• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/dlm.h>
14 #include <linux/sched.h>
15 #include <linux/raid/md_p.h>
16 #include "md.h"
17 #include "bitmap.h"
18 #include "md-cluster.h"
19 
20 #define LVB_SIZE	64
21 #define NEW_DEV_TIMEOUT 5000
22 
23 struct dlm_lock_resource {
24 	dlm_lockspace_t *ls;
25 	struct dlm_lksb lksb;
26 	char *name; /* lock name. */
27 	uint32_t flags; /* flags to pass to dlm_lock() */
28 	struct completion completion; /* completion for synchronized locking */
29 	void (*bast)(void *arg, int mode); /* blocking AST function pointer*/
30 	struct mddev *mddev; /* pointing back to mddev. */
31 	int mode;
32 };
33 
34 struct suspend_info {
35 	int slot;
36 	sector_t lo;
37 	sector_t hi;
38 	struct list_head list;
39 };
40 
41 struct resync_info {
42 	__le64 lo;
43 	__le64 hi;
44 };
45 
46 /* md_cluster_info flags */
47 #define		MD_CLUSTER_WAITING_FOR_NEWDISK		1
48 #define		MD_CLUSTER_SUSPEND_READ_BALANCING	2
49 #define		MD_CLUSTER_BEGIN_JOIN_CLUSTER		3
50 
51 
52 struct md_cluster_info {
53 	/* dlm lock space and resources for clustered raid. */
54 	dlm_lockspace_t *lockspace;
55 	int slot_number;
56 	struct completion completion;
57 	struct dlm_lock_resource *bitmap_lockres;
58 	struct dlm_lock_resource *resync_lockres;
59 	struct list_head suspend_list;
60 	spinlock_t suspend_lock;
61 	struct md_thread *recovery_thread;
62 	unsigned long recovery_map;
63 	/* communication loc resources */
64 	struct dlm_lock_resource *ack_lockres;
65 	struct dlm_lock_resource *message_lockres;
66 	struct dlm_lock_resource *token_lockres;
67 	struct dlm_lock_resource *no_new_dev_lockres;
68 	struct md_thread *recv_thread;
69 	struct completion newdisk_completion;
70 	unsigned long state;
71 };
72 
73 enum msg_type {
74 	METADATA_UPDATED = 0,
75 	RESYNCING,
76 	NEWDISK,
77 	REMOVE,
78 	RE_ADD,
79 	BITMAP_NEEDS_SYNC,
80 };
81 
82 struct cluster_msg {
83 	__le32 type;
84 	__le32 slot;
85 	/* TODO: Unionize this for smaller footprint */
86 	__le64 low;
87 	__le64 high;
88 	char uuid[16];
89 	__le32 raid_slot;
90 };
91 
sync_ast(void * arg)92 static void sync_ast(void *arg)
93 {
94 	struct dlm_lock_resource *res;
95 
96 	res = arg;
97 	complete(&res->completion);
98 }
99 
dlm_lock_sync(struct dlm_lock_resource * res,int mode)100 static int dlm_lock_sync(struct dlm_lock_resource *res, int mode)
101 {
102 	int ret = 0;
103 
104 	ret = dlm_lock(res->ls, mode, &res->lksb,
105 			res->flags, res->name, strlen(res->name),
106 			0, sync_ast, res, res->bast);
107 	if (ret)
108 		return ret;
109 	wait_for_completion(&res->completion);
110 	if (res->lksb.sb_status == 0)
111 		res->mode = mode;
112 	return res->lksb.sb_status;
113 }
114 
dlm_unlock_sync(struct dlm_lock_resource * res)115 static int dlm_unlock_sync(struct dlm_lock_resource *res)
116 {
117 	return dlm_lock_sync(res, DLM_LOCK_NL);
118 }
119 
lockres_init(struct mddev * mddev,char * name,void (* bastfn)(void * arg,int mode),int with_lvb)120 static struct dlm_lock_resource *lockres_init(struct mddev *mddev,
121 		char *name, void (*bastfn)(void *arg, int mode), int with_lvb)
122 {
123 	struct dlm_lock_resource *res = NULL;
124 	int ret, namelen;
125 	struct md_cluster_info *cinfo = mddev->cluster_info;
126 
127 	res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL);
128 	if (!res)
129 		return NULL;
130 	init_completion(&res->completion);
131 	res->ls = cinfo->lockspace;
132 	res->mddev = mddev;
133 	res->mode = DLM_LOCK_IV;
134 	namelen = strlen(name);
135 	res->name = kzalloc(namelen + 1, GFP_KERNEL);
136 	if (!res->name) {
137 		pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name);
138 		goto out_err;
139 	}
140 	strlcpy(res->name, name, namelen + 1);
141 	if (with_lvb) {
142 		res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL);
143 		if (!res->lksb.sb_lvbptr) {
144 			pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name);
145 			goto out_err;
146 		}
147 		res->flags = DLM_LKF_VALBLK;
148 	}
149 
150 	if (bastfn)
151 		res->bast = bastfn;
152 
153 	res->flags |= DLM_LKF_EXPEDITE;
154 
155 	ret = dlm_lock_sync(res, DLM_LOCK_NL);
156 	if (ret) {
157 		pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name);
158 		goto out_err;
159 	}
160 	res->flags &= ~DLM_LKF_EXPEDITE;
161 	res->flags |= DLM_LKF_CONVERT;
162 
163 	return res;
164 out_err:
165 	kfree(res->lksb.sb_lvbptr);
