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 o2hb_callback_func *f;
633 list_for_each_entry(f, &hbcall->list, hc_item) {
634 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
635 (f->hc_func)(node, idx, f->hc_data);
639 /* Will run the list in order until we process the passed event */
640 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
642 struct o2hb_callback *hbcall;
643 struct o2hb_node_event *event;
645 /* Holding callback sem assures we don't alter the callback
646 * lists when doing this, and serializes ourselves with other
647 * processes wanting callbacks. */
648 down_write(&o2hb_callback_sem);
650 spin_lock(&o2hb_live_lock);
651 while (!list_empty(&o2hb_node_events)
652 && !list_empty(&queued_event->hn_item)) {
653 event = list_entry(o2hb_node_events.next,
654 struct o2hb_node_event,
656 list_del_init(&event->hn_item);
657 spin_unlock(&o2hb_live_lock);
659 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
660 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
663 hbcall = hbcall_from_type(event->hn_event_type);
665 /* We should *never* have gotten on to the list with a
666 * bad type... This isn't something that we should try
667 * to recover from. */
668 BUG_ON(IS_ERR(hbcall));
670 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
672 spin_lock(&o2hb_live_lock);
674 spin_unlock(&o2hb_live_lock);
676 up_write(&o2hb_callback_sem);
679 static void o2hb_queue_node_event(struct o2hb_node_event *event,
680 enum o2hb_callback_type type,
681 struct o2nm_node *node,
684 assert_spin_locked(&o2hb_live_lock);
686 BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
688 event->hn_event_type = type;
689 event->hn_node = node;
690 event->hn_node_num = node_num;
692 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
693 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
695 list_add_tail(&event->hn_item, &o2hb_node_events);
698 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
700 struct o2hb_node_event event =
701 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
702 struct o2nm_node *node;
705 node = o2nm_get_node_by_num(slot->ds_node_num);
709 spin_lock(&o2hb_live_lock);
710 if (!list_empty(&slot->ds_live_item)) {
711 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
714 list_del_init(&slot->ds_live_item);
716 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
717 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
719 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
724 spin_unlock(&o2hb_live_lock);
727 o2hb_run_event_list(&event);
732 static void o2hb_set_quorum_device(struct o2hb_region *reg)
734 if (!o2hb_global_heartbeat_active())
737 /* Prevent race with o2hb_heartbeat_group_drop_item() */
738 if (kthread_should_stop())
741 /* Tag region as quorum only after thread reaches steady state */
742 if (atomic_read(®->hr_steady_iterations) != 0)
745 spin_lock(&o2hb_live_lock);
747 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
751 * A region can be added to the quorum only when it sees all
752 * live nodes heartbeat on it. In other words, the region has been
753 * added to all nodes.
755 if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
756 sizeof(o2hb_live_node_bitmap)))
759 printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
760 config_item_name(®->hr_item), reg->hr_dev_name);
762 set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
765 * If global heartbeat active, unpin all regions if the
766 * region count > CUT_OFF
768 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
769 O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
770 o2hb_region_unpin(NULL);
772 spin_unlock(&o2hb_live_lock);
775 static int o2hb_check_slot(struct o2hb_region *reg,
776 struct o2hb_disk_slot *slot)
778 int changed = 0, gen_changed = 0;
779 struct o2hb_node_event event =
780 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
781 struct o2nm_node *node;
782 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
784 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
785 unsigned int slot_dead_ms;
789 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
792 * If a node is no longer configured but is still in the livemap, we
793 * may need to clear that bit from the livemap.
795 node = o2nm_get_node_by_num(slot->ds_node_num);
797 spin_lock(&o2hb_live_lock);
798 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
799 spin_unlock(&o2hb_live_lock);
804 if (!o2hb_verify_crc(reg, hb_block)) {
805 /* all paths from here will drop o2hb_live_lock for
807 spin_lock(&o2hb_live_lock);
809 /* Don't print an error on the console in this case -
810 * a freshly formatted heartbeat area will not have a
812 if (list_empty(&slot->ds_live_item))
815 /* The node is live but pushed out a bad crc. We
816 * consider it a transient miss but don't populate any
817 * other values as they may be junk. */
818 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
819 slot->ds_node_num, reg->hr_dev_name);
820 o2hb_dump_slot(hb_block);
822 slot->ds_equal_samples++;
826 /* we don't care if these wrap.. the state transitions below
827 * clear at the right places */
828 cputime = le64_to_cpu(hb_block->hb_seq);
829 if (slot->ds_last_time != cputime)
830 slot->ds_changed_samples++;
832 slot->ds_equal_samples++;
833 slot->ds_last_time = cputime;
835 /* The node changed heartbeat generations. We assume this to
836 * mean it dropped off but came back before we timed out. We
837 * want to consider it down for the time being but don't want
838 * to lose any changed_samples state we might build up to
839 * considering it live again. */
840 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
842 slot->ds_equal_samples = 0;
843 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
844 "to 0x%llx)\n", slot->ds_node_num,
845 (long long)slot->ds_last_generation,
846 (long long)le64_to_cpu(hb_block->hb_generation));
849 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
851 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
852 "seq %llu last %llu changed %u equal %u\n",
853 slot->ds_node_num, (long long)slot->ds_last_generation,
854 le32_to_cpu(hb_block->hb_cksum),
855 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
856 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
857 slot->ds_equal_samples);
859 spin_lock(&o2hb_live_lock);
862 /* dead nodes only come to life after some number of
863 * changes at any time during their dead time */
864 if (list_empty(&slot->ds_live_item) &&
865 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
866 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
867 slot->ds_node_num, (long long)slot->ds_last_generation);
869 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
871 /* first on the list generates a callback */
872 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
873 mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
874 "bitmap\n", slot->ds_node_num);
875 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
877 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
884 list_add_tail(&slot->ds_live_item,
885 &o2hb_live_slots[slot->ds_node_num]);
887 slot->ds_equal_samples = 0;
889 /* We want to be sure that all nodes agree on the
890 * number of milliseconds before a node will be
