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 char hr_dev_name[BDEVNAME_SIZE];
240 unsigned int hr_timeout_ms;
242 /* randomized as the region goes up and down so that a node
243 * recognizes a node going up and down in one iteration */
246 struct delayed_work hr_write_timeout_work;
247 unsigned long hr_last_timeout_start;
249 /* negotiate timer, used to negotiate extending hb timeout. */
250 struct delayed_work hr_nego_timeout_work;
251 unsigned long hr_nego_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
253 /* Used during o2hb_check_slot to hold a copy of the block
254 * being checked because we temporarily have to zero out the
256 struct o2hb_disk_heartbeat_block *hr_tmp_block;
258 /* Message key for negotiate timeout message. */
260 struct list_head hr_handler_list;
262 /* last hb status, 0 for success, other value for error. */
263 int hr_last_hb_status;
266 struct o2hb_bio_wait_ctxt {
267 atomic_t wc_num_reqs;
268 struct completion wc_io_complete;
272 #define O2HB_NEGO_TIMEOUT_MS (O2HB_MAX_WRITE_TIMEOUT_MS/2)
275 O2HB_NEGO_TIMEOUT_MSG = 1,
276 O2HB_NEGO_APPROVE_MSG = 2,
279 struct o2hb_nego_msg {
283 static void o2hb_write_timeout(struct work_struct *work)
286 struct o2hb_region *reg =
287 container_of(work, struct o2hb_region,
288 hr_write_timeout_work.work);
290 mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
291 "milliseconds\n", reg->hr_dev_name,
292 jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
294 if (o2hb_global_heartbeat_active()) {
295 spin_lock(&o2hb_live_lock);
296 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
297 set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
298 failed = bitmap_weight(o2hb_failed_region_bitmap,
300 quorum = bitmap_weight(o2hb_quorum_region_bitmap,
302 spin_unlock(&o2hb_live_lock);
304 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
308 * Fence if the number of failed regions >= half the number
311 if ((failed << 1) < quorum)
315 o2quo_disk_timeout();
318 static void o2hb_arm_timeout(struct o2hb_region *reg)
320 /* Arm writeout only after thread reaches steady state */
321 if (atomic_read(®->hr_steady_iterations) != 0)
324 mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
325 O2HB_MAX_WRITE_TIMEOUT_MS);
327 if (o2hb_global_heartbeat_active()) {
328 spin_lock(&o2hb_live_lock);
329 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
330 spin_unlock(&o2hb_live_lock);
332 cancel_delayed_work(®->hr_write_timeout_work);
333 schedule_delayed_work(®->hr_write_timeout_work,
334 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
336 cancel_delayed_work(®->hr_nego_timeout_work);
337 /* negotiate timeout must be less than write timeout. */
338 schedule_delayed_work(®->hr_nego_timeout_work,
339 msecs_to_jiffies(O2HB_NEGO_TIMEOUT_MS));
340 memset(reg->hr_nego_node_bitmap, 0, sizeof(reg->hr_nego_node_bitmap));
343 static void o2hb_disarm_timeout(struct o2hb_region *reg)
345 cancel_delayed_work_sync(®->hr_write_timeout_work);
346 cancel_delayed_work_sync(®->hr_nego_timeout_work);
349 static int o2hb_send_nego_msg(int key, int type, u8 target)
351 struct o2hb_nego_msg msg;
354 msg.node_num = o2nm_this_node();
356 ret = o2net_send_message(type, key, &msg, sizeof(msg),
359 if (ret == -EAGAIN || ret == -ENOMEM) {
367 static void o2hb_nego_timeout(struct work_struct *work)
369 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
370 int master_node, i, ret;
371 struct o2hb_region *reg;
373 reg = container_of(work, struct o2hb_region, hr_nego_timeout_work.work);
374 /* don't negotiate timeout if last hb failed since it is very
375 * possible io failed. Should let write timeout fence self.
377 if (reg->hr_last_hb_status)
380 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
381 /* lowest node as master node to make negotiate decision. */
382 master_node = find_next_bit(live_node_bitmap, O2NM_MAX_NODES, 0);
384 if (master_node == o2nm_this_node()) {
385 if (!test_bit(master_node, reg->hr_nego_node_bitmap)) {
386 printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%s).\n",
387 o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000,
388 config_item_name(®->hr_item), reg->hr_dev_name);
389 set_bit(master_node, reg->hr_nego_node_bitmap);
391 if (memcmp(reg->hr_nego_node_bitmap, live_node_bitmap,
392 sizeof(reg->hr_nego_node_bitmap))) {
393 /* check negotiate bitmap every second to do timeout
396 schedule_delayed_work(®->hr_nego_timeout_work,
397 msecs_to_jiffies(1000));
402 printk(KERN_NOTICE "o2hb: all nodes hb write hung, maybe region %s (%s) is down.\n",
403 config_item_name(®->hr_item), reg->hr_dev_name);
404 /* approve negotiate timeout request. */
405 o2hb_arm_timeout(reg);
408 while ((i = find_next_bit(live_node_bitmap,
409 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
410 if (i == master_node)
413 mlog(ML_HEARTBEAT, "send NEGO_APPROVE msg to node %d\n", i);
414 ret = o2hb_send_nego_msg(reg->hr_key,
415 O2HB_NEGO_APPROVE_MSG, i);
417 mlog(ML_ERROR, "send NEGO_APPROVE msg to node %d fail %d\n",
421 /* negotiate timeout with master node. */
422 printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%s), negotiate timeout with node %d.\n",
423 o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000, config_item_name(®->hr_item),
424 reg->hr_dev_name, master_node);
425 ret = o2hb_send_nego_msg(reg->hr_key, O2HB_NEGO_TIMEOUT_MSG,
428 mlog(ML_ERROR, "send NEGO_TIMEOUT msg to node %d fail %d\n",
433 static int o2hb_nego_timeout_handler(struct o2net_msg *msg, u32 len, void *data,
436 struct o2hb_region *reg = data;
437 struct o2hb_nego_msg *nego_msg;
439 nego_msg = (struct o2hb_nego_msg *)msg->buf;
440 printk(KERN_NOTICE "o2hb: receive negotiate timeout message from node %d on region %s (%s).\n",
441 nego_msg->node_num, config_item_name(®->hr_item), reg->hr_dev_name);
442 if (nego_msg->node_num < O2NM_MAX_NODES)
443 set_bit(nego_msg->node_num, reg->hr_nego_node_bitmap);
445 mlog(ML_ERROR, "got nego timeout message from bad node.\n");
450 static int o2hb_nego_approve_handler(struct o2net_msg *msg, u32 len, void *data,
453 struct o2hb_region *reg = data;
455 printk(KERN_NOTICE "o2hb: negotiate timeout approved by master node on region %s (%s).\n",
456 config_item_name(®->hr_item), reg->hr_dev_name);
457 o2hb_arm_timeout(reg);
461 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
463 atomic_set(&wc->wc_num_reqs, 1);
464 init_completion(&wc->wc_io_complete);
468 /* Used in error paths too */
469 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
472 /* sadly atomic_sub_and_test() isn't available on all platforms. The
473 * good news is that the fast path only completes one at a time */
475 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
477 complete(&wc->wc_io_complete);
482 static void o2hb_wait_on_io(struct o2hb_bio_wait_ctxt *wc)
484 o2hb_bio_wait_dec(wc, 1);
485 wait_for_completion(&wc->wc_io_complete);
488 static void o2hb_bio_end_io(struct bio *bio)
490 struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
492 if (bio->bi_status) {
493 mlog(ML_ERROR, "IO Error %d\n", bio->bi_status);
494 wc->wc_error = blk_status_to_errno(bio->bi_status);
497 o2hb_bio_wait_dec(wc, 1);
501 /* Setup a Bio to cover I/O against num_slots slots starting at
503 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
504 struct o2hb_bio_wait_ctxt *wc,
505 unsigned int *current_slot,
506 unsigned int max_slots, int op,
509 int len, current_page;
510 unsigned int vec_len, vec_start;
511 unsigned int bits = reg->hr_block_bits;
512 unsigned int spp = reg->hr_slots_per_page;
513 unsigned int cs = *current_slot;
517 /* Testing has shown this allocation to take long enough under
518 * GFP_KERNEL that the local node can get fenced. It would be
519 * nicest if we could pre-allocate these bios and avoid this
521 bio = bio_alloc(GFP_ATOMIC, 16);
523 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
