fs/ocfs2/cluster/heartbeat.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/sched.h>
   8 #include <linux/jiffies.h>
   9 #include <linux/module.h>
  10 #include <linux/fs.h>
  11 #include <linux/bio.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/delay.h>
  14 #include <linux/file.h>
  15 #include <linux/kthread.h>
  16 #include <linux/configfs.h>
  17 #include <linux/random.h>
  18 #include <linux/crc32.h>
  19 #include <linux/time.h>
  20 #include <linux/debugfs.h>
  21 #include <linux/slab.h>
  22 #include <linux/bitmap.h>
  23 #include <linux/ktime.h>
  24 #include "heartbeat.h"
  25 #include "tcp.h"
  26 #include "nodemanager.h"
  27 #include "quorum.h"
  28
  29 #include "masklog.h"
  30
  31
  32 /*
  33  * The first heartbeat pass had one global thread that would serialize all hb
  34  * callback calls.  This global serializing sem should only be removed once
  35  * we've made sure that all callees can deal with being called concurrently
  36  * from multiple hb region threads.
  37  */
  38 static DECLARE_RWSEM(o2hb_callback_sem);
  39
  40 /*
  41  * multiple hb threads are watching multiple regions.  A node is live
  42  * whenever any of the threads sees activity from the node in its region.
  43  */
  44 static DEFINE_SPINLOCK(o2hb_live_lock);
  45 static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
  46 static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
  47 static LIST_HEAD(o2hb_node_events);
  48 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
  49
  50 /*
  51  * In global heartbeat, we maintain a series of region bitmaps.
  52  *      - o2hb_region_bitmap allows us to limit the region number to max region.
  53  *      - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
  54  *      - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
  55  *              heartbeat on it.
  56  *      - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
  57  */
  58 static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  59 static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  60 static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  61 static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  62
  63 #define O2HB_DB_TYPE_LIVENODES          0
  64 #define O2HB_DB_TYPE_LIVEREGIONS        1
  65 #define O2HB_DB_TYPE_QUORUMREGIONS      2
  66 #define O2HB_DB_TYPE_FAILEDREGIONS      3
  67 #define O2HB_DB_TYPE_REGION_LIVENODES   4
  68 #define O2HB_DB_TYPE_REGION_NUMBER      5
  69 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME        6
  70 #define O2HB_DB_TYPE_REGION_PINNED      7
  71 struct o2hb_debug_buf {
  72         int db_type;
  73         int db_size;
  74         int db_len;
  75         void *db_data;
  76 };
  77
  78 static struct o2hb_debug_buf *o2hb_db_livenodes;
  79 static struct o2hb_debug_buf *o2hb_db_liveregions;
  80 static struct o2hb_debug_buf *o2hb_db_quorumregions;
  81 static struct o2hb_debug_buf *o2hb_db_failedregions;
  82
  83 #define O2HB_DEBUG_DIR                  "o2hb"
  84 #define O2HB_DEBUG_LIVENODES            "livenodes"
  85 #define O2HB_DEBUG_LIVEREGIONS          "live_regions"
  86 #define O2HB_DEBUG_QUORUMREGIONS        "quorum_regions"
  87 #define O2HB_DEBUG_FAILEDREGIONS        "failed_regions"
  88 #define O2HB_DEBUG_REGION_NUMBER        "num"
  89 #define O2HB_DEBUG_REGION_ELAPSED_TIME  "elapsed_time_in_ms"
  90 #define O2HB_DEBUG_REGION_PINNED        "pinned"
  91
  92 static struct dentry *o2hb_debug_dir;
  93
  94 static LIST_HEAD(o2hb_all_regions);
  95
  96 static struct o2hb_callback {
  97         struct list_head list;
  98 } o2hb_callbacks[O2HB_NUM_CB];
  99
 100 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
 101
 102 enum o2hb_heartbeat_modes {
 103         O2HB_HEARTBEAT_LOCAL            = 0,
 104         O2HB_HEARTBEAT_GLOBAL,
 105         O2HB_HEARTBEAT_NUM_MODES,
 106 };
 107
 108 static const char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
 109         "local",        /* O2HB_HEARTBEAT_LOCAL */
 110         "global",       /* O2HB_HEARTBEAT_GLOBAL */
 111 };
 112
 113 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
 114 static unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
 115
 116 /*
 117  * o2hb_dependent_users tracks the number of registered callbacks that depend
 118  * on heartbeat. o2net and o2dlm are two entities that register this callback.
 119  * However only o2dlm depends on the heartbeat. It does not want the heartbeat
 120  * to stop while a dlm domain is still active.
 121  */
 122 static unsigned int o2hb_dependent_users;
 123
 124 /*
 125  * In global heartbeat mode, all regions are pinned if there are one or more
 126  * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
 127  * regions are unpinned if the region count exceeds the cut off or the number
 128  * of dependent users falls to zero.
 129  */
 130 #define O2HB_PIN_CUT_OFF                3
 131
 132 /*
 133  * In local heartbeat mode, we assume the dlm domain name to be the same as
 134  * region uuid. This is true for domains created for the file system but not
 135  * necessarily true for userdlm domains. This is a known limitation.
 136  *
 137  * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
 138  * works for both file system and userdlm domains.
 139  */
 140 static int o2hb_region_pin(const char *region_uuid);
 141 static void o2hb_region_unpin(const char *region_uuid);
 142
 143 /* Only sets a new threshold if there are no active regions.
 144  *
 145  * No locking or otherwise interesting code is required for reading
 146  * o2hb_dead_threshold as it can't change once regions are active and
 147  * it's not interesting to anyone until then anyway. */
 148 static void o2hb_dead_threshold_set(unsigned int threshold)
 149 {
 150         if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
 151                 spin_lock(&o2hb_live_lock);
 152                 if (list_empty(&o2hb_all_regions))
 153                         o2hb_dead_threshold = threshold;
 154                 spin_unlock(&o2hb_live_lock);
 155         }
 156 }
 157
 158 static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
 159 {
 160         int ret = -1;
 161
 162         if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
 163                 spin_lock(&o2hb_live_lock);
 164                 if (list_empty(&o2hb_all_regions)) {
 165                         o2hb_heartbeat_mode = hb_mode;
 166                         ret = 0;
 167                 }
 168                 spin_unlock(&o2hb_live_lock);
 169         }
 170
 171         return ret;
 172 }
 173
 174 struct o2hb_node_event {
 175         struct list_head        hn_item;
 176         enum o2hb_callback_type hn_event_type;
 177         struct o2nm_node        *hn_node;
 178         int                     hn_node_num;
 179 };
 180
 181 struct o2hb_disk_slot {
 182         struct o2hb_disk_heartbeat_block *ds_raw_block;
 183         u8                      ds_node_num;
 184         u64                     ds_last_time;
 185         u64                     ds_last_generation;
 186         u16                     ds_equal_samples;
 187         u16                     ds_changed_samples;
 188         struct list_head        ds_live_item;
 189 };
 190
 191 /* each thread owns a region.. when we're asked to tear down the region
 192  * we ask the thread to stop, who cleans up the region */
 193 struct o2hb_region {
 194         struct config_item      hr_item;
 195
 196         struct list_head        hr_all_item;
 197         unsigned                hr_unclean_stop:1,
 198                                 hr_aborted_start:1,
 199                                 hr_item_pinned:1,
 200                                 hr_item_dropped:1,
 201                                 hr_node_deleted:1;
 202
 203         /* protected by the hr_callback_sem */
 204         struct task_struct      *hr_task;
 205
 206         unsigned int            hr_blocks;
 207         unsigned long long      hr_start_block;
 208
 209         unsigned int            hr_block_bits;
 210         unsigned int            hr_block_bytes;
 211
 212         unsigned int            hr_slots_per_page;
 213         unsigned int            hr_num_pages;
 214
 215         struct page             **hr_slot_data;
 216         struct block_device     *hr_bdev;
 217         struct o2hb_disk_slot   *hr_slots;
 218
 219         /* live node map of this region */
 220         unsigned long           hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 221         unsigned int            hr_region_num;
 222
 223         struct dentry           *hr_debug_dir;
 224         struct o2hb_debug_buf   *hr_db_livenodes;
 225         struct o2hb_debug_buf   *hr_db_regnum;
 226         struct o2hb_debug_buf   *hr_db_elapsed_time;
 227         struct o2hb_debug_buf   *hr_db_pinned;
 228
 229         /* let the person setting up hb wait for it to return until it
 230          * has reached a 'steady' state.  This will be fixed when we have
 231          * a more complete api that doesn't lead to this sort of fragility. */
 232         atomic_t                hr_steady_iterations;
 233
 234         /* terminate o2hb thread if it does not reach steady state
 235          * (hr_steady_iterations == 0) within hr_unsteady_iterations */
 236         atomic_t                hr_unsteady_iterations;
 237
 238         char                    hr_dev_name[BDEVNAME_SIZE];
 239
 240         unsigned int            hr_timeout_ms;
 241
 242         /* randomized as the region goes up and down so that a node
 243          * recognizes a node going up and down in one iteration */
 244         u64                     hr_generation;
 245
 246         struct delayed_work     hr_write_timeout_work;
 247         unsigned long           hr_last_timeout_start;
 248
 249         /* negotiate timer, used to negotiate extending hb timeout. */
 250         struct delayed_work     hr_nego_timeout_work;
 251         unsigned long           hr_nego_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 252
 253         /* Used during o2hb_check_slot to hold a copy of the block
 254          * being checked because we temporarily have to zero out the
 255          * crc field. */
 256         struct o2hb_disk_heartbeat_block *hr_tmp_block;
 257
 258         /* Message key for negotiate timeout message. */
 259         unsigned int            hr_key;
 260         struct list_head        hr_handler_list;
 261
 262         /* last hb status, 0 for success, other value for error. */
 263         int                     hr_last_hb_status;
 264 };
 265
 266 struct o2hb_bio_wait_ctxt {
 267         atomic_t          wc_num_reqs;
 268         struct completion wc_io_complete;
 269         int               wc_error;
 270 };
 271
 272 #define O2HB_NEGO_TIMEOUT_MS (O2HB_MAX_WRITE_TIMEOUT_MS/2)
 273
 274 enum {
 275         O2HB_NEGO_TIMEOUT_MSG = 1,
 276         O2HB_NEGO_APPROVE_MSG = 2,
 277 };
 278
 279 struct o2hb_nego_msg {
 280         u8 node_num;
 281 };
 282
 283 static void o2hb_write_timeout(struct work_struct *work)
 284 {
 285         int failed, quorum;
 286         struct o2hb_region *reg =
 287                 container_of(work, struct o2hb_region,
 288                              hr_write_timeout_work.work);
 289
 290         mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
 291              "milliseconds\n", reg->hr_dev_name,
 292              jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
 293
 294         if (o2hb_global_heartbeat_active()) {
 295                 spin_lock(&o2hb_live_lock);
 296                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 297                         set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 298                 failed = bitmap_weight(o2hb_failed_region_bitmap,
 299                                         O2NM_MAX_REGIONS);
 300                 quorum = bitmap_weight(o2hb_quorum_region_bitmap,
 301                                         O2NM_MAX_REGIONS);
 302                 spin_unlock(&o2hb_live_lock);
 303
 304                 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
 305                      quorum, failed);
 306
 307                 /*
 308                  * Fence if the number of failed regions >= half the number
 309                  * of  quorum regions
 310                  */
 311                 if ((failed << 1) < quorum)
 312                         return;
 313         }
 314
 315         o2quo_disk_timeout();
 316 }
 317
 318 static void o2hb_arm_timeout(struct o2hb_region *reg)
 319 {
 320         /* Arm writeout only after thread reaches steady state */
 321         if (atomic_read(&reg->hr_steady_iterations) != 0)
 322                 return;
 323
 324         mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
 325              O2HB_MAX_WRITE_TIMEOUT_MS);
 326
 327         if (o2hb_global_heartbeat_active()) {
 328                 spin_lock(&o2hb_live_lock);
 329                 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 330                 spin_unlock(&o2hb_live_lock);
 331         }
 332         cancel_delayed_work(&reg->hr_write_timeout_work);
 333         schedule_delayed_work(&reg->hr_write_timeout_work,
 334                               msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
 335
 336         cancel_delayed_work(&reg->hr_nego_timeout_work);
 337         /* negotiate timeout must be less than write timeout. */
 338         schedule_delayed_work(&reg->hr_nego_timeout_work,
 339                               msecs_to_jiffies(O2HB_NEGO_TIMEOUT_MS));
 340         memset(reg->hr_nego_node_bitmap, 0, sizeof(reg->hr_nego_node_bitmap));
 341 }
 342
 343 static void o2hb_disarm_timeout(struct o2hb_region *reg)
 344 {
 345         cancel_delayed_work_sync(&reg->hr_write_timeout_work);
 346         cancel_delayed_work_sync(&reg->hr_nego_timeout_work);
 347 }
 348
 349 static int o2hb_send_nego_msg(int key, int type, u8 target)
 350 {
 351         struct o2hb_nego_msg msg;
 352         int status, ret;
 353
 354         msg.node_num = o2nm_this_node();
 355 again:
 356         ret = o2net_send_message(type, key, &msg, sizeof(msg),
 357                         target, &status);
 358
 359         if (ret == -EAGAIN || ret == -ENOMEM) {
 360                 msleep(100);
 361                 goto again;
 362         }
 363
 364         return ret;
 365 }
 366
 367 static void o2hb_nego_timeout(struct work_struct *work)
 368 {
 369         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 370         int master_node, i, ret;
 371         struct o2hb_region *reg;
 372
 373         reg = container_of(work, struct o2hb_region, hr_nego_timeout_work.work);
 374         /* don't negotiate timeout if last hb failed since it is very
 375          * possible io failed. Should let write timeout fence self.