891 * considered dead. The self-fencing timeout is
892 * computed from this value, and a discrepancy might
893 * result in heartbeat calling a node dead when it
894 * hasn't self-fenced yet. */
895 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
896 if (slot_dead_ms && slot_dead_ms != dead_ms) {
897 /* TODO: Perhaps we can fail the region here. */
898 mlog(ML_ERROR, "Node %d on device %s has a dead count "
899 "of %u ms, but our count is %u ms.\n"
900 "Please double check your configuration values "
901 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
902 slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
908 /* if the list is dead, we're done.. */
909 if (list_empty(&slot->ds_live_item))
912 /* live nodes only go dead after enough consequtive missed
913 * samples.. reset the missed counter whenever we see
915 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
916 mlog(ML_HEARTBEAT, "Node %d left my region\n",
919 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
921 /* last off the live_slot generates a callback */
922 list_del_init(&slot->ds_live_item);
923 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
924 mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
925 "nodes bitmap\n", slot->ds_node_num);
926 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
928 /* node can be null */
929 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
930 node, slot->ds_node_num);
936 /* We don't clear this because the node is still
937 * actually writing new blocks. */
939 slot->ds_changed_samples = 0;
942 if (slot->ds_changed_samples) {
943 slot->ds_changed_samples = 0;
944 slot->ds_equal_samples = 0;
947 spin_unlock(&o2hb_live_lock);
950 o2hb_run_event_list(&event);
957 /* This could be faster if we just implmented a find_last_bit, but I
958 * don't think the circumstances warrant it. */
959 static int o2hb_highest_node(unsigned long *nodes,
966 while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
976 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
978 int i, ret, highest_node;
979 int membership_change = 0, own_slot_ok = 0;
980 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
981 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
982 struct o2hb_bio_wait_ctxt write_wc;
984 ret = o2nm_configured_node_map(configured_nodes,
985 sizeof(configured_nodes));
992 * If a node is not configured but is in the livemap, we still need
993 * to read the slot so as to be able to remove it from the livemap.
995 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
997 while ((i = find_next_bit(live_node_bitmap,
998 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
999 set_bit(i, configured_nodes);
1002 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1003 if (highest_node >= O2NM_MAX_NODES) {
1004 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1009 /* No sense in reading the slots of nodes that don't exist
1010 * yet. Of course, if the node definitions have holes in them
1011 * then we're reading an empty slot anyway... Consider this
1013 ret = o2hb_read_slots(reg, highest_node + 1);
1019 /* With an up to date view of the slots, we can check that no
1020 * other node has been improperly configured to heartbeat in
1022 own_slot_ok = o2hb_check_own_slot(reg);
1024 /* fill in the proper info for our next heartbeat */
1025 o2hb_prepare_block(reg, reg->hr_generation);
1027 ret = o2hb_issue_node_write(reg, &write_wc);
1034 while((i = find_next_bit(configured_nodes,
1035 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1036 membership_change |= o2hb_check_slot(reg, ®->hr_slots[i]);
1040 * We have to be sure we've advertised ourselves on disk
1041 * before we can go to steady state. This ensures that
1042 * people we find in our steady state have seen us.
1044 o2hb_wait_on_io(reg, &write_wc);
1045 if (write_wc.wc_error) {
1046 /* Do not re-arm the write timeout on I/O error - we
1047 * can't be sure that the new block ever made it to
1049 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1050 write_wc.wc_error, reg->hr_dev_name);
1051 ret = write_wc.wc_error;
1055 /* Skip disarming the timeout if own slot has stale/bad data */
1057 o2hb_set_quorum_device(reg);
1058 o2hb_arm_write_timeout(reg);
1062 /* let the person who launched us know when things are steady */
1063 if (atomic_read(®->hr_steady_iterations) != 0) {
1064 if (!ret && own_slot_ok && !membership_change) {
1065 if (atomic_dec_and_test(®->hr_steady_iterations))
1066 wake_up(&o2hb_steady_queue);
1070 if (atomic_read(®->hr_steady_iterations) != 0) {
1071 if (atomic_dec_and_test(®->hr_unsteady_iterations)) {
1072 printk(KERN_NOTICE "o2hb: Unable to stabilize "
1073 "heartbeart on region %s (%s)\n",
1074 config_item_name(®->hr_item),
1076 atomic_set(®->hr_steady_iterations, 0);
1077 reg->hr_aborted_start = 1;
1078 wake_up(&o2hb_steady_queue);
1086 /* Subtract b from a, storing the result in a. a *must* have a larger
1088 static void o2hb_tv_subtract(struct timeval *a,
1091 /* just return 0 when a is after b */
1092 if (a->tv_sec < b->tv_sec ||
1093 (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1099 a->tv_sec -= b->tv_sec;
1100 a->tv_usec -= b->tv_usec;
1101 while ( a->tv_usec < 0 ) {
1103 a->tv_usec += 1000000;
1107 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1108 struct timeval *end)
1110 struct timeval res = *end;
1112 o2hb_tv_subtract(&res, start);
1114 return res.tv_sec * 1000 + res.tv_usec / 1000;
1118 * we ride the region ref that the region dir holds. before the region
1119 * dir is removed and drops it ref it will wait to tear down this
1122 static int o2hb_thread(void *data)
1125 struct o2hb_region *reg = data;
1126 struct o2hb_bio_wait_ctxt write_wc;
1127 struct timeval before_hb, after_hb;
1128 unsigned int elapsed_msec;
1130 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1132 set_user_nice(current, -20);
1135 o2nm_depend_this_node();
1137 while (!kthread_should_stop() &&
1138 !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1139 /* We track the time spent inside
1140 * o2hb_do_disk_heartbeat so that we avoid more than
1141 * hr_timeout_ms between disk writes. On busy systems
1142 * this should result in a heartbeat which is less
1143 * likely to time itself out. */
1144 do_gettimeofday(&before_hb);
1146 ret = o2hb_do_disk_heartbeat(reg);
1148 do_gettimeofday(&after_hb);
1149 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1152 "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
1153 before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1154 after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1157 if (!kthread_should_stop() &&
1158 elapsed_msec < reg->hr_timeout_ms) {
1159 /* the kthread api has blocked signals for us so no
1160 * need to record the return value. */
1161 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1165 o2hb_disarm_write_timeout(reg);
1167 /* unclean stop is only used in very bad situation */
1168 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1169 o2hb_shutdown_slot(®->hr_slots[i]);
1171 /* Explicit down notification - avoid forcing the other nodes
1172 * to timeout on this region when we could just as easily
1173 * write a clear generation - thus indicating to them that
1174 * this node has left this region.
