xref: /openbmc/linux/fs/ocfs2/cluster/heartbeat.c (revision 35f752be)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/kernel.h>
7 #include <linux/sched.h>
8 #include <linux/jiffies.h>
9 #include <linux/module.h>
10 #include <linux/fs.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"
25 #include "tcp.h"
26 #include "nodemanager.h"
27 #include "quorum.h"
28 
29 #include "masklog.h"
30 
31 
32 /*
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.
37  */
38 static DECLARE_RWSEM(o2hb_callback_sem);
39 
40 /*
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.
43  */
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);
49 
50 /*
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
55  * 		heartbeat on it.
56  * 	- o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
57  */
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)];
62 
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 {
72 	int db_type;
73 	int db_size;
74 	int db_len;
75 	void *db_data;
76 };
77 
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;
82 
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"
91 
92 static struct dentry *o2hb_debug_dir;
93 
94 static LIST_HEAD(o2hb_all_regions);
95 
96 static struct o2hb_callback {
97 	struct list_head list;
98 } o2hb_callbacks[O2HB_NUM_CB];
99 
100 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
101 
102 enum o2hb_heartbeat_modes {
103 	O2HB_HEARTBEAT_LOCAL		= 0,
104 	O2HB_HEARTBEAT_GLOBAL,
105 	O2HB_HEARTBEAT_NUM_MODES,
106 };
107 
108 static const char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
109 	"local",	/* O2HB_HEARTBEAT_LOCAL */
110 	"global",	/* O2HB_HEARTBEAT_GLOBAL */
111 };
112 
113 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
114 static unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
115 
116 /*
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.
121  */
122 static unsigned int o2hb_dependent_users;
123 
124 /*
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.
129  */
130 #define O2HB_PIN_CUT_OFF		3
131 
132 /*
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.
136  *
137  * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
138  * works for both file system and userdlm domains.
139  */
140 static int o2hb_region_pin(const char *region_uuid);
141 static void o2hb_region_unpin(const char *region_uuid);
142 
143 /* Only sets a new threshold if there are no active regions.
144  *
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)
149 {
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);
155 	}
156 }
157 
158 static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
159 {
160 	int ret = -1;
161 
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;
166 			ret = 0;
167 		}
168 		spin_unlock(&o2hb_live_lock);
169 	}
170 
171 	return ret;
172 }
173 
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;
178 	int                     hn_node_num;
179 };
180 
181 struct o2hb_disk_slot {
182 	struct o2hb_disk_heartbeat_block *ds_raw_block;
183 	u8			ds_node_num;
184 	u64			ds_last_time;
185 	u64			ds_last_generation;
186 	u16			ds_equal_samples;
187 	u16			ds_changed_samples;
188 	struct list_head	ds_live_item;
189 };
190 
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 */
193 struct o2hb_region {
194 	struct config_item	hr_item;
195 
196 	struct list_head	hr_all_item;
197 	unsigned		hr_unclean_stop:1,
198 				hr_aborted_start:1,
199 				hr_item_pinned:1,
200 				hr_item_dropped:1,
201 				hr_node_deleted:1;
202 
203 	/* protected by the hr_callback_sem */
204 	struct task_struct 	*hr_task;
205 
206 	unsigned int		hr_blocks;
207 	unsigned long long	hr_start_block;
208 
209 	unsigned int		hr_block_bits;
210 	unsigned int		hr_block_bytes;
211 
212 	unsigned int		hr_slots_per_page;
213 	unsigned int		hr_num_pages;
214 
215 	struct page             **hr_slot_data;
216 	struct block_device	*hr_bdev;
217 	struct o2hb_disk_slot	*hr_slots;
218 
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;
222 
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;
228 
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;
233 
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;
237 
238 	char			hr_dev_name[BDEVNAME_SIZE];
239 
240 	unsigned int		hr_timeout_ms;
241 
242 	/* randomized as the region goes up and down so that a node
243 	 * recognizes a node going up and down in one iteration */
244 	u64			hr_generation;
245 
246 	struct delayed_work	hr_write_timeout_work;
247 	unsigned long		hr_last_timeout_start;
248 
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)];
252 
253 	/* Used during o2hb_check_slot to hold a copy of the block
254 	 * being checked because we temporarily have to zero out the
255 	 * crc field. */
256 	struct o2hb_disk_heartbeat_block *hr_tmp_block;
257 
258 	/* Message key for negotiate timeout message. */
259 	unsigned int		hr_key;
260 	struct list_head	hr_handler_list;
261 
262 	/* last hb status, 0 for success, other value for error. */
263 	int			hr_last_hb_status;
264 };
265 
266 struct o2hb_bio_wait_ctxt {
267 	atomic_t          wc_num_reqs;
268 	struct completion wc_io_complete;
269 	int               wc_error;
270 };
271 
272 #define O2HB_NEGO_TIMEOUT_MS (O2HB_MAX_WRITE_TIMEOUT_MS/2)
273 
274 enum {
275 	O2HB_NEGO_TIMEOUT_MSG = 1,
276 	O2HB_NEGO_APPROVE_MSG = 2,
277 };
278 
279 struct o2hb_nego_msg {
280 	u8 node_num;
281 };
282 
283 static void o2hb_write_timeout(struct work_struct *work)
284 {
285 	int failed, quorum;
286 	struct o2hb_region *reg =
287 		container_of(work, struct o2hb_region,
288 			     hr_write_timeout_work.work);
289 
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));
293 
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,
299 					O2NM_MAX_REGIONS);
300 		quorum = bitmap_weight(o2hb_quorum_region_bitmap,
301 					O2NM_MAX_REGIONS);
302 		spin_unlock(&o2hb_live_lock);
303 
304 		mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
305 		     quorum, failed);
306 
307 		/*
308 		 * Fence if the number of failed regions >= half the number
309 		 * of  quorum regions
310 		 */
311 		if ((failed << 1) < quorum)
312 			return;
313 	}
314 
315 	o2quo_disk_timeout();
316 }
317 
318 static void o2hb_arm_timeout(struct o2hb_region *reg)
319 {
320 	/* Arm writeout only after thread reaches steady state */
321 	if (atomic_read(&reg->hr_steady_iterations) != 0)
322 		return;
323 
324 	mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
325 	     O2HB_MAX_WRITE_TIMEOUT_MS);
326 
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);
331 	}
332 	cancel_delayed_work(&reg->hr_write_timeout_work);
333 	schedule_delayed_work(&reg->hr_write_timeout_work,
334 			      msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
335 
336 	cancel_delayed_work(&reg->hr_nego_timeout_work);
337 	/* negotiate timeout must be less than write timeout. */
338 	schedule_delayed_work(&reg->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));
341 }
342 
343 static void o2hb_disarm_timeout(struct o2hb_region *reg)
344 {
345 	cancel_delayed_work_sync(&reg->hr_write_timeout_work);
346 	cancel_delayed_work_sync(&reg->hr_nego_timeout_work);
347 }
348 
349 static int o2hb_send_nego_msg(int key, int type, u8 target)
350 {
351 	struct o2hb_nego_msg msg;
352 	int status, ret;
353 
354 	msg.node_num = o2nm_this_node();
355 again:
356 	ret = o2net_send_message(type, key, &msg, sizeof(msg),
357 			target, &status);
358 
359 	if (ret == -EAGAIN || ret == -ENOMEM) {
360 		msleep(100);
361 		goto again;
362 	}
363 
364 	return ret;
365 }
366 
367 static void o2hb_nego_timeout(struct work_struct *work)
368 {
369 	unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
370 	int master_node, i, ret;
371 	struct o2hb_region *reg;
372 
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.
376 	 */
377 	if (reg->hr_last_hb_status)
378 		return;
379 
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);
383 
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(&reg->hr_item), reg->hr_dev_name);
389 			set_bit(master_node, reg->hr_nego_node_bitmap);
390 		}
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
394 			 * approve decision.
