xref: /openbmc/linux/fs/ocfs2/cluster/heartbeat.c (revision 05bdb996)
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. */
o2hb_dead_threshold_set(unsigned int threshold)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 
o2hb_global_heartbeat_mode_set(unsigned int hb_mode)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 	unsigned int		hr_timeout_ms;
239 
240 	/* randomized as the region goes up and down so that a node
241 	 * recognizes a node going up and down in one iteration */
242 	u64			hr_generation;
243 
244 	struct delayed_work	hr_write_timeout_work;
245 	unsigned long		hr_last_timeout_start;
246 
247 	/* negotiate timer, used to negotiate extending hb timeout. */
248 	struct delayed_work	hr_nego_timeout_work;
249 	unsigned long		hr_nego_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
250 
251 	/* Used during o2hb_check_slot to hold a copy of the block
252 	 * being checked because we temporarily have to zero out the
253 	 * crc field. */
254 	struct o2hb_disk_heartbeat_block *hr_tmp_block;
255 
256 	/* Message key for negotiate timeout message. */
257 	unsigned int		hr_key;
258 	struct list_head	hr_handler_list;
259 
260 	/* last hb status, 0 for success, other value for error. */
261 	int			hr_last_hb_status;
262 };
263 
264 struct o2hb_bio_wait_ctxt {
265 	atomic_t          wc_num_reqs;
266 	struct completion wc_io_complete;
267 	int               wc_error;
268 };
269 
270 #define O2HB_NEGO_TIMEOUT_MS (O2HB_MAX_WRITE_TIMEOUT_MS/2)
271 
272 enum {
273 	O2HB_NEGO_TIMEOUT_MSG = 1,
274 	O2HB_NEGO_APPROVE_MSG = 2,
275 };
276 
277 struct o2hb_nego_msg {
278 	u8 node_num;
279 };
280 
o2hb_write_timeout(struct work_struct * work)281 static void o2hb_write_timeout(struct work_struct *work)
282 {
283 	int failed, quorum;
284 	struct o2hb_region *reg =
285 		container_of(work, struct o2hb_region,
286 			     hr_write_timeout_work.work);
287 
288 	mlog(ML_ERROR, "Heartbeat write timeout to device %pg after %u "
289 	     "milliseconds\n", reg->hr_bdev,
290 	     jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
291 
292 	if (o2hb_global_heartbeat_active()) {
293 		spin_lock(&o2hb_live_lock);
294 		if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
295 			set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
296 		failed = bitmap_weight(o2hb_failed_region_bitmap,
297 					O2NM_MAX_REGIONS);
298 		quorum = bitmap_weight(o2hb_quorum_region_bitmap,
299 					O2NM_MAX_REGIONS);
300 		spin_unlock(&o2hb_live_lock);
301 
302 		mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
303 		     quorum, failed);
304 
305 		/*
306 		 * Fence if the number of failed regions >= half the number
307 		 * of  quorum regions
308 		 */
309 		if ((failed << 1) < quorum)
310 			return;
311 	}
312 
313 	o2quo_disk_timeout();
314 }
315 
o2hb_arm_timeout(struct o2hb_region * reg)316 static void o2hb_arm_timeout(struct o2hb_region *reg)
317 {
318 	/* Arm writeout only after thread reaches steady state */
319 	if (atomic_read(&reg->hr_steady_iterations) != 0)
320 		return;
321 
322 	mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
323 	     O2HB_MAX_WRITE_TIMEOUT_MS);
324 
325 	if (o2hb_global_heartbeat_active()) {
326 		spin_lock(&o2hb_live_lock);
327 		clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
328 		spin_unlock(&o2hb_live_lock);
329 	}
330 	cancel_delayed_work(&reg->hr_write_timeout_work);
331 	schedule_delayed_work(&reg->hr_write_timeout_work,
332 			      msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
333 
334 	cancel_delayed_work(&reg->hr_nego_timeout_work);
335 	/* negotiate timeout must be less than write timeout. */
336 	schedule_delayed_work(&reg->hr_nego_timeout_work,
337 			      msecs_to_jiffies(O2HB_NEGO_TIMEOUT_MS));
338 	bitmap_zero(reg->hr_nego_node_bitmap, O2NM_MAX_NODES);
339 }
340 
o2hb_disarm_timeout(struct o2hb_region * reg)341 static void o2hb_disarm_timeout(struct o2hb_region *reg)
342 {
343 	cancel_delayed_work_sync(&reg->hr_write_timeout_work);
344 	cancel_delayed_work_sync(&reg->hr_nego_timeout_work);
345 }
346 
o2hb_send_nego_msg(int key,int type,u8 target)347 static int o2hb_send_nego_msg(int key, int type, u8 target)
348 {
349 	struct o2hb_nego_msg msg;
350 	int status, ret;
351 
352 	msg.node_num = o2nm_this_node();
353 again:
354 	ret = o2net_send_message(type, key, &msg, sizeof(msg),
355 			target, &status);
356 
357 	if (ret == -EAGAIN || ret == -ENOMEM) {
358 		msleep(100);
359 		goto again;
360 	}
361 
362 	return ret;
363 }
364 
o2hb_nego_timeout(struct work_struct * work)365 static void o2hb_nego_timeout(struct work_struct *work)
366 {
367 	unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
368 	int master_node, i, ret;
369 	struct o2hb_region *reg;
370 
371 	reg = container_of(work, struct o2hb_region, hr_nego_timeout_work.work);
372 	/* don't negotiate timeout if last hb failed since it is very
373 	 * possible io failed. Should let write timeout fence self.
374 	 */
375 	if (reg->hr_last_hb_status)
376 		return;
377 
378 	o2hb_fill_node_map(live_node_bitmap, O2NM_MAX_NODES);
379 	/* lowest node as master node to make negotiate decision. */
380 	master_node = find_first_bit(live_node_bitmap, O2NM_MAX_NODES);
381 
382 	if (master_node == o2nm_this_node()) {
383 		if (!test_bit(master_node, reg->hr_nego_node_bitmap)) {
384 			printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%pg).\n",
385 				o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000,
386 				config_item_name(&reg->hr_item), reg->hr_bdev);
387 			set_bit(master_node, reg->hr_nego_node_bitmap);
388 		}
389 		if (!bitmap_equal(reg->hr_nego_node_bitmap, live_node_bitmap,
390 				  O2NM_MAX_NODES)) {
391 			/* check negotiate bitmap every second to do timeout
392 			 * approve decision.
393 			 */
394 			schedule_delayed_work(&reg->hr_nego_timeout_work,
395 				msecs_to_jiffies(1000));
396 
397 			return;
398 		}
399 
400 		printk(KERN_NOTICE "o2hb: all nodes hb write hung, maybe region %s (%pg) is down.\n",
401 			config_item_name(&reg->hr_item), reg->hr_bdev);
402 		/* approve negotiate timeout request. */
403 		o2hb_arm_timeout(reg);
404 
405 		i = -1;
406 		while ((i = find_next_bit(live_node_bitmap,
407 				O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
408 			if (i == master_node)
409 				continue;
410 
411 			mlog(ML_HEARTBEAT, "send NEGO_APPROVE msg to node %d\n", i);
412 			ret = o2hb_send_nego_msg(reg->hr_key,
413 					O2HB_NEGO_APPROVE_MSG, i);
414 			if (ret)
415 				mlog(ML_ERROR, "send NEGO_APPROVE msg to node %d fail %d\n",
416 					i, ret);
417 		}
418 	} else {
419 		/* negotiate timeout with master node. */
420 		printk(KERN_NOTICE "o2hb: node %d hb write hung for %ds on region %s (%pg), negotiate timeout with node %d.\n",
421 			o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS/1000, config_item_name(&reg->hr_item),
422 			reg->hr_bdev, master_node);
423 		ret = o2hb_send_nego_msg(reg->hr_key, O2HB_NEGO_TIMEOUT_MSG,
424 				master_node);
425 		if (ret)
426 			mlog(ML_ERROR, "send NEGO_TIMEOUT msg to node %d fail %d\n",
427 				master_node, ret);
428 	}
429 }
430 
o2hb_nego_timeout_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)431 static int o2hb_nego_timeout_handler(struct o2net_msg *msg, u32 len, void *data,
432 				void **ret_data)
433 {
434 	struct o2hb_region *reg = data;
435 	struct o2hb_nego_msg *nego_msg;
436 
437 	nego_msg = (struct o2hb_nego_msg *)msg->buf;
438 	printk(KERN_NOTICE "o2hb: receive negotiate timeout message from node %d on region %s (%pg).\n",
439 		nego_msg->node_num, config_item_name(&reg->hr_item), reg->hr_bdev);
440 	if (nego_msg->node_num < O2NM_MAX_NODES)
441 		set_bit(nego_msg->node_num, reg->hr_nego_node_bitmap);
442 	else
443 		mlog(ML_ERROR, "got nego timeout message from bad node.\n");
444 
445 	return 0;
446 }
447 
o2hb_nego_approve_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)448 static int o2hb_nego_approve_handler(struct o2net_msg *msg, u32 len, void *data,
449 				void **ret_data)
450 {
451 	struct o2hb_region *reg = data;
452 
453 	printk(KERN_NOTICE "o2hb: negotiate timeout approved by master node on region %s (%pg).\n",
454 		config_item_name(&reg->hr_item), reg->hr_bdev);
455 	o2hb_arm_timeout(reg);
456 	return 0;
457 }
458 
o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt * wc)459 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
460 {
461 	atomic_set(&wc->wc_num_reqs, 1);
462 	init_completion(&wc->wc_io_complete);
463 	wc->wc_error = 0;
464 }
465 
466 /* Used in error paths too */
o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt * wc,unsigned int num)467 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
468 				     unsigned int num)
469 {
470 	/* sadly atomic_sub_and_test() isn't available on all platforms.  The
471 	 * good news is that the fast path only completes one at a time */
472 	while(num--) {
473 		if (atomic_dec_and_test(&wc->wc_num_reqs)) {
474 			BUG_ON(num > 0);
475 			complete(&wc->wc_io_complete);
476 		}
477 	}
478 }
479 
o2hb_wait_on_io(struct o2hb_bio_wait_ctxt * wc)480 static void o2hb_wait_on_io(struct o2hb_bio_wait_ctxt *wc)
481 {
482 	o2hb_bio_wait_dec(wc, 1);
483 	wait_for_completion(&wc->wc_io_complete);
484 }
485 
o2hb_bio_end_io(struct bio * bio)486 static void o2hb_bio_end_io(struct bio *bio)
487 {
488 	struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
489 
490 	if (bio->bi_status) {
491 		mlog(ML_ERROR, "IO Error %d\n", bio->bi_status);
492 		wc->wc_error = blk_status_to_errno(bio->bi_status);
493 	}
494 
495 	o2hb_bio_wait_dec(wc, 1);
496 	bio_put(bio);
497 }
498 
499 /* Setup a Bio to cover I/O against num_slots slots starting at
500  * start_slot. */
o2hb_setup_one_bio(struct o2hb_region * reg,struct o2hb_bio_wait_ctxt * wc,unsigned int * current_slot,unsigned int max_slots,blk_opf_t opf)501 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
502 				      struct o2hb_bio_wait_ctxt *wc,
503 				      unsigned int *current_slot,
504 				      unsigned int max_slots, blk_opf_t opf)
505 {
506 	int len, current_page;
507 	unsigned int vec_len, vec_start;
508 	unsigned int bits = reg->hr_block_bits;
509 	unsigned int spp = reg->hr_slots_per_page;
510 	unsigned int cs = *current_slot;
511 	struct bio *bio;
512 	struct page *page;
513 
514 	/* Testing has shown this allocation to take long enough under
515 	 * GFP_KERNEL that the local node can get fenced. It would be
516 	 * nicest if we could pre-allocate these bios and avoid this
517 	 * all together. */
518 	bio = bio_alloc(reg->hr_bdev, 16, opf, GFP_ATOMIC);
519 	if (!bio) {
520 		mlog(ML_ERROR, "Could not alloc slots BIO!\n");
521 		bio = ERR_PTR(-ENOMEM);
522 		goto bail;
523 	}
524 
525 	/* Must put everything in 512 byte sectors for the bio... */
