node, &response);
1209                 if (status < 0) {
1210                         mlog_errno(status);
1211                         goto bail;
1212                 }
1213
1214                 /* Ok, either we got a response or the node doesn't have a
1215                  * dlm up. */
1216                 if (response == JOIN_OK)
1217                         set_bit(node, ctxt->yes_resp_map);
1218
1219                 if (dlm_should_restart_join(dlm, ctxt, response)) {
1220                         status = -EAGAIN;
1221                         goto bail;
1222                 }
1223         }
1224
1225         mlog(0, "Yay, done querying nodes!\n");
1226
1227         /* Yay, everyone agree's we can join the domain. My domain is
1228          * comprised of all nodes who were put in the
1229          * yes_resp_map. Copy that into our domain map and send a join
1230          * assert message to clean up everyone elses state. */
1231         spin_lock(&dlm->spinlock);
1232         memcpy(dlm->domain_map, ctxt->yes_resp_map,
1233                sizeof(ctxt->yes_resp_map));
1234         set_bit(dlm->node_num, dlm->domain_map);
1235         spin_unlock(&dlm->spinlock);
1236
1237         dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1238
1239         /* Joined state *must* be set before the joining node
1240          * information, otherwise the query_join handler may read no
1241          * current joiner but a state of NEW and tell joining nodes
1242          * we're not in the domain. */
1243         spin_lock(&dlm_domain_lock);
1244         dlm->dlm_state = DLM_CTXT_JOINED;
1245         dlm->num_joins++;
1246         spin_unlock(&dlm_domain_lock);
1247
1248 bail:
1249         spin_lock(&dlm->spinlock);
1250         __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1251         if (!status)
1252                 __dlm_print_nodes(dlm);
1253         spin_unlock(&dlm->spinlock);
1254
1255         if (ctxt) {
1256                 /* Do we need to send a cancel message to any nodes? */
1257                 if (status < 0) {
1258                         tmpstat = dlm_send_join_cancels(dlm,
1259                                                         ctxt->yes_resp_map,
1260                                                         sizeof(ctxt->yes_resp_map));
1261                         if (tmpstat < 0)
1262                                 mlog_errno(tmpstat);
1263                 }
1264                 kfree(ctxt);
1265         }
1266
1267         mlog(0, "returning %d\n", status);
1268         return status;
1269 }
1270
1271 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1272 {
1273         o2hb_unregister_callback(NULL, &dlm->dlm_hb_up);
1274         o2hb_unregister_callback(NULL, &dlm->dlm_hb_down);
1275         o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1276 }
1277
1278 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1279 {
1280         int status;
1281
1282         mlog(0, "registering handlers.\n");
1283
1284         o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1285                             dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1286         status = o2hb_register_callback(NULL, &dlm->dlm_hb_down);
1287         if (status)
1288                 goto bail;
1289
1290         o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1291                             dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1292         status = o2hb_register_callback(NULL, &dlm->dlm_hb_up);
1293         if (status)
1294                 goto bail;
1295
1296         status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1297                                         sizeof(struct dlm_master_request),
1298                                         dlm_master_request_handler,
1299                                         dlm, NULL, &dlm->dlm_domain_handlers);
1300         if (status)
1301                 goto bail;
1302
1303         status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1304                                         sizeof(struct dlm_assert_master),
1305                                         dlm_assert_master_handler,
1306                                         dlm, dlm_assert_master_post_handler,
1307                                         &dlm->dlm_domain_handlers);
1308         if (status)
1309                 goto bail;
1310
1311         status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1312                                         sizeof(struct dlm_create_lock),
1313                                         dlm_create_lock_handler,
1314                                         dlm, NULL, &dlm->dlm_domain_handlers);
1315         if (status)
1316                 goto bail;
1317
1318         status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1319                                         DLM_CONVERT_LOCK_MAX_LEN,
1320                                         dlm_convert_lock_handler,
1321                                         dlm, NULL, &dlm->dlm_domain_handlers);
1322         if (status)
1323                 goto bail;
1324
1325         status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1326                                         DLM_UNLOCK_LOCK_MAX_LEN,
1327                                         dlm_unlock_lock_handler,
1328                                         dlm, NULL, &dlm->dlm_domain_handlers);
1329         if (status)
1330                 goto bail;
1331
1332         status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1333                                         DLM_PROXY_AST_MAX_LEN,
1334                                         dlm_proxy_ast_handler,
1335                                         dlm, NULL, &dlm->dlm_domain_handlers);
1336         if (status)
1337                 goto bail;
