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 | |
47 | ||
48 | ||
c4ce867f GR |
49 | struct md_cluster_info { |
50 | /* dlm lock space and resources for clustered raid. */ | |
51 | dlm_lockspace_t *lockspace; | |
cf921cc1 GR |
52 | int slot_number; |
53 | struct completion completion; | |
c4ce867f GR |
54 | struct dlm_lock_resource *sb_lock; |
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, |
4664680c GR |
77 | }; |
78 | ||
79 | struct cluster_msg { | |
80 | int type; | |
81 | int slot; | |
1aee41f6 | 82 | /* TODO: Unionize this for smaller footprint */ |
4664680c GR |
83 | sector_t low; |
84 | sector_t high; | |
1aee41f6 GR |
85 | char uuid[16]; |
86 | int raid_slot; | |
47741b7c GR |
87 | }; |
88 | ||
89 | static void sync_ast(void *arg) | |
90 | { | |
91 | struct dlm_lock_resource *res; | |
92 | ||
93 | res = (struct dlm_lock_resource *) arg; | |
94 | complete(&res->completion); | |
95 | } | |
96 | ||
97 | static int dlm_lock_sync(struct dlm_lock_resource *res, int mode) | |
98 | { | |
99 | int ret = 0; | |
100 | ||
101 | init_completion(&res->completion); | |
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; | |
c4ce867f GR |
126 | res->ls = cinfo->lockspace; |
127 | res->mddev = mddev; | |
47741b7c GR |
128 | namelen = strlen(name); |
129 | res->name = kzalloc(namelen + 1, GFP_KERNEL); | |
130 | if (!res->name) { | |
131 | pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name); | |
132 | goto out_err; | |
133 | } | |
134 | strlcpy(res->name, name, namelen + 1); | |
135 | if (with_lvb) { | |
136 | res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL); | |
137 | if (!res->lksb.sb_lvbptr) { | |
138 | pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name); | |
139 | goto out_err; | |
140 | } | |
141 | res->flags = DLM_LKF_VALBLK; | |
142 | } | |
143 | ||
144 | if (bastfn) | |
145 | res->bast = bastfn; | |
146 | ||
147 | res->flags |= DLM_LKF_EXPEDITE; | |
148 | ||
149 | ret = dlm_lock_sync(res, DLM_LOCK_NL); | |
150 | if (ret) { | |
151 | pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name); | |
152 | goto out_err; | |
153 | } | |
154 | res->flags &= ~DLM_LKF_EXPEDITE; | |
155 | res->flags |= DLM_LKF_CONVERT; | |
156 | ||
157 | return res; | |
158 | out_err: | |
159 | kfree(res->lksb.sb_lvbptr); | |
160 | kfree(res->name); | |
161 | kfree(res); | |
162 | return NULL; | |
163 | } | |
164 | ||
165 | static void lockres_free(struct dlm_lock_resource *res) | |
166 | { | |
167 | if (!res) | |
168 | return; | |
169 | ||
170 | init_completion(&res->completion); | |
171 | dlm_unlock(res->ls, res->lksb.sb_lkid, 0, &res->lksb, res); | |
172 | wait_for_completion(&res->completion); | |
173 | ||
174 | kfree(res->name); | |
175 | kfree(res->lksb.sb_lvbptr); | |
176 | kfree(res); | |
177 | } | |
8e854e9c | 178 | |
c4ce867f GR |
179 | static char *pretty_uuid(char *dest, char *src) |
180 | { | |
181 | int i, len = 0; | |
182 | ||
183 | for (i = 0; i < 16; i++) { | |
184 | if (i == 4 || i == 6 || i == 8 || i == 10) | |
185 | len += sprintf(dest + len, "-"); | |
186 | len += sprintf(dest + len, "%02x", (__u8)src[i]); | |
187 | } | |
188 | return dest; | |
189 | } | |
190 | ||
96ae923a GR |
191 | static void add_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres, |
192 | sector_t lo, sector_t hi) | |
193 | { | |
194 | struct resync_info *ri; | |
195 | ||
196 | ri = (struct resync_info *)lockres->lksb.sb_lvbptr; | |
197 | ri->lo = cpu_to_le64(lo); | |
198 | ri->hi = cpu_to_le64(hi); | |
199 | } | |
200 | ||
201 | static struct suspend_info *read_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres) | |
202 | { | |
203 | struct resync_info ri; | |
204 | struct suspend_info *s = NULL; | |
205 | sector_t hi = 0; | |
206 | ||
207 | dlm_lock_sync(lockres, DLM_LOCK_CR); | |
208 | memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info)); | |
209 | hi = le64_to_cpu(ri.hi); | |
210 | if (ri.hi > 0) { | |
