fs/ocfs2/cluster/heartbeat.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* -*- mode: c; c-basic-offset: 8; -*-
   3  * vim: noexpandtab sw=8 ts=8 sts=0:
   4  *
   5  * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
   6  */
   7
   8 #include <linux/kernel.h>
   9 #include <linux/sched.h>
  10 #include <linux/jiffies.h>
  11 #include <linux/module.h>
  12 #include <linux/fs.h>
  13 #include <linux/bio.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/delay.h>
  16 #include <linux/file.h>
  17 #include <linux/kthread.h>
  18 #include <linux/configfs.h>
  19 #include <linux/random.h>
  20 #include <linux/crc32.h>
  21 #include <linux/time.h>
  22 #include <linux/debugfs.h>
  23 #include <linux/slab.h>
  24 #include <linux/bitmap.h>
  25 #include <linux/ktime.h>
  26 #include "heartbeat.h"
  27 #include "tcp.h"
  28 #include "nodemanager.h"
  29 #include "quorum.h"
  30
  31 #include "masklog.h"
  32
  33
  34 /*
  35  * The first heartbeat pass had one global thread that would serialize all hb
  36  * callback calls.  This global serializing sem should only be removed once
  37  * we've made sure that all callees can deal with being called concurrently
  38  * from multiple hb region threads.
  39  */
  40 static DECLARE_RWSEM(o2hb_callback_sem);
  41
  42 /*
  43  * multiple hb threads are watching multiple regions.  A node is live
  44  * whenever any of the threads sees activity from the node in its region.
  45  */
  46 static DEFINE_SPINLOCK(o2hb_live_lock);
  47 static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
  48 static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
  49 static LIST_HEAD(o2hb_node_events);
  50 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
  51
  52 /*
  53  * In global heartbeat, we maintain a series of region bitmaps.
  54  *      - o2hb_region_bitmap allows us to limit the region number to max region.
  55  *      - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
  56  *      - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
  57  *              heartbeat on it.
  58  *      - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
  59  */
  60 static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  61 static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  62 static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  63 static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  64
  65 #define O2HB_DB_TYPE_LIVENODES          0
  66 #define O2HB_DB_TYPE_LIVEREGIONS        1
  67 #define O2HB_DB_TYPE_QUORUMREGIONS      2
  68 #define O2HB_DB_TYPE_FAILEDREGIONS      3
  69 #define O2HB_DB_TYPE_REGION_LIVENODES   4
  70 #define O2HB_DB_TYPE_REGION_NUMBER      5
  71 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME        6
  72 #define O2HB_DB_TYPE_REGION_PINNED      7
  73 struct o2hb_debug_buf {
  74         int db_type;
  75         int db_size;
  76         int db_len;
  77         void *db_data;
  78 };
  79
  80 static struct o2hb_debug_buf *o2hb_db_livenodes;
  81 static struct o2hb_debug_buf *o2hb_db_liveregions;
  82 static struct o2hb_debug_buf *o2hb_db_quorumregions;
  83 static struct o2hb_debug_buf *o2hb_db_failedregions;
  84
  85 #define O2HB_DEBUG_DIR                  "o2hb"
  86 #define O2HB_DEBUG_LIVENODES            "livenodes"
  87 #define O2HB_DEBUG_LIVEREGIONS          "live_regions"
  88 #define O2HB_DEBUG_QUORUMREGIONS        "quorum_regions"
  89 #define O2HB_DEBUG_FAILEDREGIONS        "failed_regions"
  90 #define O2HB_DEBUG_REGION_NUMBER        "num"
  91 #define O2HB_DEBUG_REGION_ELAPSED_TIME  "elapsed_time_in_ms"
  92 #define O2HB_DEBUG_REGION_PINNED        "pinned"
  93
  94 static struct dentry *o2hb_debug_dir;
  95
  96 static LIST_HEAD(o2hb_all_regions);
  97
  98 static struct o2hb_callback {
  99         struct list_head list;
 100 } o2hb_callbacks[O2HB_NUM_CB];
 101
 102 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
 103
 104 #define O2HB_DEFAULT_BLOCK_BITS       9
 105
 106 enum o2hb_heartbeat_modes {
 107         O2HB_HEARTBEAT_LOCAL            = 0,
 108         O2HB_HEARTBEAT_GLOBAL,
 109         O2HB_HEARTBEAT_NUM_MODES,
 110 };
 111
 112 static const char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
 113         "local",        /* O2HB_HEARTBEAT_LOCAL */
 114         "global",       /* O2HB_HEARTBEAT_GLOBAL */
 115 };
 116
 117 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
 118 static unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
 119
 120 /*
 121  * o2hb_dependent_users tracks the number of registered callbacks that depend
 122  * on heartbeat. o2net and o2dlm are two entities that register this callback.
 123  * However only o2dlm depends on the heartbeat. It does not want the heartbeat
 124  * to stop while a dlm domain is still active.
 125  */
 126 static unsigned int o2hb_dependent_users;
 127
 128 /*
 129  * In global heartbeat mode, all regions are pinned if there are one or more
 130  * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
 131  * regions are unpinned if the region count exceeds the cut off or the number
 132  * of dependent users falls to zero.
 133  */
 134 #define O2HB_PIN_CUT_OFF                3
 135
 136 /*
 137  * In local heartbeat mode, we assume the dlm domain name to be the same as
 138  * region uuid. This is true for domains created for the file system but not
 139  * necessarily true for userdlm domains. This is a known limitation.
 140  *
 141  * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
 142  * works for both file system and userdlm domains.
 143  */
 144 static int o2hb_region_pin(const char *region_uuid);
 145 static void o2hb_region_unpin(const char *region_uuid);
 146
 147 /* Only sets a new threshold if there are no active regions.
 148  *
 149  * No locking or otherwise interesting code is required for reading
 150  * o2hb_dead_threshold as it can't change once regions are active and
 151  * it's not interesting to anyone until then anyway. */
 152 static void o2hb_dead_threshold_set(unsigned int threshold)
 153 {
 154         if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
 155                 spin_lock(&o2hb_live_lock);
 156                 if (list_empty(&o2hb_all_regions))
 157                         o2hb_dead_threshold = threshold;
 158                 spin_unlock(&o2hb_live_lock);
 159         }
 160 }
 161
 162 static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
 163 {
 164         int ret = -1;
 165
 166         if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
 167                 spin_lock(&o2hb_live_lock);
 168                 if (list_empty(&o2hb_all_regions)) {
 169                         o2hb_heartbeat_mode = hb_mode;
 170                         ret = 0;
 171                 }
 172                 spin_unlock(&o2hb_live_lock);
 173         }
 174
 175         return ret;
 176 }
 177
 178 struct o2hb_node_event {
 179         struct list_head        hn_item;
 180         enum o2hb_callback_type hn_event_type;
 181         struct o2nm_node        *hn_node;
 182         int                     hn_node_num;
 183 };
 184
 185 struct o2hb_disk_slot {
 186         struct o2hb_disk_heartbeat_block *ds_raw_block;
 187         u8                      ds_node_num;
 188         u64                     ds_last_time;
 189         u64                     ds_last_generation;
 190         u16                     ds_equal_samples;
 191         u16                     ds_changed_samples;
 192         struct list_head        ds_live_item;
 193 };
 194
 195 /* each thread owns a region.. when we're asked to tear down the region
 196  * we ask the thread to stop, who cleans up the region */
 197 struct o2hb_region {
 198         struct config_item      hr_item;
 199
 200         struct list_head        hr_all_item;
 201         unsigned                hr_unclean_stop:1,
 202                                 hr_aborted_start:1,
 203                                 hr_item_pinned:1,
 204                                 hr_item_dropped:1,
 205                                 hr_node_deleted:1;
 206
 207         /* protected by the hr_callback_sem */
 208         struct task_struct      *hr_task;
 209
 210         unsigned int            hr_blocks;
 211         unsigned long long      hr_start_block;
 212
 213         unsigned int            hr_block_bits;
 214         unsigned int            hr_block_bytes;
 215
 216         unsigned int            hr_slots_per_page;
 217         unsigned int            hr_num_pages;
 218
 219         struct page             **hr_slot_data;
 220         struct block_device     *hr_bdev;
 221         struct o2hb_disk_slot   *hr_slots;
 222
 223         /* live node map of this region */
 224         unsigned long           hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 225         unsigned int            hr_region_num;
 226
 227         struct dentry           *hr_debug_dir;
 228         struct dentry           *hr_debug_livenodes;
 229         struct dentry           *hr_debug_regnum;
 230         struct dentry           *hr_debug_elapsed_time;
 231         struct dentry           *hr_debug_pinned;
 232         struct o2hb_debug_buf   *hr_db_livenodes;
 233         struct o2hb_debug_buf   *hr_db_regnum;
 234         struct o2hb_debug_buf   *hr_db_elapsed_time;
 235         struct o2hb_debug_buf   *hr_db_pinned;
 236
 237         /* let the person setting up hb wait for it to return until it
 238          * has reached a 'steady' state.  This will be fixed when we have
 239          * a more complete api that doesn't lead to this sort of fragility. */
 240         atomic_t                hr_steady_iterations;
 241
 242         /* terminate o2hb thread if it does not reach steady state
 243          * (hr_steady_iterations == 0) within hr_unsteady_iterations */
 244         atomic_t                hr_unsteady_iterations;
 245
 246         char                    hr_dev_name[BDEVNAME_SIZE];
 247
 248         unsigned int            hr_timeout_ms;
 249
 250         /* randomized as the region goes up and down so that a node
 251          * recognizes a node going up and down in one iteration */
 252         u64                     hr_generation;
 253
 254         struct delayed_work     hr_write_timeout_work;
 255         unsigned long           hr_last_timeout_start;
 256
 257         /* negotiate timer, used to negotiate extending hb timeout. */
 258         struct delayed_work     hr_nego_timeout_work;
 259         unsigned long           hr_nego_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 260
 261         /* Used during o2hb_check_slot to hold a copy of the block
 262          * being checked because we temporarily have to zero out the
 263          * crc field. */
 264         struct o2hb_disk_heartbeat_block *hr_tmp_block;
 265
 266         /* Message key for negotiate timeout message. */
 267         unsigned int            hr_key;
 268         struct list_head        hr_handler_list;
 269
 270         /* last hb status, 0 for success, other value for error. */
 271         int                     hr_last_hb_status;
 272 };
 273
 274 struct o2hb_bio_wait_ctxt {
 275         atomic_t          wc_num_reqs;
 276         struct completion wc_io_complete;
 277         int               wc_error;
 278 };
 279
 280 #define O2HB_NEGO_TIMEOUT_MS (O2HB_MAX_WRITE_TIMEOUT_MS/2)
 281
 282 enum {
 283         O2HB_NEGO_TIMEOUT_MSG = 1,
 284         O2HB_NEGO_APPROVE_MSG = 2,
 285 };
 286
 287 struct o2hb_nego_msg {
 288         u8 node_num;
 289 };
 290
 291 static void o2hb_write_timeout(struct work_struct *work)
 292 {
 293         int failed, quorum;
 294         struct o2hb_region *reg =
 295                 container_of(work, struct o2hb_region,
 296                              hr_write_timeout_work.work);
 297
 298         mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
 299              "milliseconds\n", reg->hr_dev_name,
 300              jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
 301
 302         if (o2hb_global_heartbeat_active()) {
 303                 spin_lock(&o2hb_live_lock);
 304                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 305                         set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 306                 failed = bitmap_weight(o2hb_failed_region_bitmap,
 307                                         O2NM_MAX_REGIONS);
 308                 quorum = bitmap_weight(o2hb_quorum_region_bitmap,
 309                                         O2NM_MAX_REGIONS);
 310                 spin_unlock(&o2hb_live_lock);
 311
 312                 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
 313                      quorum, failed);
 314
 315                 /*
 316                  * Fence if the number of failed regions >= half the number
 317                  * of  quorum regions
 318                  */
 319                 if ((failed << 1) < quorum)
 320                         return;
 321         }
 322
 323         o2quo_disk_timeout();
 324 }
 325
 326 static void o2hb_arm_timeout(struct o2hb_region *reg)
 327 {
 328         /* Arm writeout only after thread reaches steady state */
 329         if (atomic_read(&reg->hr_steady_iterations) != 0)
 330                 return;
 331
 332         mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
 333              O2HB_MAX_WRITE_TIMEOUT_MS);
 334
