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 
39 #include "heartbeat.h"
40 #include "tcp.h"
41 #include "nodemanager.h"
42 #include "quorum.h"
43 
44 #include "masklog.h"
45 
46 
47 /*
48  * The first heartbeat pass had one global thread that would serialize all hb
49  * callback calls.  This global serializing sem should only be removed once
50  * we've made sure that all callees can deal with being called concurrently
51  * from multiple hb region threads.
52  */
53 static DECLARE_RWSEM(o2hb_callback_sem);
54 
55 /*
56  * multiple hb threads are watching multiple regions.  A node is live
57  * whenever any of the threads sees activity from the node in its region.
58  */
59 static DEFINE_SPINLOCK(o2hb_live_lock);
60 static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
61 static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
62 static LIST_HEAD(o2hb_node_events);
63 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
64 
65 /*
66  * In global heartbeat, we maintain a series of region bitmaps.
67  * 	- o2hb_region_bitmap allows us to limit the region number to max region.
68  * 	- o2hb_live_region_bitmap tracks live regions (seen steady iterations).
69  * 	- o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
70  * 		heartbeat on it.
71  * 	- o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
72  */
73 static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
74 static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
75 static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
76 static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
77 
78 #define O2HB_DB_TYPE_LIVENODES		0
79 #define O2HB_DB_TYPE_LIVEREGIONS	1
80 #define O2HB_DB_TYPE_QUORUMREGIONS	2
81 #define O2HB_DB_TYPE_FAILEDREGIONS	3
82 #define O2HB_DB_TYPE_REGION_LIVENODES	4
83 #define O2HB_DB_TYPE_REGION_NUMBER	5
84 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME	6
85 #define O2HB_DB_TYPE_REGION_PINNED	7
86 struct o2hb_debug_buf {
87 	int db_type;
88 	int db_size;
89 	int db_len;
90 	void *db_data;
91 };
92 
93 static struct o2hb_debug_buf *o2hb_db_livenodes;
94 static struct o2hb_debug_buf *o2hb_db_liveregions;
95 static struct o2hb_debug_buf *o2hb_db_quorumregions;
96 static struct o2hb_debug_buf *o2hb_db_failedregions;
97 
98 #define O2HB_DEBUG_DIR			"o2hb"
99 #define O2HB_DEBUG_LIVENODES		"livenodes"
100 #define O2HB_DEBUG_LIVEREGIONS		"live_regions"
101 #define O2HB_DEBUG_QUORUMREGIONS	"quorum_regions"
102 #define O2HB_DEBUG_FAILEDREGIONS	"failed_regions"
103 #define O2HB_DEBUG_REGION_NUMBER	"num"
104 #define O2HB_DEBUG_REGION_ELAPSED_TIME	"elapsed_time_in_ms"
105 #define O2HB_DEBUG_REGION_PINNED	"pinned"
106 
107 static struct dentry *o2hb_debug_dir;
108 static struct dentry *o2hb_debug_livenodes;
109 static struct dentry *o2hb_debug_liveregions;
110 static struct dentry *o2hb_debug_quorumregions;
111 static struct dentry *o2hb_debug_failedregions;
112 
113 static LIST_HEAD(o2hb_all_regions);
114 
115 static struct o2hb_callback {
116 	struct list_head list;
117 } o2hb_callbacks[O2HB_NUM_CB];
118 
119 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
120 
121 #define O2HB_DEFAULT_BLOCK_BITS       9
122 
123 enum o2hb_heartbeat_modes {
124 	O2HB_HEARTBEAT_LOCAL		= 0,
125 	O2HB_HEARTBEAT_GLOBAL,
126 	O2HB_HEARTBEAT_NUM_MODES,
127 };
128 
129 char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
130 		"local",	/* O2HB_HEARTBEAT_LOCAL */
131 		"global",	/* O2HB_HEARTBEAT_GLOBAL */
132 };
133 
134 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
135 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
136 
137 /*
138  * o2hb_dependent_users tracks the number of registered callbacks that depend
139  * on heartbeat. o2net and o2dlm are two entities that register this callback.
140  * However only o2dlm depends on the heartbeat. It does not want the heartbeat
141  * to stop while a dlm domain is still active.
142  */
143 unsigned int o2hb_dependent_users;
144 
145 /*
146  * In global heartbeat mode, all regions are pinned if there are one or more
147  * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
148  * regions are unpinned if the region count exceeds the cut off or the number
149  * of dependent users falls to zero.
150  */
151 #define O2HB_PIN_CUT_OFF		3
152 
153 /*
154  * In local heartbeat mode, we assume the dlm domain name to be the same as
155  * region uuid. This is true for domains created for the file system but not
156  * necessarily true for userdlm domains. This is a known limitation.
157  *
158  * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
159  * works for both file system and userdlm domains.
160  */
161 static int o2hb_region_pin(const char *region_uuid);
162 static void o2hb_region_unpin(const char *region_uuid);
163 
164 /* Only sets a new threshold if there are no active regions.
165  *
166  * No locking or otherwise interesting code is required for reading
167  * o2hb_dead_threshold as it can't change once regions are active and
168  * it's not interesting to anyone until then anyway. */
o2hb_dead_threshold_set(unsigned int threshold)169 static void o2hb_dead_threshold_set(unsigned int threshold)
170 {
171 	if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
172 		spin_lock(&o2hb_live_lock);
173 		if (list_empty(&o2hb_all_regions))
174 			o2hb_dead_threshold = threshold;
175 		spin_unlock(&o2hb_live_lock);
176 	}
177 }
178 
o2hb_global_hearbeat_mode_set(unsigned int hb_mode)179 static int o2hb_global_hearbeat_mode_set(unsigned int hb_mode)
180 {
181 	int ret = -1;
182 
183 	if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
184 		spin_lock(&o2hb_live_lock);
185 		if (list_empty(&o2hb_all_regions)) {
186 			o2hb_heartbeat_mode = hb_mode;
187 			ret = 0;
188 		}
189 		spin_unlock(&o2hb_live_lock);
190 	}
191 
192 	return ret;
193 }
194 
195 struct o2hb_node_event {
196 	struct list_head        hn_item;
197 	enum o2hb_callback_type hn_event_type;
198 	struct o2nm_node        *hn_node;
199 	int                     hn_node_num;
200 };
201 
202 struct o2hb_disk_slot {
203 	struct o2hb_disk_heartbeat_block *ds_raw_block;
204 	u8			ds_node_num;
205 	u64			ds_last_time;
206 	u64			ds_last_generation;
207 	u16			ds_equal_samples;
208 	u16			ds_changed_samples;
209 	struct list_head	ds_live_item;
210 };
211 
212 /* each thread owns a region.. when we're asked to tear down the region
213  * we ask the thread to stop, who cleans up the region */
214 struct o2hb_region {
215 	struct config_item	hr_item;
216 
217 	struct list_head	hr_all_item;
218 	unsigned		hr_unclean_stop:1,
219 				hr_aborted_start:1,
220 				hr_item_pinned:1,
221 				hr_item_dropped:1;
222 
223 	/* protected by the hr_callback_sem */
224 	struct task_struct 	*hr_task;
225 
226 	unsigned int		hr_blocks;
227 	unsigned long long	hr_start_block;
228 
229 	unsigned int		hr_block_bits;
230 	unsigned int		hr_block_bytes;
231 
232 	unsigned int		hr_slots_per_page;
233 	unsigned int		hr_num_pages;
234 
235 	struct page             **hr_slot_data;
236 	struct block_device	*hr_bdev;
237 	struct o2hb_disk_slot	*hr_slots;
238 
239 	/* live node map of this region */
240 	unsigned long		hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
241 	unsigned int		hr_region_num;
242 
243 	struct dentry		*hr_debug_dir;
244 	struct dentry		*hr_debug_livenodes;
245 	struct dentry		*hr_debug_regnum;
246 	struct dentry		*hr_debug_elapsed_time;
247 	struct dentry		*hr_debug_pinned;
248 	struct o2hb_debug_buf	*hr_db_livenodes;
249 	struct o2hb_debug_buf	*hr_db_regnum;
250 	struct o2hb_debug_buf	*hr_db_elapsed_time;
251 	struct o2hb_debug_buf	*hr_db_pinned;
252 
253 	/* let the person setting up hb wait for it to return until it
254 	 * has reached a 'steady' state.  This will be fixed when we have
255 	 * a more complete api that doesn't lead to this sort of fragility. */
256 	atomic_t		hr_steady_iterations;
257 
258 	/* terminate o2hb thread if it does not reach steady state
259 	 * (hr_steady_iterations == 0) within hr_unsteady_iterations */
260 	atomic_t		hr_unsteady_iterations;
261 
262 	char			hr_dev_name[BDEVNAME_SIZE];
263 
264 	unsigned int		hr_timeout_ms;
265 
266 	/* randomized as the region goes up and down so that a node
267 	 * recognizes a node going up and down in one iteration */
268 	u64			hr_generation;
269 
270 	struct delayed_work	hr_write_timeout_work;
271 	unsigned long		hr_last_timeout_start;
272 
273 	/* Used during o2hb_check_slot to hold a copy of the block
274 	 * being checked because we temporarily have to zero out the
275 	 * crc field. */
276 	struct o2hb_disk_heartbeat_block *hr_tmp_block;
277 };
278 
279 struct o2hb_bio_wait_ctxt {
280 	atomic_t          wc_num_reqs;
281 	struct completion wc_io_complete;
282 	int               wc_error;
283 };
284 
o2hb_pop_count(void * map,int count)285 static int o2hb_pop_count(void *map, int count)
286 {
287 	int i = -1, pop = 0;
288 
289 	while ((i = find_next_bit(map, count, i + 1)) < count)
290 		pop++;
291 	return pop;
292 }
293 
o2hb_write_timeout(struct work_struct * work)294 static void o2hb_write_timeout(struct work_struct *work)
295 {
296 	int failed, quorum;
297 	unsigned long flags;
298 	struct o2hb_region *reg =
299 		container_of(work, struct o2hb_region,
300 			     hr_write_timeout_work.work);
301 
302 	mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
303 	     "milliseconds\n", reg->hr_dev_name,
304 	     jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
305 
306 	if (o2hb_global_heartbeat_active()) {
307 		spin_lock_irqsave(&o2hb_live_lock, flags);
308 		if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
309 			set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
310 		failed = o2hb_pop_count(&o2hb_failed_region_bitmap,
311 					O2NM_MAX_REGIONS);
312 		quorum = o2hb_pop_count(&o2hb_quorum_region_bitmap,
313 					O2NM_MAX_REGIONS);
314 		spin_unlock_irqrestore(&o2hb_live_lock, flags);
315 
316 		mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
317 		     quorum, failed);
318 
319 		/*
320 		 * Fence if the number of failed regions >= half the number
321 		 * of  quorum regions
322 		 */
323 		if ((failed << 1) < quorum)
324 			return;
325 	}
326 
327 	o2quo_disk_timeout();
328 }
329 
o2hb_arm_write_timeout(struct o2hb_region * reg)330 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
331 {
332 	/* Arm writeout only after thread reaches steady state */
333 	if (atomic_read(&reg->hr_steady_iterations) != 0)
334 		return;
335 
336 	mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
337 	     O2HB_MAX_WRITE_TIMEOUT_MS);
338 
339 	if (o2hb_global_heartbeat_active()) {
340 		spin_lock(&o2hb_live_lock);
341 		clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
342 		spin_unlock(&o2hb_live_lock);
343 	}
344 	cancel_delayed_work(&reg->hr_write_timeout_work);
345 	reg->hr_last_timeout_start = jiffies;
346 	schedule_delayed_work(&reg->hr_write_timeout_work,
347 			      msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
