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>
38 #include <linux/bitmap.h>
40 #include "heartbeat.h"
42 #include "nodemanager.h"
49 * The first heartbeat pass had one global thread that would serialize all hb
50 * callback calls. This global serializing sem should only be removed once
51 * we've made sure that all callees can deal with being called concurrently
52 * from multiple hb region threads.
54 static DECLARE_RWSEM(o2hb_callback_sem);
57 * multiple hb threads are watching multiple regions. A node is live
58 * whenever any of the threads sees activity from the node in its region.
60 static DEFINE_SPINLOCK(o2hb_live_lock);
61 static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
62 static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
63 static LIST_HEAD(o2hb_node_events);
64 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
67 * In global heartbeat, we maintain a series of region bitmaps.
68 * - o2hb_region_bitmap allows us to limit the region number to max region.
69 * - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
70 * - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
72 * - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
74 static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
75 static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
76 static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
77 static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
79 #define O2HB_DB_TYPE_LIVENODES 0
80 #define O2HB_DB_TYPE_LIVEREGIONS 1
81 #define O2HB_DB_TYPE_QUORUMREGIONS 2
82 #define O2HB_DB_TYPE_FAILEDREGIONS 3
83 #define O2HB_DB_TYPE_REGION_LIVENODES 4
84 #define O2HB_DB_TYPE_REGION_NUMBER 5
85 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME 6
86 #define O2HB_DB_TYPE_REGION_PINNED 7
87 struct o2hb_debug_buf {
94 static struct o2hb_debug_buf *o2hb_db_livenodes;
95 static struct o2hb_debug_buf *o2hb_db_liveregions;
96 static struct o2hb_debug_buf *o2hb_db_quorumregions;
97 static struct o2hb_debug_buf *o2hb_db_failedregions;
99 #define O2HB_DEBUG_DIR "o2hb"
100 #define O2HB_DEBUG_LIVENODES "livenodes"
101 #define O2HB_DEBUG_LIVEREGIONS "live_regions"
102 #define O2HB_DEBUG_QUORUMREGIONS "quorum_regions"
103 #define O2HB_DEBUG_FAILEDREGIONS "failed_regions"
104 #define O2HB_DEBUG_REGION_NUMBER "num"
105 #define O2HB_DEBUG_REGION_ELAPSED_TIME "elapsed_time_in_ms"
106 #define O2HB_DEBUG_REGION_PINNED "pinned"
108 static struct dentry *o2hb_debug_dir;
109 static struct dentry *o2hb_debug_livenodes;
110 static struct dentry *o2hb_debug_liveregions;
111 static struct dentry *o2hb_debug_quorumregions;
112 static struct dentry *o2hb_debug_failedregions;
114 static LIST_HEAD(o2hb_all_regions);
116 static struct o2hb_callback {
117 struct list_head list;
118 } o2hb_callbacks[O2HB_NUM_CB];
120 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
122 #define O2HB_DEFAULT_BLOCK_BITS 9
124 enum o2hb_heartbeat_modes {
125 O2HB_HEARTBEAT_LOCAL = 0,
126 O2HB_HEARTBEAT_GLOBAL,
127 O2HB_HEARTBEAT_NUM_MODES,
130 char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
131 "local", /* O2HB_HEARTBEAT_LOCAL */
132 "global", /* O2HB_HEARTBEAT_GLOBAL */
135 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
136 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
139 * o2hb_dependent_users tracks the number of registered callbacks that depend
140 * on heartbeat. o2net and o2dlm are two entities that register this callback.
141 * However only o2dlm depends on the heartbeat. It does not want the heartbeat
142 * to stop while a dlm domain is still active.
144 unsigned int o2hb_dependent_users;
147 * In global heartbeat mode, all regions are pinned if there are one or more
148 * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
149 * regions are unpinned if the region count exceeds the cut off or the number
150 * of dependent users falls to zero.
152 #define O2HB_PIN_CUT_OFF 3
155 * In local heartbeat mode, we assume the dlm domain name to be the same as
156 * region uuid. This is true for domains created for the file system but not
157 * necessarily true for userdlm domains. This is a known limitation.
159 * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
160 * works for both file system and userdlm domains.
162 static int o2hb_region_pin(const char *region_uuid);
163 static void o2hb_region_unpin(const char *region_uuid);
165 /* Only sets a new threshold if there are no active regions.
167 * No locking or otherwise interesting code is required for reading
168 * o2hb_dead_threshold as it can't change once regions are active and
169 * it's not interesting to anyone until then anyway. */
170 static void o2hb_dead_threshold_set(unsigned int threshold)
172 if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
173 spin_lock(&o2hb_live_lock);
174 if (list_empty(&o2hb_all_regions))
175 o2hb_dead_threshold = threshold;
176 spin_unlock(&o2hb_live_lock);
180 static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
184 if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
185 spin_lock(&o2hb_live_lock);
186 if (list_empty(&o2hb_all_regions)) {
187 o2hb_heartbeat_mode = hb_mode;
190 spin_unlock(&o2hb_live_lock);
196 struct o2hb_node_event {
197 struct list_head hn_item;
198 enum o2hb_callback_type hn_event_type;
199 struct o2nm_node *hn_node;
203 struct o2hb_disk_slot {
204 struct o2hb_disk_heartbeat_block *ds_raw_block;
207 u64 ds_last_generation;
208 u16 ds_equal_samples;
209 u16 ds_changed_samples;
210 struct list_head ds_live_item;
213 /* each thread owns a region.. when we're asked to tear down the region
214 * we ask the thread to stop, who cleans up the region */
216 struct config_item hr_item;
218 struct list_head hr_all_item;
219 unsigned hr_unclean_stop:1,
224 /* protected by the hr_callback_sem */
225 struct task_struct *hr_task;
227 unsigned int hr_blocks;
228 unsigned long long hr_start_block;
230 unsigned int hr_block_bits;
231 unsigned int hr_block_bytes;
233 unsigned int hr_slots_per_page;
234 unsigned int hr_num_pages;
236 struct page **hr_slot_data;
237 struct block_device *hr_bdev;
238 struct o2hb_disk_slot *hr_slots;
240 /* live node map of this region */
241 unsigned long hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
242 unsigned int hr_region_num;
244 struct dentry *hr_debug_dir;
245 struct dentry *hr_debug_livenodes;
246 struct dentry *hr_debug_regnum;
247 struct dentry *hr_debug_elapsed_time;
248 struct dentry *hr_debug_pinned;
249 struct o2hb_debug_buf *hr_db_livenodes;