 335         if (o2hb_global_heartbeat_active()) {
 336                 spin_lock(&o2hb_live_lock);
 337                 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 338                 spin_unlock(&o2hb_live_lock);
 339         }
 340         cancel_delayed_work(&reg->hr_write_timeout_work);
 341         schedule_delayed_work(&reg->hr_write_timeout_work,
 342                               msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
 343
 344         cancel_delayed_work(&reg->hr_nego_timeout_work);
 345         /* negotiate timeout must be less than write timeout. */
 346         schedule_delayed_work(&reg->hr_nego_timeout_work,
 347                               msecs_to_jiffies(O2HB_NEGO_TIMEOUT_MS));
 348         memset(reg->hr_nego_node_bitmap, 0, sizeof(reg->hr_nego_node_bitmap));
 349 }
 350
 351 static void o2hb_disarm_timeout(struct o2hb_region *reg)
 352 {
 353         cancel_delayed_work_sync(&reg->hr_write_timeout_work);
 354         cancel_delayed_work_sync(&reg->hr_nego_timeout_work);
 355 }
 356
 357 static int o2hb_send_nego_msg(int key, int type, u8 target)
 358 {
 359         struct o2hb_nego_msg msg;
 360         int status, ret;
 361
 362         msg.node_num = o2nm_this_node();
 363 again:
 364         ret = o2net_send_message(type, key, &msg, sizeof(msg),
 365                         target, &status);
 366
 367         if (ret == -EAGAIN || ret == -ENOMEM) {
 368                 msleep(100);
 369                 goto again;
 370         }
 371
 372         return ret;
 373 }
 374
 375 static void o2hb_nego_timeout(struct work_struct *work)
 376 {
 377         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 378         int master_node, i, ret;
 379         struct o2hb_region *reg;
 380
 381         reg = container_of(work, struct o2hb_region, hr_nego_timeout_work.work);
 382         /* don't negotiate timeout if last hb failed since it is very
 383          * possible io failed. Should let write timeout fence self.
 384          */
 385         if (reg->hr_last_hb_status)
 386                 return;
 387
 388         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
 389         /* lowest node as master node to make negotiate decision. */
 390         master_node = find_next_bit(live_node_bitmap, O2NM_MAX_NODES, 0);
 391
 392         if (master_node == o2nm_this_node()) {
 393                 if (!test_bit(master_node, reg->hr_nego_node_bitmap)) {
 394                         printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%s).\n",
 395                                 o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000,
 396                                 config_item_name(&reg->hr_item), reg->hr_dev_name);
 397                         set_bit(master_node, reg->hr_nego_node_bitmap);
 398                 }
 399                 if (memcmp(reg->hr_nego_node_bitmap, live_node_bitmap,
 400                                 sizeof(reg->hr_nego_node_bitmap))) {
 401                         /* check negotiate bitmap every second to do timeout
 402                          * approve decision.
 403                          */
 404                         schedule_delayed_work(&reg->hr_nego_timeout_work,
 405                                 msecs_to_jiffies(1000));
 406
 407                         return;
 408                 }
 409
 410                 printk(KERN_NOTICE "o2hb: all nodes hb write hung, maybe region %s (%s) is down.\n",
 411                         config_item_name(&reg->hr_item), reg->hr_dev_name);
 412                 /* approve negotiate timeout request. */
 413                 o2hb_arm_timeout(reg);
 414
 415                 i = -1;
 416                 while ((i = find_next_bit(live_node_bitmap,
 417                                 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
 418                         if (i == master_node)
 419                                 continue;
 420
 421                         mlog(ML_HEARTBEAT, "send NEGO_APPROVE msg to node %d\n", i);
 422                         ret = o2hb_send_nego_msg(reg->hr_key,
 423                                         O2HB_NEGO_APPROVE_MSG, i);
 424                         if (ret)
 425                                 mlog(ML_ERROR, "send NEGO_APPROVE msg to node %d fail %d\n",
 426                                         i, ret);
 427                 }
 428         } else {
 429                 /* negotiate timeout with master node. */
 430                 printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%s), negotiate timeout with node %d.\n",
 431                         o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000, config_item_name(&reg->hr_item),
 432                         reg->hr_dev_name, master_node);
 433                 ret = o2hb_send_nego_msg(reg->hr_key, O2HB_NEGO_TIMEOUT_MSG,
 434                                 master_node);
 435                 if (ret)
 436                         mlog(ML_ERROR, "send NEGO_TIMEOUT msg to node %d fail %d\n",
 437                                 master_node, ret);
 438         }
 439 }
 440
 441 static int o2hb_nego_timeout_handler(struct o2net_msg *msg, u32 len, void *data,
 442                                 void **ret_data)
 443 {
 444         struct o2hb_region *reg = data;
 445         struct o2hb_nego_msg *nego_msg;
 446
 447         nego_msg = (struct o2hb_nego_msg *)msg->buf;
 448         printk(KERN_NOTICE "o2hb: receive negotiate timeout message from node %d on region %s (%s).\n",
 449                 nego_msg->node_num, config_item_name(&reg->hr_item), reg->hr_dev_name);
 450         if (nego_msg->node_num < O2NM_MAX_NODES)
 451                 set_bit(nego_msg->node_num, reg->hr_nego_node_bitmap);
 452         else
 453                 mlog(ML_ERROR, "got nego timeout message from bad node.\n");
 454
 455         return 0;
 456 }
 457
 458 static int o2hb_nego_approve_handler(struct o2net_msg *msg, u32 len, void *data,
 459                                 void **ret_data)
 460 {
 461         struct o2hb_region *reg = data;
 462
 463         printk(KERN_NOTICE "o2hb: negotiate timeout approved by master node on region %s (%s).\n",
 464                 config_item_name(&reg->hr_item), reg->hr_dev_name);
 465         o2hb_arm_timeout(reg);
 466         return 0;
 467 }
 468
 469 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
 470 {
 471         atomic_set(&wc->wc_num_reqs, 1);
 472         init_completion(&wc->wc_io_complete);
 473         wc->wc_error = 0;
 474 }
 475
 476 /* Used in error paths too */
 477 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
 478                                      unsigned int num)
 479 {
 480         /* sadly atomic_sub_and_test() isn't available on all platforms.  The
 481          * good news is that the fast path only completes one at a time */
 482         while(num--) {
 483                 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
 484                         BUG_ON(num > 0);
 485                         complete(&wc->wc_io_complete);
 486                 }
 487         }
 488 }
 489
 490 static void o2hb_wait_on_io(struct o2hb_bio_wait_ctxt *wc)
 491 {
 492         o2hb_bio_wait_dec(wc, 1);
 493         wait_for_completion(&wc->wc_io_complete);
 494 }
 495
 496 static void o2hb_bio_end_io(struct bio *bio)
 497 {
 498         struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
 499
 500         if (bio->bi_status) {
 501                 mlog(ML_ERROR, "IO Error %d\n", bio->bi_status);
 502                 wc->wc_error = blk_status_to_errno(bio->bi_status);
 503         }
 504
 505         o2hb_bio_wait_dec(wc, 1);
 506         bio_put(bio);
 507 }
 508
 509 /* Setup a Bio to cover I/O against num_slots slots starting at
 510  * start_slot. */
 511 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
 512                                       struct o2hb_bio_wait_ctxt *wc,
 513                                       unsigned int *current_slot,
 514                                       unsigned int max_slots, int op,
 515                                       int op_flags)
 516 {
 517         int len, current_page;
 518         unsigned int vec_len, vec_start;
 519         unsigned int bits = reg->hr_block_bits;
 520         unsigned int spp = reg->hr_slots_per_page;
 521         unsigned int cs = *current_slot;
 522         struct bio *bio;
 523         struct page *page;
 524
 525         /* Testing has shown this allocation to take long enough under
 526          * GFP_KERNEL that the local node can get fenced. It would be
 527          * nicest if we could pre-allocate these bios and avoid this
 528          * all together. */
 529         bio = bio_alloc(GFP_ATOMIC, 16);
 530         if (!bio) {
 531                 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
 532                 bio = ERR_PTR(-ENOMEM);
 533                 goto bail;
 534         }
 535
 536         /* Must put everything in 512 byte sectors for the bio... */
 537         bio->bi_iter.bi_sector = (reg->hr_start_block + cs) << (bits - 9);
 538         bio_set_dev(bio, reg->hr_bdev);
 539         bio->bi_private = wc;
 540         bio->bi_end_io = o2hb_bio_end_io;
 541         bio_set_op_attrs(bio, op, op_flags);
 542
 543         vec_start = (cs << bits) % PAGE_SIZE;
 544         while(cs < max_slots) {
 545                 current_page = cs / spp;
 546                 page = reg->hr_slot_data[current_page];
 547
 548                 vec_len = min(PAGE_SIZE - vec_start,
 549                               (max_slots-cs) * (PAGE_SIZE/spp) );
 550
 551                 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
 552                      current_page, vec_len, vec_start);
 553
 554                 len = bio_add_page(bio, page, vec_len, vec_start);
 555                 if (len != vec_len) break;
 556
 557                 cs += vec_len / (PAGE_SIZE/spp);
 558                 vec_start = 0;
 559         }
 560
 561 bail:
 562         *current_slot = cs;
 563         return bio;
 564 }
 565
 566 static int o2hb_read_slots(struct o2hb_region *reg,
 567                            unsigned int begin_slot,
 568                            unsigned int max_slots)
 569 {
 570         unsigned int current_slot = begin_slot;
 571         int status;
 572         struct o2hb_bio_wait_ctxt wc;
 573         struct bio *bio;
 574
 575         o2hb_bio_wait_init(&wc);
 576
 577         while(current_slot < max_slots) {
 578                 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots,
 579                                          REQ_OP_READ, 0);
 580                 if (IS_ERR(bio)) {
 581                         status = PTR_ERR(bio);
 582                         mlog_errno(status);
 583                         goto bail_and_wait;
 584                 }
 585
 586                 atomic_inc(&wc.wc_num_reqs);
 587                 submit_bio(bio);
 588         }
 589
 590         status = 0;
 591
 592 bail_and_wait:
 593         o2hb_wait_on_io(&wc);
 594         if (wc.wc_error && !status)
 595                 status = wc.wc_error;
 596
 597         return status;
 598 }
 599
 600 static int o2hb_issue_node_write(struct o2hb_region *reg,
 601                                  struct o2hb_bio_wait_ctxt *write_wc)
 602 {
 603         int status;
 604         unsigned int slot;
 605         struct bio *bio;
 606
 607         o2hb_bio_wait_init(write_wc);
 608
 609         slot = o2nm_this_node();
 610
 611         bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1, REQ_OP_WRITE,
 612                                  REQ_SYNC);
 613         if (IS_ERR(bio)) {
 614                 status = PTR_ERR(bio);
 615                 mlog_errno(status);
 616                 goto bail;
 617         }
 618
 619         atomic_inc(&write_wc->wc_num_reqs);
 620         submit_bio(bio);
 621
 622         status = 0;
 623 bail:
 624         return status;
 625 }
 626
 627 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
 628                                      struct o2hb_disk_heartbeat_block *hb_block)
 629 {
 630         __le32 old_cksum;
 631         u32 ret;
 632
 633         /* We want to compute the block crc with a 0 value in the
 634          * hb_cksum field. Save it off here and replace after the
 635          * crc. */
 636         old_cksum = hb_block->hb_cksum;
 637         hb_block->hb_cksum = 0;
 638
 639         ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
 640
 641         hb_block->hb_cksum = old_cksum;
 642
 643         return ret;
 644 }
 645
 646 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
 647 {
 648         mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
 649              "cksum = 0x%x, generation 0x%llx\n",
 650              (long long)le64_to_cpu(hb_block->hb_seq),
 651              hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
 652              (long long)le64_to_cpu(hb_block->hb_generation));
 653 }
 654
 655 static int o2hb_verify_crc(struct o2hb_region *reg,
 656                            struct o2hb_disk_heartbeat_block *hb_block)
 657 {
 658         u32 read, computed;
 659
 660         read = le32_to_cpu(hb_block->hb_cksum);
 661         computed = o2hb_compute_block_crc_le(reg, hb_block);
 662
 663         return read == computed;
 664 }
 665
 666 /*
 667  * Compare the slot data with what we wrote in the last iteration.
