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 struct o2hb_debug_buf {
  86         int db_type;
  87         int db_size;
  88         int db_len;
  89         void *db_data;
  90 };
  91
  92 static struct o2hb_debug_buf *o2hb_db_livenodes;
  93 static struct o2hb_debug_buf *o2hb_db_liveregions;
  94 static struct o2hb_debug_buf *o2hb_db_quorumregions;
  95 static struct o2hb_debug_buf *o2hb_db_failedregions;
  96
  97 #define O2HB_DEBUG_DIR                  "o2hb"
  98 #define O2HB_DEBUG_LIVENODES            "livenodes"
  99 #define O2HB_DEBUG_LIVEREGIONS          "live_regions"
 100 #define O2HB_DEBUG_QUORUMREGIONS        "quorum_regions"
 101 #define O2HB_DEBUG_FAILEDREGIONS        "failed_regions"
 102 #define O2HB_DEBUG_REGION_NUMBER        "num"
 103 #define O2HB_DEBUG_REGION_ELAPSED_TIME  "elapsed_time_in_ms"
 104
 105 static struct dentry *o2hb_debug_dir;
 106 static struct dentry *o2hb_debug_livenodes;
 107 static struct dentry *o2hb_debug_liveregions;
 108 static struct dentry *o2hb_debug_quorumregions;
 109 static struct dentry *o2hb_debug_failedregions;
 110
 111 static LIST_HEAD(o2hb_all_regions);
 112
 113 static struct o2hb_callback {
 114         struct list_head list;
 115 } o2hb_callbacks[O2HB_NUM_CB];
 116
 117 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
 118
 119 #define O2HB_DEFAULT_BLOCK_BITS       9
 120
 121 enum o2hb_heartbeat_modes {
 122         O2HB_HEARTBEAT_LOCAL            = 0,
 123         O2HB_HEARTBEAT_GLOBAL,
 124         O2HB_HEARTBEAT_NUM_MODES,
 125 };
 126
 127 char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
 128                 "local",        /* O2HB_HEARTBEAT_LOCAL */
 129                 "global",       /* O2HB_HEARTBEAT_GLOBAL */
 130 };
 131
 132 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
 133 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
 134
 135 /* Only sets a new threshold if there are no active regions.
 136  *
 137  * No locking or otherwise interesting code is required for reading
 138  * o2hb_dead_threshold as it can't change once regions are active and
 139  * it's not interesting to anyone until then anyway. */
 140 static void o2hb_dead_threshold_set(unsigned int threshold)
 141 {
 142         if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
 143                 spin_lock(&o2hb_live_lock);
 144                 if (list_empty(&o2hb_all_regions))
 145                         o2hb_dead_threshold = threshold;
 146                 spin_unlock(&o2hb_live_lock);
 147         }
 148 }
 149
 150 static int o2hb_global_hearbeat_mode_set(unsigned int hb_mode)
 151 {
 152         int ret = -1;
 153
 154         if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
 155                 spin_lock(&o2hb_live_lock);
 156                 if (list_empty(&o2hb_all_regions)) {
 157                         o2hb_heartbeat_mode = hb_mode;
 158                         ret = 0;
 159                 }
 160                 spin_unlock(&o2hb_live_lock);
 161         }
 162
 163         return ret;
 164 }
 165
 166 struct o2hb_node_event {
 167         struct list_head        hn_item;
 168         enum o2hb_callback_type hn_event_type;
 169         struct o2nm_node        *hn_node;
 170         int                     hn_node_num;
 171 };
 172
 173 struct o2hb_disk_slot {
 174         struct o2hb_disk_heartbeat_block *ds_raw_block;
 175         u8                      ds_node_num;
 176         u64                     ds_last_time;
 177         u64                     ds_last_generation;
 178         u16                     ds_equal_samples;
 179         u16                     ds_changed_samples;
 180         struct list_head        ds_live_item;
 181 };
 182
 183 /* each thread owns a region.. when we're asked to tear down the region
 184  * we ask the thread to stop, who cleans up the region */
 185 struct o2hb_region {
 186         struct config_item      hr_item;
 187
 188         struct list_head        hr_all_item;
 189         unsigned                hr_unclean_stop:1;
 190
 191         /* protected by the hr_callback_sem */
 192         struct task_struct      *hr_task;
 193
 194         unsigned int            hr_blocks;
 195         unsigned long long      hr_start_block;
 196
 197         unsigned int            hr_block_bits;
 198         unsigned int            hr_block_bytes;
 199
 200         unsigned int            hr_slots_per_page;
 201         unsigned int            hr_num_pages;
 202
 203         struct page             **hr_slot_data;
 204         struct block_device     *hr_bdev;
 205         struct o2hb_disk_slot   *hr_slots;
 206
 207         /* live node map of this region */
 208         unsigned long           hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 209         unsigned int            hr_region_num;
 210
 211         struct dentry           *hr_debug_dir;
 212         struct dentry           *hr_debug_livenodes;
 213         struct dentry           *hr_debug_regnum;
 214         struct dentry           *hr_debug_elapsed_time;
 215         struct o2hb_debug_buf   *hr_db_livenodes;
 216         struct o2hb_debug_buf   *hr_db_regnum;
 217         struct o2hb_debug_buf   *hr_db_elapsed_time;
 218
 219         /* let the person setting up hb wait for it to return until it
 220          * has reached a 'steady' state.  This will be fixed when we have
 221          * a more complete api that doesn't lead to this sort of fragility. */
 222         atomic_t                hr_steady_iterations;
 223
 224         char                    hr_dev_name[BDEVNAME_SIZE];
 225
 226         unsigned int            hr_timeout_ms;
 227
 228         /* randomized as the region goes up and down so that a node
 229          * recognizes a node going up and down in one iteration */
 230         u64                     hr_generation;
 231
 232         struct delayed_work     hr_write_timeout_work;
 233         unsigned long           hr_last_timeout_start;
 234
 235         /* Used during o2hb_check_slot to hold a copy of the block
 236          * being checked because we temporarily have to zero out the
 237          * crc field. */
 238         struct o2hb_disk_heartbeat_block *hr_tmp_block;
 239 };
 240
 241 struct o2hb_bio_wait_ctxt {
 242         atomic_t          wc_num_reqs;
 243         struct completion wc_io_complete;
 244         int               wc_error;
 245 };
 246
 247 static int o2hb_pop_count(void *map, int count)
 248 {
 249         int i = -1, pop = 0;
 250
 251         while ((i = find_next_bit(map, count, i + 1)) < count)
 252                 pop++;
 253         return pop;
 254 }
 255
 256 static void o2hb_write_timeout(struct work_struct *work)
 257 {
 258         int failed, quorum;
 259         unsigned long flags;
 260         struct o2hb_region *reg =
 261                 container_of(work, struct o2hb_region,
 262                              hr_write_timeout_work.work);
 263
 264         mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
 265              "milliseconds\n", reg->hr_dev_name,
 266              jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
 267
 268         if (o2hb_global_heartbeat_active()) {
 269                 spin_lock_irqsave(&o2hb_live_lock, flags);
 270                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 271                         set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 272                 failed = o2hb_pop_count(&o2hb_failed_region_bitmap,
 273                                         O2NM_MAX_REGIONS);
 274                 quorum = o2hb_pop_count(&o2hb_quorum_region_bitmap,
 275                                         O2NM_MAX_REGIONS);
 276                 spin_unlock_irqrestore(&o2hb_live_lock, flags);
 277
 278                 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
 279                      quorum, failed);
 280
 281                 /*
 282                  * Fence if the number of failed regions >= half the number
 283                  * of  quorum regions
 284                  */
 285                 if ((failed << 1) < quorum)
 286                         return;
 287         }
 288
 289         o2quo_disk_timeout();
 290 }
 291
 292 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
 293 {
 294         mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
 295              O2HB_MAX_WRITE_TIMEOUT_MS);
 296
 297         if (o2hb_global_heartbeat_active()) {
 298                 spin_lock(&o2hb_live_lock);
 299                 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
 300                 spin_unlock(&o2hb_live_lock);
 301         }
 302         cancel_delayed_work(&reg->hr_write_timeout_work);
 303         reg->hr_last_timeout_start = jiffies;
 304         schedule_delayed_work(&reg->hr_write_timeout_work,
 305                               msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
 306 }
 307
 308 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
 309 {
 310         cancel_delayed_work_sync(&reg->hr_write_timeout_work);
 311 }
 312
 313 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
 314 {
 315         atomic_set(&wc->wc_num_reqs, 1);
 316         init_completion(&wc->wc_io_complete);
 317         wc->wc_error = 0;
 318 }
 319
 320 /* Used in error paths too */
 321 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
 322                                      unsigned int num)
 323 {
 324         /* sadly atomic_sub_and_test() isn't available on all platforms.  The
 325          * good news is that the fast path only completes one at a time */
 326         while(num--) {
 327                 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
 328                         BUG_ON(num > 0);
 329                         complete(&wc->wc_io_complete);
 330                 }
 331         }
 332 }
 333
 334 static void o2hb_wait_on_io(struct o2hb_region *reg,
 335                             struct o2hb_bio_wait_ctxt *wc)
 336 {
 337         struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping;
 338
 339         blk_run_address_space(mapping);
 340         o2hb_bio_wait_dec(wc, 1);
 341
 342         wait_for_completion(&wc->wc_io_complete);
 343 }
 344
 345 static void o2hb_bio_end_io(struct bio *bio,
 346                            int error)
 347 {
 348         struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
 349
 350         if (error) {
 351                 mlog(ML_ERROR, "IO Error %d\n", error);
 352                 wc->wc_error = error;
 353         }
 354
 355         o2hb_bio_wait_dec(wc, 1);
 356         bio_put(bio);
 357 }
 358
 359 /* Setup a Bio to cover I/O against num_slots slots starting at
 360  * start_slot. */
 361 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
 362                                       struct o2hb_bio_wait_ctxt *wc,
 363                                       unsigned int *current_slot,
 364                                       unsigned int max_slots)
 365 {
 366         int len, current_page;
 367         unsigned int vec_len, vec_start;
 368         unsigned int bits = reg->hr_block_bits;
 369         unsigned int spp = reg->hr_slots_per_page;
 370         unsigned int cs = *current_slot;
 371         struct bio *bio;
 372         struct page *page;
 373
 374         /* Testing has shown this allocation to take long enough under
 375          * GFP_KERNEL that the local node can get fenced. It would be