395 			 */
396 			schedule_delayed_work(&reg->hr_nego_timeout_work,
397 				msecs_to_jiffies(1000));
398 
399 			return;
400 		}
401 
402 		printk(KERN_NOTICE "o2hb: all nodes hb write hung, maybe region %s (%s) is down.\n",
403 			config_item_name(&reg->hr_item), reg->hr_dev_name);
404 		/* approve negotiate timeout request. */
405 		o2hb_arm_timeout(reg);
406 
407 		i = -1;
408 		while ((i = find_next_bit(live_node_bitmap,
409 				O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
410 			if (i == master_node)
411 				continue;
412 
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);
416 			if (ret)
417 				mlog(ML_ERROR, "send NEGO_APPROVE msg to node %d fail %d\n",
418 					i, ret);
419 		}
420 	} else {
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(&reg->hr_item),
424 			reg->hr_dev_name, master_node);
425 		ret = o2hb_send_nego_msg(reg->hr_key, O2HB_NEGO_TIMEOUT_MSG,
426 				master_node);
427 		if (ret)
428 			mlog(ML_ERROR, "send NEGO_TIMEOUT msg to node %d fail %d\n",
429 				master_node, ret);
430 	}
431 }
432 
433 static int o2hb_nego_timeout_handler(struct o2net_msg *msg, u32 len, void *data,
434 				void **ret_data)
435 {
436 	struct o2hb_region *reg = data;
437 	struct o2hb_nego_msg *nego_msg;
438 
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(&reg->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);
444 	else
445 		mlog(ML_ERROR, "got nego timeout message from bad node.\n");
446 
447 	return 0;
448 }
449 
450 static int o2hb_nego_approve_handler(struct o2net_msg *msg, u32 len, void *data,
451 				void **ret_data)
452 {
453 	struct o2hb_region *reg = data;
454 
455 	printk(KERN_NOTICE "o2hb: negotiate timeout approved by master node on region %s (%s).\n",
456 		config_item_name(&reg->hr_item), reg->hr_dev_name);
457 	o2hb_arm_timeout(reg);
458 	return 0;
459 }
460 
461 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
462 {
463 	atomic_set(&wc->wc_num_reqs, 1);
464 	init_completion(&wc->wc_io_complete);
465 	wc->wc_error = 0;
466 }
467 
468 /* Used in error paths too */
469 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
470 				     unsigned int num)
471 {
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 */
474 	while(num--) {
475 		if (atomic_dec_and_test(&wc->wc_num_reqs)) {
476 			BUG_ON(num > 0);
477 			complete(&wc->wc_io_complete);
478 		}
479 	}
480 }
481 
482 static void o2hb_wait_on_io(struct o2hb_bio_wait_ctxt *wc)
483 {
484 	o2hb_bio_wait_dec(wc, 1);
485 	wait_for_completion(&wc->wc_io_complete);
486 }
487 
488 static void o2hb_bio_end_io(struct bio *bio)
489 {
490 	struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
491 
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);
495 	}
496 
497 	o2hb_bio_wait_dec(wc, 1);
498 	bio_put(bio);
499 }
500 
501 /* Setup a Bio to cover I/O against num_slots slots starting at
502  * start_slot. */
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,
507 				      int op_flags)
508 {
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;
514 	struct bio *bio;
515 	struct page *page;
516 
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
520 	 * all together. */
521 	bio = bio_alloc(GFP_ATOMIC, 16);
522 	if (!bio) {
523 		mlog(ML_ERROR, "Could not alloc slots BIO!\n");
524 		bio = ERR_PTR(-ENOMEM);
525 		goto bail;
526 	}
527 
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);
534 
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];
539 
540 		vec_len = min(PAGE_SIZE - vec_start,
541 			      (max_slots-cs) * (PAGE_SIZE/spp) );
542 
543 		mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
544 		     current_page, vec_len, vec_start);
545 
546 		len = bio_add_page(bio, page, vec_len, vec_start);
547 		if (len != vec_len) break;
548 
549 		cs += vec_len / (PAGE_SIZE/spp);
550 		vec_start = 0;
551 	}
552 
553 bail:
554 	*current_slot = cs;
555 	return bio;
556 }
557 
558 static int o2hb_read_slots(struct o2hb_region *reg,
559 			   unsigned int begin_slot,
560 			   unsigned int max_slots)
561 {
562 	unsigned int current_slot = begin_slot;
563 	int status;
564 	struct o2hb_bio_wait_ctxt wc;
565 	struct bio *bio;
566 
567 	o2hb_bio_wait_init(&wc);
568 
569 	while(current_slot < max_slots) {
570 		bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots,
571 					 REQ_OP_READ, 0);
572 		if (IS_ERR(bio)) {
573 			status = PTR_ERR(bio);
574 			mlog_errno(status);
575 			goto bail_and_wait;
576 		}
577 
578 		atomic_inc(&wc.wc_num_reqs);
579 		submit_bio(bio);
580 	}
581 
582 	status = 0;
583 
584 bail_and_wait:
585 	o2hb_wait_on_io(&wc);
586 	if (wc.wc_error && !status)
587 		status = wc.wc_error;
588 
589 	return status;
590 }
591 
592 static int o2hb_issue_node_write(struct o2hb_region *reg,
593 				 struct o2hb_bio_wait_ctxt *write_wc)
594 {
595 	int status;
596 	unsigned int slot;
597 	struct bio *bio;
598 
599 	o2hb_bio_wait_init(write_wc);
600 
601 	slot = o2nm_this_node();
602 
603 	bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1, REQ_OP_WRITE,
604 				 REQ_SYNC);
605 	if (IS_ERR(bio)) {
606 		status = PTR_ERR(bio);
607 		mlog_errno(status);
608 		goto bail;
609 	}
610 
611 	atomic_inc(&write_wc->wc_num_reqs);
612 	submit_bio(bio);
613 
614 	status = 0;
615 bail:
616 	return status;
617 }
618 
619 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
620 				     struct o2hb_disk_heartbeat_block *hb_block)
621 {
622 	__le32 old_cksum;
623 	u32 ret;
624 
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
627 	 * crc. */
628 	old_cksum = hb_block->hb_cksum;
629 	hb_block->hb_cksum = 0;
630 
631 	ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
632 
633 	hb_block->hb_cksum = old_cksum;
634 
635 	return ret;
636 }
637 
638 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
639 {
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));
645 }
646 
647 static int o2hb_verify_crc(struct o2hb_region *reg,
648 			   struct o2hb_disk_heartbeat_block *hb_block)
649 {
650 	u32 read, computed;
651 
652 	read = le32_to_cpu(hb_block->hb_cksum);
653 	computed = o2hb_compute_block_crc_le(reg, hb_block);
654 
655 	return read == computed;
656 }
657 
658 /*
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.
664  */
665 static int o2hb_check_own_slot(struct o2hb_region *reg)
666 {
667 	struct o2hb_disk_slot *slot;
668 	struct o2hb_disk_heartbeat_block *hb_block;
669 	char *errstr;
670 
671 	slot = &reg->hr_slots[o2nm_this_node()];
672 	/* Don't check on our 1st timestamp */
673 	if (!slot->ds_last_time)
674 		return 0;
675 
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)
680 		return 1;
681 
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"
685 
686 	if (hb_block->hb_node != slot->ds_node_num)
687 		errstr = ERRSTR1;
688 	else if (le64_to_cpu(hb_block->hb_generation) !=
689 		 slot->ds_last_generation)
690 		errstr = ERRSTR2;
691 	else
692 		errstr = ERRSTR3;
693 
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));
700 
701 	return 0;
702 }
703 
704 static inline void o2hb_prepare_block(struct o2hb_region *reg,
705 				      u64 generation)
706 {
707 	int node_num;
708 	u64 cputime;
709 	struct o2hb_disk_slot *slot;
710 	struct o2hb_disk_heartbeat_block *hb_block;
711 
712 	node_num = o2nm_this_node();
713 	slot = &reg->hr_slots[node_num];
714 
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();
719 	if (!cputime)
720 		cputime = 1;
721 
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);
726 
727 	/* This step must always happen last! */
728 	hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
729 								   hb_block));
730 
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));
734 }
735 
736 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
737 				struct o2nm_node *node,
738 				int idx)
739 {
740 	struct o2hb_callback_func *f;
741 
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);
745 	}
746 }
747 
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)
750 {
751 	struct o2hb_callback *hbcall;
752 	struct o2hb_node_event *event;
753 
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);
758 
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,
764 				   hn_item);
765 		list_del_init(&event->hn_item);
766 		spin_unlock(&o2hb_live_lock);
767 
768 		mlog(ML_HEARTBEAT, "Node %s event for %d\n",
769 		     event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
770 		     event->hn_node_num);
771 
772 		hbcall = hbcall_from_type(event->hn_event_type);
773 
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));
778 
779 		o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
780 
781 		spin_lock(&o2hb_live_lock);
782 	}
783 	spin_unlock(&o2hb_live_lock);
784 
785 	up_write(&o2hb_callback_sem);
786 }
787 
788 static void o2hb_queue_node_event(struct o2hb_node_event *event,
789 				  enum o2hb_callback_type type,
790 				  struct o2nm_node *node,
791 				  int node_num)
792 {
793 	assert_spin_locked(&o2hb_live_lock);
794 
795 	BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
796 
797 	event->hn_event_type = type;
798 	event->hn_node = node;
799 	event->hn_node_num = node_num;
800 
801 	mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
802 	     type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
803 
804 	list_add_tail(&event->hn_item, &o2hb_node_events);
805 }
806 
807 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
808 {
809 	struct o2hb_node_event event =
810 		{ .hn_item = LIST_HEAD_INIT(event.hn_item), };
811 	struct o2nm_node *node;
812 	int queued = 0;
813 
814 	node = o2nm_get_node_by_num(slot->ds_node_num);
815 	if (!node)
816 		return;
817 
818 	spin_lock(&o2hb_live_lock);
819 	if (!list_empty(&slot->ds_live_item)) {
820 		mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
821 		     slot->ds_node_num);
822 
823 		list_del_init(&slot->ds_live_item);
824 
825 		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
826 			clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
827 
828 			o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
829 					      slot->ds_node_num);
830 			queued = 1;
831 		}
832 	}
833 	spin_unlock(&o2hb_live_lock);
834 
835 	if (queued)
836 		o2hb_run_event_list(&event);
837 
838 	o2nm_node_put(node);
839 }
840 
841 static void o2hb_set_quorum_device(struct o2hb_region *reg)
842 {
843 	if (!o2hb_global_heartbeat_active())
844 		return;
845 
846 	/* Prevent race with o2hb_heartbeat_group_drop_item() */
847 	if (kthread_should_stop())
848 		return;
849 
850 	/* Tag region as quorum only after thread reaches steady state */
851 	if (atomic_read(&reg->hr_steady_iterations) != 0)
852 		return;
853 
854 	spin_lock(&o2hb_live_lock);
855 
856 	if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
857 		goto unlock;
858 
859 	/*
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.