250 struct o2hb_debug_buf *hr_db_regnum;
251 struct o2hb_debug_buf *hr_db_elapsed_time;
252 struct o2hb_debug_buf *hr_db_pinned;
254 /* let the person setting up hb wait for it to return until it
255 * has reached a 'steady' state. This will be fixed when we have
256 * a more complete api that doesn't lead to this sort of fragility. */
257 atomic_t hr_steady_iterations;
259 /* terminate o2hb thread if it does not reach steady state
260 * (hr_steady_iterations == 0) within hr_unsteady_iterations */
261 atomic_t hr_unsteady_iterations;
263 char hr_dev_name[BDEVNAME_SIZE];
265 unsigned int hr_timeout_ms;
267 /* randomized as the region goes up and down so that a node
268 * recognizes a node going up and down in one iteration */
271 struct delayed_work hr_write_timeout_work;
272 unsigned long hr_last_timeout_start;
274 /* Used during o2hb_check_slot to hold a copy of the block
275 * being checked because we temporarily have to zero out the
277 struct o2hb_disk_heartbeat_block *hr_tmp_block;
280 struct o2hb_bio_wait_ctxt {
281 atomic_t wc_num_reqs;
282 struct completion wc_io_complete;
286 static void o2hb_write_timeout(struct work_struct *work)
290 struct o2hb_region *reg =
291 container_of(work, struct o2hb_region,
292 hr_write_timeout_work.work);
294 mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
295 "milliseconds\n", reg->hr_dev_name,
296 jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
298 if (o2hb_global_heartbeat_active()) {
299 spin_lock_irqsave(&o2hb_live_lock, flags);
300 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
301 set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
302 failed = bitmap_weight(o2hb_failed_region_bitmap,
304 quorum = bitmap_weight(o2hb_quorum_region_bitmap,
306 spin_unlock_irqrestore(&o2hb_live_lock, flags);
308 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
312 * Fence if the number of failed regions >= half the number
315 if ((failed << 1) < quorum)
319 o2quo_disk_timeout();
322 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
324 /* Arm writeout only after thread reaches steady state */
325 if (atomic_read(®->hr_steady_iterations) != 0)
328 mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
329 O2HB_MAX_WRITE_TIMEOUT_MS);
331 if (o2hb_global_heartbeat_active()) {
332 spin_lock(&o2hb_live_lock);
333 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
334 spin_unlock(&o2hb_live_lock);
336 cancel_delayed_work(®->hr_write_timeout_work);
337 reg->hr_last_timeout_start = jiffies;
338 schedule_delayed_work(®->hr_write_timeout_work,
339 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
342 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
344 cancel_delayed_work_sync(®->hr_write_timeout_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 o2hb_bio_wait_dec(wc, 1);
372 wait_for_completion(&wc->wc_io_complete);
375 static void o2hb_bio_end_io(struct bio *bio,
378 struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
381 mlog(ML_ERROR, "IO Error %d\n", error);
382 wc->wc_error = error;
385 o2hb_bio_wait_dec(wc, 1);
389 /* Setup a Bio to cover I/O against num_slots slots starting at
391 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
392 struct o2hb_bio_wait_ctxt *wc,
393 unsigned int *current_slot,
394 unsigned int max_slots)
396 int len, current_page;
397 unsigned int vec_len, vec_start;
398 unsigned int bits = reg->hr_block_bits;
399 unsigned int spp = reg->hr_slots_per_page;
400 unsigned int cs = *current_slot;
404 /* Testing has shown this allocation to take long enough under
405 * GFP_KERNEL that the local node can get fenced. It would be
406 * nicest if we could pre-allocate these bios and avoid this
408 bio = bio_alloc(GFP_ATOMIC, 16);
410 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
411 bio = ERR_PTR(-ENOMEM);
415 /* Must put everything in 512 byte sectors for the bio... */
416 bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
417 bio->bi_bdev = reg->hr_bdev;
418 bio->bi_private = wc;
419 bio->bi_end_io = o2hb_bio_end_io;
421 vec_start = (cs << bits) % PAGE_CACHE_SIZE;
422 while(cs < max_slots) {
423 current_page = cs / spp;
424 page = reg->hr_slot_data[current_page];
426 vec_len = min(PAGE_CACHE_SIZE - vec_start,
427 (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
429 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
430 current_page, vec_len, vec_start);
432 len = bio_add_page(bio, page, vec_len, vec_start);
433 if (len != vec_len) break;
435 cs += vec_len / (PAGE_CACHE_SIZE/spp);
444 static int o2hb_read_slots(struct o2hb_region *reg,
445 unsigned int max_slots)
447 unsigned int current_slot=0;
449 struct o2hb_bio_wait_ctxt wc;
452 o2hb_bio_wait_init(&wc);
454 while(current_slot < max_slots) {
455 bio = o2hb_setup_one_bio(reg, &wc, ¤t_slot, max_slots);
457 status = PTR_ERR(bio);
462 atomic_inc(&wc.wc_num_reqs);
463 submit_bio(READ, bio);
469 o2hb_wait_on_io(reg, &wc);
470 if (wc.wc_error && !status)
471 status = wc.wc_error;
476 static int o2hb_issue_node_write(struct o2hb_region *reg,
477 struct o2hb_bio_wait_ctxt *write_wc)
483 o2hb_bio_wait_init(write_wc);
485 slot = o2nm_this_node();
487 bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
489 status = PTR_ERR(bio);
494 atomic_inc(&write_wc->wc_num_reqs);
495 submit_bio(WRITE_SYNC, bio);
502 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
503 struct o2hb_disk_heartbeat_block *hb_block)
508 /* We want to compute the block crc with a 0 value in the
509 * hb_cksum field. Save it off here and replace after the
511 old_cksum = hb_block->hb_cksum;
512 hb_block->hb_cksum = 0;
514 ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
516 hb_block->hb_cksum = old_cksum;
521 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
523 mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
524 "cksum = 0x%x, generation 0x%llx\n",
525 (long long)le64_to_cpu(hb_block->hb_seq),
526 hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
527 (long long)le64_to_cpu(hb_block->hb_generation));
530 static int o2hb_verify_crc(struct o2hb_region *reg,
531 struct o2hb_disk_heartbeat_block *hb_block)
535 read = le32_to_cpu(hb_block->hb_cksum);
536 computed = o2hb_compute_block_crc_le(reg, hb_block);
538 return read == computed;
542 * Compare the slot data with what we wrote in the last iteration.
543 * If the match fails, print an appropriate error message. This is to
544 * detect errors like... another node hearting on the same slot,
545 * flaky device that is losing writes, etc.
546 * Returns 1 if check succeeds, 0 otherwise.