166 	kfree(res->name);
167 	kfree(res);
168 	return NULL;
169 }
170 
lockres_free(struct dlm_lock_resource * res)171 static void lockres_free(struct dlm_lock_resource *res)
172 {
173 	int ret;
174 
175 	if (!res)
176 		return;
177 
178 	/* cancel a lock request or a conversion request that is blocked */
179 	res->flags |= DLM_LKF_CANCEL;
180 retry:
181 	ret = dlm_unlock(res->ls, res->lksb.sb_lkid, 0, &res->lksb, res);
182 	if (unlikely(ret != 0)) {
183 		pr_info("%s: failed to unlock %s return %d\n", __func__, res->name, ret);
184 
185 		/* if a lock conversion is cancelled, then the lock is put
186 		 * back to grant queue, need to ensure it is unlocked */
187 		if (ret == -DLM_ECANCEL)
188 			goto retry;
189 	}
190 	res->flags &= ~DLM_LKF_CANCEL;
191 	wait_for_completion(&res->completion);
192 
193 	kfree(res->name);
194 	kfree(res->lksb.sb_lvbptr);
195 	kfree(res);
196 }
197 
add_resync_info(struct dlm_lock_resource * lockres,sector_t lo,sector_t hi)198 static void add_resync_info(struct dlm_lock_resource *lockres,
199 			    sector_t lo, sector_t hi)
200 {
201 	struct resync_info *ri;
202 
203 	ri = (struct resync_info *)lockres->lksb.sb_lvbptr;
204 	ri->lo = cpu_to_le64(lo);
205 	ri->hi = cpu_to_le64(hi);
206 }
207 
read_resync_info(struct mddev * mddev,struct dlm_lock_resource * lockres)208 static struct suspend_info *read_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres)
209 {
210 	struct resync_info ri;
211 	struct suspend_info *s = NULL;
212 	sector_t hi = 0;
213 
214 	dlm_lock_sync(lockres, DLM_LOCK_CR);
215 	memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info));
216 	hi = le64_to_cpu(ri.hi);
217 	if (hi > 0) {
218 		s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL);
219 		if (!s)
220 			goto out;
221 		s->hi = hi;
222 		s->lo = le64_to_cpu(ri.lo);
223 	}
224 	dlm_unlock_sync(lockres);
225 out:
226 	return s;
227 }
228 
recover_bitmaps(struct md_thread * thread)229 static void recover_bitmaps(struct md_thread *thread)
230 {
231 	struct mddev *mddev = thread->mddev;
232 	struct md_cluster_info *cinfo = mddev->cluster_info;
233 	struct dlm_lock_resource *bm_lockres;
234 	char str[64];
235 	int slot, ret;
236 	struct suspend_info *s, *tmp;
237 	sector_t lo, hi;
238 
239 	while (cinfo->recovery_map) {
240 		slot = fls64((u64)cinfo->recovery_map) - 1;
241 
242 		snprintf(str, 64, "bitmap%04d", slot);
243 		bm_lockres = lockres_init(mddev, str, NULL, 1);
244 		if (!bm_lockres) {
245 			pr_err("md-cluster: Cannot initialize bitmaps\n");
246 			goto clear_bit;
247 		}
248 
249 		ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
250 		if (ret) {
251 			pr_err("md-cluster: Could not DLM lock %s: %d\n",
252 					str, ret);
253 			goto clear_bit;
254 		}
255 		ret = bitmap_copy_from_slot(mddev, slot, &lo, &hi, true);
256 		if (ret) {
257 			pr_err("md-cluster: Could not copy data from bitmap %d\n", slot);
258 			goto dlm_unlock;
259 		}
260 
261 		/* Clear suspend_area associated with the bitmap */
262 		spin_lock_irq(&cinfo->suspend_lock);
263 		list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list)
264 			if (slot == s->slot) {
265 				list_del(&s->list);
266 				kfree(s);
267 			}
268 		spin_unlock_irq(&cinfo->suspend_lock);
269 
270 		if (hi > 0) {
271 			/* TODO:Wait for current resync to get over */
272 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
273 			if (lo < mddev->recovery_cp)
274 				mddev->recovery_cp = lo;
275 			md_check_recovery(mddev);
276 		}
277 dlm_unlock:
278 		dlm_unlock_sync(bm_lockres);
279 clear_bit:
280 		clear_bit(slot, &cinfo->recovery_map);
281 	}
282 }
283 
recover_prep(void * arg)284 static void recover_prep(void *arg)
285 {
286 	struct mddev *mddev = arg;
287 	struct md_cluster_info *cinfo = mddev->cluster_info;
288 	set_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state);
289 }
290 
__recover_slot(struct mddev * mddev,int slot)291 static void __recover_slot(struct mddev *mddev, int slot)
292 {
293 	struct md_cluster_info *cinfo = mddev->cluster_info;
294 
295 	set_bit(slot, &cinfo->recovery_map);
296 	if (!cinfo->recovery_thread) {
297 		cinfo->recovery_thread = md_register_thread(recover_bitmaps,
298 				mddev, "recover");
299 		if (!cinfo->recovery_thread) {
300 			pr_warn("md-cluster: Could not create recovery thread\n");
301 			return;
302 		}
303 	}
304 	md_wakeup_thread(cinfo->recovery_thread);
