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 #define O2HB_DB_TYPE_LIVENODES          0
  66 struct o2hb_debug_buf {
  67         int db_type;
  68         int db_size;
  69         int db_len;
  70         void *db_data;
  71 };
  72
  73 static struct o2hb_debug_buf *o2hb_db_livenodes;
  74
  75 #define O2HB_DEBUG_DIR                  "o2hb"
  76 #define O2HB_DEBUG_LIVENODES            "livenodes"
  77
  78 static struct dentry *o2hb_debug_dir;
  79 static struct dentry *o2hb_debug_livenodes;
  80
  81 static LIST_HEAD(o2hb_all_regions);
  82
  83 static struct o2hb_callback {
  84         struct list_head list;
  85 } o2hb_callbacks[O2HB_NUM_CB];
  86
  87 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
  88
  89 #define O2HB_DEFAULT_BLOCK_BITS       9
  90
  91 enum o2hb_heartbeat_modes {
  92         O2HB_HEARTBEAT_LOCAL            = 0,
  93         O2HB_HEARTBEAT_GLOBAL,
  94         O2HB_HEARTBEAT_NUM_MODES,
  95 };
  96
  97 char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
  98                 "local",        /* O2HB_HEARTBEAT_LOCAL */
  99                 "global",       /* O2HB_HEARTBEAT_GLOBAL */
 100 };
 101
 102 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
 103 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
 104
 105 /* Only sets a new threshold if there are no active regions.
 106  *
 107  * No locking or otherwise interesting code is required for reading
 108  * o2hb_dead_threshold as it can't change once regions are active and
 109  * it's not interesting to anyone until then anyway. */
 110 static void o2hb_dead_threshold_set(unsigned int threshold)
 111 {
 112         if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
 113                 spin_lock(&o2hb_live_lock);
 114                 if (list_empty(&o2hb_all_regions))
 115                         o2hb_dead_threshold = threshold;
 116                 spin_unlock(&o2hb_live_lock);
 117         }
 118 }
 119
 120 static int o2hb_global_hearbeat_mode_set(unsigned int hb_mode)
 121 {
 122         int ret = -1;
 123
 124         if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
 125                 spin_lock(&o2hb_live_lock);
 126                 if (list_empty(&o2hb_all_regions)) {
 127                         o2hb_heartbeat_mode = hb_mode;
 128                         ret = 0;
 129                 }
 130                 spin_unlock(&o2hb_live_lock);
 131         }
 132
 133         return ret;
 134 }
 135
 136 struct o2hb_node_event {
 137         struct list_head        hn_item;
 138         enum o2hb_callback_type hn_event_type;
 139         struct o2nm_node        *hn_node;
 140         int                     hn_node_num;
 141 };
 142
 143 struct o2hb_disk_slot {
 144         struct o2hb_disk_heartbeat_block *ds_raw_block;
 145         u8                      ds_node_num;
 146         u64                     ds_last_time;
 147         u64                     ds_last_generation;
 148         u16                     ds_equal_samples;
 149         u16                     ds_changed_samples;
 150         struct list_head        ds_live_item;
 151 };
 152
 153 /* each thread owns a region.. when we're asked to tear down the region
 154  * we ask the thread to stop, who cleans up the region */
 155 struct o2hb_region {
 156         struct config_item      hr_item;
 157
 158         struct list_head        hr_all_item;
 159         unsigned                hr_unclean_stop:1;
 160
 161         /* protected by the hr_callback_sem */
 162         struct task_struct      *hr_task;
 163
 164         unsigned int            hr_blocks;
 165         unsigned long long      hr_start_block;
 166
 167         unsigned int            hr_block_bits;
 168         unsigned int            hr_block_bytes;
 169
 170         unsigned int            hr_slots_per_page;
 171         unsigned int            hr_num_pages;
 172
 173         struct page             **hr_slot_data;
 174         struct block_device     *hr_bdev;
 175         struct o2hb_disk_slot   *hr_slots;
 176
 177         /* live node map of this region */
 178         unsigned long           hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 179
 180         /* let the person setting up hb wait for it to return until it
 181          * has reached a 'steady' state.  This will be fixed when we have
 182          * a more complete api that doesn't lead to this sort of fragility. */
 183         atomic_t                hr_steady_iterations;
 184
 185         char                    hr_dev_name[BDEVNAME_SIZE];
 186
 187         unsigned int            hr_timeout_ms;
 188
 189         /* randomized as the region goes up and down so that a node
 190          * recognizes a node going up and down in one iteration */
 191         u64                     hr_generation;
 192
 193         struct delayed_work     hr_write_timeout_work;
 194         unsigned long           hr_last_timeout_start;
 195
 196         /* Used during o2hb_check_slot to hold a copy of the block
 197          * being checked because we temporarily have to zero out the
 198          * crc field. */
 199         struct o2hb_disk_heartbeat_block *hr_tmp_block;
 200 };
 201
 202 struct o2hb_bio_wait_ctxt {
 203         atomic_t          wc_num_reqs;
 204         struct completion wc_io_complete;
 205         int               wc_error;
 206 };
 207
 208 static void o2hb_write_timeout(struct work_struct *work)
 209 {
 210         struct o2hb_region *reg =
 211                 container_of(work, struct o2hb_region,
 212                              hr_write_timeout_work.work);
 213
 214         mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
 215              "milliseconds\n", reg->hr_dev_name,
 216              jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
 217         o2quo_disk_timeout();
 218 }
 219
 220 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
 221 {
 222         mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
 223              O2HB_MAX_WRITE_TIMEOUT_MS);
 224
 225         cancel_delayed_work(&reg->hr_write_timeout_work);
 226         reg->hr_last_timeout_start = jiffies;
 227         schedule_delayed_work(&reg->hr_write_timeout_work,
 228                               msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
 229 }
 230
 231 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
 232 {
 233         cancel_delayed_work(&reg->hr_write_timeout_work);
 234         flush_scheduled_work();
 235 }
 236
 237 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
 238 {
 239         atomic_set(&wc->wc_num_reqs, 1);
 240         init_completion(&wc->wc_io_complete);
 241         wc->wc_error = 0;
 242 }
 243
 244 /* Used in error paths too */
 245 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
 246                                      unsigned int num)
 247 {
 248         /* sadly atomic_sub_and_test() isn't available on all platforms.  The
 249          * good news is that the fast path only completes one at a time */
 250         while(num--) {
 251                 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
 252                         BUG_ON(num > 0);
 253                         complete(&wc->wc_io_complete);
 254                 }
 255         }
 256 }
 257
 258 static void o2hb_wait_on_io(struct o2hb_region *reg,
 259                             struct o2hb_bio_wait_ctxt *wc)
 260 {
 261         struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping;
 262
 263         blk_run_address_space(mapping);
 264         o2hb_bio_wait_dec(wc, 1);
 265
 266         wait_for_completion(&wc->wc_io_complete);
 267 }
 268
 269 static void o2hb_bio_end_io(struct bio *bio,
 270                            int error)
 271 {
 272         struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
 273
 274         if (error) {
 275                 mlog(ML_ERROR, "IO Error %d\n", error);
 276                 wc->wc_error = error;
 277         }
 278
 279         o2hb_bio_wait_dec(wc, 1);
 280         bio_put(bio);
 281 }
 282
 283 /* Setup a Bio to cover I/O against num_slots slots starting at
 284  * start_slot. */
 285 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
 286                                       struct o2hb_bio_wait_ctxt *wc,
 287                                       unsigned int *current_slot,
 288                                       unsigned int max_slots)
 289 {
 290         int len, current_page;
 291         unsigned int vec_len, vec_start;
 292         unsigned int bits = reg->hr_block_bits;
 293         unsigned int spp = reg->hr_slots_per_page;
 294         unsigned int cs = *current_slot;
 295         struct bio *bio;
 296         struct page *page;
 297
 298         /* Testing has shown this allocation to take long enough under
 299          * GFP_KERNEL that the local node can get fenced. It would be
 300          * nicest if we could pre-allocate these bios and avoid this
 301          * all together. */
 302         bio = bio_alloc(GFP_ATOMIC, 16);
 303         if (!bio) {
 304                 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
 305                 bio = ERR_PTR(-ENOMEM);
 306                 goto bail;
 307         }
 308
 309         /* Must put everything in 512 byte sectors for the bio... */
 310         bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
 311         bio->bi_bdev = reg->hr_bdev;
 312         bio->bi_private = wc;
 313         bio->bi_end_io = o2hb_bio_end_io;
 314
 315         vec_start = (cs << bits) % PAGE_CACHE_SIZE;
 316         while(cs < max_slots) {
 317                 current_page = cs / spp;
 318                 page = reg->hr_slot_data[current_page];
 319
 320                 vec_len = min(PAGE_CACHE_SIZE - vec_start,
 321                               (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
 322
 323                 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
 324                      current_page, vec_len, vec_start);
 325
 326                 len = bio_add_page(bio, page, vec_len, vec_start);
 327                 if (len != vec_len) break;
 328
 329                 cs += vec_len / (PAGE_CACHE_SIZE/spp);
 330                 vec_start = 0;
 331         }
 332
 333 bail:
 334         *current_slot = cs;