211 | s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL); | |
212 | if (!s) | |
213 | goto out; | |
214 | s->hi = hi; | |
215 | s->lo = le64_to_cpu(ri.lo); | |
216 | } | |
217 | dlm_unlock_sync(lockres); | |
218 | out: | |
219 | return s; | |
220 | } | |
221 | ||
6dc69c9c | 222 | static void recover_bitmaps(struct md_thread *thread) |
e94987db GR |
223 | { |
224 | struct mddev *mddev = thread->mddev; | |
225 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
226 | struct dlm_lock_resource *bm_lockres; | |
227 | char str[64]; | |
228 | int slot, ret; | |
229 | struct suspend_info *s, *tmp; | |
230 | sector_t lo, hi; | |
231 | ||
232 | while (cinfo->recovery_map) { | |
233 | slot = fls64((u64)cinfo->recovery_map) - 1; | |
234 | ||
235 | /* Clear suspend_area associated with the bitmap */ | |
236 | spin_lock_irq(&cinfo->suspend_lock); | |
237 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
238 | if (slot == s->slot) { | |
239 | list_del(&s->list); | |
240 | kfree(s); | |
241 | } | |
242 | spin_unlock_irq(&cinfo->suspend_lock); | |
243 | ||
244 | snprintf(str, 64, "bitmap%04d", slot); | |
245 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
246 | if (!bm_lockres) { | |
247 | pr_err("md-cluster: Cannot initialize bitmaps\n"); | |
248 | goto clear_bit; | |
249 | } | |
250 | ||
251 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
252 | if (ret) { | |
253 | pr_err("md-cluster: Could not DLM lock %s: %d\n", | |
254 | str, ret); | |
255 | goto clear_bit; | |
256 | } | |
97f6cd39 | 257 | ret = bitmap_copy_from_slot(mddev, slot, &lo, &hi, true); |
4b26a08a | 258 | if (ret) { |
e94987db | 259 | pr_err("md-cluster: Could not copy data from bitmap %d\n", slot); |
4b26a08a GR |
260 | goto dlm_unlock; |
261 | } | |
262 | if (hi > 0) { | |
263 | /* TODO:Wait for current resync to get over */ | |
264 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
265 | if (lo < mddev->recovery_cp) | |
266 | mddev->recovery_cp = lo; | |
267 | md_check_recovery(mddev); | |
268 | } | |
269 | dlm_unlock: | |
e94987db GR |
270 | dlm_unlock_sync(bm_lockres); |
271 | clear_bit: | |
272 | clear_bit(slot, &cinfo->recovery_map); | |
273 | } | |
274 | } | |
275 | ||
cf921cc1 GR |
276 | static void recover_prep(void *arg) |
277 | { | |
278 | } | |
279 | ||
280 | static void recover_slot(void *arg, struct dlm_slot *slot) | |
281 | { | |
282 | struct mddev *mddev = arg; | |
283 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
284 | ||
285 | pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n", | |
286 | mddev->bitmap_info.cluster_name, | |
287 | slot->nodeid, slot->slot, | |
288 | cinfo->slot_number); | |
e94987db GR |
289 | set_bit(slot->slot - 1, &cinfo->recovery_map); |
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 | ||
301 | static void recover_done(void *arg, struct dlm_slot *slots, | |
302 | int num_slots, int our_slot, | |
303 | uint32_t generation) | |
304 | { | |
305 | struct mddev *mddev = arg; | |
306 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
307 | ||
308 | cinfo->slot_number = our_slot; | |
309 | complete(&cinfo->completion); | |
310 | } | |
311 | ||
312 | static const struct dlm_lockspace_ops md_ls_ops = { | |
313 | .recover_prep = recover_prep, | |
314 | .recover_slot = recover_slot, | |
315 | .recover_done = recover_done, | |
316 | }; | |
317 | ||
4664680c GR |
318 | /* |
319 | * The BAST function for the ack lock resource | |
320 | * This function wakes up the receive thread in | |
321 | * order to receive and process the message. | |
322 | */ | |
323 | static void ack_bast(void *arg, int mode) | |
324 | { | |
325 | struct dlm_lock_resource *res = (struct dlm_lock_resource *)arg; | |
326 | struct md_cluster_info *cinfo = res->mddev->cluster_info; | |
327 | ||
328 | if (mode == DLM_LOCK_EX) | |
329 | md_wakeup_thread(cinfo->recv_thread); | |
330 | } | |
331 | ||
e59721cc GR |
332 | static void __remove_suspend_info(struct md_cluster_info *cinfo, int slot) |