305 }
306 
recover_slot(void * arg,struct dlm_slot * slot)307 static void recover_slot(void *arg, struct dlm_slot *slot)
308 {
309 	struct mddev *mddev = arg;
310 	struct md_cluster_info *cinfo = mddev->cluster_info;
311 
312 	pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n",
313 			mddev->bitmap_info.cluster_name,
314 			slot->nodeid, slot->slot,
315 			cinfo->slot_number);
316 	/* deduct one since dlm slot starts from one while the num of
317 	 * cluster-md begins with 0 */
318 	__recover_slot(mddev, slot->slot - 1);
319 }
320 
recover_done(void * arg,struct dlm_slot * slots,int num_slots,int our_slot,uint32_t generation)321 static void recover_done(void *arg, struct dlm_slot *slots,
322 		int num_slots, int our_slot,
323 		uint32_t generation)
324 {
325 	struct mddev *mddev = arg;
326 	struct md_cluster_info *cinfo = mddev->cluster_info;
327 
328 	cinfo->slot_number = our_slot;
329 	/* completion is only need to be complete when node join cluster,
330 	 * it doesn't need to run during another node's failure */
331 	if (test_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state)) {
332 		complete(&cinfo->completion);
333 		clear_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state);
334 	}
335 	clear_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state);
336 }
337 
338 /* the ops is called when node join the cluster, and do lock recovery
339  * if node failure occurs */
340 static const struct dlm_lockspace_ops md_ls_ops = {
341 	.recover_prep = recover_prep,
342 	.recover_slot = recover_slot,
343 	.recover_done = recover_done,
344 };
345 
346 /*
347  * The BAST function for the ack lock resource
348  * This function wakes up the receive thread in
349  * order to receive and process the message.
350  */
ack_bast(void * arg,int mode)351 static void ack_bast(void *arg, int mode)
352 {
353 	struct dlm_lock_resource *res = arg;
354 	struct md_cluster_info *cinfo = res->mddev->cluster_info;
355 
356 	if (mode == DLM_LOCK_EX)
357 		md_wakeup_thread(cinfo->recv_thread);
358 }
359 
__remove_suspend_info(struct md_cluster_info * cinfo,int slot)360 static void __remove_suspend_info(struct md_cluster_info *cinfo, int slot)
361 {
362 	struct suspend_info *s, *tmp;
363 
364 	list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list)
365 		if (slot == s->slot) {
366 			list_del(&s->list);
367 			kfree(s);
368 			break;
369 		}
370 }
371 
remove_suspend_info(struct mddev * mddev,int slot)372 static void remove_suspend_info(struct mddev *mddev, int slot)
373 {
374 	struct md_cluster_info *cinfo = mddev->cluster_info;
375 	spin_lock_irq(&cinfo->suspend_lock);
376 	__remove_suspend_info(cinfo, slot);
377 	spin_unlock_irq(&cinfo->suspend_lock);
378 	mddev->pers->quiesce(mddev, 2);
379 }
380 
381 
process_suspend_info(struct mddev * mddev,int slot,sector_t lo,sector_t hi)382 static void process_suspend_info(struct mddev *mddev,
383 		int slot, sector_t lo, sector_t hi)
384 {
385 	struct md_cluster_info *cinfo = mddev->cluster_info;
386 	struct suspend_info *s;
387 
388 	if (!hi) {
389 		remove_suspend_info(mddev, slot);
390 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
391 		md_wakeup_thread(mddev->thread);
392 		return;
393 	}
394 	s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL);
395 	if (!s)
396 		return;
397 	s->slot = slot;
398 	s->lo = lo;
399 	s->hi = hi;
400 	mddev->pers->quiesce(mddev, 1);
401 	mddev->pers->quiesce(mddev, 0);
402 	spin_lock_irq(&cinfo->suspend_lock);
403 	/* Remove existing entry (if exists) before adding */
404 	__remove_suspend_info(cinfo, slot);
405 	list_add(&s->list, &cinfo->suspend_list);
406 	spin_unlock_irq(&cinfo->suspend_lock);
407 	mddev->pers->quiesce(mddev, 2);
408 }
409 
process_add_new_disk(struct mddev * mddev,struct cluster_msg * cmsg)410 static void process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg)
411 {
412 	char disk_uuid[64];
413 	struct md_cluster_info *cinfo = mddev->cluster_info;
414 	char event_name[] = "EVENT=ADD_DEVICE";
415 	char raid_slot[16];
416 	char *envp[] = {event_name, disk_uuid, raid_slot, NULL};
417 	int len;
418 
419 	len = snprintf(disk_uuid, 64, "DEVICE_UUID=");
420 	sprintf(disk_uuid + len, "%pU", cmsg->uuid);
421 	snprintf(raid_slot, 16, "RAID_DISK=%d", le32_to_cpu(cmsg->raid_slot));
422 	pr_info("%s:%d Sending kobject change with %s and %s\n", __func__, __LINE__, disk_uuid, raid_slot);
