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_heartbeat_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,
223 /* protected by the hr_callback_sem */
224 struct task_struct *hr_task;
226 unsigned int hr_blocks;
227 unsigned long long hr_start_block;
229 unsigned int hr_block_bits;
230 unsigned int hr_block_bytes;
232 unsigned int hr_slots_per_page;
233 unsigned int hr_num_pages;
235 struct page **hr_slot_data;
236 struct block_device *hr_bdev;
237 struct o2hb_disk_slot *hr_slots;
239 /* live node map of this region */
240 unsigned long hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
241 unsigned int hr_region_num;
243 struct dentry *hr_debug_dir;
244 struct dentry *hr_debug_livenodes;
245 struct dentry *hr_debug_regnum;
246 struct dentry *hr_debug_elapsed_time;
247 struct dentry *hr_debug_pinned;
248 struct o2hb_debug_buf *hr_db_livenodes;
249 struct o2hb_debug_buf *hr_db_regnum;
250 struct o2hb_debug_buf *hr_db_elapsed_time;
251 struct o2hb_debug_buf *hr_db_pinned;
253 /* let the person setting up hb wait for it to return until it
254 * has reached a 'steady' state. This will be fixed when we have
255 * a more complete api that doesn't lead to this sort of fragility. */
256 atomic_t hr_steady_iterations;
258 /* terminate o2hb thread if it does not reach steady state
259 * (hr_steady_iterations == 0) within hr_unsteady_iterations */
260 atomic_t hr_unsteady_iterations;
262 char hr_dev_name[BDEVNAME_SIZE];
264 unsigned int hr_timeout_ms;
266 /* randomized as the region goes up and down so that a node
267 * recognizes a node going up and down in one iteration */
270 struct delayed_work hr_write_timeout_work;
271 unsigned long hr_last_timeout_start;
273 /* Used during o2hb_check_slot to hold a copy of the block
274 * being checked because we temporarily have to zero out the
276 struct o2hb_disk_heartbeat_block *hr_tmp_block;
279 struct o2hb_bio_wait_ctxt {
280 atomic_t wc_num_reqs;
281 struct completion wc_io_complete;
285 static int o2hb_pop_count(void *map, int count)
289 while ((i = find_next_bit(map, count, i + 1)) < count)
294 static void o2hb_write_timeout(struct work_struct *work)
298 struct o2hb_region *reg =
299 container_of(work, struct o2hb_region,
300 hr_write_timeout_work.work);
302 mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
303 "milliseconds\n", reg->hr_dev_name,
304 jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
306 if (o2hb_global_heartbeat_active()) {
307 spin_lock_irqsave(&o2hb_live_lock, flags);
308 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
309 set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
310 failed = o2hb_pop_count(&o2hb_failed_region_bitmap,
312 quorum = o2hb_pop_count(&o2hb_quorum_region_bitmap,
314 spin_unlock_irqrestore(&o2hb_live_lock, flags);
316 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
320 * Fence if the number of failed regions >= half the number
323 if ((failed << 1) < quorum)
327 o2quo_disk_timeout();
330 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
332 /* Arm writeout only after thread reaches steady state */
333 if (atomic_read(®->hr_steady_iterations) != 0)
336 mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
337 O2HB_MAX_WRITE_TIMEOUT_MS);
339 if (o2hb_global_heartbeat_active()) {
340 spin_lock(&o2hb_live_lock);
341 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
342 spin_unlock(&o2hb_live_lock);
344 cancel_delayed_work(®->hr_write_timeout_work);
345 reg->hr_last_timeout_start = jiffies;
346 schedule_delayed_work(®->hr_write_timeout_work,
347 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
350 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
352 cancel_delayed_work_sync(®->hr_write_timeout_work);
355 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
357 atomic_set(&wc->wc_num_reqs, 1);
358 init_completion(&wc->wc_io_complete);
362 /* Used in error paths too */
363 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
366 /* sadly atomic_sub_and_test() isn't available on all platforms. The
367 * good news is that the fast path only completes one at a time */
369 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
371 complete(&wc->wc_io_complete);
376 static void o2hb_wait_on_io(struct o2hb_region *reg,
377 struct o2hb_bio_wait_ctxt *wc)
379 o2hb_bio_wait_dec(wc, 1);
380 wait_for_completion(&wc->wc_io_complete);
383 static void o2hb_bio_end_io(struct bio *bio,
386 struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
389 mlog(ML_ERROR, "IO Error %d\n", error);
390 wc->wc_error = error;
393 o2hb_bio_wait_dec(wc, 1);
397 /* Setup a Bio to cover I/O against num_slots slots starting at
399 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
400 struct o2hb_bio_wait_ctxt *wc,
401 unsigned int *current_slot,
402 unsigned int max_slots)
404 int len, current_page;
405 unsigned int vec_len, vec_start;
406 unsigned int bits = reg->hr_block_bits;
407 unsigned int spp = reg->hr_slots_per_page;
408 unsigned int cs = *current_slot;
412 /* Testing has shown this allocation to take long enough under
413 * GFP_KERNEL that the local node can get fenced. It would be
414 * nicest if we could pre-allocate these bios and avoid this
416 bio = bio_alloc(GFP_ATOMIC, 16);
418 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
419 bio = ERR_PTR(-ENOMEM);
423 /* Must put everything in 512 byte sectors for the bio... */
424 bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
425 bio->bi_bdev = reg->hr_bdev;
426 bio->bi_private = wc;
427 bio->bi_end_io = o2hb_bio_end_io;
429 vec_start = (cs << bits) % PAGE_CACHE_SIZE;
430 while(cs < max_slots) {
431 current_page = cs / spp;
432 page = reg->hr_slot_data[current_page];
434 vec_len = min(PAGE_CACHE_SIZE - vec_start,
435 (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
437 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
438 current_page, vec_len, vec_start);
440 len = bio_add_page(bio, page, vec_len, vec_start);
441 if (len != vec_len) break;
443 cs += vec_len / (PAGE_CACHE_SIZE/spp);
452 static int o2hb_read_slots(struct o2hb_region *reg,
453 unsigned int max_slots)
455 unsigned int current_slot=0;
457 struct o2hb_bio_wait_ctxt wc;
460 o2hb_bio_wait_init(&wc);
462 while(current_slot < max_slots) {
463 bio = o2hb_setup_one_bio(reg, &wc, ¤t_slot, max_slots);
465 status = PTR_ERR(bio);
470 atomic_inc(&wc.wc_num_reqs);
471 submit_bio(READ, bio);
477 o2hb_wait_on_io(reg, &wc);
478 if (wc.wc_error && !status)
479 status = wc.wc_error;
484 static int o2hb_issue_node_write(struct o2hb_region *reg,
485 struct o2hb_bio_wait_ctxt *write_wc)
491 o2hb_bio_wait_init(write_wc);
493 slot = o2nm_this_node();
495 bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
497 status = PTR_ERR(bio);
502 atomic_inc(&write_wc->wc_num_reqs);
503 submit_bio(WRITE_SYNC, bio);
510 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
511 struct o2hb_disk_heartbeat_block *hb_block)
516 /* We want to compute the block crc with a 0 value in the
517 * hb_cksum field. Save it off here and replace after the
519 old_cksum = hb_block->hb_cksum;
520 hb_block->hb_cksum = 0;
522 ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
524 hb_block->hb_cksum = old_cksum;
529 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
531 mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
532 "cksum = 0x%x, generation 0x%llx\n",
533 (long long)le64_to_cpu(hb_block->hb_seq),
534 hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
535 (long long)le64_to_cpu(hb_block->hb_generation));
538 static int o2hb_verify_crc(struct o2hb_region *reg,
539 struct o2hb_disk_heartbeat_block *hb_block)
543 read = le32_to_cpu(hb_block->hb_cksum);
544 computed = o2hb_compute_block_crc_le(reg, hb_block);
546 return read == computed;
550 * Compare the slot data with what we wrote in the last iteration.