1338
1339         status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1340                                         sizeof(struct dlm_exit_domain),
1341                                         dlm_exit_domain_handler,
1342                                         dlm, NULL, &dlm->dlm_domain_handlers);
1343         if (status)
1344                 goto bail;
1345
1346         status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1347                                         sizeof(struct dlm_deref_lockres),
1348                                         dlm_deref_lockres_handler,
1349                                         dlm, NULL, &dlm->dlm_domain_handlers);
1350         if (status)
1351                 goto bail;
1352
1353         status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1354                                         sizeof(struct dlm_migrate_request),
1355                                         dlm_migrate_request_handler,
1356                                         dlm, NULL, &dlm->dlm_domain_handlers);
1357         if (status)
1358                 goto bail;
1359
1360         status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1361                                         DLM_MIG_LOCKRES_MAX_LEN,
1362                                         dlm_mig_lockres_handler,
1363                                         dlm, NULL, &dlm->dlm_domain_handlers);
1364         if (status)
1365                 goto bail;
1366
1367         status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1368                                         sizeof(struct dlm_master_requery),
1369                                         dlm_master_requery_handler,
1370                                         dlm, NULL, &dlm->dlm_domain_handlers);
1371         if (status)
1372                 goto bail;
1373
1374         status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1375                                         sizeof(struct dlm_lock_request),
1376                                         dlm_request_all_locks_handler,
1377                                         dlm, NULL, &dlm->dlm_domain_handlers);
1378         if (status)
1379                 goto bail;
1380
1381         status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1382                                         sizeof(struct dlm_reco_data_done),
1383                                         dlm_reco_data_done_handler,
1384                                         dlm, NULL, &dlm->dlm_domain_handlers);
1385         if (status)
1386                 goto bail;
1387
1388         status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1389                                         sizeof(struct dlm_begin_reco),
1390                                         dlm_begin_reco_handler,
1391                                         dlm, NULL, &dlm->dlm_domain_handlers);
1392         if (status)
1393                 goto bail;
1394
1395         status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1396                                         sizeof(struct dlm_finalize_reco),
1397                                         dlm_finalize_reco_handler,
1398                                         dlm, NULL, &dlm->dlm_domain_handlers);
1399         if (status)
1400                 goto bail;
1401
1402 bail:
1403         if (status)
1404                 dlm_unregister_domain_handlers(dlm);
1405
1406         return status;
1407 }
1408
1409 static int dlm_join_domain(struct dlm_ctxt *dlm)
1410 {
1411         int status;
1412         unsigned int backoff;
1413         unsigned int total_backoff = 0;
1414
1415         BUG_ON(!dlm);
1416
1417         mlog(0, "Join domain %s\n", dlm->name);
1418
1419         status = dlm_register_domain_handlers(dlm);
1420         if (status) {
1421                 mlog_errno(status);
1422                 goto bail;
1423         }
1424
1425         status = dlm_debug_init(dlm);
1426         if (status < 0) {
1427                 mlog_errno(status);
1428                 goto bail;
1429         }
1430
1431         status = dlm_launch_thread(dlm);
1432         if (status < 0) {
1433                 mlog_errno(status);
1434                 goto bail;
1435         }
1436
1437         status = dlm_launch_recovery_thread(dlm);
1438         if (status < 0) {
1439                 mlog_errno(status);
1440                 goto bail;
1441         }
1442
1443         dlm->dlm_worker = create_singlethread_workqueue("dlm_wq");
1444         if (!dlm->dlm_worker) {
1445                 status = -ENOMEM;
1446                 mlog_errno(status);
1447                 goto bail;
1448         }
1449
1450         do {
1451                 status = dlm_try_to_join_domain(dlm);
1452
1453                 /* If we're racing another node to the join, then we
1454                  * need to back off temporarily and let them
1455                  * complete. */
1456 #define DLM_JOIN_TIMEOUT_MSECS  90000
1457                 if (status == -EAGAIN) {
1458                         if (signal_pending(current)) {
1459                                 status = -ERESTARTSYS;
1460                                 goto bail;
1461                         }
1462
1463                         if (total_backoff >
1464                             msecs_to_jiffies(DLM_JOIN_TIMEOUT_MSECS)) {
1465                                 status = -ERESTARTSYS;
1466                                 mlog(ML_NOTICE, "Timed out joining dlm domain "
1467                                      "%s after %u msecs\n", dlm->name,
1468                                      jiffies_to_msecs(total_backoff));
1469                                 goto bail;
1470                         }
1471
1472                         /*
1473                          * <chip> After you!