333 | { | |
334 | struct suspend_info *s, *tmp; | |
335 | ||
336 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
337 | if (slot == s->slot) { | |
338 | pr_info("%s:%d Deleting suspend_info: %d\n", | |
339 | __func__, __LINE__, slot); | |
340 | list_del(&s->list); | |
341 | kfree(s); | |
342 | break; | |
343 | } | |
344 | } | |
345 | ||
346 | static void remove_suspend_info(struct md_cluster_info *cinfo, int slot) | |
347 | { | |
348 | spin_lock_irq(&cinfo->suspend_lock); | |
349 | __remove_suspend_info(cinfo, slot); | |
350 | spin_unlock_irq(&cinfo->suspend_lock); | |
351 | } | |
352 | ||
353 | ||
354 | static void process_suspend_info(struct md_cluster_info *cinfo, | |
355 | int slot, sector_t lo, sector_t hi) | |
356 | { | |
357 | struct suspend_info *s; | |
358 | ||
359 | if (!hi) { | |
360 | remove_suspend_info(cinfo, slot); | |
361 | return; | |
362 | } | |
363 | s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL); | |
364 | if (!s) | |
365 | return; | |
366 | s->slot = slot; | |
367 | s->lo = lo; | |
368 | s->hi = hi; | |
369 | spin_lock_irq(&cinfo->suspend_lock); | |
370 | /* Remove existing entry (if exists) before adding */ | |
371 | __remove_suspend_info(cinfo, slot); | |
372 | list_add(&s->list, &cinfo->suspend_list); | |
373 | spin_unlock_irq(&cinfo->suspend_lock); | |
374 | } | |
375 | ||
1aee41f6 GR |
376 | static void process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg) |
377 | { | |
378 | char disk_uuid[64]; | |
379 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
380 | char event_name[] = "EVENT=ADD_DEVICE"; | |
381 | char raid_slot[16]; | |
382 | char *envp[] = {event_name, disk_uuid, raid_slot, NULL}; | |
383 | int len; | |
384 | ||
385 | len = snprintf(disk_uuid, 64, "DEVICE_UUID="); | |
386 | pretty_uuid(disk_uuid + len, cmsg->uuid); | |
387 | snprintf(raid_slot, 16, "RAID_DISK=%d", cmsg->raid_slot); | |
388 | pr_info("%s:%d Sending kobject change with %s and %s\n", __func__, __LINE__, disk_uuid, raid_slot); | |
389 | init_completion(&cinfo->newdisk_completion); | |
fa8259da | 390 | set_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
391 | kobject_uevent_env(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE, envp); |
392 | wait_for_completion_timeout(&cinfo->newdisk_completion, | |
393 | NEW_DEV_TIMEOUT); | |
fa8259da | 394 | clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
395 | } |
396 | ||
397 | ||
398 | static void process_metadata_update(struct mddev *mddev, struct cluster_msg *msg) | |
399 | { | |
400 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
401 | ||
402 | md_reload_sb(mddev); | |
403 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); | |
404 | } | |
405 | ||
88bcfef7 GR |
406 | static void process_remove_disk(struct mddev *mddev, struct cluster_msg *msg) |
407 | { | |
408 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, msg->raid_slot); | |
409 | ||
410 | if (rdev) | |
411 | md_kick_rdev_from_array(rdev); | |
412 | else | |
413 | pr_warn("%s: %d Could not find disk(%d) to REMOVE\n", __func__, __LINE__, msg->raid_slot); | |
414 | } | |
415 | ||
97f6cd39 GR |
416 | static void process_readd_disk(struct mddev *mddev, struct cluster_msg *msg) |
417 | { | |
418 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, msg->raid_slot); | |
419 | ||
420 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
421 | clear_bit(Faulty, &rdev->flags); | |
422 | else | |
423 | pr_warn("%s: %d Could not find disk(%d) which is faulty", __func__, __LINE__, msg->raid_slot); | |
424 | } | |
425 | ||
4664680c GR |
426 | static void process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg) |
427 | { | |
428 | switch (msg->type) { | |
429 | case METADATA_UPDATED: | |
430 | pr_info("%s: %d Received message: METADATA_UPDATE from %d\n", | |
431 | __func__, __LINE__, msg->slot); | |
1aee41f6 | 432 | process_metadata_update(mddev, msg); |
4664680c GR |