348 }
349 
o2hb_disarm_write_timeout(struct o2hb_region * reg)350 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
351 {
352 	cancel_delayed_work_sync(&reg->hr_write_timeout_work);
353 }
354 
o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt * wc)355 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
356 {
357 	atomic_set(&wc->wc_num_reqs, 1);
358 	init_completion(&wc->wc_io_complete);
359 	wc->wc_error = 0;
360 }
361 
362 /* Used in error paths too */
o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt * wc,unsigned int num)363 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
364 				     unsigned int num)
365 {
366 	/* sadly atomic_sub_and_test() isn't available on all platforms.  The
367 	 * good news is that the fast path only completes one at a time */
368 	while(num--) {
369 		if (atomic_dec_and_test(&wc->wc_num_reqs)) {
370 			BUG_ON(num > 0);
371 			complete(&wc->wc_io_complete);
372 		}
373 	}
374 }
375 
o2hb_wait_on_io(struct o2hb_region * reg,struct o2hb_bio_wait_ctxt * wc)376 static void o2hb_wait_on_io(struct o2hb_region *reg,
377 			    struct o2hb_bio_wait_ctxt *wc)
378 {
379 	o2hb_bio_wait_dec(wc, 1);
380 	wait_for_completion(&wc->wc_io_complete);
381 }
382 
o2hb_bio_end_io(struct bio * bio,int error)383 static void o2hb_bio_end_io(struct bio *bio,
384 			   int error)
385 {
386 	struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
387 
388 	if (error) {
389 		mlog(ML_ERROR, "IO Error %d\n", error);
390 		wc->wc_error = error;
391 	}
392 
393 	o2hb_bio_wait_dec(wc, 1);
394 	bio_put(bio);
395 }
396 
397 /* Setup a Bio to cover I/O against num_slots slots starting at
398  * start_slot. */
o2hb_setup_one_bio(struct o2hb_region * reg,struct o2hb_bio_wait_ctxt * wc,unsigned int * current_slot,unsigned int max_slots)399 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
400 				      struct o2hb_bio_wait_ctxt *wc,
401 				      unsigned int *current_slot,
402 				      unsigned int max_slots)
403 {
404 	int len, current_page;
405 	unsigned int vec_len, vec_start;
406 	unsigned int bits = reg->hr_block_bits;
407 	unsigned int spp = reg->hr_slots_per_page;
408 	unsigned int cs = *current_slot;
409 	struct bio *bio;
410 	struct page *page;
411 
412 	/* Testing has shown this allocation to take long enough under
413 	 * GFP_KERNEL that the local node can get fenced. It would be
414 	 * nicest if we could pre-allocate these bios and avoid this
415 	 * all together. */
416 	bio = bio_alloc(GFP_ATOMIC, 16);
417 	if (!bio) {
418 		mlog(ML_ERROR, "Could not alloc slots BIO!\n");
419 		bio = ERR_PTR(-ENOMEM);
420 		goto bail;
421 	}
422 
423 	/* Must put everything in 512 byte sectors for the bio... */
424 	bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
425 	bio->bi_bdev = reg->hr_bdev;
426 	bio->bi_private = wc;
427 	bio->bi_end_io = o2hb_bio_end_io;
428 
429 	vec_start = (cs << bits) % PAGE_CACHE_SIZE;
430 	while(cs < max_slots) {
431 		current_page = cs / spp;
432 		page = reg->hr_slot_data[current_page];
433 
434 		vec_len = min(PAGE_CACHE_SIZE - vec_start,
435 			      (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
436 
437 		mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
438 		     current_page, vec_len, vec_start);
439 
440 		len = bio_add_page(bio, page, vec_len, vec_start);
441 		if (len != vec_len) break;
442 
443 		cs += vec_len / (PAGE_CACHE_SIZE/spp);
444 		vec_start = 0;
445 	}
446 
447 bail:
448 	*current_slot = cs;
449 	return bio;
450 }
451 
o2hb_read_slots(struct o2hb_region * reg,unsigned int max_slots)452 static int o2hb_read_slots(struct o2hb_region *reg,
453 			   unsigned int max_slots)
454 {
455 	unsigned int current_slot=0;
456 	int status;
457 	struct o2hb_bio_wait_ctxt wc;
458 	struct bio *bio;
459 
460 	o2hb_bio_wait_init(&wc);
461 
462 	while(current_slot < max_slots) {
463 		bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
464 		if (IS_ERR(bio)) {
465 			status = PTR_ERR(bio);
466 			mlog_errno(status);
467 			goto bail_and_wait;
468 		}
469 
470 		atomic_inc(&wc.wc_num_reqs);
471 		submit_bio(READ, bio);
472 	}
473 
474 	status = 0;
475 
476 bail_and_wait:
477 	o2hb_wait_on_io(reg, &wc);
478 	if (wc.wc_error && !status)
479 		status = wc.wc_error;
480 
481 	return status;
482 }
483 
o2hb_issue_node_write(struct o2hb_region * reg,struct o2hb_bio_wait_ctxt * write_wc)484 static int o2hb_issue_node_write(struct o2hb_region *reg,
485 				 struct o2hb_bio_wait_ctxt *write_wc)
486 {
487 	int status;
488 	unsigned int slot;
489 	struct bio *bio;
490 
491 	o2hb_bio_wait_init(write_wc);
492 
493 	slot = o2nm_this_node();
494 
495 	bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
496 	if (IS_ERR(bio)) {
497 		status = PTR_ERR(bio);
498 		mlog_errno(status);
499 		goto bail;
500 	}
501 
502 	atomic_inc(&write_wc->wc_num_reqs);
503 	submit_bio(WRITE, bio);
504 
505 	status = 0;
506 bail:
507 	return status;
508 }
509 
o2hb_compute_block_crc_le(struct o2hb_region * reg,struct o2hb_disk_heartbeat_block * hb_block)510 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
511 				     struct o2hb_disk_heartbeat_block *hb_block)
512 {
513 	__le32 old_cksum;
514 	u32 ret;
515 
516 	/* We want to compute the block crc with a 0 value in the
517 	 * hb_cksum field. Save it off here and replace after the
518 	 * crc. */
519 	old_cksum = hb_block->hb_cksum;
520 	hb_block->hb_cksum = 0;
521 
522 	ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
523 
524 	hb_block->hb_cksum = old_cksum;
525 
526 	return ret;
527 }
528 
o2hb_dump_slot(struct o2hb_disk_heartbeat_block * hb_block)529 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
530 {
531 	mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
532 	     "cksum = 0x%x, generation 0x%llx\n",
533 	     (long long)le64_to_cpu(hb_block->hb_seq),
534 	     hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
535 	     (long long)le64_to_cpu(hb_block->hb_generation));
536 }
537 
o2hb_verify_crc(struct o2hb_region * reg,struct o2hb_disk_heartbeat_block * hb_block)538 static int o2hb_verify_crc(struct o2hb_region *reg,
539 			   struct o2hb_disk_heartbeat_block *hb_block)
540 {
541 	u32 read, computed;
542 
543 	read = le32_to_cpu(hb_block->hb_cksum);
544 	computed = o2hb_compute_block_crc_le(reg, hb_block);
545 
546 	return read == computed;
547 }
548 
549 /*
550  * Compare the slot data with what we wrote in the last iteration.
551  * If the match fails, print an appropriate error message. This is to
552  * detect errors like... another node hearting on the same slot,
553  * flaky device that is losing writes, etc.
554  * Returns 1 if check succeeds, 0 otherwise.
555  */
o2hb_check_own_slot(struct o2hb_region * reg)556 static int o2hb_check_own_slot(struct o2hb_region *reg)
557 {
558 	struct o2hb_disk_slot *slot;
559 	struct o2hb_disk_heartbeat_block *hb_block;
560 	char *errstr;
561 
562 	slot = &reg->hr_slots[o2nm_this_node()];
563 	/* Don't check on our 1st timestamp */
564 	if (!slot->ds_last_time)
565 		return 0;
566 
567 	hb_block = slot->ds_raw_block;
568 	if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
569 	    le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
570 	    hb_block->hb_node == slot->ds_node_num)
571 		return 1;
572 
573 #define ERRSTR1		"Another node is heartbeating on device"
574 #define ERRSTR2		"Heartbeat generation mismatch on device"
575 #define ERRSTR3		"Heartbeat sequence mismatch on device"
576 
577 	if (hb_block->hb_node != slot->ds_node_num)
578 		errstr = ERRSTR1;
579 	else if (le64_to_cpu(hb_block->hb_generation) !=
580 		 slot->ds_last_generation)
581 		errstr = ERRSTR2;
582 	else
583 		errstr = ERRSTR3;
584 
585 	mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
586 	     "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
587 	     slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
588 	     (unsigned long long)slot->ds_last_time, hb_block->hb_node,
589 	     (unsigned long long)le64_to_cpu(hb_block->hb_generation),
590 	     (unsigned long long)le64_to_cpu(hb_block->hb_seq));
591 
592 	return 0;
593 }
594 
o2hb_prepare_block(struct o2hb_region * reg,u64 generation)595 static inline void o2hb_prepare_block(struct o2hb_region *reg,
596 				      u64 generation)
597 {
598 	int node_num;
599 	u64 cputime;
600 	struct o2hb_disk_slot *slot;
601 	struct o2hb_disk_heartbeat_block *hb_block;
602 
603 	node_num = o2nm_this_node();
604 	slot = &reg->hr_slots[node_num];
605 
606 	hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
607 	memset(hb_block, 0, reg->hr_block_bytes);
608 	/* TODO: time stuff */
609 	cputime = CURRENT_TIME.tv_sec;
610 	if (!cputime)
611 		cputime = 1;
612 
613 	hb_block->hb_seq = cpu_to_le64(cputime);
614 	hb_block->hb_node = node_num;
615 	hb_block->hb_generation = cpu_to_le64(generation);
616 	hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
617 
618 	/* This step must always happen last! */
619 	hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
620 								   hb_block));
621 
622 	mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
623 	     (long long)generation,
624 	     le32_to_cpu(hb_block->hb_cksum));
625 }
626 
o2hb_fire_callbacks(struct o2hb_callback * hbcall,struct o2nm_node * node,int idx)627 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
628 				struct o2nm_node *node,
629 				int idx)
630 {
631 	struct list_head *iter;
632 	struct o2hb_callback_func *f;
633 
634 	list_for_each(iter, &hbcall->list) {
635 		f = list_entry(iter, struct o2hb_callback_func, hc_item);
636 		mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
637 		(f->hc_func)(node, idx, f->hc_data);
638 	}
639 }
640 
641 /* Will run the list in order until we process the passed event */
o2hb_run_event_list(struct o2hb_node_event * queued_event)642 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
643 {
644 	int empty;
645 	struct o2hb_callback *hbcall;
646 	struct o2hb_node_event *event;
647 
648 	spin_lock(&o2hb_live_lock);
649 	empty = list_empty(&queued_event->hn_item);
650 	spin_unlock(&o2hb_live_lock);
651 	if (empty)
652 		return;
653 
654 	/* Holding callback sem assures we don't alter the callback
655 	 * lists when doing this, and serializes ourselves with other
656 	 * processes wanting callbacks. */
657 	down_write(&o2hb_callback_sem);
658 
659 	spin_lock(&o2hb_live_lock);
660 	while (!list_empty(&o2hb_node_events)
661 	       && !list_empty(&queued_event->hn_item)) {
662 		event = list_entry(o2hb_node_events.next,
663 				   struct o2hb_node_event,
664 				   hn_item);
665 		list_del_init(&event->hn_item);
666 		spin_unlock(&o2hb_live_lock);
667 
668 		mlog(ML_HEARTBEAT, "Node %s event for %d\n",
669 		     event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
670 		     event->hn_node_num);
671 
672 		hbcall = hbcall_from_type(event->hn_event_type);
673 
674 		/* We should *never* have gotten on to the list with a
675 		 * bad type... This isn't something that we should try