1474                          * <dale> No, after you!
1475                          * <chip> I insist!
1476                          * <dale> But you first!
1477                          * ...
1478                          */
1479                         backoff = (unsigned int)(jiffies & 0x3);
1480                         backoff *= DLM_DOMAIN_BACKOFF_MS;
1481                         total_backoff += backoff;
1482                         mlog(0, "backoff %d\n", backoff);
1483                         msleep(backoff);
1484                 }
1485         } while (status == -EAGAIN);
1486
1487         if (status < 0) {
1488                 mlog_errno(status);
1489                 goto bail;
1490         }
1491
1492         status = 0;
1493 bail:
1494         wake_up(&dlm_domain_events);
1495
1496         if (status) {
1497                 dlm_unregister_domain_handlers(dlm);
1498                 dlm_debug_shutdown(dlm);
1499                 dlm_complete_thread(dlm);
1500                 dlm_complete_recovery_thread(dlm);
1501                 dlm_destroy_dlm_worker(dlm);
1502         }
1503
1504         return status;
1505 }
1506
1507 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1508                                 u32 key)
1509 {
1510         int i;
1511         int ret;
1512         struct dlm_ctxt *dlm = NULL;
1513
1514         dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1515         if (!dlm) {
1516                 mlog_errno(-ENOMEM);
1517                 goto leave;
1518         }
1519
1520         dlm->name = kmalloc(strlen(domain) + 1, GFP_KERNEL);
1521         if (dlm->name == NULL) {
1522                 mlog_errno(-ENOMEM);
1523                 kfree(dlm);
1524                 dlm = NULL;
1525                 goto leave;
1526         }
1527
1528         dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1529         if (!dlm->lockres_hash) {
1530                 mlog_errno(-ENOMEM);
1531                 kfree(dlm->name);
1532                 kfree(dlm);
1533                 dlm = NULL;
1534                 goto leave;
1535         }
1536
1537         for (i = 0; i < DLM_HASH_BUCKETS; i++)
1538                 INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
1539
1540         dlm->master_hash = (struct hlist_head **)
1541                                 dlm_alloc_pagevec(DLM_HASH_PAGES);
1542         if (!dlm->master_hash) {
1543                 mlog_errno(-ENOMEM);
1544                 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
1545                 kfree(dlm->name);
1546                 kfree(dlm);
1547                 dlm = NULL;
1548                 goto leave;
1549         }
1550
1551         for (i = 0; i < DLM_HASH_BUCKETS; i++)
1552                 INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
1553
1554         strcpy(dlm->name, domain);
1555         dlm->key = key;
1556         dlm->node_num = o2nm_this_node();
1557
1558         ret = dlm_create_debugfs_subroot(dlm);
1559         if (ret < 0) {
1560                 dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
1561                 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
1562                 kfree(dlm->name);
1563                 kfree(dlm);
1564                 dlm = NULL;
1565                 goto leave;
1566         }
1567
1568         spin_lock_init(&dlm->spinlock);
1569         spin_lock_init(&dlm->master_lock);
1570         spin_lock_init(&dlm->ast_lock);
1571         spin_lock_init(&dlm->track_lock);
1572         INIT_LIST_HEAD(&dlm->list);