524 bio = ERR_PTR(-ENOMEM);
528 /* Must put everything in 512 byte sectors for the bio... */
529 bio->bi_iter.bi_sector = (reg->hr_start_block + cs) << (bits - 9);
530 bio_set_dev(bio, reg->hr_bdev);
531 bio->bi_private = wc;
532 bio->bi_end_io = o2hb_bio_end_io;
533 bio_set_op_attrs(bio, op, op_flags);
535 vec_start = (cs << bits) % PAGE_SIZE;
536 while(cs < max_slots) {
537 current_page = cs / spp;
538 page = reg->hr_slot_data[current_page];
540 vec_len = min(PAGE_SIZE - vec_start,
541 (max_slots-cs) * (PAGE_SIZE/spp) );
543 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
544 current_page, vec_len, vec_start);
546 len = bio_add_page(bio, page, vec_len, vec_start);
547 if (len != vec_len) break;
549 cs += vec_len / (PAGE_SIZE/spp);
558 static int o2hb_read_slots(struct o2hb_region *reg,
559 unsigned int begin_slot,
560 unsigned int max_slots)
562 unsigned int current_slot = begin_slot;
564 struct o2hb_bio_wait_ctxt wc;
567 o2hb_bio_wait_init(&wc);
569 while(current_slot < max_slots) {
570 bio = o2hb_setup_one_bio(reg, &wc, ¤t_slot, max_slots,
573 status = PTR_ERR(bio);
578 atomic_inc(&wc.wc_num_reqs);
585 o2hb_wait_on_io(&wc);
586 if (wc.wc_error && !status)
587 status = wc.wc_error;
592 static int o2hb_issue_node_write(struct o2hb_region *reg,
593 struct o2hb_bio_wait_ctxt *write_wc)
599 o2hb_bio_wait_init(write_wc);
601 slot = o2nm_this_node();
603 bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1, REQ_OP_WRITE,
606 status = PTR_ERR(bio);
611 atomic_inc(&write_wc->wc_num_reqs);
619 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
620 struct o2hb_disk_heartbeat_block *hb_block)
625 /* We want to compute the block crc with a 0 value in the
626 * hb_cksum field. Save it off here and replace after the
628 old_cksum = hb_block->hb_cksum;
629 hb_block->hb_cksum = 0;
631 ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
633 hb_block->hb_cksum = old_cksum;
638 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
640 mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
641 "cksum = 0x%x, generation 0x%llx\n",
642 (long long)le64_to_cpu(hb_block->hb_seq),
643 hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
644 (long long)le64_to_cpu(hb_block->hb_generation));
647 static int o2hb_verify_crc(struct o2hb_region *reg,
648 struct o2hb_disk_heartbeat_block *hb_block)
652 read = le32_to_cpu(hb_block->hb_cksum);
653 computed = o2hb_compute_block_crc_le(reg, hb_block);
655 return read == computed;
659 * Compare the slot data with what we wrote in the last iteration.
660 * If the match fails, print an appropriate error message. This is to
661 * detect errors like... another node hearting on the same slot,
662 * flaky device that is losing writes, etc.
663 * Returns 1 if check succeeds, 0 otherwise.
665 static int o2hb_check_own_slot(struct o2hb_region *reg)
667 struct o2hb_disk_slot *slot;
668 struct o2hb_disk_heartbeat_block *hb_block;
671 slot = ®->hr_slots[o2nm_this_node()];
672 /* Don't check on our 1st timestamp */
673 if (!slot->ds_last_time)
676 hb_block = slot->ds_raw_block;
677 if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
678 le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
679 hb_block->hb_node == slot->ds_node_num)
682 #define ERRSTR1 "Another node is heartbeating on device"
683 #define ERRSTR2 "Heartbeat generation mismatch on device"
684 #define ERRSTR3 "Heartbeat sequence mismatch on device"
686 if (hb_block->hb_node != slot->ds_node_num)
688 else if (le64_to_cpu(hb_block->hb_generation) !=
689 slot->ds_last_generation)
694 mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
695 "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
696 slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
697 (unsigned long long)slot->ds_last_time, hb_block->hb_node,
698 (unsigned long long)le64_to_cpu(hb_block->hb_generation),
699 (unsigned long long)le64_to_cpu(hb_block->hb_seq));
704 static inline void o2hb_prepare_block(struct o2hb_region *reg,
709 struct o2hb_disk_slot *slot;
710 struct o2hb_disk_heartbeat_block *hb_block;
712 node_num = o2nm_this_node();
713 slot = ®->hr_slots[node_num];
715 hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
716 memset(hb_block, 0, reg->hr_block_bytes);
717 /* TODO: time stuff */
718 cputime = ktime_get_real_seconds();
722 hb_block->hb_seq = cpu_to_le64(cputime);
723 hb_block->hb_node = node_num;
724 hb_block->hb_generation = cpu_to_le64(generation);
725 hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
727 /* This step must always happen last! */
728 hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
731 mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
732 (long long)generation,
733 le32_to_cpu(hb_block->hb_cksum));
736 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
737 struct o2nm_node *node,
740 struct o2hb_callback_func *f;
742 list_for_each_entry(f, &hbcall->list, hc_item) {
743 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
744 (f->hc_func)(node, idx, f->hc_data);
748 /* Will run the list in order until we process the passed event */
749 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
751 struct o2hb_callback *hbcall;
752 struct o2hb_node_event *event;
754 /* Holding callback sem assures we don't alter the callback
755 * lists when doing this, and serializes ourselves with other
756 * processes wanting callbacks. */
757 down_write(&o2hb_callback_sem);
759 spin_lock(&o2hb_live_lock);
760 while (!list_empty(&o2hb_node_events)
761 && !list_empty(&queued_event->hn_item)) {
762 event = list_entry(o2hb_node_events.next,
763 struct o2hb_node_event,
765 list_del_init(&event->hn_item);
766 spin_unlock(&o2hb_live_lock);
768 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
769 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
772 hbcall = hbcall_from_type(event->hn_event_type);
774 /* We should *never* have gotten on to the list with a
775 * bad type... This isn't something that we should try
776 * to recover from. */
777 BUG_ON(IS_ERR(hbcall));
779 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
781 spin_lock(&o2hb_live_lock);
783 spin_unlock(&o2hb_live_lock);
785 up_write(&o2hb_callback_sem);
788 static void o2hb_queue_node_event(struct o2hb_node_event *event,
789 enum o2hb_callback_type type,
790 struct o2nm_node *node,
793 assert_spin_locked(&o2hb_live_lock);
795 BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
797 event->hn_event_type = type;
798 event->hn_node = node;
799 event->hn_node_num = node_num;
801 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
802 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
804 list_add_tail(&event->hn_item, &o2hb_node_events);
807 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
809 struct o2hb_node_event event =
810 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
811 struct o2nm_node *node;
814 node = o2nm_get_node_by_num(slot->ds_node_num);
818 spin_lock(&o2hb_live_lock);
819 if (!list_empty(&slot->ds_live_item)) {
820 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
823 list_del_init(&slot->ds_live_item);
825 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
826 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
828 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
833 spin_unlock(&o2hb_live_lock);
836 o2hb_run_event_list(&event);
841 static void o2hb_set_quorum_device(struct o2hb_region *reg)
843 if (!o2hb_global_heartbeat_active())
846 /* Prevent race with o2hb_heartbeat_group_drop_item() */
847 if (kthread_should_stop())
850 /* Tag region as quorum only after thread reaches steady state */
851 if (atomic_read(®->hr_steady_iterations) != 0)
854 spin_lock(&o2hb_live_lock);
856 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
860 * A region can be added to the quorum only when it sees all
861 * live nodes heartbeat on it. In other words, the region has been
862 * added to all nodes.