548 static int o2hb_check_own_slot(struct o2hb_region *reg)
550 struct o2hb_disk_slot *slot;
551 struct o2hb_disk_heartbeat_block *hb_block;
554 slot = ®->hr_slots[o2nm_this_node()];
555 /* Don't check on our 1st timestamp */
556 if (!slot->ds_last_time)
559 hb_block = slot->ds_raw_block;
560 if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
561 le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
562 hb_block->hb_node == slot->ds_node_num)
565 #define ERRSTR1 "Another node is heartbeating on device"
566 #define ERRSTR2 "Heartbeat generation mismatch on device"
567 #define ERRSTR3 "Heartbeat sequence mismatch on device"
569 if (hb_block->hb_node != slot->ds_node_num)
571 else if (le64_to_cpu(hb_block->hb_generation) !=
572 slot->ds_last_generation)
577 mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
578 "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
579 slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
580 (unsigned long long)slot->ds_last_time, hb_block->hb_node,
581 (unsigned long long)le64_to_cpu(hb_block->hb_generation),
582 (unsigned long long)le64_to_cpu(hb_block->hb_seq));
587 static inline void o2hb_prepare_block(struct o2hb_region *reg,
592 struct o2hb_disk_slot *slot;
593 struct o2hb_disk_heartbeat_block *hb_block;
595 node_num = o2nm_this_node();
596 slot = ®->hr_slots[node_num];
598 hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
599 memset(hb_block, 0, reg->hr_block_bytes);
600 /* TODO: time stuff */
601 cputime = CURRENT_TIME.tv_sec;
605 hb_block->hb_seq = cpu_to_le64(cputime);
606 hb_block->hb_node = node_num;
607 hb_block->hb_generation = cpu_to_le64(generation);
608 hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
610 /* This step must always happen last! */
611 hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
614 mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
615 (long long)generation,
616 le32_to_cpu(hb_block->hb_cksum));
619 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
620 struct o2nm_node *node,
623 struct o2hb_callback_func *f;
625 list_for_each_entry(f, &hbcall->list, hc_item) {
626 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
627 (f->hc_func)(node, idx, f->hc_data);
631 /* Will run the list in order until we process the passed event */
632 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
634 struct o2hb_callback *hbcall;
635 struct o2hb_node_event *event;
637 /* Holding callback sem assures we don't alter the callback
638 * lists when doing this, and serializes ourselves with other
639 * processes wanting callbacks. */
640 down_write(&o2hb_callback_sem);
642 spin_lock(&o2hb_live_lock);
643 while (!list_empty(&o2hb_node_events)
644 && !list_empty(&queued_event->hn_item)) {
645 event = list_entry(o2hb_node_events.next,
646 struct o2hb_node_event,
648 list_del_init(&event->hn_item);
649 spin_unlock(&o2hb_live_lock);
651 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
652 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
655 hbcall = hbcall_from_type(event->hn_event_type);
657 /* We should *never* have gotten on to the list with a
658 * bad type... This isn't something that we should try
659 * to recover from. */
660 BUG_ON(IS_ERR(hbcall));
662 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
664 spin_lock(&o2hb_live_lock);
666 spin_unlock(&o2hb_live_lock);
668 up_write(&o2hb_callback_sem);
671 static void o2hb_queue_node_event(struct o2hb_node_event *event,
672 enum o2hb_callback_type type,
673 struct o2nm_node *node,
676 assert_spin_locked(&o2hb_live_lock);
678 BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
680 event->hn_event_type = type;
681 event->hn_node = node;
682 event->hn_node_num = node_num;
684 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
685 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
687 list_add_tail(&event->hn_item, &o2hb_node_events);
690 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
692 struct o2hb_node_event event =
693 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
694 struct o2nm_node *node;
697 node = o2nm_get_node_by_num(slot->ds_node_num);
701 spin_lock(&o2hb_live_lock);
702 if (!list_empty(&slot->ds_live_item)) {
703 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
706 list_del_init(&slot->ds_live_item);
708 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
709 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
711 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
716 spin_unlock(&o2hb_live_lock);
719 o2hb_run_event_list(&event);
724 static void o2hb_set_quorum_device(struct o2hb_region *reg)
726 if (!o2hb_global_heartbeat_active())
729 /* Prevent race with o2hb_heartbeat_group_drop_item() */
730 if (kthread_should_stop())
733 /* Tag region as quorum only after thread reaches steady state */
734 if (atomic_read(®->hr_steady_iterations) != 0)
737 spin_lock(&o2hb_live_lock);
739 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
743 * A region can be added to the quorum only when it sees all
744 * live nodes heartbeat on it. In other words, the region has been
745 * added to all nodes.
747 if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
748 sizeof(o2hb_live_node_bitmap)))
751 printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
752 config_item_name(®->hr_item), reg->hr_dev_name);
754 set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
757 * If global heartbeat active, unpin all regions if the
758 * region count > CUT_OFF
760 if (bitmap_weight(o2hb_quorum_region_bitmap,
761 O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
762 o2hb_region_unpin(NULL);
764 spin_unlock(&o2hb_live_lock);
767 static int o2hb_check_slot(struct o2hb_region *reg,
768 struct o2hb_disk_slot *slot)
770 int changed = 0, gen_changed = 0;
771 struct o2hb_node_event event =
772 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
773 struct o2nm_node *node;
774 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
776 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
777 unsigned int slot_dead_ms;
781 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
784 * If a node is no longer configured but is still in the livemap, we
785 * may need to clear that bit from the livemap.
787 node = o2nm_get_node_by_num(slot->ds_node_num);
789 spin_lock(&o2hb_live_lock);
790 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
791 spin_unlock(&o2hb_live_lock);
796 if (!o2hb_verify_crc(reg, hb_block)) {
797 /* all paths from here will drop o2hb_live_lock for
799 spin_lock(&o2hb_live_lock);
801 /* Don't print an error on the console in this case -
802 * a freshly formatted heartbeat area will not have a
804 if (list_empty(&slot->ds_live_item))
807 /* The node is live but pushed out a bad crc. We
808 * consider it a transient miss but don't populate any
809 * other values as they may be junk. */
810 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
811 slot->ds_node_num, reg->hr_dev_name);
812 o2hb_dump_slot(hb_block);
814 slot->ds_equal_samples++;
818 /* we don't care if these wrap.. the state transitions below
819 * clear at the right places */
820 cputime = le64_to_cpu(hb_block->hb_seq);
821 if (slot->ds_last_time != cputime)
822 slot->ds_changed_samples++;
824 slot->ds_equal_samples++;
825 slot->ds_last_time = cputime;
827 /* The node changed heartbeat generations. We assume this to
828 * mean it dropped off but came back before we timed out. We
829 * want to consider it down for the time being but don't want
830 * to lose any changed_samples state we might build up to
831 * considering it live again. */
832 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
834 slot->ds_equal_samples = 0;
835 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
836 "to 0x%llx)\n", slot->ds_node_num,
837 (long long)slot->ds_last_generation,
838 (long long)le64_to_cpu(hb_block->hb_generation));
841 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
843 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
844 "seq %llu last %llu changed %u equal %u\n",
845 slot->ds_node_num, (long long)slot->ds_last_generation,
846 le32_to_cpu(hb_block->hb_cksum),
847 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
848 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
849 slot->ds_equal_samples);
851 spin_lock(&o2hb_live_lock);
854 /* dead nodes only come to life after some number of
855 * changes at any time during their dead time */
856 if (list_empty(&slot->ds_live_item) &&
857 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
858 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
859 slot->ds_node_num, (long long)slot->ds_last_generation);
861 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
863 /* first on the list generates a callback */
864 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
865 mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
866 "bitmap\n", slot->ds_node_num);
867 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
869 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
876 list_add_tail(&slot->ds_live_item,
877 &o2hb_live_slots[slot->ds_node_num]);
879 slot->ds_equal_samples = 0;
881 /* We want to be sure that all nodes agree on the
882 * number of milliseconds before a node will be
883 * considered dead. The self-fencing timeout is
884 * computed from this value, and a discrepancy might
885 * result in heartbeat calling a node dead when it
886 * hasn't self-fenced yet. */
887 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
888 if (slot_dead_ms && slot_dead_ms != dead_ms) {
889 /* TODO: Perhaps we can fail the region here. */
890 mlog(ML_ERROR, "Node %d on device %s has a dead count "
891 "of %u ms, but our count is %u ms.\n"
892 "Please double check your configuration values "
893 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
894 slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
900 /* if the list is dead, we're done.. */
901 if (list_empty(&slot->ds_live_item))
904 /* live nodes only go dead after enough consequtive missed
905 * samples.. reset the missed counter whenever we see
907 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
908 mlog(ML_HEARTBEAT, "Node %d left my region\n",
911 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
913 /* last off the live_slot generates a callback */
914 list_del_init(&slot->ds_live_item);
915 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
916 mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
917 "nodes bitmap\n", slot->ds_node_num);
918 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
920 /* node can be null */
921 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
922 node, slot->ds_node_num);
928 /* We don't clear this because the node is still
929 * actually writing new blocks. */
931 slot->ds_changed_samples = 0;
934 if (slot->ds_changed_samples) {
935 slot->ds_changed_samples = 0;
936 slot->ds_equal_samples = 0;
939 spin_unlock(&o2hb_live_lock);
942 o2hb_run_event_list(&event);
949 /* This could be faster if we just implmented a find_last_bit, but I
950 * don't think the circumstances warrant it. */
951 static int o2hb_highest_node(unsigned long *nodes,
958 while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
968 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
970 int i, ret, highest_node;
971 int membership_change = 0, own_slot_ok = 0;
972 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
973 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
974 struct o2hb_bio_wait_ctxt write_wc;
976 ret = o2nm_configured_node_map(configured_nodes,
977 sizeof(configured_nodes));
984 * If a node is not configured but is in the livemap, we still need
985 * to read the slot so as to be able to remove it from the livemap.