676 		 * to recover from. */
677 		BUG_ON(IS_ERR(hbcall));
678 
679 		o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
680 
681 		spin_lock(&o2hb_live_lock);
682 	}
683 	spin_unlock(&o2hb_live_lock);
684 
685 	up_write(&o2hb_callback_sem);
686 }
687 
o2hb_queue_node_event(struct o2hb_node_event * event,enum o2hb_callback_type type,struct o2nm_node * node,int node_num)688 static void o2hb_queue_node_event(struct o2hb_node_event *event,
689 				  enum o2hb_callback_type type,
690 				  struct o2nm_node *node,
691 				  int node_num)
692 {
693 	assert_spin_locked(&o2hb_live_lock);
694 
695 	BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
696 
697 	event->hn_event_type = type;
698 	event->hn_node = node;
699 	event->hn_node_num = node_num;
700 
701 	mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
702 	     type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
703 
704 	list_add_tail(&event->hn_item, &o2hb_node_events);
705 }
706 
o2hb_shutdown_slot(struct o2hb_disk_slot * slot)707 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
708 {
709 	struct o2hb_node_event event =
710 		{ .hn_item = LIST_HEAD_INIT(event.hn_item), };
711 	struct o2nm_node *node;
712 
713 	node = o2nm_get_node_by_num(slot->ds_node_num);
714 	if (!node)
715 		return;
716 
717 	spin_lock(&o2hb_live_lock);
718 	if (!list_empty(&slot->ds_live_item)) {
719 		mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
720 		     slot->ds_node_num);
721 
722 		list_del_init(&slot->ds_live_item);
723 
724 		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
725 			clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
726 
727 			o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
728 					      slot->ds_node_num);
729 		}
730 	}
731 	spin_unlock(&o2hb_live_lock);
732 
733 	o2hb_run_event_list(&event);
734 
735 	o2nm_node_put(node);
736 }
737 
o2hb_set_quorum_device(struct o2hb_region * reg)738 static void o2hb_set_quorum_device(struct o2hb_region *reg)
739 {
740 	if (!o2hb_global_heartbeat_active())
741 		return;
742 
743 	/* Prevent race with o2hb_heartbeat_group_drop_item() */
744 	if (kthread_should_stop())
745 		return;
746 
747 	/* Tag region as quorum only after thread reaches steady state */
748 	if (atomic_read(&reg->hr_steady_iterations) != 0)
749 		return;
750 
751 	spin_lock(&o2hb_live_lock);
752 
753 	if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
754 		goto unlock;
755 
756 	/*
757 	 * A region can be added to the quorum only when it sees all
758 	 * live nodes heartbeat on it. In other words, the region has been
759 	 * added to all nodes.
760 	 */
761 	if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
762 		   sizeof(o2hb_live_node_bitmap)))
763 		goto unlock;
764 
765 	printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
766 	       config_item_name(&reg->hr_item), reg->hr_dev_name);
767 
768 	set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
769 
770 	/*
771 	 * If global heartbeat active, unpin all regions if the
772 	 * region count > CUT_OFF
773 	 */
774 	if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
775 			   O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
776 		o2hb_region_unpin(NULL);
777 unlock:
778 	spin_unlock(&o2hb_live_lock);
779 }
780 
o2hb_check_slot(struct o2hb_region * reg,struct o2hb_disk_slot * slot)781 static int o2hb_check_slot(struct o2hb_region *reg,
782 			   struct o2hb_disk_slot *slot)
783 {
784 	int changed = 0, gen_changed = 0;
785 	struct o2hb_node_event event =
786 		{ .hn_item = LIST_HEAD_INIT(event.hn_item), };
787 	struct o2nm_node *node;
788 	struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
789 	u64 cputime;
790 	unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
791 	unsigned int slot_dead_ms;
792 	int tmp;
793 
794 	memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
795 
796 	/*
797 	 * If a node is no longer configured but is still in the livemap, we
798 	 * may need to clear that bit from the livemap.
799 	 */
800 	node = o2nm_get_node_by_num(slot->ds_node_num);
801 	if (!node) {
802 		spin_lock(&o2hb_live_lock);
803 		tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
804 		spin_unlock(&o2hb_live_lock);
805 		if (!tmp)
806 			return 0;
807 	}
808 
809 	if (!o2hb_verify_crc(reg, hb_block)) {
810 		/* all paths from here will drop o2hb_live_lock for
811 		 * us. */
812 		spin_lock(&o2hb_live_lock);
813 
814 		/* Don't print an error on the console in this case -
815 		 * a freshly formatted heartbeat area will not have a
816 		 * crc set on it. */
817 		if (list_empty(&slot->ds_live_item))
818 			goto out;
819 
820 		/* The node is live but pushed out a bad crc. We
821 		 * consider it a transient miss but don't populate any
822 		 * other values as they may be junk. */
823 		mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
824 		     slot->ds_node_num, reg->hr_dev_name);
825 		o2hb_dump_slot(hb_block);
826 
827 		slot->ds_equal_samples++;
828 		goto fire_callbacks;
829 	}
830 
831 	/* we don't care if these wrap.. the state transitions below
832 	 * clear at the right places */
833 	cputime = le64_to_cpu(hb_block->hb_seq);
834 	if (slot->ds_last_time != cputime)
835 		slot->ds_changed_samples++;
836 	else
837 		slot->ds_equal_samples++;
838 	slot->ds_last_time = cputime;
839 
840 	/* The node changed heartbeat generations. We assume this to
841 	 * mean it dropped off but came back before we timed out. We
842 	 * want to consider it down for the time being but don't want
843 	 * to lose any changed_samples state we might build up to
844 	 * considering it live again. */
845 	if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
846 		gen_changed = 1;
847 		slot->ds_equal_samples = 0;
848 		mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
849 		     "to 0x%llx)\n", slot->ds_node_num,
850 		     (long long)slot->ds_last_generation,
851 		     (long long)le64_to_cpu(hb_block->hb_generation));
852 	}
853 
854 	slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
855 
856 	mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
857 	     "seq %llu last %llu changed %u equal %u\n",
858 	     slot->ds_node_num, (long long)slot->ds_last_generation,
859 	     le32_to_cpu(hb_block->hb_cksum),
860 	     (unsigned long long)le64_to_cpu(hb_block->hb_seq),
861 	     (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
862 	     slot->ds_equal_samples);
863 
864 	spin_lock(&o2hb_live_lock);
865 
866 fire_callbacks:
867 	/* dead nodes only come to life after some number of
868 	 * changes at any time during their dead time */
869 	if (list_empty(&slot->ds_live_item) &&
870 	    slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
871 		mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
872 		     slot->ds_node_num, (long long)slot->ds_last_generation);
873 
874 		set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
875 
876 		/* first on the list generates a callback */
877 		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
878 			mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
879 			     "bitmap\n", slot->ds_node_num);
880 			set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
881 
882 			o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
883 					      slot->ds_node_num);
884 
885 			changed = 1;
886 		}
887 
888 		list_add_tail(&slot->ds_live_item,
889 			      &o2hb_live_slots[slot->ds_node_num]);
890 
891 		slot->ds_equal_samples = 0;
892 
893 		/* We want to be sure that all nodes agree on the
894 		 * number of milliseconds before a node will be
895 		 * considered dead. The self-fencing timeout is
896 		 * computed from this value, and a discrepancy might
897 		 * result in heartbeat calling a node dead when it
898 		 * hasn't self-fenced yet. */
899 		slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
900 		if (slot_dead_ms && slot_dead_ms != dead_ms) {
901 			/* TODO: Perhaps we can fail the region here. */
902 			mlog(ML_ERROR, "Node %d on device %s has a dead count "
903 			     "of %u ms, but our count is %u ms.\n"
904 			     "Please double check your configuration values "
905 			     "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
906 			     slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
907 			     dead_ms);
908 		}
909 		goto out;
910 	}
911 
912 	/* if the list is dead, we're done.. */
913 	if (list_empty(&slot->ds_live_item))
914 		goto out;
915 
916 	/* live nodes only go dead after enough consequtive missed
917 	 * samples..  reset the missed counter whenever we see
918 	 * activity */
919 	if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
920 		mlog(ML_HEARTBEAT, "Node %d left my region\n",
921 		     slot->ds_node_num);
922 
923 		clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
924 
925 		/* last off the live_slot generates a callback */
926 		list_del_init(&slot->ds_live_item);
927 		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
928 			mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
929 			     "nodes bitmap\n", slot->ds_node_num);
930 			clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
931 
932 			/* node can be null */
933 			o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
934 					      node, slot->ds_node_num);
935 
936 			changed = 1;
937 		}
938 
939 		/* We don't clear this because the node is still
940 		 * actually writing new blocks. */
941 		if (!gen_changed)
942 			slot->ds_changed_samples = 0;
943 		goto out;
944 	}
945 	if (slot->ds_changed_samples) {
946 		slot->ds_changed_samples = 0;
947 		slot->ds_equal_samples = 0;
948 	}
949 out:
950 	spin_unlock(&o2hb_live_lock);
951 
952 	o2hb_run_event_list(&event);
953 
954 	if (node)
955 		o2nm_node_put(node);
956 	return changed;
957 }
958 
959 /* This could be faster if we just implmented a find_last_bit, but I
960  * don't think the circumstances warrant it. */
o2hb_highest_node(unsigned long * nodes,int numbits)961 static int o2hb_highest_node(unsigned long *nodes,
962 			     int numbits)
963 {
964 	int highest, node;
965 
966 	highest = numbits;
967 	node = -1;
968 	while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
969 		if (node >= numbits)
970 			break;
971 
972 		highest = node;
973 	}
974 
975 	return highest;
976 }
977 
o2hb_do_disk_heartbeat(struct o2hb_region * reg)978 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
979 {
980 	int i, ret, highest_node;
981 	int membership_change = 0, own_slot_ok = 0;
982 	unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
983 	unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
984 	struct o2hb_bio_wait_ctxt write_wc;
985 
986 	ret = o2nm_configured_node_map(configured_nodes,
987 				       sizeof(configured_nodes));
988 	if (ret) {
989 		mlog_errno(ret);
990 		goto bail;
991 	}
992 
993 	/*
994 	 * If a node is not configured but is in the livemap, we still need
995 	 * to read the slot so as to be able to remove it from the livemap.