 376          */
 377         if (reg->hr_last_hb_status)
 378                 return;
 379
 380         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
 381         /* lowest node as master node to make negotiate decision. */
 382         master_node = find_first_bit(live_node_bitmap, O2NM_MAX_NODES);
 383
 384         if (master_node == o2nm_this_node()) {
 385                 if (!test_bit(master_node, reg->hr_nego_node_bitmap)) {
 386                         printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%s).\n",
 387                                 o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000,
 388                                 config_item_name(&reg->hr_item), reg->hr_dev_name);
 389                         set_bit(master_node, reg->hr_nego_node_bitmap);
 390                 }
 391                 if (memcmp(reg->hr_nego_node_bitmap, live_node_bitmap,
 392                                 sizeof(reg->hr_nego_node_bitmap))) {
 393                         /* check negotiate bitmap every second to do timeout
 394                          * approve decision.
 395                          */
 396                         schedule_delayed_work(&reg->hr_nego_timeout_work,
 397                                 msecs_to_jiffies(1000));
 398
 399                         return;
 400                 }
 401
 402                 printk(KERN_NOTICE "o2hb: all nodes hb write hung, maybe region %s (%s) is down.\n",
 403                         config_item_name(&reg->hr_item), reg->hr_dev_name);
 404                 /* approve negotiate timeout request. */
 405                 o2hb_arm_timeout(reg);
 406
 407                 i = -1;
 408                 while ((i = find_next_bit(live_node_bitmap,
 409                                 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
 410                         if (i == master_node)
 411                                 continue;
 412
 413                         mlog(ML_HEARTBEAT, "send NEGO_APPROVE msg to node %d\n", i);
 414                         ret = o2hb_send_nego_msg(reg->hr_key,
 415                                         O2HB_NEGO_APPROVE_MSG, i);
 416                         if (ret)
 417                                 mlog(ML_ERROR, "send NEGO_APPROVE msg to node %d fail %d\n",
 418                                         i, ret);
 419                 }
 420         } else {
 421                 /* negotiate timeout with master node. */
 422                 printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%s), negotiate timeout with node %d.\n",
 423                         o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000, config_item_name(&reg->hr_item),
 424                         reg->hr_dev_name, master_node);
 425                 ret = o2hb_send_nego_msg(reg->hr_key, O2HB_NEGO_TIMEOUT_MSG,
 426                                 master_node);
 427                 if (ret)
 428                         mlog(ML_ERROR, "send NEGO_TIMEOUT msg to node %d fail %d\n",
 429                                 master_node, ret);
 430         }
 431 }
 432
 433 static int o2hb_nego_timeout_handler(struct o2net_msg *msg, u32 len, void *data,
 434                                 void **ret_data)
 435 {
 436         struct o2hb_region *reg = data;
 437         struct o2hb_nego_msg *nego_msg;
 438
 439         nego_msg = (struct o2hb_nego_msg *)msg->buf;
 440         printk(KERN_NOTICE "o2hb: receive negotiate timeout message from node %d on region %s (%s).\n",
 441                 nego_msg->node_num, config_item_name(&reg->hr_item), reg->hr_dev_name);
 442         if (nego_msg->node_num < O2NM_MAX_NODES)
 443                 set_bit(nego_msg->node_num, reg->hr_nego_node_bitmap);
 444         else
 445                 mlog(ML_ERROR, "got nego timeout message from bad node.\n");
 446
 447         return 0;
 448 }
 449
 450 static int o2hb_nego_approve_handler(struct o2net_msg *msg, u32 len, void *data,
 451                                 void **ret_data)
 452 {
 453         struct o2hb_region *reg = data;
 454
 455         printk(KERN_NOTICE "o2hb: negotiate timeout approved by master node on region %s (%s).\n",
 456                 config_item_name(&reg->hr_item), reg->hr_dev_name);
 457         o2hb_arm_timeout(reg);
 458         return 0;
 459 }
 460
 461 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
 462 {
 463         atomic_set(&wc->wc_num_reqs, 1);
 464         init_completion(&wc->wc_io_complete);
 465         wc->wc_error = 0;
 466 }
 467
 468 /* Used in error paths too */
 469 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
 470                                      unsigned int num)
 471 {
 472         /* sadly atomic_sub_and_test() isn't available on all platforms.  The
 473          * good news is that the fast path only completes one at a time */
 474         while(num--) {
 475                 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
 476                         BUG_ON(num > 0);
 477                         complete(&wc->wc_io_complete);
 478                 }
 479         }
 480 }
 481
 482 static void o2hb_wait_on_io(struct o2hb_bio_wait_ctxt *wc)
 483 {
 484         o2hb_bio_wait_dec(wc, 1);
 485         wait_for_completion(&wc->wc_io_complete);
 486 }
 487
 488 static void o2hb_bio_end_io(struct bio *bio)
 489 {
 490         struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
 491
 492         if (bio->bi_status) {
 493                 mlog(ML_ERROR, "IO Error %d\n", bio->bi_status);
 494                 wc->wc_error = blk_status_to_errno(bio->bi_status);
 495         }
 496
 497         o2hb_bio_wait_dec(wc, 1);
 498         bio_put(bio);
 499 }
 500
 501 /* Setup a Bio to cover I/O against num_slots slots starting at
 502  * start_slot. */
 503 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
 504                                       struct o2hb_bio_wait_ctxt *wc,
 505                                       unsigned int *current_slot,
 506                                       unsigned int max_slots, int op,
 507                                       int op_flags)
 508 {
 509         int len, current_page;
 510         unsigned int vec_len, vec_start;
 511         unsigned int bits = reg->hr_block_bits;
 512         unsigned int spp = reg->hr_slots_per_page;
 513         unsigned int cs = *current_slot;
 514         struct bio *bio;
 515         struct page *page;
 516
 517         /* Testing has shown this allocation to take long enough under
 518          * GFP_KERNEL that the local node can get fenced. It would be
 519          * nicest if we could pre-allocate these bios and avoid this
 520          * all together. */
 521         bio = bio_alloc(reg->hr_bdev, 16, op | op_flags, GFP_ATOMIC);
 522         if (!bio) {
 523                 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
 524                 bio = ERR_PTR(-ENOMEM);
 525                 goto bail;
 526         }
 527
 528         /* Must put everything in 512 byte sectors for the bio... */
 529         bio->bi_iter.bi_sector = (reg->hr_start_block + cs) << (bits - 9);
 530         bio->bi_private = wc;
 531         bio->bi_end_io = o2hb_bio_end_io;
 532
 533         vec_start = (cs << bits) % PAGE_SIZE;
 534         while(cs < max_slots) {
 535                 current_page = cs / spp;
 536                 page = reg->hr_slot_data[current_page];
 537
 538                 vec_len = min(PAGE_SIZE - vec_start,
 539                               (max_slots-cs) * (PAGE_SIZE/spp) );
 540
 541                 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
 542                      current_page, vec_len, vec_start);
 543
 544                 len = bio_add_page(bio, page, vec_len, vec_start);
 545                 if (len != vec_len) break;
 546
 547                 cs += vec_len / (PAGE_SIZE/spp);
 548                 vec_start = 0;
 549         }
 550
 551 bail:
 552         *current_slot = cs;
 553         return bio;
 554 }
 555
 556 static int o2hb_read_slots(struct o2hb_region *reg,
 557                            unsigned int begin_slot,
 558                            unsigned int max_slots)
 559 {
 560         unsigned int current_slot = begin_slot;
 561         int status;
 562         struct o2hb_bio_wait_ctxt wc;
 563         struct bio *bio;
 564
 565         o2hb_bio_wait_init(&wc);
 566
 567         while(current_slot < max_slots) {
 568                 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots,
 569                                          REQ_OP_READ, 0);
 570                 if (IS_ERR(bio)) {
 571                         status = PTR_ERR(bio);
 572                         mlog_errno(status);
 573                         goto bail_and_wait;
 574                 }
 575
 576                 atomic_inc(&wc.wc_num_reqs);
 577                 submit_bio(bio);
 578         }
 579
 580         status = 0;
 581
 582 bail_and_wait:
 583         o2hb_wait_on_io(&wc);
 584         if (wc.wc_error && !status)
 585                 status = wc.wc_error;
 586
 587         return status;
 588 }
 589
 590 static int o2hb_issue_node_write(struct o2hb_region *reg,
 591                                  struct o2hb_bio_wait_ctxt *write_wc)
 592 {
 593         int status;
 594         unsigned int slot;
 595         struct bio *bio;
 596
 597         o2hb_bio_wait_init(write_wc);
 598
 599         slot = o2nm_this_node();
 600
 601         bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1, REQ_OP_WRITE,
 602                                  REQ_SYNC);
 603         if (IS_ERR(bio)) {
 604                 status = PTR_ERR(bio);
 605                 mlog_errno(status);
 606                 goto bail;
 607         }
 608
 609         atomic_inc(&write_wc->wc_num_reqs);
 610         submit_bio(bio);
 611
 612         status = 0;
 613 bail:
 614         return status;
 615 }
 616
 617 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
 618                                      struct o2hb_disk_heartbeat_block *hb_block)
 619 {
 620         __le32 old_cksum;
 621         u32 ret;
 622
 623         /* We want to compute the block crc with a 0 value in the
 624          * hb_cksum field. Save it off here and replace after the
 625          * crc. */
 626         old_cksum = hb_block->hb_cksum;
 627         hb_block->hb_cksum = 0;
 628
 629         ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
 630
 631         hb_block->hb_cksum = old_cksum;
 632
 633         return ret;
 634 }
 635
 636 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
 637 {
 638         mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
 639              "cksum = 0x%x, generation 0x%llx\n",
 640              (long long)le64_to_cpu(hb_block->hb_seq),
 641              hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
 642              (long long)le64_to_cpu(hb_block->hb_generation));
 643 }
 644
 645 static int o2hb_verify_crc(struct o2hb_region *reg,
 646                            struct o2hb_disk_heartbeat_block *hb_block)
 647 {
 648         u32 read, computed;
 649
 650         read = le32_to_cpu(hb_block->hb_cksum);
 651         computed = o2hb_compute_block_crc_le(reg, hb_block);
 652
 653         return read == computed;
 654 }
 655
 656 /*
 657  * Compare the slot data with what we wrote in the last iteration.
