1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) 2004, 2005 Oracle. All rights reserved.
6 #include <linux/kernel.h>
7 #include <linux/sched.h>
8 #include <linux/jiffies.h>
9 #include <linux/module.h>
11 #include <linux/bio.h>
12 #include <linux/blkdev.h>
13 #include <linux/delay.h>
14 #include <linux/file.h>
15 #include <linux/kthread.h>
16 #include <linux/configfs.h>
17 #include <linux/random.h>
18 #include <linux/crc32.h>
19 #include <linux/time.h>
20 #include <linux/debugfs.h>
21 #include <linux/slab.h>
22 #include <linux/bitmap.h>
23 #include <linux/ktime.h>
24 #include "heartbeat.h"
26 #include "nodemanager.h"
33 * The first heartbeat pass had one global thread that would serialize all hb
34 * callback calls. This global serializing sem should only be removed once
35 * we've made sure that all callees can deal with being called concurrently
36 * from multiple hb region threads.
38 static DECLARE_RWSEM(o2hb_callback_sem);
41 * multiple hb threads are watching multiple regions. A node is live
42 * whenever any of the threads sees activity from the node in its region.
44 static DEFINE_SPINLOCK(o2hb_live_lock);
45 static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
46 static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
47 static LIST_HEAD(o2hb_node_events);
48 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
51 * In global heartbeat, we maintain a series of region bitmaps.
52 * - o2hb_region_bitmap allows us to limit the region number to max region.
53 * - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
54 * - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
56 * - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
58 static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
59 static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
60 static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
61 static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
63 #define O2HB_DB_TYPE_LIVENODES 0
64 #define O2HB_DB_TYPE_LIVEREGIONS 1
65 #define O2HB_DB_TYPE_QUORUMREGIONS 2
66 #define O2HB_DB_TYPE_FAILEDREGIONS 3
67 #define O2HB_DB_TYPE_REGION_LIVENODES 4
68 #define O2HB_DB_TYPE_REGION_NUMBER 5
69 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME 6
70 #define O2HB_DB_TYPE_REGION_PINNED 7
71 struct o2hb_debug_buf {
78 static struct o2hb_debug_buf *o2hb_db_livenodes;
79 static struct o2hb_debug_buf *o2hb_db_liveregions;
80 static struct o2hb_debug_buf *o2hb_db_quorumregions;
81 static struct o2hb_debug_buf *o2hb_db_failedregions;
83 #define O2HB_DEBUG_DIR "o2hb"
84 #define O2HB_DEBUG_LIVENODES "livenodes"
85 #define O2HB_DEBUG_LIVEREGIONS "live_regions"
86 #define O2HB_DEBUG_QUORUMREGIONS "quorum_regions"
87 #define O2HB_DEBUG_FAILEDREGIONS "failed_regions"
88 #define O2HB_DEBUG_REGION_NUMBER "num"
89 #define O2HB_DEBUG_REGION_ELAPSED_TIME "elapsed_time_in_ms"
90 #define O2HB_DEBUG_REGION_PINNED "pinned"
92 static struct dentry *o2hb_debug_dir;
94 static LIST_HEAD(o2hb_all_regions);
96 static struct o2hb_callback {
97 struct list_head list;
98 } o2hb_callbacks[O2HB_NUM_CB];
100 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
102 enum o2hb_heartbeat_modes {
103 O2HB_HEARTBEAT_LOCAL = 0,
104 O2HB_HEARTBEAT_GLOBAL,
105 O2HB_HEARTBEAT_NUM_MODES,
108 static const char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
109 "local", /* O2HB_HEARTBEAT_LOCAL */
110 "global", /* O2HB_HEARTBEAT_GLOBAL */
113 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
114 static unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
117 * o2hb_dependent_users tracks the number of registered callbacks that depend
118 * on heartbeat. o2net and o2dlm are two entities that register this callback.
119 * However only o2dlm depends on the heartbeat. It does not want the heartbeat
120 * to stop while a dlm domain is still active.
122 static unsigned int o2hb_dependent_users;
125 * In global heartbeat mode, all regions are pinned if there are one or more
126 * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
127 * regions are unpinned if the region count exceeds the cut off or the number
128 * of dependent users falls to zero.
130 #define O2HB_PIN_CUT_OFF 3
133 * In local heartbeat mode, we assume the dlm domain name to be the same as
134 * region uuid. This is true for domains created for the file system but not
135 * necessarily true for userdlm domains. This is a known limitation.
137 * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
138 * works for both file system and userdlm domains.
140 static int o2hb_region_pin(const char *region_uuid);
141 static void o2hb_region_unpin(const char *region_uuid);
143 /* Only sets a new threshold if there are no active regions.
145 * No locking or otherwise interesting code is required for reading
146 * o2hb_dead_threshold as it can't change once regions are active and
147 * it's not interesting to anyone until then anyway. */
148 static void o2hb_dead_threshold_set(unsigned int threshold)
150 if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
151 spin_lock(&o2hb_live_lock);
152 if (list_empty(&o2hb_all_regions))
153 o2hb_dead_threshold = threshold;
154 spin_unlock(&o2hb_live_lock);
158 static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
162 if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
163 spin_lock(&o2hb_live_lock);
164 if (list_empty(&o2hb_all_regions)) {
165 o2hb_heartbeat_mode = hb_mode;
168 spin_unlock(&o2hb_live_lock);
174 struct o2hb_node_event {
175 struct list_head hn_item;
176 enum o2hb_callback_type hn_event_type;
177 struct o2nm_node *hn_node;
181 struct o2hb_disk_slot {
182 struct o2hb_disk_heartbeat_block *ds_raw_block;
185 u64 ds_last_generation;
186 u16 ds_equal_samples;
187 u16 ds_changed_samples;
188 struct list_head ds_live_item;
191 /* each thread owns a region.. when we're asked to tear down the region
192 * we ask the thread to stop, who cleans up the region */
194 struct config_item hr_item;
196 struct list_head hr_all_item;
197 unsigned hr_unclean_stop:1,
203 /* protected by the hr_callback_sem */
204 struct task_struct *hr_task;
206 unsigned int hr_blocks;
207 unsigned long long hr_start_block;
209 unsigned int hr_block_bits;
210 unsigned int hr_block_bytes;
212 unsigned int hr_slots_per_page;
213 unsigned int hr_num_pages;
215 struct page **hr_slot_data;
216 struct block_device *hr_bdev;
217 struct o2hb_disk_slot *hr_slots;
219 /* live node map of this region */
220 unsigned long hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
221 unsigned int hr_region_num;
223 struct dentry *hr_debug_dir;
224 struct o2hb_debug_buf *hr_db_livenodes;
225 struct o2hb_debug_buf *hr_db_regnum;
226 struct o2hb_debug_buf *hr_db_elapsed_time;
227 struct o2hb_debug_buf *hr_db_pinned;
229 /* let the person setting up hb wait for it to return until it
230 * has reached a 'steady' state. This will be fixed when we have
231 * a more complete api that doesn't lead to this sort of fragility. */
232 atomic_t hr_steady_iterations;
234 /* terminate o2hb thread if it does not reach steady state
235 * (hr_steady_iterations == 0) within hr_unsteady_iterations */
236 atomic_t hr_unsteady_iterations;
238 unsigned int hr_timeout_ms;
240 /* randomized as the region goes up and down so that a node
241 * recognizes a node going up and down in one iteration */
244 struct delayed_work hr_write_timeout_work;
245 unsigned long hr_last_timeout_start;
247 /* negotiate timer, used to negotiate extending hb timeout. */
248 struct delayed_work hr_nego_timeout_work;
249 unsigned long hr_nego_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
251 /* Used during o2hb_check_slot to hold a copy of the block
252 * being checked because we temporarily have to zero out the
254 struct o2hb_disk_heartbeat_block *hr_tmp_block;
256 /* Message key for negotiate timeout message. */
258 struct list_head hr_handler_list;
260 /* last hb status, 0 for success, other value for error. */
261 int hr_last_hb_status;
264 struct o2hb_bio_wait_ctxt {
265 atomic_t wc_num_reqs;
266 struct completion wc_io_complete;
270 #define O2HB_NEGO_TIMEOUT_MS (O2HB_MAX_WRITE_TIMEOUT_MS/2)
273 O2HB_NEGO_TIMEOUT_MSG = 1,
274 O2HB_NEGO_APPROVE_MSG = 2,
277 struct o2hb_nego_msg {
281 static void o2hb_write_timeout(struct work_struct *work)
284 struct o2hb_region *reg =
285 container_of(work, struct o2hb_region,
286 hr_write_timeout_work.work);
288 mlog(ML_ERROR, "Heartbeat write timeout to device %pg after %u "
289 "milliseconds\n", reg->hr_bdev,
290 jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
292 if (o2hb_global_heartbeat_active()) {
293 spin_lock(&o2hb_live_lock);
294 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
