xref: /linux/fs/ocfs2/cluster/heartbeat.c (revision 9d796e66230205cd3366f5660387bd9ecca9d336)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public
8  * License as published by the Free Software Foundation; either
9  * version 2 of the License, or (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public
17  * License along with this program; if not, write to the
18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19  * Boston, MA 021110-1307, USA.
20  */
21 
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/jiffies.h>
25 #include <linux/module.h>
26 #include <linux/fs.h>
27 #include <linux/bio.h>
28 #include <linux/blkdev.h>
29 #include <linux/delay.h>
30 #include <linux/file.h>
31 #include <linux/kthread.h>
32 #include <linux/configfs.h>
33 #include <linux/random.h>
34 #include <linux/crc32.h>
35 #include <linux/time.h>
36 #include <linux/debugfs.h>
37 #include <linux/slab.h>
38 #include <linux/bitmap.h>
39 
40 #include "heartbeat.h"
41 #include "tcp.h"
42 #include "nodemanager.h"
43 #include "quorum.h"
44 
45 #include "masklog.h"
46 
47 
48 /*
49  * The first heartbeat pass had one global thread that would serialize all hb
50  * callback calls.  This global serializing sem should only be removed once
51  * we've made sure that all callees can deal with being called concurrently
52  * from multiple hb region threads.
53  */
54 static DECLARE_RWSEM(o2hb_callback_sem);
55 
56 /*
57  * multiple hb threads are watching multiple regions.  A node is live
58  * whenever any of the threads sees activity from the node in its region.
59  */
60 static DEFINE_SPINLOCK(o2hb_live_lock);
61 static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
62 static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
63 static LIST_HEAD(o2hb_node_events);
64 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
65 
66 /*
67  * In global heartbeat, we maintain a series of region bitmaps.
68  * 	- o2hb_region_bitmap allows us to limit the region number to max region.
69  * 	- o2hb_live_region_bitmap tracks live regions (seen steady iterations).
70  * 	- o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
71  * 		heartbeat on it.
72  * 	- o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
73  */
74 static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
75 static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
76 static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
77 static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
78 
79 #define O2HB_DB_TYPE_LIVENODES		0
80 #define O2HB_DB_TYPE_LIVEREGIONS	1
81 #define O2HB_DB_TYPE_QUORUMREGIONS	2
82 #define O2HB_DB_TYPE_FAILEDREGIONS	3
83 #define O2HB_DB_TYPE_REGION_LIVENODES	4
84 #define O2HB_DB_TYPE_REGION_NUMBER	5
85 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME	6
86 #define O2HB_DB_TYPE_REGION_PINNED	7
87 struct o2hb_debug_buf {
88 	int db_type;
89 	int db_size;
90 	int db_len;
91 	void *db_data;
92 };
93 
94 static struct o2hb_debug_buf *o2hb_db_livenodes;
95 static struct o2hb_debug_buf *o2hb_db_liveregions;
96 static struct o2hb_debug_buf *o2hb_db_quorumregions;
97 static struct o2hb_debug_buf *o2hb_db_failedregions;
98 
99 #define O2HB_DEBUG_DIR			"o2hb"
100 #define O2HB_DEBUG_LIVENODES		"livenodes"
101 #define O2HB_DEBUG_LIVEREGIONS		"live_regions"
102 #define O2HB_DEBUG_QUORUMREGIONS	"quorum_regions"
103 #define O2HB_DEBUG_FAILEDREGIONS	"failed_regions"
104 #define O2HB_DEBUG_REGION_NUMBER	"num"
105 #define O2HB_DEBUG_REGION_ELAPSED_TIME	"elapsed_time_in_ms"
106 #define O2HB_DEBUG_REGION_PINNED	"pinned"
107 
108 static struct dentry *o2hb_debug_dir;
109 static struct dentry *o2hb_debug_livenodes;
110 static struct dentry *o2hb_debug_liveregions;
111 static struct dentry *o2hb_debug_quorumregions;
112 static struct dentry *o2hb_debug_failedregions;
113 
114 static LIST_HEAD(o2hb_all_regions);
115 
116 static struct o2hb_callback {
117 	struct list_head list;
118 } o2hb_callbacks[O2HB_NUM_CB];
119 
120 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
121 
122 #define O2HB_DEFAULT_BLOCK_BITS       9
123 
124 enum o2hb_heartbeat_modes {
125 	O2HB_HEARTBEAT_LOCAL		= 0,
126 	O2HB_HEARTBEAT_GLOBAL,
127 	O2HB_HEARTBEAT_NUM_MODES,
128 };
129 
130 char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
131 		"local",	/* O2HB_HEARTBEAT_LOCAL */
132 		"global",	/* O2HB_HEARTBEAT_GLOBAL */
133 };
134 
135 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
136 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
137 
138 /*
139  * o2hb_dependent_users tracks the number of registered callbacks that depend
140  * on heartbeat. o2net and o2dlm are two entities that register this callback.
141  * However only o2dlm depends on the heartbeat. It does not want the heartbeat
142  * to stop while a dlm domain is still active.
143  */
144 unsigned int o2hb_dependent_users;
145 
146 /*
147  * In global heartbeat mode, all regions are pinned if there are one or more
148  * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
149  * regions are unpinned if the region count exceeds the cut off or the number
150  * of dependent users falls to zero.
151  */
152 #define O2HB_PIN_CUT_OFF		3
153 
154 /*
155  * In local heartbeat mode, we assume the dlm domain name to be the same as
156  * region uuid. This is true for domains created for the file system but not
157  * necessarily true for userdlm domains. This is a known limitation.
158  *
159  * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
160  * works for both file system and userdlm domains.
161  */
162 static int o2hb_region_pin(const char *region_uuid);
163 static void o2hb_region_unpin(const char *region_uuid);
164 
165 /* Only sets a new threshold if there are no active regions.
166  *
167  * No locking or otherwise interesting code is required for reading
168  * o2hb_dead_threshold as it can't change once regions are active and
169  * it's not interesting to anyone until then anyway. */
170 static void o2hb_dead_threshold_set(unsigned int threshold)
171 {
172 	if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
173 		spin_lock(&o2hb_live_lock);
174 		if (list_empty(&o2hb_all_regions))
175 			o2hb_dead_threshold = threshold;
176 		spin_unlock(&o2hb_live_lock);
177 	}
178 }
179 
180 static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
181 {
182 	int ret = -1;
183 
184 	if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
185 		spin_lock(&o2hb_live_lock);
186 		if (list_empty(&o2hb_all_regions)) {
187 			o2hb_heartbeat_mode = hb_mode;
188 			ret = 0;
189 		}
190 		spin_unlock(&o2hb_live_lock);
191 	}
192 
193 	return ret;
194 }
195 
196 struct o2hb_node_event {
197 	struct list_head        hn_item;
198 	enum o2hb_callback_type hn_event_type;
199 	struct o2nm_node        *hn_node;
200 	int                     hn_node_num;
201 };
202 
203 struct o2hb_disk_slot {
204 	struct o2hb_disk_heartbeat_block *ds_raw_block;
205 	u8			ds_node_num;
206 	u64			ds_last_time;
207 	u64			ds_last_generation;
208 	u16			ds_equal_samples;
209 	u16			ds_changed_samples;
210 	struct list_head	ds_live_item;
211 };
212 
213 /* each thread owns a region.. when we're asked to tear down the region
214  * we ask the thread to stop, who cleans up the region */
215 struct o2hb_region {
216 	struct config_item	hr_item;
217 
218 	struct list_head	hr_all_item;
219 	unsigned		hr_unclean_stop:1,
220 				hr_aborted_start:1,
221 				hr_item_pinned:1,
222 				hr_item_dropped:1;
223 
224 	/* protected by the hr_callback_sem */
225 	struct task_struct 	*hr_task;
226 
227 	unsigned int		hr_blocks;
228 	unsigned long long	hr_start_block;
229 
230 	unsigned int		hr_block_bits;
231 	unsigned int		hr_block_bytes;
232 
233 	unsigned int		hr_slots_per_page;
234 	unsigned int		hr_num_pages;
235 
236 	struct page             **hr_slot_data;
237 	struct block_device	*hr_bdev;
238 	struct o2hb_disk_slot	*hr_slots;
239 
240 	/* live node map of this region */
241 	unsigned long		hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
242 	unsigned int		hr_region_num;
243 
244 	struct dentry		*hr_debug_dir;
245 	struct dentry		*hr_debug_livenodes;
246 	struct dentry		*hr_debug_regnum;
247 	struct dentry		*hr_debug_elapsed_time;
248 	struct dentry		*hr_debug_pinned;
249 	struct o2hb_debug_buf	*hr_db_livenodes;
250 	struct o2hb_debug_buf	*hr_db_regnum;
251 	struct o2hb_debug_buf	*hr_db_elapsed_time;
252 	struct o2hb_debug_buf	*hr_db_pinned;
253 
254 	/* let the person setting up hb wait for it to return until it
255 	 * has reached a 'steady' state.  This will be fixed when we have
256 	 * a more complete api that doesn't lead to this sort of fragility. */
257 	atomic_t		hr_steady_iterations;
258 
259 	/* terminate o2hb thread if it does not reach steady state
260 	 * (hr_steady_iterations == 0) within hr_unsteady_iterations */
261 	atomic_t		hr_unsteady_iterations;
262 
263 	char			hr_dev_name[BDEVNAME_SIZE];
264 
265 	unsigned int		hr_timeout_ms;
266 
267 	/* randomized as the region goes up and down so that a node
268 	 * recognizes a node going up and down in one iteration */
269 	u64			hr_generation;
270 
271 	struct delayed_work	hr_write_timeout_work;
272 	unsigned long		hr_last_timeout_start;
273 
274 	/* Used during o2hb_check_slot to hold a copy of the block
275 	 * being checked because we temporarily have to zero out the
276 	 * crc field. */
277 	struct o2hb_disk_heartbeat_block *hr_tmp_block;
278 };
279 
280 struct o2hb_bio_wait_ctxt {
281 	atomic_t          wc_num_reqs;
282 	struct completion wc_io_complete;
283 	int               wc_error;
284 };
285 
286 static void o2hb_write_timeout(struct work_struct *work)
287 {
288 	int failed, quorum;
289 	unsigned long flags;
290 	struct o2hb_region *reg =
291 		container_of(work, struct o2hb_region,
292 			     hr_write_timeout_work.work);
293 
294 	mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
295 	     "milliseconds\n", reg->hr_dev_name,
296 	     jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
297 
298 	if (o2hb_global_heartbeat_active()) {
299 		spin_lock_irqsave(&o2hb_live_lock, flags);
300 		if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
301 			set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
302 		failed = bitmap_weight(o2hb_failed_region_bitmap,
303 					O2NM_MAX_REGIONS);
304 		quorum = bitmap_weight(o2hb_quorum_region_bitmap,
305 					O2NM_MAX_REGIONS);
306 		spin_unlock_irqrestore(&o2hb_live_lock, flags);
307 
308 		mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
309 		     quorum, failed);
310 
311 		/*
312 		 * Fence if the number of failed regions >= half the number
313 		 * of  quorum regions
314 		 */
315 		if ((failed << 1) < quorum)
316 			return;
317 	}
318 
319 	o2quo_disk_timeout();
320 }
321 
322 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
323 {
324 	/* Arm writeout only after thread reaches steady state */
325 	if (atomic_read(&reg->hr_steady_iterations) != 0)
326 		return;
327 
328 	mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
329 	     O2HB_MAX_WRITE_TIMEOUT_MS);
330 
331 	if (o2hb_global_heartbeat_active()) {
332 		spin_lock(&o2hb_live_lock);
333 		clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
334 		spin_unlock(&o2hb_live_lock);
335 	}
336 	cancel_delayed_work(&reg->hr_write_timeout_work);
337 	reg->hr_last_timeout_start = jiffies;
338 	schedule_delayed_work(&reg->hr_write_timeout_work,
339 			      msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
340 }
341 
342 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
343 {
344 	cancel_delayed_work_sync(&reg->hr_write_timeout_work);
345 }
346 
347 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
348 {
349 	atomic_set(&wc->wc_num_reqs, 1);
350 	init_completion(&wc->wc_io_complete);
351 	wc->wc_error = 0;
352 }
353 
