• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2015, SUSE
4  */
5 
6 
7 #include <linux/module.h>
8 #include <linux/kthread.h>
9 #include <linux/dlm.h>
10 #include <linux/sched.h>
11 #include <linux/raid/md_p.h>
12 #include "md.h"
13 #include "md-bitmap.h"
14 #include "md-cluster.h"
15 
16 #define LVB_SIZE	64
17 #define NEW_DEV_TIMEOUT 5000
18 
19 struct dlm_lock_resource {
20 	dlm_lockspace_t *ls;
21 	struct dlm_lksb lksb;
22 	char *name; /* lock name. */
23 	uint32_t flags; /* flags to pass to dlm_lock() */
24 	wait_queue_head_t sync_locking; /* wait queue for synchronized locking */
25 	bool sync_locking_done;
26 	void (*bast)(void *arg, int mode); /* blocking AST function pointer*/
27 	struct mddev *mddev; /* pointing back to mddev. */
28 	int mode;
29 };
30 
31 struct resync_info {
32 	__le64 lo;
33 	__le64 hi;
34 };
35 
36 /* md_cluster_info flags */
37 #define		MD_CLUSTER_WAITING_FOR_NEWDISK		1
38 #define		MD_CLUSTER_SUSPEND_READ_BALANCING	2
39 #define		MD_CLUSTER_BEGIN_JOIN_CLUSTER		3
40 
41 /* Lock the send communication. This is done through
42  * bit manipulation as opposed to a mutex in order to
43  * accomodate lock and hold. See next comment.
44  */
45 #define		MD_CLUSTER_SEND_LOCK			4
46 /* If cluster operations (such as adding a disk) must lock the
47  * communication channel, so as to perform extra operations
48  * (update metadata) and no other operation is allowed on the
49  * MD. Token needs to be locked and held until the operation
50  * completes witha md_update_sb(), which would eventually release
51  * the lock.
52  */
53 #define		MD_CLUSTER_SEND_LOCKED_ALREADY		5
54 /* We should receive message after node joined cluster and
55  * set up all the related infos such as bitmap and personality */
56 #define		MD_CLUSTER_ALREADY_IN_CLUSTER		6
57 #define		MD_CLUSTER_PENDING_RECV_EVENT		7
58 #define 	MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD		8
59 
60 struct md_cluster_info {
61 	struct mddev *mddev; /* the md device which md_cluster_info belongs to */
62 	/* dlm lock space and resources for clustered raid. */
63 	dlm_lockspace_t *lockspace;
64 	int slot_number;
65 	struct completion completion;
66 	struct mutex recv_mutex;
67 	struct dlm_lock_resource *bitmap_lockres;
68 	struct dlm_lock_resource **other_bitmap_lockres;
69 	struct dlm_lock_resource *resync_lockres;
70 	struct list_head suspend_list;
71 
72 	spinlock_t suspend_lock;
73 	/* record the region which write should be suspended */
74 	sector_t suspend_lo;
75 	sector_t suspend_hi;
76 	int suspend_from; /* the slot which broadcast suspend_lo/hi */
77 
78 	struct md_thread *recovery_thread;
79 	unsigned long recovery_map;
80 	/* communication loc resources */
81 	struct dlm_lock_resource *ack_lockres;
82 	struct dlm_lock_resource *message_lockres;
83 	struct dlm_lock_resource *token_lockres;
84 	struct dlm_lock_resource *no_new_dev_lockres;
85 	struct md_thread *recv_thread;
86 	struct completion newdisk_completion;
87 	wait_queue_head_t wait;
88 	unsigned long state;
89 	/* record the region in RESYNCING message */
90 	sector_t sync_low;
91 	sector_t sync_hi;
92 };
93 
94 enum msg_type {
95 	METADATA_UPDATED = 0,
96 	RESYNCING,
97 	NEWDISK,
98 	REMOVE,
99 	RE_ADD,
100 	BITMAP_NEEDS_SYNC,
101 	CHANGE_CAPACITY,
102 	BITMAP_RESIZE,
103 };
104 
105 struct cluster_msg {
106 	__le32 type;
107 	__le32 slot;
108 	/* TODO: Unionize this for smaller footprint */
109 	__le64 low;
110 	__le64 high;
111 	char uuid[16];
112 	__le32 raid_slot;
113 };
114 
sync_ast(void * arg)115 static void sync_ast(void *arg)
116 {
117 	struct dlm_lock_resource *res;
118 
119 	res = arg;
120 	res->sync_locking_done = true;
121 	wake_up(&res->sync_locking);
122 }
123 
dlm_lock_sync(struct dlm_lock_resource * res,int mode)124 static int dlm_lock_sync(struct dlm_lock_resource *res, int mode)
125 {
126 	int ret = 0;
127 
128 	ret = dlm_lock(res->ls, mode, &res->lksb,
129 			res->flags, res->name, strlen(res->name),
130 			0, sync_ast, res, res->bast);
131 	if (ret)
132 		return ret;
133 	wait_event(res->sync_locking, res->sync_locking_done);
134 	res->sync_locking_done = false;
135 	if (res->lksb.sb_status == 0)
136 		res->mode = mode;
137 	return res->lksb.sb_status;
138 }
139 
dlm_unlock_sync(struct dlm_lock_resource * res)140 static int dlm_unlock_sync(struct dlm_lock_resource *res)
141 {
142 	return dlm_lock_sync(res, DLM_LOCK_NL);
143 }
144 
145 /*
146  * An variation of dlm_lock_sync, which make lock request could
147  * be interrupted
148  */
dlm_lock_sync_interruptible(struct dlm_lock_resource * res,int mode,struct mddev * mddev)149 static int dlm_lock_sync_interruptible(struct dlm_lock_resource *res, int mode,
150 				       struct mddev *mddev)
151 {
152 	int ret = 0;
153 
154 	ret = dlm_lock(res->ls, mode, &res->lksb,
155 			res->flags, res->name, strlen(res->name),
156 			0, sync_ast, res, res->bast);
157 	if (ret)
158 		return ret;
159 
160 	wait_event(res->sync_locking, res->sync_locking_done
161 				      || kthread_should_stop()
162 				      || test_bit(MD_CLOSING, &mddev->flags));
163 	if (!res->sync_locking_done) {
164 		/*
165 		 * the convert queue contains the lock request when request is
166 		 * interrupted, and sync_ast could still be run, so need to
167 		 * cancel the request and reset completion
168 		 */
169 		ret = dlm_unlock(res->ls, res->lksb.sb_lkid, DLM_LKF_CANCEL,
170 			&res->lksb, res);
171 		res->sync_locking_done = false;
172 		if (unlikely(ret != 0))
173 			pr_info("failed to cancel previous lock request "
174 				 "%s return %d\n", res->name, ret);
175 		return -EPERM;
176 	} else
177 		res->sync_locking_done = false;
178 	if (res->lksb.sb_status == 0)
179 		res->mode = mode;
180 	return res->lksb.sb_status;
181 }
182 
lockres_init(struct mddev * mddev,char * name,void (* bastfn)(void * arg,int mode),int with_lvb)183 static struct dlm_lock_resource *lockres_init(struct mddev *mddev,
184 		char *name, void (*bastfn)(void *arg, int mode), int with_lvb)
185 {
186 	struct dlm_lock_resource *res = NULL;
187 	int ret, namelen;
188 	struct md_cluster_info *cinfo = mddev->cluster_info;
189 
190 	res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL);
191 	if (!res)
192 		return NULL;
193 	init_waitqueue_head(&res->sync_locking);
194 	res->sync_locking_done = false;
195 	res->ls = cinfo->lockspace;
196 	res->mddev = mddev;
197 	res->mode = DLM_LOCK_IV;
198 	namelen = strlen(name);
199 	res->name = kzalloc(namelen + 1, GFP_KERNEL);
200 	if (!res->name) {
201 		pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name);
202 		goto out_err;
203 	}
204 	strlcpy(res->name, name, namelen + 1);
205 	if (with_lvb) {
206 		res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL);
207 		if (!res->lksb.sb_lvbptr) {
208 			pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name);
