fs/ocfs2/cluster/heartbeat.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2 of the License, or (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public
  17  * License along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA.
  20  */
  21
  22 #include <linux/kernel.h>
  23 #include <linux/sched.h>
  24 #include <linux/jiffies.h>
  25 #include <linux/module.h>
  26 #include <linux/fs.h>
  27 #include <linux/bio.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/delay.h>
  30 #include <linux/file.h>
  31 #include <linux/kthread.h>
  32 #include <linux/configfs.h>
  33 #include <linux/random.h>
  34 #include <linux/crc32.h>
  35 #include <linux/time.h>
  36 #include <linux/debugfs.h>
  37 #include <linux/slab.h>
  38
  39 #include "heartbeat.h"
  40 #include "tcp.h"
  41 #include "nodemanager.h"
  42 #include "quorum.h"
  43
  44 #include "masklog.h"
  45
  46
  47 /*
  48  * The first heartbeat pass had one global thread that would serialize all hb
  49  * callback calls.  This global serializing sem should only be removed once
  50  * we've made sure that all callees can deal with being called concurrently
  51  * from multiple hb region threads.
  52  */
  53 static DECLARE_RWSEM(o2hb_callback_sem);
  54
  55 /*
  56  * multiple hb threads are watching multiple regions.  A node is live
  57  * whenever any of the threads sees activity from the node in its region.
  58  */
  59 static DEFINE_SPINLOCK(o2hb_live_lock);
  60 static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
  61 static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
  62 static LIST_HEAD(o2hb_node_events);
  63 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
  64
  65 /*
  66  * In global heartbeat, we maintain a series of region bitmaps.
  67  *      - o2hb_region_bitmap allows us to limit the region number to max region.
  68  *      - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
  69  *      - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
  70  *              heartbeat on it.
  71  *      - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
  72  */
  73 static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  74 static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  75 static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  76 static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  77
  78 #define O2HB_DB_TYPE_LIVENODES          0
  79 #define O2HB_DB_TYPE_LIVEREGIONS        1
  80 #define O2HB_DB_TYPE_QUORUMREGIONS      2
  81 #define O2HB_DB_TYPE_FAILEDREGIONS      3
  82 #define O2HB_DB_TYPE_REGION_LIVENODES   4
  83 #define O2HB_DB_TYPE_REGION_NUMBER      5
  84 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME        6
  85 #define O2HB_DB_TYPE_REGION_PINNED      7
  86 struct o2hb_debug_buf {
  87         int db_type;
  88         int db_size;
  89         int db_len;
  90         void *db_data;
  91 };
  92
  93 static struct o2hb_debug_buf *o2hb_db_livenodes;
  94 static struct o2hb_debug_buf *o2hb_db_liveregions;
  95 static struct o2hb_debug_buf *o2hb_db_quorumregions;
  96 static struct o2hb_debug_buf *o2hb_db_failedregions;
  97
  98 #define O2HB_DEBUG_DIR                  "o2hb"
  99 #define O2HB_DEBUG_LIVENODES            "livenodes"
 100 #define O2HB_DEBUG_LIVEREGIONS          "live_regions"
 101 #define O2HB_DEBUG_QUORUMREGIONS        "quorum_regions"
 102 #define O2HB_DEBUG_FAILEDREGIONS        "failed_regions"
 103 #define O2HB_DEBUG_REGION_NUMBER        "num"
 104 #define O2HB_DEBUG_REGION_ELAPSED_TIME  "elapsed_time_in_ms"
 105 #define O2HB_DEBUG_REGION_PINNED        "pinned"
 106
 107 static struct dentry *o2hb_debug_dir;
 108 static struct dentry *o2hb_debug_livenodes;
 109 static struct dentry *o2hb_debug_liveregions;
 110 static struct dentry *o2hb_debug_quorumregions;
 111 static struct dentry *o2hb_debug_failedregions;
 112
 113 static LIST_HEAD(o2hb_all_regions);
 114
 115 static struct o2hb_callback {
 116         struct list_head list;
 117 } o2hb_callbacks[O2HB_NUM_CB];
 118
 119 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
 120
 121 #define O2HB_DEFAULT_BLOCK_BITS       9
 122
 123 enum o2hb_heartbeat_modes {
 124         O2HB_HEARTBEAT_LOCAL            = 0,
 125         O2HB_HEARTBEAT_GLOBAL,
 126         O2HB_HEARTBEAT_NUM_MODES,
 127 };
 128
 129 char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
 130                 "local",        /* O2HB_HEARTBEAT_LOCAL */
 131                 "global",       /* O2HB_HEARTBEAT_GLOBAL */
 132 };
 133
 134 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
 135 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
 136
 137 /*
 138  * o2hb_dependent_users tracks the number of registered callbacks that depend
 139  * on heartbeat. o2net and o2dlm are two entities that register this callback.
 140  * However only o2dlm depends on the heartbeat. It does not want the heartbeat
 141  * to stop while a dlm domain is still active.
 142  */
 143 unsigned int o2hb_dependent_users;
 144
 145 /*
 146  * In global heartbeat mode, all regions are pinned if there are one or more
 147  * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
 148  * regions are unpinned if the region count exceeds the cut off or the number
 149  * of dependent users falls to zero.
 150  */
 151 #define O2HB_PIN_CUT_OFF                3
 152
 153 /*
 154  * In local heartbeat mode, we assume the dlm domain name to be the same as
 155  * region uuid. This is true for domains created for the file system but not
 156  * necessarily true for userdlm domains. This is a known limitation.
 157  *
 158  * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
 159  * works for both file system and userdlm domains.
 160  */
 161 static int o2hb_region_pin(const char *region_uuid);
 162 static void o2hb_region_unpin(const char *region_uuid);
 163
 164 /* Only sets a new threshold if there are no active regions.
 165  *
 166  * No locking or otherwise interesting code is required for reading
 167  * o2hb_dead_threshold as it can't change once regions are active and
 168  * it's not interesting to anyone until then anyway. */
 169 static void o2hb_dead_threshold_set(unsigned int threshold)
 170 {
 171         if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
 172                 spin_lock(&o2hb_live_lock);
 173                 if (list_empty(&o2hb_all_regions))
 174                         o2hb_dead_threshold = threshold;
 175                 spin_unlock(&o2hb_live_lock);
 176         }
 177 }
 178
 179 static int o2hb_global_hearbeat_mode_set(unsigned int hb_mode)
 180 {
 181         int ret = -1;
 182
 183         if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
 184                 spin_lock(&o2hb_live_lock);
 185                 if (list_empty(&o2hb_all_regions)) {
 186                         o2hb_heartbeat_mode = hb_mode;
 187                         ret = 0;
 188                 }
 189                 spin_unlock(&o2hb_live_lock);
 190         }
 191
 192         return ret;
 193 }
 194
 195 struct o2hb_node_event {
 196         struct list_head        hn_item;
 197         enum o2hb_callback_type hn_event_type;
 198         struct o2nm_node        *hn_node;
 199         int                     hn_node_num;
 200 };
 201
 202 struct o2hb_disk_slot {
 203         struct o2hb_disk_heartbeat_block *ds_raw_block;
 204         u8                      ds_node_num;
 205         u64                     ds_last_time;
 206         u64                     ds_last_generation;
 207         u16                     ds_equal_samples;
 208         u16                     ds_changed_samples;
 209         struct list_head        ds_live_item;
 210 };
 211
 212 /* each thread owns a region.. when we're asked to tear down the region
 213  * we ask the thread to stop, who cleans up the region */
 214 struct o2hb_region {
 215         struct config_item      hr_item;
 216
 217         struct list_head        hr_all_item;
 218         unsigned                hr_unclean_stop:1,
 219                                 hr_item_pinned:1,
 220                                 hr_item_dropped:1;
 221
 222         /* protected by the hr_callback_sem */
 223         struct task_struct      *hr_task;
 224
 225         unsigned int            hr_blocks;
 226         unsigned long long      hr_start_block;
 227
 228         unsigned int            hr_block_bits;
 229         unsigned int            hr_block_bytes;
 230
 231         unsigned int            hr_slots_per_page;
 232         unsigned int            hr_num_pages;
 233
 234         struct page             **hr_slot_data;
 235         struct block_device     *hr_bdev;
 236         struct o2hb_disk_slot   *hr_slots;
 237
 238         /* live node map of this region */
 239         unsigned long           hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 240         unsigned int            hr_region_num;
 241
 242         struct dentry           *hr_debug_dir;
 243         struct dentry           *hr_debug_livenodes;
 244         struct dentry           *hr_debug_regnum;
 245         struct dentry           *hr_debug_elapsed_time;
 246         struct dentry           *hr_debug_pinned;
 247         struct o2hb_debug_buf   *hr_db_livenodes;
 248         struct o2hb_debug_buf   *hr_db_regnum;
 249         struct o2hb_debug_buf   *hr_db_elapsed_time;
 250         struct o2hb_debug_buf   *hr_db_pinned;
 251
 252         /* let the person setting up hb wait for it to return until it
 253          * has reached a 'steady' state.  This will be fixed when we have
 254          * a more complete api that doesn't lead to this sort of fragility. */
 255         atomic_t                hr_steady_iterations;
 256
 257         char                    hr_dev_name[BDEVNAME_SIZE];
 258
 259         unsigned int            hr_timeout_ms;
 260
 261         /* randomized as the region goes up and down so that a node
 262          * recognizes a node going up and down in one iteration */
 263         u64                     hr_generation;
 264
 265         struct delayed_work     hr_write_timeout_work;
 266         unsigned long           hr_last_timeout_start;
 267
 268         /* Used during o2hb_check_slot to hold a copy of the block
 269          * being checked because we temporarily have to zero out the
 270          * crc field. */
 271         struct o2hb_disk_heartbeat_block *hr_tmp_block;
 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 static int o2hb_pop_count(void *map, int count)
 281 {
 282         int i = -1, pop = 0;
 283
 284         while ((i = find_next_bit(map, count, i + 1)) < count)
 285                 pop++;
 286         return pop;
 287 }
 288
 289 static void o2hb_write_timeout(struct work_struct *work)
 290 {
 291         int failed, quorum;
 292         unsigned long flags;
 293         struct o2hb_region *reg =
 294                 container_of(work, struct o2hb_region,
 295                              hr_write_timeout_work.work);
 296
 297         mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
 298              "milliseconds\n", reg->hr_dev_name,
 299              jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
 300
 301         if (o2hb_global_heartbeat_active()) {
 302                 spin_lock_irqsave(&o2hb_live_lock, flags);
 303                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 304                         set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 305                 failed = o2hb_pop_count(&o2hb_failed_region_bitmap,
 306                                         O2NM_MAX_REGIONS);
 307                 quorum = o2hb_pop_count(&o2hb_quorum_region_bitmap,
 308                                         O2NM_MAX_REGIONS);
 309                 spin_unlock_irqrestore(&o2hb_live_lock, flags);
 310
 311                 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
 312                      quorum, failed);
 313
 314                 /*
 315                  * Fence if the number of failed regions >= half the number
 316                  * of  quorum regions
 317                  */
 318                 if ((failed << 1) < quorum)
 319                         return;
 320         }
 321
 322         o2quo_disk_timeout();
 323 }
 324
 325 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
 326 {
 327         mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
 328              O2HB_MAX_WRITE_TIMEOUT_MS);
 329
 330         if (o2hb_global_heartbeat_active()) {