996 	 */
997 	o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
998 	i = -1;
999 	while ((i = find_next_bit(live_node_bitmap,
1000 				  O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1001 		set_bit(i, configured_nodes);
1002 	}
1003 
1004 	highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1005 	if (highest_node >= O2NM_MAX_NODES) {
1006 		mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1007 		ret = -EINVAL;
1008 		goto bail;
1009 	}
1010 
1011 	/* No sense in reading the slots of nodes that don't exist
1012 	 * yet. Of course, if the node definitions have holes in them
1013 	 * then we're reading an empty slot anyway... Consider this
1014 	 * best-effort. */
1015 	ret = o2hb_read_slots(reg, highest_node + 1);
1016 	if (ret < 0) {
1017 		mlog_errno(ret);
1018 		goto bail;
1019 	}
1020 
1021 	/* With an up to date view of the slots, we can check that no
1022 	 * other node has been improperly configured to heartbeat in
1023 	 * our slot. */
1024 	own_slot_ok = o2hb_check_own_slot(reg);
1025 
1026 	/* fill in the proper info for our next heartbeat */
1027 	o2hb_prepare_block(reg, reg->hr_generation);
1028 
1029 	ret = o2hb_issue_node_write(reg, &write_wc);
1030 	if (ret < 0) {
1031 		mlog_errno(ret);
1032 		goto bail;
1033 	}
1034 
1035 	i = -1;
1036 	while((i = find_next_bit(configured_nodes,
1037 				 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1038 		membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1039 	}
1040 
1041 	/*
1042 	 * We have to be sure we've advertised ourselves on disk
1043 	 * before we can go to steady state.  This ensures that
1044 	 * people we find in our steady state have seen us.
1045 	 */
1046 	o2hb_wait_on_io(reg, &write_wc);
1047 	if (write_wc.wc_error) {
1048 		/* Do not re-arm the write timeout on I/O error - we
1049 		 * can't be sure that the new block ever made it to
1050 		 * disk */
1051 		mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1052 		     write_wc.wc_error, reg->hr_dev_name);
1053 		ret = write_wc.wc_error;
1054 		goto bail;
1055 	}
1056 
1057 	/* Skip disarming the timeout if own slot has stale/bad data */
1058 	if (own_slot_ok) {
1059 		o2hb_set_quorum_device(reg);
1060 		o2hb_arm_write_timeout(reg);
1061 	}
1062 
1063 bail:
1064 	/* let the person who launched us know when things are steady */
1065 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
1066 		if (!ret && own_slot_ok && !membership_change) {
1067 			if (atomic_dec_and_test(&reg->hr_steady_iterations))
1068 				wake_up(&o2hb_steady_queue);
1069 		}
1070 	}
1071 
1072 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
1073 		if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1074 			printk(KERN_NOTICE "o2hb: Unable to stabilize "
1075 			       "heartbeart on region %s (%s)\n",
1076 			       config_item_name(&reg->hr_item),
1077 			       reg->hr_dev_name);
1078 			atomic_set(&reg->hr_steady_iterations, 0);
1079 			reg->hr_aborted_start = 1;
1080 			wake_up(&o2hb_steady_queue);
1081 			ret = -EIO;
1082 		}
1083 	}
1084 
1085 	return ret;
1086 }
1087 
1088 /* Subtract b from a, storing the result in a. a *must* have a larger
1089  * value than b. */
o2hb_tv_subtract(struct timeval * a,struct timeval * b)1090 static void o2hb_tv_subtract(struct timeval *a,
1091 			     struct timeval *b)
1092 {
1093 	/* just return 0 when a is after b */
1094 	if (a->tv_sec < b->tv_sec ||
1095 	    (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1096 		a->tv_sec = 0;
1097 		a->tv_usec = 0;
1098 		return;
1099 	}
1100 
1101 	a->tv_sec -= b->tv_sec;
1102 	a->tv_usec -= b->tv_usec;
1103 	while ( a->tv_usec < 0 ) {
1104 		a->tv_sec--;
1105 		a->tv_usec += 1000000;
1106 	}
1107 }
1108 
o2hb_elapsed_msecs(struct timeval * start,struct timeval * end)1109 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1110 				       struct timeval *end)
1111 {
1112 	struct timeval res = *end;
1113 
1114 	o2hb_tv_subtract(&res, start);
1115 
1116 	return res.tv_sec * 1000 + res.tv_usec / 1000;
1117 }
1118 
1119 /*
1120  * we ride the region ref that the region dir holds.  before the region
1121  * dir is removed and drops it ref it will wait to tear down this
1122  * thread.
1123  */
o2hb_thread(void * data)1124 static int o2hb_thread(void *data)
1125 {
1126 	int i, ret;
1127 	struct o2hb_region *reg = data;
1128 	struct o2hb_bio_wait_ctxt write_wc;
1129 	struct timeval before_hb, after_hb;
1130 	unsigned int elapsed_msec;
1131 
1132 	mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1133 
1134 	set_user_nice(current, -20);
1135 
1136 	/* Pin node */
1137 	o2nm_depend_this_node();
1138 
1139 	while (!kthread_should_stop() &&
1140 	       !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1141 		/* We track the time spent inside
1142 		 * o2hb_do_disk_heartbeat so that we avoid more than
1143 		 * hr_timeout_ms between disk writes. On busy systems
1144 		 * this should result in a heartbeat which is less
1145 		 * likely to time itself out. */
1146 		do_gettimeofday(&before_hb);
1147 
1148 		ret = o2hb_do_disk_heartbeat(reg);
1149 
1150 		do_gettimeofday(&after_hb);
1151 		elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1152 
1153 		mlog(ML_HEARTBEAT,
1154 		     "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
1155 		     before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1156 		     after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1157 		     elapsed_msec);
1158 
1159 		if (!kthread_should_stop() &&
1160 		    elapsed_msec < reg->hr_timeout_ms) {
1161 			/* the kthread api has blocked signals for us so no
1162 			 * need to record the return value. */
1163 			msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1164 		}
1165 	}
1166 
1167 	o2hb_disarm_write_timeout(reg);
1168 
1169 	/* unclean stop is only used in very bad situation */
1170 	for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1171 		o2hb_shutdown_slot(&reg->hr_slots[i]);
1172 
1173 	/* Explicit down notification - avoid forcing the other nodes
1174 	 * to timeout on this region when we could just as easily
1175 	 * write a clear generation - thus indicating to them that
1176 	 * this node has left this region.
1177 	 */
1178 	if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1179 		o2hb_prepare_block(reg, 0);
1180 		ret = o2hb_issue_node_write(reg, &write_wc);
1181 		if (ret == 0)
1182 			o2hb_wait_on_io(reg, &write_wc);
1183 		else
1184 			mlog_errno(ret);
1185 	}
1186 
1187 	/* Unpin node */
1188 	o2nm_undepend_this_node();
1189 
1190 	mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1191 
1192 	return 0;
1193 }
1194 
1195 #ifdef CONFIG_DEBUG_FS
o2hb_debug_open(struct inode * inode,struct file * file)1196 static int o2hb_debug_open(struct inode *inode, struct file *file)
1197 {
1198 	struct o2hb_debug_buf *db = inode->i_private;
1199 	struct o2hb_region *reg;
1200 	unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1201 	unsigned long lts;
1202 	char *buf = NULL;
1203 	int i = -1;
1204 	int out = 0;
1205 
1206 	/* max_nodes should be the largest bitmap we pass here */
1207 	BUG_ON(sizeof(map) < db->db_size);
1208 
1209 	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1210 	if (!buf)
1211 		goto bail;
1212 
1213 	switch (db->db_type) {
1214 	case O2HB_DB_TYPE_LIVENODES:
1215 	case O2HB_DB_TYPE_LIVEREGIONS:
1216 	case O2HB_DB_TYPE_QUORUMREGIONS:
1217 	case O2HB_DB_TYPE_FAILEDREGIONS:
1218 		spin_lock(&o2hb_live_lock);
1219 		memcpy(map, db->db_data, db->db_size);
1220 		spin_unlock(&o2hb_live_lock);
1221 		break;
1222 
1223 	case O2HB_DB_TYPE_REGION_LIVENODES:
1224 		spin_lock(&o2hb_live_lock);
1225 		reg = (struct o2hb_region *)db->db_data;
1226 		memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1227 		spin_unlock(&o2hb_live_lock);
1228 		break;
1229 
1230 	case O2HB_DB_TYPE_REGION_NUMBER:
1231 		reg = (struct o2hb_region *)db->db_data;
1232 		out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1233 				reg->hr_region_num);
1234 		goto done;
1235 
1236 	case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1237 		reg = (struct o2hb_region *)db->db_data;
1238 		lts = reg->hr_last_timeout_start;
1239 		/* If 0, it has never been set before */
1240 		if (lts)
1241 			lts = jiffies_to_msecs(jiffies - lts);
1242 		out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1243 		goto done;
1244 
1245 	case O2HB_DB_TYPE_REGION_PINNED:
1246 		reg = (struct o2hb_region *)db->db_data;
1247 		out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1248 				!!reg->hr_item_pinned);
1249 		goto done;
1250 
1251 	default:
1252 		goto done;
1253 	}
1254 
1255 	while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1256 		out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1257 	out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1258 
1259 done:
1260 	i_size_write(inode, out);
1261 
1262 	file->private_data = buf;
1263 
1264 	return 0;
1265 bail:
1266 	return -ENOMEM;
1267 }
1268 
o2hb_debug_release(struct inode * inode,struct file * file)1269 static int o2hb_debug_release(struct inode *inode, struct file *file)
1270 {
1271 	kfree(file->private_data);
1272 	return 0;
1273 }
1274 
o2hb_debug_read(struct file * file,char __user * buf,size_t nbytes,loff_t * ppos)1275 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1276 				 size_t nbytes, loff_t *ppos)
1277 {
1278 	return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1279 				       i_size_read(file->f_mapping->host));
1280 }
1281 #else
o2hb_debug_open(struct inode * inode,struct file * file)1282 static int o2hb_debug_open(struct inode *inode, struct file *file)
1283 {
1284 	return 0;
1285 }
o2hb_debug_release(struct inode * inode,struct file * file)1286 static int o2hb_debug_release(struct inode *inode, struct file *file)
1287 {
1288 	return 0;
1289 }
o2hb_debug_read(struct file * file,char __user * buf,size_t nbytes,loff_t * ppos)1290 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1291 			       size_t nbytes, loff_t *ppos)
1292 {
1293 	return 0;
1294 }
1295 #endif  /* CONFIG_DEBUG_FS */
1296 
1297 static const struct file_operations o2hb_debug_fops = {
1298 	.open =		o2hb_debug_open,
1299 	.release =	o2hb_debug_release,
1300 	.read =		o2hb_debug_read,
1301 	.llseek =	generic_file_llseek,
1302 };
1303 
o2hb_exit(void)1304 void o2hb_exit(void)
1305 {
1306 	kfree(o2hb_db_livenodes);
1307 	kfree(o2hb_db_liveregions);
1308 	kfree(o2hb_db_quorumregions);
1309 	kfree(o2hb_db_failedregions);
1310 	debugfs_remove(o2hb_debug_failedregions);
1311 	debugfs_remove(o2hb_debug_quorumregions);
1312 	debugfs_remove(o2hb_debug_liveregions);
1313 	debugfs_remove(o2hb_debug_livenodes);
1314 	debugfs_remove(o2hb_debug_dir);
1315 }
1316 
o2hb_debug_create(const char * name,struct dentry * dir,struct o2hb_debug_buf ** db,int db_len,int type,int size,int len,void * data)1317 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1318 					struct o2hb_debug_buf **db, int db_len,