209 			goto out_err;
210 		}
211 		res->flags = DLM_LKF_VALBLK;
212 	}
213 
214 	if (bastfn)
215 		res->bast = bastfn;
216 
217 	res->flags |= DLM_LKF_EXPEDITE;
218 
219 	ret = dlm_lock_sync(res, DLM_LOCK_NL);
220 	if (ret) {
221 		pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name);
222 		goto out_err;
223 	}
224 	res->flags &= ~DLM_LKF_EXPEDITE;
225 	res->flags |= DLM_LKF_CONVERT;
226 
227 	return res;
228 out_err:
229 	kfree(res->lksb.sb_lvbptr);
230 	kfree(res->name);
231 	kfree(res);
232 	return NULL;
233 }
234 
lockres_free(struct dlm_lock_resource * res)235 static void lockres_free(struct dlm_lock_resource *res)
236 {
237 	int ret = 0;
238 
239 	if (!res)
240 		return;
241 
242 	/*
243 	 * use FORCEUNLOCK flag, so we can unlock even the lock is on the
244 	 * waiting or convert queue
245 	 */
246 	ret = dlm_unlock(res->ls, res->lksb.sb_lkid, DLM_LKF_FORCEUNLOCK,
247 		&res->lksb, res);
248 	if (unlikely(ret != 0))
249 		pr_err("failed to unlock %s return %d\n", res->name, ret);
250 	else
251 		wait_event(res->sync_locking, res->sync_locking_done);
252 
253 	kfree(res->name);
254 	kfree(res->lksb.sb_lvbptr);
255 	kfree(res);
256 }
257 
add_resync_info(struct dlm_lock_resource * lockres,sector_t lo,sector_t hi)258 static void add_resync_info(struct dlm_lock_resource *lockres,
259 			    sector_t lo, sector_t hi)
260 {
261 	struct resync_info *ri;
262 
263 	ri = (struct resync_info *)lockres->lksb.sb_lvbptr;
264 	ri->lo = cpu_to_le64(lo);
265 	ri->hi = cpu_to_le64(hi);
266 }
267 
read_resync_info(struct mddev * mddev,struct dlm_lock_resource * lockres)268 static int read_resync_info(struct mddev *mddev,
269 			    struct dlm_lock_resource *lockres)
270 {
271 	struct resync_info ri;
272 	struct md_cluster_info *cinfo = mddev->cluster_info;
273 	int ret = 0;
274 
275 	dlm_lock_sync(lockres, DLM_LOCK_CR);
276 	memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info));
277 	if (le64_to_cpu(ri.hi) > 0) {
278 		cinfo->suspend_hi = le64_to_cpu(ri.hi);
279 		cinfo->suspend_lo = le64_to_cpu(ri.lo);
280 		ret = 1;
281 	}
282 	dlm_unlock_sync(lockres);
283 	return ret;
284 }
285 
recover_bitmaps(struct md_thread * thread)286 static void recover_bitmaps(struct md_thread *thread)
287 {
288 	struct mddev *mddev = thread->mddev;
289 	struct md_cluster_info *cinfo = mddev->cluster_info;
290 	struct dlm_lock_resource *bm_lockres;
291 	char str[64];
292 	int slot, ret;
293 	sector_t lo, hi;
294 
295 	while (cinfo->recovery_map) {
296 		slot = fls64((u64)cinfo->recovery_map) - 1;
297 
298 		snprintf(str, 64, "bitmap%04d", slot);
299 		bm_lockres = lockres_init(mddev, str, NULL, 1);
300 		if (!bm_lockres) {
301 			pr_err("md-cluster: Cannot initialize bitmaps\n");
302 			goto clear_bit;
303 		}
304 
305 		ret = dlm_lock_sync_interruptible(bm_lockres, DLM_LOCK_PW, mddev);
306 		if (ret) {
307 			pr_err("md-cluster: Could not DLM lock %s: %d\n",
308 					str, ret);
309 			goto clear_bit;
310 		}
311 		ret = md_bitmap_copy_from_slot(mddev, slot, &lo, &hi, true);
312 		if (ret) {
313 			pr_err("md-cluster: Could not copy data from bitmap %d\n", slot);
314 			goto clear_bit;
315 		}
316 
317 		/* Clear suspend_area associated with the bitmap */
318 		spin_lock_irq(&cinfo->suspend_lock);
319 		cinfo->suspend_hi = 0;
320 		cinfo->suspend_lo = 0;
321 		cinfo->suspend_from = -1;
322 		spin_unlock_irq(&cinfo->suspend_lock);
323 
324 		/* Kick off a reshape if needed */
325 		if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
326 		    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
327 		    mddev->reshape_position != MaxSector)
328 			md_wakeup_thread(mddev->sync_thread);
329 
330 		if (hi > 0) {
331 			if (lo < mddev->recovery_cp)
332 				mddev->recovery_cp = lo;
333 			/* wake up thread to continue resync in case resync
334 			 * is not finished */
335 			if (mddev->recovery_cp != MaxSector) {
336 				/*
337 				 * clear the REMOTE flag since we will launch
338 				 * resync thread in current node.
339 				 */
340 				clear_bit(MD_RESYNCING_REMOTE,
341 					  &mddev->recovery);
342 				set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
343 				md_wakeup_thread(mddev->thread);
344 			}
345 		}
346 clear_bit:
347 		lockres_free(bm_lockres);
348 		clear_bit(slot, &cinfo->recovery_map);
349 	}
350 }
351 
recover_prep(void * arg)352 static void recover_prep(void *arg)
353 {
354 	struct mddev *mddev = arg;
355 	struct md_cluster_info *cinfo = mddev->cluster_info;
356 	set_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state);
357 }
358 
__recover_slot(struct mddev * mddev,int slot)359 static void __recover_slot(struct mddev *mddev, int slot)
360 {
361 	struct md_cluster_info *cinfo = mddev->cluster_info;
362 
363 	set_bit(slot, &cinfo->recovery_map);
364 	if (!cinfo->recovery_thread) {
365 		cinfo->recovery_thread = md_register_thread(recover_bitmaps,
366 				mddev, "recover");
367 		if (!cinfo->recovery_thread) {
368 			pr_warn("md-cluster: Could not create recovery thread\n");
369 			return;
370 		}
371 	}
372 	md_wakeup_thread(cinfo->recovery_thread);
373 }
374 
recover_slot(void * arg,struct dlm_slot * slot)375 static void recover_slot(void *arg, struct dlm_slot *slot)
376 {
377 	struct mddev *mddev = arg;
378 	struct md_cluster_info *cinfo = mddev->cluster_info;
379 
380 	pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n",
381 			mddev->bitmap_info.cluster_name,
382 			slot->nodeid, slot->slot,
383 			cinfo->slot_number);
384 	/* deduct one since dlm slot starts from one while the num of
385 	 * cluster-md begins with 0 */
386 	__recover_slot(mddev, slot->slot - 1);
387 }
388 
recover_done(void * arg,struct dlm_slot * slots,int num_slots,int our_slot,uint32_t generation)389 static void recover_done(void *arg, struct dlm_slot *slots,
390 		int num_slots, int our_slot,
391 		uint32_t generation)
392 {
393 	struct mddev *mddev = arg;
394 	struct md_cluster_info *cinfo = mddev->cluster_info;
395 
396 	cinfo->slot_number = our_slot;
397 	/* completion is only need to be complete when node join cluster,
398 	 * it doesn't need to run during another node's failure */
399 	if (test_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state)) {
400 		complete(&cinfo->completion);
401 		clear_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state);
402 	}
403 	clear_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state);
404 }
405 
406 /* the ops is called when node join the cluster, and do lock recovery
407  * if node failure occurs */
408 static const struct dlm_lockspace_ops md_ls_ops = {
409 	.recover_prep = recover_prep,
410 	.recover_slot = recover_slot,
411 	.recover_done = recover_done,
412 };
413 
414 /*
415  * The BAST function for the ack lock resource
416  * This function wakes up the receive thread in
417  * order to receive and process the message.