 331                 spin_lock(&o2hb_live_lock);
 332                 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 333                 spin_unlock(&o2hb_live_lock);
 334         }
 335         cancel_delayed_work(&reg->hr_write_timeout_work);
 336         reg->hr_last_timeout_start = jiffies;
 337         schedule_delayed_work(&reg->hr_write_timeout_work,
 338                               msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
 339 }
 340
 341 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
 342 {
 343         cancel_delayed_work(&reg->hr_write_timeout_work);
 344         flush_scheduled_work();
 345 }
 346
 347 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
 348 {
 349         atomic_set(&wc->wc_num_reqs, 1);
 350         init_completion(&wc->wc_io_complete);
 351         wc->wc_error = 0;
 352 }
 353
 354 /* Used in error paths too */
 355 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
 356                                      unsigned int num)
 357 {
 358         /* sadly atomic_sub_and_test() isn't available on all platforms.  The
 359          * good news is that the fast path only completes one at a time */
 360         while(num--) {
 361                 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
 362                         BUG_ON(num > 0);
 363                         complete(&wc->wc_io_complete);
 364                 }
 365         }
 366 }
 367
 368 static void o2hb_wait_on_io(struct o2hb_region *reg,
 369                             struct o2hb_bio_wait_ctxt *wc)
 370 {
 371         struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping;
 372
 373         blk_run_address_space(mapping);
 374         o2hb_bio_wait_dec(wc, 1);
 375
 376         wait_for_completion(&wc->wc_io_complete);
 377 }
 378
 379 static void o2hb_bio_end_io(struct bio *bio,
 380                            int error)
 381 {
 382         struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
 383
 384         if (error) {
 385                 mlog(ML_ERROR, "IO Error %d\n", error);
 386                 wc->wc_error = error;
 387         }
 388
 389         o2hb_bio_wait_dec(wc, 1);
 390         bio_put(bio);
 391 }
 392
 393 /* Setup a Bio to cover I/O against num_slots slots starting at
 394  * start_slot. */
 395 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
 396                                       struct o2hb_bio_wait_ctxt *wc,
 397                                       unsigned int *current_slot,
 398                                       unsigned int max_slots)
 399 {
 400         int len, current_page;
 401         unsigned int vec_len, vec_start;
 402         unsigned int bits = reg->hr_block_bits;
 403         unsigned int spp = reg->hr_slots_per_page;
 404         unsigned int cs = *current_slot;
 405         struct bio *bio;
 406         struct page *page;
 407
 408         /* Testing has shown this allocation to take long enough under
 409          * GFP_KERNEL that the local node can get fenced. It would be
 410          * nicest if we could pre-allocate these bios and avoid this
 411          * all together. */
 412         bio = bio_alloc(GFP_ATOMIC, 16);
 413         if (!bio) {
 414                 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
 415                 bio = ERR_PTR(-ENOMEM);
 416                 goto bail;
 417         }
 418
 419         /* Must put everything in 512 byte sectors for the bio... */
 420         bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
 421         bio->bi_bdev = reg->hr_bdev;
 422         bio->bi_private = wc;
 423         bio->bi_end_io = o2hb_bio_end_io;
 424
 425         vec_start = (cs << bits) % PAGE_CACHE_SIZE;
 426         while(cs < max_slots) {
 427                 current_page = cs / spp;
 428                 page = reg->hr_slot_data[current_page];
 429
 430                 vec_len = min(PAGE_CACHE_SIZE - vec_start,
 431                               (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
 432
 433                 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
 434                      current_page, vec_len, vec_start);
 435
 436                 len = bio_add_page(bio, page, vec_len, vec_start);
 437                 if (len != vec_len) break;
 438
 439                 cs += vec_len / (PAGE_CACHE_SIZE/spp);
 440                 vec_start = 0;
 441         }
 442
 443 bail:
 444         *current_slot = cs;
 445         return bio;
 446 }
 447
 448 static int o2hb_read_slots(struct o2hb_region *reg,
 449                            unsigned int max_slots)
 450 {
 451         unsigned int current_slot=0;
 452         int status;
 453         struct o2hb_bio_wait_ctxt wc;
 454         struct bio *bio;
 455
 456         o2hb_bio_wait_init(&wc);
 457
 458         while(current_slot < max_slots) {
 459                 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
 460                 if (IS_ERR(bio)) {
 461                         status = PTR_ERR(bio);
 462                         mlog_errno(status);
 463                         goto bail_and_wait;
 464                 }
 465
 466                 atomic_inc(&wc.wc_num_reqs);
 467                 submit_bio(READ, bio);
 468         }
 469
 470         status = 0;
 471
 472 bail_and_wait:
 473         o2hb_wait_on_io(reg, &wc);
 474         if (wc.wc_error && !status)
 475                 status = wc.wc_error;
 476
 477         return status;
 478 }
 479
 480 static int o2hb_issue_node_write(struct o2hb_region *reg,
 481                                  struct o2hb_bio_wait_ctxt *write_wc)
 482 {
 483         int status;
 484         unsigned int slot;
 485         struct bio *bio;
 486
 487         o2hb_bio_wait_init(write_wc);
 488
 489         slot = o2nm_this_node();
 490
 491         bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
 492         if (IS_ERR(bio)) {
 493                 status = PTR_ERR(bio);
 494                 mlog_errno(status);
 495                 goto bail;
 496         }
 497
 498         atomic_inc(&write_wc->wc_num_reqs);
 499         submit_bio(WRITE, bio);
 500
 501         status = 0;
 502 bail:
 503         return status;
 504 }
 505
 506 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
 507                                      struct o2hb_disk_heartbeat_block *hb_block)
 508 {
 509         __le32 old_cksum;
 510         u32 ret;
 511
 512         /* We want to compute the block crc with a 0 value in the
 513          * hb_cksum field. Save it off here and replace after the
 514          * crc. */
 515         old_cksum = hb_block->hb_cksum;
 516         hb_block->hb_cksum = 0;
 517
 518         ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
 519
 520         hb_block->hb_cksum = old_cksum;
 521
 522         return ret;
 523 }
 524
 525 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
 526 {
 527         mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
 528              "cksum = 0x%x, generation 0x%llx\n",
 529              (long long)le64_to_cpu(hb_block->hb_seq),
 530              hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
 531              (long long)le64_to_cpu(hb_block->hb_generation));
 532 }
 533
 534 static int o2hb_verify_crc(struct o2hb_region *reg,
 535                            struct o2hb_disk_heartbeat_block *hb_block)
 536 {
 537         u32 read, computed;
 538
 539         read = le32_to_cpu(hb_block->hb_cksum);
 540         computed = o2hb_compute_block_crc_le(reg, hb_block);
 541
 542         return read == computed;
 543 }
 544
 545 /* We want to make sure that nobody is heartbeating on top of us --
 546  * this will help detect an invalid configuration. */
 547 static int o2hb_check_last_timestamp(struct o2hb_region *reg)
 548 {
 549         int node_num, ret;
 550         struct o2hb_disk_slot *slot;
 551         struct o2hb_disk_heartbeat_block *hb_block;
 552
 553         node_num = o2nm_this_node();
 554
 555         ret = 1;
 556         slot = &reg->hr_slots[node_num];
 557         /* Don't check on our 1st timestamp */
 558         if (slot->ds_last_time) {
 559                 hb_block = slot->ds_raw_block;
 560
 561                 if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time)
 562                         ret = 0;
 563         }
 564
 565         return ret;
 566 }
 567
 568 static inline void o2hb_prepare_block(struct o2hb_region *reg,
 569                                       u64 generation)
 570 {
 571         int node_num;
 572         u64 cputime;
 573         struct o2hb_disk_slot *slot;
 574         struct o2hb_disk_heartbeat_block *hb_block;
 575
 576         node_num = o2nm_this_node();
 577         slot = &reg->hr_slots[node_num];
 578
 579         hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
 580         memset(hb_block, 0, reg->hr_block_bytes);
 581         /* TODO: time stuff */
 582         cputime = CURRENT_TIME.tv_sec;
 583         if (!cputime)
 584                 cputime = 1;
 585
 586         hb_block->hb_seq = cpu_to_le64(cputime);
 587         hb_block->hb_node = node_num;
 588         hb_block->hb_generation = cpu_to_le64(generation);
 589         hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
 590
 591         /* This step must always happen last! */
 592         hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
 593                                                                    hb_block));
 594
 595         mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
 596              (long long)generation,
 597              le32_to_cpu(hb_block->hb_cksum));
 598 }
 599
 600 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
 601                                 struct o2nm_node *node,
 602                                 int idx)
 603 {
 604         struct list_head *iter;
 605         struct o2hb_callback_func *f;
 606
 607         list_for_each(iter, &hbcall->list) {
 608                 f = list_entry(iter, struct o2hb_callback_func, hc_item);
 609                 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
 610                 (f->hc_func)(node, idx, f->hc_data);
 611         }
 612 }
 613
 614 /* Will run the list in order until we process the passed event */
 615 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
 616 {
 617         int empty;
 618         struct o2hb_callback *hbcall;
 619         struct o2hb_node_event *event;
 620
 621         spin_lock(&o2hb_live_lock);
 622         empty = list_empty(&queued_event->hn_item);
 623         spin_unlock(&o2hb_live_lock);
 624         if (empty)
 625                 return;
 626
 627         /* Holding callback sem assures we don't alter the callback
 628          * lists when doing this, and serializes ourselves with other
 629          * processes wanting callbacks. */
 630         down_write(&o2hb_callback_sem);
 631
 632         spin_lock(&o2hb_live_lock);
 633         while (!list_empty(&o2hb_node_events)
 634                && !list_empty(&queued_event->hn_item)) {
 635                 event = list_entry(o2hb_node_events.next,
 636                                    struct o2hb_node_event,
 637                                    hn_item);
 638                 list_del_init(&event->hn_item);
 639                 spin_unlock(&o2hb_live_lock);
 640
 641                 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
 642                      event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
 643                      event->hn_node_num);
 644
 645                 hbcall = hbcall_from_type(event->hn_event_type);
 646
 647                 /* We should *never* have gotten on to the list with a
 648                  * bad type... This isn't something that we should try
 649                  * to recover from. */
 650                 BUG_ON(IS_ERR(hbcall));
 651
 652                 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
 653
 654                 spin_lock(&o2hb_live_lock);
 655         }
 656         spin_unlock(&o2hb_live_lock);
 657
 658         up_write(&o2hb_callback_sem);
 659 }
 660
 661 static void o2hb_queue_node_event(struct o2hb_node_event *event,
 662                                   enum o2hb_callback_type type,
 663                                   struct o2nm_node *node,
 664                                   int node_num)
 665 {
 666         assert_spin_locked(&o2hb_live_lock);
 667
 668         BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
 669
 670         event->hn_event_type = type;