 658  * If the match fails, print an appropriate error message. This is to
 659  * detect errors like... another node hearting on the same slot,
 660  * flaky device that is losing writes, etc.
 661  * Returns 1 if check succeeds, 0 otherwise.
 662  */
 663 static int o2hb_check_own_slot(struct o2hb_region *reg)
 664 {
 665         struct o2hb_disk_slot *slot;
 666         struct o2hb_disk_heartbeat_block *hb_block;
 667         char *errstr;
 668
 669         slot = &reg->hr_slots[o2nm_this_node()];
 670         /* Don't check on our 1st timestamp */
 671         if (!slot->ds_last_time)
 672                 return 0;
 673
 674         hb_block = slot->ds_raw_block;
 675         if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
 676             le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
 677             hb_block->hb_node == slot->ds_node_num)
 678                 return 1;
 679
 680 #define ERRSTR1         "Another node is heartbeating on device"
 681 #define ERRSTR2         "Heartbeat generation mismatch on device"
 682 #define ERRSTR3         "Heartbeat sequence mismatch on device"
 683
 684         if (hb_block->hb_node != slot->ds_node_num)
 685                 errstr = ERRSTR1;
 686         else if (le64_to_cpu(hb_block->hb_generation) !=
 687                  slot->ds_last_generation)
 688                 errstr = ERRSTR2;
 689         else
 690                 errstr = ERRSTR3;
 691
 692         mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
 693              "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
 694              slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
 695              (unsigned long long)slot->ds_last_time, hb_block->hb_node,
 696              (unsigned long long)le64_to_cpu(hb_block->hb_generation),
 697              (unsigned long long)le64_to_cpu(hb_block->hb_seq));
 698
 699         return 0;
 700 }
 701
 702 static inline void o2hb_prepare_block(struct o2hb_region *reg,
 703                                       u64 generation)
 704 {
 705         int node_num;
 706         u64 cputime;
 707         struct o2hb_disk_slot *slot;
 708         struct o2hb_disk_heartbeat_block *hb_block;
 709
 710         node_num = o2nm_this_node();
 711         slot = &reg->hr_slots[node_num];
 712
 713         hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
 714         memset(hb_block, 0, reg->hr_block_bytes);
 715         /* TODO: time stuff */
 716         cputime = ktime_get_real_seconds();
 717         if (!cputime)
 718                 cputime = 1;
 719
 720         hb_block->hb_seq = cpu_to_le64(cputime);
 721         hb_block->hb_node = node_num;
 722         hb_block->hb_generation = cpu_to_le64(generation);
 723         hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
 724
 725         /* This step must always happen last! */
 726         hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
 727                                                                    hb_block));
 728
 729         mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
 730              (long long)generation,
 731              le32_to_cpu(hb_block->hb_cksum));
 732 }
 733
 734 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
 735                                 struct o2nm_node *node,
 736                                 int idx)
 737 {
 738         struct o2hb_callback_func *f;
 739
 740         list_for_each_entry(f, &hbcall->list, hc_item) {
 741                 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
 742                 (f->hc_func)(node, idx, f->hc_data);
 743         }
 744 }
 745
 746 /* Will run the list in order until we process the passed event */
 747 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
 748 {
 749         struct o2hb_callback *hbcall;
 750         struct o2hb_node_event *event;
 751
 752         /* Holding callback sem assures we don't alter the callback
 753          * lists when doing this, and serializes ourselves with other
 754          * processes wanting callbacks. */
 755         down_write(&o2hb_callback_sem);
 756
 757         spin_lock(&o2hb_live_lock);
 758         while (!list_empty(&o2hb_node_events)
 759                && !list_empty(&queued_event->hn_item)) {
 760                 event = list_entry(o2hb_node_events.next,
 761                                    struct o2hb_node_event,
 762                                    hn_item);
 763                 list_del_init(&event->hn_item);
 764                 spin_unlock(&o2hb_live_lock);
 765
 766                 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
 767                      event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
 768                      event->hn_node_num);
 769
 770                 hbcall = hbcall_from_type(event->hn_event_type);
 771
 772                 /* We should *never* have gotten on to the list with a
 773                  * bad type... This isn't something that we should try
 774                  * to recover from. */
 775                 BUG_ON(IS_ERR(hbcall));
 776
 777                 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
 778
 779                 spin_lock(&o2hb_live_lock);
 780         }
 781         spin_unlock(&o2hb_live_lock);
 782
 783         up_write(&o2hb_callback_sem);
 784 }
 785
 786 static void o2hb_queue_node_event(struct o2hb_node_event *event,
 787                                   enum o2hb_callback_type type,
 788                                   struct o2nm_node *node,
 789                                   int node_num)
 790 {
 791         assert_spin_locked(&o2hb_live_lock);
 792
 793         BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
 794
 795         event->hn_event_type = type;
 796         event->hn_node = node;
 797         event->hn_node_num = node_num;
 798
 799         mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
 800              type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
 801
 802         list_add_tail(&event->hn_item, &o2hb_node_events);
 803 }
 804
 805 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
 806 {
 807         struct o2hb_node_event event =
 808                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 809         struct o2nm_node *node;
 810         int queued = 0;
 811
 812         node = o2nm_get_node_by_num(slot->ds_node_num);
 813         if (!node)
 814                 return;
 815
 816         spin_lock(&o2hb_live_lock);
 817         if (!list_empty(&slot->ds_live_item)) {
 818                 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
 819                      slot->ds_node_num);
 820
 821                 list_del_init(&slot->ds_live_item);
 822
 823                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 824                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 825
 826                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
 827                                               slot->ds_node_num);
 828                         queued = 1;
 829                 }
 830         }
 831         spin_unlock(&o2hb_live_lock);
 832
 833         if (queued)
 834                 o2hb_run_event_list(&event);
 835
 836         o2nm_node_put(node);
 837 }
 838
 839 static void o2hb_set_quorum_device(struct o2hb_region *reg)
 840 {
 841         if (!o2hb_global_heartbeat_active())
 842                 return;
 843
 844         /* Prevent race with o2hb_heartbeat_group_drop_item() */
 845         if (kthread_should_stop())
 846                 return;
 847
 848         /* Tag region as quorum only after thread reaches steady state */
 849         if (atomic_read(&reg->hr_steady_iterations) != 0)
 850                 return;
 851
 852         spin_lock(&o2hb_live_lock);
 853
 854         if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 855                 goto unlock;
 856
 857         /*
 858          * A region can be added to the quorum only when it sees all
 859          * live nodes heartbeat on it. In other words, the region has been
 860          * added to all nodes.
 861          */
 862         if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
 863                    sizeof(o2hb_live_node_bitmap)))
 864                 goto unlock;
 865
 866         printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
 867                config_item_name(&reg->hr_item), reg->hr_dev_name);
 868
 869         set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
 870
 871         /*
 872          * If global heartbeat active, unpin all regions if the
 873          * region count > CUT_OFF
 874          */
 875         if (bitmap_weight(o2hb_quorum_region_bitmap,
 876                            O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
 877                 o2hb_region_unpin(NULL);
 878 unlock:
 879         spin_unlock(&o2hb_live_lock);
 880 }
 881
 882 static int o2hb_check_slot(struct o2hb_region *reg,
 883                            struct o2hb_disk_slot *slot)
 884 {
 885         int changed = 0, gen_changed = 0;
 886         struct o2hb_node_event event =
 887                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 888         struct o2nm_node *node;
 889         struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
 890         u64 cputime;
 891         unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
 892         unsigned int slot_dead_ms;
 893         int tmp;
 894         int queued = 0;
 895
 896         memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
 897
 898         /*
 899          * If a node is no longer configured but is still in the livemap, we
 900          * may need to clear that bit from the livemap.