864 if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
865 sizeof(o2hb_live_node_bitmap)))
868 printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
869 config_item_name(®->hr_item), reg->hr_dev_name);
871 set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
874 * If global heartbeat active, unpin all regions if the
875 * region count > CUT_OFF
877 if (bitmap_weight(o2hb_quorum_region_bitmap,
878 O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
879 o2hb_region_unpin(NULL);
881 spin_unlock(&o2hb_live_lock);
884 static int o2hb_check_slot(struct o2hb_region *reg,
885 struct o2hb_disk_slot *slot)
887 int changed = 0, gen_changed = 0;
888 struct o2hb_node_event event =
889 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
890 struct o2nm_node *node;
891 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
893 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
894 unsigned int slot_dead_ms;
898 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
901 * If a node is no longer configured but is still in the livemap, we
902 * may need to clear that bit from the livemap.
904 node = o2nm_get_node_by_num(slot->ds_node_num);
906 spin_lock(&o2hb_live_lock);
907 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
908 spin_unlock(&o2hb_live_lock);
913 if (!o2hb_verify_crc(reg, hb_block)) {
914 /* all paths from here will drop o2hb_live_lock for
916 spin_lock(&o2hb_live_lock);
918 /* Don't print an error on the console in this case -
919 * a freshly formatted heartbeat area will not have a
921 if (list_empty(&slot->ds_live_item))
924 /* The node is live but pushed out a bad crc. We
925 * consider it a transient miss but don't populate any
926 * other values as they may be junk. */
927 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
928 slot->ds_node_num, reg->hr_dev_name);
929 o2hb_dump_slot(hb_block);
931 slot->ds_equal_samples++;
935 /* we don't care if these wrap.. the state transitions below
936 * clear at the right places */
937 cputime = le64_to_cpu(hb_block->hb_seq);
938 if (slot->ds_last_time != cputime)
939 slot->ds_changed_samples++;
941 slot->ds_equal_samples++;
942 slot->ds_last_time = cputime;
944 /* The node changed heartbeat generations. We assume this to
945 * mean it dropped off but came back before we timed out. We
946 * want to consider it down for the time being but don't want
947 * to lose any changed_samples state we might build up to
948 * considering it live again. */
949 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
951 slot->ds_equal_samples = 0;
952 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
953 "to 0x%llx)\n", slot->ds_node_num,
954 (long long)slot->ds_last_generation,
955 (long long)le64_to_cpu(hb_block->hb_generation));
958 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
960 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
961 "seq %llu last %llu changed %u equal %u\n",
962 slot->ds_node_num, (long long)slot->ds_last_generation,
963 le32_to_cpu(hb_block->hb_cksum),
964 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
965 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
966 slot->ds_equal_samples);
968 spin_lock(&o2hb_live_lock);
971 /* dead nodes only come to life after some number of
972 * changes at any time during their dead time */
973 if (list_empty(&slot->ds_live_item) &&
974 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
975 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
976 slot->ds_node_num, (long long)slot->ds_last_generation);
978 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
980 /* first on the list generates a callback */
981 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
982 mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
983 "bitmap\n", slot->ds_node_num);
984 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
986 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
993 list_add_tail(&slot->ds_live_item,
994 &o2hb_live_slots[slot->ds_node_num]);
996 slot->ds_equal_samples = 0;
998 /* We want to be sure that all nodes agree on the
999 * number of milliseconds before a node will be
1000 * considered dead. The self-fencing timeout is
1001 * computed from this value, and a discrepancy might
1002 * result in heartbeat calling a node dead when it
1003 * hasn't self-fenced yet. */
1004 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
1005 if (slot_dead_ms && slot_dead_ms != dead_ms) {
1006 /* TODO: Perhaps we can fail the region here. */
1007 mlog(ML_ERROR, "Node %d on device %s has a dead count "
1008 "of %u ms, but our count is %u ms.\n"
1009 "Please double check your configuration values "
1010 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
1011 slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
1017 /* if the list is dead, we're done.. */
1018 if (list_empty(&slot->ds_live_item))
1021 /* live nodes only go dead after enough consequtive missed
1022 * samples.. reset the missed counter whenever we see
1024 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
1025 mlog(ML_HEARTBEAT, "Node %d left my region\n",
1028 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1030 /* last off the live_slot generates a callback */
1031 list_del_init(&slot->ds_live_item);
1032 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
1033 mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
1034 "nodes bitmap\n", slot->ds_node_num);
1035 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
1037 /* node can be null */
1038 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
1039 node, slot->ds_node_num);
1045 /* We don't clear this because the node is still
1046 * actually writing new blocks. */
1048 slot->ds_changed_samples = 0;
1051 if (slot->ds_changed_samples) {
1052 slot->ds_changed_samples = 0;
1053 slot->ds_equal_samples = 0;
1056 spin_unlock(&o2hb_live_lock);
1059 o2hb_run_event_list(&event);
1062 o2nm_node_put(node);
1066 static int o2hb_highest_node(unsigned long *nodes, int numbits)
1068 return find_last_bit(nodes, numbits);
1071 static int o2hb_lowest_node(unsigned long *nodes, int numbits)
1073 return find_first_bit(nodes, numbits);
1076 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
1078 int i, ret, highest_node, lowest_node;
1079 int membership_change = 0, own_slot_ok = 0;
1080 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
1081 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
1082 struct o2hb_bio_wait_ctxt write_wc;
1084 ret = o2nm_configured_node_map(configured_nodes,
1085 sizeof(configured_nodes));
1092 * If a node is not configured but is in the livemap, we still need
1093 * to read the slot so as to be able to remove it from the livemap.