 671         event->hn_node = node;
 672         event->hn_node_num = node_num;
 673
 674         mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
 675              type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
 676
 677         list_add_tail(&event->hn_item, &o2hb_node_events);
 678 }
 679
 680 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
 681 {
 682         struct o2hb_node_event event =
 683                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 684         struct o2nm_node *node;
 685
 686         node = o2nm_get_node_by_num(slot->ds_node_num);
 687         if (!node)
 688                 return;
 689
 690         spin_lock(&o2hb_live_lock);
 691         if (!list_empty(&slot->ds_live_item)) {
 692                 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
 693                      slot->ds_node_num);
 694
 695                 list_del_init(&slot->ds_live_item);
 696
 697                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 698                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 699
 700                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
 701                                               slot->ds_node_num);
 702                 }
 703         }
 704         spin_unlock(&o2hb_live_lock);
 705
 706         o2hb_run_event_list(&event);
 707
 708         o2nm_node_put(node);
 709 }
 710
 711 static void o2hb_set_quorum_device(struct o2hb_region *reg,
 712                                    struct o2hb_disk_slot *slot)
 713 {
 714         assert_spin_locked(&o2hb_live_lock);
 715
 716         if (!o2hb_global_heartbeat_active())
 717                 return;
 718
 719         if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 720                 return;
 721
 722         /*
 723          * A region can be added to the quorum only when it sees all
 724          * live nodes heartbeat on it. In other words, the region has been
 725          * added to all nodes.
 726          */
 727         if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
 728                    sizeof(o2hb_live_node_bitmap)))
 729                 return;
 730
 731         if (slot->ds_changed_samples < O2HB_LIVE_THRESHOLD)
 732                 return;
 733
 734         printk(KERN_NOTICE "o2hb: Region %s is now a quorum device\n",
 735                config_item_name(&reg->hr_item));
 736
 737         set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
 738
 739         /*
 740          * If global heartbeat active, unpin all regions if the
 741          * region count > CUT_OFF
 742          */
 743         if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
 744                            O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
 745                 o2hb_region_unpin(NULL);
 746 }
 747
 748 static int o2hb_check_slot(struct o2hb_region *reg,
 749                            struct o2hb_disk_slot *slot)
 750 {
 751         int changed = 0, gen_changed = 0;
 752         struct o2hb_node_event event =
 753                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 754         struct o2nm_node *node;
 755         struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
 756         u64 cputime;
 757         unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
 758         unsigned int slot_dead_ms;
 759         int tmp;
 760
 761         memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
 762
 763         /*
 764          * If a node is no longer configured but is still in the livemap, we
 765          * may need to clear that bit from the livemap.
 766          */
 767         node = o2nm_get_node_by_num(slot->ds_node_num);
 768         if (!node) {
 769                 spin_lock(&o2hb_live_lock);
 770                 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 771                 spin_unlock(&o2hb_live_lock);
 772                 if (!tmp)
 773                         return 0;
 774         }
 775
 776         if (!o2hb_verify_crc(reg, hb_block)) {
 777                 /* all paths from here will drop o2hb_live_lock for
 778                  * us. */
 779                 spin_lock(&o2hb_live_lock);
 780
 781                 /* Don't print an error on the console in this case -
 782                  * a freshly formatted heartbeat area will not have a
 783                  * crc set on it. */
 784                 if (list_empty(&slot->ds_live_item))
 785                         goto out;
 786
 787                 /* The node is live but pushed out a bad crc. We
 788                  * consider it a transient miss but don't populate any
 789                  * other values as they may be junk. */
 790                 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
 791                      slot->ds_node_num, reg->hr_dev_name);
 792                 o2hb_dump_slot(hb_block);
 793
 794                 slot->ds_equal_samples++;
 795                 goto fire_callbacks;
 796         }
 797
 798         /* we don't care if these wrap.. the state transitions below
 799          * clear at the right places */
 800         cputime = le64_to_cpu(hb_block->hb_seq);
 801         if (slot->ds_last_time != cputime)
 802                 slot->ds_changed_samples++;
 803         else
 804                 slot->ds_equal_samples++;
 805         slot->ds_last_time = cputime;
 806
 807         /* The node changed heartbeat generations. We assume this to
 808          * mean it dropped off but came back before we timed out. We
 809          * want to consider it down for the time being but don't want
 810          * to lose any changed_samples state we might build up to
 811          * considering it live again. */
 812         if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
 813                 gen_changed = 1;
 814                 slot->ds_equal_samples = 0;
 815                 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
 816                      "to 0x%llx)\n", slot->ds_node_num,
 817                      (long long)slot->ds_last_generation,
 818                      (long long)le64_to_cpu(hb_block->hb_generation));
 819         }
 820
 821         slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
 822
 823         mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
 824              "seq %llu last %llu changed %u equal %u\n",
 825              slot->ds_node_num, (long long)slot->ds_last_generation,
 826              le32_to_cpu(hb_block->hb_cksum),
 827              (unsigned long long)le64_to_cpu(hb_block->hb_seq),
 828              (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
 829              slot->ds_equal_samples);
 830
 831         spin_lock(&o2hb_live_lock);
 832
 833 fire_callbacks:
 834         /* dead nodes only come to life after some number of
 835          * changes at any time during their dead time */
 836         if (list_empty(&slot->ds_live_item) &&
 837             slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
 838                 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
 839                      slot->ds_node_num, (long long)slot->ds_last_generation);
 840
 841                 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 842
 843                 /* first on the list generates a callback */
 844                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 845                         mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
 846                              "bitmap\n", slot->ds_node_num);
 847                         set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 848
 849                         o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
 850                                               slot->ds_node_num);
 851
 852                         changed = 1;
 853                 }
 854
 855                 list_add_tail(&slot->ds_live_item,
 856                               &o2hb_live_slots[slot->ds_node_num]);
 857
 858                 slot->ds_equal_samples = 0;
 859
 860                 /* We want to be sure that all nodes agree on the
 861                  * number of milliseconds before a node will be
 862                  * considered dead. The self-fencing timeout is
 863                  * computed from this value, and a discrepancy might
 864                  * result in heartbeat calling a node dead when it
 865                  * hasn't self-fenced yet. */
 866                 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
 867                 if (slot_dead_ms && slot_dead_ms != dead_ms) {
 868                         /* TODO: Perhaps we can fail the region here. */
 869                         mlog(ML_ERROR, "Node %d on device %s has a dead count "
 870                              "of %u ms, but our count is %u ms.\n"
 871                              "Please double check your configuration values "
 872                              "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
 873                              slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
 874                              dead_ms);
 875                 }
 876                 goto out;
 877         }
 878
 879         /* if the list is dead, we're done.. */
 880         if (list_empty(&slot->ds_live_item))
 881                 goto out;
 882
 883         /* live nodes only go dead after enough consequtive missed
 884          * samples..  reset the missed counter whenever we see
 885          * activity */
 886         if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
 887                 mlog(ML_HEARTBEAT, "Node %d left my region\n",
 888                      slot->ds_node_num);
 889
 890                 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 891
 892                 /* last off the live_slot generates a callback */
 893                 list_del_init(&slot->ds_live_item);
 894                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 895                         mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
 896                              "nodes bitmap\n", slot->ds_node_num);
 897                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 898
 899                         /* node can be null */
 900                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
 901                                               node, slot->ds_node_num);
 902
 903                         changed = 1;
 904                 }
 905
 906                 /* We don't clear this because the node is still
 907                  * actually writing new blocks. */
 908                 if (!gen_changed)
 909                         slot->ds_changed_samples = 0;
 910                 goto out;
 911         }
 912         if (slot->ds_changed_samples) {
 913                 slot->ds_changed_samples = 0;
 914                 slot->ds_equal_samples = 0;
 915         }
 916 out:
 917         o2hb_set_quorum_device(reg, slot);
 918
 919         spin_unlock(&o2hb_live_lock);
 920
 921         o2hb_run_event_list(&event);
 922
 923         if (node)
 924                 o2nm_node_put(node);
 925         return changed;
 926 }
 927
 928 /* This could be faster if we just implmented a find_last_bit, but I
 929  * don't think the circumstances warrant it. */
 930 static int o2hb_highest_node(unsigned long *nodes,
 931                              int numbits)
 932 {
 933         int highest, node;
 934
 935         highest = numbits;
 936         node = -1;
 937         while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
 938                 if (node >= numbits)
 939                         break;
 940
 941                 highest = node;
 942         }
 943
 944         return highest;
 945 }
 946
 947 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
 948 {
 949         int i, ret, highest_node, change = 0;
 950         unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
 951         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 952         struct o2hb_bio_wait_ctxt write_wc;
 953
 954         ret = o2nm_configured_node_map(configured_nodes,
 955                                        sizeof(configured_nodes));
 956         if (ret) {
 957                 mlog_errno(ret);
 958                 return ret;
 959         }
 960
 961         /*
 962          * If a node is not configured but is in the livemap, we still need
 963          * to read the slot so as to be able to remove it from the livemap.