1573         INIT_LIST_HEAD(&dlm->dirty_list);
1574         INIT_LIST_HEAD(&dlm->reco.resources);
1575         INIT_LIST_HEAD(&dlm->reco.received);
1576         INIT_LIST_HEAD(&dlm->reco.node_data);
1577         INIT_LIST_HEAD(&dlm->purge_list);
1578         INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
1579         INIT_LIST_HEAD(&dlm->tracking_list);
1580         dlm->reco.state = 0;
1581
1582         INIT_LIST_HEAD(&dlm->pending_asts);
1583         INIT_LIST_HEAD(&dlm->pending_basts);
1584
1585         mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
1586                   dlm->recovery_map, &(dlm->recovery_map[0]));
1587
1588         memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
1589         memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
1590         memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
1591
1592         dlm->dlm_thread_task = NULL;
1593         dlm->dlm_reco_thread_task = NULL;
1594         dlm->dlm_worker = NULL;
1595         init_waitqueue_head(&dlm->dlm_thread_wq);
1596         init_waitqueue_head(&dlm->dlm_reco_thread_wq);
1597         init_waitqueue_head(&dlm->reco.event);
1598         init_waitqueue_head(&dlm->ast_wq);
1599         init_waitqueue_head(&dlm->migration_wq);
1600         INIT_LIST_HEAD(&dlm->mle_hb_events);
1601
1602         dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
1603         init_waitqueue_head(&dlm->dlm_join_events);
1604
1605         dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
1606         dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
1607
1608         atomic_set(&dlm->res_tot_count, 0);
1609         atomic_set(&dlm->res_cur_count, 0);
1610         for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
1611                 atomic_set(&dlm->mle_tot_count[i], 0);
1612                 atomic_set(&dlm->mle_cur_count[i], 0);
1613         }
1614
1615         spin_lock_init(&dlm->work_lock);
1616         INIT_LIST_HEAD(&dlm->work_list);
1617         INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
1618
1619         kref_init(&dlm->dlm_refs);
1620         dlm->dlm_state = DLM_CTXT_NEW;
1621
1622         INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
1623
1624         mlog(0, "context init: refcount %u\n",
1625                   atomic_read(&dlm->dlm_refs.refcount));
1626
1627 leave:
1628         return dlm;
1629 }
1630
1631 /*
1632  * Compare a requested locking protocol version against the current one.
1633  *
1634  * If the major numbers are different, they are incompatible.
1635  * If the current minor is greater than the request, they are incompatible.
1636  * If the current minor is less than or equal to the request, they are
1637  * compatible, and the requester should run at the current minor version.
1638  */
1639 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
1640                                 struct dlm_protocol_version *request)
1641 {
1642         if (existing->pv_major != request->pv_major)
1643                 return 1;
1644
1645         if (existing->pv_minor > request->pv_minor)
1646                 return 1;
1647
1648         if (existing->pv_minor < request->pv_minor)
1649                 request->pv_minor = existing->pv_minor;
1650
1651         return 0;
1652 }
1653
1654 /*
1655  * dlm_register_domain: one-time setup per "domain".
1656  *
1657  * The filesystem passes in the requested locking version via proto.
1658  * If registration was successful, proto will contain the negotiated
1659  * locking protocol.
