]>
Commit | Line | Data |
---|---|---|
8e854e9c GR |
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> | |
47741b7c GR |
13 | #include <linux/dlm.h> |
14 | #include <linux/sched.h> | |
1aee41f6 | 15 | #include <linux/raid/md_p.h> |
47741b7c | 16 | #include "md.h" |
e94987db | 17 | #include "bitmap.h" |
edb39c9d | 18 | #include "md-cluster.h" |
47741b7c GR |
19 | |
20 | #define LVB_SIZE 64 | |
1aee41f6 | 21 | #define NEW_DEV_TIMEOUT 5000 |
47741b7c GR |
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() */ | |
47741b7c | 28 | struct completion completion; /* completion for synchronized locking */ |
c4ce867f GR |
29 | void (*bast)(void *arg, int mode); /* blocking AST function pointer*/ |
30 | struct mddev *mddev; /* pointing back to mddev. */ | |
31 | }; | |
32 | ||
96ae923a GR |
33 | struct suspend_info { |
34 | int slot; | |
35 | sector_t lo; | |
36 | sector_t hi; | |
37 | struct list_head list; | |
38 | }; | |
39 | ||
40 | struct resync_info { | |
41 | __le64 lo; | |
42 | __le64 hi; | |
43 | }; | |
44 | ||
fa8259da GR |
45 | /* md_cluster_info flags */ |
46 | #define MD_CLUSTER_WAITING_FOR_NEWDISK 1 | |
90382ed9 | 47 | #define MD_CLUSTER_SUSPEND_READ_BALANCING 2 |
fa8259da GR |
48 | |
49 | ||
c4ce867f GR |
50 | struct md_cluster_info { |
51 | /* dlm lock space and resources for clustered raid. */ | |
52 | dlm_lockspace_t *lockspace; | |
cf921cc1 GR |
53 | int slot_number; |
54 | struct completion completion; | |
c4ce867f | 55 | struct mutex sb_mutex; |
54519c5f | 56 | struct dlm_lock_resource *bitmap_lockres; |
96ae923a GR |
57 | struct list_head suspend_list; |
58 | spinlock_t suspend_lock; | |
e94987db GR |
59 | struct md_thread *recovery_thread; |
60 | unsigned long recovery_map; | |
4664680c GR |
61 | /* communication loc resources */ |
62 | struct dlm_lock_resource *ack_lockres; | |
63 | struct dlm_lock_resource *message_lockres; | |
64 | struct dlm_lock_resource *token_lockres; | |
1aee41f6 | 65 | struct dlm_lock_resource *no_new_dev_lockres; |
4664680c | 66 | struct md_thread *recv_thread; |
1aee41f6 | 67 | struct completion newdisk_completion; |
fa8259da | 68 | unsigned long state; |
4664680c GR |
69 | }; |
70 | ||
71 | enum msg_type { | |
72 | METADATA_UPDATED = 0, | |
73 | RESYNCING, | |
1aee41f6 | 74 | NEWDISK, |
88bcfef7 | 75 | REMOVE, |
97f6cd39 | 76 | RE_ADD, |
dc737d7c | 77 | BITMAP_NEEDS_SYNC, |
4664680c GR |
78 | }; |
79 | ||
80 | struct cluster_msg { | |
81 | int type; | |
82 | int slot; | |
1aee41f6 | 83 | /* TODO: Unionize this for smaller footprint */ |
4664680c GR |
84 | sector_t low; |
85 | sector_t high; | |
1aee41f6 GR |
86 | char uuid[16]; |
87 | int raid_slot; | |
47741b7c GR |
88 | }; |
89 | ||
90 | static void sync_ast(void *arg) | |
91 | { | |
92 | struct dlm_lock_resource *res; | |
93 | ||
94 | res = (struct dlm_lock_resource *) arg; | |
95 | complete(&res->completion); | |
96 | } | |
97 | ||
98 | static int dlm_lock_sync(struct dlm_lock_resource *res, int mode) | |
99 | { | |
100 | int ret = 0; | |
101 | ||
47741b7c GR |
102 | ret = dlm_lock(res->ls, mode, &res->lksb, |
103 | res->flags, res->name, strlen(res->name), | |
104 | 0, sync_ast, res, res->bast); | |
105 | if (ret) | |
106 | return ret; | |
107 | wait_for_completion(&res->completion); | |
108 | return res->lksb.sb_status; | |
109 | } | |
110 | ||
111 | static int dlm_unlock_sync(struct dlm_lock_resource *res) | |
112 | { | |
113 | return dlm_lock_sync(res, DLM_LOCK_NL); | |
114 | } | |
115 | ||
c4ce867f | 116 | static struct dlm_lock_resource *lockres_init(struct mddev *mddev, |
47741b7c GR |
117 | char *name, void (*bastfn)(void *arg, int mode), int with_lvb) |
118 | { | |
119 | struct dlm_lock_resource *res = NULL; | |
120 | int ret, namelen; | |
c4ce867f | 121 | struct md_cluster_info *cinfo = mddev->cluster_info; |
47741b7c GR |
122 | |
123 | res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL); | |
124 | if (!res) | |
125 | return NULL; | |
b83d51c0 | 126 | init_completion(&res->completion); |
c4ce867f GR |
127 | res->ls = cinfo->lockspace; |
128 | res->mddev = mddev; | |
47741b7c GR |
129 | namelen = strlen(name); |
130 | res->name = kzalloc(namelen + 1, GFP_KERNEL); | |
131 | if (!res->name) { | |
132 | pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name); | |
133 | goto out_err; | |
134 | } | |
135 | strlcpy(res->name, name, namelen + 1); | |
136 | if (with_lvb) { | |
137 | res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL); | |
138 | if (!res->lksb.sb_lvbptr) { | |
139 | pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name); | |
140 | goto out_err; | |
141 | } | |
142 | res->flags = DLM_LKF_VALBLK; | |