 668  * If the match fails, print an appropriate error message. This is to
 669  * detect errors like... another node hearting on the same slot,
 670  * flaky device that is losing writes, etc.
 671  * Returns 1 if check succeeds, 0 otherwise.
 672  */
 673 static int o2hb_check_own_slot(struct o2hb_region *reg)
 674 {
 675         struct o2hb_disk_slot *slot;
 676         struct o2hb_disk_heartbeat_block *hb_block;
 677         char *errstr;
 678
 679         slot = &reg->hr_slots[o2nm_this_node()];
 680         /* Don't check on our 1st timestamp */
 681         if (!slot->ds_last_time)
 682                 return 0;
 683
 684         hb_block = slot->ds_raw_block;
 685         if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
 686             le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
 687             hb_block->hb_node == slot->ds_node_num)
 688                 return 1;
 689
 690 #define ERRSTR1         "Another node is heartbeating on device"
 691 #define ERRSTR2         "Heartbeat generation mismatch on device"
 692 #define ERRSTR3         "Heartbeat sequence mismatch on device"
 693
 694         if (hb_block->hb_node != slot->ds_node_num)
 695                 errstr = ERRSTR1;
 696         else if (le64_to_cpu(hb_block->hb_generation) !=
 697                  slot->ds_last_generation)
 698                 errstr = ERRSTR2;
 699         else
 700                 errstr = ERRSTR3;
 701
 702         mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
 703              "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
 704              slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
 705              (unsigned long long)slot->ds_last_time, hb_block->hb_node,
 706              (unsigned long long)le64_to_cpu(hb_block->hb_generation),
 707              (unsigned long long)le64_to_cpu(hb_block->hb_seq));
 708
 709         return 0;
 710 }
 711
 712 static inline void o2hb_prepare_block(struct o2hb_region *reg,
 713                                       u64 generation)
 714 {
 715         int node_num;
 716         u64 cputime;
 717         struct o2hb_disk_slot *slot;
 718         struct o2hb_disk_heartbeat_block *hb_block;
 719
 720         node_num = o2nm_this_node();
 721         slot = &reg->hr_slots[node_num];
 722
 723         hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
 724         memset(hb_block, 0, reg->hr_block_bytes);
 725         /* TODO: time stuff */
 726         cputime = ktime_get_real_seconds();
 727         if (!cputime)
 728                 cputime = 1;
 729
 730         hb_block->hb_seq = cpu_to_le64(cputime);
 731         hb_block->hb_node = node_num;
 732         hb_block->hb_generation = cpu_to_le64(generation);
 733         hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
 734
 735         /* This step must always happen last! */
 736         hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
 737                                                                    hb_block));
 738
 739         mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
 740              (long long)generation,
 741              le32_to_cpu(hb_block->hb_cksum));
 742 }
 743
 744 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
 745                                 struct o2nm_node *node,
 746                                 int idx)
 747 {
 748         struct o2hb_callback_func *f;
 749
 750         list_for_each_entry(f, &hbcall->list, hc_item) {
 751                 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
 752                 (f->hc_func)(node, idx, f->hc_data);
 753         }
 754 }
 755
 756 /* Will run the list in order until we process the passed event */
 757 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
 758 {
 759         struct o2hb_callback *hbcall;
 760         struct o2hb_node_event *event;
 761
 762         /* Holding callback sem assures we don't alter the callback
 763          * lists when doing this, and serializes ourselves with other
 764          * processes wanting callbacks. */
 765         down_write(&o2hb_callback_sem);
 766
 767         spin_lock(&o2hb_live_lock);
 768         while (!list_empty(&o2hb_node_events)
 769                && !list_empty(&queued_event->hn_item)) {
 770                 event = list_entry(o2hb_node_events.next,
 771                                    struct o2hb_node_event,
 772                                    hn_item);
 773                 list_del_init(&event->hn_item);
 774                 spin_unlock(&o2hb_live_lock);
 775
 776                 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
 777                      event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
 778                      event->hn_node_num);
 779
 780                 hbcall = hbcall_from_type(event->hn_event_type);
 781
 782                 /* We should *never* have gotten on to the list with a
 783                  * bad type... This isn't something that we should try
 784                  * to recover from. */
 785                 BUG_ON(IS_ERR(hbcall));
 786
 787                 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
 788
 789                 spin_lock(&o2hb_live_lock);
 790         }
 791         spin_unlock(&o2hb_live_lock);
 792
 793         up_write(&o2hb_callback_sem);
 794 }
 795
 796 static void o2hb_queue_node_event(struct o2hb_node_event *event,
 797                                   enum o2hb_callback_type type,
 798                                   struct o2nm_node *node,
 799                                   int node_num)
 800 {
 801         assert_spin_locked(&o2hb_live_lock);
 802
 803         BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
 804
 805         event->hn_event_type = type;
 806         event->hn_node = node;
 807         event->hn_node_num = node_num;
 808
 809         mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
 810              type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
 811
 812         list_add_tail(&event->hn_item, &o2hb_node_events);
 813 }
 814
 815 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
 816 {
 817         struct o2hb_node_event event =
 818                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 819         struct o2nm_node *node;
 820         int queued = 0;
 821
 822         node = o2nm_get_node_by_num(slot->ds_node_num);
 823         if (!node)
 824                 return;
 825
 826         spin_lock(&o2hb_live_lock);
 827         if (!list_empty(&slot->ds_live_item)) {
 828                 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
 829                      slot->ds_node_num);
 830
 831                 list_del_init(&slot->ds_live_item);
 832
 833                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 834                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 835
 836                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
 837                                               slot->ds_node_num);
 838                         queued = 1;
 839                 }
 840         }
 841         spin_unlock(&o2hb_live_lock);
 842
 843         if (queued)
 844                 o2hb_run_event_list(&event);
 845
 846         o2nm_node_put(node);
 847 }
 848
 849 static void o2hb_set_quorum_device(struct o2hb_region *reg)
 850 {
 851         if (!o2hb_global_heartbeat_active())
 852                 return;
 853
 854         /* Prevent race with o2hb_heartbeat_group_drop_item() */
 855         if (kthread_should_stop())
 856                 return;
 857
 858         /* Tag region as quorum only after thread reaches steady state */
 859         if (atomic_read(&reg->hr_steady_iterations) != 0)
 860                 return;
 861
 862         spin_lock(&o2hb_live_lock);
 863
 864         if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 865                 goto unlock;
 866
 867         /*
 868          * A region can be added to the quorum only when it sees all
 869          * live nodes heartbeat on it. In other words, the region has been
 870          * added to all nodes.
 871          */
 872         if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
 873                    sizeof(o2hb_live_node_bitmap)))
 874                 goto unlock;
 875
 876         printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
 877                config_item_name(&reg->hr_item), reg->hr_dev_name);
 878
 879         set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
 880
 881         /*
 882          * If global heartbeat active, unpin all regions if the
 883          * region count > CUT_OFF
 884          */
 885         if (bitmap_weight(o2hb_quorum_region_bitmap,
 886                            O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
 887                 o2hb_region_unpin(NULL);
 888 unlock:
 889         spin_unlock(&o2hb_live_lock);
 890 }
 891
 892 static int o2hb_check_slot(struct o2hb_region *reg,
 893                            struct o2hb_disk_slot *slot)
 894 {
 895         int changed = 0, gen_changed = 0;
 896         struct o2hb_node_event event =
 897                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 898         struct o2nm_node *node;
 899         struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
 900         u64 cputime;
 901         unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
 902         unsigned int slot_dead_ms;
 903         int tmp;
 904         int queued = 0;
 905
 906         memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
 907
 908         /*
 909          * If a node is no longer configured but is still in the livemap, we
 910          * may need to clear that bit from the livemap.
