fs/ocfs2/cluster/heartbeat.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2 of the License, or (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public
  17  * License along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA.
  20  */
  21
  22 #include <linux/kernel.h>
  23 #include <linux/sched.h>
  24 #include <linux/jiffies.h>
  25 #include <linux/module.h>
  26 #include <linux/fs.h>
  27 #include <linux/bio.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/delay.h>
  30 #include <linux/file.h>
  31 #include <linux/kthread.h>
  32 #include <linux/configfs.h>
  33 #include <linux/random.h>
  34 #include <linux/crc32.h>
  35 #include <linux/time.h>
  36 #include <linux/debugfs.h>
  37 #include <linux/slab.h>
  38
  39 #include "heartbeat.h"
  40 #include "tcp.h"
  41 #include "nodemanager.h"
  42 #include "quorum.h"
  43
  44 #include "masklog.h"
  45
  46
  47 /*
  48  * The first heartbeat pass had one global thread that would serialize all hb
  49  * callback calls.  This global serializing sem should only be removed once
  50  * we've made sure that all callees can deal with being called concurrently
  51  * from multiple hb region threads.
  52  */
  53 static DECLARE_RWSEM(o2hb_callback_sem);
  54
  55 /*
  56  * multiple hb threads are watching multiple regions.  A node is live
  57  * whenever any of the threads sees activity from the node in its region.
  58  */
  59 static DEFINE_SPINLOCK(o2hb_live_lock);
  60 static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
  61 static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
  62 static LIST_HEAD(o2hb_node_events);
  63 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
  64
  65 /*
  66  * In global heartbeat, we maintain a series of region bitmaps.
  67  *      - o2hb_region_bitmap allows us to limit the region number to max region.
  68  *      - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
  69  *      - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
  70  *              heartbeat on it.
  71  *      - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
  72  */
  73 static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  74 static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  75 static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  76 static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
  77
  78 #define O2HB_DB_TYPE_LIVENODES          0
  79 #define O2HB_DB_TYPE_LIVEREGIONS        1
  80 #define O2HB_DB_TYPE_QUORUMREGIONS      2
  81 #define O2HB_DB_TYPE_FAILEDREGIONS      3
  82 #define O2HB_DB_TYPE_REGION_LIVENODES   4
  83 #define O2HB_DB_TYPE_REGION_NUMBER      5
  84 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME        6
  85 #define O2HB_DB_TYPE_REGION_PINNED      7
  86 struct o2hb_debug_buf {
  87         int db_type;
  88         int db_size;
  89         int db_len;
  90         void *db_data;
  91 };
  92
  93 static struct o2hb_debug_buf *o2hb_db_livenodes;
  94 static struct o2hb_debug_buf *o2hb_db_liveregions;
  95 static struct o2hb_debug_buf *o2hb_db_quorumregions;
  96 static struct o2hb_debug_buf *o2hb_db_failedregions;
  97
  98 #define O2HB_DEBUG_DIR                  "o2hb"
  99 #define O2HB_DEBUG_LIVENODES            "livenodes"
 100 #define O2HB_DEBUG_LIVEREGIONS          "live_regions"
 101 #define O2HB_DEBUG_QUORUMREGIONS        "quorum_regions"
 102 #define O2HB_DEBUG_FAILEDREGIONS        "failed_regions"
 103 #define O2HB_DEBUG_REGION_NUMBER        "num"
 104 #define O2HB_DEBUG_REGION_ELAPSED_TIME  "elapsed_time_in_ms"
 105 #define O2HB_DEBUG_REGION_PINNED        "pinned"
 106
 107 static struct dentry *o2hb_debug_dir;
 108 static struct dentry *o2hb_debug_livenodes;
 109 static struct dentry *o2hb_debug_liveregions;
 110 static struct dentry *o2hb_debug_quorumregions;
 111 static struct dentry *o2hb_debug_failedregions;
 112
 113 static LIST_HEAD(o2hb_all_regions);
 114
 115 static struct o2hb_callback {
 116         struct list_head list;
 117 } o2hb_callbacks[O2HB_NUM_CB];
 118
 119 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
 120
 121 #define O2HB_DEFAULT_BLOCK_BITS       9
 122
 123 enum o2hb_heartbeat_modes {
 124         O2HB_HEARTBEAT_LOCAL            = 0,
 125         O2HB_HEARTBEAT_GLOBAL,
 126         O2HB_HEARTBEAT_NUM_MODES,
 127 };
 128
 129 char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
 130                 "local",        /* O2HB_HEARTBEAT_LOCAL */
 131                 "global",       /* O2HB_HEARTBEAT_GLOBAL */
 132 };
 133
 134 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
 135 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
 136
 137 /*
 138  * o2hb_dependent_users tracks the number of registered callbacks that depend
 139  * on heartbeat. o2net and o2dlm are two entities that register this callback.
 140  * However only o2dlm depends on the heartbeat. It does not want the heartbeat
 141  * to stop while a dlm domain is still active.
 142  */
 143 unsigned int o2hb_dependent_users;
 144
 145 /*
 146  * In global heartbeat mode, all regions are pinned if there are one or more
 147  * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
 148  * regions are unpinned if the region count exceeds the cut off or the number
 149  * of dependent users falls to zero.
 150  */
 151 #define O2HB_PIN_CUT_OFF                3
 152
 153 /*
 154  * In local heartbeat mode, we assume the dlm domain name to be the same as
 155  * region uuid. This is true for domains created for the file system but not
 156  * necessarily true for userdlm domains. This is a known limitation.
 157  *
 158  * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
 159  * works for both file system and userdlm domains.
 160  */
 161 static int o2hb_region_pin(const char *region_uuid);
 162 static void o2hb_region_unpin(const char *region_uuid);
 163
 164 /* Only sets a new threshold if there are no active regions.
 165  *
 166  * No locking or otherwise interesting code is required for reading
 167  * o2hb_dead_threshold as it can't change once regions are active and
 168  * it's not interesting to anyone until then anyway. */
 169 static void o2hb_dead_threshold_set(unsigned int threshold)
 170 {
 171         if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
 172                 spin_lock(&o2hb_live_lock);
 173                 if (list_empty(&o2hb_all_regions))
 174                         o2hb_dead_threshold = threshold;
 175                 spin_unlock(&o2hb_live_lock);
 176         }
 177 }
 178
 179 static int o2hb_global_hearbeat_mode_set(unsigned int hb_mode)
 180 {
 181         int ret = -1;
 182
 183         if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
 184                 spin_lock(&o2hb_live_lock);
 185                 if (list_empty(&o2hb_all_regions)) {
 186                         o2hb_heartbeat_mode = hb_mode;
 187                         ret = 0;
 188                 }
 189                 spin_unlock(&o2hb_live_lock);
 190         }
 191
 192         return ret;
 193 }
 194
 195 struct o2hb_node_event {
 196         struct list_head        hn_item;
 197         enum o2hb_callback_type hn_event_type;
 198         struct o2nm_node        *hn_node;
 199         int                     hn_node_num;
 200 };
 201
 202 struct o2hb_disk_slot {
 203         struct o2hb_disk_heartbeat_block *ds_raw_block;
 204         u8                      ds_node_num;
 205         u64                     ds_last_time;
 206         u64                     ds_last_generation;
 207         u16                     ds_equal_samples;
 208         u16                     ds_changed_samples;
 209         struct list_head        ds_live_item;
 210 };
 211
 212 /* each thread owns a region.. when we're asked to tear down the region
 213  * we ask the thread to stop, who cleans up the region */
 214 struct o2hb_region {
 215         struct config_item      hr_item;
 216
 217         struct list_head        hr_all_item;
 218         unsigned                hr_unclean_stop:1,
 219                                 hr_item_pinned:1,
 220                                 hr_item_dropped:1;
 221
 222         /* protected by the hr_callback_sem */
 223         struct task_struct      *hr_task;
 224
 225         unsigned int            hr_blocks;
 226         unsigned long long      hr_start_block;
 227
 228         unsigned int            hr_block_bits;
 229         unsigned int            hr_block_bytes;
 230
 231         unsigned int            hr_slots_per_page;
 232         unsigned int            hr_num_pages;
 233
 234         struct page             **hr_slot_data;
 235         struct block_device     *hr_bdev;
 236         struct o2hb_disk_slot   *hr_slots;
 237
 238         /* live node map of this region */
 239         unsigned long           hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 240         unsigned int            hr_region_num;
 241
 242         struct dentry           *hr_debug_dir;
 243         struct dentry           *hr_debug_livenodes;
 244         struct dentry           *hr_debug_regnum;
 245         struct dentry           *hr_debug_elapsed_time;
 246         struct dentry           *hr_debug_pinned;
 247         struct o2hb_debug_buf   *hr_db_livenodes;
 248         struct o2hb_debug_buf   *hr_db_regnum;
 249         struct o2hb_debug_buf   *hr_db_elapsed_time;
 250         struct o2hb_debug_buf   *hr_db_pinned;
 251
 252         /* let the person setting up hb wait for it to return until it
 253          * has reached a 'steady' state.  This will be fixed when we have
 254          * a more complete api that doesn't lead to this sort of fragility. */
 255         atomic_t                hr_steady_iterations;
 256
 257         char                    hr_dev_name[BDEVNAME_SIZE];
 258
 259         unsigned int            hr_timeout_ms;
 260
 261         /* randomized as the region goes up and down so that a node
 262          * recognizes a node going up and down in one iteration */
 263         u64                     hr_generation;
 264
 265         struct delayed_work     hr_write_timeout_work;
 266         unsigned long           hr_last_timeout_start;
 267
 268         /* Used during o2hb_check_slot to hold a copy of the block
 269          * being checked because we temporarily have to zero out the
 270          * crc field. */
 271         struct o2hb_disk_heartbeat_block *hr_tmp_block;
 272 };
 273
 274 struct o2hb_bio_wait_ctxt {
 275         atomic_t          wc_num_reqs;
 276         struct completion wc_io_complete;
 277         int               wc_error;
 278 };
 279
 280 static int o2hb_pop_count(void *map, int count)
 281 {
 282         int i = -1, pop = 0;
 283
 284         while ((i = find_next_bit(map, count, i + 1)) < count)
 285                 pop++;
 286         return pop;
 287 }
 288
 289 static void o2hb_write_timeout(struct work_struct *work)
 290 {
 291         int failed, quorum;
 292         unsigned long flags;
 293         struct o2hb_region *reg =
 294                 container_of(work, struct o2hb_region,
 295                              hr_write_timeout_work.work);
 296
 297         mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
 298              "milliseconds\n", reg->hr_dev_name,
 299              jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
 300
 301         if (o2hb_global_heartbeat_active()) {
 302                 spin_lock_irqsave(&o2hb_live_lock, flags);
 303                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 304                         set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 305                 failed = o2hb_pop_count(&o2hb_failed_region_bitmap,
 306                                         O2NM_MAX_REGIONS);
 307                 quorum = o2hb_pop_count(&o2hb_quorum_region_bitmap,
 308                                         O2NM_MAX_REGIONS);
 309                 spin_unlock_irqrestore(&o2hb_live_lock, flags);
 310
 311                 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
 312                      quorum, failed);
 313
 314                 /*
 315                  * Fence if the number of failed regions >= half the number
 316                  * of  quorum regions
 317                  */
 318                 if ((failed << 1) < quorum)
 319                         return;
 320         }
 321
 322         o2quo_disk_timeout();
 323 }
 324
 325 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
 326 {
 327         mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