987 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
989 while ((i = find_next_bit(live_node_bitmap,
990 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
991 set_bit(i, configured_nodes);
994 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
995 if (highest_node >= O2NM_MAX_NODES) {
996 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1001 /* No sense in reading the slots of nodes that don't exist
1002 * yet. Of course, if the node definitions have holes in them
1003 * then we're reading an empty slot anyway... Consider this
1005 ret = o2hb_read_slots(reg, highest_node + 1);
1011 /* With an up to date view of the slots, we can check that no
1012 * other node has been improperly configured to heartbeat in
1014 own_slot_ok = o2hb_check_own_slot(reg);
1016 /* fill in the proper info for our next heartbeat */
1017 o2hb_prepare_block(reg, reg->hr_generation);
1019 ret = o2hb_issue_node_write(reg, &write_wc);
1026 while((i = find_next_bit(configured_nodes,
1027 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1028 membership_change |= o2hb_check_slot(reg, ®->hr_slots[i]);
1032 * We have to be sure we've advertised ourselves on disk
1033 * before we can go to steady state. This ensures that
1034 * people we find in our steady state have seen us.
1036 o2hb_wait_on_io(reg, &write_wc);
1037 if (write_wc.wc_error) {
1038 /* Do not re-arm the write timeout on I/O error - we
1039 * can't be sure that the new block ever made it to
1041 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1042 write_wc.wc_error, reg->hr_dev_name);
1043 ret = write_wc.wc_error;
1047 /* Skip disarming the timeout if own slot has stale/bad data */
1049 o2hb_set_quorum_device(reg);
1050 o2hb_arm_write_timeout(reg);
1054 /* let the person who launched us know when things are steady */
1055 if (atomic_read(®->hr_steady_iterations) != 0) {
1056 if (!ret && own_slot_ok && !membership_change) {
1057 if (atomic_dec_and_test(®->hr_steady_iterations))
1058 wake_up(&o2hb_steady_queue);
1062 if (atomic_read(®->hr_steady_iterations) != 0) {
1063 if (atomic_dec_and_test(®->hr_unsteady_iterations)) {
1064 printk(KERN_NOTICE "o2hb: Unable to stabilize "
1065 "heartbeart on region %s (%s)\n",
1066 config_item_name(®->hr_item),
1068 atomic_set(®->hr_steady_iterations, 0);
1069 reg->hr_aborted_start = 1;
1070 wake_up(&o2hb_steady_queue);
1078 /* Subtract b from a, storing the result in a. a *must* have a larger
1080 static void o2hb_tv_subtract(struct timeval *a,
1083 /* just return 0 when a is after b */
1084 if (a->tv_sec < b->tv_sec ||
1085 (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1091 a->tv_sec -= b->tv_sec;
1092 a->tv_usec -= b->tv_usec;
1093 while ( a->tv_usec < 0 ) {
1095 a->tv_usec += 1000000;
1099 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1100 struct timeval *end)
1102 struct timeval res = *end;
1104 o2hb_tv_subtract(&res, start);
1106 return res.tv_sec * 1000 + res.tv_usec / 1000;
1110 * we ride the region ref that the region dir holds. before the region
1111 * dir is removed and drops it ref it will wait to tear down this
1114 static int o2hb_thread(void *data)
1117 struct o2hb_region *reg = data;
1118 struct o2hb_bio_wait_ctxt write_wc;
1119 struct timeval before_hb, after_hb;
1120 unsigned int elapsed_msec;
1122 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1124 set_user_nice(current, -20);
1127 o2nm_depend_this_node();
1129 while (!kthread_should_stop() &&
1130 !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1131 /* We track the time spent inside
1132 * o2hb_do_disk_heartbeat so that we avoid more than
1133 * hr_timeout_ms between disk writes. On busy systems
1134 * this should result in a heartbeat which is less
1135 * likely to time itself out. */
1136 do_gettimeofday(&before_hb);
1138 ret = o2hb_do_disk_heartbeat(reg);
1140 do_gettimeofday(&after_hb);
1141 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1144 "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
1145 before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1146 after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1149 if (!kthread_should_stop() &&
1150 elapsed_msec < reg->hr_timeout_ms) {
1151 /* the kthread api has blocked signals for us so no
1152 * need to record the return value. */
1153 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1157 o2hb_disarm_write_timeout(reg);
1159 /* unclean stop is only used in very bad situation */
1160 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1161 o2hb_shutdown_slot(®->hr_slots[i]);
1163 /* Explicit down notification - avoid forcing the other nodes
1164 * to timeout on this region when we could just as easily
1165 * write a clear generation - thus indicating to them that
1166 * this node has left this region.