 335         return bio;
 336 }
 337
 338 static int o2hb_read_slots(struct o2hb_region *reg,
 339                            unsigned int max_slots)
 340 {
 341         unsigned int current_slot=0;
 342         int status;
 343         struct o2hb_bio_wait_ctxt wc;
 344         struct bio *bio;
 345
 346         o2hb_bio_wait_init(&wc);
 347
 348         while(current_slot < max_slots) {
 349                 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
 350                 if (IS_ERR(bio)) {
 351                         status = PTR_ERR(bio);
 352                         mlog_errno(status);
 353                         goto bail_and_wait;
 354                 }
 355
 356                 atomic_inc(&wc.wc_num_reqs);
 357                 submit_bio(READ, bio);
 358         }
 359
 360         status = 0;
 361
 362 bail_and_wait:
 363         o2hb_wait_on_io(reg, &wc);
 364         if (wc.wc_error && !status)
 365                 status = wc.wc_error;
 366
 367         return status;
 368 }
 369
 370 static int o2hb_issue_node_write(struct o2hb_region *reg,
 371                                  struct o2hb_bio_wait_ctxt *write_wc)
 372 {
 373         int status;
 374         unsigned int slot;
 375         struct bio *bio;
 376
 377         o2hb_bio_wait_init(write_wc);
 378
 379         slot = o2nm_this_node();
 380
 381         bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
 382         if (IS_ERR(bio)) {
 383                 status = PTR_ERR(bio);
 384                 mlog_errno(status);
 385                 goto bail;
 386         }
 387
 388         atomic_inc(&write_wc->wc_num_reqs);
 389         submit_bio(WRITE, bio);
 390
 391         status = 0;
 392 bail:
 393         return status;
 394 }
 395
 396 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
 397                                      struct o2hb_disk_heartbeat_block *hb_block)
 398 {
 399         __le32 old_cksum;
 400         u32 ret;
 401
 402         /* We want to compute the block crc with a 0 value in the
 403          * hb_cksum field. Save it off here and replace after the
 404          * crc. */
 405         old_cksum = hb_block->hb_cksum;
 406         hb_block->hb_cksum = 0;
 407
 408         ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
 409
 410         hb_block->hb_cksum = old_cksum;
 411
 412         return ret;
 413 }
 414
 415 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
 416 {
 417         mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
 418              "cksum = 0x%x, generation 0x%llx\n",
 419              (long long)le64_to_cpu(hb_block->hb_seq),
 420              hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
 421              (long long)le64_to_cpu(hb_block->hb_generation));
 422 }
 423
 424 static int o2hb_verify_crc(struct o2hb_region *reg,
 425                            struct o2hb_disk_heartbeat_block *hb_block)
 426 {
 427         u32 read, computed;
 428
 429         read = le32_to_cpu(hb_block->hb_cksum);
 430         computed = o2hb_compute_block_crc_le(reg, hb_block);
 431
 432         return read == computed;
 433 }
 434
 435 /* We want to make sure that nobody is heartbeating on top of us --
 436  * this will help detect an invalid configuration. */
 437 static int o2hb_check_last_timestamp(struct o2hb_region *reg)
 438 {
 439         int node_num, ret;
 440         struct o2hb_disk_slot *slot;
 441         struct o2hb_disk_heartbeat_block *hb_block;
 442
 443         node_num = o2nm_this_node();
 444
 445         ret = 1;
 446         slot = &reg->hr_slots[node_num];
 447         /* Don't check on our 1st timestamp */
 448         if (slot->ds_last_time) {
 449                 hb_block = slot->ds_raw_block;
 450
 451                 if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time)
 452                         ret = 0;
 453         }
 454
 455         return ret;
 456 }
 457
 458 static inline void o2hb_prepare_block(struct o2hb_region *reg,
 459                                       u64 generation)
 460 {
 461         int node_num;
 462         u64 cputime;
 463         struct o2hb_disk_slot *slot;
 464         struct o2hb_disk_heartbeat_block *hb_block;
 465
 466         node_num = o2nm_this_node();
 467         slot = &reg->hr_slots[node_num];
 468
 469         hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
 470         memset(hb_block, 0, reg->hr_block_bytes);
 471         /* TODO: time stuff */
 472         cputime = CURRENT_TIME.tv_sec;
 473         if (!cputime)
 474                 cputime = 1;
 475
 476         hb_block->hb_seq = cpu_to_le64(cputime);
 477         hb_block->hb_node = node_num;
 478         hb_block->hb_generation = cpu_to_le64(generation);
 479         hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
 480
 481         /* This step must always happen last! */
 482         hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
 483                                                                    hb_block));
 484
 485         mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
 486              (long long)generation,
 487              le32_to_cpu(hb_block->hb_cksum));
 488 }
 489
 490 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
 491                                 struct o2nm_node *node,
 492                                 int idx)
 493 {
 494         struct list_head *iter;
 495         struct o2hb_callback_func *f;
 496
 497         list_for_each(iter, &hbcall->list) {
 498                 f = list_entry(iter, struct o2hb_callback_func, hc_item);
 499                 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
 500                 (f->hc_func)(node, idx, f->hc_data);
 501         }
 502 }
 503
 504 /* Will run the list in order until we process the passed event */
 505 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
 506 {
 507         int empty;
 508         struct o2hb_callback *hbcall;
 509         struct o2hb_node_event *event;
 510
 511         spin_lock(&o2hb_live_lock);
 512         empty = list_empty(&queued_event->hn_item);
 513         spin_unlock(&o2hb_live_lock);
 514         if (empty)
 515                 return;
 516
 517         /* Holding callback sem assures we don't alter the callback
 518          * lists when doing this, and serializes ourselves with other
 519          * processes wanting callbacks. */
 520         down_write(&o2hb_callback_sem);
 521
 522         spin_lock(&o2hb_live_lock);
 523         while (!list_empty(&o2hb_node_events)
 524                && !list_empty(&queued_event->hn_item)) {
 525                 event = list_entry(o2hb_node_events.next,
 526                                    struct o2hb_node_event,
 527                                    hn_item);
 528                 list_del_init(&event->hn_item);
 529                 spin_unlock(&o2hb_live_lock);
 530
 531                 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
 532                      event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
 533                      event->hn_node_num);
 534
 535                 hbcall = hbcall_from_type(event->hn_event_type);
 536
 537                 /* We should *never* have gotten on to the list with a
 538                  * bad type... This isn't something that we should try
 539                  * to recover from. */
 540                 BUG_ON(IS_ERR(hbcall));
 541
 542                 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
 543
 544                 spin_lock(&o2hb_live_lock);
 545         }
 546         spin_unlock(&o2hb_live_lock);
 547
 548         up_write(&o2hb_callback_sem);
 549 }
 550
 551 static void o2hb_queue_node_event(struct o2hb_node_event *event,
 552                                   enum o2hb_callback_type type,
 553                                   struct o2nm_node *node,
 554                                   int node_num)
 555 {
 556         assert_spin_locked(&o2hb_live_lock);
 557
 558         BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
 559
 560         event->hn_event_type = type;
 561         event->hn_node = node;
 562         event->hn_node_num = node_num;
 563
 564         mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
 565              type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
 566
 567         list_add_tail(&event->hn_item, &o2hb_node_events);
 568 }
 569
 570 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
 571 {
 572         struct o2hb_node_event event =
 573                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 574         struct o2nm_node *node;
 575
 576         node = o2nm_get_node_by_num(slot->ds_node_num);
 577         if (!node)
 578                 return;
 579
 580         spin_lock(&o2hb_live_lock);
 581         if (!list_empty(&slot->ds_live_item)) {
 582                 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
 583                      slot->ds_node_num);
 584
 585                 list_del_init(&slot->ds_live_item);
 586
 587                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 588                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 589
 590                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
 591                                               slot->ds_node_num);
 592                 }
 593         }
 594         spin_unlock(&o2hb_live_lock);
 595
 596         o2hb_run_event_list(&event);
 597
 598         o2nm_node_put(node);
 599 }
 600
 601 static int o2hb_check_slot(struct o2hb_region *reg,
 602                            struct o2hb_disk_slot *slot)
 603 {
 604         int changed = 0, gen_changed = 0;
 605         struct o2hb_node_event event =
 606                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 607         struct o2nm_node *node;
 608         struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
 609         u64 cputime;
 610         unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
 611         unsigned int slot_dead_ms;
 612         int tmp;
 613
 614         memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
 615
 616         /*
 617          * If a node is no longer configured but is still in the livemap, we
 618          * may need to clear that bit from the livemap.
