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 | 50 | |
8b9277c8 GJ |
51 | /* Lock the send communication. This is done through |
52 | * bit manipulation as opposed to a mutex in order to | |
53 | * accomodate lock and hold. See next comment. | |
54 | */ | |
55 | #define MD_CLUSTER_SEND_LOCK 4 | |
e19508fa GJ |
56 | /* If cluster operations (such as adding a disk) must lock the |
57 | * communication channel, so as to perform extra operations | |
58 | * (update metadata) and no other operation is allowed on the | |
59 | * MD. Token needs to be locked and held until the operation | |
60 | * completes witha md_update_sb(), which would eventually release | |
61 | * the lock. | |
8b9277c8 GJ |
62 | */ |
63 | #define MD_CLUSTER_SEND_LOCKED_ALREADY 5 | |
64 | ||
fa8259da | 65 | |
c4ce867f GR |
66 | struct md_cluster_info { |
67 | /* dlm lock space and resources for clustered raid. */ | |
68 | dlm_lockspace_t *lockspace; | |
cf921cc1 GR |
69 | int slot_number; |
70 | struct completion completion; | |
8b9277c8 | 71 | struct mutex recv_mutex; |
54519c5f | 72 | struct dlm_lock_resource *bitmap_lockres; |
f6a2dc64 | 73 | struct dlm_lock_resource **other_bitmap_lockres; |
c186b128 | 74 | struct dlm_lock_resource *resync_lockres; |
96ae923a GR |
75 | struct list_head suspend_list; |
76 | spinlock_t suspend_lock; | |
e94987db GR |
77 | struct md_thread *recovery_thread; |
78 | unsigned long recovery_map; | |
4664680c GR |
79 | /* communication loc resources */ |
80 | struct dlm_lock_resource *ack_lockres; | |
81 | struct dlm_lock_resource *message_lockres; | |
82 | struct dlm_lock_resource *token_lockres; | |
1aee41f6 | 83 | struct dlm_lock_resource *no_new_dev_lockres; |
4664680c | 84 | struct md_thread *recv_thread; |
1aee41f6 | 85 | struct completion newdisk_completion; |
8b9277c8 | 86 | wait_queue_head_t wait; |
fa8259da | 87 | unsigned long state; |
4664680c GR |
88 | }; |
89 | ||
90 | enum msg_type { | |
91 | METADATA_UPDATED = 0, | |
92 | RESYNCING, | |
1aee41f6 | 93 | NEWDISK, |
88bcfef7 | 94 | REMOVE, |
97f6cd39 | 95 | RE_ADD, |
dc737d7c | 96 | BITMAP_NEEDS_SYNC, |
4664680c GR |
97 | }; |
98 | ||
99 | struct cluster_msg { | |
cf97a348 GJ |
100 | __le32 type; |
101 | __le32 slot; | |
1aee41f6 | 102 | /* TODO: Unionize this for smaller footprint */ |
cf97a348 GJ |
103 | __le64 low; |
104 | __le64 high; | |
1aee41f6 | 105 | char uuid[16]; |
cf97a348 | 106 | __le32 raid_slot; |
47741b7c GR |
107 | }; |
108 | ||
109 | static void sync_ast(void *arg) | |
110 | { | |
111 | struct dlm_lock_resource *res; | |
112 | ||
2e2a7cd9 | 113 | res = arg; |
47741b7c GR |
114 | complete(&res->completion); |
115 | } | |
116 | ||
117 | static int dlm_lock_sync(struct dlm_lock_resource *res, int mode) | |
118 | { | |
119 | int ret = 0; | |
120 | ||
47741b7c GR |
121 | ret = dlm_lock(res->ls, mode, &res->lksb, |
122 | res->flags, res->name, strlen(res->name), | |
123 | 0, sync_ast, res, res->bast); | |
124 | if (ret) | |
125 | return ret; | |
126 | wait_for_completion(&res->completion); | |
dbb64f86 GR |
127 | if (res->lksb.sb_status == 0) |
128 | res->mode = mode; | |
47741b7c GR |
129 | return res->lksb.sb_status; |
130 | } | |
131 | ||
132 | static int dlm_unlock_sync(struct dlm_lock_resource *res) | |
133 | { | |
134 | return dlm_lock_sync(res, DLM_LOCK_NL); | |
135 | } | |
136 | ||
c4ce867f | 137 | static struct dlm_lock_resource *lockres_init(struct mddev *mddev, |