1319 					int type, int size, int len, void *data)
1320 {
1321 	*db = kmalloc(db_len, GFP_KERNEL);
1322 	if (!*db)
1323 		return NULL;
1324 
1325 	(*db)->db_type = type;
1326 	(*db)->db_size = size;
1327 	(*db)->db_len = len;
1328 	(*db)->db_data = data;
1329 
1330 	return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1331 				   &o2hb_debug_fops);
1332 }
1333 
o2hb_debug_init(void)1334 static int o2hb_debug_init(void)
1335 {
1336 	int ret = -ENOMEM;
1337 
1338 	o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1339 	if (!o2hb_debug_dir) {
1340 		mlog_errno(ret);
1341 		goto bail;
1342 	}
1343 
1344 	o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1345 						 o2hb_debug_dir,
1346 						 &o2hb_db_livenodes,
1347 						 sizeof(*o2hb_db_livenodes),
1348 						 O2HB_DB_TYPE_LIVENODES,
1349 						 sizeof(o2hb_live_node_bitmap),
1350 						 O2NM_MAX_NODES,
1351 						 o2hb_live_node_bitmap);
1352 	if (!o2hb_debug_livenodes) {
1353 		mlog_errno(ret);
1354 		goto bail;
1355 	}
1356 
1357 	o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1358 						   o2hb_debug_dir,
1359 						   &o2hb_db_liveregions,
1360 						   sizeof(*o2hb_db_liveregions),
1361 						   O2HB_DB_TYPE_LIVEREGIONS,
1362 						   sizeof(o2hb_live_region_bitmap),
1363 						   O2NM_MAX_REGIONS,
1364 						   o2hb_live_region_bitmap);
1365 	if (!o2hb_debug_liveregions) {
1366 		mlog_errno(ret);
1367 		goto bail;
1368 	}
1369 
1370 	o2hb_debug_quorumregions =
1371 			o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1372 					  o2hb_debug_dir,
1373 					  &o2hb_db_quorumregions,
1374 					  sizeof(*o2hb_db_quorumregions),
1375 					  O2HB_DB_TYPE_QUORUMREGIONS,
1376 					  sizeof(o2hb_quorum_region_bitmap),
1377 					  O2NM_MAX_REGIONS,
1378 					  o2hb_quorum_region_bitmap);
1379 	if (!o2hb_debug_quorumregions) {
1380 		mlog_errno(ret);
1381 		goto bail;
1382 	}
1383 
1384 	o2hb_debug_failedregions =
1385 			o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1386 					  o2hb_debug_dir,
1387 					  &o2hb_db_failedregions,
1388 					  sizeof(*o2hb_db_failedregions),
1389 					  O2HB_DB_TYPE_FAILEDREGIONS,
1390 					  sizeof(o2hb_failed_region_bitmap),
1391 					  O2NM_MAX_REGIONS,
1392 					  o2hb_failed_region_bitmap);
1393 	if (!o2hb_debug_failedregions) {
1394 		mlog_errno(ret);
1395 		goto bail;
1396 	}
1397 
1398 	ret = 0;
1399 bail:
1400 	if (ret)
1401 		o2hb_exit();
1402 
1403 	return ret;
1404 }
1405 
o2hb_init(void)1406 int o2hb_init(void)
1407 {
1408 	int i;
1409 
1410 	for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1411 		INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1412 
1413 	for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1414 		INIT_LIST_HEAD(&o2hb_live_slots[i]);
1415 
1416 	INIT_LIST_HEAD(&o2hb_node_events);
1417 
1418 	memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1419 	memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1420 	memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1421 	memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1422 	memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1423 
1424 	o2hb_dependent_users = 0;
1425 
1426 	return o2hb_debug_init();
1427 }
1428 
1429 /* if we're already in a callback then we're already serialized by the sem */
o2hb_fill_node_map_from_callback(unsigned long * map,unsigned bytes)1430 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1431 					     unsigned bytes)
1432 {
1433 	BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1434 
1435 	memcpy(map, &o2hb_live_node_bitmap, bytes);
1436 }
1437 
1438 /*
1439  * get a map of all nodes that are heartbeating in any regions
1440  */
o2hb_fill_node_map(unsigned long * map,unsigned bytes)1441 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1442 {
1443 	/* callers want to serialize this map and callbacks so that they
1444 	 * can trust that they don't miss nodes coming to the party */
1445 	down_read(&o2hb_callback_sem);
1446 	spin_lock(&o2hb_live_lock);
1447 	o2hb_fill_node_map_from_callback(map, bytes);
1448 	spin_unlock(&o2hb_live_lock);
1449 	up_read(&o2hb_callback_sem);
1450 }
1451 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1452 
1453 /*
1454  * heartbeat configfs bits.  The heartbeat set is a default set under
1455  * the cluster set in nodemanager.c.
1456  */
1457 
to_o2hb_region(struct config_item * item)1458 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1459 {
1460 	return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1461 }
1462 
1463 /* drop_item only drops its ref after killing the thread, nothing should
1464  * be using the region anymore.  this has to clean up any state that
1465  * attributes might have built up. */
o2hb_region_release(struct config_item * item)1466 static void o2hb_region_release(struct config_item *item)
1467 {
1468 	int i;
1469 	struct page *page;
1470 	struct o2hb_region *reg = to_o2hb_region(item);
1471 
1472 	mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1473 
1474 	if (reg->hr_tmp_block)
1475 		kfree(reg->hr_tmp_block);
1476 
1477 	if (reg->hr_slot_data) {
1478 		for (i = 0; i < reg->hr_num_pages; i++) {
1479 			page = reg->hr_slot_data[i];
1480 			if (page)
1481 				__free_page(page);
1482 		}
1483 		kfree(reg->hr_slot_data);
1484 	}
1485 
1486 	if (reg->hr_bdev)
1487 		blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1488 
1489 	if (reg->hr_slots)
1490 		kfree(reg->hr_slots);
1491 
1492 	kfree(reg->hr_db_regnum);
1493 	kfree(reg->hr_db_livenodes);
1494 	debugfs_remove(reg->hr_debug_livenodes);
1495 	debugfs_remove(reg->hr_debug_regnum);
1496 	debugfs_remove(reg->hr_debug_elapsed_time);
1497 	debugfs_remove(reg->hr_debug_pinned);
1498 	debugfs_remove(reg->hr_debug_dir);
1499 
1500 	spin_lock(&o2hb_live_lock);
1501 	list_del(&reg->hr_all_item);
1502 	spin_unlock(&o2hb_live_lock);
1503 
1504 	kfree(reg);
1505 }
1506 
o2hb_read_block_input(struct o2hb_region * reg,const char * page,size_t count,unsigned long * ret_bytes,unsigned int * ret_bits)1507 static int o2hb_read_block_input(struct o2hb_region *reg,
1508 				 const char *page,
1509 				 size_t count,
1510 				 unsigned long *ret_bytes,
1511 				 unsigned int *ret_bits)
1512 {
1513 	unsigned long bytes;
1514 	char *p = (char *)page;
1515 
1516 	bytes = simple_strtoul(p, &p, 0);
1517 	if (!p || (*p && (*p != '\n')))
1518 		return -EINVAL;
1519 
1520 	/* Heartbeat and fs min / max block sizes are the same. */
1521 	if (bytes > 4096 || bytes < 512)
1522 		return -ERANGE;
1523 	if (hweight16(bytes) != 1)
1524 		return -EINVAL;
1525 
1526 	if (ret_bytes)
1527 		*ret_bytes = bytes;
1528 	if (ret_bits)
1529 		*ret_bits = ffs(bytes) - 1;
1530 
1531 	return 0;
1532 }
1533 
o2hb_region_block_bytes_read(struct o2hb_region * reg,char * page)1534 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1535 					    char *page)
1536 {
1537 	return sprintf(page, "%u\n", reg->hr_block_bytes);
1538 }
1539 
o2hb_region_block_bytes_write(struct o2hb_region * reg,const char * page,size_t count)1540 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1541 					     const char *page,
1542 					     size_t count)
1543 {
1544 	int status;
1545 	unsigned long block_bytes;
1546 	unsigned int block_bits;
1547 
1548 	if (reg->hr_bdev)
1549 		return -EINVAL;
1550 
1551 	status = o2hb_read_block_input(reg, page, count,
1552 				       &block_bytes, &block_bits);
1553 	if (status)
1554 		return status;
1555 
1556 	reg->hr_block_bytes = (unsigned int)block_bytes;
1557 	reg->hr_block_bits = block_bits;
1558 
1559 	return count;
1560 }
1561 
o2hb_region_start_block_read(struct o2hb_region * reg,char * page)1562 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1563 					    char *page)
1564 {
1565 	return sprintf(page, "%llu\n", reg->hr_start_block);
1566 }
1567 
o2hb_region_start_block_write(struct o2hb_region * reg,const char * page,size_t count)1568 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1569 					     const char *page,
1570 					     size_t count)
1571 {
1572 	unsigned long long tmp;
1573 	char *p = (char *)page;
1574 
1575 	if (reg->hr_bdev)
1576 		return -EINVAL;
1577 
1578 	tmp = simple_strtoull(p, &p, 0);
1579 	if (!p || (*p && (*p != '\n')))
1580 		return -EINVAL;
1581 
1582 	reg->hr_start_block = tmp;
1583 
1584 	return count;
1585 }
1586 
o2hb_region_blocks_read(struct o2hb_region * reg,char * page)1587 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1588 				       char *page)
1589 {
1590 	return sprintf(page, "%d\n", reg->hr_blocks);
1591 }
1592 
o2hb_region_blocks_write(struct o2hb_region * reg,const char * page,size_t count)1593 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1594 					const char *page,
1595 					size_t count)
1596 {
1597 	unsigned long tmp;
1598 	char *p = (char *)page;
1599 
1600 	if (reg->hr_bdev)
1601 		return -EINVAL;
1602 
1603 	tmp = simple_strtoul(p, &p, 0);
1604 	if (!p || (*p && (*p != '\n')))
1605 		return -EINVAL;
1606 
1607 	if (tmp > O2NM_MAX_NODES || tmp == 0)
1608 		return -ERANGE;
1609 
1610 	reg->hr_blocks = (unsigned int)tmp;
1611 
1612 	return count;
1613 }
1614 
o2hb_region_dev_read(struct o2hb_region * reg,char * page)1615 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1616 				    char *page)
1617 {
1618 	unsigned int ret = 0;
1619 
1620 	if (reg->hr_bdev)
1621 		ret = sprintf(page, "%s\n", reg->hr_dev_name);
1622 
1623 	return ret;
1624 }
1625 
o2hb_init_region_params(struct o2hb_region * reg)1626 static void o2hb_init_region_params(struct o2hb_region *reg)
1627 {
1628 	reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1629 	reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1630 
1631 	mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1632 	     reg->hr_start_block, reg->hr_blocks);
1633 	mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1634 	     reg->hr_block_bytes, reg->hr_block_bits);
1635 	mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1636 	mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1637 }
1638 
o2hb_map_slot_data(struct o2hb_region * reg)1639 static int o2hb_map_slot_data(struct o2hb_region *reg)
1640 {
1641 	int i, j;
1642 	unsigned int last_slot;
1643 	unsigned int spp = reg->hr_slots_per_page;
1644 	struct page *page;
1645 	char *raw;
1646 	struct o2hb_disk_slot *slot;
1647 
1648 	reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1649 	if (reg->hr_tmp_block == NULL) {
1650 		mlog_errno(-ENOMEM);
1651 		return -ENOMEM;
1652 	}
1653 
1654 	reg->hr_slots = kcalloc(reg->hr_blocks,
1655 				sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1656 	if (reg->hr_slots == NULL) {
1657 		mlog_errno(-ENOMEM);
1658 		return -ENOMEM;
1659 	}
1660 
1661 	for(i = 0; i < reg->hr_blocks; i++) {
1662 		slot = &reg->hr_slots[i];
1663 		slot->ds_node_num = i;
1664 		INIT_LIST_HEAD(&slot->ds_live_item);
1665 		slot->ds_raw_block = NULL;