143 | } | |
144 | ||
145 | if (bastfn) | |
146 | res->bast = bastfn; | |
147 | ||
148 | res->flags |= DLM_LKF_EXPEDITE; | |
149 | ||
150 | ret = dlm_lock_sync(res, DLM_LOCK_NL); | |
151 | if (ret) { | |
152 | pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name); | |
153 | goto out_err; | |
154 | } | |
155 | res->flags &= ~DLM_LKF_EXPEDITE; | |
156 | res->flags |= DLM_LKF_CONVERT; | |
157 | ||
158 | return res; | |
159 | out_err: | |
160 | kfree(res->lksb.sb_lvbptr); | |
161 | kfree(res->name); | |
162 | kfree(res); | |
163 | return NULL; | |
164 | } | |
165 | ||
166 | static void lockres_free(struct dlm_lock_resource *res) | |
167 | { | |
b5ef5678 GJ |
168 | int ret; |
169 | ||
47741b7c GR |
170 | if (!res) |
171 | return; | |
172 | ||
b5ef5678 GJ |
173 | /* cancel a lock request or a conversion request that is blocked */ |
174 | res->flags |= DLM_LKF_CANCEL; | |
175 | retry: | |
176 | ret = dlm_unlock(res->ls, res->lksb.sb_lkid, 0, &res->lksb, res); | |
177 | if (unlikely(ret != 0)) { | |
178 | pr_info("%s: failed to unlock %s return %d\n", __func__, res->name, ret); | |
179 | ||
180 | /* if a lock conversion is cancelled, then the lock is put | |
181 | * back to grant queue, need to ensure it is unlocked */ | |
182 | if (ret == -DLM_ECANCEL) | |
183 | goto retry; | |
184 | } | |
185 | res->flags &= ~DLM_LKF_CANCEL; | |
47741b7c GR |
186 | wait_for_completion(&res->completion); |
187 | ||
188 | kfree(res->name); | |
189 | kfree(res->lksb.sb_lvbptr); | |
190 | kfree(res); | |
191 | } | |
8e854e9c | 192 | |
96ae923a GR |
193 | static void add_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres, |
194 | sector_t lo, sector_t hi) | |
195 | { | |
196 | struct resync_info *ri; | |
197 | ||
198 | ri = (struct resync_info *)lockres->lksb.sb_lvbptr; | |
199 | ri->lo = cpu_to_le64(lo); | |
200 | ri->hi = cpu_to_le64(hi); | |
201 | } | |
202 | ||
203 | static struct suspend_info *read_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres) | |
204 | { | |
205 | struct resync_info ri; | |
206 | struct suspend_info *s = NULL; | |
207 | sector_t hi = 0; | |
208 | ||
209 | dlm_lock_sync(lockres, DLM_LOCK_CR); | |
210 | memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info)); | |
211 | hi = le64_to_cpu(ri.hi); | |
212 | if (ri.hi > 0) { | |
213 | s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL); | |
214 | if (!s) | |
215 | goto out; | |
216 | s->hi = hi; | |
217 | s->lo = le64_to_cpu(ri.lo); | |
218 | } | |
219 | dlm_unlock_sync(lockres); | |
220 | out: | |
221 | return s; | |
222 | } | |
223 | ||
6dc69c9c | 224 | static void recover_bitmaps(struct md_thread *thread) |
e94987db GR |
225 | { |
226 | struct mddev *mddev = thread->mddev; | |
227 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
228 | struct dlm_lock_resource *bm_lockres; | |
229 | char str[64]; | |
230 | int slot, ret; | |
231 | struct suspend_info *s, *tmp; | |
232 | sector_t lo, hi; | |
233 | ||
234 | while (cinfo->recovery_map) { | |
235 | slot = fls64((u64)cinfo->recovery_map) - 1; | |
236 | ||
237 | /* Clear suspend_area associated with the bitmap */ | |
238 | spin_lock_irq(&cinfo->suspend_lock); | |
239 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
240 | if (slot == s->slot) { | |
241 | list_del(&s->list); | |
242 | kfree(s); | |
243 | } | |
244 | spin_unlock_irq(&cinfo->suspend_lock); | |
245 | ||
246 | snprintf(str, 64, "bitmap%04d", slot); | |
247 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
248 | if (!bm_lockres) { | |
249 | pr_err("md-cluster: Cannot initialize bitmaps\n"); | |
250 | goto clear_bit; | |
251 | } | |
252 | ||
253 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
254 | if (ret) { | |
255 | pr_err("md-cluster: Could not DLM lock %s: %d\n", | |
256 | str, ret); | |
257 | goto clear_bit; | |
258 | } | |
97f6cd39 | 259 | ret = bitmap_copy_from_slot(mddev, slot, &lo, &hi, true); |
4b26a08a | 260 | if (ret) { |
e94987db | 261 | pr_err("md-cluster: Could not copy data from bitmap %d\n", slot); |
4b26a08a GR |
262 | goto dlm_unlock; |
263 | } | |
264 | if (hi > 0) { | |
265 | /* TODO:Wait for current resync to get over */ | |
266 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
267 | if (lo < mddev->recovery_cp) | |
268 | mddev->recovery_cp = lo; | |
269 | md_check_recovery(mddev); | |
270 | } | |
271 | dlm_unlock: | |
e94987db GR |
272 | dlm_unlock_sync(bm_lockres); |
273 | clear_bit: | |
274 | clear_bit(slot, &cinfo->recovery_map); | |
275 | } | |
276 | } | |
277 | ||
cf921cc1 GR |
278 | static void recover_prep(void *arg) |
279 | { | |
90382ed9 GR |
280 | struct mddev *mddev = arg; |