 619          */
 620         node = o2nm_get_node_by_num(slot->ds_node_num);
 621         if (!node) {
 622                 spin_lock(&o2hb_live_lock);
 623                 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 624                 spin_unlock(&o2hb_live_lock);
 625                 if (!tmp)
 626                         return 0;
 627         }
 628
 629         if (!o2hb_verify_crc(reg, hb_block)) {
 630                 /* all paths from here will drop o2hb_live_lock for
 631                  * us. */
 632                 spin_lock(&o2hb_live_lock);
 633
 634                 /* Don't print an error on the console in this case -
 635                  * a freshly formatted heartbeat area will not have a
 636                  * crc set on it. */
 637                 if (list_empty(&slot->ds_live_item))
 638                         goto out;
 639
 640                 /* The node is live but pushed out a bad crc. We
 641                  * consider it a transient miss but don't populate any
 642                  * other values as they may be junk. */
 643                 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
 644                      slot->ds_node_num, reg->hr_dev_name);
 645                 o2hb_dump_slot(hb_block);
 646
 647                 slot->ds_equal_samples++;
 648                 goto fire_callbacks;
 649         }
 650
 651         /* we don't care if these wrap.. the state transitions below
 652          * clear at the right places */
 653         cputime = le64_to_cpu(hb_block->hb_seq);
 654         if (slot->ds_last_time != cputime)
 655                 slot->ds_changed_samples++;
 656         else
 657                 slot->ds_equal_samples++;
 658         slot->ds_last_time = cputime;
 659
 660         /* The node changed heartbeat generations. We assume this to
 661          * mean it dropped off but came back before we timed out. We
 662          * want to consider it down for the time being but don't want
 663          * to lose any changed_samples state we might build up to
 664          * considering it live again. */
 665         if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
 666                 gen_changed = 1;
 667                 slot->ds_equal_samples = 0;
 668                 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
 669                      "to 0x%llx)\n", slot->ds_node_num,
 670                      (long long)slot->ds_last_generation,
 671                      (long long)le64_to_cpu(hb_block->hb_generation));
 672         }
 673
 674         slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
 675
 676         mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
 677              "seq %llu last %llu changed %u equal %u\n",
 678              slot->ds_node_num, (long long)slot->ds_last_generation,
 679              le32_to_cpu(hb_block->hb_cksum),
 680              (unsigned long long)le64_to_cpu(hb_block->hb_seq),
 681              (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
 682              slot->ds_equal_samples);
 683
 684         spin_lock(&o2hb_live_lock);
 685
 686 fire_callbacks:
 687         /* dead nodes only come to life after some number of
 688          * changes at any time during their dead time */
 689         if (list_empty(&slot->ds_live_item) &&
 690             slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
 691                 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
 692                      slot->ds_node_num, (long long)slot->ds_last_generation);
 693
 694                 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 695
 696                 /* first on the list generates a callback */
 697                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 698                         set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 699
 700                         o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
 701                                               slot->ds_node_num);
 702
 703                         changed = 1;
 704                 }
 705
 706                 list_add_tail(&slot->ds_live_item,
 707                               &o2hb_live_slots[slot->ds_node_num]);
 708
 709                 slot->ds_equal_samples = 0;
 710
 711                 /* We want to be sure that all nodes agree on the
 712                  * number of milliseconds before a node will be
 713                  * considered dead. The self-fencing timeout is
 714                  * computed from this value, and a discrepancy might
 715                  * result in heartbeat calling a node dead when it
 716                  * hasn't self-fenced yet. */
 717                 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
 718                 if (slot_dead_ms && slot_dead_ms != dead_ms) {
 719                         /* TODO: Perhaps we can fail the region here. */
 720                         mlog(ML_ERROR, "Node %d on device %s has a dead count "
 721                              "of %u ms, but our count is %u ms.\n"
 722                              "Please double check your configuration values "
 723                              "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
 724                              slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
 725                              dead_ms);
 726                 }
 727                 goto out;
 728         }
 729
 730         /* if the list is dead, we're done.. */
 731         if (list_empty(&slot->ds_live_item))
 732                 goto out;
 733
 734         /* live nodes only go dead after enough consequtive missed
 735          * samples..  reset the missed counter whenever we see
 736          * activity */
 737         if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
 738                 mlog(ML_HEARTBEAT, "Node %d left my region\n",
 739                      slot->ds_node_num);
 740
 741                 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 742
 743                 /* last off the live_slot generates a callback */
 744                 list_del_init(&slot->ds_live_item);
 745                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 746                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 747
 748                         /* node can be null */
 749                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
 750                                               node, slot->ds_node_num);
 751
 752                         changed = 1;
 753                 }
 754
 755                 /* We don't clear this because the node is still
 756                  * actually writing new blocks. */
 757                 if (!gen_changed)
 758                         slot->ds_changed_samples = 0;
 759                 goto out;
 760         }
 761         if (slot->ds_changed_samples) {
 762                 slot->ds_changed_samples = 0;
 763                 slot->ds_equal_samples = 0;
 764         }
 765 out:
 766         spin_unlock(&o2hb_live_lock);
 767
 768         o2hb_run_event_list(&event);
 769
 770         if (node)
 771                 o2nm_node_put(node);
 772         return changed;
 773 }
 774
 775 /* This could be faster if we just implmented a find_last_bit, but I
 776  * don't think the circumstances warrant it. */
 777 static int o2hb_highest_node(unsigned long *nodes,
 778                              int numbits)
 779 {
 780         int highest, node;
 781
 782         highest = numbits;
 783         node = -1;
 784         while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
 785                 if (node >= numbits)
 786                         break;
 787
 788                 highest = node;
 789         }
 790
 791         return highest;
 792 }
 793
 794 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
 795 {
 796         int i, ret, highest_node, change = 0;
 797         unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
 798         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 799         struct o2hb_bio_wait_ctxt write_wc;
 800
 801         ret = o2nm_configured_node_map(configured_nodes,
 802                                        sizeof(configured_nodes));
 803         if (ret) {
 804                 mlog_errno(ret);
 805                 return ret;
 806         }
 807
 808         /*
 809          * If a node is not configured but is in the livemap, we still need
 810          * to read the slot so as to be able to remove it from the livemap.