1176 if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1177 o2hb_prepare_block(reg, 0);
1178 ret = o2hb_issue_node_write(reg, &write_wc);
1180 o2hb_wait_on_io(reg, &write_wc);
1186 o2nm_undepend_this_node();
1188 mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1193 #ifdef CONFIG_DEBUG_FS
1194 static int o2hb_debug_open(struct inode *inode, struct file *file)
1196 struct o2hb_debug_buf *db = inode->i_private;
1197 struct o2hb_region *reg;
1198 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1204 /* max_nodes should be the largest bitmap we pass here */
1205 BUG_ON(sizeof(map) < db->db_size);
1207 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1211 switch (db->db_type) {
1212 case O2HB_DB_TYPE_LIVENODES:
1213 case O2HB_DB_TYPE_LIVEREGIONS:
1214 case O2HB_DB_TYPE_QUORUMREGIONS:
1215 case O2HB_DB_TYPE_FAILEDREGIONS:
1216 spin_lock(&o2hb_live_lock);
1217 memcpy(map, db->db_data, db->db_size);
1218 spin_unlock(&o2hb_live_lock);
1221 case O2HB_DB_TYPE_REGION_LIVENODES:
1222 spin_lock(&o2hb_live_lock);
1223 reg = (struct o2hb_region *)db->db_data;
1224 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1225 spin_unlock(&o2hb_live_lock);
1228 case O2HB_DB_TYPE_REGION_NUMBER:
1229 reg = (struct o2hb_region *)db->db_data;
1230 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1231 reg->hr_region_num);
1234 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1235 reg = (struct o2hb_region *)db->db_data;
1236 lts = reg->hr_last_timeout_start;
1237 /* If 0, it has never been set before */
1239 lts = jiffies_to_msecs(jiffies - lts);
1240 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1243 case O2HB_DB_TYPE_REGION_PINNED:
1244 reg = (struct o2hb_region *)db->db_data;
1245 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1246 !!reg->hr_item_pinned);
1253 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1254 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1255 out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1258 i_size_write(inode, out);
1260 file->private_data = buf;
1267 static int o2hb_debug_release(struct inode *inode, struct file *file)
1269 kfree(file->private_data);
1273 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1274 size_t nbytes, loff_t *ppos)
1276 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1277 i_size_read(file->f_mapping->host));
1280 static int o2hb_debug_open(struct inode *inode, struct file *file)
1284 static int o2hb_debug_release(struct inode *inode, struct file *file)
1288 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1289 size_t nbytes, loff_t *ppos)
1293 #endif /* CONFIG_DEBUG_FS */
1295 static const struct file_operations o2hb_debug_fops = {
1296 .open = o2hb_debug_open,
1297 .release = o2hb_debug_release,
1298 .read = o2hb_debug_read,
1299 .llseek = generic_file_llseek,
1302 void o2hb_exit(void)
1304 kfree(o2hb_db_livenodes);
1305 kfree(o2hb_db_liveregions);
1306 kfree(o2hb_db_quorumregions);
1307 kfree(o2hb_db_failedregions);
1308 debugfs_remove(o2hb_debug_failedregions);
1309 debugfs_remove(o2hb_debug_quorumregions);
1310 debugfs_remove(o2hb_debug_liveregions);
1311 debugfs_remove(o2hb_debug_livenodes);
1312 debugfs_remove(o2hb_debug_dir);
1315 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1316 struct o2hb_debug_buf **db, int db_len,
1317 int type, int size, int len, void *data)
1319 *db = kmalloc(db_len, GFP_KERNEL);
1323 (*db)->db_type = type;
1324 (*db)->db_size = size;
1325 (*db)->db_len = len;
1326 (*db)->db_data = data;
1328 return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1332 static int o2hb_debug_init(void)
1336 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1337 if (!o2hb_debug_dir) {
1342 o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1345 sizeof(*o2hb_db_livenodes),
1346 O2HB_DB_TYPE_LIVENODES,
1347 sizeof(o2hb_live_node_bitmap),
1349 o2hb_live_node_bitmap);
1350 if (!o2hb_debug_livenodes) {
1355 o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1357 &o2hb_db_liveregions,
1358 sizeof(*o2hb_db_liveregions),
1359 O2HB_DB_TYPE_LIVEREGIONS,
1360 sizeof(o2hb_live_region_bitmap),
1362 o2hb_live_region_bitmap);
1363 if (!o2hb_debug_liveregions) {
1368 o2hb_debug_quorumregions =
1369 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1371 &o2hb_db_quorumregions,
1372 sizeof(*o2hb_db_quorumregions),
1373 O2HB_DB_TYPE_QUORUMREGIONS,
1374 sizeof(o2hb_quorum_region_bitmap),
1376 o2hb_quorum_region_bitmap);
1377 if (!o2hb_debug_quorumregions) {
1382 o2hb_debug_failedregions =
1383 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1385 &o2hb_db_failedregions,
1386 sizeof(*o2hb_db_failedregions),
1387 O2HB_DB_TYPE_FAILEDREGIONS,
1388 sizeof(o2hb_failed_region_bitmap),
1390 o2hb_failed_region_bitmap);
1391 if (!o2hb_debug_failedregions) {
1408 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1409 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1411 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1412 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1414 INIT_LIST_HEAD(&o2hb_node_events);
1416 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1417 memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1418 memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1419 memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1420 memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1422 o2hb_dependent_users = 0;