418  */
ack_bast(void * arg,int mode)419 static void ack_bast(void *arg, int mode)
420 {
421 	struct dlm_lock_resource *res = arg;
422 	struct md_cluster_info *cinfo = res->mddev->cluster_info;
423 
424 	if (mode == DLM_LOCK_EX) {
425 		if (test_bit(MD_CLUSTER_ALREADY_IN_CLUSTER, &cinfo->state))
426 			md_wakeup_thread(cinfo->recv_thread);
427 		else
428 			set_bit(MD_CLUSTER_PENDING_RECV_EVENT, &cinfo->state);
429 	}
430 }
431 
remove_suspend_info(struct mddev * mddev,int slot)432 static void remove_suspend_info(struct mddev *mddev, int slot)
433 {
434 	struct md_cluster_info *cinfo = mddev->cluster_info;
435 	mddev->pers->quiesce(mddev, 1);
436 	spin_lock_irq(&cinfo->suspend_lock);
437 	cinfo->suspend_hi = 0;
438 	cinfo->suspend_lo = 0;
439 	spin_unlock_irq(&cinfo->suspend_lock);
440 	mddev->pers->quiesce(mddev, 0);
441 }
442 
process_suspend_info(struct mddev * mddev,int slot,sector_t lo,sector_t hi)443 static void process_suspend_info(struct mddev *mddev,
444 		int slot, sector_t lo, sector_t hi)
445 {
446 	struct md_cluster_info *cinfo = mddev->cluster_info;
447 	struct mdp_superblock_1 *sb = NULL;
448 	struct md_rdev *rdev;
449 
450 	if (!hi) {
451 		/*
452 		 * clear the REMOTE flag since resync or recovery is finished
453 		 * in remote node.
454 		 */
455 		clear_bit(MD_RESYNCING_REMOTE, &mddev->recovery);
456 		remove_suspend_info(mddev, slot);
457 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
458 		md_wakeup_thread(mddev->thread);
459 		return;
460 	}
461 
462 	rdev_for_each(rdev, mddev)
463 		if (rdev->raid_disk > -1 && !test_bit(Faulty, &rdev->flags)) {
464 			sb = page_address(rdev->sb_page);
465 			break;
466 		}
467 
468 	/*
469 	 * The bitmaps are not same for different nodes
470 	 * if RESYNCING is happening in one node, then
471 	 * the node which received the RESYNCING message
472 	 * probably will perform resync with the region
473 	 * [lo, hi] again, so we could reduce resync time
474 	 * a lot if we can ensure that the bitmaps among
475 	 * different nodes are match up well.
476 	 *
477 	 * sync_low/hi is used to record the region which
478 	 * arrived in the previous RESYNCING message,
479 	 *
480 	 * Call md_bitmap_sync_with_cluster to clear NEEDED_MASK
481 	 * and set RESYNC_MASK since  resync thread is running
482 	 * in another node, so we don't need to do the resync
483 	 * again with the same section.
484 	 *
485 	 * Skip md_bitmap_sync_with_cluster in case reshape
486 	 * happening, because reshaping region is small and
487 	 * we don't want to trigger lots of WARN.
488 	 */
489 	if (sb && !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE))
490 		md_bitmap_sync_with_cluster(mddev, cinfo->sync_low,
491 					    cinfo->sync_hi, lo, hi);
492 	cinfo->sync_low = lo;
493 	cinfo->sync_hi = hi;
494 
495 	mddev->pers->quiesce(mddev, 1);
496 	spin_lock_irq(&cinfo->suspend_lock);
497 	cinfo->suspend_from = slot;
498 	cinfo->suspend_lo = lo;
499 	cinfo->suspend_hi = hi;
500 	spin_unlock_irq(&cinfo->suspend_lock);
501 	mddev->pers->quiesce(mddev, 0);
502 }
503 
process_add_new_disk(struct mddev * mddev,struct cluster_msg * cmsg)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 
process_metadata_update(struct mddev * mddev,struct cluster_msg * msg)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 
process_remove_disk(struct mddev * mddev,struct cluster_msg * msg)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 
process_readd_disk(struct mddev * mddev,struct cluster_msg * msg)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 
process_recvd_msg(struct mddev * mddev,struct cluster_msg * msg)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_size(mddev->gendisk, true);
586 		break;
587 	case RESYNCING:
588 		set_bit(MD_RESYNCING_REMOTE, &mddev->recovery);
589 		process_suspend_info(mddev, le32_to_cpu(msg->slot),
590 				     le64_to_cpu(msg->low),
591 				     le64_to_cpu(msg->high));
592 		break;
593 	case NEWDISK:
594 		process_add_new_disk(mddev, msg);
595 		break;
596 	case REMOVE:
597 		process_remove_disk(mddev, msg);
598 		break;
599 	case RE_ADD:
600 		process_readd_disk(mddev, msg);
601 		break;
602 	case BITMAP_NEEDS_SYNC:
603 		__recover_slot(mddev, le32_to_cpu(msg->slot));
604 		break;
605 	case BITMAP_RESIZE:
606 		if (le64_to_cpu(msg->high) != mddev->pers->size(mddev, 0, 0))
607 			ret = md_bitmap_resize(mddev->bitmap,
608 					    le64_to_cpu(msg->high), 0, 0);
609 		break;
610 	default:
611 		ret = -1;
612 		pr_warn("%s:%d Received unknown message from %d\n",
613 			__func__, __LINE__, msg->slot);
614 	}
615 	return ret;
616 }
617 
618 /*
619  * thread for receiving message
620  */
recv_daemon(struct md_thread * thread)621 static void recv_daemon(struct md_thread *thread)
622 {
623 	struct md_cluster_info *cinfo = thread->mddev->cluster_info;
624 	struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres;
625 	struct dlm_lock_resource *message_lockres = cinfo->message_lockres;
626 	struct cluster_msg msg;
627 	int ret;
628 
629 	mutex_lock(&cinfo->recv_mutex);
630 	/*get CR on Message*/
631 	if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) {
632 		pr_err("md/raid1:failed to get CR on MESSAGE\n");
633 		mutex_unlock(&cinfo->recv_mutex);
634 		return;
635 	}
636 
637 	/* read lvb and wake up thread to process this message_lockres */
638 	memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg));
639 	ret = process_recvd_msg(thread->mddev, &msg);
640 	if (ret)
641 		goto out;
642 
643 	/*release CR on ack_lockres*/
644 	ret = dlm_unlock_sync(ack_lockres);
645 	if (unlikely(ret != 0))
646 		pr_info("unlock ack failed return %d\n", ret);
647 	/*up-convert to PR on message_lockres*/
648 	ret = dlm_lock_sync(message_lockres, DLM_LOCK_PR);
649 	if (unlikely(ret != 0))
650 		pr_info("lock PR on msg failed return %d\n", ret);
651 	/*get CR on ack_lockres again*/
652 	ret = dlm_lock_sync(ack_lockres, DLM_LOCK_CR);
653 	if (unlikely(ret != 0))
654 		pr_info("lock CR on ack failed return %d\n", ret);
655 out:
656 	/*release CR on message_lockres*/
657 	ret = dlm_unlock_sync(message_lockres);
658 	if (unlikely(ret != 0))
659 		pr_info("unlock msg failed return %d\n", ret);
660 	mutex_unlock(&cinfo->recv_mutex);
661 }
662 
663 /* lock_token()
664  * Takes the lock on the TOKEN lock resource so no other
665  * node can communicate while the operation is underway.