354 /* Used in error paths too */
355 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
356 				     unsigned int num)
357 {
358 	/* sadly atomic_sub_and_test() isn't available on all platforms.  The
359 	 * good news is that the fast path only completes one at a time */
360 	while(num--) {
361 		if (atomic_dec_and_test(&wc->wc_num_reqs)) {
362 			BUG_ON(num > 0);
363 			complete(&wc->wc_io_complete);
364 		}
365 	}
366 }
367 
368 static void o2hb_wait_on_io(struct o2hb_region *reg,
369 			    struct o2hb_bio_wait_ctxt *wc)
370 {
371 	o2hb_bio_wait_dec(wc, 1);
372 	wait_for_completion(&wc->wc_io_complete);
373 }
374 
375 static void o2hb_bio_end_io(struct bio *bio,
376 			   int error)
377 {
378 	struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
379 
380 	if (error) {
381 		mlog(ML_ERROR, "IO Error %d\n", error);
382 		wc->wc_error = error;
383 	}
384 
385 	o2hb_bio_wait_dec(wc, 1);
386 	bio_put(bio);
387 }
388 
389 /* Setup a Bio to cover I/O against num_slots slots starting at
390  * start_slot. */
391 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
392 				      struct o2hb_bio_wait_ctxt *wc,
393 				      unsigned int *current_slot,
394 				      unsigned int max_slots)
395 {
396 	int len, current_page;
397 	unsigned int vec_len, vec_start;
398 	unsigned int bits = reg->hr_block_bits;
399 	unsigned int spp = reg->hr_slots_per_page;
400 	unsigned int cs = *current_slot;
401 	struct bio *bio;
402 	struct page *page;
403 
404 	/* Testing has shown this allocation to take long enough under
405 	 * GFP_KERNEL that the local node can get fenced. It would be
406 	 * nicest if we could pre-allocate these bios and avoid this
407 	 * all together. */
408 	bio = bio_alloc(GFP_ATOMIC, 16);
409 	if (!bio) {
410 		mlog(ML_ERROR, "Could not alloc slots BIO!\n");
411 		bio = ERR_PTR(-ENOMEM);
412 		goto bail;
413 	}
414 
415 	/* Must put everything in 512 byte sectors for the bio... */
416 	bio->bi_iter.bi_sector = (reg->hr_start_block + cs) << (bits - 9);
417 	bio->bi_bdev = reg->hr_bdev;
418 	bio->bi_private = wc;
419 	bio->bi_end_io = o2hb_bio_end_io;
420 
421 	vec_start = (cs << bits) % PAGE_CACHE_SIZE;
422 	while(cs < max_slots) {
423 		current_page = cs / spp;
424 		page = reg->hr_slot_data[current_page];
425 
426 		vec_len = min(PAGE_CACHE_SIZE - vec_start,
427 			      (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
428 
429 		mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
430 		     current_page, vec_len, vec_start);
431 
432 		len = bio_add_page(bio, page, vec_len, vec_start);
433 		if (len != vec_len) break;
434 
435 		cs += vec_len / (PAGE_CACHE_SIZE/spp);
436 		vec_start = 0;
437 	}
438 
439 bail:
440 	*current_slot = cs;
441 	return bio;
442 }
443 
444 static int o2hb_read_slots(struct o2hb_region *reg,
445 			   unsigned int max_slots)
446 {
447 	unsigned int current_slot=0;
448 	int status;
449 	struct o2hb_bio_wait_ctxt wc;
450 	struct bio *bio;
451 
452 	o2hb_bio_wait_init(&wc);
453 
454 	while(current_slot < max_slots) {
455 		bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
456 		if (IS_ERR(bio)) {
457 			status = PTR_ERR(bio);
458 			mlog_errno(status);
459 			goto bail_and_wait;
460 		}
461 
462 		atomic_inc(&wc.wc_num_reqs);
463 		submit_bio(READ, bio);
464 	}
465 
466 	status = 0;
467 
468 bail_and_wait:
469 	o2hb_wait_on_io(reg, &wc);
470 	if (wc.wc_error && !status)
471 		status = wc.wc_error;
472 
473 	return status;
474 }
475 
476 static int o2hb_issue_node_write(struct o2hb_region *reg,
477 				 struct o2hb_bio_wait_ctxt *write_wc)
478 {
479 	int status;
480 	unsigned int slot;
481 	struct bio *bio;
482 
483 	o2hb_bio_wait_init(write_wc);
484 
485 	slot = o2nm_this_node();
486 
487 	bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
488 	if (IS_ERR(bio)) {
489 		status = PTR_ERR(bio);
490 		mlog_errno(status);
491 		goto bail;
492 	}
493 
494 	atomic_inc(&write_wc->wc_num_reqs);
495 	submit_bio(WRITE_SYNC, bio);
496 
497 	status = 0;
498 bail:
499 	return status;
500 }
501 
502 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
503 				     struct o2hb_disk_heartbeat_block *hb_block)
504 {
505 	__le32 old_cksum;
506 	u32 ret;
507 
508 	/* We want to compute the block crc with a 0 value in the
509 	 * hb_cksum field. Save it off here and replace after the
510 	 * crc. */
511 	old_cksum = hb_block->hb_cksum;
512 	hb_block->hb_cksum = 0;
513 
514 	ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
515 
516 	hb_block->hb_cksum = old_cksum;
517 
518 	return ret;
519 }
520 
521 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
522 {
523 	mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
524 	     "cksum = 0x%x, generation 0x%llx\n",
525 	     (long long)le64_to_cpu(hb_block->hb_seq),
526 	     hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
527 	     (long long)le64_to_cpu(hb_block->hb_generation));
528 }
529 
530 static int o2hb_verify_crc(struct o2hb_region *reg,
531 			   struct o2hb_disk_heartbeat_block *hb_block)
532 {
533 	u32 read, computed;
534 
535 	read = le32_to_cpu(hb_block->hb_cksum);
536 	computed = o2hb_compute_block_crc_le(reg, hb_block);
537 
538 	return read == computed;
539 }
540 
541 /*
542  * Compare the slot data with what we wrote in the last iteration.
543  * If the match fails, print an appropriate error message. This is to
544  * detect errors like... another node hearting on the same slot,
545  * flaky device that is losing writes, etc.
546  * Returns 1 if check succeeds, 0 otherwise.
547  */
548 static int o2hb_check_own_slot(struct o2hb_region *reg)
549 {
550 	struct o2hb_disk_slot *slot;
551 	struct o2hb_disk_heartbeat_block *hb_block;
552 	char *errstr;
553 
554 	slot = &reg->hr_slots[o2nm_this_node()];
555 	/* Don't check on our 1st timestamp */
556 	if (!slot->ds_last_time)
557 		return 0;
558 
559 	hb_block = slot->ds_raw_block;
560 	if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
561 	    le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
562 	    hb_block->hb_node == slot->ds_node_num)
563 		return 1;
564 
565 #define ERRSTR1		"Another node is heartbeating on device"
566 #define ERRSTR2		"Heartbeat generation mismatch on device"
567 #define ERRSTR3		"Heartbeat sequence mismatch on device"
568 
569 	if (hb_block->hb_node != slot->ds_node_num)
570 		errstr = ERRSTR1;
571 	else if (le64_to_cpu(hb_block->hb_generation) !=
572 		 slot->ds_last_generation)
573 		errstr = ERRSTR2;
574 	else
575 		errstr = ERRSTR3;
576 
577 	mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
578 	     "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
579 	     slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
580 	     (unsigned long long)slot->ds_last_time, hb_block->hb_node,
581 	     (unsigned long long)le64_to_cpu(hb_block->hb_generation),
582 	     (unsigned long long)le64_to_cpu(hb_block->hb_seq));
583 
584 	return 0;
585 }
586 
587 static inline void o2hb_prepare_block(struct o2hb_region *reg,
588 				      u64 generation)
589 {
590 	int node_num;
591 	u64 cputime;
592 	struct o2hb_disk_slot *slot;
593 	struct o2hb_disk_heartbeat_block *hb_block;
594 
595 	node_num = o2nm_this_node();
596 	slot = &reg->hr_slots[node_num];
597 
598 	hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
599 	memset(hb_block, 0, reg->hr_block_bytes);
600 	/* TODO: time stuff */
601 	cputime = CURRENT_TIME.tv_sec;
602 	if (!cputime)
603 		cputime = 1;
604 
605 	hb_block->hb_seq = cpu_to_le64(cputime);
606 	hb_block->hb_node = node_num;
607 	hb_block->hb_generation = cpu_to_le64(generation);
608 	hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
609 
610 	/* This step must always happen last! */
611 	hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
612 								   hb_block));
613 
614 	mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
615 	     (long long)generation,
616 	     le32_to_cpu(hb_block->hb_cksum));
617 }
618 
619 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
620 				struct o2nm_node *node,
621 				int idx)
622 {
623 	struct o2hb_callback_func *f;
624 
625 	list_for_each_entry(f, &hbcall->list, hc_item) {
626 		mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
627 		(f->hc_func)(node, idx, f->hc_data);
628 	}
629 }
630 
631 /* Will run the list in order until we process the passed event */
632 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
633 {
634 	struct o2hb_callback *hbcall;
635 	struct o2hb_node_event *event;
636 
637 	/* Holding callback sem assures we don't alter the callback
638 	 * lists when doing this, and serializes ourselves with other
639 	 * processes wanting callbacks. */
640 	down_write(&o2hb_callback_sem);
641 
642 	spin_lock(&o2hb_live_lock);
643 	while (!list_empty(&o2hb_node_events)
644 	       && !list_empty(&queued_event->hn_item)) {
645 		event = list_entry(o2hb_node_events.next,
646 				   struct o2hb_node_event,
647 				   hn_item);
648 		list_del_init(&event->hn_item);
649 		spin_unlock(&o2hb_live_lock);
650 
651 		mlog(ML_HEARTBEAT, "Node %s event for %d\n",
652 		     event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
653 		     event->hn_node_num);
654 
655 		hbcall = hbcall_from_type(event->hn_event_type);
656 
657 		/* We should *never* have gotten on to the list with a
658 		 * bad type... This isn't something that we should try
659 		 * to recover from. */
660 		BUG_ON(IS_ERR(hbcall));
661 
662 		o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
663 
664 		spin_lock(&o2hb_live_lock);
665 	}
666 	spin_unlock(&o2hb_live_lock);
667 
668 	up_write(&o2hb_callback_sem);
669 }
670 
671 static void o2hb_queue_node_event(struct o2hb_node_event *event,
672 				  enum o2hb_callback_type type,
673 				  struct o2nm_node *node,
674 				  int node_num)
675 {
676 	assert_spin_locked(&o2hb_live_lock);
677 
678 	BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
679 
680 	event->hn_event_type = type;
681 	event->hn_node = node;
682 	event->hn_node_num = node_num;
683 
684 	mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
685 	     type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
686 
687 	list_add_tail(&event->hn_item, &o2hb_node_events);
688 }
689 
690 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
691 {
692 	struct o2hb_node_event event =
693 		{ .hn_item = LIST_HEAD_INIT(event.hn_item), };
694 	struct o2nm_node *node;
695 	int queued = 0;
696 
697 	node = o2nm_get_node_by_num(slot->ds_node_num);
698 	if (!node)
699 		return;
700 
701 	spin_lock(&o2hb_live_lock);
702 	if (!list_empty(&slot->ds_live_item)) {
703 		mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
704 		     slot->ds_node_num);
705 
706 		list_del_init(&slot->ds_live_item);
707 
708 		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
709 			clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
710 
711 			o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
712 					      slot->ds_node_num);
713 			queued = 1;
714 		}
715 	}
716 	spin_unlock(&o2hb_live_lock);
717 
718 	if (queued)
719 		o2hb_run_event_list(&event);
720 
721 	o2nm_node_put(node);
722 }
723 
724 static void o2hb_set_quorum_device(struct o2hb_region *reg)
725 {
726 	if (!o2hb_global_heartbeat_active())
727 		return;
728 
729 	/* Prevent race with o2hb_heartbeat_group_drop_item() */
730 	if (kthread_should_stop())
731 		return;
732 
733 	/* Tag region as quorum only after thread reaches steady state */
734 	if (atomic_read(&reg->hr_steady_iterations) != 0)
735 		return;
736 
737 	spin_lock(&o2hb_live_lock);
738 
739 	if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
740 		goto unlock;
741 
742 	/*
743 	 * A region can be added to the quorum only when it sees all
744 	 * live nodes heartbeat on it. In other words, the region has been
745 	 * added to all nodes.