1424 return o2hb_debug_init();
1427 /* if we're already in a callback then we're already serialized by the sem */
1428 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1431 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1433 memcpy(map, &o2hb_live_node_bitmap, bytes);
1437 * get a map of all nodes that are heartbeating in any regions
1439 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1441 /* callers want to serialize this map and callbacks so that they
1442 * can trust that they don't miss nodes coming to the party */
1443 down_read(&o2hb_callback_sem);
1444 spin_lock(&o2hb_live_lock);
1445 o2hb_fill_node_map_from_callback(map, bytes);
1446 spin_unlock(&o2hb_live_lock);
1447 up_read(&o2hb_callback_sem);
1449 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1452 * heartbeat configfs bits. The heartbeat set is a default set under
1453 * the cluster set in nodemanager.c.
1456 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1458 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1461 /* drop_item only drops its ref after killing the thread, nothing should
1462 * be using the region anymore. this has to clean up any state that
1463 * attributes might have built up. */
1464 static void o2hb_region_release(struct config_item *item)
1468 struct o2hb_region *reg = to_o2hb_region(item);
1470 mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1472 kfree(reg->hr_tmp_block);
1474 if (reg->hr_slot_data) {
1475 for (i = 0; i < reg->hr_num_pages; i++) {
1476 page = reg->hr_slot_data[i];
1480 kfree(reg->hr_slot_data);
1484 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1486 kfree(reg->hr_slots);
1488 kfree(reg->hr_db_regnum);
1489 kfree(reg->hr_db_livenodes);
1490 debugfs_remove(reg->hr_debug_livenodes);
1491 debugfs_remove(reg->hr_debug_regnum);
1492 debugfs_remove(reg->hr_debug_elapsed_time);
1493 debugfs_remove(reg->hr_debug_pinned);
1494 debugfs_remove(reg->hr_debug_dir);
1496 spin_lock(&o2hb_live_lock);
1497 list_del(®->hr_all_item);
1498 spin_unlock(&o2hb_live_lock);
1503 static int o2hb_read_block_input(struct o2hb_region *reg,
1506 unsigned long *ret_bytes,
1507 unsigned int *ret_bits)
1509 unsigned long bytes;
1510 char *p = (char *)page;
1512 bytes = simple_strtoul(p, &p, 0);
1513 if (!p || (*p && (*p != '\n')))
1516 /* Heartbeat and fs min / max block sizes are the same. */
1517 if (bytes > 4096 || bytes < 512)
1519 if (hweight16(bytes) != 1)
1525 *ret_bits = ffs(bytes) - 1;
1530 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1533 return sprintf(page, "%u\n", reg->hr_block_bytes);
1536 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1541 unsigned long block_bytes;
1542 unsigned int block_bits;
1547 status = o2hb_read_block_input(reg, page, count,
1548 &block_bytes, &block_bits);
1552 reg->hr_block_bytes = (unsigned int)block_bytes;
1553 reg->hr_block_bits = block_bits;
1558 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1561 return sprintf(page, "%llu\n", reg->hr_start_block);
1564 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1568 unsigned long long tmp;
1569 char *p = (char *)page;
1574 tmp = simple_strtoull(p, &p, 0);
1575 if (!p || (*p && (*p != '\n')))
1578 reg->hr_start_block = tmp;
1583 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1586 return sprintf(page, "%d\n", reg->hr_blocks);
1589 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1594 char *p = (char *)page;
1599 tmp = simple_strtoul(p, &p, 0);
1600 if (!p || (*p && (*p != '\n')))
1603 if (tmp > O2NM_MAX_NODES || tmp == 0)
1606 reg->hr_blocks = (unsigned int)tmp;
1611 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1614 unsigned int ret = 0;
1617 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1622 static void o2hb_init_region_params(struct o2hb_region *reg)
1624 reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1625 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1627 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1628 reg->hr_start_block, reg->hr_blocks);
1629 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1630 reg->hr_block_bytes, reg->hr_block_bits);
1631 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1632 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1635 static int o2hb_map_slot_data(struct o2hb_region *reg)
1638 unsigned int last_slot;
1639 unsigned int spp = reg->hr_slots_per_page;
1642 struct o2hb_disk_slot *slot;
1644 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1645 if (reg->hr_tmp_block == NULL) {
1646 mlog_errno(-ENOMEM);
1650 reg->hr_slots = kcalloc(reg->hr_blocks,
1651 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1652 if (reg->hr_slots == NULL) {
1653 mlog_errno(-ENOMEM);
1657 for(i = 0; i < reg->hr_blocks; i++) {
1658 slot = ®->hr_slots[i];
1659 slot->ds_node_num = i;
1660 INIT_LIST_HEAD(&slot->ds_live_item);
1661 slot->ds_raw_block = NULL;
1664 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1665 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1666 "at %u blocks per page\n",
1667 reg->hr_num_pages, reg->hr_blocks, spp);
1669 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1671 if (!reg->hr_slot_data) {
