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_heartbeat_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_aborted_start:1,
 220                                 hr_item_pinned:1,
 221                                 hr_item_dropped:1;
 222
 223         /* protected by the hr_callback_sem */
 224         struct task_struct      *hr_task;
 225
 226         unsigned int            hr_blocks;
 227         unsigned long long      hr_start_block;
 228
 229         unsigned int            hr_block_bits;
 230         unsigned int            hr_block_bytes;
 231
 232         unsigned int            hr_slots_per_page;
 233         unsigned int            hr_num_pages;
 234
 235         struct page             **hr_slot_data;
 236         struct block_device     *hr_bdev;
 237         struct o2hb_disk_slot   *hr_slots;
 238
 239         /* live node map of this region */
 240         unsigned long           hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 241         unsigned int            hr_region_num;
 242
 243         struct dentry           *hr_debug_dir;
 244         struct dentry           *hr_debug_livenodes;
 245         struct dentry           *hr_debug_regnum;
 246         struct dentry           *hr_debug_elapsed_time;
 247         struct dentry           *hr_debug_pinned;
 248         struct o2hb_debug_buf   *hr_db_livenodes;
 249         struct o2hb_debug_buf   *hr_db_regnum;
 250         struct o2hb_debug_buf   *hr_db_elapsed_time;
 251         struct o2hb_debug_buf   *hr_db_pinned;
 252
 253         /* let the person setting up hb wait for it to return until it
 254          * has reached a 'steady' state.  This will be fixed when we have
 255          * a more complete api that doesn't lead to this sort of fragility. */
 256         atomic_t                hr_steady_iterations;
 257
 258         /* terminate o2hb thread if it does not reach steady state
 259          * (hr_steady_iterations == 0) within hr_unsteady_iterations */
 260         atomic_t                hr_unsteady_iterations;
 261
 262         char                    hr_dev_name[BDEVNAME_SIZE];
 263
 264         unsigned int            hr_timeout_ms;
 265
 266         /* randomized as the region goes up and down so that a node
 267          * recognizes a node going up and down in one iteration */
 268         u64                     hr_generation;
 269
 270         struct delayed_work     hr_write_timeout_work;
 271         unsigned long           hr_last_timeout_start;
 272
 273         /* Used during o2hb_check_slot to hold a copy of the block
 274          * being checked because we temporarily have to zero out the
 275          * crc field. */
 276         struct o2hb_disk_heartbeat_block *hr_tmp_block;
 277 };
 278
 279 struct o2hb_bio_wait_ctxt {
 280         atomic_t          wc_num_reqs;
 281         struct completion wc_io_complete;
 282         int               wc_error;
 283 };
 284
 285 static int o2hb_pop_count(void *map, int count)
 286 {
 287         int i = -1, pop = 0;
 288
 289         while ((i = find_next_bit(map, count, i + 1)) < count)
 290                 pop++;
 291         return pop;
 292 }
 293
 294 static void o2hb_write_timeout(struct work_struct *work)
 295 {
 296         int failed, quorum;
 297         unsigned long flags;
 298         struct o2hb_region *reg =
 299                 container_of(work, struct o2hb_region,
 300                              hr_write_timeout_work.work);
 301
 302         mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
 303              "milliseconds\n", reg->hr_dev_name,
 304              jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
 305
 306         if (o2hb_global_heartbeat_active()) {
 307                 spin_lock_irqsave(&o2hb_live_lock, flags);
 308                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 309                         set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 310                 failed = o2hb_pop_count(&o2hb_failed_region_bitmap,
 311                                         O2NM_MAX_REGIONS);
 312                 quorum = o2hb_pop_count(&o2hb_quorum_region_bitmap,
 313                                         O2NM_MAX_REGIONS);
 314                 spin_unlock_irqrestore(&o2hb_live_lock, flags);
 315
 316                 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
 317                      quorum, failed);
 318
 319                 /*
 320                  * Fence if the number of failed regions >= half the number
 321                  * of  quorum regions
 322                  */
 323                 if ((failed << 1) < quorum)
 324                         return;
 325         }
 326
 327         o2quo_disk_timeout();
 328 }
 329
 330 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
 331 {
 332         /* Arm writeout only after thread reaches steady state */
 333         if (atomic_read(&reg->hr_steady_iterations) != 0)
 334                 return;
 335
 336         mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
 337              O2HB_MAX_WRITE_TIMEOUT_MS);
 338
 339         if (o2hb_global_heartbeat_active()) {
 340                 spin_lock(&o2hb_live_lock);
 341                 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 342                 spin_unlock(&o2hb_live_lock);
 343         }
 344         cancel_delayed_work(&reg->hr_write_timeout_work);
 345         reg->hr_last_timeout_start = jiffies;
 346         schedule_delayed_work(&reg->hr_write_timeout_work,
 347                               msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
 348 }
 349
 350 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
 351 {
 352         cancel_delayed_work_sync(&reg->hr_write_timeout_work);
 353 }
 354
 355 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
 356 {
 357         atomic_set(&wc->wc_num_reqs, 1);
 358         init_completion(&wc->wc_io_complete);
 359         wc->wc_error = 0;
 360 }
 361
 362 /* Used in error paths too */
 363 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
 364                                      unsigned int num)
 365 {
 366         /* sadly atomic_sub_and_test() isn't available on all platforms.  The
 367          * good news is that the fast path only completes one at a time */
 368         while(num--) {
 369                 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
 370                         BUG_ON(num > 0);
 371                         complete(&wc->wc_io_complete);
 372                 }
 373         }
 374 }
 375
 376 static void o2hb_wait_on_io(struct o2hb_region *reg,
 377                             struct o2hb_bio_wait_ctxt *wc)
 378 {
 379         o2hb_bio_wait_dec(wc, 1);
 380         wait_for_completion(&wc->wc_io_complete);
 381 }
 382
 383 static void o2hb_bio_end_io(struct bio *bio,
 384                            int error)
 385 {
 386         struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
 387
 388         if (error) {
 389                 mlog(ML_ERROR, "IO Error %d\n", error);
 390                 wc->wc_error = error;
 391         }
 392
 393         o2hb_bio_wait_dec(wc, 1);
 394         bio_put(bio);
 395 }
 396
 397 /* Setup a Bio to cover I/O against num_slots slots starting at
 398  * start_slot. */
 399 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
 400                                       struct o2hb_bio_wait_ctxt *wc,
 401                                       unsigned int *current_slot,
 402                                       unsigned int max_slots)
 403 {
 404         int len, current_page;
 405         unsigned int vec_len, vec_start;
 406         unsigned int bits = reg->hr_block_bits;
 407         unsigned int spp = reg->hr_slots_per_page;
 408         unsigned int cs = *current_slot;
 409         struct bio *bio;
 410         struct page *page;
 411
 412         /* Testing has shown this allocation to take long enough under
 413          * GFP_KERNEL that the local node can get fenced. It would be
 414          * nicest if we could pre-allocate these bios and avoid this
 415          * all together. */
 416         bio = bio_alloc(GFP_ATOMIC, 16);
 417         if (!bio) {
 418                 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
 419                 bio = ERR_PTR(-ENOMEM);
 420                 goto bail;
 421         }
 422
 423         /* Must put everything in 512 byte sectors for the bio... */
 424         bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
 425         bio->bi_bdev = reg->hr_bdev;
 426         bio->bi_private = wc;
 427         bio->bi_end_io = o2hb_bio_end_io;
 428
 429         vec_start = (cs << bits) % PAGE_CACHE_SIZE;
 430         while(cs < max_slots) {
 431                 current_page = cs / spp;
 432                 page = reg->hr_slot_data[current_page];
 433
 434                 vec_len = min(PAGE_CACHE_SIZE - vec_start,
 435                               (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
 436
 437                 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
 438                      current_page, vec_len, vec_start);
 439
 440                 len = bio_add_page(bio, page, vec_len, vec_start);
 441                 if (len != vec_len) break;
 442
 443                 cs += vec_len / (PAGE_CACHE_SIZE/spp);
 444                 vec_start = 0;
 445         }
 446
 447 bail:
 448         *current_slot = cs;
 449         return bio;
 450 }
 451
 452 static int o2hb_read_slots(struct o2hb_region *reg,
 453                            unsigned int max_slots)
 454 {
 455         unsigned int current_slot=0;
 456         int status;
 457         struct o2hb_bio_wait_ctxt wc;
 458         struct bio *bio;
 459
 460         o2hb_bio_wait_init(&wc);
 461
 462         while(current_slot < max_slots) {
 463                 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
 464                 if (IS_ERR(bio)) {
 465                         status = PTR_ERR(bio);
 466                         mlog_errno(status);
 467                         goto bail_and_wait;
 468                 }
 469
 470                 atomic_inc(&wc.wc_num_reqs);
 471                 submit_bio(READ, bio);
 472         }
 473
 474         status = 0;
 475
 476 bail_and_wait:
 477         o2hb_wait_on_io(reg, &wc);
 478         if (wc.wc_error && !status)
 479                 status = wc.wc_error;
 480
 481         return status;
 482 }
 483
 484 static int o2hb_issue_node_write(struct o2hb_region *reg,
 485                                  struct o2hb_bio_wait_ctxt *write_wc)
 486 {
 487         int status;
 488         unsigned int slot;
 489         struct bio *bio;
 490
 491         o2hb_bio_wait_init(write_wc);
 492
 493         slot = o2nm_this_node();
 494
 495         bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
 496         if (IS_ERR(bio)) {
 497                 status = PTR_ERR(bio);
 498                 mlog_errno(status);
 499                 goto bail;
 500         }
 501
 502         atomic_inc(&write_wc->wc_num_reqs);
 503         submit_bio(WRITE_SYNC, bio);
 504
 505         status = 0;
 506 bail:
 507         return status;
 508 }
 509
 510 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
 511                                      struct o2hb_disk_heartbeat_block *hb_block)
 512 {
 513         __le32 old_cksum;
 514         u32 ret;
 515
 516         /* We want to compute the block crc with a 0 value in the
 517          * hb_cksum field. Save it off here and replace after the
 518          * crc. */
 519         old_cksum = hb_block->hb_cksum;
 520         hb_block->hb_cksum = 0;
 521
 522         ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
 523
 524         hb_block->hb_cksum = old_cksum;
 525
 526         return ret;
 527 }
 528
 529 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
 530 {
 531         mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
 532              "cksum = 0x%x, generation 0x%llx\n",
 533              (long long)le64_to_cpu(hb_block->hb_seq),
 534              hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
 535              (long long)le64_to_cpu(hb_block->hb_generation));
 536 }
 537
 538 static int o2hb_verify_crc(struct o2hb_region *reg,
 539                            struct o2hb_disk_heartbeat_block *hb_block)
 540 {
 541         u32 read, computed;
 542
 543         read = le32_to_cpu(hb_block->hb_cksum);
 544         computed = o2hb_compute_block_crc_le(reg, hb_block);
 545
 546         return read == computed;
 547 }
 548
 549 /*
 550  * Compare the slot data with what we wrote in the last iteration.