863 	 */
864 	if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
865 		   sizeof(o2hb_live_node_bitmap)))
866 		goto unlock;
867 
868 	printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
869 	       config_item_name(&reg->hr_item), reg->hr_dev_name);
870 
871 	set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
872 
873 	/*
874 	 * If global heartbeat active, unpin all regions if the
875 	 * region count > CUT_OFF
876 	 */
877 	if (bitmap_weight(o2hb_quorum_region_bitmap,
878 			   O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
879 		o2hb_region_unpin(NULL);
880 unlock:
881 	spin_unlock(&o2hb_live_lock);
882 }
883 
884 static int o2hb_check_slot(struct o2hb_region *reg,
885 			   struct o2hb_disk_slot *slot)
886 {
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;
892 	u64 cputime;
893 	unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
894 	unsigned int slot_dead_ms;
895 	int tmp;
896 	int queued = 0;
897 
898 	memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
899 
900 	/*
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.
903 	 */
904 	node = o2nm_get_node_by_num(slot->ds_node_num);
905 	if (!node) {
906 		spin_lock(&o2hb_live_lock);
907 		tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
908 		spin_unlock(&o2hb_live_lock);
909 		if (!tmp)
910 			return 0;
911 	}
912 
913 	if (!o2hb_verify_crc(reg, hb_block)) {
914 		/* all paths from here will drop o2hb_live_lock for
915 		 * us. */
916 		spin_lock(&o2hb_live_lock);
917 
918 		/* Don't print an error on the console in this case -
919 		 * a freshly formatted heartbeat area will not have a
920 		 * crc set on it. */
921 		if (list_empty(&slot->ds_live_item))
922 			goto out;
923 
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);
930 
931 		slot->ds_equal_samples++;
932 		goto fire_callbacks;
933 	}
934 
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++;
940 	else
941 		slot->ds_equal_samples++;
942 	slot->ds_last_time = cputime;
943 
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)) {
950 		gen_changed = 1;
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));
956 	}
957 
958 	slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
959 
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);
967 
968 	spin_lock(&o2hb_live_lock);
969 
970 fire_callbacks:
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);
977 
978 		set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
979 
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);
985 
986 			o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
987 					      slot->ds_node_num);
988 
989 			changed = 1;
990 			queued = 1;
991 		}
992 
993 		list_add_tail(&slot->ds_live_item,
994 			      &o2hb_live_slots[slot->ds_node_num]);
995 
996 		slot->ds_equal_samples = 0;
997 
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,
1012 			     dead_ms);
1013 		}
1014 		goto out;
1015 	}
1016 
1017 	/* if the list is dead, we're done.. */
1018 	if (list_empty(&slot->ds_live_item))
1019 		goto out;
1020 
1021 	/* live nodes only go dead after enough consequtive missed
1022 	 * samples..  reset the missed counter whenever we see
1023 	 * activity */
1024 	if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
1025 		mlog(ML_HEARTBEAT, "Node %d left my region\n",
1026 		     slot->ds_node_num);
1027 
1028 		clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1029 
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);
1036 
1037 			/* node can be null */
1038 			o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
1039 					      node, slot->ds_node_num);
1040 
1041 			changed = 1;
1042 			queued = 1;
1043 		}
1044 
1045 		/* We don't clear this because the node is still
1046 		 * actually writing new blocks. */
1047 		if (!gen_changed)
1048 			slot->ds_changed_samples = 0;
1049 		goto out;
1050 	}
1051 	if (slot->ds_changed_samples) {
1052 		slot->ds_changed_samples = 0;
1053 		slot->ds_equal_samples = 0;
1054 	}
1055 out:
1056 	spin_unlock(&o2hb_live_lock);
1057 
1058 	if (queued)
1059 		o2hb_run_event_list(&event);
1060 
1061 	if (node)
1062 		o2nm_node_put(node);
1063 	return changed;
1064 }
1065 
1066 static int o2hb_highest_node(unsigned long *nodes, int numbits)
1067 {
1068 	return find_last_bit(nodes, numbits);
1069 }
1070 
1071 static int o2hb_lowest_node(unsigned long *nodes, int numbits)
1072 {
1073 	return find_first_bit(nodes, numbits);
1074 }
1075 
1076 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
1077 {
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;
1083 
1084 	ret = o2nm_configured_node_map(configured_nodes,
1085 				       sizeof(configured_nodes));
1086 	if (ret) {
1087 		mlog_errno(ret);
1088 		goto bail;
1089 	}
1090 
1091 	/*
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.
1094 	 */
1095 	o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
1096 	i = -1;
1097 	while ((i = find_next_bit(live_node_bitmap,
1098 				  O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1099 		set_bit(i, configured_nodes);
1100 	}
1101 
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");
1106 		ret = -EINVAL;
1107 		goto bail;
1108 	}
1109 
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
1113 	 * best-effort. */
1114 	ret = o2hb_read_slots(reg, lowest_node, highest_node + 1);
1115 	if (ret < 0) {
1116 		mlog_errno(ret);
1117 		goto bail;
1118 	}
1119 
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
1122 	 * our slot. */
1123 	own_slot_ok = o2hb_check_own_slot(reg);
1124 
1125 	/* fill in the proper info for our next heartbeat */
1126 	o2hb_prepare_block(reg, reg->hr_generation);
1127 
1128 	ret = o2hb_issue_node_write(reg, &write_wc);
1129 	if (ret < 0) {
1130 		mlog_errno(ret);
1131 		goto bail;
1132 	}
1133 
1134 	i = -1;
1135 	while((i = find_next_bit(configured_nodes,
1136 				 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1137 		membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1138 	}
1139 
1140 	/*
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.
1144 	 */
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
1149 		 * disk */
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;
1153 		goto bail;
1154 	}
1155 
1156 	/* Skip disarming the timeout if own slot has stale/bad data */
1157 	if (own_slot_ok) {
1158 		o2hb_set_quorum_device(reg);
1159 		o2hb_arm_timeout(reg);
1160 		reg->hr_last_timeout_start = jiffies;
1161 	}
1162 
1163 bail:
1164 	/* let the person who launched us know when things are steady */
1165 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
1166 		if (!ret && own_slot_ok && !membership_change) {
1167 			if (atomic_dec_and_test(&reg->hr_steady_iterations))
1168 				wake_up(&o2hb_steady_queue);
1169 		}
1170 	}
1171 
1172 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
1173 		if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1174 			printk(KERN_NOTICE "o2hb: Unable to stabilize "
1175 			       "heartbeat on region %s (%s)\n",
1176 			       config_item_name(&reg->hr_item),
1177 			       reg->hr_dev_name);
1178 			atomic_set(&reg->hr_steady_iterations, 0);
1179 			reg->hr_aborted_start = 1;
1180 			wake_up(&o2hb_steady_queue);
1181 			ret = -EIO;
1182 		}
1183 	}
1184 
1185 	return ret;
1186 }
1187 
1188 /*
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
1191  * thread.