433 | break; |
434 | case RESYNCING: | |
435 | pr_info("%s: %d Received message: RESYNCING from %d\n", | |
436 | __func__, __LINE__, msg->slot); | |
e59721cc GR |
437 | process_suspend_info(mddev->cluster_info, msg->slot, |
438 | msg->low, msg->high); | |
4664680c | 439 | break; |
1aee41f6 GR |
440 | case NEWDISK: |
441 | pr_info("%s: %d Received message: NEWDISK from %d\n", | |
442 | __func__, __LINE__, msg->slot); | |
443 | process_add_new_disk(mddev, msg); | |
88bcfef7 GR |
444 | break; |
445 | case REMOVE: | |
446 | pr_info("%s: %d Received REMOVE from %d\n", | |
447 | __func__, __LINE__, msg->slot); | |
448 | process_remove_disk(mddev, msg); | |
449 | break; | |
97f6cd39 GR |
450 | case RE_ADD: |
451 | pr_info("%s: %d Received RE_ADD from %d\n", | |
452 | __func__, __LINE__, msg->slot); | |
453 | process_readd_disk(mddev, msg); | |
454 | break; | |
88bcfef7 GR |
455 | default: |
456 | pr_warn("%s:%d Received unknown message from %d\n", | |
457 | __func__, __LINE__, msg->slot); | |
09dd1af2 | 458 | } |
4664680c GR |
459 | } |
460 | ||
461 | /* | |
462 | * thread for receiving message | |
463 | */ | |
464 | static void recv_daemon(struct md_thread *thread) | |
465 | { | |
466 | struct md_cluster_info *cinfo = thread->mddev->cluster_info; | |
467 | struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres; | |
468 | struct dlm_lock_resource *message_lockres = cinfo->message_lockres; | |
469 | struct cluster_msg msg; | |
470 | ||
471 | /*get CR on Message*/ | |
472 | if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) { | |
473 | pr_err("md/raid1:failed to get CR on MESSAGE\n"); | |
474 | return; | |
475 | } | |
476 | ||
477 | /* read lvb and wake up thread to process this message_lockres */ | |
478 | memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg)); | |
479 | process_recvd_msg(thread->mddev, &msg); | |
480 | ||
481 | /*release CR on ack_lockres*/ | |
482 | dlm_unlock_sync(ack_lockres); | |
483 | /*up-convert to EX on message_lockres*/ | |
484 | dlm_lock_sync(message_lockres, DLM_LOCK_EX); | |
485 | /*get CR on ack_lockres again*/ | |
486 | dlm_lock_sync(ack_lockres, DLM_LOCK_CR); | |
487 | /*release CR on message_lockres*/ | |
488 | dlm_unlock_sync(message_lockres); | |
489 | } | |
490 | ||
601b515c GR |
491 | /* lock_comm() |
492 | * Takes the lock on the TOKEN lock resource so no other | |
493 | * node can communicate while the operation is underway. | |
494 | */ | |
495 | static int lock_comm(struct md_cluster_info *cinfo) | |
496 | { | |
497 | int error; | |
498 | ||
499 | error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX); | |
500 | if (error) | |
501 | pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n", | |
502 | __func__, __LINE__, error); | |
503 | return error; | |
504 | } | |
505 | ||
506 | static void unlock_comm(struct md_cluster_info *cinfo) | |
507 | { | |
508 | dlm_unlock_sync(cinfo->token_lockres); | |
509 | } | |
510 | ||
511 | /* __sendmsg() | |
512 | * This function performs the actual sending of the message. This function is | |
513 | * usually called after performing the encompassing operation | |
514 | * The function: | |
515 | * 1. Grabs the message lockresource in EX mode | |
516 | * 2. Copies the message to the message LVB | |
517 | * 3. Downconverts message lockresource to CR | |
518 | * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes | |
519 | * and the other nodes read the message. The thread will wait here until all other | |
520 | * nodes have released ack lock resource. | |
521 | * 5. Downconvert ack lockresource to CR | |
522 | */ | |
523 | static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
524 | { | |
525 | int error; | |
526 | int slot = cinfo->slot_number - 1; | |
527 | ||
528 | cmsg->slot = cpu_to_le32(slot); | |
529 | /*get EX on Message*/ | |
530 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX); | |
531 | if (error) { | |
532 | pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error); | |