 328              O2HB_MAX_WRITE_TIMEOUT_MS);
 329
 330         if (o2hb_global_heartbeat_active()) {
 331                 spin_lock(&o2hb_live_lock);
 332                 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 333                 spin_unlock(&o2hb_live_lock);
 334         }
 335         cancel_delayed_work(&reg->hr_write_timeout_work);
 336         reg->hr_last_timeout_start = jiffies;
 337         schedule_delayed_work(&reg->hr_write_timeout_work,
 338                               msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
 339 }
 340
 341 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
 342 {
 343         cancel_delayed_work_sync(&reg->hr_write_timeout_work);
 344 }
 345
 346 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
 347 {
 348         atomic_set(&wc->wc_num_reqs, 1);
 349         init_completion(&wc->wc_io_complete);
 350         wc->wc_error = 0;
 351 }
 352
 353 /* Used in error paths too */
 354 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
 355                                      unsigned int num)
 356 {
 357         /* sadly atomic_sub_and_test() isn't available on all platforms.  The
 358          * good news is that the fast path only completes one at a time */
 359         while(num--) {
 360                 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
 361                         BUG_ON(num > 0);
 362                         complete(&wc->wc_io_complete);
 363                 }
 364         }
 365 }
 366
 367 static void o2hb_wait_on_io(struct o2hb_region *reg,
 368                             struct o2hb_bio_wait_ctxt *wc)
 369 {
 370         struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping;
 371
 372         blk_run_address_space(mapping);
 373         o2hb_bio_wait_dec(wc, 1);
 374
 375         wait_for_completion(&wc->wc_io_complete);
 376 }
 377
 378 static void o2hb_bio_end_io(struct bio *bio,
 379                            int error)
 380 {
 381         struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
 382
 383         if (error) {
 384                 mlog(ML_ERROR, "IO Error %d\n", error);
 385                 wc->wc_error = error;
 386         }
 387
 388         o2hb_bio_wait_dec(wc, 1);
 389         bio_put(bio);
 390 }
 391
 392 /* Setup a Bio to cover I/O against num_slots slots starting at
 393  * start_slot. */
 394 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
 395                                       struct o2hb_bio_wait_ctxt *wc,
 396                                       unsigned int *current_slot,
 397                                       unsigned int max_slots)
 398 {
 399         int len, current_page;
 400         unsigned int vec_len, vec_start;
 401         unsigned int bits = reg->hr_block_bits;
 402         unsigned int spp = reg->hr_slots_per_page;
 403         unsigned int cs = *current_slot;
 404         struct bio *bio;
 405         struct page *page;
 406
 407         /* Testing has shown this allocation to take long enough under
 408          * GFP_KERNEL that the local node can get fenced. It would be
 409          * nicest if we could pre-allocate these bios and avoid this
 410          * all together. */
 411         bio = bio_alloc(GFP_ATOMIC, 16);
 412         if (!bio) {
 413                 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
 414                 bio = ERR_PTR(-ENOMEM);
 415                 goto bail;
 416         }
 417
 418         /* Must put everything in 512 byte sectors for the bio... */
 419         bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
 420         bio->bi_bdev = reg->hr_bdev;
 421         bio->bi_private = wc;
 422         bio->bi_end_io = o2hb_bio_end_io;
 423
 424         vec_start = (cs << bits) % PAGE_CACHE_SIZE;
 425         while(cs < max_slots) {
 426                 current_page = cs / spp;
 427                 page = reg->hr_slot_data[current_page];
 428
 429                 vec_len = min(PAGE_CACHE_SIZE - vec_start,
 430                               (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
 431
 432                 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
 433                      current_page, vec_len, vec_start);
 434
 435                 len = bio_add_page(bio, page, vec_len, vec_start);
 436                 if (len != vec_len) break;
 437
 438                 cs += vec_len / (PAGE_CACHE_SIZE/spp);
 439                 vec_start = 0;
 440         }
 441
 442 bail:
 443         *current_slot = cs;
 444         return bio;
 445 }
 446
 447 static int o2hb_read_slots(struct o2hb_region *reg,
 448                            unsigned int max_slots)
 449 {
 450         unsigned int current_slot=0;
 451         int status;
 452         struct o2hb_bio_wait_ctxt wc;
 453         struct bio *bio;
 454
 455         o2hb_bio_wait_init(&wc);
 456
 457         while(current_slot < max_slots) {
 458                 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
 459                 if (IS_ERR(bio)) {
 460                         status = PTR_ERR(bio);
 461                         mlog_errno(status);
 462                         goto bail_and_wait;
 463                 }
 464
 465                 atomic_inc(&wc.wc_num_reqs);
 466                 submit_bio(READ, bio);
 467         }
 468
 469         status = 0;
 470
 471 bail_and_wait:
 472         o2hb_wait_on_io(reg, &wc);
 473         if (wc.wc_error && !status)
 474                 status = wc.wc_error;
 475
 476         return status;
 477 }
 478
 479 static int o2hb_issue_node_write(struct o2hb_region *reg,
 480                                  struct o2hb_bio_wait_ctxt *write_wc)
 481 {
 482         int status;
 483         unsigned int slot;
 484         struct bio *bio;
 485
 486         o2hb_bio_wait_init(write_wc);
 487
 488         slot = o2nm_this_node();
 489
 490         bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
 491         if (IS_ERR(bio)) {
 492                 status = PTR_ERR(bio);
 493                 mlog_errno(status);
 494                 goto bail;
 495         }
 496
 497         atomic_inc(&write_wc->wc_num_reqs);
 498         submit_bio(WRITE, bio);
 499
 500         status = 0;
 501 bail:
 502         return status;
 503 }
 504
 505 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
 506                                      struct o2hb_disk_heartbeat_block *hb_block)
 507 {
 508         __le32 old_cksum;
 509         u32 ret;
 510
 511         /* We want to compute the block crc with a 0 value in the
 512          * hb_cksum field. Save it off here and replace after the
 513          * crc. */
 514         old_cksum = hb_block->hb_cksum;
 515         hb_block->hb_cksum = 0;
 516
 517         ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
 518
 519         hb_block->hb_cksum = old_cksum;
 520
 521         return ret;
 522 }
 523
 524 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
 525 {
 526         mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
 527              "cksum = 0x%x, generation 0x%llx\n",
 528              (long long)le64_to_cpu(hb_block->hb_seq),
 529              hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
 530              (long long)le64_to_cpu(hb_block->hb_generation));
 531 }
 532
 533 static int o2hb_verify_crc(struct o2hb_region *reg,
 534                            struct o2hb_disk_heartbeat_block *hb_block)
 535 {
 536         u32 read, computed;
 537
 538         read = le32_to_cpu(hb_block->hb_cksum);
 539         computed = o2hb_compute_block_crc_le(reg, hb_block);
 540
 541         return read == computed;
 542 }
 543
 544 /* We want to make sure that nobody is heartbeating on top of us --
 545  * this will help detect an invalid configuration. */
 546 static int o2hb_check_last_timestamp(struct o2hb_region *reg)
 547 {
 548         int node_num, ret;
 549         struct o2hb_disk_slot *slot;
 550         struct o2hb_disk_heartbeat_block *hb_block;
 551
 552         node_num = o2nm_this_node();
 553
 554         ret = 1;
 555         slot = &reg->hr_slots[node_num];
 556         /* Don't check on our 1st timestamp */
 557         if (slot->ds_last_time) {
 558                 hb_block = slot->ds_raw_block;
 559
 560                 if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time)
 561                         ret = 0;
 562         }
 563
 564         return ret;
 565 }
 566
 567 static inline void o2hb_prepare_block(struct o2hb_region *reg,
 568                                       u64 generation)
 569 {
 570         int node_num;
 571         u64 cputime;
 572         struct o2hb_disk_slot *slot;
 573         struct o2hb_disk_heartbeat_block *hb_block;
 574
 575         node_num = o2nm_this_node();
 576         slot = &reg->hr_slots[node_num];
 577
 578         hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
 579         memset(hb_block, 0, reg->hr_block_bytes);
 580         /* TODO: time stuff */
 581         cputime = CURRENT_TIME.tv_sec;
 582         if (!cputime)
 583                 cputime = 1;
 584
 585         hb_block->hb_seq = cpu_to_le64(cputime);
 586         hb_block->hb_node = node_num;
 587         hb_block->hb_generation = cpu_to_le64(generation);
 588         hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
 589
 590         /* This step must always happen last! */
 591         hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
 592                                                                    hb_block));
 593
 594         mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
 595              (long long)generation,
 596              le32_to_cpu(hb_block->hb_cksum));
 597 }
 598
 599 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
 600                                 struct o2nm_node *node,
 601                                 int idx)
 602 {
 603         struct list_head *iter;
 604         struct o2hb_callback_func *f;
 605
 606         list_for_each(iter, &hbcall->list) {
 607                 f = list_entry(iter, struct o2hb_callback_func, hc_item);
 608                 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
 609                 (f->hc_func)(node, idx, f->hc_data);
 610         }
 611 }
 612
 613 /* Will run the list in order until we process the passed event */
 614 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
 615 {
 616         int empty;
 617         struct o2hb_callback *hbcall;
 618         struct o2hb_node_event *event;
 619
 620         spin_lock(&o2hb_live_lock);
 621         empty = list_empty(&queued_event->hn_item);
 622         spin_unlock(&o2hb_live_lock);
 623         if (empty)
 624                 return;
 625
 626         /* Holding callback sem assures we don't alter the callback
 627          * lists when doing this, and serializes ourselves with other
 628          * processes wanting callbacks. */
 629         down_write(&o2hb_callback_sem);
 630
 631         spin_lock(&o2hb_live_lock);
 632         while (!list_empty(&o2hb_node_events)
 633                && !list_empty(&queued_event->hn_item)) {
 634                 event = list_entry(o2hb_node_events.next,
 635                                    struct o2hb_node_event,
 636                                    hn_item);
 637                 list_del_init(&event->hn_item);
 638                 spin_unlock(&o2hb_live_lock);
 639
 640                 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
 641                      event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
 642                      event->hn_node_num);
 643
 644                 hbcall = hbcall_from_type(event->hn_event_type);
 645
 646                 /* We should *never* have gotten on to the list with a
 647                  * bad type... This isn't something that we should try
 648                  * to recover from. */
 649                 BUG_ON(IS_ERR(hbcall));
 650
 651                 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
 652
 653                 spin_lock(&o2hb_live_lock);
 654         }
 655         spin_unlock(&o2hb_live_lock);
 656
 657         up_write(&o2hb_callback_sem);
 658 }
 659
 660 static void o2hb_queue_node_event(struct o2hb_node_event *event,
 661                                   enum o2hb_callback_type type,
 662                                   struct o2nm_node *node,
 663                                   int node_num)
 664 {
 665         assert_spin_locked(&o2hb_live_lock);
 666
 667         BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
 668