 551  * If the match fails, print an appropriate error message. This is to
 552  * detect errors like... another node hearting on the same slot,
 553  * flaky device that is losing writes, etc.
 554  * Returns 1 if check succeeds, 0 otherwise.
 555  */
 556 static int o2hb_check_own_slot(struct o2hb_region *reg)
 557 {
 558         struct o2hb_disk_slot *slot;
 559         struct o2hb_disk_heartbeat_block *hb_block;
 560         char *errstr;
 561
 562         slot = &reg->hr_slots[o2nm_this_node()];
 563         /* Don't check on our 1st timestamp */
 564         if (!slot->ds_last_time)
 565                 return 0;
 566
 567         hb_block = slot->ds_raw_block;
 568         if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
 569             le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
 570             hb_block->hb_node == slot->ds_node_num)
 571                 return 1;
 572
 573 #define ERRSTR1         "Another node is heartbeating on device"
 574 #define ERRSTR2         "Heartbeat generation mismatch on device"
 575 #define ERRSTR3         "Heartbeat sequence mismatch on device"
 576
 577         if (hb_block->hb_node != slot->ds_node_num)
 578                 errstr = ERRSTR1;
 579         else if (le64_to_cpu(hb_block->hb_generation) !=
 580                  slot->ds_last_generation)
 581                 errstr = ERRSTR2;
 582         else
 583                 errstr = ERRSTR3;
 584
 585         mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
 586              "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
 587              slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
 588              (unsigned long long)slot->ds_last_time, hb_block->hb_node,
 589              (unsigned long long)le64_to_cpu(hb_block->hb_generation),
 590              (unsigned long long)le64_to_cpu(hb_block->hb_seq));
 591
 592         return 0;
 593 }
 594
 595 static inline void o2hb_prepare_block(struct o2hb_region *reg,
 596                                       u64 generation)
 597 {
 598         int node_num;
 599         u64 cputime;
 600         struct o2hb_disk_slot *slot;
 601         struct o2hb_disk_heartbeat_block *hb_block;
 602
 603         node_num = o2nm_this_node();
 604         slot = &reg->hr_slots[node_num];
 605
 606         hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
 607         memset(hb_block, 0, reg->hr_block_bytes);
 608         /* TODO: time stuff */
 609         cputime = CURRENT_TIME.tv_sec;
 610         if (!cputime)
 611                 cputime = 1;
 612
 613         hb_block->hb_seq = cpu_to_le64(cputime);
 614         hb_block->hb_node = node_num;
 615         hb_block->hb_generation = cpu_to_le64(generation);
 616         hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
 617
 618         /* This step must always happen last! */
 619         hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
 620                                                                    hb_block));
 621
 622         mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
 623              (long long)generation,
 624              le32_to_cpu(hb_block->hb_cksum));
 625 }
 626
 627 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
 628                                 struct o2nm_node *node,
 629                                 int idx)
 630 {
 631         struct o2hb_callback_func *f;
 632
 633         list_for_each_entry(f, &hbcall->list, hc_item) {
 634                 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
 635                 (f->hc_func)(node, idx, f->hc_data);
 636         }
 637 }
 638
 639 /* Will run the list in order until we process the passed event */
 640 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
 641 {
 642         int empty;
 643         struct o2hb_callback *hbcall;
 644         struct o2hb_node_event *event;
 645
 646         spin_lock(&o2hb_live_lock);
 647         empty = list_empty(&queued_event->hn_item);
 648         spin_unlock(&o2hb_live_lock);
 649         if (empty)
 650                 return;
 651
 652         /* Holding callback sem assures we don't alter the callback
 653          * lists when doing this, and serializes ourselves with other
 654          * processes wanting callbacks. */
 655         down_write(&o2hb_callback_sem);
 656
 657         spin_lock(&o2hb_live_lock);
 658         while (!list_empty(&o2hb_node_events)
 659                && !list_empty(&queued_event->hn_item)) {
 660                 event = list_entry(o2hb_node_events.next,
 661                                    struct o2hb_node_event,
 662                                    hn_item);
 663                 list_del_init(&event->hn_item);
 664                 spin_unlock(&o2hb_live_lock);
 665
 666                 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
 667                      event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
 668                      event->hn_node_num);
 669
 670                 hbcall = hbcall_from_type(event->hn_event_type);
 671
 672                 /* We should *never* have gotten on to the list with a
 673                  * bad type... This isn't something that we should try
 674                  * to recover from. */
 675                 BUG_ON(IS_ERR(hbcall));
 676
 677                 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
 678
 679                 spin_lock(&o2hb_live_lock);
 680         }
 681         spin_unlock(&o2hb_live_lock);
 682
 683         up_write(&o2hb_callback_sem);
 684 }
 685
 686 static void o2hb_queue_node_event(struct o2hb_node_event *event,
 687                                   enum o2hb_callback_type type,
 688                                   struct o2nm_node *node,
 689                                   int node_num)
 690 {
 691         assert_spin_locked(&o2hb_live_lock);
 692
 693         BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
 694
 695         event->hn_event_type = type;
 696         event->hn_node = node;
 697         event->hn_node_num = node_num;
 698
 699         mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
 700              type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
 701
 702         list_add_tail(&event->hn_item, &o2hb_node_events);
 703 }
 704
 705 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
 706 {
 707         struct o2hb_node_event event =
 708                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 709         struct o2nm_node *node;
 710
 711         node = o2nm_get_node_by_num(slot->ds_node_num);
 712         if (!node)
 713                 return;
 714
 715         spin_lock(&o2hb_live_lock);
 716         if (!list_empty(&slot->ds_live_item)) {
 717                 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
 718                      slot->ds_node_num);
 719
 720                 list_del_init(&slot->ds_live_item);
 721
 722                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 723                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 724
 725                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
 726                                               slot->ds_node_num);
 727                 }
 728         }
 729         spin_unlock(&o2hb_live_lock);
 730
 731         o2hb_run_event_list(&event);
 732
 733         o2nm_node_put(node);
 734 }
 735
 736 static void o2hb_set_quorum_device(struct o2hb_region *reg)
 737 {
 738         if (!o2hb_global_heartbeat_active())
 739                 return;
 740
 741         /* Prevent race with o2hb_heartbeat_group_drop_item() */
 742         if (kthread_should_stop())
 743                 return;
 744
 745         /* Tag region as quorum only after thread reaches steady state */
 746         if (atomic_read(&reg->hr_steady_iterations) != 0)
 747                 return;
 748
 749         spin_lock(&o2hb_live_lock);
 750
 751         if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 752                 goto unlock;
 753
 754         /*
 755          * A region can be added to the quorum only when it sees all
 756          * live nodes heartbeat on it. In other words, the region has been
 757          * added to all nodes.
 758          */
 759         if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
 760                    sizeof(o2hb_live_node_bitmap)))
 761                 goto unlock;
 762
 763         printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
 764                config_item_name(&reg->hr_item), reg->hr_dev_name);
 765
 766         set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
 767
 768         /*
 769          * If global heartbeat active, unpin all regions if the
 770          * region count > CUT_OFF
 771          */
 772         if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
 773                            O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
 774                 o2hb_region_unpin(NULL);
 775 unlock:
 776         spin_unlock(&o2hb_live_lock);
 777 }
 778
 779 static int o2hb_check_slot(struct o2hb_region *reg,
 780                            struct o2hb_disk_slot *slot)
 781 {
 782         int changed = 0, gen_changed = 0;
 783         struct o2hb_node_event event =
 784                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 785         struct o2nm_node *node;
 786         struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
 787         u64 cputime;
 788         unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
 789         unsigned int slot_dead_ms;
 790         int tmp;
 791
 792         memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
 793
 794         /*
 795          * If a node is no longer configured but is still in the livemap, we
 796          * may need to clear that bit from the livemap.