526 	bio->bi_iter.bi_sector = (reg->hr_start_block + cs) << (bits - 9);
527 	bio->bi_private = wc;
528 	bio->bi_end_io = o2hb_bio_end_io;
529 
530 	vec_start = (cs << bits) % PAGE_SIZE;
531 	while(cs < max_slots) {
532 		current_page = cs / spp;
533 		page = reg->hr_slot_data[current_page];
534 
535 		vec_len = min(PAGE_SIZE - vec_start,
536 			      (max_slots-cs) * (PAGE_SIZE/spp) );
537 
538 		mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
539 		     current_page, vec_len, vec_start);
540 
541 		len = bio_add_page(bio, page, vec_len, vec_start);
542 		if (len != vec_len) break;
543 
544 		cs += vec_len / (PAGE_SIZE/spp);
545 		vec_start = 0;
546 	}
547 
548 bail:
549 	*current_slot = cs;
550 	return bio;
551 }
552 
o2hb_read_slots(struct o2hb_region * reg,unsigned int begin_slot,unsigned int max_slots)553 static int o2hb_read_slots(struct o2hb_region *reg,
554 			   unsigned int begin_slot,
555 			   unsigned int max_slots)
556 {
557 	unsigned int current_slot = begin_slot;
558 	int status;
559 	struct o2hb_bio_wait_ctxt wc;
560 	struct bio *bio;
561 
562 	o2hb_bio_wait_init(&wc);
563 
564 	while(current_slot < max_slots) {
565 		bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots,
566 					 REQ_OP_READ);
567 		if (IS_ERR(bio)) {
568 			status = PTR_ERR(bio);
569 			mlog_errno(status);
570 			goto bail_and_wait;
571 		}
572 
573 		atomic_inc(&wc.wc_num_reqs);
574 		submit_bio(bio);
575 	}
576 
577 	status = 0;
578 
579 bail_and_wait:
580 	o2hb_wait_on_io(&wc);
581 	if (wc.wc_error && !status)
582 		status = wc.wc_error;
583 
584 	return status;
585 }
586 
o2hb_issue_node_write(struct o2hb_region * reg,struct o2hb_bio_wait_ctxt * write_wc)587 static int o2hb_issue_node_write(struct o2hb_region *reg,
588 				 struct o2hb_bio_wait_ctxt *write_wc)
589 {
590 	int status;
591 	unsigned int slot;
592 	struct bio *bio;
593 
594 	o2hb_bio_wait_init(write_wc);
595 
596 	slot = o2nm_this_node();
597 
598 	bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1,
599 				 REQ_OP_WRITE | REQ_SYNC);
600 	if (IS_ERR(bio)) {
601 		status = PTR_ERR(bio);
602 		mlog_errno(status);
603 		goto bail;
604 	}
605 
606 	atomic_inc(&write_wc->wc_num_reqs);
607 	submit_bio(bio);
608 
609 	status = 0;
610 bail:
611 	return status;
612 }
613 
o2hb_compute_block_crc_le(struct o2hb_region * reg,struct o2hb_disk_heartbeat_block * hb_block)614 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
615 				     struct o2hb_disk_heartbeat_block *hb_block)
616 {
617 	__le32 old_cksum;
618 	u32 ret;
619 
620 	/* We want to compute the block crc with a 0 value in the
621 	 * hb_cksum field. Save it off here and replace after the
622 	 * crc. */
623 	old_cksum = hb_block->hb_cksum;
624 	hb_block->hb_cksum = 0;
625 
626 	ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
627 
628 	hb_block->hb_cksum = old_cksum;
629 
630 	return ret;
631 }
632 
o2hb_dump_slot(struct o2hb_disk_heartbeat_block * hb_block)633 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
634 {
635 	mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
636 	     "cksum = 0x%x, generation 0x%llx\n",
637 	     (long long)le64_to_cpu(hb_block->hb_seq),
638 	     hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
639 	     (long long)le64_to_cpu(hb_block->hb_generation));
640 }
641 
o2hb_verify_crc(struct o2hb_region * reg,struct o2hb_disk_heartbeat_block * hb_block)642 static int o2hb_verify_crc(struct o2hb_region *reg,
643 			   struct o2hb_disk_heartbeat_block *hb_block)
644 {
645 	u32 read, computed;
646 
647 	read = le32_to_cpu(hb_block->hb_cksum);
648 	computed = o2hb_compute_block_crc_le(reg, hb_block);
649 
650 	return read == computed;
651 }
652 
653 /*
654  * Compare the slot data with what we wrote in the last iteration.
655  * If the match fails, print an appropriate error message. This is to
656  * detect errors like... another node hearting on the same slot,
657  * flaky device that is losing writes, etc.
658  * Returns 1 if check succeeds, 0 otherwise.
659  */
o2hb_check_own_slot(struct o2hb_region * reg)660 static int o2hb_check_own_slot(struct o2hb_region *reg)
661 {
662 	struct o2hb_disk_slot *slot;
663 	struct o2hb_disk_heartbeat_block *hb_block;
664 	char *errstr;
665 
666 	slot = &reg->hr_slots[o2nm_this_node()];
667 	/* Don't check on our 1st timestamp */
668 	if (!slot->ds_last_time)
669 		return 0;
670 
671 	hb_block = slot->ds_raw_block;
672 	if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
673 	    le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
674 	    hb_block->hb_node == slot->ds_node_num)
675 		return 1;
676 
677 #define ERRSTR1		"Another node is heartbeating on device"
678 #define ERRSTR2		"Heartbeat generation mismatch on device"
679 #define ERRSTR3		"Heartbeat sequence mismatch on device"
680 
681 	if (hb_block->hb_node != slot->ds_node_num)
682 		errstr = ERRSTR1;
683 	else if (le64_to_cpu(hb_block->hb_generation) !=
684 		 slot->ds_last_generation)
685 		errstr = ERRSTR2;
686 	else
687 		errstr = ERRSTR3;
688 
689 	mlog(ML_ERROR, "%s (%pg): expected(%u:0x%llx, 0x%llx), "
690 	     "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_bdev,
691 	     slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
692 	     (unsigned long long)slot->ds_last_time, hb_block->hb_node,
693 	     (unsigned long long)le64_to_cpu(hb_block->hb_generation),
694 	     (unsigned long long)le64_to_cpu(hb_block->hb_seq));
695 
696 	return 0;
697 }
698 
o2hb_prepare_block(struct o2hb_region * reg,u64 generation)699 static inline void o2hb_prepare_block(struct o2hb_region *reg,
700 				      u64 generation)
701 {
702 	int node_num;
703 	u64 cputime;
704 	struct o2hb_disk_slot *slot;
705 	struct o2hb_disk_heartbeat_block *hb_block;
706 
707 	node_num = o2nm_this_node();
708 	slot = &reg->hr_slots[node_num];
709 
710 	hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
711 	memset(hb_block, 0, reg->hr_block_bytes);
712 	/* TODO: time stuff */
713 	cputime = ktime_get_real_seconds();
714 	if (!cputime)
715 		cputime = 1;
716 
717 	hb_block->hb_seq = cpu_to_le64(cputime);
718 	hb_block->hb_node = node_num;
719 	hb_block->hb_generation = cpu_to_le64(generation);
720 	hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
721 
722 	/* This step must always happen last! */
723 	hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
724 								   hb_block));
725 
726 	mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
727 	     (long long)generation,
728 	     le32_to_cpu(hb_block->hb_cksum));
729 }
730 
o2hb_fire_callbacks(struct o2hb_callback * hbcall,struct o2nm_node * node,int idx)731 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
732 				struct o2nm_node *node,
733 				int idx)
734 {
735 	struct o2hb_callback_func *f;
736 
737 	list_for_each_entry(f, &hbcall->list, hc_item) {
738 		mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
739 		(f->hc_func)(node, idx, f->hc_data);
740 	}
741 }
742 
743 /* Will run the list in order until we process the passed event */
o2hb_run_event_list(struct o2hb_node_event * queued_event)744 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
745 {
746 	struct o2hb_callback *hbcall;
747 	struct o2hb_node_event *event;
748 
749 	/* Holding callback sem assures we don't alter the callback
750 	 * lists when doing this, and serializes ourselves with other
751 	 * processes wanting callbacks. */
752 	down_write(&o2hb_callback_sem);
753 
754 	spin_lock(&o2hb_live_lock);
755 	while (!list_empty(&o2hb_node_events)
756 	       && !list_empty(&queued_event->hn_item)) {
757 		event = list_entry(o2hb_node_events.next,
758 				   struct o2hb_node_event,
759 				   hn_item);
760 		list_del_init(&event->hn_item);
761 		spin_unlock(&o2hb_live_lock);
762 
763 		mlog(ML_HEARTBEAT, "Node %s event for %d\n",
764 		     event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
765 		     event->hn_node_num);
766 
767 		hbcall = hbcall_from_type(event->hn_event_type);
768 
769 		/* We should *never* have gotten on to the list with a
770 		 * bad type... This isn't something that we should try
771 		 * to recover from. */
772 		BUG_ON(IS_ERR(hbcall));
773 
774 		o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
775 
776 		spin_lock(&o2hb_live_lock);
777 	}
778 	spin_unlock(&o2hb_live_lock);
779 
780 	up_write(&o2hb_callback_sem);
781 }
782 
o2hb_queue_node_event(struct o2hb_node_event * event,enum o2hb_callback_type type,struct o2nm_node * node,int node_num)783 static void o2hb_queue_node_event(struct o2hb_node_event *event,
784 				  enum o2hb_callback_type type,
785 				  struct o2nm_node *node,
786 				  int node_num)
787 {
788 	assert_spin_locked(&o2hb_live_lock);
789 
790 	BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
791 
792 	event->hn_event_type = type;
793 	event->hn_node = node;
794 	event->hn_node_num = node_num;
795 
796 	mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
797 	     type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
798 
799 	list_add_tail(&event->hn_item, &o2hb_node_events);
800 }
801 
o2hb_shutdown_slot(struct o2hb_disk_slot * slot)802 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
803 {
804 	struct o2hb_node_event event =
805 		{ .hn_item = LIST_HEAD_INIT(event.hn_item), };
806 	struct o2nm_node *node;
807 	int queued = 0;
808 
809 	node = o2nm_get_node_by_num(slot->ds_node_num);
810 	if (!node)
811 		return;
812 
813 	spin_lock(&o2hb_live_lock);
814 	if (!list_empty(&slot->ds_live_item)) {
815 		mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
816 		     slot->ds_node_num);
817 
818 		list_del_init(&slot->ds_live_item);
819 
820 		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
821 			clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
822 
823 			o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
824 					      slot->ds_node_num);
825 			queued = 1;
826 		}
827 	}
828 	spin_unlock(&o2hb_live_lock);
829 
830 	if (queued)
831 		o2hb_run_event_list(&event);
832 
833 	o2nm_node_put(node);
834 }
835 
o2hb_set_quorum_device(struct o2hb_region * reg)836 static void o2hb_set_quorum_device(struct o2hb_region *reg)
837 {
838 	if (!o2hb_global_heartbeat_active())
839 		return;
840 
841 	/* Prevent race with o2hb_heartbeat_group_drop_item() */
842 	if (kthread_should_stop())
843 		return;
844 
845 	/* Tag region as quorum only after thread reaches steady state */
846 	if (atomic_read(&reg->hr_steady_iterations) != 0)
847 		return;
848 
849 	spin_lock(&o2hb_live_lock);
850 
851 	if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
852 		goto unlock;
853 
854 	/*
855 	 * A region can be added to the quorum only when it sees all
856 	 * live nodes heartbeat on it. In other words, the region has been
857 	 * added to all nodes.