423 	init_completion(&cinfo->newdisk_completion);
424 	set_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state);
425 	kobject_uevent_env(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE, envp);
426 	wait_for_completion_timeout(&cinfo->newdisk_completion,
427 			NEW_DEV_TIMEOUT);
428 	clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state);
429 }
430 
431 
process_metadata_update(struct mddev * mddev,struct cluster_msg * msg)432 static void process_metadata_update(struct mddev *mddev, struct cluster_msg *msg)
433 {
434 	struct md_cluster_info *cinfo = mddev->cluster_info;
435 	md_reload_sb(mddev, le32_to_cpu(msg->raid_slot));
436 	dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR);
437 }
438 
process_remove_disk(struct mddev * mddev,struct cluster_msg * msg)439 static void process_remove_disk(struct mddev *mddev, struct cluster_msg *msg)
440 {
441 	struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev,
442 						   le32_to_cpu(msg->raid_slot));
443 
444 	if (rdev)
445 		md_kick_rdev_from_array(rdev);
446 	else
447 		pr_warn("%s: %d Could not find disk(%d) to REMOVE\n",
448 			__func__, __LINE__, le32_to_cpu(msg->raid_slot));
449 }
450 
process_readd_disk(struct mddev * mddev,struct cluster_msg * msg)451 static void process_readd_disk(struct mddev *mddev, struct cluster_msg *msg)
452 {
453 	struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev,
454 						   le32_to_cpu(msg->raid_slot));
455 
456 	if (rdev && test_bit(Faulty, &rdev->flags))
457 		clear_bit(Faulty, &rdev->flags);
458 	else
459 		pr_warn("%s: %d Could not find disk(%d) which is faulty",
460 			__func__, __LINE__, le32_to_cpu(msg->raid_slot));
461 }
462 
process_recvd_msg(struct mddev * mddev,struct cluster_msg * msg)463 static void process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg)
464 {
465 	if (WARN(mddev->cluster_info->slot_number - 1 == le32_to_cpu(msg->slot),
466 		"node %d received it's own msg\n", le32_to_cpu(msg->slot)))
467 		return;
468 	switch (le32_to_cpu(msg->type)) {
469 	case METADATA_UPDATED:
470 		process_metadata_update(mddev, msg);
471 		break;
472 	case RESYNCING:
473 		process_suspend_info(mddev, le32_to_cpu(msg->slot),
474 				     le64_to_cpu(msg->low),
475 				     le64_to_cpu(msg->high));
476 		break;
477 	case NEWDISK:
478 		process_add_new_disk(mddev, msg);
479 		break;
480 	case REMOVE:
481 		process_remove_disk(mddev, msg);
482 		break;
483 	case RE_ADD:
484 		process_readd_disk(mddev, msg);
485 		break;
486 	case BITMAP_NEEDS_SYNC:
487 		__recover_slot(mddev, le32_to_cpu(msg->slot));
488 		break;
489 	default:
490 		pr_warn("%s:%d Received unknown message from %d\n",
491 			__func__, __LINE__, msg->slot);
492 	}
493 }
494 
495 /*
496  * thread for receiving message
497  */
recv_daemon(struct md_thread * thread)498 static void recv_daemon(struct md_thread *thread)
499 {
500 	struct md_cluster_info *cinfo = thread->mddev->cluster_info;
501 	struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres;
502 	struct dlm_lock_resource *message_lockres = cinfo->message_lockres;
503 	struct cluster_msg msg;
504 	int ret;
505 
506 	/*get CR on Message*/
507 	if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) {
508 		pr_err("md/raid1:failed to get CR on MESSAGE\n");
509 		return;
510 	}
511 
512 	/* read lvb and wake up thread to process this message_lockres */
513 	memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg));
514 	process_recvd_msg(thread->mddev, &msg);
515 
516 	/*release CR on ack_lockres*/
517 	ret = dlm_unlock_sync(ack_lockres);
518 	if (unlikely(ret != 0))
519 		pr_info("unlock ack failed return %d\n", ret);
520 	/*up-convert to PR on message_lockres*/
521 	ret = dlm_lock_sync(message_lockres, DLM_LOCK_PR);
522 	if (unlikely(ret != 0))
523 		pr_info("lock PR on msg failed return %d\n", ret);
524 	/*get CR on ack_lockres again*/
525 	ret = dlm_lock_sync(ack_lockres, DLM_LOCK_CR);
526 	if (unlikely(ret != 0))
527 		pr_info("lock CR on ack failed return %d\n", ret);
528 	/*release CR on message_lockres*/
529 	ret = dlm_unlock_sync(message_lockres);
530 	if (unlikely(ret != 0))
531 		pr_info("unlock msg failed return %d\n", ret);
532 }
533 
534 /* lock_comm()
535  * Takes the lock on the TOKEN lock resource so no other
536  * node can communicate while the operation is underway.