1192  */
1193 static int o2hb_thread(void *data)
1194 {
1195 	int i, ret;
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;
1200 
1201 	mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1202 
1203 	set_user_nice(current, MIN_NICE);
1204 
1205 	/* Pin node */
1206 	ret = o2nm_depend_this_node();
1207 	if (ret) {
1208 		mlog(ML_ERROR, "Node has been deleted, ret = %d\n", ret);
1209 		reg->hr_node_deleted = 1;
1210 		wake_up(&o2hb_steady_queue);
1211 		return 0;
1212 	}
1213 
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();
1222 
1223 		ret = o2hb_do_disk_heartbeat(reg);
1224 		reg->hr_last_hb_status = ret;
1225 
1226 		after_hb = ktime_get_real();
1227 
1228 		elapsed_msec = (unsigned int)
1229 				ktime_ms_delta(after_hb, before_hb);
1230 
1231 		mlog(ML_HEARTBEAT,
1232 		     "start = %lld, end = %lld, msec = %u, ret = %d\n",
1233 		     before_hb, after_hb, elapsed_msec, ret);
1234 
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);
1240 		}
1241 	}
1242 
1243 	o2hb_disarm_timeout(reg);
1244 
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(&reg->hr_slots[i]);
1248 
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.
1253 	 */
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);
1257 		if (ret == 0)
1258 			o2hb_wait_on_io(&write_wc);
1259 		else
1260 			mlog_errno(ret);
1261 	}
1262 
1263 	/* Unpin node */
1264 	o2nm_undepend_this_node();
1265 
1266 	mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1267 
1268 	return 0;
1269 }
1270 
1271 #ifdef CONFIG_DEBUG_FS
1272 static int o2hb_debug_open(struct inode *inode, struct file *file)
1273 {
1274 	struct o2hb_debug_buf *db = inode->i_private;
1275 	struct o2hb_region *reg;
1276 	unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1277 	unsigned long lts;
1278 	char *buf = NULL;
1279 	int i = -1;
1280 	int out = 0;
1281 
1282 	/* max_nodes should be the largest bitmap we pass here */
1283 	BUG_ON(sizeof(map) < db->db_size);
1284 
1285 	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1286 	if (!buf)
1287 		goto bail;
1288 
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);
1297 		break;
1298 
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);
1304 		break;
1305 
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);
1310 		goto done;
1311 
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 */
1316 		if (lts)
1317 			lts = jiffies_to_msecs(jiffies - lts);
1318 		out += scnprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1319 		goto done;
1320 
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);
1325 		goto done;
1326 
1327 	default:
1328 		goto done;
1329 	}
1330 
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");
1334 
1335 done:
1336 	i_size_write(inode, out);
1337 
1338 	file->private_data = buf;
1339 
1340 	return 0;
1341 bail:
1342 	return -ENOMEM;
1343 }
1344 
1345 static int o2hb_debug_release(struct inode *inode, struct file *file)
1346 {
1347 	kfree(file->private_data);
1348 	return 0;
1349 }
1350 
1351 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1352 				 size_t nbytes, loff_t *ppos)
1353 {
1354 	return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1355 				       i_size_read(file->f_mapping->host));
1356 }
1357 #else
1358 static int o2hb_debug_open(struct inode *inode, struct file *file)
1359 {
1360 	return 0;
1361 }
1362 static int o2hb_debug_release(struct inode *inode, struct file *file)
1363 {
1364 	return 0;
1365 }
1366 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1367 			       size_t nbytes, loff_t *ppos)
1368 {
1369 	return 0;
1370 }
1371 #endif  /* CONFIG_DEBUG_FS */
1372 
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,
1378 };
1379 
1380 void o2hb_exit(void)
1381 {
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);
1387 }
1388 
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)
1392 {
1393 	*db = kmalloc(db_len, GFP_KERNEL);
1394 	if (!*db)
1395 		return;
1396 
1397 	(*db)->db_type = type;
1398 	(*db)->db_size = size;
1399 	(*db)->db_len = len;
1400 	(*db)->db_data = data;
1401 
1402 	debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db, &o2hb_debug_fops);
1403 }
1404 
1405 static void o2hb_debug_init(void)
1406 {
1407 	o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1408 
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);
1413 
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);
1419 
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);
1426 
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);
1433 }
1434 
1435 void o2hb_init(void)
1436 {
1437 	int i;
1438 
1439 	for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1440 		INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1441 
1442 	for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1443 		INIT_LIST_HEAD(&o2hb_live_slots[i]);
1444 
1445 	INIT_LIST_HEAD(&o2hb_node_events);
1446 
1447 	memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1448 	memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1449 	memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1450 	memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1451 	memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1452 
1453 	o2hb_dependent_users = 0;
1454 
1455 	o2hb_debug_init();
1456 }
1457 
1458 /* if we're already in a callback then we're already serialized by the sem */
1459 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1460 					     unsigned bytes)
1461 {
1462 	BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1463 
1464 	memcpy(map, &o2hb_live_node_bitmap, bytes);
1465 }
1466 
1467 /*
1468  * get a map of all nodes that are heartbeating in any regions
1469  */
1470 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1471 {
1472 	/* callers want to serialize this map and callbacks so that they
1473 	 * can trust that they don't miss nodes coming to the party */
1474 	down_read(&o2hb_callback_sem);
1475 	spin_lock(&o2hb_live_lock);
1476 	o2hb_fill_node_map_from_callback(map, bytes);
1477 	spin_unlock(&o2hb_live_lock);
1478 	up_read(&o2hb_callback_sem);
1479 }
1480 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1481 
1482 /*
1483  * heartbeat configfs bits.  The heartbeat set is a default set under
1484  * the cluster set in nodemanager.c.