533 | goto failed_message; | |
534 | } | |
535 | ||
536 | memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg, | |
537 | sizeof(struct cluster_msg)); | |
538 | /*down-convert EX to CR on Message*/ | |
539 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CR); | |
540 | if (error) { | |
541 | pr_err("md-cluster: failed to convert EX to CR on MESSAGE(%d)\n", | |
542 | error); | |
543 | goto failed_message; | |
544 | } | |
545 | ||
546 | /*up-convert CR to EX on Ack*/ | |
547 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX); | |
548 | if (error) { | |
549 | pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n", | |
550 | error); | |
551 | goto failed_ack; | |
552 | } | |
553 | ||
554 | /*down-convert EX to CR on Ack*/ | |
555 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR); | |
556 | if (error) { | |
557 | pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n", | |
558 | error); | |
559 | goto failed_ack; | |
560 | } | |
561 | ||
562 | failed_ack: | |
563 | dlm_unlock_sync(cinfo->message_lockres); | |
564 | failed_message: | |
565 | return error; | |
566 | } | |
567 | ||
568 | static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
569 | { | |
570 | int ret; | |
571 | ||
572 | lock_comm(cinfo); | |
573 | ret = __sendmsg(cinfo, cmsg); | |
574 | unlock_comm(cinfo); | |
575 | return ret; | |
576 | } | |
577 | ||
96ae923a GR |
578 | static int gather_all_resync_info(struct mddev *mddev, int total_slots) |
579 | { | |
580 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
581 | int i, ret = 0; | |
582 | struct dlm_lock_resource *bm_lockres; | |
583 | struct suspend_info *s; | |
584 | char str[64]; | |
585 | ||
586 | ||
587 | for (i = 0; i < total_slots; i++) { | |
588 | memset(str, '\0', 64); | |
589 | snprintf(str, 64, "bitmap%04d", i); | |
590 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
591 | if (!bm_lockres) | |
592 | return -ENOMEM; | |
593 | if (i == (cinfo->slot_number - 1)) | |
594 | continue; | |
595 | ||
596 | bm_lockres->flags |= DLM_LKF_NOQUEUE; | |
597 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
598 | if (ret == -EAGAIN) { | |
599 | memset(bm_lockres->lksb.sb_lvbptr, '\0', LVB_SIZE); | |
600 | s = read_resync_info(mddev, bm_lockres); | |
601 | if (s) { | |
602 | pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n", | |
603 | __func__, __LINE__, | |
604 | (unsigned long long) s->lo, | |
605 | (unsigned long long) s->hi, i); | |
606 | spin_lock_irq(&cinfo->suspend_lock); | |
607 | s->slot = i; | |
608 | list_add(&s->list, &cinfo->suspend_list); | |
609 | spin_unlock_irq(&cinfo->suspend_lock); | |
610 | } | |
611 | ret = 0; | |
612 | lockres_free(bm_lockres); | |
613 | continue; | |
614 | } | |
615 | if (ret) | |
616 | goto out; | |
617 | /* TODO: Read the disk bitmap sb and check if it needs recovery */ | |
618 | dlm_unlock_sync(bm_lockres); | |
619 | lockres_free(bm_lockres); | |
620 | } | |
621 | out: | |
622 | return ret; | |
623 | } | |
624 | ||
edb39c9d GR |
625 | static int join(struct mddev *mddev, int nodes) |
626 | { | |
c4ce867f | 627 | struct md_cluster_info *cinfo; |
cf921cc1 | 628 | int ret, ops_rv; |
c4ce867f GR |
629 | char str[64]; |
630 | ||
631 | if (!try_module_get(THIS_MODULE)) | |
632 | return -ENOENT; | |
633 | ||
634 | cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL); | |
635 | if (!cinfo) | |
636 | return -ENOMEM; | |
637 | ||
cf921cc1 GR |
638 | init_completion(&cinfo->completion); |
639 | ||
640 | mutex_init(&cinfo->sb_mutex); | |
641 | mddev->cluster_info = cinfo; | |
642 | ||
c4ce867f GR |
643 | memset(str, 0, 64); |
644 | pretty_uuid(str, mddev->uuid); | |
cf921cc1 GR |
645 | ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name, |
646 | DLM_LSFL_FS, LVB_SIZE, | |
647 | &md_ls_ops, mddev, &ops_rv, &cinfo->lockspace); | |
c4ce867f GR |
648 | if (ret) |
649 | goto err; | |
cf921cc1 | 650 | wait_for_completion(&cinfo->completion); |