 911          */
 912         node = o2nm_get_node_by_num(slot->ds_node_num);
 913         if (!node) {
 914                 spin_lock(&o2hb_live_lock);
 915                 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 916                 spin_unlock(&o2hb_live_lock);
 917                 if (!tmp)
 918                         return 0;
 919         }
 920
 921         if (!o2hb_verify_crc(reg, hb_block)) {
 922                 /* all paths from here will drop o2hb_live_lock for
 923                  * us. */
 924                 spin_lock(&o2hb_live_lock);
 925
 926                 /* Don't print an error on the console in this case -
 927                  * a freshly formatted heartbeat area will not have a
 928                  * crc set on it. */
 929                 if (list_empty(&slot->ds_live_item))
 930                         goto out;
 931
 932                 /* The node is live but pushed out a bad crc. We
 933                  * consider it a transient miss but don't populate any
 934                  * other values as they may be junk. */
 935                 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
 936                      slot->ds_node_num, reg->hr_dev_name);
 937                 o2hb_dump_slot(hb_block);
 938
 939                 slot->ds_equal_samples++;
 940                 goto fire_callbacks;
 941         }
 942
 943         /* we don't care if these wrap.. the state transitions below
 944          * clear at the right places */
 945         cputime = le64_to_cpu(hb_block->hb_seq);
 946         if (slot->ds_last_time != cputime)
 947                 slot->ds_changed_samples++;
 948         else
 949                 slot->ds_equal_samples++;
 950         slot->ds_last_time = cputime;
 951
 952         /* The node changed heartbeat generations. We assume this to
 953          * mean it dropped off but came back before we timed out. We
 954          * want to consider it down for the time being but don't want
 955          * to lose any changed_samples state we might build up to
 956          * considering it live again. */
 957         if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
 958                 gen_changed = 1;
 959                 slot->ds_equal_samples = 0;
 960                 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
 961                      "to 0x%llx)\n", slot->ds_node_num,
 962                      (long long)slot->ds_last_generation,
 963                      (long long)le64_to_cpu(hb_block->hb_generation));
 964         }
 965
 966         slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
 967
 968         mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
 969              "seq %llu last %llu changed %u equal %u\n",
 970              slot->ds_node_num, (long long)slot->ds_last_generation,
 971              le32_to_cpu(hb_block->hb_cksum),
 972              (unsigned long long)le64_to_cpu(hb_block->hb_seq),
 973              (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
 974              slot->ds_equal_samples);
 975
 976         spin_lock(&o2hb_live_lock);
 977
 978 fire_callbacks:
 979         /* dead nodes only come to life after some number of
 980          * changes at any time during their dead time */
 981         if (list_empty(&slot->ds_live_item) &&
 982             slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
 983                 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
 984                      slot->ds_node_num, (long long)slot->ds_last_generation);
 985
 986                 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 987
 988                 /* first on the list generates a callback */
 989                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 990                         mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
 991                              "bitmap\n", slot->ds_node_num);
 992                         set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 993
 994                         o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
 995                                               slot->ds_node_num);
 996
 997                         changed = 1;
 998                         queued = 1;
 999                 }
1000
1001                 list_add_tail(&slot->ds_live_item,
1002                               &o2hb_live_slots[slot->ds_node_num]);
1003
1004                 slot->ds_equal_samples = 0;
1005
1006                 /* We want to be sure that all nodes agree on the
1007                  * number of milliseconds before a node will be
1008                  * considered dead. The self-fencing timeout is
1009                  * computed from this value, and a discrepancy might
1010                  * result in heartbeat calling a node dead when it
1011                  * hasn't self-fenced yet. */
1012                 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
1013                 if (slot_dead_ms && slot_dead_ms != dead_ms) {
1014                         /* TODO: Perhaps we can fail the region here. */
1015                         mlog(ML_ERROR, "Node %d on device %s has a dead count "
1016                              "of %u ms, but our count is %u ms.\n"
1017                              "Please double check your configuration values "
1018                              "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
1019                              slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
1020                              dead_ms);
1021                 }
1022                 goto out;
1023         }
1024
1025         /* if the list is dead, we're done.. */
1026         if (list_empty(&slot->ds_live_item))
1027                 goto out;
1028
1029         /* live nodes only go dead after enough consequtive missed
1030          * samples..  reset the missed counter whenever we see
1031          * activity */
1032         if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
1033                 mlog(ML_HEARTBEAT, "Node %d left my region\n",
1034                      slot->ds_node_num);
1035
1036                 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1037
1038                 /* last off the live_slot generates a callback */
1039                 list_del_init(&slot->ds_live_item);
1040                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
1041                         mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
1042                              "nodes bitmap\n", slot->ds_node_num);
1043                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
1044
1045                         /* node can be null */
1046                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
1047                                               node, slot->ds_node_num);
1048
1049                         changed = 1;
1050                         queued = 1;
1051                 }
1052
1053                 /* We don't clear this because the node is still
1054                  * actually writing new blocks. */
1055                 if (!gen_changed)
1056                         slot->ds_changed_samples = 0;
1057                 goto out;
1058         }
1059         if (slot->ds_changed_samples) {
1060                 slot->ds_changed_samples = 0;
1061                 slot->ds_equal_samples = 0;
1062         }
1063 out:
1064         spin_unlock(&o2hb_live_lock);
1065
1066         if (queued)
1067                 o2hb_run_event_list(&event);
1068
1069         if (node)
1070                 o2nm_node_put(node);
1071         return changed;
1072 }
1073
1074 static int o2hb_highest_node(unsigned long *nodes, int numbits)
1075 {
1076         return find_last_bit(nodes, numbits);
1077 }
1078
1079 static int o2hb_lowest_node(unsigned long *nodes, int numbits)
1080 {
1081         return find_first_bit(nodes, numbits);
1082 }
1083
1084 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
1085 {
1086         int i, ret, highest_node, lowest_node;
1087         int membership_change = 0, own_slot_ok = 0;
1088         unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
1089         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
1090         struct o2hb_bio_wait_ctxt write_wc;
1091
1092         ret = o2nm_configured_node_map(configured_nodes,
1093                                        sizeof(configured_nodes));
1094         if (ret) {
1095                 mlog_errno(ret);
1096                 goto bail;
1097         }
1098
1099         /*
1100          * If a node is not configured but is in the livemap, we still need
1101          * to read the slot so as to be able to remove it from the livemap.
1102          */
1103         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
1104         i = -1;
1105         while ((i = find_next_bit(live_node_bitmap,
1106                                   O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1107                 set_bit(i, configured_nodes);
1108         }
1109
1110         highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1111         lowest_node = o2hb_lowest_node(configured_nodes, O2NM_MAX_NODES);
1112         if (highest_node >= O2NM_MAX_NODES || lowest_node >= O2NM_MAX_NODES) {
1113                 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1114                 ret = -EINVAL;
1115                 goto bail;
1116         }
1117
1118         /* No sense in reading the slots of nodes that don't exist
1119          * yet. Of course, if the node definitions have holes in them
1120          * then we're reading an empty slot anyway... Consider this
1121          * best-effort. */
1122         ret = o2hb_read_slots(reg, lowest_node, highest_node + 1);
1123         if (ret < 0) {
1124                 mlog_errno(ret);
1125                 goto bail;
1126         }
1127
1128         /* With an up to date view of the slots, we can check that no
1129          * other node has been improperly configured to heartbeat in
1130          * our slot. */
1131         own_slot_ok = o2hb_check_own_slot(reg);
1132
1133         /* fill in the proper info for our next heartbeat */
1134         o2hb_prepare_block(reg, reg->hr_generation);
1135
1136         ret = o2hb_issue_node_write(reg, &write_wc);
1137         if (ret < 0) {
1138                 mlog_errno(ret);
1139                 goto bail;
1140         }
1141
1142         i = -1;
1143         while((i = find_next_bit(configured_nodes,
1144                                  O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1145                 membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1146         }
1147
1148         /*
1149          * We have to be sure we've advertised ourselves on disk
1150          * before we can go to steady state.  This ensures that
1151          * people we find in our steady state have seen us.
1152          */
1153         o2hb_wait_on_io(&write_wc);
1154         if (write_wc.wc_error) {
1155                 /* Do not re-arm the write timeout on I/O error - we
1156                  * can't be sure that the new block ever made it to
1157                  * disk */
1158                 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1159                      write_wc.wc_error, reg->hr_dev_name);
1160                 ret = write_wc.wc_error;
1161                 goto bail;
1162         }
1163
1164         /* Skip disarming the timeout if own slot has stale/bad data */
1165         if (own_slot_ok) {
1166                 o2hb_set_quorum_device(reg);
1167                 o2hb_arm_timeout(reg);
1168                 reg->hr_last_timeout_start = jiffies;
1169         }
1170
1171 bail:
1172         /* let the person who launched us know when things are steady */
1173         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1174                 if (!ret && own_slot_ok && !membership_change) {
1175                         if (atomic_dec_and_test(&reg->hr_steady_iterations))
1176                                 wake_up(&o2hb_steady_queue);
1177                 }
1178         }
1179
1180         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1181                 if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1182                         printk(KERN_NOTICE "o2hb: Unable to stabilize "
1183                                "heartbeat on region %s (%s)\n",
1184                                config_item_name(&reg->hr_item),
1185                                reg->hr_dev_name);
1186                         atomic_set(&reg->hr_steady_iterations, 0);
1187                         reg->hr_aborted_start = 1;
1188                         wake_up(&o2hb_steady_queue);
1189                         ret = -EIO;
1190                 }
1191         }
1192
1193         return ret;
1194 }
1195
1196 /*
1197  * we ride the region ref that the region dir holds.  before the region
1198  * dir is removed and drops it ref it will wait to tear down this
1199  * thread.
1200  */
1201 static int o2hb_thread(void *data)
1202 {
1203         int i, ret;
1204         struct o2hb_region *reg = data;
1205         struct o2hb_bio_wait_ctxt write_wc;
1206         ktime_t before_hb, after_hb;
1207         unsigned int elapsed_msec;
1208
1209         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1210
1211         set_user_nice(current, MIN_NICE);
1212
1213         /* Pin node */
1214         ret = o2nm_depend_this_node();
1215         if (ret) {
1216                 mlog(ML_ERROR, "Node has been deleted, ret = %d\n", ret);
1217                 reg->hr_node_deleted = 1;
1218                 wake_up(&o2hb_steady_queue);
1219                 return 0;
1220         }
1221
1222         while (!kthread_should_stop() &&
1223                !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1224                 /* We track the time spent inside
1225                  * o2hb_do_disk_heartbeat so that we avoid more than
1226                  * hr_timeout_ms between disk writes. On busy systems
1227                  * this should result in a heartbeat which is less
1228                  * likely to time itself out. */
1229                 before_hb = ktime_get_real();
1230
1231                 ret = o2hb_do_disk_heartbeat(reg);
1232                 reg->hr_last_hb_status = ret;
1233
1234                 after_hb = ktime_get_real();
1235
1236                 elapsed_msec = (unsigned int)
1237                                 ktime_ms_delta(after_hb, before_hb);
1238
1239                 mlog(ML_HEARTBEAT,
1240                      "start = %lld, end = %lld, msec = %u, ret = %d\n",
1241                      before_hb, after_hb, elapsed_msec, ret);
1242
1243                 if (!kthread_should_stop() &&
1244                     elapsed_msec < reg->hr_timeout_ms) {
1245                         /* the kthread api has blocked signals for us so no
1246                          * need to record the return value. */
1247                         msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1248                 }
1249         }
1250
1251         o2hb_disarm_timeout(reg);
1252
1253         /* unclean stop is only used in very bad situation */
1254         for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1255                 o2hb_shutdown_slot(&reg->hr_slots[i]);
1256
1257         /* Explicit down notification - avoid forcing the other nodes
1258          * to timeout on this region when we could just as easily
1259          * write a clear generation - thus indicating to them that
1260          * this node has left this region.
