xref: /openbmc/linux/drivers/md/md-cluster.c (revision 82e6fdd6)
1 /*
2  * Copyright (C) 2015, SUSE
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License as published by
6  * the Free Software Foundation; either version 2, or (at your option)
7  * any later version.
8  *
9  */
10 
11 
12 #include <linux/module.h>
13 #include <linux/kthread.h>
14 #include <linux/dlm.h>
15 #include <linux/sched.h>
16 #include <linux/raid/md_p.h>
17 #include "md.h"
18 #include "md-bitmap.h"
19 #include "md-cluster.h"
20 
21 #define LVB_SIZE	64
22 #define NEW_DEV_TIMEOUT 5000
23 
24 struct dlm_lock_resource {
25 	dlm_lockspace_t *ls;
26 	struct dlm_lksb lksb;
27 	char *name; /* lock name. */
28 	uint32_t flags; /* flags to pass to dlm_lock() */
29 	wait_queue_head_t sync_locking; /* wait queue for synchronized locking */
30 	bool sync_locking_done;
31 	void (*bast)(void *arg, int mode); /* blocking AST function pointer*/
32 	struct mddev *mddev; /* pointing back to mddev. */
33 	int mode;
34 };
35 
36 struct suspend_info {
37 	int slot;
38 	sector_t lo;
39 	sector_t hi;
40 	struct list_head list;
41 };
42 
43 struct resync_info {
44 	__le64 lo;
45 	__le64 hi;
46 };
47 
48 /* md_cluster_info flags */
49 #define		MD_CLUSTER_WAITING_FOR_NEWDISK		1
50 #define		MD_CLUSTER_SUSPEND_READ_BALANCING	2
51 #define		MD_CLUSTER_BEGIN_JOIN_CLUSTER		3
52 
53 /* Lock the send communication. This is done through
54  * bit manipulation as opposed to a mutex in order to
55  * accomodate lock and hold. See next comment.
56  */
57 #define		MD_CLUSTER_SEND_LOCK			4
58 /* If cluster operations (such as adding a disk) must lock the
59  * communication channel, so as to perform extra operations
60  * (update metadata) and no other operation is allowed on the
61  * MD. Token needs to be locked and held until the operation
62  * completes witha md_update_sb(), which would eventually release
63  * the lock.
64  */
65 #define		MD_CLUSTER_SEND_LOCKED_ALREADY		5
66 /* We should receive message after node joined cluster and
67  * set up all the related infos such as bitmap and personality */
68 #define		MD_CLUSTER_ALREADY_IN_CLUSTER		6
69 #define		MD_CLUSTER_PENDING_RECV_EVENT		7
70 #define 	MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD		8
71 
72 struct md_cluster_info {
73 	struct mddev *mddev; /* the md device which md_cluster_info belongs to */
74 	/* dlm lock space and resources for clustered raid. */
75 	dlm_lockspace_t *lockspace;
76 	int slot_number;
77 	struct completion completion;
78 	struct mutex recv_mutex;
79 	struct dlm_lock_resource *bitmap_lockres;
80 	struct dlm_lock_resource **other_bitmap_lockres;
81 	struct dlm_lock_resource *resync_lockres;
82 	struct list_head suspend_list;
83 	spinlock_t suspend_lock;
84 	struct md_thread *recovery_thread;
85 	unsigned long recovery_map;
86 	/* communication loc resources */
87 	struct dlm_lock_resource *ack_lockres;
88 	struct dlm_lock_resource *message_lockres;
89 	struct dlm_lock_resource *token_lockres;
90 	struct dlm_lock_resource *no_new_dev_lockres;
91 	struct md_thread *recv_thread;
92 	struct completion newdisk_completion;
93 	wait_queue_head_t wait;
94 	unsigned long state;
95 	/* record the region in RESYNCING message */
96 	sector_t sync_low;
97 	sector_t sync_hi;
98 };
99 
100 enum msg_type {
101 	METADATA_UPDATED = 0,
102 	RESYNCING,
103 	NEWDISK,
104 	REMOVE,
105 	RE_ADD,
106 	BITMAP_NEEDS_SYNC,
107 	CHANGE_CAPACITY,
108 };
109 
110 struct cluster_msg {
111 	__le32 type;
112 	__le32 slot;
113 	/* TODO: Unionize this for smaller footprint */
114 	__le64 low;
115 	__le64 high;
116 	char uuid[16];
117 	__le32 raid_slot;
118 };
119 
120 static void sync_ast(void *arg)
121 {
122 	struct dlm_lock_resource *res;
123 
124 	res = arg;
125 	res->sync_locking_done = true;
126 	wake_up(&res->sync_locking);
127 }
128 
129 static int dlm_lock_sync(struct dlm_lock_resource *res, int mode)
130 {
131 	int ret = 0;
132 
133 	ret = dlm_lock(res->ls, mode, &res->lksb,
134 			res->flags, res->name, strlen(res->name),
135 			0, sync_ast, res, res->bast);
136 	if (ret)
137 		return ret;
138 	wait_event(res->sync_locking, res->sync_locking_done);
139 	res->sync_locking_done = false;
140 	if (res->lksb.sb_status == 0)
141 		res->mode = mode;
142 	return res->lksb.sb_status;
143 }
144 
145 static int dlm_unlock_sync(struct dlm_lock_resource *res)
146 {
147 	return dlm_lock_sync(res, DLM_LOCK_NL);
148 }
149 
150 /*
151  * An variation of dlm_lock_sync, which make lock request could
152  * be interrupted
153  */
154 static int dlm_lock_sync_interruptible(struct dlm_lock_resource *res, int mode,
155 				       struct mddev *mddev)
156 {
157 	int ret = 0;
158 
159 	ret = dlm_lock(res->ls, mode, &res->lksb,
160 			res->flags, res->name, strlen(res->name),
161 			0, sync_ast, res, res->bast);
162 	if (ret)
163 		return ret;
164 
165 	wait_event(res->sync_locking, res->sync_locking_done
166 				      || kthread_should_stop()
167 				      || test_bit(MD_CLOSING, &mddev->flags));
168 	if (!res->sync_locking_done) {
169 		/*
170 		 * the convert queue contains the lock request when request is
171 		 * interrupted, and sync_ast could still be run, so need to
172 		 * cancel the request and reset completion
173 		 */
174 		ret = dlm_unlock(res->ls, res->lksb.sb_lkid, DLM_LKF_CANCEL,
175 			&res->lksb, res);
176 		res->sync_locking_done = false;
177 		if (unlikely(ret != 0))
178 			pr_info("failed to cancel previous lock request "
179 				 "%s return %d\n", res->name, ret);
180 		return -EPERM;
181 	} else
182 		res->sync_locking_done = false;
183 	if (res->lksb.sb_status == 0)
184 		res->mode = mode;
185 	return res->lksb.sb_status;
186 }
187 
188 static struct dlm_lock_resource *lockres_init(struct mddev *mddev,
189 		char *name, void (*bastfn)(void *arg, int mode), int with_lvb)
190 {
191 	struct dlm_lock_resource *res = NULL;
192 	int ret, namelen;
193 	struct md_cluster_info *cinfo = mddev->cluster_info;
194 
195 	res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL);
196 	if (!res)
197 		return NULL;
198 	init_waitqueue_head(&res->sync_locking);
199 	res->sync_locking_done = false;
200 	res->ls = cinfo->lockspace;
201 	res->mddev = mddev;
202 	res->mode = DLM_LOCK_IV;
203 	namelen = strlen(name);
204 	res->name = kzalloc(namelen + 1, GFP_KERNEL);
205 	if (!res->name) {
206 		pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name);