 669         event->hn_event_type = type;
 670         event->hn_node = node;
 671         event->hn_node_num = node_num;
 672
 673         mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
 674              type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
 675
 676         list_add_tail(&event->hn_item, &o2hb_node_events);
 677 }
 678
 679 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
 680 {
 681         struct o2hb_node_event event =
 682                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 683         struct o2nm_node *node;
 684
 685         node = o2nm_get_node_by_num(slot->ds_node_num);
 686         if (!node)
 687                 return;
 688
 689         spin_lock(&o2hb_live_lock);
 690         if (!list_empty(&slot->ds_live_item)) {
 691                 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
 692                      slot->ds_node_num);
 693
 694                 list_del_init(&slot->ds_live_item);
 695
 696                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 697                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 698
 699                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
 700                                               slot->ds_node_num);
 701                 }
 702         }
 703         spin_unlock(&o2hb_live_lock);
 704
 705         o2hb_run_event_list(&event);
 706
 707         o2nm_node_put(node);
 708 }
 709
 710 static void o2hb_set_quorum_device(struct o2hb_region *reg,
 711                                    struct o2hb_disk_slot *slot)
 712 {
 713         assert_spin_locked(&o2hb_live_lock);
 714
 715         if (!o2hb_global_heartbeat_active())
 716                 return;
 717
 718         if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 719                 return;
 720
 721         /*
 722          * A region can be added to the quorum only when it sees all
 723          * live nodes heartbeat on it. In other words, the region has been
 724          * added to all nodes.
 725          */
 726         if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
 727                    sizeof(o2hb_live_node_bitmap)))
 728                 return;
 729
 730         if (slot->ds_changed_samples < O2HB_LIVE_THRESHOLD)
 731                 return;
 732
 733         printk(KERN_NOTICE "o2hb: Region %s is now a quorum device\n",
 734                config_item_name(&reg->hr_item));
 735
 736         set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
 737
 738         /*
 739          * If global heartbeat active, unpin all regions if the
 740          * region count > CUT_OFF
 741          */
 742         if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
 743                            O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
 744                 o2hb_region_unpin(NULL);
 745 }
 746
 747 static int o2hb_check_slot(struct o2hb_region *reg,
 748                            struct o2hb_disk_slot *slot)
 749 {
 750         int changed = 0, gen_changed = 0;
 751         struct o2hb_node_event event =
 752                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 753         struct o2nm_node *node;
 754         struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
 755         u64 cputime;
 756         unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
 757         unsigned int slot_dead_ms;
 758         int tmp;
 759
 760         memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
 761
 762         /*
 763          * If a node is no longer configured but is still in the livemap, we
 764          * may need to clear that bit from the livemap.
 765          */
 766         node = o2nm_get_node_by_num(slot->ds_node_num);
 767         if (!node) {
 768                 spin_lock(&o2hb_live_lock);
 769                 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 770                 spin_unlock(&o2hb_live_lock);
 771                 if (!tmp)
 772                         return 0;
 773         }
 774
 775         if (!o2hb_verify_crc(reg, hb_block)) {
 776                 /* all paths from here will drop o2hb_live_lock for
 777                  * us. */
 778                 spin_lock(&o2hb_live_lock);
 779
 780                 /* Don't print an error on the console in this case -
 781                  * a freshly formatted heartbeat area will not have a
 782                  * crc set on it. */
 783                 if (list_empty(&slot->ds_live_item))
 784                         goto out;
 785
 786                 /* The node is live but pushed out a bad crc. We
 787                  * consider it a transient miss but don't populate any
 788                  * other values as they may be junk. */
 789                 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
 790                      slot->ds_node_num, reg->hr_dev_name);
 791                 o2hb_dump_slot(hb_block);
 792
 793                 slot->ds_equal_samples++;
 794                 goto fire_callbacks;
 795         }
 796
 797         /* we don't care if these wrap.. the state transitions below
 798          * clear at the right places */
 799         cputime = le64_to_cpu(hb_block->hb_seq);
 800         if (slot->ds_last_time != cputime)
 801                 slot->ds_changed_samples++;
 802         else
 803                 slot->ds_equal_samples++;
 804         slot->ds_last_time = cputime;
 805
 806         /* The node changed heartbeat generations. We assume this to
 807          * mean it dropped off but came back before we timed out. We
 808          * want to consider it down for the time being but don't want
 809          * to lose any changed_samples state we might build up to
 810          * considering it live again. */
 811         if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
 812                 gen_changed = 1;
 813                 slot->ds_equal_samples = 0;
 814                 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
 815                      "to 0x%llx)\n", slot->ds_node_num,
 816                      (long long)slot->ds_last_generation,
 817                      (long long)le64_to_cpu(hb_block->hb_generation));
 818         }
 819
 820         slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
 821
 822         mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
 823              "seq %llu last %llu changed %u equal %u\n",
 824              slot->ds_node_num, (long long)slot->ds_last_generation,
 825              le32_to_cpu(hb_block->hb_cksum),
 826              (unsigned long long)le64_to_cpu(hb_block->hb_seq),
 827              (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
 828              slot->ds_equal_samples);
 829
 830         spin_lock(&o2hb_live_lock);
 831
 832 fire_callbacks:
 833         /* dead nodes only come to life after some number of
 834          * changes at any time during their dead time */
 835         if (list_empty(&slot->ds_live_item) &&
 836             slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
 837                 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
 838                      slot->ds_node_num, (long long)slot->ds_last_generation);
 839
 840                 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 841
 842                 /* first on the list generates a callback */
 843                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 844                         mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
 845                              "bitmap\n", slot->ds_node_num);
 846                         set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 847
 848                         o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
 849                                               slot->ds_node_num);
 850
 851                         changed = 1;
 852                 }
 853
 854                 list_add_tail(&slot->ds_live_item,
 855                               &o2hb_live_slots[slot->ds_node_num]);
 856
 857                 slot->ds_equal_samples = 0;
 858
 859                 /* We want to be sure that all nodes agree on the
 860                  * number of milliseconds before a node will be
 861                  * considered dead. The self-fencing timeout is
 862                  * computed from this value, and a discrepancy might
 863                  * result in heartbeat calling a node dead when it
 864                  * hasn't self-fenced yet. */
 865                 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
 866                 if (slot_dead_ms && slot_dead_ms != dead_ms) {
 867                         /* TODO: Perhaps we can fail the region here. */
 868                         mlog(ML_ERROR, "Node %d on device %s has a dead count "
 869                              "of %u ms, but our count is %u ms.\n"
 870                              "Please double check your configuration values "
 871                              "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
 872                              slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
 873                              dead_ms);
 874                 }
 875                 goto out;
 876         }
 877
 878         /* if the list is dead, we're done.. */
 879         if (list_empty(&slot->ds_live_item))
 880                 goto out;
 881
 882         /* live nodes only go dead after enough consequtive missed
 883          * samples..  reset the missed counter whenever we see
 884          * activity */
 885         if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
 886                 mlog(ML_HEARTBEAT, "Node %d left my region\n",
 887                      slot->ds_node_num);
 888
 889                 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 890
 891                 /* last off the live_slot generates a callback */
 892                 list_del_init(&slot->ds_live_item);
 893                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 894                         mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
 895                              "nodes bitmap\n", slot->ds_node_num);
 896                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 897
 898                         /* node can be null */
 899                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
 900                                               node, slot->ds_node_num);
 901
 902                         changed = 1;
 903                 }
 904
 905                 /* We don't clear this because the node is still
 906                  * actually writing new blocks. */
 907                 if (!gen_changed)
 908                         slot->ds_changed_samples = 0;
 909                 goto out;
 910         }
 911         if (slot->ds_changed_samples) {
 912                 slot->ds_changed_samples = 0;
 913                 slot->ds_equal_samples = 0;
 914         }
 915 out:
 916         o2hb_set_quorum_device(reg, slot);
 917
 918         spin_unlock(&o2hb_live_lock);
 919
 920         o2hb_run_event_list(&event);
 921
 922         if (node)
 923                 o2nm_node_put(node);
 924         return changed;
 925 }
 926
 927 /* This could be faster if we just implmented a find_last_bit, but I
 928  * don't think the circumstances warrant it. */
 929 static int o2hb_highest_node(unsigned long *nodes,
 930                              int numbits)
 931 {
 932         int highest, node;
 933
 934         highest = numbits;
 935         node = -1;
 936         while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
 937                 if (node >= numbits)
 938                         break;
 939
 940                 highest = node;
 941         }
 942
 943         return highest;
 944 }
 945
 946 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
 947 {
 948         int i, ret, highest_node, change = 0;
 949         unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
 950         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 951         struct o2hb_bio_wait_ctxt write_wc;
 952
 953         ret = o2nm_configured_node_map(configured_nodes,
 954                                        sizeof(configured_nodes));
 955         if (ret) {
 956                 mlog_errno(ret);
 957                 return ret;
 958         }
 959
 960         /*
 961          * If a node is not configured but is in the livemap, we still need
 962          * to read the slot so as to be able to remove it from the livemap.