 901          */
 902         node = o2nm_get_node_by_num(slot->ds_node_num);
 903         if (!node) {
 904                 spin_lock(&o2hb_live_lock);
 905                 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 906                 spin_unlock(&o2hb_live_lock);
 907                 if (!tmp)
 908                         return 0;
 909         }
 910
 911         if (!o2hb_verify_crc(reg, hb_block)) {
 912                 /* all paths from here will drop o2hb_live_lock for
 913                  * us. */
 914                 spin_lock(&o2hb_live_lock);
 915
 916                 /* Don't print an error on the console in this case -
 917                  * a freshly formatted heartbeat area will not have a
 918                  * crc set on it. */
 919                 if (list_empty(&slot->ds_live_item))
 920                         goto out;
 921
 922                 /* The node is live but pushed out a bad crc. We
 923                  * consider it a transient miss but don't populate any
 924                  * other values as they may be junk. */
 925                 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
 926                      slot->ds_node_num, reg->hr_dev_name);
 927                 o2hb_dump_slot(hb_block);
 928
 929                 slot->ds_equal_samples++;
 930                 goto fire_callbacks;
 931         }
 932
 933         /* we don't care if these wrap.. the state transitions below
 934          * clear at the right places */
 935         cputime = le64_to_cpu(hb_block->hb_seq);
 936         if (slot->ds_last_time != cputime)
 937                 slot->ds_changed_samples++;
 938         else
 939                 slot->ds_equal_samples++;
 940         slot->ds_last_time = cputime;
 941
 942         /* The node changed heartbeat generations. We assume this to
 943          * mean it dropped off but came back before we timed out. We
 944          * want to consider it down for the time being but don't want
 945          * to lose any changed_samples state we might build up to
 946          * considering it live again. */
 947         if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
 948                 gen_changed = 1;
 949                 slot->ds_equal_samples = 0;
 950                 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
 951                      "to 0x%llx)\n", slot->ds_node_num,
 952                      (long long)slot->ds_last_generation,
 953                      (long long)le64_to_cpu(hb_block->hb_generation));
 954         }
 955
 956         slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
 957
 958         mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
 959              "seq %llu last %llu changed %u equal %u\n",
 960              slot->ds_node_num, (long long)slot->ds_last_generation,
 961              le32_to_cpu(hb_block->hb_cksum),
 962              (unsigned long long)le64_to_cpu(hb_block->hb_seq),
 963              (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
 964              slot->ds_equal_samples);
 965
 966         spin_lock(&o2hb_live_lock);
 967
 968 fire_callbacks:
 969         /* dead nodes only come to life after some number of
 970          * changes at any time during their dead time */
 971         if (list_empty(&slot->ds_live_item) &&
 972             slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
 973                 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
 974                      slot->ds_node_num, (long long)slot->ds_last_generation);
 975
 976                 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 977
 978                 /* first on the list generates a callback */
 979                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 980                         mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
 981                              "bitmap\n", slot->ds_node_num);
 982                         set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 983
 984                         o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
 985                                               slot->ds_node_num);
 986
 987                         changed = 1;
 988                         queued = 1;
 989                 }
 990
 991                 list_add_tail(&slot->ds_live_item,
 992                               &o2hb_live_slots[slot->ds_node_num]);
 993
 994                 slot->ds_equal_samples = 0;
 995
 996                 /* We want to be sure that all nodes agree on the
 997                  * number of milliseconds before a node will be
 998                  * considered dead. The self-fencing timeout is
 999                  * computed from this value, and a discrepancy might
1000                  * result in heartbeat calling a node dead when it
1001                  * hasn't self-fenced yet. */
1002                 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
1003                 if (slot_dead_ms && slot_dead_ms != dead_ms) {
1004                         /* TODO: Perhaps we can fail the region here. */
1005                         mlog(ML_ERROR, "Node %d on device %s has a dead count "
1006                              "of %u ms, but our count is %u ms.\n"
1007                              "Please double check your configuration values "
1008                              "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
1009                              slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
1010                              dead_ms);
1011                 }
1012                 goto out;
1013         }
1014
1015         /* if the list is dead, we're done.. */
1016         if (list_empty(&slot->ds_live_item))
1017                 goto out;
1018
1019         /* live nodes only go dead after enough consequtive missed
1020          * samples..  reset the missed counter whenever we see
1021          * activity */
1022         if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
1023                 mlog(ML_HEARTBEAT, "Node %d left my region\n",
1024                      slot->ds_node_num);
1025
1026                 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1027
1028                 /* last off the live_slot generates a callback */
1029                 list_del_init(&slot->ds_live_item);
1030                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
1031                         mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
1032                              "nodes bitmap\n", slot->ds_node_num);
1033                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
1034
1035                         /* node can be null */
1036                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
1037                                               node, slot->ds_node_num);
1038
1039                         changed = 1;
1040                         queued = 1;
1041                 }
1042
1043                 /* We don't clear this because the node is still
1044                  * actually writing new blocks. */
1045                 if (!gen_changed)
1046                         slot->ds_changed_samples = 0;
1047                 goto out;
1048         }
1049         if (slot->ds_changed_samples) {
1050                 slot->ds_changed_samples = 0;
1051                 slot->ds_equal_samples = 0;
1052         }
1053 out:
1054         spin_unlock(&o2hb_live_lock);
1055
1056         if (queued)
1057                 o2hb_run_event_list(&event);
1058
1059         if (node)
1060                 o2nm_node_put(node);
1061         return changed;
1062 }
1063
1064 static int o2hb_highest_node(unsigned long *nodes, int numbits)
1065 {
1066         return find_last_bit(nodes, numbits);
1067 }
1068
1069 static int o2hb_lowest_node(unsigned long *nodes, int numbits)
1070 {
1071         return find_first_bit(nodes, numbits);
1072 }
1073
1074 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
1075 {
1076         int i, ret, highest_node, lowest_node;
1077         int membership_change = 0, own_slot_ok = 0;
1078         unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
1079         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
1080         struct o2hb_bio_wait_ctxt write_wc;
1081
1082         ret = o2nm_configured_node_map(configured_nodes,
1083                                        sizeof(configured_nodes));
1084         if (ret) {
1085                 mlog_errno(ret);
1086                 goto bail;
1087         }
1088
1089         /*
1090          * If a node is not configured but is in the livemap, we still need
1091          * to read the slot so as to be able to remove it from the livemap.
1092          */
1093         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
1094         i = -1;
1095         while ((i = find_next_bit(live_node_bitmap,
1096                                   O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1097                 set_bit(i, configured_nodes);
1098         }
1099
1100         highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1101         lowest_node = o2hb_lowest_node(configured_nodes, O2NM_MAX_NODES);
1102         if (highest_node >= O2NM_MAX_NODES || lowest_node >= O2NM_MAX_NODES) {
1103                 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1104                 ret = -EINVAL;
1105                 goto bail;
1106         }
1107
1108         /* No sense in reading the slots of nodes that don't exist
1109          * yet. Of course, if the node definitions have holes in them
1110          * then we're reading an empty slot anyway... Consider this
1111          * best-effort. */
1112         ret = o2hb_read_slots(reg, lowest_node, highest_node + 1);
1113         if (ret < 0) {
1114                 mlog_errno(ret);
1115                 goto bail;
1116         }
1117
1118         /* With an up to date view of the slots, we can check that no
1119          * other node has been improperly configured to heartbeat in
1120          * our slot. */
1121         own_slot_ok = o2hb_check_own_slot(reg);
1122
1123         /* fill in the proper info for our next heartbeat */
1124         o2hb_prepare_block(reg, reg->hr_generation);
1125
1126         ret = o2hb_issue_node_write(reg, &write_wc);
1127         if (ret < 0) {
1128                 mlog_errno(ret);
1129                 goto bail;
1130         }
1131
1132         i = -1;
1133         while((i = find_next_bit(configured_nodes,
1134                                  O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1135                 membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1136         }
1137
1138         /*
1139          * We have to be sure we've advertised ourselves on disk
1140          * before we can go to steady state.  This ensures that
1141          * people we find in our steady state have seen us.
1142          */
1143         o2hb_wait_on_io(&write_wc);
1144         if (write_wc.wc_error) {
1145                 /* Do not re-arm the write timeout on I/O error - we
1146                  * can't be sure that the new block ever made it to
1147                  * disk */
1148                 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1149                      write_wc.wc_error, reg->hr_dev_name);
1150                 ret = write_wc.wc_error;
1151                 goto bail;
1152         }
1153
1154         /* Skip disarming the timeout if own slot has stale/bad data */
1155         if (own_slot_ok) {
1156                 o2hb_set_quorum_device(reg);
1157                 o2hb_arm_timeout(reg);
1158                 reg->hr_last_timeout_start = jiffies;
1159         }
1160
1161 bail:
1162         /* let the person who launched us know when things are steady */
1163         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1164                 if (!ret && own_slot_ok && !membership_change) {
1165                         if (atomic_dec_and_test(&reg->hr_steady_iterations))
1166                                 wake_up(&o2hb_steady_queue);
1167                 }
1168         }
1169
1170         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1171                 if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1172                         printk(KERN_NOTICE "o2hb: Unable to stabilize "
1173                                "heartbeat on region %s (%s)\n",
1174                                config_item_name(&reg->hr_item),
1175                                reg->hr_dev_name);
1176                         atomic_set(&reg->hr_steady_iterations, 0);
1177                         reg->hr_aborted_start = 1;
1178                         wake_up(&o2hb_steady_queue);
1179                         ret = -EIO;
1180                 }
1181         }
1182
1183         return ret;
1184 }
1185
1186 /*
1187  * we ride the region ref that the region dir holds.  before the region
1188  * dir is removed and drops it ref it will wait to tear down this
1189  * thread.
1190  */
1191 static int o2hb_thread(void *data)
1192 {
1193         int i, ret;
1194         struct o2hb_region *reg = data;
1195         struct o2hb_bio_wait_ctxt write_wc;
1196         ktime_t before_hb, after_hb;
1197         unsigned int elapsed_msec;
1198
1199         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1200
1201         set_user_nice(current, MIN_NICE);
1202
1203         /* Pin node */
1204         ret = o2nm_depend_this_node();
1205         if (ret) {
1206                 mlog(ML_ERROR, "Node has been deleted, ret = %d\n", ret);
1207                 reg->hr_node_deleted = 1;
1208                 wake_up(&o2hb_steady_queue);
1209                 return 0;
1210         }
1211
1212         while (!kthread_should_stop() &&
1213                !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1214                 /* We track the time spent inside
1215                  * o2hb_do_disk_heartbeat so that we avoid more than
1216                  * hr_timeout_ms between disk writes. On busy systems
1217                  * this should result in a heartbeat which is less
1218                  * likely to time itself out. */
1219                 before_hb = ktime_get_real();
1220
1221                 ret = o2hb_do_disk_heartbeat(reg);
1222                 reg->hr_last_hb_status = ret;
1223
1224                 after_hb = ktime_get_real();
1225
1226                 elapsed_msec = (unsigned int)
1227                                 ktime_ms_delta(after_hb, before_hb);
1228
1229                 mlog(ML_HEARTBEAT,
1230                      "start = %lld, end = %lld, msec = %u, ret = %d\n",
1231                      before_hb, after_hb, elapsed_msec, ret);
1232
1233                 if (!kthread_should_stop() &&
1234                     elapsed_msec < reg->hr_timeout_ms) {
1235                         /* the kthread api has blocked signals for us so no
1236                          * need to record the return value. */
1237                         msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1238                 }
1239         }
1240
1241         o2hb_disarm_timeout(reg);
1242
1243         /* unclean stop is only used in very bad situation */
1244         for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1245                 o2hb_shutdown_slot(&reg->hr_slots[i]);
1246
1247         /* Explicit down notification - avoid forcing the other nodes
1248          * to timeout on this region when we could just as easily
1249          * write a clear generation - thus indicating to them that
1250          * this node has left this region.