666  */
lock_token(struct md_cluster_info * cinfo)667 static int lock_token(struct md_cluster_info *cinfo)
668 {
669 	int error;
670 
671 	error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX);
672 	if (error) {
673 		pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n",
674 				__func__, __LINE__, error);
675 	} else {
676 		/* Lock the receive sequence */
677 		mutex_lock(&cinfo->recv_mutex);
678 	}
679 	return error;
680 }
681 
682 /* lock_comm()
683  * Sets the MD_CLUSTER_SEND_LOCK bit to lock the send channel.
684  */
lock_comm(struct md_cluster_info * cinfo,bool mddev_locked)685 static int lock_comm(struct md_cluster_info *cinfo, bool mddev_locked)
686 {
687 	int rv, set_bit = 0;
688 	struct mddev *mddev = cinfo->mddev;
689 
690 	/*
691 	 * If resync thread run after raid1d thread, then process_metadata_update
692 	 * could not continue if raid1d held reconfig_mutex (and raid1d is blocked
693 	 * since another node already got EX on Token and waitting the EX of Ack),
694 	 * so let resync wake up thread in case flag is set.
695 	 */
696 	if (mddev_locked && !test_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD,
697 				      &cinfo->state)) {
698 		rv = test_and_set_bit_lock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD,
699 					      &cinfo->state);
700 		WARN_ON_ONCE(rv);
701 		md_wakeup_thread(mddev->thread);
702 		set_bit = 1;
703 	}
704 
705 	wait_event(cinfo->wait,
706 		   !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state));
707 	rv = lock_token(cinfo);
708 	if (set_bit)
709 		clear_bit_unlock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state);
710 	return rv;
711 }
712 
unlock_comm(struct md_cluster_info * cinfo)713 static void unlock_comm(struct md_cluster_info *cinfo)
714 {
715 	WARN_ON(cinfo->token_lockres->mode != DLM_LOCK_EX);
716 	mutex_unlock(&cinfo->recv_mutex);
717 	dlm_unlock_sync(cinfo->token_lockres);
718 	clear_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state);
719 	wake_up(&cinfo->wait);
720 }
721 
722 /* __sendmsg()
723  * This function performs the actual sending of the message. This function is
724  * usually called after performing the encompassing operation
725  * The function:
726  * 1. Grabs the message lockresource in EX mode
727  * 2. Copies the message to the message LVB
728  * 3. Downconverts message lockresource to CW
729  * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes
730  *    and the other nodes read the message. The thread will wait here until all other
731  *    nodes have released ack lock resource.
732  * 5. Downconvert ack lockresource to CR
733  */
__sendmsg(struct md_cluster_info * cinfo,struct cluster_msg * cmsg)734 static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg)
735 {
736 	int error;
737 	int slot = cinfo->slot_number - 1;
738 
739 	cmsg->slot = cpu_to_le32(slot);
740 	/*get EX on Message*/
741 	error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX);
742 	if (error) {
743 		pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error);
744 		goto failed_message;
745 	}
746 
747 	memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg,
748 			sizeof(struct cluster_msg));
749 	/*down-convert EX to CW on Message*/
750 	error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CW);
751 	if (error) {
752 		pr_err("md-cluster: failed to convert EX to CW on MESSAGE(%d)\n",
753 				error);
754 		goto failed_ack;
755 	}
756 
757 	/*up-convert CR to EX on Ack*/
758 	error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX);
759 	if (error) {
760 		pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n",
761 				error);
762 		goto failed_ack;
763 	}
764 
765 	/*down-convert EX to CR on Ack*/
766 	error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR);
767 	if (error) {
768 		pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n",
769 				error);
770 		goto failed_ack;
771 	}
772 
773 failed_ack:
774 	error = dlm_unlock_sync(cinfo->message_lockres);
775 	if (unlikely(error != 0)) {
776 		pr_err("md-cluster: failed convert to NL on MESSAGE(%d)\n",
777 			error);
778 		/* in case the message can't be released due to some reason */
779 		goto failed_ack;
780 	}
781 failed_message:
782 	return error;
783 }
784 
sendmsg(struct md_cluster_info * cinfo,struct cluster_msg * cmsg,bool mddev_locked)785 static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg,
786 		   bool mddev_locked)
787 {
788 	int ret;
789 
790 	ret = lock_comm(cinfo, mddev_locked);
791 	if (!ret) {
792 		ret = __sendmsg(cinfo, cmsg);
793 		unlock_comm(cinfo);
794 	}
795 	return ret;
796 }
797 
gather_all_resync_info(struct mddev * mddev,int total_slots)798 static int gather_all_resync_info(struct mddev *mddev, int total_slots)
799 {
800 	struct md_cluster_info *cinfo = mddev->cluster_info;
801 	int i, ret = 0;
802 	struct dlm_lock_resource *bm_lockres;
803 	char str[64];
804 	sector_t lo, hi;
805 
806 
807 	for (i = 0; i < total_slots; i++) {
808 		memset(str, '\0', 64);
809 		snprintf(str, 64, "bitmap%04d", i);
810 		bm_lockres = lockres_init(mddev, str, NULL, 1);
811 		if (!bm_lockres)
812 			return -ENOMEM;
813 		if (i == (cinfo->slot_number - 1)) {
814 			lockres_free(bm_lockres);
815 			continue;
816 		}
817 
818 		bm_lockres->flags |= DLM_LKF_NOQUEUE;
819 		ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
820 		if (ret == -EAGAIN) {
821 			if (read_resync_info(mddev, bm_lockres)) {
822 				pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n",
823 						__func__, __LINE__,
824 					(unsigned long long) cinfo->suspend_lo,
825 					(unsigned long long) cinfo->suspend_hi,
826 					i);
827 				cinfo->suspend_from = i;
828 			}
829 			ret = 0;
830 			lockres_free(bm_lockres);
831 			continue;
832 		}
833 		if (ret) {
834 			lockres_free(bm_lockres);
835 			goto out;
836 		}
837 
838 		/* Read the disk bitmap sb and check if it needs recovery */
839 		ret = md_bitmap_copy_from_slot(mddev, i, &lo, &hi, false);
840 		if (ret) {
841 			pr_warn("md-cluster: Could not gather bitmaps from slot %d", i);
842 			lockres_free(bm_lockres);
843 			continue;
844 		}
845 		if ((hi > 0) && (lo < mddev->recovery_cp)) {
846 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
847 			mddev->recovery_cp = lo;
848 			md_check_recovery(mddev);
849 		}
850 
851 		lockres_free(bm_lockres);
852 	}
853 out:
854 	return ret;
855 }
856 
join(struct mddev * mddev,int nodes)857 static int join(struct mddev *mddev, int nodes)
858 {
859 	struct md_cluster_info *cinfo;
860 	int ret, ops_rv;
861 	char str[64];
862 
863 	cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL);
864 	if (!cinfo)
865 		return -ENOMEM;
866 
867 	INIT_LIST_HEAD(&cinfo->suspend_list);
868 	spin_lock_init(&cinfo->suspend_lock);
869 	init_completion(&cinfo->completion);
870 	set_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state);
871 	init_waitqueue_head(&cinfo->wait);
872 	mutex_init(&cinfo->recv_mutex);
873 
874 	mddev->cluster_info = cinfo;
875 	cinfo->mddev = mddev;
876 
877 	memset(str, 0, 64);
878 	sprintf(str, "%pU", mddev->uuid);
879 	ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name,
880 				DLM_LSFL_FS, LVB_SIZE,
881 				&md_ls_ops, mddev, &ops_rv, &cinfo->lockspace);
882 	if (ret)
883 		goto err;