858 	 */
859 	if (!bitmap_equal(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
860 			  O2NM_MAX_NODES))
861 		goto unlock;
862 
863 	printk(KERN_NOTICE "o2hb: Region %s (%pg) is now a quorum device\n",
864 	       config_item_name(&reg->hr_item), reg->hr_bdev);
865 
866 	set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
867 
868 	/*
869 	 * If global heartbeat active, unpin all regions if the
870 	 * region count > CUT_OFF
871 	 */
872 	if (bitmap_weight(o2hb_quorum_region_bitmap,
873 			   O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
874 		o2hb_region_unpin(NULL);
875 unlock:
876 	spin_unlock(&o2hb_live_lock);
877 }
878 
o2hb_check_slot(struct o2hb_region * reg,struct o2hb_disk_slot * slot)879 static int o2hb_check_slot(struct o2hb_region *reg,
880 			   struct o2hb_disk_slot *slot)
881 {
882 	int changed = 0, gen_changed = 0;
883 	struct o2hb_node_event event =
884 		{ .hn_item = LIST_HEAD_INIT(event.hn_item), };
885 	struct o2nm_node *node;
886 	struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
887 	u64 cputime;
888 	unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
889 	unsigned int slot_dead_ms;
890 	int tmp;
891 	int queued = 0;
892 
893 	memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
894 
895 	/*
896 	 * If a node is no longer configured but is still in the livemap, we
897 	 * may need to clear that bit from the livemap.
898 	 */
899 	node = o2nm_get_node_by_num(slot->ds_node_num);
900 	if (!node) {
901 		spin_lock(&o2hb_live_lock);
902 		tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
903 		spin_unlock(&o2hb_live_lock);
904 		if (!tmp)
905 			return 0;
906 	}
907 
908 	if (!o2hb_verify_crc(reg, hb_block)) {
909 		/* all paths from here will drop o2hb_live_lock for
910 		 * us. */
911 		spin_lock(&o2hb_live_lock);
912 
913 		/* Don't print an error on the console in this case -
914 		 * a freshly formatted heartbeat area will not have a
915 		 * crc set on it. */
916 		if (list_empty(&slot->ds_live_item))
917 			goto out;
918 
919 		/* The node is live but pushed out a bad crc. We
920 		 * consider it a transient miss but don't populate any
921 		 * other values as they may be junk. */
922 		mlog(ML_ERROR, "Node %d has written a bad crc to %pg\n",
923 		     slot->ds_node_num, reg->hr_bdev);
924 		o2hb_dump_slot(hb_block);
925 
926 		slot->ds_equal_samples++;
927 		goto fire_callbacks;
928 	}
929 
930 	/* we don't care if these wrap.. the state transitions below
931 	 * clear at the right places */
932 	cputime = le64_to_cpu(hb_block->hb_seq);
933 	if (slot->ds_last_time != cputime)
934 		slot->ds_changed_samples++;
935 	else
936 		slot->ds_equal_samples++;
937 	slot->ds_last_time = cputime;
938 
939 	/* The node changed heartbeat generations. We assume this to
940 	 * mean it dropped off but came back before we timed out. We
941 	 * want to consider it down for the time being but don't want
942 	 * to lose any changed_samples state we might build up to
943 	 * considering it live again. */
944 	if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
945 		gen_changed = 1;
946 		slot->ds_equal_samples = 0;
947 		mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
948 		     "to 0x%llx)\n", slot->ds_node_num,
949 		     (long long)slot->ds_last_generation,
950 		     (long long)le64_to_cpu(hb_block->hb_generation));
951 	}
952 
953 	slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
954 
955 	mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
956 	     "seq %llu last %llu changed %u equal %u\n",
957 	     slot->ds_node_num, (long long)slot->ds_last_generation,
958 	     le32_to_cpu(hb_block->hb_cksum),
959 	     (unsigned long long)le64_to_cpu(hb_block->hb_seq),
960 	     (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
961 	     slot->ds_equal_samples);
962 
963 	spin_lock(&o2hb_live_lock);
964 
965 fire_callbacks:
966 	/* dead nodes only come to life after some number of
967 	 * changes at any time during their dead time */
968 	if (list_empty(&slot->ds_live_item) &&
969 	    slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
970 		mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
971 		     slot->ds_node_num, (long long)slot->ds_last_generation);
972 
973 		set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
974 
975 		/* first on the list generates a callback */
976 		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
977 			mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
978 			     "bitmap\n", slot->ds_node_num);
979 			set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
980 
981 			o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
982 					      slot->ds_node_num);
983 
984 			changed = 1;
985 			queued = 1;
986 		}
987 
988 		list_add_tail(&slot->ds_live_item,
989 			      &o2hb_live_slots[slot->ds_node_num]);
990 
991 		slot->ds_equal_samples = 0;
992 
993 		/* We want to be sure that all nodes agree on the
994 		 * number of milliseconds before a node will be
995 		 * considered dead. The self-fencing timeout is
996 		 * computed from this value, and a discrepancy might
997 		 * result in heartbeat calling a node dead when it
998 		 * hasn't self-fenced yet. */
999 		slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
1000 		if (slot_dead_ms && slot_dead_ms != dead_ms) {
1001 			/* TODO: Perhaps we can fail the region here. */
1002 			mlog(ML_ERROR, "Node %d on device %pg has a dead count "
1003 			     "of %u ms, but our count is %u ms.\n"
1004 			     "Please double check your configuration values "
1005 			     "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
1006 			     slot->ds_node_num, reg->hr_bdev, slot_dead_ms,
1007 			     dead_ms);
1008 		}
1009 		goto out;
1010 	}
1011 
1012 	/* if the list is dead, we're done.. */
1013 	if (list_empty(&slot->ds_live_item))
1014 		goto out;
1015 
1016 	/* live nodes only go dead after enough consequtive missed
1017 	 * samples..  reset the missed counter whenever we see
1018 	 * activity */
1019 	if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
1020 		mlog(ML_HEARTBEAT, "Node %d left my region\n",
1021 		     slot->ds_node_num);
1022 
1023 		clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
1024 
1025 		/* last off the live_slot generates a callback */
1026 		list_del_init(&slot->ds_live_item);
1027 		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
1028 			mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
1029 			     "nodes bitmap\n", slot->ds_node_num);
1030 			clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
1031 
1032 			/* node can be null */
1033 			o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
1034 					      node, slot->ds_node_num);
1035 
1036 			changed = 1;
1037 			queued = 1;
1038 		}
1039 
1040 		/* We don't clear this because the node is still
1041 		 * actually writing new blocks. */
1042 		if (!gen_changed)
1043 			slot->ds_changed_samples = 0;
1044 		goto out;
1045 	}
1046 	if (slot->ds_changed_samples) {
1047 		slot->ds_changed_samples = 0;
1048 		slot->ds_equal_samples = 0;
1049 	}
1050 out:
1051 	spin_unlock(&o2hb_live_lock);
1052 
1053 	if (queued)
1054 		o2hb_run_event_list(&event);
1055 
1056 	if (node)
1057 		o2nm_node_put(node);
1058 	return changed;
1059 }
1060 
o2hb_highest_node(unsigned long * nodes,int numbits)1061 static int o2hb_highest_node(unsigned long *nodes, int numbits)
1062 {
1063 	return find_last_bit(nodes, numbits);
1064 }
1065 
o2hb_lowest_node(unsigned long * nodes,int numbits)1066 static int o2hb_lowest_node(unsigned long *nodes, int numbits)
1067 {
1068 	return find_first_bit(nodes, numbits);
1069 }
1070 
o2hb_do_disk_heartbeat(struct o2hb_region * reg)1071 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
1072 {
1073 	int i, ret, highest_node, lowest_node;
1074 	int membership_change = 0, own_slot_ok = 0;
1075 	unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
1076 	unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
1077 	struct o2hb_bio_wait_ctxt write_wc;
1078 
1079 	ret = o2nm_configured_node_map(configured_nodes,
1080 				       sizeof(configured_nodes));
1081 	if (ret) {
1082 		mlog_errno(ret);
1083 		goto bail;
1084 	}
1085 
1086 	/*
1087 	 * If a node is not configured but is in the livemap, we still need
1088 	 * to read the slot so as to be able to remove it from the livemap.
