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. */ | |
dbb64f86 | 31 | int mode; |
c4ce867f GR |
32 | }; |
33 | ||
96ae923a GR |
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 | ||
fa8259da GR |
46 | /* md_cluster_info flags */ |
47 | #define MD_CLUSTER_WAITING_FOR_NEWDISK 1 | |
90382ed9 | 48 | #define MD_CLUSTER_SUSPEND_READ_BALANCING 2 |
eece075c | 49 | #define MD_CLUSTER_BEGIN_JOIN_CLUSTER 3 |
fa8259da GR |
50 | |
51 | ||
c4ce867f GR |
52 | struct md_cluster_info { |
53 | /* dlm lock space and resources for clustered raid. */ | |
54 | dlm_lockspace_t *lockspace; | |
cf921cc1 GR |
55 | int slot_number; |
56 | struct completion completion; | |
54519c5f | 57 | struct dlm_lock_resource *bitmap_lockres; |
c186b128 | 58 | struct dlm_lock_resource *resync_lockres; |
96ae923a GR |
59 | struct list_head suspend_list; |
60 | spinlock_t suspend_lock; | |
e94987db GR |
61 | struct md_thread *recovery_thread; |
62 | unsigned long recovery_map; | |
4664680c GR |
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; | |
1aee41f6 | 67 | struct dlm_lock_resource *no_new_dev_lockres; |
4664680c | 68 | struct md_thread *recv_thread; |
1aee41f6 | 69 | struct completion newdisk_completion; |
fa8259da | 70 | unsigned long state; |
4664680c GR |
71 | }; |
72 | ||
73 | enum msg_type { | |
74 | METADATA_UPDATED = 0, | |
75 | RESYNCING, | |
1aee41f6 | 76 | NEWDISK, |
88bcfef7 | 77 | REMOVE, |
97f6cd39 | 78 | RE_ADD, |
dc737d7c | 79 | BITMAP_NEEDS_SYNC, |
4664680c GR |
80 | }; |
81 | ||
82 | struct cluster_msg { | |
cf97a348 GJ |
83 | __le32 type; |
84 | __le32 slot; | |
1aee41f6 | 85 | /* TODO: Unionize this for smaller footprint */ |
cf97a348 GJ |
86 | __le64 low; |
87 | __le64 high; | |
1aee41f6 | 88 | char uuid[16]; |
cf97a348 | 89 | __le32 raid_slot; |
47741b7c GR |
90 | }; |
91 | ||
92 | static void sync_ast(void *arg) | |
93 | { | |
94 | struct dlm_lock_resource *res; | |
95 | ||
2e2a7cd9 | 96 | res = arg; |
47741b7c GR |
97 | complete(&res->completion); |
98 | } | |
99 | ||
100 | static int dlm_lock_sync(struct dlm_lock_resource *res, int mode) | |
101 | { | |
102 | int ret = 0; | |
103 | ||
47741b7c GR |
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); | |
dbb64f86 GR |
110 | if (res->lksb.sb_status == 0) |
111 | res->mode = mode; | |
47741b7c GR |
112 | return res->lksb.sb_status; |
113 | } | |
114 | ||
115 | static int dlm_unlock_sync(struct dlm_lock_resource *res) | |
116 | { | |
117 | return dlm_lock_sync(res, DLM_LOCK_NL); | |
118 | } | |
119 | ||
c4ce867f | 120 | static struct dlm_lock_resource *lockres_init(struct mddev *mddev, |
47741b7c GR |
121 | char *name, void (*bastfn)(void *arg, int mode), int with_lvb) |
122 | { | |
123 | struct dlm_lock_resource *res = NULL; | |
124 | int ret, namelen; | |
c4ce867f | 125 | struct md_cluster_info *cinfo = mddev->cluster_info; |
47741b7c GR |
126 | |
127 | res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL); | |
128 | if (!res) | |
129 | return NULL; | |
b83d51c0 | 130 | init_completion(&res->completion); |
c4ce867f GR |
131 | res->ls = cinfo->lockspace; |
132 | res->mddev = mddev; | |
dbb64f86 | 133 | res->mode = DLM_LOCK_IV; |
47741b7c GR |
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 | ||
171 | static void lockres_free(struct dlm_lock_resource *res) | |
172 | { | |
b5ef5678 GJ |
173 | int ret; |
174 | ||
47741b7c GR |
175 | if (!res) |
176 | return; | |
177 | ||
b5ef5678 GJ |
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; | |
47741b7c GR |
191 | wait_for_completion(&res->completion); |
192 | ||
193 | kfree(res->name); | |
194 | kfree(res->lksb.sb_lvbptr); | |
195 | kfree(res); | |
196 | } | |
8e854e9c | 197 | |
30661b49 N |
198 | static void add_resync_info(struct dlm_lock_resource *lockres, |
199 | sector_t lo, sector_t hi) | |
96ae923a GR |
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 | ||
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); | |
cf97a348 | 217 | if (hi > 0) { |
96ae923a GR |
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 | ||
6dc69c9c | 229 | static void recover_bitmaps(struct md_thread *thread) |
e94987db GR |
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 | /* Clear suspend_area associated with the bitmap */ | |
243 | spin_lock_irq(&cinfo->suspend_lock); | |
244 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
245 | if (slot == s->slot) { | |
246 | list_del(&s->list); | |
247 | kfree(s); | |
248 | } | |
249 | spin_unlock_irq(&cinfo->suspend_lock); | |
250 | ||
251 | snprintf(str, 64, "bitmap%04d", slot); | |
252 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
253 | if (!bm_lockres) { | |