884 	wait_for_completion(&cinfo->completion);
885 	if (nodes < cinfo->slot_number) {
886 		pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).",
887 			cinfo->slot_number, nodes);
888 		ret = -ERANGE;
889 		goto err;
890 	}
891 	/* Initiate the communication resources */
892 	ret = -ENOMEM;
893 	cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv");
894 	if (!cinfo->recv_thread) {
895 		pr_err("md-cluster: cannot allocate memory for recv_thread!\n");
896 		goto err;
897 	}
898 	cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1);
899 	if (!cinfo->message_lockres)
900 		goto err;
901 	cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0);
902 	if (!cinfo->token_lockres)
903 		goto err;
904 	cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0);
905 	if (!cinfo->no_new_dev_lockres)
906 		goto err;
907 
908 	ret = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX);
909 	if (ret) {
910 		ret = -EAGAIN;
911 		pr_err("md-cluster: can't join cluster to avoid lock issue\n");
912 		goto err;
913 	}
914 	cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0);
915 	if (!cinfo->ack_lockres) {
916 		ret = -ENOMEM;
917 		goto err;
918 	}
919 	/* get sync CR lock on ACK. */
920 	if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR))
921 		pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n",
922 				ret);
923 	dlm_unlock_sync(cinfo->token_lockres);
924 	/* get sync CR lock on no-new-dev. */
925 	if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR))
926 		pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret);
927 
928 
929 	pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number);
930 	snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1);
931 	cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1);
932 	if (!cinfo->bitmap_lockres) {
933 		ret = -ENOMEM;
934 		goto err;
935 	}
936 	if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) {
937 		pr_err("Failed to get bitmap lock\n");
938 		ret = -EINVAL;
939 		goto err;
940 	}
941 
942 	cinfo->resync_lockres = lockres_init(mddev, "resync", NULL, 0);
943 	if (!cinfo->resync_lockres) {
944 		ret = -ENOMEM;
945 		goto err;
946 	}
947 
948 	return 0;
949 err:
950 	set_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state);
951 	md_unregister_thread(&cinfo->recovery_thread);
952 	md_unregister_thread(&cinfo->recv_thread);
953 	lockres_free(cinfo->message_lockres);
954 	lockres_free(cinfo->token_lockres);
955 	lockres_free(cinfo->ack_lockres);
956 	lockres_free(cinfo->no_new_dev_lockres);
957 	lockres_free(cinfo->resync_lockres);
958 	lockres_free(cinfo->bitmap_lockres);
959 	if (cinfo->lockspace)
960 		dlm_release_lockspace(cinfo->lockspace, 2);
961 	mddev->cluster_info = NULL;
962 	kfree(cinfo);
963 	return ret;
964 }
965 
load_bitmaps(struct mddev * mddev,int total_slots)966 static void load_bitmaps(struct mddev *mddev, int total_slots)
967 {
968 	struct md_cluster_info *cinfo = mddev->cluster_info;
969 
970 	/* load all the node's bitmap info for resync */
971 	if (gather_all_resync_info(mddev, total_slots))
972 		pr_err("md-cluster: failed to gather all resyn infos\n");
973 	set_bit(MD_CLUSTER_ALREADY_IN_CLUSTER, &cinfo->state);
974 	/* wake up recv thread in case something need to be handled */
975 	if (test_and_clear_bit(MD_CLUSTER_PENDING_RECV_EVENT, &cinfo->state))
976 		md_wakeup_thread(cinfo->recv_thread);
977 }
978 
resync_bitmap(struct mddev * mddev)979 static void resync_bitmap(struct mddev *mddev)
980 {
981 	struct md_cluster_info *cinfo = mddev->cluster_info;
982 	struct cluster_msg cmsg = {0};
983 	int err;
984 
985 	cmsg.type = cpu_to_le32(BITMAP_NEEDS_SYNC);
986 	err = sendmsg(cinfo, &cmsg, 1);
987 	if (err)
988 		pr_err("%s:%d: failed to send BITMAP_NEEDS_SYNC message (%d)\n",
989 			__func__, __LINE__, err);
990 }
991 
992 static void unlock_all_bitmaps(struct mddev *mddev);
leave(struct mddev * mddev)993 static int leave(struct mddev *mddev)
994 {
995 	struct md_cluster_info *cinfo = mddev->cluster_info;
996 
997 	if (!cinfo)
998 		return 0;
999 
1000 	/*
1001 	 * BITMAP_NEEDS_SYNC message should be sent when node
1002 	 * is leaving the cluster with dirty bitmap, also we
1003 	 * can only deliver it when dlm connection is available.
1004 	 *
1005 	 * Also, we should send BITMAP_NEEDS_SYNC message in
1006 	 * case reshaping is interrupted.
1007 	 */
1008 	if ((cinfo->slot_number > 0 && mddev->recovery_cp != MaxSector) ||
1009 	    (mddev->reshape_position != MaxSector &&
1010 	     test_bit(MD_CLOSING, &mddev->flags)))
1011 		resync_bitmap(mddev);
1012 
1013 	set_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state);
1014 	md_unregister_thread(&cinfo->recovery_thread);
1015 	md_unregister_thread(&cinfo->recv_thread);
1016 	lockres_free(cinfo->message_lockres);
1017 	lockres_free(cinfo->token_lockres);
1018 	lockres_free(cinfo->ack_lockres);
1019 	lockres_free(cinfo->no_new_dev_lockres);
1020 	lockres_free(cinfo->resync_lockres);
1021 	lockres_free(cinfo->bitmap_lockres);
1022 	unlock_all_bitmaps(mddev);
1023 	dlm_release_lockspace(cinfo->lockspace, 2);
1024 	kfree(cinfo);
1025 	return 0;
1026 }
1027 
1028 /* slot_number(): Returns the MD slot number to use
1029  * DLM starts the slot numbers from 1, wheras cluster-md
1030  * wants the number to be from zero, so we deduct one
1031  */
slot_number(struct mddev * mddev)1032 static int slot_number(struct mddev *mddev)
1033 {
1034 	struct md_cluster_info *cinfo = mddev->cluster_info;
1035 
1036 	return cinfo->slot_number - 1;
1037 }
1038 
1039 /*
1040  * Check if the communication is already locked, else lock the communication
1041  * channel.
1042  * If it is already locked, token is in EX mode, and hence lock_token()
1043  * should not be called.
1044  */
metadata_update_start(struct mddev * mddev)1045 static int metadata_update_start(struct mddev *mddev)
1046 {
1047 	struct md_cluster_info *cinfo = mddev->cluster_info;
1048 	int ret;
1049 
1050 	/*
1051 	 * metadata_update_start is always called with the protection of
1052 	 * reconfig_mutex, so set WAITING_FOR_TOKEN here.
1053 	 */
1054 	ret = test_and_set_bit_lock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD,
1055 				    &cinfo->state);
1056 	WARN_ON_ONCE(ret);
1057 	md_wakeup_thread(mddev->thread);
1058 
1059 	wait_event(cinfo->wait,
1060 		   !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state) ||
1061 		   test_and_clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state));
1062 
1063 	/* If token is already locked, return 0 */
1064 	if (cinfo->token_lockres->mode == DLM_LOCK_EX) {
1065 		clear_bit_unlock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state);
1066 		return 0;
1067 	}
1068 
1069 	ret = lock_token(cinfo);
1070 	clear_bit_unlock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state);
1071 	return ret;
1072 }
1073 
metadata_update_finish(struct mddev * mddev)1074 static int metadata_update_finish(struct mddev *mddev)
1075 {
1076 	struct md_cluster_info *cinfo = mddev->cluster_info;
1077 	struct cluster_msg cmsg;
1078 	struct md_rdev *rdev;
1079 	int ret = 0;
1080 	int raid_slot = -1;
1081 
1082 	memset(&cmsg, 0, sizeof(cmsg));
1083 	cmsg.type = cpu_to_le32(METADATA_UPDATED);
1084 	/* Pick up a good active device number to send.