295 set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
296 failed = bitmap_weight(o2hb_failed_region_bitmap,
298 quorum = bitmap_weight(o2hb_quorum_region_bitmap,
300 spin_unlock(&o2hb_live_lock);
302 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
306 * Fence if the number of failed regions >= half the number
309 if ((failed << 1) < quorum)
313 o2quo_disk_timeout();
316 static void o2hb_arm_timeout(struct o2hb_region *reg)
318 /* Arm writeout only after thread reaches steady state */
319 if (atomic_read(®->hr_steady_iterations) != 0)
322 mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
323 O2HB_MAX_WRITE_TIMEOUT_MS);
325 if (o2hb_global_heartbeat_active()) {
326 spin_lock(&o2hb_live_lock);
327 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
328 spin_unlock(&o2hb_live_lock);
330 cancel_delayed_work(®->hr_write_timeout_work);
331 schedule_delayed_work(®->hr_write_timeout_work,
332 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
334 cancel_delayed_work(®->hr_nego_timeout_work);
335 /* negotiate timeout must be less than write timeout. */
336 schedule_delayed_work(®->hr_nego_timeout_work,
337 msecs_to_jiffies(O2HB_NEGO_TIMEOUT_MS));
338 bitmap_zero(reg->hr_nego_node_bitmap, O2NM_MAX_NODES);
341 static void o2hb_disarm_timeout(struct o2hb_region *reg)
343 cancel_delayed_work_sync(®->hr_write_timeout_work);
344 cancel_delayed_work_sync(®->hr_nego_timeout_work);
347 static int o2hb_send_nego_msg(int key, int type, u8 target)
349 struct o2hb_nego_msg msg;
352 msg.node_num = o2nm_this_node();
354 ret = o2net_send_message(type, key, &msg, sizeof(msg),
357 if (ret == -EAGAIN || ret == -ENOMEM) {
365 static void o2hb_nego_timeout(struct work_struct *work)
367 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
368 int master_node, i, ret;
369 struct o2hb_region *reg;
371 reg = container_of(work, struct o2hb_region, hr_nego_timeout_work.work);
372 /* don't negotiate timeout if last hb failed since it is very
373 * possible io failed. Should let write timeout fence self.
375 if (reg->hr_last_hb_status)
378 o2hb_fill_node_map(live_node_bitmap, O2NM_MAX_NODES);
379 /* lowest node as master node to make negotiate decision. */
380 master_node = find_first_bit(live_node_bitmap, O2NM_MAX_NODES);
382 if (master_node == o2nm_this_node()) {
383 if (!test_bit(master_node, reg->hr_nego_node_bitmap)) {
384 printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%pg).\n",
385 o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000,
386 config_item_name(®->hr_item), reg->hr_bdev);
387 set_bit(master_node, reg->hr_nego_node_bitmap);
389 if (!bitmap_equal(reg->hr_nego_node_bitmap, live_node_bitmap,
391 /* check negotiate bitmap every second to do timeout
394 schedule_delayed_work(®->hr_nego_timeout_work,
395 msecs_to_jiffies(1000));
400 printk(KERN_NOTICE "o2hb: all nodes hb write hung, maybe region %s (%pg) is down.\n",
401 config_item_name(®->hr_item), reg->hr_bdev);
402 /* approve negotiate timeout request. */
403 o2hb_arm_timeout(reg);
406 while ((i = find_next_bit(live_node_bitmap,
407 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
408 if (i == master_node)
411 mlog(ML_HEARTBEAT, "send NEGO_APPROVE msg to node %d\n", i);
412 ret = o2hb_send_nego_msg(reg->hr_key,
413 O2HB_NEGO_APPROVE_MSG, i);
415 mlog(ML_ERROR, "send NEGO_APPROVE msg to node %d fail %d\n",
419 /* negotiate timeout with master node. */
420 printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%pg), negotiate timeout with node %d.\n",
421 o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000, config_item_name(®->hr_item),
422 reg->hr_bdev, master_node);
423 ret = o2hb_send_nego_msg(reg->hr_key, O2HB_NEGO_TIMEOUT_MSG,
426 mlog(ML_ERROR, "send NEGO_TIMEOUT msg to node %d fail %d\n",
431 static int o2hb_nego_timeout_handler(struct o2net_msg *msg, u32 len, void *data,
434 struct o2hb_region *reg = data;
435 struct o2hb_nego_msg *nego_msg;
437 nego_msg = (struct o2hb_nego_msg *)msg->buf;
438 printk(KERN_NOTICE "o2hb: receive negotiate timeout message from node %d on region %s (%pg).\n",
439 nego_msg->node_num, config_item_name(®->hr_item), reg->hr_bdev);
440 if (nego_msg->node_num < O2NM_MAX_NODES)
441 set_bit(nego_msg->node_num, reg->hr_nego_node_bitmap);
443 mlog(ML_ERROR, "got nego timeout message from bad node.\n");
448 static int o2hb_nego_approve_handler(struct o2net_msg *msg, u32 len, void *data,
451 struct o2hb_region *reg = data;
453 printk(KERN_NOTICE "o2hb: negotiate timeout approved by master node on region %s (%pg).\n",
454 config_item_name(®->hr_item), reg->hr_bdev);
455 o2hb_arm_timeout(reg);
459 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
461 atomic_set(&wc->wc_num_reqs, 1);
462 init_completion(&wc->wc_io_complete);
466 /* Used in error paths too */
467 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
470 /* sadly atomic_sub_and_test() isn't available on all platforms. The
471 * good news is that the fast path only completes one at a time */
473 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
475 complete(&wc->wc_io_complete);
480 static void o2hb_wait_on_io(struct o2hb_bio_wait_ctxt *wc)
482 o2hb_bio_wait_dec(wc, 1);
483 wait_for_completion(&wc->wc_io_complete);
486 static void o2hb_bio_end_io(struct bio *bio)
488 struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
490 if (bio->bi_status) {
491 mlog(ML_ERROR, "IO Error %d\n", bio->bi_status);
492 wc->wc_error = blk_status_to_errno(bio->bi_status);
495 o2hb_bio_wait_dec(wc, 1);
499 /* Setup a Bio to cover I/O against num_slots slots starting at
501 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
502 struct o2hb_bio_wait_ctxt *wc,
503 unsigned int *current_slot,
504 unsigned int max_slots, blk_opf_t opf)
506 int len, current_page;
507 unsigned int vec_len, vec_start;
508 unsigned int bits = reg->hr_block_bits;
509 unsigned int spp = reg->hr_slots_per_page;
510 unsigned int cs = *current_slot;
514 /* Testing has shown this allocation to take long enough under
515 * GFP_KERNEL that the local node can get fenced. It would be
516 * nicest if we could pre-allocate these bios and avoid this
518 bio = bio_alloc(reg->hr_bdev, 16, opf, GFP_ATOMIC);
520 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
521 bio = ERR_PTR(-ENOMEM);
525 /* Must put everything in 512 byte sectors for the bio... */
526 bio->bi_iter.bi_sector = (reg->hr_start_block + cs) << (bits - 9);
527 bio->bi_private = wc;
528 bio->bi_end_io = o2hb_bio_end_io;
530 vec_start = (cs << bits) % PAGE_SIZE;
531 while(cs < max_slots) {
532 current_page = cs / spp;
533 page = reg->hr_slot_data[current_page];
535 vec_len = min(PAGE_SIZE - vec_start,
536 (max_slots-cs) * (PAGE_SIZE/spp) );
538 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
539 current_page, vec_len, vec_start);
541 len = bio_add_page(bio, page, vec_len, vec_start);
542 if (len != vec_len) break;
544 cs += vec_len / (PAGE_SIZE/spp);
553 static int o2hb_read_slots(struct o2hb_region *reg,
554 unsigned int begin_slot,
555 unsigned int max_slots)
557 unsigned int current_slot = begin_slot;
559 struct o2hb_bio_wait_ctxt wc;
562 o2hb_bio_wait_init(&wc);
564 while(current_slot < max_slots) {
565 bio = o2hb_setup_one_bio(reg, &wc, ¤t_slot, max_slots,
568 status = PTR_ERR(bio);
573 atomic_inc(&wc.wc_num_reqs);
580 o2hb_wait_on_io(&wc);
581 if (wc.wc_error && !status)
582 status = wc.wc_error;
587 static int o2hb_issue_node_write(struct o2hb_region *reg,
588 struct o2hb_bio_wait_ctxt *write_wc)
594 o2hb_bio_wait_init(write_wc);
596 slot = o2nm_this_node();
598 bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1,
599 REQ_OP_WRITE | REQ_SYNC);
601 status = PTR_ERR(bio);
606 atomic_inc(&write_wc->wc_num_reqs);
614 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
615 struct o2hb_disk_heartbeat_block *hb_block)
620 /* We want to compute the block crc with a 0 value in the
621 * hb_cksum field. Save it off here and replace after the
623 old_cksum = hb_block->hb_cksum;
624 hb_block->hb_cksum = 0;
626 ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
628 hb_block->hb_cksum = old_cksum;
633 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
635 mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
636 "cksum = 0x%x, generation 0x%llx\n",
637 (long long)le64_to_cpu(hb_block->hb_seq),
638 hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
639 (long long)le64_to_cpu(hb_block->hb_generation));
642 static int o2hb_verify_crc(struct o2hb_region *reg,
643 struct o2hb_disk_heartbeat_block *hb_block)
647 read = le32_to_cpu(hb_block->hb_cksum);
648 computed = o2hb_compute_block_crc_le(reg, hb_block);
650 return read == computed;
654 * Compare the slot data with what we wrote in the last iteration.