551 * If the match fails, print an appropriate error message. This is to
552 * detect errors like... another node hearting on the same slot,
553 * flaky device that is losing writes, etc.
554 * Returns 1 if check succeeds, 0 otherwise.
556 static int o2hb_check_own_slot(struct o2hb_region *reg)
558 struct o2hb_disk_slot *slot;
559 struct o2hb_disk_heartbeat_block *hb_block;
562 slot = ®->hr_slots[o2nm_this_node()];
563 /* Don't check on our 1st timestamp */
564 if (!slot->ds_last_time)
567 hb_block = slot->ds_raw_block;
568 if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
569 le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
570 hb_block->hb_node == slot->ds_node_num)
573 #define ERRSTR1 "Another node is heartbeating on device"
574 #define ERRSTR2 "Heartbeat generation mismatch on device"
575 #define ERRSTR3 "Heartbeat sequence mismatch on device"
577 if (hb_block->hb_node != slot->ds_node_num)
579 else if (le64_to_cpu(hb_block->hb_generation) !=
580 slot->ds_last_generation)
585 mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
586 "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
587 slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
588 (unsigned long long)slot->ds_last_time, hb_block->hb_node,
589 (unsigned long long)le64_to_cpu(hb_block->hb_generation),
590 (unsigned long long)le64_to_cpu(hb_block->hb_seq));
595 static inline void o2hb_prepare_block(struct o2hb_region *reg,
600 struct o2hb_disk_slot *slot;
601 struct o2hb_disk_heartbeat_block *hb_block;
603 node_num = o2nm_this_node();
604 slot = ®->hr_slots[node_num];
606 hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
607 memset(hb_block, 0, reg->hr_block_bytes);
608 /* TODO: time stuff */
609 cputime = CURRENT_TIME.tv_sec;
613 hb_block->hb_seq = cpu_to_le64(cputime);
614 hb_block->hb_node = node_num;
615 hb_block->hb_generation = cpu_to_le64(generation);
616 hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
618 /* This step must always happen last! */
619 hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
622 mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
623 (long long)generation,
624 le32_to_cpu(hb_block->hb_cksum));
627 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
628 struct o2nm_node *node,
631 struct list_head *iter;
632 struct o2hb_callback_func *f;
634 list_for_each(iter, &hbcall->list) {
635 f = list_entry(iter, struct o2hb_callback_func, hc_item);
636 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
637 (f->hc_func)(node, idx, f->hc_data);
641 /* Will run the list in order until we process the passed event */
642 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
645 struct o2hb_callback *hbcall;
646 struct o2hb_node_event *event;
648 spin_lock(&o2hb_live_lock);
649 empty = list_empty(&queued_event->hn_item);
650 spin_unlock(&o2hb_live_lock);
654 /* Holding callback sem assures we don't alter the callback
655 * lists when doing this, and serializes ourselves with other
656 * processes wanting callbacks. */
657 down_write(&o2hb_callback_sem);
659 spin_lock(&o2hb_live_lock);
660 while (!list_empty(&o2hb_node_events)
661 && !list_empty(&queued_event->hn_item)) {
662 event = list_entry(o2hb_node_events.next,
663 struct o2hb_node_event,
665 list_del_init(&event->hn_item);
666 spin_unlock(&o2hb_live_lock);
668 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
669 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
672 hbcall = hbcall_from_type(event->hn_event_type);
674 /* We should *never* have gotten on to the list with a
675 * bad type... This isn't something that we should try
676 * to recover from. */
677 BUG_ON(IS_ERR(hbcall));
679 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
681 spin_lock(&o2hb_live_lock);
683 spin_unlock(&o2hb_live_lock);
685 up_write(&o2hb_callback_sem);
688 static void o2hb_queue_node_event(struct o2hb_node_event *event,
689 enum o2hb_callback_type type,
690 struct o2nm_node *node,
693 assert_spin_locked(&o2hb_live_lock);
695 BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
697 event->hn_event_type = type;
698 event->hn_node = node;
699 event->hn_node_num = node_num;
701 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
702 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
704 list_add_tail(&event->hn_item, &o2hb_node_events);
707 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
709 struct o2hb_node_event event =
710 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
711 struct o2nm_node *node;
713 node = o2nm_get_node_by_num(slot->ds_node_num);
717 spin_lock(&o2hb_live_lock);
718 if (!list_empty(&slot->ds_live_item)) {
719 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
722 list_del_init(&slot->ds_live_item);
724 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
725 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
727 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
731 spin_unlock(&o2hb_live_lock);
733 o2hb_run_event_list(&event);
738 static void o2hb_set_quorum_device(struct o2hb_region *reg)
740 if (!o2hb_global_heartbeat_active())
743 /* Prevent race with o2hb_heartbeat_group_drop_item() */
744 if (kthread_should_stop())
747 /* Tag region as quorum only after thread reaches steady state */
748 if (atomic_read(®->hr_steady_iterations) != 0)
751 spin_lock(&o2hb_live_lock);
753 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
757 * A region can be added to the quorum only when it sees all
758 * live nodes heartbeat on it. In other words, the region has been
759 * added to all nodes.
761 if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
762 sizeof(o2hb_live_node_bitmap)))
765 printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
766 config_item_name(®->hr_item), reg->hr_dev_name);
768 set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
771 * If global heartbeat active, unpin all regions if the
772 * region count > CUT_OFF
774 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
775 O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
776 o2hb_region_unpin(NULL);
778 spin_unlock(&o2hb_live_lock);
781 static int o2hb_check_slot(struct o2hb_region *reg,
782 struct o2hb_disk_slot *slot)
784 int changed = 0, gen_changed = 0;
785 struct o2hb_node_event event =
786 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
787 struct o2nm_node *node;
788 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
790 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
791 unsigned int slot_dead_ms;
794 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
797 * If a node is no longer configured but is still in the livemap, we
798 * may need to clear that bit from the livemap.