281 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
282 | set_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); | |
cf921cc1 GR |
283 | } |
284 | ||
05cd0e51 | 285 | static void __recover_slot(struct mddev *mddev, int slot) |
cf921cc1 | 286 | { |
cf921cc1 GR |
287 | struct md_cluster_info *cinfo = mddev->cluster_info; |
288 | ||
05cd0e51 | 289 | set_bit(slot, &cinfo->recovery_map); |
e94987db GR |
290 | if (!cinfo->recovery_thread) { |
291 | cinfo->recovery_thread = md_register_thread(recover_bitmaps, | |
292 | mddev, "recover"); | |
293 | if (!cinfo->recovery_thread) { | |
294 | pr_warn("md-cluster: Could not create recovery thread\n"); | |
295 | return; | |
296 | } | |
297 | } | |
298 | md_wakeup_thread(cinfo->recovery_thread); | |
cf921cc1 GR |
299 | } |
300 | ||
05cd0e51 GJ |
301 | static void recover_slot(void *arg, struct dlm_slot *slot) |
302 | { | |
303 | struct mddev *mddev = arg; | |
304 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
305 | ||
306 | pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n", | |
307 | mddev->bitmap_info.cluster_name, | |
308 | slot->nodeid, slot->slot, | |
309 | cinfo->slot_number); | |
310 | /* deduct one since dlm slot starts from one while the num of | |
311 | * cluster-md begins with 0 */ | |
312 | __recover_slot(mddev, slot->slot - 1); | |
313 | } | |
314 | ||
cf921cc1 GR |
315 | static void recover_done(void *arg, struct dlm_slot *slots, |
316 | int num_slots, int our_slot, | |
317 | uint32_t generation) | |
318 | { | |
319 | struct mddev *mddev = arg; | |
320 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
321 | ||
322 | cinfo->slot_number = our_slot; | |
323 | complete(&cinfo->completion); | |
90382ed9 | 324 | clear_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); |
cf921cc1 GR |
325 | } |
326 | ||
327 | static const struct dlm_lockspace_ops md_ls_ops = { | |
328 | .recover_prep = recover_prep, | |
329 | .recover_slot = recover_slot, | |
330 | .recover_done = recover_done, | |
331 | }; | |
332 | ||
4664680c GR |
333 | /* |
334 | * The BAST function for the ack lock resource | |
335 | * This function wakes up the receive thread in | |
336 | * order to receive and process the message. | |
337 | */ | |
338 | static void ack_bast(void *arg, int mode) | |
339 | { | |
340 | struct dlm_lock_resource *res = (struct dlm_lock_resource *)arg; | |
341 | struct md_cluster_info *cinfo = res->mddev->cluster_info; | |
342 | ||
343 | if (mode == DLM_LOCK_EX) | |
344 | md_wakeup_thread(cinfo->recv_thread); | |
345 | } | |
346 | ||
e59721cc GR |
347 | static void __remove_suspend_info(struct md_cluster_info *cinfo, int slot) |
348 | { | |
349 | struct suspend_info *s, *tmp; | |
350 | ||
351 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
352 | if (slot == s->slot) { | |
353 | pr_info("%s:%d Deleting suspend_info: %d\n", | |
354 | __func__, __LINE__, slot); | |
355 | list_del(&s->list); | |
356 | kfree(s); | |
357 | break; | |
358 | } | |
359 | } | |
360 | ||
361 | static void remove_suspend_info(struct md_cluster_info *cinfo, int slot) | |
362 | { | |
363 | spin_lock_irq(&cinfo->suspend_lock); | |
364 | __remove_suspend_info(cinfo, slot); | |
365 | spin_unlock_irq(&cinfo->suspend_lock); | |
366 | } | |
367 | ||
368 | ||
369 | static void process_suspend_info(struct md_cluster_info *cinfo, | |
370 | int slot, sector_t lo, sector_t hi) | |
371 | { | |
372 | struct suspend_info *s; | |
373 | ||
374 | if (!hi) { | |
375 | remove_suspend_info(cinfo, slot); | |
376 | return; | |
377 | } | |
378 | s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL); | |
379 | if (!s) | |
380 | return; | |
381 | s->slot = slot; | |
382 | s->lo = lo; | |
383 | s->hi = hi; | |
384 | spin_lock_irq(&cinfo->suspend_lock); | |
385 | /* Remove existing entry (if exists) before adding */ | |
386 | __remove_suspend_info(cinfo, slot); | |
387 | list_add(&s->list, &cinfo->suspend_list); | |
388 | spin_unlock_irq(&cinfo->suspend_lock); | |
389 | } | |
390 | ||
1aee41f6 GR |
391 | static void process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg) |
392 | { | |
393 | char disk_uuid[64]; | |
394 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
395 | char event_name[] = "EVENT=ADD_DEVICE"; | |
396 | char raid_slot[16]; | |
397 | char *envp[] = {event_name, disk_uuid, raid_slot, NULL}; | |
398 | int len; | |
399 | ||
400 | len = snprintf(disk_uuid, 64, "DEVICE_UUID="); | |
b89f704a | 401 | sprintf(disk_uuid + len, "%pU", cmsg->uuid); |
1aee41f6 GR |
402 | snprintf(raid_slot, 16, "RAID_DISK=%d", cmsg->raid_slot); |
403 | pr_info("%s:%d Sending kobject change with %s and %s\n", __func__, __LINE__, disk_uuid, raid_slot); | |