1261          */
1262         if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1263                 o2hb_prepare_block(reg, 0);
1264                 ret = o2hb_issue_node_write(reg, &write_wc);
1265                 if (ret == 0)
1266                         o2hb_wait_on_io(&write_wc);
1267                 else
1268                         mlog_errno(ret);
1269         }
1270
1271         /* Unpin node */
1272         o2nm_undepend_this_node();
1273
1274         mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1275
1276         return 0;
1277 }
1278
1279 #ifdef CONFIG_DEBUG_FS
1280 static int o2hb_debug_open(struct inode *inode, struct file *file)
1281 {
1282         struct o2hb_debug_buf *db = inode->i_private;
1283         struct o2hb_region *reg;
1284         unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1285         unsigned long lts;
1286         char *buf = NULL;
1287         int i = -1;
1288         int out = 0;
1289
1290         /* max_nodes should be the largest bitmap we pass here */
1291         BUG_ON(sizeof(map) < db->db_size);
1292
1293         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1294         if (!buf)
1295                 goto bail;
1296
1297         switch (db->db_type) {
1298         case O2HB_DB_TYPE_LIVENODES:
1299         case O2HB_DB_TYPE_LIVEREGIONS:
1300         case O2HB_DB_TYPE_QUORUMREGIONS:
1301         case O2HB_DB_TYPE_FAILEDREGIONS:
1302                 spin_lock(&o2hb_live_lock);
1303                 memcpy(map, db->db_data, db->db_size);
1304                 spin_unlock(&o2hb_live_lock);
1305                 break;
1306
1307         case O2HB_DB_TYPE_REGION_LIVENODES:
1308                 spin_lock(&o2hb_live_lock);
1309                 reg = (struct o2hb_region *)db->db_data;
1310                 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1311                 spin_unlock(&o2hb_live_lock);
1312                 break;
1313
1314         case O2HB_DB_TYPE_REGION_NUMBER:
1315                 reg = (struct o2hb_region *)db->db_data;
1316                 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1317                                 reg->hr_region_num);
1318                 goto done;
1319
1320         case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1321                 reg = (struct o2hb_region *)db->db_data;
1322                 lts = reg->hr_last_timeout_start;
1323                 /* If 0, it has never been set before */
1324                 if (lts)
1325                         lts = jiffies_to_msecs(jiffies - lts);
1326                 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1327                 goto done;
1328
1329         case O2HB_DB_TYPE_REGION_PINNED:
1330                 reg = (struct o2hb_region *)db->db_data;
1331                 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1332                                 !!reg->hr_item_pinned);
1333                 goto done;
1334
1335         default:
1336                 goto done;
1337         }
1338
1339         while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1340                 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1341         out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1342
1343 done:
1344         i_size_write(inode, out);
1345
1346         file->private_data = buf;
1347
1348         return 0;
1349 bail:
1350         return -ENOMEM;
1351 }
1352
1353 static int o2hb_debug_release(struct inode *inode, struct file *file)
1354 {
1355         kfree(file->private_data);
1356         return 0;
1357 }
1358
1359 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1360                                  size_t nbytes, loff_t *ppos)
1361 {
1362         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1363                                        i_size_read(file->f_mapping->host));
1364 }
1365 #else
1366 static int o2hb_debug_open(struct inode *inode, struct file *file)
1367 {
1368         return 0;
1369 }
1370 static int o2hb_debug_release(struct inode *inode, struct file *file)
1371 {
1372         return 0;
1373 }
1374 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1375                                size_t nbytes, loff_t *ppos)
1376 {
1377         return 0;
1378 }
1379 #endif  /* CONFIG_DEBUG_FS */
1380
1381 static const struct file_operations o2hb_debug_fops = {
1382         .open =         o2hb_debug_open,
1383         .release =      o2hb_debug_release,
1384         .read =         o2hb_debug_read,
1385         .llseek =       generic_file_llseek,
1386 };
1387
1388 void o2hb_exit(void)
1389 {
1390         debugfs_remove_recursive(o2hb_debug_dir);
1391         kfree(o2hb_db_livenodes);
1392         kfree(o2hb_db_liveregions);
1393         kfree(o2hb_db_quorumregions);
1394         kfree(o2hb_db_failedregions);
1395 }
1396
1397 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1398                                         struct o2hb_debug_buf **db, int db_len,
1399                                         int type, int size, int len, void *data)
1400 {
1401         *db = kmalloc(db_len, GFP_KERNEL);
1402         if (!*db)
1403                 return NULL;
1404
1405         (*db)->db_type = type;
1406         (*db)->db_size = size;
1407         (*db)->db_len = len;
1408         (*db)->db_data = data;
1409
1410         return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1411                                    &o2hb_debug_fops);
1412 }
1413
1414 static void o2hb_debug_init(void)
1415 {
1416         o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1417
1418         o2hb_debug_create(O2HB_DEBUG_LIVENODES, o2hb_debug_dir,
1419                           &o2hb_db_livenodes, sizeof(*o2hb_db_livenodes),
1420                           O2HB_DB_TYPE_LIVENODES, sizeof(o2hb_live_node_bitmap),
1421                           O2NM_MAX_NODES, o2hb_live_node_bitmap);
1422
1423         o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS, o2hb_debug_dir,
1424                           &o2hb_db_liveregions, sizeof(*o2hb_db_liveregions),
1425                           O2HB_DB_TYPE_LIVEREGIONS,
1426                           sizeof(o2hb_live_region_bitmap), O2NM_MAX_REGIONS,
1427                           o2hb_live_region_bitmap);
1428
1429         o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS, o2hb_debug_dir,
1430                           &o2hb_db_quorumregions,
1431                           sizeof(*o2hb_db_quorumregions),
1432                           O2HB_DB_TYPE_QUORUMREGIONS,
1433                           sizeof(o2hb_quorum_region_bitmap), O2NM_MAX_REGIONS,
1434                           o2hb_quorum_region_bitmap);
1435
1436         o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS, o2hb_debug_dir,
1437                           &o2hb_db_failedregions,
1438                           sizeof(*o2hb_db_failedregions),
1439                           O2HB_DB_TYPE_FAILEDREGIONS,
1440                           sizeof(o2hb_failed_region_bitmap), O2NM_MAX_REGIONS,
1441                           o2hb_failed_region_bitmap);
1442 }
1443
1444 void o2hb_init(void)
1445 {
1446         int i;
1447
1448         for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1449                 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1450
1451         for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1452                 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1453
1454         INIT_LIST_HEAD(&o2hb_node_events);
1455
1456         memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1457         memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1458         memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1459         memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1460         memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1461
1462         o2hb_dependent_users = 0;
1463
1464         o2hb_debug_init();
1465 }
1466
1467 /* if we're already in a callback then we're already serialized by the sem */
1468 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1469                                              unsigned bytes)
1470 {
1471         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1472
1473         memcpy(map, &o2hb_live_node_bitmap, bytes);
1474 }
1475
1476 /*
1477  * get a map of all nodes that are heartbeating in any regions
1478  */
1479 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1480 {
1481         /* callers want to serialize this map and callbacks so that they
1482          * can trust that they don't miss nodes coming to the party */
1483         down_read(&o2hb_callback_sem);
1484         spin_lock(&o2hb_live_lock);
1485         o2hb_fill_node_map_from_callback(map, bytes);
1486         spin_unlock(&o2hb_live_lock);
1487         up_read(&o2hb_callback_sem);
1488 }
1489 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1490
1491 /*
1492  * heartbeat configfs bits.  The heartbeat set is a default set under
1493  * the cluster set in nodemanager.c.