1672 mlog_errno(-ENOMEM);
1676 for(i = 0; i < reg->hr_num_pages; i++) {
1677 page = alloc_page(GFP_KERNEL);
1679 mlog_errno(-ENOMEM);
1683 reg->hr_slot_data[i] = page;
1685 last_slot = i * spp;
1686 raw = page_address(page);
1688 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1690 BUG_ON((j + last_slot) >= reg->hr_blocks);
1692 slot = ®->hr_slots[j + last_slot];
1693 slot->ds_raw_block =
1694 (struct o2hb_disk_heartbeat_block *) raw;
1696 raw += reg->hr_block_bytes;
1703 /* Read in all the slots available and populate the tracking
1704 * structures so that we can start with a baseline idea of what's
1706 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1709 struct o2hb_disk_slot *slot;
1710 struct o2hb_disk_heartbeat_block *hb_block;
1712 ret = o2hb_read_slots(reg, reg->hr_blocks);
1718 /* We only want to get an idea of the values initially in each
1719 * slot, so we do no verification - o2hb_check_slot will
1720 * actually determine if each configured slot is valid and
1721 * whether any values have changed. */
1722 for(i = 0; i < reg->hr_blocks; i++) {
1723 slot = ®->hr_slots[i];
1724 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1726 /* Only fill the values that o2hb_check_slot uses to
1727 * determine changing slots */
1728 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1729 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1736 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1737 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1741 struct task_struct *hb_task;
1744 char *p = (char *)page;
1746 struct inode *inode;
1747 ssize_t ret = -EINVAL;
1753 /* We can't heartbeat without having had our node number
1754 * configured yet. */
1755 if (o2nm_this_node() == O2NM_MAX_NODES)
1758 fd = simple_strtol(p, &p, 0);
1759 if (!p || (*p && (*p != '\n')))
1762 if (fd < 0 || fd >= INT_MAX)
1769 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1770 reg->hr_block_bytes == 0)
1773 inode = igrab(f.file->f_mapping->host);
1777 if (!S_ISBLK(inode->i_mode))
1780 reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
1781 ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
1783 reg->hr_bdev = NULL;
1788 bdevname(reg->hr_bdev, reg->hr_dev_name);
1790 sectsize = bdev_logical_block_size(reg->hr_bdev);
1791 if (sectsize != reg->hr_block_bytes) {
1793 "blocksize %u incorrect for device, expected %d",
1794 reg->hr_block_bytes, sectsize);
1799 o2hb_init_region_params(reg);
1801 /* Generation of zero is invalid */
1803 get_random_bytes(®->hr_generation,
1804 sizeof(reg->hr_generation));
1805 } while (reg->hr_generation == 0);
1807 ret = o2hb_map_slot_data(reg);
1813 ret = o2hb_populate_slot_data(reg);
1819 INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout);
1822 * A node is considered live after it has beat LIVE_THRESHOLD
1823 * times. We're not steady until we've given them a chance
1824 * _after_ our first read.
1825 * The default threshold is bare minimum so as to limit the delay
1826 * during mounts. For global heartbeat, the threshold doubled for the
1829 live_threshold = O2HB_LIVE_THRESHOLD;
1830 if (o2hb_global_heartbeat_active()) {
1831 spin_lock(&o2hb_live_lock);
1832 if (o2hb_pop_count(&o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1833 live_threshold <<= 1;
1834 spin_unlock(&o2hb_live_lock);
1837 atomic_set(®->hr_steady_iterations, live_threshold);
1838 /* unsteady_iterations is double the steady_iterations */
1839 atomic_set(®->hr_unsteady_iterations, (live_threshold << 1));
1841 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1842 reg->hr_item.ci_name);
1843 if (IS_ERR(hb_task)) {
1844 ret = PTR_ERR(hb_task);
1849 spin_lock(&o2hb_live_lock);
1850 reg->hr_task = hb_task;
1851 spin_unlock(&o2hb_live_lock);
1853 ret = wait_event_interruptible(o2hb_steady_queue,
1854 atomic_read(®->hr_steady_iterations) == 0);
1856 atomic_set(®->hr_steady_iterations, 0);
1857 reg->hr_aborted_start = 1;
1860 if (reg->hr_aborted_start) {
1865 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1866 spin_lock(&o2hb_live_lock);
1867 hb_task = reg->hr_task;
1868 if (o2hb_global_heartbeat_active())
1869 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1870 spin_unlock(&o2hb_live_lock);
1877 if (hb_task && o2hb_global_heartbeat_active())
1878 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1879 config_item_name(®->hr_item), reg->hr_dev_name);
1888 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1889 reg->hr_bdev = NULL;
1895 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1900 spin_lock(&o2hb_live_lock);
1902 pid = task_pid_nr(reg->hr_task);
1903 spin_unlock(&o2hb_live_lock);
1908 return sprintf(page, "%u\n", pid);
1911 struct o2hb_region_attribute {
1912 struct configfs_attribute attr;
1913 ssize_t (*show)(struct o2hb_region *, char *);
1914 ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1917 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1918 .attr = { .ca_owner = THIS_MODULE,
1919 .ca_name = "block_bytes",
1920 .ca_mode = S_IRUGO | S_IWUSR },
1921 .show = o2hb_region_block_bytes_read,
1922 .store = o2hb_region_block_bytes_write,
1925 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1926 .attr = { .ca_owner = THIS_MODULE,
1927 .ca_name = "start_block",
1928 .ca_mode = S_IRUGO | S_IWUSR },