207 		goto out_err;
208 	}
209 	strlcpy(res->name, name, namelen + 1);
210 	if (with_lvb) {
211 		res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL);
212 		if (!res->lksb.sb_lvbptr) {
213 			pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name);
214 			goto out_err;
215 		}
216 		res->flags = DLM_LKF_VALBLK;
217 	}
218 
219 	if (bastfn)
220 		res->bast = bastfn;
221 
222 	res->flags |= DLM_LKF_EXPEDITE;
223 
224 	ret = dlm_lock_sync(res, DLM_LOCK_NL);
225 	if (ret) {
226 		pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name);
227 		goto out_err;
228 	}
229 	res->flags &= ~DLM_LKF_EXPEDITE;
230 	res->flags |= DLM_LKF_CONVERT;
231 
232 	return res;
233 out_err:
234 	kfree(res->lksb.sb_lvbptr);
235 	kfree(res->name);
236 	kfree(res);
237 	return NULL;
238 }
239 
240 static void lockres_free(struct dlm_lock_resource *res)
241 {
242 	int ret = 0;
243 
244 	if (!res)
245 		return;
246 
247 	/*
248 	 * use FORCEUNLOCK flag, so we can unlock even the lock is on the
249 	 * waiting or convert queue
250 	 */
251 	ret = dlm_unlock(res->ls, res->lksb.sb_lkid, DLM_LKF_FORCEUNLOCK,
252 		&res->lksb, res);
253 	if (unlikely(ret != 0))
254 		pr_err("failed to unlock %s return %d\n", res->name, ret);
255 	else
256 		wait_event(res->sync_locking, res->sync_locking_done);
257 
258 	kfree(res->name);
259 	kfree(res->lksb.sb_lvbptr);
260 	kfree(res);
261 }
262 
263 static void add_resync_info(struct dlm_lock_resource *lockres,
264 			    sector_t lo, sector_t hi)
265 {
266 	struct resync_info *ri;
267 
268 	ri = (struct resync_info *)lockres->lksb.sb_lvbptr;
269 	ri->lo = cpu_to_le64(lo);
270 	ri->hi = cpu_to_le64(hi);
271 }
272 
273 static struct suspend_info *read_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres)
274 {
275 	struct resync_info ri;
276 	struct suspend_info *s = NULL;
277 	sector_t hi = 0;
278 
279 	dlm_lock_sync(lockres, DLM_LOCK_CR);
280 	memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info));
281 	hi = le64_to_cpu(ri.hi);
282 	if (hi > 0) {
283 		s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL);
284 		if (!s)
285 			goto out;
286 		s->hi = hi;
287 		s->lo = le64_to_cpu(ri.lo);
288 	}
289 	dlm_unlock_sync(lockres);
290 out:
291 	return s;
292 }
293 
294 static void recover_bitmaps(struct md_thread *thread)
295 {
296 	struct mddev *mddev = thread->mddev;
297 	struct md_cluster_info *cinfo = mddev->cluster_info;
298 	struct dlm_lock_resource *bm_lockres;
299 	char str[64];
300 	int slot, ret;
301 	struct suspend_info *s, *tmp;
302 	sector_t lo, hi;
303 
304 	while (cinfo->recovery_map) {
305 		slot = fls64((u64)cinfo->recovery_map) - 1;
306 
307 		/* Clear suspend_area associated with the bitmap */
308 		spin_lock_irq(&cinfo->suspend_lock);
309 		list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list)
310 			if (slot == s->slot) {
311 				list_del(&s->list);
312 				kfree(s);
313 			}
314 		spin_unlock_irq(&cinfo->suspend_lock);
315 
316 		snprintf(str, 64, "bitmap%04d", slot);
317 		bm_lockres = lockres_init(mddev, str, NULL, 1);
318 		if (!bm_lockres) {
319 			pr_err("md-cluster: Cannot initialize bitmaps\n");
320 			goto clear_bit;
321 		}
322 
323 		ret = dlm_lock_sync_interruptible(bm_lockres, DLM_LOCK_PW, mddev);
324 		if (ret) {
325 			pr_err("md-cluster: Could not DLM lock %s: %d\n",
326 					str, ret);
327 			goto clear_bit;
328 		}
329 		ret = bitmap_copy_from_slot(mddev, slot, &lo, &hi, true);
330 		if (ret) {
331 			pr_err("md-cluster: Could not copy data from bitmap %d\n", slot);
332 			goto clear_bit;
333 		}
334 		if (hi > 0) {
335 			if (lo < mddev->recovery_cp)
336 				mddev->recovery_cp = lo;
337 			/* wake up thread to continue resync in case resync
338 			 * is not finished */
339 			if (mddev->recovery_cp != MaxSector) {
340 			    set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
341 			    md_wakeup_thread(mddev->thread);
342 			}
343 		}
344 clear_bit:
345 		lockres_free(bm_lockres);
346 		clear_bit(slot, &cinfo->recovery_map);
347 	}
348 }
349 
350 static void recover_prep(void *arg)
351 {
352 	struct mddev *mddev = arg;
353 	struct md_cluster_info *cinfo = mddev->cluster_info;
354 	set_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state);
355 }
356 
357 static void __recover_slot(struct mddev *mddev, int slot)
358 {
359 	struct md_cluster_info *cinfo = mddev->cluster_info;
360 
361 	set_bit(slot, &cinfo->recovery_map);
362 	if (!cinfo->recovery_thread) {
363 		cinfo->recovery_thread = md_register_thread(recover_bitmaps,
364 				mddev, "recover");
365 		if (!cinfo->recovery_thread) {
366 			pr_warn("md-cluster: Could not create recovery thread\n");
367 			return;
368 		}
369 	}
370 	md_wakeup_thread(cinfo->recovery_thread);
371 }
372 
373 static void recover_slot(void *arg, struct dlm_slot *slot)
374 {
375 	struct mddev *mddev = arg;
376 	struct md_cluster_info *cinfo = mddev->cluster_info;
377 
378 	pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n",
379 			mddev->bitmap_info.cluster_name,
380 			slot->nodeid, slot->slot,
381 			cinfo->slot_number);
382 	/* deduct one since dlm slot starts from one while the num of
383 	 * cluster-md begins with 0 */
384 	__recover_slot(mddev, slot->slot - 1);
385 }
386 
387 static void recover_done(void *arg, struct dlm_slot *slots,
388 		int num_slots, int our_slot,
389 		uint32_t generation)
390 {
391 	struct mddev *mddev = arg;
392 	struct md_cluster_info *cinfo = mddev->cluster_info;
393 
394 	cinfo->slot_number = our_slot;
395 	/* completion is only need to be complete when node join cluster,
396 	 * it doesn't need to run during another node's failure */
397 	if (test_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state)) {
398 		complete(&cinfo->completion);
399 		clear_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state);
400 	}
401 	clear_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state);
402 }
403 
404 /* the ops is called when node join the cluster, and do lock recovery
405  * if node failure occurs */
406 static const struct dlm_lockspace_ops md_ls_ops = {
407 	.recover_prep = recover_prep,
408 	.recover_slot = recover_slot,
409 	.recover_done = recover_done,
410 };
411 
412 /*
413  * The BAST function for the ack lock resource
414  * This function wakes up the receive thread in
415  * order to receive and process the message.