254 | pr_err("md-cluster: Cannot initialize bitmaps\n"); | |
255 | goto clear_bit; | |
256 | } | |
257 | ||
258 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
259 | if (ret) { | |
260 | pr_err("md-cluster: Could not DLM lock %s: %d\n", | |
261 | str, ret); | |
262 | goto clear_bit; | |
263 | } | |
97f6cd39 | 264 | ret = bitmap_copy_from_slot(mddev, slot, &lo, &hi, true); |
4b26a08a | 265 | if (ret) { |
e94987db | 266 | pr_err("md-cluster: Could not copy data from bitmap %d\n", slot); |
4b26a08a GR |
267 | goto dlm_unlock; |
268 | } | |
269 | if (hi > 0) { | |
270 | /* TODO:Wait for current resync to get over */ | |
271 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
272 | if (lo < mddev->recovery_cp) | |
273 | mddev->recovery_cp = lo; | |
274 | md_check_recovery(mddev); | |
275 | } | |
276 | dlm_unlock: | |
e94987db GR |
277 | dlm_unlock_sync(bm_lockres); |
278 | clear_bit: | |
279 | clear_bit(slot, &cinfo->recovery_map); | |
280 | } | |
281 | } | |
282 | ||
cf921cc1 GR |
283 | static void recover_prep(void *arg) |
284 | { | |
90382ed9 GR |
285 | struct mddev *mddev = arg; |
286 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
287 | set_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); | |
cf921cc1 GR |
288 | } |
289 | ||
05cd0e51 | 290 | static void __recover_slot(struct mddev *mddev, int slot) |
cf921cc1 | 291 | { |
cf921cc1 GR |
292 | struct md_cluster_info *cinfo = mddev->cluster_info; |
293 | ||
05cd0e51 | 294 | set_bit(slot, &cinfo->recovery_map); |
e94987db GR |
295 | if (!cinfo->recovery_thread) { |
296 | cinfo->recovery_thread = md_register_thread(recover_bitmaps, | |
297 | mddev, "recover"); | |
298 | if (!cinfo->recovery_thread) { | |
299 | pr_warn("md-cluster: Could not create recovery thread\n"); | |
300 | return; | |
301 | } | |
302 | } | |
303 | md_wakeup_thread(cinfo->recovery_thread); | |
cf921cc1 GR |
304 | } |
305 | ||
05cd0e51 GJ |
306 | static void recover_slot(void *arg, struct dlm_slot *slot) |
307 | { | |
308 | struct mddev *mddev = arg; | |
309 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
310 | ||
311 | pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n", | |
312 | mddev->bitmap_info.cluster_name, | |
313 | slot->nodeid, slot->slot, | |
314 | cinfo->slot_number); | |
315 | /* deduct one since dlm slot starts from one while the num of | |
316 | * cluster-md begins with 0 */ | |
317 | __recover_slot(mddev, slot->slot - 1); | |
318 | } | |
319 | ||
cf921cc1 GR |
320 | static void recover_done(void *arg, struct dlm_slot *slots, |
321 | int num_slots, int our_slot, | |
322 | uint32_t generation) | |
323 | { | |
324 | struct mddev *mddev = arg; | |
325 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
326 | ||
327 | cinfo->slot_number = our_slot; | |
eece075c GJ |
328 | /* completion is only need to be complete when node join cluster, |
329 | * it doesn't need to run during another node's failure */ | |
330 | if (test_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state)) { | |
331 | complete(&cinfo->completion); | |
332 | clear_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state); | |
333 | } | |
90382ed9 | 334 | clear_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); |
cf921cc1 GR |
335 | } |
336 | ||
eece075c GJ |
337 | /* the ops is called when node join the cluster, and do lock recovery |
338 | * if node failure occurs */ | |
cf921cc1 GR |
339 | static const struct dlm_lockspace_ops md_ls_ops = { |
340 | .recover_prep = recover_prep, | |
341 | .recover_slot = recover_slot, | |
342 | .recover_done = recover_done, | |
343 | }; | |
344 | ||
4664680c GR |
345 | /* |
346 | * The BAST function for the ack lock resource | |
347 | * This function wakes up the receive thread in | |
348 | * order to receive and process the message. | |
349 | */ | |
350 | static void ack_bast(void *arg, int mode) | |
351 | { | |
2e2a7cd9 | 352 | struct dlm_lock_resource *res = arg; |
4664680c GR |
353 | struct md_cluster_info *cinfo = res->mddev->cluster_info; |
354 | ||
355 | if (mode == DLM_LOCK_EX) | |
356 | md_wakeup_thread(cinfo->recv_thread); | |
357 | } | |
358 | ||
e59721cc GR |
359 | static void __remove_suspend_info(struct md_cluster_info *cinfo, int slot) |
360 | { | |
361 | struct suspend_info *s, *tmp; | |
362 | ||
363 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
364 | if (slot == s->slot) { | |
e59721cc GR |
365 | list_del(&s->list); |
366 | kfree(s); | |
367 | break; | |
368 | } | |
369 | } | |
370 | ||
b8ca846e | 371 | static void remove_suspend_info(struct mddev *mddev, int slot) |
e59721cc | 372 | { |
b8ca846e | 373 | struct md_cluster_info *cinfo = mddev->cluster_info; |
