1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * Copyright (C) 2004, 2005 Oracle. All rights reserved.
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.
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.
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.
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/jiffies.h>
25 #include <linux/module.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>
39 #include "heartbeat.h"
41 #include "nodemanager.h"
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.
53 static DECLARE_RWSEM(o2hb_callback_sem);
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.
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);
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
71 * - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
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)];
78 #define O2HB_DB_TYPE_LIVENODES 0
79 #define O2HB_DB_TYPE_LIVEREGIONS 1
80 #define O2HB_DB_TYPE_QUORUMREGIONS 2
81 #define O2HB_DB_TYPE_FAILEDREGIONS 3
82 #define O2HB_DB_TYPE_REGION_LIVENODES 4
83 #define O2HB_DB_TYPE_REGION_NUMBER 5
84 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME 6
85 #define O2HB_DB_TYPE_REGION_PINNED 7
86 struct o2hb_debug_buf {
93 static struct o2hb_debug_buf *o2hb_db_livenodes;
94 static struct o2hb_debug_buf *o2hb_db_liveregions;
95 static struct o2hb_debug_buf *o2hb_db_quorumregions;
96 static struct o2hb_debug_buf *o2hb_db_failedregions;
98 #define O2HB_DEBUG_DIR "o2hb"
99 #define O2HB_DEBUG_LIVENODES "livenodes"
100 #define O2HB_DEBUG_LIVEREGIONS "live_regions"
101 #define O2HB_DEBUG_QUORUMREGIONS "quorum_regions"
102 #define O2HB_DEBUG_FAILEDREGIONS "failed_regions"
103 #define O2HB_DEBUG_REGION_NUMBER "num"
104 #define O2HB_DEBUG_REGION_ELAPSED_TIME "elapsed_time_in_ms"
105 #define O2HB_DEBUG_REGION_PINNED "pinned"
107 static struct dentry *o2hb_debug_dir;
108 static struct dentry *o2hb_debug_livenodes;
109 static struct dentry *o2hb_debug_liveregions;
110 static struct dentry *o2hb_debug_quorumregions;
111 static struct dentry *o2hb_debug_failedregions;
113 static LIST_HEAD(o2hb_all_regions);
115 static struct o2hb_callback {
116 struct list_head list;
117 } o2hb_callbacks[O2HB_NUM_CB];
119 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
121 #define O2HB_DEFAULT_BLOCK_BITS 9
123 enum o2hb_heartbeat_modes {
124 O2HB_HEARTBEAT_LOCAL = 0,
125 O2HB_HEARTBEAT_GLOBAL,
126 O2HB_HEARTBEAT_NUM_MODES,
129 char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
130 "local", /* O2HB_HEARTBEAT_LOCAL */
131 "global", /* O2HB_HEARTBEAT_GLOBAL */
134 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
135 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
138 * o2hb_dependent_users tracks the number of registered callbacks that depend
139 * on heartbeat. o2net and o2dlm are two entities that register this callback.
140 * However only o2dlm depends on the heartbeat. It does not want the heartbeat
141 * to stop while a dlm domain is still active.
143 unsigned int o2hb_dependent_users;
146 * In global heartbeat mode, all regions are pinned if there are one or more
147 * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
148 * regions are unpinned if the region count exceeds the cut off or the number
149 * of dependent users falls to zero.
151 #define O2HB_PIN_CUT_OFF 3
154 * In local heartbeat mode, we assume the dlm domain name to be the same as
155 * region uuid. This is true for domains created for the file system but not
156 * necessarily true for userdlm domains. This is a known limitation.
158 * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
159 * works for both file system and userdlm domains.
161 static int o2hb_region_pin(const char *region_uuid);
162 static void o2hb_region_unpin(const char *region_uuid);
164 /* Only sets a new threshold if there are no active regions.
166 * No locking or otherwise interesting code is required for reading
167 * o2hb_dead_threshold as it can't change once regions are active and
168 * it's not interesting to anyone until then anyway. */
169 static void o2hb_dead_threshold_set(unsigned int threshold)
171 if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
172 spin_lock(&o2hb_live_lock);
173 if (list_empty(&o2hb_all_regions))
174 o2hb_dead_threshold = threshold;
175 spin_unlock(&o2hb_live_lock);
179 static int o2hb_global_hearbeat_mode_set(unsigned int hb_mode)
183 if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
184 spin_lock(&o2hb_live_lock);
185 if (list_empty(&o2hb_all_regions)) {
186 o2hb_heartbeat_mode = hb_mode;
189 spin_unlock(&o2hb_live_lock);
195 struct o2hb_node_event {
196 struct list_head hn_item;
197 enum o2hb_callback_type hn_event_type;
198 struct o2nm_node *hn_node;
202 struct o2hb_disk_slot {
203 struct o2hb_disk_heartbeat_block *ds_raw_block;
206 u64 ds_last_generation;
207 u16 ds_equal_samples;
208 u16 ds_changed_samples;
209 struct list_head ds_live_item;
212 /* each thread owns a region.. when we're asked to tear down the region
213 * we ask the thread to stop, who cleans up the region */
215 struct config_item hr_item;
217 struct list_head hr_all_item;
218 unsigned hr_unclean_stop:1,
222 /* protected by the hr_callback_sem */
223 struct task_struct *hr_task;
225 unsigned int hr_blocks;
226 unsigned long long hr_start_block;
228 unsigned int hr_block_bits;
229 unsigned int hr_block_bytes;
231 unsigned int hr_slots_per_page;
232 unsigned int hr_num_pages;
234 struct page **hr_slot_data;
235 struct block_device *hr_bdev;
236 struct o2hb_disk_slot *hr_slots;
238 /* live node map of this region */
239 unsigned long hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
240 unsigned int hr_region_num;
242 struct dentry *hr_debug_dir;
243 struct dentry *hr_debug_livenodes;
244 struct dentry *hr_debug_regnum;
245 struct dentry *hr_debug_elapsed_time;
246 struct dentry *hr_debug_pinned;
247 struct o2hb_debug_buf *hr_db_livenodes;
248 struct o2hb_debug_buf *hr_db_regnum;
249 struct o2hb_debug_buf *hr_db_elapsed_time;
250 struct o2hb_debug_buf *hr_db_pinned;
252 /* let the person setting up hb wait for it to return until it
253 * has reached a 'steady' state. This will be fixed when we have
254 * a more complete api that doesn't lead to this sort of fragility. */
255 atomic_t hr_steady_iterations;
257 char hr_dev_name[BDEVNAME_SIZE];
259 unsigned int hr_timeout_ms;
261 /* randomized as the region goes up and down so that a node
262 * recognizes a node going up and down in one iteration */
265 struct delayed_work hr_write_timeout_work;
266 unsigned long hr_last_timeout_start;
268 /* Used during o2hb_check_slot to hold a copy of the block
269 * being checked because we temporarily have to zero out the
271 struct o2hb_disk_heartbeat_block *hr_tmp_block;
274 struct o2hb_bio_wait_ctxt {
275 atomic_t wc_num_reqs;
276 struct completion wc_io_complete;
280 static int o2hb_pop_count(void *map, int count)
284 while ((i = find_next_bit(map, count, i + 1)) < count)
289 static void o2hb_write_timeout(struct work_struct *work)
293 struct o2hb_region *reg =
294 container_of(work, struct o2hb_region,
295 hr_write_timeout_work.work);
297 mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
298 "milliseconds\n", reg->hr_dev_name,
299 jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
301 if (o2hb_global_heartbeat_active()) {
302 spin_lock_irqsave(&o2hb_live_lock, flags);
303 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
304 set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
305 failed = o2hb_pop_count(&o2hb_failed_region_bitmap,
307 quorum = o2hb_pop_count(&o2hb_quorum_region_bitmap,
309 spin_unlock_irqrestore(&o2hb_live_lock, flags);
311 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
315 * Fence if the number of failed regions >= half the number
318 if ((failed << 1) < quorum)
322 o2quo_disk_timeout();
325 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