1485  */
1486 
1487 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1488 {
1489 	return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1490 }
1491 
1492 /* drop_item only drops its ref after killing the thread, nothing should
1493  * be using the region anymore.  this has to clean up any state that
1494  * attributes might have built up. */
1495 static void o2hb_region_release(struct config_item *item)
1496 {
1497 	int i;
1498 	struct page *page;
1499 	struct o2hb_region *reg = to_o2hb_region(item);
1500 
1501 	mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1502 
1503 	kfree(reg->hr_tmp_block);
1504 
1505 	if (reg->hr_slot_data) {
1506 		for (i = 0; i < reg->hr_num_pages; i++) {
1507 			page = reg->hr_slot_data[i];
1508 			if (page)
1509 				__free_page(page);
1510 		}
1511 		kfree(reg->hr_slot_data);
1512 	}
1513 
1514 	if (reg->hr_bdev)
1515 		blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1516 
1517 	kfree(reg->hr_slots);
1518 
1519 	debugfs_remove_recursive(reg->hr_debug_dir);
1520 	kfree(reg->hr_db_livenodes);
1521 	kfree(reg->hr_db_regnum);
1522 	kfree(reg->hr_db_elapsed_time);
1523 	kfree(reg->hr_db_pinned);
1524 
1525 	spin_lock(&o2hb_live_lock);
1526 	list_del(&reg->hr_all_item);
1527 	spin_unlock(&o2hb_live_lock);
1528 
1529 	o2net_unregister_handler_list(&reg->hr_handler_list);
1530 	kfree(reg);
1531 }
1532 
1533 static int o2hb_read_block_input(struct o2hb_region *reg,
1534 				 const char *page,
1535 				 unsigned long *ret_bytes,
1536 				 unsigned int *ret_bits)
1537 {
1538 	unsigned long bytes;
1539 	char *p = (char *)page;
1540 
1541 	bytes = simple_strtoul(p, &p, 0);
1542 	if (!p || (*p && (*p != '\n')))
1543 		return -EINVAL;
1544 
1545 	/* Heartbeat and fs min / max block sizes are the same. */
1546 	if (bytes > 4096 || bytes < 512)
1547 		return -ERANGE;
1548 	if (hweight16(bytes) != 1)
1549 		return -EINVAL;
1550 
1551 	if (ret_bytes)
1552 		*ret_bytes = bytes;
1553 	if (ret_bits)
1554 		*ret_bits = ffs(bytes) - 1;
1555 
1556 	return 0;
1557 }
1558 
1559 static ssize_t o2hb_region_block_bytes_show(struct config_item *item,
1560 					    char *page)
1561 {
1562 	return sprintf(page, "%u\n", to_o2hb_region(item)->hr_block_bytes);
1563 }
1564 
1565 static ssize_t o2hb_region_block_bytes_store(struct config_item *item,
1566 					     const char *page,
1567 					     size_t count)
1568 {
1569 	struct o2hb_region *reg = to_o2hb_region(item);
1570 	int status;
1571 	unsigned long block_bytes;
1572 	unsigned int block_bits;
1573 
1574 	if (reg->hr_bdev)
1575 		return -EINVAL;
1576 
1577 	status = o2hb_read_block_input(reg, page, &block_bytes,
1578 				       &block_bits);
1579 	if (status)
1580 		return status;
1581 
1582 	reg->hr_block_bytes = (unsigned int)block_bytes;
1583 	reg->hr_block_bits = block_bits;
1584 
1585 	return count;
1586 }
1587 
1588 static ssize_t o2hb_region_start_block_show(struct config_item *item,
1589 					    char *page)
1590 {
1591 	return sprintf(page, "%llu\n", to_o2hb_region(item)->hr_start_block);
1592 }
1593 
1594 static ssize_t o2hb_region_start_block_store(struct config_item *item,
1595 					     const char *page,
1596 					     size_t count)
1597 {
1598 	struct o2hb_region *reg = to_o2hb_region(item);
1599 	unsigned long long tmp;
1600 	char *p = (char *)page;
1601 
1602 	if (reg->hr_bdev)
1603 		return -EINVAL;
1604 
1605 	tmp = simple_strtoull(p, &p, 0);
1606 	if (!p || (*p && (*p != '\n')))
1607 		return -EINVAL;
1608 
1609 	reg->hr_start_block = tmp;
1610 
1611 	return count;
1612 }
1613 
1614 static ssize_t o2hb_region_blocks_show(struct config_item *item, char *page)
1615 {
1616 	return sprintf(page, "%d\n", to_o2hb_region(item)->hr_blocks);
1617 }
1618 
1619 static ssize_t o2hb_region_blocks_store(struct config_item *item,
1620 					const char *page,
1621 					size_t count)
1622 {
1623 	struct o2hb_region *reg = to_o2hb_region(item);
1624 	unsigned long tmp;
1625 	char *p = (char *)page;
1626 
1627 	if (reg->hr_bdev)
1628 		return -EINVAL;
1629 
1630 	tmp = simple_strtoul(p, &p, 0);
1631 	if (!p || (*p && (*p != '\n')))
1632 		return -EINVAL;
1633 
1634 	if (tmp > O2NM_MAX_NODES || tmp == 0)
1635 		return -ERANGE;
1636 
1637 	reg->hr_blocks = (unsigned int)tmp;
1638 
1639 	return count;
1640 }
1641 
1642 static ssize_t o2hb_region_dev_show(struct config_item *item, char *page)
1643 {
1644 	unsigned int ret = 0;
1645 
1646 	if (to_o2hb_region(item)->hr_bdev)
1647 		ret = sprintf(page, "%s\n", to_o2hb_region(item)->hr_dev_name);
1648 
1649 	return ret;
1650 }
1651 
1652 static void o2hb_init_region_params(struct o2hb_region *reg)
1653 {
1654 	reg->hr_slots_per_page = PAGE_SIZE >> reg->hr_block_bits;
1655 	reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1656 
1657 	mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1658 	     reg->hr_start_block, reg->hr_blocks);
1659 	mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1660 	     reg->hr_block_bytes, reg->hr_block_bits);
1661 	mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1662 	mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1663 }
1664 
1665 static int o2hb_map_slot_data(struct o2hb_region *reg)
1666 {
1667 	int i, j;
1668 	unsigned int last_slot;
1669 	unsigned int spp = reg->hr_slots_per_page;
1670 	struct page *page;
1671 	char *raw;
1672 	struct o2hb_disk_slot *slot;
1673 
1674 	reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1675 	if (reg->hr_tmp_block == NULL)
1676 		return -ENOMEM;
1677 
1678 	reg->hr_slots = kcalloc(reg->hr_blocks,
1679 				sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1680 	if (reg->hr_slots == NULL)
1681 		return -ENOMEM;
1682 
1683 	for(i = 0; i < reg->hr_blocks; i++) {
1684 		slot = &reg->hr_slots[i];
1685 		slot->ds_node_num = i;
1686 		INIT_LIST_HEAD(&slot->ds_live_item);
1687 		slot->ds_raw_block = NULL;
1688 	}
1689 
1690 	reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1691 	mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1692 			   "at %u blocks per page\n",
1693 	     reg->hr_num_pages, reg->hr_blocks, spp);
1694 
1695 	reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1696 				    GFP_KERNEL);
1697 	if (!reg->hr_slot_data)
1698 		return -ENOMEM;
1699 
1700 	for(i = 0; i < reg->hr_num_pages; i++) {
1701 		page = alloc_page(GFP_KERNEL);
1702 		if (!page)
1703 			return -ENOMEM;
1704 
1705 		reg->hr_slot_data[i] = page;
1706 
1707 		last_slot = i * spp;
1708 		raw = page_address(page);
1709 		for (j = 0;
1710 		     (j < spp) && ((j + last_slot) < reg->hr_blocks);
1711 		     j++) {
1712 			BUG_ON((j + last_slot) >= reg->hr_blocks);
1713 
1714 			slot = &reg->hr_slots[j + last_slot];
1715 			slot->ds_raw_block =
1716 				(struct o2hb_disk_heartbeat_block *) raw;
1717 
1718 			raw += reg->hr_block_bytes;
1719 		}
1720 	}
1721 
1722 	return 0;
1723 }
1724 
1725 /* Read in all the slots available and populate the tracking
1726  * structures so that we can start with a baseline idea of what's
1727  * there. */
1728 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1729 {
1730 	int ret, i;
1731 	struct o2hb_disk_slot *slot;
1732 	struct o2hb_disk_heartbeat_block *hb_block;
1733 
1734 	ret = o2hb_read_slots(reg, 0, reg->hr_blocks);
1735 	if (ret)
1736 		goto out;
1737 
1738 	/* We only want to get an idea of the values initially in each
1739 	 * slot, so we do no verification - o2hb_check_slot will
1740 	 * actually determine if each configured slot is valid and
1741 	 * whether any values have changed. */
1742 	for(i = 0; i < reg->hr_blocks; i++) {
1743 		slot = &reg->hr_slots[i];
1744 		hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1745 
1746 		/* Only fill the values that o2hb_check_slot uses to
1747 		 * determine changing slots */
1748 		slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1749 		slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1750 	}
1751 
1752 out:
1753 	return ret;
1754 }
1755 
1756 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1757 static ssize_t o2hb_region_dev_store(struct config_item *item,
1758 				     const char *page,
1759 				     size_t count)
1760 {
1761 	struct o2hb_region *reg = to_o2hb_region(item);
1762 	struct task_struct *hb_task;
1763 	long fd;
1764 	int sectsize;
1765 	char *p = (char *)page;
1766 	struct fd f;
1767 	ssize_t ret = -EINVAL;
1768 	int live_threshold;
1769 
1770 	if (reg->hr_bdev)
1771 		goto out;
1772 
1773 	/* We can't heartbeat without having had our node number
1774 	 * configured yet. */
1775 	if (o2nm_this_node() == O2NM_MAX_NODES)
1776 		goto out;
1777 
1778 	fd = simple_strtol(p, &p, 0);
1779 	if (!p || (*p && (*p != '\n')))
1780 		goto out;
1781 
1782 	if (fd < 0 || fd >= INT_MAX)
1783 		goto out;
1784 
1785 	f = fdget(fd);
1786 	if (f.file == NULL)
1787 		goto out;
1788 
1789 	if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1790 	    reg->hr_block_bytes == 0)
1791 		goto out2;
1792 
1793 	if (!S_ISBLK(f.file->f_mapping->host->i_mode))
1794 		goto out2;
1795 
1796 	reg->hr_bdev = blkdev_get_by_dev(f.file->f_mapping->host->i_rdev,
1797 					 FMODE_WRITE | FMODE_READ, NULL);
1798 	if (IS_ERR(reg->hr_bdev)) {
1799 		ret = PTR_ERR(reg->hr_bdev);
1800 		reg->hr_bdev = NULL;
1801 		goto out2;
1802 	}
1803 
1804 	bdevname(reg->hr_bdev, reg->hr_dev_name);
1805 
1806 	sectsize = bdev_logical_block_size(reg->hr_bdev);
1807 	if (sectsize != reg->hr_block_bytes) {
1808 		mlog(ML_ERROR,
1809 		     "blocksize %u incorrect for device, expected %d",
1810 		     reg->hr_block_bytes, sectsize);
1811 		ret = -EINVAL;
1812 		goto out3;
1813 	}
1814 
1815 	o2hb_init_region_params(reg);
1816 
1817 	/* Generation of zero is invalid */
1818 	do {
1819 		get_random_bytes(&reg->hr_generation,
1820 				 sizeof(reg->hr_generation));
1821 	} while (reg->hr_generation == 0);
1822 
1823 	ret = o2hb_map_slot_data(reg);
1824 	if (ret) {
1825 		mlog_errno(ret);
1826 		goto out3;
1827 	}
1828 
1829 	ret = o2hb_populate_slot_data(reg);
1830 	if (ret) {
1831 		mlog_errno(ret);
1832 		goto out3;
1833 	}
1834 
1835 	INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1836 	INIT_DELAYED_WORK(&reg->hr_nego_timeout_work, o2hb_nego_timeout);
1837 
1838 	/*
1839 	 * A node is considered live after it has beat LIVE_THRESHOLD
1840 	 * times.  We're not steady until we've given them a chance
1841 	 * _after_ our first read.