746 	 */
747 	if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
748 		   sizeof(o2hb_live_node_bitmap)))
749 		goto unlock;
750 
751 	printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
752 	       config_item_name(&reg->hr_item), reg->hr_dev_name);
753 
754 	set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
755 
756 	/*
757 	 * If global heartbeat active, unpin all regions if the
758 	 * region count > CUT_OFF
759 	 */
760 	if (bitmap_weight(o2hb_quorum_region_bitmap,
761 			   O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
762 		o2hb_region_unpin(NULL);
763 unlock:
764 	spin_unlock(&o2hb_live_lock);
765 }
766 
767 static int o2hb_check_slot(struct o2hb_region *reg,
768 			   struct o2hb_disk_slot *slot)
769 {
770 	int changed = 0, gen_changed = 0;
771 	struct o2hb_node_event event =
772 		{ .hn_item = LIST_HEAD_INIT(event.hn_item), };
773 	struct o2nm_node *node;
774 	struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
775 	u64 cputime;
776 	unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
777 	unsigned int slot_dead_ms;
778 	int tmp;
779 	int queued = 0;
780 
781 	memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
782 
783 	/*
784 	 * If a node is no longer configured but is still in the livemap, we
785 	 * may need to clear that bit from the livemap.
786 	 */
787 	node = o2nm_get_node_by_num(slot->ds_node_num);
788 	if (!node) {
789 		spin_lock(&o2hb_live_lock);
790 		tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
791 		spin_unlock(&o2hb_live_lock);
792 		if (!tmp)
793 			return 0;
794 	}
795 
796 	if (!o2hb_verify_crc(reg, hb_block)) {
797 		/* all paths from here will drop o2hb_live_lock for
798 		 * us. */
799 		spin_lock(&o2hb_live_lock);
800 
801 		/* Don't print an error on the console in this case -
802 		 * a freshly formatted heartbeat area will not have a
803 		 * crc set on it. */
804 		if (list_empty(&slot->ds_live_item))
805 			goto out;
806 
807 		/* The node is live but pushed out a bad crc. We
808 		 * consider it a transient miss but don't populate any
809 		 * other values as they may be junk. */
810 		mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
811 		     slot->ds_node_num, reg->hr_dev_name);
812 		o2hb_dump_slot(hb_block);
813 
814 		slot->ds_equal_samples++;
815 		goto fire_callbacks;
816 	}
817 
818 	/* we don't care if these wrap.. the state transitions below
819 	 * clear at the right places */
820 	cputime = le64_to_cpu(hb_block->hb_seq);
821 	if (slot->ds_last_time != cputime)
822 		slot->ds_changed_samples++;
823 	else
824 		slot->ds_equal_samples++;
825 	slot->ds_last_time = cputime;
826 
827 	/* The node changed heartbeat generations. We assume this to
828 	 * mean it dropped off but came back before we timed out. We
829 	 * want to consider it down for the time being but don't want
830 	 * to lose any changed_samples state we might build up to
831 	 * considering it live again. */
832 	if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
833 		gen_changed = 1;
834 		slot->ds_equal_samples = 0;
835 		mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
836 		     "to 0x%llx)\n", slot->ds_node_num,
837 		     (long long)slot->ds_last_generation,
838 		     (long long)le64_to_cpu(hb_block->hb_generation));
839 	}
840 
841 	slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
842 
843 	mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
844 	     "seq %llu last %llu changed %u equal %u\n",
845 	     slot->ds_node_num, (long long)slot->ds_last_generation,
846 	     le32_to_cpu(hb_block->hb_cksum),
847 	     (unsigned long long)le64_to_cpu(hb_block->hb_seq),
848 	     (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
849 	     slot->ds_equal_samples);
850 
851 	spin_lock(&o2hb_live_lock);
852 
853 fire_callbacks:
854 	/* dead nodes only come to life after some number of
855 	 * changes at any time during their dead time */
856 	if (list_empty(&slot->ds_live_item) &&
857 	    slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
858 		mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
859 		     slot->ds_node_num, (long long)slot->ds_last_generation);
860 
861 		set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
862 
863 		/* first on the list generates a callback */
864 		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
865 			mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
866 			     "bitmap\n", slot->ds_node_num);
867 			set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
868 
869 			o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
870 					      slot->ds_node_num);
871 
872 			changed = 1;
873 			queued = 1;
874 		}
875 
876 		list_add_tail(&slot->ds_live_item,
877 			      &o2hb_live_slots[slot->ds_node_num]);
878 
879 		slot->ds_equal_samples = 0;
880 
881 		/* We want to be sure that all nodes agree on the
882 		 * number of milliseconds before a node will be
883 		 * considered dead. The self-fencing timeout is
884 		 * computed from this value, and a discrepancy might
885 		 * result in heartbeat calling a node dead when it
886 		 * hasn't self-fenced yet. */
887 		slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
888 		if (slot_dead_ms && slot_dead_ms != dead_ms) {
889 			/* TODO: Perhaps we can fail the region here. */
890 			mlog(ML_ERROR, "Node %d on device %s has a dead count "
891 			     "of %u ms, but our count is %u ms.\n"
892 			     "Please double check your configuration values "
893 			     "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
894 			     slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
895 			     dead_ms);
896 		}
897 		goto out;
898 	}
899 
900 	/* if the list is dead, we're done.. */
901 	if (list_empty(&slot->ds_live_item))
902 		goto out;
903 
904 	/* live nodes only go dead after enough consequtive missed
905 	 * samples..  reset the missed counter whenever we see
906 	 * activity */
907 	if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
908 		mlog(ML_HEARTBEAT, "Node %d left my region\n",
909 		     slot->ds_node_num);
910 
911 		clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
912 
913 		/* last off the live_slot generates a callback */
914 		list_del_init(&slot->ds_live_item);
915 		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
916 			mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
917 			     "nodes bitmap\n", slot->ds_node_num);
918 			clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
919 
920 			/* node can be null */
921 			o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
922 					      node, slot->ds_node_num);
923 
924 			changed = 1;
925 			queued = 1;
926 		}
927 
928 		/* We don't clear this because the node is still
929 		 * actually writing new blocks. */
930 		if (!gen_changed)
931 			slot->ds_changed_samples = 0;
932 		goto out;
933 	}
934 	if (slot->ds_changed_samples) {
935 		slot->ds_changed_samples = 0;
936 		slot->ds_equal_samples = 0;
937 	}
938 out:
939 	spin_unlock(&o2hb_live_lock);
940 
941 	if (queued)
942 		o2hb_run_event_list(&event);
943 
944 	if (node)
945 		o2nm_node_put(node);
946 	return changed;
947 }
948 
949 static int o2hb_highest_node(unsigned long *nodes, int numbits)
950 {
951 	return find_last_bit(nodes, numbits);
952 }
953 
954 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
955 {
956 	int i, ret, highest_node;
957 	int membership_change = 0, own_slot_ok = 0;
958 	unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
959 	unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
960 	struct o2hb_bio_wait_ctxt write_wc;
961 
962 	ret = o2nm_configured_node_map(configured_nodes,
963 				       sizeof(configured_nodes));
964 	if (ret) {
965 		mlog_errno(ret);
966 		goto bail;
967 	}
968 
969 	/*
970 	 * If a node is not configured but is in the livemap, we still need
971 	 * to read the slot so as to be able to remove it from the livemap.
972 	 */
973 	o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
974 	i = -1;
975 	while ((i = find_next_bit(live_node_bitmap,
976 				  O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
977 		set_bit(i, configured_nodes);
978 	}
979 
980 	highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
981 	if (highest_node >= O2NM_MAX_NODES) {
982 		mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
983 		ret = -EINVAL;
984 		goto bail;
985 	}
986 
987 	/* No sense in reading the slots of nodes that don't exist
988 	 * yet. Of course, if the node definitions have holes in them
989 	 * then we're reading an empty slot anyway... Consider this
990 	 * best-effort. */
991 	ret = o2hb_read_slots(reg, highest_node + 1);
992 	if (ret < 0) {
993 		mlog_errno(ret);
994 		goto bail;
995 	}
996 
997 	/* With an up to date view of the slots, we can check that no
998 	 * other node has been improperly configured to heartbeat in
999 	 * our slot. */
1000 	own_slot_ok = o2hb_check_own_slot(reg);
1001 
1002 	/* fill in the proper info for our next heartbeat */
1003 	o2hb_prepare_block(reg, reg->hr_generation);
1004 
1005 	ret = o2hb_issue_node_write(reg, &write_wc);
1006 	if (ret < 0) {
1007 		mlog_errno(ret);
1008 		goto bail;
1009 	}
1010 
1011 	i = -1;
1012 	while((i = find_next_bit(configured_nodes,
1013 				 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1014 		membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1015 	}
1016 
1017 	/*
1018 	 * We have to be sure we've advertised ourselves on disk
1019 	 * before we can go to steady state.  This ensures that
1020 	 * people we find in our steady state have seen us.