800 node = o2nm_get_node_by_num(slot->ds_node_num);
802 spin_lock(&o2hb_live_lock);
803 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
804 spin_unlock(&o2hb_live_lock);
809 if (!o2hb_verify_crc(reg, hb_block)) {
810 /* all paths from here will drop o2hb_live_lock for
812 spin_lock(&o2hb_live_lock);
814 /* Don't print an error on the console in this case -
815 * a freshly formatted heartbeat area will not have a
817 if (list_empty(&slot->ds_live_item))
820 /* The node is live but pushed out a bad crc. We
821 * consider it a transient miss but don't populate any
822 * other values as they may be junk. */
823 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
824 slot->ds_node_num, reg->hr_dev_name);
825 o2hb_dump_slot(hb_block);
827 slot->ds_equal_samples++;
831 /* we don't care if these wrap.. the state transitions below
832 * clear at the right places */
833 cputime = le64_to_cpu(hb_block->hb_seq);
834 if (slot->ds_last_time != cputime)
835 slot->ds_changed_samples++;
837 slot->ds_equal_samples++;
838 slot->ds_last_time = cputime;
840 /* The node changed heartbeat generations. We assume this to
841 * mean it dropped off but came back before we timed out. We
842 * want to consider it down for the time being but don't want
843 * to lose any changed_samples state we might build up to
844 * considering it live again. */
845 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
847 slot->ds_equal_samples = 0;
848 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
849 "to 0x%llx)\n", slot->ds_node_num,
850 (long long)slot->ds_last_generation,
851 (long long)le64_to_cpu(hb_block->hb_generation));
854 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
856 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
857 "seq %llu last %llu changed %u equal %u\n",
858 slot->ds_node_num, (long long)slot->ds_last_generation,
859 le32_to_cpu(hb_block->hb_cksum),
860 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
861 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
862 slot->ds_equal_samples);
864 spin_lock(&o2hb_live_lock);
867 /* dead nodes only come to life after some number of
868 * changes at any time during their dead time */
869 if (list_empty(&slot->ds_live_item) &&
870 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
871 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
872 slot->ds_node_num, (long long)slot->ds_last_generation);
874 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
876 /* first on the list generates a callback */
877 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
878 mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
879 "bitmap\n", slot->ds_node_num);
880 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
882 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
888 list_add_tail(&slot->ds_live_item,
889 &o2hb_live_slots[slot->ds_node_num]);
891 slot->ds_equal_samples = 0;
893 /* We want to be sure that all nodes agree on the
894 * number of milliseconds before a node will be
895 * considered dead. The self-fencing timeout is
896 * computed from this value, and a discrepancy might
897 * result in heartbeat calling a node dead when it
898 * hasn't self-fenced yet. */
899 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
900 if (slot_dead_ms && slot_dead_ms != dead_ms) {
901 /* TODO: Perhaps we can fail the region here. */
902 mlog(ML_ERROR, "Node %d on device %s has a dead count "
903 "of %u ms, but our count is %u ms.\n"
904 "Please double check your configuration values "
905 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
906 slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
912 /* if the list is dead, we're done.. */
913 if (list_empty(&slot->ds_live_item))
916 /* live nodes only go dead after enough consequtive missed
917 * samples.. reset the missed counter whenever we see
919 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
920 mlog(ML_HEARTBEAT, "Node %d left my region\n",
923 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
925 /* last off the live_slot generates a callback */
926 list_del_init(&slot->ds_live_item);
927 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
928 mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
929 "nodes bitmap\n", slot->ds_node_num);
930 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
932 /* node can be null */
933 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
934 node, slot->ds_node_num);
939 /* We don't clear this because the node is still
940 * actually writing new blocks. */
942 slot->ds_changed_samples = 0;
945 if (slot->ds_changed_samples) {
946 slot->ds_changed_samples = 0;
947 slot->ds_equal_samples = 0;
950 spin_unlock(&o2hb_live_lock);
952 o2hb_run_event_list(&event);
959 /* This could be faster if we just implmented a find_last_bit, but I
960 * don't think the circumstances warrant it. */
961 static int o2hb_highest_node(unsigned long *nodes,
968 while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
978 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
980 int i, ret, highest_node;
981 int membership_change = 0, own_slot_ok = 0;
982 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
983 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
984 struct o2hb_bio_wait_ctxt write_wc;
986 ret = o2nm_configured_node_map(configured_nodes,
987 sizeof(configured_nodes));
994 * If a node is not configured but is in the livemap, we still need
995 * to read the slot so as to be able to remove it from the livemap.
997 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
999 while ((i = find_next_bit(live_node_bitmap,
1000 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1001 set_bit(i, configured_nodes);
1004 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1005 if (highest_node >= O2NM_MAX_NODES) {
1006 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1011 /* No sense in reading the slots of nodes that don't exist
1012 * yet. Of course, if the node definitions have holes in them
1013 * then we're reading an empty slot anyway... Consider this
1015 ret = o2hb_read_slots(reg, highest_node + 1);
1021 /* With an up to date view of the slots, we can check that no
1022 * other node has been improperly configured to heartbeat in
1024 own_slot_ok = o2hb_check_own_slot(reg);
1026 /* fill in the proper info for our next heartbeat */
1027 o2hb_prepare_block(reg, reg->hr_generation);
1029 ret = o2hb_issue_node_write(reg, &write_wc);
1036 while((i = find_next_bit(configured_nodes,
1037 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1038 membership_change |= o2hb_check_slot(reg, ®->hr_slots[i]);
1042 * We have to be sure we've advertised ourselves on disk
1043 * before we can go to steady state. This ensures that
1044 * people we find in our steady state have seen us.