 376          * nicest if we could pre-allocate these bios and avoid this
 377          * all together. */
 378         bio = bio_alloc(GFP_ATOMIC, 16);
 379         if (!bio) {
 380                 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
 381                 bio = ERR_PTR(-ENOMEM);
 382                 goto bail;
 383         }
 384
 385         /* Must put everything in 512 byte sectors for the bio... */
 386         bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
 387         bio->bi_bdev = reg->hr_bdev;
 388         bio->bi_private = wc;
 389         bio->bi_end_io = o2hb_bio_end_io;
 390
 391         vec_start = (cs << bits) % PAGE_CACHE_SIZE;
 392         while(cs < max_slots) {
 393                 current_page = cs / spp;
 394                 page = reg->hr_slot_data[current_page];
 395
 396                 vec_len = min(PAGE_CACHE_SIZE - vec_start,
 397                               (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
 398
 399                 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
 400                      current_page, vec_len, vec_start);
 401
 402                 len = bio_add_page(bio, page, vec_len, vec_start);
 403                 if (len != vec_len) break;
 404
 405                 cs += vec_len / (PAGE_CACHE_SIZE/spp);
 406                 vec_start = 0;
 407         }
 408
 409 bail:
 410         *current_slot = cs;
 411         return bio;
 412 }
 413
 414 static int o2hb_read_slots(struct o2hb_region *reg,
 415                            unsigned int max_slots)
 416 {
 417         unsigned int current_slot=0;
 418         int status;
 419         struct o2hb_bio_wait_ctxt wc;
 420         struct bio *bio;
 421
 422         o2hb_bio_wait_init(&wc);
 423
 424         while(current_slot < max_slots) {
 425                 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
 426                 if (IS_ERR(bio)) {
 427                         status = PTR_ERR(bio);
 428                         mlog_errno(status);
 429                         goto bail_and_wait;
 430                 }
 431
 432                 atomic_inc(&wc.wc_num_reqs);
 433                 submit_bio(READ, bio);
 434         }
 435
 436         status = 0;
 437
 438 bail_and_wait:
 439         o2hb_wait_on_io(reg, &wc);
 440         if (wc.wc_error && !status)
 441                 status = wc.wc_error;
 442
 443         return status;
 444 }
 445
 446 static int o2hb_issue_node_write(struct o2hb_region *reg,
 447                                  struct o2hb_bio_wait_ctxt *write_wc)
 448 {
 449         int status;
 450         unsigned int slot;
 451         struct bio *bio;
 452
 453         o2hb_bio_wait_init(write_wc);
 454
 455         slot = o2nm_this_node();
 456
 457         bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
 458         if (IS_ERR(bio)) {
 459                 status = PTR_ERR(bio);
 460                 mlog_errno(status);
 461                 goto bail;
 462         }
 463
 464         atomic_inc(&write_wc->wc_num_reqs);
 465         submit_bio(WRITE, bio);
 466
 467         status = 0;
 468 bail:
 469         return status;
 470 }
 471
 472 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
 473                                      struct o2hb_disk_heartbeat_block *hb_block)
 474 {
 475         __le32 old_cksum;
 476         u32 ret;
 477
 478         /* We want to compute the block crc with a 0 value in the
 479          * hb_cksum field. Save it off here and replace after the
 480          * crc. */
 481         old_cksum = hb_block->hb_cksum;
 482         hb_block->hb_cksum = 0;
 483
 484         ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
 485
 486         hb_block->hb_cksum = old_cksum;
 487
 488         return ret;
 489 }
 490
 491 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
 492 {
 493         mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
 494              "cksum = 0x%x, generation 0x%llx\n",
 495              (long long)le64_to_cpu(hb_block->hb_seq),
 496              hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
 497              (long long)le64_to_cpu(hb_block->hb_generation));
 498 }
 499
 500 static int o2hb_verify_crc(struct o2hb_region *reg,
 501                            struct o2hb_disk_heartbeat_block *hb_block)
 502 {
 503         u32 read, computed;
 504
 505         read = le32_to_cpu(hb_block->hb_cksum);
 506         computed = o2hb_compute_block_crc_le(reg, hb_block);
 507
 508         return read == computed;
 509 }
 510
 511 /* We want to make sure that nobody is heartbeating on top of us --
 512  * this will help detect an invalid configuration. */
 513 static int o2hb_check_last_timestamp(struct o2hb_region *reg)
 514 {
 515         int node_num, ret;
 516         struct o2hb_disk_slot *slot;
 517         struct o2hb_disk_heartbeat_block *hb_block;
 518
 519         node_num = o2nm_this_node();
 520
 521         ret = 1;
 522         slot = &reg->hr_slots[node_num];
 523         /* Don't check on our 1st timestamp */
 524         if (slot->ds_last_time) {
 525                 hb_block = slot->ds_raw_block;
 526
 527                 if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time)
 528                         ret = 0;
 529         }
 530
 531         return ret;
 532 }
 533
 534 static inline void o2hb_prepare_block(struct o2hb_region *reg,
 535                                       u64 generation)
 536 {
 537         int node_num;
 538         u64 cputime;
 539         struct o2hb_disk_slot *slot;
 540         struct o2hb_disk_heartbeat_block *hb_block;
 541
 542         node_num = o2nm_this_node();
 543         slot = &reg->hr_slots[node_num];
 544
 545         hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
 546         memset(hb_block, 0, reg->hr_block_bytes);
 547         /* TODO: time stuff */
 548         cputime = CURRENT_TIME.tv_sec;
 549         if (!cputime)
 550                 cputime = 1;
 551
 552         hb_block->hb_seq = cpu_to_le64(cputime);
 553         hb_block->hb_node = node_num;
 554         hb_block->hb_generation = cpu_to_le64(generation);
 555         hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
 556
 557         /* This step must always happen last! */
 558         hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
 559                                                                    hb_block));
 560
 561         mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
 562              (long long)generation,
 563              le32_to_cpu(hb_block->hb_cksum));
 564 }
 565
 566 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
 567                                 struct o2nm_node *node,
 568                                 int idx)
 569 {
 570         struct list_head *iter;
 571         struct o2hb_callback_func *f;
 572
 573         list_for_each(iter, &hbcall->list) {
 574                 f = list_entry(iter, struct o2hb_callback_func, hc_item);
 575                 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
 576                 (f->hc_func)(node, idx, f->hc_data);
 577         }
 578 }
 579
 580 /* Will run the list in order until we process the passed event */
 581 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
 582 {
 583         int empty;
 584         struct o2hb_callback *hbcall;
 585         struct o2hb_node_event *event;
 586
 587         spin_lock(&o2hb_live_lock);
 588         empty = list_empty(&queued_event->hn_item);
 589         spin_unlock(&o2hb_live_lock);
 590         if (empty)
 591                 return;
 592
 593         /* Holding callback sem assures we don't alter the callback
 594          * lists when doing this, and serializes ourselves with other
 595          * processes wanting callbacks. */
 596         down_write(&o2hb_callback_sem);
 597
 598         spin_lock(&o2hb_live_lock);
 599         while (!list_empty(&o2hb_node_events)
 600                && !list_empty(&queued_event->hn_item)) {
 601                 event = list_entry(o2hb_node_events.next,
 602                                    struct o2hb_node_event,
 603                                    hn_item);
 604                 list_del_init(&event->hn_item);
 605                 spin_unlock(&o2hb_live_lock);
 606
 607                 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
 608                      event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
 609                      event->hn_node_num);
 610
 611                 hbcall = hbcall_from_type(event->hn_event_type);
 612
 613                 /* We should *never* have gotten on to the list with a
 614                  * bad type... This isn't something that we should try
 615                  * to recover from. */
 616                 BUG_ON(IS_ERR(hbcall));
 617
 618                 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
 619
 620                 spin_lock(&o2hb_live_lock);
 621         }
 622         spin_unlock(&o2hb_live_lock);
 623
 624         up_write(&o2hb_callback_sem);
 625 }
 626
 627 static void o2hb_queue_node_event(struct o2hb_node_event *event,
 628                                   enum o2hb_callback_type type,
 629                                   struct o2nm_node *node,
 630                                   int node_num)
 631 {
 632         assert_spin_locked(&o2hb_live_lock);
 633
 634         BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
 635
 636         event->hn_event_type = type;
 637         event->hn_node = node;
 638         event->hn_node_num = node_num;
 639
 640         mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
 641              type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
 642
 643         list_add_tail(&event->hn_item, &o2hb_node_events);
 644 }
 645
 646 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
 647 {
 648         struct o2hb_node_event event =
 649                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 650         struct o2nm_node *node;
 651
 652         node = o2nm_get_node_by_num(slot->ds_node_num);
 653         if (!node)
 654                 return;
 655
 656         spin_lock(&o2hb_live_lock);
 657         if (!list_empty(&slot->ds_live_item)) {
 658                 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
 659                      slot->ds_node_num);
 660
 661                 list_del_init(&slot->ds_live_item);
 662
 663                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 664                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 665
 666                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
 667                                               slot->ds_node_num);
 668                 }
 669         }
 670         spin_unlock(&o2hb_live_lock);
 671
 672         o2hb_run_event_list(&event);
 673
 674         o2nm_node_put(node);
 675 }
 676
 677 static void o2hb_set_quorum_device(struct o2hb_region *reg,
 678                                    struct o2hb_disk_slot *slot)
 679 {
 680         assert_spin_locked(&o2hb_live_lock);
 681
 682         if (!o2hb_global_heartbeat_active())
 683                 return;
 684
 685         if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
 686                 return;
 687
 688         /*
 689          * A region can be added to the quorum only when it sees all
 690          * live nodes heartbeat on it. In other words, the region has been
 691          * added to all nodes.