655 * If the match fails, print an appropriate error message. This is to
656 * detect errors like... another node hearting on the same slot,
657 * flaky device that is losing writes, etc.
658 * Returns 1 if check succeeds, 0 otherwise.
660 static int o2hb_check_own_slot(struct o2hb_region *reg)
662 struct o2hb_disk_slot *slot;
663 struct o2hb_disk_heartbeat_block *hb_block;
666 slot = ®->hr_slots[o2nm_this_node()];
667 /* Don't check on our 1st timestamp */
668 if (!slot->ds_last_time)
671 hb_block = slot->ds_raw_block;
672 if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
673 le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
674 hb_block->hb_node == slot->ds_node_num)
677 #define ERRSTR1 "Another node is heartbeating on device"
678 #define ERRSTR2 "Heartbeat generation mismatch on device"
679 #define ERRSTR3 "Heartbeat sequence mismatch on device"
681 if (hb_block->hb_node != slot->ds_node_num)
683 else if (le64_to_cpu(hb_block->hb_generation) !=
684 slot->ds_last_generation)
689 mlog(ML_ERROR, "%s (%pg): expected(%u:0x%llx, 0x%llx), "
690 "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_bdev,
691 slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
692 (unsigned long long)slot->ds_last_time, hb_block->hb_node,
693 (unsigned long long)le64_to_cpu(hb_block->hb_generation),
694 (unsigned long long)le64_to_cpu(hb_block->hb_seq));
699 static inline void o2hb_prepare_block(struct o2hb_region *reg,
704 struct o2hb_disk_slot *slot;
705 struct o2hb_disk_heartbeat_block *hb_block;
707 node_num = o2nm_this_node();
708 slot = ®->hr_slots[node_num];
710 hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
711 memset(hb_block, 0, reg->hr_block_bytes);
712 /* TODO: time stuff */
713 cputime = ktime_get_real_seconds();
717 hb_block->hb_seq = cpu_to_le64(cputime);
718 hb_block->hb_node = node_num;
719 hb_block->hb_generation = cpu_to_le64(generation);
720 hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
722 /* This step must always happen last! */
723 hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
726 mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
727 (long long)generation,
728 le32_to_cpu(hb_block->hb_cksum));
731 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
732 struct o2nm_node *node,
735 struct o2hb_callback_func *f;
737 list_for_each_entry(f, &hbcall->list, hc_item) {
738 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
739 (f->hc_func)(node, idx, f->hc_data);
743 /* Will run the list in order until we process the passed event */
744 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
746 struct o2hb_callback *hbcall;
747 struct o2hb_node_event *event;
749 /* Holding callback sem assures we don't alter the callback
750 * lists when doing this, and serializes ourselves with other
751 * processes wanting callbacks. */
752 down_write(&o2hb_callback_sem);
754 spin_lock(&o2hb_live_lock);
755 while (!list_empty(&o2hb_node_events)
756 && !list_empty(&queued_event->hn_item)) {
757 event = list_entry(o2hb_node_events.next,
758 struct o2hb_node_event,
760 list_del_init(&event->hn_item);
761 spin_unlock(&o2hb_live_lock);
763 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
764 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
767 hbcall = hbcall_from_type(event->hn_event_type);
769 /* We should *never* have gotten on to the list with a
770 * bad type... This isn't something that we should try
771 * to recover from. */
772 BUG_ON(IS_ERR(hbcall));
774 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
776 spin_lock(&o2hb_live_lock);
778 spin_unlock(&o2hb_live_lock);
780 up_write(&o2hb_callback_sem);
783 static void o2hb_queue_node_event(struct o2hb_node_event *event,
784 enum o2hb_callback_type type,
785 struct o2nm_node *node,
788 assert_spin_locked(&o2hb_live_lock);
790 BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
792 event->hn_event_type = type;
793 event->hn_node = node;
794 event->hn_node_num = node_num;
796 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
797 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
799 list_add_tail(&event->hn_item, &o2hb_node_events);
802 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
804 struct o2hb_node_event event =
805 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
806 struct o2nm_node *node;
809 node = o2nm_get_node_by_num(slot->ds_node_num);
813 spin_lock(&o2hb_live_lock);
814 if (!list_empty(&slot->ds_live_item)) {
815 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
818 list_del_init(&slot->ds_live_item);
820 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
821 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
823 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
828 spin_unlock(&o2hb_live_lock);
831 o2hb_run_event_list(&event);
836 static void o2hb_set_quorum_device(struct o2hb_region *reg)
838 if (!o2hb_global_heartbeat_active())
841 /* Prevent race with o2hb_heartbeat_group_drop_item() */
842 if (kthread_should_stop())
845 /* Tag region as quorum only after thread reaches steady state */
846 if (atomic_read(®->hr_steady_iterations) != 0)
849 spin_lock(&o2hb_live_lock);
851 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
855 * A region can be added to the quorum only when it sees all
856 * live nodes heartbeat on it. In other words, the region has been
857 * added to all nodes.
859 if (!bitmap_equal(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
863 printk(KERN_NOTICE "o2hb: Region %s (%pg) is now a quorum device\n",
864 config_item_name(®->hr_item), reg->hr_bdev);
866 set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
869 * If global heartbeat active, unpin all regions if the
870 * region count > CUT_OFF
872 if (bitmap_weight(o2hb_quorum_region_bitmap,
873 O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
874 o2hb_region_unpin(NULL);
876 spin_unlock(&o2hb_live_lock);
879 static int o2hb_check_slot(struct o2hb_region *reg,
880 struct o2hb_disk_slot *slot)
882 int changed = 0, gen_changed = 0;
883 struct o2hb_node_event event =
884 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
885 struct o2nm_node *node;
886 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
888 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
889 unsigned int slot_dead_ms;
893 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
896 * If a node is no longer configured but is still in the livemap, we
897 * may need to clear that bit from the livemap.