 963          */
 964         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
 965         i = -1;
 966         while ((i = find_next_bit(live_node_bitmap,
 967                                   O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
 968                 set_bit(i, configured_nodes);
 969         }
 970
 971         highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
 972         if (highest_node >= O2NM_MAX_NODES) {
 973                 mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n");
 974                 return -EINVAL;
 975         }
 976
 977         /* No sense in reading the slots of nodes that don't exist
 978          * yet. Of course, if the node definitions have holes in them
 979          * then we're reading an empty slot anyway... Consider this
 980          * best-effort. */
 981         ret = o2hb_read_slots(reg, highest_node + 1);
 982         if (ret < 0) {
 983                 mlog_errno(ret);
 984                 return ret;
 985         }
 986
 987         /* With an up to date view of the slots, we can check that no
 988          * other node has been improperly configured to heartbeat in
 989          * our slot. */
 990         if (!o2hb_check_last_timestamp(reg))
 991                 mlog(ML_ERROR, "Device \"%s\": another node is heartbeating "
 992                      "in our slot!\n", reg->hr_dev_name);
 993
 994         /* fill in the proper info for our next heartbeat */
 995         o2hb_prepare_block(reg, reg->hr_generation);
 996
 997         /* And fire off the write. Note that we don't wait on this I/O
 998          * until later. */
 999         ret = o2hb_issue_node_write(reg, &write_wc);
1000         if (ret < 0) {
1001                 mlog_errno(ret);
1002                 return ret;
1003         }
1004
1005         i = -1;
1006         while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1007
1008                 change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1009         }
1010
1011         /*
1012          * We have to be sure we've advertised ourselves on disk
1013          * before we can go to steady state.  This ensures that
1014          * people we find in our steady state have seen us.
1015          */
1016         o2hb_wait_on_io(reg, &write_wc);
1017         if (write_wc.wc_error) {
1018                 /* Do not re-arm the write timeout on I/O error - we
1019                  * can't be sure that the new block ever made it to
1020                  * disk */
1021                 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1022                      write_wc.wc_error, reg->hr_dev_name);
1023                 return write_wc.wc_error;
1024         }
1025
1026         o2hb_arm_write_timeout(reg);
1027
1028         /* let the person who launched us know when things are steady */
1029         if (!change && (atomic_read(&reg->hr_steady_iterations) != 0)) {
1030                 if (atomic_dec_and_test(&reg->hr_steady_iterations))
1031                         wake_up(&o2hb_steady_queue);
1032         }
1033
1034         return 0;
1035 }
1036
1037 /* Subtract b from a, storing the result in a. a *must* have a larger
1038  * value than b. */
1039 static void o2hb_tv_subtract(struct timeval *a,
1040                              struct timeval *b)
1041 {
1042         /* just return 0 when a is after b */
1043         if (a->tv_sec < b->tv_sec ||
1044             (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1045                 a->tv_sec = 0;
1046                 a->tv_usec = 0;
1047                 return;
1048         }
1049
1050         a->tv_sec -= b->tv_sec;
1051         a->tv_usec -= b->tv_usec;
1052         while ( a->tv_usec < 0 ) {
1053                 a->tv_sec--;
1054                 a->tv_usec += 1000000;
1055         }
1056 }
1057
1058 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1059                                        struct timeval *end)
1060 {
1061         struct timeval res = *end;
1062
1063         o2hb_tv_subtract(&res, start);
1064
1065         return res.tv_sec * 1000 + res.tv_usec / 1000;
1066 }
1067
1068 /*
1069  * we ride the region ref that the region dir holds.  before the region
1070  * dir is removed and drops it ref it will wait to tear down this
1071  * thread.
1072  */
1073 static int o2hb_thread(void *data)
1074 {
1075         int i, ret;
1076         struct o2hb_region *reg = data;
1077         struct o2hb_bio_wait_ctxt write_wc;
1078         struct timeval before_hb, after_hb;
1079         unsigned int elapsed_msec;
1080
1081         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1082
1083         set_user_nice(current, -20);
1084
1085         /* Pin node */
1086         o2nm_depend_this_node();
1087
1088         while (!kthread_should_stop() && !reg->hr_unclean_stop) {
1089                 /* We track the time spent inside
1090                  * o2hb_do_disk_heartbeat so that we avoid more than
1091                  * hr_timeout_ms between disk writes. On busy systems
1092                  * this should result in a heartbeat which is less
1093                  * likely to time itself out. */
1094                 do_gettimeofday(&before_hb);
1095
1096                 i = 0;
1097                 do {
1098                         ret = o2hb_do_disk_heartbeat(reg);
1099                 } while (ret && ++i < 2);
1100
1101                 do_gettimeofday(&after_hb);
1102                 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1103
1104                 mlog(ML_HEARTBEAT,
1105                      "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
1106                      before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1107                      after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1108                      elapsed_msec);
1109
1110                 if (elapsed_msec < reg->hr_timeout_ms) {
1111                         /* the kthread api has blocked signals for us so no
1112                          * need to record the return value. */
1113                         msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1114                 }
1115         }
1116
1117         o2hb_disarm_write_timeout(reg);
1118
1119         /* unclean stop is only used in very bad situation */
1120         for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1121                 o2hb_shutdown_slot(&reg->hr_slots[i]);
1122
1123         /* Explicit down notification - avoid forcing the other nodes
1124          * to timeout on this region when we could just as easily
1125          * write a clear generation - thus indicating to them that
1126          * this node has left this region.
1127          *
1128          * XXX: Should we skip this on unclean_stop? */
1129         o2hb_prepare_block(reg, 0);
1130         ret = o2hb_issue_node_write(reg, &write_wc);
1131         if (ret == 0) {
1132                 o2hb_wait_on_io(reg, &write_wc);
1133         } else {
1134                 mlog_errno(ret);
1135         }
1136
1137         /* Unpin node */
1138         o2nm_undepend_this_node();
1139
1140         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n");
1141
1142         return 0;
1143 }
1144
1145 #ifdef CONFIG_DEBUG_FS
1146 static int o2hb_debug_open(struct inode *inode, struct file *file)
1147 {
1148         struct o2hb_debug_buf *db = inode->i_private;
1149         struct o2hb_region *reg;
1150         unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1151         char *buf = NULL;
1152         int i = -1;
1153         int out = 0;
1154
1155         /* max_nodes should be the largest bitmap we pass here */
1156         BUG_ON(sizeof(map) < db->db_size);
1157
1158         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1159         if (!buf)
1160                 goto bail;
1161
1162         switch (db->db_type) {
1163         case O2HB_DB_TYPE_LIVENODES:
1164         case O2HB_DB_TYPE_LIVEREGIONS:
1165         case O2HB_DB_TYPE_QUORUMREGIONS:
1166         case O2HB_DB_TYPE_FAILEDREGIONS:
1167                 spin_lock(&o2hb_live_lock);
1168                 memcpy(map, db->db_data, db->db_size);
1169                 spin_unlock(&o2hb_live_lock);
1170                 break;
1171
1172         case O2HB_DB_TYPE_REGION_LIVENODES:
1173                 spin_lock(&o2hb_live_lock);
1174                 reg = (struct o2hb_region *)db->db_data;
1175                 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1176                 spin_unlock(&o2hb_live_lock);
1177                 break;
1178
1179         case O2HB_DB_TYPE_REGION_NUMBER:
1180                 reg = (struct o2hb_region *)db->db_data;
1181                 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1182                                 reg->hr_region_num);
1183                 goto done;
1184
1185         case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1186                 reg = (struct o2hb_region *)db->db_data;
1187                 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1188                                 jiffies_to_msecs(jiffies -
1189                                                  reg->hr_last_timeout_start));
1190                 goto done;
1191
1192         case O2HB_DB_TYPE_REGION_PINNED:
1193                 reg = (struct o2hb_region *)db->db_data;
1194                 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1195                                 !!reg->hr_item_pinned);
1196                 goto done;
1197
1198         default:
1199                 goto done;
1200         }
1201
1202         while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1203                 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1204         out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1205
1206 done:
1207         i_size_write(inode, out);
1208
1209         file->private_data = buf;
1210
1211         return 0;
1212 bail:
1213         return -ENOMEM;
1214 }
1215
1216 static int o2hb_debug_release(struct inode *inode, struct file *file)
1217 {
1218         kfree(file->private_data);
1219         return 0;
1220 }
1221
1222 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1223                                  size_t nbytes, loff_t *ppos)
1224 {
1225         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1226                                        i_size_read(file->f_mapping->host));
1227 }
1228 #else
1229 static int o2hb_debug_open(struct inode *inode, struct file *file)
1230 {
1231         return 0;
1232 }
1233 static int o2hb_debug_release(struct inode *inode, struct file *file)
1234 {
1235         return 0;
1236 }
1237 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1238                                size_t nbytes, loff_t *ppos)
1239 {
1240         return 0;
1241 }
1242 #endif  /* CONFIG_DEBUG_FS */
1243
1244 static const struct file_operations o2hb_debug_fops = {
1245         .open =         o2hb_debug_open,
1246         .release =      o2hb_debug_release,
1247         .read =         o2hb_debug_read,
1248         .llseek =       generic_file_llseek,
1249 };
1250
1251 void o2hb_exit(void)
1252 {
1253         kfree(o2hb_db_livenodes);
1254         kfree(o2hb_db_liveregions);
1255         kfree(o2hb_db_quorumregions);
1256         kfree(o2hb_db_failedregions);
1257         debugfs_remove(o2hb_debug_failedregions);
1258         debugfs_remove(o2hb_debug_quorumregions);
1259         debugfs_remove(o2hb_debug_liveregions);
1260         debugfs_remove(o2hb_debug_livenodes);
1261         debugfs_remove(o2hb_debug_dir);
1262 }
1263
1264 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1265                                         struct o2hb_debug_buf **db, int db_len,
1266                                         int type, int size, int len, void *data)
1267 {
1268         *db = kmalloc(db_len, GFP_KERNEL);
1269         if (!*db)
1270                 return NULL;
1271
1272         (*db)->db_type = type;
1273         (*db)->db_size = size;
1274         (*db)->db_len = len;
1275         (*db)->db_data = data;
1276
1277         return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1278                                    &o2hb_debug_fops);
1279 }
1280
1281 static int o2hb_debug_init(void)
1282 {
1283         int ret = -ENOMEM;
1284
1285         o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1286         if (!o2hb_debug_dir) {