e59721cc GR |
374 | spin_lock_irq(&cinfo->suspend_lock); |
375 | __remove_suspend_info(cinfo, slot); | |
376 | spin_unlock_irq(&cinfo->suspend_lock); | |
b8ca846e | 377 | mddev->pers->quiesce(mddev, 2); |
e59721cc GR |
378 | } |
379 | ||
380 | ||
9ed38ff5 | 381 | static void process_suspend_info(struct mddev *mddev, |
e59721cc GR |
382 | int slot, sector_t lo, sector_t hi) |
383 | { | |
9ed38ff5 | 384 | struct md_cluster_info *cinfo = mddev->cluster_info; |
e59721cc GR |
385 | struct suspend_info *s; |
386 | ||
387 | if (!hi) { | |
b8ca846e | 388 | remove_suspend_info(mddev, slot); |
c186b128 GR |
389 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
390 | md_wakeup_thread(mddev->thread); | |
e59721cc GR |
391 | return; |
392 | } | |
393 | s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL); | |
394 | if (!s) | |
395 | return; | |
396 | s->slot = slot; | |
397 | s->lo = lo; | |
398 | s->hi = hi; | |
9ed38ff5 GR |
399 | mddev->pers->quiesce(mddev, 1); |
400 | mddev->pers->quiesce(mddev, 0); | |
e59721cc GR |
401 | spin_lock_irq(&cinfo->suspend_lock); |
402 | /* Remove existing entry (if exists) before adding */ | |
403 | __remove_suspend_info(cinfo, slot); | |
404 | list_add(&s->list, &cinfo->suspend_list); | |
405 | spin_unlock_irq(&cinfo->suspend_lock); | |
b8ca846e | 406 | mddev->pers->quiesce(mddev, 2); |
e59721cc GR |
407 | } |
408 | ||
1aee41f6 GR |
409 | static void process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg) |
410 | { | |
411 | char disk_uuid[64]; | |
412 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
413 | char event_name[] = "EVENT=ADD_DEVICE"; | |
414 | char raid_slot[16]; | |
415 | char *envp[] = {event_name, disk_uuid, raid_slot, NULL}; | |
416 | int len; | |
417 | ||
418 | len = snprintf(disk_uuid, 64, "DEVICE_UUID="); | |
b89f704a | 419 | sprintf(disk_uuid + len, "%pU", cmsg->uuid); |
faeff83f | 420 | snprintf(raid_slot, 16, "RAID_DISK=%d", le32_to_cpu(cmsg->raid_slot)); |
1aee41f6 GR |
421 | pr_info("%s:%d Sending kobject change with %s and %s\n", __func__, __LINE__, disk_uuid, raid_slot); |
422 | init_completion(&cinfo->newdisk_completion); | |
fa8259da | 423 | set_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
424 | kobject_uevent_env(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE, envp); |
425 | wait_for_completion_timeout(&cinfo->newdisk_completion, | |
426 | NEW_DEV_TIMEOUT); | |
fa8259da | 427 | clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
428 | } |
429 | ||
430 | ||
431 | static void process_metadata_update(struct mddev *mddev, struct cluster_msg *msg) | |
432 | { | |
433 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
70bcecdb | 434 | md_reload_sb(mddev, le32_to_cpu(msg->raid_slot)); |
1aee41f6 GR |
435 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); |
436 | } | |
437 | ||
88bcfef7 GR |
438 | static void process_remove_disk(struct mddev *mddev, struct cluster_msg *msg) |
439 | { | |
faeff83f GJ |
440 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, |
441 | le32_to_cpu(msg->raid_slot)); | |
88bcfef7 | 442 | |
659b254f GJ |
443 | if (rdev) { |
444 | set_bit(ClusterRemove, &rdev->flags); | |
445 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
446 | md_wakeup_thread(mddev->thread); | |
447 | } | |
88bcfef7 | 448 | else |
faeff83f GJ |
449 | pr_warn("%s: %d Could not find disk(%d) to REMOVE\n", |
450 | __func__, __LINE__, le32_to_cpu(msg->raid_slot)); | |
88bcfef7 GR |
451 | } |
452 | ||
97f6cd39 GR |
453 | static void process_readd_disk(struct mddev *mddev, struct cluster_msg *msg) |
454 | { | |
faeff83f GJ |
455 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, |
456 | le32_to_cpu(msg->raid_slot)); | |
97f6cd39 GR |
457 | |
458 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
459 | clear_bit(Faulty, &rdev->flags); | |
460 | else | |
faeff83f GJ |
461 | pr_warn("%s: %d Could not find disk(%d) which is faulty", |
462 | __func__, __LINE__, le32_to_cpu(msg->raid_slot)); | |
97f6cd39 GR |
463 | } |
464 | ||
4664680c GR |
465 | static void process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg) |
466 | { | |
256f5b24 GJ |
467 | if (WARN(mddev->cluster_info->slot_number - 1 == le32_to_cpu(msg->slot), |
468 | "node %d received it's own msg\n", le32_to_cpu(msg->slot))) | |
469 | return; | |
cf97a348 | 470 | switch (le32_to_cpu(msg->type)) { |
4664680c | 471 | case METADATA_UPDATED: |
1aee41f6 | 472 | process_metadata_update(mddev, msg); |
4664680c GR |
473 | break; |
474 | case RESYNCING: | |
cf97a348 GJ |
475 | process_suspend_info(mddev, le32_to_cpu(msg->slot), |
476 | le64_to_cpu(msg->low), | |
477 | le64_to_cpu(msg->high)); | |