899 node = o2nm_get_node_by_num(slot->ds_node_num);
901 spin_lock(&o2hb_live_lock);
902 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
903 spin_unlock(&o2hb_live_lock);
908 if (!o2hb_verify_crc(reg, hb_block)) {
909 /* all paths from here will drop o2hb_live_lock for
911 spin_lock(&o2hb_live_lock);
913 /* Don't print an error on the console in this case -
914 * a freshly formatted heartbeat area will not have a
916 if (list_empty(&slot->ds_live_item))
919 /* The node is live but pushed out a bad crc. We
920 * consider it a transient miss but don't populate any
921 * other values as they may be junk. */
922 mlog(ML_ERROR, "Node %d has written a bad crc to %pg\n",
923 slot->ds_node_num, reg->hr_bdev);
924 o2hb_dump_slot(hb_block);
926 slot->ds_equal_samples++;
930 /* we don't care if these wrap.. the state transitions below
931 * clear at the right places */
932 cputime = le64_to_cpu(hb_block->hb_seq);
933 if (slot->ds_last_time != cputime)
934 slot->ds_changed_samples++;
936 slot->ds_equal_samples++;
937 slot->ds_last_time = cputime;
939 /* The node changed heartbeat generations. We assume this to
940 * mean it dropped off but came back before we timed out. We
941 * want to consider it down for the time being but don't want
942 * to lose any changed_samples state we might build up to
943 * considering it live again. */
944 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
946 slot->ds_equal_samples = 0;
947 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
948 "to 0x%llx)\n", slot->ds_node_num,
949 (long long)slot->ds_last_generation,
950 (long long)le64_to_cpu(hb_block->hb_generation));
953 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
955 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
956 "seq %llu last %llu changed %u equal %u\n",
957 slot->ds_node_num, (long long)slot->ds_last_generation,
958 le32_to_cpu(hb_block->hb_cksum),
959 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
960 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
961 slot->ds_equal_samples);
963 spin_lock(&o2hb_live_lock);
966 /* dead nodes only come to life after some number of
967 * changes at any time during their dead time */
968 if (list_empty(&slot->ds_live_item) &&
969 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
970 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
971 slot->ds_node_num, (long long)slot->ds_last_generation);
973 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
975 /* first on the list generates a callback */
976 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
977 mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
978 "bitmap\n", slot->ds_node_num);
979 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
981 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
988 list_add_tail(&slot->ds_live_item,
989 &o2hb_live_slots[slot->ds_node_num]);
991 slot->ds_equal_samples = 0;
993 /* We want to be sure that all nodes agree on the
994 * number of milliseconds before a node will be
995 * considered dead. The self-fencing timeout is
996 * computed from this value, and a discrepancy might
997 * result in heartbeat calling a node dead when it
998 * hasn't self-fenced yet. */
999 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
1000 if (slot_dead_ms && slot_dead_ms != dead_ms) {
1001 /* TODO: Perhaps we can fail the region here. */
1002 mlog(ML_ERROR, "Node %d on device %pg has a dead count "
1003 "of %u ms, but our count is %u ms.\n"
1004 "Please double check your configuration values "
1005 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
1006 slot->ds_node_num, reg->hr_bdev, slot_dead_ms,
1012 /* if the list is dead, we're done.. */
1013 if (list_empty(&slot->ds_live_item))
1016 /* live nodes only go dead after enough consequtive missed
1017 * samples.. reset the missed counter whenever we see
1019 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
1020 mlog(ML_HEARTBEAT, "Node %d left my region\n",
1023 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1025 /* last off the live_slot generates a callback */
1026 list_del_init(&slot->ds_live_item);
1027 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
1028 mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
1029 "nodes bitmap\n", slot->ds_node_num);
1030 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
1032 /* node can be null */
1033 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
1034 node, slot->ds_node_num);
1040 /* We don't clear this because the node is still
1041 * actually writing new blocks. */
1043 slot->ds_changed_samples = 0;
1046 if (slot->ds_changed_samples) {
1047 slot->ds_changed_samples = 0;
1048 slot->ds_equal_samples = 0;
1051 spin_unlock(&o2hb_live_lock);
1054 o2hb_run_event_list(&event);
1057 o2nm_node_put(node);
1061 static int o2hb_highest_node(unsigned long *nodes, int numbits)
1063 return find_last_bit(nodes, numbits);
1066 static int o2hb_lowest_node(unsigned long *nodes, int numbits)
1068 return find_first_bit(nodes, numbits);
1071 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
1073 int i, ret, highest_node, lowest_node;
1074 int membership_change = 0, own_slot_ok = 0;
1075 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
1076 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
1077 struct o2hb_bio_wait_ctxt write_wc;
1079 ret = o2nm_configured_node_map(configured_nodes,
1080 sizeof(configured_nodes));
1087 * If a node is not configured but is in the livemap, we still need
1088 * to read the slot so as to be able to remove it from the livemap.
1090 o2hb_fill_node_map(live_node_bitmap, O2NM_MAX_NODES);
1092 while ((i = find_next_bit(live_node_bitmap,
1093 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1094 set_bit(i, configured_nodes);
1097 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1098 lowest_node = o2hb_lowest_node(configured_nodes, O2NM_MAX_NODES);
1099 if (highest_node >= O2NM_MAX_NODES || lowest_node >= O2NM_MAX_NODES) {
1100 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1105 /* No sense in reading the slots of nodes that don't exist
1106 * yet. Of course, if the node definitions have holes in them
1107 * then we're reading an empty slot anyway... Consider this
1109 ret = o2hb_read_slots(reg, lowest_node, highest_node + 1);
1115 /* With an up to date view of the slots, we can check that no
1116 * other node has been improperly configured to heartbeat in
1118 own_slot_ok = o2hb_check_own_slot(reg);
1120 /* fill in the proper info for our next heartbeat */
1121 o2hb_prepare_block(reg, reg->hr_generation);
1123 ret = o2hb_issue_node_write(reg, &write_wc);
1130 while((i = find_next_bit(configured_nodes,
1131 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1132 membership_change |= o2hb_check_slot(reg, ®->hr_slots[i]);
1136 * We have to be sure we've advertised ourselves on disk
1137 * before we can go to steady state. This ensures that
1138 * people we find in our steady state have seen us.
1140 o2hb_wait_on_io(&write_wc);
1141 if (write_wc.wc_error) {
1142 /* Do not re-arm the write timeout on I/O error - we
1143 * can't be sure that the new block ever made it to
1145 mlog(ML_ERROR, "Write error %d on device \"%pg\"\n",
1146 write_wc.wc_error, reg->hr_bdev);
1147 ret = write_wc.wc_error;
1151 /* Skip disarming the timeout if own slot has stale/bad data */
1153 o2hb_set_quorum_device(reg);
1154 o2hb_arm_timeout(reg);
1155 reg->hr_last_timeout_start = jiffies;
1159 /* let the person who launched us know when things are steady */
1160 if (atomic_read(®->hr_steady_iterations) != 0) {
1161 if (!ret && own_slot_ok && !membership_change) {
1162 if (atomic_dec_and_test(®->hr_steady_iterations))
1163 wake_up(&o2hb_steady_queue);
1167 if (atomic_read(®->hr_steady_iterations) != 0) {
1168 if (atomic_dec_and_test(®->hr_unsteady_iterations)) {
1169 printk(KERN_NOTICE "o2hb: Unable to stabilize "
1170 "heartbeat on region %s (%pg)\n",
1171 config_item_name(®->hr_item),
1173 atomic_set(®->hr_steady_iterations, 0);
1174 reg->hr_aborted_start = 1;
1175 wake_up(&o2hb_steady_queue);
1184 * we ride the region ref that the region dir holds. before the region
1185 * dir is removed and drops it ref it will wait to tear down this
1188 static int o2hb_thread(void *data)
1191 struct o2hb_region *reg = data;
1192 struct o2hb_bio_wait_ctxt write_wc;
1193 ktime_t before_hb, after_hb;
1194 unsigned int elapsed_msec;
1196 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1198 set_user_nice(current, MIN_NICE);
1201 ret = o2nm_depend_this_node();
1203 mlog(ML_ERROR, "Node has been deleted, ret = %d\n", ret);
1204 reg->hr_node_deleted = 1;
1205 wake_up(&o2hb_steady_queue);
1209 while (!kthread_should_stop() &&
1210 !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1211 /* We track the time spent inside
1212 * o2hb_do_disk_heartbeat so that we avoid more than
1213 * hr_timeout_ms between disk writes. On busy systems
1214 * this should result in a heartbeat which is less
1215 * likely to time itself out. */
1216 before_hb = ktime_get_real();
1218 ret = o2hb_do_disk_heartbeat(reg);
1219 reg->hr_last_hb_status = ret;
1221 after_hb = ktime_get_real();
1223 elapsed_msec = (unsigned int)
1224 ktime_ms_delta(after_hb, before_hb);
1227 "start = %lld, end = %lld, msec = %u, ret = %d\n",
1228 before_hb, after_hb, elapsed_msec, ret);
1230 if (!kthread_should_stop() &&
1231 elapsed_msec < reg->hr_timeout_ms) {
1232 /* the kthread api has blocked signals for us so no
1233 * need to record the return value. */
1234 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1238 o2hb_disarm_timeout(reg);
1240 /* unclean stop is only used in very bad situation */
1241 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1242 o2hb_shutdown_slot(®->hr_slots[i]);
1244 /* Explicit down notification - avoid forcing the other nodes
1245 * to timeout on this region when we could just as easily
1246 * write a clear generation - thus indicating to them that
1247 * this node has left this region.