327 mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
328 O2HB_MAX_WRITE_TIMEOUT_MS);
330 if (o2hb_global_heartbeat_active()) {
331 spin_lock(&o2hb_live_lock);
332 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
333 spin_unlock(&o2hb_live_lock);
335 cancel_delayed_work(®->hr_write_timeout_work);
336 reg->hr_last_timeout_start = jiffies;
337 schedule_delayed_work(®->hr_write_timeout_work,
338 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
341 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
343 cancel_delayed_work(®->hr_write_timeout_work);
344 flush_scheduled_work();
347 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
349 atomic_set(&wc->wc_num_reqs, 1);
350 init_completion(&wc->wc_io_complete);
354 /* Used in error paths too */
355 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
358 /* sadly atomic_sub_and_test() isn't available on all platforms. The
359 * good news is that the fast path only completes one at a time */
361 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
363 complete(&wc->wc_io_complete);
368 static void o2hb_wait_on_io(struct o2hb_region *reg,
369 struct o2hb_bio_wait_ctxt *wc)
371 struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping;
373 blk_run_address_space(mapping);
374 o2hb_bio_wait_dec(wc, 1);
376 wait_for_completion(&wc->wc_io_complete);
379 static void o2hb_bio_end_io(struct bio *bio,
382 struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
385 mlog(ML_ERROR, "IO Error %d\n", error);
386 wc->wc_error = error;
389 o2hb_bio_wait_dec(wc, 1);
393 /* Setup a Bio to cover I/O against num_slots slots starting at
395 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
396 struct o2hb_bio_wait_ctxt *wc,
397 unsigned int *current_slot,
398 unsigned int max_slots)
400 int len, current_page;
401 unsigned int vec_len, vec_start;
402 unsigned int bits = reg->hr_block_bits;
403 unsigned int spp = reg->hr_slots_per_page;
404 unsigned int cs = *current_slot;
408 /* Testing has shown this allocation to take long enough under
409 * GFP_KERNEL that the local node can get fenced. It would be
410 * nicest if we could pre-allocate these bios and avoid this
412 bio = bio_alloc(GFP_ATOMIC, 16);
414 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
415 bio = ERR_PTR(-ENOMEM);
419 /* Must put everything in 512 byte sectors for the bio... */
420 bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
421 bio->bi_bdev = reg->hr_bdev;
422 bio->bi_private = wc;
423 bio->bi_end_io = o2hb_bio_end_io;
425 vec_start = (cs << bits) % PAGE_CACHE_SIZE;
426 while(cs < max_slots) {
427 current_page = cs / spp;
428 page = reg->hr_slot_data[current_page];
430 vec_len = min(PAGE_CACHE_SIZE - vec_start,
431 (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
433 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
434 current_page, vec_len, vec_start);
436 len = bio_add_page(bio, page, vec_len, vec_start);
437 if (len != vec_len) break;
439 cs += vec_len / (PAGE_CACHE_SIZE/spp);
448 static int o2hb_read_slots(struct o2hb_region *reg,
449 unsigned int max_slots)
451 unsigned int current_slot=0;
453 struct o2hb_bio_wait_ctxt wc;
456 o2hb_bio_wait_init(&wc);
458 while(current_slot < max_slots) {
459 bio = o2hb_setup_one_bio(reg, &wc, ¤t_slot, max_slots);
461 status = PTR_ERR(bio);
466 atomic_inc(&wc.wc_num_reqs);
467 submit_bio(READ, bio);
473 o2hb_wait_on_io(reg, &wc);
474 if (wc.wc_error && !status)
475 status = wc.wc_error;
480 static int o2hb_issue_node_write(struct o2hb_region *reg,
481 struct o2hb_bio_wait_ctxt *write_wc)
487 o2hb_bio_wait_init(write_wc);
489 slot = o2nm_this_node();
491 bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
493 status = PTR_ERR(bio);
498 atomic_inc(&write_wc->wc_num_reqs);
499 submit_bio(WRITE, bio);
506 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
507 struct o2hb_disk_heartbeat_block *hb_block)
512 /* We want to compute the block crc with a 0 value in the
513 * hb_cksum field. Save it off here and replace after the
515 old_cksum = hb_block->hb_cksum;
516 hb_block->hb_cksum = 0;
518 ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
520 hb_block->hb_cksum = old_cksum;
525 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
527 mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
528 "cksum = 0x%x, generation 0x%llx\n",
529 (long long)le64_to_cpu(hb_block->hb_seq),
530 hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
531 (long long)le64_to_cpu(hb_block->hb_generation));
534 static int o2hb_verify_crc(struct o2hb_region *reg,
535 struct o2hb_disk_heartbeat_block *hb_block)
539 read = le32_to_cpu(hb_block->hb_cksum);
540 computed = o2hb_compute_block_crc_le(reg, hb_block);
542 return read == computed;
545 /* We want to make sure that nobody is heartbeating on top of us --
546 * this will help detect an invalid configuration. */
547 static int o2hb_check_last_timestamp(struct o2hb_region *reg)
550 struct o2hb_disk_slot *slot;
551 struct o2hb_disk_heartbeat_block *hb_block;
553 node_num = o2nm_this_node();
556 slot = ®->hr_slots[node_num];
557 /* Don't check on our 1st timestamp */
558 if (slot->ds_last_time) {
559 hb_block = slot->ds_raw_block;
561 if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time)
568 static inline void o2hb_prepare_block(struct o2hb_region *reg,
573 struct o2hb_disk_slot *slot;
574 struct o2hb_disk_heartbeat_block *hb_block;
576 node_num = o2nm_this_node();
577 slot = ®->hr_slots[node_num];
579 hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
580 memset(hb_block, 0, reg->hr_block_bytes);
581 /* TODO: time stuff */
582 cputime = CURRENT_TIME.tv_sec;
586 hb_block->hb_seq = cpu_to_le64(cputime);
587 hb_block->hb_node = node_num;
588 hb_block->hb_generation = cpu_to_le64(generation);
589 hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
591 /* This step must always happen last! */
592 hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
595 mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
596 (long long)generation,
597 le32_to_cpu(hb_block->hb_cksum));
600 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
601 struct o2nm_node *node,
604 struct list_head *iter;
605 struct o2hb_callback_func *f;
607 list_for_each(iter, &hbcall->list) {
608 f = list_entry(iter, struct o2hb_callback_func, hc_item);
609 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
610 (f->hc_func)(node, idx, f->hc_data);
614 /* Will run the list in order until we process the passed event */
615 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
618 struct o2hb_callback *hbcall;
619 struct o2hb_node_event *event;
621 spin_lock(&o2hb_live_lock);
622 empty = list_empty(&queued_event->hn_item);
623 spin_unlock(&o2hb_live_lock);
627 /* Holding callback sem assures we don't alter the callback
628 * lists when doing this, and serializes ourselves with other
629 * processes wanting callbacks. */
630 down_write(&o2hb_callback_sem);
632 spin_lock(&o2hb_live_lock);
633 while (!list_empty(&o2hb_node_events)
634 && !list_empty(&queued_event->hn_item)) {
635 event = list_entry(o2hb_node_events.next,
636 struct o2hb_node_event,
638 list_del_init(&event->hn_item);
639 spin_unlock(&o2hb_live_lock);
641 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
642 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
645 hbcall = hbcall_from_type(event->hn_event_type);
647 /* We should *never* have gotten on to the list with a
648 * bad type... This isn't something that we should try
649 * to recover from. */
650 BUG_ON(IS_ERR(hbcall));
652 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
654 spin_lock(&o2hb_live_lock);
656 spin_unlock(&o2hb_live_lock);
658 up_write(&o2hb_callback_sem);
661 static void o2hb_queue_node_event(struct o2hb_node_event *event,
662 enum o2hb_callback_type type,
663 struct o2nm_node *node,
666 assert_spin_locked(&o2hb_live_lock);
668 BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
670 event->hn_event_type = type;