 692          */
 693         if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
 694                    sizeof(o2hb_live_node_bitmap)))
 695                 return;
 696
 697         if (slot->ds_changed_samples < O2HB_LIVE_THRESHOLD)
 698                 return;
 699
 700         printk(KERN_NOTICE "o2hb: Region %s is now a quorum device\n",
 701                config_item_name(&reg->hr_item));
 702
 703         set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
 704 }
 705
 706 static int o2hb_check_slot(struct o2hb_region *reg,
 707                            struct o2hb_disk_slot *slot)
 708 {
 709         int changed = 0, gen_changed = 0;
 710         struct o2hb_node_event event =
 711                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
 712         struct o2nm_node *node;
 713         struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
 714         u64 cputime;
 715         unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
 716         unsigned int slot_dead_ms;
 717         int tmp;
 718
 719         memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
 720
 721         /*
 722          * If a node is no longer configured but is still in the livemap, we
 723          * may need to clear that bit from the livemap.
 724          */
 725         node = o2nm_get_node_by_num(slot->ds_node_num);
 726         if (!node) {
 727                 spin_lock(&o2hb_live_lock);
 728                 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 729                 spin_unlock(&o2hb_live_lock);
 730                 if (!tmp)
 731                         return 0;
 732         }
 733
 734         if (!o2hb_verify_crc(reg, hb_block)) {
 735                 /* all paths from here will drop o2hb_live_lock for
 736                  * us. */
 737                 spin_lock(&o2hb_live_lock);
 738
 739                 /* Don't print an error on the console in this case -
 740                  * a freshly formatted heartbeat area will not have a
 741                  * crc set on it. */
 742                 if (list_empty(&slot->ds_live_item))
 743                         goto out;
 744
 745                 /* The node is live but pushed out a bad crc. We
 746                  * consider it a transient miss but don't populate any
 747                  * other values as they may be junk. */
 748                 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
 749                      slot->ds_node_num, reg->hr_dev_name);
 750                 o2hb_dump_slot(hb_block);
 751
 752                 slot->ds_equal_samples++;
 753                 goto fire_callbacks;
 754         }
 755
 756         /* we don't care if these wrap.. the state transitions below
 757          * clear at the right places */
 758         cputime = le64_to_cpu(hb_block->hb_seq);
 759         if (slot->ds_last_time != cputime)
 760                 slot->ds_changed_samples++;
 761         else
 762                 slot->ds_equal_samples++;
 763         slot->ds_last_time = cputime;
 764
 765         /* The node changed heartbeat generations. We assume this to
 766          * mean it dropped off but came back before we timed out. We
 767          * want to consider it down for the time being but don't want
 768          * to lose any changed_samples state we might build up to
 769          * considering it live again. */
 770         if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
 771                 gen_changed = 1;
 772                 slot->ds_equal_samples = 0;
 773                 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
 774                      "to 0x%llx)\n", slot->ds_node_num,
 775                      (long long)slot->ds_last_generation,
 776                      (long long)le64_to_cpu(hb_block->hb_generation));
 777         }
 778
 779         slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
 780
 781         mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
 782              "seq %llu last %llu changed %u equal %u\n",
 783              slot->ds_node_num, (long long)slot->ds_last_generation,
 784              le32_to_cpu(hb_block->hb_cksum),
 785              (unsigned long long)le64_to_cpu(hb_block->hb_seq),
 786              (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
 787              slot->ds_equal_samples);
 788
 789         spin_lock(&o2hb_live_lock);
 790
 791 fire_callbacks:
 792         /* dead nodes only come to life after some number of
 793          * changes at any time during their dead time */
 794         if (list_empty(&slot->ds_live_item) &&
 795             slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
 796                 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
 797                      slot->ds_node_num, (long long)slot->ds_last_generation);
 798
 799                 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 800
 801                 /* first on the list generates a callback */
 802                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 803                         mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
 804                              "bitmap\n", slot->ds_node_num);
 805                         set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 806
 807                         o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
 808                                               slot->ds_node_num);
 809
 810                         changed = 1;
 811                 }
 812
 813                 list_add_tail(&slot->ds_live_item,
 814                               &o2hb_live_slots[slot->ds_node_num]);
 815
 816                 slot->ds_equal_samples = 0;
 817
 818                 /* We want to be sure that all nodes agree on the
 819                  * number of milliseconds before a node will be
 820                  * considered dead. The self-fencing timeout is
 821                  * computed from this value, and a discrepancy might
 822                  * result in heartbeat calling a node dead when it
 823                  * hasn't self-fenced yet. */
 824                 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
 825                 if (slot_dead_ms && slot_dead_ms != dead_ms) {
 826                         /* TODO: Perhaps we can fail the region here. */
 827                         mlog(ML_ERROR, "Node %d on device %s has a dead count "
 828                              "of %u ms, but our count is %u ms.\n"
 829                              "Please double check your configuration values "
 830                              "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
 831                              slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
 832                              dead_ms);
 833                 }
 834                 goto out;
 835         }
 836
 837         /* if the list is dead, we're done.. */
 838         if (list_empty(&slot->ds_live_item))
 839                 goto out;
 840
 841         /* live nodes only go dead after enough consequtive missed
 842          * samples..  reset the missed counter whenever we see
 843          * activity */
 844         if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
 845                 mlog(ML_HEARTBEAT, "Node %d left my region\n",
 846                      slot->ds_node_num);
 847
 848                 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
 849
 850                 /* last off the live_slot generates a callback */
 851                 list_del_init(&slot->ds_live_item);
 852                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
 853                         mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
 854                              "nodes bitmap\n", slot->ds_node_num);
 855                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
 856
 857                         /* node can be null */
 858                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
 859                                               node, slot->ds_node_num);
 860
 861                         changed = 1;
 862                 }
 863
 864                 /* We don't clear this because the node is still
 865                  * actually writing new blocks. */
 866                 if (!gen_changed)
 867                         slot->ds_changed_samples = 0;
 868                 goto out;
 869         }
 870         if (slot->ds_changed_samples) {
 871                 slot->ds_changed_samples = 0;
 872                 slot->ds_equal_samples = 0;
 873         }
 874 out:
 875         o2hb_set_quorum_device(reg, slot);
 876
 877         spin_unlock(&o2hb_live_lock);
 878
 879         o2hb_run_event_list(&event);
 880
 881         if (node)
 882                 o2nm_node_put(node);
 883         return changed;
 884 }
 885
 886 /* This could be faster if we just implmented a find_last_bit, but I
 887  * don't think the circumstances warrant it. */
 888 static int o2hb_highest_node(unsigned long *nodes,
 889                              int numbits)
 890 {
 891         int highest, node;
 892
 893         highest = numbits;
 894         node = -1;
 895         while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
 896                 if (node >= numbits)
 897                         break;
 898
 899                 highest = node;
 900         }
 901
 902         return highest;
 903 }
 904
 905 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
 906 {
 907         int i, ret, highest_node, change = 0;
 908         unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
 909         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
 910         struct o2hb_bio_wait_ctxt write_wc;
 911
 912         ret = o2nm_configured_node_map(configured_nodes,
 913                                        sizeof(configured_nodes));
 914         if (ret) {
 915                 mlog_errno(ret);
 916                 return ret;
 917         }
 918
 919         /*
 920          * If a node is not configured but is in the livemap, we still need
 921          * to read the slot so as to be able to remove it from the livemap.