1249 if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1250 o2hb_prepare_block(reg, 0);
1251 ret = o2hb_issue_node_write(reg, &write_wc);
1253 o2hb_wait_on_io(&write_wc);
1259 o2nm_undepend_this_node();
1261 mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1266 #ifdef CONFIG_DEBUG_FS
1267 static int o2hb_debug_open(struct inode *inode, struct file *file)
1269 struct o2hb_debug_buf *db = inode->i_private;
1270 struct o2hb_region *reg;
1271 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1277 /* max_nodes should be the largest bitmap we pass here */
1278 BUG_ON(sizeof(map) < db->db_size);
1280 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1284 switch (db->db_type) {
1285 case O2HB_DB_TYPE_LIVENODES:
1286 case O2HB_DB_TYPE_LIVEREGIONS:
1287 case O2HB_DB_TYPE_QUORUMREGIONS:
1288 case O2HB_DB_TYPE_FAILEDREGIONS:
1289 spin_lock(&o2hb_live_lock);
1290 memcpy(map, db->db_data, db->db_size);
1291 spin_unlock(&o2hb_live_lock);
1294 case O2HB_DB_TYPE_REGION_LIVENODES:
1295 spin_lock(&o2hb_live_lock);
1296 reg = (struct o2hb_region *)db->db_data;
1297 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1298 spin_unlock(&o2hb_live_lock);
1301 case O2HB_DB_TYPE_REGION_NUMBER:
1302 reg = (struct o2hb_region *)db->db_data;
1303 out += scnprintf(buf + out, PAGE_SIZE - out, "%d\n",
1304 reg->hr_region_num);
1307 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1308 reg = (struct o2hb_region *)db->db_data;
1309 lts = reg->hr_last_timeout_start;
1310 /* If 0, it has never been set before */
1312 lts = jiffies_to_msecs(jiffies - lts);
1313 out += scnprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1316 case O2HB_DB_TYPE_REGION_PINNED:
1317 reg = (struct o2hb_region *)db->db_data;
1318 out += scnprintf(buf + out, PAGE_SIZE - out, "%u\n",
1319 !!reg->hr_item_pinned);
1326 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1327 out += scnprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1328 out += scnprintf(buf + out, PAGE_SIZE - out, "\n");
1331 i_size_write(inode, out);
1333 file->private_data = buf;
1340 static int o2hb_debug_release(struct inode *inode, struct file *file)
1342 kfree(file->private_data);
1346 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1347 size_t nbytes, loff_t *ppos)
1349 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1350 i_size_read(file->f_mapping->host));
1353 static int o2hb_debug_open(struct inode *inode, struct file *file)
1357 static int o2hb_debug_release(struct inode *inode, struct file *file)
1361 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1362 size_t nbytes, loff_t *ppos)
1366 #endif /* CONFIG_DEBUG_FS */
1368 static const struct file_operations o2hb_debug_fops = {
1369 .open = o2hb_debug_open,
1370 .release = o2hb_debug_release,
1371 .read = o2hb_debug_read,
1372 .llseek = generic_file_llseek,
1375 void o2hb_exit(void)
1377 debugfs_remove_recursive(o2hb_debug_dir);
1378 kfree(o2hb_db_livenodes);
1379 kfree(o2hb_db_liveregions);
1380 kfree(o2hb_db_quorumregions);
1381 kfree(o2hb_db_failedregions);
1384 static void o2hb_debug_create(const char *name, struct dentry *dir,
1385 struct o2hb_debug_buf **db, int db_len, int type,
1386 int size, int len, void *data)
1388 *db = kmalloc(db_len, GFP_KERNEL);
1392 (*db)->db_type = type;
1393 (*db)->db_size = size;
1394 (*db)->db_len = len;
1395 (*db)->db_data = data;
1397 debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db, &o2hb_debug_fops);
1400 static void o2hb_debug_init(void)
1402 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1404 o2hb_debug_create(O2HB_DEBUG_LIVENODES, o2hb_debug_dir,
1405 &o2hb_db_livenodes, sizeof(*o2hb_db_livenodes),
1406 O2HB_DB_TYPE_LIVENODES, sizeof(o2hb_live_node_bitmap),
1407 O2NM_MAX_NODES, o2hb_live_node_bitmap);
1409 o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS, o2hb_debug_dir,
1410 &o2hb_db_liveregions, sizeof(*o2hb_db_liveregions),
1411 O2HB_DB_TYPE_LIVEREGIONS,
1412 sizeof(o2hb_live_region_bitmap), O2NM_MAX_REGIONS,
1413 o2hb_live_region_bitmap);
1415 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS, o2hb_debug_dir,
1416 &o2hb_db_quorumregions,
1417 sizeof(*o2hb_db_quorumregions),
1418 O2HB_DB_TYPE_QUORUMREGIONS,
1419 sizeof(o2hb_quorum_region_bitmap), O2NM_MAX_REGIONS,
1420 o2hb_quorum_region_bitmap);
1422 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS, o2hb_debug_dir,
1423 &o2hb_db_failedregions,
1424 sizeof(*o2hb_db_failedregions),
1425 O2HB_DB_TYPE_FAILEDREGIONS,
1426 sizeof(o2hb_failed_region_bitmap), O2NM_MAX_REGIONS,
1427 o2hb_failed_region_bitmap);
1430 void o2hb_init(void)
1434 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1435 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1437 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1438 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1440 bitmap_zero(o2hb_live_node_bitmap, O2NM_MAX_NODES);
1441 bitmap_zero(o2hb_region_bitmap, O2NM_MAX_REGIONS);
1442 bitmap_zero(o2hb_live_region_bitmap, O2NM_MAX_REGIONS);
1443 bitmap_zero(o2hb_quorum_region_bitmap, O2NM_MAX_REGIONS);
1444 bitmap_zero(o2hb_failed_region_bitmap, O2NM_MAX_REGIONS);
1446 o2hb_dependent_users = 0;
1451 /* if we're already in a callback then we're already serialized by the sem */
1452 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1455 bitmap_copy(map, o2hb_live_node_bitmap, bits);
1459 * get a map of all nodes that are heartbeating in any regions
1461 void o2hb_fill_node_map(unsigned long *map, unsigned int bits)
1463 /* callers want to serialize this map and callbacks so that they
1464 * can trust that they don't miss nodes coming to the party */
1465 down_read(&o2hb_callback_sem);
1466 spin_lock(&o2hb_live_lock);
1467 o2hb_fill_node_map_from_callback(map, bits);
1468 spin_unlock(&o2hb_live_lock);
1469 up_read(&o2hb_callback_sem);
1471 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1474 * heartbeat configfs bits. The heartbeat set is a default set under
1475 * the cluster set in nodemanager.c.