1021 	 */
1022 	o2hb_wait_on_io(reg, &write_wc);
1023 	if (write_wc.wc_error) {
1024 		/* Do not re-arm the write timeout on I/O error - we
1025 		 * can't be sure that the new block ever made it to
1026 		 * disk */
1027 		mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1028 		     write_wc.wc_error, reg->hr_dev_name);
1029 		ret = write_wc.wc_error;
1030 		goto bail;
1031 	}
1032 
1033 	/* Skip disarming the timeout if own slot has stale/bad data */
1034 	if (own_slot_ok) {
1035 		o2hb_set_quorum_device(reg);
1036 		o2hb_arm_write_timeout(reg);
1037 	}
1038 
1039 bail:
1040 	/* let the person who launched us know when things are steady */
1041 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
1042 		if (!ret && own_slot_ok && !membership_change) {
1043 			if (atomic_dec_and_test(&reg->hr_steady_iterations))
1044 				wake_up(&o2hb_steady_queue);
1045 		}
1046 	}
1047 
1048 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
1049 		if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1050 			printk(KERN_NOTICE "o2hb: Unable to stabilize "
1051 			       "heartbeart on region %s (%s)\n",
1052 			       config_item_name(&reg->hr_item),
1053 			       reg->hr_dev_name);
1054 			atomic_set(&reg->hr_steady_iterations, 0);
1055 			reg->hr_aborted_start = 1;
1056 			wake_up(&o2hb_steady_queue);
1057 			ret = -EIO;
1058 		}
1059 	}
1060 
1061 	return ret;
1062 }
1063 
1064 /* Subtract b from a, storing the result in a. a *must* have a larger
1065  * value than b. */
1066 static void o2hb_tv_subtract(struct timeval *a,
1067 			     struct timeval *b)
1068 {
1069 	/* just return 0 when a is after b */
1070 	if (a->tv_sec < b->tv_sec ||
1071 	    (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1072 		a->tv_sec = 0;
1073 		a->tv_usec = 0;
1074 		return;
1075 	}
1076 
1077 	a->tv_sec -= b->tv_sec;
1078 	a->tv_usec -= b->tv_usec;
1079 	while ( a->tv_usec < 0 ) {
1080 		a->tv_sec--;
1081 		a->tv_usec += 1000000;
1082 	}
1083 }
1084 
1085 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1086 				       struct timeval *end)
1087 {
1088 	struct timeval res = *end;
1089 
1090 	o2hb_tv_subtract(&res, start);
1091 
1092 	return res.tv_sec * 1000 + res.tv_usec / 1000;
1093 }
1094 
1095 /*
1096  * we ride the region ref that the region dir holds.  before the region
1097  * dir is removed and drops it ref it will wait to tear down this
1098  * thread.
1099  */
1100 static int o2hb_thread(void *data)
1101 {
1102 	int i, ret;
1103 	struct o2hb_region *reg = data;
1104 	struct o2hb_bio_wait_ctxt write_wc;
1105 	struct timeval before_hb, after_hb;
1106 	unsigned int elapsed_msec;
1107 
1108 	mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1109 
1110 	set_user_nice(current, MIN_NICE);
1111 
1112 	/* Pin node */
1113 	o2nm_depend_this_node();
1114 
1115 	while (!kthread_should_stop() &&
1116 	       !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1117 		/* We track the time spent inside
1118 		 * o2hb_do_disk_heartbeat so that we avoid more than
1119 		 * hr_timeout_ms between disk writes. On busy systems
1120 		 * this should result in a heartbeat which is less
1121 		 * likely to time itself out. */
1122 		do_gettimeofday(&before_hb);
1123 
1124 		ret = o2hb_do_disk_heartbeat(reg);
1125 
1126 		do_gettimeofday(&after_hb);
1127 		elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1128 
1129 		mlog(ML_HEARTBEAT,
1130 		     "start = %lu.%lu, end = %lu.%lu, msec = %u, ret = %d\n",
1131 		     before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1132 		     after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1133 		     elapsed_msec, ret);
1134 
1135 		if (!kthread_should_stop() &&
1136 		    elapsed_msec < reg->hr_timeout_ms) {
1137 			/* the kthread api has blocked signals for us so no
1138 			 * need to record the return value. */
1139 			msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1140 		}
1141 	}
1142 
1143 	o2hb_disarm_write_timeout(reg);
1144 
1145 	/* unclean stop is only used in very bad situation */
1146 	for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1147 		o2hb_shutdown_slot(&reg->hr_slots[i]);
1148 
1149 	/* Explicit down notification - avoid forcing the other nodes
1150 	 * to timeout on this region when we could just as easily
1151 	 * write a clear generation - thus indicating to them that
1152 	 * this node has left this region.
1153 	 */
1154 	if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1155 		o2hb_prepare_block(reg, 0);
1156 		ret = o2hb_issue_node_write(reg, &write_wc);
1157 		if (ret == 0)
1158 			o2hb_wait_on_io(reg, &write_wc);
1159 		else
1160 			mlog_errno(ret);
1161 	}
1162 
1163 	/* Unpin node */
1164 	o2nm_undepend_this_node();
1165 
1166 	mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1167 
1168 	return 0;
1169 }
1170 
1171 #ifdef CONFIG_DEBUG_FS
1172 static int o2hb_debug_open(struct inode *inode, struct file *file)
1173 {
1174 	struct o2hb_debug_buf *db = inode->i_private;
1175 	struct o2hb_region *reg;
1176 	unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1177 	unsigned long lts;
1178 	char *buf = NULL;
1179 	int i = -1;
1180 	int out = 0;
1181 
1182 	/* max_nodes should be the largest bitmap we pass here */
1183 	BUG_ON(sizeof(map) < db->db_size);
1184 
1185 	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1186 	if (!buf)
1187 		goto bail;
1188 
1189 	switch (db->db_type) {
1190 	case O2HB_DB_TYPE_LIVENODES:
1191 	case O2HB_DB_TYPE_LIVEREGIONS:
1192 	case O2HB_DB_TYPE_QUORUMREGIONS:
1193 	case O2HB_DB_TYPE_FAILEDREGIONS:
1194 		spin_lock(&o2hb_live_lock);
1195 		memcpy(map, db->db_data, db->db_size);
1196 		spin_unlock(&o2hb_live_lock);
1197 		break;
1198 
1199 	case O2HB_DB_TYPE_REGION_LIVENODES:
1200 		spin_lock(&o2hb_live_lock);
1201 		reg = (struct o2hb_region *)db->db_data;
1202 		memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1203 		spin_unlock(&o2hb_live_lock);
1204 		break;
1205 
1206 	case O2HB_DB_TYPE_REGION_NUMBER:
1207 		reg = (struct o2hb_region *)db->db_data;
1208 		out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1209 				reg->hr_region_num);
1210 		goto done;
1211 
1212 	case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1213 		reg = (struct o2hb_region *)db->db_data;
1214 		lts = reg->hr_last_timeout_start;
1215 		/* If 0, it has never been set before */
1216 		if (lts)
1217 			lts = jiffies_to_msecs(jiffies - lts);
1218 		out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1219 		goto done;
1220 
1221 	case O2HB_DB_TYPE_REGION_PINNED:
1222 		reg = (struct o2hb_region *)db->db_data;
1223 		out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1224 				!!reg->hr_item_pinned);
1225 		goto done;
1226 
1227 	default:
1228 		goto done;
1229 	}
1230 
1231 	while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1232 		out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1233 	out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1234 
1235 done:
1236 	i_size_write(inode, out);
1237 
1238 	file->private_data = buf;
1239 
1240 	return 0;
1241 bail:
1242 	return -ENOMEM;
1243 }
1244 
1245 static int o2hb_debug_release(struct inode *inode, struct file *file)
1246 {
1247 	kfree(file->private_data);
1248 	return 0;
1249 }
1250 
1251 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1252 				 size_t nbytes, loff_t *ppos)
1253 {
1254 	return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1255 				       i_size_read(file->f_mapping->host));
1256 }
1257 #else
1258 static int o2hb_debug_open(struct inode *inode, struct file *file)
1259 {
1260 	return 0;
1261 }
1262 static int o2hb_debug_release(struct inode *inode, struct file *file)
1263 {
1264 	return 0;
1265 }
1266 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1267 			       size_t nbytes, loff_t *ppos)
1268 {
1269 	return 0;
1270 }
1271 #endif  /* CONFIG_DEBUG_FS */
1272 
1273 static const struct file_operations o2hb_debug_fops = {
1274 	.open =		o2hb_debug_open,
1275 	.release =	o2hb_debug_release,
1276 	.read =		o2hb_debug_read,
1277 	.llseek =	generic_file_llseek,
1278 };
1279 
1280 void o2hb_exit(void)
1281 {
1282 	kfree(o2hb_db_livenodes);
1283 	kfree(o2hb_db_liveregions);
1284 	kfree(o2hb_db_quorumregions);
1285 	kfree(o2hb_db_failedregions);
1286 	debugfs_remove(o2hb_debug_failedregions);
1287 	debugfs_remove(o2hb_debug_quorumregions);
1288 	debugfs_remove(o2hb_debug_liveregions);
1289 	debugfs_remove(o2hb_debug_livenodes);
1290 	debugfs_remove(o2hb_debug_dir);
1291 }
1292 
1293 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1294 					struct o2hb_debug_buf **db, int db_len,
1295 					int type, int size, int len, void *data)
1296 {
1297 	*db = kmalloc(db_len, GFP_KERNEL);
1298 	if (!*db)
1299 		return NULL;
1300 
1301 	(*db)->db_type = type;
1302 	(*db)->db_size = size;
1303 	(*db)->db_len = len;
1304 	(*db)->db_data = data;
1305 
1306 	return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1307 				   &o2hb_debug_fops);
1308 }
1309 
1310 static int o2hb_debug_init(void)
1311 {
1312 	int ret = -ENOMEM;
1313 
1314 	o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1315 	if (IS_ERR_OR_NULL(o2hb_debug_dir)) {
1316 		ret = o2hb_debug_dir ?
1317 			PTR_ERR(o2hb_debug_dir) : -ENOMEM;
1318 		mlog_errno(ret);
1319 		goto bail;
1320 	}
1321 
1322 	o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1323 						 o2hb_debug_dir,
1324 						 &o2hb_db_livenodes,
1325 						 sizeof(*o2hb_db_livenodes),
1326 						 O2HB_DB_TYPE_LIVENODES,
1327 						 sizeof(o2hb_live_node_bitmap),
1328 						 O2NM_MAX_NODES,
1329 						 o2hb_live_node_bitmap);
1330 	if (IS_ERR_OR_NULL(o2hb_debug_livenodes)) {
1331 		ret = o2hb_debug_livenodes ?