4664680c | 478 | break; |
1aee41f6 | 479 | case NEWDISK: |
1aee41f6 | 480 | process_add_new_disk(mddev, msg); |
88bcfef7 GR |
481 | break; |
482 | case REMOVE: | |
88bcfef7 GR |
483 | process_remove_disk(mddev, msg); |
484 | break; | |
97f6cd39 | 485 | case RE_ADD: |
97f6cd39 GR |
486 | process_readd_disk(mddev, msg); |
487 | break; | |
dc737d7c | 488 | case BITMAP_NEEDS_SYNC: |
cf97a348 | 489 | __recover_slot(mddev, le32_to_cpu(msg->slot)); |
dc737d7c | 490 | break; |
88bcfef7 GR |
491 | default: |
492 | pr_warn("%s:%d Received unknown message from %d\n", | |
493 | __func__, __LINE__, msg->slot); | |
09dd1af2 | 494 | } |
4664680c GR |
495 | } |
496 | ||
497 | /* | |
498 | * thread for receiving message | |
499 | */ | |
500 | static void recv_daemon(struct md_thread *thread) | |
501 | { | |
502 | struct md_cluster_info *cinfo = thread->mddev->cluster_info; | |
503 | struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres; | |
504 | struct dlm_lock_resource *message_lockres = cinfo->message_lockres; | |
505 | struct cluster_msg msg; | |
b5ef5678 | 506 | int ret; |
4664680c GR |
507 | |
508 | /*get CR on Message*/ | |
509 | if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) { | |
510 | pr_err("md/raid1:failed to get CR on MESSAGE\n"); | |
511 | return; | |
512 | } | |
513 | ||
514 | /* read lvb and wake up thread to process this message_lockres */ | |
515 | memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg)); | |
516 | process_recvd_msg(thread->mddev, &msg); | |
517 | ||
518 | /*release CR on ack_lockres*/ | |
b5ef5678 GJ |
519 | ret = dlm_unlock_sync(ack_lockres); |
520 | if (unlikely(ret != 0)) | |
521 | pr_info("unlock ack failed return %d\n", ret); | |
66099bb0 | 522 | /*up-convert to PR on message_lockres*/ |
b5ef5678 GJ |
523 | ret = dlm_lock_sync(message_lockres, DLM_LOCK_PR); |
524 | if (unlikely(ret != 0)) | |
525 | pr_info("lock PR on msg failed return %d\n", ret); | |
4664680c | 526 | /*get CR on ack_lockres again*/ |
b5ef5678 GJ |
527 | ret = dlm_lock_sync(ack_lockres, DLM_LOCK_CR); |
528 | if (unlikely(ret != 0)) | |
529 | pr_info("lock CR on ack failed return %d\n", ret); | |
4664680c | 530 | /*release CR on message_lockres*/ |
b5ef5678 GJ |
531 | ret = dlm_unlock_sync(message_lockres); |
532 | if (unlikely(ret != 0)) | |
533 | pr_info("unlock msg failed return %d\n", ret); | |
4664680c GR |
534 | } |
535 | ||
601b515c GR |
536 | /* lock_comm() |
537 | * Takes the lock on the TOKEN lock resource so no other | |
538 | * node can communicate while the operation is underway. | |
dbb64f86 GR |
539 | * If called again, and the TOKEN lock is alread in EX mode |
540 | * return success. However, care must be taken that unlock_comm() | |
541 | * is called only once. | |
601b515c GR |
542 | */ |
543 | static int lock_comm(struct md_cluster_info *cinfo) | |
544 | { | |
545 | int error; | |
546 | ||
dbb64f86 GR |
547 | if (cinfo->token_lockres->mode == DLM_LOCK_EX) |
548 | return 0; | |
549 | ||
601b515c GR |
550 | error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX); |
551 | if (error) | |
552 | pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n", | |
553 | __func__, __LINE__, error); | |
554 | return error; | |
555 | } | |
556 | ||
557 | static void unlock_comm(struct md_cluster_info *cinfo) | |
558 | { | |
dbb64f86 | 559 | WARN_ON(cinfo->token_lockres->mode != DLM_LOCK_EX); |
601b515c GR |
560 | dlm_unlock_sync(cinfo->token_lockres); |
561 | } | |
562 | ||
563 | /* __sendmsg() | |
564 | * This function performs the actual sending of the message. This function is | |
565 | * usually called after performing the encompassing operation | |
566 | * The function: | |
567 | * 1. Grabs the message lockresource in EX mode | |
568 | * 2. Copies the message to the message LVB | |
66099bb0 | 569 | * 3. Downconverts message lockresource to CW |
601b515c GR |
570 | * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes |
571 | * and the other nodes read the message. The thread will wait here until all other | |
572 | * nodes have released ack lock resource. | |
573 | * 5. Downconvert ack lockresource to CR | |
574 | */ | |
575 | static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
576 | { | |
577 | int error; | |
578 | int slot = cinfo->slot_number - 1; | |
579 | ||
580 | cmsg->slot = cpu_to_le32(slot); | |
581 | /*get EX on Message*/ | |
582 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX); | |
583 | if (error) { | |
584 | pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error); | |
585 | goto failed_message; | |
586 | } | |
587 | ||
588 | memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg, | |