1095 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
1097 while ((i = find_next_bit(live_node_bitmap,
1098 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1099 set_bit(i, configured_nodes);
1102 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1103 lowest_node = o2hb_lowest_node(configured_nodes, O2NM_MAX_NODES);
1104 if (highest_node >= O2NM_MAX_NODES || lowest_node >= O2NM_MAX_NODES) {
1105 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1110 /* No sense in reading the slots of nodes that don't exist
1111 * yet. Of course, if the node definitions have holes in them
1112 * then we're reading an empty slot anyway... Consider this
1114 ret = o2hb_read_slots(reg, lowest_node, highest_node + 1);
1120 /* With an up to date view of the slots, we can check that no
1121 * other node has been improperly configured to heartbeat in
1123 own_slot_ok = o2hb_check_own_slot(reg);
1125 /* fill in the proper info for our next heartbeat */
1126 o2hb_prepare_block(reg, reg->hr_generation);
1128 ret = o2hb_issue_node_write(reg, &write_wc);
1135 while((i = find_next_bit(configured_nodes,
1136 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1137 membership_change |= o2hb_check_slot(reg, ®->hr_slots[i]);
1141 * We have to be sure we've advertised ourselves on disk
1142 * before we can go to steady state. This ensures that
1143 * people we find in our steady state have seen us.
1145 o2hb_wait_on_io(&write_wc);
1146 if (write_wc.wc_error) {
1147 /* Do not re-arm the write timeout on I/O error - we
1148 * can't be sure that the new block ever made it to
1150 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1151 write_wc.wc_error, reg->hr_dev_name);
1152 ret = write_wc.wc_error;
1156 /* Skip disarming the timeout if own slot has stale/bad data */
1158 o2hb_set_quorum_device(reg);
1159 o2hb_arm_timeout(reg);
1160 reg->hr_last_timeout_start = jiffies;
1164 /* let the person who launched us know when things are steady */
1165 if (atomic_read(®->hr_steady_iterations) != 0) {
1166 if (!ret && own_slot_ok && !membership_change) {
1167 if (atomic_dec_and_test(®->hr_steady_iterations))
1168 wake_up(&o2hb_steady_queue);
1172 if (atomic_read(®->hr_steady_iterations) != 0) {
1173 if (atomic_dec_and_test(®->hr_unsteady_iterations)) {
1174 printk(KERN_NOTICE "o2hb: Unable to stabilize "
1175 "heartbeat on region %s (%s)\n",
1176 config_item_name(®->hr_item),
1178 atomic_set(®->hr_steady_iterations, 0);
1179 reg->hr_aborted_start = 1;
1180 wake_up(&o2hb_steady_queue);
1189 * we ride the region ref that the region dir holds. before the region
1190 * dir is removed and drops it ref it will wait to tear down this
1193 static int o2hb_thread(void *data)
1196 struct o2hb_region *reg = data;
1197 struct o2hb_bio_wait_ctxt write_wc;
1198 ktime_t before_hb, after_hb;
1199 unsigned int elapsed_msec;
1201 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1203 set_user_nice(current, MIN_NICE);
1206 ret = o2nm_depend_this_node();
1208 mlog(ML_ERROR, "Node has been deleted, ret = %d\n", ret);
1209 reg->hr_node_deleted = 1;
1210 wake_up(&o2hb_steady_queue);
1214 while (!kthread_should_stop() &&
1215 !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1216 /* We track the time spent inside
1217 * o2hb_do_disk_heartbeat so that we avoid more than
1218 * hr_timeout_ms between disk writes. On busy systems
1219 * this should result in a heartbeat which is less
1220 * likely to time itself out. */
1221 before_hb = ktime_get_real();
1223 ret = o2hb_do_disk_heartbeat(reg);
1224 reg->hr_last_hb_status = ret;
1226 after_hb = ktime_get_real();
1228 elapsed_msec = (unsigned int)
1229 ktime_ms_delta(after_hb, before_hb);
1232 "start = %lld, end = %lld, msec = %u, ret = %d\n",
1233 before_hb, after_hb, elapsed_msec, ret);
1235 if (!kthread_should_stop() &&
1236 elapsed_msec < reg->hr_timeout_ms) {
1237 /* the kthread api has blocked signals for us so no
1238 * need to record the return value. */
1239 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1243 o2hb_disarm_timeout(reg);
1245 /* unclean stop is only used in very bad situation */
1246 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1247 o2hb_shutdown_slot(®->hr_slots[i]);
1249 /* Explicit down notification - avoid forcing the other nodes
1250 * to timeout on this region when we could just as easily
1251 * write a clear generation - thus indicating to them that
1252 * this node has left this region.
1254 if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1255 o2hb_prepare_block(reg, 0);
1256 ret = o2hb_issue_node_write(reg, &write_wc);
1258 o2hb_wait_on_io(&write_wc);
1264 o2nm_undepend_this_node();
1266 mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1271 #ifdef CONFIG_DEBUG_FS
1272 static int o2hb_debug_open(struct inode *inode, struct file *file)
1274 struct o2hb_debug_buf *db = inode->i_private;
1275 struct o2hb_region *reg;
1276 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1282 /* max_nodes should be the largest bitmap we pass here */
1283 BUG_ON(sizeof(map) < db->db_size);
1285 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1289 switch (db->db_type) {
1290 case O2HB_DB_TYPE_LIVENODES:
1291 case O2HB_DB_TYPE_LIVEREGIONS:
1292 case O2HB_DB_TYPE_QUORUMREGIONS:
1293 case O2HB_DB_TYPE_FAILEDREGIONS:
1294 spin_lock(&o2hb_live_lock);
1295 memcpy(map, db->db_data, db->db_size);
1296 spin_unlock(&o2hb_live_lock);
1299 case O2HB_DB_TYPE_REGION_LIVENODES:
1300 spin_lock(&o2hb_live_lock);
1301 reg = (struct o2hb_region *)db->db_data;
1302 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1303 spin_unlock(&o2hb_live_lock);
1306 case O2HB_DB_TYPE_REGION_NUMBER:
1307 reg = (struct o2hb_region *)db->db_data;
1308 out += scnprintf(buf + out, PAGE_SIZE - out, "%d\n",
1309 reg->hr_region_num);
1312 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1313 reg = (struct o2hb_region *)db->db_data;
1314 lts = reg->hr_last_timeout_start;
1315 /* If 0, it has never been set before */
1317 lts = jiffies_to_msecs(jiffies - lts);
1318 out += scnprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1321 case O2HB_DB_TYPE_REGION_PINNED:
1322 reg = (struct o2hb_region *)db->db_data;
1323 out += scnprintf(buf + out, PAGE_SIZE - out, "%u\n",
1324 !!reg->hr_item_pinned);
1331 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1332 out += scnprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1333 out += scnprintf(buf + out, PAGE_SIZE - out, "\n");
1336 i_size_write(inode, out);
1338 file->private_data = buf;
1345 static int o2hb_debug_release(struct inode *inode, struct file *file)
1347 kfree(file->private_data);
1351 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1352 size_t nbytes, loff_t *ppos)
1354 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1355 i_size_read(file->f_mapping->host));
1358 static int o2hb_debug_open(struct inode *inode, struct file *file)
1362 static int o2hb_debug_release(struct inode *inode, struct file *file)