1332 			PTR_ERR(o2hb_debug_livenodes) : -ENOMEM;
1333 		mlog_errno(ret);
1334 		goto bail;
1335 	}
1336 
1337 	o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1338 						   o2hb_debug_dir,
1339 						   &o2hb_db_liveregions,
1340 						   sizeof(*o2hb_db_liveregions),
1341 						   O2HB_DB_TYPE_LIVEREGIONS,
1342 						   sizeof(o2hb_live_region_bitmap),
1343 						   O2NM_MAX_REGIONS,
1344 						   o2hb_live_region_bitmap);
1345 	if (IS_ERR_OR_NULL(o2hb_debug_liveregions)) {
1346 		ret = o2hb_debug_liveregions ?
1347 			PTR_ERR(o2hb_debug_liveregions) : -ENOMEM;
1348 		mlog_errno(ret);
1349 		goto bail;
1350 	}
1351 
1352 	o2hb_debug_quorumregions =
1353 			o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1354 					  o2hb_debug_dir,
1355 					  &o2hb_db_quorumregions,
1356 					  sizeof(*o2hb_db_quorumregions),
1357 					  O2HB_DB_TYPE_QUORUMREGIONS,
1358 					  sizeof(o2hb_quorum_region_bitmap),
1359 					  O2NM_MAX_REGIONS,
1360 					  o2hb_quorum_region_bitmap);
1361 	if (IS_ERR_OR_NULL(o2hb_debug_quorumregions)) {
1362 		ret = o2hb_debug_quorumregions ?
1363 			PTR_ERR(o2hb_debug_quorumregions) : -ENOMEM;
1364 		mlog_errno(ret);
1365 		goto bail;
1366 	}
1367 
1368 	o2hb_debug_failedregions =
1369 			o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1370 					  o2hb_debug_dir,
1371 					  &o2hb_db_failedregions,
1372 					  sizeof(*o2hb_db_failedregions),
1373 					  O2HB_DB_TYPE_FAILEDREGIONS,
1374 					  sizeof(o2hb_failed_region_bitmap),
1375 					  O2NM_MAX_REGIONS,
1376 					  o2hb_failed_region_bitmap);
1377 	if (IS_ERR_OR_NULL(o2hb_debug_failedregions)) {
1378 		ret = o2hb_debug_failedregions ?
1379 			PTR_ERR(o2hb_debug_failedregions) : -ENOMEM;
1380 		mlog_errno(ret);
1381 		goto bail;
1382 	}
1383 
1384 	ret = 0;
1385 bail:
1386 	if (ret)
1387 		o2hb_exit();
1388 
1389 	return ret;
1390 }
1391 
1392 int o2hb_init(void)
1393 {
1394 	int i;
1395 
1396 	for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1397 		INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1398 
1399 	for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1400 		INIT_LIST_HEAD(&o2hb_live_slots[i]);
1401 
1402 	INIT_LIST_HEAD(&o2hb_node_events);
1403 
1404 	memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1405 	memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1406 	memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1407 	memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1408 	memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1409 
1410 	o2hb_dependent_users = 0;
1411 
1412 	return o2hb_debug_init();
1413 }
1414 
1415 /* if we're already in a callback then we're already serialized by the sem */
1416 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1417 					     unsigned bytes)
1418 {
1419 	BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1420 
1421 	memcpy(map, &o2hb_live_node_bitmap, bytes);
1422 }
1423 
1424 /*
1425  * get a map of all nodes that are heartbeating in any regions
1426  */
1427 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1428 {
1429 	/* callers want to serialize this map and callbacks so that they
1430 	 * can trust that they don't miss nodes coming to the party */
1431 	down_read(&o2hb_callback_sem);
1432 	spin_lock(&o2hb_live_lock);
1433 	o2hb_fill_node_map_from_callback(map, bytes);
1434 	spin_unlock(&o2hb_live_lock);
1435 	up_read(&o2hb_callback_sem);
1436 }
1437 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1438 
1439 /*
1440  * heartbeat configfs bits.  The heartbeat set is a default set under
1441  * the cluster set in nodemanager.c.
1442  */
1443 
1444 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1445 {
1446 	return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1447 }
1448 
1449 /* drop_item only drops its ref after killing the thread, nothing should
1450  * be using the region anymore.  this has to clean up any state that
1451  * attributes might have built up. */
1452 static void o2hb_region_release(struct config_item *item)
1453 {
1454 	int i;
1455 	struct page *page;
1456 	struct o2hb_region *reg = to_o2hb_region(item);
1457 
1458 	mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1459 
1460 	kfree(reg->hr_tmp_block);
1461 
1462 	if (reg->hr_slot_data) {
1463 		for (i = 0; i < reg->hr_num_pages; i++) {
1464 			page = reg->hr_slot_data[i];
1465 			if (page)
1466 				__free_page(page);
1467 		}
1468 		kfree(reg->hr_slot_data);
1469 	}
1470 
1471 	if (reg->hr_bdev)
1472 		blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1473 
1474 	kfree(reg->hr_slots);
1475 
1476 	kfree(reg->hr_db_regnum);
1477 	kfree(reg->hr_db_livenodes);
1478 	debugfs_remove(reg->hr_debug_livenodes);
1479 	debugfs_remove(reg->hr_debug_regnum);
1480 	debugfs_remove(reg->hr_debug_elapsed_time);
1481 	debugfs_remove(reg->hr_debug_pinned);
1482 	debugfs_remove(reg->hr_debug_dir);
1483 
1484 	spin_lock(&o2hb_live_lock);
1485 	list_del(&reg->hr_all_item);
1486 	spin_unlock(&o2hb_live_lock);
1487 
1488 	kfree(reg);
1489 }
1490 
1491 static int o2hb_read_block_input(struct o2hb_region *reg,
1492 				 const char *page,
1493 				 size_t count,
1494 				 unsigned long *ret_bytes,
1495 				 unsigned int *ret_bits)
1496 {
1497 	unsigned long bytes;
1498 	char *p = (char *)page;
1499 
1500 	bytes = simple_strtoul(p, &p, 0);
1501 	if (!p || (*p && (*p != '\n')))
1502 		return -EINVAL;
1503 
1504 	/* Heartbeat and fs min / max block sizes are the same. */
1505 	if (bytes > 4096 || bytes < 512)
1506 		return -ERANGE;
1507 	if (hweight16(bytes) != 1)
1508 		return -EINVAL;
1509 
1510 	if (ret_bytes)
1511 		*ret_bytes = bytes;
1512 	if (ret_bits)
1513 		*ret_bits = ffs(bytes) - 1;
1514 
1515 	return 0;
1516 }
1517 
1518 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1519 					    char *page)
1520 {
1521 	return sprintf(page, "%u\n", reg->hr_block_bytes);
1522 }
1523 
1524 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1525 					     const char *page,
1526 					     size_t count)
1527 {
1528 	int status;
1529 	unsigned long block_bytes;
1530 	unsigned int block_bits;
1531 
1532 	if (reg->hr_bdev)
1533 		return -EINVAL;
1534 
1535 	status = o2hb_read_block_input(reg, page, count,
1536 				       &block_bytes, &block_bits);
1537 	if (status)
1538 		return status;
1539 
1540 	reg->hr_block_bytes = (unsigned int)block_bytes;
1541 	reg->hr_block_bits = block_bits;
1542 
1543 	return count;
1544 }
1545 
1546 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1547 					    char *page)
1548 {
1549 	return sprintf(page, "%llu\n", reg->hr_start_block);
1550 }
1551 
1552 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1553 					     const char *page,
1554 					     size_t count)
1555 {
1556 	unsigned long long tmp;
1557 	char *p = (char *)page;
1558 
1559 	if (reg->hr_bdev)
1560 		return -EINVAL;
1561 
1562 	tmp = simple_strtoull(p, &p, 0);
1563 	if (!p || (*p && (*p != '\n')))
1564 		return -EINVAL;
1565 
1566 	reg->hr_start_block = tmp;
1567 
1568 	return count;
1569 }
1570 
1571 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1572 				       char *page)
1573 {
1574 	return sprintf(page, "%d\n", reg->hr_blocks);
1575 }
1576 
1577 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1578 					const char *page,
1579 					size_t count)
1580 {
1581 	unsigned long tmp;
1582 	char *p = (char *)page;
1583 
1584 	if (reg->hr_bdev)
1585 		return -EINVAL;
1586 
1587 	tmp = simple_strtoul(p, &p, 0);
1588 	if (!p || (*p && (*p != '\n')))
1589 		return -EINVAL;
1590 
1591 	if (tmp > O2NM_MAX_NODES || tmp == 0)
1592 		return -ERANGE;
1593 
1594 	reg->hr_blocks = (unsigned int)tmp;
1595 
1596 	return count;
1597 }
1598 
1599 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1600 				    char *page)
1601 {
1602 	unsigned int ret = 0;
1603 
1604 	if (reg->hr_bdev)
1605 		ret = sprintf(page, "%s\n", reg->hr_dev_name);
1606 
1607 	return ret;
1608 }
1609 
1610 static void o2hb_init_region_params(struct o2hb_region *reg)
1611 {
1612 	reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1613 	reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1614 
1615 	mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1616 	     reg->hr_start_block, reg->hr_blocks);
1617 	mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1618 	     reg->hr_block_bytes, reg->hr_block_bits);
1619 	mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1620 	mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1621 }
1622 
1623 static int o2hb_map_slot_data(struct o2hb_region *reg)
1624 {
1625 	int i, j;
1626 	unsigned int last_slot;
1627 	unsigned int spp = reg->hr_slots_per_page;
1628 	struct page *page;
1629 	char *raw;
1630 	struct o2hb_disk_slot *slot;
1631 
1632 	reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1633 	if (reg->hr_tmp_block == NULL) {
1634 		mlog_errno(-ENOMEM);
1635 		return -ENOMEM;
1636 	}
1637 
1638 	reg->hr_slots = kcalloc(reg->hr_blocks,
1639 				sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1640 	if (reg->hr_slots == NULL) {
1641 		mlog_errno(-ENOMEM);
1642 		return -ENOMEM;
1643 	}
1644 
1645 	for(i = 0; i < reg->hr_blocks; i++) {
1646 		slot = &reg->hr_slots[i];
1647 		slot->ds_node_num = i;
1648 		INIT_LIST_HEAD(&slot->ds_live_item);
1649 		slot->ds_raw_block = NULL;
1650 	}
1651 
1652 	reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1653 	mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1654 			   "at %u blocks per page\n",
1655 	     reg->hr_num_pages, reg->hr_blocks, spp);
1656 
1657 	reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1658 				    GFP_KERNEL);
1659 	if (!reg->hr_slot_data) {
1660 		mlog_errno(-ENOMEM);
1661 		return -ENOMEM;
1662 	}
1663 
1664 	for(i = 0; i < reg->hr_num_pages; i++) {
1665 		page = alloc_page(GFP_KERNEL);
1666 		if (!page) {
1667 			mlog_errno(-ENOMEM);
1668 			return -ENOMEM;
1669 		}