1085 	 */
1086 	rdev_for_each(rdev, mddev)
1087 		if (rdev->raid_disk > -1 && !test_bit(Faulty, &rdev->flags)) {
1088 			raid_slot = rdev->desc_nr;
1089 			break;
1090 		}
1091 	if (raid_slot >= 0) {
1092 		cmsg.raid_slot = cpu_to_le32(raid_slot);
1093 		ret = __sendmsg(cinfo, &cmsg);
1094 	} else
1095 		pr_warn("md-cluster: No good device id found to send\n");
1096 	clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state);
1097 	unlock_comm(cinfo);
1098 	return ret;
1099 }
1100 
metadata_update_cancel(struct mddev * mddev)1101 static void metadata_update_cancel(struct mddev *mddev)
1102 {
1103 	struct md_cluster_info *cinfo = mddev->cluster_info;
1104 	clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state);
1105 	unlock_comm(cinfo);
1106 }
1107 
update_bitmap_size(struct mddev * mddev,sector_t size)1108 static int update_bitmap_size(struct mddev *mddev, sector_t size)
1109 {
1110 	struct md_cluster_info *cinfo = mddev->cluster_info;
1111 	struct cluster_msg cmsg = {0};
1112 	int ret;
1113 
1114 	cmsg.type = cpu_to_le32(BITMAP_RESIZE);
1115 	cmsg.high = cpu_to_le64(size);
1116 	ret = sendmsg(cinfo, &cmsg, 0);
1117 	if (ret)
1118 		pr_err("%s:%d: failed to send BITMAP_RESIZE message (%d)\n",
1119 			__func__, __LINE__, ret);
1120 	return ret;
1121 }
1122 
resize_bitmaps(struct mddev * mddev,sector_t newsize,sector_t oldsize)1123 static int resize_bitmaps(struct mddev *mddev, sector_t newsize, sector_t oldsize)
1124 {
1125 	struct bitmap_counts *counts;
1126 	char str[64];
1127 	struct dlm_lock_resource *bm_lockres;
1128 	struct bitmap *bitmap = mddev->bitmap;
1129 	unsigned long my_pages = bitmap->counts.pages;
1130 	int i, rv;
1131 
1132 	/*
1133 	 * We need to ensure all the nodes can grow to a larger
1134 	 * bitmap size before make the reshaping.
1135 	 */
1136 	rv = update_bitmap_size(mddev, newsize);
1137 	if (rv)
1138 		return rv;
1139 
1140 	for (i = 0; i < mddev->bitmap_info.nodes; i++) {
1141 		if (i == md_cluster_ops->slot_number(mddev))
1142 			continue;
1143 
1144 		bitmap = get_bitmap_from_slot(mddev, i);
1145 		if (IS_ERR(bitmap)) {
1146 			pr_err("can't get bitmap from slot %d\n", i);
1147 			bitmap = NULL;
1148 			goto out;
1149 		}
1150 		counts = &bitmap->counts;
1151 
1152 		/*
1153 		 * If we can hold the bitmap lock of one node then
1154 		 * the slot is not occupied, update the pages.
1155 		 */
1156 		snprintf(str, 64, "bitmap%04d", i);
1157 		bm_lockres = lockres_init(mddev, str, NULL, 1);
1158 		if (!bm_lockres) {
1159 			pr_err("Cannot initialize %s lock\n", str);
1160 			goto out;
1161 		}
1162 		bm_lockres->flags |= DLM_LKF_NOQUEUE;
1163 		rv = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
1164 		if (!rv)
1165 			counts->pages = my_pages;
1166 		lockres_free(bm_lockres);
1167 
1168 		if (my_pages != counts->pages)
1169 			/*
1170 			 * Let's revert the bitmap size if one node
1171 			 * can't resize bitmap
1172 			 */
1173 			goto out;
1174 		md_bitmap_free(bitmap);
1175 	}
1176 
1177 	return 0;
1178 out:
1179 	md_bitmap_free(bitmap);
1180 	update_bitmap_size(mddev, oldsize);
1181 	return -1;
1182 }
1183 
1184 /*
1185  * return 0 if all the bitmaps have the same sync_size
1186  */
cluster_check_sync_size(struct mddev * mddev)1187 static int cluster_check_sync_size(struct mddev *mddev)
1188 {
1189 	int i, rv;
1190 	bitmap_super_t *sb;
1191 	unsigned long my_sync_size, sync_size = 0;
1192 	int node_num = mddev->bitmap_info.nodes;
1193 	int current_slot = md_cluster_ops->slot_number(mddev);
1194 	struct bitmap *bitmap = mddev->bitmap;
1195 	char str[64];
1196 	struct dlm_lock_resource *bm_lockres;
1197 
1198 	sb = kmap_atomic(bitmap->storage.sb_page);
1199 	my_sync_size = sb->sync_size;
1200 	kunmap_atomic(sb);
1201 
1202 	for (i = 0; i < node_num; i++) {
1203 		if (i == current_slot)
1204 			continue;
1205 
1206 		bitmap = get_bitmap_from_slot(mddev, i);
1207 		if (IS_ERR(bitmap)) {
1208 			pr_err("can't get bitmap from slot %d\n", i);
1209 			return -1;
1210 		}
1211 
1212 		/*
1213 		 * If we can hold the bitmap lock of one node then
1214 		 * the slot is not occupied, update the sb.
1215 		 */
1216 		snprintf(str, 64, "bitmap%04d", i);
1217 		bm_lockres = lockres_init(mddev, str, NULL, 1);
1218 		if (!bm_lockres) {
1219 			pr_err("md-cluster: Cannot initialize %s\n", str);
1220 			md_bitmap_free(bitmap);
1221 			return -1;
1222 		}
1223 		bm_lockres->flags |= DLM_LKF_NOQUEUE;
1224 		rv = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
1225 		if (!rv)
1226 			md_bitmap_update_sb(bitmap);
1227 		lockres_free(bm_lockres);
1228 
1229 		sb = kmap_atomic(bitmap->storage.sb_page);
1230 		if (sync_size == 0)
1231 			sync_size = sb->sync_size;
1232 		else if (sync_size != sb->sync_size) {
1233 			kunmap_atomic(sb);
1234 			md_bitmap_free(bitmap);
1235 			return -1;
1236 		}
1237 		kunmap_atomic(sb);
1238 		md_bitmap_free(bitmap);
1239 	}
1240 
1241 	return (my_sync_size == sync_size) ? 0 : -1;
1242 }
1243 
1244 /*
1245  * Update the size for cluster raid is a little more complex, we perform it
1246  * by the steps:
1247  * 1. hold token lock and update superblock in initiator node.
1248  * 2. send METADATA_UPDATED msg to other nodes.
1249  * 3. The initiator node continues to check each bitmap's sync_size, if all
1250  *    bitmaps have the same value of sync_size, then we can set capacity and
1251  *    let other nodes to perform it. If one node can't update sync_size
1252  *    accordingly, we need to revert to previous value.