671 event->hn_node = node;
672 event->hn_node_num = node_num;
674 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
675 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
677 list_add_tail(&event->hn_item, &o2hb_node_events);
680 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
682 struct o2hb_node_event event =
683 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
684 struct o2nm_node *node;
686 node = o2nm_get_node_by_num(slot->ds_node_num);
690 spin_lock(&o2hb_live_lock);
691 if (!list_empty(&slot->ds_live_item)) {
692 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
695 list_del_init(&slot->ds_live_item);
697 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
698 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
700 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
704 spin_unlock(&o2hb_live_lock);
706 o2hb_run_event_list(&event);
711 static void o2hb_set_quorum_device(struct o2hb_region *reg,
712 struct o2hb_disk_slot *slot)
714 assert_spin_locked(&o2hb_live_lock);
716 if (!o2hb_global_heartbeat_active())
719 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
723 * A region can be added to the quorum only when it sees all
724 * live nodes heartbeat on it. In other words, the region has been
725 * added to all nodes.
727 if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
728 sizeof(o2hb_live_node_bitmap)))
731 if (slot->ds_changed_samples < O2HB_LIVE_THRESHOLD)
734 printk(KERN_NOTICE "o2hb: Region %s is now a quorum device\n",
735 config_item_name(®->hr_item));
737 set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
740 * If global heartbeat active, unpin all regions if the
741 * region count > CUT_OFF
743 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
744 O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
745 o2hb_region_unpin(NULL);
748 static int o2hb_check_slot(struct o2hb_region *reg,
749 struct o2hb_disk_slot *slot)
751 int changed = 0, gen_changed = 0;
752 struct o2hb_node_event event =
753 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
754 struct o2nm_node *node;
755 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
757 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
758 unsigned int slot_dead_ms;
761 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
764 * If a node is no longer configured but is still in the livemap, we
765 * may need to clear that bit from the livemap.
767 node = o2nm_get_node_by_num(slot->ds_node_num);
769 spin_lock(&o2hb_live_lock);
770 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
771 spin_unlock(&o2hb_live_lock);
776 if (!o2hb_verify_crc(reg, hb_block)) {
777 /* all paths from here will drop o2hb_live_lock for
779 spin_lock(&o2hb_live_lock);
781 /* Don't print an error on the console in this case -
782 * a freshly formatted heartbeat area will not have a
784 if (list_empty(&slot->ds_live_item))
787 /* The node is live but pushed out a bad crc. We
788 * consider it a transient miss but don't populate any
789 * other values as they may be junk. */
790 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
791 slot->ds_node_num, reg->hr_dev_name);
792 o2hb_dump_slot(hb_block);
794 slot->ds_equal_samples++;
798 /* we don't care if these wrap.. the state transitions below
799 * clear at the right places */
800 cputime = le64_to_cpu(hb_block->hb_seq);
801 if (slot->ds_last_time != cputime)
802 slot->ds_changed_samples++;
804 slot->ds_equal_samples++;
805 slot->ds_last_time = cputime;
807 /* The node changed heartbeat generations. We assume this to
808 * mean it dropped off but came back before we timed out. We
809 * want to consider it down for the time being but don't want
810 * to lose any changed_samples state we might build up to
811 * considering it live again. */
812 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
814 slot->ds_equal_samples = 0;
815 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
816 "to 0x%llx)\n", slot->ds_node_num,
817 (long long)slot->ds_last_generation,
818 (long long)le64_to_cpu(hb_block->hb_generation));
821 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
823 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
824 "seq %llu last %llu changed %u equal %u\n",
825 slot->ds_node_num, (long long)slot->ds_last_generation,
826 le32_to_cpu(hb_block->hb_cksum),
827 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
828 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
829 slot->ds_equal_samples);
831 spin_lock(&o2hb_live_lock);
834 /* dead nodes only come to life after some number of
835 * changes at any time during their dead time */
836 if (list_empty(&slot->ds_live_item) &&
837 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
838 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
839 slot->ds_node_num, (long long)slot->ds_last_generation);
841 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
843 /* first on the list generates a callback */
844 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
845 mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
846 "bitmap\n", slot->ds_node_num);
847 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
849 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
855 list_add_tail(&slot->ds_live_item,
856 &o2hb_live_slots[slot->ds_node_num]);
858 slot->ds_equal_samples = 0;
860 /* We want to be sure that all nodes agree on the
861 * number of milliseconds before a node will be
862 * considered dead. The self-fencing timeout is
863 * computed from this value, and a discrepancy might
864 * result in heartbeat calling a node dead when it
865 * hasn't self-fenced yet. */
866 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
867 if (slot_dead_ms && slot_dead_ms != dead_ms) {
868 /* TODO: Perhaps we can fail the region here. */
869 mlog(ML_ERROR, "Node %d on device %s has a dead count "
870 "of %u ms, but our count is %u ms.\n"
871 "Please double check your configuration values "
872 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
873 slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
879 /* if the list is dead, we're done.. */
880 if (list_empty(&slot->ds_live_item))
883 /* live nodes only go dead after enough consequtive missed
884 * samples.. reset the missed counter whenever we see
886 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
887 mlog(ML_HEARTBEAT, "Node %d left my region\n",
890 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
892 /* last off the live_slot generates a callback */
893 list_del_init(&slot->ds_live_item);
894 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
895 mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
896 "nodes bitmap\n", slot->ds_node_num);
897 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
899 /* node can be null */
900 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
901 node, slot->ds_node_num);
906 /* We don't clear this because the node is still
907 * actually writing new blocks. */
909 slot->ds_changed_samples = 0;
912 if (slot->ds_changed_samples) {
913 slot->ds_changed_samples = 0;
914 slot->ds_equal_samples = 0;
917 o2hb_set_quorum_device(reg, slot);
919 spin_unlock(&o2hb_live_lock);
921 o2hb_run_event_list(&event);
928 /* This could be faster if we just implmented a find_last_bit, but I
929 * don't think the circumstances warrant it. */
930 static int o2hb_highest_node(unsigned long *nodes,
937 while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
947 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
949 int i, ret, highest_node, change = 0;
950 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
951 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
952 struct o2hb_bio_wait_ctxt write_wc;
954 ret = o2nm_configured_node_map(configured_nodes,
955 sizeof(configured_nodes));
962 * If a node is not configured but is in the livemap, we still need
963 * to read the slot so as to be able to remove it from the livemap.