589 | sizeof(struct cluster_msg)); | |
66099bb0 GJ |
590 | /*down-convert EX to CW on Message*/ |
591 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CW); | |
601b515c | 592 | if (error) { |
66099bb0 | 593 | pr_err("md-cluster: failed to convert EX to CW on MESSAGE(%d)\n", |
601b515c | 594 | error); |
66099bb0 | 595 | goto failed_ack; |
601b515c GR |
596 | } |
597 | ||
598 | /*up-convert CR to EX on Ack*/ | |
599 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX); | |
600 | if (error) { | |
601 | pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n", | |
602 | error); | |
603 | goto failed_ack; | |
604 | } | |
605 | ||
606 | /*down-convert EX to CR on Ack*/ | |
607 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR); | |
608 | if (error) { | |
609 | pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n", | |
610 | error); | |
611 | goto failed_ack; | |
612 | } | |
613 | ||
614 | failed_ack: | |
b5ef5678 GJ |
615 | error = dlm_unlock_sync(cinfo->message_lockres); |
616 | if (unlikely(error != 0)) { | |
617 | pr_err("md-cluster: failed convert to NL on MESSAGE(%d)\n", | |
618 | error); | |
619 | /* in case the message can't be released due to some reason */ | |
620 | goto failed_ack; | |
621 | } | |
601b515c GR |
622 | failed_message: |
623 | return error; | |
624 | } | |
625 | ||
626 | static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
627 | { | |
628 | int ret; | |
629 | ||
630 | lock_comm(cinfo); | |
631 | ret = __sendmsg(cinfo, cmsg); | |
632 | unlock_comm(cinfo); | |
633 | return ret; | |
634 | } | |
635 | ||
96ae923a GR |
636 | static int gather_all_resync_info(struct mddev *mddev, int total_slots) |
637 | { | |
638 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
639 | int i, ret = 0; | |
640 | struct dlm_lock_resource *bm_lockres; | |
641 | struct suspend_info *s; | |
642 | char str[64]; | |
abb9b22a | 643 | sector_t lo, hi; |
96ae923a GR |
644 | |
645 | ||
646 | for (i = 0; i < total_slots; i++) { | |
647 | memset(str, '\0', 64); | |
648 | snprintf(str, 64, "bitmap%04d", i); | |
649 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
650 | if (!bm_lockres) | |
651 | return -ENOMEM; | |
652 | if (i == (cinfo->slot_number - 1)) | |
653 | continue; | |
654 | ||
655 | bm_lockres->flags |= DLM_LKF_NOQUEUE; | |
656 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
657 | if (ret == -EAGAIN) { | |
658 | memset(bm_lockres->lksb.sb_lvbptr, '\0', LVB_SIZE); | |
659 | s = read_resync_info(mddev, bm_lockres); | |
660 | if (s) { | |
661 | pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n", | |
662 | __func__, __LINE__, | |
663 | (unsigned long long) s->lo, | |
664 | (unsigned long long) s->hi, i); | |
665 | spin_lock_irq(&cinfo->suspend_lock); | |
666 | s->slot = i; | |
667 | list_add(&s->list, &cinfo->suspend_list); | |
668 | spin_unlock_irq(&cinfo->suspend_lock); | |
669 | } | |
670 | ret = 0; | |
671 | lockres_free(bm_lockres); | |
672 | continue; | |
673 | } | |
6e6d9f2c GJ |
674 | if (ret) { |
675 | lockres_free(bm_lockres); | |
96ae923a | 676 | goto out; |
6e6d9f2c | 677 | } |
abb9b22a GJ |
678 | |
679 | /* Read the disk bitmap sb and check if it needs recovery */ | |
680 | ret = bitmap_copy_from_slot(mddev, i, &lo, &hi, false); | |
681 | if (ret) { | |
682 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", i); | |
683 | lockres_free(bm_lockres); | |
684 | continue; | |
685 | } | |
686 | if ((hi > 0) && (lo < mddev->recovery_cp)) { | |
687 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
688 | mddev->recovery_cp = lo; | |
689 | md_check_recovery(mddev); | |
690 | } | |
691 | ||
96ae923a GR |
692 | dlm_unlock_sync(bm_lockres); |
693 | lockres_free(bm_lockres); | |
694 | } | |
695 | out: | |
696 | return ret; | |
697 | } | |
698 | ||
edb39c9d GR |
699 | static int join(struct mddev *mddev, int nodes) |
700 | { | |
c4ce867f | 701 | struct md_cluster_info *cinfo; |
cf921cc1 | 702 | int ret, ops_rv; |
c4ce867f GR |
703 | char str[64]; |
704 | ||
c4ce867f GR |
705 | cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL); |
706 | if (!cinfo) | |
707 | return -ENOMEM; | |
708 | ||
9e3072e3 GJ |
709 | INIT_LIST_HEAD(&cinfo->suspend_list); |
710 | spin_lock_init(&cinfo->suspend_lock); | |
cf921cc1 | 711 | init_completion(&cinfo->completion); |
eece075c | 712 | set_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state); |
cf921cc1 | 713 | |
cf921cc1 GR |
714 | mddev->cluster_info = cinfo; |
715 | ||
c4ce867f | 716 | memset(str, 0, 64); |
b89f704a | 717 | sprintf(str, "%pU", mddev->uuid); |
cf921cc1 GR |
718 | ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name, |
719 | DLM_LSFL_FS, LVB_SIZE, | |