537  * If called again, and the TOKEN lock is alread in EX mode
538  * return success. However, care must be taken that unlock_comm()
539  * is called only once.
540  */
lock_comm(struct md_cluster_info * cinfo)541 static int lock_comm(struct md_cluster_info *cinfo)
542 {
543 	int error;
544 
545 	if (cinfo->token_lockres->mode == DLM_LOCK_EX)
546 		return 0;
547 
548 	error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX);
549 	if (error)
550 		pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n",
551 				__func__, __LINE__, error);
552 	return error;
553 }
554 
unlock_comm(struct md_cluster_info * cinfo)555 static void unlock_comm(struct md_cluster_info *cinfo)
556 {
557 	WARN_ON(cinfo->token_lockres->mode != DLM_LOCK_EX);
558 	dlm_unlock_sync(cinfo->token_lockres);
559 }
560 
561 /* __sendmsg()
562  * This function performs the actual sending of the message. This function is
563  * usually called after performing the encompassing operation
564  * The function:
565  * 1. Grabs the message lockresource in EX mode
566  * 2. Copies the message to the message LVB
567  * 3. Downconverts message lockresource to CW
568  * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes
569  *    and the other nodes read the message. The thread will wait here until all other
570  *    nodes have released ack lock resource.
571  * 5. Downconvert ack lockresource to CR
572  */
__sendmsg(struct md_cluster_info * cinfo,struct cluster_msg * cmsg)573 static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg)
574 {
575 	int error;
576 	int slot = cinfo->slot_number - 1;
577 
578 	cmsg->slot = cpu_to_le32(slot);
579 	/*get EX on Message*/
580 	error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX);
581 	if (error) {
582 		pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error);
583 		goto failed_message;
584 	}
585 
586 	memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg,
587 			sizeof(struct cluster_msg));
588 	/*down-convert EX to CW on Message*/
589 	error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CW);
590 	if (error) {
591 		pr_err("md-cluster: failed to convert EX to CW on MESSAGE(%d)\n",
592 				error);
593 		goto failed_ack;
594 	}
595 
596 	/*up-convert CR to EX on Ack*/
597 	error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX);
598 	if (error) {
599 		pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n",
600 				error);
601 		goto failed_ack;
602 	}
603 
604 	/*down-convert EX to CR on Ack*/
605 	error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR);
606 	if (error) {
607 		pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n",
608 				error);
609 		goto failed_ack;
610 	}
611 
612 failed_ack:
613 	error = dlm_unlock_sync(cinfo->message_lockres);
614 	if (unlikely(error != 0)) {
615 		pr_err("md-cluster: failed convert to NL on MESSAGE(%d)\n",
616 			error);
617 		/* in case the message can't be released due to some reason */
618 		goto failed_ack;
619 	}
620 failed_message:
621 	return error;
622 }
623 
sendmsg(struct md_cluster_info * cinfo,struct cluster_msg * cmsg)624 static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg)
625 {
626 	int ret;
627 
628 	lock_comm(cinfo);
629 	ret = __sendmsg(cinfo, cmsg);
630 	unlock_comm(cinfo);
631 	return ret;
632 }
633 
gather_all_resync_info(struct mddev * mddev,int total_slots)634 static int gather_all_resync_info(struct mddev *mddev, int total_slots)
635 {
636 	struct md_cluster_info *cinfo = mddev->cluster_info;
637 	int i, ret = 0;
638 	struct dlm_lock_resource *bm_lockres;
639 	struct suspend_info *s;
640 	char str[64];
641 	sector_t lo, hi;
642 
643 
644 	for (i = 0; i < total_slots; i++) {
645 		memset(str, '\0', 64);
646 		snprintf(str, 64, "bitmap%04d", i);
647 		bm_lockres = lockres_init(mddev, str, NULL, 1);
648 		if (!bm_lockres)
649 			return -ENOMEM;
650 		if (i == (cinfo->slot_number - 1))
651 			continue;
652 
653 		bm_lockres->flags |= DLM_LKF_NOQUEUE;
654 		ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
655 		if (ret == -EAGAIN) {
656 			memset(bm_lockres->lksb.sb_lvbptr, '\0', LVB_SIZE);
657 			s = read_resync_info(mddev, bm_lockres);
658 			if (s) {
659 				pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n",
660 						__func__, __LINE__,
661 						(unsigned long long) s->lo,
662 						(unsigned long long) s->hi, i);
663 				spin_lock_irq(&cinfo->suspend_lock);