1089 	 */
1090 	o2hb_fill_node_map(live_node_bitmap, O2NM_MAX_NODES);
1091 	i = -1;
1092 	while ((i = find_next_bit(live_node_bitmap,
1093 				  O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1094 		set_bit(i, configured_nodes);
1095 	}
1096 
1097 	highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
1098 	lowest_node = o2hb_lowest_node(configured_nodes, O2NM_MAX_NODES);
1099 	if (highest_node >= O2NM_MAX_NODES || lowest_node >= O2NM_MAX_NODES) {
1100 		mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
1101 		ret = -EINVAL;
1102 		goto bail;
1103 	}
1104 
1105 	/* No sense in reading the slots of nodes that don't exist
1106 	 * yet. Of course, if the node definitions have holes in them
1107 	 * then we're reading an empty slot anyway... Consider this
1108 	 * best-effort. */
1109 	ret = o2hb_read_slots(reg, lowest_node, highest_node + 1);
1110 	if (ret < 0) {
1111 		mlog_errno(ret);
1112 		goto bail;
1113 	}
1114 
1115 	/* With an up to date view of the slots, we can check that no
1116 	 * other node has been improperly configured to heartbeat in
1117 	 * our slot. */
1118 	own_slot_ok = o2hb_check_own_slot(reg);
1119 
1120 	/* fill in the proper info for our next heartbeat */
1121 	o2hb_prepare_block(reg, reg->hr_generation);
1122 
1123 	ret = o2hb_issue_node_write(reg, &write_wc);
1124 	if (ret < 0) {
1125 		mlog_errno(ret);
1126 		goto bail;
1127 	}
1128 
1129 	i = -1;
1130 	while((i = find_next_bit(configured_nodes,
1131 				 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1132 		membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1133 	}
1134 
1135 	/*
1136 	 * We have to be sure we've advertised ourselves on disk
1137 	 * before we can go to steady state.  This ensures that
1138 	 * people we find in our steady state have seen us.
1139 	 */
1140 	o2hb_wait_on_io(&write_wc);
1141 	if (write_wc.wc_error) {
1142 		/* Do not re-arm the write timeout on I/O error - we
1143 		 * can't be sure that the new block ever made it to
1144 		 * disk */
1145 		mlog(ML_ERROR, "Write error %d on device \"%pg\"\n",
1146 		     write_wc.wc_error, reg->hr_bdev);
1147 		ret = write_wc.wc_error;
1148 		goto bail;
1149 	}
1150 
1151 	/* Skip disarming the timeout if own slot has stale/bad data */
1152 	if (own_slot_ok) {
1153 		o2hb_set_quorum_device(reg);
1154 		o2hb_arm_timeout(reg);
1155 		reg->hr_last_timeout_start = jiffies;
1156 	}
1157 
1158 bail:
1159 	/* let the person who launched us know when things are steady */
1160 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
1161 		if (!ret && own_slot_ok && !membership_change) {
1162 			if (atomic_dec_and_test(&reg->hr_steady_iterations))
1163 				wake_up(&o2hb_steady_queue);
1164 		}
1165 	}
1166 
1167 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
1168 		if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1169 			printk(KERN_NOTICE "o2hb: Unable to stabilize "
1170 			       "heartbeat on region %s (%pg)\n",
1171 			       config_item_name(&reg->hr_item),
1172 			       reg->hr_bdev);
1173 			atomic_set(&reg->hr_steady_iterations, 0);
1174 			reg->hr_aborted_start = 1;
1175 			wake_up(&o2hb_steady_queue);
1176 			ret = -EIO;
1177 		}
1178 	}
1179 
1180 	return ret;
1181 }
1182 
1183 /*
1184  * we ride the region ref that the region dir holds.  before the region
1185  * dir is removed and drops it ref it will wait to tear down this
1186  * thread.
1187  */
o2hb_thread(void * data)1188 static int o2hb_thread(void *data)
1189 {
1190 	int i, ret;
1191 	struct o2hb_region *reg = data;
1192 	struct o2hb_bio_wait_ctxt write_wc;
1193 	ktime_t before_hb, after_hb;
1194 	unsigned int elapsed_msec;
1195 
1196 	mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1197 
1198 	set_user_nice(current, MIN_NICE);
1199 
1200 	/* Pin node */
1201 	ret = o2nm_depend_this_node();
1202 	if (ret) {
1203 		mlog(ML_ERROR, "Node has been deleted, ret = %d\n", ret);
1204 		reg->hr_node_deleted = 1;
1205 		wake_up(&o2hb_steady_queue);
1206 		return 0;
1207 	}
1208 
1209 	while (!kthread_should_stop() &&
1210 	       !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1211 		/* We track the time spent inside
1212 		 * o2hb_do_disk_heartbeat so that we avoid more than
1213 		 * hr_timeout_ms between disk writes. On busy systems
1214 		 * this should result in a heartbeat which is less
1215 		 * likely to time itself out. */
1216 		before_hb = ktime_get_real();
1217 
1218 		ret = o2hb_do_disk_heartbeat(reg);
1219 		reg->hr_last_hb_status = ret;
1220 
1221 		after_hb = ktime_get_real();
1222 
1223 		elapsed_msec = (unsigned int)
1224 				ktime_ms_delta(after_hb, before_hb);
1225 
1226 		mlog(ML_HEARTBEAT,
1227 		     "start = %lld, end = %lld, msec = %u, ret = %d\n",
1228 		     before_hb, after_hb, elapsed_msec, ret);
1229 
1230 		if (!kthread_should_stop() &&
1231 		    elapsed_msec < reg->hr_timeout_ms) {
1232 			/* the kthread api has blocked signals for us so no
1233 			 * need to record the return value. */
1234 			msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1235 		}
1236 	}
1237 
1238 	o2hb_disarm_timeout(reg);
1239 
1240 	/* unclean stop is only used in very bad situation */
1241 	for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1242 		o2hb_shutdown_slot(&reg->hr_slots[i]);
1243 
1244 	/* Explicit down notification - avoid forcing the other nodes
1245 	 * to timeout on this region when we could just as easily
1246 	 * write a clear generation - thus indicating to them that
1247 	 * this node has left this region.
1248 	 */
1249 	if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1250 		o2hb_prepare_block(reg, 0);
1251 		ret = o2hb_issue_node_write(reg, &write_wc);
1252 		if (ret == 0)
1253 			o2hb_wait_on_io(&write_wc);
1254 		else
1255 			mlog_errno(ret);
1256 	}
1257 
1258 	/* Unpin node */
1259 	o2nm_undepend_this_node();
1260 
1261 	mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1262 
1263 	return 0;
1264 }
1265 
1266 #ifdef CONFIG_DEBUG_FS
o2hb_debug_open(struct inode * inode,struct file * file)1267 static int o2hb_debug_open(struct inode *inode, struct file *file)
1268 {
1269 	struct o2hb_debug_buf *db = inode->i_private;
1270 	struct o2hb_region *reg;
1271 	unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1272 	unsigned long lts;
1273 	char *buf = NULL;
1274 	int i = -1;
1275 	int out = 0;
1276 
1277 	/* max_nodes should be the largest bitmap we pass here */
1278 	BUG_ON(sizeof(map) < db->db_size);
1279 
1280 	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1281 	if (!buf)
1282 		goto bail;
1283 
1284 	switch (db->db_type) {
1285 	case O2HB_DB_TYPE_LIVENODES:
1286 	case O2HB_DB_TYPE_LIVEREGIONS:
1287 	case O2HB_DB_TYPE_QUORUMREGIONS:
1288 	case O2HB_DB_TYPE_FAILEDREGIONS:
1289 		spin_lock(&o2hb_live_lock);
1290 		memcpy(map, db->db_data, db->db_size);
1291 		spin_unlock(&o2hb_live_lock);
1292 		break;
1293 
1294 	case O2HB_DB_TYPE_REGION_LIVENODES:
1295 		spin_lock(&o2hb_live_lock);
1296 		reg = (struct o2hb_region *)db->db_data;
1297 		memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1298 		spin_unlock(&o2hb_live_lock);
1299 		break;
1300 
1301 	case O2HB_DB_TYPE_REGION_NUMBER:
1302 		reg = (struct o2hb_region *)db->db_data;
1303 		out += scnprintf(buf + out, PAGE_SIZE - out, "%d\n",
1304 				reg->hr_region_num);
1305 		goto done;
1306 
1307 	case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1308 		reg = (struct o2hb_region *)db->db_data;
1309 		lts = reg->hr_last_timeout_start;
1310 		/* If 0, it has never been set before */
1311 		if (lts)
1312 			lts = jiffies_to_msecs(jiffies - lts);
1313 		out += scnprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1314 		goto done;
1315 
1316 	case O2HB_DB_TYPE_REGION_PINNED:
1317 		reg = (struct o2hb_region *)db->db_data;
1318 		out += scnprintf(buf + out, PAGE_SIZE - out, "%u\n",
1319 				!!reg->hr_item_pinned);
1320 		goto done;
1321 
1322 	default:
1323 		goto done;
1324 	}
1325 
1326 	while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1327 		out += scnprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1328 	out += scnprintf(buf + out, PAGE_SIZE - out, "\n");
1329 
1330 done:
1331 	i_size_write(inode, out);
1332 
1333 	file->private_data = buf;
1334 
1335 	return 0;
1336 bail:
1337 	return -ENOMEM;
1338 }
1339 
o2hb_debug_release(struct inode * inode,struct file * file)1340 static int o2hb_debug_release(struct inode *inode, struct file *file)
1341 {
1342 	kfree(file->private_data);
1343 	return 0;
1344 }
1345 
o2hb_debug_read(struct file * file,char __user * buf,size_t nbytes,loff_t * ppos)1346 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1347 				 size_t nbytes, loff_t *ppos)
1348 {
1349 	return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1350 				       i_size_read(file->f_mapping->host));
1351 }
1352 #else
o2hb_debug_open(struct inode * inode,struct file * file)1353 static int o2hb_debug_open(struct inode *inode, struct file *file)
1354 {
1355 	return 0;
1356 }
o2hb_debug_release(struct inode * inode,struct file * file)1357 static int o2hb_debug_release(struct inode *inode, struct file *file)
1358 {
1359 	return 0;
1360 }
o2hb_debug_read(struct file * file,char __user * buf,size_t nbytes,loff_t * ppos)1361 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1362 			       size_t nbytes, loff_t *ppos)
1363 {
1364 	return 0;
1365 }
1366 #endif  /* CONFIG_DEBUG_FS */
1367 