1287                 mlog_errno(ret);
1288                 goto bail;
1289         }
1290
1291         o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1292                                                  o2hb_debug_dir,
1293                                                  &o2hb_db_livenodes,
1294                                                  sizeof(*o2hb_db_livenodes),
1295                                                  O2HB_DB_TYPE_LIVENODES,
1296                                                  sizeof(o2hb_live_node_bitmap),
1297                                                  O2NM_MAX_NODES,
1298                                                  o2hb_live_node_bitmap);
1299         if (!o2hb_debug_livenodes) {
1300                 mlog_errno(ret);
1301                 goto bail;
1302         }
1303
1304         o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1305                                                    o2hb_debug_dir,
1306                                                    &o2hb_db_liveregions,
1307                                                    sizeof(*o2hb_db_liveregions),
1308                                                    O2HB_DB_TYPE_LIVEREGIONS,
1309                                                    sizeof(o2hb_live_region_bitmap),
1310                                                    O2NM_MAX_REGIONS,
1311                                                    o2hb_live_region_bitmap);
1312         if (!o2hb_debug_liveregions) {
1313                 mlog_errno(ret);
1314                 goto bail;
1315         }
1316
1317         o2hb_debug_quorumregions =
1318                         o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1319                                           o2hb_debug_dir,
1320                                           &o2hb_db_quorumregions,
1321                                           sizeof(*o2hb_db_quorumregions),
1322                                           O2HB_DB_TYPE_QUORUMREGIONS,
1323                                           sizeof(o2hb_quorum_region_bitmap),
1324                                           O2NM_MAX_REGIONS,
1325                                           o2hb_quorum_region_bitmap);
1326         if (!o2hb_debug_quorumregions) {
1327                 mlog_errno(ret);
1328                 goto bail;
1329         }
1330
1331         o2hb_debug_failedregions =
1332                         o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1333                                           o2hb_debug_dir,
1334                                           &o2hb_db_failedregions,
1335                                           sizeof(*o2hb_db_failedregions),
1336                                           O2HB_DB_TYPE_FAILEDREGIONS,
1337                                           sizeof(o2hb_failed_region_bitmap),
1338                                           O2NM_MAX_REGIONS,
1339                                           o2hb_failed_region_bitmap);
1340         if (!o2hb_debug_failedregions) {
1341                 mlog_errno(ret);
1342                 goto bail;
1343         }
1344
1345         ret = 0;
1346 bail:
1347         if (ret)
1348                 o2hb_exit();
1349
1350         return ret;
1351 }
1352
1353 int o2hb_init(void)
1354 {
1355         int i;
1356
1357         for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1358                 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1359
1360         for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1361                 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1362
1363         INIT_LIST_HEAD(&o2hb_node_events);
1364
1365         memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1366         memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1367         memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1368         memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1369         memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1370
1371         o2hb_dependent_users = 0;
1372
1373         return o2hb_debug_init();
1374 }
1375
1376 /* if we're already in a callback then we're already serialized by the sem */
1377 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1378                                              unsigned bytes)
1379 {
1380         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1381
1382         memcpy(map, &o2hb_live_node_bitmap, bytes);
1383 }
1384
1385 /*
1386  * get a map of all nodes that are heartbeating in any regions
1387  */
1388 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1389 {
1390         /* callers want to serialize this map and callbacks so that they
1391          * can trust that they don't miss nodes coming to the party */
1392         down_read(&o2hb_callback_sem);
1393         spin_lock(&o2hb_live_lock);
1394         o2hb_fill_node_map_from_callback(map, bytes);
1395         spin_unlock(&o2hb_live_lock);
1396         up_read(&o2hb_callback_sem);
1397 }
1398 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1399
1400 /*
1401  * heartbeat configfs bits.  The heartbeat set is a default set under
1402  * the cluster set in nodemanager.c.
1403  */
1404
1405 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1406 {
1407         return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1408 }
1409
1410 /* drop_item only drops its ref after killing the thread, nothing should
1411  * be using the region anymore.  this has to clean up any state that
1412  * attributes might have built up. */
1413 static void o2hb_region_release(struct config_item *item)
1414 {
1415         int i;
1416         struct page *page;
1417         struct o2hb_region *reg = to_o2hb_region(item);
1418
1419         if (reg->hr_tmp_block)
1420                 kfree(reg->hr_tmp_block);
1421
1422         if (reg->hr_slot_data) {
1423                 for (i = 0; i < reg->hr_num_pages; i++) {
1424                         page = reg->hr_slot_data[i];
1425                         if (page)
1426                                 __free_page(page);
1427                 }
1428                 kfree(reg->hr_slot_data);
1429         }
1430
1431         if (reg->hr_bdev)
1432                 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1433
1434         if (reg->hr_slots)
1435                 kfree(reg->hr_slots);
1436
1437         kfree(reg->hr_db_regnum);
1438         kfree(reg->hr_db_livenodes);
1439         debugfs_remove(reg->hr_debug_livenodes);
1440         debugfs_remove(reg->hr_debug_regnum);
1441         debugfs_remove(reg->hr_debug_elapsed_time);
1442         debugfs_remove(reg->hr_debug_pinned);
1443         debugfs_remove(reg->hr_debug_dir);
1444
1445         spin_lock(&o2hb_live_lock);
1446         list_del(&reg->hr_all_item);
1447         spin_unlock(&o2hb_live_lock);
1448
1449         kfree(reg);
1450 }
1451
1452 static int o2hb_read_block_input(struct o2hb_region *reg,
1453                                  const char *page,
1454                                  size_t count,
1455                                  unsigned long *ret_bytes,
1456                                  unsigned int *ret_bits)
1457 {
1458         unsigned long bytes;
1459         char *p = (char *)page;
1460
1461         bytes = simple_strtoul(p, &p, 0);
1462         if (!p || (*p && (*p != '\n')))
1463                 return -EINVAL;
1464
1465         /* Heartbeat and fs min / max block sizes are the same. */
1466         if (bytes > 4096 || bytes < 512)
1467                 return -ERANGE;
1468         if (hweight16(bytes) != 1)
1469                 return -EINVAL;
1470
1471         if (ret_bytes)
1472                 *ret_bytes = bytes;
1473         if (ret_bits)
1474                 *ret_bits = ffs(bytes) - 1;
1475
1476         return 0;
1477 }
1478
1479 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1480                                             char *page)
1481 {
1482         return sprintf(page, "%u\n", reg->hr_block_bytes);
1483 }
1484
1485 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1486                                              const char *page,
1487                                              size_t count)
1488 {
1489         int status;
1490         unsigned long block_bytes;
1491         unsigned int block_bits;
1492
1493         if (reg->hr_bdev)
1494                 return -EINVAL;
1495
1496         status = o2hb_read_block_input(reg, page, count,
1497                                        &block_bytes, &block_bits);
1498         if (status)
1499                 return status;
1500
1501         reg->hr_block_bytes = (unsigned int)block_bytes;
1502         reg->hr_block_bits = block_bits;
1503
1504         return count;
1505 }
1506
1507 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1508                                             char *page)
1509 {
1510         return sprintf(page, "%llu\n", reg->hr_start_block);
1511 }
1512
1513 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1514                                              const char *page,
1515                                              size_t count)
1516 {
1517         unsigned long long tmp;
1518         char *p = (char *)page;
1519
1520         if (reg->hr_bdev)
1521                 return -EINVAL;
1522
1523         tmp = simple_strtoull(p, &p, 0);
1524         if (!p || (*p && (*p != '\n')))
1525                 return -EINVAL;
1526
1527         reg->hr_start_block = tmp;
1528
1529         return count;
1530 }
1531
1532 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1533                                        char *page)
1534 {
1535         return sprintf(page, "%d\n", reg->hr_blocks);
1536 }
1537
1538 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1539                                         const char *page,
1540                                         size_t count)
1541 {
1542         unsigned long tmp;
1543         char *p = (char *)page;
1544
1545         if (reg->hr_bdev)
1546                 return -EINVAL;
1547
1548         tmp = simple_strtoul(p, &p, 0);
1549         if (!p || (*p && (*p != '\n')))
1550                 return -EINVAL;
1551
1552         if (tmp > O2NM_MAX_NODES || tmp == 0)
1553                 return -ERANGE;
1554
1555         reg->hr_blocks = (unsigned int)tmp;
1556
1557         return count;
1558 }
1559
1560 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1561                                     char *page)
1562 {
1563         unsigned int ret = 0;
1564
1565         if (reg->hr_bdev)
1566                 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1567
1568         return ret;
1569 }
1570
1571 static void o2hb_init_region_params(struct o2hb_region *reg)
1572 {
1573         reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1574         reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1575
1576         mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1577              reg->hr_start_block, reg->hr_blocks);
1578         mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1579              reg->hr_block_bytes, reg->hr_block_bits);
1580         mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1581         mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1582 }
1583
1584 static int o2hb_map_slot_data(struct o2hb_region *reg)
1585 {
1586         int i, j;
1587         unsigned int last_slot;
1588         unsigned int spp = reg->hr_slots_per_page;
1589         struct page *page;
1590         char *raw;
1591         struct o2hb_disk_slot *slot;
1592
1593         reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1594         if (reg->hr_tmp_block == NULL) {
1595                 mlog_errno(-ENOMEM);
1596                 return -ENOMEM;
1597         }
1598
1599         reg->hr_slots = kcalloc(reg->hr_blocks,
1600                                 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1601         if (reg->hr_slots == NULL) {
1602                 mlog_errno(-ENOMEM);
1603                 return -ENOMEM;
1604         }
1605
1606         for(i = 0; i < reg->hr_blocks; i++) {
1607                 slot = &reg->hr_slots[i];
1608                 slot->ds_node_num = i;
1609                 INIT_LIST_HEAD(&slot->ds_live_item);
1610                 slot->ds_raw_block = NULL;
1611         }
1612