 811          */
 812         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
 813         i = -1;
 814         while ((i = find_next_bit(live_node_bitmap,
 815                                   O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
 816                 set_bit(i, configured_nodes);
 817         }
 818
 819         highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
 820         if (highest_node >= O2NM_MAX_NODES) {
 821                 mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n");
 822                 return -EINVAL;
 823         }
 824
 825         /* No sense in reading the slots of nodes that don't exist
 826          * yet. Of course, if the node definitions have holes in them
 827          * then we're reading an empty slot anyway... Consider this
 828          * best-effort. */
 829         ret = o2hb_read_slots(reg, highest_node + 1);
 830         if (ret < 0) {
 831                 mlog_errno(ret);
 832                 return ret;
 833         }
 834
 835         /* With an up to date view of the slots, we can check that no
 836          * other node has been improperly configured to heartbeat in
 837          * our slot. */
 838         if (!o2hb_check_last_timestamp(reg))
 839                 mlog(ML_ERROR, "Device \"%s\": another node is heartbeating "
 840                      "in our slot!\n", reg->hr_dev_name);
 841
 842         /* fill in the proper info for our next heartbeat */
 843         o2hb_prepare_block(reg, reg->hr_generation);
 844
 845         /* And fire off the write. Note that we don't wait on this I/O
 846          * until later. */
 847         ret = o2hb_issue_node_write(reg, &write_wc);
 848         if (ret < 0) {
 849                 mlog_errno(ret);
 850                 return ret;
 851         }
 852
 853         i = -1;
 854         while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
 855
 856                 change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
 857         }
 858
 859         /*
 860          * We have to be sure we've advertised ourselves on disk
 861          * before we can go to steady state.  This ensures that
 862          * people we find in our steady state have seen us.
 863          */
 864         o2hb_wait_on_io(reg, &write_wc);
 865         if (write_wc.wc_error) {
 866                 /* Do not re-arm the write timeout on I/O error - we
 867                  * can't be sure that the new block ever made it to
 868                  * disk */
 869                 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
 870                      write_wc.wc_error, reg->hr_dev_name);
 871                 return write_wc.wc_error;
 872         }
 873
 874         o2hb_arm_write_timeout(reg);
 875
 876         /* let the person who launched us know when things are steady */
 877         if (!change && (atomic_read(&reg->hr_steady_iterations) != 0)) {
 878                 if (atomic_dec_and_test(&reg->hr_steady_iterations))
 879                         wake_up(&o2hb_steady_queue);
 880         }
 881
 882         return 0;
 883 }
 884
 885 /* Subtract b from a, storing the result in a. a *must* have a larger
 886  * value than b. */
 887 static void o2hb_tv_subtract(struct timeval *a,
 888                              struct timeval *b)
 889 {
 890         /* just return 0 when a is after b */
 891         if (a->tv_sec < b->tv_sec ||
 892             (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
 893                 a->tv_sec = 0;
 894                 a->tv_usec = 0;
 895                 return;
 896         }
 897
 898         a->tv_sec -= b->tv_sec;
 899         a->tv_usec -= b->tv_usec;
 900         while ( a->tv_usec < 0 ) {
 901                 a->tv_sec--;
 902                 a->tv_usec += 1000000;
 903         }
 904 }
 905
 906 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
 907                                        struct timeval *end)
 908 {
 909         struct timeval res = *end;
 910
 911         o2hb_tv_subtract(&res, start);
 912
 913         return res.tv_sec * 1000 + res.tv_usec / 1000;
 914 }
 915
 916 /*
 917  * we ride the region ref that the region dir holds.  before the region
 918  * dir is removed and drops it ref it will wait to tear down this
 919  * thread.
 920  */
 921 static int o2hb_thread(void *data)
 922 {
 923         int i, ret;
 924         struct o2hb_region *reg = data;
 925         struct o2hb_bio_wait_ctxt write_wc;
 926         struct timeval before_hb, after_hb;
 927         unsigned int elapsed_msec;
 928
 929         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
 930
 931         set_user_nice(current, -20);
 932
 933         while (!kthread_should_stop() && !reg->hr_unclean_stop) {
 934                 /* We track the time spent inside
 935                  * o2hb_do_disk_heartbeat so that we avoid more than
 936                  * hr_timeout_ms between disk writes. On busy systems
 937                  * this should result in a heartbeat which is less
 938                  * likely to time itself out. */
 939                 do_gettimeofday(&before_hb);
 940
 941                 i = 0;
 942                 do {
 943                         ret = o2hb_do_disk_heartbeat(reg);
 944                 } while (ret && ++i < 2);
 945
 946                 do_gettimeofday(&after_hb);
 947                 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
 948
 949                 mlog(ML_HEARTBEAT,
 950                      "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
 951                      before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
 952                      after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
 953                      elapsed_msec);
 954
 955                 if (elapsed_msec < reg->hr_timeout_ms) {
 956                         /* the kthread api has blocked signals for us so no
 957                          * need to record the return value. */
 958                         msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
 959                 }
 960         }
 961
 962         o2hb_disarm_write_timeout(reg);
 963
 964         /* unclean stop is only used in very bad situation */
 965         for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
 966                 o2hb_shutdown_slot(&reg->hr_slots[i]);
 967
 968         /* Explicit down notification - avoid forcing the other nodes
 969          * to timeout on this region when we could just as easily
 970          * write a clear generation - thus indicating to them that
 971          * this node has left this region.
 972          *
 973          * XXX: Should we skip this on unclean_stop? */
 974         o2hb_prepare_block(reg, 0);
 975         ret = o2hb_issue_node_write(reg, &write_wc);
 976         if (ret == 0) {
 977                 o2hb_wait_on_io(reg, &write_wc);
 978         } else {
 979                 mlog_errno(ret);
 980         }
 981
 982         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n");
 983
 984         return 0;
 985 }
 986
 987 #ifdef CONFIG_DEBUG_FS
 988 static int o2hb_debug_open(struct inode *inode, struct file *file)
 989 {
 990         struct o2hb_debug_buf *db = inode->i_private;
 991         unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
 992         char *buf = NULL;
 993         int i = -1;
 994         int out = 0;
 995
 996         /* max_nodes should be the largest bitmap we pass here */
 997         BUG_ON(sizeof(map) < db->db_size);
 998
 999         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1000         if (!buf)
1001                 goto bail;
1002
1003         switch (db->db_type) {
1004         case O2HB_DB_TYPE_LIVENODES:
1005                 spin_lock(&o2hb_live_lock);
1006                 memcpy(map, db->db_data, db->db_size);
1007                 spin_unlock(&o2hb_live_lock);
1008                 break;
1009
1010         default:
1011                 goto done;
1012         }
1013
1014         while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1015                 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1016         out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1017
1018 done:
1019         i_size_write(inode, out);
1020
1021         file->private_data = buf;
1022
1023         return 0;
1024 bail:
1025         return -ENOMEM;
1026 }
1027
1028 static int o2hb_debug_release(struct inode *inode, struct file *file)
1029 {
1030         kfree(file->private_data);
1031         return 0;
1032 }
1033
1034 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1035                                  size_t nbytes, loff_t *ppos)
1036 {
1037         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1038                                        i_size_read(file->f_mapping->host));
1039 }
1040 #else
1041 static int o2hb_debug_open(struct inode *inode, struct file *file)
1042 {
1043         return 0;
1044 }
1045 static int o2hb_debug_release(struct inode *inode, struct file *file)
1046 {
1047         return 0;
1048 }
1049 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1050                                size_t nbytes, loff_t *ppos)
1051 {
1052         return 0;
1053 }
1054 #endif  /* CONFIG_DEBUG_FS */
1055
1056 static const struct file_operations o2hb_debug_fops = {
1057         .open =         o2hb_debug_open,
1058         .release =      o2hb_debug_release,
1059         .read =         o2hb_debug_read,
1060         .llseek =       generic_file_llseek,
1061 };
1062
1063 void o2hb_exit(void)
1064 {
1065         kfree(o2hb_db_livenodes);
1066         debugfs_remove(o2hb_debug_livenodes);
1067         debugfs_remove(o2hb_debug_dir);
1068 }
1069
1070 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1071                                         struct o2hb_debug_buf **db, int db_len,
1072                                         int type, int size, int len, void *data)
1073 {
1074         *db = kmalloc(db_len, GFP_KERNEL);
1075         if (!*db)
1076                 return NULL;
1077
1078         (*db)->db_type = type;
1079         (*db)->db_size = size;
1080         (*db)->db_len = len;
1081         (*db)->db_data = data;
1082
1083         return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1084                                    &o2hb_debug_fops);
1085 }
1086
1087 static int o2hb_debug_init(void)
1088 {
1089         int ret = -ENOMEM;
1090
1091         o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1092         if (!o2hb_debug_dir) {
1093                 mlog_errno(ret);
1094                 goto bail;
1095         }
1096
1097         o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1098                                                  o2hb_debug_dir,
1099                                                  &o2hb_db_livenodes,
1100                                                  sizeof(*o2hb_db_livenodes),
1101                                                  O2HB_DB_TYPE_LIVENODES,
1102                                                  sizeof(o2hb_live_node_bitmap),
1103                                                  O2NM_MAX_NODES,
1104                                                  o2hb_live_node_bitmap);
1105         if (!o2hb_debug_livenodes) {
1106                 mlog_errno(ret);
1107                 goto bail;
1108         }
1109         ret = 0;
1110 bail:
1111         if (ret)
1112                 o2hb_exit();
1113
1114         return ret;
1115 }
1116
1117 int o2hb_init(void)
1118 {
1119         int i;
1120
1121         for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1122                 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1123
1124         for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1125                 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1126
1127         INIT_LIST_HEAD(&o2hb_node_events);
1128
1129         memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1130
1131         return o2hb_debug_init();
1132 }
1133
1134 /* if we're already in a callback then we're already serialized by the sem */
1135 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1136                                              unsigned bytes)
1137 {
1138         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1139
1140         memcpy(map, &o2hb_live_node_bitmap, bytes);
1141 }
1142
1143 /*
1144  * get a map of all nodes that are heartbeating in any regions
1145  */
1146 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1147 {
1148         /* callers want to serialize this map and callbacks so that they
1149          * can trust that they don't miss nodes coming to the party */
1150         down_read(&o2hb_callback_sem);
1151         spin_lock(&o2hb_live_lock);
1152         o2hb_fill_node_map_from_callback(map, bytes);
1153         spin_unlock(&o2hb_live_lock);
1154         up_read(&o2hb_callback_sem);
1155 }
1156 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1157
1158 /*
1159  * heartbeat configfs bits.  The heartbeat set is a default set under
1160  * the cluster set in nodemanager.c.