1368 static const struct file_operations o2hb_debug_fops = {
1369 	.open =		o2hb_debug_open,
1370 	.release =	o2hb_debug_release,
1371 	.read =		o2hb_debug_read,
1372 	.llseek =	generic_file_llseek,
1373 };
1374 
o2hb_exit(void)1375 void o2hb_exit(void)
1376 {
1377 	debugfs_remove_recursive(o2hb_debug_dir);
1378 	kfree(o2hb_db_livenodes);
1379 	kfree(o2hb_db_liveregions);
1380 	kfree(o2hb_db_quorumregions);
1381 	kfree(o2hb_db_failedregions);
1382 }
1383 
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)1384 static void o2hb_debug_create(const char *name, struct dentry *dir,
1385 			      struct o2hb_debug_buf **db, int db_len, int type,
1386 			      int size, int len, void *data)
1387 {
1388 	*db = kmalloc(db_len, GFP_KERNEL);
1389 	if (!*db)
1390 		return;
1391 
1392 	(*db)->db_type = type;
1393 	(*db)->db_size = size;
1394 	(*db)->db_len = len;
1395 	(*db)->db_data = data;
1396 
1397 	debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db, &o2hb_debug_fops);
1398 }
1399 
o2hb_debug_init(void)1400 static void o2hb_debug_init(void)
1401 {
1402 	o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1403 
1404 	o2hb_debug_create(O2HB_DEBUG_LIVENODES, o2hb_debug_dir,
1405 			  &o2hb_db_livenodes, sizeof(*o2hb_db_livenodes),
1406 			  O2HB_DB_TYPE_LIVENODES, sizeof(o2hb_live_node_bitmap),
1407 			  O2NM_MAX_NODES, o2hb_live_node_bitmap);
1408 
1409 	o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS, o2hb_debug_dir,
1410 			  &o2hb_db_liveregions, sizeof(*o2hb_db_liveregions),
1411 			  O2HB_DB_TYPE_LIVEREGIONS,
1412 			  sizeof(o2hb_live_region_bitmap), O2NM_MAX_REGIONS,
1413 			  o2hb_live_region_bitmap);
1414 
1415 	o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS, o2hb_debug_dir,
1416 			  &o2hb_db_quorumregions,
1417 			  sizeof(*o2hb_db_quorumregions),
1418 			  O2HB_DB_TYPE_QUORUMREGIONS,
1419 			  sizeof(o2hb_quorum_region_bitmap), O2NM_MAX_REGIONS,
1420 			  o2hb_quorum_region_bitmap);
1421 
1422 	o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS, o2hb_debug_dir,
1423 			  &o2hb_db_failedregions,
1424 			  sizeof(*o2hb_db_failedregions),
1425 			  O2HB_DB_TYPE_FAILEDREGIONS,
1426 			  sizeof(o2hb_failed_region_bitmap), O2NM_MAX_REGIONS,
1427 			  o2hb_failed_region_bitmap);
1428 }
1429 
o2hb_init(void)1430 void o2hb_init(void)
1431 {
1432 	int i;
1433 
1434 	for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1435 		INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1436 
1437 	for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1438 		INIT_LIST_HEAD(&o2hb_live_slots[i]);
1439 
1440 	bitmap_zero(o2hb_live_node_bitmap, O2NM_MAX_NODES);
1441 	bitmap_zero(o2hb_region_bitmap, O2NM_MAX_REGIONS);
1442 	bitmap_zero(o2hb_live_region_bitmap, O2NM_MAX_REGIONS);
1443 	bitmap_zero(o2hb_quorum_region_bitmap, O2NM_MAX_REGIONS);
1444 	bitmap_zero(o2hb_failed_region_bitmap, O2NM_MAX_REGIONS);
1445 
1446 	o2hb_dependent_users = 0;
1447 
1448 	o2hb_debug_init();
1449 }
1450 
1451 /* 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 int bits)1452 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1453 					     unsigned int bits)
1454 {
1455 	bitmap_copy(map, o2hb_live_node_bitmap, bits);
1456 }
1457 
1458 /*
1459  * get a map of all nodes that are heartbeating in any regions
1460  */
o2hb_fill_node_map(unsigned long * map,unsigned int bits)1461 void o2hb_fill_node_map(unsigned long *map, unsigned int bits)
1462 {
1463 	/* callers want to serialize this map and callbacks so that they
1464 	 * can trust that they don't miss nodes coming to the party */
1465 	down_read(&o2hb_callback_sem);
1466 	spin_lock(&o2hb_live_lock);
1467 	o2hb_fill_node_map_from_callback(map, bits);
1468 	spin_unlock(&o2hb_live_lock);
1469 	up_read(&o2hb_callback_sem);
1470 }
1471 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1472 
1473 /*
1474  * heartbeat configfs bits.  The heartbeat set is a default set under
1475  * the cluster set in nodemanager.c.
1476  */
1477 
to_o2hb_region(struct config_item * item)1478 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1479 {
1480 	return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1481 }
1482 
1483 /* drop_item only drops its ref after killing the thread, nothing should
1484  * be using the region anymore.  this has to clean up any state that
1485  * attributes might have built up. */
o2hb_region_release(struct config_item * item)1486 static void o2hb_region_release(struct config_item *item)
1487 {
1488 	int i;
1489 	struct page *page;
1490 	struct o2hb_region *reg = to_o2hb_region(item);
1491 
1492 	mlog(ML_HEARTBEAT, "hb region release (%pg)\n", reg->hr_bdev);
1493 
1494 	kfree(reg->hr_tmp_block);
1495 
1496 	if (reg->hr_slot_data) {
1497 		for (i = 0; i < reg->hr_num_pages; i++) {
1498 			page = reg->hr_slot_data[i];
1499 			if (page)
1500 				__free_page(page);
1501 		}
1502 		kfree(reg->hr_slot_data);
1503 	}
1504 
1505 	if (reg->hr_bdev)
1506 		blkdev_put(reg->hr_bdev, NULL);
1507 
1508 	kfree(reg->hr_slots);
1509 
1510 	debugfs_remove_recursive(reg->hr_debug_dir);
1511 	kfree(reg->hr_db_livenodes);
1512 	kfree(reg->hr_db_regnum);
1513 	kfree(reg->hr_db_elapsed_time);
1514 	kfree(reg->hr_db_pinned);
1515 
1516 	spin_lock(&o2hb_live_lock);
1517 	list_del(&reg->hr_all_item);
1518 	spin_unlock(&o2hb_live_lock);
1519 
1520 	o2net_unregister_handler_list(&reg->hr_handler_list);
1521 	kfree(reg);
1522 }
1523 
o2hb_read_block_input(struct o2hb_region * reg,const char * page,unsigned long * ret_bytes,unsigned int * ret_bits)1524 static int o2hb_read_block_input(struct o2hb_region *reg,
1525 				 const char *page,
1526 				 unsigned long *ret_bytes,
1527 				 unsigned int *ret_bits)
1528 {
1529 	unsigned long bytes;
1530 	char *p = (char *)page;
1531 
1532 	bytes = simple_strtoul(p, &p, 0);
1533 	if (!p || (*p && (*p != '\n')))
1534 		return -EINVAL;
1535 
1536 	/* Heartbeat and fs min / max block sizes are the same. */
1537 	if (bytes > 4096 || bytes < 512)
1538 		return -ERANGE;
1539 	if (hweight16(bytes) != 1)
1540 		return -EINVAL;
1541 
1542 	if (ret_bytes)
1543 		*ret_bytes = bytes;
1544 	if (ret_bits)
1545 		*ret_bits = ffs(bytes) - 1;
1546 
1547 	return 0;
1548 }
1549 
o2hb_region_block_bytes_show(struct config_item * item,char * page)1550 static ssize_t o2hb_region_block_bytes_show(struct config_item *item,
1551 					    char *page)
1552 {
1553 	return sprintf(page, "%u\n", to_o2hb_region(item)->hr_block_bytes);
1554 }
1555 
o2hb_region_block_bytes_store(struct config_item * item,const char * page,size_t count)1556 static ssize_t o2hb_region_block_bytes_store(struct config_item *item,
1557 					     const char *page,
1558 					     size_t count)
1559 {
1560 	struct o2hb_region *reg = to_o2hb_region(item);
1561 	int status;
1562 	unsigned long block_bytes;
1563 	unsigned int block_bits;
1564 
1565 	if (reg->hr_bdev)
1566 		return -EINVAL;
1567 
1568 	status = o2hb_read_block_input(reg, page, &block_bytes,
1569 				       &block_bits);
1570 	if (status)
1571 		return status;
1572 
1573 	reg->hr_block_bytes = (unsigned int)block_bytes;
1574 	reg->hr_block_bits = block_bits;
1575 
1576 	return count;
1577 }
1578 
o2hb_region_start_block_show(struct config_item * item,char * page)1579 static ssize_t o2hb_region_start_block_show(struct config_item *item,
1580 					    char *page)
1581 {
1582 	return sprintf(page, "%llu\n", to_o2hb_region(item)->hr_start_block);
1583 }
1584 
o2hb_region_start_block_store(struct config_item * item,const char * page,size_t count)1585 static ssize_t o2hb_region_start_block_store(struct config_item *item,
1586 					     const char *page,
1587 					     size_t count)
1588 {
1589 	struct o2hb_region *reg = to_o2hb_region(item);
1590 	unsigned long long tmp;
1591 	char *p = (char *)page;
1592 	ssize_t ret;
1593 
1594 	if (reg->hr_bdev)
1595 		return -EINVAL;
1596 
1597 	ret = kstrtoull(p, 0, &tmp);
1598 	if (ret)
1599 		return -EINVAL;
1600 
1601 	reg->hr_start_block = tmp;
1602 
1603 	return count;
1604 }
1605 
o2hb_region_blocks_show(struct config_item * item,char * page)1606 static ssize_t o2hb_region_blocks_show(struct config_item *item, char *page)
1607 {
1608 	return sprintf(page, "%d\n", to_o2hb_region(item)->hr_blocks);
1609 }
1610 
o2hb_region_blocks_store(struct config_item * item,const char * page,size_t count)1611 static ssize_t o2hb_region_blocks_store(struct config_item *item,
1612 					const char *page,
1613 					size_t count)
1614 {
1615 	struct o2hb_region *reg = to_o2hb_region(item);
1616 	unsigned long tmp;
1617 	char *p = (char *)page;
1618 
1619 	if (reg->hr_bdev)
1620 		return -EINVAL;
1621 
1622 	tmp = simple_strtoul(p, &p, 0);
1623 	if (!p || (*p && (*p != '\n')))
1624 		return -EINVAL;
1625 
1626 	if (tmp > O2NM_MAX_NODES || tmp == 0)
1627 		return -ERANGE;
1628 
1629 	reg->hr_blocks = (unsigned int)tmp;
1630 
1631 	return count;
1632 }
1633 
o2hb_region_dev_show(struct config_item * item,char * page)1634 static ssize_t o2hb_region_dev_show(struct config_item *item, char *page)
1635 {
1636 	unsigned int ret = 0;
1637 
1638 	if (to_o2hb_region(item)->hr_bdev)
1639 		ret = sprintf(page, "%pg\n", to_o2hb_region(item)->hr_bdev);
1640 
1641 	return ret;
1642 }
1643 
o2hb_init_region_params(struct o2hb_region * reg)1644 static void o2hb_init_region_params(struct o2hb_region *reg)
1645 {
1646 	reg->hr_slots_per_page = PAGE_SIZE >> reg->hr_block_bits;
1647 	reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1648 