1251          */
1252         if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1253                 o2hb_prepare_block(reg, 0);
1254                 ret = o2hb_issue_node_write(reg, &write_wc);
1255                 if (ret == 0)
1256                         o2hb_wait_on_io(&write_wc);
1257                 else
1258                         mlog_errno(ret);
1259         }
1260
1261         /* Unpin node */
1262         o2nm_undepend_this_node();
1263
1264         mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1265
1266         return 0;
1267 }
1268
1269 #ifdef CONFIG_DEBUG_FS
1270 static int o2hb_debug_open(struct inode *inode, struct file *file)
1271 {
1272         struct o2hb_debug_buf *db = inode->i_private;
1273         struct o2hb_region *reg;
1274         unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1275         unsigned long lts;
1276         char *buf = NULL;
1277         int i = -1;
1278         int out = 0;
1279
1280         /* max_nodes should be the largest bitmap we pass here */
1281         BUG_ON(sizeof(map) < db->db_size);
1282
1283         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1284         if (!buf)
1285                 goto bail;
1286
1287         switch (db->db_type) {
1288         case O2HB_DB_TYPE_LIVENODES:
1289         case O2HB_DB_TYPE_LIVEREGIONS:
1290         case O2HB_DB_TYPE_QUORUMREGIONS:
1291         case O2HB_DB_TYPE_FAILEDREGIONS:
1292                 spin_lock(&o2hb_live_lock);
1293                 memcpy(map, db->db_data, db->db_size);
1294                 spin_unlock(&o2hb_live_lock);
1295                 break;
1296
1297         case O2HB_DB_TYPE_REGION_LIVENODES:
1298                 spin_lock(&o2hb_live_lock);
1299                 reg = (struct o2hb_region *)db->db_data;
1300                 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1301                 spin_unlock(&o2hb_live_lock);
1302                 break;
1303
1304         case O2HB_DB_TYPE_REGION_NUMBER:
1305                 reg = (struct o2hb_region *)db->db_data;
1306                 out += scnprintf(buf + out, PAGE_SIZE - out, "%d\n",
1307                                 reg->hr_region_num);
1308                 goto done;
1309
1310         case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1311                 reg = (struct o2hb_region *)db->db_data;
1312                 lts = reg->hr_last_timeout_start;
1313                 /* If 0, it has never been set before */
1314                 if (lts)
1315                         lts = jiffies_to_msecs(jiffies - lts);
1316                 out += scnprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1317                 goto done;
1318
1319         case O2HB_DB_TYPE_REGION_PINNED:
1320                 reg = (struct o2hb_region *)db->db_data;
1321                 out += scnprintf(buf + out, PAGE_SIZE - out, "%u\n",
1322                                 !!reg->hr_item_pinned);
1323                 goto done;
1324
1325         default:
1326                 goto done;
1327         }
1328
1329         while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1330                 out += scnprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1331         out += scnprintf(buf + out, PAGE_SIZE - out, "\n");
1332
1333 done:
1334         i_size_write(inode, out);
1335
1336         file->private_data = buf;
1337
1338         return 0;
1339 bail:
1340         return -ENOMEM;
1341 }
1342
1343 static int o2hb_debug_release(struct inode *inode, struct file *file)
1344 {
1345         kfree(file->private_data);
1346         return 0;
1347 }
1348
1349 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1350                                  size_t nbytes, loff_t *ppos)
1351 {
1352         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1353                                        i_size_read(file->f_mapping->host));
1354 }
1355 #else
1356 static int o2hb_debug_open(struct inode *inode, struct file *file)
1357 {
1358         return 0;
1359 }
1360 static int o2hb_debug_release(struct inode *inode, struct file *file)
1361 {
1362         return 0;
1363 }
1364 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1365                                size_t nbytes, loff_t *ppos)
1366 {
1367         return 0;
1368 }
1369 #endif  /* CONFIG_DEBUG_FS */
1370
1371 static const struct file_operations o2hb_debug_fops = {
1372         .open =         o2hb_debug_open,
1373         .release =      o2hb_debug_release,
1374         .read =         o2hb_debug_read,
1375         .llseek =       generic_file_llseek,
1376 };
1377
1378 void o2hb_exit(void)
1379 {
1380         debugfs_remove_recursive(o2hb_debug_dir);
1381         kfree(o2hb_db_livenodes);
1382         kfree(o2hb_db_liveregions);
1383         kfree(o2hb_db_quorumregions);
1384         kfree(o2hb_db_failedregions);
1385 }
1386
1387 static void o2hb_debug_create(const char *name, struct dentry *dir,
1388                               struct o2hb_debug_buf **db, int db_len, int type,
1389                               int size, int len, void *data)
1390 {
1391         *db = kmalloc(db_len, GFP_KERNEL);
1392         if (!*db)
1393                 return;
1394
1395         (*db)->db_type = type;
1396         (*db)->db_size = size;
1397         (*db)->db_len = len;
1398         (*db)->db_data = data;
1399
1400         debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db, &o2hb_debug_fops);
1401 }
1402
1403 static void o2hb_debug_init(void)
1404 {
1405         o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1406
1407         o2hb_debug_create(O2HB_DEBUG_LIVENODES, o2hb_debug_dir,
1408                           &o2hb_db_livenodes, sizeof(*o2hb_db_livenodes),
1409                           O2HB_DB_TYPE_LIVENODES, sizeof(o2hb_live_node_bitmap),
1410                           O2NM_MAX_NODES, o2hb_live_node_bitmap);
1411
1412         o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS, o2hb_debug_dir,
1413                           &o2hb_db_liveregions, sizeof(*o2hb_db_liveregions),
1414                           O2HB_DB_TYPE_LIVEREGIONS,
1415                           sizeof(o2hb_live_region_bitmap), O2NM_MAX_REGIONS,
1416                           o2hb_live_region_bitmap);
1417
1418         o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS, o2hb_debug_dir,
1419                           &o2hb_db_quorumregions,
1420                           sizeof(*o2hb_db_quorumregions),
1421                           O2HB_DB_TYPE_QUORUMREGIONS,
1422                           sizeof(o2hb_quorum_region_bitmap), O2NM_MAX_REGIONS,
1423                           o2hb_quorum_region_bitmap);
1424
1425         o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS, o2hb_debug_dir,
1426                           &o2hb_db_failedregions,
1427                           sizeof(*o2hb_db_failedregions),
1428                           O2HB_DB_TYPE_FAILEDREGIONS,
1429                           sizeof(o2hb_failed_region_bitmap), O2NM_MAX_REGIONS,
1430                           o2hb_failed_region_bitmap);
1431 }
1432
1433 void o2hb_init(void)
1434 {
1435         int i;
1436
1437         for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1438                 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1439
1440         for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1441                 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1442
1443         memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1444         memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1445         memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1446         memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1447         memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1448
1449         o2hb_dependent_users = 0;
1450
1451         o2hb_debug_init();
1452 }
1453
1454 /* if we're already in a callback then we're already serialized by the sem */
1455 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1456                                              unsigned bytes)
1457 {
1458         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1459
1460         memcpy(map, &o2hb_live_node_bitmap, bytes);
1461 }
1462
1463 /*
1464  * get a map of all nodes that are heartbeating in any regions
1465  */
1466 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1467 {
1468         /* callers want to serialize this map and callbacks so that they
1469          * can trust that they don't miss nodes coming to the party */
1470         down_read(&o2hb_callback_sem);
1471         spin_lock(&o2hb_live_lock);
1472         o2hb_fill_node_map_from_callback(map, bytes);
1473         spin_unlock(&o2hb_live_lock);
1474         up_read(&o2hb_callback_sem);
1475 }
1476 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1477
1478 /*
1479  * heartbeat configfs bits.  The heartbeat set is a default set under
1480  * the cluster set in nodemanager.c.