416  */
417 static void ack_bast(void *arg, int mode)
418 {
419 	struct dlm_lock_resource *res = arg;
420 	struct md_cluster_info *cinfo = res->mddev->cluster_info;
421 
422 	if (mode == DLM_LOCK_EX) {
423 		if (test_bit(MD_CLUSTER_ALREADY_IN_CLUSTER, &cinfo->state))
424 			md_wakeup_thread(cinfo->recv_thread);
425 		else
426 			set_bit(MD_CLUSTER_PENDING_RECV_EVENT, &cinfo->state);
427 	}
428 }
429 
430 static void __remove_suspend_info(struct md_cluster_info *cinfo, int slot)
431 {
432 	struct suspend_info *s, *tmp;
433 
434 	list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list)
435 		if (slot == s->slot) {
436 			list_del(&s->list);
437 			kfree(s);
438 			break;
439 		}
440 }
441 
442 static void remove_suspend_info(struct mddev *mddev, int slot)
443 {
444 	struct md_cluster_info *cinfo = mddev->cluster_info;
445 	mddev->pers->quiesce(mddev, 1);
446 	spin_lock_irq(&cinfo->suspend_lock);
447 	__remove_suspend_info(cinfo, slot);
448 	spin_unlock_irq(&cinfo->suspend_lock);
449 	mddev->pers->quiesce(mddev, 0);
450 }
451 
452 
453 static void process_suspend_info(struct mddev *mddev,
454 		int slot, sector_t lo, sector_t hi)
455 {
456 	struct md_cluster_info *cinfo = mddev->cluster_info;
457 	struct suspend_info *s;
458 
459 	if (!hi) {
460 		remove_suspend_info(mddev, slot);
461 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
462 		md_wakeup_thread(mddev->thread);
463 		return;
464 	}
465 
466 	/*
467 	 * The bitmaps are not same for different nodes
468 	 * if RESYNCING is happening in one node, then
469 	 * the node which received the RESYNCING message
470 	 * probably will perform resync with the region
471 	 * [lo, hi] again, so we could reduce resync time
472 	 * a lot if we can ensure that the bitmaps among
473 	 * different nodes are match up well.
474 	 *
475 	 * sync_low/hi is used to record the region which
476 	 * arrived in the previous RESYNCING message,
477 	 *
478 	 * Call bitmap_sync_with_cluster to clear
479 	 * NEEDED_MASK and set RESYNC_MASK since
480 	 * resync thread is running in another node,
481 	 * so we don't need to do the resync again
482 	 * with the same section */
483 	bitmap_sync_with_cluster(mddev, cinfo->sync_low,
484 					cinfo->sync_hi,
485 					lo, hi);
486 	cinfo->sync_low = lo;
487 	cinfo->sync_hi = hi;
488 
489 	s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL);
490 	if (!s)
491 		return;
492 	s->slot = slot;
493 	s->lo = lo;
494 	s->hi = hi;
495 	mddev->pers->quiesce(mddev, 1);
496 	spin_lock_irq(&cinfo->suspend_lock);
497 	/* Remove existing entry (if exists) before adding */
498 	__remove_suspend_info(cinfo, slot);
499 	list_add(&s->list, &cinfo->suspend_list);
500 	spin_unlock_irq(&cinfo->suspend_lock);
501 	mddev->pers->quiesce(mddev, 0);
502 }
503 
504 static void process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg)
505 {
506 	char disk_uuid[64];
507 	struct md_cluster_info *cinfo = mddev->cluster_info;
508 	char event_name[] = "EVENT=ADD_DEVICE";
509 	char raid_slot[16];
510 	char *envp[] = {event_name, disk_uuid, raid_slot, NULL};
511 	int len;
512 
513 	len = snprintf(disk_uuid, 64, "DEVICE_UUID=");
514 	sprintf(disk_uuid + len, "%pU", cmsg->uuid);
515 	snprintf(raid_slot, 16, "RAID_DISK=%d", le32_to_cpu(cmsg->raid_slot));
516 	pr_info("%s:%d Sending kobject change with %s and %s\n", __func__, __LINE__, disk_uuid, raid_slot);
517 	init_completion(&cinfo->newdisk_completion);
518 	set_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state);
519 	kobject_uevent_env(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE, envp);
520 	wait_for_completion_timeout(&cinfo->newdisk_completion,
521 			NEW_DEV_TIMEOUT);
522 	clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state);
523 }
524 
525 
526 static void process_metadata_update(struct mddev *mddev, struct cluster_msg *msg)
527 {
528 	int got_lock = 0;
529 	struct md_cluster_info *cinfo = mddev->cluster_info;
530 	mddev->good_device_nr = le32_to_cpu(msg->raid_slot);
531 
532 	dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR);
533 	wait_event(mddev->thread->wqueue,
534 		   (got_lock = mddev_trylock(mddev)) ||
535 		    test_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state));
536 	md_reload_sb(mddev, mddev->good_device_nr);
537 	if (got_lock)
538 		mddev_unlock(mddev);
539 }
540 
541 static void process_remove_disk(struct mddev *mddev, struct cluster_msg *msg)
542 {
543 	struct md_rdev *rdev;
544 
545 	rcu_read_lock();
546 	rdev = md_find_rdev_nr_rcu(mddev, le32_to_cpu(msg->raid_slot));
547 	if (rdev) {
548 		set_bit(ClusterRemove, &rdev->flags);
549 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
550 		md_wakeup_thread(mddev->thread);
551 	}
552 	else
553 		pr_warn("%s: %d Could not find disk(%d) to REMOVE\n",
554 			__func__, __LINE__, le32_to_cpu(msg->raid_slot));
555 	rcu_read_unlock();
556 }
557 
558 static void process_readd_disk(struct mddev *mddev, struct cluster_msg *msg)
559 {
560 	struct md_rdev *rdev;
561 
562 	rcu_read_lock();
563 	rdev = md_find_rdev_nr_rcu(mddev, le32_to_cpu(msg->raid_slot));
564 	if (rdev && test_bit(Faulty, &rdev->flags))
565 		clear_bit(Faulty, &rdev->flags);
566 	else
567 		pr_warn("%s: %d Could not find disk(%d) which is faulty",
568 			__func__, __LINE__, le32_to_cpu(msg->raid_slot));
569 	rcu_read_unlock();