1670 
1671 		reg->hr_slot_data[i] = page;
1672 
1673 		last_slot = i * spp;
1674 		raw = page_address(page);
1675 		for (j = 0;
1676 		     (j < spp) && ((j + last_slot) < reg->hr_blocks);
1677 		     j++) {
1678 			BUG_ON((j + last_slot) >= reg->hr_blocks);
1679 
1680 			slot = &reg->hr_slots[j + last_slot];
1681 			slot->ds_raw_block =
1682 				(struct o2hb_disk_heartbeat_block *) raw;
1683 
1684 			raw += reg->hr_block_bytes;
1685 		}
1686 	}
1687 
1688 	return 0;
1689 }
1690 
1691 /* Read in all the slots available and populate the tracking
1692  * structures so that we can start with a baseline idea of what's
1693  * there. */
1694 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1695 {
1696 	int ret, i;
1697 	struct o2hb_disk_slot *slot;
1698 	struct o2hb_disk_heartbeat_block *hb_block;
1699 
1700 	ret = o2hb_read_slots(reg, reg->hr_blocks);
1701 	if (ret) {
1702 		mlog_errno(ret);
1703 		goto out;
1704 	}
1705 
1706 	/* We only want to get an idea of the values initially in each
1707 	 * slot, so we do no verification - o2hb_check_slot will
1708 	 * actually determine if each configured slot is valid and
1709 	 * whether any values have changed. */
1710 	for(i = 0; i < reg->hr_blocks; i++) {
1711 		slot = &reg->hr_slots[i];
1712 		hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1713 
1714 		/* Only fill the values that o2hb_check_slot uses to
1715 		 * determine changing slots */
1716 		slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1717 		slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1718 	}
1719 
1720 out:
1721 	return ret;
1722 }
1723 
1724 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1725 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1726 				     const char *page,
1727 				     size_t count)
1728 {
1729 	struct task_struct *hb_task;
1730 	long fd;
1731 	int sectsize;
1732 	char *p = (char *)page;
1733 	struct fd f;
1734 	struct inode *inode;
1735 	ssize_t ret = -EINVAL;
1736 	int live_threshold;
1737 
1738 	if (reg->hr_bdev)
1739 		goto out;
1740 
1741 	/* We can't heartbeat without having had our node number
1742 	 * configured yet. */
1743 	if (o2nm_this_node() == O2NM_MAX_NODES)
1744 		goto out;
1745 
1746 	fd = simple_strtol(p, &p, 0);
1747 	if (!p || (*p && (*p != '\n')))
1748 		goto out;
1749 
1750 	if (fd < 0 || fd >= INT_MAX)
1751 		goto out;
1752 
1753 	f = fdget(fd);
1754 	if (f.file == NULL)
1755 		goto out;
1756 
1757 	if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1758 	    reg->hr_block_bytes == 0)
1759 		goto out2;
1760 
1761 	inode = igrab(f.file->f_mapping->host);
1762 	if (inode == NULL)
1763 		goto out2;
1764 
1765 	if (!S_ISBLK(inode->i_mode))
1766 		goto out3;
1767 
1768 	reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
1769 	ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
1770 	if (ret) {
1771 		reg->hr_bdev = NULL;
1772 		goto out3;
1773 	}
1774 	inode = NULL;
1775 
1776 	bdevname(reg->hr_bdev, reg->hr_dev_name);
1777 
1778 	sectsize = bdev_logical_block_size(reg->hr_bdev);
1779 	if (sectsize != reg->hr_block_bytes) {
1780 		mlog(ML_ERROR,
1781 		     "blocksize %u incorrect for device, expected %d",
1782 		     reg->hr_block_bytes, sectsize);
1783 		ret = -EINVAL;
1784 		goto out3;
1785 	}
1786 
1787 	o2hb_init_region_params(reg);
1788 
1789 	/* Generation of zero is invalid */
1790 	do {
1791 		get_random_bytes(&reg->hr_generation,
1792 				 sizeof(reg->hr_generation));
1793 	} while (reg->hr_generation == 0);
1794 
1795 	ret = o2hb_map_slot_data(reg);
1796 	if (ret) {
1797 		mlog_errno(ret);
1798 		goto out3;
1799 	}
1800 
1801 	ret = o2hb_populate_slot_data(reg);
1802 	if (ret) {
1803 		mlog_errno(ret);
1804 		goto out3;
1805 	}
1806 
1807 	INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1808 
1809 	/*
1810 	 * A node is considered live after it has beat LIVE_THRESHOLD
1811 	 * times.  We're not steady until we've given them a chance
1812 	 * _after_ our first read.
1813 	 * The default threshold is bare minimum so as to limit the delay
1814 	 * during mounts. For global heartbeat, the threshold doubled for the
1815 	 * first region.
1816 	 */
1817 	live_threshold = O2HB_LIVE_THRESHOLD;
1818 	if (o2hb_global_heartbeat_active()) {
1819 		spin_lock(&o2hb_live_lock);
1820 		if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1821 			live_threshold <<= 1;
1822 		spin_unlock(&o2hb_live_lock);
1823 	}
1824 	++live_threshold;
1825 	atomic_set(&reg->hr_steady_iterations, live_threshold);
1826 	/* unsteady_iterations is double the steady_iterations */
1827 	atomic_set(&reg->hr_unsteady_iterations, (live_threshold << 1));
1828 
1829 	hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1830 			      reg->hr_item.ci_name);
1831 	if (IS_ERR(hb_task)) {
1832 		ret = PTR_ERR(hb_task);
1833 		mlog_errno(ret);
1834 		goto out3;
1835 	}
1836 
1837 	spin_lock(&o2hb_live_lock);
1838 	reg->hr_task = hb_task;
1839 	spin_unlock(&o2hb_live_lock);
1840 
1841 	ret = wait_event_interruptible(o2hb_steady_queue,
1842 				atomic_read(&reg->hr_steady_iterations) == 0);
1843 	if (ret) {
1844 		atomic_set(&reg->hr_steady_iterations, 0);
1845 		reg->hr_aborted_start = 1;
1846 	}
1847 
1848 	if (reg->hr_aborted_start) {
1849 		ret = -EIO;
1850 		goto out3;
1851 	}
1852 
1853 	/* Ok, we were woken.  Make sure it wasn't by drop_item() */
1854 	spin_lock(&o2hb_live_lock);
1855 	hb_task = reg->hr_task;
1856 	if (o2hb_global_heartbeat_active())
1857 		set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1858 	spin_unlock(&o2hb_live_lock);
1859 
1860 	if (hb_task)
1861 		ret = count;
1862 	else
1863 		ret = -EIO;
1864 
1865 	if (hb_task && o2hb_global_heartbeat_active())
1866 		printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1867 		       config_item_name(&reg->hr_item), reg->hr_dev_name);
1868 
1869 out3:
1870 	iput(inode);
1871 out2:
1872 	fdput(f);
1873 out:
1874 	if (ret < 0) {
1875 		if (reg->hr_bdev) {
1876 			blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1877 			reg->hr_bdev = NULL;
1878 		}
1879 	}
1880 	return ret;
1881 }
1882 
1883 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1884                                       char *page)
1885 {
1886 	pid_t pid = 0;
1887 
1888 	spin_lock(&o2hb_live_lock);
1889 	if (reg->hr_task)
1890 		pid = task_pid_nr(reg->hr_task);
1891 	spin_unlock(&o2hb_live_lock);
1892 
1893 	if (!pid)
1894 		return 0;
1895 
1896 	return sprintf(page, "%u\n", pid);
1897 }
1898 
1899 struct o2hb_region_attribute {
1900 	struct configfs_attribute attr;
1901 	ssize_t (*show)(struct o2hb_region *, char *);
1902 	ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1903 };
1904 
1905 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1906 	.attr	= { .ca_owner = THIS_MODULE,
1907 		    .ca_name = "block_bytes",
1908 		    .ca_mode = S_IRUGO | S_IWUSR },
1909 	.show	= o2hb_region_block_bytes_read,
1910 	.store	= o2hb_region_block_bytes_write,
1911 };
1912 
1913 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1914 	.attr	= { .ca_owner = THIS_MODULE,
1915 		    .ca_name = "start_block",
1916 		    .ca_mode = S_IRUGO | S_IWUSR },
1917 	.show	= o2hb_region_start_block_read,
1918 	.store	= o2hb_region_start_block_write,
1919 };
1920 
1921 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1922 	.attr	= { .ca_owner = THIS_MODULE,
1923 		    .ca_name = "blocks",
1924 		    .ca_mode = S_IRUGO | S_IWUSR },
1925 	.show	= o2hb_region_blocks_read,
1926 	.store	= o2hb_region_blocks_write,
1927 };
1928 
1929 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1930 	.attr	= { .ca_owner = THIS_MODULE,
1931 		    .ca_name = "dev",
1932 		    .ca_mode = S_IRUGO | S_IWUSR },
1933 	.show	= o2hb_region_dev_read,
1934 	.store	= o2hb_region_dev_write,
1935 };
1936 
1937 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1938        .attr   = { .ca_owner = THIS_MODULE,
1939                    .ca_name = "pid",
1940                    .ca_mode = S_IRUGO | S_IRUSR },
1941        .show   = o2hb_region_pid_read,
1942 };
1943 
1944 static struct configfs_attribute *o2hb_region_attrs[] = {
1945 	&o2hb_region_attr_block_bytes.attr,
1946 	&o2hb_region_attr_start_block.attr,
1947 	&o2hb_region_attr_blocks.attr,
1948 	&o2hb_region_attr_dev.attr,
1949 	&o2hb_region_attr_pid.attr,
1950 	NULL,
1951 };
1952 
1953 static ssize_t o2hb_region_show(struct config_item *item,
1954 				struct configfs_attribute *attr,
1955 				char *page)
1956 {
1957 	struct o2hb_region *reg = to_o2hb_region(item);
1958 	struct o2hb_region_attribute *o2hb_region_attr =
1959 		container_of(attr, struct o2hb_region_attribute, attr);
1960 	ssize_t ret = 0;
1961 
1962 	if (o2hb_region_attr->show)
1963 		ret = o2hb_region_attr->show(reg, page);
1964 	return ret;
1965 }
1966 
1967 static ssize_t o2hb_region_store(struct config_item *item,
1968 				 struct configfs_attribute *attr,
1969 				 const char *page, size_t count)
1970 {
1971 	struct o2hb_region *reg = to_o2hb_region(item);
1972 	struct o2hb_region_attribute *o2hb_region_attr =
1973 		container_of(attr, struct o2hb_region_attribute, attr);
1974 	ssize_t ret = -EINVAL;
1975 
1976 	if (o2hb_region_attr->store)
1977 		ret = o2hb_region_attr->store(reg, page, count);
1978 	return ret;
1979 }
1980 
1981 static struct configfs_item_operations o2hb_region_item_ops = {
1982 	.release		= o2hb_region_release,
1983 	.show_attribute		= o2hb_region_show,
1984 	.store_attribute	= o2hb_region_store,
1985 };
1986 
1987 static struct config_item_type o2hb_region_type = {
1988 	.ct_item_ops	= &o2hb_region_item_ops,
1989 	.ct_attrs	= o2hb_region_attrs,
1990 	.ct_owner	= THIS_MODULE,
1991 };
1992 
1993 /* heartbeat set */
1994 
1995 struct o2hb_heartbeat_group {
1996 	struct config_group hs_group;
1997 	/* some stuff? */
1998 };
1999 
2000 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
2001 {
2002 	return group ?