1366 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1367 size_t nbytes, loff_t *ppos)
1371 #endif /* CONFIG_DEBUG_FS */
1373 static const struct file_operations o2hb_debug_fops = {
1374 .open = o2hb_debug_open,
1375 .release = o2hb_debug_release,
1376 .read = o2hb_debug_read,
1377 .llseek = generic_file_llseek,
1380 void o2hb_exit(void)
1382 debugfs_remove_recursive(o2hb_debug_dir);
1383 kfree(o2hb_db_livenodes);
1384 kfree(o2hb_db_liveregions);
1385 kfree(o2hb_db_quorumregions);
1386 kfree(o2hb_db_failedregions);
1389 static void o2hb_debug_create(const char *name, struct dentry *dir,
1390 struct o2hb_debug_buf **db, int db_len, int type,
1391 int size, int len, void *data)
1393 *db = kmalloc(db_len, GFP_KERNEL);
1397 (*db)->db_type = type;
1398 (*db)->db_size = size;
1399 (*db)->db_len = len;
1400 (*db)->db_data = data;
1402 debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db, &o2hb_debug_fops);
1405 static void o2hb_debug_init(void)
1407 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1409 o2hb_debug_create(O2HB_DEBUG_LIVENODES, o2hb_debug_dir,
1410 &o2hb_db_livenodes, sizeof(*o2hb_db_livenodes),
1411 O2HB_DB_TYPE_LIVENODES, sizeof(o2hb_live_node_bitmap),
1412 O2NM_MAX_NODES, o2hb_live_node_bitmap);
1414 o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS, o2hb_debug_dir,
1415 &o2hb_db_liveregions, sizeof(*o2hb_db_liveregions),
1416 O2HB_DB_TYPE_LIVEREGIONS,
1417 sizeof(o2hb_live_region_bitmap), O2NM_MAX_REGIONS,
1418 o2hb_live_region_bitmap);
1420 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS, o2hb_debug_dir,
1421 &o2hb_db_quorumregions,
1422 sizeof(*o2hb_db_quorumregions),
1423 O2HB_DB_TYPE_QUORUMREGIONS,
1424 sizeof(o2hb_quorum_region_bitmap), O2NM_MAX_REGIONS,
1425 o2hb_quorum_region_bitmap);
1427 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS, o2hb_debug_dir,
1428 &o2hb_db_failedregions,
1429 sizeof(*o2hb_db_failedregions),
1430 O2HB_DB_TYPE_FAILEDREGIONS,
1431 sizeof(o2hb_failed_region_bitmap), O2NM_MAX_REGIONS,
1432 o2hb_failed_region_bitmap);
1435 void o2hb_init(void)
1439 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1440 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1442 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1443 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1445 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1446 memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1447 memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1448 memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1449 memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1451 o2hb_dependent_users = 0;
1456 /* if we're already in a callback then we're already serialized by the sem */
1457 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1460 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1462 memcpy(map, &o2hb_live_node_bitmap, bytes);
1466 * get a map of all nodes that are heartbeating in any regions
1468 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1470 /* callers want to serialize this map and callbacks so that they
1471 * can trust that they don't miss nodes coming to the party */
1472 down_read(&o2hb_callback_sem);
1473 spin_lock(&o2hb_live_lock);
1474 o2hb_fill_node_map_from_callback(map, bytes);
1475 spin_unlock(&o2hb_live_lock);
1476 up_read(&o2hb_callback_sem);
1478 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1481 * heartbeat configfs bits. The heartbeat set is a default set under
1482 * the cluster set in nodemanager.c.
1485 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1487 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1490 /* drop_item only drops its ref after killing the thread, nothing should
1491 * be using the region anymore. this has to clean up any state that
1492 * attributes might have built up. */
1493 static void o2hb_region_release(struct config_item *item)
1497 struct o2hb_region *reg = to_o2hb_region(item);
1499 mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1501 kfree(reg->hr_tmp_block);
1503 if (reg->hr_slot_data) {
1504 for (i = 0; i < reg->hr_num_pages; i++) {
1505 page = reg->hr_slot_data[i];
1509 kfree(reg->hr_slot_data);
1513 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1515 kfree(reg->hr_slots);
1517 debugfs_remove_recursive(reg->hr_debug_dir);
1518 kfree(reg->hr_db_livenodes);
1519 kfree(reg->hr_db_regnum);
1520 kfree(reg->hr_db_elapsed_time);
1521 kfree(reg->hr_db_pinned);
1523 spin_lock(&o2hb_live_lock);
1524 list_del(®->hr_all_item);
1525 spin_unlock(&o2hb_live_lock);
1527 o2net_unregister_handler_list(®->hr_handler_list);
1531 static int o2hb_read_block_input(struct o2hb_region *reg,
1533 unsigned long *ret_bytes,
1534 unsigned int *ret_bits)
1536 unsigned long bytes;
1537 char *p = (char *)page;
1539 bytes = simple_strtoul(p, &p, 0);
1540 if (!p || (*p && (*p != '\n')))
1543 /* Heartbeat and fs min / max block sizes are the same. */
1544 if (bytes > 4096 || bytes < 512)
1546 if (hweight16(bytes) != 1)
1552 *ret_bits = ffs(bytes) - 1;
1557 static ssize_t o2hb_region_block_bytes_show(struct config_item *item,
1560 return sprintf(page, "%u\n", to_o2hb_region(item)->hr_block_bytes);
1563 static ssize_t o2hb_region_block_bytes_store(struct config_item *item,
1567 struct o2hb_region *reg = to_o2hb_region(item);
1569 unsigned long block_bytes;
1570 unsigned int block_bits;
1575 status = o2hb_read_block_input(reg, page, &block_bytes,
1580 reg->hr_block_bytes = (unsigned int)block_bytes;
1581 reg->hr_block_bits = block_bits;
1586 static ssize_t o2hb_region_start_block_show(struct config_item *item,
1589 return sprintf(page, "%llu\n", to_o2hb_region(item)->hr_start_block);
1592 static ssize_t o2hb_region_start_block_store(struct config_item *item,
1596 struct o2hb_region *reg = to_o2hb_region(item);
1597 unsigned long long tmp;
1598 char *p = (char *)page;
1603 tmp = simple_strtoull(p, &p, 0);
1604 if (!p || (*p && (*p != '\n')))
1607 reg->hr_start_block = tmp;
1612 static ssize_t o2hb_region_blocks_show(struct config_item *item, char *page)
1614 return sprintf(page, "%d\n", to_o2hb_region(item)->hr_blocks);
1617 static ssize_t o2hb_region_blocks_store(struct config_item *item,
1621 struct o2hb_region *reg = to_o2hb_region(item);
1623 char *p = (char *)page;
1628 tmp = simple_strtoul(p, &p, 0);
1629 if (!p || (*p && (*p != '\n')))
1632 if (tmp > O2NM_MAX_NODES || tmp == 0)
1635 reg->hr_blocks = (unsigned int)tmp;
1640 static ssize_t o2hb_region_dev_show(struct config_item *item, char *page)
1642 unsigned int ret = 0;
1644 if (to_o2hb_region(item)->hr_bdev)
1645 ret = sprintf(page, "%s\n", to_o2hb_region(item)->hr_dev_name);
1650 static void o2hb_init_region_params(struct o2hb_region *reg)
1652 reg->hr_slots_per_page = PAGE_SIZE >> reg->hr_block_bits;
1653 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1655 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1656 reg->hr_start_block, reg->hr_blocks);
1657 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1658 reg->hr_block_bytes, reg->hr_block_bits);