 922          */
 923         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
 924         i = -1;
 925         while ((i = find_next_bit(live_node_bitmap,
 926                                   O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
 927                 set_bit(i, configured_nodes);
 928         }
 929
 930         highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
 931         if (highest_node >= O2NM_MAX_NODES) {
 932                 mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n");
 933                 return -EINVAL;
 934         }
 935
 936         /* No sense in reading the slots of nodes that don't exist
 937          * yet. Of course, if the node definitions have holes in them
 938          * then we're reading an empty slot anyway... Consider this
 939          * best-effort. */
 940         ret = o2hb_read_slots(reg, highest_node + 1);
 941         if (ret < 0) {
 942                 mlog_errno(ret);
 943                 return ret;
 944         }
 945
 946         /* With an up to date view of the slots, we can check that no
 947          * other node has been improperly configured to heartbeat in
 948          * our slot. */
 949         if (!o2hb_check_last_timestamp(reg))
 950                 mlog(ML_ERROR, "Device \"%s\": another node is heartbeating "
 951                      "in our slot!\n", reg->hr_dev_name);
 952
 953         /* fill in the proper info for our next heartbeat */
 954         o2hb_prepare_block(reg, reg->hr_generation);
 955
 956         /* And fire off the write. Note that we don't wait on this I/O
 957          * until later. */
 958         ret = o2hb_issue_node_write(reg, &write_wc);
 959         if (ret < 0) {
 960                 mlog_errno(ret);
 961                 return ret;
 962         }
 963
 964         i = -1;
 965         while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
 966
 967                 change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
 968         }
 969
 970         /*
 971          * We have to be sure we've advertised ourselves on disk
 972          * before we can go to steady state.  This ensures that
 973          * people we find in our steady state have seen us.
 974          */
 975         o2hb_wait_on_io(reg, &write_wc);
 976         if (write_wc.wc_error) {
 977                 /* Do not re-arm the write timeout on I/O error - we
 978                  * can't be sure that the new block ever made it to
 979                  * disk */
 980                 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
 981                      write_wc.wc_error, reg->hr_dev_name);
 982                 return write_wc.wc_error;
 983         }
 984
 985         o2hb_arm_write_timeout(reg);
 986
 987         /* let the person who launched us know when things are steady */
 988         if (!change && (atomic_read(&reg->hr_steady_iterations) != 0)) {
 989                 if (atomic_dec_and_test(&reg->hr_steady_iterations))
 990                         wake_up(&o2hb_steady_queue);
 991         }
 992
 993         return 0;
 994 }
 995
 996 /* Subtract b from a, storing the result in a. a *must* have a larger
 997  * value than b. */
 998 static void o2hb_tv_subtract(struct timeval *a,
 999                              struct timeval *b)
1000 {
1001         /* just return 0 when a is after b */
1002         if (a->tv_sec < b->tv_sec ||
1003             (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1004                 a->tv_sec = 0;
1005                 a->tv_usec = 0;
1006                 return;
1007         }
1008
1009         a->tv_sec -= b->tv_sec;
1010         a->tv_usec -= b->tv_usec;
1011         while ( a->tv_usec < 0 ) {
1012                 a->tv_sec--;
1013                 a->tv_usec += 1000000;
1014         }
1015 }
1016
1017 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1018                                        struct timeval *end)
1019 {
1020         struct timeval res = *end;
1021
1022         o2hb_tv_subtract(&res, start);
1023
1024         return res.tv_sec * 1000 + res.tv_usec / 1000;
1025 }
1026
1027 /*
1028  * we ride the region ref that the region dir holds.  before the region
1029  * dir is removed and drops it ref it will wait to tear down this
1030  * thread.
1031  */
1032 static int o2hb_thread(void *data)
1033 {
1034         int i, ret;
1035         struct o2hb_region *reg = data;
1036         struct o2hb_bio_wait_ctxt write_wc;
1037         struct timeval before_hb, after_hb;
1038         unsigned int elapsed_msec;
1039
1040         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1041
1042         set_user_nice(current, -20);
1043
1044         while (!kthread_should_stop() && !reg->hr_unclean_stop) {
1045                 /* We track the time spent inside
1046                  * o2hb_do_disk_heartbeat so that we avoid more than
1047                  * hr_timeout_ms between disk writes. On busy systems
1048                  * this should result in a heartbeat which is less
1049                  * likely to time itself out. */
1050                 do_gettimeofday(&before_hb);
1051
1052                 i = 0;
1053                 do {
1054                         ret = o2hb_do_disk_heartbeat(reg);
1055                 } while (ret && ++i < 2);
1056
1057                 do_gettimeofday(&after_hb);
1058                 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1059
1060                 mlog(ML_HEARTBEAT,
1061                      "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
1062                      before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1063                      after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1064                      elapsed_msec);
1065
1066                 if (elapsed_msec < reg->hr_timeout_ms) {
1067                         /* the kthread api has blocked signals for us so no
1068                          * need to record the return value. */
1069                         msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1070                 }
1071         }
1072
1073         o2hb_disarm_write_timeout(reg);
1074
1075         /* unclean stop is only used in very bad situation */
1076         for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1077                 o2hb_shutdown_slot(&reg->hr_slots[i]);
1078
1079         /* Explicit down notification - avoid forcing the other nodes
1080          * to timeout on this region when we could just as easily
1081          * write a clear generation - thus indicating to them that
1082          * this node has left this region.
1083          *
1084          * XXX: Should we skip this on unclean_stop? */
1085         o2hb_prepare_block(reg, 0);
1086         ret = o2hb_issue_node_write(reg, &write_wc);
1087         if (ret == 0) {
1088                 o2hb_wait_on_io(reg, &write_wc);
1089         } else {
1090                 mlog_errno(ret);
1091         }
1092
1093         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n");
1094
1095         return 0;
1096 }
1097
1098 #ifdef CONFIG_DEBUG_FS
1099 static int o2hb_debug_open(struct inode *inode, struct file *file)
1100 {
1101         struct o2hb_debug_buf *db = inode->i_private;
1102         struct o2hb_region *reg;
1103         unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1104         char *buf = NULL;
1105         int i = -1;
1106         int out = 0;
1107
1108         /* max_nodes should be the largest bitmap we pass here */
1109         BUG_ON(sizeof(map) < db->db_size);
1110
1111         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1112         if (!buf)
1113                 goto bail;
1114
1115         switch (db->db_type) {
1116         case O2HB_DB_TYPE_LIVENODES:
1117         case O2HB_DB_TYPE_LIVEREGIONS:
1118         case O2HB_DB_TYPE_QUORUMREGIONS:
1119         case O2HB_DB_TYPE_FAILEDREGIONS:
1120                 spin_lock(&o2hb_live_lock);
1121                 memcpy(map, db->db_data, db->db_size);
1122                 spin_unlock(&o2hb_live_lock);
1123                 break;
1124
1125         case O2HB_DB_TYPE_REGION_LIVENODES:
1126                 spin_lock(&o2hb_live_lock);
1127                 reg = (struct o2hb_region *)db->db_data;
1128                 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1129                 spin_unlock(&o2hb_live_lock);
1130                 break;
1131
1132         case O2HB_DB_TYPE_REGION_NUMBER:
1133                 reg = (struct o2hb_region *)db->db_data;
1134                 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1135                                 reg->hr_region_num);
1136                 goto done;
1137
1138         case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1139                 reg = (struct o2hb_region *)db->db_data;
1140                 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1141                                 jiffies_to_msecs(jiffies -
1142                                                  reg->hr_last_timeout_start));
1143                 goto done;
1144
1145         default:
1146                 goto done;
1147         }
1148
1149         while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1150                 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1151         out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1152
1153 done:
1154         i_size_write(inode, out);
1155
1156         file->private_data = buf;
1157
1158         return 0;
1159 bail:
1160         return -ENOMEM;
1161 }
1162
1163 static int o2hb_debug_release(struct inode *inode, struct file *file)
1164 {
1165         kfree(file->private_data);
1166         return 0;
1167 }
1168
1169 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1170                                  size_t nbytes, loff_t *ppos)
1171 {
1172         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1173                                        i_size_read(file->f_mapping->host));
1174 }
1175 #else
1176 static int o2hb_debug_open(struct inode *inode, struct file *file)
1177 {
1178         return 0;
1179 }
1180 static int o2hb_debug_release(struct inode *inode, struct file *file)
1181 {
1182         return 0;
1183 }
1184 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1185                                size_t nbytes, loff_t *ppos)
1186 {
1187         return 0;
1188 }
1189 #endif  /* CONFIG_DEBUG_FS */
1190
1191 static const struct file_operations o2hb_debug_fops = {
1192         .open =         o2hb_debug_open,
1193         .release =      o2hb_debug_release,
1194         .read =         o2hb_debug_read,
1195         .llseek =       generic_file_llseek,
1196 };
1197
1198 void o2hb_exit(void)
1199 {
1200         kfree(o2hb_db_livenodes);
1201         kfree(o2hb_db_liveregions);
1202         kfree(o2hb_db_quorumregions);
1203         kfree(o2hb_db_failedregions);
1204         debugfs_remove(o2hb_debug_failedregions);
1205         debugfs_remove(o2hb_debug_quorumregions);
1206         debugfs_remove(o2hb_debug_liveregions);
1207         debugfs_remove(o2hb_debug_livenodes);
1208         debugfs_remove(o2hb_debug_dir);
1209 }
1210
1211 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1212                                         struct o2hb_debug_buf **db, int db_len,
1213                                         int type, int size, int len, void *data)
1214 {
1215         *db = kmalloc(db_len, GFP_KERNEL);
1216         if (!*db)
1217                 return NULL;
1218
1219         (*db)->db_type = type;
1220         (*db)->db_size = size;
1221         (*db)->db_len = len;
1222         (*db)->db_data = data;
1223
1224         return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1225                                    &o2hb_debug_fops);
1226 }
1227
1228 static int o2hb_debug_init(void)
1229 {
1230         int ret = -ENOMEM;
1231
1232         o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1233         if (!o2hb_debug_dir) {
1234                 mlog_errno(ret);
1235                 goto bail;
1236         }
1237
1238         o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1239                                                  o2hb_debug_dir,