1481  */
1482
1483 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1484 {
1485         return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1486 }
1487
1488 /* drop_item only drops its ref after killing the thread, nothing should
1489  * be using the region anymore.  this has to clean up any state that
1490  * attributes might have built up. */
1491 static void o2hb_region_release(struct config_item *item)
1492 {
1493         int i;
1494         struct page *page;
1495         struct o2hb_region *reg = to_o2hb_region(item);
1496
1497         mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1498
1499         kfree(reg->hr_tmp_block);
1500
1501         if (reg->hr_slot_data) {
1502                 for (i = 0; i < reg->hr_num_pages; i++) {
1503                         page = reg->hr_slot_data[i];
1504                         if (page)
1505                                 __free_page(page);
1506                 }
1507                 kfree(reg->hr_slot_data);
1508         }
1509
1510         if (reg->hr_bdev)
1511                 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1512
1513         kfree(reg->hr_slots);
1514
1515         debugfs_remove_recursive(reg->hr_debug_dir);
1516         kfree(reg->hr_db_livenodes);
1517         kfree(reg->hr_db_regnum);
1518         kfree(reg->hr_db_elapsed_time);
1519         kfree(reg->hr_db_pinned);
1520
1521         spin_lock(&o2hb_live_lock);
1522         list_del(&reg->hr_all_item);
1523         spin_unlock(&o2hb_live_lock);
1524
1525         o2net_unregister_handler_list(&reg->hr_handler_list);
1526         kfree(reg);
1527 }
1528
1529 static int o2hb_read_block_input(struct o2hb_region *reg,
1530                                  const char *page,
1531                                  unsigned long *ret_bytes,
1532                                  unsigned int *ret_bits)
1533 {
1534         unsigned long bytes;
1535         char *p = (char *)page;
1536
1537         bytes = simple_strtoul(p, &p, 0);
1538         if (!p || (*p && (*p != '\n')))
1539                 return -EINVAL;
1540
1541         /* Heartbeat and fs min / max block sizes are the same. */
1542         if (bytes > 4096 || bytes < 512)
1543                 return -ERANGE;
1544         if (hweight16(bytes) != 1)
1545                 return -EINVAL;
1546
1547         if (ret_bytes)
1548                 *ret_bytes = bytes;
1549         if (ret_bits)
1550                 *ret_bits = ffs(bytes) - 1;
1551
1552         return 0;
1553 }
1554
1555 static ssize_t o2hb_region_block_bytes_show(struct config_item *item,
1556                                             char *page)
1557 {
1558         return sprintf(page, "%u\n", to_o2hb_region(item)->hr_block_bytes);
1559 }
1560
1561 static ssize_t o2hb_region_block_bytes_store(struct config_item *item,
1562                                              const char *page,
1563                                              size_t count)
1564 {
1565         struct o2hb_region *reg = to_o2hb_region(item);
1566         int status;
1567         unsigned long block_bytes;
1568         unsigned int block_bits;
1569
1570         if (reg->hr_bdev)
1571                 return -EINVAL;
1572
1573         status = o2hb_read_block_input(reg, page, &block_bytes,
1574                                        &block_bits);
1575         if (status)
1576                 return status;
1577
1578         reg->hr_block_bytes = (unsigned int)block_bytes;
1579         reg->hr_block_bits = block_bits;
1580
1581         return count;
1582 }
1583
1584 static ssize_t o2hb_region_start_block_show(struct config_item *item,
1585                                             char *page)
1586 {
1587         return sprintf(page, "%llu\n", to_o2hb_region(item)->hr_start_block);
1588 }
1589
1590 static ssize_t o2hb_region_start_block_store(struct config_item *item,
1591                                              const char *page,
1592                                              size_t count)
1593 {
1594         struct o2hb_region *reg = to_o2hb_region(item);
1595         unsigned long long tmp;
1596         char *p = (char *)page;
1597         ssize_t ret;
1598
1599         if (reg->hr_bdev)
1600                 return -EINVAL;
1601
1602         ret = kstrtoull(p, 0, &tmp);
1603         if (ret)
1604                 return -EINVAL;
1605
1606         reg->hr_start_block = tmp;
1607
1608         return count;
1609 }
1610
1611 static ssize_t o2hb_region_blocks_show(struct config_item *item, char *page)
1612 {
1613         return sprintf(page, "%d\n", to_o2hb_region(item)->hr_blocks);
1614 }
1615
1616 static ssize_t o2hb_region_blocks_store(struct config_item *item,
1617                                         const char *page,
1618                                         size_t count)
1619 {
1620         struct o2hb_region *reg = to_o2hb_region(item);
1621         unsigned long tmp;
1622         char *p = (char *)page;
1623
1624         if (reg->hr_bdev)
1625                 return -EINVAL;
1626
1627         tmp = simple_strtoul(p, &p, 0);
1628         if (!p || (*p && (*p != '\n')))
1629                 return -EINVAL;
1630
1631         if (tmp > O2NM_MAX_NODES || tmp == 0)
1632                 return -ERANGE;
1633
1634         reg->hr_blocks = (unsigned int)tmp;
1635
1636         return count;
1637 }
1638
1639 static ssize_t o2hb_region_dev_show(struct config_item *item, char *page)
1640 {
1641         unsigned int ret = 0;
1642
1643         if (to_o2hb_region(item)->hr_bdev)
1644                 ret = sprintf(page, "%s\n", to_o2hb_region(item)->hr_dev_name);
1645
1646         return ret;
1647 }
1648
1649 static void o2hb_init_region_params(struct o2hb_region *reg)
1650 {
1651         reg->hr_slots_per_page = PAGE_SIZE >> reg->hr_block_bits;
1652         reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1653
1654         mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1655              reg->hr_start_block, reg->hr_blocks);
1656         mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1657              reg->hr_block_bytes, reg->hr_block_bits);
1658         mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1659         mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1660 }
1661
1662 static int o2hb_map_slot_data(struct o2hb_region *reg)
1663 {
1664         int i, j;
1665         unsigned int last_slot;
1666         unsigned int spp = reg->hr_slots_per_page;
1667         struct page *page;
1668         char *raw;
1669         struct o2hb_disk_slot *slot;
1670
1671         reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1672         if (reg->hr_tmp_block == NULL)
1673                 return -ENOMEM;
1674
1675         reg->hr_slots = kcalloc(reg->hr_blocks,
1676                                 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1677         if (reg->hr_slots == NULL)
1678                 return -ENOMEM;
1679
1680         for(i = 0; i < reg->hr_blocks; i++) {
1681                 slot = &reg->hr_slots[i];
1682                 slot->ds_node_num = i;
1683                 INIT_LIST_HEAD(&slot->ds_live_item);
1684                 slot->ds_raw_block = NULL;
1685         }
1686
1687         reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1688         mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1689                            "at %u blocks per page\n",
1690              reg->hr_num_pages, reg->hr_blocks, spp);
1691
1692         reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1693                                     GFP_KERNEL);
1694         if (!reg->hr_slot_data)
1695                 return -ENOMEM;
1696
1697         for(i = 0; i < reg->hr_num_pages; i++) {
1698                 page = alloc_page(GFP_KERNEL);
1699                 if (!page)
1700                         return -ENOMEM;
1701
1702                 reg->hr_slot_data[i] = page;
1703
1704                 last_slot = i * spp;
1705                 raw = page_address(page);
1706                 for (j = 0;
1707                      (j < spp) && ((j + last_slot) < reg->hr_blocks);
1708                      j++) {
1709                         BUG_ON((j + last_slot) >= reg->hr_blocks);
1710
1711                         slot = &reg->hr_slots[j + last_slot];
1712                         slot->ds_raw_block =
1713                                 (struct o2hb_disk_heartbeat_block *) raw;
1714
1715                         raw += reg->hr_block_bytes;
1716                 }
1717         }
1718
1719         return 0;
1720 }
1721
1722 /* Read in all the slots available and populate the tracking
1723  * structures so that we can start with a baseline idea of what's
1724  * there. */
1725 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1726 {
1727         int ret, i;
1728         struct o2hb_disk_slot *slot;
1729         struct o2hb_disk_heartbeat_block *hb_block;
1730
1731         ret = o2hb_read_slots(reg, 0, reg->hr_blocks);
1732         if (ret)
1733                 goto out;
1734
1735         /* We only want to get an idea of the values initially in each
1736          * slot, so we do no verification - o2hb_check_slot will
1737          * actually determine if each configured slot is valid and
1738          * whether any values have changed. */
1739         for(i = 0; i < reg->hr_blocks; i++) {
1740                 slot = &reg->hr_slots[i];
1741                 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1742
1743                 /* Only fill the values that o2hb_check_slot uses to
1744                  * determine changing slots */
1745                 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1746                 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1747         }
1748
1749 out:
1750         return ret;
1751 }
1752
1753 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1754 static ssize_t o2hb_region_dev_store(struct config_item *item,
1755                                      const char *page,
1756                                      size_t count)
1757 {
1758         struct o2hb_region *reg = to_o2hb_region(item);
1759         struct task_struct *hb_task;
1760         long fd;
1761         int sectsize;
1762         char *p = (char *)page;
1763         struct fd f;
1764         ssize_t ret = -EINVAL;
1765         int live_threshold;
1766
1767         if (reg->hr_bdev)
1768                 goto out;
1769
1770         /* We can't heartbeat without having had our node number
1771          * configured yet. */
1772         if (o2nm_this_node() == O2NM_MAX_NODES)
1773                 goto out;
1774
1775         fd = simple_strtol(p, &p, 0);
1776         if (!p || (*p && (*p != '\n')))
1777                 goto out;
1778
1779         if (fd < 0 || fd >= INT_MAX)
1780                 goto out;
1781
1782         f = fdget(fd);
1783         if (f.file == NULL)
1784                 goto out;
1785
1786         if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1787             reg->hr_block_bytes == 0)
1788                 goto out2;
1789
1790         if (!S_ISBLK(f.file->f_mapping->host->i_mode))
1791                 goto out2;
1792
1793         reg->hr_bdev = blkdev_get_by_dev(f.file->f_mapping->host->i_rdev,
1794                                          FMODE_WRITE | FMODE_READ, NULL);
1795         if (IS_ERR(reg->hr_bdev)) {
1796                 ret = PTR_ERR(reg->hr_bdev);
1797                 reg->hr_bdev = NULL;
1798                 goto out2;
1799         }
1800
1801         bdevname(reg->hr_bdev, reg->hr_dev_name);
1802
1803         sectsize = bdev_logical_block_size(reg->hr_bdev);
1804         if (sectsize != reg->hr_block_bytes) {
1805                 mlog(ML_ERROR,
1806                      "blocksize %u incorrect for device, expected %d",
1807                      reg->hr_block_bytes, sectsize);
1808                 ret = -EINVAL;
1809                 goto out3;
1810         }
1811
1812         o2hb_init_region_params(reg);
1813
1814         /* Generation of zero is invalid */
1815         do {
1816                 get_random_bytes(&reg->hr_generation,
1817                                  sizeof(reg->hr_generation));
1818         } while (reg->hr_generation == 0);
1819
1820         ret = o2hb_map_slot_data(reg);
1821         if (ret) {
1822                 mlog_errno(ret);
1823                 goto out3;
1824         }
1825
1826         ret = o2hb_populate_slot_data(reg);
1827         if (ret) {
1828                 mlog_errno(ret);
1829                 goto out3;
1830         }
1831
1832         INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1833         INIT_DELAYED_WORK(&reg->hr_nego_timeout_work, o2hb_nego_timeout);
1834
1835         /*
1836          * A node is considered live after it has beat LIVE_THRESHOLD
1837          * times.  We're not steady until we've given them a chance
1838          * _after_ our first read.
1839          * The default threshold is bare minimum so as to limit the delay
1840          * during mounts. For global heartbeat, the threshold doubled for the
1841          * first region.