1659 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1660 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1663 static int o2hb_map_slot_data(struct o2hb_region *reg)
1666 unsigned int last_slot;
1667 unsigned int spp = reg->hr_slots_per_page;
1670 struct o2hb_disk_slot *slot;
1672 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1673 if (reg->hr_tmp_block == NULL)
1676 reg->hr_slots = kcalloc(reg->hr_blocks,
1677 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1678 if (reg->hr_slots == NULL)
1681 for(i = 0; i < reg->hr_blocks; i++) {
1682 slot = ®->hr_slots[i];
1683 slot->ds_node_num = i;
1684 INIT_LIST_HEAD(&slot->ds_live_item);
1685 slot->ds_raw_block = NULL;
1688 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1689 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1690 "at %u blocks per page\n",
1691 reg->hr_num_pages, reg->hr_blocks, spp);
1693 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1695 if (!reg->hr_slot_data)
1698 for(i = 0; i < reg->hr_num_pages; i++) {
1699 page = alloc_page(GFP_KERNEL);
1703 reg->hr_slot_data[i] = page;
1705 last_slot = i * spp;
1706 raw = page_address(page);
1708 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1710 BUG_ON((j + last_slot) >= reg->hr_blocks);
1712 slot = ®->hr_slots[j + last_slot];
1713 slot->ds_raw_block =
1714 (struct o2hb_disk_heartbeat_block *) raw;
1716 raw += reg->hr_block_bytes;
1723 /* Read in all the slots available and populate the tracking
1724 * structures so that we can start with a baseline idea of what's
1726 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1729 struct o2hb_disk_slot *slot;
1730 struct o2hb_disk_heartbeat_block *hb_block;
1732 ret = o2hb_read_slots(reg, 0, reg->hr_blocks);
1736 /* We only want to get an idea of the values initially in each
1737 * slot, so we do no verification - o2hb_check_slot will
1738 * actually determine if each configured slot is valid and
1739 * whether any values have changed. */
1740 for(i = 0; i < reg->hr_blocks; i++) {
1741 slot = ®->hr_slots[i];
1742 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1744 /* Only fill the values that o2hb_check_slot uses to
1745 * determine changing slots */
1746 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1747 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1754 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1755 static ssize_t o2hb_region_dev_store(struct config_item *item,
1759 struct o2hb_region *reg = to_o2hb_region(item);
1760 struct task_struct *hb_task;
1763 char *p = (char *)page;
1765 ssize_t ret = -EINVAL;
1771 /* We can't heartbeat without having had our node number
1772 * configured yet. */
1773 if (o2nm_this_node() == O2NM_MAX_NODES)
1776 fd = simple_strtol(p, &p, 0);
1777 if (!p || (*p && (*p != '\n')))
1780 if (fd < 0 || fd >= INT_MAX)
1787 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1788 reg->hr_block_bytes == 0)
1791 if (!S_ISBLK(f.file->f_mapping->host->i_mode))
1794 reg->hr_bdev = blkdev_get_by_dev(f.file->f_mapping->host->i_rdev,
1795 FMODE_WRITE | FMODE_READ, NULL);
1796 if (IS_ERR(reg->hr_bdev)) {
1797 ret = PTR_ERR(reg->hr_bdev);
1798 reg->hr_bdev = NULL;
1802 bdevname(reg->hr_bdev, reg->hr_dev_name);
1804 sectsize = bdev_logical_block_size(reg->hr_bdev);
1805 if (sectsize != reg->hr_block_bytes) {
1807 "blocksize %u incorrect for device, expected %d",
1808 reg->hr_block_bytes, sectsize);
1813 o2hb_init_region_params(reg);
1815 /* Generation of zero is invalid */
1817 get_random_bytes(®->hr_generation,
1818 sizeof(reg->hr_generation));
1819 } while (reg->hr_generation == 0);
1821 ret = o2hb_map_slot_data(reg);
1827 ret = o2hb_populate_slot_data(reg);
1833 INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout);
1834 INIT_DELAYED_WORK(®->hr_nego_timeout_work, o2hb_nego_timeout);
1837 * A node is considered live after it has beat LIVE_THRESHOLD
1838 * times. We're not steady until we've given them a chance
1839 * _after_ our first read.
1840 * The default threshold is bare minimum so as to limit the delay
1841 * during mounts. For global heartbeat, the threshold doubled for the
1844 live_threshold = O2HB_LIVE_THRESHOLD;
1845 if (o2hb_global_heartbeat_active()) {
1846 spin_lock(&o2hb_live_lock);
1847 if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1848 live_threshold <<= 1;
1849 spin_unlock(&o2hb_live_lock);
1852 atomic_set(®->hr_steady_iterations, live_threshold);
1853 /* unsteady_iterations is triple the steady_iterations */
1854 atomic_set(®->hr_unsteady_iterations, (live_threshold * 3));
1856 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1857 reg->hr_item.ci_name);
1858 if (IS_ERR(hb_task)) {
1859 ret = PTR_ERR(hb_task);
1864 spin_lock(&o2hb_live_lock);
1865 reg->hr_task = hb_task;
1866 spin_unlock(&o2hb_live_lock);
1868 ret = wait_event_interruptible(o2hb_steady_queue,
1869 atomic_read(®->hr_steady_iterations) == 0 ||
1870 reg->hr_node_deleted);
1872 atomic_set(®->hr_steady_iterations, 0);
1873 reg->hr_aborted_start = 1;
1876 if (reg->hr_aborted_start) {
1881 if (reg->hr_node_deleted) {
1886 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1887 spin_lock(&o2hb_live_lock);
1888 hb_task = reg->hr_task;
1889 if (o2hb_global_heartbeat_active())
1890 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1891 spin_unlock(&o2hb_live_lock);
1898 if (hb_task && o2hb_global_heartbeat_active())
1899 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1900 config_item_name(®->hr_item), reg->hr_dev_name);
1904 blkdev_put(reg->hr_bdev, FMODE_READ | FMODE_WRITE);
1905 reg->hr_bdev = NULL;
1913 static ssize_t o2hb_region_pid_show(struct config_item *item, char *page)
1915 struct o2hb_region *reg = to_o2hb_region(item);
1918 spin_lock(&o2hb_live_lock);
1920 pid = task_pid_nr(reg->hr_task);
1921 spin_unlock(&o2hb_live_lock);
1926 return sprintf(page, "%u\n", pid);
1929 CONFIGFS_ATTR(o2hb_region_, block_bytes);
1930 CONFIGFS_ATTR(o2hb_region_, start_block);
1931 CONFIGFS_ATTR(o2hb_region_, blocks);
1932 CONFIGFS_ATTR(o2hb_region_, dev);
1933 CONFIGFS_ATTR_RO(o2hb_region_, pid);
1935 static struct configfs_attribute *o2hb_region_attrs[] = {
1936 &o2hb_region_attr_block_bytes,
1937 &o2hb_region_attr_start_block,
1938 &o2hb_region_attr_blocks,
1939 &o2hb_region_attr_dev,
1940 &o2hb_region_attr_pid,
1944 static struct configfs_item_operations o2hb_region_item_ops = {
1945 .release = o2hb_region_release,
1948 static const struct config_item_type o2hb_region_type = {
1949 .ct_item_ops = &o2hb_region_item_ops,
1950 .ct_attrs = o2hb_region_attrs,
1951 .ct_owner = THIS_MODULE,
1956 struct o2hb_heartbeat_group {
1957 struct config_group hs_group;
1961 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1964 container_of(group, struct o2hb_heartbeat_group, hs_group)
1968 static void o2hb_debug_region_init(struct o2hb_region *reg,
1969 struct dentry *parent)
1973 dir = debugfs_create_dir(config_item_name(®->hr_item), parent);
1974 reg->hr_debug_dir = dir;
1976 o2hb_debug_create(O2HB_DEBUG_LIVENODES, dir, &(reg->hr_db_livenodes),