1478 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1480 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1483 /* drop_item only drops its ref after killing the thread, nothing should
1484 * be using the region anymore. this has to clean up any state that
1485 * attributes might have built up. */
1486 static void o2hb_region_release(struct config_item *item)
1490 struct o2hb_region *reg = to_o2hb_region(item);
1492 mlog(ML_HEARTBEAT, "hb region release (%pg)\n", reg->hr_bdev);
1494 kfree(reg->hr_tmp_block);
1496 if (reg->hr_slot_data) {
1497 for (i = 0; i < reg->hr_num_pages; i++) {
1498 page = reg->hr_slot_data[i];
1502 kfree(reg->hr_slot_data);
1506 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1508 kfree(reg->hr_slots);
1510 debugfs_remove_recursive(reg->hr_debug_dir);
1511 kfree(reg->hr_db_livenodes);
1512 kfree(reg->hr_db_regnum);
1513 kfree(reg->hr_db_elapsed_time);
1514 kfree(reg->hr_db_pinned);
1516 spin_lock(&o2hb_live_lock);
1517 list_del(®->hr_all_item);
1518 spin_unlock(&o2hb_live_lock);
1520 o2net_unregister_handler_list(®->hr_handler_list);
1524 static int o2hb_read_block_input(struct o2hb_region *reg,
1526 unsigned long *ret_bytes,
1527 unsigned int *ret_bits)
1529 unsigned long bytes;
1530 char *p = (char *)page;
1532 bytes = simple_strtoul(p, &p, 0);
1533 if (!p || (*p && (*p != '\n')))
1536 /* Heartbeat and fs min / max block sizes are the same. */
1537 if (bytes > 4096 || bytes < 512)
1539 if (hweight16(bytes) != 1)
1545 *ret_bits = ffs(bytes) - 1;
1550 static ssize_t o2hb_region_block_bytes_show(struct config_item *item,
1553 return sprintf(page, "%u\n", to_o2hb_region(item)->hr_block_bytes);
1556 static ssize_t o2hb_region_block_bytes_store(struct config_item *item,
1560 struct o2hb_region *reg = to_o2hb_region(item);
1562 unsigned long block_bytes;
1563 unsigned int block_bits;
1568 status = o2hb_read_block_input(reg, page, &block_bytes,
1573 reg->hr_block_bytes = (unsigned int)block_bytes;
1574 reg->hr_block_bits = block_bits;
1579 static ssize_t o2hb_region_start_block_show(struct config_item *item,
1582 return sprintf(page, "%llu\n", to_o2hb_region(item)->hr_start_block);
1585 static ssize_t o2hb_region_start_block_store(struct config_item *item,
1589 struct o2hb_region *reg = to_o2hb_region(item);
1590 unsigned long long tmp;
1591 char *p = (char *)page;
1597 ret = kstrtoull(p, 0, &tmp);
1601 reg->hr_start_block = tmp;
1606 static ssize_t o2hb_region_blocks_show(struct config_item *item, char *page)
1608 return sprintf(page, "%d\n", to_o2hb_region(item)->hr_blocks);
1611 static ssize_t o2hb_region_blocks_store(struct config_item *item,
1615 struct o2hb_region *reg = to_o2hb_region(item);
1617 char *p = (char *)page;
1622 tmp = simple_strtoul(p, &p, 0);
1623 if (!p || (*p && (*p != '\n')))
1626 if (tmp > O2NM_MAX_NODES || tmp == 0)
1629 reg->hr_blocks = (unsigned int)tmp;
1634 static ssize_t o2hb_region_dev_show(struct config_item *item, char *page)
1636 unsigned int ret = 0;
1638 if (to_o2hb_region(item)->hr_bdev)
1639 ret = sprintf(page, "%pg\n", to_o2hb_region(item)->hr_bdev);
1644 static void o2hb_init_region_params(struct o2hb_region *reg)
1646 reg->hr_slots_per_page = PAGE_SIZE >> reg->hr_block_bits;
1647 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1649 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1650 reg->hr_start_block, reg->hr_blocks);
1651 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1652 reg->hr_block_bytes, reg->hr_block_bits);
1653 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1654 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1657 static int o2hb_map_slot_data(struct o2hb_region *reg)
1660 unsigned int last_slot;
1661 unsigned int spp = reg->hr_slots_per_page;
1664 struct o2hb_disk_slot *slot;
1666 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1667 if (reg->hr_tmp_block == NULL)
1670 reg->hr_slots = kcalloc(reg->hr_blocks,
1671 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1672 if (reg->hr_slots == NULL)
1675 for(i = 0; i < reg->hr_blocks; i++) {
1676 slot = ®->hr_slots[i];
1677 slot->ds_node_num = i;
1678 INIT_LIST_HEAD(&slot->ds_live_item);
1679 slot->ds_raw_block = NULL;
1682 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1683 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1684 "at %u blocks per page\n",
1685 reg->hr_num_pages, reg->hr_blocks, spp);
1687 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1689 if (!reg->hr_slot_data)
1692 for(i = 0; i < reg->hr_num_pages; i++) {
1693 page = alloc_page(GFP_KERNEL);
1697 reg->hr_slot_data[i] = page;
1699 last_slot = i * spp;
1700 raw = page_address(page);
1702 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1704 BUG_ON((j + last_slot) >= reg->hr_blocks);
1706 slot = ®->hr_slots[j + last_slot];
1707 slot->ds_raw_block =
1708 (struct o2hb_disk_heartbeat_block *) raw;
1710 raw += reg->hr_block_bytes;
1717 /* Read in all the slots available and populate the tracking
1718 * structures so that we can start with a baseline idea of what's
1720 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1723 struct o2hb_disk_slot *slot;
1724 struct o2hb_disk_heartbeat_block *hb_block;
1726 ret = o2hb_read_slots(reg, 0, reg->hr_blocks);
1730 /* We only want to get an idea of the values initially in each
1731 * slot, so we do no verification - o2hb_check_slot will
1732 * actually determine if each configured slot is valid and
1733 * whether any values have changed. */
1734 for(i = 0; i < reg->hr_blocks; i++) {
1735 slot = ®->hr_slots[i];
1736 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1738 /* Only fill the values that o2hb_check_slot uses to
1739 * determine changing slots */
1740 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1741 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1748 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1749 static ssize_t o2hb_region_dev_store(struct config_item *item,
1753 struct o2hb_region *reg = to_o2hb_region(item);
1754 struct task_struct *hb_task;
1757 char *p = (char *)page;
1759 ssize_t ret = -EINVAL;
1765 /* We can't heartbeat without having had our node number
1766 * configured yet. */
1767 if (o2nm_this_node() == O2NM_MAX_NODES)
1770 fd = simple_strtol(p, &p, 0);
1771 if (!p || (*p && (*p != '\n')))
1774 if (fd < 0 || fd >= INT_MAX)
1781 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1782 reg->hr_block_bytes == 0)
1785 if (!S_ISBLK(f.file->f_mapping->host->i_mode))
1788 reg->hr_bdev = blkdev_get_by_dev(f.file->f_mapping->host->i_rdev,
1789 FMODE_WRITE | FMODE_READ, NULL);
1790 if (IS_ERR(reg->hr_bdev)) {
1791 ret = PTR_ERR(reg->hr_bdev);
1792 reg->hr_bdev = NULL;
1796 sectsize = bdev_logical_block_size(reg->hr_bdev);
1797 if (sectsize != reg->hr_block_bytes) {
1799 "blocksize %u incorrect for device, expected %d",
1800 reg->hr_block_bytes, sectsize);
1805 o2hb_init_region_params(reg);
1807 /* Generation of zero is invalid */
1809 get_random_bytes(®->hr_generation,
1810 sizeof(reg->hr_generation));
1811 } while (reg->hr_generation == 0);
1813 ret = o2hb_map_slot_data(reg);
1819 ret = o2hb_populate_slot_data(reg);
1825 INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout);
1826 INIT_DELAYED_WORK(®->hr_nego_timeout_work, o2hb_nego_timeout);
1829 * A node is considered live after it has beat LIVE_THRESHOLD
1830 * times. We're not steady until we've given them a chance
1831 * _after_ our first read.