1649 	mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1650 	     reg->hr_start_block, reg->hr_blocks);
1651 	mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1652 	     reg->hr_block_bytes, reg->hr_block_bits);
1653 	mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1654 	mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1655 }
1656 
o2hb_map_slot_data(struct o2hb_region * reg)1657 static int o2hb_map_slot_data(struct o2hb_region *reg)
1658 {
1659 	int i, j;
1660 	unsigned int last_slot;
1661 	unsigned int spp = reg->hr_slots_per_page;
1662 	struct page *page;
1663 	char *raw;
1664 	struct o2hb_disk_slot *slot;
1665 
1666 	reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1667 	if (reg->hr_tmp_block == NULL)
1668 		return -ENOMEM;
1669 
1670 	reg->hr_slots = kcalloc(reg->hr_blocks,
1671 				sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1672 	if (reg->hr_slots == NULL)
1673 		return -ENOMEM;
1674 
1675 	for(i = 0; i < reg->hr_blocks; i++) {
1676 		slot = &reg->hr_slots[i];
1677 		slot->ds_node_num = i;
1678 		INIT_LIST_HEAD(&slot->ds_live_item);
1679 		slot->ds_raw_block = NULL;
1680 	}
1681 
1682 	reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1683 	mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1684 			   "at %u blocks per page\n",
1685 	     reg->hr_num_pages, reg->hr_blocks, spp);
1686 
1687 	reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1688 				    GFP_KERNEL);
1689 	if (!reg->hr_slot_data)
1690 		return -ENOMEM;
1691 
1692 	for(i = 0; i < reg->hr_num_pages; i++) {
1693 		page = alloc_page(GFP_KERNEL);
1694 		if (!page)
1695 			return -ENOMEM;
1696 
1697 		reg->hr_slot_data[i] = page;
1698 
1699 		last_slot = i * spp;
1700 		raw = page_address(page);
1701 		for (j = 0;
1702 		     (j < spp) && ((j + last_slot) < reg->hr_blocks);
1703 		     j++) {
1704 			BUG_ON((j + last_slot) >= reg->hr_blocks);
1705 
1706 			slot = &reg->hr_slots[j + last_slot];
1707 			slot->ds_raw_block =
1708 				(struct o2hb_disk_heartbeat_block *) raw;
1709 
1710 			raw += reg->hr_block_bytes;
1711 		}
1712 	}
1713 
1714 	return 0;
1715 }
1716 
1717 /* Read in all the slots available and populate the tracking
1718  * structures so that we can start with a baseline idea of what's
1719  * there. */
o2hb_populate_slot_data(struct o2hb_region * reg)1720 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1721 {
1722 	int ret, i;
1723 	struct o2hb_disk_slot *slot;
1724 	struct o2hb_disk_heartbeat_block *hb_block;
1725 
1726 	ret = o2hb_read_slots(reg, 0, reg->hr_blocks);
1727 	if (ret)
1728 		goto out;
1729 
1730 	/* We only want to get an idea of the values initially in each
1731 	 * slot, so we do no verification - o2hb_check_slot will
1732 	 * actually determine if each configured slot is valid and
1733 	 * whether any values have changed. */
1734 	for(i = 0; i < reg->hr_blocks; i++) {
1735 		slot = &reg->hr_slots[i];
1736 		hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1737 
1738 		/* Only fill the values that o2hb_check_slot uses to
1739 		 * determine changing slots */
1740 		slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1741 		slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1742 	}
1743 
1744 out:
1745 	return ret;
1746 }
1747 
1748 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
o2hb_region_dev_store(struct config_item * item,const char * page,size_t count)1749 static ssize_t o2hb_region_dev_store(struct config_item *item,
1750 				     const char *page,
1751 				     size_t count)
1752 {
1753 	struct o2hb_region *reg = to_o2hb_region(item);
1754 	struct task_struct *hb_task;
1755 	long fd;
1756 	int sectsize;
1757 	char *p = (char *)page;
1758 	struct fd f;
1759 	ssize_t ret = -EINVAL;
1760 	int live_threshold;
1761 
1762 	if (reg->hr_bdev)
1763 		goto out;
1764 
1765 	/* We can't heartbeat without having had our node number
1766 	 * configured yet. */
1767 	if (o2nm_this_node() == O2NM_MAX_NODES)
1768 		goto out;
1769 
1770 	fd = simple_strtol(p, &p, 0);
1771 	if (!p || (*p && (*p != '\n')))
1772 		goto out;
1773 
1774 	if (fd < 0 || fd >= INT_MAX)
1775 		goto out;
1776 
1777 	f = fdget(fd);
1778 	if (f.file == NULL)
1779 		goto out;
1780 
1781 	if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1782 	    reg->hr_block_bytes == 0)
1783 		goto out2;
1784 
1785 	if (!S_ISBLK(f.file->f_mapping->host->i_mode))
1786 		goto out2;
1787 
1788 	reg->hr_bdev = blkdev_get_by_dev(f.file->f_mapping->host->i_rdev,
1789 					 BLK_OPEN_WRITE | BLK_OPEN_READ, NULL,
1790 					 NULL);
1791 	if (IS_ERR(reg->hr_bdev)) {
1792 		ret = PTR_ERR(reg->hr_bdev);
1793 		reg->hr_bdev = NULL;
1794 		goto out2;
1795 	}
1796 
1797 	sectsize = bdev_logical_block_size(reg->hr_bdev);
1798 	if (sectsize != reg->hr_block_bytes) {
1799 		mlog(ML_ERROR,
1800 		     "blocksize %u incorrect for device, expected %d",
1801 		     reg->hr_block_bytes, sectsize);
1802 		ret = -EINVAL;
1803 		goto out3;
1804 	}
1805 
1806 	o2hb_init_region_params(reg);
1807 
1808 	/* Generation of zero is invalid */
1809 	do {
1810 		get_random_bytes(&reg->hr_generation,
1811 				 sizeof(reg->hr_generation));
1812 	} while (reg->hr_generation == 0);
1813 
1814 	ret = o2hb_map_slot_data(reg);
1815 	if (ret) {
1816 		mlog_errno(ret);
1817 		goto out3;
1818 	}
1819 
1820 	ret = o2hb_populate_slot_data(reg);
1821 	if (ret) {
1822 		mlog_errno(ret);
1823 		goto out3;
1824 	}
1825 
1826 	INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1827 	INIT_DELAYED_WORK(&reg->hr_nego_timeout_work, o2hb_nego_timeout);
1828 
1829 	/*
1830 	 * A node is considered live after it has beat LIVE_THRESHOLD
1831 	 * times.  We're not steady until we've given them a chance
1832 	 * _after_ our first read.
1833 	 * The default threshold is bare minimum so as to limit the delay
1834 	 * during mounts. For global heartbeat, the threshold doubled for the
1835 	 * first region.
1836 	 */
1837 	live_threshold = O2HB_LIVE_THRESHOLD;
1838 	if (o2hb_global_heartbeat_active()) {
1839 		spin_lock(&o2hb_live_lock);
1840 		if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1841 			live_threshold <<= 1;
1842 		spin_unlock(&o2hb_live_lock);
1843 	}
1844 	++live_threshold;
1845 	atomic_set(&reg->hr_steady_iterations, live_threshold);
1846 	/* unsteady_iterations is triple the steady_iterations */
1847 	atomic_set(&reg->hr_unsteady_iterations, (live_threshold * 3));
1848 
1849 	hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1850 			      reg->hr_item.ci_name);
1851 	if (IS_ERR(hb_task)) {
1852 		ret = PTR_ERR(hb_task);
1853 		mlog_errno(ret);
1854 		goto out3;
1855 	}
1856 
1857 	spin_lock(&o2hb_live_lock);
1858 	reg->hr_task = hb_task;
1859 	spin_unlock(&o2hb_live_lock);
1860 
1861 	ret = wait_event_interruptible(o2hb_steady_queue,
1862 				atomic_read(&reg->hr_steady_iterations) == 0 ||
1863 				reg->hr_node_deleted);
1864 	if (ret) {
1865 		atomic_set(&reg->hr_steady_iterations, 0);
1866 		reg->hr_aborted_start = 1;
1867 	}
1868 
1869 	if (reg->hr_aborted_start) {
1870 		ret = -EIO;
1871 		goto out3;
1872 	}
1873 
1874 	if (reg->hr_node_deleted) {
1875 		ret = -EINVAL;
1876 		goto out3;
1877 	}
1878 
1879 	/* Ok, we were woken.  Make sure it wasn't by drop_item() */
1880 	spin_lock(&o2hb_live_lock);
1881 	hb_task = reg->hr_task;
1882 	if (o2hb_global_heartbeat_active())
1883 		set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1884 	spin_unlock(&o2hb_live_lock);
1885 
1886 	if (hb_task)
1887 		ret = count;
1888 	else
1889 		ret = -EIO;
1890 
1891 	if (hb_task && o2hb_global_heartbeat_active())
1892 		printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%pg)\n",
1893 		       config_item_name(&reg->hr_item), reg->hr_bdev);
1894 
1895 out3:
1896 	if (ret < 0) {
1897 		blkdev_put(reg->hr_bdev, NULL);
1898 		reg->hr_bdev = NULL;
1899 	}
1900 out2:
1901 	fdput(f);
1902 out:
1903 	return ret;
1904 }
1905 
o2hb_region_pid_show(struct config_item * item,char * page)1906 static ssize_t o2hb_region_pid_show(struct config_item *item, char *page)
1907 {
1908 	struct o2hb_region *reg = to_o2hb_region(item);
1909 	pid_t pid = 0;
1910 
1911 	spin_lock(&o2hb_live_lock);
1912 	if (reg->hr_task)
1913 		pid = task_pid_nr(reg->hr_task);
1914 	spin_unlock(&o2hb_live_lock);
1915 
1916 	if (!pid)
1917 		return 0;
1918 
1919 	return sprintf(page, "%u\n", pid);
1920 }
1921 
1922 CONFIGFS_ATTR(o2hb_region_, block_bytes);
1923 CONFIGFS_ATTR(o2hb_region_, start_block);
1924 CONFIGFS_ATTR(o2hb_region_, blocks);
1925 CONFIGFS_ATTR(o2hb_region_, dev);
1926 CONFIGFS_ATTR_RO(o2hb_region_, pid);
1927 
1928 static struct configfs_attribute *o2hb_region_attrs[] = {
1929 	&o2hb_region_attr_block_bytes,
1930 	&o2hb_region_attr_start_block,
1931 	&o2hb_region_attr_blocks,
1932 	&o2hb_region_attr_dev,
1933 	&o2hb_region_attr_pid,
1934 	NULL,
1935 };
1936 
1937 static struct configfs_item_operations o2hb_region_item_ops = {
1938 	.release		= o2hb_region_release,
1939 };
1940 
1941 static const struct config_item_type o2hb_region_type = {
1942 	.ct_item_ops	= &o2hb_region_item_ops,
1943 	.ct_attrs	= o2hb_region_attrs,
1944 	.ct_owner	= THIS_MODULE,
1945 };
1946 
1947 /* heartbeat set */
1948 
1949 struct o2hb_heartbeat_group {
1950 	struct config_group hs_group;
1951 	/* some stuff? */
1952 };
1953 
to_o2hb_heartbeat_group(struct config_group * group)1954 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1955 {
1956 	return group ?