965 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
967 while ((i = find_next_bit(live_node_bitmap,
968 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
969 set_bit(i, configured_nodes);
972 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
973 if (highest_node >= O2NM_MAX_NODES) {
974 mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n");
978 /* No sense in reading the slots of nodes that don't exist
979 * yet. Of course, if the node definitions have holes in them
980 * then we're reading an empty slot anyway... Consider this
982 ret = o2hb_read_slots(reg, highest_node + 1);
988 /* With an up to date view of the slots, we can check that no
989 * other node has been improperly configured to heartbeat in
991 if (!o2hb_check_last_timestamp(reg))
992 mlog(ML_ERROR, "Device \"%s\": another node is heartbeating "
993 "in our slot!\n", reg->hr_dev_name);
995 /* fill in the proper info for our next heartbeat */
996 o2hb_prepare_block(reg, reg->hr_generation);
998 /* And fire off the write. Note that we don't wait on this I/O
1000 ret = o2hb_issue_node_write(reg, &write_wc);
1007 while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1009 change |= o2hb_check_slot(reg, ®->hr_slots[i]);
1013 * We have to be sure we've advertised ourselves on disk
1014 * before we can go to steady state. This ensures that
1015 * people we find in our steady state have seen us.
1017 o2hb_wait_on_io(reg, &write_wc);
1018 if (write_wc.wc_error) {
1019 /* Do not re-arm the write timeout on I/O error - we
1020 * can't be sure that the new block ever made it to
1022 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1023 write_wc.wc_error, reg->hr_dev_name);
1024 return write_wc.wc_error;
1027 o2hb_arm_write_timeout(reg);
1029 /* let the person who launched us know when things are steady */
1030 if (!change && (atomic_read(®->hr_steady_iterations) != 0)) {
1031 if (atomic_dec_and_test(®->hr_steady_iterations))
1032 wake_up(&o2hb_steady_queue);
1038 /* Subtract b from a, storing the result in a. a *must* have a larger
1040 static void o2hb_tv_subtract(struct timeval *a,
1043 /* just return 0 when a is after b */
1044 if (a->tv_sec < b->tv_sec ||
1045 (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1051 a->tv_sec -= b->tv_sec;
1052 a->tv_usec -= b->tv_usec;
1053 while ( a->tv_usec < 0 ) {
1055 a->tv_usec += 1000000;
1059 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1060 struct timeval *end)
1062 struct timeval res = *end;
1064 o2hb_tv_subtract(&res, start);
1066 return res.tv_sec * 1000 + res.tv_usec / 1000;
1070 * we ride the region ref that the region dir holds. before the region
1071 * dir is removed and drops it ref it will wait to tear down this
1074 static int o2hb_thread(void *data)
1077 struct o2hb_region *reg = data;
1078 struct o2hb_bio_wait_ctxt write_wc;
1079 struct timeval before_hb, after_hb;
1080 unsigned int elapsed_msec;
1082 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1084 set_user_nice(current, -20);
1086 while (!kthread_should_stop() && !reg->hr_unclean_stop) {
1087 /* We track the time spent inside
1088 * o2hb_do_disk_heartbeat so that we avoid more than
1089 * hr_timeout_ms between disk writes. On busy systems
1090 * this should result in a heartbeat which is less
1091 * likely to time itself out. */
1092 do_gettimeofday(&before_hb);
1096 ret = o2hb_do_disk_heartbeat(reg);
1097 } while (ret && ++i < 2);
1099 do_gettimeofday(&after_hb);
1100 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1103 "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
1104 before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1105 after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1108 if (elapsed_msec < reg->hr_timeout_ms) {
1109 /* the kthread api has blocked signals for us so no
1110 * need to record the return value. */
1111 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1115 o2hb_disarm_write_timeout(reg);
1117 /* unclean stop is only used in very bad situation */
1118 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1119 o2hb_shutdown_slot(®->hr_slots[i]);
1121 /* Explicit down notification - avoid forcing the other nodes
1122 * to timeout on this region when we could just as easily
1123 * write a clear generation - thus indicating to them that
1124 * this node has left this region.
1126 * XXX: Should we skip this on unclean_stop? */
1127 o2hb_prepare_block(reg, 0);
1128 ret = o2hb_issue_node_write(reg, &write_wc);
1130 o2hb_wait_on_io(reg, &write_wc);
1135 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n");
1140 #ifdef CONFIG_DEBUG_FS
1141 static int o2hb_debug_open(struct inode *inode, struct file *file)
1143 struct o2hb_debug_buf *db = inode->i_private;
1144 struct o2hb_region *reg;
1145 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1150 /* max_nodes should be the largest bitmap we pass here */
1151 BUG_ON(sizeof(map) < db->db_size);
1153 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1157 switch (db->db_type) {
1158 case O2HB_DB_TYPE_LIVENODES:
1159 case O2HB_DB_TYPE_LIVEREGIONS:
1160 case O2HB_DB_TYPE_QUORUMREGIONS:
1161 case O2HB_DB_TYPE_FAILEDREGIONS:
1162 spin_lock(&o2hb_live_lock);
1163 memcpy(map, db->db_data, db->db_size);
1164 spin_unlock(&o2hb_live_lock);
1167 case O2HB_DB_TYPE_REGION_LIVENODES:
1168 spin_lock(&o2hb_live_lock);
1169 reg = (struct o2hb_region *)db->db_data;
1170 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1171 spin_unlock(&o2hb_live_lock);
1174 case O2HB_DB_TYPE_REGION_NUMBER:
1175 reg = (struct o2hb_region *)db->db_data;
1176 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1177 reg->hr_region_num);
1180 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1181 reg = (struct o2hb_region *)db->db_data;
1182 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1183 jiffies_to_msecs(jiffies -
1184 reg->hr_last_timeout_start));
1187 case O2HB_DB_TYPE_REGION_PINNED:
1188 reg = (struct o2hb_region *)db->db_data;
1189 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1190 !!reg->hr_item_pinned);
1197 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1198 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1199 out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1202 i_size_write(inode, out);
1204 file->private_data = buf;
1211 static int o2hb_debug_release(struct inode *inode, struct file *file)
1213 kfree(file->private_data);
1217 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1218 size_t nbytes, loff_t *ppos)
1220 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1221 i_size_read(file->f_mapping->host));
1224 static int o2hb_debug_open(struct inode *inode, struct file *file)
1228 static int o2hb_debug_release(struct inode *inode, struct file *file)
1232 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1233 size_t nbytes, loff_t *ppos)
1237 #endif /* CONFIG_DEBUG_FS */
1239 static const struct file_operations o2hb_debug_fops = {
1240 .open = o2hb_debug_open,
1241 .release = o2hb_debug_release,
1242 .read = o2hb_debug_read,
1243 .llseek = generic_file_llseek,
1246 void o2hb_exit(void)
1248 kfree(o2hb_db_livenodes);
1249 kfree(o2hb_db_liveregions);
1250 kfree(o2hb_db_quorumregions);
1251 kfree(o2hb_db_failedregions);
1252 debugfs_remove(o2hb_debug_failedregions);
1253 debugfs_remove(o2hb_debug_quorumregions);
1254 debugfs_remove(o2hb_debug_liveregions);
1255 debugfs_remove(o2hb_debug_livenodes);
1256 debugfs_remove(o2hb_debug_dir);