404 | init_completion(&cinfo->newdisk_completion); | |
fa8259da | 405 | set_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
406 | kobject_uevent_env(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE, envp); |
407 | wait_for_completion_timeout(&cinfo->newdisk_completion, | |
408 | NEW_DEV_TIMEOUT); | |
fa8259da | 409 | clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
410 | } |
411 | ||
412 | ||
413 | static void process_metadata_update(struct mddev *mddev, struct cluster_msg *msg) | |
414 | { | |
415 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
416 | ||
417 | md_reload_sb(mddev); | |
418 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); | |
419 | } | |
420 | ||
88bcfef7 GR |
421 | static void process_remove_disk(struct mddev *mddev, struct cluster_msg *msg) |
422 | { | |
423 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, msg->raid_slot); | |
424 | ||
425 | if (rdev) | |
426 | md_kick_rdev_from_array(rdev); | |
427 | else | |
428 | pr_warn("%s: %d Could not find disk(%d) to REMOVE\n", __func__, __LINE__, msg->raid_slot); | |
429 | } | |
430 | ||
97f6cd39 GR |
431 | static void process_readd_disk(struct mddev *mddev, struct cluster_msg *msg) |
432 | { | |
433 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, msg->raid_slot); | |
434 | ||
435 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
436 | clear_bit(Faulty, &rdev->flags); | |
437 | else | |
438 | pr_warn("%s: %d Could not find disk(%d) which is faulty", __func__, __LINE__, msg->raid_slot); | |
439 | } | |
440 | ||
4664680c GR |
441 | static void process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg) |
442 | { | |
443 | switch (msg->type) { | |
444 | case METADATA_UPDATED: | |
445 | pr_info("%s: %d Received message: METADATA_UPDATE from %d\n", | |
446 | __func__, __LINE__, msg->slot); | |
1aee41f6 | 447 | process_metadata_update(mddev, msg); |
4664680c GR |
448 | break; |
449 | case RESYNCING: | |
450 | pr_info("%s: %d Received message: RESYNCING from %d\n", | |
451 | __func__, __LINE__, msg->slot); | |
e59721cc GR |
452 | process_suspend_info(mddev->cluster_info, msg->slot, |
453 | msg->low, msg->high); | |
4664680c | 454 | break; |
1aee41f6 GR |
455 | case NEWDISK: |
456 | pr_info("%s: %d Received message: NEWDISK from %d\n", | |
457 | __func__, __LINE__, msg->slot); | |
458 | process_add_new_disk(mddev, msg); | |
88bcfef7 GR |
459 | break; |
460 | case REMOVE: | |
461 | pr_info("%s: %d Received REMOVE from %d\n", | |
462 | __func__, __LINE__, msg->slot); | |
463 | process_remove_disk(mddev, msg); | |
464 | break; | |
97f6cd39 GR |
465 | case RE_ADD: |
466 | pr_info("%s: %d Received RE_ADD from %d\n", | |
467 | __func__, __LINE__, msg->slot); | |
468 | process_readd_disk(mddev, msg); | |
469 | break; | |
dc737d7c GJ |
470 | case BITMAP_NEEDS_SYNC: |
471 | pr_info("%s: %d Received BITMAP_NEEDS_SYNC from %d\n", | |
472 | __func__, __LINE__, msg->slot); | |
473 | __recover_slot(mddev, msg->slot); | |
474 | break; | |
88bcfef7 GR |
475 | default: |
476 | pr_warn("%s:%d Received unknown message from %d\n", | |
477 | __func__, __LINE__, msg->slot); | |
09dd1af2 | 478 | } |
4664680c GR |
479 | } |
480 | ||
481 | /* | |
482 | * thread for receiving message | |
483 | */ | |
484 | static void recv_daemon(struct md_thread *thread) | |
485 | { | |
486 | struct md_cluster_info *cinfo = thread->mddev->cluster_info; | |
487 | struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres; | |
488 | struct dlm_lock_resource *message_lockres = cinfo->message_lockres; | |
489 | struct cluster_msg msg; | |
b5ef5678 | 490 | int ret; |
4664680c GR |
491 | |
492 | /*get CR on Message*/ | |
493 | if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) { | |
494 | pr_err("md/raid1:failed to get CR on MESSAGE\n"); | |
495 | return; | |
496 | } | |
497 | ||
498 | /* read lvb and wake up thread to process this message_lockres */ | |
499 | memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg)); | |
500 | process_recvd_msg(thread->mddev, &msg); | |
501 | ||
502 | /*release CR on ack_lockres*/ | |
b5ef5678 GJ |
503 | ret = dlm_unlock_sync(ack_lockres); |
504 | if (unlikely(ret != 0)) | |
505 | pr_info("unlock ack failed return %d\n", ret); | |
66099bb0 | 506 | /*up-convert to PR on message_lockres*/ |
b5ef5678 GJ |
507 | ret = dlm_lock_sync(message_lockres, DLM_LOCK_PR); |
508 | if (unlikely(ret != 0)) | |
509 | pr_info("lock PR on msg failed return %d\n", ret); | |
4664680c | 510 | /*get CR on ack_lockres again*/ |
b5ef5678 GJ |
511 | ret = dlm_lock_sync(ack_lockres, DLM_LOCK_CR); |
512 | if (unlikely(ret != 0)) | |