1660  */
1661 struct dlm_ctxt * dlm_register_domain(const char *domain,
1662                                u32 key,
1663                                struct dlm_protocol_version *fs_proto)
1664 {
1665         int ret;
1666         struct dlm_ctxt *dlm = NULL;
1667         struct dlm_ctxt *new_ctxt = NULL;
1668
1669         if (strlen(domain) > O2NM_MAX_NAME_LEN) {
1670                 ret = -ENAMETOOLONG;
1671                 mlog(ML_ERROR, "domain name length too long\n");
1672                 goto leave;
1673         }
1674
1675         if (!o2hb_check_local_node_heartbeating()) {
1676                 mlog(ML_ERROR, "the local node has not been configured, or is "
1677                      "not heartbeating\n");
1678                 ret = -EPROTO;
1679                 goto leave;
1680         }
1681
1682         mlog(0, "register called for domain \"%s\"\n", domain);
1683
1684 retry:
1685         dlm = NULL;
1686         if (signal_pending(current)) {
1687                 ret = -ERESTARTSYS;
1688                 mlog_errno(ret);
1689                 goto leave;
1690         }
1691
1692         spin_lock(&dlm_domain_lock);
1693
1694         dlm = __dlm_lookup_domain(domain);
1695         if (dlm) {
1696                 if (dlm->dlm_state != DLM_CTXT_JOINED) {
1697                         spin_unlock(&dlm_domain_lock);
1698
1699                         mlog(0, "This ctxt is not joined yet!\n");
1700                         wait_event_interruptible(dlm_domain_events,
1701                                                  dlm_wait_on_domain_helper(
1702                                                          domain));
1703                         goto retry;
1704                 }
1705
1706                 if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
1707                         mlog(ML_ERROR,
1708                              "Requested locking protocol version is not "
1709                              "compatible with already registered domain "
1710                              "\"%s\"\n", domain);
1711                         ret = -EPROTO;
1712                         goto leave;
1713                 }
1714
1715                 __dlm_get(dlm);
1716                 dlm->num_joins++;
1717
1718                 spin_unlock(&dlm_domain_lock);
1719
1720                 ret = 0;
1721                 goto leave;
1722         }
1723
1724         /* doesn't exist */
1725         if (!new_ctxt) {
1726                 spin_unlock(&dlm_domain_lock);
1727
1728                 new_ctxt = dlm_alloc_ctxt(domain, key);
1729                 if (new_ctxt)
1730                         goto retry;
1731
1732                 ret = -ENOMEM;
1733                 mlog_errno(ret);
1734                 goto leave;
1735         }
1736
1737         /* a little variable switch-a-roo here... */
1738         dlm = new_ctxt;
1739         new_ctxt = NULL;
1740
1741         /* add the new domain */
1742         list_add_tail(&dlm->list, &dlm_domains);
1743         spin_unlock(&dlm_domain_lock);
1744
1745         /*
1746          * Pass the locking protocol version into the join.  If the join
1747          * succeeds, it will have the negotiated protocol set.
1748          */
1749         dlm->dlm_locking_proto = dlm_protocol;
1750         dlm->fs_locking_proto = *fs_proto;
1751
1752         ret = dlm_join_domain(dlm);
1753         if (ret) {
1754                 mlog_errno(ret);
1755                 dlm_put(dlm);
1756                 goto leave;
1757         }
1758
1759         /* Tell the caller what locking protocol we negotiated */
1760         *fs_proto = dlm->fs_locking_proto;
1761
1762         ret = 0;
1763 leave:
1764         if (new_ctxt)
1765                 dlm_free_ctxt_mem(new_ctxt);
1766
1767         if (ret < 0)
1768                 dlm = ERR_PTR(ret);
1769
1770         return dlm;
1771 }
1772 EXPORT_SYMBOL_GPL(dlm_register_domain);
1773
1774 static LIST_HEAD(dlm_join_handlers);
1775
1776 static void dlm_unregister_net_handlers(void)
1777 {
1778         o2net_unregister_handler_list(&dlm_join_handlers);
1779 }
1780
1781 static int dlm_register_net_handlers(void)
1782 {
1783         int status = 0;
1784
1785         status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1786                                         sizeof(struct dlm_query_join_request),
1787                                         dlm_query_join_handler,
1788                                         NULL, NULL, &dlm_join_handlers);
1789         if (status)
1790                 goto bail;
1791
1792         status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1793                                         sizeof(struct dlm_assert_joined),
1794                                         dlm_assert_joined_handler,
1795                                         NULL, NULL, &dlm_join_handlers);
1796         if (status)
1797                 goto bail;
1798
1799         status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1800                                         sizeof(struct dlm_cancel_join),
1801                                         dlm_cancel_join_handler,
1802                                         NULL, NULL, &dlm_join_handlers);
1803
1804 bail:
1805         if (status < 0)
1806                 dlm_unregister_net_handlers();
1807
1808         return status;
1809 }
1810
1811 /* Domain eviction callback handling.