1161  */
1162
1163 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1164 {
1165         return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1166 }
1167
1168 /* drop_item only drops its ref after killing the thread, nothing should
1169  * be using the region anymore.  this has to clean up any state that
1170  * attributes might have built up. */
1171 static void o2hb_region_release(struct config_item *item)
1172 {
1173         int i;
1174         struct page *page;
1175         struct o2hb_region *reg = to_o2hb_region(item);
1176
1177         if (reg->hr_tmp_block)
1178                 kfree(reg->hr_tmp_block);
1179
1180         if (reg->hr_slot_data) {
1181                 for (i = 0; i < reg->hr_num_pages; i++) {
1182                         page = reg->hr_slot_data[i];
1183                         if (page)
1184                                 __free_page(page);
1185                 }
1186                 kfree(reg->hr_slot_data);
1187         }
1188
1189         if (reg->hr_bdev)
1190                 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1191
1192         if (reg->hr_slots)
1193                 kfree(reg->hr_slots);
1194
1195         spin_lock(&o2hb_live_lock);
1196         list_del(&reg->hr_all_item);
1197         spin_unlock(&o2hb_live_lock);
1198
1199         kfree(reg);
1200 }
1201
1202 static int o2hb_read_block_input(struct o2hb_region *reg,
1203                                  const char *page,
1204                                  size_t count,
1205                                  unsigned long *ret_bytes,
1206                                  unsigned int *ret_bits)
1207 {
1208         unsigned long bytes;
1209         char *p = (char *)page;
1210
1211         bytes = simple_strtoul(p, &p, 0);
1212         if (!p || (*p && (*p != '\n')))
1213                 return -EINVAL;
1214
1215         /* Heartbeat and fs min / max block sizes are the same. */
1216         if (bytes > 4096 || bytes < 512)
1217                 return -ERANGE;
1218         if (hweight16(bytes) != 1)
1219                 return -EINVAL;
1220
1221         if (ret_bytes)
1222                 *ret_bytes = bytes;
1223         if (ret_bits)
1224                 *ret_bits = ffs(bytes) - 1;
1225
1226         return 0;
1227 }
1228
1229 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1230                                             char *page)
1231 {
1232         return sprintf(page, "%u\n", reg->hr_block_bytes);
1233 }
1234
1235 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1236                                              const char *page,
1237                                              size_t count)
1238 {
1239         int status;
1240         unsigned long block_bytes;
1241         unsigned int block_bits;
1242
1243         if (reg->hr_bdev)
1244                 return -EINVAL;
1245
1246         status = o2hb_read_block_input(reg, page, count,
1247                                        &block_bytes, &block_bits);
1248         if (status)
1249                 return status;
1250
1251         reg->hr_block_bytes = (unsigned int)block_bytes;
1252         reg->hr_block_bits = block_bits;
1253
1254         return count;
1255 }
1256
1257 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1258                                             char *page)
1259 {
1260         return sprintf(page, "%llu\n", reg->hr_start_block);
1261 }
1262
1263 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1264                                              const char *page,
1265                                              size_t count)
1266 {
1267         unsigned long long tmp;
1268         char *p = (char *)page;
1269
1270         if (reg->hr_bdev)
1271                 return -EINVAL;
1272
1273         tmp = simple_strtoull(p, &p, 0);
1274         if (!p || (*p && (*p != '\n')))
1275                 return -EINVAL;
1276
1277         reg->hr_start_block = tmp;
1278
1279         return count;
1280 }
1281
1282 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1283                                        char *page)
1284 {
1285         return sprintf(page, "%d\n", reg->hr_blocks);
1286 }
1287
1288 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1289                                         const char *page,
1290                                         size_t count)
1291 {
1292         unsigned long tmp;
1293         char *p = (char *)page;
1294
1295         if (reg->hr_bdev)
1296                 return -EINVAL;
1297
1298         tmp = simple_strtoul(p, &p, 0);
1299         if (!p || (*p && (*p != '\n')))
1300                 return -EINVAL;
1301
1302         if (tmp > O2NM_MAX_NODES || tmp == 0)
1303                 return -ERANGE;
1304
1305         reg->hr_blocks = (unsigned int)tmp;
1306
1307         return count;
1308 }
1309
1310 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1311                                     char *page)
1312 {
1313         unsigned int ret = 0;
1314
1315         if (reg->hr_bdev)
1316                 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1317
1318         return ret;
1319 }
1320
1321 static void o2hb_init_region_params(struct o2hb_region *reg)
1322 {
1323         reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1324         reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1325
1326         mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1327              reg->hr_start_block, reg->hr_blocks);
1328         mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1329              reg->hr_block_bytes, reg->hr_block_bits);
1330         mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1331         mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1332 }
1333
1334 static int o2hb_map_slot_data(struct o2hb_region *reg)
1335 {
1336         int i, j;
1337         unsigned int last_slot;
1338         unsigned int spp = reg->hr_slots_per_page;
1339         struct page *page;
1340         char *raw;
1341         struct o2hb_disk_slot *slot;
1342
1343         reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1344         if (reg->hr_tmp_block == NULL) {
1345                 mlog_errno(-ENOMEM);
1346                 return -ENOMEM;
1347         }
1348
1349         reg->hr_slots = kcalloc(reg->hr_blocks,
1350                                 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1351         if (reg->hr_slots == NULL) {
1352                 mlog_errno(-ENOMEM);
1353                 return -ENOMEM;
1354         }
1355
1356         for(i = 0; i < reg->hr_blocks; i++) {
1357                 slot = &reg->hr_slots[i];
1358                 slot->ds_node_num = i;
1359                 INIT_LIST_HEAD(&slot->ds_live_item);
1360                 slot->ds_raw_block = NULL;
1361         }
1362
1363         reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1364         mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1365                            "at %u blocks per page\n",
1366              reg->hr_num_pages, reg->hr_blocks, spp);
1367
1368         reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1369                                     GFP_KERNEL);
1370         if (!reg->hr_slot_data) {
1371                 mlog_errno(-ENOMEM);
1372                 return -ENOMEM;
1373         }
1374
1375         for(i = 0; i < reg->hr_num_pages; i++) {
1376                 page = alloc_page(GFP_KERNEL);
1377                 if (!page) {
1378                         mlog_errno(-ENOMEM);
1379                         return -ENOMEM;
1380                 }
1381
1382                 reg->hr_slot_data[i] = page;
1383
1384                 last_slot = i * spp;
1385                 raw = page_address(page);
1386                 for (j = 0;
1387                      (j < spp) && ((j + last_slot) < reg->hr_blocks);
1388                      j++) {
1389                         BUG_ON((j + last_slot) >= reg->hr_blocks);
1390
1391                         slot = &reg->hr_slots[j + last_slot];
1392                         slot->ds_raw_block =
1393                                 (struct o2hb_disk_heartbeat_block *) raw;
1394
1395                         raw += reg->hr_block_bytes;
1396                 }
1397         }
1398
1399         return 0;
1400 }
1401
1402 /* Read in all the slots available and populate the tracking
1403  * structures so that we can start with a baseline idea of what's
1404  * there. */
1405 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1406 {
1407         int ret, i;
1408         struct o2hb_disk_slot *slot;
1409         struct o2hb_disk_heartbeat_block *hb_block;
1410
1411         mlog_entry_void();
1412
1413         ret = o2hb_read_slots(reg, reg->hr_blocks);
1414         if (ret) {
1415                 mlog_errno(ret);