1259 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1260 struct o2hb_debug_buf **db, int db_len,
1261 int type, int size, int len, void *data)
1263 *db = kmalloc(db_len, GFP_KERNEL);
1267 (*db)->db_type = type;
1268 (*db)->db_size = size;
1269 (*db)->db_len = len;
1270 (*db)->db_data = data;
1272 return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1276 static int o2hb_debug_init(void)
1280 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1281 if (!o2hb_debug_dir) {
1286 o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1289 sizeof(*o2hb_db_livenodes),
1290 O2HB_DB_TYPE_LIVENODES,
1291 sizeof(o2hb_live_node_bitmap),
1293 o2hb_live_node_bitmap);
1294 if (!o2hb_debug_livenodes) {
1299 o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1301 &o2hb_db_liveregions,
1302 sizeof(*o2hb_db_liveregions),
1303 O2HB_DB_TYPE_LIVEREGIONS,
1304 sizeof(o2hb_live_region_bitmap),
1306 o2hb_live_region_bitmap);
1307 if (!o2hb_debug_liveregions) {
1312 o2hb_debug_quorumregions =
1313 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1315 &o2hb_db_quorumregions,
1316 sizeof(*o2hb_db_quorumregions),
1317 O2HB_DB_TYPE_QUORUMREGIONS,
1318 sizeof(o2hb_quorum_region_bitmap),
1320 o2hb_quorum_region_bitmap);
1321 if (!o2hb_debug_quorumregions) {
1326 o2hb_debug_failedregions =
1327 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1329 &o2hb_db_failedregions,
1330 sizeof(*o2hb_db_failedregions),
1331 O2HB_DB_TYPE_FAILEDREGIONS,
1332 sizeof(o2hb_failed_region_bitmap),
1334 o2hb_failed_region_bitmap);
1335 if (!o2hb_debug_failedregions) {
1352 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1353 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1355 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1356 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1358 INIT_LIST_HEAD(&o2hb_node_events);
1360 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1361 memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1362 memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1363 memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1364 memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1366 o2hb_dependent_users = 0;
1368 return o2hb_debug_init();
1371 /* if we're already in a callback then we're already serialized by the sem */
1372 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1375 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1377 memcpy(map, &o2hb_live_node_bitmap, bytes);
1381 * get a map of all nodes that are heartbeating in any regions
1383 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1385 /* callers want to serialize this map and callbacks so that they
1386 * can trust that they don't miss nodes coming to the party */
1387 down_read(&o2hb_callback_sem);
1388 spin_lock(&o2hb_live_lock);
1389 o2hb_fill_node_map_from_callback(map, bytes);
1390 spin_unlock(&o2hb_live_lock);
1391 up_read(&o2hb_callback_sem);
1393 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1396 * heartbeat configfs bits. The heartbeat set is a default set under
1397 * the cluster set in nodemanager.c.
1400 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1402 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1405 /* drop_item only drops its ref after killing the thread, nothing should
1406 * be using the region anymore. this has to clean up any state that
1407 * attributes might have built up. */
1408 static void o2hb_region_release(struct config_item *item)
1412 struct o2hb_region *reg = to_o2hb_region(item);
1414 if (reg->hr_tmp_block)
1415 kfree(reg->hr_tmp_block);
1417 if (reg->hr_slot_data) {
1418 for (i = 0; i < reg->hr_num_pages; i++) {
1419 page = reg->hr_slot_data[i];
1423 kfree(reg->hr_slot_data);
1427 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1430 kfree(reg->hr_slots);
1432 kfree(reg->hr_db_regnum);
1433 kfree(reg->hr_db_livenodes);
1434 debugfs_remove(reg->hr_debug_livenodes);
1435 debugfs_remove(reg->hr_debug_regnum);
1436 debugfs_remove(reg->hr_debug_elapsed_time);
1437 debugfs_remove(reg->hr_debug_pinned);
1438 debugfs_remove(reg->hr_debug_dir);
1440 spin_lock(&o2hb_live_lock);
1441 list_del(®->hr_all_item);
1442 spin_unlock(&o2hb_live_lock);
1447 static int o2hb_read_block_input(struct o2hb_region *reg,
1450 unsigned long *ret_bytes,
1451 unsigned int *ret_bits)
1453 unsigned long bytes;
1454 char *p = (char *)page;
1456 bytes = simple_strtoul(p, &p, 0);
1457 if (!p || (*p && (*p != '\n')))
1460 /* Heartbeat and fs min / max block sizes are the same. */
1461 if (bytes > 4096 || bytes < 512)
1463 if (hweight16(bytes) != 1)
1469 *ret_bits = ffs(bytes) - 1;
1474 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1477 return sprintf(page, "%u\n", reg->hr_block_bytes);
1480 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1485 unsigned long block_bytes;
1486 unsigned int block_bits;
1491 status = o2hb_read_block_input(reg, page, count,
1492 &block_bytes, &block_bits);
1496 reg->hr_block_bytes = (unsigned int)block_bytes;
1497 reg->hr_block_bits = block_bits;
1502 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1505 return sprintf(page, "%llu\n", reg->hr_start_block);
1508 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1512 unsigned long long tmp;
1513 char *p = (char *)page;
1518 tmp = simple_strtoull(p, &p, 0);
1519 if (!p || (*p && (*p != '\n')))
1522 reg->hr_start_block = tmp;
1527 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1530 return sprintf(page, "%d\n", reg->hr_blocks);
1533 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1538 char *p = (char *)page;
1543 tmp = simple_strtoul(p, &p, 0);
1544 if (!p || (*p && (*p != '\n')))
1547 if (tmp > O2NM_MAX_NODES || tmp == 0)
1550 reg->hr_blocks = (unsigned int)tmp;
1555 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1558 unsigned int ret = 0;
1561 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1566 static void o2hb_init_region_params(struct o2hb_region *reg)
1568 reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1569 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1571 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1572 reg->hr_start_block, reg->hr_blocks);
1573 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1574 reg->hr_block_bytes, reg->hr_block_bits);
1575 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1576 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1579 static int o2hb_map_slot_data(struct o2hb_region *reg)
1582 unsigned int last_slot;
1583 unsigned int spp = reg->hr_slots_per_page;
1586 struct o2hb_disk_slot *slot;
1588 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1589 if (reg->hr_tmp_block == NULL) {
1590 mlog_errno(-ENOMEM);
1594 reg->hr_slots = kcalloc(reg->hr_blocks,
1595 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1596 if (reg->hr_slots == NULL) {
1597 mlog_errno(-ENOMEM);
1601 for(i = 0; i < reg->hr_blocks; i++) {
1602 slot = ®->hr_slots[i];
1603 slot->ds_node_num = i;
1604 INIT_LIST_HEAD(&slot->ds_live_item);
1605 slot->ds_raw_block = NULL;
1608 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1609 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1610 "at %u blocks per page\n",
1611 reg->hr_num_pages, reg->hr_blocks, spp);
1613 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1615 if (!reg->hr_slot_data) {
1616 mlog_errno(-ENOMEM);
1620 for(i = 0; i < reg->hr_num_pages; i++) {
1621 page = alloc_page(GFP_KERNEL);
1623 mlog_errno(-ENOMEM);
1627 reg->hr_slot_data[i] = page;
