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