1929 .show = o2hb_region_start_block_read,
1930 .store = o2hb_region_start_block_write,
1933 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1934 .attr = { .ca_owner = THIS_MODULE,
1935 .ca_name = "blocks",
1936 .ca_mode = S_IRUGO | S_IWUSR },
1937 .show = o2hb_region_blocks_read,
1938 .store = o2hb_region_blocks_write,
1941 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1942 .attr = { .ca_owner = THIS_MODULE,
1944 .ca_mode = S_IRUGO | S_IWUSR },
1945 .show = o2hb_region_dev_read,
1946 .store = o2hb_region_dev_write,
1949 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1950 .attr = { .ca_owner = THIS_MODULE,
1952 .ca_mode = S_IRUGO | S_IRUSR },
1953 .show = o2hb_region_pid_read,
1956 static struct configfs_attribute *o2hb_region_attrs[] = {
1957 &o2hb_region_attr_block_bytes.attr,
1958 &o2hb_region_attr_start_block.attr,
1959 &o2hb_region_attr_blocks.attr,
1960 &o2hb_region_attr_dev.attr,
1961 &o2hb_region_attr_pid.attr,
1965 static ssize_t o2hb_region_show(struct config_item *item,
1966 struct configfs_attribute *attr,
1969 struct o2hb_region *reg = to_o2hb_region(item);
1970 struct o2hb_region_attribute *o2hb_region_attr =
1971 container_of(attr, struct o2hb_region_attribute, attr);
1974 if (o2hb_region_attr->show)
1975 ret = o2hb_region_attr->show(reg, page);
1979 static ssize_t o2hb_region_store(struct config_item *item,
1980 struct configfs_attribute *attr,
1981 const char *page, size_t count)
1983 struct o2hb_region *reg = to_o2hb_region(item);
1984 struct o2hb_region_attribute *o2hb_region_attr =
1985 container_of(attr, struct o2hb_region_attribute, attr);
1986 ssize_t ret = -EINVAL;
1988 if (o2hb_region_attr->store)
1989 ret = o2hb_region_attr->store(reg, page, count);
1993 static struct configfs_item_operations o2hb_region_item_ops = {
1994 .release = o2hb_region_release,
1995 .show_attribute = o2hb_region_show,
1996 .store_attribute = o2hb_region_store,
1999 static struct config_item_type o2hb_region_type = {
2000 .ct_item_ops = &o2hb_region_item_ops,
2001 .ct_attrs = o2hb_region_attrs,
2002 .ct_owner = THIS_MODULE,
2007 struct o2hb_heartbeat_group {
2008 struct config_group hs_group;
2012 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
2015 container_of(group, struct o2hb_heartbeat_group, hs_group)
2019 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
2024 debugfs_create_dir(config_item_name(®->hr_item), dir);
2025 if (!reg->hr_debug_dir) {
2030 reg->hr_debug_livenodes =
2031 o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2033 &(reg->hr_db_livenodes),
2034 sizeof(*(reg->hr_db_livenodes)),
2035 O2HB_DB_TYPE_REGION_LIVENODES,
2036 sizeof(reg->hr_live_node_bitmap),
2037 O2NM_MAX_NODES, reg);
2038 if (!reg->hr_debug_livenodes) {
2043 reg->hr_debug_regnum =
2044 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2046 &(reg->hr_db_regnum),
2047 sizeof(*(reg->hr_db_regnum)),
2048 O2HB_DB_TYPE_REGION_NUMBER,
2049 0, O2NM_MAX_NODES, reg);
2050 if (!reg->hr_debug_regnum) {
2055 reg->hr_debug_elapsed_time =
2056 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2058 &(reg->hr_db_elapsed_time),
2059 sizeof(*(reg->hr_db_elapsed_time)),
2060 O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2062 if (!reg->hr_debug_elapsed_time) {
2067 reg->hr_debug_pinned =
2068 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2070 &(reg->hr_db_pinned),
2071 sizeof(*(reg->hr_db_pinned)),
2072 O2HB_DB_TYPE_REGION_PINNED,
2074 if (!reg->hr_debug_pinned) {
2084 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2087 struct o2hb_region *reg = NULL;
2090 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2092 return ERR_PTR(-ENOMEM);
2094 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2095 ret = -ENAMETOOLONG;
2099 spin_lock(&o2hb_live_lock);
2100 reg->hr_region_num = 0;
2101 if (o2hb_global_heartbeat_active()) {
2102 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2104 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2105 spin_unlock(&o2hb_live_lock);
2109 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2111 list_add_tail(®->hr_all_item, &o2hb_all_regions);
2112 spin_unlock(&o2hb_live_lock);
2114 config_item_init_type_name(®->hr_item, name, &o2hb_region_type);
2116 ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2118 config_item_put(®->hr_item);
2122 return ®->hr_item;
2125 return ERR_PTR(ret);
2128 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2129 struct config_item *item)
2131 struct task_struct *hb_task;
2132 struct o2hb_region *reg = to_o2hb_region(item);
2133 int quorum_region = 0;
2135 /* stop the thread when the user removes the region dir */
2136 spin_lock(&o2hb_live_lock);
2137 hb_task = reg->hr_task;
2138 reg->hr_task = NULL;
2139 reg->hr_item_dropped = 1;
2140 spin_unlock(&o2hb_live_lock);
2143 kthread_stop(hb_task);
2145 if (o2hb_global_heartbeat_active()) {
2146 spin_lock(&o2hb_live_lock);
2147 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2148 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2149 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2151 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2152 spin_unlock(&o2hb_live_lock);
2153 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2154 ((atomic_read(®->hr_steady_iterations) == 0) ?