1494  */
1495
1496 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1497 {
1498         return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1499 }
1500
1501 /* drop_item only drops its ref after killing the thread, nothing should
1502  * be using the region anymore.  this has to clean up any state that
1503  * attributes might have built up. */
1504 static void o2hb_region_release(struct config_item *item)
1505 {
1506         int i;
1507         struct page *page;
1508         struct o2hb_region *reg = to_o2hb_region(item);
1509
1510         mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1511
1512         kfree(reg->hr_tmp_block);
1513
1514         if (reg->hr_slot_data) {
1515                 for (i = 0; i < reg->hr_num_pages; i++) {
1516                         page = reg->hr_slot_data[i];
1517                         if (page)
1518                                 __free_page(page);
1519                 }
1520                 kfree(reg->hr_slot_data);
1521         }
1522
1523         if (reg->hr_bdev)
1524                 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1525
1526         kfree(reg->hr_slots);
1527
1528         debugfs_remove(reg->hr_debug_livenodes);
1529         debugfs_remove(reg->hr_debug_regnum);
1530         debugfs_remove(reg->hr_debug_elapsed_time);
1531         debugfs_remove(reg->hr_debug_pinned);
1532         debugfs_remove(reg->hr_debug_dir);
1533         kfree(reg->hr_db_livenodes);
1534         kfree(reg->hr_db_regnum);
1535         kfree(reg->hr_db_elapsed_time);
1536         kfree(reg->hr_db_pinned);
1537
1538         spin_lock(&o2hb_live_lock);
1539         list_del(&reg->hr_all_item);
1540         spin_unlock(&o2hb_live_lock);
1541
1542         o2net_unregister_handler_list(&reg->hr_handler_list);
1543         kfree(reg);
1544 }
1545
1546 static int o2hb_read_block_input(struct o2hb_region *reg,
1547                                  const char *page,
1548                                  unsigned long *ret_bytes,
1549                                  unsigned int *ret_bits)
1550 {
1551         unsigned long bytes;
1552         char *p = (char *)page;
1553
1554         bytes = simple_strtoul(p, &p, 0);
1555         if (!p || (*p && (*p != '\n')))
1556                 return -EINVAL;
1557
1558         /* Heartbeat and fs min / max block sizes are the same. */
1559         if (bytes > 4096 || bytes < 512)
1560                 return -ERANGE;
1561         if (hweight16(bytes) != 1)
1562                 return -EINVAL;
1563
1564         if (ret_bytes)
1565                 *ret_bytes = bytes;
1566         if (ret_bits)
1567                 *ret_bits = ffs(bytes) - 1;
1568
1569         return 0;
1570 }
1571
1572 static ssize_t o2hb_region_block_bytes_show(struct config_item *item,
1573                                             char *page)
1574 {
1575         return sprintf(page, "%u\n", to_o2hb_region(item)->hr_block_bytes);
1576 }
1577
1578 static ssize_t o2hb_region_block_bytes_store(struct config_item *item,
1579                                              const char *page,
1580                                              size_t count)
1581 {
1582         struct o2hb_region *reg = to_o2hb_region(item);
1583         int status;
1584         unsigned long block_bytes;
1585         unsigned int block_bits;
1586
1587         if (reg->hr_bdev)
1588                 return -EINVAL;
1589
1590         status = o2hb_read_block_input(reg, page, &block_bytes,
1591                                        &block_bits);
1592         if (status)
1593                 return status;
1594
1595         reg->hr_block_bytes = (unsigned int)block_bytes;
1596         reg->hr_block_bits = block_bits;
1597
1598         return count;
1599 }
1600
1601 static ssize_t o2hb_region_start_block_show(struct config_item *item,
1602                                             char *page)
1603 {
1604         return sprintf(page, "%llu\n", to_o2hb_region(item)->hr_start_block);
1605 }
1606
1607 static ssize_t o2hb_region_start_block_store(struct config_item *item,
1608                                              const char *page,
1609                                              size_t count)
1610 {
1611         struct o2hb_region *reg = to_o2hb_region(item);
1612         unsigned long long tmp;
1613         char *p = (char *)page;
1614
1615         if (reg->hr_bdev)
1616                 return -EINVAL;
1617
1618         tmp = simple_strtoull(p, &p, 0);
1619         if (!p || (*p && (*p != '\n')))
1620                 return -EINVAL;
1621
1622         reg->hr_start_block = tmp;
1623
1624         return count;
1625 }
1626
1627 static ssize_t o2hb_region_blocks_show(struct config_item *item, char *page)
1628 {
1629         return sprintf(page, "%d\n", to_o2hb_region(item)->hr_blocks);
1630 }
1631
1632 static ssize_t o2hb_region_blocks_store(struct config_item *item,
1633                                         const char *page,
1634                                         size_t count)
1635 {
1636         struct o2hb_region *reg = to_o2hb_region(item);
1637         unsigned long tmp;
1638         char *p = (char *)page;
1639
1640         if (reg->hr_bdev)
1641                 return -EINVAL;
1642
1643         tmp = simple_strtoul(p, &p, 0);
1644         if (!p || (*p && (*p != '\n')))
1645                 return -EINVAL;
1646
1647         if (tmp > O2NM_MAX_NODES || tmp == 0)
1648                 return -ERANGE;
1649
1650         reg->hr_blocks = (unsigned int)tmp;
1651
1652         return count;
1653 }
1654
1655 static ssize_t o2hb_region_dev_show(struct config_item *item, char *page)
1656 {
1657         unsigned int ret = 0;
1658
1659         if (to_o2hb_region(item)->hr_bdev)
1660                 ret = sprintf(page, "%s\n", to_o2hb_region(item)->hr_dev_name);
1661
1662         return ret;
1663 }
1664
1665 static void o2hb_init_region_params(struct o2hb_region *reg)
1666 {
1667         reg->hr_slots_per_page = PAGE_SIZE >> reg->hr_block_bits;
1668         reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1669
1670         mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1671              reg->hr_start_block, reg->hr_blocks);
1672         mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1673              reg->hr_block_bytes, reg->hr_block_bits);
1674         mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1675         mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1676 }
1677
1678 static int o2hb_map_slot_data(struct o2hb_region *reg)
1679 {
1680         int i, j;
1681         unsigned int last_slot;
1682         unsigned int spp = reg->hr_slots_per_page;
1683         struct page *page;
1684         char *raw;
1685         struct o2hb_disk_slot *slot;
1686
1687         reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1688         if (reg->hr_tmp_block == NULL)
1689                 return -ENOMEM;
1690
1691         reg->hr_slots = kcalloc(reg->hr_blocks,
1692                                 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1693         if (reg->hr_slots == NULL)
1694                 return -ENOMEM;
1695
1696         for(i = 0; i < reg->hr_blocks; i++) {
1697                 slot = &reg->hr_slots[i];
1698                 slot->ds_node_num = i;
1699                 INIT_LIST_HEAD(&slot->ds_live_item);
1700                 slot->ds_raw_block = NULL;
1701         }
1702
1703         reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1704         mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1705                            "at %u blocks per page\n",
1706              reg->hr_num_pages, reg->hr_blocks, spp);
1707
1708         reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1709                                     GFP_KERNEL);
1710         if (!reg->hr_slot_data)
1711                 return -ENOMEM;
1712
1713         for(i = 0; i < reg->hr_num_pages; i++) {
1714                 page = alloc_page(GFP_KERNEL);
1715                 if (!page)
1716                         return -ENOMEM;
1717
1718                 reg->hr_slot_data[i] = page;
1719
1720                 last_slot = i * spp;
1721                 raw = page_address(page);
1722                 for (j = 0;
1723                      (j < spp) && ((j + last_slot) < reg->hr_blocks);
1724                      j++) {
1725                         BUG_ON((j + last_slot) >= reg->hr_blocks);
1726
1727                         slot = &reg->hr_slots[j + last_slot];
1728                         slot->ds_raw_block =
1729                                 (struct o2hb_disk_heartbeat_block *) raw;
1730
1731                         raw += reg->hr_block_bytes;
1732                 }
1733         }
1734
1735         return 0;
1736 }
1737
1738 /* Read in all the slots available and populate the tracking
1739  * structures so that we can start with a baseline idea of what's
1740  * there. */
1741 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1742 {
1743         int ret, i;
1744         struct o2hb_disk_slot *slot;
1745         struct o2hb_disk_heartbeat_block *hb_block;
1746
1747         ret = o2hb_read_slots(reg, 0, reg->hr_blocks);
1748         if (ret)
1749                 goto out;
1750
1751         /* We only want to get an idea of the values initially in each
1752          * slot, so we do no verification - o2hb_check_slot will
1753          * actually determine if each configured slot is valid and
1754          * whether any values have changed. */
1755         for(i = 0; i < reg->hr_blocks; i++) {
1756                 slot = &reg->hr_slots[i];
1757                 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1758
1759                 /* Only fill the values that o2hb_check_slot uses to
1760                  * determine changing slots */
1761                 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1762                 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1763         }
1764
1765 out:
1766         return ret;
1767 }
1768
1769 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1770 static ssize_t o2hb_region_dev_store(struct config_item *item,
1771                                      const char *page,
1772                                      size_t count)
1773 {
1774         struct o2hb_region *reg = to_o2hb_region(item);
1775         struct task_struct *hb_task;
1776         long fd;
1777         int sectsize;
1778         char *p = (char *)page;
1779         struct fd f;
1780         struct inode *inode;
1781         ssize_t ret = -EINVAL;
1782         int live_threshold;
1783
1784         if (reg->hr_bdev)
1785                 goto out;
1786
1787         /* We can't heartbeat without having had our node number
1788          * configured yet. */
1789         if (o2nm_this_node() == O2NM_MAX_NODES)
1790                 goto out;
1791
1792         fd = simple_strtol(p, &p, 0);
1793         if (!p || (*p && (*p != '\n')))
1794                 goto out;
1795
1796         if (fd < 0 || fd >= INT_MAX)
1797                 goto out;
1798
1799         f = fdget(fd);
1800         if (f.file == NULL)
1801                 goto out;
1802
1803         if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1804             reg->hr_block_bytes == 0)
1805                 goto out2;
1806
1807         inode = igrab(f.file->f_mapping->host);
1808         if (inode == NULL)
1809                 goto out2;
1810
1811         if (!S_ISBLK(inode->i_mode))
1812                 goto out3;
1813
1814         reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
1815         ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
1816         if (ret) {
1817                 reg->hr_bdev = NULL;
1818                 goto out3;
1819         }
1820         inode = NULL;
1821
1822         bdevname(reg->hr_bdev, reg->hr_dev_name);
1823
1824         sectsize = bdev_logical_block_size(reg->hr_bdev);
1825         if (sectsize != reg->hr_block_bytes) {
1826                 mlog(ML_ERROR,
1827                      "blocksize %u incorrect for device, expected %d",
1828                      reg->hr_block_bytes, sectsize);
1829                 ret = -EINVAL;
1830                 goto out3;
1831         }
1832
1833         o2hb_init_region_params(reg);
1834
1835         /* Generation of zero is invalid */
1836         do {
1837                 get_random_bytes(&reg->hr_generation,
1838                                  sizeof(reg->hr_generation));
1839         } while (reg->hr_generation == 0);
1840
1841         ret = o2hb_map_slot_data(reg);
1842         if (ret) {
1843                 mlog_errno(ret);
1844                 goto out3;
1845         }
1846
1847         ret = o2hb_populate_slot_data(reg);
1848         if (ret) {
1849                 mlog_errno(ret);
1850                 goto out3;
1851         }
1852
1853         INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1854         INIT_DELAYED_WORK(&reg->hr_nego_timeout_work, o2hb_nego_timeout);
1855
1856         /*
1857          * A node is considered live after it has beat LIVE_THRESHOLD
1858          * times.  We're not steady until we've given them a chance
1859          * _after_ our first read.
1860          * The default threshold is bare minimum so as to limit the delay
1861          * during mounts. For global heartbeat, the threshold doubled for the
1862          * first region.