 797          */
 798         node = o2nm_get_node_by_num(slot->ds_node_num);
 799         if (!node) {
 800                 spin_lock(&o2hb_live_lock);
 801                 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 802                 spin_unlock(&o2hb_live_lock);
 803                 if (!tmp)
 804                         return 0;
 805         }
 806
 807         if (!o2hb_verify_crc(reg, hb_block)) {
 808                 /* all paths from here will drop o2hb_live_lock for
 809                  * us. */
 810                 spin_lock(&o2hb_live_lock);
 811
 812                 /* Don't print an error on the console in this case -
 813                  * a freshly formatted heartbeat area will not have a
 814                  * crc set on it. */
 815                 if (list_empty(&slot->ds_live_item))
 816                         goto out;
 817
 818                 /* The node is live but pushed out a bad crc. We
 819                  * consider it a transient miss but don't populate any
 820                  * other values as they may be junk. */
 821                 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
 822                      slot->ds_node_num, reg->hr_dev_name);
 823                 o2hb_dump_slot(hb_block);
 824
 825                 slot->ds_equal_samples++;
 826                 goto fire_callbacks;
 827         }
 828
 829         /* we don't care if these wrap.. the state transitions below
 830          * clear at the right places */
 831         cputime = le64_to_cpu(hb_block->hb_seq);
 832         if (slot->ds_last_time != cputime)
 833                 slot->ds_changed_samples++;
 834         else
 835                 slot->ds_equal_samples++;
 836         slot->ds_last_time = cputime;
 837
 838         /* The node changed heartbeat generations. We assume this to
 839          * mean it dropped off but came back before we timed out. We
 840          * want to consider it down for the time being but don't want
 841          * to lose any changed_samples state we might build up to
 842          * considering it live again. */
 843         if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
 844                 gen_changed = 1;
 845                 slot->ds_equal_samples = 0;
 846                 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
 847                      "to 0x%llx)\n", slot->ds_node_num,
 848                      (long long)slot->ds_last_generation,
 849                      (long long)le64_to_cpu(hb_block->hb_generation));
 850         }
 851
 852         slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
 853
 854         mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
 855              "seq %llu last %llu changed %u equal %u\n",
 856              slot->ds_node_num, (long long)slot->ds_last_generation,
 857              le32_to_cpu(hb_block->hb_cksum),
 858              (unsigned long long)le64_to_cpu(hb_block->hb_seq),
 859              (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
 860              slot->ds_equal_samples);
 861
 862         spin_lock(&o2hb_live_lock);
 863
 864 fire_callbacks:
 865         /* dead nodes only come to life after some number of
 866          * changes at any time during their dead time */
 867         if (list_empty(&slot->ds_live_item) &&
 868             slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
 869                 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
 870                      slot->ds_node_num, (long long)slot->ds_last_generation);
 871
 872                 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 873
 874                 /* first on the list generates a callback */
 875                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 876                         mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
 877                              "bitmap\n", slot->ds_node_num);
 878                         set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 879
 880                         o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
 881                                               slot->ds_node_num);
 882
 883                         changed = 1;
 884                 }
 885
 886                 list_add_tail(&slot->ds_live_item,
 887                               &o2hb_live_slots[slot->ds_node_num]);
 888
 889                 slot->ds_equal_samples = 0;
 890
 891                 /* We want to be sure that all nodes agree on the
 892                  * number of milliseconds before a node will be
 893                  * considered dead. The self-fencing timeout is
 894                  * computed from this value, and a discrepancy might
 895                  * result in heartbeat calling a node dead when it
 896                  * hasn't self-fenced yet. */
 897                 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
 898                 if (slot_dead_ms && slot_dead_ms != dead_ms) {
 899                         /* TODO: Perhaps we can fail the region here. */
 900                         mlog(ML_ERROR, "Node %d on device %s has a dead count "
 901                              "of %u ms, but our count is %u ms.\n"
 902                              "Please double check your configuration values "
 903                              "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
 904                              slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
 905                              dead_ms);
 906                 }
 907                 goto out;
 908         }
 909
 910         /* if the list is dead, we're done.. */
 911         if (list_empty(&slot->ds_live_item))
 912                 goto out;
 913
 914         /* live nodes only go dead after enough consequtive missed
 915          * samples..  reset the missed counter whenever we see
 916          * activity */
 917         if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
 918                 mlog(ML_HEARTBEAT, "Node %d left my region\n",
 919                      slot->ds_node_num);
 920
 921                 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 922
 923                 /* last off the live_slot generates a callback */
 924                 list_del_init(&slot->ds_live_item);
 925                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 926                         mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
 927                              "nodes bitmap\n", slot->ds_node_num);
 928                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 929
 930                         /* node can be null */
 931                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
 932                                               node, slot->ds_node_num);
 933
 934                         changed = 1;
 935                 }
 936
 937                 /* We don't clear this because the node is still
 938                  * actually writing new blocks. */
 939                 if (!gen_changed)
 940                         slot->ds_changed_samples = 0;
 941                 goto out;
 942         }
 943         if (slot->ds_changed_samples) {
 944                 slot->ds_changed_samples = 0;
 945                 slot->ds_equal_samples = 0;
 946         }
 947 out:
 948         spin_unlock(&o2hb_live_lock);
 949
 950         o2hb_run_event_list(&event);
 951
 952         if (node)
 953                 o2nm_node_put(node);
 954         return changed;
 955 }
 956
 957 /* This could be faster if we just implmented a find_last_bit, but I
 958  * don't think the circumstances warrant it. */
 959 static int o2hb_highest_node(unsigned long *nodes,
 960                              int numbits)
 961 {
 962         int highest, node;
 963
 964         highest = numbits;
 965         node = -1;
 966         while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
 967                 if (node >= numbits)
 968                         break;
 969
 970                 highest = node;
 971         }
 972
 973         return highest;
 974 }
 975
 976 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
 977 {
 978         int i, ret, highest_node;
 979         int membership_change = 0, own_slot_ok = 0;
 980         unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
 981         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 982         struct o2hb_bio_wait_ctxt write_wc;
 983
 984         ret = o2nm_configured_node_map(configured_nodes,
 985                                        sizeof(configured_nodes));
 986         if (ret) {
 987                 mlog_errno(ret);
 988                 goto bail;
 989         }
 990
 991         /*
 992          * If a node is not configured but is in the livemap, we still need
 993          * to read the slot so as to be able to remove it from the livemap.
 994          */
 995         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
 996         i = -1;
 997         while ((i = find_next_bit(live_node_bitmap,
 998                                   O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
 999                 set_bit(i, configured_nodes);
1000         }
1001
1002         highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1003         if (highest_node >= O2NM_MAX_NODES) {
1004                 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1005                 ret = -EINVAL;
1006                 goto bail;
1007         }
1008
1009         /* No sense in reading the slots of nodes that don't exist
1010          * yet. Of course, if the node definitions have holes in them
1011          * then we're reading an empty slot anyway... Consider this
1012          * best-effort. */
1013         ret = o2hb_read_slots(reg, highest_node + 1);
1014         if (ret < 0) {
1015                 mlog_errno(ret);
1016                 goto bail;
1017         }
1018
1019         /* With an up to date view of the slots, we can check that no
1020          * other node has been improperly configured to heartbeat in
1021          * our slot. */
1022         own_slot_ok = o2hb_check_own_slot(reg);
1023
1024         /* fill in the proper info for our next heartbeat */
1025         o2hb_prepare_block(reg, reg->hr_generation);
1026
1027         ret = o2hb_issue_node_write(reg, &write_wc);
1028         if (ret < 0) {
1029                 mlog_errno(ret);
1030                 goto bail;
1031         }
1032
1033         i = -1;
1034         while((i = find_next_bit(configured_nodes,
1035                                  O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1036                 membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1037         }
1038
1039         /*
1040          * We have to be sure we've advertised ourselves on disk
1041          * before we can go to steady state.  This ensures that
1042          * people we find in our steady state have seen us.
1043          */
1044         o2hb_wait_on_io(reg, &write_wc);
1045         if (write_wc.wc_error) {
1046                 /* Do not re-arm the write timeout on I/O error - we
1047                  * can't be sure that the new block ever made it to
1048                  * disk */
1049                 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1050                      write_wc.wc_error, reg->hr_dev_name);
1051                 ret = write_wc.wc_error;
1052                 goto bail;
1053         }
1054
1055         /* Skip disarming the timeout if own slot has stale/bad data */
1056         if (own_slot_ok) {
1057                 o2hb_set_quorum_device(reg);
1058                 o2hb_arm_write_timeout(reg);
1059         }
1060
1061 bail:
1062         /* let the person who launched us know when things are steady */
1063         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1064                 if (!ret && own_slot_ok && !membership_change) {
1065                         if (atomic_dec_and_test(&reg->hr_steady_iterations))
1066                                 wake_up(&o2hb_steady_queue);
1067                 }
1068         }
1069
1070         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1071                 if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1072                         printk(KERN_NOTICE "o2hb: Unable to stabilize "
1073                                "heartbeart on region %s (%s)\n",
1074                                config_item_name(&reg->hr_item),
1075                                reg->hr_dev_name);
1076                         atomic_set(&reg->hr_steady_iterations, 0);
1077                         reg->hr_aborted_start = 1;
1078                         wake_up(&o2hb_steady_queue);
1079                         ret = -EIO;
1080                 }
1081         }
1082
1083         return ret;
1084 }
1085
1086 /* Subtract b from a, storing the result in a. a *must* have a larger
1087  * value than b. */
1088 static void o2hb_tv_subtract(struct timeval *a,
1089                              struct timeval *b)
1090 {
1091         /* just return 0 when a is after b */
1092         if (a->tv_sec < b->tv_sec ||
1093             (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1094                 a->tv_sec = 0;
1095                 a->tv_usec = 0;
1096                 return;
1097         }
1098
1099         a->tv_sec -= b->tv_sec;
1100         a->tv_usec -= b->tv_usec;
1101         while ( a->tv_usec < 0 ) {
1102                 a->tv_sec--;
1103                 a->tv_usec += 1000000;
1104         }
1105 }
1106
1107 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1108                                        struct timeval *end)
1109 {
1110         struct timeval res = *end;
1111
1112         o2hb_tv_subtract(&res, start);
1113
1114         return res.tv_sec * 1000 + res.tv_usec / 1000;
1115 }
1116
1117 /*
1118  * we ride the region ref that the region dir holds.  before the region
1119  * dir is removed and drops it ref it will wait to tear down this
1120  * thread.