1629 last_slot = i * spp;
1630 raw = page_address(page);
1632 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1634 BUG_ON((j + last_slot) >= reg->hr_blocks);
1636 slot = ®->hr_slots[j + last_slot];
1637 slot->ds_raw_block =
1638 (struct o2hb_disk_heartbeat_block *) raw;
1640 raw += reg->hr_block_bytes;
1647 /* Read in all the slots available and populate the tracking
1648 * structures so that we can start with a baseline idea of what's
1650 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1653 struct o2hb_disk_slot *slot;
1654 struct o2hb_disk_heartbeat_block *hb_block;
1658 ret = o2hb_read_slots(reg, reg->hr_blocks);
1664 /* We only want to get an idea of the values initially in each
1665 * slot, so we do no verification - o2hb_check_slot will
1666 * actually determine if each configured slot is valid and
1667 * whether any values have changed. */
1668 for(i = 0; i < reg->hr_blocks; i++) {
1669 slot = ®->hr_slots[i];
1670 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1672 /* Only fill the values that o2hb_check_slot uses to
1673 * determine changing slots */
1674 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1675 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1683 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1684 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1688 struct task_struct *hb_task;
1691 char *p = (char *)page;
1692 struct file *filp = NULL;
1693 struct inode *inode = NULL;
1694 ssize_t ret = -EINVAL;
1699 /* We can't heartbeat without having had our node number
1700 * configured yet. */
1701 if (o2nm_this_node() == O2NM_MAX_NODES)
1704 fd = simple_strtol(p, &p, 0);
1705 if (!p || (*p && (*p != '\n')))
1708 if (fd < 0 || fd >= INT_MAX)
1715 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1716 reg->hr_block_bytes == 0)
1719 inode = igrab(filp->f_mapping->host);
1723 if (!S_ISBLK(inode->i_mode))
1726 reg->hr_bdev = I_BDEV(filp->f_mapping->host);
1727 ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ);
1729 reg->hr_bdev = NULL;
1734 bdevname(reg->hr_bdev, reg->hr_dev_name);
1736 sectsize = bdev_logical_block_size(reg->hr_bdev);
1737 if (sectsize != reg->hr_block_bytes) {
1739 "blocksize %u incorrect for device, expected %d",
1740 reg->hr_block_bytes, sectsize);
1745 o2hb_init_region_params(reg);
1747 /* Generation of zero is invalid */
1749 get_random_bytes(®->hr_generation,
1750 sizeof(reg->hr_generation));
1751 } while (reg->hr_generation == 0);
1753 ret = o2hb_map_slot_data(reg);
1759 ret = o2hb_populate_slot_data(reg);
1765 INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout);
1768 * A node is considered live after it has beat LIVE_THRESHOLD
1769 * times. We're not steady until we've given them a chance
1770 * _after_ our first read.
1772 atomic_set(®->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1);
1774 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1775 reg->hr_item.ci_name);
1776 if (IS_ERR(hb_task)) {
1777 ret = PTR_ERR(hb_task);
1782 spin_lock(&o2hb_live_lock);
1783 reg->hr_task = hb_task;
1784 spin_unlock(&o2hb_live_lock);
1786 ret = wait_event_interruptible(o2hb_steady_queue,
1787 atomic_read(®->hr_steady_iterations) == 0);
1789 /* We got interrupted (hello ptrace!). Clean up */
1790 spin_lock(&o2hb_live_lock);
1791 hb_task = reg->hr_task;
1792 reg->hr_task = NULL;
1793 spin_unlock(&o2hb_live_lock);
1796 kthread_stop(hb_task);
1800 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1801 spin_lock(&o2hb_live_lock);
1802 hb_task = reg->hr_task;
1803 if (o2hb_global_heartbeat_active())
1804 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1805 spin_unlock(&o2hb_live_lock);
1812 if (hb_task && o2hb_global_heartbeat_active())
1813 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s\n",
1814 config_item_name(®->hr_item));
1823 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1824 reg->hr_bdev = NULL;
1830 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1835 spin_lock(&o2hb_live_lock);
1837 pid = task_pid_nr(reg->hr_task);
1838 spin_unlock(&o2hb_live_lock);
1843 return sprintf(page, "%u\n", pid);
1846 struct o2hb_region_attribute {
1847 struct configfs_attribute attr;
1848 ssize_t (*show)(struct o2hb_region *, char *);
1849 ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1852 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1853 .attr = { .ca_owner = THIS_MODULE,
1854 .ca_name = "block_bytes",
1855 .ca_mode = S_IRUGO | S_IWUSR },
1856 .show = o2hb_region_block_bytes_read,
1857 .store = o2hb_region_block_bytes_write,
1860 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1861 .attr = { .ca_owner = THIS_MODULE,
1862 .ca_name = "start_block",
1863 .ca_mode = S_IRUGO | S_IWUSR },
1864 .show = o2hb_region_start_block_read,
1865 .store = o2hb_region_start_block_write,
1868 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1869 .attr = { .ca_owner = THIS_MODULE,
1870 .ca_name = "blocks",
1871 .ca_mode = S_IRUGO | S_IWUSR },
1872 .show = o2hb_region_blocks_read,
1873 .store = o2hb_region_blocks_write,
1876 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1877 .attr = { .ca_owner = THIS_MODULE,
1879 .ca_mode = S_IRUGO | S_IWUSR },
1880 .show = o2hb_region_dev_read,
1881 .store = o2hb_region_dev_write,
1884 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1885 .attr = { .ca_owner = THIS_MODULE,
1887 .ca_mode = S_IRUGO | S_IRUSR },
1888 .show = o2hb_region_pid_read,
1891 static struct configfs_attribute *o2hb_region_attrs[] = {
1892 &o2hb_region_attr_block_bytes.attr,
1893 &o2hb_region_attr_start_block.attr,
1894 &o2hb_region_attr_blocks.attr,
1895 &o2hb_region_attr_dev.attr,
1896 &o2hb_region_attr_pid.attr,
1900 static ssize_t o2hb_region_show(struct config_item *item,
1901 struct configfs_attribute *attr,
1904 struct o2hb_region *reg = to_o2hb_region(item);
1905 struct o2hb_region_attribute *o2hb_region_attr =
1906 container_of(attr, struct o2hb_region_attribute, attr);
1909 if (o2hb_region_attr->show)
1910 ret = o2hb_region_attr->show(reg, page);
1914 static ssize_t o2hb_region_store(struct config_item *item,
1915 struct configfs_attribute *attr,
1916 const char *page, size_t count)
1918 struct o2hb_region *reg = to_o2hb_region(item);
1919 struct o2hb_region_attribute *o2hb_region_attr =
1920 container_of(attr, struct o2hb_region_attribute, attr);
1921 ssize_t ret = -EINVAL;
1923 if (o2hb_region_attr->store)
1924 ret = o2hb_region_attr->store(reg, page, count);
1928 static struct configfs_item_operations o2hb_region_item_ops = {
1929 .release = o2hb_region_release,
1930 .show_attribute = o2hb_region_show,
1931 .store_attribute = o2hb_region_store,
1934 static struct config_item_type o2hb_region_type = {
1935 .ct_item_ops = &o2hb_region_item_ops,
1936 .ct_attrs = o2hb_region_attrs,
1937 .ct_owner = THIS_MODULE,
1942 struct o2hb_heartbeat_group {
1943 struct config_group hs_group;
1947 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1950 container_of(group, struct o2hb_heartbeat_group, hs_group)
1954 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
1959 debugfs_create_dir(config_item_name(®->hr_item), dir);
1960 if (!reg->hr_debug_dir) {
1965 reg->hr_debug_livenodes =
1966 o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1968 &(reg->hr_db_livenodes),
1969 sizeof(*(reg->hr_db_livenodes)),
1970 O2HB_DB_TYPE_REGION_LIVENODES,
1971 sizeof(reg->hr_live_node_bitmap),
1972 O2NM_MAX_NODES, reg);
1973 if (!reg->hr_debug_livenodes) {