513 | pr_info("lock CR on ack failed return %d\n", ret); | |
4664680c | 514 | /*release CR on message_lockres*/ |
b5ef5678 GJ |
515 | ret = dlm_unlock_sync(message_lockres); |
516 | if (unlikely(ret != 0)) | |
517 | pr_info("unlock msg failed return %d\n", ret); | |
4664680c GR |
518 | } |
519 | ||
601b515c GR |
520 | /* lock_comm() |
521 | * Takes the lock on the TOKEN lock resource so no other | |
522 | * node can communicate while the operation is underway. | |
523 | */ | |
524 | static int lock_comm(struct md_cluster_info *cinfo) | |
525 | { | |
526 | int error; | |
527 | ||
528 | error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX); | |
529 | if (error) | |
530 | pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n", | |
531 | __func__, __LINE__, error); | |
532 | return error; | |
533 | } | |
534 | ||
535 | static void unlock_comm(struct md_cluster_info *cinfo) | |
536 | { | |
537 | dlm_unlock_sync(cinfo->token_lockres); | |
538 | } | |
539 | ||
540 | /* __sendmsg() | |
541 | * This function performs the actual sending of the message. This function is | |
542 | * usually called after performing the encompassing operation | |
543 | * The function: | |
544 | * 1. Grabs the message lockresource in EX mode | |
545 | * 2. Copies the message to the message LVB | |
66099bb0 | 546 | * 3. Downconverts message lockresource to CW |
601b515c GR |
547 | * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes |
548 | * and the other nodes read the message. The thread will wait here until all other | |
549 | * nodes have released ack lock resource. | |
550 | * 5. Downconvert ack lockresource to CR | |
551 | */ | |
552 | static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
553 | { | |
554 | int error; | |
555 | int slot = cinfo->slot_number - 1; | |
556 | ||
557 | cmsg->slot = cpu_to_le32(slot); | |
558 | /*get EX on Message*/ | |
559 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX); | |
560 | if (error) { | |
561 | pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error); | |
562 | goto failed_message; | |
563 | } | |
564 | ||
565 | memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg, | |
566 | sizeof(struct cluster_msg)); | |
66099bb0 GJ |
567 | /*down-convert EX to CW on Message*/ |
568 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CW); | |
601b515c | 569 | if (error) { |
66099bb0 | 570 | pr_err("md-cluster: failed to convert EX to CW on MESSAGE(%d)\n", |
601b515c | 571 | error); |
66099bb0 | 572 | goto failed_ack; |
601b515c GR |
573 | } |
574 | ||
575 | /*up-convert CR to EX on Ack*/ | |
576 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX); | |
577 | if (error) { | |
578 | pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n", | |
579 | error); | |
580 | goto failed_ack; | |
581 | } | |
582 | ||
583 | /*down-convert EX to CR on Ack*/ | |
584 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR); | |
585 | if (error) { | |
586 | pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n", | |
587 | error); | |
588 | goto failed_ack; | |
589 | } | |
590 | ||
591 | failed_ack: | |
b5ef5678 GJ |
592 | error = dlm_unlock_sync(cinfo->message_lockres); |
593 | if (unlikely(error != 0)) { | |
594 | pr_err("md-cluster: failed convert to NL on MESSAGE(%d)\n", | |
595 | error); | |
596 | /* in case the message can't be released due to some reason */ | |
597 | goto failed_ack; | |
598 | } | |
601b515c GR |
599 | failed_message: |
600 | return error; | |
601 | } | |
602 | ||
603 | static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
604 | { | |
605 | int ret; | |
606 | ||
607 | lock_comm(cinfo); | |
608 | ret = __sendmsg(cinfo, cmsg); | |
609 | unlock_comm(cinfo); | |
610 | return ret; | |
611 | } | |
612 | ||
96ae923a GR |
613 | static int gather_all_resync_info(struct mddev *mddev, int total_slots) |
614 | { | |
615 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
616 | int i, ret = 0; | |
617 | struct dlm_lock_resource *bm_lockres; | |
618 | struct suspend_info *s; | |
619 | char str[64]; | |
620 | ||
621 | ||
622 | for (i = 0; i < total_slots; i++) { | |
623 | memset(str, '\0', 64); | |
624 | snprintf(str, 64, "bitmap%04d", i); | |
625 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
626 | if (!bm_lockres) | |
627 | return -ENOMEM; | |
628 | if (i == (cinfo->slot_number - 1)) | |
629 | continue; | |
630 | ||
631 | bm_lockres->flags |= DLM_LKF_NOQUEUE; | |
632 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
633 | if (ret == -EAGAIN) { | |
634 | memset(bm_lockres->lksb.sb_lvbptr, '\0', LVB_SIZE); | |
635 | s = read_resync_info(mddev, bm_lockres); | |
636 | if (s) { | |