1863          */
1864         live_threshold = O2HB_LIVE_THRESHOLD;
1865         if (o2hb_global_heartbeat_active()) {
1866                 spin_lock(&o2hb_live_lock);
1867                 if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1868                         live_threshold <<= 1;
1869                 spin_unlock(&o2hb_live_lock);
1870         }
1871         ++live_threshold;
1872         atomic_set(&reg->hr_steady_iterations, live_threshold);
1873         /* unsteady_iterations is triple the steady_iterations */
1874         atomic_set(&reg->hr_unsteady_iterations, (live_threshold * 3));
1875
1876         hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1877                               reg->hr_item.ci_name);
1878         if (IS_ERR(hb_task)) {
1879                 ret = PTR_ERR(hb_task);
1880                 mlog_errno(ret);
1881                 goto out3;
1882         }
1883
1884         spin_lock(&o2hb_live_lock);
1885         reg->hr_task = hb_task;
1886         spin_unlock(&o2hb_live_lock);
1887
1888         ret = wait_event_interruptible(o2hb_steady_queue,
1889                                 atomic_read(&reg->hr_steady_iterations) == 0 ||
1890                                 reg->hr_node_deleted);
1891         if (ret) {
1892                 atomic_set(&reg->hr_steady_iterations, 0);
1893                 reg->hr_aborted_start = 1;
1894         }
1895
1896         if (reg->hr_aborted_start) {
1897                 ret = -EIO;
1898                 goto out3;
1899         }
1900
1901         if (reg->hr_node_deleted) {
1902                 ret = -EINVAL;
1903                 goto out3;
1904         }
1905
1906         /* Ok, we were woken.  Make sure it wasn't by drop_item() */
1907         spin_lock(&o2hb_live_lock);
1908         hb_task = reg->hr_task;
1909         if (o2hb_global_heartbeat_active())
1910                 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1911         spin_unlock(&o2hb_live_lock);
1912
1913         if (hb_task)
1914                 ret = count;
1915         else
1916                 ret = -EIO;
1917
1918         if (hb_task && o2hb_global_heartbeat_active())
1919                 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1920                        config_item_name(&reg->hr_item), reg->hr_dev_name);
1921
1922 out3:
1923         iput(inode);
1924 out2:
1925         fdput(f);
1926 out:
1927         if (ret < 0) {
1928                 if (reg->hr_bdev) {
1929                         blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1930                         reg->hr_bdev = NULL;
1931                 }
1932         }
1933         return ret;
1934 }
1935
1936 static ssize_t o2hb_region_pid_show(struct config_item *item, char *page)
1937 {
1938         struct o2hb_region *reg = to_o2hb_region(item);
1939         pid_t pid = 0;
1940
1941         spin_lock(&o2hb_live_lock);
1942         if (reg->hr_task)
1943                 pid = task_pid_nr(reg->hr_task);
1944         spin_unlock(&o2hb_live_lock);
1945
1946         if (!pid)
1947                 return 0;
1948
1949         return sprintf(page, "%u\n", pid);
1950 }
1951
1952 CONFIGFS_ATTR(o2hb_region_, block_bytes);
1953 CONFIGFS_ATTR(o2hb_region_, start_block);
1954 CONFIGFS_ATTR(o2hb_region_, blocks);
1955 CONFIGFS_ATTR(o2hb_region_, dev);
1956 CONFIGFS_ATTR_RO(o2hb_region_, pid);
1957
1958 static struct configfs_attribute *o2hb_region_attrs[] = {
1959         &o2hb_region_attr_block_bytes,
1960         &o2hb_region_attr_start_block,
1961         &o2hb_region_attr_blocks,
1962         &o2hb_region_attr_dev,
1963         &o2hb_region_attr_pid,
1964         NULL,
1965 };
1966
1967 static struct configfs_item_operations o2hb_region_item_ops = {
1968         .release                = o2hb_region_release,
1969 };
1970
1971 static const struct config_item_type o2hb_region_type = {
1972         .ct_item_ops    = &o2hb_region_item_ops,
1973         .ct_attrs       = o2hb_region_attrs,
1974         .ct_owner       = THIS_MODULE,
1975 };
1976
1977 /* heartbeat set */
1978
1979 struct o2hb_heartbeat_group {
1980         struct config_group hs_group;
1981         /* some stuff? */
1982 };
1983
1984 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1985 {
1986         return group ?
1987                 container_of(group, struct o2hb_heartbeat_group, hs_group)
1988                 : NULL;
1989 }
1990
1991 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
1992 {
1993         int ret = -ENOMEM;
1994
1995         reg->hr_debug_dir =
1996                 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
1997         if (!reg->hr_debug_dir) {
1998                 mlog_errno(ret);
1999                 goto bail;
2000         }
2001
2002         reg->hr_debug_livenodes =
2003                         o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2004                                           reg->hr_debug_dir,
2005                                           &(reg->hr_db_livenodes),
2006                                           sizeof(*(reg->hr_db_livenodes)),
2007                                           O2HB_DB_TYPE_REGION_LIVENODES,
2008                                           sizeof(reg->hr_live_node_bitmap),
2009                                           O2NM_MAX_NODES, reg);
2010         if (!reg->hr_debug_livenodes) {
2011                 mlog_errno(ret);
2012                 goto bail;
2013         }
2014
2015         reg->hr_debug_regnum =
2016                         o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2017                                           reg->hr_debug_dir,
2018                                           &(reg->hr_db_regnum),
2019                                           sizeof(*(reg->hr_db_regnum)),
2020                                           O2HB_DB_TYPE_REGION_NUMBER,
2021                                           0, O2NM_MAX_NODES, reg);
2022         if (!reg->hr_debug_regnum) {
2023                 mlog_errno(ret);
2024                 goto bail;
2025         }
2026
2027         reg->hr_debug_elapsed_time =
2028                         o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2029                                           reg->hr_debug_dir,
2030                                           &(reg->hr_db_elapsed_time),
2031                                           sizeof(*(reg->hr_db_elapsed_time)),
2032                                           O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2033                                           0, 0, reg);
2034         if (!reg->hr_debug_elapsed_time) {
2035                 mlog_errno(ret);
2036                 goto bail;
2037         }
2038
2039         reg->hr_debug_pinned =
2040                         o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2041                                           reg->hr_debug_dir,
2042                                           &(reg->hr_db_pinned),
2043                                           sizeof(*(reg->hr_db_pinned)),
2044                                           O2HB_DB_TYPE_REGION_PINNED,
2045                                           0, 0, reg);
2046         if (!reg->hr_debug_pinned) {
2047                 mlog_errno(ret);
2048                 goto bail;
2049         }
2050
2051         ret = 0;
2052 bail:
2053         return ret;
2054 }
2055
2056 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2057                                                           const char *name)
2058 {
2059         struct o2hb_region *reg = NULL;
2060         int ret;
2061
2062         reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2063         if (reg == NULL)
2064                 return ERR_PTR(-ENOMEM);
2065
2066         if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2067                 ret = -ENAMETOOLONG;
2068                 goto free;
2069         }
2070
2071         spin_lock(&o2hb_live_lock);
2072         reg->hr_region_num = 0;
2073         if (o2hb_global_heartbeat_active()) {
2074                 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2075                                                          O2NM_MAX_REGIONS);
2076                 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2077                         spin_unlock(&o2hb_live_lock);
2078                         ret = -EFBIG;
2079                         goto free;
2080                 }
2081                 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2082         }
2083         list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2084         spin_unlock(&o2hb_live_lock);
2085
2086         config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2087
2088         /* this is the same way to generate msg key as dlm, for local heartbeat,
2089          * name is also the same, so make initial crc value different to avoid
2090          * message key conflict.
2091          */
2092         reg->hr_key = crc32_le(reg->hr_region_num + O2NM_MAX_REGIONS,
2093                 name, strlen(name));
2094         INIT_LIST_HEAD(&reg->hr_handler_list);
2095         ret = o2net_register_handler(O2HB_NEGO_TIMEOUT_MSG, reg->hr_key,
2096                         sizeof(struct o2hb_nego_msg),
2097                         o2hb_nego_timeout_handler,
2098                         reg, NULL, &reg->hr_handler_list);
2099         if (ret)
2100                 goto free;
2101
2102         ret = o2net_register_handler(O2HB_NEGO_APPROVE_MSG, reg->hr_key,
2103                         sizeof(struct o2hb_nego_msg),
2104                         o2hb_nego_approve_handler,
2105                         reg, NULL, &reg->hr_handler_list);
2106         if (ret)
2107                 goto unregister_handler;
2108
2109         ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2110         if (ret) {
2111                 config_item_put(&reg->hr_item);
2112                 goto unregister_handler;
2113         }
2114
2115         return &reg->hr_item;
2116
2117 unregister_handler:
2118         o2net_unregister_handler_list(&reg->hr_handler_list);
2119 free:
2120         kfree(reg);
2121         return ERR_PTR(ret);
2122 }
2123
2124 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2125                                            struct config_item *item)
2126 {
2127         struct task_struct *hb_task;
2128         struct o2hb_region *reg = to_o2hb_region(item);
2129         int quorum_region = 0;
2130
2131         /* stop the thread when the user removes the region dir */
2132         spin_lock(&o2hb_live_lock);
2133         hb_task = reg->hr_task;
2134         reg->hr_task = NULL;
2135         reg->hr_item_dropped = 1;
2136         spin_unlock(&o2hb_live_lock);
2137
2138         if (hb_task)
2139                 kthread_stop(hb_task);
2140
2141         if (o2hb_global_heartbeat_active()) {
2142                 spin_lock(&o2hb_live_lock);
2143                 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2144                 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2145                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2146                         quorum_region = 1;
2147                 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2148                 spin_unlock(&o2hb_live_lock);
2149                 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2150                        ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2151                         "stopped" : "start aborted"), config_item_name(item),
2152                        reg->hr_dev_name);
2153         }
2154
2155         /*
2156          * If we're racing a dev_write(), we need to wake them.  They will
2157          * check reg->hr_task
2158          */
2159         if (atomic_read(&reg->hr_steady_iterations) != 0) {
2160                 reg->hr_aborted_start = 1;
2161                 atomic_set(&reg->hr_steady_iterations, 0);
2162                 wake_up(&o2hb_steady_queue);
2163         }
2164
2165         config_item_put(item);
2166
2167         if (!o2hb_global_heartbeat_active() || !quorum_region)
2168                 return;
2169
2170         /*
2171          * If global heartbeat active and there are dependent users,
2172          * pin all regions if quorum region count <= CUT_OFF
2173          */
2174         spin_lock(&o2hb_live_lock);
2175
2176         if (!o2hb_dependent_users)
2177                 goto unlock;
2178
2179         if (bitmap_weight(o2hb_quorum_region_bitmap,
2180                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2181                 o2hb_region_pin(NULL);
2182
2183 unlock:
2184         spin_unlock(&o2hb_live_lock);
2185 }
2186
2187 static ssize_t o2hb_heartbeat_group_dead_threshold_show(struct config_item *item,
2188                 char *page)
2189 {
2190         return sprintf(page, "%u\n", o2hb_dead_threshold);
2191 }
2192
2193 static ssize_t o2hb_heartbeat_group_dead_threshold_store(struct config_item *item,
2194                 const char *page, size_t count)
2195 {
2196         unsigned long tmp;
2197         char *p = (char *)page;
2198
2199         tmp = simple_strtoul(p, &p, 10);
2200         if (!p || (*p && (*p != '\n')))
2201                 return -EINVAL;
2202
2203         /* this will validate ranges for us. */
2204         o2hb_dead_threshold_set((unsigned int) tmp);
2205
2206         return count;
2207 }
2208
2209 static ssize_t o2hb_heartbeat_group_mode_show(struct config_item *item,
2210                 char *page)
2211 {
2212         return sprintf(page, "%s\n",
2213                        o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2214 }
2215
2216 static ssize_t o2hb_heartbeat_group_mode_store(struct config_item *item,
2217                 const char *page, size_t count)
2218 {
2219         unsigned int i;
2220         int ret;
2221         size_t len;
2222
2223         len = (page[count - 1] == '\n') ? count - 1 : count;
2224         if (!len)
2225                 return -EINVAL;
2226
2227         for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2228                 if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
2229                         continue;
2230
2231                 ret = o2hb_global_heartbeat_mode_set(i);
2232                 if (!ret)
2233                         printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2234                                o2hb_heartbeat_mode_desc[i]);
2235                 return count;
2236         }
2237
2238         return -EINVAL;
2239
2240 }
2241
2242 CONFIGFS_ATTR(o2hb_heartbeat_group_, dead_threshold);
2243 CONFIGFS_ATTR(o2hb_heartbeat_group_, mode);
2244
2245 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2246         &o2hb_heartbeat_group_attr_dead_threshold,
2247         &o2hb_heartbeat_group_attr_mode,
2248         NULL,
2249 };
2250
2251 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2252         .make_item      = o2hb_heartbeat_group_make_item,
2253         .drop_item      = o2hb_heartbeat_group_drop_item,
2254 };
2255
2256 static const struct config_item_type o2hb_heartbeat_group_type = {
2257         .ct_group_ops   = &o2hb_heartbeat_group_group_ops,
2258         .ct_attrs       = o2hb_heartbeat_group_attrs,
2259         .ct_owner       = THIS_MODULE,
2260 };
2261
2262 /* this is just here to avoid touching group in heartbeat.h which the
2263  * entire damn world #includes */
2264 struct config_group *o2hb_alloc_hb_set(void)
2265 {
2266         struct o2hb_heartbeat_group *hs = NULL;
2267         struct config_group *ret = NULL;
2268
2269         hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2270         if (hs == NULL)
2271                 goto out;
2272
2273         config_group_init_type_name(&hs->hs_group, "heartbeat",
2274                                     &o2hb_heartbeat_group_type);
2275
2276         ret = &hs->hs_group;
2277 out:
2278         if (ret == NULL)
2279                 kfree(hs);
2280         return ret;
2281 }
2282
2283 void o2hb_free_hb_set(struct config_group *group)
2284 {
2285         struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2286         kfree(hs);
2287 }
2288
2289 /* hb callback registration and issuing */
2290
2291 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2292 {
2293         if (type == O2HB_NUM_CB)
2294                 return ERR_PTR(-EINVAL);
2295
2296         return &o2hb_callbacks[type];
2297 }
2298
2299 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2300                          enum o2hb_callback_type type,
2301                          o2hb_cb_func *func,
2302                          void *data,
2303                          int priority)
2304 {
2305         INIT_LIST_HEAD(&hc->hc_item);
2306         hc->hc_func = func;
2307         hc->hc_data = data;
2308         hc->hc_priority = priority;
2309         hc->hc_type = type;
2310         hc->hc_magic = O2HB_CB_MAGIC;
2311 }
2312 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2313
2314 /*
2315  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2316  * In global heartbeat mode, region_uuid passed is NULL.