1832 * The default threshold is bare minimum so as to limit the delay
1833 * during mounts. For global heartbeat, the threshold doubled for the
1836 live_threshold = O2HB_LIVE_THRESHOLD;
1837 if (o2hb_global_heartbeat_active()) {
1838 spin_lock(&o2hb_live_lock);
1839 if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1840 live_threshold <<= 1;
1841 spin_unlock(&o2hb_live_lock);
1844 atomic_set(®->hr_steady_iterations, live_threshold);
1845 /* unsteady_iterations is triple the steady_iterations */
1846 atomic_set(®->hr_unsteady_iterations, (live_threshold * 3));
1848 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1849 reg->hr_item.ci_name);
1850 if (IS_ERR(hb_task)) {
1851 ret = PTR_ERR(hb_task);
1856 spin_lock(&o2hb_live_lock);
1857 reg->hr_task = hb_task;
1858 spin_unlock(&o2hb_live_lock);
1860 ret = wait_event_interruptible(o2hb_steady_queue,
1861 atomic_read(®->hr_steady_iterations) == 0 ||
1862 reg->hr_node_deleted);
1864 atomic_set(®->hr_steady_iterations, 0);
1865 reg->hr_aborted_start = 1;
1868 if (reg->hr_aborted_start) {
1873 if (reg->hr_node_deleted) {
1878 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1879 spin_lock(&o2hb_live_lock);
1880 hb_task = reg->hr_task;
1881 if (o2hb_global_heartbeat_active())
1882 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1883 spin_unlock(&o2hb_live_lock);
1890 if (hb_task && o2hb_global_heartbeat_active())
1891 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%pg)\n",
1892 config_item_name(®->hr_item), reg->hr_bdev);
1896 blkdev_put(reg->hr_bdev, FMODE_READ | FMODE_WRITE);
1897 reg->hr_bdev = NULL;
1905 static ssize_t o2hb_region_pid_show(struct config_item *item, char *page)
1907 struct o2hb_region *reg = to_o2hb_region(item);
1910 spin_lock(&o2hb_live_lock);
1912 pid = task_pid_nr(reg->hr_task);
1913 spin_unlock(&o2hb_live_lock);
1918 return sprintf(page, "%u\n", pid);
1921 CONFIGFS_ATTR(o2hb_region_, block_bytes);
1922 CONFIGFS_ATTR(o2hb_region_, start_block);
1923 CONFIGFS_ATTR(o2hb_region_, blocks);
1924 CONFIGFS_ATTR(o2hb_region_, dev);
1925 CONFIGFS_ATTR_RO(o2hb_region_, pid);
1927 static struct configfs_attribute *o2hb_region_attrs[] = {
1928 &o2hb_region_attr_block_bytes,
1929 &o2hb_region_attr_start_block,
1930 &o2hb_region_attr_blocks,
1931 &o2hb_region_attr_dev,
1932 &o2hb_region_attr_pid,
1936 static struct configfs_item_operations o2hb_region_item_ops = {
1937 .release = o2hb_region_release,
1940 static const struct config_item_type o2hb_region_type = {
1941 .ct_item_ops = &o2hb_region_item_ops,
1942 .ct_attrs = o2hb_region_attrs,
1943 .ct_owner = THIS_MODULE,
1948 struct o2hb_heartbeat_group {
1949 struct config_group hs_group;
1953 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1956 container_of(group, struct o2hb_heartbeat_group, hs_group)
1960 static void o2hb_debug_region_init(struct o2hb_region *reg,
1961 struct dentry *parent)
1965 dir = debugfs_create_dir(config_item_name(®->hr_item), parent);
1966 reg->hr_debug_dir = dir;
1968 o2hb_debug_create(O2HB_DEBUG_LIVENODES, dir, &(reg->hr_db_livenodes),
1969 sizeof(*(reg->hr_db_livenodes)),
1970 O2HB_DB_TYPE_REGION_LIVENODES,
1971 sizeof(reg->hr_live_node_bitmap), O2NM_MAX_NODES,
1974 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER, dir, &(reg->hr_db_regnum),
1975 sizeof(*(reg->hr_db_regnum)),
1976 O2HB_DB_TYPE_REGION_NUMBER, 0, O2NM_MAX_NODES, reg);
1978 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME, dir,
1979 &(reg->hr_db_elapsed_time),
1980 sizeof(*(reg->hr_db_elapsed_time)),
1981 O2HB_DB_TYPE_REGION_ELAPSED_TIME, 0, 0, reg);
1983 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED, dir, &(reg->hr_db_pinned),
1984 sizeof(*(reg->hr_db_pinned)),
1985 O2HB_DB_TYPE_REGION_PINNED, 0, 0, reg);
1989 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
1992 struct o2hb_region *reg = NULL;
1995 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
1997 return ERR_PTR(-ENOMEM);
1999 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2000 ret = -ENAMETOOLONG;
2004 spin_lock(&o2hb_live_lock);
2005 reg->hr_region_num = 0;
2006 if (o2hb_global_heartbeat_active()) {
2007 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2009 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2010 spin_unlock(&o2hb_live_lock);
2014 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2016 list_add_tail(®->hr_all_item, &o2hb_all_regions);
2017 spin_unlock(&o2hb_live_lock);
2019 config_item_init_type_name(®->hr_item, name, &o2hb_region_type);
2021 /* this is the same way to generate msg key as dlm, for local heartbeat,
2022 * name is also the same, so make initial crc value different to avoid
2023 * message key conflict.
2025 reg->hr_key = crc32_le(reg->hr_region_num + O2NM_MAX_REGIONS,
2026 name, strlen(name));
2027 INIT_LIST_HEAD(®->hr_handler_list);
2028 ret = o2net_register_handler(O2HB_NEGO_TIMEOUT_MSG, reg->hr_key,
2029 sizeof(struct o2hb_nego_msg),
2030 o2hb_nego_timeout_handler,
2031 reg, NULL, ®->hr_handler_list);
2035 ret = o2net_register_handler(O2HB_NEGO_APPROVE_MSG, reg->hr_key,
2036 sizeof(struct o2hb_nego_msg),
2037 o2hb_nego_approve_handler,
2038 reg, NULL, ®->hr_handler_list);
2040 goto unregister_handler;
2042 o2hb_debug_region_init(reg, o2hb_debug_dir);
2044 return ®->hr_item;
2047 o2net_unregister_handler_list(®->hr_handler_list);
2049 spin_lock(&o2hb_live_lock);
2050 list_del(®->hr_all_item);
2051 if (o2hb_global_heartbeat_active())
2052 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2053 spin_unlock(&o2hb_live_lock);
2056 return ERR_PTR(ret);
2059 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2060 struct config_item *item)
2062 struct task_struct *hb_task;
2063 struct o2hb_region *reg = to_o2hb_region(item);
2064 int quorum_region = 0;
2066 /* stop the thread when the user removes the region dir */
2067 spin_lock(&o2hb_live_lock);
2068 hb_task = reg->hr_task;
2069 reg->hr_task = NULL;
2070 reg->hr_item_dropped = 1;
2071 spin_unlock(&o2hb_live_lock);
2074 kthread_stop(hb_task);
2076 if (o2hb_global_heartbeat_active()) {
2077 spin_lock(&o2hb_live_lock);
2078 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2079 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2080 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2082 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2083 spin_unlock(&o2hb_live_lock);
2084 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%pg)\n",
2085 ((atomic_read(®->hr_steady_iterations) == 0) ?