1168 if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1169 o2hb_prepare_block(reg, 0);
1170 ret = o2hb_issue_node_write(reg, &write_wc);
1172 o2hb_wait_on_io(reg, &write_wc);
1178 o2nm_undepend_this_node();
1180 mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1185 #ifdef CONFIG_DEBUG_FS
1186 static int o2hb_debug_open(struct inode *inode, struct file *file)
1188 struct o2hb_debug_buf *db = inode->i_private;
1189 struct o2hb_region *reg;
1190 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1196 /* max_nodes should be the largest bitmap we pass here */
1197 BUG_ON(sizeof(map) < db->db_size);
1199 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1203 switch (db->db_type) {
1204 case O2HB_DB_TYPE_LIVENODES:
1205 case O2HB_DB_TYPE_LIVEREGIONS:
1206 case O2HB_DB_TYPE_QUORUMREGIONS:
1207 case O2HB_DB_TYPE_FAILEDREGIONS:
1208 spin_lock(&o2hb_live_lock);
1209 memcpy(map, db->db_data, db->db_size);
1210 spin_unlock(&o2hb_live_lock);
1213 case O2HB_DB_TYPE_REGION_LIVENODES:
1214 spin_lock(&o2hb_live_lock);
1215 reg = (struct o2hb_region *)db->db_data;
1216 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1217 spin_unlock(&o2hb_live_lock);
1220 case O2HB_DB_TYPE_REGION_NUMBER:
1221 reg = (struct o2hb_region *)db->db_data;
1222 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1223 reg->hr_region_num);
1226 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1227 reg = (struct o2hb_region *)db->db_data;
1228 lts = reg->hr_last_timeout_start;
1229 /* If 0, it has never been set before */
1231 lts = jiffies_to_msecs(jiffies - lts);
1232 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1235 case O2HB_DB_TYPE_REGION_PINNED:
1236 reg = (struct o2hb_region *)db->db_data;
1237 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1238 !!reg->hr_item_pinned);
1245 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1246 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1247 out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1250 i_size_write(inode, out);
1252 file->private_data = buf;
1259 static int o2hb_debug_release(struct inode *inode, struct file *file)
1261 kfree(file->private_data);
1265 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1266 size_t nbytes, loff_t *ppos)
1268 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1269 i_size_read(file->f_mapping->host));
1272 static int o2hb_debug_open(struct inode *inode, struct file *file)
1276 static int o2hb_debug_release(struct inode *inode, struct file *file)
1280 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1281 size_t nbytes, loff_t *ppos)
1285 #endif /* CONFIG_DEBUG_FS */
1287 static const struct file_operations o2hb_debug_fops = {
1288 .open = o2hb_debug_open,
1289 .release = o2hb_debug_release,
1290 .read = o2hb_debug_read,
1291 .llseek = generic_file_llseek,
1294 void o2hb_exit(void)
1296 kfree(o2hb_db_livenodes);
1297 kfree(o2hb_db_liveregions);
1298 kfree(o2hb_db_quorumregions);
1299 kfree(o2hb_db_failedregions);
1300 debugfs_remove(o2hb_debug_failedregions);
1301 debugfs_remove(o2hb_debug_quorumregions);
1302 debugfs_remove(o2hb_debug_liveregions);
1303 debugfs_remove(o2hb_debug_livenodes);
1304 debugfs_remove(o2hb_debug_dir);
1307 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1308 struct o2hb_debug_buf **db, int db_len,
1309 int type, int size, int len, void *data)
1311 *db = kmalloc(db_len, GFP_KERNEL);
1315 (*db)->db_type = type;
1316 (*db)->db_size = size;
1317 (*db)->db_len = len;
1318 (*db)->db_data = data;
1320 return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1324 static int o2hb_debug_init(void)
1328 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1329 if (!o2hb_debug_dir) {
1334 o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1337 sizeof(*o2hb_db_livenodes),
1338 O2HB_DB_TYPE_LIVENODES,
1339 sizeof(o2hb_live_node_bitmap),
1341 o2hb_live_node_bitmap);
1342 if (!o2hb_debug_livenodes) {
1347 o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1349 &o2hb_db_liveregions,
1350 sizeof(*o2hb_db_liveregions),
1351 O2HB_DB_TYPE_LIVEREGIONS,
1352 sizeof(o2hb_live_region_bitmap),
1354 o2hb_live_region_bitmap);
1355 if (!o2hb_debug_liveregions) {
1360 o2hb_debug_quorumregions =
1361 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1363 &o2hb_db_quorumregions,
1364 sizeof(*o2hb_db_quorumregions),
1365 O2HB_DB_TYPE_QUORUMREGIONS,
1366 sizeof(o2hb_quorum_region_bitmap),
1368 o2hb_quorum_region_bitmap);
1369 if (!o2hb_debug_quorumregions) {
1374 o2hb_debug_failedregions =
1375 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1377 &o2hb_db_failedregions,
1378 sizeof(*o2hb_db_failedregions),
1379 O2HB_DB_TYPE_FAILEDREGIONS,
1380 sizeof(o2hb_failed_region_bitmap),
1382 o2hb_failed_region_bitmap);
1383 if (!o2hb_debug_failedregions) {
1400 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1401 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1403 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1404 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1406 INIT_LIST_HEAD(&o2hb_node_events);
1408 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1409 memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1410 memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1411 memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1412 memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1414 o2hb_dependent_users = 0;
1416 return o2hb_debug_init();
1419 /* if we're already in a callback then we're already serialized by the sem */
1420 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1423 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1425 memcpy(map, &o2hb_live_node_bitmap, bytes);
1429 * get a map of all nodes that are heartbeating in any regions
1431 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1433 /* callers want to serialize this map and callbacks so that they
1434 * can trust that they don't miss nodes coming to the party */
1435 down_read(&o2hb_callback_sem);
1436 spin_lock(&o2hb_live_lock);
1437 o2hb_fill_node_map_from_callback(map, bytes);
1438 spin_unlock(&o2hb_live_lock);
1439 up_read(&o2hb_callback_sem);
1441 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1444 * heartbeat configfs bits. The heartbeat set is a default set under
1445 * the cluster set in nodemanager.c.
1448 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1450 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1453 /* drop_item only drops its ref after killing the thread, nothing should
1454 * be using the region anymore. this has to clean up any state that
1455 * attributes might have built up. */
1456 static void o2hb_region_release(struct config_item *item)
1460 struct o2hb_region *reg = to_o2hb_region(item);
1462 mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1464 kfree(reg->hr_tmp_block);
1466 if (reg->hr_slot_data) {
1467 for (i = 0; i < reg->hr_num_pages; i++) {
1468 page = reg->hr_slot_data[i];
1472 kfree(reg->hr_slot_data);
1476 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1478 kfree(reg->hr_slots);
1480 kfree(reg->hr_db_regnum);
1481 kfree(reg->hr_db_livenodes);
1482 debugfs_remove(reg->hr_debug_livenodes);
1483 debugfs_remove(reg->hr_debug_regnum);
1484 debugfs_remove(reg->hr_debug_elapsed_time);
1485 debugfs_remove(reg->hr_debug_pinned);
1486 debugfs_remove(reg->hr_debug_dir);
1488 spin_lock(&o2hb_live_lock);
1489 list_del(®->hr_all_item);
1490 spin_unlock(&o2hb_live_lock);
1495 static int o2hb_read_block_input(struct o2hb_region *reg,
1498 unsigned long *ret_bytes,
1499 unsigned int *ret_bits)
1501 unsigned long bytes;
1502 char *p = (char *)page;
1504 bytes = simple_strtoul(p, &p, 0);
1505 if (!p || (*p && (*p != '\n')))
1508 /* Heartbeat and fs min / max block sizes are the same. */
1509 if (bytes > 4096 || bytes < 512)
1511 if (hweight16(bytes) != 1)
1517 *ret_bits = ffs(bytes) - 1;
1522 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1525 return sprintf(page, "%u\n", reg->hr_block_bytes);
1528 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1533 unsigned long block_bytes;
1534 unsigned int block_bits;
1539 status = o2hb_read_block_input(reg, page, count,
1540 &block_bytes, &block_bits);
1544 reg->hr_block_bytes = (unsigned int)block_bytes;
1545 reg->hr_block_bits = block_bits;
1550 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1553 return sprintf(page, "%llu\n", reg->hr_start_block);
1556 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1560 unsigned long long tmp;
1561 char *p = (char *)page;
1566 tmp = simple_strtoull(p, &p, 0);
1567 if (!p || (*p && (*p != '\n')))
1570 reg->hr_start_block = tmp;
1575 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1578 return sprintf(page, "%d\n", reg->hr_blocks);