2155 "stopped" : "start aborted"), config_item_name(item),
2160 * If we're racing a dev_write(), we need to wake them. They will
2161 * check reg->hr_task
2163 if (atomic_read(®->hr_steady_iterations) != 0) {
2164 reg->hr_aborted_start = 1;
2165 atomic_set(®->hr_steady_iterations, 0);
2166 wake_up(&o2hb_steady_queue);
2169 config_item_put(item);
2171 if (!o2hb_global_heartbeat_active() || !quorum_region)
2175 * If global heartbeat active and there are dependent users,
2176 * pin all regions if quorum region count <= CUT_OFF
2178 spin_lock(&o2hb_live_lock);
2180 if (!o2hb_dependent_users)
2183 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2184 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2185 o2hb_region_pin(NULL);
2188 spin_unlock(&o2hb_live_lock);
2191 struct o2hb_heartbeat_group_attribute {
2192 struct configfs_attribute attr;
2193 ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2194 ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2197 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2198 struct configfs_attribute *attr,
2201 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2202 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2203 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2206 if (o2hb_heartbeat_group_attr->show)
2207 ret = o2hb_heartbeat_group_attr->show(reg, page);
2211 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2212 struct configfs_attribute *attr,
2213 const char *page, size_t count)
2215 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2216 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2217 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2218 ssize_t ret = -EINVAL;
2220 if (o2hb_heartbeat_group_attr->store)
2221 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2225 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2228 return sprintf(page, "%u\n", o2hb_dead_threshold);
2231 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2236 char *p = (char *)page;
2238 tmp = simple_strtoul(p, &p, 10);
2239 if (!p || (*p && (*p != '\n')))
2242 /* this will validate ranges for us. */
2243 o2hb_dead_threshold_set((unsigned int) tmp);
2249 ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2252 return sprintf(page, "%s\n",
2253 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2257 ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2258 const char *page, size_t count)
2264 len = (page[count - 1] == '\n') ? count - 1 : count;
2268 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2269 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2272 ret = o2hb_global_heartbeat_mode_set(i);
2274 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2275 o2hb_heartbeat_mode_desc[i]);
2283 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2284 .attr = { .ca_owner = THIS_MODULE,
2285 .ca_name = "dead_threshold",
2286 .ca_mode = S_IRUGO | S_IWUSR },
2287 .show = o2hb_heartbeat_group_threshold_show,
2288 .store = o2hb_heartbeat_group_threshold_store,
2291 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2292 .attr = { .ca_owner = THIS_MODULE,
2294 .ca_mode = S_IRUGO | S_IWUSR },
2295 .show = o2hb_heartbeat_group_mode_show,
2296 .store = o2hb_heartbeat_group_mode_store,
2299 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2300 &o2hb_heartbeat_group_attr_threshold.attr,
2301 &o2hb_heartbeat_group_attr_mode.attr,
2305 static struct configfs_item_operations o2hb_heartbeat_group_item_ops = {
2306 .show_attribute = o2hb_heartbeat_group_show,
2307 .store_attribute = o2hb_heartbeat_group_store,
2310 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2311 .make_item = o2hb_heartbeat_group_make_item,
2312 .drop_item = o2hb_heartbeat_group_drop_item,
2315 static struct config_item_type o2hb_heartbeat_group_type = {
2316 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
2317 .ct_item_ops = &o2hb_heartbeat_group_item_ops,
2318 .ct_attrs = o2hb_heartbeat_group_attrs,
2319 .ct_owner = THIS_MODULE,
2322 /* this is just here to avoid touching group in heartbeat.h which the
2323 * entire damn world #includes */
2324 struct config_group *o2hb_alloc_hb_set(void)
2326 struct o2hb_heartbeat_group *hs = NULL;
2327 struct config_group *ret = NULL;
2329 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2333 config_group_init_type_name(&hs->hs_group, "heartbeat",
2334 &o2hb_heartbeat_group_type);
2336 ret = &hs->hs_group;
2343 void o2hb_free_hb_set(struct config_group *group)
2345 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2349 /* hb callback registration and issuing */
2351 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2353 if (type == O2HB_NUM_CB)
2354 return ERR_PTR(-EINVAL);
2356 return &o2hb_callbacks[type];
2359 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2360 enum o2hb_callback_type type,
2365 INIT_LIST_HEAD(&hc->hc_item);
2368 hc->hc_priority = priority;
2370 hc->hc_magic = O2HB_CB_MAGIC;
2372 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2375 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2376 * In global heartbeat mode, region_uuid passed is NULL.