1613         reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1614         mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1615                            "at %u blocks per page\n",
1616              reg->hr_num_pages, reg->hr_blocks, spp);
1617
1618         reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1619                                     GFP_KERNEL);
1620         if (!reg->hr_slot_data) {
1621                 mlog_errno(-ENOMEM);
1622                 return -ENOMEM;
1623         }
1624
1625         for(i = 0; i < reg->hr_num_pages; i++) {
1626                 page = alloc_page(GFP_KERNEL);
1627                 if (!page) {
1628                         mlog_errno(-ENOMEM);
1629                         return -ENOMEM;
1630                 }
1631
1632                 reg->hr_slot_data[i] = page;
1633
1634                 last_slot = i * spp;
1635                 raw = page_address(page);
1636                 for (j = 0;
1637                      (j < spp) && ((j + last_slot) < reg->hr_blocks);
1638                      j++) {
1639                         BUG_ON((j + last_slot) >= reg->hr_blocks);
1640
1641                         slot = &reg->hr_slots[j + last_slot];
1642                         slot->ds_raw_block =
1643                                 (struct o2hb_disk_heartbeat_block *) raw;
1644
1645                         raw += reg->hr_block_bytes;
1646                 }
1647         }
1648
1649         return 0;
1650 }
1651
1652 /* Read in all the slots available and populate the tracking
1653  * structures so that we can start with a baseline idea of what's
1654  * there. */
1655 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1656 {
1657         int ret, i;
1658         struct o2hb_disk_slot *slot;
1659         struct o2hb_disk_heartbeat_block *hb_block;
1660
1661         mlog_entry_void();
1662
1663         ret = o2hb_read_slots(reg, reg->hr_blocks);
1664         if (ret) {
1665                 mlog_errno(ret);
1666                 goto out;
1667         }
1668
1669         /* We only want to get an idea of the values initially in each
1670          * slot, so we do no verification - o2hb_check_slot will
1671          * actually determine if each configured slot is valid and
1672          * whether any values have changed. */
1673         for(i = 0; i < reg->hr_blocks; i++) {
1674                 slot = &reg->hr_slots[i];
1675                 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1676
1677                 /* Only fill the values that o2hb_check_slot uses to
1678                  * determine changing slots */
1679                 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1680                 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1681         }
1682
1683 out:
1684         mlog_exit(ret);
1685         return ret;
1686 }
1687
1688 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1689 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1690                                      const char *page,
1691                                      size_t count)
1692 {
1693         struct task_struct *hb_task;
1694         long fd;
1695         int sectsize;
1696         char *p = (char *)page;
1697         struct file *filp = NULL;
1698         struct inode *inode = NULL;
1699         ssize_t ret = -EINVAL;
1700
1701         if (reg->hr_bdev)
1702                 goto out;
1703
1704         /* We can't heartbeat without having had our node number
1705          * configured yet. */
1706         if (o2nm_this_node() == O2NM_MAX_NODES)
1707                 goto out;
1708
1709         fd = simple_strtol(p, &p, 0);
1710         if (!p || (*p && (*p != '\n')))
1711                 goto out;
1712
1713         if (fd < 0 || fd >= INT_MAX)
1714                 goto out;
1715
1716         filp = fget(fd);
1717         if (filp == NULL)
1718                 goto out;
1719
1720         if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1721             reg->hr_block_bytes == 0)
1722                 goto out;
1723
1724         inode = igrab(filp->f_mapping->host);
1725         if (inode == NULL)
1726                 goto out;
1727
1728         if (!S_ISBLK(inode->i_mode))
1729                 goto out;
1730
1731         reg->hr_bdev = I_BDEV(filp->f_mapping->host);
1732         ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ);
1733         if (ret) {
1734                 reg->hr_bdev = NULL;
1735                 goto out;
1736         }
1737         inode = NULL;
1738
1739         bdevname(reg->hr_bdev, reg->hr_dev_name);
1740
1741         sectsize = bdev_logical_block_size(reg->hr_bdev);
1742         if (sectsize != reg->hr_block_bytes) {
1743                 mlog(ML_ERROR,
1744                      "blocksize %u incorrect for device, expected %d",
1745                      reg->hr_block_bytes, sectsize);
1746                 ret = -EINVAL;
1747                 goto out;
1748         }
1749
1750         o2hb_init_region_params(reg);
1751
1752         /* Generation of zero is invalid */
1753         do {
1754                 get_random_bytes(&reg->hr_generation,
1755                                  sizeof(reg->hr_generation));
1756         } while (reg->hr_generation == 0);
1757
1758         ret = o2hb_map_slot_data(reg);
1759         if (ret) {
1760                 mlog_errno(ret);
1761                 goto out;
1762         }
1763
1764         ret = o2hb_populate_slot_data(reg);
1765         if (ret) {
1766                 mlog_errno(ret);
1767                 goto out;
1768         }
1769
1770         INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1771
1772         /*
1773          * A node is considered live after it has beat LIVE_THRESHOLD
1774          * times.  We're not steady until we've given them a chance
1775          * _after_ our first read.
1776          */
1777         atomic_set(&reg->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1);
1778
1779         hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1780                               reg->hr_item.ci_name);
1781         if (IS_ERR(hb_task)) {
1782                 ret = PTR_ERR(hb_task);
1783                 mlog_errno(ret);
1784                 goto out;
1785         }
1786
1787         spin_lock(&o2hb_live_lock);
1788         reg->hr_task = hb_task;
1789         spin_unlock(&o2hb_live_lock);
1790
1791         ret = wait_event_interruptible(o2hb_steady_queue,
1792                                 atomic_read(&reg->hr_steady_iterations) == 0);
1793         if (ret) {
1794                 /* We got interrupted (hello ptrace!).  Clean up */
1795                 spin_lock(&o2hb_live_lock);
1796                 hb_task = reg->hr_task;
1797                 reg->hr_task = NULL;
1798                 spin_unlock(&o2hb_live_lock);
1799
1800                 if (hb_task)
1801                         kthread_stop(hb_task);
1802                 goto out;
1803         }
1804
1805         /* Ok, we were woken.  Make sure it wasn't by drop_item() */
1806         spin_lock(&o2hb_live_lock);
1807         hb_task = reg->hr_task;
1808         if (o2hb_global_heartbeat_active())
1809                 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1810         spin_unlock(&o2hb_live_lock);
1811
1812         if (hb_task)
1813                 ret = count;
1814         else
1815                 ret = -EIO;
1816
1817         if (hb_task && o2hb_global_heartbeat_active())
1818                 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s\n",
1819                        config_item_name(&reg->hr_item));
1820
1821 out:
1822         if (filp)
1823                 fput(filp);
1824         if (inode)
1825                 iput(inode);
1826         if (ret < 0) {
1827                 if (reg->hr_bdev) {
1828                         blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1829                         reg->hr_bdev = NULL;
1830                 }
1831         }
1832         return ret;
1833 }
1834
1835 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1836                                       char *page)
1837 {
1838         pid_t pid = 0;
1839
1840         spin_lock(&o2hb_live_lock);
1841         if (reg->hr_task)
1842                 pid = task_pid_nr(reg->hr_task);
1843         spin_unlock(&o2hb_live_lock);
1844
1845         if (!pid)
1846                 return 0;
1847
1848         return sprintf(page, "%u\n", pid);
1849 }
1850
1851 struct o2hb_region_attribute {
1852         struct configfs_attribute attr;
1853         ssize_t (*show)(struct o2hb_region *, char *);
1854         ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1855 };
1856
1857 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1858         .attr   = { .ca_owner = THIS_MODULE,
1859                     .ca_name = "block_bytes",
1860                     .ca_mode = S_IRUGO | S_IWUSR },
1861         .show   = o2hb_region_block_bytes_read,
1862         .store  = o2hb_region_block_bytes_write,
1863 };
1864
1865 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1866         .attr   = { .ca_owner = THIS_MODULE,
1867                     .ca_name = "start_block",
1868                     .ca_mode = S_IRUGO | S_IWUSR },
1869         .show   = o2hb_region_start_block_read,
1870         .store  = o2hb_region_start_block_write,
1871 };
1872
1873 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1874         .attr   = { .ca_owner = THIS_MODULE,
1875                     .ca_name = "blocks",
1876                     .ca_mode = S_IRUGO | S_IWUSR },
1877         .show   = o2hb_region_blocks_read,
1878         .store  = o2hb_region_blocks_write,
1879 };
1880
1881 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1882         .attr   = { .ca_owner = THIS_MODULE,
1883                     .ca_name = "dev",
1884                     .ca_mode = S_IRUGO | S_IWUSR },
1885         .show   = o2hb_region_dev_read,
1886         .store  = o2hb_region_dev_write,
1887 };
1888
1889 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1890        .attr   = { .ca_owner = THIS_MODULE,
1891                    .ca_name = "pid",
1892                    .ca_mode = S_IRUGO | S_IRUSR },
1893        .show   = o2hb_region_pid_read,
1894 };
1895
1896 static struct configfs_attribute *o2hb_region_attrs[] = {
1897         &o2hb_region_attr_block_bytes.attr,
1898         &o2hb_region_attr_start_block.attr,
1899         &o2hb_region_attr_blocks.attr,
1900         &o2hb_region_attr_dev.attr,
1901         &o2hb_region_attr_pid.attr,
1902         NULL,
1903 };
1904
1905 static ssize_t o2hb_region_show(struct config_item *item,
1906                                 struct configfs_attribute *attr,
1907                                 char *page)
1908 {
1909         struct o2hb_region *reg = to_o2hb_region(item);
1910         struct o2hb_region_attribute *o2hb_region_attr =
1911                 container_of(attr, struct o2hb_region_attribute, attr);
1912         ssize_t ret = 0;
1913
1914         if (o2hb_region_attr->show)
1915                 ret = o2hb_region_attr->show(reg, page);
1916         return ret;
1917 }
1918
1919 static ssize_t o2hb_region_store(struct config_item *item,
1920                                  struct configfs_attribute *attr,
1921                                  const char *page, size_t count)
1922 {
1923         struct o2hb_region *reg = to_o2hb_region(item);
1924         struct o2hb_region_attribute *o2hb_region_attr =
1925                 container_of(attr, struct o2hb_region_attribute, attr);
1926         ssize_t ret = -EINVAL;
1927
1928         if (o2hb_region_attr->store)
1929                 ret = o2hb_region_attr->store(reg, page, count);
1930         return ret;
1931 }
1932
1933 static struct configfs_item_operations o2hb_region_item_ops = {
1934         .release                = o2hb_region_release,
1935         .show_attribute         = o2hb_region_show,
1936         .store_attribute        = o2hb_region_store,
1937 };
1938
1939 static struct config_item_type o2hb_region_type = {
1940         .ct_item_ops    = &o2hb_region_item_ops,
1941         .ct_attrs       = o2hb_region_attrs,
1942         .ct_owner       = THIS_MODULE,
1943 };
1944
1945 /* heartbeat set */
1946
1947 struct o2hb_heartbeat_group {
1948         struct config_group hs_group;
1949         /* some stuff? */
1950 };
1951
1952 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1953 {
1954         return group ?