1581 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1586 char *p = (char *)page;
1591 tmp = simple_strtoul(p, &p, 0);
1592 if (!p || (*p && (*p != '\n')))
1595 if (tmp > O2NM_MAX_NODES || tmp == 0)
1598 reg->hr_blocks = (unsigned int)tmp;
1603 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1606 unsigned int ret = 0;
1609 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1614 static void o2hb_init_region_params(struct o2hb_region *reg)
1616 reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1617 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1619 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1620 reg->hr_start_block, reg->hr_blocks);
1621 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1622 reg->hr_block_bytes, reg->hr_block_bits);
1623 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1624 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1627 static int o2hb_map_slot_data(struct o2hb_region *reg)
1630 unsigned int last_slot;
1631 unsigned int spp = reg->hr_slots_per_page;
1634 struct o2hb_disk_slot *slot;
1636 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1637 if (reg->hr_tmp_block == NULL) {
1638 mlog_errno(-ENOMEM);
1642 reg->hr_slots = kcalloc(reg->hr_blocks,
1643 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1644 if (reg->hr_slots == NULL) {
1645 mlog_errno(-ENOMEM);
1649 for(i = 0; i < reg->hr_blocks; i++) {
1650 slot = ®->hr_slots[i];
1651 slot->ds_node_num = i;
1652 INIT_LIST_HEAD(&slot->ds_live_item);
1653 slot->ds_raw_block = NULL;
1656 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1657 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1658 "at %u blocks per page\n",
1659 reg->hr_num_pages, reg->hr_blocks, spp);
1661 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1663 if (!reg->hr_slot_data) {
1664 mlog_errno(-ENOMEM);
1668 for(i = 0; i < reg->hr_num_pages; i++) {
1669 page = alloc_page(GFP_KERNEL);
1671 mlog_errno(-ENOMEM);
1675 reg->hr_slot_data[i] = page;
1677 last_slot = i * spp;
1678 raw = page_address(page);
1680 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1682 BUG_ON((j + last_slot) >= reg->hr_blocks);
1684 slot = ®->hr_slots[j + last_slot];
1685 slot->ds_raw_block =
1686 (struct o2hb_disk_heartbeat_block *) raw;
1688 raw += reg->hr_block_bytes;
1695 /* Read in all the slots available and populate the tracking
1696 * structures so that we can start with a baseline idea of what's
1698 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1701 struct o2hb_disk_slot *slot;
1702 struct o2hb_disk_heartbeat_block *hb_block;
1704 ret = o2hb_read_slots(reg, reg->hr_blocks);
1710 /* We only want to get an idea of the values initially in each
1711 * slot, so we do no verification - o2hb_check_slot will
1712 * actually determine if each configured slot is valid and
1713 * whether any values have changed. */
1714 for(i = 0; i < reg->hr_blocks; i++) {
1715 slot = ®->hr_slots[i];
1716 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1718 /* Only fill the values that o2hb_check_slot uses to
1719 * determine changing slots */
1720 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1721 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1728 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1729 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1733 struct task_struct *hb_task;
1736 char *p = (char *)page;
1738 struct inode *inode;
1739 ssize_t ret = -EINVAL;
1745 /* We can't heartbeat without having had our node number
1746 * configured yet. */
1747 if (o2nm_this_node() == O2NM_MAX_NODES)
1750 fd = simple_strtol(p, &p, 0);
1751 if (!p || (*p && (*p != '\n')))
1754 if (fd < 0 || fd >= INT_MAX)
1761 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1762 reg->hr_block_bytes == 0)
1765 inode = igrab(f.file->f_mapping->host);
1769 if (!S_ISBLK(inode->i_mode))
1772 reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
1773 ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
1775 reg->hr_bdev = NULL;
1780 bdevname(reg->hr_bdev, reg->hr_dev_name);
1782 sectsize = bdev_logical_block_size(reg->hr_bdev);
1783 if (sectsize != reg->hr_block_bytes) {
1785 "blocksize %u incorrect for device, expected %d",
1786 reg->hr_block_bytes, sectsize);
1791 o2hb_init_region_params(reg);
1793 /* Generation of zero is invalid */
1795 get_random_bytes(®->hr_generation,
1796 sizeof(reg->hr_generation));
1797 } while (reg->hr_generation == 0);
1799 ret = o2hb_map_slot_data(reg);
1805 ret = o2hb_populate_slot_data(reg);
1811 INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout);
1814 * A node is considered live after it has beat LIVE_THRESHOLD
1815 * times. We're not steady until we've given them a chance
1816 * _after_ our first read.
1817 * The default threshold is bare minimum so as to limit the delay
1818 * during mounts. For global heartbeat, the threshold doubled for the
1821 live_threshold = O2HB_LIVE_THRESHOLD;
1822 if (o2hb_global_heartbeat_active()) {
1823 spin_lock(&o2hb_live_lock);
1824 if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1825 live_threshold <<= 1;
1826 spin_unlock(&o2hb_live_lock);
1829 atomic_set(®->hr_steady_iterations, live_threshold);
1830 /* unsteady_iterations is double the steady_iterations */
1831 atomic_set(®->hr_unsteady_iterations, (live_threshold << 1));
1833 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1834 reg->hr_item.ci_name);
1835 if (IS_ERR(hb_task)) {
1836 ret = PTR_ERR(hb_task);
1841 spin_lock(&o2hb_live_lock);
1842 reg->hr_task = hb_task;
1843 spin_unlock(&o2hb_live_lock);
1845 ret = wait_event_interruptible(o2hb_steady_queue,
1846 atomic_read(®->hr_steady_iterations) == 0);
1848 atomic_set(®->hr_steady_iterations, 0);
1849 reg->hr_aborted_start = 1;
1852 if (reg->hr_aborted_start) {
1857 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1858 spin_lock(&o2hb_live_lock);
1859 hb_task = reg->hr_task;
1860 if (o2hb_global_heartbeat_active())
1861 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1862 spin_unlock(&o2hb_live_lock);
1869 if (hb_task && o2hb_global_heartbeat_active())
1870 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1871 config_item_name(®->hr_item), reg->hr_dev_name);
1880 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1881 reg->hr_bdev = NULL;
1887 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1892 spin_lock(&o2hb_live_lock);
1894 pid = task_pid_nr(reg->hr_task);
1895 spin_unlock(&o2hb_live_lock);
1900 return sprintf(page, "%u\n", pid);
1903 struct o2hb_region_attribute {
1904 struct configfs_attribute attr;
1905 ssize_t (*show)(struct o2hb_region *, char *);
1906 ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1909 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1910 .attr = { .ca_owner = THIS_MODULE,
1911 .ca_name = "block_bytes",
1912 .ca_mode = S_IRUGO | S_IWUSR },
1913 .show = o2hb_region_block_bytes_read,
1914 .store = o2hb_region_block_bytes_write,
1917 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1918 .attr = { .ca_owner = THIS_MODULE,
1919 .ca_name = "start_block",
1920 .ca_mode = S_IRUGO | S_IWUSR },
1921 .show = o2hb_region_start_block_read,
1922 .store = o2hb_region_start_block_write,
1925 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1926 .attr = { .ca_owner = THIS_MODULE,
1927 .ca_name = "blocks",
1928 .ca_mode = S_IRUGO | S_IWUSR },
1929 .show = o2hb_region_blocks_read,
1930 .store = o2hb_region_blocks_write,
1933 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1934 .attr = { .ca_owner = THIS_MODULE,
1936 .ca_mode = S_IRUGO | S_IWUSR },
1937 .show = o2hb_region_dev_read,
1938 .store = o2hb_region_dev_write,
1941 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1942 .attr = { .ca_owner = THIS_MODULE,
1944 .ca_mode = S_IRUGO | S_IRUSR },
1945 .show = o2hb_region_pid_read,
1948 static struct configfs_attribute *o2hb_region_attrs[] = {
1949 &o2hb_region_attr_block_bytes.attr,
1950 &o2hb_region_attr_start_block.attr,
1951 &o2hb_region_attr_blocks.attr,
1952 &o2hb_region_attr_dev.attr,
1953 &o2hb_region_attr_pid.attr,
1957 static ssize_t o2hb_region_show(struct config_item *item,
1958 struct configfs_attribute *attr,
1961 struct o2hb_region *reg = to_o2hb_region(item);
1962 struct o2hb_region_attribute *o2hb_region_attr =
1963 container_of(attr, struct o2hb_region_attribute, attr);
1966 if (o2hb_region_attr->show)
1967 ret = o2hb_region_attr->show(reg, page);
1971 static ssize_t o2hb_region_store(struct config_item *item,
1972 struct configfs_attribute *attr,
1973 const char *page, size_t count)
1975 struct o2hb_region *reg = to_o2hb_region(item);
1976 struct o2hb_region_attribute *o2hb_region_attr =
1977 container_of(attr, struct o2hb_region_attribute, attr);