8c58f02e GJ |
651 | if (nodes < cinfo->slot_number) { |
652 | pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).", | |
653 | cinfo->slot_number, nodes); | |
b97e9257 GR |
654 | ret = -ERANGE; |
655 | goto err; | |
656 | } | |
c4ce867f GR |
657 | cinfo->sb_lock = lockres_init(mddev, "cmd-super", |
658 | NULL, 0); | |
659 | if (!cinfo->sb_lock) { | |
660 | ret = -ENOMEM; | |
661 | goto err; | |
662 | } | |
4664680c GR |
663 | /* Initiate the communication resources */ |
664 | ret = -ENOMEM; | |
665 | cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv"); | |
666 | if (!cinfo->recv_thread) { | |
667 | pr_err("md-cluster: cannot allocate memory for recv_thread!\n"); | |
668 | goto err; | |
669 | } | |
670 | cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1); | |
671 | if (!cinfo->message_lockres) | |
672 | goto err; | |
673 | cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0); | |
674 | if (!cinfo->token_lockres) | |
675 | goto err; | |
676 | cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0); | |
677 | if (!cinfo->ack_lockres) | |
678 | goto err; | |
1aee41f6 GR |
679 | cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0); |
680 | if (!cinfo->no_new_dev_lockres) | |
681 | goto err; | |
682 | ||
4664680c GR |
683 | /* get sync CR lock on ACK. */ |
684 | if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR)) | |
685 | pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n", | |
686 | ret); | |
1aee41f6 GR |
687 | /* get sync CR lock on no-new-dev. */ |
688 | if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR)) | |
689 | pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret); | |
690 | ||
54519c5f GR |
691 | |
692 | pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number); | |
693 | snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1); | |
694 | cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1); | |
695 | if (!cinfo->bitmap_lockres) | |
696 | goto err; | |
697 | if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) { | |
698 | pr_err("Failed to get bitmap lock\n"); | |
699 | ret = -EINVAL; | |
700 | goto err; | |
701 | } | |
702 | ||
96ae923a GR |
703 | INIT_LIST_HEAD(&cinfo->suspend_list); |
704 | spin_lock_init(&cinfo->suspend_lock); | |
705 | ||
706 | ret = gather_all_resync_info(mddev, nodes); | |
707 | if (ret) | |
708 | goto err; | |
709 | ||
edb39c9d | 710 | return 0; |
c4ce867f | 711 | err: |
4664680c GR |
712 | lockres_free(cinfo->message_lockres); |
713 | lockres_free(cinfo->token_lockres); | |
714 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 715 | lockres_free(cinfo->no_new_dev_lockres); |
96ae923a GR |
716 | lockres_free(cinfo->bitmap_lockres); |
717 | lockres_free(cinfo->sb_lock); | |
c4ce867f GR |
718 | if (cinfo->lockspace) |
719 | dlm_release_lockspace(cinfo->lockspace, 2); | |
cf921cc1 | 720 | mddev->cluster_info = NULL; |
c4ce867f GR |
721 | kfree(cinfo); |
722 | module_put(THIS_MODULE); | |
723 | return ret; | |
edb39c9d GR |
724 | } |
725 | ||
726 | static int leave(struct mddev *mddev) | |
727 | { | |
c4ce867f GR |
728 | struct md_cluster_info *cinfo = mddev->cluster_info; |
729 | ||
730 | if (!cinfo) | |
731 | return 0; | |
e94987db | 732 | md_unregister_thread(&cinfo->recovery_thread); |
4664680c GR |
733 | md_unregister_thread(&cinfo->recv_thread); |
734 | lockres_free(cinfo->message_lockres); | |
735 | lockres_free(cinfo->token_lockres); | |
736 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 737 | lockres_free(cinfo->no_new_dev_lockres); |
c4ce867f | 738 | lockres_free(cinfo->sb_lock); |
54519c5f | 739 | lockres_free(cinfo->bitmap_lockres); |
c4ce867f | 740 | dlm_release_lockspace(cinfo->lockspace, 2); |
edb39c9d GR |
741 | return 0; |
742 | } | |
743 | ||
cf921cc1 GR |
744 | /* slot_number(): Returns the MD slot number to use |
745 | * DLM starts the slot numbers from 1, wheras cluster-md | |
746 | * wants the number to be from zero, so we deduct one | |