 964          */
 965         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
 966         i = -1;
 967         while ((i = find_next_bit(live_node_bitmap,
 968                                   O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
 969                 set_bit(i, configured_nodes);
 970         }
 971
 972         highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
 973         if (highest_node >= O2NM_MAX_NODES) {
 974                 mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n");
 975                 return -EINVAL;
 976         }
 977
 978         /* No sense in reading the slots of nodes that don't exist
 979          * yet. Of course, if the node definitions have holes in them
 980          * then we're reading an empty slot anyway... Consider this
 981          * best-effort. */
 982         ret = o2hb_read_slots(reg, highest_node + 1);
 983         if (ret < 0) {
 984                 mlog_errno(ret);
 985                 return ret;
 986         }
 987
 988         /* With an up to date view of the slots, we can check that no
 989          * other node has been improperly configured to heartbeat in
 990          * our slot. */
 991         if (!o2hb_check_last_timestamp(reg))
 992                 mlog(ML_ERROR, "Device \"%s\": another node is heartbeating "
 993                      "in our slot!\n", reg->hr_dev_name);
 994
 995         /* fill in the proper info for our next heartbeat */
 996         o2hb_prepare_block(reg, reg->hr_generation);
 997
 998         /* And fire off the write. Note that we don't wait on this I/O
 999          * until later. */
1000         ret = o2hb_issue_node_write(reg, &write_wc);
1001         if (ret < 0) {
1002                 mlog_errno(ret);
1003                 return ret;
1004         }
1005
1006         i = -1;
1007         while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1008
1009                 change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1010         }
1011
1012         /*
1013          * We have to be sure we've advertised ourselves on disk
1014          * before we can go to steady state.  This ensures that
1015          * people we find in our steady state have seen us.
1016          */
1017         o2hb_wait_on_io(reg, &write_wc);
1018         if (write_wc.wc_error) {
1019                 /* Do not re-arm the write timeout on I/O error - we
1020                  * can't be sure that the new block ever made it to
1021                  * disk */
1022                 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1023                      write_wc.wc_error, reg->hr_dev_name);
1024                 return write_wc.wc_error;
1025         }
1026
1027         o2hb_arm_write_timeout(reg);
1028
1029         /* let the person who launched us know when things are steady */
1030         if (!change && (atomic_read(&reg->hr_steady_iterations) != 0)) {
1031                 if (atomic_dec_and_test(&reg->hr_steady_iterations))
1032                         wake_up(&o2hb_steady_queue);
1033         }
1034
1035         return 0;
1036 }
1037
1038 /* Subtract b from a, storing the result in a. a *must* have a larger
1039  * value than b. */
1040 static void o2hb_tv_subtract(struct timeval *a,
1041                              struct timeval *b)
1042 {
1043         /* just return 0 when a is after b */
1044         if (a->tv_sec < b->tv_sec ||
1045             (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1046                 a->tv_sec = 0;
1047                 a->tv_usec = 0;
1048                 return;
1049         }
1050
1051         a->tv_sec -= b->tv_sec;
1052         a->tv_usec -= b->tv_usec;
1053         while ( a->tv_usec < 0 ) {
1054                 a->tv_sec--;
1055                 a->tv_usec += 1000000;
1056         }
1057 }
1058
1059 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1060                                        struct timeval *end)
1061 {
1062         struct timeval res = *end;
1063
1064         o2hb_tv_subtract(&res, start);
1065
1066         return res.tv_sec * 1000 + res.tv_usec / 1000;
1067 }
1068
1069 /*
1070  * we ride the region ref that the region dir holds.  before the region
1071  * dir is removed and drops it ref it will wait to tear down this
1072  * thread.
1073  */
1074 static int o2hb_thread(void *data)
1075 {
1076         int i, ret;
1077         struct o2hb_region *reg = data;
1078         struct o2hb_bio_wait_ctxt write_wc;
1079         struct timeval before_hb, after_hb;
1080         unsigned int elapsed_msec;
1081
1082         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1083
1084         set_user_nice(current, -20);
1085
1086         /* Pin node */
1087         o2nm_depend_this_node();
1088
1089         while (!kthread_should_stop() && !reg->hr_unclean_stop) {
1090                 /* We track the time spent inside
1091                  * o2hb_do_disk_heartbeat so that we avoid more than
1092                  * hr_timeout_ms between disk writes. On busy systems
1093                  * this should result in a heartbeat which is less
1094                  * likely to time itself out. */
1095                 do_gettimeofday(&before_hb);
1096
1097                 i = 0;
1098                 do {
1099                         ret = o2hb_do_disk_heartbeat(reg);
1100                 } while (ret && ++i < 2);
1101
1102                 do_gettimeofday(&after_hb);
1103                 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1104
1105                 mlog(ML_HEARTBEAT,
1106                      "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
1107                      before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1108                      after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1109                      elapsed_msec);
1110
1111                 if (elapsed_msec < reg->hr_timeout_ms) {
1112                         /* the kthread api has blocked signals for us so no
1113                          * need to record the return value. */
1114                         msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1115                 }
1116         }
1117
1118         o2hb_disarm_write_timeout(reg);
1119
1120         /* unclean stop is only used in very bad situation */
1121         for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1122                 o2hb_shutdown_slot(&reg->hr_slots[i]);
1123
1124         /* Explicit down notification - avoid forcing the other nodes
1125          * to timeout on this region when we could just as easily
1126          * write a clear generation - thus indicating to them that
1127          * this node has left this region.
1128          *
1129          * XXX: Should we skip this on unclean_stop? */
1130         o2hb_prepare_block(reg, 0);
1131         ret = o2hb_issue_node_write(reg, &write_wc);
1132         if (ret == 0) {
1133                 o2hb_wait_on_io(reg, &write_wc);
1134         } else {
1135                 mlog_errno(ret);
1136         }
1137
1138         /* Unpin node */
1139         o2nm_undepend_this_node();
1140
1141         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n");
1142
1143         return 0;
1144 }
1145
1146 #ifdef CONFIG_DEBUG_FS
1147 static int o2hb_debug_open(struct inode *inode, struct file *file)
1148 {
1149         struct o2hb_debug_buf *db = inode->i_private;
1150         struct o2hb_region *reg;
1151         unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1152         char *buf = NULL;
1153         int i = -1;
1154         int out = 0;
1155
1156         /* max_nodes should be the largest bitmap we pass here */
1157         BUG_ON(sizeof(map) < db->db_size);
1158
1159         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1160         if (!buf)
1161                 goto bail;
1162
1163         switch (db->db_type) {
1164         case O2HB_DB_TYPE_LIVENODES:
1165         case O2HB_DB_TYPE_LIVEREGIONS:
1166         case O2HB_DB_TYPE_QUORUMREGIONS:
1167         case O2HB_DB_TYPE_FAILEDREGIONS:
1168                 spin_lock(&o2hb_live_lock);
1169                 memcpy(map, db->db_data, db->db_size);
1170                 spin_unlock(&o2hb_live_lock);
1171                 break;
1172
1173         case O2HB_DB_TYPE_REGION_LIVENODES:
1174                 spin_lock(&o2hb_live_lock);
1175                 reg = (struct o2hb_region *)db->db_data;
1176                 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1177                 spin_unlock(&o2hb_live_lock);
1178                 break;
1179
1180         case O2HB_DB_TYPE_REGION_NUMBER:
1181                 reg = (struct o2hb_region *)db->db_data;
1182                 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1183                                 reg->hr_region_num);
1184                 goto done;
1185
1186         case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1187                 reg = (struct o2hb_region *)db->db_data;
1188                 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1189                                 jiffies_to_msecs(jiffies -
1190                                                  reg->hr_last_timeout_start));
1191                 goto done;
1192
1193         case O2HB_DB_TYPE_REGION_PINNED:
1194                 reg = (struct o2hb_region *)db->db_data;
1195                 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1196                                 !!reg->hr_item_pinned);
1197                 goto done;
1198
1199         default:
1200                 goto done;
1201         }
1202
1203         while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1204                 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1205         out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1206
1207 done:
1208         i_size_write(inode, out);
1209
1210         file->private_data = buf;
1211
1212         return 0;
1213 bail:
1214         return -ENOMEM;
1215 }
1216
1217 static int o2hb_debug_release(struct inode *inode, struct file *file)
1218 {
1219         kfree(file->private_data);
1220         return 0;
1221 }
1222
1223 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1224                                  size_t nbytes, loff_t *ppos)
1225 {
1226         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1227                                        i_size_read(file->f_mapping->host));
1228 }
1229 #else
1230 static int o2hb_debug_open(struct inode *inode, struct file *file)
1231 {
1232         return 0;
1233 }
1234 static int o2hb_debug_release(struct inode *inode, struct file *file)
1235 {
1236         return 0;
1237 }
1238 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1239                                size_t nbytes, loff_t *ppos)
1240 {
1241         return 0;
1242 }
1243 #endif  /* CONFIG_DEBUG_FS */
1244
1245 static const struct file_operations o2hb_debug_fops = {
1246         .open =         o2hb_debug_open,
1247         .release =      o2hb_debug_release,
1248         .read =         o2hb_debug_read,
1249         .llseek =       generic_file_llseek,
1250 };
1251
1252 void o2hb_exit(void)
1253 {
1254         kfree(o2hb_db_livenodes);
1255         kfree(o2hb_db_liveregions);
1256         kfree(o2hb_db_quorumregions);
1257         kfree(o2hb_db_failedregions);
1258         debugfs_remove(o2hb_debug_failedregions);
1259         debugfs_remove(o2hb_debug_quorumregions);
1260         debugfs_remove(o2hb_debug_liveregions);
1261         debugfs_remove(o2hb_debug_livenodes);
1262         debugfs_remove(o2hb_debug_dir);
1263 }
1264
1265 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1266                                         struct o2hb_debug_buf **db, int db_len,
1267                                         int type, int size, int len, void *data)
1268 {
1269         *db = kmalloc(db_len, GFP_KERNEL);
1270         if (!*db)
1271                 return NULL;
1272
1273         (*db)->db_type = type;
1274         (*db)->db_size = size;
1275         (*db)->db_len = len;
1276         (*db)->db_data = data;
1277
1278         return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1279                                    &o2hb_debug_fops);
1280 }
1281
1282 static int o2hb_debug_init(void)
1283 {
1284         int ret = -ENOMEM;
1285
1286         o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1287         if (!o2hb_debug_dir) {