1978 ssize_t ret = -EINVAL;
1980 if (o2hb_region_attr->store)
1981 ret = o2hb_region_attr->store(reg, page, count);
1985 static struct configfs_item_operations o2hb_region_item_ops = {
1986 .release = o2hb_region_release,
1987 .show_attribute = o2hb_region_show,
1988 .store_attribute = o2hb_region_store,
1991 static struct config_item_type o2hb_region_type = {
1992 .ct_item_ops = &o2hb_region_item_ops,
1993 .ct_attrs = o2hb_region_attrs,
1994 .ct_owner = THIS_MODULE,
1999 struct o2hb_heartbeat_group {
2000 struct config_group hs_group;
2004 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
2007 container_of(group, struct o2hb_heartbeat_group, hs_group)
2011 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
2016 debugfs_create_dir(config_item_name(®->hr_item), dir);
2017 if (!reg->hr_debug_dir) {
2022 reg->hr_debug_livenodes =
2023 o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2025 &(reg->hr_db_livenodes),
2026 sizeof(*(reg->hr_db_livenodes)),
2027 O2HB_DB_TYPE_REGION_LIVENODES,
2028 sizeof(reg->hr_live_node_bitmap),
2029 O2NM_MAX_NODES, reg);
2030 if (!reg->hr_debug_livenodes) {
2035 reg->hr_debug_regnum =
2036 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2038 &(reg->hr_db_regnum),
2039 sizeof(*(reg->hr_db_regnum)),
2040 O2HB_DB_TYPE_REGION_NUMBER,
2041 0, O2NM_MAX_NODES, reg);
2042 if (!reg->hr_debug_regnum) {
2047 reg->hr_debug_elapsed_time =
2048 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2050 &(reg->hr_db_elapsed_time),
2051 sizeof(*(reg->hr_db_elapsed_time)),
2052 O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2054 if (!reg->hr_debug_elapsed_time) {
2059 reg->hr_debug_pinned =
2060 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2062 &(reg->hr_db_pinned),
2063 sizeof(*(reg->hr_db_pinned)),
2064 O2HB_DB_TYPE_REGION_PINNED,
2066 if (!reg->hr_debug_pinned) {
2076 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2079 struct o2hb_region *reg = NULL;
2082 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2084 return ERR_PTR(-ENOMEM);
2086 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2087 ret = -ENAMETOOLONG;
2091 spin_lock(&o2hb_live_lock);
2092 reg->hr_region_num = 0;
2093 if (o2hb_global_heartbeat_active()) {
2094 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2096 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2097 spin_unlock(&o2hb_live_lock);
2101 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2103 list_add_tail(®->hr_all_item, &o2hb_all_regions);
2104 spin_unlock(&o2hb_live_lock);
2106 config_item_init_type_name(®->hr_item, name, &o2hb_region_type);
2108 ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2110 config_item_put(®->hr_item);
2114 return ®->hr_item;
2117 return ERR_PTR(ret);
2120 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2121 struct config_item *item)
2123 struct task_struct *hb_task;
2124 struct o2hb_region *reg = to_o2hb_region(item);
2125 int quorum_region = 0;
2127 /* stop the thread when the user removes the region dir */
2128 spin_lock(&o2hb_live_lock);
2129 hb_task = reg->hr_task;
2130 reg->hr_task = NULL;
2131 reg->hr_item_dropped = 1;
2132 spin_unlock(&o2hb_live_lock);
2135 kthread_stop(hb_task);
2137 if (o2hb_global_heartbeat_active()) {
2138 spin_lock(&o2hb_live_lock);
2139 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2140 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2141 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2143 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2144 spin_unlock(&o2hb_live_lock);
2145 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2146 ((atomic_read(®->hr_steady_iterations) == 0) ?
2147 "stopped" : "start aborted"), config_item_name(item),
2152 * If we're racing a dev_write(), we need to wake them. They will
2153 * check reg->hr_task
2155 if (atomic_read(®->hr_steady_iterations) != 0) {
2156 reg->hr_aborted_start = 1;
2157 atomic_set(®->hr_steady_iterations, 0);
2158 wake_up(&o2hb_steady_queue);
2161 config_item_put(item);
2163 if (!o2hb_global_heartbeat_active() || !quorum_region)
2167 * If global heartbeat active and there are dependent users,
2168 * pin all regions if quorum region count <= CUT_OFF
2170 spin_lock(&o2hb_live_lock);
2172 if (!o2hb_dependent_users)
2175 if (bitmap_weight(o2hb_quorum_region_bitmap,
2176 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2177 o2hb_region_pin(NULL);
2180 spin_unlock(&o2hb_live_lock);
2183 struct o2hb_heartbeat_group_attribute {
2184 struct configfs_attribute attr;
2185 ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2186 ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2189 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2190 struct configfs_attribute *attr,
2193 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2194 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2195 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2198 if (o2hb_heartbeat_group_attr->show)
2199 ret = o2hb_heartbeat_group_attr->show(reg, page);
2203 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2204 struct configfs_attribute *attr,
2205 const char *page, size_t count)
2207 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2208 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2209 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2210 ssize_t ret = -EINVAL;
2212 if (o2hb_heartbeat_group_attr->store)
2213 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2217 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2220 return sprintf(page, "%u\n", o2hb_dead_threshold);
2223 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2228 char *p = (char *)page;
2230 tmp = simple_strtoul(p, &p, 10);
2231 if (!p || (*p && (*p != '\n')))
2234 /* this will validate ranges for us. */
2235 o2hb_dead_threshold_set((unsigned int) tmp);
2241 ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2244 return sprintf(page, "%s\n",
2245 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2249 ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2250 const char *page, size_t count)
2256 len = (page[count - 1] == '\n') ? count - 1 : count;
2260 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2261 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2264 ret = o2hb_global_heartbeat_mode_set(i);
2266 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2267 o2hb_heartbeat_mode_desc[i]);
2275 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2276 .attr = { .ca_owner = THIS_MODULE,
2277 .ca_name = "dead_threshold",
2278 .ca_mode = S_IRUGO | S_IWUSR },
2279 .show = o2hb_heartbeat_group_threshold_show,
2280 .store = o2hb_heartbeat_group_threshold_store,
2283 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2284 .attr = { .ca_owner = THIS_MODULE,
2286 .ca_mode = S_IRUGO | S_IWUSR },
2287 .show = o2hb_heartbeat_group_mode_show,
2288 .store = o2hb_heartbeat_group_mode_store,
2291 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2292 &o2hb_heartbeat_group_attr_threshold.attr,
2293 &o2hb_heartbeat_group_attr_mode.attr,
2297 static struct configfs_item_operations o2hb_heartbeat_group_item_ops = {
2298 .show_attribute = o2hb_heartbeat_group_show,
2299 .store_attribute = o2hb_heartbeat_group_store,
2302 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2303 .make_item = o2hb_heartbeat_group_make_item,
2304 .drop_item = o2hb_heartbeat_group_drop_item,
2307 static struct config_item_type o2hb_heartbeat_group_type = {
2308 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
2309 .ct_item_ops = &o2hb_heartbeat_group_item_ops,
2310 .ct_attrs = o2hb_heartbeat_group_attrs,
2311 .ct_owner = THIS_MODULE,
2314 /* this is just here to avoid touching group in heartbeat.h which the
2315 * entire damn world #includes */
2316 struct config_group *o2hb_alloc_hb_set(void)
2318 struct o2hb_heartbeat_group *hs = NULL;
2319 struct config_group *ret = NULL;
2321 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2325 config_group_init_type_name(&hs->hs_group, "heartbeat",
2326 &o2hb_heartbeat_group_type);
2328 ret = &hs->hs_group;
2335 void o2hb_free_hb_set(struct config_group *group)
2337 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2341 /* hb callback registration and issuing */
2343 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2345 if (type == O2HB_NUM_CB)
2346 return ERR_PTR(-EINVAL);
2348 return &o2hb_callbacks[type];
2351 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2352 enum o2hb_callback_type type,
2357 INIT_LIST_HEAD(&hc->hc_item);
2360 hc->hc_priority = priority;
2362 hc->hc_magic = O2HB_CB_MAGIC;
2364 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2367 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2368 * In global heartbeat mode, region_uuid passed is NULL.