1666 	}
1667 
1668 	reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1669 	mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1670 			   "at %u blocks per page\n",
1671 	     reg->hr_num_pages, reg->hr_blocks, spp);
1672 
1673 	reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1674 				    GFP_KERNEL);
1675 	if (!reg->hr_slot_data) {
1676 		mlog_errno(-ENOMEM);
1677 		return -ENOMEM;
1678 	}
1679 
1680 	for(i = 0; i < reg->hr_num_pages; i++) {
1681 		page = alloc_page(GFP_KERNEL);
1682 		if (!page) {
1683 			mlog_errno(-ENOMEM);
1684 			return -ENOMEM;
1685 		}
1686 
1687 		reg->hr_slot_data[i] = page;
1688 
1689 		last_slot = i * spp;
1690 		raw = page_address(page);
1691 		for (j = 0;
1692 		     (j < spp) && ((j + last_slot) < reg->hr_blocks);
1693 		     j++) {
1694 			BUG_ON((j + last_slot) >= reg->hr_blocks);
1695 
1696 			slot = &reg->hr_slots[j + last_slot];
1697 			slot->ds_raw_block =
1698 				(struct o2hb_disk_heartbeat_block *) raw;
1699 
1700 			raw += reg->hr_block_bytes;
1701 		}
1702 	}
1703 
1704 	return 0;
1705 }
1706 
1707 /* Read in all the slots available and populate the tracking
1708  * structures so that we can start with a baseline idea of what's
1709  * there. */
o2hb_populate_slot_data(struct o2hb_region * reg)1710 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1711 {
1712 	int ret, i;
1713 	struct o2hb_disk_slot *slot;
1714 	struct o2hb_disk_heartbeat_block *hb_block;
1715 
1716 	ret = o2hb_read_slots(reg, reg->hr_blocks);
1717 	if (ret) {
1718 		mlog_errno(ret);
1719 		goto out;
1720 	}
1721 
1722 	/* We only want to get an idea of the values initially in each
1723 	 * slot, so we do no verification - o2hb_check_slot will
1724 	 * actually determine if each configured slot is valid and
1725 	 * whether any values have changed. */
1726 	for(i = 0; i < reg->hr_blocks; i++) {
1727 		slot = &reg->hr_slots[i];
1728 		hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1729 
1730 		/* Only fill the values that o2hb_check_slot uses to
1731 		 * determine changing slots */
1732 		slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1733 		slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1734 	}
1735 
1736 out:
1737 	return ret;
1738 }
1739 
1740 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
o2hb_region_dev_write(struct o2hb_region * reg,const char * page,size_t count)1741 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1742 				     const char *page,
1743 				     size_t count)
1744 {
1745 	struct task_struct *hb_task;
1746 	long fd;
1747 	int sectsize;
1748 	char *p = (char *)page;
1749 	struct file *filp = NULL;
1750 	struct inode *inode = NULL;
1751 	ssize_t ret = -EINVAL;
1752 	int live_threshold;
1753 
1754 	if (reg->hr_bdev)
1755 		goto out;
1756 
1757 	/* We can't heartbeat without having had our node number
1758 	 * configured yet. */
1759 	if (o2nm_this_node() == O2NM_MAX_NODES)
1760 		goto out;
1761 
1762 	fd = simple_strtol(p, &p, 0);
1763 	if (!p || (*p && (*p != '\n')))
1764 		goto out;
1765 
1766 	if (fd < 0 || fd >= INT_MAX)
1767 		goto out;
1768 
1769 	filp = fget(fd);
1770 	if (filp == NULL)
1771 		goto out;
1772 
1773 	if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1774 	    reg->hr_block_bytes == 0)
1775 		goto out;
1776 
1777 	inode = igrab(filp->f_mapping->host);
1778 	if (inode == NULL)
1779 		goto out;
1780 
1781 	if (!S_ISBLK(inode->i_mode))
1782 		goto out;
1783 
1784 	reg->hr_bdev = I_BDEV(filp->f_mapping->host);
1785 	ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
1786 	if (ret) {
1787 		reg->hr_bdev = NULL;
1788 		goto out;
1789 	}
1790 	inode = NULL;
1791 
1792 	bdevname(reg->hr_bdev, reg->hr_dev_name);
1793 
1794 	sectsize = bdev_logical_block_size(reg->hr_bdev);
1795 	if (sectsize != reg->hr_block_bytes) {
1796 		mlog(ML_ERROR,
1797 		     "blocksize %u incorrect for device, expected %d",
1798 		     reg->hr_block_bytes, sectsize);
1799 		ret = -EINVAL;
1800 		goto out;
1801 	}
1802 
1803 	o2hb_init_region_params(reg);
1804 
1805 	/* Generation of zero is invalid */
1806 	do {
1807 		get_random_bytes(&reg->hr_generation,
1808 				 sizeof(reg->hr_generation));
1809 	} while (reg->hr_generation == 0);
1810 
1811 	ret = o2hb_map_slot_data(reg);
1812 	if (ret) {
1813 		mlog_errno(ret);
1814 		goto out;
1815 	}
1816 
1817 	ret = o2hb_populate_slot_data(reg);
1818 	if (ret) {
1819 		mlog_errno(ret);
1820 		goto out;
1821 	}
1822 
1823 	INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1824 
1825 	/*
1826 	 * A node is considered live after it has beat LIVE_THRESHOLD
1827 	 * times.  We're not steady until we've given them a chance
1828 	 * _after_ our first read.
1829 	 * The default threshold is bare minimum so as to limit the delay
1830 	 * during mounts. For global heartbeat, the threshold doubled for the
1831 	 * first region.
1832 	 */
1833 	live_threshold = O2HB_LIVE_THRESHOLD;
1834 	if (o2hb_global_heartbeat_active()) {
1835 		spin_lock(&o2hb_live_lock);
1836 		if (o2hb_pop_count(&o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1837 			live_threshold <<= 1;
1838 		spin_unlock(&o2hb_live_lock);
1839 	}
1840 	++live_threshold;
1841 	atomic_set(&reg->hr_steady_iterations, live_threshold);
1842 	/* unsteady_iterations is double the steady_iterations */
1843 	atomic_set(&reg->hr_unsteady_iterations, (live_threshold << 1));
1844 
1845 	hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1846 			      reg->hr_item.ci_name);
1847 	if (IS_ERR(hb_task)) {
1848 		ret = PTR_ERR(hb_task);
1849 		mlog_errno(ret);
1850 		goto out;
1851 	}
1852 
1853 	spin_lock(&o2hb_live_lock);
1854 	reg->hr_task = hb_task;
1855 	spin_unlock(&o2hb_live_lock);
1856 
1857 	ret = wait_event_interruptible(o2hb_steady_queue,
1858 				atomic_read(&reg->hr_steady_iterations) == 0);
1859 	if (ret) {
1860 		atomic_set(&reg->hr_steady_iterations, 0);
1861 		reg->hr_aborted_start = 1;
1862 	}
1863 
1864 	if (reg->hr_aborted_start) {
1865 		ret = -EIO;
1866 		goto out;
1867 	}
1868 
1869 	/* Ok, we were woken.  Make sure it wasn't by drop_item() */
1870 	spin_lock(&o2hb_live_lock);
1871 	hb_task = reg->hr_task;
1872 	if (o2hb_global_heartbeat_active())
1873 		set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1874 	spin_unlock(&o2hb_live_lock);
1875 
1876 	if (hb_task)
1877 		ret = count;
1878 	else
1879 		ret = -EIO;
1880 
1881 	if (hb_task && o2hb_global_heartbeat_active())
1882 		printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1883 		       config_item_name(&reg->hr_item), reg->hr_dev_name);
1884 
1885 out:
1886 	if (filp)
1887 		fput(filp);
1888 	if (inode)
1889 		iput(inode);
1890 	if (ret < 0) {
1891 		if (reg->hr_bdev) {
1892 			blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1893 			reg->hr_bdev = NULL;
1894 		}
1895 	}
1896 	return ret;
1897 }
1898 
o2hb_region_pid_read(struct o2hb_region * reg,char * page)1899 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1900                                       char *page)
1901 {
1902 	pid_t pid = 0;
1903 
1904 	spin_lock(&o2hb_live_lock);
1905 	if (reg->hr_task)
1906 		pid = task_pid_nr(reg->hr_task);
1907 	spin_unlock(&o2hb_live_lock);
1908 
1909 	if (!pid)
1910 		return 0;
1911 
1912 	return sprintf(page, "%u\n", pid);
1913 }
1914 
1915 struct o2hb_region_attribute {
1916 	struct configfs_attribute attr;
1917 	ssize_t (*show)(struct o2hb_region *, char *);
1918 	ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1919 };
1920 
1921 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1922 	.attr	= { .ca_owner = THIS_MODULE,
1923 		    .ca_name = "block_bytes",
1924 		    .ca_mode = S_IRUGO | S_IWUSR },
1925 	.show	= o2hb_region_block_bytes_read,
1926 	.store	= o2hb_region_block_bytes_write,
1927 };
1928 
1929 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1930 	.attr	= { .ca_owner = THIS_MODULE,
1931 		    .ca_name = "start_block",
1932 		    .ca_mode = S_IRUGO | S_IWUSR },
1933 	.show	= o2hb_region_start_block_read,
1934 	.store	= o2hb_region_start_block_write,
1935 };
1936 
1937 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1938 	.attr	= { .ca_owner = THIS_MODULE,
1939 		    .ca_name = "blocks",
1940 		    .ca_mode = S_IRUGO | S_IWUSR },
1941 	.show	= o2hb_region_blocks_read,
1942 	.store	= o2hb_region_blocks_write,
1943 };
1944 
1945 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1946 	.attr	= { .ca_owner = THIS_MODULE,
1947 		    .ca_name = "dev",
1948 		    .ca_mode = S_IRUGO | S_IWUSR },
1949 	.show	= o2hb_region_dev_read,
1950 	.store	= o2hb_region_dev_write,
1951 };
1952 
1953 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1954        .attr   = { .ca_owner = THIS_MODULE,
1955                    .ca_name = "pid",
1956                    .ca_mode = S_IRUGO | S_IRUSR },
1957        .show   = o2hb_region_pid_read,
1958 };
1959 
1960 static struct configfs_attribute *o2hb_region_attrs[] = {
1961 	&o2hb_region_attr_block_bytes.attr,
1962 	&o2hb_region_attr_start_block.attr,
1963 	&o2hb_region_attr_blocks.attr,
1964 	&o2hb_region_attr_dev.attr,
1965 	&o2hb_region_attr_pid.attr,
1966 	NULL,
1967 };
1968 
o2hb_region_show(struct config_item * item,struct configfs_attribute * attr,char * page)1969 static ssize_t o2hb_region_show(struct config_item *item,
1970 				struct configfs_attribute *attr,
1971 				char *page)
1972 {
1973 	struct o2hb_region *reg = to_o2hb_region(item);
1974 	struct o2hb_region_attribute *o2hb_region_attr =
1975 		container_of(attr, struct o2hb_region_attribute, attr);
1976 	ssize_t ret = 0;
1977 
1978 	if (o2hb_region_attr->show)
1979 		ret = o2hb_region_attr->show(reg, page);
1980 	return ret;
1981 }
1982 
o2hb_region_store(struct config_item * item,struct configfs_attribute * attr,const char * page,size_t count)1983 static ssize_t o2hb_region_store(struct config_item *item,
1984 				 struct configfs_attribute *attr,
1985 				 const char *page, size_t count)
1986 {
1987 	struct o2hb_region *reg = to_o2hb_region(item);
1988 	struct o2hb_region_attribute *o2hb_region_attr =
1989 		container_of(attr, struct o2hb_region_attribute, attr);
1990 	ssize_t ret = -EINVAL;
1991 
1992 	if (o2hb_region_attr->store)
1993 		ret = o2hb_region_attr->store(reg, page, count);
1994 	return ret;
1995 }
1996 
1997 static struct configfs_item_operations o2hb_region_item_ops = {
1998 	.release		= o2hb_region_release,
1999 	.show_attribute		= o2hb_region_show,
2000 	.store_attribute	= o2hb_region_store,
2001 };
2002 
2003 static struct config_item_type o2hb_region_type = {
2004 	.ct_item_ops	= &o2hb_region_item_ops,
2005 	.ct_attrs	= o2hb_region_attrs,
2006 	.ct_owner	= THIS_MODULE,
2007 };
2008 
2009 /* heartbeat set */
2010 
2011 struct o2hb_heartbeat_group {
2012 	struct config_group hs_group;
2013 	/* some stuff? */
2014 };
2015 
to_o2hb_heartbeat_group(struct config_group * group)2016 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
2017 {
2018 	return group ?