1957 		container_of(group, struct o2hb_heartbeat_group, hs_group)
1958 		: NULL;
1959 }
1960 
o2hb_debug_region_init(struct o2hb_region * reg,struct dentry * parent)1961 static void o2hb_debug_region_init(struct o2hb_region *reg,
1962 				   struct dentry *parent)
1963 {
1964 	struct dentry *dir;
1965 
1966 	dir = debugfs_create_dir(config_item_name(&reg->hr_item), parent);
1967 	reg->hr_debug_dir = dir;
1968 
1969 	o2hb_debug_create(O2HB_DEBUG_LIVENODES, dir, &(reg->hr_db_livenodes),
1970 			  sizeof(*(reg->hr_db_livenodes)),
1971 			  O2HB_DB_TYPE_REGION_LIVENODES,
1972 			  sizeof(reg->hr_live_node_bitmap), O2NM_MAX_NODES,
1973 			  reg);
1974 
1975 	o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER, dir, &(reg->hr_db_regnum),
1976 			  sizeof(*(reg->hr_db_regnum)),
1977 			  O2HB_DB_TYPE_REGION_NUMBER, 0, O2NM_MAX_NODES, reg);
1978 
1979 	o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME, dir,
1980 			  &(reg->hr_db_elapsed_time),
1981 			  sizeof(*(reg->hr_db_elapsed_time)),
1982 			  O2HB_DB_TYPE_REGION_ELAPSED_TIME, 0, 0, reg);
1983 
1984 	o2hb_debug_create(O2HB_DEBUG_REGION_PINNED, dir, &(reg->hr_db_pinned),
1985 			  sizeof(*(reg->hr_db_pinned)),
1986 			  O2HB_DB_TYPE_REGION_PINNED, 0, 0, reg);
1987 
1988 }
1989 
o2hb_heartbeat_group_make_item(struct config_group * group,const char * name)1990 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
1991 							  const char *name)
1992 {
1993 	struct o2hb_region *reg = NULL;
1994 	int ret;
1995 
1996 	reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
1997 	if (reg == NULL)
1998 		return ERR_PTR(-ENOMEM);
1999 
2000 	if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
2001 		ret = -ENAMETOOLONG;
2002 		goto free;
2003 	}
2004 
2005 	spin_lock(&o2hb_live_lock);
2006 	reg->hr_region_num = 0;
2007 	if (o2hb_global_heartbeat_active()) {
2008 		reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
2009 							 O2NM_MAX_REGIONS);
2010 		if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
2011 			spin_unlock(&o2hb_live_lock);
2012 			ret = -EFBIG;
2013 			goto free;
2014 		}
2015 		set_bit(reg->hr_region_num, o2hb_region_bitmap);
2016 	}
2017 	list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
2018 	spin_unlock(&o2hb_live_lock);
2019 
2020 	config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
2021 
2022 	/* this is the same way to generate msg key as dlm, for local heartbeat,
2023 	 * name is also the same, so make initial crc value different to avoid
2024 	 * message key conflict.
2025 	 */
2026 	reg->hr_key = crc32_le(reg->hr_region_num + O2NM_MAX_REGIONS,
2027 		name, strlen(name));
2028 	INIT_LIST_HEAD(&reg->hr_handler_list);
2029 	ret = o2net_register_handler(O2HB_NEGO_TIMEOUT_MSG, reg->hr_key,
2030 			sizeof(struct o2hb_nego_msg),
2031 			o2hb_nego_timeout_handler,
2032 			reg, NULL, &reg->hr_handler_list);
2033 	if (ret)
2034 		goto remove_item;
2035 
2036 	ret = o2net_register_handler(O2HB_NEGO_APPROVE_MSG, reg->hr_key,
2037 			sizeof(struct o2hb_nego_msg),
2038 			o2hb_nego_approve_handler,
2039 			reg, NULL, &reg->hr_handler_list);
2040 	if (ret)
2041 		goto unregister_handler;
2042 
2043 	o2hb_debug_region_init(reg, o2hb_debug_dir);
2044 
2045 	return &reg->hr_item;
2046 
2047 unregister_handler:
2048 	o2net_unregister_handler_list(&reg->hr_handler_list);
2049 remove_item:
2050 	spin_lock(&o2hb_live_lock);
2051 	list_del(&reg->hr_all_item);
2052 	if (o2hb_global_heartbeat_active())
2053 		clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2054 	spin_unlock(&o2hb_live_lock);
2055 free:
2056 	kfree(reg);
2057 	return ERR_PTR(ret);
2058 }
2059 
o2hb_heartbeat_group_drop_item(struct config_group * group,struct config_item * item)2060 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2061 					   struct config_item *item)
2062 {
2063 	struct task_struct *hb_task;
2064 	struct o2hb_region *reg = to_o2hb_region(item);
2065 	int quorum_region = 0;
2066 
2067 	/* stop the thread when the user removes the region dir */
2068 	spin_lock(&o2hb_live_lock);
2069 	hb_task = reg->hr_task;
2070 	reg->hr_task = NULL;
2071 	reg->hr_item_dropped = 1;
2072 	spin_unlock(&o2hb_live_lock);
2073 
2074 	if (hb_task)
2075 		kthread_stop(hb_task);
2076 
2077 	if (o2hb_global_heartbeat_active()) {
2078 		spin_lock(&o2hb_live_lock);
2079 		clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2080 		clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2081 		if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2082 			quorum_region = 1;
2083 		clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2084 		spin_unlock(&o2hb_live_lock);
2085 		printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%pg)\n",
2086 		       ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2087 			"stopped" : "start aborted"), config_item_name(item),
2088 		       reg->hr_bdev);
2089 	}
2090 
2091 	/*
2092 	 * If we're racing a dev_write(), we need to wake them.  They will
2093 	 * check reg->hr_task
2094 	 */
2095 	if (atomic_read(&reg->hr_steady_iterations) != 0) {
2096 		reg->hr_aborted_start = 1;
2097 		atomic_set(&reg->hr_steady_iterations, 0);
2098 		wake_up(&o2hb_steady_queue);
2099 	}
2100 
2101 	config_item_put(item);
2102 
2103 	if (!o2hb_global_heartbeat_active() || !quorum_region)
2104 		return;
2105 
2106 	/*
2107 	 * If global heartbeat active and there are dependent users,
2108 	 * pin all regions if quorum region count <= CUT_OFF
2109 	 */
2110 	spin_lock(&o2hb_live_lock);
2111 
2112 	if (!o2hb_dependent_users)
2113 		goto unlock;
2114 
2115 	if (bitmap_weight(o2hb_quorum_region_bitmap,
2116 			   O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2117 		o2hb_region_pin(NULL);
2118 
2119 unlock:
2120 	spin_unlock(&o2hb_live_lock);
2121 }
2122 
o2hb_heartbeat_group_dead_threshold_show(struct config_item * item,char * page)2123 static ssize_t o2hb_heartbeat_group_dead_threshold_show(struct config_item *item,
2124 		char *page)
2125 {
2126 	return sprintf(page, "%u\n", o2hb_dead_threshold);
2127 }
2128 
o2hb_heartbeat_group_dead_threshold_store(struct config_item * item,const char * page,size_t count)2129 static ssize_t o2hb_heartbeat_group_dead_threshold_store(struct config_item *item,
2130 		const char *page, size_t count)
2131 {
2132 	unsigned long tmp;
2133 	char *p = (char *)page;
2134 
2135 	tmp = simple_strtoul(p, &p, 10);
2136 	if (!p || (*p && (*p != '\n')))
2137                 return -EINVAL;
2138 
2139 	/* this will validate ranges for us. */
2140 	o2hb_dead_threshold_set((unsigned int) tmp);
2141 
2142 	return count;
2143 }
2144 
o2hb_heartbeat_group_mode_show(struct config_item * item,char * page)2145 static ssize_t o2hb_heartbeat_group_mode_show(struct config_item *item,
2146 		char *page)
2147 {
2148 	return sprintf(page, "%s\n",
2149 		       o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2150 }
2151 
o2hb_heartbeat_group_mode_store(struct config_item * item,const char * page,size_t count)2152 static ssize_t o2hb_heartbeat_group_mode_store(struct config_item *item,
2153 		const char *page, size_t count)
2154 {
2155 	unsigned int i;
2156 	int ret;
2157 	size_t len;
2158 
2159 	len = (page[count - 1] == '\n') ? count - 1 : count;
2160 	if (!len)
2161 		return -EINVAL;
2162 
2163 	for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2164 		if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
2165 			continue;
2166 
2167 		ret = o2hb_global_heartbeat_mode_set(i);
2168 		if (!ret)
2169 			printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2170 			       o2hb_heartbeat_mode_desc[i]);
2171 		return count;
2172 	}
2173 
2174 	return -EINVAL;
2175 
2176 }
2177 
2178 CONFIGFS_ATTR(o2hb_heartbeat_group_, dead_threshold);
2179 CONFIGFS_ATTR(o2hb_heartbeat_group_, mode);
2180 
2181 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2182 	&o2hb_heartbeat_group_attr_dead_threshold,
2183 	&o2hb_heartbeat_group_attr_mode,
2184 	NULL,
2185 };
2186 
2187 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2188 	.make_item	= o2hb_heartbeat_group_make_item,
2189 	.drop_item	= o2hb_heartbeat_group_drop_item,
2190 };
2191 
2192 static const struct config_item_type o2hb_heartbeat_group_type = {
2193 	.ct_group_ops	= &o2hb_heartbeat_group_group_ops,
2194 	.ct_attrs	= o2hb_heartbeat_group_attrs,
2195 	.ct_owner	= THIS_MODULE,
2196 };
2197 
2198 /* this is just here to avoid touching group in heartbeat.h which the
2199  * entire damn world #includes */
o2hb_alloc_hb_set(void)2200 struct config_group *o2hb_alloc_hb_set(void)
2201 {
2202 	struct o2hb_heartbeat_group *hs = NULL;
2203 	struct config_group *ret = NULL;
2204 
2205 	hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2206 	if (hs == NULL)
2207 		goto out;
2208 
2209 	config_group_init_type_name(&hs->hs_group, "heartbeat",
2210 				    &o2hb_heartbeat_group_type);
2211 
2212 	ret = &hs->hs_group;
2213 out:
2214 	if (ret == NULL)
2215 		kfree(hs);
2216 	return ret;
2217 }
2218 
o2hb_free_hb_set(struct config_group * group)2219 void o2hb_free_hb_set(struct config_group *group)
2220 {
2221 	struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2222 	kfree(hs);
2223 }
2224 
2225 /* hb callback registration and issuing */
2226 
hbcall_from_type(enum o2hb_callback_type type)2227 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2228 {
2229 	if (type == O2HB_NUM_CB)
2230 		return ERR_PTR(-EINVAL);
2231 
2232 	return &o2hb_callbacks[type];
2233 }
2234 
o2hb_setup_callback(struct o2hb_callback_func * hc,enum o2hb_callback_type type,o2hb_cb_func * func,void * data,int priority)2235 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2236 			 enum o2hb_callback_type type,
2237 			 o2hb_cb_func *func,
2238 			 void *data,
2239 			 int priority)
2240 {
2241 	INIT_LIST_HEAD(&hc->hc_item);
2242 	hc->hc_func = func;
2243 	hc->hc_data = data;
2244 	hc->hc_priority = priority;
2245 	hc->hc_type = type;
2246 	hc->hc_magic = O2HB_CB_MAGIC;
2247 }
2248 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2249 
2250 /*
2251  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2252  * In global heartbeat mode, region_uuid passed is NULL.