1977 sizeof(*(reg->hr_db_livenodes)),
1978 O2HB_DB_TYPE_REGION_LIVENODES,
1979 sizeof(reg->hr_live_node_bitmap), O2NM_MAX_NODES,
1982 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER, dir, &(reg->hr_db_regnum),
1983 sizeof(*(reg->hr_db_regnum)),
1984 O2HB_DB_TYPE_REGION_NUMBER, 0, O2NM_MAX_NODES, reg);
1986 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME, dir,
1987 &(reg->hr_db_elapsed_time),
1988 sizeof(*(reg->hr_db_elapsed_time)),
1989 O2HB_DB_TYPE_REGION_ELAPSED_TIME, 0, 0, reg);
1991 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED, dir, &(reg->hr_db_pinned),
1992 sizeof(*(reg->hr_db_pinned)),
1993 O2HB_DB_TYPE_REGION_PINNED, 0, 0, reg);
1997 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2000 struct o2hb_region *reg = NULL;
2003 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2005 return ERR_PTR(-ENOMEM);
2007 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2008 ret = -ENAMETOOLONG;
2012 spin_lock(&o2hb_live_lock);
2013 reg->hr_region_num = 0;
2014 if (o2hb_global_heartbeat_active()) {
2015 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2017 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2018 spin_unlock(&o2hb_live_lock);
2022 set_bit(reg->hr_region_num, o2hb_region_bitmap);
2024 list_add_tail(®->hr_all_item, &o2hb_all_regions);
2025 spin_unlock(&o2hb_live_lock);
2027 config_item_init_type_name(®->hr_item, name, &o2hb_region_type);
2029 /* this is the same way to generate msg key as dlm, for local heartbeat,
2030 * name is also the same, so make initial crc value different to avoid
2031 * message key conflict.
2033 reg->hr_key = crc32_le(reg->hr_region_num + O2NM_MAX_REGIONS,
2034 name, strlen(name));
2035 INIT_LIST_HEAD(®->hr_handler_list);
2036 ret = o2net_register_handler(O2HB_NEGO_TIMEOUT_MSG, reg->hr_key,
2037 sizeof(struct o2hb_nego_msg),
2038 o2hb_nego_timeout_handler,
2039 reg, NULL, ®->hr_handler_list);
2043 ret = o2net_register_handler(O2HB_NEGO_APPROVE_MSG, reg->hr_key,
2044 sizeof(struct o2hb_nego_msg),
2045 o2hb_nego_approve_handler,
2046 reg, NULL, ®->hr_handler_list);
2048 goto unregister_handler;
2050 o2hb_debug_region_init(reg, o2hb_debug_dir);
2052 return ®->hr_item;
2055 o2net_unregister_handler_list(®->hr_handler_list);
2057 spin_lock(&o2hb_live_lock);
2058 list_del(®->hr_all_item);
2059 if (o2hb_global_heartbeat_active())
2060 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2061 spin_unlock(&o2hb_live_lock);
2064 return ERR_PTR(ret);
2067 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2068 struct config_item *item)
2070 struct task_struct *hb_task;
2071 struct o2hb_region *reg = to_o2hb_region(item);
2072 int quorum_region = 0;
2074 /* stop the thread when the user removes the region dir */
2075 spin_lock(&o2hb_live_lock);
2076 hb_task = reg->hr_task;
2077 reg->hr_task = NULL;
2078 reg->hr_item_dropped = 1;
2079 spin_unlock(&o2hb_live_lock);
2082 kthread_stop(hb_task);
2084 if (o2hb_global_heartbeat_active()) {
2085 spin_lock(&o2hb_live_lock);
2086 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2087 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2088 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2090 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2091 spin_unlock(&o2hb_live_lock);
2092 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2093 ((atomic_read(®->hr_steady_iterations) == 0) ?
2094 "stopped" : "start aborted"), config_item_name(item),
2099 * If we're racing a dev_write(), we need to wake them. They will
2100 * check reg->hr_task
2102 if (atomic_read(®->hr_steady_iterations) != 0) {
2103 reg->hr_aborted_start = 1;
2104 atomic_set(®->hr_steady_iterations, 0);
2105 wake_up(&o2hb_steady_queue);
2108 config_item_put(item);
2110 if (!o2hb_global_heartbeat_active() || !quorum_region)
2114 * If global heartbeat active and there are dependent users,
2115 * pin all regions if quorum region count <= CUT_OFF
2117 spin_lock(&o2hb_live_lock);
2119 if (!o2hb_dependent_users)
2122 if (bitmap_weight(o2hb_quorum_region_bitmap,
2123 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2124 o2hb_region_pin(NULL);
2127 spin_unlock(&o2hb_live_lock);
2130 static ssize_t o2hb_heartbeat_group_dead_threshold_show(struct config_item *item,
2133 return sprintf(page, "%u\n", o2hb_dead_threshold);
2136 static ssize_t o2hb_heartbeat_group_dead_threshold_store(struct config_item *item,
2137 const char *page, size_t count)
2140 char *p = (char *)page;
2142 tmp = simple_strtoul(p, &p, 10);
2143 if (!p || (*p && (*p != '\n')))
2146 /* this will validate ranges for us. */
2147 o2hb_dead_threshold_set((unsigned int) tmp);
2152 static ssize_t o2hb_heartbeat_group_mode_show(struct config_item *item,
2155 return sprintf(page, "%s\n",
2156 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2159 static ssize_t o2hb_heartbeat_group_mode_store(struct config_item *item,
2160 const char *page, size_t count)
2166 len = (page[count - 1] == '\n') ? count - 1 : count;
2170 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2171 if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
2174 ret = o2hb_global_heartbeat_mode_set(i);
2176 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2177 o2hb_heartbeat_mode_desc[i]);
2185 CONFIGFS_ATTR(o2hb_heartbeat_group_, dead_threshold);
2186 CONFIGFS_ATTR(o2hb_heartbeat_group_, mode);
2188 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2189 &o2hb_heartbeat_group_attr_dead_threshold,
2190 &o2hb_heartbeat_group_attr_mode,
2194 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2195 .make_item = o2hb_heartbeat_group_make_item,
2196 .drop_item = o2hb_heartbeat_group_drop_item,
2199 static const struct config_item_type o2hb_heartbeat_group_type = {
2200 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
2201 .ct_attrs = o2hb_heartbeat_group_attrs,
2202 .ct_owner = THIS_MODULE,
2205 /* this is just here to avoid touching group in heartbeat.h which the
2206 * entire damn world #includes */
2207 struct config_group *o2hb_alloc_hb_set(void)
2209 struct o2hb_heartbeat_group *hs = NULL;
2210 struct config_group *ret = NULL;
2212 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2216 config_group_init_type_name(&hs->hs_group, "heartbeat",
2217 &o2hb_heartbeat_group_type);
2219 ret = &hs->hs_group;
2226 void o2hb_free_hb_set(struct config_group *group)
2228 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2232 /* hb callback registration and issuing */
2234 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2236 if (type == O2HB_NUM_CB)
2237 return ERR_PTR(-EINVAL);
2239 return &o2hb_callbacks[type];
2242 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2243 enum o2hb_callback_type type,
2248 INIT_LIST_HEAD(&hc->hc_item);
2251 hc->hc_priority = priority;
2253 hc->hc_magic = O2HB_CB_MAGIC;
2255 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2258 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2259 * In global heartbeat mode, region_uuid passed is NULL.