1240                                                  &o2hb_db_livenodes,
1241                                                  sizeof(*o2hb_db_livenodes),
1242                                                  O2HB_DB_TYPE_LIVENODES,
1243                                                  sizeof(o2hb_live_node_bitmap),
1244                                                  O2NM_MAX_NODES,
1245                                                  o2hb_live_node_bitmap);
1246         if (!o2hb_debug_livenodes) {
1247                 mlog_errno(ret);
1248                 goto bail;
1249         }
1250
1251         o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1252                                                    o2hb_debug_dir,
1253                                                    &o2hb_db_liveregions,
1254                                                    sizeof(*o2hb_db_liveregions),
1255                                                    O2HB_DB_TYPE_LIVEREGIONS,
1256                                                    sizeof(o2hb_live_region_bitmap),
1257                                                    O2NM_MAX_REGIONS,
1258                                                    o2hb_live_region_bitmap);
1259         if (!o2hb_debug_liveregions) {
1260                 mlog_errno(ret);
1261                 goto bail;
1262         }
1263
1264         o2hb_debug_quorumregions =
1265                         o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1266                                           o2hb_debug_dir,
1267                                           &o2hb_db_quorumregions,
1268                                           sizeof(*o2hb_db_quorumregions),
1269                                           O2HB_DB_TYPE_QUORUMREGIONS,
1270                                           sizeof(o2hb_quorum_region_bitmap),
1271                                           O2NM_MAX_REGIONS,
1272                                           o2hb_quorum_region_bitmap);
1273         if (!o2hb_debug_quorumregions) {
1274                 mlog_errno(ret);
1275                 goto bail;
1276         }
1277
1278         o2hb_debug_failedregions =
1279                         o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1280                                           o2hb_debug_dir,
1281                                           &o2hb_db_failedregions,
1282                                           sizeof(*o2hb_db_failedregions),
1283                                           O2HB_DB_TYPE_FAILEDREGIONS,
1284                                           sizeof(o2hb_failed_region_bitmap),
1285                                           O2NM_MAX_REGIONS,
1286                                           o2hb_failed_region_bitmap);
1287         if (!o2hb_debug_failedregions) {
1288                 mlog_errno(ret);
1289                 goto bail;
1290         }
1291
1292         ret = 0;
1293 bail:
1294         if (ret)
1295                 o2hb_exit();
1296
1297         return ret;
1298 }
1299
1300 int o2hb_init(void)
1301 {
1302         int i;
1303
1304         for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1305                 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1306
1307         for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1308                 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1309
1310         INIT_LIST_HEAD(&o2hb_node_events);
1311
1312         memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1313         memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1314         memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1315         memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1316         memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1317
1318         return o2hb_debug_init();
1319 }
1320
1321 /* if we're already in a callback then we're already serialized by the sem */
1322 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1323                                              unsigned bytes)
1324 {
1325         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1326
1327         memcpy(map, &o2hb_live_node_bitmap, bytes);
1328 }
1329
1330 /*
1331  * get a map of all nodes that are heartbeating in any regions
1332  */
1333 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1334 {
1335         /* callers want to serialize this map and callbacks so that they
1336          * can trust that they don't miss nodes coming to the party */
1337         down_read(&o2hb_callback_sem);
1338         spin_lock(&o2hb_live_lock);
1339         o2hb_fill_node_map_from_callback(map, bytes);
1340         spin_unlock(&o2hb_live_lock);
1341         up_read(&o2hb_callback_sem);
1342 }
1343 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1344
1345 /*
1346  * heartbeat configfs bits.  The heartbeat set is a default set under
1347  * the cluster set in nodemanager.c.
1348  */
1349
1350 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1351 {
1352         return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1353 }
1354
1355 /* drop_item only drops its ref after killing the thread, nothing should
1356  * be using the region anymore.  this has to clean up any state that
1357  * attributes might have built up. */
1358 static void o2hb_region_release(struct config_item *item)
1359 {
1360         int i;
1361         struct page *page;
1362         struct o2hb_region *reg = to_o2hb_region(item);
1363
1364         if (reg->hr_tmp_block)
1365                 kfree(reg->hr_tmp_block);
1366
1367         if (reg->hr_slot_data) {
1368                 for (i = 0; i < reg->hr_num_pages; i++) {
1369                         page = reg->hr_slot_data[i];
1370                         if (page)
1371                                 __free_page(page);
1372                 }
1373                 kfree(reg->hr_slot_data);
1374         }
1375
1376         if (reg->hr_bdev)
1377                 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1378
1379         if (reg->hr_slots)
1380                 kfree(reg->hr_slots);
1381
1382         kfree(reg->hr_db_regnum);
1383         kfree(reg->hr_db_livenodes);
1384         debugfs_remove(reg->hr_debug_livenodes);
1385         debugfs_remove(reg->hr_debug_regnum);
1386         debugfs_remove(reg->hr_debug_elapsed_time);
1387         debugfs_remove(reg->hr_debug_dir);
1388
1389         spin_lock(&o2hb_live_lock);
1390         list_del(&reg->hr_all_item);
1391         spin_unlock(&o2hb_live_lock);
1392
1393         kfree(reg);
1394 }
1395
1396 static int o2hb_read_block_input(struct o2hb_region *reg,
1397                                  const char *page,
1398                                  size_t count,
1399                                  unsigned long *ret_bytes,
1400                                  unsigned int *ret_bits)
1401 {
1402         unsigned long bytes;
1403         char *p = (char *)page;
1404
1405         bytes = simple_strtoul(p, &p, 0);
1406         if (!p || (*p && (*p != '\n')))
1407                 return -EINVAL;
1408
1409         /* Heartbeat and fs min / max block sizes are the same. */
1410         if (bytes > 4096 || bytes < 512)
1411                 return -ERANGE;
1412         if (hweight16(bytes) != 1)
1413                 return -EINVAL;
1414
1415         if (ret_bytes)
1416                 *ret_bytes = bytes;
1417         if (ret_bits)
1418                 *ret_bits = ffs(bytes) - 1;
1419
1420         return 0;
1421 }
1422
1423 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1424                                             char *page)
1425 {
1426         return sprintf(page, "%u\n", reg->hr_block_bytes);
1427 }
1428
1429 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1430                                              const char *page,
1431                                              size_t count)
1432 {
1433         int status;
1434         unsigned long block_bytes;
1435         unsigned int block_bits;
1436
1437         if (reg->hr_bdev)
1438                 return -EINVAL;
1439
1440         status = o2hb_read_block_input(reg, page, count,
1441                                        &block_bytes, &block_bits);
1442         if (status)
1443                 return status;
1444
1445         reg->hr_block_bytes = (unsigned int)block_bytes;
1446         reg->hr_block_bits = block_bits;
1447
1448         return count;
1449 }
1450
1451 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1452                                             char *page)
1453 {
1454         return sprintf(page, "%llu\n", reg->hr_start_block);
1455 }
1456
1457 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1458                                              const char *page,
1459                                              size_t count)
1460 {
1461         unsigned long long tmp;
1462         char *p = (char *)page;
1463
1464         if (reg->hr_bdev)
1465                 return -EINVAL;
1466
1467         tmp = simple_strtoull(p, &p, 0);
1468         if (!p || (*p && (*p != '\n')))
1469                 return -EINVAL;
1470
1471         reg->hr_start_block = tmp;
1472
1473         return count;
1474 }
1475
1476 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1477                                        char *page)
1478 {
1479         return sprintf(page, "%d\n", reg->hr_blocks);
1480 }
1481
1482 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1483                                         const char *page,
1484                                         size_t count)
1485 {
1486         unsigned long tmp;
1487         char *p = (char *)page;
1488
1489         if (reg->hr_bdev)
1490                 return -EINVAL;
1491
1492         tmp = simple_strtoul(p, &p, 0);
1493         if (!p || (*p && (*p != '\n')))
1494                 return -EINVAL;
1495
1496         if (tmp > O2NM_MAX_NODES || tmp == 0)
1497                 return -ERANGE;
1498
1499         reg->hr_blocks = (unsigned int)tmp;
1500
1501         return count;
1502 }
1503
1504 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1505                                     char *page)
1506 {
1507         unsigned int ret = 0;
1508
1509         if (reg->hr_bdev)
1510                 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1511
1512         return ret;
1513 }
1514
1515 static void o2hb_init_region_params(struct o2hb_region *reg)
1516 {
1517         reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1518         reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1519
1520         mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1521              reg->hr_start_block, reg->hr_blocks);
1522         mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1523              reg->hr_block_bytes, reg->hr_block_bits);
1524         mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1525         mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1526 }
1527
1528 static int o2hb_map_slot_data(struct o2hb_region *reg)
1529 {
1530         int i, j;
1531         unsigned int last_slot;
1532         unsigned int spp = reg->hr_slots_per_page;
1533         struct page *page;
1534         char *raw;
1535         struct o2hb_disk_slot *slot;
1536
1537         reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1538         if (reg->hr_tmp_block == NULL) {
1539                 mlog_errno(-ENOMEM);
1540                 return -ENOMEM;
1541         }
1542
1543         reg->hr_slots = kcalloc(reg->hr_blocks,
1544                                 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1545         if (reg->hr_slots == NULL) {
1546                 mlog_errno(-ENOMEM);
1547                 return -ENOMEM;
1548         }
1549
1550         for(i = 0; i < reg->hr_blocks; i++) {
1551                 slot = &reg->hr_slots[i];
1552                 slot->ds_node_num = i;
1553                 INIT_LIST_HEAD(&slot->ds_live_item);
1554                 slot->ds_raw_block = NULL;
1555         }
1556
1557         reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1558         mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1559                            "at %u blocks per page\n",