664 				s->slot = i;
665 				list_add(&s->list, &cinfo->suspend_list);
666 				spin_unlock_irq(&cinfo->suspend_lock);
667 			}
668 			ret = 0;
669 			lockres_free(bm_lockres);
670 			continue;
671 		}
672 		if (ret) {
673 			lockres_free(bm_lockres);
674 			goto out;
675 		}
676 
677 		/* Read the disk bitmap sb and check if it needs recovery */
678 		ret = bitmap_copy_from_slot(mddev, i, &lo, &hi, false);
679 		if (ret) {
680 			pr_warn("md-cluster: Could not gather bitmaps from slot %d", i);
681 			lockres_free(bm_lockres);
682 			continue;
683 		}
684 		if ((hi > 0) && (lo < mddev->recovery_cp)) {
685 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
686 			mddev->recovery_cp = lo;
687 			md_check_recovery(mddev);
688 		}
689 
690 		dlm_unlock_sync(bm_lockres);
691 		lockres_free(bm_lockres);
692 	}
693 out:
694 	return ret;
695 }
696 
join(struct mddev * mddev,int nodes)697 static int join(struct mddev *mddev, int nodes)
698 {
699 	struct md_cluster_info *cinfo;
700 	int ret, ops_rv;
701 	char str[64];
702 
703 	cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL);
704 	if (!cinfo)
705 		return -ENOMEM;
706 
707 	INIT_LIST_HEAD(&cinfo->suspend_list);
708 	spin_lock_init(&cinfo->suspend_lock);
709 	init_completion(&cinfo->completion);
710 	set_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state);
711 
712 	mddev->cluster_info = cinfo;
713 
714 	memset(str, 0, 64);
715 	sprintf(str, "%pU", mddev->uuid);
716 	ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name,
717 				DLM_LSFL_FS, LVB_SIZE,
718 				&md_ls_ops, mddev, &ops_rv, &cinfo->lockspace);
719 	if (ret)
720 		goto err;
721 	wait_for_completion(&cinfo->completion);
722 	if (nodes < cinfo->slot_number) {
723 		pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).",
724 			cinfo->slot_number, nodes);
725 		ret = -ERANGE;
726 		goto err;
727 	}
728 	/* Initiate the communication resources */
729 	ret = -ENOMEM;
730 	cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv");
731 	if (!cinfo->recv_thread) {
732 		pr_err("md-cluster: cannot allocate memory for recv_thread!\n");
733 		goto err;
734 	}
735 	cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1);
736 	if (!cinfo->message_lockres)
737 		goto err;
738 	cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0);
739 	if (!cinfo->token_lockres)
740 		goto err;
741 	cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0);
742 	if (!cinfo->ack_lockres)
743 		goto err;
744 	cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0);
745 	if (!cinfo->no_new_dev_lockres)
746 		goto err;
747 
748 	/* get sync CR lock on ACK. */
749 	if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR))
750 		pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n",
751 				ret);
752 	/* get sync CR lock on no-new-dev. */
753 	if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR))
754 		pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret);
755 
756 
757 	pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number);
758 	snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1);
759 	cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1);
760 	if (!cinfo->bitmap_lockres)
761 		goto err;
762 	if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) {
763 		pr_err("Failed to get bitmap lock\n");
764 		ret = -EINVAL;
765 		goto err;
766 	}
767 
768 	cinfo->resync_lockres = lockres_init(mddev, "resync", NULL, 0);
769 	if (!cinfo->resync_lockres)
770 		goto err;
771 
772 	ret = gather_all_resync_info(mddev, nodes);
773 	if (ret)
774 		goto err;
775 
776 	return 0;
777 err:
778 	lockres_free(cinfo->message_lockres);
779 	lockres_free(cinfo->token_lockres);
780 	lockres_free(cinfo->ack_lockres);
781 	lockres_free(cinfo->no_new_dev_lockres);
782 	lockres_free(cinfo->resync_lockres);
783 	lockres_free(cinfo->bitmap_lockres);
784 	if (cinfo->lockspace)
785 		dlm_release_lockspace(cinfo->lockspace, 2);
786 	mddev->cluster_info = NULL;
787 	kfree(cinfo);
788 	return ret;
789 }
790 
resync_bitmap(struct mddev * mddev)791 static void resync_bitmap(struct mddev *mddev)
792 {
793 	struct md_cluster_info *cinfo = mddev->cluster_info;
794 	struct cluster_msg cmsg = {0};
795 	int err;
796 
797 	cmsg.type = cpu_to_le32(BITMAP_NEEDS_SYNC);
798 	err = sendmsg(cinfo, &cmsg);
799 	if (err)
800 		pr_err("%s:%d: failed to send BITMAP_NEEDS_SYNC message (%d)\n",