2261 * In local, we only pin the matching region. In global we pin all the active
2264 static int o2hb_region_pin(const char *region_uuid)
2266 int ret = 0, found = 0;
2267 struct o2hb_region *reg;
2270 assert_spin_locked(&o2hb_live_lock);
2272 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2273 if (reg->hr_item_dropped)
2276 uuid = config_item_name(®->hr_item);
2278 /* local heartbeat */
2280 if (strcmp(region_uuid, uuid))
2285 if (reg->hr_item_pinned || reg->hr_item_dropped)
2288 /* Ignore ENOENT only for local hb (userdlm domain) */
2289 ret = o2nm_depend_item(®->hr_item);
2291 mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2292 reg->hr_item_pinned = 1;
2294 if (ret == -ENOENT && found)
2297 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2311 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2312 * In global heartbeat mode, region_uuid passed is NULL.
2314 * In local, we only unpin the matching region. In global we unpin all the
2317 static void o2hb_region_unpin(const char *region_uuid)
2319 struct o2hb_region *reg;
2323 assert_spin_locked(&o2hb_live_lock);
2325 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2326 if (reg->hr_item_dropped)
2329 uuid = config_item_name(®->hr_item);
2331 if (strcmp(region_uuid, uuid))
2336 if (reg->hr_item_pinned) {
2337 mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2338 o2nm_undepend_item(®->hr_item);
2339 reg->hr_item_pinned = 0;
2346 static int o2hb_region_inc_user(const char *region_uuid)
2350 spin_lock(&o2hb_live_lock);
2352 /* local heartbeat */
2353 if (!o2hb_global_heartbeat_active()) {
2354 ret = o2hb_region_pin(region_uuid);
2359 * if global heartbeat active and this is the first dependent user,
2360 * pin all regions if quorum region count <= CUT_OFF
2362 o2hb_dependent_users++;
2363 if (o2hb_dependent_users > 1)
2366 if (bitmap_weight(o2hb_quorum_region_bitmap,
2367 O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2368 ret = o2hb_region_pin(NULL);
2371 spin_unlock(&o2hb_live_lock);
2375 static void o2hb_region_dec_user(const char *region_uuid)
2377 spin_lock(&o2hb_live_lock);
2379 /* local heartbeat */
2380 if (!o2hb_global_heartbeat_active()) {
2381 o2hb_region_unpin(region_uuid);
2386 * if global heartbeat active and there are no dependent users,
2387 * unpin all quorum regions
2389 o2hb_dependent_users--;
2390 if (!o2hb_dependent_users)
2391 o2hb_region_unpin(NULL);
2394 spin_unlock(&o2hb_live_lock);
2397 int o2hb_register_callback(const char *region_uuid,
2398 struct o2hb_callback_func *hc)
2400 struct o2hb_callback_func *f;
2401 struct o2hb_callback *hbcall;
2404 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2405 BUG_ON(!list_empty(&hc->hc_item));
2407 hbcall = hbcall_from_type(hc->hc_type);
2408 if (IS_ERR(hbcall)) {
2409 ret = PTR_ERR(hbcall);
2414 ret = o2hb_region_inc_user(region_uuid);
2421 down_write(&o2hb_callback_sem);
2423 list_for_each_entry(f, &hbcall->list, hc_item) {
2424 if (hc->hc_priority < f->hc_priority) {
2425 list_add_tail(&hc->hc_item, &f->hc_item);
2429 if (list_empty(&hc->hc_item))
2430 list_add_tail(&hc->hc_item, &hbcall->list);
2432 up_write(&o2hb_callback_sem);
2435 mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2436 ret, __builtin_return_address(0), hc);
2439 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2441 void o2hb_unregister_callback(const char *region_uuid,
2442 struct o2hb_callback_func *hc)
2444 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2446 mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2447 __builtin_return_address(0), hc);
2449 /* XXX Can this happen _with_ a region reference? */
2450 if (list_empty(&hc->hc_item))
2454 o2hb_region_dec_user(region_uuid);
2456 down_write(&o2hb_callback_sem);
2458 list_del_init(&hc->hc_item);
2460 up_write(&o2hb_callback_sem);
2462 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2464 int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2466 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2468 spin_lock(&o2hb_live_lock);
2469 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2470 spin_unlock(&o2hb_live_lock);
2471 if (!test_bit(node_num, testing_map)) {
2473 "node (%u) does not have heartbeating enabled.\n",
2480 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2482 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2484 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2486 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2487 if (!test_bit(node_num, testing_map)) {
2489 "node (%u) does not have heartbeating enabled.\n",
2496 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2499 * this is just a hack until we get the plumbing which flips file systems
2500 * read only and drops the hb ref instead of killing the node dead.
2502 void o2hb_stop_all_regions(void)
2504 struct o2hb_region *reg;
2506 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2508 spin_lock(&o2hb_live_lock);
2510 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2511 reg->hr_unclean_stop = 1;
2513 spin_unlock(&o2hb_live_lock);
2515 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2517 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2519 struct o2hb_region *reg;
2523 spin_lock(&o2hb_live_lock);
2526 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2527 if (reg->hr_item_dropped)
2530 mlog(0, "Region: %s\n", config_item_name(®->hr_item));
2531 if (numregs < max_regions) {
2532 memcpy(p, config_item_name(®->hr_item),
2533 O2HB_MAX_REGION_NAME_LEN);
2534 p += O2HB_MAX_REGION_NAME_LEN;
2539 spin_unlock(&o2hb_live_lock);
2543 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2545 int o2hb_global_heartbeat_active(void)
2547 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2549 EXPORT_SYMBOL(o2hb_global_heartbeat_active);