637 | pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n", | |
638 | __func__, __LINE__, | |
639 | (unsigned long long) s->lo, | |
640 | (unsigned long long) s->hi, i); | |
641 | spin_lock_irq(&cinfo->suspend_lock); | |
642 | s->slot = i; | |
643 | list_add(&s->list, &cinfo->suspend_list); | |
644 | spin_unlock_irq(&cinfo->suspend_lock); | |
645 | } | |
646 | ret = 0; | |
647 | lockres_free(bm_lockres); | |
648 | continue; | |
649 | } | |
6e6d9f2c GJ |
650 | if (ret) { |
651 | lockres_free(bm_lockres); | |
96ae923a | 652 | goto out; |
6e6d9f2c | 653 | } |
96ae923a GR |
654 | /* TODO: Read the disk bitmap sb and check if it needs recovery */ |
655 | dlm_unlock_sync(bm_lockres); | |
656 | lockres_free(bm_lockres); | |
657 | } | |
658 | out: | |
659 | return ret; | |
660 | } | |
661 | ||
edb39c9d GR |
662 | static int join(struct mddev *mddev, int nodes) |
663 | { | |
c4ce867f | 664 | struct md_cluster_info *cinfo; |
cf921cc1 | 665 | int ret, ops_rv; |
c4ce867f GR |
666 | char str[64]; |
667 | ||
668 | if (!try_module_get(THIS_MODULE)) | |
669 | return -ENOENT; | |
670 | ||
671 | cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL); | |
672 | if (!cinfo) | |
673 | return -ENOMEM; | |
674 | ||
9e3072e3 GJ |
675 | INIT_LIST_HEAD(&cinfo->suspend_list); |
676 | spin_lock_init(&cinfo->suspend_lock); | |
cf921cc1 GR |
677 | init_completion(&cinfo->completion); |
678 | ||
679 | mutex_init(&cinfo->sb_mutex); | |
680 | mddev->cluster_info = cinfo; | |
681 | ||
c4ce867f | 682 | memset(str, 0, 64); |
b89f704a | 683 | sprintf(str, "%pU", mddev->uuid); |
cf921cc1 GR |
684 | ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name, |
685 | DLM_LSFL_FS, LVB_SIZE, | |
686 | &md_ls_ops, mddev, &ops_rv, &cinfo->lockspace); | |
c4ce867f GR |
687 | if (ret) |
688 | goto err; | |
cf921cc1 | 689 | wait_for_completion(&cinfo->completion); |
8c58f02e GJ |
690 | if (nodes < cinfo->slot_number) { |
691 | pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).", | |
692 | cinfo->slot_number, nodes); | |
b97e9257 GR |
693 | ret = -ERANGE; |
694 | goto err; | |
695 | } | |
4664680c GR |
696 | /* Initiate the communication resources */ |
697 | ret = -ENOMEM; | |
698 | cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv"); | |
699 | if (!cinfo->recv_thread) { | |
700 | pr_err("md-cluster: cannot allocate memory for recv_thread!\n"); | |
701 | goto err; | |
702 | } | |
703 | cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1); | |
704 | if (!cinfo->message_lockres) | |
705 | goto err; | |
706 | cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0); | |
707 | if (!cinfo->token_lockres) | |
708 | goto err; | |
709 | cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0); | |
710 | if (!cinfo->ack_lockres) | |
711 | goto err; | |
1aee41f6 GR |
712 | cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0); |
713 | if (!cinfo->no_new_dev_lockres) | |
714 | goto err; | |
715 | ||
4664680c GR |
716 | /* get sync CR lock on ACK. */ |
717 | if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR)) | |
718 | pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n", | |
719 | ret); | |
1aee41f6 GR |
720 | /* get sync CR lock on no-new-dev. */ |
721 | if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR)) | |
722 | pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret); | |
723 | ||
54519c5f GR |
724 | |
725 | pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number); | |
726 | snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1); | |
727 | cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1); | |
728 | if (!cinfo->bitmap_lockres) | |
729 | goto err; | |
730 | if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) { | |
731 | pr_err("Failed to get bitmap lock\n"); | |
732 | ret = -EINVAL; | |
733 | goto err; | |
734 | } | |
735 | ||
96ae923a GR |
736 | ret = gather_all_resync_info(mddev, nodes); |
737 | if (ret) | |
738 | goto err; | |
739 | ||
edb39c9d | 740 | return 0; |
c4ce867f | 741 | err: |
4664680c GR |
742 | lockres_free(cinfo->message_lockres); |
743 | lockres_free(cinfo->token_lockres); | |
744 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 745 | lockres_free(cinfo->no_new_dev_lockres); |
96ae923a | 746 | lockres_free(cinfo->bitmap_lockres); |
c4ce867f GR |
747 | if (cinfo->lockspace) |
748 | dlm_release_lockspace(cinfo->lockspace, 2); | |
cf921cc1 | 749 | mddev->cluster_info = NULL; |
c4ce867f GR |
750 | kfree(cinfo); |
751 | module_put(THIS_MODULE); | |
752 | return ret; | |
edb39c9d GR |
753 | } |
754 | ||
755 | static int leave(struct mddev *mddev) | |
756 | { | |
c4ce867f GR |