1121  */
1122 static int o2hb_thread(void *data)
1123 {
1124         int i, ret;
1125         struct o2hb_region *reg = data;
1126         struct o2hb_bio_wait_ctxt write_wc;
1127         struct timeval before_hb, after_hb;
1128         unsigned int elapsed_msec;
1129
1130         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1131
1132         set_user_nice(current, -20);
1133
1134         /* Pin node */
1135         o2nm_depend_this_node();
1136
1137         while (!kthread_should_stop() &&
1138                !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1139                 /* We track the time spent inside
1140                  * o2hb_do_disk_heartbeat so that we avoid more than
1141                  * hr_timeout_ms between disk writes. On busy systems
1142                  * this should result in a heartbeat which is less
1143                  * likely to time itself out. */
1144                 do_gettimeofday(&before_hb);
1145
1146                 ret = o2hb_do_disk_heartbeat(reg);
1147
1148                 do_gettimeofday(&after_hb);
1149                 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1150
1151                 mlog(ML_HEARTBEAT,
1152                      "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
1153                      before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1154                      after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1155                      elapsed_msec);
1156
1157                 if (!kthread_should_stop() &&
1158                     elapsed_msec < reg->hr_timeout_ms) {
1159                         /* the kthread api has blocked signals for us so no
1160                          * need to record the return value. */
1161                         msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1162                 }
1163         }
1164
1165         o2hb_disarm_write_timeout(reg);
1166
1167         /* unclean stop is only used in very bad situation */
1168         for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1169                 o2hb_shutdown_slot(&reg->hr_slots[i]);
1170
1171         /* Explicit down notification - avoid forcing the other nodes
1172          * to timeout on this region when we could just as easily
1173          * write a clear generation - thus indicating to them that
1174          * this node has left this region.
1175          */
1176         if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1177                 o2hb_prepare_block(reg, 0);
1178                 ret = o2hb_issue_node_write(reg, &write_wc);
1179                 if (ret == 0)
1180                         o2hb_wait_on_io(reg, &write_wc);
1181                 else
1182                         mlog_errno(ret);
1183         }
1184
1185         /* Unpin node */
1186         o2nm_undepend_this_node();
1187
1188         mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1189
1190         return 0;
1191 }
1192
1193 #ifdef CONFIG_DEBUG_FS
1194 static int o2hb_debug_open(struct inode *inode, struct file *file)
1195 {
1196         struct o2hb_debug_buf *db = inode->i_private;
1197         struct o2hb_region *reg;
1198         unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1199         unsigned long lts;
1200         char *buf = NULL;
1201         int i = -1;
1202         int out = 0;
1203
1204         /* max_nodes should be the largest bitmap we pass here */
1205         BUG_ON(sizeof(map) < db->db_size);
1206
1207         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1208         if (!buf)
1209                 goto bail;
1210
1211         switch (db->db_type) {
1212         case O2HB_DB_TYPE_LIVENODES:
1213         case O2HB_DB_TYPE_LIVEREGIONS:
1214         case O2HB_DB_TYPE_QUORUMREGIONS:
1215         case O2HB_DB_TYPE_FAILEDREGIONS:
1216                 spin_lock(&o2hb_live_lock);
1217                 memcpy(map, db->db_data, db->db_size);
1218                 spin_unlock(&o2hb_live_lock);
1219                 break;
1220
1221         case O2HB_DB_TYPE_REGION_LIVENODES:
1222                 spin_lock(&o2hb_live_lock);
1223                 reg = (struct o2hb_region *)db->db_data;
1224                 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1225                 spin_unlock(&o2hb_live_lock);
1226                 break;
1227
1228         case O2HB_DB_TYPE_REGION_NUMBER:
1229                 reg = (struct o2hb_region *)db->db_data;
1230                 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1231                                 reg->hr_region_num);
1232                 goto done;
1233
1234         case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1235                 reg = (struct o2hb_region *)db->db_data;
1236                 lts = reg->hr_last_timeout_start;
1237                 /* If 0, it has never been set before */
1238                 if (lts)
1239                         lts = jiffies_to_msecs(jiffies - lts);
1240                 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1241                 goto done;
1242
1243         case O2HB_DB_TYPE_REGION_PINNED:
1244                 reg = (struct o2hb_region *)db->db_data;
1245                 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1246                                 !!reg->hr_item_pinned);
1247                 goto done;
1248
1249         default:
1250                 goto done;
1251         }
1252
1253         while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1254                 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1255         out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1256
1257 done:
1258         i_size_write(inode, out);
1259
1260         file->private_data = buf;
1261
1262         return 0;
1263 bail:
1264         return -ENOMEM;
1265 }
1266
1267 static int o2hb_debug_release(struct inode *inode, struct file *file)
1268 {
1269         kfree(file->private_data);
1270         return 0;
1271 }
1272
1273 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1274                                  size_t nbytes, loff_t *ppos)
1275 {
1276         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1277                                        i_size_read(file->f_mapping->host));
1278 }
1279 #else
1280 static int o2hb_debug_open(struct inode *inode, struct file *file)
1281 {
1282         return 0;
1283 }
1284 static int o2hb_debug_release(struct inode *inode, struct file *file)
1285 {
1286         return 0;
1287 }
1288 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1289                                size_t nbytes, loff_t *ppos)
1290 {
1291         return 0;
1292 }
1293 #endif  /* CONFIG_DEBUG_FS */
1294
1295 static const struct file_operations o2hb_debug_fops = {
1296         .open =         o2hb_debug_open,
1297         .release =      o2hb_debug_release,
1298         .read =         o2hb_debug_read,
1299         .llseek =       generic_file_llseek,
1300 };
1301
1302 void o2hb_exit(void)
1303 {
1304         kfree(o2hb_db_livenodes);
1305         kfree(o2hb_db_liveregions);
1306         kfree(o2hb_db_quorumregions);
1307         kfree(o2hb_db_failedregions);
1308         debugfs_remove(o2hb_debug_failedregions);
1309         debugfs_remove(o2hb_debug_quorumregions);
1310         debugfs_remove(o2hb_debug_liveregions);
1311         debugfs_remove(o2hb_debug_livenodes);
1312         debugfs_remove(o2hb_debug_dir);
1313 }
1314
1315 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1316                                         struct o2hb_debug_buf **db, int db_len,
1317                                         int type, int size, int len, void *data)
1318 {
1319         *db = kmalloc(db_len, GFP_KERNEL);
1320         if (!*db)
1321                 return NULL;
1322
1323         (*db)->db_type = type;
1324         (*db)->db_size = size;
1325         (*db)->db_len = len;
1326         (*db)->db_data = data;
1327
1328         return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1329                                    &o2hb_debug_fops);
1330 }
1331
1332 static int o2hb_debug_init(void)
1333 {
1334         int ret = -ENOMEM;
1335
1336         o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1337         if (!o2hb_debug_dir) {
1338                 mlog_errno(ret);
1339                 goto bail;
1340         }
1341
1342         o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1343                                                  o2hb_debug_dir,
1344                                                  &o2hb_db_livenodes,
1345                                                  sizeof(*o2hb_db_livenodes),
1346                                                  O2HB_DB_TYPE_LIVENODES,
1347                                                  sizeof(o2hb_live_node_bitmap),
1348                                                  O2NM_MAX_NODES,
1349                                                  o2hb_live_node_bitmap);
1350         if (!o2hb_debug_livenodes) {
1351                 mlog_errno(ret);
1352                 goto bail;
1353         }
1354
1355         o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1356                                                    o2hb_debug_dir,
1357                                                    &o2hb_db_liveregions,
1358                                                    sizeof(*o2hb_db_liveregions),
1359                                                    O2HB_DB_TYPE_LIVEREGIONS,
1360                                                    sizeof(o2hb_live_region_bitmap),
1361                                                    O2NM_MAX_REGIONS,
1362                                                    o2hb_live_region_bitmap);
1363         if (!o2hb_debug_liveregions) {
1364                 mlog_errno(ret);
1365                 goto bail;
1366         }
1367
1368         o2hb_debug_quorumregions =
1369                         o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1370                                           o2hb_debug_dir,
1371                                           &o2hb_db_quorumregions,
1372                                           sizeof(*o2hb_db_quorumregions),
1373                                           O2HB_DB_TYPE_QUORUMREGIONS,
1374                                           sizeof(o2hb_quorum_region_bitmap),
1375                                           O2NM_MAX_REGIONS,
1376                                           o2hb_quorum_region_bitmap);
1377         if (!o2hb_debug_quorumregions) {
1378                 mlog_errno(ret);
1379                 goto bail;
1380         }
1381
1382         o2hb_debug_failedregions =
1383                         o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1384                                           o2hb_debug_dir,
1385                                           &o2hb_db_failedregions,
1386                                           sizeof(*o2hb_db_failedregions),
1387                                           O2HB_DB_TYPE_FAILEDREGIONS,
1388                                           sizeof(o2hb_failed_region_bitmap),
1389                                           O2NM_MAX_REGIONS,
1390                                           o2hb_failed_region_bitmap);
1391         if (!o2hb_debug_failedregions) {
1392                 mlog_errno(ret);
1393                 goto bail;
1394         }
1395
1396         ret = 0;
1397 bail:
1398         if (ret)
1399                 o2hb_exit();
1400
1401         return ret;
1402 }
1403
1404 int o2hb_init(void)
1405 {
1406         int i;
1407
1408         for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1409                 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1410
1411         for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1412                 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1413
1414         INIT_LIST_HEAD(&o2hb_node_events);
1415
1416         memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1417         memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1418         memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1419         memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1420         memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1421
1422         o2hb_dependent_users = 0;
1423
1424         return o2hb_debug_init();
1425 }
1426
1427 /* if we're already in a callback then we're already serialized by the sem */
1428 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1429                                              unsigned bytes)
1430 {
1431         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1432
1433         memcpy(map, &o2hb_live_node_bitmap, bytes);
1434 }
1435
1436 /*
1437  * get a map of all nodes that are heartbeating in any regions
1438  */
1439 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1440 {
1441         /* callers want to serialize this map and callbacks so that they
1442          * can trust that they don't miss nodes coming to the party */
1443         down_read(&o2hb_callback_sem);
1444         spin_lock(&o2hb_live_lock);
1445         o2hb_fill_node_map_from_callback(map, bytes);
1446         spin_unlock(&o2hb_live_lock);
1447         up_read(&o2hb_callback_sem);
1448 }
1449 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1450
1451 /*
1452  * heartbeat configfs bits.  The heartbeat set is a default set under
1453  * the cluster set in nodemanager.c.