1842 	 * The default threshold is bare minimum so as to limit the delay
1843 	 * during mounts. For global heartbeat, the threshold doubled for the
1844 	 * first region.
1845 	 */
1846 	live_threshold = O2HB_LIVE_THRESHOLD;
1847 	if (o2hb_global_heartbeat_active()) {
1848 		spin_lock(&o2hb_live_lock);
1849 		if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1850 			live_threshold <<= 1;
1851 		spin_unlock(&o2hb_live_lock);
1852 	}
1853 	++live_threshold;
1854 	atomic_set(&reg->hr_steady_iterations, live_threshold);
1855 	/* unsteady_iterations is triple the steady_iterations */
1856 	atomic_set(&reg->hr_unsteady_iterations, (live_threshold * 3));
1857 
1858 	hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1859 			      reg->hr_item.ci_name);
1860 	if (IS_ERR(hb_task)) {
1861 		ret = PTR_ERR(hb_task);
1862 		mlog_errno(ret);
1863 		goto out3;
1864 	}
1865 
1866 	spin_lock(&o2hb_live_lock);
1867 	reg->hr_task = hb_task;
1868 	spin_unlock(&o2hb_live_lock);
1869 
1870 	ret = wait_event_interruptible(o2hb_steady_queue,
1871 				atomic_read(&reg->hr_steady_iterations) == 0 ||
1872 				reg->hr_node_deleted);
1873 	if (ret) {
1874 		atomic_set(&reg->hr_steady_iterations, 0);
1875 		reg->hr_aborted_start = 1;
1876 	}
1877 
1878 	if (reg->hr_aborted_start) {
1879 		ret = -EIO;
1880 		goto out3;
1881 	}
1882 
1883 	if (reg->hr_node_deleted) {
1884 		ret = -EINVAL;
1885 		goto out3;
1886 	}
1887 
1888 	/* Ok, we were woken.  Make sure it wasn't by drop_item() */
1889 	spin_lock(&o2hb_live_lock);
1890 	hb_task = reg->hr_task;
1891 	if (o2hb_global_heartbeat_active())
1892 		set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1893 	spin_unlock(&o2hb_live_lock);
1894 
1895 	if (hb_task)
1896 		ret = count;
1897 	else
1898 		ret = -EIO;
1899 
1900 	if (hb_task && o2hb_global_heartbeat_active())
1901 		printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1902 		       config_item_name(&reg->hr_item), reg->hr_dev_name);
1903 
1904 out3:
1905 	if (ret < 0) {
1906 		blkdev_put(reg->hr_bdev, FMODE_READ | FMODE_WRITE);
1907 		reg->hr_bdev = NULL;
1908 	}
1909 out2:
1910 	fdput(f);
1911 out:
1912 	return ret;
1913 }
1914 
1915 static ssize_t o2hb_region_pid_show(struct config_item *item, char *page)
1916 {
1917 	struct o2hb_region *reg = to_o2hb_region(item);
1918 	pid_t pid = 0;
1919 
1920 	spin_lock(&o2hb_live_lock);
1921 	if (reg->hr_task)
1922 		pid = task_pid_nr(reg->hr_task);
1923 	spin_unlock(&o2hb_live_lock);
1924 
1925 	if (!pid)
1926 		return 0;
1927 
1928 	return sprintf(page, "%u\n", pid);
1929 }
1930 
1931 CONFIGFS_ATTR(o2hb_region_, block_bytes);
1932 CONFIGFS_ATTR(o2hb_region_, start_block);
1933 CONFIGFS_ATTR(o2hb_region_, blocks);
1934 CONFIGFS_ATTR(o2hb_region_, dev);
1935 CONFIGFS_ATTR_RO(o2hb_region_, pid);
1936 
1937 static struct configfs_attribute *o2hb_region_attrs[] = {
1938 	&o2hb_region_attr_block_bytes,
1939 	&o2hb_region_attr_start_block,
1940 	&o2hb_region_attr_blocks,
1941 	&o2hb_region_attr_dev,
1942 	&o2hb_region_attr_pid,
1943 	NULL,
1944 };
1945 
1946 static struct configfs_item_operations o2hb_region_item_ops = {
1947 	.release		= o2hb_region_release,
1948 };
1949 
1950 static const struct config_item_type o2hb_region_type = {
1951 	.ct_item_ops	= &o2hb_region_item_ops,
1952 	.ct_attrs	= o2hb_region_attrs,
1953 	.ct_owner	= THIS_MODULE,
1954 };
1955 
1956 /* heartbeat set */
1957 
1958 struct o2hb_heartbeat_group {
1959 	struct config_group hs_group;
1960 	/* some stuff? */
1961 };
1962 
1963 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1964 {
1965 	return group ?
1966 		container_of(group, struct o2hb_heartbeat_group, hs_group)
1967 		: NULL;
1968 }
1969 
1970 static void o2hb_debug_region_init(struct o2hb_region *reg,
1971 				   struct dentry *parent)
1972 {
1973 	struct dentry *dir;
1974 
1975 	dir = debugfs_create_dir(config_item_name(&reg->hr_item), parent);
1976 	reg->hr_debug_dir = dir;
1977 
1978 	o2hb_debug_create(O2HB_DEBUG_LIVENODES, dir, &(reg->hr_db_livenodes),
1979 			  sizeof(*(reg->hr_db_livenodes)),
1980 			  O2HB_DB_TYPE_REGION_LIVENODES,
1981 			  sizeof(reg->hr_live_node_bitmap), O2NM_MAX_NODES,
1982 			  reg);
1983 
1984 	o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER, dir, &(reg->hr_db_regnum),
1985 			  sizeof(*(reg->hr_db_regnum)),
1986 			  O2HB_DB_TYPE_REGION_NUMBER, 0, O2NM_MAX_NODES, reg);
1987 
1988 	o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME, dir,
1989 			  &(reg->hr_db_elapsed_time),
1990 			  sizeof(*(reg->hr_db_elapsed_time)),
1991 			  O2HB_DB_TYPE_REGION_ELAPSED_TIME, 0, 0, reg);
1992 
1993 	o2hb_debug_create(O2HB_DEBUG_REGION_PINNED, dir, &(reg->hr_db_pinned),
1994 			  sizeof(*(reg->hr_db_pinned)),
1995 			  O2HB_DB_TYPE_REGION_PINNED, 0, 0, reg);
1996 
1997 }
1998 
1999 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2000 							  const char *name)
2001 {
2002 	struct o2hb_region *reg = NULL;
2003 	int ret;
2004 
2005 	reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2006 	if (reg == NULL)
2007 		return ERR_PTR(-ENOMEM);
2008 
2009 	if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2010 		ret = -ENAMETOOLONG;
2011 		goto free;
2012 	}
2013 
2014 	spin_lock(&o2hb_live_lock);
2015 	reg->hr_region_num = 0;
2016 	if (o2hb_global_heartbeat_active()) {
2017 		reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2018 							 O2NM_MAX_REGIONS);
2019 		if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2020 			spin_unlock(&o2hb_live_lock);
2021 			ret = -EFBIG;
2022 			goto free;
2023 		}
2024 		set_bit(reg->hr_region_num, o2hb_region_bitmap);
2025 	}
2026 	list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2027 	spin_unlock(&o2hb_live_lock);
2028 
2029 	config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2030 
2031 	/* this is the same way to generate msg key as dlm, for local heartbeat,
2032 	 * name is also the same, so make initial crc value different to avoid
2033 	 * message key conflict.