1842          */
1843         live_threshold = O2HB_LIVE_THRESHOLD;
1844         if (o2hb_global_heartbeat_active()) {
1845                 spin_lock(&o2hb_live_lock);
1846                 if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1847                         live_threshold <<= 1;
1848                 spin_unlock(&o2hb_live_lock);
1849         }
1850         ++live_threshold;
1851         atomic_set(&reg->hr_steady_iterations, live_threshold);
1852         /* unsteady_iterations is triple the steady_iterations */
1853         atomic_set(&reg->hr_unsteady_iterations, (live_threshold * 3));
1854
1855         hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1856                               reg->hr_item.ci_name);
1857         if (IS_ERR(hb_task)) {
1858                 ret = PTR_ERR(hb_task);
1859                 mlog_errno(ret);
1860                 goto out3;
1861         }
1862
1863         spin_lock(&o2hb_live_lock);
1864         reg->hr_task = hb_task;
1865         spin_unlock(&o2hb_live_lock);
1866
1867         ret = wait_event_interruptible(o2hb_steady_queue,
1868                                 atomic_read(&reg->hr_steady_iterations) == 0 ||
1869                                 reg->hr_node_deleted);
1870         if (ret) {
1871                 atomic_set(&reg->hr_steady_iterations, 0);
1872                 reg->hr_aborted_start = 1;
1873         }
1874
1875         if (reg->hr_aborted_start) {
1876                 ret = -EIO;
1877                 goto out3;
1878         }
1879
1880         if (reg->hr_node_deleted) {
1881                 ret = -EINVAL;
1882                 goto out3;
1883         }
1884
1885         /* Ok, we were woken.  Make sure it wasn't by drop_item() */
1886         spin_lock(&o2hb_live_lock);
1887         hb_task = reg->hr_task;
1888         if (o2hb_global_heartbeat_active())
1889                 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1890         spin_unlock(&o2hb_live_lock);
1891
1892         if (hb_task)
1893                 ret = count;
1894         else
1895                 ret = -EIO;
1896
1897         if (hb_task && o2hb_global_heartbeat_active())
1898                 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1899                        config_item_name(&reg->hr_item), reg->hr_dev_name);
1900
1901 out3:
1902         if (ret < 0) {
1903                 blkdev_put(reg->hr_bdev, FMODE_READ | FMODE_WRITE);
1904                 reg->hr_bdev = NULL;
1905         }
1906 out2:
1907         fdput(f);
1908 out:
1909         return ret;
1910 }
1911
1912 static ssize_t o2hb_region_pid_show(struct config_item *item, char *page)
1913 {
1914         struct o2hb_region *reg = to_o2hb_region(item);
1915         pid_t pid = 0;
1916
1917         spin_lock(&o2hb_live_lock);
1918         if (reg->hr_task)
1919                 pid = task_pid_nr(reg->hr_task);
1920         spin_unlock(&o2hb_live_lock);
1921
1922         if (!pid)
1923                 return 0;
1924
1925         return sprintf(page, "%u\n", pid);
1926 }
1927
1928 CONFIGFS_ATTR(o2hb_region_, block_bytes);
1929 CONFIGFS_ATTR(o2hb_region_, start_block);
1930 CONFIGFS_ATTR(o2hb_region_, blocks);
1931 CONFIGFS_ATTR(o2hb_region_, dev);
1932 CONFIGFS_ATTR_RO(o2hb_region_, pid);
1933
1934 static struct configfs_attribute *o2hb_region_attrs[] = {
1935         &o2hb_region_attr_block_bytes,
1936         &o2hb_region_attr_start_block,
1937         &o2hb_region_attr_blocks,
1938         &o2hb_region_attr_dev,
1939         &o2hb_region_attr_pid,
1940         NULL,
1941 };
1942
1943 static struct configfs_item_operations o2hb_region_item_ops = {
1944         .release                = o2hb_region_release,
1945 };
1946
1947 static const struct config_item_type o2hb_region_type = {
1948         .ct_item_ops    = &o2hb_region_item_ops,
1949         .ct_attrs       = o2hb_region_attrs,
1950         .ct_owner       = THIS_MODULE,
1951 };
1952
1953 /* heartbeat set */
1954
1955 struct o2hb_heartbeat_group {
1956         struct config_group hs_group;
1957         /* some stuff? */
1958 };
1959
1960 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1961 {
1962         return group ?
1963                 container_of(group, struct o2hb_heartbeat_group, hs_group)
1964                 : NULL;
1965 }
1966
1967 static void o2hb_debug_region_init(struct o2hb_region *reg,
1968                                    struct dentry *parent)
1969 {
1970         struct dentry *dir;
1971
1972         dir = debugfs_create_dir(config_item_name(&reg->hr_item), parent);
1973         reg->hr_debug_dir = dir;
1974
1975         o2hb_debug_create(O2HB_DEBUG_LIVENODES, dir, &(reg->hr_db_livenodes),
1976                           sizeof(*(reg->hr_db_livenodes)),
1977                           O2HB_DB_TYPE_REGION_LIVENODES,
1978                           sizeof(reg->hr_live_node_bitmap), O2NM_MAX_NODES,
1979                           reg);
1980
1981         o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER, dir, &(reg->hr_db_regnum),
1982                           sizeof(*(reg->hr_db_regnum)),
1983                           O2HB_DB_TYPE_REGION_NUMBER, 0, O2NM_MAX_NODES, reg);
1984
1985         o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME, dir,
1986                           &(reg->hr_db_elapsed_time),
1987                           sizeof(*(reg->hr_db_elapsed_time)),
1988                           O2HB_DB_TYPE_REGION_ELAPSED_TIME, 0, 0, reg);
1989
1990         o2hb_debug_create(O2HB_DEBUG_REGION_PINNED, dir, &(reg->hr_db_pinned),
1991                           sizeof(*(reg->hr_db_pinned)),
1992                           O2HB_DB_TYPE_REGION_PINNED, 0, 0, reg);
1993
1994 }
1995
1996 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
1997                                                           const char *name)
1998 {
1999         struct o2hb_region *reg = NULL;
2000         int ret;
2001
2002         reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2003         if (reg == NULL)
2004                 return ERR_PTR(-ENOMEM);
2005
2006         if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2007                 ret = -ENAMETOOLONG;
2008                 goto free;
2009         }
2010
2011         spin_lock(&o2hb_live_lock);
2012         reg->hr_region_num = 0;
2013         if (o2hb_global_heartbeat_active()) {
2014                 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2015                                                          O2NM_MAX_REGIONS);
2016                 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2017                         spin_unlock(&o2hb_live_lock);
2018                         ret = -EFBIG;
2019                         goto free;
2020                 }
2021                 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2022         }
2023         list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2024         spin_unlock(&o2hb_live_lock);
2025
2026         config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2027
2028         /* this is the same way to generate msg key as dlm, for local heartbeat,
2029          * name is also the same, so make initial crc value different to avoid
2030          * message key conflict.
2031          */
2032         reg->hr_key = crc32_le(reg->hr_region_num + O2NM_MAX_REGIONS,
2033                 name, strlen(name));
2034         INIT_LIST_HEAD(&reg->hr_handler_list);
2035         ret = o2net_register_handler(O2HB_NEGO_TIMEOUT_MSG, reg->hr_key,
2036                         sizeof(struct o2hb_nego_msg),
2037                         o2hb_nego_timeout_handler,
2038                         reg, NULL, &reg->hr_handler_list);
2039         if (ret)
2040                 goto remove_item;
2041
2042         ret = o2net_register_handler(O2HB_NEGO_APPROVE_MSG, reg->hr_key,
2043                         sizeof(struct o2hb_nego_msg),
2044                         o2hb_nego_approve_handler,
2045                         reg, NULL, &reg->hr_handler_list);
2046         if (ret)
2047                 goto unregister_handler;
2048
2049         o2hb_debug_region_init(reg, o2hb_debug_dir);
2050
2051         return &reg->hr_item;
2052
2053 unregister_handler:
2054         o2net_unregister_handler_list(&reg->hr_handler_list);
2055 remove_item:
2056         spin_lock(&o2hb_live_lock);
2057         list_del(&reg->hr_all_item);
2058         if (o2hb_global_heartbeat_active())
2059                 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2060         spin_unlock(&o2hb_live_lock);
2061 free:
2062         kfree(reg);
2063         return ERR_PTR(ret);
2064 }
2065
2066 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2067                                            struct config_item *item)
2068 {
2069         struct task_struct *hb_task;
2070         struct o2hb_region *reg = to_o2hb_region(item);
2071         int quorum_region = 0;
2072
2073         /* stop the thread when the user removes the region dir */
2074         spin_lock(&o2hb_live_lock);
2075         hb_task = reg->hr_task;
2076         reg->hr_task = NULL;
2077         reg->hr_item_dropped = 1;
2078         spin_unlock(&o2hb_live_lock);
2079
2080         if (hb_task)
2081                 kthread_stop(hb_task);
2082
2083         if (o2hb_global_heartbeat_active()) {
2084                 spin_lock(&o2hb_live_lock);
2085                 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2086                 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2087                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2088                         quorum_region = 1;
2089                 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2090                 spin_unlock(&o2hb_live_lock);
2091                 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2092                        ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2093                         "stopped" : "start aborted"), config_item_name(item),
2094                        reg->hr_dev_name);
2095         }
2096
2097         /*
2098          * If we're racing a dev_write(), we need to wake them.  They will
2099          * check reg->hr_task
2100          */
2101         if (atomic_read(&reg->hr_steady_iterations) != 0) {
2102                 reg->hr_aborted_start = 1;
2103                 atomic_set(&reg->hr_steady_iterations, 0);
2104                 wake_up(&o2hb_steady_queue);
2105         }
2106
2107         config_item_put(item);
2108
2109         if (!o2hb_global_heartbeat_active() || !quorum_region)
2110                 return;
2111
2112         /*
2113          * If global heartbeat active and there are dependent users,
2114          * pin all regions if quorum region count <= CUT_OFF
2115          */
2116         spin_lock(&o2hb_live_lock);
2117
2118         if (!o2hb_dependent_users)
2119                 goto unlock;
2120
2121         if (bitmap_weight(o2hb_quorum_region_bitmap,
2122                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2123                 o2hb_region_pin(NULL);
2124
2125 unlock:
2126         spin_unlock(&o2hb_live_lock);
2127 }
2128
2129 static ssize_t o2hb_heartbeat_group_dead_threshold_show(struct config_item *item,
2130                 char *page)
2131 {
2132         return sprintf(page, "%u\n", o2hb_dead_threshold);
2133 }
2134
2135 static ssize_t o2hb_heartbeat_group_dead_threshold_store(struct config_item *item,
2136                 const char *page, size_t count)
2137 {
2138         unsigned long tmp;
2139         char *p = (char *)page;
2140
2141         tmp = simple_strtoul(p, &p, 10);
2142         if (!p || (*p && (*p != '\n')))
2143                 return -EINVAL;
2144
2145         /* this will validate ranges for us. */
2146         o2hb_dead_threshold_set((unsigned int) tmp);
2147
2148         return count;
2149 }
2150
2151 static ssize_t o2hb_heartbeat_group_mode_show(struct config_item *item,
2152                 char *page)
2153 {
2154         return sprintf(page, "%s\n",
2155                        o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2156 }
2157
2158 static ssize_t o2hb_heartbeat_group_mode_store(struct config_item *item,
2159                 const char *page, size_t count)
2160 {
2161         unsigned int i;
2162         int ret;
2163         size_t len;
2164
2165         len = (page[count - 1] == '\n') ? count - 1 : count;
2166         if (!len)
2167                 return -EINVAL;
2168
2169         for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2170                 if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
2171                         continue;
2172
2173                 ret = o2hb_global_heartbeat_mode_set(i);
2174                 if (!ret)
2175                         printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2176                                o2hb_heartbeat_mode_desc[i]);
2177                 return count;
2178         }
2179
2180         return -EINVAL;
2181
2182 }
2183
2184 CONFIGFS_ATTR(o2hb_heartbeat_group_, dead_threshold);
2185 CONFIGFS_ATTR(o2hb_heartbeat_group_, mode);
2186
2187 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2188         &o2hb_heartbeat_group_attr_dead_threshold,
2189         &o2hb_heartbeat_group_attr_mode,
2190         NULL,
2191 };
2192
2193 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2194         .make_item      = o2hb_heartbeat_group_make_item,
2195         .drop_item      = o2hb_heartbeat_group_drop_item,
2196 };
2197
2198 static const struct config_item_type o2hb_heartbeat_group_type = {
2199         .ct_group_ops   = &o2hb_heartbeat_group_group_ops,
2200         .ct_attrs       = o2hb_heartbeat_group_attrs,
2201         .ct_owner       = THIS_MODULE,
2202 };
2203
2204 /* this is just here to avoid touching group in heartbeat.h which the
2205  * entire damn world #includes */
2206 struct config_group *o2hb_alloc_hb_set(void)
2207 {
2208         struct o2hb_heartbeat_group *hs = NULL;
2209         struct config_group *ret = NULL;
2210
2211         hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2212         if (hs == NULL)
2213                 goto out;
2214
2215         config_group_init_type_name(&hs->hs_group, "heartbeat",
2216                                     &o2hb_heartbeat_group_type);
2217
2218         ret = &hs->hs_group;
2219 out:
2220         if (ret == NULL)
2221                 kfree(hs);
2222         return ret;
2223 }
2224
2225 void o2hb_free_hb_set(struct config_group *group)
2226 {
2227         struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2228         kfree(hs);
2229 }
2230
2231 /* hb callback registration and issuing */
2232
2233 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2234 {
2235         if (type == O2HB_NUM_CB)
2236                 return ERR_PTR(-EINVAL);
2237
2238         return &o2hb_callbacks[type];
2239 }
2240
2241 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2242                          enum o2hb_callback_type type,
2243                          o2hb_cb_func *func,
2244                          void *data,
2245                          int priority)
2246 {
2247         INIT_LIST_HEAD(&hc->hc_item);
2248         hc->hc_func = func;
2249         hc->hc_data = data;
2250         hc->hc_priority = priority;
2251         hc->hc_type = type;
2252         hc->hc_magic = O2HB_CB_MAGIC;
2253 }
2254 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2255
2256 /*
2257  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2258  * In global heartbeat mode, region_uuid passed is NULL.