747 | */ | |
748 | static int slot_number(struct mddev *mddev) | |
749 | { | |
750 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
751 | ||
752 | return cinfo->slot_number - 1; | |
753 | } | |
754 | ||
96ae923a GR |
755 | static void resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi) |
756 | { | |
757 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
758 | ||
759 | add_resync_info(mddev, cinfo->bitmap_lockres, lo, hi); | |
760 | /* Re-acquire the lock to refresh LVB */ | |
761 | dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW); | |
762 | } | |
763 | ||
293467aa GR |
764 | static int metadata_update_start(struct mddev *mddev) |
765 | { | |
766 | return lock_comm(mddev->cluster_info); | |
767 | } | |
768 | ||
769 | static int metadata_update_finish(struct mddev *mddev) | |
770 | { | |
771 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
772 | struct cluster_msg cmsg; | |
773 | int ret; | |
774 | ||
775 | memset(&cmsg, 0, sizeof(cmsg)); | |
776 | cmsg.type = cpu_to_le32(METADATA_UPDATED); | |
777 | ret = __sendmsg(cinfo, &cmsg); | |
778 | unlock_comm(cinfo); | |
779 | return ret; | |
780 | } | |
781 | ||
782 | static int metadata_update_cancel(struct mddev *mddev) | |
783 | { | |
784 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
785 | ||
786 | return dlm_unlock_sync(cinfo->token_lockres); | |
787 | } | |
788 | ||
965400eb GR |
789 | static int resync_send(struct mddev *mddev, enum msg_type type, |
790 | sector_t lo, sector_t hi) | |
791 | { | |
792 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
793 | struct cluster_msg cmsg; | |
794 | int slot = cinfo->slot_number - 1; | |
795 | ||
796 | pr_info("%s:%d lo: %llu hi: %llu\n", __func__, __LINE__, | |
797 | (unsigned long long)lo, | |
798 | (unsigned long long)hi); | |
799 | resync_info_update(mddev, lo, hi); | |
800 | cmsg.type = cpu_to_le32(type); | |
801 | cmsg.slot = cpu_to_le32(slot); | |
802 | cmsg.low = cpu_to_le64(lo); | |
803 | cmsg.high = cpu_to_le64(hi); | |
804 | return sendmsg(cinfo, &cmsg); | |
805 | } | |
806 | ||
807 | static int resync_start(struct mddev *mddev, sector_t lo, sector_t hi) | |
808 | { | |
809 | pr_info("%s:%d\n", __func__, __LINE__); | |
810 | return resync_send(mddev, RESYNCING, lo, hi); | |
811 | } | |
812 | ||
813 | static void resync_finish(struct mddev *mddev) | |
814 | { | |
815 | pr_info("%s:%d\n", __func__, __LINE__); | |
816 | resync_send(mddev, RESYNCING, 0, 0); | |
817 | } | |
818 | ||
589a1c49 GR |
819 | static int area_resyncing(struct mddev *mddev, sector_t lo, sector_t hi) |
820 | { | |
821 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
822 | int ret = 0; | |
823 | struct suspend_info *s; | |
824 | ||
825 | spin_lock_irq(&cinfo->suspend_lock); | |
826 | if (list_empty(&cinfo->suspend_list)) | |
827 | goto out; | |
828 | list_for_each_entry(s, &cinfo->suspend_list, list) | |
829 | if (hi > s->lo && lo < s->hi) { | |
830 | ret = 1; | |
831 | break; | |
832 | } | |
833 | out: | |
834 | spin_unlock_irq(&cinfo->suspend_lock); | |
835 | return ret; | |
836 | } | |
837 | ||
1aee41f6 GR |
838 | static int add_new_disk_start(struct mddev *mddev, struct md_rdev *rdev) |
839 | { | |
840 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
841 | struct cluster_msg cmsg; | |
842 | int ret = 0; | |
843 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); | |
844 | char *uuid = sb->device_uuid; | |
845 | ||
846 | memset(&cmsg, 0, sizeof(cmsg)); | |
847 | cmsg.type = cpu_to_le32(NEWDISK); | |
848 | memcpy(cmsg.uuid, uuid, 16); | |
849 | cmsg.raid_slot = rdev->desc_nr; | |
850 | lock_comm(cinfo); | |
851 | ret = __sendmsg(cinfo, &cmsg); | |
852 | if (ret) | |
853 | return ret; | |
854 | cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE; | |
855 | ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX); | |
856 | cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE; | |