570 }
571 
572 static int process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg)
573 {
574 	int ret = 0;
575 
576 	if (WARN(mddev->cluster_info->slot_number - 1 == le32_to_cpu(msg->slot),
577 		"node %d received it's own msg\n", le32_to_cpu(msg->slot)))
578 		return -1;
579 	switch (le32_to_cpu(msg->type)) {
580 	case METADATA_UPDATED:
581 		process_metadata_update(mddev, msg);
582 		break;
583 	case CHANGE_CAPACITY:
584 		set_capacity(mddev->gendisk, mddev->array_sectors);
585 		revalidate_disk(mddev->gendisk);
586 		break;
587 	case RESYNCING:
588 		process_suspend_info(mddev, le32_to_cpu(msg->slot),
589 				     le64_to_cpu(msg->low),
590 				     le64_to_cpu(msg->high));
591 		break;
592 	case NEWDISK:
593 		process_add_new_disk(mddev, msg);
594 		break;
595 	case REMOVE:
596 		process_remove_disk(mddev, msg);
597 		break;
598 	case RE_ADD:
599 		process_readd_disk(mddev, msg);
600 		break;
601 	case BITMAP_NEEDS_SYNC:
602 		__recover_slot(mddev, le32_to_cpu(msg->slot));
603 		break;
604 	default:
605 		ret = -1;
606 		pr_warn("%s:%d Received unknown message from %d\n",
607 			__func__, __LINE__, msg->slot);
608 	}
609 	return ret;
610 }
611 
612 /*
613  * thread for receiving message
614  */
615 static void recv_daemon(struct md_thread *thread)
616 {
617 	struct md_cluster_info *cinfo = thread->mddev->cluster_info;
618 	struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres;
619 	struct dlm_lock_resource *message_lockres = cinfo->message_lockres;
620 	struct cluster_msg msg;
621 	int ret;
622 
623 	mutex_lock(&cinfo->recv_mutex);
624 	/*get CR on Message*/
625 	if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) {
626 		pr_err("md/raid1:failed to get CR on MESSAGE\n");
627 		mutex_unlock(&cinfo->recv_mutex);
628 		return;
629 	}
630 
631 	/* read lvb and wake up thread to process this message_lockres */
632 	memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg));
633 	ret = process_recvd_msg(thread->mddev, &msg);
634 	if (ret)
635 		goto out;
636 
637 	/*release CR on ack_lockres*/
638 	ret = dlm_unlock_sync(ack_lockres);
639 	if (unlikely(ret != 0))
640 		pr_info("unlock ack failed return %d\n", ret);
641 	/*up-convert to PR on message_lockres*/
642 	ret = dlm_lock_sync(message_lockres, DLM_LOCK_PR);
643 	if (unlikely(ret != 0))
644 		pr_info("lock PR on msg failed return %d\n", ret);
645 	/*get CR on ack_lockres again*/
646 	ret = dlm_lock_sync(ack_lockres, DLM_LOCK_CR);
647 	if (unlikely(ret != 0))
648 		pr_info("lock CR on ack failed return %d\n", ret);
649 out:
650 	/*release CR on message_lockres*/
651 	ret = dlm_unlock_sync(message_lockres);
652 	if (unlikely(ret != 0))
653 		pr_info("unlock msg failed return %d\n", ret);
654 	mutex_unlock(&cinfo->recv_mutex);
655 }
656 
657 /* lock_token()
658  * Takes the lock on the TOKEN lock resource so no other
659  * node can communicate while the operation is underway.
660  */
661 static int lock_token(struct md_cluster_info *cinfo, bool mddev_locked)
662 {
663 	int error, set_bit = 0;
664 	struct mddev *mddev = cinfo->mddev;
665 
666 	/*
667 	 * If resync thread run after raid1d thread, then process_metadata_update
668 	 * could not continue if raid1d held reconfig_mutex (and raid1d is blocked
669 	 * since another node already got EX on Token and waitting the EX of Ack),
670 	 * so let resync wake up thread in case flag is set.
671 	 */
672 	if (mddev_locked && !test_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD,
673 				      &cinfo->state)) {
674 		error = test_and_set_bit_lock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD,
675 					      &cinfo->state);
676 		WARN_ON_ONCE(error);
677 		md_wakeup_thread(mddev->thread);
678 		set_bit = 1;
679 	}
680 	error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX);
681 	if (set_bit)
682 		clear_bit_unlock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state);
683 
684 	if (error)
685 		pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n",
686 				__func__, __LINE__, error);
687 
688 	/* Lock the receive sequence */
689 	mutex_lock(&cinfo->recv_mutex);
690 	return error;
691 }
692 
693 /* lock_comm()
694  * Sets the MD_CLUSTER_SEND_LOCK bit to lock the send channel.
695  */
696 static int lock_comm(struct md_cluster_info *cinfo, bool mddev_locked)
697 {
698 	wait_event(cinfo->wait,
699 		   !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state));
700 
701 	return lock_token(cinfo, mddev_locked);
702 }
703 
704 static void unlock_comm(struct md_cluster_info *cinfo)
705 {
706 	WARN_ON(cinfo->token_lockres->mode != DLM_LOCK_EX);
707 	mutex_unlock(&cinfo->recv_mutex);
708 	dlm_unlock_sync(cinfo->token_lockres);
709 	clear_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state);
710 	wake_up(&cinfo->wait);
711 }
712 
713 /* __sendmsg()
714  * This function performs the actual sending of the message. This function is
715  * usually called after performing the encompassing operation
716  * The function:
717  * 1. Grabs the message lockresource in EX mode
718  * 2. Copies the message to the message LVB
719  * 3. Downconverts message lockresource to CW
720  * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes
721  *    and the other nodes read the message. The thread will wait here until all other
722  *    nodes have released ack lock resource.