801 			__func__, __LINE__, err);
802 }
803 
leave(struct mddev * mddev)804 static int leave(struct mddev *mddev)
805 {
806 	struct md_cluster_info *cinfo = mddev->cluster_info;
807 
808 	if (!cinfo)
809 		return 0;
810 
811 	/* BITMAP_NEEDS_SYNC message should be sent when node
812 	 * is leaving the cluster with dirty bitmap, also we
813 	 * can only deliver it when dlm connection is available */
814 	if (cinfo->slot_number > 0 && mddev->recovery_cp != MaxSector)
815 		resync_bitmap(mddev);
816 
817 	md_unregister_thread(&cinfo->recovery_thread);
818 	md_unregister_thread(&cinfo->recv_thread);
819 	lockres_free(cinfo->message_lockres);
820 	lockres_free(cinfo->token_lockres);
821 	lockres_free(cinfo->ack_lockres);
822 	lockres_free(cinfo->no_new_dev_lockres);
823 	lockres_free(cinfo->bitmap_lockres);
824 	dlm_release_lockspace(cinfo->lockspace, 2);
825 	kfree(cinfo);
826 	return 0;
827 }
828 
829 /* slot_number(): Returns the MD slot number to use
830  * DLM starts the slot numbers from 1, wheras cluster-md
831  * wants the number to be from zero, so we deduct one
832  */
slot_number(struct mddev * mddev)833 static int slot_number(struct mddev *mddev)
834 {
835 	struct md_cluster_info *cinfo = mddev->cluster_info;
836 
837 	return cinfo->slot_number - 1;
838 }
839 
metadata_update_start(struct mddev * mddev)840 static int metadata_update_start(struct mddev *mddev)
841 {
842 	return lock_comm(mddev->cluster_info);
843 }
844 
metadata_update_finish(struct mddev * mddev)845 static int metadata_update_finish(struct mddev *mddev)
846 {
847 	struct md_cluster_info *cinfo = mddev->cluster_info;
848 	struct cluster_msg cmsg;
849 	struct md_rdev *rdev;
850 	int ret = 0;
851 	int raid_slot = -1;
852 
853 	memset(&cmsg, 0, sizeof(cmsg));
854 	cmsg.type = cpu_to_le32(METADATA_UPDATED);
855 	/* Pick up a good active device number to send.
856 	 */
857 	rdev_for_each(rdev, mddev)
858 		if (rdev->raid_disk > -1 && !test_bit(Faulty, &rdev->flags)) {
859 			raid_slot = rdev->desc_nr;
860 			break;
861 		}
862 	if (raid_slot >= 0) {
863 		cmsg.raid_slot = cpu_to_le32(raid_slot);
864 		ret = __sendmsg(cinfo, &cmsg);
865 	} else
866 		pr_warn("md-cluster: No good device id found to send\n");
867 	unlock_comm(cinfo);
868 	return ret;
869 }
870 
metadata_update_cancel(struct mddev * mddev)871 static void metadata_update_cancel(struct mddev *mddev)
872 {
873 	struct md_cluster_info *cinfo = mddev->cluster_info;
874 	unlock_comm(cinfo);
875 }
876 
resync_start(struct mddev * mddev)877 static int resync_start(struct mddev *mddev)
878 {
879 	struct md_cluster_info *cinfo = mddev->cluster_info;
880 	cinfo->resync_lockres->flags |= DLM_LKF_NOQUEUE;
881 	return dlm_lock_sync(cinfo->resync_lockres, DLM_LOCK_EX);
882 }
883 
resync_info_update(struct mddev * mddev,sector_t lo,sector_t hi)884 static int resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi)
885 {
886 	struct md_cluster_info *cinfo = mddev->cluster_info;
887 	struct cluster_msg cmsg = {0};
888 
889 	add_resync_info(cinfo->bitmap_lockres, lo, hi);
890 	/* Re-acquire the lock to refresh LVB */
891 	dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW);
892 	cmsg.type = cpu_to_le32(RESYNCING);
893 	cmsg.low = cpu_to_le64(lo);
894 	cmsg.high = cpu_to_le64(hi);
895 
896 	return sendmsg(cinfo, &cmsg);
897 }
898 
resync_finish(struct mddev * mddev)899 static int resync_finish(struct mddev *mddev)
900 {
901 	struct md_cluster_info *cinfo = mddev->cluster_info;
902 	cinfo->resync_lockres->flags &= ~DLM_LKF_NOQUEUE;
903 	dlm_unlock_sync(cinfo->resync_lockres);
904 	return resync_info_update(mddev, 0, 0);
905 }
906 
area_resyncing(struct mddev * mddev,int direction,sector_t lo,sector_t hi)907 static int area_resyncing(struct mddev *mddev, int direction,
908 		sector_t lo, sector_t hi)
909 {
910 	struct md_cluster_info *cinfo = mddev->cluster_info;
911 	int ret = 0;
912 	struct suspend_info *s;
913 
914 	if ((direction == READ) &&
915 		test_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state))
916 		return 1;
917 
918 	spin_lock_irq(&cinfo->suspend_lock);
919 	if (list_empty(&cinfo->suspend_list))
920 		goto out;
921 	list_for_each_entry(s, &cinfo->suspend_list, list)
922 		if (hi > s->lo && lo < s->hi) {
923 			ret = 1;
924 			break;
925 		}
926 out:
927 	spin_unlock_irq(&cinfo->suspend_lock);
928 	return ret;
929 }
930 