1955                 container_of(group, struct o2hb_heartbeat_group, hs_group)
1956                 : NULL;
1957 }
1958
1959 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
1960 {
1961         int ret = -ENOMEM;
1962
1963         reg->hr_debug_dir =
1964                 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
1965         if (!reg->hr_debug_dir) {
1966                 mlog_errno(ret);
1967                 goto bail;
1968         }
1969
1970         reg->hr_debug_livenodes =
1971                         o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1972                                           reg->hr_debug_dir,
1973                                           &(reg->hr_db_livenodes),
1974                                           sizeof(*(reg->hr_db_livenodes)),
1975                                           O2HB_DB_TYPE_REGION_LIVENODES,
1976                                           sizeof(reg->hr_live_node_bitmap),
1977                                           O2NM_MAX_NODES, reg);
1978         if (!reg->hr_debug_livenodes) {
1979                 mlog_errno(ret);
1980                 goto bail;
1981         }
1982
1983         reg->hr_debug_regnum =
1984                         o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
1985                                           reg->hr_debug_dir,
1986                                           &(reg->hr_db_regnum),
1987                                           sizeof(*(reg->hr_db_regnum)),
1988                                           O2HB_DB_TYPE_REGION_NUMBER,
1989                                           0, O2NM_MAX_NODES, reg);
1990         if (!reg->hr_debug_regnum) {
1991                 mlog_errno(ret);
1992                 goto bail;
1993         }
1994
1995         reg->hr_debug_elapsed_time =
1996                         o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
1997                                           reg->hr_debug_dir,
1998                                           &(reg->hr_db_elapsed_time),
1999                                           sizeof(*(reg->hr_db_elapsed_time)),
2000                                           O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2001                                           0, 0, reg);
2002         if (!reg->hr_debug_elapsed_time) {
2003                 mlog_errno(ret);
2004                 goto bail;
2005         }
2006
2007         reg->hr_debug_pinned =
2008                         o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2009                                           reg->hr_debug_dir,
2010                                           &(reg->hr_db_pinned),
2011                                           sizeof(*(reg->hr_db_pinned)),
2012                                           O2HB_DB_TYPE_REGION_PINNED,
2013                                           0, 0, reg);
2014         if (!reg->hr_debug_pinned) {
2015                 mlog_errno(ret);
2016                 goto bail;
2017         }
2018
2019         ret = 0;
2020 bail:
2021         return ret;
2022 }
2023
2024 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2025                                                           const char *name)
2026 {
2027         struct o2hb_region *reg = NULL;
2028         int ret;
2029
2030         reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2031         if (reg == NULL)
2032                 return ERR_PTR(-ENOMEM);
2033
2034         if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2035                 ret = -ENAMETOOLONG;
2036                 goto free;
2037         }
2038
2039         spin_lock(&o2hb_live_lock);
2040         reg->hr_region_num = 0;
2041         if (o2hb_global_heartbeat_active()) {
2042                 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2043                                                          O2NM_MAX_REGIONS);
2044                 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2045                         spin_unlock(&o2hb_live_lock);
2046                         ret = -EFBIG;
2047                         goto free;
2048                 }
2049                 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2050         }
2051         list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2052         spin_unlock(&o2hb_live_lock);
2053
2054         config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2055
2056         ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2057         if (ret) {
2058                 config_item_put(&reg->hr_item);
2059                 goto free;
2060         }
2061
2062         return &reg->hr_item;
2063 free:
2064         kfree(reg);
2065         return ERR_PTR(ret);
2066 }
2067
2068 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2069                                            struct config_item *item)
2070 {
2071         struct task_struct *hb_task;
2072         struct o2hb_region *reg = to_o2hb_region(item);
2073         int quorum_region = 0;
2074
2075         /* stop the thread when the user removes the region dir */
2076         spin_lock(&o2hb_live_lock);
2077         if (o2hb_global_heartbeat_active()) {
2078                 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2079                 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2080                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2081                         quorum_region = 1;
2082                 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2083         }
2084         hb_task = reg->hr_task;
2085         reg->hr_task = NULL;
2086         reg->hr_item_dropped = 1;
2087         spin_unlock(&o2hb_live_lock);
2088
2089         if (hb_task)
2090                 kthread_stop(hb_task);
2091
2092         /*
2093          * If we're racing a dev_write(), we need to wake them.  They will
2094          * check reg->hr_task
2095          */
2096         if (atomic_read(&reg->hr_steady_iterations) != 0) {
2097                 atomic_set(&reg->hr_steady_iterations, 0);
2098                 wake_up(&o2hb_steady_queue);
2099         }
2100
2101         if (o2hb_global_heartbeat_active())
2102                 printk(KERN_NOTICE "o2hb: Heartbeat stopped on region %s\n",
2103                        config_item_name(&reg->hr_item));
2104
2105         config_item_put(item);
2106
2107         if (!o2hb_global_heartbeat_active() || !quorum_region)
2108                 return;
2109
2110         /*
2111          * If global heartbeat active and there are dependent users,
2112          * pin all regions if quorum region count <= CUT_OFF
2113          */
2114         spin_lock(&o2hb_live_lock);
2115
2116         if (!o2hb_dependent_users)
2117                 goto unlock;
2118
2119         if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2120                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2121                 o2hb_region_pin(NULL);
2122
2123 unlock:
2124         spin_unlock(&o2hb_live_lock);
2125 }
2126
2127 struct o2hb_heartbeat_group_attribute {
2128         struct configfs_attribute attr;
2129         ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2130         ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2131 };
2132
2133 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2134                                          struct configfs_attribute *attr,
2135                                          char *page)
2136 {
2137         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2138         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2139                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2140         ssize_t ret = 0;
2141
2142         if (o2hb_heartbeat_group_attr->show)
2143                 ret = o2hb_heartbeat_group_attr->show(reg, page);
2144         return ret;
2145 }
2146
2147 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2148                                           struct configfs_attribute *attr,
2149                                           const char *page, size_t count)
2150 {
2151         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2152         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2153                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2154         ssize_t ret = -EINVAL;
2155
2156         if (o2hb_heartbeat_group_attr->store)
2157                 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2158         return ret;
2159 }
2160
2161 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2162                                                      char *page)
2163 {
2164         return sprintf(page, "%u\n", o2hb_dead_threshold);
2165 }
2166
2167 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2168                                                     const char *page,
2169                                                     size_t count)
2170 {
2171         unsigned long tmp;
2172         char *p = (char *)page;
2173
2174         tmp = simple_strtoul(p, &p, 10);
2175         if (!p || (*p && (*p != '\n')))
2176                 return -EINVAL;
2177
2178         /* this will validate ranges for us. */
2179         o2hb_dead_threshold_set((unsigned int) tmp);
2180
2181         return count;
2182 }
2183
2184 static
2185 ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2186                                        char *page)
2187 {
2188         return sprintf(page, "%s\n",
2189                        o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2190 }
2191
2192 static
2193 ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2194                                         const char *page, size_t count)
2195 {
2196         unsigned int i;
2197         int ret;
2198         size_t len;
2199
2200         len = (page[count - 1] == '\n') ? count - 1 : count;
2201         if (!len)
2202                 return -EINVAL;
2203
2204         for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2205                 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2206                         continue;
2207
2208                 ret = o2hb_global_hearbeat_mode_set(i);
2209                 if (!ret)
2210                         printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2211                                o2hb_heartbeat_mode_desc[i]);
2212                 return count;
2213         }
2214
2215         return -EINVAL;
2216
2217 }
2218
2219 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2220         .attr   = { .ca_owner = THIS_MODULE,
2221                     .ca_name = "dead_threshold",
2222                     .ca_mode = S_IRUGO | S_IWUSR },
2223         .show   = o2hb_heartbeat_group_threshold_show,
2224         .store  = o2hb_heartbeat_group_threshold_store,
2225 };
2226
2227 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2228         .attr   = { .ca_owner = THIS_MODULE,
2229                 .ca_name = "mode",
2230                 .ca_mode = S_IRUGO | S_IWUSR },
2231         .show   = o2hb_heartbeat_group_mode_show,
2232         .store  = o2hb_heartbeat_group_mode_store,
2233 };
2234
2235 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2236         &o2hb_heartbeat_group_attr_threshold.attr,
2237         &o2hb_heartbeat_group_attr_mode.attr,
2238         NULL,
2239 };
2240
2241 static struct configfs_item_operations o2hb_hearbeat_group_item_ops = {
2242         .show_attribute         = o2hb_heartbeat_group_show,
2243         .store_attribute        = o2hb_heartbeat_group_store,
2244 };
2245
2246 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2247         .make_item      = o2hb_heartbeat_group_make_item,
2248         .drop_item      = o2hb_heartbeat_group_drop_item,
2249 };
2250
2251 static struct config_item_type o2hb_heartbeat_group_type = {
2252         .ct_group_ops   = &o2hb_heartbeat_group_group_ops,
2253         .ct_item_ops    = &o2hb_hearbeat_group_item_ops,
2254         .ct_attrs       = o2hb_heartbeat_group_attrs,
2255         .ct_owner       = THIS_MODULE,
2256 };
2257
2258 /* this is just here to avoid touching group in heartbeat.h which the
2259  * entire damn world #includes */
2260 struct config_group *o2hb_alloc_hb_set(void)
2261 {
2262         struct o2hb_heartbeat_group *hs = NULL;
2263         struct config_group *ret = NULL;
2264
2265         hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2266         if (hs == NULL)
2267                 goto out;
2268
2269         config_group_init_type_name(&hs->hs_group, "heartbeat",
2270                                     &o2hb_heartbeat_group_type);
2271
2272         ret = &hs->hs_group;
2273 out:
2274         if (ret == NULL)
2275                 kfree(hs);
2276         return ret;
2277 }
2278
2279 void o2hb_free_hb_set(struct config_group *group)
2280 {
2281         struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2282         kfree(hs);
2283 }
2284
2285 /* hb callback registration and issueing */
2286
2287 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2288 {
2289         if (type == O2HB_NUM_CB)
2290                 return ERR_PTR(-EINVAL);
2291
2292         return &o2hb_callbacks[type];
2293 }
2294
2295 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2296                          enum o2hb_callback_type type,
2297                          o2hb_cb_func *func,
2298                          void *data,
2299                          int priority)
2300 {
2301         INIT_LIST_HEAD(&hc->hc_item);
2302         hc->hc_func = func;
2303         hc->hc_data = data;
2304         hc->hc_priority = priority;
2305         hc->hc_type = type;
2306         hc->hc_magic = O2HB_CB_MAGIC;
2307 }
2308 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2309
2310 /*
2311  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2312  * In global heartbeat mode, region_uuid passed is NULL.