723  * 5. Downconvert ack lockresource to CR
724  */
725 static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg)
726 {
727 	int error;
728 	int slot = cinfo->slot_number - 1;
729 
730 	cmsg->slot = cpu_to_le32(slot);
731 	/*get EX on Message*/
732 	error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX);
733 	if (error) {
734 		pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error);
735 		goto failed_message;
736 	}
737 
738 	memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg,
739 			sizeof(struct cluster_msg));
740 	/*down-convert EX to CW on Message*/
741 	error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CW);
742 	if (error) {
743 		pr_err("md-cluster: failed to convert EX to CW on MESSAGE(%d)\n",
744 				error);
745 		goto failed_ack;
746 	}
747 
748 	/*up-convert CR to EX on Ack*/
749 	error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX);
750 	if (error) {
751 		pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n",
752 				error);
753 		goto failed_ack;
754 	}
755 
756 	/*down-convert EX to CR on Ack*/
757 	error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR);
758 	if (error) {
759 		pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n",
760 				error);
761 		goto failed_ack;
762 	}
763 
764 failed_ack:
765 	error = dlm_unlock_sync(cinfo->message_lockres);
766 	if (unlikely(error != 0)) {
767 		pr_err("md-cluster: failed convert to NL on MESSAGE(%d)\n",
768 			error);
769 		/* in case the message can't be released due to some reason */
770 		goto failed_ack;
771 	}
772 failed_message:
773 	return error;
774 }
775 
776 static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg,
777 		   bool mddev_locked)
778 {
779 	int ret;
780 
781 	lock_comm(cinfo, mddev_locked);
782 	ret = __sendmsg(cinfo, cmsg);
783 	unlock_comm(cinfo);
784 	return ret;
785 }
786 
787 static int gather_all_resync_info(struct mddev *mddev, int total_slots)
788 {
789 	struct md_cluster_info *cinfo = mddev->cluster_info;
790 	int i, ret = 0;
791 	struct dlm_lock_resource *bm_lockres;
792 	struct suspend_info *s;
793 	char str[64];
794 	sector_t lo, hi;
795 
796 
797 	for (i = 0; i < total_slots; i++) {
798 		memset(str, '\0', 64);
799 		snprintf(str, 64, "bitmap%04d", i);
800 		bm_lockres = lockres_init(mddev, str, NULL, 1);
801 		if (!bm_lockres)
802 			return -ENOMEM;
803 		if (i == (cinfo->slot_number - 1)) {
804 			lockres_free(bm_lockres);
805 			continue;
806 		}
807 
808 		bm_lockres->flags |= DLM_LKF_NOQUEUE;
809 		ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
810 		if (ret == -EAGAIN) {
811 			s = read_resync_info(mddev, bm_lockres);
812 			if (s) {
813 				pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n",
814 						__func__, __LINE__,
815 						(unsigned long long) s->lo,
816 						(unsigned long long) s->hi, i);
817 				spin_lock_irq(&cinfo->suspend_lock);
818 				s->slot = i;
819 				list_add(&s->list, &cinfo->suspend_list);
820 				spin_unlock_irq(&cinfo->suspend_lock);
821 			}
822 			ret = 0;
823 			lockres_free(bm_lockres);
824 			continue;
825 		}
826 		if (ret) {
827 			lockres_free(bm_lockres);
828 			goto out;
829 		}
830 
831 		/* Read the disk bitmap sb and check if it needs recovery */
832 		ret = bitmap_copy_from_slot(mddev, i, &lo, &hi, false);
833 		if (ret) {
834 			pr_warn("md-cluster: Could not gather bitmaps from slot %d", i);
835 			lockres_free(bm_lockres);
836 			continue;
837 		}
838 		if ((hi > 0) && (lo < mddev->recovery_cp)) {
839 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
840 			mddev->recovery_cp = lo;
841 			md_check_recovery(mddev);
842 		}
843 
844 		lockres_free(bm_lockres);
845 	}
846 out:
847 	return ret;
848 }
849 
850 static int join(struct mddev *mddev, int nodes)
851 {
852 	struct md_cluster_info *cinfo;
853 	int ret, ops_rv;
854 	char str[64];
855 
856 	cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL);
857 	if (!cinfo)
858 		return -ENOMEM;
859 
860 	INIT_LIST_HEAD(&cinfo->suspend_list);
861 	spin_lock_init(&cinfo->suspend_lock);
862 	init_completion(&cinfo->completion);
863 	set_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state);
864 	init_waitqueue_head(&cinfo->wait);
865 	mutex_init(&cinfo->recv_mutex);
866 
867 	mddev->cluster_info = cinfo;
868 	cinfo->mddev = mddev;
869 
870 	memset(str, 0, 64);
871 	sprintf(str, "%pU", mddev->uuid);
872 	ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name,
873 				DLM_LSFL_FS, LVB_SIZE,
874 				&md_ls_ops, mddev, &ops_rv, &cinfo->lockspace);
875 	if (ret)
876 		goto err;
877 	wait_for_completion(&cinfo->completion);
878 	if (nodes < cinfo->slot_number) {
879 		pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).",
880 			cinfo->slot_number, nodes);
881 		ret = -ERANGE;
882 		goto err;
883 	}
884 	/* Initiate the communication resources */
885 	ret = -ENOMEM;
886 	cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv");
887 	if (!cinfo->recv_thread) {
888 		pr_err("md-cluster: cannot allocate memory for recv_thread!\n");
889 		goto err;
890 	}
891 	cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1);
892 	if (!cinfo->message_lockres)
893 		goto err;
894 	cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0);
895 	if (!cinfo->token_lockres)
896 		goto err;
897 	cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0);
898 	if (!cinfo->no_new_dev_lockres)
899 		goto err;
900 
901 	ret = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX);
902 	if (ret) {
903 		ret = -EAGAIN;
904 		pr_err("md-cluster: can't join cluster to avoid lock issue\n");
905 		goto err;
906 	}
907 	cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0);
908 	if (!cinfo->ack_lockres) {
909 		ret = -ENOMEM;
910 		goto err;
911 	}
912 	/* get sync CR lock on ACK. */
913 	if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR))
914 		pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n",
915 				ret);
916 	dlm_unlock_sync(cinfo->token_lockres);
917 	/* get sync CR lock on no-new-dev. */
918 	if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR))
919 		pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret);
920 
921 
922 	pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number);
923 	snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1);
924 	cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1);
925 	if (!cinfo->bitmap_lockres) {
926 		ret = -ENOMEM;
927 		goto err;
928 	}
929 	if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) {
930 		pr_err("Failed to get bitmap lock\n");
931 		ret = -EINVAL;
932 		goto err;
933 	}
934 
935 	cinfo->resync_lockres = lockres_init(mddev, "resync", NULL, 0);
936 	if (!cinfo->resync_lockres) {
937 		ret = -ENOMEM;
938 		goto err;
939 	}
940 
941 	return 0;
942 err:
943 	set_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state);
944 	md_unregister_thread(&cinfo->recovery_thread);
945 	md_unregister_thread(&cinfo->recv_thread);
946 	lockres_free(cinfo->message_lockres);
947 	lockres_free(cinfo->token_lockres);
948 	lockres_free(cinfo->ack_lockres);
949 	lockres_free(cinfo->no_new_dev_lockres);
950 	lockres_free(cinfo->resync_lockres);
951 	lockres_free(cinfo->bitmap_lockres);
952 	if (cinfo->lockspace)
953 		dlm_release_lockspace(cinfo->lockspace, 2);
954 	mddev->cluster_info = NULL;
955 	kfree(cinfo);
956 	return ret;
957 }
958 
959 static void load_bitmaps(struct mddev *mddev, int total_slots)
960 {
961 	struct md_cluster_info *cinfo = mddev->cluster_info;
962 
963 	/* load all the node's bitmap info for resync */
964 	if (gather_all_resync_info(mddev, total_slots))
965 		pr_err("md-cluster: failed to gather all resyn infos\n");
966 	set_bit(MD_CLUSTER_ALREADY_IN_CLUSTER, &cinfo->state);
967 	/* wake up recv thread in case something need to be handled */
968 	if (test_and_clear_bit(MD_CLUSTER_PENDING_RECV_EVENT, &cinfo->state))
969 		md_wakeup_thread(cinfo->recv_thread);
970 }
971 
972 static void resync_bitmap(struct mddev *mddev)
973 {
974 	struct md_cluster_info *cinfo = mddev->cluster_info;
975 	struct cluster_msg cmsg = {0};
976 	int err;
977 
978 	cmsg.type = cpu_to_le32(BITMAP_NEEDS_SYNC);
979 	err = sendmsg(cinfo, &cmsg, 1);
980 	if (err)
981 		pr_err("%s:%d: failed to send BITMAP_NEEDS_SYNC message (%d)\n",
982 			__func__, __LINE__, err);
983 }
984 
985 static void unlock_all_bitmaps(struct mddev *mddev);
986 static int leave(struct mddev *mddev)
987 {
988 	struct md_cluster_info *cinfo = mddev->cluster_info;
989 
990 	if (!cinfo)
991 		return 0;
992 
993 	/* BITMAP_NEEDS_SYNC message should be sent when node
994 	 * is leaving the cluster with dirty bitmap, also we
995 	 * can only deliver it when dlm connection is available */
996 	if (cinfo->slot_number > 0 && mddev->recovery_cp != MaxSector)
997 		resync_bitmap(mddev);
998 
999 	set_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state);
1000 	md_unregister_thread(&cinfo->recovery_thread);
1001 	md_unregister_thread(&cinfo->recv_thread);
1002 	lockres_free(cinfo->message_lockres);
1003 	lockres_free(cinfo->token_lockres);
1004 	lockres_free(cinfo->ack_lockres);
1005 	lockres_free(cinfo->no_new_dev_lockres);
1006 	lockres_free(cinfo->resync_lockres);
1007 	lockres_free(cinfo->bitmap_lockres);
1008 	unlock_all_bitmaps(mddev);
1009 	dlm_release_lockspace(cinfo->lockspace, 2);
1010 	kfree(cinfo);
1011 	return 0;
1012 }
1013 
1014 /* slot_number(): Returns the MD slot number to use
1015  * DLM starts the slot numbers from 1, wheras cluster-md
1016  * wants the number to be from zero, so we deduct one
1017  */
1018 static int slot_number(struct mddev *mddev)
1019 {
1020 	struct md_cluster_info *cinfo = mddev->cluster_info;
1021 
1022 	return cinfo->slot_number - 1;
1023 }
1024 
1025 /*
1026  * Check if the communication is already locked, else lock the communication
1027  * channel.