720 | &md_ls_ops, mddev, &ops_rv, &cinfo->lockspace); | |
c4ce867f GR |
721 | if (ret) |
722 | goto err; | |
cf921cc1 | 723 | wait_for_completion(&cinfo->completion); |
8c58f02e GJ |
724 | if (nodes < cinfo->slot_number) { |
725 | pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).", | |
726 | cinfo->slot_number, nodes); | |
b97e9257 GR |
727 | ret = -ERANGE; |
728 | goto err; | |
729 | } | |
4664680c GR |
730 | /* Initiate the communication resources */ |
731 | ret = -ENOMEM; | |
732 | cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv"); | |
733 | if (!cinfo->recv_thread) { | |
734 | pr_err("md-cluster: cannot allocate memory for recv_thread!\n"); | |
735 | goto err; | |
736 | } | |
737 | cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1); | |
738 | if (!cinfo->message_lockres) | |
739 | goto err; | |
740 | cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0); | |
741 | if (!cinfo->token_lockres) | |
742 | goto err; | |
743 | cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0); | |
744 | if (!cinfo->ack_lockres) | |
745 | goto err; | |
1aee41f6 GR |
746 | cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0); |
747 | if (!cinfo->no_new_dev_lockres) | |
748 | goto err; | |
749 | ||
4664680c GR |
750 | /* get sync CR lock on ACK. */ |
751 | if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR)) | |
752 | pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n", | |
753 | ret); | |
1aee41f6 GR |
754 | /* get sync CR lock on no-new-dev. */ |
755 | if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR)) | |
756 | pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret); | |
757 | ||
54519c5f GR |
758 | |
759 | pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number); | |
760 | snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1); | |
761 | cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1); | |
762 | if (!cinfo->bitmap_lockres) | |
763 | goto err; | |
764 | if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) { | |
765 | pr_err("Failed to get bitmap lock\n"); | |
766 | ret = -EINVAL; | |
767 | goto err; | |
768 | } | |
769 | ||
c186b128 GR |
770 | cinfo->resync_lockres = lockres_init(mddev, "resync", NULL, 0); |
771 | if (!cinfo->resync_lockres) | |
772 | goto err; | |
773 | ||
96ae923a GR |
774 | ret = gather_all_resync_info(mddev, nodes); |
775 | if (ret) | |
776 | goto err; | |
777 | ||
edb39c9d | 778 | return 0; |
c4ce867f | 779 | err: |
4664680c GR |
780 | lockres_free(cinfo->message_lockres); |
781 | lockres_free(cinfo->token_lockres); | |
782 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 783 | lockres_free(cinfo->no_new_dev_lockres); |
c186b128 | 784 | lockres_free(cinfo->resync_lockres); |
96ae923a | 785 | lockres_free(cinfo->bitmap_lockres); |
c4ce867f GR |
786 | if (cinfo->lockspace) |
787 | dlm_release_lockspace(cinfo->lockspace, 2); | |
cf921cc1 | 788 | mddev->cluster_info = NULL; |
c4ce867f | 789 | kfree(cinfo); |
c4ce867f | 790 | return ret; |
edb39c9d GR |
791 | } |
792 | ||
09995411 GJ |
793 | static void resync_bitmap(struct mddev *mddev) |
794 | { | |
795 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
796 | struct cluster_msg cmsg = {0}; | |
797 | int err; | |
798 | ||
799 | cmsg.type = cpu_to_le32(BITMAP_NEEDS_SYNC); | |
800 | err = sendmsg(cinfo, &cmsg); | |
801 | if (err) | |
802 | pr_err("%s:%d: failed to send BITMAP_NEEDS_SYNC message (%d)\n", | |
803 | __func__, __LINE__, err); | |
804 | } | |
805 | ||
edb39c9d GR |
806 | static int leave(struct mddev *mddev) |
807 | { | |
c4ce867f GR |
808 | struct md_cluster_info *cinfo = mddev->cluster_info; |
809 | ||
810 | if (!cinfo) | |
811 | return 0; | |
09995411 GJ |
812 | |
813 | /* BITMAP_NEEDS_SYNC message should be sent when node | |
814 | * is leaving the cluster with dirty bitmap, also we | |
815 | * can only deliver it when dlm connection is available */ | |
816 | if (cinfo->slot_number > 0 && mddev->recovery_cp != MaxSector) | |
817 | resync_bitmap(mddev); | |
818 | ||
e94987db | 819 | md_unregister_thread(&cinfo->recovery_thread); |
4664680c GR |
820 | md_unregister_thread(&cinfo->recv_thread); |
821 | lockres_free(cinfo->message_lockres); | |
822 | lockres_free(cinfo->token_lockres); | |
823 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 824 | lockres_free(cinfo->no_new_dev_lockres); |
54519c5f | 825 | lockres_free(cinfo->bitmap_lockres); |
c4ce867f | 826 | dlm_release_lockspace(cinfo->lockspace, 2); |
edb39c9d GR |
827 | return 0; |
828 | } | |
829 | ||
cf921cc1 GR |
830 | /* slot_number(): Returns the MD slot number to use |
831 | * DLM starts the slot numbers from 1, wheras cluster-md | |