1560              reg->hr_num_pages, reg->hr_blocks, spp);
1561
1562         reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1563                                     GFP_KERNEL);
1564         if (!reg->hr_slot_data) {
1565                 mlog_errno(-ENOMEM);
1566                 return -ENOMEM;
1567         }
1568
1569         for(i = 0; i < reg->hr_num_pages; i++) {
1570                 page = alloc_page(GFP_KERNEL);
1571                 if (!page) {
1572                         mlog_errno(-ENOMEM);
1573                         return -ENOMEM;
1574                 }
1575
1576                 reg->hr_slot_data[i] = page;
1577
1578                 last_slot = i * spp;
1579                 raw = page_address(page);
1580                 for (j = 0;
1581                      (j < spp) && ((j + last_slot) < reg->hr_blocks);
1582                      j++) {
1583                         BUG_ON((j + last_slot) >= reg->hr_blocks);
1584
1585                         slot = &reg->hr_slots[j + last_slot];
1586                         slot->ds_raw_block =
1587                                 (struct o2hb_disk_heartbeat_block *) raw;
1588
1589                         raw += reg->hr_block_bytes;
1590                 }
1591         }
1592
1593         return 0;
1594 }
1595
1596 /* Read in all the slots available and populate the tracking
1597  * structures so that we can start with a baseline idea of what's
1598  * there. */
1599 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1600 {
1601         int ret, i;
1602         struct o2hb_disk_slot *slot;
1603         struct o2hb_disk_heartbeat_block *hb_block;
1604
1605         mlog_entry_void();
1606
1607         ret = o2hb_read_slots(reg, reg->hr_blocks);
1608         if (ret) {
1609                 mlog_errno(ret);
1610                 goto out;
1611         }
1612
1613         /* We only want to get an idea of the values initially in each
1614          * slot, so we do no verification - o2hb_check_slot will
1615          * actually determine if each configured slot is valid and
1616          * whether any values have changed. */
1617         for(i = 0; i < reg->hr_blocks; i++) {
1618                 slot = &reg->hr_slots[i];
1619                 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1620
1621                 /* Only fill the values that o2hb_check_slot uses to
1622                  * determine changing slots */
1623                 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1624                 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1625         }
1626
1627 out:
1628         mlog_exit(ret);
1629         return ret;
1630 }
1631
1632 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1633 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1634                                      const char *page,
1635                                      size_t count)
1636 {
1637         struct task_struct *hb_task;
1638         long fd;
1639         int sectsize;
1640         char *p = (char *)page;
1641         struct file *filp = NULL;
1642         struct inode *inode = NULL;
1643         ssize_t ret = -EINVAL;
1644
1645         if (reg->hr_bdev)
1646                 goto out;
1647
1648         /* We can't heartbeat without having had our node number
1649          * configured yet. */
1650         if (o2nm_this_node() == O2NM_MAX_NODES)
1651                 goto out;
1652
1653         fd = simple_strtol(p, &p, 0);
1654         if (!p || (*p && (*p != '\n')))
1655                 goto out;
1656
1657         if (fd < 0 || fd >= INT_MAX)
1658                 goto out;
1659
1660         filp = fget(fd);
1661         if (filp == NULL)
1662                 goto out;
1663
1664         if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1665             reg->hr_block_bytes == 0)
1666                 goto out;
1667
1668         inode = igrab(filp->f_mapping->host);
1669         if (inode == NULL)
1670                 goto out;
1671
1672         if (!S_ISBLK(inode->i_mode))
1673                 goto out;
1674
1675         reg->hr_bdev = I_BDEV(filp->f_mapping->host);
1676         ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ);
1677         if (ret) {
1678                 reg->hr_bdev = NULL;
1679                 goto out;
1680         }
1681         inode = NULL;
1682
1683         bdevname(reg->hr_bdev, reg->hr_dev_name);
1684
1685         sectsize = bdev_logical_block_size(reg->hr_bdev);
1686         if (sectsize != reg->hr_block_bytes) {
1687                 mlog(ML_ERROR,
1688                      "blocksize %u incorrect for device, expected %d",
1689                      reg->hr_block_bytes, sectsize);
1690                 ret = -EINVAL;
1691                 goto out;
1692         }
1693
1694         o2hb_init_region_params(reg);
1695
1696         /* Generation of zero is invalid */
1697         do {
1698                 get_random_bytes(&reg->hr_generation,
1699                                  sizeof(reg->hr_generation));
1700         } while (reg->hr_generation == 0);
1701
1702         ret = o2hb_map_slot_data(reg);
1703         if (ret) {
1704                 mlog_errno(ret);
1705                 goto out;
1706         }
1707
1708         ret = o2hb_populate_slot_data(reg);
1709         if (ret) {
1710                 mlog_errno(ret);
1711                 goto out;
1712         }
1713
1714         INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1715
1716         /*
1717          * A node is considered live after it has beat LIVE_THRESHOLD
1718          * times.  We're not steady until we've given them a chance
1719          * _after_ our first read.
1720          */
1721         atomic_set(&reg->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1);
1722
1723         hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1724                               reg->hr_item.ci_name);
1725         if (IS_ERR(hb_task)) {
1726                 ret = PTR_ERR(hb_task);
1727                 mlog_errno(ret);
1728                 goto out;
1729         }
1730
1731         spin_lock(&o2hb_live_lock);
1732         reg->hr_task = hb_task;
1733         spin_unlock(&o2hb_live_lock);
1734
1735         ret = wait_event_interruptible(o2hb_steady_queue,
1736                                 atomic_read(&reg->hr_steady_iterations) == 0);
1737         if (ret) {
1738                 /* We got interrupted (hello ptrace!).  Clean up */
1739                 spin_lock(&o2hb_live_lock);
1740                 hb_task = reg->hr_task;
1741                 reg->hr_task = NULL;
1742                 spin_unlock(&o2hb_live_lock);
1743
1744                 if (hb_task)
1745                         kthread_stop(hb_task);
1746                 goto out;
1747         }
1748
1749         /* Ok, we were woken.  Make sure it wasn't by drop_item() */
1750         spin_lock(&o2hb_live_lock);
1751         hb_task = reg->hr_task;
1752         if (o2hb_global_heartbeat_active())
1753                 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1754         spin_unlock(&o2hb_live_lock);
1755
1756         if (hb_task)
1757                 ret = count;
1758         else
1759                 ret = -EIO;
1760
1761         if (hb_task && o2hb_global_heartbeat_active())
1762                 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s\n",
1763                        config_item_name(&reg->hr_item));
1764
1765 out:
1766         if (filp)
1767                 fput(filp);
1768         if (inode)
1769                 iput(inode);
1770         if (ret < 0) {
1771                 if (reg->hr_bdev) {
1772                         blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1773                         reg->hr_bdev = NULL;
1774                 }
1775         }
1776         return ret;
1777 }
1778
1779 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1780                                       char *page)
1781 {
1782         pid_t pid = 0;
1783
1784         spin_lock(&o2hb_live_lock);
1785         if (reg->hr_task)
1786                 pid = task_pid_nr(reg->hr_task);
1787         spin_unlock(&o2hb_live_lock);
1788
1789         if (!pid)
1790                 return 0;
1791
1792         return sprintf(page, "%u\n", pid);
1793 }
1794
1795 struct o2hb_region_attribute {
1796         struct configfs_attribute attr;
1797         ssize_t (*show)(struct o2hb_region *, char *);
1798         ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1799 };
1800
1801 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1802         .attr   = { .ca_owner = THIS_MODULE,
1803                     .ca_name = "block_bytes",
1804                     .ca_mode = S_IRUGO | S_IWUSR },
1805         .show   = o2hb_region_block_bytes_read,
1806         .store  = o2hb_region_block_bytes_write,
1807 };
1808
1809 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1810         .attr   = { .ca_owner = THIS_MODULE,
1811                     .ca_name = "start_block",
1812                     .ca_mode = S_IRUGO | S_IWUSR },
1813         .show   = o2hb_region_start_block_read,
1814         .store  = o2hb_region_start_block_write,
1815 };
1816
1817 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1818         .attr   = { .ca_owner = THIS_MODULE,
1819                     .ca_name = "blocks",
1820                     .ca_mode = S_IRUGO | S_IWUSR },
1821         .show   = o2hb_region_blocks_read,
1822         .store  = o2hb_region_blocks_write,
1823 };
1824
1825 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1826         .attr   = { .ca_owner = THIS_MODULE,
1827                     .ca_name = "dev",
1828                     .ca_mode = S_IRUGO | S_IWUSR },
1829         .show   = o2hb_region_dev_read,
1830         .store  = o2hb_region_dev_write,
1831 };
1832
1833 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1834        .attr   = { .ca_owner = THIS_MODULE,
1835                    .ca_name = "pid",
1836                    .ca_mode = S_IRUGO | S_IRUSR },
1837        .show   = o2hb_region_pid_read,
1838 };
1839
1840 static struct configfs_attribute *o2hb_region_attrs[] = {
1841         &o2hb_region_attr_block_bytes.attr,
1842         &o2hb_region_attr_start_block.attr,
1843         &o2hb_region_attr_blocks.attr,
1844         &o2hb_region_attr_dev.attr,
1845         &o2hb_region_attr_pid.attr,
1846         NULL,
1847 };
1848
1849 static ssize_t o2hb_region_show(struct config_item *item,
1850                                 struct configfs_attribute *attr,
1851                                 char *page)
1852 {
1853         struct o2hb_region *reg = to_o2hb_region(item);
1854         struct o2hb_region_attribute *o2hb_region_attr =
1855                 container_of(attr, struct o2hb_region_attribute, attr);
1856         ssize_t ret = 0;
1857
1858         if (o2hb_region_attr->show)
1859                 ret = o2hb_region_attr->show(reg, page);
1860         return ret;
1861 }
1862
1863 static ssize_t o2hb_region_store(struct config_item *item,
1864                                  struct configfs_attribute *attr,
1865                                  const char *page, size_t count)
1866 {
1867         struct o2hb_region *reg = to_o2hb_region(item);
1868         struct o2hb_region_attribute *o2hb_region_attr =
1869                 container_of(attr, struct o2hb_region_attribute, attr);
1870         ssize_t ret = -EINVAL;
1871
1872         if (o2hb_region_attr->store)
1873                 ret = o2hb_region_attr->store(reg, page, count);
1874         return ret;
1875 }
1876
1877 static struct configfs_item_operations o2hb_region_item_ops = {
1878         .release                = o2hb_region_release,
1879         .show_attribute         = o2hb_region_show,
1880         .store_attribute        = o2hb_region_store,
1881 };
1882
1883 static struct config_item_type o2hb_region_type = {
1884         .ct_item_ops    = &o2hb_region_item_ops,
1885         .ct_attrs       = o2hb_region_attrs,
1886         .ct_owner       = THIS_MODULE,
1887 };
1888
1889 /* heartbeat set */
1890
1891 struct o2hb_heartbeat_group {
1892         struct config_group hs_group;
1893         /* some stuff? */
1894 };
1895
1896 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1897 {
1898         return group ?