1454  */
1455
1456 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1457 {
1458         return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1459 }
1460
1461 /* drop_item only drops its ref after killing the thread, nothing should
1462  * be using the region anymore.  this has to clean up any state that
1463  * attributes might have built up. */
1464 static void o2hb_region_release(struct config_item *item)
1465 {
1466         int i;
1467         struct page *page;
1468         struct o2hb_region *reg = to_o2hb_region(item);
1469
1470         mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1471
1472         kfree(reg->hr_tmp_block);
1473
1474         if (reg->hr_slot_data) {
1475                 for (i = 0; i < reg->hr_num_pages; i++) {
1476                         page = reg->hr_slot_data[i];
1477                         if (page)
1478                                 __free_page(page);
1479                 }
1480                 kfree(reg->hr_slot_data);
1481         }
1482
1483         if (reg->hr_bdev)
1484                 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1485
1486         kfree(reg->hr_slots);
1487
1488         kfree(reg->hr_db_regnum);
1489         kfree(reg->hr_db_livenodes);
1490         debugfs_remove(reg->hr_debug_livenodes);
1491         debugfs_remove(reg->hr_debug_regnum);
1492         debugfs_remove(reg->hr_debug_elapsed_time);
1493         debugfs_remove(reg->hr_debug_pinned);
1494         debugfs_remove(reg->hr_debug_dir);
1495
1496         spin_lock(&o2hb_live_lock);
1497         list_del(&reg->hr_all_item);
1498         spin_unlock(&o2hb_live_lock);
1499
1500         kfree(reg);
1501 }
1502
1503 static int o2hb_read_block_input(struct o2hb_region *reg,
1504                                  const char *page,
1505                                  size_t count,
1506                                  unsigned long *ret_bytes,
1507                                  unsigned int *ret_bits)
1508 {
1509         unsigned long bytes;
1510         char *p = (char *)page;
1511
1512         bytes = simple_strtoul(p, &p, 0);
1513         if (!p || (*p && (*p != '\n')))
1514                 return -EINVAL;
1515
1516         /* Heartbeat and fs min / max block sizes are the same. */
1517         if (bytes > 4096 || bytes < 512)
1518                 return -ERANGE;
1519         if (hweight16(bytes) != 1)
1520                 return -EINVAL;
1521
1522         if (ret_bytes)
1523                 *ret_bytes = bytes;
1524         if (ret_bits)
1525                 *ret_bits = ffs(bytes) - 1;
1526
1527         return 0;
1528 }
1529
1530 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1531                                             char *page)
1532 {
1533         return sprintf(page, "%u\n", reg->hr_block_bytes);
1534 }
1535
1536 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1537                                              const char *page,
1538                                              size_t count)
1539 {
1540         int status;
1541         unsigned long block_bytes;
1542         unsigned int block_bits;
1543
1544         if (reg->hr_bdev)
1545                 return -EINVAL;
1546
1547         status = o2hb_read_block_input(reg, page, count,
1548                                        &block_bytes, &block_bits);
1549         if (status)
1550                 return status;
1551
1552         reg->hr_block_bytes = (unsigned int)block_bytes;
1553         reg->hr_block_bits = block_bits;
1554
1555         return count;
1556 }
1557
1558 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1559                                             char *page)
1560 {
1561         return sprintf(page, "%llu\n", reg->hr_start_block);
1562 }
1563
1564 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1565                                              const char *page,
1566                                              size_t count)
1567 {
1568         unsigned long long tmp;
1569         char *p = (char *)page;
1570
1571         if (reg->hr_bdev)
1572                 return -EINVAL;
1573
1574         tmp = simple_strtoull(p, &p, 0);
1575         if (!p || (*p && (*p != '\n')))
1576                 return -EINVAL;
1577
1578         reg->hr_start_block = tmp;
1579
1580         return count;
1581 }
1582
1583 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1584                                        char *page)
1585 {
1586         return sprintf(page, "%d\n", reg->hr_blocks);
1587 }
1588
1589 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1590                                         const char *page,
1591                                         size_t count)
1592 {
1593         unsigned long tmp;
1594         char *p = (char *)page;
1595
1596         if (reg->hr_bdev)
1597                 return -EINVAL;
1598
1599         tmp = simple_strtoul(p, &p, 0);
1600         if (!p || (*p && (*p != '\n')))
1601                 return -EINVAL;
1602
1603         if (tmp > O2NM_MAX_NODES || tmp == 0)
1604                 return -ERANGE;
1605
1606         reg->hr_blocks = (unsigned int)tmp;
1607
1608         return count;
1609 }
1610
1611 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1612                                     char *page)
1613 {
1614         unsigned int ret = 0;
1615
1616         if (reg->hr_bdev)
1617                 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1618
1619         return ret;
1620 }
1621
1622 static void o2hb_init_region_params(struct o2hb_region *reg)
1623 {
1624         reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1625         reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1626
1627         mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1628              reg->hr_start_block, reg->hr_blocks);
1629         mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1630              reg->hr_block_bytes, reg->hr_block_bits);
1631         mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1632         mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1633 }
1634
1635 static int o2hb_map_slot_data(struct o2hb_region *reg)
1636 {
1637         int i, j;
1638         unsigned int last_slot;
1639         unsigned int spp = reg->hr_slots_per_page;
1640         struct page *page;
1641         char *raw;
1642         struct o2hb_disk_slot *slot;
1643
1644         reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1645         if (reg->hr_tmp_block == NULL) {
1646                 mlog_errno(-ENOMEM);
1647                 return -ENOMEM;
1648         }
1649
1650         reg->hr_slots = kcalloc(reg->hr_blocks,
1651                                 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1652         if (reg->hr_slots == NULL) {
1653                 mlog_errno(-ENOMEM);
1654                 return -ENOMEM;
1655         }
1656
1657         for(i = 0; i < reg->hr_blocks; i++) {
1658                 slot = &reg->hr_slots[i];
1659                 slot->ds_node_num = i;
1660                 INIT_LIST_HEAD(&slot->ds_live_item);
1661                 slot->ds_raw_block = NULL;
1662         }
1663
1664         reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1665         mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1666                            "at %u blocks per page\n",
1667              reg->hr_num_pages, reg->hr_blocks, spp);
1668
1669         reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1670                                     GFP_KERNEL);
1671         if (!reg->hr_slot_data) {
1672                 mlog_errno(-ENOMEM);
1673                 return -ENOMEM;
1674         }
1675
1676         for(i = 0; i < reg->hr_num_pages; i++) {
1677                 page = alloc_page(GFP_KERNEL);
1678                 if (!page) {
1679                         mlog_errno(-ENOMEM);
1680                         return -ENOMEM;
1681                 }
1682
1683                 reg->hr_slot_data[i] = page;
1684
1685                 last_slot = i * spp;
1686                 raw = page_address(page);
1687                 for (j = 0;
1688                      (j < spp) && ((j + last_slot) < reg->hr_blocks);
1689                      j++) {
1690                         BUG_ON((j + last_slot) >= reg->hr_blocks);
1691
1692                         slot = &reg->hr_slots[j + last_slot];
1693                         slot->ds_raw_block =
1694                                 (struct o2hb_disk_heartbeat_block *) raw;
1695
1696                         raw += reg->hr_block_bytes;
1697                 }
1698         }
1699
1700         return 0;
1701 }
1702
1703 /* Read in all the slots available and populate the tracking
1704  * structures so that we can start with a baseline idea of what's
1705  * there. */
1706 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1707 {
1708         int ret, i;
1709         struct o2hb_disk_slot *slot;
1710         struct o2hb_disk_heartbeat_block *hb_block;
1711
1712         ret = o2hb_read_slots(reg, reg->hr_blocks);
1713         if (ret) {
1714                 mlog_errno(ret);
1715                 goto out;
1716         }
1717
1718         /* We only want to get an idea of the values initially in each
1719          * slot, so we do no verification - o2hb_check_slot will
1720          * actually determine if each configured slot is valid and
1721          * whether any values have changed. */
1722         for(i = 0; i < reg->hr_blocks; i++) {
1723                 slot = &reg->hr_slots[i];
1724                 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1725
1726                 /* Only fill the values that o2hb_check_slot uses to
1727                  * determine changing slots */
1728                 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1729                 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1730         }
1731
1732 out:
1733         return ret;
1734 }
1735
1736 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1737 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1738                                      const char *page,
1739                                      size_t count)
1740 {
1741         struct task_struct *hb_task;
1742         long fd;
1743         int sectsize;
1744         char *p = (char *)page;
1745         struct fd f;
1746         struct inode *inode;
1747         ssize_t ret = -EINVAL;
1748         int live_threshold;
1749
1750         if (reg->hr_bdev)
1751                 goto out;
1752
1753         /* We can't heartbeat without having had our node number
1754          * configured yet. */
1755         if (o2nm_this_node() == O2NM_MAX_NODES)
1756                 goto out;
1757
1758         fd = simple_strtol(p, &p, 0);
1759         if (!p || (*p && (*p != '\n')))
1760                 goto out;
1761
1762         if (fd < 0 || fd >= INT_MAX)
1763                 goto out;
1764
1765         f = fdget(fd);
1766         if (f.file == NULL)
1767                 goto out;
1768
1769         if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1770             reg->hr_block_bytes == 0)
1771                 goto out2;
1772
1773         inode = igrab(f.file->f_mapping->host);
1774         if (inode == NULL)
1775                 goto out2;
1776
1777         if (!S_ISBLK(inode->i_mode))
1778                 goto out3;
1779
1780         reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
1781         ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
1782         if (ret) {
1783                 reg->hr_bdev = NULL;
1784                 goto out3;
1785         }
1786         inode = NULL;
1787
1788         bdevname(reg->hr_bdev, reg->hr_dev_name);
1789
1790         sectsize = bdev_logical_block_size(reg->hr_bdev);
1791         if (sectsize != reg->hr_block_bytes) {
1792                 mlog(ML_ERROR,
1793                      "blocksize %u incorrect for device, expected %d",
1794                      reg->hr_block_bytes, sectsize);
1795                 ret = -EINVAL;
1796                 goto out3;
1797         }
1798
1799         o2hb_init_region_params(reg);
1800
1801         /* Generation of zero is invalid */
1802         do {
1803                 get_random_bytes(&reg->hr_generation,
1804                                  sizeof(reg->hr_generation));
1805         } while (reg->hr_generation == 0);
1806
1807         ret = o2hb_map_slot_data(reg);
1808         if (ret) {
1809                 mlog_errno(ret);
1810                 goto out3;
1811         }
1812
1813         ret = o2hb_populate_slot_data(reg);
1814         if (ret) {
1815                 mlog_errno(ret);
1816                 goto out3;
1817         }
1818
1819         INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1820
1821         /*
1822          * A node is considered live after it has beat LIVE_THRESHOLD
1823          * times.  We're not steady until we've given them a chance
1824          * _after_ our first read.