2034 	 */
2035 	reg->hr_key = crc32_le(reg->hr_region_num + O2NM_MAX_REGIONS,
2036 		name, strlen(name));
2037 	INIT_LIST_HEAD(&reg->hr_handler_list);
2038 	ret = o2net_register_handler(O2HB_NEGO_TIMEOUT_MSG, reg->hr_key,
2039 			sizeof(struct o2hb_nego_msg),
2040 			o2hb_nego_timeout_handler,
2041 			reg, NULL, &reg->hr_handler_list);
2042 	if (ret)
2043 		goto remove_item;
2044 
2045 	ret = o2net_register_handler(O2HB_NEGO_APPROVE_MSG, reg->hr_key,
2046 			sizeof(struct o2hb_nego_msg),
2047 			o2hb_nego_approve_handler,
2048 			reg, NULL, &reg->hr_handler_list);
2049 	if (ret)
2050 		goto unregister_handler;
2051 
2052 	o2hb_debug_region_init(reg, o2hb_debug_dir);
2053 
2054 	return &reg->hr_item;
2055 
2056 unregister_handler:
2057 	o2net_unregister_handler_list(&reg->hr_handler_list);
2058 remove_item:
2059 	spin_lock(&o2hb_live_lock);
2060 	list_del(&reg->hr_all_item);
2061 	if (o2hb_global_heartbeat_active())
2062 		clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2063 	spin_unlock(&o2hb_live_lock);
2064 free:
2065 	kfree(reg);
2066 	return ERR_PTR(ret);
2067 }
2068 
2069 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2070 					   struct config_item *item)
2071 {
2072 	struct task_struct *hb_task;
2073 	struct o2hb_region *reg = to_o2hb_region(item);
2074 	int quorum_region = 0;
2075 
2076 	/* stop the thread when the user removes the region dir */
2077 	spin_lock(&o2hb_live_lock);
2078 	hb_task = reg->hr_task;
2079 	reg->hr_task = NULL;
2080 	reg->hr_item_dropped = 1;
2081 	spin_unlock(&o2hb_live_lock);
2082 
2083 	if (hb_task)
2084 		kthread_stop(hb_task);
2085 
2086 	if (o2hb_global_heartbeat_active()) {
2087 		spin_lock(&o2hb_live_lock);
2088 		clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2089 		clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2090 		if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2091 			quorum_region = 1;
2092 		clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2093 		spin_unlock(&o2hb_live_lock);
2094 		printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2095 		       ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2096 			"stopped" : "start aborted"), config_item_name(item),
2097 		       reg->hr_dev_name);
2098 	}
2099 
2100 	/*
2101 	 * If we're racing a dev_write(), we need to wake them.  They will
2102 	 * check reg->hr_task
2103 	 */
2104 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
2105 		reg->hr_aborted_start = 1;
2106 		atomic_set(&reg->hr_steady_iterations, 0);
2107 		wake_up(&o2hb_steady_queue);
2108 	}
2109 
2110 	config_item_put(item);
2111 
2112 	if (!o2hb_global_heartbeat_active() || !quorum_region)
2113 		return;
2114 
2115 	/*
2116 	 * If global heartbeat active and there are dependent users,
2117 	 * pin all regions if quorum region count <= CUT_OFF
2118 	 */
2119 	spin_lock(&o2hb_live_lock);
2120 
2121 	if (!o2hb_dependent_users)
2122 		goto unlock;
2123 
2124 	if (bitmap_weight(o2hb_quorum_region_bitmap,
2125 			   O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2126 		o2hb_region_pin(NULL);
2127 
2128 unlock:
2129 	spin_unlock(&o2hb_live_lock);
2130 }
2131 
2132 static ssize_t o2hb_heartbeat_group_dead_threshold_show(struct config_item *item,
2133 		char *page)
2134 {
2135 	return sprintf(page, "%u\n", o2hb_dead_threshold);
2136 }
2137 
2138 static ssize_t o2hb_heartbeat_group_dead_threshold_store(struct config_item *item,
2139 		const char *page, size_t count)
2140 {
2141 	unsigned long tmp;
2142 	char *p = (char *)page;
2143 
2144 	tmp = simple_strtoul(p, &p, 10);
2145 	if (!p || (*p && (*p != '\n')))
2146                 return -EINVAL;
2147 
2148 	/* this will validate ranges for us. */
2149 	o2hb_dead_threshold_set((unsigned int) tmp);
2150 
2151 	return count;
2152 }
2153 
2154 static ssize_t o2hb_heartbeat_group_mode_show(struct config_item *item,
2155 		char *page)
2156 {
2157 	return sprintf(page, "%s\n",
2158 		       o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2159 }
2160 
2161 static ssize_t o2hb_heartbeat_group_mode_store(struct config_item *item,
2162 		const char *page, size_t count)
2163 {
2164 	unsigned int i;
2165 	int ret;
2166 	size_t len;
2167 
2168 	len = (page[count - 1] == '\n') ? count - 1 : count;
2169 	if (!len)
2170 		return -EINVAL;
2171 
2172 	for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2173 		if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
2174 			continue;
2175 
2176 		ret = o2hb_global_heartbeat_mode_set(i);
2177 		if (!ret)
2178 			printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2179 			       o2hb_heartbeat_mode_desc[i]);
2180 		return count;
2181 	}
2182 
2183 	return -EINVAL;
2184 
2185 }
2186 
2187 CONFIGFS_ATTR(o2hb_heartbeat_group_, dead_threshold);
2188 CONFIGFS_ATTR(o2hb_heartbeat_group_, mode);
2189 
2190 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2191 	&o2hb_heartbeat_group_attr_dead_threshold,
2192 	&o2hb_heartbeat_group_attr_mode,
2193 	NULL,
2194 };
2195 
2196 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2197 	.make_item	= o2hb_heartbeat_group_make_item,
2198 	.drop_item	= o2hb_heartbeat_group_drop_item,
2199 };
2200 
2201 static const struct config_item_type o2hb_heartbeat_group_type = {
2202 	.ct_group_ops	= &o2hb_heartbeat_group_group_ops,
2203 	.ct_attrs	= o2hb_heartbeat_group_attrs,
2204 	.ct_owner	= THIS_MODULE,
2205 };
2206 
2207 /* this is just here to avoid touching group in heartbeat.h which the
2208  * entire damn world #includes */
2209 struct config_group *o2hb_alloc_hb_set(void)
2210 {
2211 	struct o2hb_heartbeat_group *hs = NULL;
2212 	struct config_group *ret = NULL;
2213 
2214 	hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2215 	if (hs == NULL)
2216 		goto out;
2217 
2218 	config_group_init_type_name(&hs->hs_group, "heartbeat",
2219 				    &o2hb_heartbeat_group_type);
2220 
2221 	ret = &hs->hs_group;
2222 out:
2223 	if (ret == NULL)
2224 		kfree(hs);
2225 	return ret;
2226 }
2227 
2228 void o2hb_free_hb_set(struct config_group *group)
2229 {
2230 	struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2231 	kfree(hs);
2232 }
2233 
2234 /* hb callback registration and issuing */
2235 
2236 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2237 {
2238 	if (type == O2HB_NUM_CB)
2239 		return ERR_PTR(-EINVAL);
2240 
2241 	return &o2hb_callbacks[type];
2242 }
2243 
2244 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2245 			 enum o2hb_callback_type type,
2246 			 o2hb_cb_func *func,
2247 			 void *data,
2248 			 int priority)
2249 {
2250 	INIT_LIST_HEAD(&hc->hc_item);
2251 	hc->hc_func = func;
2252 	hc->hc_data = data;
2253 	hc->hc_priority = priority;
2254 	hc->hc_type = type;
2255 	hc->hc_magic = O2HB_CB_MAGIC;
2256 }
2257 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2258 
2259 /*
2260  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2261  * In global heartbeat mode, region_uuid passed is NULL.
2262  *
2263  * In local, we only pin the matching region. In global we pin all the active
2264  * regions.