1046 o2hb_wait_on_io(reg, &write_wc);
1047 if (write_wc.wc_error) {
1048 /* Do not re-arm the write timeout on I/O error - we
1049 * can't be sure that the new block ever made it to
1051 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1052 write_wc.wc_error, reg->hr_dev_name);
1053 ret = write_wc.wc_error;
1057 /* Skip disarming the timeout if own slot has stale/bad data */
1059 o2hb_set_quorum_device(reg);
1060 o2hb_arm_write_timeout(reg);
1064 /* let the person who launched us know when things are steady */
1065 if (atomic_read(®->hr_steady_iterations) != 0) {
1066 if (!ret && own_slot_ok && !membership_change) {
1067 if (atomic_dec_and_test(®->hr_steady_iterations))
1068 wake_up(&o2hb_steady_queue);
1072 if (atomic_read(®->hr_steady_iterations) != 0) {
1073 if (atomic_dec_and_test(®->hr_unsteady_iterations)) {
1074 printk(KERN_NOTICE "o2hb: Unable to stabilize "
1075 "heartbeart on region %s (%s)\n",
1076 config_item_name(®->hr_item),
1078 atomic_set(®->hr_steady_iterations, 0);
1079 reg->hr_aborted_start = 1;
1080 wake_up(&o2hb_steady_queue);
1088 /* Subtract b from a, storing the result in a. a *must* have a larger
1090 static void o2hb_tv_subtract(struct timeval *a,
1093 /* just return 0 when a is after b */
1094 if (a->tv_sec < b->tv_sec ||
1095 (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1101 a->tv_sec -= b->tv_sec;
1102 a->tv_usec -= b->tv_usec;
1103 while ( a->tv_usec < 0 ) {
1105 a->tv_usec += 1000000;
1109 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1110 struct timeval *end)
1112 struct timeval res = *end;
1114 o2hb_tv_subtract(&res, start);
1116 return res.tv_sec * 1000 + res.tv_usec / 1000;
1120 * we ride the region ref that the region dir holds. before the region
1121 * dir is removed and drops it ref it will wait to tear down this
1124 static int o2hb_thread(void *data)
1127 struct o2hb_region *reg = data;
1128 struct o2hb_bio_wait_ctxt write_wc;
1129 struct timeval before_hb, after_hb;
1130 unsigned int elapsed_msec;
1132 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1134 set_user_nice(current, -20);
1137 o2nm_depend_this_node();
1139 while (!kthread_should_stop() &&
1140 !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1141 /* We track the time spent inside
1142 * o2hb_do_disk_heartbeat so that we avoid more than
1143 * hr_timeout_ms between disk writes. On busy systems
1144 * this should result in a heartbeat which is less
1145 * likely to time itself out. */
1146 do_gettimeofday(&before_hb);
1148 ret = o2hb_do_disk_heartbeat(reg);
1150 do_gettimeofday(&after_hb);
1151 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1154 "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
1155 before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1156 after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1159 if (!kthread_should_stop() &&
1160 elapsed_msec < reg->hr_timeout_ms) {
1161 /* the kthread api has blocked signals for us so no
1162 * need to record the return value. */
1163 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1167 o2hb_disarm_write_timeout(reg);
1169 /* unclean stop is only used in very bad situation */
1170 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1171 o2hb_shutdown_slot(®->hr_slots[i]);
1173 /* Explicit down notification - avoid forcing the other nodes
1174 * to timeout on this region when we could just as easily
1175 * write a clear generation - thus indicating to them that
1176 * this node has left this region.
1178 if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1179 o2hb_prepare_block(reg, 0);
1180 ret = o2hb_issue_node_write(reg, &write_wc);
1182 o2hb_wait_on_io(reg, &write_wc);
1188 o2nm_undepend_this_node();
1190 mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1195 #ifdef CONFIG_DEBUG_FS
1196 static int o2hb_debug_open(struct inode *inode, struct file *file)
1198 struct o2hb_debug_buf *db = inode->i_private;
1199 struct o2hb_region *reg;
1200 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1206 /* max_nodes should be the largest bitmap we pass here */
1207 BUG_ON(sizeof(map) < db->db_size);
1209 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1213 switch (db->db_type) {
1214 case O2HB_DB_TYPE_LIVENODES:
1215 case O2HB_DB_TYPE_LIVEREGIONS:
1216 case O2HB_DB_TYPE_QUORUMREGIONS:
1217 case O2HB_DB_TYPE_FAILEDREGIONS:
1218 spin_lock(&o2hb_live_lock);
1219 memcpy(map, db->db_data, db->db_size);
1220 spin_unlock(&o2hb_live_lock);
1223 case O2HB_DB_TYPE_REGION_LIVENODES:
1224 spin_lock(&o2hb_live_lock);
1225 reg = (struct o2hb_region *)db->db_data;
1226 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1227 spin_unlock(&o2hb_live_lock);
1230 case O2HB_DB_TYPE_REGION_NUMBER:
1231 reg = (struct o2hb_region *)db->db_data;
1232 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1233 reg->hr_region_num);
1236 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1237 reg = (struct o2hb_region *)db->db_data;
1238 lts = reg->hr_last_timeout_start;
1239 /* If 0, it has never been set before */
1241 lts = jiffies_to_msecs(jiffies - lts);
1242 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1245 case O2HB_DB_TYPE_REGION_PINNED:
1246 reg = (struct o2hb_region *)db->db_data;
1247 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1248 !!reg->hr_item_pinned);
1255 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1256 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1257 out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1260 i_size_write(inode, out);
1262 file->private_data = buf;
1269 static int o2hb_debug_release(struct inode *inode, struct file *file)
1271 kfree(file->private_data);
1275 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1276 size_t nbytes, loff_t *ppos)
1278 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1279 i_size_read(file->f_mapping->host));
1282 static int o2hb_debug_open(struct inode *inode, struct file *file)
1286 static int o2hb_debug_release(struct inode *inode, struct file *file)
1290 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1291 size_t nbytes, loff_t *ppos)
1295 #endif /* CONFIG_DEBUG_FS */
1297 static const struct file_operations o2hb_debug_fops = {
1298 .open = o2hb_debug_open,
1299 .release = o2hb_debug_release,
1300 .read = o2hb_debug_read,
1301 .llseek = generic_file_llseek,
1304 void o2hb_exit(void)
1306 kfree(o2hb_db_livenodes);
1307 kfree(o2hb_db_liveregions);
1308 kfree(o2hb_db_quorumregions);
1309 kfree(o2hb_db_failedregions);
1310 debugfs_remove(o2hb_debug_failedregions);
1311 debugfs_remove(o2hb_debug_quorumregions);
1312 debugfs_remove(o2hb_debug_liveregions);
1313 debugfs_remove(o2hb_debug_livenodes);
1314 debugfs_remove(o2hb_debug_dir);
1317 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1318 struct o2hb_debug_buf **db, int db_len,
1319 int type, int size, int len, void *data)
1321 *db = kmalloc(db_len, GFP_KERNEL);
1325 (*db)->db_type = type;
1326 (*db)->db_size = size;
1327 (*db)->db_len = len;
1328 (*db)->db_data = data;
1330 return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1334 static int o2hb_debug_init(void)
1338 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1339 if (!o2hb_debug_dir) {
1344 o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1347 sizeof(*o2hb_db_livenodes),
1348 O2HB_DB_TYPE_LIVENODES,
1349 sizeof(o2hb_live_node_bitmap),
1351 o2hb_live_node_bitmap);
1352 if (!o2hb_debug_livenodes) {
1357 o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1359 &o2hb_db_liveregions,
1360 sizeof(*o2hb_db_liveregions),
1361 O2HB_DB_TYPE_LIVEREGIONS,
1362 sizeof(o2hb_live_region_bitmap),
1364 o2hb_live_region_bitmap);
1365 if (!o2hb_debug_liveregions) {
1370 o2hb_debug_quorumregions =
1371 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1373 &o2hb_db_quorumregions,
1374 sizeof(*o2hb_db_quorumregions),
1375 O2HB_DB_TYPE_QUORUMREGIONS,
1376 sizeof(o2hb_quorum_region_bitmap),
1378 o2hb_quorum_region_bitmap);
1379 if (!o2hb_debug_quorumregions) {
1384 o2hb_debug_failedregions =
1385 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1387 &o2hb_db_failedregions,
1388 sizeof(*o2hb_db_failedregions),
1389 O2HB_DB_TYPE_FAILEDREGIONS,
1390 sizeof(o2hb_failed_region_bitmap),
1392 o2hb_failed_region_bitmap);
1393 if (!o2hb_debug_failedregions) {
1410 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1411 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1413 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1414 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1416 INIT_LIST_HEAD(&o2hb_node_events);
1418 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1419 memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1420 memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1421 memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1422 memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1424 o2hb_dependent_users = 0;
1426 return o2hb_debug_init();
1429 /* if we're already in a callback then we're already serialized by the sem */
1430 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1433 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1435 memcpy(map, &o2hb_live_node_bitmap, bytes);
1439 * get a map of all nodes that are heartbeating in any regions
1441 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1443 /* callers want to serialize this map and callbacks so that they
1444 * can trust that they don't miss nodes coming to the party */
1445 down_read(&o2hb_callback_sem);
1446 spin_lock(&o2hb_live_lock);
1447 o2hb_fill_node_map_from_callback(map, bytes);
1448 spin_unlock(&o2hb_live_lock);
1449 up_read(&o2hb_callback_sem);
1451 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1454 * heartbeat configfs bits. The heartbeat set is a default set under
1455 * the cluster set in nodemanager.c.