1028  * If it is already locked, token is in EX mode, and hence lock_token()
1029  * should not be called.
1030  */
1031 static int metadata_update_start(struct mddev *mddev)
1032 {
1033 	struct md_cluster_info *cinfo = mddev->cluster_info;
1034 	int ret;
1035 
1036 	/*
1037 	 * metadata_update_start is always called with the protection of
1038 	 * reconfig_mutex, so set WAITING_FOR_TOKEN here.
1039 	 */
1040 	ret = test_and_set_bit_lock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD,
1041 				    &cinfo->state);
1042 	WARN_ON_ONCE(ret);
1043 	md_wakeup_thread(mddev->thread);
1044 
1045 	wait_event(cinfo->wait,
1046 		   !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state) ||
1047 		   test_and_clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state));
1048 
1049 	/* If token is already locked, return 0 */
1050 	if (cinfo->token_lockres->mode == DLM_LOCK_EX) {
1051 		clear_bit_unlock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state);
1052 		return 0;
1053 	}
1054 
1055 	ret = lock_token(cinfo, 1);
1056 	clear_bit_unlock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state);
1057 	return ret;
1058 }
1059 
1060 static int metadata_update_finish(struct mddev *mddev)
1061 {
1062 	struct md_cluster_info *cinfo = mddev->cluster_info;
1063 	struct cluster_msg cmsg;
1064 	struct md_rdev *rdev;
1065 	int ret = 0;
1066 	int raid_slot = -1;
1067 
1068 	memset(&cmsg, 0, sizeof(cmsg));
1069 	cmsg.type = cpu_to_le32(METADATA_UPDATED);
1070 	/* Pick up a good active device number to send.
1071 	 */
1072 	rdev_for_each(rdev, mddev)
1073 		if (rdev->raid_disk > -1 && !test_bit(Faulty, &rdev->flags)) {
1074 			raid_slot = rdev->desc_nr;
1075 			break;
1076 		}
1077 	if (raid_slot >= 0) {
1078 		cmsg.raid_slot = cpu_to_le32(raid_slot);
1079 		ret = __sendmsg(cinfo, &cmsg);
1080 	} else
1081 		pr_warn("md-cluster: No good device id found to send\n");
1082 	clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state);
1083 	unlock_comm(cinfo);
1084 	return ret;
1085 }
1086 
1087 static void metadata_update_cancel(struct mddev *mddev)
1088 {
1089 	struct md_cluster_info *cinfo = mddev->cluster_info;
1090 	clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state);
1091 	unlock_comm(cinfo);
1092 }
1093 
1094 /*
1095  * return 0 if all the bitmaps have the same sync_size
1096  */
1097 static int cluster_check_sync_size(struct mddev *mddev)
1098 {
1099 	int i, rv;
1100 	bitmap_super_t *sb;
1101 	unsigned long my_sync_size, sync_size = 0;
1102 	int node_num = mddev->bitmap_info.nodes;
1103 	int current_slot = md_cluster_ops->slot_number(mddev);
1104 	struct bitmap *bitmap = mddev->bitmap;
1105 	char str[64];
1106 	struct dlm_lock_resource *bm_lockres;
1107 
1108 	sb = kmap_atomic(bitmap->storage.sb_page);
1109 	my_sync_size = sb->sync_size;
1110 	kunmap_atomic(sb);
1111 
1112 	for (i = 0; i < node_num; i++) {
1113 		if (i == current_slot)
1114 			continue;
1115 
1116 		bitmap = get_bitmap_from_slot(mddev, i);
1117 		if (IS_ERR(bitmap)) {
1118 			pr_err("can't get bitmap from slot %d\n", i);
1119 			return -1;
1120 		}
1121 
1122 		/*
1123 		 * If we can hold the bitmap lock of one node then
1124 		 * the slot is not occupied, update the sb.
1125 		 */
1126 		snprintf(str, 64, "bitmap%04d", i);
1127 		bm_lockres = lockres_init(mddev, str, NULL, 1);
1128 		if (!bm_lockres) {
1129 			pr_err("md-cluster: Cannot initialize %s\n", str);
1130 			bitmap_free(bitmap);
1131 			return -1;
1132 		}
1133 		bm_lockres->flags |= DLM_LKF_NOQUEUE;
1134 		rv = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
1135 		if (!rv)
1136 			bitmap_update_sb(bitmap);
1137 		lockres_free(bm_lockres);
1138 
1139 		sb = kmap_atomic(bitmap->storage.sb_page);
1140 		if (sync_size == 0)
1141 			sync_size = sb->sync_size;
1142 		else if (sync_size != sb->sync_size) {
1143 			kunmap_atomic(sb);
1144 			bitmap_free(bitmap);
1145 			return -1;
1146 		}
1147 		kunmap_atomic(sb);
1148 		bitmap_free(bitmap);
1149 	}
1150 
1151 	return (my_sync_size == sync_size) ? 0 : -1;
1152 }
1153 
1154 /*
1155  * Update the size for cluster raid is a little more complex, we perform it
1156  * by the steps:
1157  * 1. hold token lock and update superblock in initiator node.