1288                 mlog_errno(ret);
1289                 goto bail;
1290         }
1291
1292         o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1293                                                  o2hb_debug_dir,
1294                                                  &o2hb_db_livenodes,
1295                                                  sizeof(*o2hb_db_livenodes),
1296                                                  O2HB_DB_TYPE_LIVENODES,
1297                                                  sizeof(o2hb_live_node_bitmap),
1298                                                  O2NM_MAX_NODES,
1299                                                  o2hb_live_node_bitmap);
1300         if (!o2hb_debug_livenodes) {
1301                 mlog_errno(ret);
1302                 goto bail;
1303         }
1304
1305         o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1306                                                    o2hb_debug_dir,
1307                                                    &o2hb_db_liveregions,
1308                                                    sizeof(*o2hb_db_liveregions),
1309                                                    O2HB_DB_TYPE_LIVEREGIONS,
1310                                                    sizeof(o2hb_live_region_bitmap),
1311                                                    O2NM_MAX_REGIONS,
1312                                                    o2hb_live_region_bitmap);
1313         if (!o2hb_debug_liveregions) {
1314                 mlog_errno(ret);
1315                 goto bail;
1316         }
1317
1318         o2hb_debug_quorumregions =
1319                         o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1320                                           o2hb_debug_dir,
1321                                           &o2hb_db_quorumregions,
1322                                           sizeof(*o2hb_db_quorumregions),
1323                                           O2HB_DB_TYPE_QUORUMREGIONS,
1324                                           sizeof(o2hb_quorum_region_bitmap),
1325                                           O2NM_MAX_REGIONS,
1326                                           o2hb_quorum_region_bitmap);
1327         if (!o2hb_debug_quorumregions) {
1328                 mlog_errno(ret);
1329                 goto bail;
1330         }
1331
1332         o2hb_debug_failedregions =
1333                         o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1334                                           o2hb_debug_dir,
1335                                           &o2hb_db_failedregions,
1336                                           sizeof(*o2hb_db_failedregions),
1337                                           O2HB_DB_TYPE_FAILEDREGIONS,
1338                                           sizeof(o2hb_failed_region_bitmap),
1339                                           O2NM_MAX_REGIONS,
1340                                           o2hb_failed_region_bitmap);
1341         if (!o2hb_debug_failedregions) {
1342                 mlog_errno(ret);
1343                 goto bail;
1344         }
1345
1346         ret = 0;
1347 bail:
1348         if (ret)
1349                 o2hb_exit();
1350
1351         return ret;
1352 }
1353
1354 int o2hb_init(void)
1355 {
1356         int i;
1357
1358         for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1359                 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1360
1361         for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1362                 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1363
1364         INIT_LIST_HEAD(&o2hb_node_events);
1365
1366         memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1367         memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1368         memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1369         memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1370         memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1371
1372         o2hb_dependent_users = 0;
1373
1374         return o2hb_debug_init();
1375 }
1376
1377 /* if we're already in a callback then we're already serialized by the sem */
1378 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1379                                              unsigned bytes)
1380 {
1381         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1382
1383         memcpy(map, &o2hb_live_node_bitmap, bytes);
1384 }
1385
1386 /*
1387  * get a map of all nodes that are heartbeating in any regions
1388  */
1389 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1390 {
1391         /* callers want to serialize this map and callbacks so that they
1392          * can trust that they don't miss nodes coming to the party */
1393         down_read(&o2hb_callback_sem);
1394         spin_lock(&o2hb_live_lock);
1395         o2hb_fill_node_map_from_callback(map, bytes);
1396         spin_unlock(&o2hb_live_lock);
1397         up_read(&o2hb_callback_sem);
1398 }
1399 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1400
1401 /*
1402  * heartbeat configfs bits.  The heartbeat set is a default set under
1403  * the cluster set in nodemanager.c.
1404  */
1405
1406 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1407 {
1408         return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1409 }
1410
1411 /* drop_item only drops its ref after killing the thread, nothing should
1412  * be using the region anymore.  this has to clean up any state that
1413  * attributes might have built up. */
1414 static void o2hb_region_release(struct config_item *item)
1415 {
1416         int i;
1417         struct page *page;
1418         struct o2hb_region *reg = to_o2hb_region(item);
1419
1420         if (reg->hr_tmp_block)
1421                 kfree(reg->hr_tmp_block);
1422
1423         if (reg->hr_slot_data) {
1424                 for (i = 0; i < reg->hr_num_pages; i++) {
1425                         page = reg->hr_slot_data[i];
1426                         if (page)
1427                                 __free_page(page);
1428                 }
1429                 kfree(reg->hr_slot_data);
1430         }
1431
1432         if (reg->hr_bdev)
1433                 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1434
1435         if (reg->hr_slots)
1436                 kfree(reg->hr_slots);
1437
1438         kfree(reg->hr_db_regnum);
1439         kfree(reg->hr_db_livenodes);
1440         debugfs_remove(reg->hr_debug_livenodes);
1441         debugfs_remove(reg->hr_debug_regnum);
1442         debugfs_remove(reg->hr_debug_elapsed_time);
1443         debugfs_remove(reg->hr_debug_pinned);
1444         debugfs_remove(reg->hr_debug_dir);
1445
1446         spin_lock(&o2hb_live_lock);
1447         list_del(&reg->hr_all_item);
1448         spin_unlock(&o2hb_live_lock);
1449
1450         kfree(reg);
1451 }
1452
1453 static int o2hb_read_block_input(struct o2hb_region *reg,
1454                                  const char *page,
1455                                  size_t count,
1456                                  unsigned long *ret_bytes,
1457                                  unsigned int *ret_bits)
1458 {
1459         unsigned long bytes;
1460         char *p = (char *)page;
1461
1462         bytes = simple_strtoul(p, &p, 0);
1463         if (!p || (*p && (*p != '\n')))
1464                 return -EINVAL;
1465
1466         /* Heartbeat and fs min / max block sizes are the same. */
1467         if (bytes > 4096 || bytes < 512)
1468                 return -ERANGE;
1469         if (hweight16(bytes) != 1)
1470                 return -EINVAL;
1471
1472         if (ret_bytes)
1473                 *ret_bytes = bytes;
1474         if (ret_bits)
1475                 *ret_bits = ffs(bytes) - 1;
1476
1477         return 0;
1478 }
1479
1480 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1481                                             char *page)
1482 {
1483         return sprintf(page, "%u\n", reg->hr_block_bytes);
1484 }
1485
1486 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1487                                              const char *page,
1488                                              size_t count)
1489 {
1490         int status;
1491         unsigned long block_bytes;
1492         unsigned int block_bits;
1493
1494         if (reg->hr_bdev)
1495                 return -EINVAL;
1496
1497         status = o2hb_read_block_input(reg, page, count,
1498                                        &block_bytes, &block_bits);
1499         if (status)
1500                 return status;
1501
1502         reg->hr_block_bytes = (unsigned int)block_bytes;
1503         reg->hr_block_bits = block_bits;
1504
1505         return count;
1506 }
1507
1508 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1509                                             char *page)
1510 {
1511         return sprintf(page, "%llu\n", reg->hr_start_block);
1512 }
1513
1514 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1515                                              const char *page,
1516                                              size_t count)
1517 {
1518         unsigned long long tmp;
1519         char *p = (char *)page;
1520
1521         if (reg->hr_bdev)
1522                 return -EINVAL;
1523
1524         tmp = simple_strtoull(p, &p, 0);
1525         if (!p || (*p && (*p != '\n')))
1526                 return -EINVAL;
1527
1528         reg->hr_start_block = tmp;
1529
1530         return count;
1531 }
1532
1533 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1534                                        char *page)
1535 {
1536         return sprintf(page, "%d\n", reg->hr_blocks);
1537 }
1538
1539 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1540                                         const char *page,
1541                                         size_t count)
1542 {
1543         unsigned long tmp;
1544         char *p = (char *)page;
1545
1546         if (reg->hr_bdev)
1547                 return -EINVAL;
1548
1549         tmp = simple_strtoul(p, &p, 0);
1550         if (!p || (*p && (*p != '\n')))
1551                 return -EINVAL;
1552
1553         if (tmp > O2NM_MAX_NODES || tmp == 0)
1554                 return -ERANGE;
1555
1556         reg->hr_blocks = (unsigned int)tmp;
1557
1558         return count;
1559 }
1560
1561 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1562                                     char *page)
1563 {
1564         unsigned int ret = 0;
1565
1566         if (reg->hr_bdev)
1567                 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1568
1569         return ret;
1570 }
1571
1572 static void o2hb_init_region_params(struct o2hb_region *reg)
1573 {
1574         reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1575         reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1576
1577         mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1578              reg->hr_start_block, reg->hr_blocks);
1579         mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1580              reg->hr_block_bytes, reg->hr_block_bits);
1581         mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1582         mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1583 }
1584
1585 static int o2hb_map_slot_data(struct o2hb_region *reg)
1586 {
1587         int i, j;
1588         unsigned int last_slot;
1589         unsigned int spp = reg->hr_slots_per_page;
1590         struct page *page;
1591         char *raw;
1592         struct o2hb_disk_slot *slot;
1593
1594         reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1595         if (reg->hr_tmp_block == NULL) {
1596                 mlog_errno(-ENOMEM);
1597                 return -ENOMEM;
1598         }
1599
1600         reg->hr_slots = kcalloc(reg->hr_blocks,
1601                                 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1602         if (reg->hr_slots == NULL) {
1603                 mlog_errno(-ENOMEM);
1604                 return -ENOMEM;
1605         }
1606
1607         for(i = 0; i < reg->hr_blocks; i++) {
1608                 slot = &reg->hr_slots[i];
1609                 slot->ds_node_num = i;
1610                 INIT_LIST_HEAD(&slot->ds_live_item);
1611                 slot->ds_raw_block = NULL;
1612         }
1613