1458 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1460 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1463 /* drop_item only drops its ref after killing the thread, nothing should
1464 * be using the region anymore. this has to clean up any state that
1465 * attributes might have built up. */
1466 static void o2hb_region_release(struct config_item *item)
1470 struct o2hb_region *reg = to_o2hb_region(item);
1472 mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1474 kfree(reg->hr_tmp_block);
1476 if (reg->hr_slot_data) {
1477 for (i = 0; i < reg->hr_num_pages; i++) {
1478 page = reg->hr_slot_data[i];
1482 kfree(reg->hr_slot_data);
1486 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1488 kfree(reg->hr_slots);
1490 kfree(reg->hr_db_regnum);
1491 kfree(reg->hr_db_livenodes);
1492 debugfs_remove(reg->hr_debug_livenodes);
1493 debugfs_remove(reg->hr_debug_regnum);
1494 debugfs_remove(reg->hr_debug_elapsed_time);
1495 debugfs_remove(reg->hr_debug_pinned);
1496 debugfs_remove(reg->hr_debug_dir);
1498 spin_lock(&o2hb_live_lock);
1499 list_del(®->hr_all_item);
1500 spin_unlock(&o2hb_live_lock);
1505 static int o2hb_read_block_input(struct o2hb_region *reg,
1508 unsigned long *ret_bytes,
1509 unsigned int *ret_bits)
1511 unsigned long bytes;
1512 char *p = (char *)page;
1514 bytes = simple_strtoul(p, &p, 0);
1515 if (!p || (*p && (*p != '\n')))
1518 /* Heartbeat and fs min / max block sizes are the same. */
1519 if (bytes > 4096 || bytes < 512)
1521 if (hweight16(bytes) != 1)
1527 *ret_bits = ffs(bytes) - 1;
1532 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1535 return sprintf(page, "%u\n", reg->hr_block_bytes);
1538 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1543 unsigned long block_bytes;
1544 unsigned int block_bits;
1549 status = o2hb_read_block_input(reg, page, count,
1550 &block_bytes, &block_bits);
1554 reg->hr_block_bytes = (unsigned int)block_bytes;
1555 reg->hr_block_bits = block_bits;
1560 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1563 return sprintf(page, "%llu\n", reg->hr_start_block);
1566 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1570 unsigned long long tmp;
1571 char *p = (char *)page;
1576 tmp = simple_strtoull(p, &p, 0);
1577 if (!p || (*p && (*p != '\n')))
1580 reg->hr_start_block = tmp;
1585 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1588 return sprintf(page, "%d\n", reg->hr_blocks);
1591 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1596 char *p = (char *)page;
1601 tmp = simple_strtoul(p, &p, 0);
1602 if (!p || (*p && (*p != '\n')))
1605 if (tmp > O2NM_MAX_NODES || tmp == 0)
1608 reg->hr_blocks = (unsigned int)tmp;
1613 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1616 unsigned int ret = 0;
1619 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1624 static void o2hb_init_region_params(struct o2hb_region *reg)
1626 reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1627 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1629 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1630 reg->hr_start_block, reg->hr_blocks);
1631 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1632 reg->hr_block_bytes, reg->hr_block_bits);
1633 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1634 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1637 static int o2hb_map_slot_data(struct o2hb_region *reg)
1640 unsigned int last_slot;
1641 unsigned int spp = reg->hr_slots_per_page;
1644 struct o2hb_disk_slot *slot;
1646 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1647 if (reg->hr_tmp_block == NULL) {
1648 mlog_errno(-ENOMEM);
1652 reg->hr_slots = kcalloc(reg->hr_blocks,
1653 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1654 if (reg->hr_slots == NULL) {
1655 mlog_errno(-ENOMEM);
1659 for(i = 0; i < reg->hr_blocks; i++) {
1660 slot = ®->hr_slots[i];
1661 slot->ds_node_num = i;
1662 INIT_LIST_HEAD(&slot->ds_live_item);
1663 slot->ds_raw_block = NULL;
1666 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1667 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1668 "at %u blocks per page\n",
1669 reg->hr_num_pages, reg->hr_blocks, spp);
1671 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1673 if (!reg->hr_slot_data) {
1674 mlog_errno(-ENOMEM);
1678 for(i = 0; i < reg->hr_num_pages; i++) {
1679 page = alloc_page(GFP_KERNEL);
1681 mlog_errno(-ENOMEM);
1685 reg->hr_slot_data[i] = page;
1687 last_slot = i * spp;
1688 raw = page_address(page);
1690 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1692 BUG_ON((j + last_slot) >= reg->hr_blocks);
1694 slot = ®->hr_slots[j + last_slot];
1695 slot->ds_raw_block =
1696 (struct o2hb_disk_heartbeat_block *) raw;
1698 raw += reg->hr_block_bytes;
1705 /* Read in all the slots available and populate the tracking
1706 * structures so that we can start with a baseline idea of what's
1708 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1711 struct o2hb_disk_slot *slot;
1712 struct o2hb_disk_heartbeat_block *hb_block;
1714 ret = o2hb_read_slots(reg, reg->hr_blocks);
1720 /* We only want to get an idea of the values initially in each
1721 * slot, so we do no verification - o2hb_check_slot will
1722 * actually determine if each configured slot is valid and
1723 * whether any values have changed. */
1724 for(i = 0; i < reg->hr_blocks; i++) {
1725 slot = ®->hr_slots[i];
1726 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1728 /* Only fill the values that o2hb_check_slot uses to
1729 * determine changing slots */
1730 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1731 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1738 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1739 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1743 struct task_struct *hb_task;
1746 char *p = (char *)page;
1748 struct inode *inode;
1749 ssize_t ret = -EINVAL;
1755 /* We can't heartbeat without having had our node number
1756 * configured yet. */
1757 if (o2nm_this_node() == O2NM_MAX_NODES)
1760 fd = simple_strtol(p, &p, 0);
1761 if (!p || (*p && (*p != '\n')))
1764 if (fd < 0 || fd >= INT_MAX)
1771 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1772 reg->hr_block_bytes == 0)
1775 inode = igrab(f.file->f_mapping->host);
1779 if (!S_ISBLK(inode->i_mode))
1782 reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
1783 ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
1785 reg->hr_bdev = NULL;
1790 bdevname(reg->hr_bdev, reg->hr_dev_name);
1792 sectsize = bdev_logical_block_size(reg->hr_bdev);
1793 if (sectsize != reg->hr_block_bytes) {
1795 "blocksize %u incorrect for device, expected %d",
1796 reg->hr_block_bytes, sectsize);
1801 o2hb_init_region_params(reg);
1803 /* Generation of zero is invalid */
1805 get_random_bytes(®->hr_generation,
1806 sizeof(reg->hr_generation));
1807 } while (reg->hr_generation == 0);
1809 ret = o2hb_map_slot_data(reg);
1815 ret = o2hb_populate_slot_data(reg);
1821 INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout);
1824 * A node is considered live after it has beat LIVE_THRESHOLD
1825 * times. We're not steady until we've given them a chance
1826 * _after_ our first read.