2265  */
2266 static int o2hb_region_pin(const char *region_uuid)
2267 {
2268 	int ret = 0, found = 0;
2269 	struct o2hb_region *reg;
2270 	char *uuid;
2271 
2272 	assert_spin_locked(&o2hb_live_lock);
2273 
2274 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2275 		if (reg->hr_item_dropped)
2276 			continue;
2277 
2278 		uuid = config_item_name(&reg->hr_item);
2279 
2280 		/* local heartbeat */
2281 		if (region_uuid) {
2282 			if (strcmp(region_uuid, uuid))
2283 				continue;
2284 			found = 1;
2285 		}
2286 
2287 		if (reg->hr_item_pinned || reg->hr_item_dropped)
2288 			goto skip_pin;
2289 
2290 		/* Ignore ENOENT only for local hb (userdlm domain) */
2291 		ret = o2nm_depend_item(&reg->hr_item);
2292 		if (!ret) {
2293 			mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2294 			reg->hr_item_pinned = 1;
2295 		} else {
2296 			if (ret == -ENOENT && found)
2297 				ret = 0;
2298 			else {
2299 				mlog(ML_ERROR, "Pin region %s fails with %d\n",
2300 				     uuid, ret);
2301 				break;
2302 			}
2303 		}
2304 skip_pin:
2305 		if (found)
2306 			break;
2307 	}
2308 
2309 	return ret;
2310 }
2311 
2312 /*
2313  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2314  * In global heartbeat mode, region_uuid passed is NULL.
2315  *
2316  * In local, we only unpin the matching region. In global we unpin all the
2317  * active regions.
2318  */
2319 static void o2hb_region_unpin(const char *region_uuid)
2320 {
2321 	struct o2hb_region *reg;
2322 	char *uuid;
2323 	int found = 0;
2324 
2325 	assert_spin_locked(&o2hb_live_lock);
2326 
2327 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2328 		if (reg->hr_item_dropped)
2329 			continue;
2330 
2331 		uuid = config_item_name(&reg->hr_item);
2332 		if (region_uuid) {
2333 			if (strcmp(region_uuid, uuid))
2334 				continue;
2335 			found = 1;
2336 		}
2337 
2338 		if (reg->hr_item_pinned) {
2339 			mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2340 			o2nm_undepend_item(&reg->hr_item);
2341 			reg->hr_item_pinned = 0;
2342 		}
2343 		if (found)
2344 			break;
2345 	}
2346 }
2347 
2348 static int o2hb_region_inc_user(const char *region_uuid)
2349 {
2350 	int ret = 0;
2351 
2352 	spin_lock(&o2hb_live_lock);
2353 
2354 	/* local heartbeat */
2355 	if (!o2hb_global_heartbeat_active()) {
2356 	    ret = o2hb_region_pin(region_uuid);
2357 	    goto unlock;
2358 	}
2359 
2360 	/*
2361 	 * if global heartbeat active and this is the first dependent user,
2362 	 * pin all regions if quorum region count <= CUT_OFF
2363 	 */
2364 	o2hb_dependent_users++;
2365 	if (o2hb_dependent_users > 1)
2366 		goto unlock;
2367 
2368 	if (bitmap_weight(o2hb_quorum_region_bitmap,
2369 			   O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2370 		ret = o2hb_region_pin(NULL);
2371 
2372 unlock:
2373 	spin_unlock(&o2hb_live_lock);
2374 	return ret;
2375 }
2376 
2377 static void o2hb_region_dec_user(const char *region_uuid)
2378 {
2379 	spin_lock(&o2hb_live_lock);
2380 
2381 	/* local heartbeat */
2382 	if (!o2hb_global_heartbeat_active()) {
2383 	    o2hb_region_unpin(region_uuid);
2384 	    goto unlock;
2385 	}
2386 
2387 	/*
2388 	 * if global heartbeat active and there are no dependent users,
2389 	 * unpin all quorum regions
2390 	 */
2391 	o2hb_dependent_users--;
2392 	if (!o2hb_dependent_users)
2393 		o2hb_region_unpin(NULL);
2394 
2395 unlock:
2396 	spin_unlock(&o2hb_live_lock);
2397 }
2398 
2399 int o2hb_register_callback(const char *region_uuid,
2400 			   struct o2hb_callback_func *hc)
2401 {
2402 	struct o2hb_callback_func *f;
2403 	struct o2hb_callback *hbcall;
2404 	int ret;
2405 
2406 	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2407 	BUG_ON(!list_empty(&hc->hc_item));
2408 
2409 	hbcall = hbcall_from_type(hc->hc_type);
2410 	if (IS_ERR(hbcall)) {
2411 		ret = PTR_ERR(hbcall);
2412 		goto out;
2413 	}
2414 
2415 	if (region_uuid) {
2416 		ret = o2hb_region_inc_user(region_uuid);
2417 		if (ret) {
2418 			mlog_errno(ret);
2419 			goto out;
2420 		}
2421 	}
2422 
2423 	down_write(&o2hb_callback_sem);
2424 
2425 	list_for_each_entry(f, &hbcall->list, hc_item) {
2426 		if (hc->hc_priority < f->hc_priority) {
2427 			list_add_tail(&hc->hc_item, &f->hc_item);
2428 			break;
2429 		}
2430 	}
2431 	if (list_empty(&hc->hc_item))
2432 		list_add_tail(&hc->hc_item, &hbcall->list);
2433 
2434 	up_write(&o2hb_callback_sem);
2435 	ret = 0;
2436 out:
2437 	mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2438 	     ret, __builtin_return_address(0), hc);
2439 	return ret;
2440 }
2441 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2442 
2443 void o2hb_unregister_callback(const char *region_uuid,
2444 			      struct o2hb_callback_func *hc)
2445 {
2446 	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2447 
2448 	mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2449 	     __builtin_return_address(0), hc);
2450 
2451 	/* XXX Can this happen _with_ a region reference? */
2452 	if (list_empty(&hc->hc_item))
2453 		return;
2454 
2455 	if (region_uuid)
2456 		o2hb_region_dec_user(region_uuid);
2457 
2458 	down_write(&o2hb_callback_sem);
2459 
2460 	list_del_init(&hc->hc_item);
2461 
2462 	up_write(&o2hb_callback_sem);
2463 }
2464 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2465 
2466 int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2467 {
2468 	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2469 
2470 	spin_lock(&o2hb_live_lock);
2471 	o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2472 	spin_unlock(&o2hb_live_lock);
2473 	if (!test_bit(node_num, testing_map)) {
2474 		mlog(ML_HEARTBEAT,
2475 		     "node (%u) does not have heartbeating enabled.\n",
2476 		     node_num);
2477 		return 0;
2478 	}
2479 
2480 	return 1;
2481 }
2482 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2483 
2484 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2485 {
2486 	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2487 
2488 	o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2489 	if (!test_bit(node_num, testing_map)) {
2490 		mlog(ML_HEARTBEAT,
2491 		     "node (%u) does not have heartbeating enabled.\n",
2492 		     node_num);
2493 		return 0;
2494 	}
2495 
2496 	return 1;
2497 }
2498 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2499 
2500 /*
2501  * this is just a hack until we get the plumbing which flips file systems
2502  * read only and drops the hb ref instead of killing the node dead.
2503  */
2504 void o2hb_stop_all_regions(void)
2505 {
2506 	struct o2hb_region *reg;
2507 
2508 	mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2509 
2510 	spin_lock(&o2hb_live_lock);
2511 
2512 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2513 		reg->hr_unclean_stop = 1;
2514 
2515 	spin_unlock(&o2hb_live_lock);
2516 }
2517 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2518 
2519 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2520 {
2521 	struct o2hb_region *reg;
2522 	int numregs = 0;
2523 	char *p;
2524 
2525 	spin_lock(&o2hb_live_lock);
2526 
2527 	p = region_uuids;
2528 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2529 		if (reg->hr_item_dropped)
2530 			continue;
2531 
2532 		mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2533 		if (numregs < max_regions) {
2534 			memcpy(p, config_item_name(&reg->hr_item),
2535 			       O2HB_MAX_REGION_NAME_LEN);
2536 			p += O2HB_MAX_REGION_NAME_LEN;
2537 		}
2538 		numregs++;
2539 	}
2540 
2541 	spin_unlock(&o2hb_live_lock);
2542 
2543 	return numregs;
2544 }
2545 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2546 
2547 int o2hb_global_heartbeat_active(void)
2548 {
2549 	return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2550 }
2551 EXPORT_SYMBOL(o2hb_global_heartbeat_active);
2552