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