1978 reg->hr_debug_regnum =
1979 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
1981 &(reg->hr_db_regnum),
1982 sizeof(*(reg->hr_db_regnum)),
1983 O2HB_DB_TYPE_REGION_NUMBER,
1984 0, O2NM_MAX_NODES, reg);
1985 if (!reg->hr_debug_regnum) {
1990 reg->hr_debug_elapsed_time =
1991 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
1993 &(reg->hr_db_elapsed_time),
1994 sizeof(*(reg->hr_db_elapsed_time)),
1995 O2HB_DB_TYPE_REGION_ELAPSED_TIME,
1997 if (!reg->hr_debug_elapsed_time) {
2002 reg->hr_debug_pinned =
2003 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2005 &(reg->hr_db_pinned),
2006 sizeof(*(reg->hr_db_pinned)),
2007 O2HB_DB_TYPE_REGION_PINNED,
2009 if (!reg->hr_debug_pinned) {
2019 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2022 struct o2hb_region *reg = NULL;
2025 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2027 return ERR_PTR(-ENOMEM);
2029 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2030 ret = -ENAMETOOLONG;
2034 spin_lock(&o2hb_live_lock);
2035 reg->hr_region_num = 0;
2036 if (o2hb_global_heartbeat_active()) {
2037 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2039 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2040 spin_unlock(&o2hb_live_lock);
2044 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2046 list_add_tail(®->hr_all_item, &o2hb_all_regions);
2047 spin_unlock(&o2hb_live_lock);
2049 config_item_init_type_name(®->hr_item, name, &o2hb_region_type);
2051 ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2053 config_item_put(®->hr_item);
2057 return ®->hr_item;
2060 return ERR_PTR(ret);
2063 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2064 struct config_item *item)
2066 struct task_struct *hb_task;
2067 struct o2hb_region *reg = to_o2hb_region(item);
2068 int quorum_region = 0;
2070 /* stop the thread when the user removes the region dir */
2071 spin_lock(&o2hb_live_lock);
2072 if (o2hb_global_heartbeat_active()) {
2073 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2074 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2075 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2077 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2079 hb_task = reg->hr_task;
2080 reg->hr_task = NULL;
2081 reg->hr_item_dropped = 1;
2082 spin_unlock(&o2hb_live_lock);
2085 kthread_stop(hb_task);
2088 * If we're racing a dev_write(), we need to wake them. They will
2089 * check reg->hr_task
2091 if (atomic_read(®->hr_steady_iterations) != 0) {
2092 atomic_set(®->hr_steady_iterations, 0);
2093 wake_up(&o2hb_steady_queue);
2096 if (o2hb_global_heartbeat_active())
2097 printk(KERN_NOTICE "o2hb: Heartbeat stopped on region %s\n",
2098 config_item_name(®->hr_item));
2100 config_item_put(item);
2102 if (!o2hb_global_heartbeat_active() || !quorum_region)
2106 * If global heartbeat active and there are dependent users,
2107 * pin all regions if quorum region count <= CUT_OFF
2109 spin_lock(&o2hb_live_lock);
2111 if (!o2hb_dependent_users)
2114 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2115 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2116 o2hb_region_pin(NULL);
2119 spin_unlock(&o2hb_live_lock);
2122 struct o2hb_heartbeat_group_attribute {
2123 struct configfs_attribute attr;
2124 ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2125 ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2128 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2129 struct configfs_attribute *attr,
2132 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2133 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2134 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2137 if (o2hb_heartbeat_group_attr->show)
2138 ret = o2hb_heartbeat_group_attr->show(reg, page);
2142 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2143 struct configfs_attribute *attr,
2144 const char *page, size_t count)
2146 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2147 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2148 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2149 ssize_t ret = -EINVAL;
2151 if (o2hb_heartbeat_group_attr->store)
2152 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2156 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2159 return sprintf(page, "%u\n", o2hb_dead_threshold);
2162 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2167 char *p = (char *)page;
2169 tmp = simple_strtoul(p, &p, 10);
2170 if (!p || (*p && (*p != '\n')))
2173 /* this will validate ranges for us. */
2174 o2hb_dead_threshold_set((unsigned int) tmp);
2180 ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2183 return sprintf(page, "%s\n",
2184 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2188 ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2189 const char *page, size_t count)
2195 len = (page[count - 1] == '\n') ? count - 1 : count;
2199 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2200 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2203 ret = o2hb_global_hearbeat_mode_set(i);
2205 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2206 o2hb_heartbeat_mode_desc[i]);
2214 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2215 .attr = { .ca_owner = THIS_MODULE,
2216 .ca_name = "dead_threshold",
2217 .ca_mode = S_IRUGO | S_IWUSR },
2218 .show = o2hb_heartbeat_group_threshold_show,
2219 .store = o2hb_heartbeat_group_threshold_store,
2222 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2223 .attr = { .ca_owner = THIS_MODULE,
2225 .ca_mode = S_IRUGO | S_IWUSR },
2226 .show = o2hb_heartbeat_group_mode_show,
2227 .store = o2hb_heartbeat_group_mode_store,
2230 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2231 &o2hb_heartbeat_group_attr_threshold.attr,
2232 &o2hb_heartbeat_group_attr_mode.attr,
2236 static struct configfs_item_operations o2hb_hearbeat_group_item_ops = {
2237 .show_attribute = o2hb_heartbeat_group_show,
2238 .store_attribute = o2hb_heartbeat_group_store,
2241 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2242 .make_item = o2hb_heartbeat_group_make_item,
2243 .drop_item = o2hb_heartbeat_group_drop_item,
2246 static struct config_item_type o2hb_heartbeat_group_type = {
2247 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
2248 .ct_item_ops = &o2hb_hearbeat_group_item_ops,
2249 .ct_attrs = o2hb_heartbeat_group_attrs,
2250 .ct_owner = THIS_MODULE,
2253 /* this is just here to avoid touching group in heartbeat.h which the
2254 * entire damn world #includes */
2255 struct config_group *o2hb_alloc_hb_set(void)
2257 struct o2hb_heartbeat_group *hs = NULL;
2258 struct config_group *ret = NULL;
2260 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2264 config_group_init_type_name(&hs->hs_group, "heartbeat",
2265 &o2hb_heartbeat_group_type);
2267 ret = &hs->hs_group;
2274 void o2hb_free_hb_set(struct config_group *group)
2276 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2280 /* hb callback registration and issueing */
2282 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2284 if (type == O2HB_NUM_CB)
2285 return ERR_PTR(-EINVAL);
2287 return &o2hb_callbacks[type];
2290 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2291 enum o2hb_callback_type type,
2296 INIT_LIST_HEAD(&hc->hc_item);
2299 hc->hc_priority = priority;
2301 hc->hc_magic = O2HB_CB_MAGIC;
2303 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2306 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2307 * In global heartbeat mode, region_uuid passed is NULL.