2317  *
2318  * In local, we only pin the matching region. In global we pin all the active
2319  * regions.
2320  */
2321 static int o2hb_region_pin(const char *region_uuid)
2322 {
2323         int ret = 0, found = 0;
2324         struct o2hb_region *reg;
2325         char *uuid;
2326
2327         assert_spin_locked(&o2hb_live_lock);
2328
2329         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2330                 if (reg->hr_item_dropped)
2331                         continue;
2332
2333                 uuid = config_item_name(&reg->hr_item);
2334
2335                 /* local heartbeat */
2336                 if (region_uuid) {
2337                         if (strcmp(region_uuid, uuid))
2338                                 continue;
2339                         found = 1;
2340                 }
2341
2342                 if (reg->hr_item_pinned || reg->hr_item_dropped)
2343                         goto skip_pin;
2344
2345                 /* Ignore ENOENT only for local hb (userdlm domain) */
2346                 ret = o2nm_depend_item(&reg->hr_item);
2347                 if (!ret) {
2348                         mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2349                         reg->hr_item_pinned = 1;
2350                 } else {
2351                         if (ret == -ENOENT && found)
2352                                 ret = 0;
2353                         else {
2354                                 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2355                                      uuid, ret);
2356                                 break;
2357                         }
2358                 }
2359 skip_pin:
2360                 if (found)
2361                         break;
2362         }
2363
2364         return ret;
2365 }
2366
2367 /*
2368  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2369  * In global heartbeat mode, region_uuid passed is NULL.
2370  *
2371  * In local, we only unpin the matching region. In global we unpin all the
2372  * active regions.
2373  */
2374 static void o2hb_region_unpin(const char *region_uuid)
2375 {
2376         struct o2hb_region *reg;
2377         char *uuid;
2378         int found = 0;
2379
2380         assert_spin_locked(&o2hb_live_lock);
2381
2382         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2383                 if (reg->hr_item_dropped)
2384                         continue;
2385
2386                 uuid = config_item_name(&reg->hr_item);
2387                 if (region_uuid) {
2388                         if (strcmp(region_uuid, uuid))
2389                                 continue;
2390                         found = 1;
2391                 }
2392
2393                 if (reg->hr_item_pinned) {
2394                         mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2395                         o2nm_undepend_item(&reg->hr_item);
2396                         reg->hr_item_pinned = 0;
2397                 }
2398                 if (found)
2399                         break;
2400         }
2401 }
2402
2403 static int o2hb_region_inc_user(const char *region_uuid)
2404 {
2405         int ret = 0;
2406
2407         spin_lock(&o2hb_live_lock);
2408
2409         /* local heartbeat */
2410         if (!o2hb_global_heartbeat_active()) {
2411             ret = o2hb_region_pin(region_uuid);
2412             goto unlock;
2413         }
2414
2415         /*
2416          * if global heartbeat active and this is the first dependent user,
2417          * pin all regions if quorum region count <= CUT_OFF
2418          */
2419         o2hb_dependent_users++;
2420         if (o2hb_dependent_users > 1)
2421                 goto unlock;
2422
2423         if (bitmap_weight(o2hb_quorum_region_bitmap,
2424                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2425                 ret = o2hb_region_pin(NULL);
2426
2427 unlock:
2428         spin_unlock(&o2hb_live_lock);
2429         return ret;
2430 }
2431
2432 static void o2hb_region_dec_user(const char *region_uuid)
2433 {
2434         spin_lock(&o2hb_live_lock);
2435
2436         /* local heartbeat */
2437         if (!o2hb_global_heartbeat_active()) {
2438             o2hb_region_unpin(region_uuid);
2439             goto unlock;
2440         }
2441
2442         /*
2443          * if global heartbeat active and there are no dependent users,
2444          * unpin all quorum regions
2445          */
2446         o2hb_dependent_users--;
2447         if (!o2hb_dependent_users)
2448                 o2hb_region_unpin(NULL);
2449
2450 unlock:
2451         spin_unlock(&o2hb_live_lock);
2452 }
2453
2454 int o2hb_register_callback(const char *region_uuid,
2455                            struct o2hb_callback_func *hc)
2456 {
2457         struct o2hb_callback_func *f;
2458         struct o2hb_callback *hbcall;
2459         int ret;
2460
2461         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2462         BUG_ON(!list_empty(&hc->hc_item));
2463
2464         hbcall = hbcall_from_type(hc->hc_type);
2465         if (IS_ERR(hbcall)) {
2466                 ret = PTR_ERR(hbcall);
2467                 goto out;
2468         }
2469
2470         if (region_uuid) {
2471                 ret = o2hb_region_inc_user(region_uuid);
2472                 if (ret) {
2473                         mlog_errno(ret);
2474                         goto out;
2475                 }
2476         }
2477
2478         down_write(&o2hb_callback_sem);
2479
2480         list_for_each_entry(f, &hbcall->list, hc_item) {
2481                 if (hc->hc_priority < f->hc_priority) {
2482                         list_add_tail(&hc->hc_item, &f->hc_item);
2483                         break;
2484                 }
2485         }
2486         if (list_empty(&hc->hc_item))
2487                 list_add_tail(&hc->hc_item, &hbcall->list);
2488
2489         up_write(&o2hb_callback_sem);
2490         ret = 0;
2491 out:
2492         mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2493              ret, __builtin_return_address(0), hc);
2494         return ret;
2495 }
2496 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2497
2498 void o2hb_unregister_callback(const char *region_uuid,
2499                               struct o2hb_callback_func *hc)
2500 {
2501         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2502
2503         mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2504              __builtin_return_address(0), hc);
2505
2506         /* XXX Can this happen _with_ a region reference? */
2507         if (list_empty(&hc->hc_item))
2508                 return;
2509
2510         if (region_uuid)
2511                 o2hb_region_dec_user(region_uuid);
2512
2513         down_write(&o2hb_callback_sem);
2514
2515         list_del_init(&hc->hc_item);
2516
2517         up_write(&o2hb_callback_sem);
2518 }
2519 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2520
2521 int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2522 {
2523         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2524
2525         spin_lock(&o2hb_live_lock);
2526         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2527         spin_unlock(&o2hb_live_lock);
2528         if (!test_bit(node_num, testing_map)) {
2529                 mlog(ML_HEARTBEAT,
2530                      "node (%u) does not have heartbeating enabled.\n",
2531                      node_num);
2532                 return 0;
2533         }
2534
2535         return 1;
2536 }
2537 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2538
2539 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2540 {
2541         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2542
2543         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2544         if (!test_bit(node_num, testing_map)) {
2545                 mlog(ML_HEARTBEAT,
2546                      "node (%u) does not have heartbeating enabled.\n",
2547                      node_num);
2548                 return 0;
2549         }
2550
2551         return 1;
2552 }
2553 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2554
2555 /*
2556  * this is just a hack until we get the plumbing which flips file systems
2557  * read only and drops the hb ref instead of killing the node dead.
2558  */
2559 void o2hb_stop_all_regions(void)
2560 {
2561         struct o2hb_region *reg;
2562
2563         mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2564
2565         spin_lock(&o2hb_live_lock);
2566
2567         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2568                 reg->hr_unclean_stop = 1;
2569
2570         spin_unlock(&o2hb_live_lock);
2571 }
2572 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2573
2574 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2575 {
2576         struct o2hb_region *reg;
2577         int numregs = 0;
2578         char *p;
2579
2580         spin_lock(&o2hb_live_lock);
2581
2582         p = region_uuids;
2583         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2584                 if (reg->hr_item_dropped)
2585                         continue;
2586
2587                 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2588                 if (numregs < max_regions) {
2589                         memcpy(p, config_item_name(&reg->hr_item),
2590                                O2HB_MAX_REGION_NAME_LEN);
2591                         p += O2HB_MAX_REGION_NAME_LEN;
2592                 }
2593                 numregs++;
2594         }
2595
2596         spin_unlock(&o2hb_live_lock);
2597
2598         return numregs;
2599 }
2600 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2601
2602 int o2hb_global_heartbeat_active(void)
2603 {
2604         return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2605 }
2606 EXPORT_SYMBOL(o2hb_global_heartbeat_active);