857 | /* Some node does not "see" the device */ | |
858 | if (ret == -EAGAIN) | |
859 | ret = -ENOENT; | |
860 | else | |
861 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); | |
862 | return ret; | |
863 | } | |
864 | ||
865 | static int add_new_disk_finish(struct mddev *mddev) | |
866 | { | |
867 | struct cluster_msg cmsg; | |
868 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
869 | int ret; | |
870 | /* Write sb and inform others */ | |
871 | md_update_sb(mddev, 1); | |
872 | cmsg.type = METADATA_UPDATED; | |
873 | ret = __sendmsg(cinfo, &cmsg); | |
874 | unlock_comm(cinfo); | |
875 | return ret; | |
876 | } | |
877 | ||
fa8259da | 878 | static int new_disk_ack(struct mddev *mddev, bool ack) |
1aee41f6 GR |
879 | { |
880 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
881 | ||
fa8259da GR |
882 | if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) { |
883 | pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev)); | |
884 | return -EINVAL; | |
885 | } | |
886 | ||
1aee41f6 GR |
887 | if (ack) |
888 | dlm_unlock_sync(cinfo->no_new_dev_lockres); | |
889 | complete(&cinfo->newdisk_completion); | |
fa8259da | 890 | return 0; |
1aee41f6 GR |
891 | } |
892 | ||
88bcfef7 GR |
893 | static int remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
894 | { | |
895 | struct cluster_msg cmsg; | |
896 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
897 | cmsg.type = REMOVE; | |
898 | cmsg.raid_slot = rdev->desc_nr; | |
899 | return __sendmsg(cinfo, &cmsg); | |
900 | } | |
901 | ||
97f6cd39 GR |
902 | static int gather_bitmaps(struct md_rdev *rdev) |
903 | { | |
904 | int sn, err; | |
905 | sector_t lo, hi; | |
906 | struct cluster_msg cmsg; | |
907 | struct mddev *mddev = rdev->mddev; | |
908 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
909 | ||
910 | cmsg.type = RE_ADD; | |
911 | cmsg.raid_slot = rdev->desc_nr; | |
912 | err = sendmsg(cinfo, &cmsg); | |
913 | if (err) | |
914 | goto out; | |
915 | ||
916 | for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) { | |
917 | if (sn == (cinfo->slot_number - 1)) | |
918 | continue; | |
919 | err = bitmap_copy_from_slot(mddev, sn, &lo, &hi, false); | |
920 | if (err) { | |
921 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn); | |
922 | goto out; | |
923 | } | |
924 | if ((hi > 0) && (lo < mddev->recovery_cp)) | |
925 | mddev->recovery_cp = lo; | |
926 | } | |
927 | out: | |
928 | return err; | |
929 | } | |
930 | ||
edb39c9d GR |
931 | static struct md_cluster_operations cluster_ops = { |
932 | .join = join, | |
933 | .leave = leave, | |
cf921cc1 | 934 | .slot_number = slot_number, |
96ae923a | 935 | .resync_info_update = resync_info_update, |
965400eb GR |
936 | .resync_start = resync_start, |
937 | .resync_finish = resync_finish, | |
293467aa GR |
938 | .metadata_update_start = metadata_update_start, |
939 | .metadata_update_finish = metadata_update_finish, | |
940 | .metadata_update_cancel = metadata_update_cancel, | |
589a1c49 | 941 | .area_resyncing = area_resyncing, |
1aee41f6 GR |
942 | .add_new_disk_start = add_new_disk_start, |
943 | .add_new_disk_finish = add_new_disk_finish, | |
944 | .new_disk_ack = new_disk_ack, | |
88bcfef7 | 945 | .remove_disk = remove_disk, |
97f6cd39 | 946 | .gather_bitmaps = gather_bitmaps, |
edb39c9d GR |
947 | }; |
948 | ||
8e854e9c GR |
949 | static int __init cluster_init(void) |
950 | { | |
951 | pr_warn("md-cluster: EXPERIMENTAL. Use with caution\n"); | |
952 | pr_info("Registering Cluster MD functions\n"); | |
edb39c9d | 953 | register_md_cluster_operations(&cluster_ops, THIS_MODULE); |
8e854e9c GR |
954 | return 0; |
955 | } | |
956 | ||
957 | static void cluster_exit(void) | |
958 | { | |
edb39c9d | 959 | unregister_md_cluster_operations(); |
8e854e9c GR |
960 | } |
961 | ||
962 | module_init(cluster_init); | |
963 | module_exit(cluster_exit); | |
964 | MODULE_LICENSE("GPL"); | |
965 | MODULE_DESCRIPTION("Clustering support for MD"); |