1416                 goto out;
1417         }
1418
1419         /* We only want to get an idea of the values initially in each
1420          * slot, so we do no verification - o2hb_check_slot will
1421          * actually determine if each configured slot is valid and
1422          * whether any values have changed. */
1423         for(i = 0; i < reg->hr_blocks; i++) {
1424                 slot = &reg->hr_slots[i];
1425                 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1426
1427                 /* Only fill the values that o2hb_check_slot uses to
1428                  * determine changing slots */
1429                 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1430                 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1431         }
1432
1433 out:
1434         mlog_exit(ret);
1435         return ret;
1436 }
1437
1438 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1439 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1440                                      const char *page,
1441                                      size_t count)
1442 {
1443         struct task_struct *hb_task;
1444         long fd;
1445         int sectsize;
1446         char *p = (char *)page;
1447         struct file *filp = NULL;
1448         struct inode *inode = NULL;
1449         ssize_t ret = -EINVAL;
1450
1451         if (reg->hr_bdev)
1452                 goto out;
1453
1454         /* We can't heartbeat without having had our node number
1455          * configured yet. */
1456         if (o2nm_this_node() == O2NM_MAX_NODES)
1457                 goto out;
1458
1459         fd = simple_strtol(p, &p, 0);
1460         if (!p || (*p && (*p != '\n')))
1461                 goto out;
1462
1463         if (fd < 0 || fd >= INT_MAX)
1464                 goto out;
1465
1466         filp = fget(fd);
1467         if (filp == NULL)
1468                 goto out;
1469
1470         if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1471             reg->hr_block_bytes == 0)
1472                 goto out;
1473
1474         inode = igrab(filp->f_mapping->host);
1475         if (inode == NULL)
1476                 goto out;
1477
1478         if (!S_ISBLK(inode->i_mode))
1479                 goto out;
1480
1481         reg->hr_bdev = I_BDEV(filp->f_mapping->host);
1482         ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ);
1483         if (ret) {
1484                 reg->hr_bdev = NULL;
1485                 goto out;
1486         }
1487         inode = NULL;
1488
1489         bdevname(reg->hr_bdev, reg->hr_dev_name);
1490
1491         sectsize = bdev_logical_block_size(reg->hr_bdev);
1492         if (sectsize != reg->hr_block_bytes) {
1493                 mlog(ML_ERROR,
1494                      "blocksize %u incorrect for device, expected %d",
1495                      reg->hr_block_bytes, sectsize);
1496                 ret = -EINVAL;
1497                 goto out;
1498         }
1499
1500         o2hb_init_region_params(reg);
1501
1502         /* Generation of zero is invalid */
1503         do {
1504                 get_random_bytes(&reg->hr_generation,
1505                                  sizeof(reg->hr_generation));
1506         } while (reg->hr_generation == 0);
1507
1508         ret = o2hb_map_slot_data(reg);
1509         if (ret) {
1510                 mlog_errno(ret);
1511                 goto out;
1512         }
1513
1514         ret = o2hb_populate_slot_data(reg);
1515         if (ret) {
1516                 mlog_errno(ret);
1517                 goto out;
1518         }
1519
1520         INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1521
1522         /*
1523          * A node is considered live after it has beat LIVE_THRESHOLD
1524          * times.  We're not steady until we've given them a chance
1525          * _after_ our first read.
1526          */
1527         atomic_set(&reg->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1);
1528
1529         hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1530                               reg->hr_item.ci_name);
1531         if (IS_ERR(hb_task)) {
1532                 ret = PTR_ERR(hb_task);
1533                 mlog_errno(ret);
1534                 goto out;
1535         }
1536
1537         spin_lock(&o2hb_live_lock);
1538         reg->hr_task = hb_task;
1539         spin_unlock(&o2hb_live_lock);
1540
1541         ret = wait_event_interruptible(o2hb_steady_queue,
1542                                 atomic_read(&reg->hr_steady_iterations) == 0);
1543         if (ret) {
1544                 /* We got interrupted (hello ptrace!).  Clean up */
1545                 spin_lock(&o2hb_live_lock);
1546                 hb_task = reg->hr_task;
1547                 reg->hr_task = NULL;
1548                 spin_unlock(&o2hb_live_lock);
1549
1550                 if (hb_task)
1551                         kthread_stop(hb_task);
1552                 goto out;
1553         }
1554
1555         /* Ok, we were woken.  Make sure it wasn't by drop_item() */
1556         spin_lock(&o2hb_live_lock);
1557         hb_task = reg->hr_task;
1558         spin_unlock(&o2hb_live_lock);
1559
1560         if (hb_task)
1561                 ret = count;
1562         else
1563                 ret = -EIO;
1564
1565         if (hb_task && o2hb_global_heartbeat_active())
1566                 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s\n",
1567                        config_item_name(&reg->hr_item));
1568
1569 out:
1570         if (filp)
1571                 fput(filp);
1572         if (inode)
1573                 iput(inode);
1574         if (ret < 0) {
1575                 if (reg->hr_bdev) {
1576                         blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1577                         reg->hr_bdev = NULL;
1578                 }
1579         }
1580         return ret;
1581 }
1582
1583 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1584                                       char *page)
1585 {
1586         pid_t pid = 0;
1587
1588         spin_lock(&o2hb_live_lock);
1589         if (reg->hr_task)
1590                 pid = task_pid_nr(reg->hr_task);
1591         spin_unlock(&o2hb_live_lock);
1592
1593         if (!pid)
1594                 return 0;
1595
1596         return sprintf(page, "%u\n", pid);
1597 }
1598
1599 struct o2hb_region_attribute {
1600         struct configfs_attribute attr;
1601         ssize_t (*show)(struct o2hb_region *, char *);
1602         ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1603 };
1604
1605 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1606         .attr   = { .ca_owner = THIS_MODULE,
1607                     .ca_name = "block_bytes",
1608                     .ca_mode = S_IRUGO | S_IWUSR },
1609         .show   = o2hb_region_block_bytes_read,
1610         .store  = o2hb_region_block_bytes_write,
1611 };
1612
1613 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1614         .attr   = { .ca_owner = THIS_MODULE,
1615                     .ca_name = "start_block",
1616                     .ca_mode = S_IRUGO | S_IWUSR },
1617         .show   = o2hb_region_start_block_read,
1618         .store  = o2hb_region_start_block_write,
1619 };
1620
1621 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1622         .attr   = { .ca_owner = THIS_MODULE,
1623                     .ca_name = "blocks",
1624                     .ca_mode = S_IRUGO | S_IWUSR },
1625         .show   = o2hb_region_blocks_read,
1626         .store  = o2hb_region_blocks_write,
1627 };
1628
1629 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1630         .attr   = { .ca_owner = THIS_MODULE,
1631                     .ca_name = "dev",
1632                     .ca_mode = S_IRUGO | S_IWUSR },
1633         .show   = o2hb_region_dev_read,
1634         .store  = o2hb_region_dev_write,
1635 };
1636
1637 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1638        .attr   = { .ca_owner = THIS_MODULE,
1639                    .ca_name = "pid",
1640                    .ca_mode = S_IRUGO | S_IRUSR },
1641        .show   = o2hb_region_pid_read,
1642 };
1643
1644 static struct configfs_attribute *o2hb_region_attrs[] = {
1645         &o2hb_region_attr_block_bytes.attr,
1646         &o2hb_region_attr_start_block.attr,
1647         &o2hb_region_attr_blocks.attr,
1648         &o2hb_region_attr_dev.attr,
1649         &o2hb_region_attr_pid.attr,
1650         NULL,
1651 };
1652
1653 static ssize_t o2hb_region_show(struct config_item *item,
1654                                 struct configfs_attribute *attr,
1655                                 char *page)
1656 {
1657         struct o2hb_region *reg = to_o2hb_region(item);
1658         struct o2hb_region_attribute *o2hb_region_attr =