1253  */
update_size(struct mddev * mddev,sector_t old_dev_sectors)1254 static void update_size(struct mddev *mddev, sector_t old_dev_sectors)
1255 {
1256 	struct md_cluster_info *cinfo = mddev->cluster_info;
1257 	struct cluster_msg cmsg;
1258 	struct md_rdev *rdev;
1259 	int ret = 0;
1260 	int raid_slot = -1;
1261 
1262 	md_update_sb(mddev, 1);
1263 	if (lock_comm(cinfo, 1)) {
1264 		pr_err("%s: lock_comm failed\n", __func__);
1265 		return;
1266 	}
1267 
1268 	memset(&cmsg, 0, sizeof(cmsg));
1269 	cmsg.type = cpu_to_le32(METADATA_UPDATED);
1270 	rdev_for_each(rdev, mddev)
1271 		if (rdev->raid_disk >= 0 && !test_bit(Faulty, &rdev->flags)) {
1272 			raid_slot = rdev->desc_nr;
1273 			break;
1274 		}
1275 	if (raid_slot >= 0) {
1276 		cmsg.raid_slot = cpu_to_le32(raid_slot);
1277 		/*
1278 		 * We can only change capiticy after all the nodes can do it,
1279 		 * so need to wait after other nodes already received the msg
1280 		 * and handled the change
1281 		 */
1282 		ret = __sendmsg(cinfo, &cmsg);
1283 		if (ret) {
1284 			pr_err("%s:%d: failed to send METADATA_UPDATED msg\n",
1285 			       __func__, __LINE__);
1286 			unlock_comm(cinfo);
1287 			return;
1288 		}
1289 	} else {
1290 		pr_err("md-cluster: No good device id found to send\n");
1291 		unlock_comm(cinfo);
1292 		return;
1293 	}
1294 
1295 	/*
1296 	 * check the sync_size from other node's bitmap, if sync_size
1297 	 * have already updated in other nodes as expected, send an
1298 	 * empty metadata msg to permit the change of capacity
1299 	 */
1300 	if (cluster_check_sync_size(mddev) == 0) {
1301 		memset(&cmsg, 0, sizeof(cmsg));
1302 		cmsg.type = cpu_to_le32(CHANGE_CAPACITY);
1303 		ret = __sendmsg(cinfo, &cmsg);
1304 		if (ret)
1305 			pr_err("%s:%d: failed to send CHANGE_CAPACITY msg\n",
1306 			       __func__, __LINE__);
1307 		set_capacity(mddev->gendisk, mddev->array_sectors);
1308 		revalidate_disk_size(mddev->gendisk, true);
1309 	} else {
1310 		/* revert to previous sectors */
1311 		ret = mddev->pers->resize(mddev, old_dev_sectors);
1312 		if (!ret)
1313 			revalidate_disk_size(mddev->gendisk, true);
1314 		ret = __sendmsg(cinfo, &cmsg);
1315 		if (ret)
1316 			pr_err("%s:%d: failed to send METADATA_UPDATED msg\n",
1317 			       __func__, __LINE__);
1318 	}
1319 	unlock_comm(cinfo);
1320 }
1321 
resync_start(struct mddev * mddev)1322 static int resync_start(struct mddev *mddev)
1323 {
1324 	struct md_cluster_info *cinfo = mddev->cluster_info;
1325 	return dlm_lock_sync_interruptible(cinfo->resync_lockres, DLM_LOCK_EX, mddev);
1326 }
1327 
resync_info_get(struct mddev * mddev,sector_t * lo,sector_t * hi)1328 static void resync_info_get(struct mddev *mddev, sector_t *lo, sector_t *hi)
1329 {
1330 	struct md_cluster_info *cinfo = mddev->cluster_info;
1331 
1332 	spin_lock_irq(&cinfo->suspend_lock);
1333 	*lo = cinfo->suspend_lo;
1334 	*hi = cinfo->suspend_hi;
1335 	spin_unlock_irq(&cinfo->suspend_lock);
1336 }
1337 
resync_info_update(struct mddev * mddev,sector_t lo,sector_t hi)1338 static int resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi)
1339 {
1340 	struct md_cluster_info *cinfo = mddev->cluster_info;
1341 	struct resync_info ri;
1342 	struct cluster_msg cmsg = {0};
1343 
1344 	/* do not send zero again, if we have sent before */
1345 	if (hi == 0) {
1346 		memcpy(&ri, cinfo->bitmap_lockres->lksb.sb_lvbptr, sizeof(struct resync_info));
1347 		if (le64_to_cpu(ri.hi) == 0)
1348 			return 0;
1349 	}
1350 
1351 	add_resync_info(cinfo->bitmap_lockres, lo, hi);
1352 	/* Re-acquire the lock to refresh LVB */
1353 	dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW);
1354 	cmsg.type = cpu_to_le32(RESYNCING);
1355 	cmsg.low = cpu_to_le64(lo);
1356 	cmsg.high = cpu_to_le64(hi);
1357 
1358 	/*
1359 	 * mddev_lock is held if resync_info_update is called from
1360 	 * resync_finish (md_reap_sync_thread -> resync_finish)
1361 	 */
1362 	if (lo == 0 && hi == 0)
1363 		return sendmsg(cinfo, &cmsg, 1);
1364 	else
1365 		return sendmsg(cinfo, &cmsg, 0);
1366 }
1367 
resync_finish(struct mddev * mddev)1368 static int resync_finish(struct mddev *mddev)
1369 {
1370 	struct md_cluster_info *cinfo = mddev->cluster_info;
1371 	int ret = 0;
1372 
1373 	clear_bit(MD_RESYNCING_REMOTE, &mddev->recovery);
1374 
1375 	/*
1376 	 * If resync thread is interrupted so we can't say resync is finished,
1377 	 * another node will launch resync thread to continue.
1378 	 */
1379 	if (!test_bit(MD_CLOSING, &mddev->flags))
1380 		ret = resync_info_update(mddev, 0, 0);
1381 	dlm_unlock_sync(cinfo->resync_lockres);
1382 	return ret;
1383 }
1384 
area_resyncing(struct mddev * mddev,int direction,sector_t lo,sector_t hi)1385 static int area_resyncing(struct mddev *mddev, int direction,
1386 		sector_t lo, sector_t hi)
1387 {
1388 	struct md_cluster_info *cinfo = mddev->cluster_info;
1389 	int ret = 0;
1390 
1391 	if ((direction == READ) &&
1392 		test_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state))
1393 		return 1;
1394 
1395 	spin_lock_irq(&cinfo->suspend_lock);
1396 	if (hi > cinfo->suspend_lo && lo < cinfo->suspend_hi)
1397 		ret = 1;
1398 	spin_unlock_irq(&cinfo->suspend_lock);
1399 	return ret;
1400 }
1401 
1402 /* add_new_disk() - initiates a disk add
1403  * However, if this fails before writing md_update_sb(),
1404  * add_new_disk_cancel() must be called to release token lock
1405  */
add_new_disk(struct mddev * mddev,struct md_rdev * rdev)1406 static int add_new_disk(struct mddev *mddev, struct md_rdev *rdev)
1407 {
1408 	struct md_cluster_info *cinfo = mddev->cluster_info;
1409 	struct cluster_msg cmsg;
1410 	int ret = 0;
1411 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1412 	char *uuid = sb->device_uuid;
1413 
1414 	memset(&cmsg, 0, sizeof(cmsg));
1415 	cmsg.type = cpu_to_le32(NEWDISK);
1416 	memcpy(cmsg.uuid, uuid, 16);
1417 	cmsg.raid_slot = cpu_to_le32(rdev->desc_nr);
1418 	if (lock_comm(cinfo, 1))
1419 		return -EAGAIN;
1420 	ret = __sendmsg(cinfo, &cmsg);
1421 	if (ret) {
1422 		unlock_comm(cinfo);
1423 		return ret;
1424 	}
1425 	cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE;
1426 	ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX);
1427 	cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE;
1428 	/* Some node does not "see" the device */
1429 	if (ret == -EAGAIN)
1430 		ret = -ENOENT;
1431 	if (ret)
1432 		unlock_comm(cinfo);
1433 	else {
1434 		dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR);
1435 		/* Since MD_CHANGE_DEVS will be set in add_bound_rdev which
1436 		 * will run soon after add_new_disk, the below path will be
1437 		 * invoked:
1438 		 *   md_wakeup_thread(mddev->thread)
1439 		 *	-> conf->thread (raid1d)
1440 		 *	-> md_check_recovery -> md_update_sb
1441 		 *	-> metadata_update_start/finish
1442 		 * MD_CLUSTER_SEND_LOCKED_ALREADY will be cleared eventually.