2019 		container_of(group, struct o2hb_heartbeat_group, hs_group)
2020 		: NULL;
2021 }
2022 
o2hb_debug_region_init(struct o2hb_region * reg,struct dentry * dir)2023 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
2024 {
2025 	int ret = -ENOMEM;
2026 
2027 	reg->hr_debug_dir =
2028 		debugfs_create_dir(config_item_name(&reg->hr_item), dir);
2029 	if (!reg->hr_debug_dir) {
2030 		mlog_errno(ret);
2031 		goto bail;
2032 	}
2033 
2034 	reg->hr_debug_livenodes =
2035 			o2hb_debug_create(O2HB_DEBUG_LIVENODES,
2036 					  reg->hr_debug_dir,
2037 					  &(reg->hr_db_livenodes),
2038 					  sizeof(*(reg->hr_db_livenodes)),
2039 					  O2HB_DB_TYPE_REGION_LIVENODES,
2040 					  sizeof(reg->hr_live_node_bitmap),
2041 					  O2NM_MAX_NODES, reg);
2042 	if (!reg->hr_debug_livenodes) {
2043 		mlog_errno(ret);
2044 		goto bail;
2045 	}
2046 
2047 	reg->hr_debug_regnum =
2048 			o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
2049 					  reg->hr_debug_dir,
2050 					  &(reg->hr_db_regnum),
2051 					  sizeof(*(reg->hr_db_regnum)),
2052 					  O2HB_DB_TYPE_REGION_NUMBER,
2053 					  0, O2NM_MAX_NODES, reg);
2054 	if (!reg->hr_debug_regnum) {
2055 		mlog_errno(ret);
2056 		goto bail;
2057 	}
2058 
2059 	reg->hr_debug_elapsed_time =
2060 			o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
2061 					  reg->hr_debug_dir,
2062 					  &(reg->hr_db_elapsed_time),
2063 					  sizeof(*(reg->hr_db_elapsed_time)),
2064 					  O2HB_DB_TYPE_REGION_ELAPSED_TIME,
2065 					  0, 0, reg);
2066 	if (!reg->hr_debug_elapsed_time) {
2067 		mlog_errno(ret);
2068 		goto bail;
2069 	}
2070 
2071 	reg->hr_debug_pinned =
2072 			o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
2073 					  reg->hr_debug_dir,
2074 					  &(reg->hr_db_pinned),
2075 					  sizeof(*(reg->hr_db_pinned)),
2076 					  O2HB_DB_TYPE_REGION_PINNED,
2077 					  0, 0, reg);
2078 	if (!reg->hr_debug_pinned) {
2079 		mlog_errno(ret);
2080 		goto bail;
2081 	}
2082 
2083 	ret = 0;
2084 bail:
2085 	return ret;
2086 }
2087 
o2hb_heartbeat_group_make_item(struct config_group * group,const char * name)2088 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
2089 							  const char *name)
2090 {
2091 	struct o2hb_region *reg = NULL;
2092 	int ret;
2093 
2094 	reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
2095 	if (reg == NULL)
2096 		return ERR_PTR(-ENOMEM);
2097 
2098 	if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2099 		ret = -ENAMETOOLONG;
2100 		goto free;
2101 	}
2102 
2103 	spin_lock(&o2hb_live_lock);
2104 	reg->hr_region_num = 0;
2105 	if (o2hb_global_heartbeat_active()) {
2106 		reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2107 							 O2NM_MAX_REGIONS);
2108 		if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2109 			spin_unlock(&o2hb_live_lock);
2110 			ret = -EFBIG;
2111 			goto free;
2112 		}
2113 		set_bit(reg->hr_region_num, o2hb_region_bitmap);
2114 	}
2115 	list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2116 	spin_unlock(&o2hb_live_lock);
2117 
2118 	config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2119 
2120 	ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
2121 	if (ret) {
2122 		config_item_put(&reg->hr_item);
2123 		goto free;
2124 	}
2125 
2126 	return &reg->hr_item;
2127 free:
2128 	kfree(reg);
2129 	return ERR_PTR(ret);
2130 }
2131 
o2hb_heartbeat_group_drop_item(struct config_group * group,struct config_item * item)2132 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2133 					   struct config_item *item)
2134 {
2135 	struct task_struct *hb_task;
2136 	struct o2hb_region *reg = to_o2hb_region(item);
2137 	int quorum_region = 0;
2138 
2139 	/* stop the thread when the user removes the region dir */
2140 	spin_lock(&o2hb_live_lock);
2141 	hb_task = reg->hr_task;
2142 	reg->hr_task = NULL;
2143 	reg->hr_item_dropped = 1;
2144 	spin_unlock(&o2hb_live_lock);
2145 
2146 	if (hb_task)
2147 		kthread_stop(hb_task);
2148 
2149 	if (o2hb_global_heartbeat_active()) {
2150 		spin_lock(&o2hb_live_lock);
2151 		clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2152 		clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2153 		if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2154 			quorum_region = 1;
2155 		clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2156 		spin_unlock(&o2hb_live_lock);
2157 		printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2158 		       ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2159 			"stopped" : "start aborted"), config_item_name(item),
2160 		       reg->hr_dev_name);
2161 	}
2162 
2163 	/*
2164 	 * If we're racing a dev_write(), we need to wake them.  They will
2165 	 * check reg->hr_task
2166 	 */
2167 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
2168 		reg->hr_aborted_start = 1;
2169 		atomic_set(&reg->hr_steady_iterations, 0);
2170 		wake_up(&o2hb_steady_queue);
2171 	}
2172 
2173 	config_item_put(item);
2174 
2175 	if (!o2hb_global_heartbeat_active() || !quorum_region)
2176 		return;
2177 
2178 	/*
2179 	 * If global heartbeat active and there are dependent users,
2180 	 * pin all regions if quorum region count <= CUT_OFF
2181 	 */
2182 	spin_lock(&o2hb_live_lock);
2183 
2184 	if (!o2hb_dependent_users)
2185 		goto unlock;
2186 
2187 	if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
2188 			   O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2189 		o2hb_region_pin(NULL);
2190 
2191 unlock:
2192 	spin_unlock(&o2hb_live_lock);
2193 }
2194 
2195 struct o2hb_heartbeat_group_attribute {
2196 	struct configfs_attribute attr;
2197 	ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2198 	ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2199 };
2200 
o2hb_heartbeat_group_show(struct config_item * item,struct configfs_attribute * attr,char * page)2201 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2202 					 struct configfs_attribute *attr,
2203 					 char *page)
2204 {
2205 	struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2206 	struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2207 		container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2208 	ssize_t ret = 0;
2209 
2210 	if (o2hb_heartbeat_group_attr->show)
2211 		ret = o2hb_heartbeat_group_attr->show(reg, page);
2212 	return ret;
2213 }
2214 
o2hb_heartbeat_group_store(struct config_item * item,struct configfs_attribute * attr,const char * page,size_t count)2215 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2216 					  struct configfs_attribute *attr,
2217 					  const char *page, size_t count)
2218 {
2219 	struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2220 	struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2221 		container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2222 	ssize_t ret = -EINVAL;
2223 
2224 	if (o2hb_heartbeat_group_attr->store)
2225 		ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2226 	return ret;
2227 }
2228 
o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group * group,char * page)2229 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2230 						     char *page)
2231 {
2232 	return sprintf(page, "%u\n", o2hb_dead_threshold);
2233 }
2234 
o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group * group,const char * page,size_t count)2235 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2236 						    const char *page,
2237 						    size_t count)
2238 {
2239 	unsigned long tmp;
2240 	char *p = (char *)page;
2241 
2242 	tmp = simple_strtoul(p, &p, 10);
2243 	if (!p || (*p && (*p != '\n')))
2244                 return -EINVAL;
2245 
2246 	/* this will validate ranges for us. */
2247 	o2hb_dead_threshold_set((unsigned int) tmp);
2248 
2249 	return count;
2250 }
2251 
2252 static
o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group * group,char * page)2253 ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2254 				       char *page)
2255 {
2256 	return sprintf(page, "%s\n",
2257 		       o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2258 }
2259 
2260 static
o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group * group,const char * page,size_t count)2261 ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2262 					const char *page, size_t count)
2263 {
2264 	unsigned int i;
2265 	int ret;
2266 	size_t len;
2267 
2268 	len = (page[count - 1] == '\n') ? count - 1 : count;
2269 	if (!len)
2270 		return -EINVAL;
2271 
2272 	for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2273 		if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2274 			continue;
2275 
2276 		ret = o2hb_global_hearbeat_mode_set(i);
2277 		if (!ret)
2278 			printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2279 			       o2hb_heartbeat_mode_desc[i]);
2280 		return count;
2281 	}
2282 
2283 	return -EINVAL;
2284 
2285 }
2286 
2287 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2288 	.attr	= { .ca_owner = THIS_MODULE,
2289 		    .ca_name = "dead_threshold",
2290 		    .ca_mode = S_IRUGO | S_IWUSR },
2291 	.show	= o2hb_heartbeat_group_threshold_show,
2292 	.store	= o2hb_heartbeat_group_threshold_store,
2293 };
2294 
2295 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2296 	.attr   = { .ca_owner = THIS_MODULE,
2297 		.ca_name = "mode",
2298 		.ca_mode = S_IRUGO | S_IWUSR },
2299 	.show   = o2hb_heartbeat_group_mode_show,
2300 	.store  = o2hb_heartbeat_group_mode_store,
2301 };
2302 
2303 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2304 	&o2hb_heartbeat_group_attr_threshold.attr,
2305 	&o2hb_heartbeat_group_attr_mode.attr,
2306 	NULL,
2307 };
2308 
2309 static struct configfs_item_operations o2hb_hearbeat_group_item_ops = {