1659                 container_of(attr, struct o2hb_region_attribute, attr);
1660         ssize_t ret = 0;
1661
1662         if (o2hb_region_attr->show)
1663                 ret = o2hb_region_attr->show(reg, page);
1664         return ret;
1665 }
1666
1667 static ssize_t o2hb_region_store(struct config_item *item,
1668                                  struct configfs_attribute *attr,
1669                                  const char *page, size_t count)
1670 {
1671         struct o2hb_region *reg = to_o2hb_region(item);
1672         struct o2hb_region_attribute *o2hb_region_attr =
1673                 container_of(attr, struct o2hb_region_attribute, attr);
1674         ssize_t ret = -EINVAL;
1675
1676         if (o2hb_region_attr->store)
1677                 ret = o2hb_region_attr->store(reg, page, count);
1678         return ret;
1679 }
1680
1681 static struct configfs_item_operations o2hb_region_item_ops = {
1682         .release                = o2hb_region_release,
1683         .show_attribute         = o2hb_region_show,
1684         .store_attribute        = o2hb_region_store,
1685 };
1686
1687 static struct config_item_type o2hb_region_type = {
1688         .ct_item_ops    = &o2hb_region_item_ops,
1689         .ct_attrs       = o2hb_region_attrs,
1690         .ct_owner       = THIS_MODULE,
1691 };
1692
1693 /* heartbeat set */
1694
1695 struct o2hb_heartbeat_group {
1696         struct config_group hs_group;
1697         /* some stuff? */
1698 };
1699
1700 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1701 {
1702         return group ?
1703                 container_of(group, struct o2hb_heartbeat_group, hs_group)
1704                 : NULL;
1705 }
1706
1707 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
1708                                                           const char *name)
1709 {
1710         struct o2hb_region *reg = NULL;
1711
1712         reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
1713         if (reg == NULL)
1714                 return ERR_PTR(-ENOMEM);
1715
1716         if (strlen(name) > O2HB_MAX_REGION_NAME_LEN)
1717                 return ERR_PTR(-ENAMETOOLONG);
1718
1719         config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
1720
1721         spin_lock(&o2hb_live_lock);
1722         list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
1723         spin_unlock(&o2hb_live_lock);
1724
1725         return &reg->hr_item;
1726 }
1727
1728 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
1729                                            struct config_item *item)
1730 {
1731         struct task_struct *hb_task;
1732         struct o2hb_region *reg = to_o2hb_region(item);
1733
1734         /* stop the thread when the user removes the region dir */
1735         spin_lock(&o2hb_live_lock);
1736         hb_task = reg->hr_task;
1737         reg->hr_task = NULL;
1738         spin_unlock(&o2hb_live_lock);
1739
1740         if (hb_task)
1741                 kthread_stop(hb_task);
1742
1743         /*
1744          * If we're racing a dev_write(), we need to wake them.  They will
1745          * check reg->hr_task
1746          */
1747         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1748                 atomic_set(&reg->hr_steady_iterations, 0);
1749                 wake_up(&o2hb_steady_queue);
1750         }
1751
1752         if (o2hb_global_heartbeat_active())
1753                 printk(KERN_NOTICE "o2hb: Heartbeat stopped on region %s\n",
1754                        config_item_name(&reg->hr_item));
1755         config_item_put(item);
1756 }
1757
1758 struct o2hb_heartbeat_group_attribute {
1759         struct configfs_attribute attr;
1760         ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
1761         ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
1762 };
1763
1764 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
1765                                          struct configfs_attribute *attr,
1766                                          char *page)
1767 {
1768         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
1769         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
1770                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
1771         ssize_t ret = 0;
1772
1773         if (o2hb_heartbeat_group_attr->show)
1774                 ret = o2hb_heartbeat_group_attr->show(reg, page);
1775         return ret;
1776 }
1777
1778 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
1779                                           struct configfs_attribute *attr,
1780                                           const char *page, size_t count)
1781 {
1782         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
1783         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
1784                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
1785         ssize_t ret = -EINVAL;
1786
1787         if (o2hb_heartbeat_group_attr->store)
1788                 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
1789         return ret;
1790 }
1791
1792 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
1793                                                      char *page)
1794 {
1795         return sprintf(page, "%u\n", o2hb_dead_threshold);
1796 }
1797
1798 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
1799                                                     const char *page,
1800                                                     size_t count)
1801 {
1802         unsigned long tmp;
1803         char *p = (char *)page;
1804
1805         tmp = simple_strtoul(p, &p, 10);
1806         if (!p || (*p && (*p != '\n')))
1807                 return -EINVAL;
1808
1809         /* this will validate ranges for us. */
1810         o2hb_dead_threshold_set((unsigned int) tmp);
1811
1812         return count;
1813 }
1814
1815 static
1816 ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
1817                                        char *page)
1818 {
1819         return sprintf(page, "%s\n",
1820                        o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
1821 }
1822
1823 static
1824 ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
1825                                         const char *page, size_t count)
1826 {
1827         unsigned int i;
1828         int ret;
1829         size_t len;
1830
1831         len = (page[count - 1] == '\n') ? count - 1 : count;
1832         if (!len)
1833                 return -EINVAL;
1834
1835         for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
1836                 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
1837                         continue;
1838
1839                 ret = o2hb_global_hearbeat_mode_set(i);
1840                 if (!ret)
1841                         printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
1842                                o2hb_heartbeat_mode_desc[i]);
1843                 return count;
1844         }
1845
1846         return -EINVAL;
1847
1848 }
1849
1850 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
1851         .attr   = { .ca_owner = THIS_MODULE,
1852                     .ca_name = "dead_threshold",
1853                     .ca_mode = S_IRUGO | S_IWUSR },
1854         .show   = o2hb_heartbeat_group_threshold_show,
1855         .store  = o2hb_heartbeat_group_threshold_store,
1856 };
1857
1858 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
1859         .attr   = { .ca_owner = THIS_MODULE,
1860                 .ca_name = "mode",
1861                 .ca_mode = S_IRUGO | S_IWUSR },
1862         .show   = o2hb_heartbeat_group_mode_show,
1863         .store  = o2hb_heartbeat_group_mode_store,
1864 };
1865
1866 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
1867         &o2hb_heartbeat_group_attr_threshold.attr,
1868         &o2hb_heartbeat_group_attr_mode.attr,
1869         NULL,
1870 };
1871
1872 static struct configfs_item_operations o2hb_hearbeat_group_item_ops = {
1873         .show_attribute         = o2hb_heartbeat_group_show,
1874         .store_attribute        = o2hb_heartbeat_group_store,
1875 };
1876
1877 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
1878         .make_item      = o2hb_heartbeat_group_make_item,
1879         .drop_item      = o2hb_heartbeat_group_drop_item,
1880 };
1881
1882 static struct config_item_type o2hb_heartbeat_group_type = {
1883         .ct_group_ops   = &o2hb_heartbeat_group_group_ops,
1884         .ct_item_ops    = &o2hb_hearbeat_group_item_ops,
1885         .ct_attrs       = o2hb_heartbeat_group_attrs,
1886         .ct_owner       = THIS_MODULE,
1887 };
1888
1889 /* this is just here to avoid touching group in heartbeat.h which the
1890  * entire damn world #includes */