832 | * wants the number to be from zero, so we deduct one | |
833 | */ | |
834 | static int slot_number(struct mddev *mddev) | |
835 | { | |
836 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
837 | ||
838 | return cinfo->slot_number - 1; | |
839 | } | |
840 | ||
293467aa GR |
841 | static int metadata_update_start(struct mddev *mddev) |
842 | { | |
843 | return lock_comm(mddev->cluster_info); | |
844 | } | |
845 | ||
846 | static int metadata_update_finish(struct mddev *mddev) | |
847 | { | |
848 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
849 | struct cluster_msg cmsg; | |
70bcecdb GR |
850 | struct md_rdev *rdev; |
851 | int ret = 0; | |
ba2746b0 | 852 | int raid_slot = -1; |
293467aa GR |
853 | |
854 | memset(&cmsg, 0, sizeof(cmsg)); | |
855 | cmsg.type = cpu_to_le32(METADATA_UPDATED); | |
70bcecdb GR |
856 | /* Pick up a good active device number to send. |
857 | */ | |
858 | rdev_for_each(rdev, mddev) | |
859 | if (rdev->raid_disk > -1 && !test_bit(Faulty, &rdev->flags)) { | |
ba2746b0 | 860 | raid_slot = rdev->desc_nr; |
70bcecdb GR |
861 | break; |
862 | } | |
ba2746b0 N |
863 | if (raid_slot >= 0) { |
864 | cmsg.raid_slot = cpu_to_le32(raid_slot); | |
70bcecdb | 865 | ret = __sendmsg(cinfo, &cmsg); |
ba2746b0 | 866 | } else |
70bcecdb | 867 | pr_warn("md-cluster: No good device id found to send\n"); |
293467aa GR |
868 | unlock_comm(cinfo); |
869 | return ret; | |
870 | } | |
871 | ||
dbb64f86 | 872 | static void metadata_update_cancel(struct mddev *mddev) |
293467aa GR |
873 | { |
874 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
dbb64f86 | 875 | unlock_comm(cinfo); |
293467aa GR |
876 | } |
877 | ||
c186b128 GR |
878 | static int resync_start(struct mddev *mddev) |
879 | { | |
880 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
881 | cinfo->resync_lockres->flags |= DLM_LKF_NOQUEUE; | |
882 | return dlm_lock_sync(cinfo->resync_lockres, DLM_LOCK_EX); | |
883 | } | |
884 | ||
c40f341f | 885 | static int resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi) |
965400eb GR |
886 | { |
887 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
ac277c6a | 888 | struct resync_info ri; |
aee177ac | 889 | struct cluster_msg cmsg = {0}; |
965400eb | 890 | |
ac277c6a GR |
891 | /* do not send zero again, if we have sent before */ |
892 | if (hi == 0) { | |
893 | memcpy(&ri, cinfo->bitmap_lockres->lksb.sb_lvbptr, sizeof(struct resync_info)); | |
894 | if (le64_to_cpu(ri.hi) == 0) | |
895 | return 0; | |
896 | } | |
897 | ||
30661b49 | 898 | add_resync_info(cinfo->bitmap_lockres, lo, hi); |
c40f341f GR |
899 | /* Re-acquire the lock to refresh LVB */ |
900 | dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW); | |
c40f341f | 901 | cmsg.type = cpu_to_le32(RESYNCING); |
965400eb GR |
902 | cmsg.low = cpu_to_le64(lo); |
903 | cmsg.high = cpu_to_le64(hi); | |
c186b128 | 904 | |
965400eb GR |
905 | return sendmsg(cinfo, &cmsg); |
906 | } | |
907 | ||
c186b128 GR |
908 | static int resync_finish(struct mddev *mddev) |
909 | { | |
910 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
911 | cinfo->resync_lockres->flags &= ~DLM_LKF_NOQUEUE; | |
912 | dlm_unlock_sync(cinfo->resync_lockres); | |
913 | return resync_info_update(mddev, 0, 0); | |
914 | } | |
915 | ||
90382ed9 GR |
916 | static int area_resyncing(struct mddev *mddev, int direction, |
917 | sector_t lo, sector_t hi) | |
589a1c49 GR |
918 | { |
919 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
920 | int ret = 0; | |
921 | struct suspend_info *s; | |
922 | ||
90382ed9 GR |
923 | if ((direction == READ) && |
924 | test_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state)) | |
925 | return 1; | |
926 | ||
589a1c49 GR |
927 | spin_lock_irq(&cinfo->suspend_lock); |
928 | if (list_empty(&cinfo->suspend_list)) | |
929 | goto out; | |
930 | list_for_each_entry(s, &cinfo->suspend_list, list) | |
931 | if (hi > s->lo && lo < s->hi) { | |
932 | ret = 1; | |
933 | break; | |
934 | } | |
935 | out: | |
936 | spin_unlock_irq(&cinfo->suspend_lock); | |
937 | return ret; | |
938 | } | |
939 | ||
dbb64f86 GR |
940 | /* add_new_disk() - initiates a disk add |
941 | * However, if this fails before writing md_update_sb(), | |
942 | * add_new_disk_cancel() must be called to release token lock | |
943 | */ | |
944 | static int add_new_disk(struct mddev *mddev, struct md_rdev *rdev) | |
1aee41f6 GR |
945 | { |
946 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
947 | struct cluster_msg cmsg; | |
948 | int ret = 0; | |
949 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); | |
950 | char *uuid = sb->device_uuid; | |
951 | ||