1827 * The default threshold is bare minimum so as to limit the delay
1828 * during mounts. For global heartbeat, the threshold doubled for the
1831 live_threshold = O2HB_LIVE_THRESHOLD;
1832 if (o2hb_global_heartbeat_active()) {
1833 spin_lock(&o2hb_live_lock);
1834 if (o2hb_pop_count(&o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1835 live_threshold <<= 1;
1836 spin_unlock(&o2hb_live_lock);
1839 atomic_set(®->hr_steady_iterations, live_threshold);
1840 /* unsteady_iterations is double the steady_iterations */
1841 atomic_set(®->hr_unsteady_iterations, (live_threshold << 1));
1843 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1844 reg->hr_item.ci_name);
1845 if (IS_ERR(hb_task)) {
1846 ret = PTR_ERR(hb_task);
1851 spin_lock(&o2hb_live_lock);
1852 reg->hr_task = hb_task;
1853 spin_unlock(&o2hb_live_lock);
1855 ret = wait_event_interruptible(o2hb_steady_queue,
1856 atomic_read(®->hr_steady_iterations) == 0);
1858 atomic_set(®->hr_steady_iterations, 0);
1859 reg->hr_aborted_start = 1;
1862 if (reg->hr_aborted_start) {
1867 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1868 spin_lock(&o2hb_live_lock);
1869 hb_task = reg->hr_task;
1870 if (o2hb_global_heartbeat_active())
1871 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1872 spin_unlock(&o2hb_live_lock);
1879 if (hb_task && o2hb_global_heartbeat_active())
1880 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1881 config_item_name(®->hr_item), reg->hr_dev_name);
1890 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1891 reg->hr_bdev = NULL;
1897 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1902 spin_lock(&o2hb_live_lock);
1904 pid = task_pid_nr(reg->hr_task);
1905 spin_unlock(&o2hb_live_lock);
1910 return sprintf(page, "%u\n", pid);
1913 struct o2hb_region_attribute {
1914 struct configfs_attribute attr;
1915 ssize_t (*show)(struct o2hb_region *, char *);
1916 ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1919 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1920 .attr = { .ca_owner = THIS_MODULE,
1921 .ca_name = "block_bytes",
1922 .ca_mode = S_IRUGO | S_IWUSR },
1923 .show = o2hb_region_block_bytes_read,
1924 .store = o2hb_region_block_bytes_write,
1927 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1928 .attr = { .ca_owner = THIS_MODULE,
1929 .ca_name = "start_block",
1930 .ca_mode = S_IRUGO | S_IWUSR },
1931 .show = o2hb_region_start_block_read,
1932 .store = o2hb_region_start_block_write,
1935 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1936 .attr = { .ca_owner = THIS_MODULE,
1937 .ca_name = "blocks",
1938 .ca_mode = S_IRUGO | S_IWUSR },
1939 .show = o2hb_region_blocks_read,
1940 .store = o2hb_region_blocks_write,
1943 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1944 .attr = { .ca_owner = THIS_MODULE,
1946 .ca_mode = S_IRUGO | S_IWUSR },
1947 .show = o2hb_region_dev_read,
1948 .store = o2hb_region_dev_write,
1951 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1952 .attr = { .ca_owner = THIS_MODULE,
1954 .ca_mode = S_IRUGO | S_IRUSR },
1955 .show = o2hb_region_pid_read,
1958 static struct configfs_attribute *o2hb_region_attrs[] = {
1959 &o2hb_region_attr_block_bytes.attr,
1960 &o2hb_region_attr_start_block.attr,
1961 &o2hb_region_attr_blocks.attr,
1962 &o2hb_region_attr_dev.attr,
1963 &o2hb_region_attr_pid.attr,
1967 static ssize_t o2hb_region_show(struct config_item *item,
1968 struct configfs_attribute *attr,
1971 struct o2hb_region *reg = to_o2hb_region(item);
1972 struct o2hb_region_attribute *o2hb_region_attr =
1973 container_of(attr, struct o2hb_region_attribute, attr);
1976 if (o2hb_region_attr->show)
1977 ret = o2hb_region_attr->show(reg, page);
1981 static ssize_t o2hb_region_store(struct config_item *item,
1982 struct configfs_attribute *attr,
1983 const char *page, size_t count)
1985 struct o2hb_region *reg = to_o2hb_region(item);
1986 struct o2hb_region_attribute *o2hb_region_attr =
1987 container_of(attr, struct o2hb_region_attribute, attr);
1988 ssize_t ret = -EINVAL;
1990 if (o2hb_region_attr->store)
1991 ret = o2hb_region_attr->store(reg, page, count);
1995 static struct configfs_item_operations o2hb_region_item_ops = {
1996 .release = o2hb_region_release,
1997 .show_attribute = o2hb_region_show,
1998 .store_attribute = o2hb_region_store,
2001 static struct config_item_type o2hb_region_type = {
2002 .ct_item_ops = &o2hb_region_item_ops,
2003 .ct_attrs = o2hb_region_attrs,
2004 .ct_owner = THIS_MODULE,
2009 struct o2hb_heartbeat_group {
2010 struct config_group hs_group;
2014 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
2017 container_of(group, struct o2hb_heartbeat_group, hs_group)
2021 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
2026 debugfs_create_dir(config_item_name(®->hr_item), dir);
2027 if (!reg->hr_debug_dir) {
2032 reg->hr_debug_livenodes =
2033 o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2035 &(reg->hr_db_livenodes),
2036 sizeof(*(reg->hr_db_livenodes)),
2037 O2HB_DB_TYPE_REGION_LIVENODES,
2038 sizeof(reg->hr_live_node_bitmap),
2039 O2NM_MAX_NODES, reg);
2040 if (!reg->hr_debug_livenodes) {
2045 reg->hr_debug_regnum =
2046 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2048 &(reg->hr_db_regnum),
2049 sizeof(*(reg->hr_db_regnum)),
2050 O2HB_DB_TYPE_REGION_NUMBER,
2051 0, O2NM_MAX_NODES, reg);
2052 if (!reg->hr_debug_regnum) {
2057 reg->hr_debug_elapsed_time =
2058 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2060 &(reg->hr_db_elapsed_time),
2061 sizeof(*(reg->hr_db_elapsed_time)),
2062 O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2064 if (!reg->hr_debug_elapsed_time) {
2069 reg->hr_debug_pinned =
2070 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2072 &(reg->hr_db_pinned),
2073 sizeof(*(reg->hr_db_pinned)),
2074 O2HB_DB_TYPE_REGION_PINNED,
2076 if (!reg->hr_debug_pinned) {
2086 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2089 struct o2hb_region *reg = NULL;
2092 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2094 return ERR_PTR(-ENOMEM);
2096 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2097 ret = -ENAMETOOLONG;
2101 spin_lock(&o2hb_live_lock);
2102 reg->hr_region_num = 0;
2103 if (o2hb_global_heartbeat_active()) {
2104 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2106 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2107 spin_unlock(&o2hb_live_lock);