1891 struct config_group *o2hb_alloc_hb_set(void)
1892 {
1893         struct o2hb_heartbeat_group *hs = NULL;
1894         struct config_group *ret = NULL;
1895
1896         hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
1897         if (hs == NULL)
1898                 goto out;
1899
1900         config_group_init_type_name(&hs->hs_group, "heartbeat",
1901                                     &o2hb_heartbeat_group_type);
1902
1903         ret = &hs->hs_group;
1904 out:
1905         if (ret == NULL)
1906                 kfree(hs);
1907         return ret;
1908 }
1909
1910 void o2hb_free_hb_set(struct config_group *group)
1911 {
1912         struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
1913         kfree(hs);
1914 }
1915
1916 /* hb callback registration and issueing */
1917
1918 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
1919 {
1920         if (type == O2HB_NUM_CB)
1921                 return ERR_PTR(-EINVAL);
1922
1923         return &o2hb_callbacks[type];
1924 }
1925
1926 void o2hb_setup_callback(struct o2hb_callback_func *hc,
1927                          enum o2hb_callback_type type,
1928                          o2hb_cb_func *func,
1929                          void *data,
1930                          int priority)
1931 {
1932         INIT_LIST_HEAD(&hc->hc_item);
1933         hc->hc_func = func;
1934         hc->hc_data = data;
1935         hc->hc_priority = priority;
1936         hc->hc_type = type;
1937         hc->hc_magic = O2HB_CB_MAGIC;
1938 }
1939 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
1940
1941 static struct o2hb_region *o2hb_find_region(const char *region_uuid)
1942 {
1943         struct o2hb_region *p, *reg = NULL;
1944
1945         assert_spin_locked(&o2hb_live_lock);
1946
1947         list_for_each_entry(p, &o2hb_all_regions, hr_all_item) {
1948                 if (!strcmp(region_uuid, config_item_name(&p->hr_item))) {
1949                         reg = p;
1950                         break;
1951                 }
1952         }
1953
1954         return reg;
1955 }
1956
1957 static int o2hb_region_get(const char *region_uuid)
1958 {
1959         int ret = 0;
1960         struct o2hb_region *reg;
1961
1962         spin_lock(&o2hb_live_lock);
1963
1964         reg = o2hb_find_region(region_uuid);
1965         if (!reg)
1966                 ret = -ENOENT;
1967         spin_unlock(&o2hb_live_lock);
1968
1969         if (ret)
1970                 goto out;
1971
1972         ret = o2nm_depend_this_node();
1973         if (ret)
1974                 goto out;
1975
1976         ret = o2nm_depend_item(&reg->hr_item);
1977         if (ret)
1978                 o2nm_undepend_this_node();
1979
1980 out:
1981         return ret;
1982 }
1983
1984 static void o2hb_region_put(const char *region_uuid)
1985 {
1986         struct o2hb_region *reg;
1987
1988         spin_lock(&o2hb_live_lock);
1989
1990         reg = o2hb_find_region(region_uuid);
1991
1992         spin_unlock(&o2hb_live_lock);
1993
1994         if (reg) {
1995                 o2nm_undepend_item(&reg->hr_item);
1996                 o2nm_undepend_this_node();
1997         }
1998 }
1999
2000 int o2hb_register_callback(const char *region_uuid,
2001                            struct o2hb_callback_func *hc)
2002 {
2003         struct o2hb_callback_func *tmp;
2004         struct list_head *iter;
2005         struct o2hb_callback *hbcall;
2006         int ret;
2007
2008         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2009         BUG_ON(!list_empty(&hc->hc_item));
2010
2011         hbcall = hbcall_from_type(hc->hc_type);
2012         if (IS_ERR(hbcall)) {
2013                 ret = PTR_ERR(hbcall);
2014                 goto out;
2015         }
2016
2017         if (region_uuid) {
2018                 ret = o2hb_region_get(region_uuid);
2019                 if (ret)
2020                         goto out;
2021         }
2022
2023         down_write(&o2hb_callback_sem);
2024
2025         list_for_each(iter, &hbcall->list) {
2026                 tmp = list_entry(iter, struct o2hb_callback_func, hc_item);
2027                 if (hc->hc_priority < tmp->hc_priority) {
2028                         list_add_tail(&hc->hc_item, iter);
2029                         break;
2030                 }
2031         }
2032         if (list_empty(&hc->hc_item))
2033                 list_add_tail(&hc->hc_item, &hbcall->list);
2034
2035         up_write(&o2hb_callback_sem);
2036         ret = 0;
2037 out:
2038         mlog(ML_HEARTBEAT, "returning %d on behalf of %p for funcs %p\n",
2039              ret, __builtin_return_address(0), hc);
2040         return ret;
2041 }
2042 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2043
2044 void o2hb_unregister_callback(const char *region_uuid,
2045                               struct o2hb_callback_func *hc)
2046 {
2047         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2048
2049         mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n",
2050              __builtin_return_address(0), hc);
2051
2052         /* XXX Can this happen _with_ a region reference? */
2053         if (list_empty(&hc->hc_item))
2054                 return;
2055
2056         if (region_uuid)
2057                 o2hb_region_put(region_uuid);
2058
2059         down_write(&o2hb_callback_sem);
2060
2061         list_del_init(&hc->hc_item);
2062
2063         up_write(&o2hb_callback_sem);
2064 }
2065 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2066
2067 int o2hb_check_node_heartbeating(u8 node_num)
2068 {
2069         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2070
2071         o2hb_fill_node_map(testing_map, sizeof(testing_map));
2072         if (!test_bit(node_num, testing_map)) {
2073                 mlog(ML_HEARTBEAT,
2074                      "node (%u) does not have heartbeating enabled.\n",
2075                      node_num);
2076                 return 0;
2077         }
2078
2079         return 1;
2080 }
2081 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2082
2083 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2084 {
2085         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2086
2087         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2088         if (!test_bit(node_num, testing_map)) {
2089                 mlog(ML_HEARTBEAT,
2090                      "node (%u) does not have heartbeating enabled.\n",
2091                      node_num);
2092                 return 0;
2093         }
2094
2095         return 1;
2096 }
2097 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2098
2099 /* Makes sure our local node is configured with a node number, and is
2100  * heartbeating. */
2101 int o2hb_check_local_node_heartbeating(void)
2102 {
2103         u8 node_num;
2104
2105         /* if this node was set then we have networking */
2106         node_num = o2nm_this_node();
2107         if (node_num == O2NM_MAX_NODES) {
2108                 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2109                 return 0;
2110         }
2111
2112         return o2hb_check_node_heartbeating(node_num);
2113 }
2114 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2115
2116 /*
2117  * this is just a hack until we get the plumbing which flips file systems
2118  * read only and drops the hb ref instead of killing the node dead.
2119  */
2120 void o2hb_stop_all_regions(void)
2121 {
2122         struct o2hb_region *reg;
2123
2124         mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2125
2126         spin_lock(&o2hb_live_lock);
2127
2128         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2129                 reg->hr_unclean_stop = 1;
2130
2131         spin_unlock(&o2hb_live_lock);
2132 }
2133 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2134
2135 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2136 {
2137         struct o2hb_region *reg;
2138         int numregs = 0;
2139         char *p;
2140
2141         spin_lock(&o2hb_live_lock);
2142
2143         p = region_uuids;
2144         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2145                 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2146                 if (numregs < max_regions) {
2147                         memcpy(p, config_item_name(&reg->hr_item),
2148                                O2HB_MAX_REGION_NAME_LEN);
2149                         p += O2HB_MAX_REGION_NAME_LEN;
2150                 }
2151                 numregs++;
2152         }
2153
2154         spin_unlock(&o2hb_live_lock);
2155
2156         return numregs;
2157 }
2158 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2159
2160 int o2hb_global_heartbeat_active(void)
2161 {
2162         return 0;
2163 }
2164 EXPORT_SYMBOL(o2hb_global_heartbeat_active);