1614         reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1615         mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1616                            "at %u blocks per page\n",
1617              reg->hr_num_pages, reg->hr_blocks, spp);
1618
1619         reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1620                                     GFP_KERNEL);
1621         if (!reg->hr_slot_data) {
1622                 mlog_errno(-ENOMEM);
1623                 return -ENOMEM;
1624         }
1625
1626         for(i = 0; i < reg->hr_num_pages; i++) {
1627                 page = alloc_page(GFP_KERNEL);
1628                 if (!page) {
1629                         mlog_errno(-ENOMEM);
1630                         return -ENOMEM;
1631                 }
1632
1633                 reg->hr_slot_data[i] = page;
1634
1635                 last_slot = i * spp;
1636                 raw = page_address(page);
1637                 for (j = 0;
1638                      (j < spp) && ((j + last_slot) < reg->hr_blocks);
1639                      j++) {
1640                         BUG_ON((j + last_slot) >= reg->hr_blocks);
1641
1642                         slot = &reg->hr_slots[j + last_slot];
1643                         slot->ds_raw_block =
1644                                 (struct o2hb_disk_heartbeat_block *) raw;
1645
1646                         raw += reg->hr_block_bytes;
1647                 }
1648         }
1649
1650         return 0;
1651 }
1652
1653 /* Read in all the slots available and populate the tracking
1654  * structures so that we can start with a baseline idea of what's
1655  * there. */
1656 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1657 {
1658         int ret, i;
1659         struct o2hb_disk_slot *slot;
1660         struct o2hb_disk_heartbeat_block *hb_block;
1661
1662         mlog_entry_void();
1663
1664         ret = o2hb_read_slots(reg, reg->hr_blocks);
1665         if (ret) {
1666                 mlog_errno(ret);
1667                 goto out;
1668         }
1669
1670         /* We only want to get an idea of the values initially in each
1671          * slot, so we do no verification - o2hb_check_slot will
1672          * actually determine if each configured slot is valid and
1673          * whether any values have changed. */
1674         for(i = 0; i < reg->hr_blocks; i++) {
1675                 slot = &reg->hr_slots[i];
1676                 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1677
1678                 /* Only fill the values that o2hb_check_slot uses to
1679                  * determine changing slots */
1680                 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1681                 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1682         }
1683
1684 out:
1685         mlog_exit(ret);
1686         return ret;
1687 }
1688
1689 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1690 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1691                                      const char *page,
1692                                      size_t count)
1693 {
1694         struct task_struct *hb_task;
1695         long fd;
1696         int sectsize;
1697         char *p = (char *)page;
1698         struct file *filp = NULL;
1699         struct inode *inode = NULL;
1700         ssize_t ret = -EINVAL;
1701
1702         if (reg->hr_bdev)
1703                 goto out;
1704
1705         /* We can't heartbeat without having had our node number
1706          * configured yet. */
1707         if (o2nm_this_node() == O2NM_MAX_NODES)
1708                 goto out;
1709
1710         fd = simple_strtol(p, &p, 0);
1711         if (!p || (*p && (*p != '\n')))
1712                 goto out;
1713
1714         if (fd < 0 || fd >= INT_MAX)
1715                 goto out;
1716
1717         filp = fget(fd);
1718         if (filp == NULL)
1719                 goto out;
1720
1721         if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1722             reg->hr_block_bytes == 0)
1723                 goto out;
1724
1725         inode = igrab(filp->f_mapping->host);
1726         if (inode == NULL)
1727                 goto out;
1728
1729         if (!S_ISBLK(inode->i_mode))
1730                 goto out;
1731
1732         reg->hr_bdev = I_BDEV(filp->f_mapping->host);
1733         ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ);
1734         if (ret) {
1735                 reg->hr_bdev = NULL;
1736                 goto out;
1737         }
1738         inode = NULL;
1739
1740         bdevname(reg->hr_bdev, reg->hr_dev_name);
1741
1742         sectsize = bdev_logical_block_size(reg->hr_bdev);
1743         if (sectsize != reg->hr_block_bytes) {
1744                 mlog(ML_ERROR,
1745                      "blocksize %u incorrect for device, expected %d",
1746                      reg->hr_block_bytes, sectsize);
1747                 ret = -EINVAL;
1748                 goto out;
1749         }
1750
1751         o2hb_init_region_params(reg);
1752
1753         /* Generation of zero is invalid */
1754         do {
1755                 get_random_bytes(&reg->hr_generation,
1756                                  sizeof(reg->hr_generation));
1757         } while (reg->hr_generation == 0);
1758
1759         ret = o2hb_map_slot_data(reg);
1760         if (ret) {
1761                 mlog_errno(ret);
1762                 goto out;
1763         }
1764
1765         ret = o2hb_populate_slot_data(reg);
1766         if (ret) {
1767                 mlog_errno(ret);
1768                 goto out;
1769         }
1770
1771         INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1772
1773         /*
1774          * A node is considered live after it has beat LIVE_THRESHOLD
1775          * times.  We're not steady until we've given them a chance
1776          * _after_ our first read.
1777          */
1778         atomic_set(&reg->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1);
1779
1780         hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1781                               reg->hr_item.ci_name);
1782         if (IS_ERR(hb_task)) {
1783                 ret = PTR_ERR(hb_task);
1784                 mlog_errno(ret);
1785                 goto out;
1786         }
1787
1788         spin_lock(&o2hb_live_lock);
1789         reg->hr_task = hb_task;
1790         spin_unlock(&o2hb_live_lock);
1791
1792         ret = wait_event_interruptible(o2hb_steady_queue,
1793                                 atomic_read(&reg->hr_steady_iterations) == 0);
1794         if (ret) {
1795                 /* We got interrupted (hello ptrace!).  Clean up */
1796                 spin_lock(&o2hb_live_lock);
1797                 hb_task = reg->hr_task;
1798                 reg->hr_task = NULL;
1799                 spin_unlock(&o2hb_live_lock);
1800
1801                 if (hb_task)
1802                         kthread_stop(hb_task);
1803                 goto out;
1804         }
1805
1806         /* Ok, we were woken.  Make sure it wasn't by drop_item() */
1807         spin_lock(&o2hb_live_lock);
1808         hb_task = reg->hr_task;
1809         if (o2hb_global_heartbeat_active())
1810                 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1811         spin_unlock(&o2hb_live_lock);
1812
1813         if (hb_task)
1814                 ret = count;
1815         else
1816                 ret = -EIO;
1817
1818         if (hb_task && o2hb_global_heartbeat_active())
1819                 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s\n",
1820                        config_item_name(&reg->hr_item));
1821
1822 out:
1823         if (filp)
1824                 fput(filp);
1825         if (inode)
1826                 iput(inode);
1827         if (ret < 0) {
1828                 if (reg->hr_bdev) {
1829                         blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1830                         reg->hr_bdev = NULL;
1831                 }
1832         }
1833         return ret;
1834 }
1835
1836 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1837                                       char *page)
1838 {
1839         pid_t pid = 0;
1840
1841         spin_lock(&o2hb_live_lock);
1842         if (reg->hr_task)
1843                 pid = task_pid_nr(reg->hr_task);
1844         spin_unlock(&o2hb_live_lock);
1845
1846         if (!pid)
1847                 return 0;
1848
1849         return sprintf(page, "%u\n", pid);
1850 }
1851
1852 struct o2hb_region_attribute {
1853         struct configfs_attribute attr;
1854         ssize_t (*show)(struct o2hb_region *, char *);
1855         ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1856 };
1857
1858 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1859         .attr   = { .ca_owner = THIS_MODULE,
1860                     .ca_name = "block_bytes",
1861                     .ca_mode = S_IRUGO | S_IWUSR },
1862         .show   = o2hb_region_block_bytes_read,
1863         .store  = o2hb_region_block_bytes_write,
1864 };
1865
1866 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1867         .attr   = { .ca_owner = THIS_MODULE,
1868                     .ca_name = "start_block",
1869                     .ca_mode = S_IRUGO | S_IWUSR },
1870         .show   = o2hb_region_start_block_read,
1871         .store  = o2hb_region_start_block_write,
1872 };
1873
1874 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1875         .attr   = { .ca_owner = THIS_MODULE,
1876                     .ca_name = "blocks",
1877                     .ca_mode = S_IRUGO | S_IWUSR },
1878         .show   = o2hb_region_blocks_read,
1879         .store  = o2hb_region_blocks_write,
1880 };
1881
1882 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1883         .attr   = { .ca_owner = THIS_MODULE,
1884                     .ca_name = "dev",
1885                     .ca_mode = S_IRUGO | S_IWUSR },
1886         .show   = o2hb_region_dev_read,
1887         .store  = o2hb_region_dev_write,
1888 };
1889
1890 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1891        .attr   = { .ca_owner = THIS_MODULE,
1892                    .ca_name = "pid",
1893                    .ca_mode = S_IRUGO | S_IRUSR },
1894        .show   = o2hb_region_pid_read,
1895 };
1896
1897 static struct configfs_attribute *o2hb_region_attrs[] = {
1898         &o2hb_region_attr_block_bytes.attr,
1899         &o2hb_region_attr_start_block.attr,
1900         &o2hb_region_attr_blocks.attr,
1901         &o2hb_region_attr_dev.attr,
1902         &o2hb_region_attr_pid.attr,
1903         NULL,
1904 };
1905
1906 static ssize_t o2hb_region_show(struct config_item *item,
1907                                 struct configfs_attribute *attr,
1908                                 char *page)
1909 {
1910         struct o2hb_region *reg = to_o2hb_region(item);
1911         struct o2hb_region_attribute *o2hb_region_attr =
1912                 container_of(attr, struct o2hb_region_attribute, attr);
1913         ssize_t ret = 0;
1914
1915         if (o2hb_region_attr->show)
1916                 ret = o2hb_region_attr->show(reg, page);
1917         return ret;
1918 }
1919
1920 static ssize_t o2hb_region_store(struct config_item *item,
1921                                  struct configfs_attribute *attr,
1922                                  const char *page, size_t count)
1923 {
1924         struct o2hb_region *reg = to_o2hb_region(item);
1925         struct o2hb_region_attribute *o2hb_region_attr =
1926                 container_of(attr, struct o2hb_region_attribute, attr);
1927         ssize_t ret = -EINVAL;
1928
1929         if (o2hb_region_attr->store)
1930                 ret = o2hb_region_attr->store(reg, page, count);
1931         return ret;
1932 }
1933
1934 static struct configfs_item_operations o2hb_region_item_ops = {
1935         .release                = o2hb_region_release,
1936         .show_attribute         = o2hb_region_show,
1937         .store_attribute        = o2hb_region_store,
1938 };
1939
1940 static struct config_item_type o2hb_region_type = {
1941         .ct_item_ops    = &o2hb_region_item_ops,
1942         .ct_attrs       = o2hb_region_attrs,
1943         .ct_owner       = THIS_MODULE,
1944 };
1945
1946 /* heartbeat set */
1947
1948 struct o2hb_heartbeat_group {
1949         struct config_group hs_group;
1950         /* some stuff? */
1951 };
1952
1953 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1954 {
1955         return group ?