2086 "stopped" : "start aborted"), config_item_name(item),
2091 * If we're racing a dev_write(), we need to wake them. They will
2092 * check reg->hr_task
2094 if (atomic_read(®->hr_steady_iterations) != 0) {
2095 reg->hr_aborted_start = 1;
2096 atomic_set(®->hr_steady_iterations, 0);
2097 wake_up(&o2hb_steady_queue);
2100 config_item_put(item);
2102 if (!o2hb_global_heartbeat_active() || !quorum_region)
2106 * If global heartbeat active and there are dependent users,
2107 * pin all regions if quorum region count <= CUT_OFF
2109 spin_lock(&o2hb_live_lock);
2111 if (!o2hb_dependent_users)
2114 if (bitmap_weight(o2hb_quorum_region_bitmap,
2115 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2116 o2hb_region_pin(NULL);
2119 spin_unlock(&o2hb_live_lock);
2122 static ssize_t o2hb_heartbeat_group_dead_threshold_show(struct config_item *item,
2125 return sprintf(page, "%u\n", o2hb_dead_threshold);
2128 static ssize_t o2hb_heartbeat_group_dead_threshold_store(struct config_item *item,
2129 const char *page, size_t count)
2132 char *p = (char *)page;
2134 tmp = simple_strtoul(p, &p, 10);
2135 if (!p || (*p && (*p != '\n')))
2138 /* this will validate ranges for us. */
2139 o2hb_dead_threshold_set((unsigned int) tmp);
2144 static ssize_t o2hb_heartbeat_group_mode_show(struct config_item *item,
2147 return sprintf(page, "%s\n",
2148 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2151 static ssize_t o2hb_heartbeat_group_mode_store(struct config_item *item,
2152 const char *page, size_t count)
2158 len = (page[count - 1] == '\n') ? count - 1 : count;
2162 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2163 if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
2166 ret = o2hb_global_heartbeat_mode_set(i);
2168 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2169 o2hb_heartbeat_mode_desc[i]);
2177 CONFIGFS_ATTR(o2hb_heartbeat_group_, dead_threshold);
2178 CONFIGFS_ATTR(o2hb_heartbeat_group_, mode);
2180 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2181 &o2hb_heartbeat_group_attr_dead_threshold,
2182 &o2hb_heartbeat_group_attr_mode,
2186 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2187 .make_item = o2hb_heartbeat_group_make_item,
2188 .drop_item = o2hb_heartbeat_group_drop_item,
2191 static const struct config_item_type o2hb_heartbeat_group_type = {
2192 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
2193 .ct_attrs = o2hb_heartbeat_group_attrs,
2194 .ct_owner = THIS_MODULE,
2197 /* this is just here to avoid touching group in heartbeat.h which the
2198 * entire damn world #includes */
2199 struct config_group *o2hb_alloc_hb_set(void)
2201 struct o2hb_heartbeat_group *hs = NULL;
2202 struct config_group *ret = NULL;
2204 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2208 config_group_init_type_name(&hs->hs_group, "heartbeat",
2209 &o2hb_heartbeat_group_type);
2211 ret = &hs->hs_group;
2218 void o2hb_free_hb_set(struct config_group *group)
2220 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2224 /* hb callback registration and issuing */
2226 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2228 if (type == O2HB_NUM_CB)
2229 return ERR_PTR(-EINVAL);
2231 return &o2hb_callbacks[type];
2234 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2235 enum o2hb_callback_type type,
2240 INIT_LIST_HEAD(&hc->hc_item);
2243 hc->hc_priority = priority;
2245 hc->hc_magic = O2HB_CB_MAGIC;
2247 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2250 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2251 * In global heartbeat mode, region_uuid passed is NULL.
2253 * In local, we only pin the matching region. In global we pin all the active
2256 static int o2hb_region_pin(const char *region_uuid)
2258 int ret = 0, found = 0;
2259 struct o2hb_region *reg;
2262 assert_spin_locked(&o2hb_live_lock);
2264 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2265 if (reg->hr_item_dropped)
2268 uuid = config_item_name(®->hr_item);
2270 /* local heartbeat */
2272 if (strcmp(region_uuid, uuid))
2277 if (reg->hr_item_pinned || reg->hr_item_dropped)
2280 /* Ignore ENOENT only for local hb (userdlm domain) */
2281 ret = o2nm_depend_item(®->hr_item);
2283 mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2284 reg->hr_item_pinned = 1;
2286 if (ret == -ENOENT && found)
2289 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2303 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2304 * In global heartbeat mode, region_uuid passed is NULL.
2306 * In local, we only unpin the matching region. In global we unpin all the
2309 static void o2hb_region_unpin(const char *region_uuid)
2311 struct o2hb_region *reg;
2315 assert_spin_locked(&o2hb_live_lock);
2317 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2318 if (reg->hr_item_dropped)
2321 uuid = config_item_name(®->hr_item);
2323 if (strcmp(region_uuid, uuid))
2328 if (reg->hr_item_pinned) {
2329 mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2330 o2nm_undepend_item(®->hr_item);
2331 reg->hr_item_pinned = 0;
2338 static int o2hb_region_inc_user(const char *region_uuid)
2342 spin_lock(&o2hb_live_lock);
2344 /* local heartbeat */
2345 if (!o2hb_global_heartbeat_active()) {
2346 ret = o2hb_region_pin(region_uuid);
2351 * if global heartbeat active and this is the first dependent user,
2352 * pin all regions if quorum region count <= CUT_OFF
2354 o2hb_dependent_users++;
2355 if (o2hb_dependent_users > 1)
2358 if (bitmap_weight(o2hb_quorum_region_bitmap,
2359 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2360 ret = o2hb_region_pin(NULL);
2363 spin_unlock(&o2hb_live_lock);
2367 static void o2hb_region_dec_user(const char *region_uuid)
2369 spin_lock(&o2hb_live_lock);
2371 /* local heartbeat */
2372 if (!o2hb_global_heartbeat_active()) {
2373 o2hb_region_unpin(region_uuid);
2378 * if global heartbeat active and there are no dependent users,
2379 * unpin all quorum regions
2381 o2hb_dependent_users--;
2382 if (!o2hb_dependent_users)
2383 o2hb_region_unpin(NULL);
2386 spin_unlock(&o2hb_live_lock);
2389 int o2hb_register_callback(const char *region_uuid,
2390 struct o2hb_callback_func *hc)
2392 struct o2hb_callback_func *f;
2393 struct o2hb_callback *hbcall;
2396 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2397 BUG_ON(!list_empty(&hc->hc_item));
2399 hbcall = hbcall_from_type(hc->hc_type);
2400 if (IS_ERR(hbcall)) {
2401 ret = PTR_ERR(hbcall);
2406 ret = o2hb_region_inc_user(region_uuid);
2413 down_write(&o2hb_callback_sem);
2415 list_for_each_entry(f, &hbcall->list, hc_item) {
2416 if (hc->hc_priority < f->hc_priority) {
2417 list_add_tail(&hc->hc_item, &f->hc_item);
2421 if (list_empty(&hc->hc_item))
2422 list_add_tail(&hc->hc_item, &hbcall->list);
2424 up_write(&o2hb_callback_sem);
2427 mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2428 ret, __builtin_return_address(0), hc);
2431 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2433 void o2hb_unregister_callback(const char *region_uuid,
2434 struct o2hb_callback_func *hc)
2436 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2438 mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2439 __builtin_return_address(0), hc);
2441 /* XXX Can this happen _with_ a region reference? */
2442 if (list_empty(&hc->hc_item))
2446 o2hb_region_dec_user(region_uuid);
2448 down_write(&o2hb_callback_sem);
2450 list_del_init(&hc->hc_item);
2452 up_write(&o2hb_callback_sem);
2454 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2456 int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2458 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2460 spin_lock(&o2hb_live_lock);
2461 o2hb_fill_node_map_from_callback(testing_map, O2NM_MAX_NODES);
2462 spin_unlock(&o2hb_live_lock);
2463 if (!test_bit(node_num, testing_map)) {
2465 "node (%u) does not have heartbeating enabled.\n",
2472 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2474 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2476 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2478 o2hb_fill_node_map_from_callback(testing_map, O2NM_MAX_NODES);
2479 if (!test_bit(node_num, testing_map)) {
2481 "node (%u) does not have heartbeating enabled.\n",
2488 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2491 * this is just a hack until we get the plumbing which flips file systems
2492 * read only and drops the hb ref instead of killing the node dead.
2494 void o2hb_stop_all_regions(void)
2496 struct o2hb_region *reg;
2498 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2500 spin_lock(&o2hb_live_lock);
2502 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2503 reg->hr_unclean_stop = 1;
2505 spin_unlock(&o2hb_live_lock);
2507 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2509 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2511 struct o2hb_region *reg;
2515 spin_lock(&o2hb_live_lock);
2518 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2519 if (reg->hr_item_dropped)
2522 mlog(0, "Region: %s\n", config_item_name(®->hr_item));
2523 if (numregs < max_regions) {
2524 memcpy(p, config_item_name(®->hr_item),
2525 O2HB_MAX_REGION_NAME_LEN);
2526 p += O2HB_MAX_REGION_NAME_LEN;
2531 spin_unlock(&o2hb_live_lock);
2535 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2537 int o2hb_global_heartbeat_active(void)
2539 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2541 EXPORT_SYMBOL(o2hb_global_heartbeat_active);