757 | struct md_cluster_info *cinfo = mddev->cluster_info; |
758 | ||
759 | if (!cinfo) | |
760 | return 0; | |
e94987db | 761 | md_unregister_thread(&cinfo->recovery_thread); |
4664680c GR |
762 | md_unregister_thread(&cinfo->recv_thread); |
763 | lockres_free(cinfo->message_lockres); | |
764 | lockres_free(cinfo->token_lockres); | |
765 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 766 | lockres_free(cinfo->no_new_dev_lockres); |
54519c5f | 767 | lockres_free(cinfo->bitmap_lockres); |
c4ce867f | 768 | dlm_release_lockspace(cinfo->lockspace, 2); |
edb39c9d GR |
769 | return 0; |
770 | } | |
771 | ||
cf921cc1 GR |
772 | /* slot_number(): Returns the MD slot number to use |
773 | * DLM starts the slot numbers from 1, wheras cluster-md | |
774 | * wants the number to be from zero, so we deduct one | |
775 | */ | |
776 | static int slot_number(struct mddev *mddev) | |
777 | { | |
778 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
779 | ||
780 | return cinfo->slot_number - 1; | |
781 | } | |
782 | ||
96ae923a GR |
783 | static void resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi) |
784 | { | |
785 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
786 | ||
787 | add_resync_info(mddev, cinfo->bitmap_lockres, lo, hi); | |
788 | /* Re-acquire the lock to refresh LVB */ | |
789 | dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW); | |
790 | } | |
791 | ||
293467aa GR |
792 | static int metadata_update_start(struct mddev *mddev) |
793 | { | |
794 | return lock_comm(mddev->cluster_info); | |
795 | } | |
796 | ||
797 | static int metadata_update_finish(struct mddev *mddev) | |
798 | { | |
799 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
800 | struct cluster_msg cmsg; | |
801 | int ret; | |
802 | ||
803 | memset(&cmsg, 0, sizeof(cmsg)); | |
804 | cmsg.type = cpu_to_le32(METADATA_UPDATED); | |
805 | ret = __sendmsg(cinfo, &cmsg); | |
806 | unlock_comm(cinfo); | |
807 | return ret; | |
808 | } | |
809 | ||
810 | static int metadata_update_cancel(struct mddev *mddev) | |
811 | { | |
812 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
813 | ||
814 | return dlm_unlock_sync(cinfo->token_lockres); | |
815 | } | |
816 | ||
965400eb GR |
817 | static int resync_send(struct mddev *mddev, enum msg_type type, |
818 | sector_t lo, sector_t hi) | |
819 | { | |
820 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
821 | struct cluster_msg cmsg; | |
822 | int slot = cinfo->slot_number - 1; | |
823 | ||
824 | pr_info("%s:%d lo: %llu hi: %llu\n", __func__, __LINE__, | |
825 | (unsigned long long)lo, | |
826 | (unsigned long long)hi); | |
827 | resync_info_update(mddev, lo, hi); | |
828 | cmsg.type = cpu_to_le32(type); | |
829 | cmsg.slot = cpu_to_le32(slot); | |
830 | cmsg.low = cpu_to_le64(lo); | |
831 | cmsg.high = cpu_to_le64(hi); | |
832 | return sendmsg(cinfo, &cmsg); | |
833 | } | |
834 | ||
835 | static int resync_start(struct mddev *mddev, sector_t lo, sector_t hi) | |
836 | { | |
837 | pr_info("%s:%d\n", __func__, __LINE__); | |
838 | return resync_send(mddev, RESYNCING, lo, hi); | |
839 | } | |
840 | ||
841 | static void resync_finish(struct mddev *mddev) | |
842 | { | |
dc737d7c GJ |
843 | struct md_cluster_info *cinfo = mddev->cluster_info; |
844 | struct cluster_msg cmsg; | |
845 | int slot = cinfo->slot_number - 1; | |
846 | ||
965400eb GR |
847 | pr_info("%s:%d\n", __func__, __LINE__); |
848 | resync_send(mddev, RESYNCING, 0, 0); | |
dc737d7c GJ |
849 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
850 | cmsg.type = cpu_to_le32(BITMAP_NEEDS_SYNC); | |
851 | cmsg.slot = cpu_to_le32(slot); | |
852 | sendmsg(cinfo, &cmsg); | |
853 | } | |
965400eb GR |
854 | } |
855 | ||
90382ed9 GR |
856 | static int area_resyncing(struct mddev *mddev, int direction, |
857 | sector_t lo, sector_t hi) | |
589a1c49 GR |
858 | { |
859 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
860 | int ret = 0; | |
861 | struct suspend_info *s; | |
862 | ||
90382ed9 GR |
863 | if ((direction == READ) && |
864 | test_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state)) | |
865 | return 1; | |
866 | ||
589a1c49 GR |
867 | spin_lock_irq(&cinfo->suspend_lock); |
868 | if (list_empty(&cinfo->suspend_list)) | |
869 | goto out; | |
870 | list_for_each_entry(s, &cinfo->suspend_list, list) | |
871 | if (hi > s->lo && lo < s->hi) { | |
872 | ret = 1; | |
873 | break; | |
874 | } | |
875 | out: | |
876 | spin_unlock_irq(&cinfo->suspend_lock); | |
877 | return ret; | |
878 | } | |
879 | ||
1aee41f6 GR |
880 | static int add_new_disk_start(struct mddev *mddev, struct md_rdev *rdev) |
881 | { | |
882 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