1825          * The default threshold is bare minimum so as to limit the delay
1826          * during mounts. For global heartbeat, the threshold doubled for the
1827          * first region.
1828          */
1829         live_threshold = O2HB_LIVE_THRESHOLD;
1830         if (o2hb_global_heartbeat_active()) {
1831                 spin_lock(&o2hb_live_lock);
1832                 if (o2hb_pop_count(&o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1833                         live_threshold <<= 1;
1834                 spin_unlock(&o2hb_live_lock);
1835         }
1836         ++live_threshold;
1837         atomic_set(&reg->hr_steady_iterations, live_threshold);
1838         /* unsteady_iterations is double the steady_iterations */
1839         atomic_set(&reg->hr_unsteady_iterations, (live_threshold << 1));
1840
1841         hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1842                               reg->hr_item.ci_name);
1843         if (IS_ERR(hb_task)) {
1844                 ret = PTR_ERR(hb_task);
1845                 mlog_errno(ret);
1846                 goto out3;
1847         }
1848
1849         spin_lock(&o2hb_live_lock);
1850         reg->hr_task = hb_task;
1851         spin_unlock(&o2hb_live_lock);
1852
1853         ret = wait_event_interruptible(o2hb_steady_queue,
1854                                 atomic_read(&reg->hr_steady_iterations) == 0);
1855         if (ret) {
1856                 atomic_set(&reg->hr_steady_iterations, 0);
1857                 reg->hr_aborted_start = 1;
1858         }
1859
1860         if (reg->hr_aborted_start) {
1861                 ret = -EIO;
1862                 goto out3;
1863         }
1864
1865         /* Ok, we were woken.  Make sure it wasn't by drop_item() */
1866         spin_lock(&o2hb_live_lock);
1867         hb_task = reg->hr_task;
1868         if (o2hb_global_heartbeat_active())
1869                 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1870         spin_unlock(&o2hb_live_lock);
1871
1872         if (hb_task)
1873                 ret = count;
1874         else
1875                 ret = -EIO;
1876
1877         if (hb_task && o2hb_global_heartbeat_active())
1878                 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1879                        config_item_name(&reg->hr_item), reg->hr_dev_name);
1880
1881 out3:
1882         iput(inode);
1883 out2:
1884         fdput(f);
1885 out:
1886         if (ret < 0) {
1887                 if (reg->hr_bdev) {
1888                         blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1889                         reg->hr_bdev = NULL;
1890                 }
1891         }
1892         return ret;
1893 }
1894
1895 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1896                                       char *page)
1897 {
1898         pid_t pid = 0;
1899
1900         spin_lock(&o2hb_live_lock);
1901         if (reg->hr_task)
1902                 pid = task_pid_nr(reg->hr_task);
1903         spin_unlock(&o2hb_live_lock);
1904
1905         if (!pid)
1906                 return 0;
1907
1908         return sprintf(page, "%u\n", pid);
1909 }
1910
1911 struct o2hb_region_attribute {
1912         struct configfs_attribute attr;
1913         ssize_t (*show)(struct o2hb_region *, char *);
1914         ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1915 };
1916
1917 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1918         .attr   = { .ca_owner = THIS_MODULE,
1919                     .ca_name = "block_bytes",
1920                     .ca_mode = S_IRUGO | S_IWUSR },
1921         .show   = o2hb_region_block_bytes_read,
1922         .store  = o2hb_region_block_bytes_write,
1923 };
1924
1925 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1926         .attr   = { .ca_owner = THIS_MODULE,
1927                     .ca_name = "start_block",
1928                     .ca_mode = S_IRUGO | S_IWUSR },
1929         .show   = o2hb_region_start_block_read,
1930         .store  = o2hb_region_start_block_write,
1931 };
1932
1933 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1934         .attr   = { .ca_owner = THIS_MODULE,
1935                     .ca_name = "blocks",
1936                     .ca_mode = S_IRUGO | S_IWUSR },
1937         .show   = o2hb_region_blocks_read,
1938         .store  = o2hb_region_blocks_write,
1939 };
1940
1941 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1942         .attr   = { .ca_owner = THIS_MODULE,
1943                     .ca_name = "dev",
1944                     .ca_mode = S_IRUGO | S_IWUSR },
1945         .show   = o2hb_region_dev_read,
1946         .store  = o2hb_region_dev_write,
1947 };
1948
1949 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1950        .attr   = { .ca_owner = THIS_MODULE,
1951                    .ca_name = "pid",
1952                    .ca_mode = S_IRUGO | S_IRUSR },
1953        .show   = o2hb_region_pid_read,
1954 };
1955
1956 static struct configfs_attribute *o2hb_region_attrs[] = {
1957         &o2hb_region_attr_block_bytes.attr,
1958         &o2hb_region_attr_start_block.attr,
1959         &o2hb_region_attr_blocks.attr,
1960         &o2hb_region_attr_dev.attr,
1961         &o2hb_region_attr_pid.attr,
1962         NULL,
1963 };
1964
1965 static ssize_t o2hb_region_show(struct config_item *item,
1966                                 struct configfs_attribute *attr,
1967                                 char *page)
1968 {
1969         struct o2hb_region *reg = to_o2hb_region(item);
1970         struct o2hb_region_attribute *o2hb_region_attr =
1971                 container_of(attr, struct o2hb_region_attribute, attr);
1972         ssize_t ret = 0;
1973
1974         if (o2hb_region_attr->show)
1975                 ret = o2hb_region_attr->show(reg, page);
1976         return ret;
1977 }
1978
1979 static ssize_t o2hb_region_store(struct config_item *item,
1980                                  struct configfs_attribute *attr,
1981                                  const char *page, size_t count)
1982 {
1983         struct o2hb_region *reg = to_o2hb_region(item);
1984         struct o2hb_region_attribute *o2hb_region_attr =
1985                 container_of(attr, struct o2hb_region_attribute, attr);
1986         ssize_t ret = -EINVAL;
1987
1988         if (o2hb_region_attr->store)
1989                 ret = o2hb_region_attr->store(reg, page, count);
1990         return ret;
1991 }
1992
1993 static struct configfs_item_operations o2hb_region_item_ops = {
1994         .release                = o2hb_region_release,
1995         .show_attribute         = o2hb_region_show,
1996         .store_attribute        = o2hb_region_store,
1997 };
1998
1999 static struct config_item_type o2hb_region_type = {
2000         .ct_item_ops    = &o2hb_region_item_ops,
2001         .ct_attrs       = o2hb_region_attrs,
2002         .ct_owner       = THIS_MODULE,
2003 };
2004
2005 /* heartbeat set */
2006
2007 struct o2hb_heartbeat_group {
2008         struct config_group hs_group;
2009         /* some stuff? */
2010 };
2011
2012 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
2013 {
2014         return group ?