2111 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2113 list_add_tail(®->hr_all_item, &o2hb_all_regions);
2114 spin_unlock(&o2hb_live_lock);
2116 config_item_init_type_name(®->hr_item, name, &o2hb_region_type);
2118 ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2120 config_item_put(®->hr_item);
2124 return ®->hr_item;
2127 return ERR_PTR(ret);
2130 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2131 struct config_item *item)
2133 struct task_struct *hb_task;
2134 struct o2hb_region *reg = to_o2hb_region(item);
2135 int quorum_region = 0;
2137 /* stop the thread when the user removes the region dir */
2138 spin_lock(&o2hb_live_lock);
2139 hb_task = reg->hr_task;
2140 reg->hr_task = NULL;
2141 reg->hr_item_dropped = 1;
2142 spin_unlock(&o2hb_live_lock);
2145 kthread_stop(hb_task);
2147 if (o2hb_global_heartbeat_active()) {
2148 spin_lock(&o2hb_live_lock);
2149 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2150 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2151 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2153 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2154 spin_unlock(&o2hb_live_lock);
2155 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2156 ((atomic_read(®->hr_steady_iterations) == 0) ?
2157 "stopped" : "start aborted"), config_item_name(item),
2162 * If we're racing a dev_write(), we need to wake them. They will
2163 * check reg->hr_task
2165 if (atomic_read(®->hr_steady_iterations) != 0) {
2166 reg->hr_aborted_start = 1;
2167 atomic_set(®->hr_steady_iterations, 0);
2168 wake_up(&o2hb_steady_queue);
2171 config_item_put(item);
2173 if (!o2hb_global_heartbeat_active() || !quorum_region)
2177 * If global heartbeat active and there are dependent users,
2178 * pin all regions if quorum region count <= CUT_OFF
2180 spin_lock(&o2hb_live_lock);
2182 if (!o2hb_dependent_users)
2185 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2186 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2187 o2hb_region_pin(NULL);
2190 spin_unlock(&o2hb_live_lock);
2193 struct o2hb_heartbeat_group_attribute {
2194 struct configfs_attribute attr;
2195 ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2196 ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2199 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2200 struct configfs_attribute *attr,
2203 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2204 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2205 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2208 if (o2hb_heartbeat_group_attr->show)
2209 ret = o2hb_heartbeat_group_attr->show(reg, page);
2213 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2214 struct configfs_attribute *attr,
2215 const char *page, size_t count)
2217 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2218 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2219 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2220 ssize_t ret = -EINVAL;
2222 if (o2hb_heartbeat_group_attr->store)
2223 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2227 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2230 return sprintf(page, "%u\n", o2hb_dead_threshold);
2233 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2238 char *p = (char *)page;
2240 tmp = simple_strtoul(p, &p, 10);
2241 if (!p || (*p && (*p != '\n')))
2244 /* this will validate ranges for us. */
2245 o2hb_dead_threshold_set((unsigned int) tmp);
2251 ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2254 return sprintf(page, "%s\n",
2255 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2259 ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2260 const char *page, size_t count)
2266 len = (page[count - 1] == '\n') ? count - 1 : count;
2270 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2271 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2274 ret = o2hb_global_heartbeat_mode_set(i);
2276 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2277 o2hb_heartbeat_mode_desc[i]);
2285 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2286 .attr = { .ca_owner = THIS_MODULE,
2287 .ca_name = "dead_threshold",
2288 .ca_mode = S_IRUGO | S_IWUSR },
2289 .show = o2hb_heartbeat_group_threshold_show,
2290 .store = o2hb_heartbeat_group_threshold_store,
2293 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2294 .attr = { .ca_owner = THIS_MODULE,
2296 .ca_mode = S_IRUGO | S_IWUSR },
2297 .show = o2hb_heartbeat_group_mode_show,
2298 .store = o2hb_heartbeat_group_mode_store,
2301 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2302 &o2hb_heartbeat_group_attr_threshold.attr,
2303 &o2hb_heartbeat_group_attr_mode.attr,
2307 static struct configfs_item_operations o2hb_heartbeat_group_item_ops = {
2308 .show_attribute = o2hb_heartbeat_group_show,
2309 .store_attribute = o2hb_heartbeat_group_store,
2312 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2313 .make_item = o2hb_heartbeat_group_make_item,
2314 .drop_item = o2hb_heartbeat_group_drop_item,
2317 static struct config_item_type o2hb_heartbeat_group_type = {
2318 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
2319 .ct_item_ops = &o2hb_heartbeat_group_item_ops,
2320 .ct_attrs = o2hb_heartbeat_group_attrs,
2321 .ct_owner = THIS_MODULE,
2324 /* this is just here to avoid touching group in heartbeat.h which the
2325 * entire damn world #includes */
2326 struct config_group *o2hb_alloc_hb_set(void)
2328 struct o2hb_heartbeat_group *hs = NULL;
2329 struct config_group *ret = NULL;
2331 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2335 config_group_init_type_name(&hs->hs_group, "heartbeat",
2336 &o2hb_heartbeat_group_type);
2338 ret = &hs->hs_group;
2345 void o2hb_free_hb_set(struct config_group *group)
2347 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2351 /* hb callback registration and issuing */
2353 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2355 if (type == O2HB_NUM_CB)
2356 return ERR_PTR(-EINVAL);
2358 return &o2hb_callbacks[type];
2361 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2362 enum o2hb_callback_type type,
2367 INIT_LIST_HEAD(&hc->hc_item);
2370 hc->hc_priority = priority;
2372 hc->hc_magic = O2HB_CB_MAGIC;
2374 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2377 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2378 * In global heartbeat mode, region_uuid passed is NULL.