2313  *
2314  * In local, we only pin the matching region. In global we pin all the active
2315  * regions.
2316  */
2317 static int o2hb_region_pin(const char *region_uuid)
2318 {
2319         int ret = 0, found = 0;
2320         struct o2hb_region *reg;
2321         char *uuid;
2322
2323         assert_spin_locked(&o2hb_live_lock);
2324
2325         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2326                 uuid = config_item_name(&reg->hr_item);
2327
2328                 /* local heartbeat */
2329                 if (region_uuid) {
2330                         if (strcmp(region_uuid, uuid))
2331                                 continue;
2332                         found = 1;
2333                 }
2334
2335                 if (reg->hr_item_pinned || reg->hr_item_dropped)
2336                         goto skip_pin;
2337
2338                 /* Ignore ENOENT only for local hb (userdlm domain) */
2339                 ret = o2nm_depend_item(&reg->hr_item);
2340                 if (!ret) {
2341                         mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2342                         reg->hr_item_pinned = 1;
2343                 } else {
2344                         if (ret == -ENOENT && found)
2345                                 ret = 0;
2346                         else {
2347                                 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2348                                      uuid, ret);
2349                                 break;
2350                         }
2351                 }
2352 skip_pin:
2353                 if (found)
2354                         break;
2355         }
2356
2357         return ret;
2358 }
2359
2360 /*
2361  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2362  * In global heartbeat mode, region_uuid passed is NULL.
2363  *
2364  * In local, we only unpin the matching region. In global we unpin all the
2365  * active regions.
2366  */
2367 static void o2hb_region_unpin(const char *region_uuid)
2368 {
2369         struct o2hb_region *reg;
2370         char *uuid;
2371         int found = 0;
2372
2373         assert_spin_locked(&o2hb_live_lock);
2374
2375         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2376                 uuid = config_item_name(&reg->hr_item);
2377                 if (region_uuid) {
2378                         if (strcmp(region_uuid, uuid))
2379                                 continue;
2380                         found = 1;
2381                 }
2382
2383                 if (reg->hr_item_pinned) {
2384                         mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2385                         o2nm_undepend_item(&reg->hr_item);
2386                         reg->hr_item_pinned = 0;
2387                 }
2388                 if (found)
2389                         break;
2390         }
2391 }
2392
2393 static int o2hb_region_inc_user(const char *region_uuid)
2394 {
2395         int ret = 0;
2396
2397         spin_lock(&o2hb_live_lock);
2398
2399         /* local heartbeat */
2400         if (!o2hb_global_heartbeat_active()) {
2401             ret = o2hb_region_pin(region_uuid);
2402             goto unlock;
2403         }
2404
2405         /*
2406          * if global heartbeat active and this is the first dependent user,
2407          * pin all regions if quorum region count <= CUT_OFF
2408          */
2409         o2hb_dependent_users++;
2410         if (o2hb_dependent_users > 1)
2411                 goto unlock;
2412
2413         if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2414                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2415                 ret = o2hb_region_pin(NULL);
2416
2417 unlock:
2418         spin_unlock(&o2hb_live_lock);
2419         return ret;
2420 }
2421
2422 void o2hb_region_dec_user(const char *region_uuid)
2423 {
2424         spin_lock(&o2hb_live_lock);
2425
2426         /* local heartbeat */
2427         if (!o2hb_global_heartbeat_active()) {
2428             o2hb_region_unpin(region_uuid);
2429             goto unlock;
2430         }
2431
2432         /*
2433          * if global heartbeat active and there are no dependent users,
2434          * unpin all quorum regions
2435          */
2436         o2hb_dependent_users--;
2437         if (!o2hb_dependent_users)
2438                 o2hb_region_unpin(NULL);
2439
2440 unlock:
2441         spin_unlock(&o2hb_live_lock);
2442 }
2443
2444 int o2hb_register_callback(const char *region_uuid,
2445                            struct o2hb_callback_func *hc)
2446 {
2447         struct o2hb_callback_func *tmp;
2448         struct list_head *iter;
2449         struct o2hb_callback *hbcall;
2450         int ret;
2451
2452         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2453         BUG_ON(!list_empty(&hc->hc_item));
2454
2455         hbcall = hbcall_from_type(hc->hc_type);
2456         if (IS_ERR(hbcall)) {
2457                 ret = PTR_ERR(hbcall);
2458                 goto out;
2459         }
2460
2461         if (region_uuid) {
2462                 ret = o2hb_region_inc_user(region_uuid);
2463                 if (ret) {
2464                         mlog_errno(ret);
2465                         goto out;
2466                 }
2467         }
2468
2469         down_write(&o2hb_callback_sem);
2470
2471         list_for_each(iter, &hbcall->list) {
2472                 tmp = list_entry(iter, struct o2hb_callback_func, hc_item);
2473                 if (hc->hc_priority < tmp->hc_priority) {
2474                         list_add_tail(&hc->hc_item, iter);
2475                         break;
2476                 }
2477         }
2478         if (list_empty(&hc->hc_item))
2479                 list_add_tail(&hc->hc_item, &hbcall->list);
2480
2481         up_write(&o2hb_callback_sem);
2482         ret = 0;
2483 out:
2484         mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2485              ret, __builtin_return_address(0), hc);
2486         return ret;
2487 }
2488 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2489
2490 void o2hb_unregister_callback(const char *region_uuid,
2491                               struct o2hb_callback_func *hc)
2492 {
2493         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2494
2495         mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2496              __builtin_return_address(0), hc);
2497
2498         /* XXX Can this happen _with_ a region reference? */
2499         if (list_empty(&hc->hc_item))
2500                 return;
2501
2502         if (region_uuid)
2503                 o2hb_region_dec_user(region_uuid);
2504
2505         down_write(&o2hb_callback_sem);
2506
2507         list_del_init(&hc->hc_item);
2508
2509         up_write(&o2hb_callback_sem);
2510 }
2511 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2512
2513 int o2hb_check_node_heartbeating(u8 node_num)
2514 {
2515         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2516
2517         o2hb_fill_node_map(testing_map, sizeof(testing_map));
2518         if (!test_bit(node_num, testing_map)) {
2519                 mlog(ML_HEARTBEAT,
2520                      "node (%u) does not have heartbeating enabled.\n",
2521                      node_num);
2522                 return 0;
2523         }
2524
2525         return 1;
2526 }
2527 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2528
2529 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2530 {
2531         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2532
2533         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2534         if (!test_bit(node_num, testing_map)) {
2535                 mlog(ML_HEARTBEAT,
2536                      "node (%u) does not have heartbeating enabled.\n",
2537                      node_num);
2538                 return 0;
2539         }
2540
2541         return 1;
2542 }
2543 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2544
2545 /* Makes sure our local node is configured with a node number, and is
2546  * heartbeating. */
2547 int o2hb_check_local_node_heartbeating(void)
2548 {
2549         u8 node_num;
2550
2551         /* if this node was set then we have networking */
2552         node_num = o2nm_this_node();
2553         if (node_num == O2NM_MAX_NODES) {
2554                 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2555                 return 0;
2556         }
2557
2558         return o2hb_check_node_heartbeating(node_num);
2559 }
2560 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2561
2562 /*
2563  * this is just a hack until we get the plumbing which flips file systems
2564  * read only and drops the hb ref instead of killing the node dead.
2565  */
2566 void o2hb_stop_all_regions(void)
2567 {
2568         struct o2hb_region *reg;
2569
2570         mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2571
2572         spin_lock(&o2hb_live_lock);
2573
2574         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2575                 reg->hr_unclean_stop = 1;
2576
2577         spin_unlock(&o2hb_live_lock);
2578 }
2579 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2580
2581 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2582 {
2583         struct o2hb_region *reg;
2584         int numregs = 0;
2585         char *p;
2586
2587         spin_lock(&o2hb_live_lock);
2588
2589         p = region_uuids;
2590         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2591                 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2592                 if (numregs < max_regions) {
2593                         memcpy(p, config_item_name(&reg->hr_item),
2594                                O2HB_MAX_REGION_NAME_LEN);
2595                         p += O2HB_MAX_REGION_NAME_LEN;
2596                 }
2597                 numregs++;
2598         }
2599
2600         spin_unlock(&o2hb_live_lock);
2601
2602         return numregs;
2603 }
2604 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2605
2606 int o2hb_global_heartbeat_active(void)
2607 {
2608         return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2609 }
2610 EXPORT_SYMBOL(o2hb_global_heartbeat_active);