2015                 container_of(group, struct o2hb_heartbeat_group, hs_group)
2016                 : NULL;
2017 }
2018
2019 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
2020 {
2021         int ret = -ENOMEM;
2022
2023         reg->hr_debug_dir =
2024                 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
2025         if (!reg->hr_debug_dir) {
2026                 mlog_errno(ret);
2027                 goto bail;
2028         }
2029
2030         reg->hr_debug_livenodes =
2031                         o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2032                                           reg->hr_debug_dir,
2033                                           &(reg->hr_db_livenodes),
2034                                           sizeof(*(reg->hr_db_livenodes)),
2035                                           O2HB_DB_TYPE_REGION_LIVENODES,
2036                                           sizeof(reg->hr_live_node_bitmap),
2037                                           O2NM_MAX_NODES, reg);
2038         if (!reg->hr_debug_livenodes) {
2039                 mlog_errno(ret);
2040                 goto bail;
2041         }
2042
2043         reg->hr_debug_regnum =
2044                         o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2045                                           reg->hr_debug_dir,
2046                                           &(reg->hr_db_regnum),
2047                                           sizeof(*(reg->hr_db_regnum)),
2048                                           O2HB_DB_TYPE_REGION_NUMBER,
2049                                           0, O2NM_MAX_NODES, reg);
2050         if (!reg->hr_debug_regnum) {
2051                 mlog_errno(ret);
2052                 goto bail;
2053         }
2054
2055         reg->hr_debug_elapsed_time =
2056                         o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2057                                           reg->hr_debug_dir,
2058                                           &(reg->hr_db_elapsed_time),
2059                                           sizeof(*(reg->hr_db_elapsed_time)),
2060                                           O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2061                                           0, 0, reg);
2062         if (!reg->hr_debug_elapsed_time) {
2063                 mlog_errno(ret);
2064                 goto bail;
2065         }
2066
2067         reg->hr_debug_pinned =
2068                         o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2069                                           reg->hr_debug_dir,
2070                                           &(reg->hr_db_pinned),
2071                                           sizeof(*(reg->hr_db_pinned)),
2072                                           O2HB_DB_TYPE_REGION_PINNED,
2073                                           0, 0, reg);
2074         if (!reg->hr_debug_pinned) {
2075                 mlog_errno(ret);
2076                 goto bail;
2077         }
2078
2079         ret = 0;
2080 bail:
2081         return ret;
2082 }
2083
2084 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2085                                                           const char *name)
2086 {
2087         struct o2hb_region *reg = NULL;
2088         int ret;
2089
2090         reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2091         if (reg == NULL)
2092                 return ERR_PTR(-ENOMEM);
2093
2094         if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2095                 ret = -ENAMETOOLONG;
2096                 goto free;
2097         }
2098
2099         spin_lock(&o2hb_live_lock);
2100         reg->hr_region_num = 0;
2101         if (o2hb_global_heartbeat_active()) {
2102                 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2103                                                          O2NM_MAX_REGIONS);
2104                 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2105                         spin_unlock(&o2hb_live_lock);
2106                         ret = -EFBIG;
2107                         goto free;
2108                 }
2109                 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2110         }
2111         list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2112         spin_unlock(&o2hb_live_lock);
2113
2114         config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2115
2116         ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2117         if (ret) {
2118                 config_item_put(&reg->hr_item);
2119                 goto free;
2120         }
2121
2122         return &reg->hr_item;
2123 free:
2124         kfree(reg);
2125         return ERR_PTR(ret);
2126 }
2127
2128 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2129                                            struct config_item *item)
2130 {
2131         struct task_struct *hb_task;
2132         struct o2hb_region *reg = to_o2hb_region(item);
2133         int quorum_region = 0;
2134
2135         /* stop the thread when the user removes the region dir */
2136         spin_lock(&o2hb_live_lock);
2137         hb_task = reg->hr_task;
2138         reg->hr_task = NULL;
2139         reg->hr_item_dropped = 1;
2140         spin_unlock(&o2hb_live_lock);
2141
2142         if (hb_task)
2143                 kthread_stop(hb_task);
2144
2145         if (o2hb_global_heartbeat_active()) {
2146                 spin_lock(&o2hb_live_lock);
2147                 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2148                 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2149                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2150                         quorum_region = 1;
2151                 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2152                 spin_unlock(&o2hb_live_lock);
2153                 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2154                        ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2155                         "stopped" : "start aborted"), config_item_name(item),
2156                        reg->hr_dev_name);
2157         }
2158
2159         /*
2160          * If we're racing a dev_write(), we need to wake them.  They will
2161          * check reg->hr_task
2162          */
2163         if (atomic_read(&reg->hr_steady_iterations) != 0) {
2164                 reg->hr_aborted_start = 1;
2165                 atomic_set(&reg->hr_steady_iterations, 0);
2166                 wake_up(&o2hb_steady_queue);
2167         }
2168
2169         config_item_put(item);
2170
2171         if (!o2hb_global_heartbeat_active() || !quorum_region)
2172                 return;
2173
2174         /*
2175          * If global heartbeat active and there are dependent users,
2176          * pin all regions if quorum region count <= CUT_OFF
2177          */
2178         spin_lock(&o2hb_live_lock);
2179
2180         if (!o2hb_dependent_users)
2181                 goto unlock;
2182
2183         if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2184                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2185                 o2hb_region_pin(NULL);
2186
2187 unlock:
2188         spin_unlock(&o2hb_live_lock);
2189 }
2190
2191 struct o2hb_heartbeat_group_attribute {
2192         struct configfs_attribute attr;
2193         ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2194         ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2195 };
2196
2197 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2198                                          struct configfs_attribute *attr,
2199                                          char *page)
2200 {
2201         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2202         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2203                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2204         ssize_t ret = 0;
2205
2206         if (o2hb_heartbeat_group_attr->show)
2207                 ret = o2hb_heartbeat_group_attr->show(reg, page);
2208         return ret;
2209 }
2210
2211 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2212                                           struct configfs_attribute *attr,
2213                                           const char *page, size_t count)
2214 {
2215         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2216         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2217                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2218         ssize_t ret = -EINVAL;
2219
2220         if (o2hb_heartbeat_group_attr->store)
2221                 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2222         return ret;
2223 }
2224
2225 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2226                                                      char *page)
2227 {
2228         return sprintf(page, "%u\n", o2hb_dead_threshold);
2229 }
2230
2231 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2232                                                     const char *page,
2233                                                     size_t count)
2234 {
2235         unsigned long tmp;
2236         char *p = (char *)page;
2237
2238         tmp = simple_strtoul(p, &p, 10);
2239         if (!p || (*p && (*p != '\n')))
2240                 return -EINVAL;
2241
2242         /* this will validate ranges for us. */
2243         o2hb_dead_threshold_set((unsigned int) tmp);
2244
2245         return count;
2246 }
2247
2248 static
2249 ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2250                                        char *page)
2251 {
2252         return sprintf(page, "%s\n",
2253                        o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2254 }
2255
2256 static
2257 ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2258                                         const char *page, size_t count)
2259 {
2260         unsigned int i;
2261         int ret;
2262         size_t len;
2263
2264         len = (page[count - 1] == '\n') ? count - 1 : count;
2265         if (!len)
2266                 return -EINVAL;
2267
2268         for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2269                 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2270                         continue;
2271
2272                 ret = o2hb_global_heartbeat_mode_set(i);
2273                 if (!ret)
2274                         printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2275                                o2hb_heartbeat_mode_desc[i]);
2276                 return count;
2277         }
2278
2279         return -EINVAL;
2280
2281 }
2282
2283 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2284         .attr   = { .ca_owner = THIS_MODULE,
2285                     .ca_name = "dead_threshold",
2286                     .ca_mode = S_IRUGO | S_IWUSR },
2287         .show   = o2hb_heartbeat_group_threshold_show,
2288         .store  = o2hb_heartbeat_group_threshold_store,
2289 };
2290
2291 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2292         .attr   = { .ca_owner = THIS_MODULE,
2293                 .ca_name = "mode",
2294                 .ca_mode = S_IRUGO | S_IWUSR },
2295         .show   = o2hb_heartbeat_group_mode_show,
2296         .store  = o2hb_heartbeat_group_mode_store,
2297 };
2298
2299 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2300         &o2hb_heartbeat_group_attr_threshold.attr,
2301         &o2hb_heartbeat_group_attr_mode.attr,
2302         NULL,
2303 };
2304
2305 static struct configfs_item_operations o2hb_heartbeat_group_item_ops = {
2306         .show_attribute         = o2hb_heartbeat_group_show,
2307         .store_attribute        = o2hb_heartbeat_group_store,
2308 };
2309
2310 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2311         .make_item      = o2hb_heartbeat_group_make_item,
2312         .drop_item      = o2hb_heartbeat_group_drop_item,
2313 };
2314
2315 static struct config_item_type o2hb_heartbeat_group_type = {
2316         .ct_group_ops   = &o2hb_heartbeat_group_group_ops,
2317         .ct_item_ops    = &o2hb_heartbeat_group_item_ops,
2318         .ct_attrs       = o2hb_heartbeat_group_attrs,
2319         .ct_owner       = THIS_MODULE,
2320 };
2321
2322 /* this is just here to avoid touching group in heartbeat.h which the
2323  * entire damn world #includes */
2324 struct config_group *o2hb_alloc_hb_set(void)
2325 {
2326         struct o2hb_heartbeat_group *hs = NULL;
2327         struct config_group *ret = NULL;
2328
2329         hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2330         if (hs == NULL)
2331                 goto out;
2332
2333         config_group_init_type_name(&hs->hs_group, "heartbeat",
2334                                     &o2hb_heartbeat_group_type);
2335
2336         ret = &hs->hs_group;
2337 out:
2338         if (ret == NULL)
2339                 kfree(hs);
2340         return ret;
2341 }
2342
2343 void o2hb_free_hb_set(struct config_group *group)
2344 {
2345         struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2346         kfree(hs);
2347 }
2348
2349 /* hb callback registration and issuing */
2350
2351 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2352 {
2353         if (type == O2HB_NUM_CB)
2354                 return ERR_PTR(-EINVAL);
2355
2356         return &o2hb_callbacks[type];
2357 }
2358
2359 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2360                          enum o2hb_callback_type type,
2361                          o2hb_cb_func *func,
2362                          void *data,
2363                          int priority)
2364 {
2365         INIT_LIST_HEAD(&hc->hc_item);
2366         hc->hc_func = func;
2367         hc->hc_data = data;
2368         hc->hc_priority = priority;
2369         hc->hc_type = type;
2370         hc->hc_magic = O2HB_CB_MAGIC;
2371 }
2372 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2373
2374 /*
2375  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2376  * In global heartbeat mode, region_uuid passed is NULL.