1956                 container_of(group, struct o2hb_heartbeat_group, hs_group)
1957                 : NULL;
1958 }
1959
1960 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
1961 {
1962         int ret = -ENOMEM;
1963
1964         reg->hr_debug_dir =
1965                 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
1966         if (!reg->hr_debug_dir) {
1967                 mlog_errno(ret);
1968                 goto bail;
1969         }
1970
1971         reg->hr_debug_livenodes =
1972                         o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1973                                           reg->hr_debug_dir,
1974                                           &(reg->hr_db_livenodes),
1975                                           sizeof(*(reg->hr_db_livenodes)),
1976                                           O2HB_DB_TYPE_REGION_LIVENODES,
1977                                           sizeof(reg->hr_live_node_bitmap),
1978                                           O2NM_MAX_NODES, reg);
1979         if (!reg->hr_debug_livenodes) {
1980                 mlog_errno(ret);
1981                 goto bail;
1982         }
1983
1984         reg->hr_debug_regnum =
1985                         o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
1986                                           reg->hr_debug_dir,
1987                                           &(reg->hr_db_regnum),
1988                                           sizeof(*(reg->hr_db_regnum)),
1989                                           O2HB_DB_TYPE_REGION_NUMBER,
1990                                           0, O2NM_MAX_NODES, reg);
1991         if (!reg->hr_debug_regnum) {
1992                 mlog_errno(ret);
1993                 goto bail;
1994         }
1995
1996         reg->hr_debug_elapsed_time =
1997                         o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
1998                                           reg->hr_debug_dir,
1999                                           &(reg->hr_db_elapsed_time),
2000                                           sizeof(*(reg->hr_db_elapsed_time)),
2001                                           O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2002                                           0, 0, reg);
2003         if (!reg->hr_debug_elapsed_time) {
2004                 mlog_errno(ret);
2005                 goto bail;
2006         }
2007
2008         reg->hr_debug_pinned =
2009                         o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2010                                           reg->hr_debug_dir,
2011                                           &(reg->hr_db_pinned),
2012                                           sizeof(*(reg->hr_db_pinned)),
2013                                           O2HB_DB_TYPE_REGION_PINNED,
2014                                           0, 0, reg);
2015         if (!reg->hr_debug_pinned) {
2016                 mlog_errno(ret);
2017                 goto bail;
2018         }
2019
2020         ret = 0;
2021 bail:
2022         return ret;
2023 }
2024
2025 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2026                                                           const char *name)
2027 {
2028         struct o2hb_region *reg = NULL;
2029         int ret;
2030
2031         reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2032         if (reg == NULL)
2033                 return ERR_PTR(-ENOMEM);
2034
2035         if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2036                 ret = -ENAMETOOLONG;
2037                 goto free;
2038         }
2039
2040         spin_lock(&o2hb_live_lock);
2041         reg->hr_region_num = 0;
2042         if (o2hb_global_heartbeat_active()) {
2043                 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2044                                                          O2NM_MAX_REGIONS);
2045                 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2046                         spin_unlock(&o2hb_live_lock);
2047                         ret = -EFBIG;
2048                         goto free;
2049                 }
2050                 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2051         }
2052         list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2053         spin_unlock(&o2hb_live_lock);
2054
2055         config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2056
2057         ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2058         if (ret) {
2059                 config_item_put(&reg->hr_item);
2060                 goto free;
2061         }
2062
2063         return &reg->hr_item;
2064 free:
2065         kfree(reg);
2066         return ERR_PTR(ret);
2067 }
2068
2069 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2070                                            struct config_item *item)
2071 {
2072         struct task_struct *hb_task;
2073         struct o2hb_region *reg = to_o2hb_region(item);
2074         int quorum_region = 0;
2075
2076         /* stop the thread when the user removes the region dir */
2077         spin_lock(&o2hb_live_lock);
2078         if (o2hb_global_heartbeat_active()) {
2079                 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2080                 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2081                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2082                         quorum_region = 1;
2083                 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2084         }
2085         hb_task = reg->hr_task;
2086         reg->hr_task = NULL;
2087         reg->hr_item_dropped = 1;
2088         spin_unlock(&o2hb_live_lock);
2089
2090         if (hb_task)
2091                 kthread_stop(hb_task);
2092
2093         /*
2094          * If we're racing a dev_write(), we need to wake them.  They will
2095          * check reg->hr_task
2096          */
2097         if (atomic_read(&reg->hr_steady_iterations) != 0) {
2098                 atomic_set(&reg->hr_steady_iterations, 0);
2099                 wake_up(&o2hb_steady_queue);
2100         }
2101
2102         if (o2hb_global_heartbeat_active())
2103                 printk(KERN_NOTICE "o2hb: Heartbeat stopped on region %s\n",
2104                        config_item_name(&reg->hr_item));
2105
2106         config_item_put(item);
2107
2108         if (!o2hb_global_heartbeat_active() || !quorum_region)
2109                 return;
2110
2111         /*
2112          * If global heartbeat active and there are dependent users,
2113          * pin all regions if quorum region count <= CUT_OFF
2114          */
2115         spin_lock(&o2hb_live_lock);
2116
2117         if (!o2hb_dependent_users)
2118                 goto unlock;
2119
2120         if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2121                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2122                 o2hb_region_pin(NULL);
2123
2124 unlock:
2125         spin_unlock(&o2hb_live_lock);
2126 }
2127
2128 struct o2hb_heartbeat_group_attribute {
2129         struct configfs_attribute attr;
2130         ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2131         ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2132 };
2133
2134 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2135                                          struct configfs_attribute *attr,
2136                                          char *page)
2137 {
2138         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2139         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2140                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2141         ssize_t ret = 0;
2142
2143         if (o2hb_heartbeat_group_attr->show)
2144                 ret = o2hb_heartbeat_group_attr->show(reg, page);
2145         return ret;
2146 }
2147
2148 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2149                                           struct configfs_attribute *attr,
2150                                           const char *page, size_t count)
2151 {
2152         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2153         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2154                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2155         ssize_t ret = -EINVAL;
2156
2157         if (o2hb_heartbeat_group_attr->store)
2158                 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2159         return ret;
2160 }
2161
2162 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2163                                                      char *page)
2164 {
2165         return sprintf(page, "%u\n", o2hb_dead_threshold);
2166 }
2167
2168 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2169                                                     const char *page,
2170                                                     size_t count)
2171 {
2172         unsigned long tmp;
2173         char *p = (char *)page;
2174
2175         tmp = simple_strtoul(p, &p, 10);
2176         if (!p || (*p && (*p != '\n')))
2177                 return -EINVAL;
2178
2179         /* this will validate ranges for us. */
2180         o2hb_dead_threshold_set((unsigned int) tmp);
2181
2182         return count;
2183 }
2184
2185 static
2186 ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2187                                        char *page)
2188 {
2189         return sprintf(page, "%s\n",
2190                        o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2191 }
2192
2193 static
2194 ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2195                                         const char *page, size_t count)
2196 {
2197         unsigned int i;
2198         int ret;
2199         size_t len;
2200
2201         len = (page[count - 1] == '\n') ? count - 1 : count;
2202         if (!len)
2203                 return -EINVAL;
2204
2205         for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2206                 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2207                         continue;
2208
2209                 ret = o2hb_global_hearbeat_mode_set(i);
2210                 if (!ret)
2211                         printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2212                                o2hb_heartbeat_mode_desc[i]);
2213                 return count;
2214         }
2215
2216         return -EINVAL;
2217
2218 }
2219
2220 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2221         .attr   = { .ca_owner = THIS_MODULE,
2222                     .ca_name = "dead_threshold",
2223                     .ca_mode = S_IRUGO | S_IWUSR },
2224         .show   = o2hb_heartbeat_group_threshold_show,
2225         .store  = o2hb_heartbeat_group_threshold_store,
2226 };
2227
2228 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2229         .attr   = { .ca_owner = THIS_MODULE,
2230                 .ca_name = "mode",
2231                 .ca_mode = S_IRUGO | S_IWUSR },
2232         .show   = o2hb_heartbeat_group_mode_show,
2233         .store  = o2hb_heartbeat_group_mode_store,
2234 };
2235
2236 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2237         &o2hb_heartbeat_group_attr_threshold.attr,
2238         &o2hb_heartbeat_group_attr_mode.attr,
2239         NULL,
2240 };
2241
2242 static struct configfs_item_operations o2hb_hearbeat_group_item_ops = {
2243         .show_attribute         = o2hb_heartbeat_group_show,
2244         .store_attribute        = o2hb_heartbeat_group_store,
2245 };
2246
2247 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2248         .make_item      = o2hb_heartbeat_group_make_item,
2249         .drop_item      = o2hb_heartbeat_group_drop_item,
2250 };
2251
2252 static struct config_item_type o2hb_heartbeat_group_type = {
2253         .ct_group_ops   = &o2hb_heartbeat_group_group_ops,
2254         .ct_item_ops    = &o2hb_hearbeat_group_item_ops,
2255         .ct_attrs       = o2hb_heartbeat_group_attrs,
2256         .ct_owner       = THIS_MODULE,
2257 };
2258
2259 /* this is just here to avoid touching group in heartbeat.h which the
2260  * entire damn world #includes */
2261 struct config_group *o2hb_alloc_hb_set(void)
2262 {
2263         struct o2hb_heartbeat_group *hs = NULL;
2264         struct config_group *ret = NULL;
2265
2266         hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2267         if (hs == NULL)
2268                 goto out;
2269
2270         config_group_init_type_name(&hs->hs_group, "heartbeat",
2271                                     &o2hb_heartbeat_group_type);
2272
2273         ret = &hs->hs_group;
2274 out:
2275         if (ret == NULL)
2276                 kfree(hs);
2277         return ret;
2278 }
2279
2280 void o2hb_free_hb_set(struct config_group *group)
2281 {
2282         struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2283         kfree(hs);
2284 }
2285
2286 /* hb callback registration and issueing */
2287
2288 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2289 {
2290         if (type == O2HB_NUM_CB)
2291                 return ERR_PTR(-EINVAL);
2292
2293         return &o2hb_callbacks[type];
2294 }
2295
2296 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2297                          enum o2hb_callback_type type,
2298                          o2hb_cb_func *func,
2299                          void *data,
2300                          int priority)
2301 {
2302         INIT_LIST_HEAD(&hc->hc_item);
2303         hc->hc_func = func;
2304         hc->hc_data = data;
2305         hc->hc_priority = priority;
2306         hc->hc_type = type;
2307         hc->hc_magic = O2HB_CB_MAGIC;
2308 }
2309 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2310
2311 /*
2312  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2313  * In global heartbeat mode, region_uuid passed is NULL.