931 /* add_new_disk() - initiates a disk add
932  * However, if this fails before writing md_update_sb(),
933  * add_new_disk_cancel() must be called to release token lock
934  */
add_new_disk(struct mddev * mddev,struct md_rdev * rdev)935 static int add_new_disk(struct mddev *mddev, struct md_rdev *rdev)
936 {
937 	struct md_cluster_info *cinfo = mddev->cluster_info;
938 	struct cluster_msg cmsg;
939 	int ret = 0;
940 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
941 	char *uuid = sb->device_uuid;
942 
943 	memset(&cmsg, 0, sizeof(cmsg));
944 	cmsg.type = cpu_to_le32(NEWDISK);
945 	memcpy(cmsg.uuid, uuid, 16);
946 	cmsg.raid_slot = cpu_to_le32(rdev->desc_nr);
947 	lock_comm(cinfo);
948 	ret = __sendmsg(cinfo, &cmsg);
949 	if (ret) {
950 		unlock_comm(cinfo);
951 		return ret;
952 	}
953 	cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE;
954 	ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX);
955 	cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE;
956 	/* Some node does not "see" the device */
957 	if (ret == -EAGAIN)
958 		ret = -ENOENT;
959 	if (ret)
960 		unlock_comm(cinfo);
961 	else
962 		dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR);
963 	return ret;
964 }
965 
add_new_disk_cancel(struct mddev * mddev)966 static void add_new_disk_cancel(struct mddev *mddev)
967 {
968 	struct md_cluster_info *cinfo = mddev->cluster_info;
969 	unlock_comm(cinfo);
970 }
971 
new_disk_ack(struct mddev * mddev,bool ack)972 static int new_disk_ack(struct mddev *mddev, bool ack)
973 {
974 	struct md_cluster_info *cinfo = mddev->cluster_info;
975 
976 	if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) {
977 		pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev));
978 		return -EINVAL;
979 	}
980 
981 	if (ack)
982 		dlm_unlock_sync(cinfo->no_new_dev_lockres);
983 	complete(&cinfo->newdisk_completion);
984 	return 0;
985 }
986 
remove_disk(struct mddev * mddev,struct md_rdev * rdev)987 static int remove_disk(struct mddev *mddev, struct md_rdev *rdev)
988 {
989 	struct cluster_msg cmsg = {0};
990 	struct md_cluster_info *cinfo = mddev->cluster_info;
991 	cmsg.type = cpu_to_le32(REMOVE);
992 	cmsg.raid_slot = cpu_to_le32(rdev->desc_nr);
993 	return __sendmsg(cinfo, &cmsg);
994 }
995 
gather_bitmaps(struct md_rdev * rdev)996 static int gather_bitmaps(struct md_rdev *rdev)
997 {
998 	int sn, err;
999 	sector_t lo, hi;
1000 	struct cluster_msg cmsg = {0};
1001 	struct mddev *mddev = rdev->mddev;
1002 	struct md_cluster_info *cinfo = mddev->cluster_info;
1003 
1004 	cmsg.type = cpu_to_le32(RE_ADD);
1005 	cmsg.raid_slot = cpu_to_le32(rdev->desc_nr);
1006 	err = sendmsg(cinfo, &cmsg);
1007 	if (err)
1008 		goto out;
1009 
1010 	for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) {
1011 		if (sn == (cinfo->slot_number - 1))
1012 			continue;
1013 		err = bitmap_copy_from_slot(mddev, sn, &lo, &hi, false);
1014 		if (err) {
1015 			pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn);
1016 			goto out;
1017 		}
1018 		if ((hi > 0) && (lo < mddev->recovery_cp))
1019 			mddev->recovery_cp = lo;
1020 	}
1021 out:
1022 	return err;
1023 }
1024 
1025 static struct md_cluster_operations cluster_ops = {
1026 	.join   = join,
1027 	.leave  = leave,
1028 	.slot_number = slot_number,
1029 	.resync_start = resync_start,
1030 	.resync_finish = resync_finish,
1031 	.resync_info_update = resync_info_update,
1032 	.metadata_update_start = metadata_update_start,
1033 	.metadata_update_finish = metadata_update_finish,
1034 	.metadata_update_cancel = metadata_update_cancel,
1035 	.area_resyncing = area_resyncing,
1036 	.add_new_disk = add_new_disk,
1037 	.add_new_disk_cancel = add_new_disk_cancel,
1038 	.new_disk_ack = new_disk_ack,
1039 	.remove_disk = remove_disk,
1040 	.gather_bitmaps = gather_bitmaps,
1041 };
1042 
cluster_init(void)1043 static int __init cluster_init(void)
1044 {
1045 	pr_warn("md-cluster: EXPERIMENTAL. Use with caution\n");
1046 	pr_info("Registering Cluster MD functions\n");
1047 	register_md_cluster_operations(&cluster_ops, THIS_MODULE);
1048 	return 0;
1049 }
1050 
cluster_exit(void)1051 static void cluster_exit(void)
1052 {
1053 	unregister_md_cluster_operations();
1054 }
1055 
1056 module_init(cluster_init);
1057 module_exit(cluster_exit);
1058 MODULE_AUTHOR("SUSE");
1059 MODULE_LICENSE("GPL");
1060 MODULE_DESCRIPTION("Clustering support for MD");
1061