2253  *
2254  * In local, we only pin the matching region. In global we pin all the active
2255  * regions.
2256  */
o2hb_region_pin(const char * region_uuid)2257 static int o2hb_region_pin(const char *region_uuid)
2258 {
2259 	int ret = 0, found = 0;
2260 	struct o2hb_region *reg;
2261 	char *uuid;
2262 
2263 	assert_spin_locked(&o2hb_live_lock);
2264 
2265 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2266 		if (reg->hr_item_dropped)
2267 			continue;
2268 
2269 		uuid = config_item_name(&reg->hr_item);
2270 
2271 		/* local heartbeat */
2272 		if (region_uuid) {
2273 			if (strcmp(region_uuid, uuid))
2274 				continue;
2275 			found = 1;
2276 		}
2277 
2278 		if (reg->hr_item_pinned || reg->hr_item_dropped)
2279 			goto skip_pin;
2280 
2281 		/* Ignore ENOENT only for local hb (userdlm domain) */
2282 		ret = o2nm_depend_item(&reg->hr_item);
2283 		if (!ret) {
2284 			mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2285 			reg->hr_item_pinned = 1;
2286 		} else {
2287 			if (ret == -ENOENT && found)
2288 				ret = 0;
2289 			else {
2290 				mlog(ML_ERROR, "Pin region %s fails with %d\n",
2291 				     uuid, ret);
2292 				break;
2293 			}
2294 		}
2295 skip_pin:
2296 		if (found)
2297 			break;
2298 	}
2299 
2300 	return ret;
2301 }
2302 
2303 /*
2304  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2305  * In global heartbeat mode, region_uuid passed is NULL.
2306  *
2307  * In local, we only unpin the matching region. In global we unpin all the
2308  * active regions.
2309  */
o2hb_region_unpin(const char * region_uuid)2310 static void o2hb_region_unpin(const char *region_uuid)
2311 {
2312 	struct o2hb_region *reg;
2313 	char *uuid;
2314 	int found = 0;
2315 
2316 	assert_spin_locked(&o2hb_live_lock);
2317 
2318 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2319 		if (reg->hr_item_dropped)
2320 			continue;
2321 
2322 		uuid = config_item_name(&reg->hr_item);
2323 		if (region_uuid) {
2324 			if (strcmp(region_uuid, uuid))
2325 				continue;
2326 			found = 1;
2327 		}
2328 
2329 		if (reg->hr_item_pinned) {
2330 			mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2331 			o2nm_undepend_item(&reg->hr_item);
2332 			reg->hr_item_pinned = 0;
2333 		}
2334 		if (found)
2335 			break;
2336 	}
2337 }
2338 
o2hb_region_inc_user(const char * region_uuid)2339 static int o2hb_region_inc_user(const char *region_uuid)
2340 {
2341 	int ret = 0;
2342 
2343 	spin_lock(&o2hb_live_lock);
2344 
2345 	/* local heartbeat */
2346 	if (!o2hb_global_heartbeat_active()) {
2347 	    ret = o2hb_region_pin(region_uuid);
2348 	    goto unlock;
2349 	}
2350 
2351 	/*
2352 	 * if global heartbeat active and this is the first dependent user,
2353 	 * pin all regions if quorum region count <= CUT_OFF
2354 	 */
2355 	o2hb_dependent_users++;
2356 	if (o2hb_dependent_users > 1)
2357 		goto unlock;
2358 
2359 	if (bitmap_weight(o2hb_quorum_region_bitmap,
2360 			   O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2361 		ret = o2hb_region_pin(NULL);
2362 
2363 unlock:
2364 	spin_unlock(&o2hb_live_lock);
2365 	return ret;
2366 }
2367 
o2hb_region_dec_user(const char * region_uuid)2368 static void o2hb_region_dec_user(const char *region_uuid)
2369 {
2370 	spin_lock(&o2hb_live_lock);
2371 
2372 	/* local heartbeat */
2373 	if (!o2hb_global_heartbeat_active()) {
2374 	    o2hb_region_unpin(region_uuid);
2375 	    goto unlock;
2376 	}
2377 
2378 	/*
2379 	 * if global heartbeat active and there are no dependent users,
2380 	 * unpin all quorum regions
2381 	 */
2382 	o2hb_dependent_users--;
2383 	if (!o2hb_dependent_users)
2384 		o2hb_region_unpin(NULL);
2385 
2386 unlock:
2387 	spin_unlock(&o2hb_live_lock);
2388 }
2389 
o2hb_register_callback(const char * region_uuid,struct o2hb_callback_func * hc)2390 int o2hb_register_callback(const char *region_uuid,
2391 			   struct o2hb_callback_func *hc)
2392 {
2393 	struct o2hb_callback_func *f;
2394 	struct o2hb_callback *hbcall;
2395 	int ret;
2396 
2397 	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2398 	BUG_ON(!list_empty(&hc->hc_item));
2399 
2400 	hbcall = hbcall_from_type(hc->hc_type);
2401 	if (IS_ERR(hbcall)) {
2402 		ret = PTR_ERR(hbcall);
2403 		goto out;
2404 	}
2405 
2406 	if (region_uuid) {
2407 		ret = o2hb_region_inc_user(region_uuid);
2408 		if (ret) {
2409 			mlog_errno(ret);
2410 			goto out;
2411 		}
2412 	}
2413 
2414 	down_write(&o2hb_callback_sem);
2415 
2416 	list_for_each_entry(f, &hbcall->list, hc_item) {
2417 		if (hc->hc_priority < f->hc_priority) {
2418 			list_add_tail(&hc->hc_item, &f->hc_item);
2419 			break;
2420 		}
2421 	}
2422 	if (list_empty(&hc->hc_item))
2423 		list_add_tail(&hc->hc_item, &hbcall->list);
2424 
2425 	up_write(&o2hb_callback_sem);
2426 	ret = 0;
2427 out:
2428 	mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2429 	     ret, __builtin_return_address(0), hc);
2430 	return ret;
2431 }
2432 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2433 
o2hb_unregister_callback(const char * region_uuid,struct o2hb_callback_func * hc)2434 void o2hb_unregister_callback(const char *region_uuid,
2435 			      struct o2hb_callback_func *hc)
2436 {
2437 	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2438 
2439 	mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2440 	     __builtin_return_address(0), hc);
2441 
2442 	/* XXX Can this happen _with_ a region reference? */
2443 	if (list_empty(&hc->hc_item))
2444 		return;
2445 
2446 	if (region_uuid)
2447 		o2hb_region_dec_user(region_uuid);
2448 
2449 	down_write(&o2hb_callback_sem);
2450 
2451 	list_del_init(&hc->hc_item);
2452 
2453 	up_write(&o2hb_callback_sem);
2454 }
2455 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2456 
o2hb_check_node_heartbeating_no_sem(u8 node_num)2457 int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2458 {
2459 	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2460 
2461 	spin_lock(&o2hb_live_lock);
2462 	o2hb_fill_node_map_from_callback(testing_map, O2NM_MAX_NODES);
2463 	spin_unlock(&o2hb_live_lock);
2464 	if (!test_bit(node_num, testing_map)) {
2465 		mlog(ML_HEARTBEAT,
2466 		     "node (%u) does not have heartbeating enabled.\n",
2467 		     node_num);
2468 		return 0;
2469 	}
2470 
2471 	return 1;
2472 }
2473 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2474 
o2hb_check_node_heartbeating_from_callback(u8 node_num)2475 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2476 {
2477 	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2478 
2479 	o2hb_fill_node_map_from_callback(testing_map, O2NM_MAX_NODES);
2480 	if (!test_bit(node_num, testing_map)) {
2481 		mlog(ML_HEARTBEAT,
2482 		     "node (%u) does not have heartbeating enabled.\n",
2483 		     node_num);
2484 		return 0;
2485 	}
2486 
2487 	return 1;
2488 }
2489 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2490 
2491 /*
2492  * this is just a hack until we get the plumbing which flips file systems
2493  * read only and drops the hb ref instead of killing the node dead.
2494  */
o2hb_stop_all_regions(void)2495 void o2hb_stop_all_regions(void)
2496 {
2497 	struct o2hb_region *reg;
2498 
2499 	mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2500 
2501 	spin_lock(&o2hb_live_lock);
2502 
2503 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2504 		reg->hr_unclean_stop = 1;
2505 
2506 	spin_unlock(&o2hb_live_lock);
2507 }
2508 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2509 
o2hb_get_all_regions(char * region_uuids,u8 max_regions)2510 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2511 {
2512 	struct o2hb_region *reg;
2513 	int numregs = 0;
2514 	char *p;
2515 
2516 	spin_lock(&o2hb_live_lock);
2517 
2518 	p = region_uuids;
2519 	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2520 		if (reg->hr_item_dropped)
2521 			continue;
2522 
2523 		mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2524 		if (numregs < max_regions) {
2525 			memcpy(p, config_item_name(&reg->hr_item),
2526 			       O2HB_MAX_REGION_NAME_LEN);
2527 			p += O2HB_MAX_REGION_NAME_LEN;
2528 		}
2529 		numregs++;
2530 	}
2531 
2532 	spin_unlock(&o2hb_live_lock);
2533 
2534 	return numregs;
2535 }
2536 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2537 
o2hb_global_heartbeat_active(void)2538 int o2hb_global_heartbeat_active(void)
2539 {
2540 	return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2541 }
2542 EXPORT_SYMBOL(o2hb_global_heartbeat_active);
2543