2003 		container_of(group, struct o2hb_heartbeat_group, hs_group)
2004 		: NULL;
2005 }
2006 
2007 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
2008 {
2009 	int ret = -ENOMEM;
2010 
2011 	reg->hr_debug_dir =
2012 		debugfs_create_dir(config_item_name(&reg->hr_item), dir);
2013 	if (IS_ERR_OR_NULL(reg->hr_debug_dir)) {
2014 		ret = reg->hr_debug_dir ? PTR_ERR(reg->hr_debug_dir) : -ENOMEM;
2015 		mlog_errno(ret);
2016 		goto bail;
2017 	}
2018 
2019 	reg->hr_debug_livenodes =
2020 			o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2021 					  reg->hr_debug_dir,
2022 					  &(reg->hr_db_livenodes),
2023 					  sizeof(*(reg->hr_db_livenodes)),
2024 					  O2HB_DB_TYPE_REGION_LIVENODES,
2025 					  sizeof(reg->hr_live_node_bitmap),
2026 					  O2NM_MAX_NODES, reg);
2027 	if (IS_ERR_OR_NULL(reg->hr_debug_livenodes)) {
2028 		ret = reg->hr_debug_livenodes ?
2029 			PTR_ERR(reg->hr_debug_livenodes) : -ENOMEM;
2030 		mlog_errno(ret);
2031 		goto bail;
2032 	}
2033 
2034 	reg->hr_debug_regnum =
2035 			o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2036 					  reg->hr_debug_dir,
2037 					  &(reg->hr_db_regnum),
2038 					  sizeof(*(reg->hr_db_regnum)),
2039 					  O2HB_DB_TYPE_REGION_NUMBER,
2040 					  0, O2NM_MAX_NODES, reg);
2041 	if (IS_ERR_OR_NULL(reg->hr_debug_regnum)) {
2042 		ret = reg->hr_debug_regnum ?
2043 			PTR_ERR(reg->hr_debug_regnum) : -ENOMEM;
2044 		mlog_errno(ret);
2045 		goto bail;
2046 	}
2047 
2048 	reg->hr_debug_elapsed_time =
2049 			o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2050 					  reg->hr_debug_dir,
2051 					  &(reg->hr_db_elapsed_time),
2052 					  sizeof(*(reg->hr_db_elapsed_time)),
2053 					  O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2054 					  0, 0, reg);
2055 	if (IS_ERR_OR_NULL(reg->hr_debug_elapsed_time)) {
2056 		ret = reg->hr_debug_elapsed_time ?
2057 			PTR_ERR(reg->hr_debug_elapsed_time) : -ENOMEM;
2058 		mlog_errno(ret);
2059 		goto bail;
2060 	}
2061 
2062 	reg->hr_debug_pinned =
2063 			o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2064 					  reg->hr_debug_dir,
2065 					  &(reg->hr_db_pinned),
2066 					  sizeof(*(reg->hr_db_pinned)),
2067 					  O2HB_DB_TYPE_REGION_PINNED,
2068 					  0, 0, reg);
2069 	if (IS_ERR_OR_NULL(reg->hr_debug_pinned)) {
2070 		ret = reg->hr_debug_pinned ?
2071 			PTR_ERR(reg->hr_debug_pinned) : -ENOMEM;
2072 		mlog_errno(ret);
2073 		goto bail;
2074 	}
2075 
2076 	return 0;
2077 bail:
2078 	debugfs_remove_recursive(reg->hr_debug_dir);
2079 	return ret;
2080 }
2081 
2082 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2083 							  const char *name)
2084 {
2085 	struct o2hb_region *reg = NULL;
2086 	int ret;
2087 
2088 	reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2089 	if (reg == NULL)
2090 		return ERR_PTR(-ENOMEM);
2091 
2092 	if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2093 		ret = -ENAMETOOLONG;
2094 		goto free;
2095 	}
2096 
2097 	spin_lock(&o2hb_live_lock);
2098 	reg->hr_region_num = 0;
2099 	if (o2hb_global_heartbeat_active()) {
2100 		reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2101 							 O2NM_MAX_REGIONS);
2102 		if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2103 			spin_unlock(&o2hb_live_lock);
2104 			ret = -EFBIG;
2105 			goto free;
2106 		}
2107 		set_bit(reg->hr_region_num, o2hb_region_bitmap);
2108 	}
2109 	list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2110 	spin_unlock(&o2hb_live_lock);
2111 
2112 	config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2113 
2114 	ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2115 	if (ret) {
2116 		config_item_put(&reg->hr_item);
2117 		goto free;
2118 	}
2119 
2120 	return &reg->hr_item;
2121 free:
2122 	kfree(reg);
2123 	return ERR_PTR(ret);
2124 }
2125 
2126 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2127 					   struct config_item *item)
2128 {
2129 	struct task_struct *hb_task;
2130 	struct o2hb_region *reg = to_o2hb_region(item);
2131 	int quorum_region = 0;
2132 
2133 	/* stop the thread when the user removes the region dir */
2134 	spin_lock(&o2hb_live_lock);
2135 	hb_task = reg->hr_task;
2136 	reg->hr_task = NULL;
2137 	reg->hr_item_dropped = 1;
2138 	spin_unlock(&o2hb_live_lock);
2139 
2140 	if (hb_task)
2141 		kthread_stop(hb_task);
2142 
2143 	if (o2hb_global_heartbeat_active()) {
2144 		spin_lock(&o2hb_live_lock);
2145 		clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2146 		clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2147 		if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2148 			quorum_region = 1;
2149 		clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2150 		spin_unlock(&o2hb_live_lock);
2151 		printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2152 		       ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2153 			"stopped" : "start aborted"), config_item_name(item),
2154 		       reg->hr_dev_name);
2155 	}
2156 
2157 	/*
2158 	 * If we're racing a dev_write(), we need to wake them.  They will
2159 	 * check reg->hr_task
2160 	 */
2161 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
2162 		reg->hr_aborted_start = 1;
2163 		atomic_set(&reg->hr_steady_iterations, 0);
2164 		wake_up(&o2hb_steady_queue);
2165 	}
2166 
2167 	config_item_put(item);
2168 
2169 	if (!o2hb_global_heartbeat_active() || !quorum_region)
2170 		return;
2171 
2172 	/*
2173 	 * If global heartbeat active and there are dependent users,
2174 	 * pin all regions if quorum region count <= CUT_OFF
2175 	 */
2176 	spin_lock(&o2hb_live_lock);
2177 
2178 	if (!o2hb_dependent_users)
2179 		goto unlock;
2180 
2181 	if (bitmap_weight(o2hb_quorum_region_bitmap,
2182 			   O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2183 		o2hb_region_pin(NULL);
2184 
2185 unlock:
2186 	spin_unlock(&o2hb_live_lock);
2187 }
2188 
2189 struct o2hb_heartbeat_group_attribute {
2190 	struct configfs_attribute attr;
2191 	ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2192 	ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2193 };
2194 
2195 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2196 					 struct configfs_attribute *attr,
2197 					 char *page)
2198 {
2199 	struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2200 	struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2201 		container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2202 	ssize_t ret = 0;
2203 
2204 	if (o2hb_heartbeat_group_attr->show)
2205 		ret = o2hb_heartbeat_group_attr->show(reg, page);
2206 	return ret;
2207 }
2208 
2209 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2210 					  struct configfs_attribute *attr,
2211 					  const char *page, size_t count)
2212 {
2213 	struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2214 	struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2215 		container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2216 	ssize_t ret = -EINVAL;
2217 
2218 	if (o2hb_heartbeat_group_attr->store)
2219 		ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2220 	return ret;
2221 }
2222 
2223 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2224 						     char *page)
2225 {
2226 	return sprintf(page, "%u\n", o2hb_dead_threshold);
2227 }
2228 
2229 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2230 						    const char *page,
2231 						    size_t count)
2232 {
2233 	unsigned long tmp;
2234 	char *p = (char *)page;
2235 
2236 	tmp = simple_strtoul(p, &p, 10);
2237 	if (!p || (*p && (*p != '\n')))
2238                 return -EINVAL;
2239 
2240 	/* this will validate ranges for us. */
2241 	o2hb_dead_threshold_set((unsigned int) tmp);
2242 
2243 	return count;
2244 }
2245 
2246 static
2247 ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2248 				       char *page)
2249 {
2250 	return sprintf(page, "%s\n",
2251 		       o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2252 }
2253 
2254 static
2255 ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2256 					const char *page, size_t count)
2257 {
2258 	unsigned int i;
2259 	int ret;
2260 	size_t len;
2261 
2262 	len = (page[count - 1] == '\n') ? count - 1 : count;
2263 	if (!len)
2264 		return -EINVAL;
2265 
2266 	for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2267 		if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
2268 			continue;
2269 
2270 		ret = o2hb_global_heartbeat_mode_set(i);
2271 		if (!ret)
2272 			printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2273 			       o2hb_heartbeat_mode_desc[i]);
2274 		return count;
2275 	}
2276 
2277 	return -EINVAL;
2278 
2279 }
2280 
2281 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2282 	.attr	= { .ca_owner = THIS_MODULE,
2283 		    .ca_name = "dead_threshold",
2284 		    .ca_mode = S_IRUGO | S_IWUSR },
2285 	.show	= o2hb_heartbeat_group_threshold_show,
2286 	.store	= o2hb_heartbeat_group_threshold_store,
2287 };
2288 
2289 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2290 	.attr   = { .ca_owner = THIS_MODULE,
2291 		.ca_name = "mode",
2292 		.ca_mode = S_IRUGO | S_IWUSR },
2293 	.show   = o2hb_heartbeat_group_mode_show,
2294 	.store  = o2hb_heartbeat_group_mode_store,
2295 };
2296 
2297 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2298 	&o2hb_heartbeat_group_attr_threshold.attr,
2299 	&o2hb_heartbeat_group_attr_mode.attr,
2300 	NULL,
2301 };
2302 
2303 static struct configfs_item_operations o2hb_heartbeat_group_item_ops = {
2304 	.show_attribute		= o2hb_heartbeat_group_show,
2305 	.store_attribute	= o2hb_heartbeat_group_store,
2306 };
2307 
2308 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2309 	.make_item	= o2hb_heartbeat_group_make_item,
2310 	.drop_item	= o2hb_heartbeat_group_drop_item,
2311 };
2312 
2313 static struct config_item_type o2hb_heartbeat_group_type = {
2314 	.ct_group_ops	= &o2hb_heartbeat_group_group_ops,
2315 	.ct_item_ops	= &o2hb_heartbeat_group_item_ops,
2316 	.ct_attrs	= o2hb_heartbeat_group_attrs,
2317 	.ct_owner	= THIS_MODULE,
2318 };
2319 
2320 /* this is just here to avoid touching group in heartbeat.h which the
2321  * entire damn world #includes */
2322 struct config_group *o2hb_alloc_hb_set(void)
2323 {
2324 	struct o2hb_heartbeat_group *hs = NULL;
2325 	struct config_group *ret = NULL;
2326 
2327 	hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2328 	if (hs == NULL)
2329 		goto out;
2330 
2331 	config_group_init_type_name(&hs->hs_group, "heartbeat",
2332 				    &o2hb_heartbeat_group_type);
2333 
2334 	ret = &hs->hs_group;
2335 out:
2336 	if (ret == NULL)
2337 		kfree(hs);
2338 	return ret;
2339 }
2340 
2341 void o2hb_free_hb_set(struct config_group *group)
2342 {
2343 	struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2344 	kfree(hs);
2345 }
2346 
2347 /* hb callback registration and issuing */
2348 
2349 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2350 {
2351 	if (type == O2HB_NUM_CB)
2352 		return ERR_PTR(-EINVAL);
2353 
2354 	return &o2hb_callbacks[type];
2355 }
2356 
2357 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2358 			 enum o2hb_callback_type type,
2359 			 o2hb_cb_func *func,
2360 			 void *data,
2361 			 int priority)
2362 {
2363 	INIT_LIST_HEAD(&hc->hc_item);
2364 	hc->hc_func = func;
2365 	hc->hc_data = data;
2366 	hc->hc_priority = priority;
2367 	hc->hc_type = type;
2368 	hc->hc_magic = O2HB_CB_MAGIC;
2369 }
2370 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2371 
2372 /*
2373  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2374  * In global heartbeat mode, region_uuid passed is NULL.