1812  *
1813  * The file system requires notification of node death *before* the
1814  * dlm completes it's recovery work, otherwise it may be able to
1815  * acquire locks on resources requiring recovery. Since the dlm can
1816  * evict a node from it's domain *before* heartbeat fires, a similar
1817  * mechanism is required. */
1818
1819 /* Eviction is not expected to happen often, so a per-domain lock is
1820  * not necessary. Eviction callbacks are allowed to sleep for short
1821  * periods of time. */
1822 static DECLARE_RWSEM(dlm_callback_sem);
1823
1824 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
1825                                         int node_num)
1826 {
1827         struct list_head *iter;
1828         struct dlm_eviction_cb *cb;
1829
1830         down_read(&dlm_callback_sem);
1831         list_for_each(iter, &dlm->dlm_eviction_callbacks) {
1832                 cb = list_entry(iter, struct dlm_eviction_cb, ec_item);
1833
1834                 cb->ec_func(node_num, cb->ec_data);
1835         }
1836         up_read(&dlm_callback_sem);
1837 }
1838
1839 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
1840                            dlm_eviction_func *f,
1841                            void *data)
1842 {
1843         INIT_LIST_HEAD(&cb->ec_item);
1844         cb->ec_func = f;
1845         cb->ec_data = data;
1846 }
1847 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
1848
1849 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
1850                               struct dlm_eviction_cb *cb)
1851 {
1852         down_write(&dlm_callback_sem);
1853         list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
1854         up_write(&dlm_callback_sem);
1855 }
1856 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
1857
1858 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
1859 {
1860         down_write(&dlm_callback_sem);
1861         list_del_init(&cb->ec_item);
1862         up_write(&dlm_callback_sem);
1863 }
1864 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
1865
1866 static int __init dlm_init(void)
1867 {
1868         int status;
1869
1870         dlm_print_version();
1871
1872         status = dlm_init_mle_cache();
1873         if (status) {
1874                 mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
1875                 goto error;
1876         }
1877
1878         status = dlm_init_master_caches();
1879         if (status) {
1880                 mlog(ML_ERROR, "Could not create o2dlm_lockres and "
1881                      "o2dlm_lockname slabcaches\n");
1882                 goto error;
1883         }
1884
1885         status = dlm_init_lock_cache();
1886         if (status) {
1887                 mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
1888                 goto error;
1889         }
1890
1891         status = dlm_register_net_handlers();
1892         if (status) {
1893                 mlog(ML_ERROR, "Unable to register network handlers\n");
1894                 goto error;
1895         }
1896
1897         status = dlm_create_debugfs_root();
1898         if (status)
1899                 goto error;
1900
1901         return 0;
1902 error:
1903         dlm_unregister_net_handlers();
1904         dlm_destroy_lock_cache();
1905         dlm_destroy_master_caches();
1906         dlm_destroy_mle_cache();
1907         return -1;
1908 }
1909
1910 static void __exit dlm_exit (void)
1911 {
1912         dlm_destroy_debugfs_root();
1913         dlm_unregister_net_handlers();
1914         dlm_destroy_lock_cache();
1915         dlm_destroy_master_caches();
1916         dlm_destroy_mle_cache();
1917 }
1918
1919 MODULE_AUTHOR("Oracle");
1920 MODULE_LICENSE("GPL");
1921
1922 module_init(dlm_init);
1923 module_exit(dlm_exit);