2377  *
2378  * In local, we only pin the matching region. In global we pin all the active
2379  * regions.
2380  */
2381 static int o2hb_region_pin(const char *region_uuid)
2382 {
2383         int ret = 0, found = 0;
2384         struct o2hb_region *reg;
2385         char *uuid;
2386
2387         assert_spin_locked(&o2hb_live_lock);
2388
2389         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2390                 if (reg->hr_item_dropped)
2391                         continue;
2392
2393                 uuid = config_item_name(&reg->hr_item);
2394
2395                 /* local heartbeat */
2396                 if (region_uuid) {
2397                         if (strcmp(region_uuid, uuid))
2398                                 continue;
2399                         found = 1;
2400                 }
2401
2402                 if (reg->hr_item_pinned || reg->hr_item_dropped)
2403                         goto skip_pin;
2404
2405                 /* Ignore ENOENT only for local hb (userdlm domain) */
2406                 ret = o2nm_depend_item(&reg->hr_item);
2407                 if (!ret) {
2408                         mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2409                         reg->hr_item_pinned = 1;
2410                 } else {
2411                         if (ret == -ENOENT && found)
2412                                 ret = 0;
2413                         else {
2414                                 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2415                                      uuid, ret);
2416                                 break;
2417                         }
2418                 }
2419 skip_pin:
2420                 if (found)
2421                         break;
2422         }
2423
2424         return ret;
2425 }
2426
2427 /*
2428  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2429  * In global heartbeat mode, region_uuid passed is NULL.
2430  *
2431  * In local, we only unpin the matching region. In global we unpin all the
2432  * active regions.
2433  */
2434 static void o2hb_region_unpin(const char *region_uuid)
2435 {
2436         struct o2hb_region *reg;
2437         char *uuid;
2438         int found = 0;
2439
2440         assert_spin_locked(&o2hb_live_lock);
2441
2442         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2443                 if (reg->hr_item_dropped)
2444                         continue;
2445
2446                 uuid = config_item_name(&reg->hr_item);
2447                 if (region_uuid) {
2448                         if (strcmp(region_uuid, uuid))
2449                                 continue;
2450                         found = 1;
2451                 }
2452
2453                 if (reg->hr_item_pinned) {
2454                         mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2455                         o2nm_undepend_item(&reg->hr_item);
2456                         reg->hr_item_pinned = 0;
2457                 }
2458                 if (found)
2459                         break;
2460         }
2461 }
2462
2463 static int o2hb_region_inc_user(const char *region_uuid)
2464 {
2465         int ret = 0;
2466
2467         spin_lock(&o2hb_live_lock);
2468
2469         /* local heartbeat */
2470         if (!o2hb_global_heartbeat_active()) {
2471             ret = o2hb_region_pin(region_uuid);
2472             goto unlock;
2473         }
2474
2475         /*
2476          * if global heartbeat active and this is the first dependent user,
2477          * pin all regions if quorum region count <= CUT_OFF
2478          */
2479         o2hb_dependent_users++;
2480         if (o2hb_dependent_users > 1)
2481                 goto unlock;
2482
2483         if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2484                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2485                 ret = o2hb_region_pin(NULL);
2486
2487 unlock:
2488         spin_unlock(&o2hb_live_lock);
2489         return ret;
2490 }
2491
2492 void o2hb_region_dec_user(const char *region_uuid)
2493 {
2494         spin_lock(&o2hb_live_lock);
2495
2496         /* local heartbeat */
2497         if (!o2hb_global_heartbeat_active()) {
2498             o2hb_region_unpin(region_uuid);
2499             goto unlock;
2500         }
2501
2502         /*
2503          * if global heartbeat active and there are no dependent users,
2504          * unpin all quorum regions
2505          */
2506         o2hb_dependent_users--;
2507         if (!o2hb_dependent_users)
2508                 o2hb_region_unpin(NULL);
2509
2510 unlock:
2511         spin_unlock(&o2hb_live_lock);
2512 }
2513
2514 int o2hb_register_callback(const char *region_uuid,
2515                            struct o2hb_callback_func *hc)
2516 {
2517         struct o2hb_callback_func *f;
2518         struct o2hb_callback *hbcall;
2519         int ret;
2520
2521         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2522         BUG_ON(!list_empty(&hc->hc_item));
2523
2524         hbcall = hbcall_from_type(hc->hc_type);
2525         if (IS_ERR(hbcall)) {
2526                 ret = PTR_ERR(hbcall);
2527                 goto out;
2528         }
2529
2530         if (region_uuid) {
2531                 ret = o2hb_region_inc_user(region_uuid);
2532                 if (ret) {
2533                         mlog_errno(ret);
2534                         goto out;
2535                 }
2536         }
2537
2538         down_write(&o2hb_callback_sem);
2539
2540         list_for_each_entry(f, &hbcall->list, hc_item) {
2541                 if (hc->hc_priority < f->hc_priority) {
2542                         list_add_tail(&hc->hc_item, &f->hc_item);
2543                         break;
2544                 }
2545         }
2546         if (list_empty(&hc->hc_item))
2547                 list_add_tail(&hc->hc_item, &hbcall->list);
2548
2549         up_write(&o2hb_callback_sem);
2550         ret = 0;
2551 out:
2552         mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2553              ret, __builtin_return_address(0), hc);
2554         return ret;
2555 }
2556 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2557
2558 void o2hb_unregister_callback(const char *region_uuid,
2559                               struct o2hb_callback_func *hc)
2560 {
2561         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2562
2563         mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2564              __builtin_return_address(0), hc);
2565
2566         /* XXX Can this happen _with_ a region reference? */
2567         if (list_empty(&hc->hc_item))
2568                 return;
2569
2570         if (region_uuid)
2571                 o2hb_region_dec_user(region_uuid);
2572
2573         down_write(&o2hb_callback_sem);
2574
2575         list_del_init(&hc->hc_item);
2576
2577         up_write(&o2hb_callback_sem);
2578 }
2579 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2580
2581 int o2hb_check_node_heartbeating(u8 node_num)
2582 {
2583         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2584
2585         o2hb_fill_node_map(testing_map, sizeof(testing_map));
2586         if (!test_bit(node_num, testing_map)) {
2587                 mlog(ML_HEARTBEAT,
2588                      "node (%u) does not have heartbeating enabled.\n",
2589                      node_num);
2590                 return 0;
2591         }
2592
2593         return 1;
2594 }
2595 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2596
2597 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2598 {
2599         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2600
2601         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2602         if (!test_bit(node_num, testing_map)) {
2603                 mlog(ML_HEARTBEAT,
2604                      "node (%u) does not have heartbeating enabled.\n",
2605                      node_num);
2606                 return 0;
2607         }
2608
2609         return 1;
2610 }
2611 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2612
2613 /* Makes sure our local node is configured with a node number, and is
2614  * heartbeating. */
2615 int o2hb_check_local_node_heartbeating(void)
2616 {
2617         u8 node_num;
2618
2619         /* if this node was set then we have networking */
2620         node_num = o2nm_this_node();
2621         if (node_num == O2NM_MAX_NODES) {
2622                 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2623                 return 0;
2624         }
2625
2626         return o2hb_check_node_heartbeating(node_num);
2627 }
2628 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2629
2630 /*
2631  * this is just a hack until we get the plumbing which flips file systems
2632  * read only and drops the hb ref instead of killing the node dead.
2633  */
2634 void o2hb_stop_all_regions(void)
2635 {
2636         struct o2hb_region *reg;
2637
2638         mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2639
2640         spin_lock(&o2hb_live_lock);
2641
2642         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2643                 reg->hr_unclean_stop = 1;
2644
2645         spin_unlock(&o2hb_live_lock);
2646 }
2647 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2648
2649 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2650 {
2651         struct o2hb_region *reg;
2652         int numregs = 0;
2653         char *p;
2654
2655         spin_lock(&o2hb_live_lock);
2656
2657         p = region_uuids;
2658         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2659                 if (reg->hr_item_dropped)
2660                         continue;
2661
2662                 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2663                 if (numregs < max_regions) {
2664                         memcpy(p, config_item_name(&reg->hr_item),
2665                                O2HB_MAX_REGION_NAME_LEN);
2666                         p += O2HB_MAX_REGION_NAME_LEN;
2667                 }
2668                 numregs++;
2669         }
2670
2671         spin_unlock(&o2hb_live_lock);
2672
2673         return numregs;
2674 }
2675 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2676
2677 int o2hb_global_heartbeat_active(void)
2678 {
2679         return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2680 }
2681 EXPORT_SYMBOL(o2hb_global_heartbeat_active);