2259  *
2260  * In local, we only pin the matching region. In global we pin all the active
2261  * regions.
2262  */
2263 static int o2hb_region_pin(const char *region_uuid)
2264 {
2265         int ret = 0, found = 0;
2266         struct o2hb_region *reg;
2267         char *uuid;
2268
2269         assert_spin_locked(&o2hb_live_lock);
2270
2271         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2272                 if (reg->hr_item_dropped)
2273                         continue;
2274
2275                 uuid = config_item_name(&reg->hr_item);
2276
2277                 /* local heartbeat */
2278                 if (region_uuid) {
2279                         if (strcmp(region_uuid, uuid))
2280                                 continue;
2281                         found = 1;
2282                 }
2283
2284                 if (reg->hr_item_pinned || reg->hr_item_dropped)
2285                         goto skip_pin;
2286
2287                 /* Ignore ENOENT only for local hb (userdlm domain) */
2288                 ret = o2nm_depend_item(&reg->hr_item);
2289                 if (!ret) {
2290                         mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2291                         reg->hr_item_pinned = 1;
2292                 } else {
2293                         if (ret == -ENOENT && found)
2294                                 ret = 0;
2295                         else {
2296                                 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2297                                      uuid, ret);
2298                                 break;
2299                         }
2300                 }
2301 skip_pin:
2302                 if (found)
2303                         break;
2304         }
2305
2306         return ret;
2307 }
2308
2309 /*
2310  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2311  * In global heartbeat mode, region_uuid passed is NULL.
2312  *
2313  * In local, we only unpin the matching region. In global we unpin all the
2314  * active regions.
2315  */
2316 static void o2hb_region_unpin(const char *region_uuid)
2317 {
2318         struct o2hb_region *reg;
2319         char *uuid;
2320         int found = 0;
2321
2322         assert_spin_locked(&o2hb_live_lock);
2323
2324         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2325                 if (reg->hr_item_dropped)
2326                         continue;
2327
2328                 uuid = config_item_name(&reg->hr_item);
2329                 if (region_uuid) {
2330                         if (strcmp(region_uuid, uuid))
2331                                 continue;
2332                         found = 1;
2333                 }
2334
2335                 if (reg->hr_item_pinned) {
2336                         mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2337                         o2nm_undepend_item(&reg->hr_item);
2338                         reg->hr_item_pinned = 0;
2339                 }
2340                 if (found)
2341                         break;
2342         }
2343 }
2344
2345 static int o2hb_region_inc_user(const char *region_uuid)
2346 {
2347         int ret = 0;
2348
2349         spin_lock(&o2hb_live_lock);
2350
2351         /* local heartbeat */
2352         if (!o2hb_global_heartbeat_active()) {
2353             ret = o2hb_region_pin(region_uuid);
2354             goto unlock;
2355         }
2356
2357         /*
2358          * if global heartbeat active and this is the first dependent user,
2359          * pin all regions if quorum region count <= CUT_OFF
2360          */
2361         o2hb_dependent_users++;
2362         if (o2hb_dependent_users > 1)
2363                 goto unlock;
2364
2365         if (bitmap_weight(o2hb_quorum_region_bitmap,
2366                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2367                 ret = o2hb_region_pin(NULL);
2368
2369 unlock:
2370         spin_unlock(&o2hb_live_lock);
2371         return ret;
2372 }
2373
2374 static void o2hb_region_dec_user(const char *region_uuid)
2375 {
2376         spin_lock(&o2hb_live_lock);
2377
2378         /* local heartbeat */
2379         if (!o2hb_global_heartbeat_active()) {
2380             o2hb_region_unpin(region_uuid);
2381             goto unlock;
2382         }
2383
2384         /*
2385          * if global heartbeat active and there are no dependent users,
2386          * unpin all quorum regions
2387          */
2388         o2hb_dependent_users--;
2389         if (!o2hb_dependent_users)
2390                 o2hb_region_unpin(NULL);
2391
2392 unlock:
2393         spin_unlock(&o2hb_live_lock);
2394 }
2395
2396 int o2hb_register_callback(const char *region_uuid,
2397                            struct o2hb_callback_func *hc)
2398 {
2399         struct o2hb_callback_func *f;
2400         struct o2hb_callback *hbcall;
2401         int ret;
2402
2403         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2404         BUG_ON(!list_empty(&hc->hc_item));
2405
2406         hbcall = hbcall_from_type(hc->hc_type);
2407         if (IS_ERR(hbcall)) {
2408                 ret = PTR_ERR(hbcall);
2409                 goto out;
2410         }
2411
2412         if (region_uuid) {
2413                 ret = o2hb_region_inc_user(region_uuid);
2414                 if (ret) {
2415                         mlog_errno(ret);
2416                         goto out;
2417                 }
2418         }
2419
2420         down_write(&o2hb_callback_sem);
2421
2422         list_for_each_entry(f, &hbcall->list, hc_item) {
2423                 if (hc->hc_priority < f->hc_priority) {
2424                         list_add_tail(&hc->hc_item, &f->hc_item);
2425                         break;
2426                 }
2427         }
2428         if (list_empty(&hc->hc_item))
2429                 list_add_tail(&hc->hc_item, &hbcall->list);
2430
2431         up_write(&o2hb_callback_sem);
2432         ret = 0;
2433 out:
2434         mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2435              ret, __builtin_return_address(0), hc);
2436         return ret;
2437 }
2438 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2439
2440 void o2hb_unregister_callback(const char *region_uuid,
2441                               struct o2hb_callback_func *hc)
2442 {
2443         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2444
2445         mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2446              __builtin_return_address(0), hc);
2447
2448         /* XXX Can this happen _with_ a region reference? */
2449         if (list_empty(&hc->hc_item))
2450                 return;
2451
2452         if (region_uuid)
2453                 o2hb_region_dec_user(region_uuid);
2454
2455         down_write(&o2hb_callback_sem);
2456
2457         list_del_init(&hc->hc_item);
2458
2459         up_write(&o2hb_callback_sem);
2460 }
2461 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2462
2463 int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2464 {
2465         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2466
2467         spin_lock(&o2hb_live_lock);
2468         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2469         spin_unlock(&o2hb_live_lock);
2470         if (!test_bit(node_num, testing_map)) {
2471                 mlog(ML_HEARTBEAT,
2472                      "node (%u) does not have heartbeating enabled.\n",
2473                      node_num);
2474                 return 0;
2475         }
2476
2477         return 1;
2478 }
2479 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2480
2481 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2482 {
2483         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2484
2485         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2486         if (!test_bit(node_num, testing_map)) {
2487                 mlog(ML_HEARTBEAT,
2488                      "node (%u) does not have heartbeating enabled.\n",
2489                      node_num);
2490                 return 0;
2491         }
2492
2493         return 1;
2494 }
2495 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2496
2497 /*
2498  * this is just a hack until we get the plumbing which flips file systems
2499  * read only and drops the hb ref instead of killing the node dead.
2500  */
2501 void o2hb_stop_all_regions(void)
2502 {
2503         struct o2hb_region *reg;
2504
2505         mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2506
2507         spin_lock(&o2hb_live_lock);
2508
2509         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2510                 reg->hr_unclean_stop = 1;
2511
2512         spin_unlock(&o2hb_live_lock);
2513 }
2514 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2515
2516 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2517 {
2518         struct o2hb_region *reg;
2519         int numregs = 0;
2520         char *p;
2521
2522         spin_lock(&o2hb_live_lock);
2523
2524         p = region_uuids;
2525         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2526                 if (reg->hr_item_dropped)
2527                         continue;
2528
2529                 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2530                 if (numregs < max_regions) {
2531                         memcpy(p, config_item_name(&reg->hr_item),
2532                                O2HB_MAX_REGION_NAME_LEN);
2533                         p += O2HB_MAX_REGION_NAME_LEN;
2534                 }
2535                 numregs++;
2536         }
2537
2538         spin_unlock(&o2hb_live_lock);
2539
2540         return numregs;
2541 }
2542 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2543
2544 int o2hb_global_heartbeat_active(void)
2545 {
2546         return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2547 }
2548 EXPORT_SYMBOL(o2hb_global_heartbeat_active);