1158  * 2. send METADATA_UPDATED msg to other nodes.
1159  * 3. The initiator node continues to check each bitmap's sync_size, if all
1160  *    bitmaps have the same value of sync_size, then we can set capacity and
1161  *    let other nodes to perform it. If one node can't update sync_size
1162  *    accordingly, we need to revert to previous value.
1163  */
1164 static void update_size(struct mddev *mddev, sector_t old_dev_sectors)
1165 {
1166 	struct md_cluster_info *cinfo = mddev->cluster_info;
1167 	struct cluster_msg cmsg;
1168 	struct md_rdev *rdev;
1169 	int ret = 0;
1170 	int raid_slot = -1;
1171 
1172 	md_update_sb(mddev, 1);
1173 	lock_comm(cinfo, 1);
1174 
1175 	memset(&cmsg, 0, sizeof(cmsg));
1176 	cmsg.type = cpu_to_le32(METADATA_UPDATED);
1177 	rdev_for_each(rdev, mddev)
1178 		if (rdev->raid_disk >= 0 && !test_bit(Faulty, &rdev->flags)) {
1179 			raid_slot = rdev->desc_nr;
1180 			break;
1181 		}
1182 	if (raid_slot >= 0) {
1183 		cmsg.raid_slot = cpu_to_le32(raid_slot);
1184 		/*
1185 		 * We can only change capiticy after all the nodes can do it,
1186 		 * so need to wait after other nodes already received the msg
1187 		 * and handled the change
1188 		 */
1189 		ret = __sendmsg(cinfo, &cmsg);
1190 		if (ret) {
1191 			pr_err("%s:%d: failed to send METADATA_UPDATED msg\n",
1192 			       __func__, __LINE__);
1193 			unlock_comm(cinfo);
1194 			return;
1195 		}
1196 	} else {
1197 		pr_err("md-cluster: No good device id found to send\n");
1198 		unlock_comm(cinfo);
1199 		return;
1200 	}
1201 
1202 	/*
1203 	 * check the sync_size from other node's bitmap, if sync_size
1204 	 * have already updated in other nodes as expected, send an
1205 	 * empty metadata msg to permit the change of capacity
1206 	 */
1207 	if (cluster_check_sync_size(mddev) == 0) {
1208 		memset(&cmsg, 0, sizeof(cmsg));
1209 		cmsg.type = cpu_to_le32(CHANGE_CAPACITY);
1210 		ret = __sendmsg(cinfo, &cmsg);
1211 		if (ret)
1212 			pr_err("%s:%d: failed to send CHANGE_CAPACITY msg\n",
1213 			       __func__, __LINE__);
1214 		set_capacity(mddev->gendisk, mddev->array_sectors);
1215 		revalidate_disk(mddev->gendisk);
1216 	} else {
1217 		/* revert to previous sectors */
1218 		ret = mddev->pers->resize(mddev, old_dev_sectors);
1219 		if (!ret)
1220 			revalidate_disk(mddev->gendisk);
1221 		ret = __sendmsg(cinfo, &cmsg);
1222 		if (ret)
1223 			pr_err("%s:%d: failed to send METADATA_UPDATED msg\n",
1224 			       __func__, __LINE__);
1225 	}
1226 	unlock_comm(cinfo);
1227 }
1228 
1229 static int resync_start(struct mddev *mddev)
1230 {
1231 	struct md_cluster_info *cinfo = mddev->cluster_info;
1232 	return dlm_lock_sync_interruptible(cinfo->resync_lockres, DLM_LOCK_EX, mddev);
1233 }
1234 
1235 static int resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi)
1236 {
1237 	struct md_cluster_info *cinfo = mddev->cluster_info;
1238 	struct resync_info ri;
1239 	struct cluster_msg cmsg = {0};
1240 
1241 	/* do not send zero again, if we have sent before */
1242 	if (hi == 0) {
1243 		memcpy(&ri, cinfo->bitmap_lockres->lksb.sb_lvbptr, sizeof(struct resync_info));
1244 		if (le64_to_cpu(ri.hi) == 0)
1245 			return 0;
1246 	}
1247 
1248 	add_resync_info(cinfo->bitmap_lockres, lo, hi);
1249 	/* Re-acquire the lock to refresh LVB */
1250 	dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW);
1251 	cmsg.type = cpu_to_le32(RESYNCING);
1252 	cmsg.low = cpu_to_le64(lo);
1253 	cmsg.high = cpu_to_le64(hi);
1254 
1255 	/*
1256 	 * mddev_lock is held if resync_info_update is called from
1257 	 * resync_finish (md_reap_sync_thread -> resync_finish)
1258 	 */
1259 	if (lo == 0 && hi == 0)
1260 		return sendmsg(cinfo, &cmsg, 1);
1261 	else
1262 		return sendmsg(cinfo, &cmsg, 0);
1263 }
1264 
1265 static int resync_finish(struct mddev *mddev)
1266 {
1267 	struct md_cluster_info *cinfo = mddev->cluster_info;
1268 	dlm_unlock_sync(cinfo->resync_lockres);
1269 	return resync_info_update(mddev, 0, 0);
1270 }
1271 
1272 static int area_resyncing(struct mddev *mddev, int direction,
1273 		sector_t lo, sector_t hi)
1274 {
1275 	struct md_cluster_info *cinfo = mddev->cluster_info;
1276 	int ret = 0;
1277 	struct suspend_info *s;
1278 
1279 	if ((direction == READ) &&
1280 		test_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state))
1281 		return 1;
1282 
1283 	spin_lock_irq(&cinfo->suspend_lock);
1284 	if (list_empty(&cinfo->suspend_list))
1285 		goto out;
1286 	list_for_each_entry(s, &cinfo->suspend_list, list)
1287 		if (hi > s->lo && lo < s->hi) {
1288 			ret = 1;
1289 			break;
1290 		}
1291 out:
1292 	spin_unlock_irq(&cinfo->suspend_lock);
1293 	return ret;
1294 }
1295 
1296 /* add_new_disk() - initiates a disk add
1297  * However, if this fails before writing md_update_sb(),
1298  * add_new_disk_cancel() must be called to release token lock
1299  */
1300 static int add_new_disk(struct mddev *mddev, struct md_rdev *rdev)
1301 {
1302 	struct md_cluster_info *cinfo = mddev->cluster_info;
1303 	struct cluster_msg cmsg;
1304 	int ret = 0;
1305 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1306 	char *uuid = sb->device_uuid;
1307 
1308 	memset(&cmsg, 0, sizeof(cmsg));
1309 	cmsg.type = cpu_to_le32(NEWDISK);
1310 	memcpy(cmsg.uuid, uuid, 16);
1311 	cmsg.raid_slot = cpu_to_le32(rdev->desc_nr);
1312 	lock_comm(cinfo, 1);
1313 	ret = __sendmsg(cinfo, &cmsg);
1314 	if (ret) {
1315 		unlock_comm(cinfo);
1316 		return ret;
1317 	}
1318 	cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE;
1319 	ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX);
1320 	cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE;
1321 	/* Some node does not "see" the device */
1322 	if (ret == -EAGAIN)
1323 		ret = -ENOENT;
1324 	if (ret)
1325 		unlock_comm(cinfo);
1326 	else {
1327 		dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR);
1328 		/* Since MD_CHANGE_DEVS will be set in add_bound_rdev which
1329 		 * will run soon after add_new_disk, the below path will be
1330 		 * invoked:
1331 		 *   md_wakeup_thread(mddev->thread)
1332 		 *	-> conf->thread (raid1d)
1333 		 *	-> md_check_recovery -> md_update_sb
1334 		 *	-> metadata_update_start/finish
1335 		 * MD_CLUSTER_SEND_LOCKED_ALREADY will be cleared eventually.