2378 * In local, we only pin the matching region. In global we pin all the active
2381 static int o2hb_region_pin(const char *region_uuid)
2383 int ret = 0, found = 0;
2384 struct o2hb_region *reg;
2387 assert_spin_locked(&o2hb_live_lock);
2389 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2390 if (reg->hr_item_dropped)
2393 uuid = config_item_name(®->hr_item);
2395 /* local heartbeat */
2397 if (strcmp(region_uuid, uuid))
2402 if (reg->hr_item_pinned || reg->hr_item_dropped)
2405 /* Ignore ENOENT only for local hb (userdlm domain) */
2406 ret = o2nm_depend_item(®->hr_item);
2408 mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2409 reg->hr_item_pinned = 1;
2411 if (ret == -ENOENT && found)
2414 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2428 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2429 * In global heartbeat mode, region_uuid passed is NULL.
2431 * In local, we only unpin the matching region. In global we unpin all the
2434 static void o2hb_region_unpin(const char *region_uuid)
2436 struct o2hb_region *reg;
2440 assert_spin_locked(&o2hb_live_lock);
2442 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2443 if (reg->hr_item_dropped)
2446 uuid = config_item_name(®->hr_item);
2448 if (strcmp(region_uuid, uuid))
2453 if (reg->hr_item_pinned) {
2454 mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2455 o2nm_undepend_item(®->hr_item);
2456 reg->hr_item_pinned = 0;
2463 static int o2hb_region_inc_user(const char *region_uuid)
2467 spin_lock(&o2hb_live_lock);
2469 /* local heartbeat */
2470 if (!o2hb_global_heartbeat_active()) {
2471 ret = o2hb_region_pin(region_uuid);
2476 * if global heartbeat active and this is the first dependent user,
2477 * pin all regions if quorum region count <= CUT_OFF
2479 o2hb_dependent_users++;
2480 if (o2hb_dependent_users > 1)
2483 if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2484 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2485 ret = o2hb_region_pin(NULL);
2488 spin_unlock(&o2hb_live_lock);
2492 void o2hb_region_dec_user(const char *region_uuid)
2494 spin_lock(&o2hb_live_lock);
2496 /* local heartbeat */
2497 if (!o2hb_global_heartbeat_active()) {
2498 o2hb_region_unpin(region_uuid);
2503 * if global heartbeat active and there are no dependent users,
2504 * unpin all quorum regions
2506 o2hb_dependent_users--;
2507 if (!o2hb_dependent_users)
2508 o2hb_region_unpin(NULL);
2511 spin_unlock(&o2hb_live_lock);
2514 int o2hb_register_callback(const char *region_uuid,
2515 struct o2hb_callback_func *hc)
2517 struct o2hb_callback_func *f;
2518 struct o2hb_callback *hbcall;
2521 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2522 BUG_ON(!list_empty(&hc->hc_item));
2524 hbcall = hbcall_from_type(hc->hc_type);
2525 if (IS_ERR(hbcall)) {
2526 ret = PTR_ERR(hbcall);
2531 ret = o2hb_region_inc_user(region_uuid);
2538 down_write(&o2hb_callback_sem);
2540 list_for_each_entry(f, &hbcall->list, hc_item) {
2541 if (hc->hc_priority < f->hc_priority) {
2542 list_add_tail(&hc->hc_item, &f->hc_item);
2546 if (list_empty(&hc->hc_item))
2547 list_add_tail(&hc->hc_item, &hbcall->list);
2549 up_write(&o2hb_callback_sem);
2552 mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2553 ret, __builtin_return_address(0), hc);
2556 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2558 void o2hb_unregister_callback(const char *region_uuid,
2559 struct o2hb_callback_func *hc)
2561 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2563 mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2564 __builtin_return_address(0), hc);
2566 /* XXX Can this happen _with_ a region reference? */
2567 if (list_empty(&hc->hc_item))
2571 o2hb_region_dec_user(region_uuid);
2573 down_write(&o2hb_callback_sem);
2575 list_del_init(&hc->hc_item);
2577 up_write(&o2hb_callback_sem);
2579 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2581 int o2hb_check_node_heartbeating(u8 node_num)
2583 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2585 o2hb_fill_node_map(testing_map, sizeof(testing_map));
2586 if (!test_bit(node_num, testing_map)) {
2588 "node (%u) does not have heartbeating enabled.\n",
2595 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2597 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2599 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2601 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2602 if (!test_bit(node_num, testing_map)) {
2604 "node (%u) does not have heartbeating enabled.\n",
2611 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2613 /* Makes sure our local node is configured with a node number, and is
2615 int o2hb_check_local_node_heartbeating(void)
2619 /* if this node was set then we have networking */
2620 node_num = o2nm_this_node();
2621 if (node_num == O2NM_MAX_NODES) {
2622 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2626 return o2hb_check_node_heartbeating(node_num);
2628 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2631 * this is just a hack until we get the plumbing which flips file systems
2632 * read only and drops the hb ref instead of killing the node dead.
2634 void o2hb_stop_all_regions(void)
2636 struct o2hb_region *reg;
2638 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2640 spin_lock(&o2hb_live_lock);
2642 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2643 reg->hr_unclean_stop = 1;
2645 spin_unlock(&o2hb_live_lock);
2647 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2649 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2651 struct o2hb_region *reg;
2655 spin_lock(&o2hb_live_lock);
2658 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2659 if (reg->hr_item_dropped)
2662 mlog(0, "Region: %s\n", config_item_name(®->hr_item));
2663 if (numregs < max_regions) {
2664 memcpy(p, config_item_name(®->hr_item),
2665 O2HB_MAX_REGION_NAME_LEN);
2666 p += O2HB_MAX_REGION_NAME_LEN;
2671 spin_unlock(&o2hb_live_lock);
2675 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2677 int o2hb_global_heartbeat_active(void)
2679 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2681 EXPORT_SYMBOL(o2hb_global_heartbeat_active);