952 | memset(&cmsg, 0, sizeof(cmsg)); | |
953 | cmsg.type = cpu_to_le32(NEWDISK); | |
954 | memcpy(cmsg.uuid, uuid, 16); | |
faeff83f | 955 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); |
1aee41f6 GR |
956 | lock_comm(cinfo); |
957 | ret = __sendmsg(cinfo, &cmsg); | |
958 | if (ret) | |
959 | return ret; | |
960 | cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE; | |
961 | ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX); | |
962 | cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE; | |
963 | /* Some node does not "see" the device */ | |
964 | if (ret == -EAGAIN) | |
965 | ret = -ENOENT; | |
dbb64f86 GR |
966 | if (ret) |
967 | unlock_comm(cinfo); | |
1aee41f6 GR |
968 | else |
969 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); | |
970 | return ret; | |
971 | } | |
972 | ||
dbb64f86 | 973 | static void add_new_disk_cancel(struct mddev *mddev) |
1aee41f6 | 974 | { |
dbb64f86 GR |
975 | struct md_cluster_info *cinfo = mddev->cluster_info; |
976 | unlock_comm(cinfo); | |
1aee41f6 GR |
977 | } |
978 | ||
fa8259da | 979 | static int new_disk_ack(struct mddev *mddev, bool ack) |
1aee41f6 GR |
980 | { |
981 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
982 | ||
fa8259da GR |
983 | if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) { |
984 | pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev)); | |
985 | return -EINVAL; | |
986 | } | |
987 | ||
1aee41f6 GR |
988 | if (ack) |
989 | dlm_unlock_sync(cinfo->no_new_dev_lockres); | |
990 | complete(&cinfo->newdisk_completion); | |
fa8259da | 991 | return 0; |
1aee41f6 GR |
992 | } |
993 | ||
88bcfef7 GR |
994 | static int remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
995 | { | |
aee177ac | 996 | struct cluster_msg cmsg = {0}; |
88bcfef7 | 997 | struct md_cluster_info *cinfo = mddev->cluster_info; |
faeff83f GJ |
998 | cmsg.type = cpu_to_le32(REMOVE); |
999 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); | |
54a88392 | 1000 | return sendmsg(cinfo, &cmsg); |
88bcfef7 GR |
1001 | } |
1002 | ||
97f6cd39 GR |
1003 | static int gather_bitmaps(struct md_rdev *rdev) |
1004 | { | |
1005 | int sn, err; | |
1006 | sector_t lo, hi; | |
aee177ac | 1007 | struct cluster_msg cmsg = {0}; |
97f6cd39 GR |
1008 | struct mddev *mddev = rdev->mddev; |
1009 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1010 | ||
faeff83f GJ |
1011 | cmsg.type = cpu_to_le32(RE_ADD); |
1012 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); | |
97f6cd39 GR |
1013 | err = sendmsg(cinfo, &cmsg); |
1014 | if (err) | |
1015 | goto out; | |
1016 | ||
1017 | for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) { | |
1018 | if (sn == (cinfo->slot_number - 1)) | |
1019 | continue; | |
1020 | err = bitmap_copy_from_slot(mddev, sn, &lo, &hi, false); | |
1021 | if (err) { | |
1022 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn); | |
1023 | goto out; | |
1024 | } | |
1025 | if ((hi > 0) && (lo < mddev->recovery_cp)) | |
1026 | mddev->recovery_cp = lo; | |
1027 | } | |
1028 | out: | |
1029 | return err; | |
1030 | } | |
1031 | ||
edb39c9d GR |
1032 | static struct md_cluster_operations cluster_ops = { |
1033 | .join = join, | |
1034 | .leave = leave, | |
cf921cc1 | 1035 | .slot_number = slot_number, |
c186b128 GR |
1036 | .resync_start = resync_start, |
1037 | .resync_finish = resync_finish, | |
96ae923a | 1038 | .resync_info_update = resync_info_update, |
293467aa GR |
1039 | .metadata_update_start = metadata_update_start, |
1040 | .metadata_update_finish = metadata_update_finish, | |
1041 | .metadata_update_cancel = metadata_update_cancel, | |
589a1c49 | 1042 | .area_resyncing = area_resyncing, |
dbb64f86 GR |
1043 | .add_new_disk = add_new_disk, |
1044 | .add_new_disk_cancel = add_new_disk_cancel, | |
1aee41f6 | 1045 | .new_disk_ack = new_disk_ack, |
88bcfef7 | 1046 | .remove_disk = remove_disk, |
97f6cd39 | 1047 | .gather_bitmaps = gather_bitmaps, |
edb39c9d GR |
1048 | }; |
1049 | ||
8e854e9c GR |
1050 | static int __init cluster_init(void) |
1051 | { | |
1052 | pr_warn("md-cluster: EXPERIMENTAL. Use with caution\n"); | |
1053 | pr_info("Registering Cluster MD functions\n"); | |
edb39c9d | 1054 | register_md_cluster_operations(&cluster_ops, THIS_MODULE); |
8e854e9c GR |
1055 | return 0; |
1056 | } | |
1057 | ||
1058 | static void cluster_exit(void) | |
1059 | { | |
edb39c9d | 1060 | unregister_md_cluster_operations(); |
8e854e9c GR |
1061 | } |
1062 | ||
1063 | module_init(cluster_init); | |
1064 | module_exit(cluster_exit); | |
86b57277 | 1065 | MODULE_AUTHOR("SUSE"); |
8e854e9c GR |
1066 | MODULE_LICENSE("GPL"); |
1067 | MODULE_DESCRIPTION("Clustering support for MD"); |