1899                 container_of(group, struct o2hb_heartbeat_group, hs_group)
1900                 : NULL;
1901 }
1902
1903 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
1904 {
1905         int ret = -ENOMEM;
1906
1907         reg->hr_debug_dir =
1908                 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
1909         if (!reg->hr_debug_dir) {
1910                 mlog_errno(ret);
1911                 goto bail;
1912         }
1913
1914         reg->hr_debug_livenodes =
1915                         o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1916                                           reg->hr_debug_dir,
1917                                           &(reg->hr_db_livenodes),
1918                                           sizeof(*(reg->hr_db_livenodes)),
1919                                           O2HB_DB_TYPE_REGION_LIVENODES,
1920                                           sizeof(reg->hr_live_node_bitmap),
1921                                           O2NM_MAX_NODES, reg);
1922         if (!reg->hr_debug_livenodes) {
1923                 mlog_errno(ret);
1924                 goto bail;
1925         }
1926
1927         reg->hr_debug_regnum =
1928                         o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
1929                                           reg->hr_debug_dir,
1930                                           &(reg->hr_db_regnum),
1931                                           sizeof(*(reg->hr_db_regnum)),
1932                                           O2HB_DB_TYPE_REGION_NUMBER,
1933                                           0, O2NM_MAX_NODES, reg);
1934         if (!reg->hr_debug_regnum) {
1935                 mlog_errno(ret);
1936                 goto bail;
1937         }
1938
1939         reg->hr_debug_elapsed_time =
1940                         o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
1941                                           reg->hr_debug_dir,
1942                                           &(reg->hr_db_elapsed_time),
1943                                           sizeof(*(reg->hr_db_elapsed_time)),
1944                                           O2HB_DB_TYPE_REGION_ELAPSED_TIME,
1945                                           0, 0, reg);
1946         if (!reg->hr_debug_elapsed_time) {
1947                 mlog_errno(ret);
1948                 goto bail;
1949         }
1950
1951         ret = 0;
1952 bail:
1953         return ret;
1954 }
1955
1956 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
1957                                                           const char *name)
1958 {
1959         struct o2hb_region *reg = NULL;
1960         int ret;
1961
1962         reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
1963         if (reg == NULL)
1964                 return ERR_PTR(-ENOMEM);
1965
1966         if (strlen(name) > O2HB_MAX_REGION_NAME_LEN)
1967                 return ERR_PTR(-ENAMETOOLONG);
1968
1969         spin_lock(&o2hb_live_lock);
1970         reg->hr_region_num = 0;
1971         if (o2hb_global_heartbeat_active()) {
1972                 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
1973                                                          O2NM_MAX_REGIONS);
1974                 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
1975                         spin_unlock(&o2hb_live_lock);
1976                         return ERR_PTR(-EFBIG);
1977                 }
1978                 set_bit(reg->hr_region_num, o2hb_region_bitmap);
1979         }
1980         list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
1981         spin_unlock(&o2hb_live_lock);
1982
1983         config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
1984
1985         ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
1986         if (ret) {
1987                 config_item_put(&reg->hr_item);
1988                 return ERR_PTR(ret);
1989         }
1990
1991         return &reg->hr_item;
1992 }
1993
1994 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
1995                                            struct config_item *item)
1996 {
1997         struct task_struct *hb_task;
1998         struct o2hb_region *reg = to_o2hb_region(item);
1999
2000         /* stop the thread when the user removes the region dir */
2001         spin_lock(&o2hb_live_lock);
2002         if (o2hb_global_heartbeat_active()) {
2003                 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2004                 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2005         }
2006         hb_task = reg->hr_task;
2007         reg->hr_task = NULL;
2008         spin_unlock(&o2hb_live_lock);
2009
2010         if (hb_task)
2011                 kthread_stop(hb_task);
2012
2013         /*
2014          * If we're racing a dev_write(), we need to wake them.  They will
2015          * check reg->hr_task
2016          */
2017         if (atomic_read(&reg->hr_steady_iterations) != 0) {
2018                 atomic_set(&reg->hr_steady_iterations, 0);
2019                 wake_up(&o2hb_steady_queue);
2020         }
2021
2022         if (o2hb_global_heartbeat_active())
2023                 printk(KERN_NOTICE "o2hb: Heartbeat stopped on region %s\n",
2024                        config_item_name(&reg->hr_item));
2025         config_item_put(item);
2026 }
2027
2028 struct o2hb_heartbeat_group_attribute {
2029         struct configfs_attribute attr;
2030         ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2031         ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2032 };
2033
2034 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2035                                          struct configfs_attribute *attr,
2036                                          char *page)
2037 {
2038         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2039         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2040                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2041         ssize_t ret = 0;
2042
2043         if (o2hb_heartbeat_group_attr->show)
2044                 ret = o2hb_heartbeat_group_attr->show(reg, page);
2045         return ret;
2046 }
2047
2048 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2049                                           struct configfs_attribute *attr,
2050                                           const char *page, size_t count)
2051 {
2052         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2053         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2054                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2055         ssize_t ret = -EINVAL;
2056
2057         if (o2hb_heartbeat_group_attr->store)
2058                 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2059         return ret;
2060 }
2061
2062 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2063                                                      char *page)
2064 {
2065         return sprintf(page, "%u\n", o2hb_dead_threshold);
2066 }
2067
2068 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2069                                                     const char *page,
2070                                                     size_t count)
2071 {
2072         unsigned long tmp;
2073         char *p = (char *)page;
2074
2075         tmp = simple_strtoul(p, &p, 10);
2076         if (!p || (*p && (*p != '\n')))
2077                 return -EINVAL;
2078
2079         /* this will validate ranges for us. */
2080         o2hb_dead_threshold_set((unsigned int) tmp);
2081
2082         return count;
2083 }
2084
2085 static
2086 ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2087                                        char *page)
2088 {
2089         return sprintf(page, "%s\n",
2090                        o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2091 }
2092
2093 static
2094 ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2095                                         const char *page, size_t count)
2096 {
2097         unsigned int i;
2098         int ret;
2099         size_t len;
2100
2101         len = (page[count - 1] == '\n') ? count - 1 : count;
2102         if (!len)
2103                 return -EINVAL;
2104
2105         for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2106                 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2107                         continue;
2108
2109                 ret = o2hb_global_hearbeat_mode_set(i);
2110                 if (!ret)
2111                         printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2112                                o2hb_heartbeat_mode_desc[i]);
2113                 return count;
2114         }
2115
2116         return -EINVAL;
2117
2118 }
2119
2120 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2121         .attr   = { .ca_owner = THIS_MODULE,
2122                     .ca_name = "dead_threshold",
2123                     .ca_mode = S_IRUGO | S_IWUSR },
2124         .show   = o2hb_heartbeat_group_threshold_show,
2125         .store  = o2hb_heartbeat_group_threshold_store,
2126 };
2127
2128 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2129         .attr   = { .ca_owner = THIS_MODULE,
2130                 .ca_name = "mode",
2131                 .ca_mode = S_IRUGO | S_IWUSR },
2132         .show   = o2hb_heartbeat_group_mode_show,
2133         .store  = o2hb_heartbeat_group_mode_store,
2134 };
2135
2136 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2137         &o2hb_heartbeat_group_attr_threshold.attr,
2138         &o2hb_heartbeat_group_attr_mode.attr,
2139         NULL,
2140 };
2141