883 | struct cluster_msg cmsg; | |
884 | int ret = 0; | |
885 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); | |
886 | char *uuid = sb->device_uuid; | |
887 | ||
888 | memset(&cmsg, 0, sizeof(cmsg)); | |
889 | cmsg.type = cpu_to_le32(NEWDISK); | |
890 | memcpy(cmsg.uuid, uuid, 16); | |
891 | cmsg.raid_slot = rdev->desc_nr; | |
892 | lock_comm(cinfo); | |
893 | ret = __sendmsg(cinfo, &cmsg); | |
894 | if (ret) | |
895 | return ret; | |
896 | cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE; | |
897 | ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX); | |
898 | cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE; | |
899 | /* Some node does not "see" the device */ | |
900 | if (ret == -EAGAIN) | |
901 | ret = -ENOENT; | |
902 | else | |
903 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); | |
904 | return ret; | |
905 | } | |
906 | ||
907 | static int add_new_disk_finish(struct mddev *mddev) | |
908 | { | |
909 | struct cluster_msg cmsg; | |
910 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
911 | int ret; | |
912 | /* Write sb and inform others */ | |
913 | md_update_sb(mddev, 1); | |
914 | cmsg.type = METADATA_UPDATED; | |
915 | ret = __sendmsg(cinfo, &cmsg); | |
916 | unlock_comm(cinfo); | |
917 | return ret; | |
918 | } | |
919 | ||
fa8259da | 920 | static int new_disk_ack(struct mddev *mddev, bool ack) |
1aee41f6 GR |
921 | { |
922 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
923 | ||
fa8259da GR |
924 | if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) { |
925 | pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev)); | |
926 | return -EINVAL; | |
927 | } | |
928 | ||
1aee41f6 GR |
929 | if (ack) |
930 | dlm_unlock_sync(cinfo->no_new_dev_lockres); | |
931 | complete(&cinfo->newdisk_completion); | |
fa8259da | 932 | return 0; |
1aee41f6 GR |
933 | } |
934 | ||
88bcfef7 GR |
935 | static int remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
936 | { | |
937 | struct cluster_msg cmsg; | |
938 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
939 | cmsg.type = REMOVE; | |
940 | cmsg.raid_slot = rdev->desc_nr; | |
941 | return __sendmsg(cinfo, &cmsg); | |
942 | } | |
943 | ||
97f6cd39 GR |
944 | static int gather_bitmaps(struct md_rdev *rdev) |
945 | { | |
946 | int sn, err; | |
947 | sector_t lo, hi; | |
948 | struct cluster_msg cmsg; | |
949 | struct mddev *mddev = rdev->mddev; | |
950 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
951 | ||
952 | cmsg.type = RE_ADD; | |
953 | cmsg.raid_slot = rdev->desc_nr; | |
954 | err = sendmsg(cinfo, &cmsg); | |
955 | if (err) | |
956 | goto out; | |
957 | ||
958 | for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) { | |
959 | if (sn == (cinfo->slot_number - 1)) | |
960 | continue; | |
961 | err = bitmap_copy_from_slot(mddev, sn, &lo, &hi, false); | |
962 | if (err) { | |
963 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn); | |
964 | goto out; | |
965 | } | |
966 | if ((hi > 0) && (lo < mddev->recovery_cp)) | |
967 | mddev->recovery_cp = lo; | |
968 | } | |
969 | out: | |
970 | return err; | |
971 | } | |
972 | ||
edb39c9d GR |
973 | static struct md_cluster_operations cluster_ops = { |
974 | .join = join, | |
975 | .leave = leave, | |
cf921cc1 | 976 | .slot_number = slot_number, |
96ae923a | 977 | .resync_info_update = resync_info_update, |
965400eb GR |
978 | .resync_start = resync_start, |
979 | .resync_finish = resync_finish, | |
293467aa GR |
980 | .metadata_update_start = metadata_update_start, |
981 | .metadata_update_finish = metadata_update_finish, | |
982 | .metadata_update_cancel = metadata_update_cancel, | |
589a1c49 | 983 | .area_resyncing = area_resyncing, |
1aee41f6 GR |
984 | .add_new_disk_start = add_new_disk_start, |
985 | .add_new_disk_finish = add_new_disk_finish, | |
986 | .new_disk_ack = new_disk_ack, | |
88bcfef7 | 987 | .remove_disk = remove_disk, |
97f6cd39 | 988 | .gather_bitmaps = gather_bitmaps, |
edb39c9d GR |
989 | }; |
990 | ||
8e854e9c GR |
991 | static int __init cluster_init(void) |
992 | { | |
993 | pr_warn("md-cluster: EXPERIMENTAL. Use with caution\n"); | |
994 | pr_info("Registering Cluster MD functions\n"); | |
edb39c9d | 995 | register_md_cluster_operations(&cluster_ops, THIS_MODULE); |
8e854e9c GR |
996 | return 0; |
997 | } | |
998 | ||
999 | static void cluster_exit(void) | |
1000 | { | |
edb39c9d | 1001 | unregister_md_cluster_operations(); |
8e854e9c GR |
1002 | } |
1003 | ||
1004 | module_init(cluster_init); | |
1005 | module_exit(cluster_exit); | |
1006 | MODULE_LICENSE("GPL"); | |
1007 | MODULE_DESCRIPTION("Clustering support for MD"); |