2375  *
2376  * In local, we only pin the matching region. In global we pin all the active
2377  * regions.
2378  */
2379 static int o2hb_region_pin(const char *region_uuid)
2380 {
2381 	int ret = 0, found = 0;
2382 	struct o2hb_region *reg;
2383 	char *uuid;
2384 
2385 	assert_spin_locked(&o2hb_live_lock);
2386 
2387 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2388 		if (reg->hr_item_dropped)
2389 			continue;
2390 
2391 		uuid = config_item_name(&reg->hr_item);
2392 
2393 		/* local heartbeat */
2394 		if (region_uuid) {
2395 			if (strcmp(region_uuid, uuid))
2396 				continue;
2397 			found = 1;
2398 		}
2399 
2400 		if (reg->hr_item_pinned || reg->hr_item_dropped)
2401 			goto skip_pin;
2402 
2403 		/* Ignore ENOENT only for local hb (userdlm domain) */
2404 		ret = o2nm_depend_item(&reg->hr_item);
2405 		if (!ret) {
2406 			mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2407 			reg->hr_item_pinned = 1;
2408 		} else {
2409 			if (ret == -ENOENT && found)
2410 				ret = 0;
2411 			else {
2412 				mlog(ML_ERROR, "Pin region %s fails with %d\n",
2413 				     uuid, ret);
2414 				break;
2415 			}
2416 		}
2417 skip_pin:
2418 		if (found)
2419 			break;
2420 	}
2421 
2422 	return ret;
2423 }
2424 
2425 /*
2426  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2427  * In global heartbeat mode, region_uuid passed is NULL.
2428  *
2429  * In local, we only unpin the matching region. In global we unpin all the
2430  * active regions.
2431  */
2432 static void o2hb_region_unpin(const char *region_uuid)
2433 {
2434 	struct o2hb_region *reg;
2435 	char *uuid;
2436 	int found = 0;
2437 
2438 	assert_spin_locked(&o2hb_live_lock);
2439 
2440 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2441 		if (reg->hr_item_dropped)
2442 			continue;
2443 
2444 		uuid = config_item_name(&reg->hr_item);
2445 		if (region_uuid) {
2446 			if (strcmp(region_uuid, uuid))
2447 				continue;
2448 			found = 1;
2449 		}
2450 
2451 		if (reg->hr_item_pinned) {
2452 			mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2453 			o2nm_undepend_item(&reg->hr_item);
2454 			reg->hr_item_pinned = 0;
2455 		}
2456 		if (found)
2457 			break;
2458 	}
2459 }
2460 
2461 static int o2hb_region_inc_user(const char *region_uuid)
2462 {
2463 	int ret = 0;
2464 
2465 	spin_lock(&o2hb_live_lock);
2466 
2467 	/* local heartbeat */
2468 	if (!o2hb_global_heartbeat_active()) {
2469 	    ret = o2hb_region_pin(region_uuid);
2470 	    goto unlock;
2471 	}
2472 
2473 	/*
2474 	 * if global heartbeat active and this is the first dependent user,
2475 	 * pin all regions if quorum region count <= CUT_OFF
2476 	 */
2477 	o2hb_dependent_users++;
2478 	if (o2hb_dependent_users > 1)
2479 		goto unlock;
2480 
2481 	if (bitmap_weight(o2hb_quorum_region_bitmap,
2482 			   O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2483 		ret = o2hb_region_pin(NULL);
2484 
2485 unlock:
2486 	spin_unlock(&o2hb_live_lock);
2487 	return ret;
2488 }
2489 
2490 void o2hb_region_dec_user(const char *region_uuid)
2491 {
2492 	spin_lock(&o2hb_live_lock);
2493 
2494 	/* local heartbeat */
2495 	if (!o2hb_global_heartbeat_active()) {
2496 	    o2hb_region_unpin(region_uuid);
2497 	    goto unlock;
2498 	}
2499 
2500 	/*
2501 	 * if global heartbeat active and there are no dependent users,
2502 	 * unpin all quorum regions
2503 	 */
2504 	o2hb_dependent_users--;
2505 	if (!o2hb_dependent_users)
2506 		o2hb_region_unpin(NULL);
2507 
2508 unlock:
2509 	spin_unlock(&o2hb_live_lock);
2510 }
2511 
2512 int o2hb_register_callback(const char *region_uuid,
2513 			   struct o2hb_callback_func *hc)
2514 {
2515 	struct o2hb_callback_func *f;
2516 	struct o2hb_callback *hbcall;
2517 	int ret;
2518 
2519 	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2520 	BUG_ON(!list_empty(&hc->hc_item));
2521 
2522 	hbcall = hbcall_from_type(hc->hc_type);
2523 	if (IS_ERR(hbcall)) {
2524 		ret = PTR_ERR(hbcall);
2525 		goto out;
2526 	}
2527 
2528 	if (region_uuid) {
2529 		ret = o2hb_region_inc_user(region_uuid);
2530 		if (ret) {
2531 			mlog_errno(ret);
2532 			goto out;
2533 		}
2534 	}
2535 
2536 	down_write(&o2hb_callback_sem);
2537 
2538 	list_for_each_entry(f, &hbcall->list, hc_item) {
2539 		if (hc->hc_priority < f->hc_priority) {
2540 			list_add_tail(&hc->hc_item, &f->hc_item);
2541 			break;
2542 		}
2543 	}
2544 	if (list_empty(&hc->hc_item))
2545 		list_add_tail(&hc->hc_item, &hbcall->list);
2546 
2547 	up_write(&o2hb_callback_sem);
2548 	ret = 0;
2549 out:
2550 	mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2551 	     ret, __builtin_return_address(0), hc);
2552 	return ret;
2553 }
2554 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2555 
2556 void o2hb_unregister_callback(const char *region_uuid,
2557 			      struct o2hb_callback_func *hc)
2558 {
2559 	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2560 
2561 	mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2562 	     __builtin_return_address(0), hc);
2563 
2564 	/* XXX Can this happen _with_ a region reference? */
2565 	if (list_empty(&hc->hc_item))
2566 		return;
2567 
2568 	if (region_uuid)
2569 		o2hb_region_dec_user(region_uuid);
2570 
2571 	down_write(&o2hb_callback_sem);
2572 
2573 	list_del_init(&hc->hc_item);
2574 
2575 	up_write(&o2hb_callback_sem);
2576 }
2577 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2578 
2579 int o2hb_check_node_heartbeating(u8 node_num)
2580 {
2581 	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2582 
2583 	o2hb_fill_node_map(testing_map, sizeof(testing_map));
2584 	if (!test_bit(node_num, testing_map)) {
2585 		mlog(ML_HEARTBEAT,
2586 		     "node (%u) does not have heartbeating enabled.\n",
2587 		     node_num);
2588 		return 0;
2589 	}
2590 
2591 	return 1;
2592 }
2593 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2594 
2595 int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2596 {
2597 	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2598 	unsigned long flags;
2599 
2600 	spin_lock_irqsave(&o2hb_live_lock, flags);
2601 	o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2602 	spin_unlock_irqrestore(&o2hb_live_lock, flags);
2603 	if (!test_bit(node_num, testing_map)) {
2604 		mlog(ML_HEARTBEAT,
2605 		     "node (%u) does not have heartbeating enabled.\n",
2606 		     node_num);
2607 		return 0;
2608 	}
2609 
2610 	return 1;
2611 }
2612 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2613 
2614 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2615 {
2616 	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2617 
2618 	o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2619 	if (!test_bit(node_num, testing_map)) {
2620 		mlog(ML_HEARTBEAT,
2621 		     "node (%u) does not have heartbeating enabled.\n",
2622 		     node_num);
2623 		return 0;
2624 	}
2625 
2626 	return 1;
2627 }
2628 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2629 
2630 /* Makes sure our local node is configured with a node number, and is
2631  * heartbeating. */
2632 int o2hb_check_local_node_heartbeating(void)
2633 {
2634 	u8 node_num;
2635 
2636 	/* if this node was set then we have networking */
2637 	node_num = o2nm_this_node();
2638 	if (node_num == O2NM_MAX_NODES) {
2639 		mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2640 		return 0;
2641 	}
2642 
2643 	return o2hb_check_node_heartbeating(node_num);
2644 }
2645 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2646 
2647 /*
2648  * this is just a hack until we get the plumbing which flips file systems
2649  * read only and drops the hb ref instead of killing the node dead.
2650  */
2651 void o2hb_stop_all_regions(void)
2652 {
2653 	struct o2hb_region *reg;
2654 
2655 	mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2656 
2657 	spin_lock(&o2hb_live_lock);
2658 
2659 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2660 		reg->hr_unclean_stop = 1;
2661 
2662 	spin_unlock(&o2hb_live_lock);
2663 }
2664 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2665 
2666 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2667 {
2668 	struct o2hb_region *reg;
2669 	int numregs = 0;
2670 	char *p;
2671 
2672 	spin_lock(&o2hb_live_lock);
2673 
2674 	p = region_uuids;
2675 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2676 		if (reg->hr_item_dropped)
2677 			continue;
2678 
2679 		mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2680 		if (numregs < max_regions) {
2681 			memcpy(p, config_item_name(&reg->hr_item),
2682 			       O2HB_MAX_REGION_NAME_LEN);
2683 			p += O2HB_MAX_REGION_NAME_LEN;
2684 		}
2685 		numregs++;
2686 	}
2687 
2688 	spin_unlock(&o2hb_live_lock);
2689 
2690 	return numregs;
2691 }
2692 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2693 
2694 int o2hb_global_heartbeat_active(void)
2695 {
2696 	return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2697 }
2698 EXPORT_SYMBOL(o2hb_global_heartbeat_active);
2699