2310 	.show_attribute		= o2hb_heartbeat_group_show,
2311 	.store_attribute	= o2hb_heartbeat_group_store,
2312 };
2313 
2314 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2315 	.make_item	= o2hb_heartbeat_group_make_item,
2316 	.drop_item	= o2hb_heartbeat_group_drop_item,
2317 };
2318 
2319 static struct config_item_type o2hb_heartbeat_group_type = {
2320 	.ct_group_ops	= &o2hb_heartbeat_group_group_ops,
2321 	.ct_item_ops	= &o2hb_hearbeat_group_item_ops,
2322 	.ct_attrs	= o2hb_heartbeat_group_attrs,
2323 	.ct_owner	= THIS_MODULE,
2324 };
2325 
2326 /* this is just here to avoid touching group in heartbeat.h which the
2327  * entire damn world #includes */
o2hb_alloc_hb_set(void)2328 struct config_group *o2hb_alloc_hb_set(void)
2329 {
2330 	struct o2hb_heartbeat_group *hs = NULL;
2331 	struct config_group *ret = NULL;
2332 
2333 	hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2334 	if (hs == NULL)
2335 		goto out;
2336 
2337 	config_group_init_type_name(&hs->hs_group, "heartbeat",
2338 				    &o2hb_heartbeat_group_type);
2339 
2340 	ret = &hs->hs_group;
2341 out:
2342 	if (ret == NULL)
2343 		kfree(hs);
2344 	return ret;
2345 }
2346 
o2hb_free_hb_set(struct config_group * group)2347 void o2hb_free_hb_set(struct config_group *group)
2348 {
2349 	struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2350 	kfree(hs);
2351 }
2352 
2353 /* hb callback registration and issuing */
2354 
hbcall_from_type(enum o2hb_callback_type type)2355 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2356 {
2357 	if (type == O2HB_NUM_CB)
2358 		return ERR_PTR(-EINVAL);
2359 
2360 	return &o2hb_callbacks[type];
2361 }
2362 
o2hb_setup_callback(struct o2hb_callback_func * hc,enum o2hb_callback_type type,o2hb_cb_func * func,void * data,int priority)2363 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2364 			 enum o2hb_callback_type type,
2365 			 o2hb_cb_func *func,
2366 			 void *data,
2367 			 int priority)
2368 {
2369 	INIT_LIST_HEAD(&hc->hc_item);
2370 	hc->hc_func = func;
2371 	hc->hc_data = data;
2372 	hc->hc_priority = priority;
2373 	hc->hc_type = type;
2374 	hc->hc_magic = O2HB_CB_MAGIC;
2375 }
2376 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2377 
2378 /*
2379  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2380  * In global heartbeat mode, region_uuid passed is NULL.
2381  *
2382  * In local, we only pin the matching region. In global we pin all the active
2383  * regions.
2384  */
o2hb_region_pin(const char * region_uuid)2385 static int o2hb_region_pin(const char *region_uuid)
2386 {
2387 	int ret = 0, found = 0;
2388 	struct o2hb_region *reg;
2389 	char *uuid;
2390 
2391 	assert_spin_locked(&o2hb_live_lock);
2392 
2393 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2394 		uuid = config_item_name(&reg->hr_item);
2395 
2396 		/* local heartbeat */
2397 		if (region_uuid) {
2398 			if (strcmp(region_uuid, uuid))
2399 				continue;
2400 			found = 1;
2401 		}
2402 
2403 		if (reg->hr_item_pinned || reg->hr_item_dropped)
2404 			goto skip_pin;
2405 
2406 		/* Ignore ENOENT only for local hb (userdlm domain) */
2407 		ret = o2nm_depend_item(&reg->hr_item);
2408 		if (!ret) {
2409 			mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2410 			reg->hr_item_pinned = 1;
2411 		} else {
2412 			if (ret == -ENOENT && found)
2413 				ret = 0;
2414 			else {
2415 				mlog(ML_ERROR, "Pin region %s fails with %d\n",
2416 				     uuid, ret);
2417 				break;
2418 			}
2419 		}
2420 skip_pin:
2421 		if (found)
2422 			break;
2423 	}
2424 
2425 	return ret;
2426 }
2427 
2428 /*
2429  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2430  * In global heartbeat mode, region_uuid passed is NULL.
2431  *
2432  * In local, we only unpin the matching region. In global we unpin all the
2433  * active regions.
2434  */
o2hb_region_unpin(const char * region_uuid)2435 static void o2hb_region_unpin(const char *region_uuid)
2436 {
2437 	struct o2hb_region *reg;
2438 	char *uuid;
2439 	int found = 0;
2440 
2441 	assert_spin_locked(&o2hb_live_lock);
2442 
2443 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
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 
o2hb_region_inc_user(const char * region_uuid)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 (o2hb_pop_count(&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 
o2hb_region_dec_user(const char * region_uuid)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 
o2hb_register_callback(const char * region_uuid,struct o2hb_callback_func * hc)2512 int o2hb_register_callback(const char *region_uuid,
2513 			   struct o2hb_callback_func *hc)
2514 {
2515 	struct o2hb_callback_func *tmp;
2516 	struct list_head *iter;
2517 	struct o2hb_callback *hbcall;
2518 	int ret;
2519 
2520 	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2521 	BUG_ON(!list_empty(&hc->hc_item));
2522 
2523 	hbcall = hbcall_from_type(hc->hc_type);
2524 	if (IS_ERR(hbcall)) {
2525 		ret = PTR_ERR(hbcall);
2526 		goto out;
2527 	}
2528 
2529 	if (region_uuid) {
2530 		ret = o2hb_region_inc_user(region_uuid);
2531 		if (ret) {
2532 			mlog_errno(ret);
2533 			goto out;
2534 		}
2535 	}
2536 
2537 	down_write(&o2hb_callback_sem);
2538 
2539 	list_for_each(iter, &hbcall->list) {
2540 		tmp = list_entry(iter, struct o2hb_callback_func, hc_item);
2541 		if (hc->hc_priority < tmp->hc_priority) {
2542 			list_add_tail(&hc->hc_item, iter);
2543 			break;
2544 		}
2545 	}
2546 	if (list_empty(&hc->hc_item))
2547 		list_add_tail(&hc->hc_item, &hbcall->list);
2548 
2549 	up_write(&o2hb_callback_sem);
2550 	ret = 0;
2551 out:
2552 	mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2553 	     ret, __builtin_return_address(0), hc);
2554 	return ret;
2555 }
2556 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2557 
o2hb_unregister_callback(const char * region_uuid,struct o2hb_callback_func * hc)2558 void o2hb_unregister_callback(const char *region_uuid,
2559 			      struct o2hb_callback_func *hc)
2560 {
2561 	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2562 
2563 	mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2564 	     __builtin_return_address(0), hc);
2565 
2566 	/* XXX Can this happen _with_ a region reference? */
2567 	if (list_empty(&hc->hc_item))
2568 		return;
2569 
2570 	if (region_uuid)
2571 		o2hb_region_dec_user(region_uuid);
2572 
2573 	down_write(&o2hb_callback_sem);
2574 
2575 	list_del_init(&hc->hc_item);
2576 
2577 	up_write(&o2hb_callback_sem);
2578 }
2579 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2580 
o2hb_check_node_heartbeating(u8 node_num)2581 int o2hb_check_node_heartbeating(u8 node_num)
2582 {
2583 	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2584 
2585 	o2hb_fill_node_map(testing_map, sizeof(testing_map));
2586 	if (!test_bit(node_num, testing_map)) {
2587 		mlog(ML_HEARTBEAT,
2588 		     "node (%u) does not have heartbeating enabled.\n",
2589 		     node_num);
2590 		return 0;
2591 	}
2592 
2593 	return 1;
2594 }
2595 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2596 
o2hb_check_node_heartbeating_from_callback(u8 node_num)2597 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2598 {
2599 	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2600 
2601 	o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2602 	if (!test_bit(node_num, testing_map)) {
2603 		mlog(ML_HEARTBEAT,
2604 		     "node (%u) does not have heartbeating enabled.\n",
2605 		     node_num);
2606 		return 0;
2607 	}
2608 
2609 	return 1;
2610 }
2611 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2612 
2613 /* Makes sure our local node is configured with a node number, and is
2614  * heartbeating. */
o2hb_check_local_node_heartbeating(void)2615 int o2hb_check_local_node_heartbeating(void)
2616 {
2617 	u8 node_num;
2618 
2619 	/* if this node was set then we have networking */
2620 	node_num = o2nm_this_node();
2621 	if (node_num == O2NM_MAX_NODES) {
2622 		mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2623 		return 0;
2624 	}
2625 
2626 	return o2hb_check_node_heartbeating(node_num);
2627 }
2628 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2629 
2630 /*
2631  * this is just a hack until we get the plumbing which flips file systems
2632  * read only and drops the hb ref instead of killing the node dead.
2633  */
o2hb_stop_all_regions(void)2634 void o2hb_stop_all_regions(void)
2635 {
2636 	struct o2hb_region *reg;
2637 
2638 	mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2639 
2640 	spin_lock(&o2hb_live_lock);
2641 
2642 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2643 		reg->hr_unclean_stop = 1;
2644 
2645 	spin_unlock(&o2hb_live_lock);
2646 }
2647 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2648 
o2hb_get_all_regions(char * region_uuids,u8 max_regions)2649 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2650 {
2651 	struct o2hb_region *reg;
2652 	int numregs = 0;
2653 	char *p;
2654 
2655 	spin_lock(&o2hb_live_lock);
2656 
2657 	p = region_uuids;
2658 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2659 		mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2660 		if (numregs < max_regions) {
2661 			memcpy(p, config_item_name(&reg->hr_item),
2662 			       O2HB_MAX_REGION_NAME_LEN);
2663 			p += O2HB_MAX_REGION_NAME_LEN;
2664 		}
2665 		numregs++;
2666 	}
2667 
2668 	spin_unlock(&o2hb_live_lock);
2669 
2670 	return numregs;
2671 }
2672 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2673 
o2hb_global_heartbeat_active(void)2674 int o2hb_global_heartbeat_active(void)
2675 {
2676 	return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2677 }
2678 EXPORT_SYMBOL(o2hb_global_heartbeat_active);
2679