2309 * In local, we only pin the matching region. In global we pin all the active
2312 static int o2hb_region_pin(const char *region_uuid)
2314 int ret = 0, found = 0;
2315 struct o2hb_region *reg;
2318 assert_spin_locked(&o2hb_live_lock);
2320 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2321 uuid = config_item_name(®->hr_item);
2323 /* local heartbeat */
2325 if (strcmp(region_uuid, uuid))
2330 if (reg->hr_item_pinned || reg->hr_item_dropped)
2333 /* Ignore ENOENT only for local hb (userdlm domain) */
2334 ret = o2nm_depend_item(®->hr_item);
2336 mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2337 reg->hr_item_pinned = 1;
2339 if (ret == -ENOENT && found)
2342 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2356 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2357 * In global heartbeat mode, region_uuid passed is NULL.
2359 * In local, we only unpin the matching region. In global we unpin all the
2362 static void o2hb_region_unpin(const char *region_uuid)
2364 struct o2hb_region *reg;
2368 assert_spin_locked(&o2hb_live_lock);
2370 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2371 uuid = config_item_name(®->hr_item);
2373 if (strcmp(region_uuid, uuid))
2378 if (reg->hr_item_pinned) {
2379 mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2380 o2nm_undepend_item(®->hr_item);
2381 reg->hr_item_pinned = 0;
2388 static int o2hb_region_inc_user(const char *region_uuid)
2392 spin_lock(&o2hb_live_lock);
2394 /* local heartbeat */
2395 if (!o2hb_global_heartbeat_active()) {
2396 ret = o2hb_region_pin(region_uuid);
2401 * if global heartbeat active and this is the first dependent user,
2402 * pin all regions if quorum region count <= CUT_OFF
2404 o2hb_dependent_users++;
2405 if (o2hb_dependent_users > 1)
2408 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2409 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2410 ret = o2hb_region_pin(NULL);
2413 spin_unlock(&o2hb_live_lock);
2417 void o2hb_region_dec_user(const char *region_uuid)
2419 spin_lock(&o2hb_live_lock);
2421 /* local heartbeat */
2422 if (!o2hb_global_heartbeat_active()) {
2423 o2hb_region_unpin(region_uuid);
2428 * if global heartbeat active and there are no dependent users,
2429 * unpin all quorum regions
2431 o2hb_dependent_users--;
2432 if (!o2hb_dependent_users)
2433 o2hb_region_unpin(NULL);
2436 spin_unlock(&o2hb_live_lock);
2439 int o2hb_register_callback(const char *region_uuid,
2440 struct o2hb_callback_func *hc)
2442 struct o2hb_callback_func *tmp;
2443 struct list_head *iter;
2444 struct o2hb_callback *hbcall;
2447 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2448 BUG_ON(!list_empty(&hc->hc_item));
2450 hbcall = hbcall_from_type(hc->hc_type);
2451 if (IS_ERR(hbcall)) {
2452 ret = PTR_ERR(hbcall);
2457 ret = o2hb_region_inc_user(region_uuid);
2464 down_write(&o2hb_callback_sem);
2466 list_for_each(iter, &hbcall->list) {
2467 tmp = list_entry(iter, struct o2hb_callback_func, hc_item);
2468 if (hc->hc_priority < tmp->hc_priority) {
2469 list_add_tail(&hc->hc_item, iter);
2473 if (list_empty(&hc->hc_item))
2474 list_add_tail(&hc->hc_item, &hbcall->list);
2476 up_write(&o2hb_callback_sem);
2479 mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2480 ret, __builtin_return_address(0), hc);
2483 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2485 void o2hb_unregister_callback(const char *region_uuid,
2486 struct o2hb_callback_func *hc)
2488 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2490 mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2491 __builtin_return_address(0), hc);
2493 /* XXX Can this happen _with_ a region reference? */
2494 if (list_empty(&hc->hc_item))
2498 o2hb_region_dec_user(region_uuid);
2500 down_write(&o2hb_callback_sem);
2502 list_del_init(&hc->hc_item);
2504 up_write(&o2hb_callback_sem);
2506 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2508 int o2hb_check_node_heartbeating(u8 node_num)
2510 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2512 o2hb_fill_node_map(testing_map, sizeof(testing_map));
2513 if (!test_bit(node_num, testing_map)) {
2515 "node (%u) does not have heartbeating enabled.\n",
2522 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2524 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2526 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2528 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2529 if (!test_bit(node_num, testing_map)) {
2531 "node (%u) does not have heartbeating enabled.\n",
2538 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2540 /* Makes sure our local node is configured with a node number, and is
2542 int o2hb_check_local_node_heartbeating(void)
2546 /* if this node was set then we have networking */
2547 node_num = o2nm_this_node();
2548 if (node_num == O2NM_MAX_NODES) {
2549 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2553 return o2hb_check_node_heartbeating(node_num);
2555 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2558 * this is just a hack until we get the plumbing which flips file systems
2559 * read only and drops the hb ref instead of killing the node dead.
2561 void o2hb_stop_all_regions(void)
2563 struct o2hb_region *reg;
2565 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2567 spin_lock(&o2hb_live_lock);
2569 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2570 reg->hr_unclean_stop = 1;
2572 spin_unlock(&o2hb_live_lock);
2574 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2576 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2578 struct o2hb_region *reg;
2582 spin_lock(&o2hb_live_lock);
2585 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2586 mlog(0, "Region: %s\n", config_item_name(®->hr_item));
2587 if (numregs < max_regions) {
2588 memcpy(p, config_item_name(®->hr_item),
2589 O2HB_MAX_REGION_NAME_LEN);
2590 p += O2HB_MAX_REGION_NAME_LEN;
2595 spin_unlock(&o2hb_live_lock);
2599 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2601 int o2hb_global_heartbeat_active(void)
2603 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2605 EXPORT_SYMBOL(o2hb_global_heartbeat_active);