2380 * In local, we only pin the matching region. In global we pin all the active
2383 static int o2hb_region_pin(const char *region_uuid)
2385 int ret = 0, found = 0;
2386 struct o2hb_region *reg;
2389 assert_spin_locked(&o2hb_live_lock);
2391 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2392 if (reg->hr_item_dropped)
2395 uuid = config_item_name(®->hr_item);
2397 /* local heartbeat */
2399 if (strcmp(region_uuid, uuid))
2404 if (reg->hr_item_pinned || reg->hr_item_dropped)
2407 /* Ignore ENOENT only for local hb (userdlm domain) */
2408 ret = o2nm_depend_item(®->hr_item);
2410 mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2411 reg->hr_item_pinned = 1;
2413 if (ret == -ENOENT && found)
2416 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2430 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2431 * In global heartbeat mode, region_uuid passed is NULL.
2433 * In local, we only unpin the matching region. In global we unpin all the
2436 static void o2hb_region_unpin(const char *region_uuid)
2438 struct o2hb_region *reg;
2442 assert_spin_locked(&o2hb_live_lock);
2444 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2445 if (reg->hr_item_dropped)
2448 uuid = config_item_name(®->hr_item);
2450 if (strcmp(region_uuid, uuid))
2455 if (reg->hr_item_pinned) {
2456 mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2457 o2nm_undepend_item(®->hr_item);
2458 reg->hr_item_pinned = 0;
2465 static int o2hb_region_inc_user(const char *region_uuid)
2469 spin_lock(&o2hb_live_lock);
2471 /* local heartbeat */
2472 if (!o2hb_global_heartbeat_active()) {
2473 ret = o2hb_region_pin(region_uuid);
2478 * if global heartbeat active and this is the first dependent user,
2479 * pin all regions if quorum region count <= CUT_OFF
2481 o2hb_dependent_users++;
2482 if (o2hb_dependent_users > 1)
2485 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2486 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2487 ret = o2hb_region_pin(NULL);
2490 spin_unlock(&o2hb_live_lock);
2494 void o2hb_region_dec_user(const char *region_uuid)
2496 spin_lock(&o2hb_live_lock);
2498 /* local heartbeat */
2499 if (!o2hb_global_heartbeat_active()) {
2500 o2hb_region_unpin(region_uuid);
2505 * if global heartbeat active and there are no dependent users,
2506 * unpin all quorum regions
2508 o2hb_dependent_users--;
2509 if (!o2hb_dependent_users)
2510 o2hb_region_unpin(NULL);
2513 spin_unlock(&o2hb_live_lock);
2516 int o2hb_register_callback(const char *region_uuid,
2517 struct o2hb_callback_func *hc)
2519 struct o2hb_callback_func *tmp;
2520 struct list_head *iter;
2521 struct o2hb_callback *hbcall;
2524 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2525 BUG_ON(!list_empty(&hc->hc_item));
2527 hbcall = hbcall_from_type(hc->hc_type);
2528 if (IS_ERR(hbcall)) {
2529 ret = PTR_ERR(hbcall);
2534 ret = o2hb_region_inc_user(region_uuid);
2541 down_write(&o2hb_callback_sem);
2543 list_for_each(iter, &hbcall->list) {
2544 tmp = list_entry(iter, struct o2hb_callback_func, hc_item);
2545 if (hc->hc_priority < tmp->hc_priority) {
2546 list_add_tail(&hc->hc_item, iter);
2550 if (list_empty(&hc->hc_item))
2551 list_add_tail(&hc->hc_item, &hbcall->list);
2553 up_write(&o2hb_callback_sem);
2556 mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2557 ret, __builtin_return_address(0), hc);
2560 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2562 void o2hb_unregister_callback(const char *region_uuid,
2563 struct o2hb_callback_func *hc)
2565 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2567 mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2568 __builtin_return_address(0), hc);
2570 /* XXX Can this happen _with_ a region reference? */
2571 if (list_empty(&hc->hc_item))
2575 o2hb_region_dec_user(region_uuid);
2577 down_write(&o2hb_callback_sem);
2579 list_del_init(&hc->hc_item);
2581 up_write(&o2hb_callback_sem);
2583 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2585 int o2hb_check_node_heartbeating(u8 node_num)
2587 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2589 o2hb_fill_node_map(testing_map, sizeof(testing_map));
2590 if (!test_bit(node_num, testing_map)) {
2592 "node (%u) does not have heartbeating enabled.\n",
2599 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2601 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2603 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2605 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2606 if (!test_bit(node_num, testing_map)) {
2608 "node (%u) does not have heartbeating enabled.\n",
2615 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2617 /* Makes sure our local node is configured with a node number, and is
2619 int o2hb_check_local_node_heartbeating(void)
2623 /* if this node was set then we have networking */
2624 node_num = o2nm_this_node();
2625 if (node_num == O2NM_MAX_NODES) {
2626 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2630 return o2hb_check_node_heartbeating(node_num);
2632 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2635 * this is just a hack until we get the plumbing which flips file systems
2636 * read only and drops the hb ref instead of killing the node dead.
2638 void o2hb_stop_all_regions(void)
2640 struct o2hb_region *reg;
2642 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2644 spin_lock(&o2hb_live_lock);
2646 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2647 reg->hr_unclean_stop = 1;
2649 spin_unlock(&o2hb_live_lock);
2651 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2653 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2655 struct o2hb_region *reg;
2659 spin_lock(&o2hb_live_lock);
2662 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2663 if (reg->hr_item_dropped)
2666 mlog(0, "Region: %s\n", config_item_name(®->hr_item));
2667 if (numregs < max_regions) {
2668 memcpy(p, config_item_name(®->hr_item),
2669 O2HB_MAX_REGION_NAME_LEN);
2670 p += O2HB_MAX_REGION_NAME_LEN;
2675 spin_unlock(&o2hb_live_lock);
2679 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2681 int o2hb_global_heartbeat_active(void)
2683 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2685 EXPORT_SYMBOL(o2hb_global_heartbeat_active);