2314  *
2315  * In local, we only pin the matching region. In global we pin all the active
2316  * regions.
2317  */
2318 static int o2hb_region_pin(const char *region_uuid)
2319 {
2320         int ret = 0, found = 0;
2321         struct o2hb_region *reg;
2322         char *uuid;
2323
2324         assert_spin_locked(&o2hb_live_lock);
2325
2326         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2327                 uuid = config_item_name(&reg->hr_item);
2328
2329                 /* local heartbeat */
2330                 if (region_uuid) {
2331                         if (strcmp(region_uuid, uuid))
2332                                 continue;
2333                         found = 1;
2334                 }
2335
2336                 if (reg->hr_item_pinned || reg->hr_item_dropped)
2337                         goto skip_pin;
2338
2339                 /* Ignore ENOENT only for local hb (userdlm domain) */
2340                 ret = o2nm_depend_item(&reg->hr_item);
2341                 if (!ret) {
2342                         mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2343                         reg->hr_item_pinned = 1;
2344                 } else {
2345                         if (ret == -ENOENT && found)
2346                                 ret = 0;
2347                         else {
2348                                 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2349                                      uuid, ret);
2350                                 break;
2351                         }
2352                 }
2353 skip_pin:
2354                 if (found)
2355                         break;
2356         }
2357
2358         return ret;
2359 }
2360
2361 /*
2362  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2363  * In global heartbeat mode, region_uuid passed is NULL.
2364  *
2365  * In local, we only unpin the matching region. In global we unpin all the
2366  * active regions.
2367  */
2368 static void o2hb_region_unpin(const char *region_uuid)
2369 {
2370         struct o2hb_region *reg;
2371         char *uuid;
2372         int found = 0;
2373
2374         assert_spin_locked(&o2hb_live_lock);
2375
2376         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2377                 uuid = config_item_name(&reg->hr_item);
2378                 if (region_uuid) {
2379                         if (strcmp(region_uuid, uuid))
2380                                 continue;
2381                         found = 1;
2382                 }
2383
2384                 if (reg->hr_item_pinned) {
2385                         mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2386                         o2nm_undepend_item(&reg->hr_item);
2387                         reg->hr_item_pinned = 0;
2388                 }
2389                 if (found)
2390                         break;
2391         }
2392 }
2393
2394 static int o2hb_region_inc_user(const char *region_uuid)
2395 {
2396         int ret = 0;
2397
2398         spin_lock(&o2hb_live_lock);
2399
2400         /* local heartbeat */
2401         if (!o2hb_global_heartbeat_active()) {
2402             ret = o2hb_region_pin(region_uuid);
2403             goto unlock;
2404         }
2405
2406         /*
2407          * if global heartbeat active and this is the first dependent user,
2408          * pin all regions if quorum region count <= CUT_OFF
2409          */
2410         o2hb_dependent_users++;
2411         if (o2hb_dependent_users > 1)
2412                 goto unlock;
2413
2414         if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2415                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2416                 ret = o2hb_region_pin(NULL);
2417
2418 unlock:
2419         spin_unlock(&o2hb_live_lock);
2420         return ret;
2421 }
2422
2423 void o2hb_region_dec_user(const char *region_uuid)
2424 {
2425         spin_lock(&o2hb_live_lock);
2426
2427         /* local heartbeat */
2428         if (!o2hb_global_heartbeat_active()) {
2429             o2hb_region_unpin(region_uuid);
2430             goto unlock;
2431         }
2432
2433         /*
2434          * if global heartbeat active and there are no dependent users,
2435          * unpin all quorum regions
2436          */
2437         o2hb_dependent_users--;
2438         if (!o2hb_dependent_users)
2439                 o2hb_region_unpin(NULL);
2440
2441 unlock:
2442         spin_unlock(&o2hb_live_lock);
2443 }
2444
2445 int o2hb_register_callback(const char *region_uuid,
2446                            struct o2hb_callback_func *hc)
2447 {
2448         struct o2hb_callback_func *tmp;
2449         struct list_head *iter;
2450         struct o2hb_callback *hbcall;
2451         int ret;
2452
2453         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2454         BUG_ON(!list_empty(&hc->hc_item));
2455
2456         hbcall = hbcall_from_type(hc->hc_type);
2457         if (IS_ERR(hbcall)) {
2458                 ret = PTR_ERR(hbcall);
2459                 goto out;
2460         }
2461
2462         if (region_uuid) {
2463                 ret = o2hb_region_inc_user(region_uuid);
2464                 if (ret) {
2465                         mlog_errno(ret);
2466                         goto out;
2467                 }
2468         }
2469
2470         down_write(&o2hb_callback_sem);
2471
2472         list_for_each(iter, &hbcall->list) {
2473                 tmp = list_entry(iter, struct o2hb_callback_func, hc_item);
2474                 if (hc->hc_priority < tmp->hc_priority) {
2475                         list_add_tail(&hc->hc_item, iter);
2476                         break;
2477                 }
2478         }
2479         if (list_empty(&hc->hc_item))
2480                 list_add_tail(&hc->hc_item, &hbcall->list);
2481
2482         up_write(&o2hb_callback_sem);
2483         ret = 0;
2484 out:
2485         mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2486              ret, __builtin_return_address(0), hc);
2487         return ret;
2488 }
2489 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2490
2491 void o2hb_unregister_callback(const char *region_uuid,
2492                               struct o2hb_callback_func *hc)
2493 {
2494         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2495
2496         mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2497              __builtin_return_address(0), hc);
2498
2499         /* XXX Can this happen _with_ a region reference? */
2500         if (list_empty(&hc->hc_item))
2501                 return;
2502
2503         if (region_uuid)
2504                 o2hb_region_dec_user(region_uuid);
2505
2506         down_write(&o2hb_callback_sem);
2507
2508         list_del_init(&hc->hc_item);
2509
2510         up_write(&o2hb_callback_sem);
2511 }
2512 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2513
2514 int o2hb_check_node_heartbeating(u8 node_num)
2515 {
2516         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2517
2518         o2hb_fill_node_map(testing_map, sizeof(testing_map));
2519         if (!test_bit(node_num, testing_map)) {
2520                 mlog(ML_HEARTBEAT,
2521                      "node (%u) does not have heartbeating enabled.\n",
2522                      node_num);
2523                 return 0;
2524         }
2525
2526         return 1;
2527 }
2528 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2529
2530 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2531 {
2532         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2533
2534         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2535         if (!test_bit(node_num, testing_map)) {
2536                 mlog(ML_HEARTBEAT,
2537                      "node (%u) does not have heartbeating enabled.\n",
2538                      node_num);
2539                 return 0;
2540         }
2541
2542         return 1;
2543 }
2544 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2545
2546 /* Makes sure our local node is configured with a node number, and is
2547  * heartbeating. */
2548 int o2hb_check_local_node_heartbeating(void)
2549 {
2550         u8 node_num;
2551
2552         /* if this node was set then we have networking */
2553         node_num = o2nm_this_node();
2554         if (node_num == O2NM_MAX_NODES) {
2555                 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2556                 return 0;
2557         }
2558
2559         return o2hb_check_node_heartbeating(node_num);
2560 }
2561 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2562
2563 /*
2564  * this is just a hack until we get the plumbing which flips file systems
2565  * read only and drops the hb ref instead of killing the node dead.
2566  */
2567 void o2hb_stop_all_regions(void)
2568 {
2569         struct o2hb_region *reg;
2570
2571         mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2572
2573         spin_lock(&o2hb_live_lock);
2574
2575         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2576                 reg->hr_unclean_stop = 1;
2577
2578         spin_unlock(&o2hb_live_lock);
2579 }
2580 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2581
2582 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2583 {
2584         struct o2hb_region *reg;
2585         int numregs = 0;
2586         char *p;
2587
2588         spin_lock(&o2hb_live_lock);
2589
2590         p = region_uuids;
2591         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2592                 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2593                 if (numregs < max_regions) {
2594                         memcpy(p, config_item_name(&reg->hr_item),
2595                                O2HB_MAX_REGION_NAME_LEN);
2596                         p += O2HB_MAX_REGION_NAME_LEN;
2597                 }
2598                 numregs++;
2599         }
2600
2601         spin_unlock(&o2hb_live_lock);
2602
2603         return numregs;
2604 }
2605 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2606
2607 int o2hb_global_heartbeat_active(void)
2608 {
2609         return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2610 }
2611 EXPORT_SYMBOL(o2hb_global_heartbeat_active);