1443 		 *
1444 		 * For other failure cases, metadata_update_cancel and
1445 		 * add_new_disk_cancel also clear below bit as well.
1446 		 * */
1447 		set_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state);
1448 		wake_up(&cinfo->wait);
1449 	}
1450 	return ret;
1451 }
1452 
add_new_disk_cancel(struct mddev * mddev)1453 static void add_new_disk_cancel(struct mddev *mddev)
1454 {
1455 	struct md_cluster_info *cinfo = mddev->cluster_info;
1456 	clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state);
1457 	unlock_comm(cinfo);
1458 }
1459 
new_disk_ack(struct mddev * mddev,bool ack)1460 static int new_disk_ack(struct mddev *mddev, bool ack)
1461 {
1462 	struct md_cluster_info *cinfo = mddev->cluster_info;
1463 
1464 	if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) {
1465 		pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev));
1466 		return -EINVAL;
1467 	}
1468 
1469 	if (ack)
1470 		dlm_unlock_sync(cinfo->no_new_dev_lockres);
1471 	complete(&cinfo->newdisk_completion);
1472 	return 0;
1473 }
1474 
remove_disk(struct mddev * mddev,struct md_rdev * rdev)1475 static int remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1476 {
1477 	struct cluster_msg cmsg = {0};
1478 	struct md_cluster_info *cinfo = mddev->cluster_info;
1479 	cmsg.type = cpu_to_le32(REMOVE);
1480 	cmsg.raid_slot = cpu_to_le32(rdev->desc_nr);
1481 	return sendmsg(cinfo, &cmsg, 1);
1482 }
1483 
lock_all_bitmaps(struct mddev * mddev)1484 static int lock_all_bitmaps(struct mddev *mddev)
1485 {
1486 	int slot, my_slot, ret, held = 1, i = 0;
1487 	char str[64];
1488 	struct md_cluster_info *cinfo = mddev->cluster_info;
1489 
1490 	cinfo->other_bitmap_lockres =
1491 		kcalloc(mddev->bitmap_info.nodes - 1,
1492 			sizeof(struct dlm_lock_resource *), GFP_KERNEL);
1493 	if (!cinfo->other_bitmap_lockres) {
1494 		pr_err("md: can't alloc mem for other bitmap locks\n");
1495 		return 0;
1496 	}
1497 
1498 	my_slot = slot_number(mddev);
1499 	for (slot = 0; slot < mddev->bitmap_info.nodes; slot++) {
1500 		if (slot == my_slot)
1501 			continue;
1502 
1503 		memset(str, '\0', 64);
1504 		snprintf(str, 64, "bitmap%04d", slot);
1505 		cinfo->other_bitmap_lockres[i] = lockres_init(mddev, str, NULL, 1);
1506 		if (!cinfo->other_bitmap_lockres[i])
1507 			return -ENOMEM;
1508 
1509 		cinfo->other_bitmap_lockres[i]->flags |= DLM_LKF_NOQUEUE;
1510 		ret = dlm_lock_sync(cinfo->other_bitmap_lockres[i], DLM_LOCK_PW);
1511 		if (ret)
1512 			held = -1;
1513 		i++;
1514 	}
1515 
1516 	return held;
1517 }
1518 
unlock_all_bitmaps(struct mddev * mddev)1519 static void unlock_all_bitmaps(struct mddev *mddev)
1520 {
1521 	struct md_cluster_info *cinfo = mddev->cluster_info;
1522 	int i;
1523 
1524 	/* release other node's bitmap lock if they are existed */
1525 	if (cinfo->other_bitmap_lockres) {
1526 		for (i = 0; i < mddev->bitmap_info.nodes - 1; i++) {
1527 			if (cinfo->other_bitmap_lockres[i]) {
1528 				lockres_free(cinfo->other_bitmap_lockres[i]);
1529 			}
1530 		}
1531 		kfree(cinfo->other_bitmap_lockres);
1532 		cinfo->other_bitmap_lockres = NULL;
1533 	}
1534 }
1535 
gather_bitmaps(struct md_rdev * rdev)1536 static int gather_bitmaps(struct md_rdev *rdev)
1537 {
1538 	int sn, err;
1539 	sector_t lo, hi;
1540 	struct cluster_msg cmsg = {0};
1541 	struct mddev *mddev = rdev->mddev;
1542 	struct md_cluster_info *cinfo = mddev->cluster_info;
1543 
1544 	cmsg.type = cpu_to_le32(RE_ADD);
1545 	cmsg.raid_slot = cpu_to_le32(rdev->desc_nr);
1546 	err = sendmsg(cinfo, &cmsg, 1);
1547 	if (err)
1548 		goto out;
1549 
1550 	for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) {
1551 		if (sn == (cinfo->slot_number - 1))
1552 			continue;
1553 		err = md_bitmap_copy_from_slot(mddev, sn, &lo, &hi, false);
1554 		if (err) {
1555 			pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn);
1556 			goto out;
1557 		}
1558 		if ((hi > 0) && (lo < mddev->recovery_cp))
1559 			mddev->recovery_cp = lo;
1560 	}
1561 out:
1562 	return err;
1563 }
1564 
1565 static struct md_cluster_operations cluster_ops = {
1566 	.join   = join,
1567 	.leave  = leave,
1568 	.slot_number = slot_number,
1569 	.resync_start = resync_start,
1570 	.resync_finish = resync_finish,
1571 	.resync_info_update = resync_info_update,
1572 	.resync_info_get = resync_info_get,
1573 	.metadata_update_start = metadata_update_start,
1574 	.metadata_update_finish = metadata_update_finish,
1575 	.metadata_update_cancel = metadata_update_cancel,
1576 	.area_resyncing = area_resyncing,
1577 	.add_new_disk = add_new_disk,
1578 	.add_new_disk_cancel = add_new_disk_cancel,
1579 	.new_disk_ack = new_disk_ack,
1580 	.remove_disk = remove_disk,
1581 	.load_bitmaps = load_bitmaps,
1582 	.gather_bitmaps = gather_bitmaps,
1583 	.resize_bitmaps = resize_bitmaps,
1584 	.lock_all_bitmaps = lock_all_bitmaps,
1585 	.unlock_all_bitmaps = unlock_all_bitmaps,
1586 	.update_size = update_size,
1587 };
1588 
cluster_init(void)1589 static int __init cluster_init(void)
1590 {
1591 	pr_warn("md-cluster: support raid1 and raid10 (limited support)\n");
1592 	pr_info("Registering Cluster MD functions\n");
1593 	register_md_cluster_operations(&cluster_ops, THIS_MODULE);
1594 	return 0;
1595 }
1596 
cluster_exit(void)1597 static void cluster_exit(void)
1598 {
1599 	unregister_md_cluster_operations();
1600 }
1601 
1602 module_init(cluster_init);
1603 module_exit(cluster_exit);
1604 MODULE_AUTHOR("SUSE");
1605 MODULE_LICENSE("GPL");
1606 MODULE_DESCRIPTION("Clustering support for MD");
1607