2370 * In local, we only pin the matching region. In global we pin all the active
2373 static int o2hb_region_pin(const char *region_uuid)
2375 int ret = 0, found = 0;
2376 struct o2hb_region *reg;
2379 assert_spin_locked(&o2hb_live_lock);
2381 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2382 if (reg->hr_item_dropped)
2385 uuid = config_item_name(®->hr_item);
2387 /* local heartbeat */
2389 if (strcmp(region_uuid, uuid))
2394 if (reg->hr_item_pinned || reg->hr_item_dropped)
2397 /* Ignore ENOENT only for local hb (userdlm domain) */
2398 ret = o2nm_depend_item(®->hr_item);
2400 mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2401 reg->hr_item_pinned = 1;
2403 if (ret == -ENOENT && found)
2406 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2420 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2421 * In global heartbeat mode, region_uuid passed is NULL.
2423 * In local, we only unpin the matching region. In global we unpin all the
2426 static void o2hb_region_unpin(const char *region_uuid)
2428 struct o2hb_region *reg;
2432 assert_spin_locked(&o2hb_live_lock);
2434 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2435 if (reg->hr_item_dropped)
2438 uuid = config_item_name(®->hr_item);
2440 if (strcmp(region_uuid, uuid))
2445 if (reg->hr_item_pinned) {
2446 mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2447 o2nm_undepend_item(®->hr_item);
2448 reg->hr_item_pinned = 0;
2455 static int o2hb_region_inc_user(const char *region_uuid)
2459 spin_lock(&o2hb_live_lock);
2461 /* local heartbeat */
2462 if (!o2hb_global_heartbeat_active()) {
2463 ret = o2hb_region_pin(region_uuid);
2468 * if global heartbeat active and this is the first dependent user,
2469 * pin all regions if quorum region count <= CUT_OFF
2471 o2hb_dependent_users++;
2472 if (o2hb_dependent_users > 1)
2475 if (bitmap_weight(o2hb_quorum_region_bitmap,
2476 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2477 ret = o2hb_region_pin(NULL);
2480 spin_unlock(&o2hb_live_lock);
2484 void o2hb_region_dec_user(const char *region_uuid)
2486 spin_lock(&o2hb_live_lock);
2488 /* local heartbeat */
2489 if (!o2hb_global_heartbeat_active()) {
2490 o2hb_region_unpin(region_uuid);
2495 * if global heartbeat active and there are no dependent users,
2496 * unpin all quorum regions
2498 o2hb_dependent_users--;
2499 if (!o2hb_dependent_users)
2500 o2hb_region_unpin(NULL);
2503 spin_unlock(&o2hb_live_lock);
2506 int o2hb_register_callback(const char *region_uuid,
2507 struct o2hb_callback_func *hc)
2509 struct o2hb_callback_func *f;
2510 struct o2hb_callback *hbcall;
2513 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2514 BUG_ON(!list_empty(&hc->hc_item));
2516 hbcall = hbcall_from_type(hc->hc_type);
2517 if (IS_ERR(hbcall)) {
2518 ret = PTR_ERR(hbcall);
2523 ret = o2hb_region_inc_user(region_uuid);
2530 down_write(&o2hb_callback_sem);
2532 list_for_each_entry(f, &hbcall->list, hc_item) {
2533 if (hc->hc_priority < f->hc_priority) {
2534 list_add_tail(&hc->hc_item, &f->hc_item);
2538 if (list_empty(&hc->hc_item))
2539 list_add_tail(&hc->hc_item, &hbcall->list);
2541 up_write(&o2hb_callback_sem);
2544 mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2545 ret, __builtin_return_address(0), hc);
2548 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2550 void o2hb_unregister_callback(const char *region_uuid,
2551 struct o2hb_callback_func *hc)
2553 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2555 mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2556 __builtin_return_address(0), hc);
2558 /* XXX Can this happen _with_ a region reference? */
2559 if (list_empty(&hc->hc_item))
2563 o2hb_region_dec_user(region_uuid);
2565 down_write(&o2hb_callback_sem);
2567 list_del_init(&hc->hc_item);
2569 up_write(&o2hb_callback_sem);
2571 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2573 int o2hb_check_node_heartbeating(u8 node_num)
2575 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2577 o2hb_fill_node_map(testing_map, sizeof(testing_map));
2578 if (!test_bit(node_num, testing_map)) {
2580 "node (%u) does not have heartbeating enabled.\n",
2587 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2589 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2591 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2593 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2594 if (!test_bit(node_num, testing_map)) {
2596 "node (%u) does not have heartbeating enabled.\n",
2603 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2605 /* Makes sure our local node is configured with a node number, and is
2607 int o2hb_check_local_node_heartbeating(void)
2611 /* if this node was set then we have networking */
2612 node_num = o2nm_this_node();
2613 if (node_num == O2NM_MAX_NODES) {
2614 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2618 return o2hb_check_node_heartbeating(node_num);
2620 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2623 * this is just a hack until we get the plumbing which flips file systems
2624 * read only and drops the hb ref instead of killing the node dead.
2626 void o2hb_stop_all_regions(void)
2628 struct o2hb_region *reg;
2630 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2632 spin_lock(&o2hb_live_lock);
2634 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2635 reg->hr_unclean_stop = 1;
2637 spin_unlock(&o2hb_live_lock);
2639 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2641 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2643 struct o2hb_region *reg;
2647 spin_lock(&o2hb_live_lock);
2650 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2651 if (reg->hr_item_dropped)
2654 mlog(0, "Region: %s\n", config_item_name(®->hr_item));
2655 if (numregs < max_regions) {
2656 memcpy(p, config_item_name(®->hr_item),
2657 O2HB_MAX_REGION_NAME_LEN);
2658 p += O2HB_MAX_REGION_NAME_LEN;
2663 spin_unlock(&o2hb_live_lock);
2667 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2669 int o2hb_global_heartbeat_active(void)
2671 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2673 EXPORT_SYMBOL(o2hb_global_heartbeat_active);