1336 		 *
1337 		 * For other failure cases, metadata_update_cancel and
1338 		 * add_new_disk_cancel also clear below bit as well.
1339 		 * */
1340 		set_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state);
1341 		wake_up(&cinfo->wait);
1342 	}
1343 	return ret;
1344 }
1345 
1346 static void add_new_disk_cancel(struct mddev *mddev)
1347 {
1348 	struct md_cluster_info *cinfo = mddev->cluster_info;
1349 	clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state);
1350 	unlock_comm(cinfo);
1351 }
1352 
1353 static int new_disk_ack(struct mddev *mddev, bool ack)
1354 {
1355 	struct md_cluster_info *cinfo = mddev->cluster_info;
1356 
1357 	if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) {
1358 		pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev));
1359 		return -EINVAL;
1360 	}
1361 
1362 	if (ack)
1363 		dlm_unlock_sync(cinfo->no_new_dev_lockres);
1364 	complete(&cinfo->newdisk_completion);
1365 	return 0;
1366 }
1367 
1368 static int remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1369 {
1370 	struct cluster_msg cmsg = {0};
1371 	struct md_cluster_info *cinfo = mddev->cluster_info;
1372 	cmsg.type = cpu_to_le32(REMOVE);
1373 	cmsg.raid_slot = cpu_to_le32(rdev->desc_nr);
1374 	return sendmsg(cinfo, &cmsg, 1);
1375 }
1376 
1377 static int lock_all_bitmaps(struct mddev *mddev)
1378 {
1379 	int slot, my_slot, ret, held = 1, i = 0;
1380 	char str[64];
1381 	struct md_cluster_info *cinfo = mddev->cluster_info;
1382 
1383 	cinfo->other_bitmap_lockres = kzalloc((mddev->bitmap_info.nodes - 1) *
1384 					     sizeof(struct dlm_lock_resource *),
1385 					     GFP_KERNEL);
1386 	if (!cinfo->other_bitmap_lockres) {
1387 		pr_err("md: can't alloc mem for other bitmap locks\n");
1388 		return 0;
1389 	}
1390 
1391 	my_slot = slot_number(mddev);
1392 	for (slot = 0; slot < mddev->bitmap_info.nodes; slot++) {
1393 		if (slot == my_slot)
1394 			continue;
1395 
1396 		memset(str, '\0', 64);
1397 		snprintf(str, 64, "bitmap%04d", slot);
1398 		cinfo->other_bitmap_lockres[i] = lockres_init(mddev, str, NULL, 1);
1399 		if (!cinfo->other_bitmap_lockres[i])
1400 			return -ENOMEM;
1401 
1402 		cinfo->other_bitmap_lockres[i]->flags |= DLM_LKF_NOQUEUE;
1403 		ret = dlm_lock_sync(cinfo->other_bitmap_lockres[i], DLM_LOCK_PW);
1404 		if (ret)
1405 			held = -1;
1406 		i++;
1407 	}
1408 
1409 	return held;
1410 }
1411 
1412 static void unlock_all_bitmaps(struct mddev *mddev)
1413 {
1414 	struct md_cluster_info *cinfo = mddev->cluster_info;
1415 	int i;
1416 
1417 	/* release other node's bitmap lock if they are existed */
1418 	if (cinfo->other_bitmap_lockres) {
1419 		for (i = 0; i < mddev->bitmap_info.nodes - 1; i++) {
1420 			if (cinfo->other_bitmap_lockres[i]) {
1421 				lockres_free(cinfo->other_bitmap_lockres[i]);
1422 			}
1423 		}
1424 		kfree(cinfo->other_bitmap_lockres);
1425 	}
1426 }
1427 
1428 static int gather_bitmaps(struct md_rdev *rdev)
1429 {
1430 	int sn, err;
1431 	sector_t lo, hi;
1432 	struct cluster_msg cmsg = {0};
1433 	struct mddev *mddev = rdev->mddev;
1434 	struct md_cluster_info *cinfo = mddev->cluster_info;
1435 
1436 	cmsg.type = cpu_to_le32(RE_ADD);
1437 	cmsg.raid_slot = cpu_to_le32(rdev->desc_nr);
1438 	err = sendmsg(cinfo, &cmsg, 1);
1439 	if (err)
1440 		goto out;
1441 
1442 	for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) {
1443 		if (sn == (cinfo->slot_number - 1))
1444 			continue;
1445 		err = bitmap_copy_from_slot(mddev, sn, &lo, &hi, false);
1446 		if (err) {
1447 			pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn);
1448 			goto out;
1449 		}
1450 		if ((hi > 0) && (lo < mddev->recovery_cp))
1451 			mddev->recovery_cp = lo;
1452 	}
1453 out:
1454 	return err;
1455 }
1456 
1457 static struct md_cluster_operations cluster_ops = {
1458 	.join   = join,
1459 	.leave  = leave,
1460 	.slot_number = slot_number,
1461 	.resync_start = resync_start,
1462 	.resync_finish = resync_finish,
1463 	.resync_info_update = resync_info_update,
1464 	.metadata_update_start = metadata_update_start,
1465 	.metadata_update_finish = metadata_update_finish,
1466 	.metadata_update_cancel = metadata_update_cancel,
1467 	.area_resyncing = area_resyncing,
1468 	.add_new_disk = add_new_disk,
1469 	.add_new_disk_cancel = add_new_disk_cancel,
1470 	.new_disk_ack = new_disk_ack,
1471 	.remove_disk = remove_disk,
1472 	.load_bitmaps = load_bitmaps,
1473 	.gather_bitmaps = gather_bitmaps,
1474 	.lock_all_bitmaps = lock_all_bitmaps,
1475 	.unlock_all_bitmaps = unlock_all_bitmaps,
1476 	.update_size = update_size,
1477 };
1478 
1479 static int __init cluster_init(void)
1480 {
1481 	pr_warn("md-cluster: support raid1 and raid10 (limited support)\n");
1482 	pr_info("Registering Cluster MD functions\n");
1483 	register_md_cluster_operations(&cluster_ops, THIS_MODULE);
1484 	return 0;
1485 }
1486 
1487 static void cluster_exit(void)
1488 {
1489 	unregister_md_cluster_operations();
1490 }
1491 
1492 module_init(cluster_init);
1493 module_exit(cluster_exit);
1494 MODULE_AUTHOR("SUSE");
1495 MODULE_LICENSE("GPL");
1496 MODULE_DESCRIPTION("Clustering support for MD");
1497