2142 static struct configfs_item_operations o2hb_hearbeat_group_item_ops = {
2143         .show_attribute         = o2hb_heartbeat_group_show,
2144         .store_attribute        = o2hb_heartbeat_group_store,
2145 };
2146
2147 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2148         .make_item      = o2hb_heartbeat_group_make_item,
2149         .drop_item      = o2hb_heartbeat_group_drop_item,
2150 };
2151
2152 static struct config_item_type o2hb_heartbeat_group_type = {
2153         .ct_group_ops   = &o2hb_heartbeat_group_group_ops,
2154         .ct_item_ops    = &o2hb_hearbeat_group_item_ops,
2155         .ct_attrs       = o2hb_heartbeat_group_attrs,
2156         .ct_owner       = THIS_MODULE,
2157 };
2158
2159 /* this is just here to avoid touching group in heartbeat.h which the
2160  * entire damn world #includes */
2161 struct config_group *o2hb_alloc_hb_set(void)
2162 {
2163         struct o2hb_heartbeat_group *hs = NULL;
2164         struct config_group *ret = NULL;
2165
2166         hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2167         if (hs == NULL)
2168                 goto out;
2169
2170         config_group_init_type_name(&hs->hs_group, "heartbeat",
2171                                     &o2hb_heartbeat_group_type);
2172
2173         ret = &hs->hs_group;
2174 out:
2175         if (ret == NULL)
2176                 kfree(hs);
2177         return ret;
2178 }
2179
2180 void o2hb_free_hb_set(struct config_group *group)
2181 {
2182         struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2183         kfree(hs);
2184 }
2185
2186 /* hb callback registration and issueing */
2187
2188 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2189 {
2190         if (type == O2HB_NUM_CB)
2191                 return ERR_PTR(-EINVAL);
2192
2193         return &o2hb_callbacks[type];
2194 }
2195
2196 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2197                          enum o2hb_callback_type type,
2198                          o2hb_cb_func *func,
2199                          void *data,
2200                          int priority)
2201 {
2202         INIT_LIST_HEAD(&hc->hc_item);
2203         hc->hc_func = func;
2204         hc->hc_data = data;
2205         hc->hc_priority = priority;
2206         hc->hc_type = type;
2207         hc->hc_magic = O2HB_CB_MAGIC;
2208 }
2209 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2210
2211 static struct o2hb_region *o2hb_find_region(const char *region_uuid)
2212 {
2213         struct o2hb_region *p, *reg = NULL;
2214
2215         assert_spin_locked(&o2hb_live_lock);
2216
2217         list_for_each_entry(p, &o2hb_all_regions, hr_all_item) {
2218                 if (!strcmp(region_uuid, config_item_name(&p->hr_item))) {
2219                         reg = p;
2220                         break;
2221                 }
2222         }
2223
2224         return reg;
2225 }
2226
2227 static int o2hb_region_get(const char *region_uuid)
2228 {
2229         int ret = 0;
2230         struct o2hb_region *reg;
2231
2232         spin_lock(&o2hb_live_lock);
2233
2234         reg = o2hb_find_region(region_uuid);
2235         if (!reg)
2236                 ret = -ENOENT;
2237         spin_unlock(&o2hb_live_lock);
2238
2239         if (ret)
2240                 goto out;
2241
2242         ret = o2nm_depend_this_node();
2243         if (ret)
2244                 goto out;
2245
2246         ret = o2nm_depend_item(&reg->hr_item);
2247         if (ret)
2248                 o2nm_undepend_this_node();
2249
2250 out:
2251         return ret;
2252 }
2253
2254 static void o2hb_region_put(const char *region_uuid)
2255 {
2256         struct o2hb_region *reg;
2257
2258         spin_lock(&o2hb_live_lock);
2259
2260         reg = o2hb_find_region(region_uuid);
2261
2262         spin_unlock(&o2hb_live_lock);
2263
2264         if (reg) {
2265                 o2nm_undepend_item(&reg->hr_item);
2266                 o2nm_undepend_this_node();
2267         }
2268 }
2269
2270 int o2hb_register_callback(const char *region_uuid,
2271                            struct o2hb_callback_func *hc)
2272 {
2273         struct o2hb_callback_func *tmp;
2274         struct list_head *iter;
2275         struct o2hb_callback *hbcall;
2276         int ret;
2277
2278         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2279         BUG_ON(!list_empty(&hc->hc_item));
2280
2281         hbcall = hbcall_from_type(hc->hc_type);
2282         if (IS_ERR(hbcall)) {
2283                 ret = PTR_ERR(hbcall);
2284                 goto out;
2285         }
2286
2287         if (region_uuid) {
2288                 ret = o2hb_region_get(region_uuid);
2289                 if (ret)
2290                         goto out;
2291         }
2292
2293         down_write(&o2hb_callback_sem);
2294
2295         list_for_each(iter, &hbcall->list) {
2296                 tmp = list_entry(iter, struct o2hb_callback_func, hc_item);
2297                 if (hc->hc_priority < tmp->hc_priority) {
2298                         list_add_tail(&hc->hc_item, iter);
2299                         break;
2300                 }
2301         }
2302         if (list_empty(&hc->hc_item))
2303                 list_add_tail(&hc->hc_item, &hbcall->list);
2304
2305         up_write(&o2hb_callback_sem);
2306         ret = 0;
2307 out:
2308         mlog(ML_HEARTBEAT, "returning %d on behalf of %p for funcs %p\n",
2309              ret, __builtin_return_address(0), hc);
2310         return ret;
2311 }
2312 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2313
2314 void o2hb_unregister_callback(const char *region_uuid,
2315                               struct o2hb_callback_func *hc)
2316 {
2317         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2318
2319         mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n",
2320              __builtin_return_address(0), hc);
2321
2322         /* XXX Can this happen _with_ a region reference? */
2323         if (list_empty(&hc->hc_item))
2324                 return;
2325
2326         if (region_uuid)
2327                 o2hb_region_put(region_uuid);
2328
2329         down_write(&o2hb_callback_sem);
2330
2331         list_del_init(&hc->hc_item);
2332
2333         up_write(&o2hb_callback_sem);
2334 }
2335 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2336
2337 int o2hb_check_node_heartbeating(u8 node_num)
2338 {
2339         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2340
2341         o2hb_fill_node_map(testing_map, sizeof(testing_map));
2342         if (!test_bit(node_num, testing_map)) {
2343                 mlog(ML_HEARTBEAT,
2344                      "node (%u) does not have heartbeating enabled.\n",
2345                      node_num);
2346                 return 0;
2347         }
2348
2349         return 1;
2350 }
2351 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2352
2353 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2354 {
2355         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2356
2357         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2358         if (!test_bit(node_num, testing_map)) {
2359                 mlog(ML_HEARTBEAT,
2360                      "node (%u) does not have heartbeating enabled.\n",
2361                      node_num);
2362                 return 0;
2363         }
2364
2365         return 1;
2366 }
2367 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2368
2369 /* Makes sure our local node is configured with a node number, and is
2370  * heartbeating. */
2371 int o2hb_check_local_node_heartbeating(void)
2372 {
2373         u8 node_num;
2374
2375         /* if this node was set then we have networking */
2376         node_num = o2nm_this_node();
2377         if (node_num == O2NM_MAX_NODES) {
2378                 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2379                 return 0;
2380         }
2381
2382         return o2hb_check_node_heartbeating(node_num);
2383 }
2384 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2385
2386 /*
2387  * this is just a hack until we get the plumbing which flips file systems
2388  * read only and drops the hb ref instead of killing the node dead.
2389  */
2390 void o2hb_stop_all_regions(void)
2391 {
2392         struct o2hb_region *reg;
2393
2394         mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2395
2396         spin_lock(&o2hb_live_lock);
2397
2398         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2399                 reg->hr_unclean_stop = 1;
2400
2401         spin_unlock(&o2hb_live_lock);
2402 }
2403 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2404
2405 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2406 {
2407         struct o2hb_region *reg;
2408         int numregs = 0;
2409         char *p;
2410
2411         spin_lock(&o2hb_live_lock);
2412
2413         p = region_uuids;
2414         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2415                 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2416                 if (numregs < max_regions) {
2417                         memcpy(p, config_item_name(&reg->hr_item),
2418                                O2HB_MAX_REGION_NAME_LEN);
2419                         p += O2HB_MAX_REGION_NAME_LEN;
2420                 }
2421                 numregs++;
2422         }
2423
2424         spin_unlock(&o2hb_live_lock);
2425
2426         return numregs;
2427 }
2428 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2429
2430 int o2hb_global_heartbeat_active(void)
2431 {
2432         return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2433 }
2434 EXPORT_SYMBOL(o2hb_global_heartbeat_active);