47741b7c GR |
138 | char *name, void (*bastfn)(void *arg, int mode), int with_lvb) |
139 | { | |
140 | struct dlm_lock_resource *res = NULL; | |
141 | int ret, namelen; | |
c4ce867f | 142 | struct md_cluster_info *cinfo = mddev->cluster_info; |
47741b7c GR |
143 | |
144 | res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL); | |
145 | if (!res) | |
146 | return NULL; | |
b83d51c0 | 147 | init_completion(&res->completion); |
c4ce867f GR |
148 | res->ls = cinfo->lockspace; |
149 | res->mddev = mddev; | |
dbb64f86 | 150 | res->mode = DLM_LOCK_IV; |
47741b7c GR |
151 | namelen = strlen(name); |
152 | res->name = kzalloc(namelen + 1, GFP_KERNEL); | |
153 | if (!res->name) { | |
154 | pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name); | |
155 | goto out_err; | |
156 | } | |
157 | strlcpy(res->name, name, namelen + 1); | |
158 | if (with_lvb) { | |
159 | res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL); | |
160 | if (!res->lksb.sb_lvbptr) { | |
161 | pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name); | |
162 | goto out_err; | |
163 | } | |
164 | res->flags = DLM_LKF_VALBLK; | |
165 | } | |
166 | ||
167 | if (bastfn) | |
168 | res->bast = bastfn; | |
169 | ||
170 | res->flags |= DLM_LKF_EXPEDITE; | |
171 | ||
172 | ret = dlm_lock_sync(res, DLM_LOCK_NL); | |
173 | if (ret) { | |
174 | pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name); | |
175 | goto out_err; | |
176 | } | |
177 | res->flags &= ~DLM_LKF_EXPEDITE; | |
178 | res->flags |= DLM_LKF_CONVERT; | |
179 | ||
180 | return res; | |
181 | out_err: | |
182 | kfree(res->lksb.sb_lvbptr); | |
183 | kfree(res->name); | |
184 | kfree(res); | |
185 | return NULL; | |
186 | } | |
187 | ||
188 | static void lockres_free(struct dlm_lock_resource *res) | |
189 | { | |
b5ef5678 GJ |
190 | int ret; |
191 | ||
47741b7c GR |
192 | if (!res) |
193 | return; | |
194 | ||
b5ef5678 GJ |
195 | /* cancel a lock request or a conversion request that is blocked */ |
196 | res->flags |= DLM_LKF_CANCEL; | |
197 | retry: | |
198 | ret = dlm_unlock(res->ls, res->lksb.sb_lkid, 0, &res->lksb, res); | |
199 | if (unlikely(ret != 0)) { | |
200 | pr_info("%s: failed to unlock %s return %d\n", __func__, res->name, ret); | |
201 | ||
202 | /* if a lock conversion is cancelled, then the lock is put | |
203 | * back to grant queue, need to ensure it is unlocked */ | |
204 | if (ret == -DLM_ECANCEL) | |
205 | goto retry; | |
206 | } | |
207 | res->flags &= ~DLM_LKF_CANCEL; | |
47741b7c GR |
208 | wait_for_completion(&res->completion); |
209 | ||
210 | kfree(res->name); | |
211 | kfree(res->lksb.sb_lvbptr); | |
212 | kfree(res); | |
213 | } | |
8e854e9c | 214 | |
30661b49 N |
215 | static void add_resync_info(struct dlm_lock_resource *lockres, |
216 | sector_t lo, sector_t hi) | |
96ae923a GR |
217 | { |
218 | struct resync_info *ri; | |
219 | ||
220 | ri = (struct resync_info *)lockres->lksb.sb_lvbptr; | |
221 | ri->lo = cpu_to_le64(lo); | |
222 | ri->hi = cpu_to_le64(hi); | |
223 | } | |
224 | ||
225 | static struct suspend_info *read_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres) | |
226 | { | |
227 | struct resync_info ri; | |
228 | struct suspend_info *s = NULL; | |
229 | sector_t hi = 0; | |
230 | ||
231 | dlm_lock_sync(lockres, DLM_LOCK_CR); | |
232 | memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info)); | |
233 | hi = le64_to_cpu(ri.hi); | |
cf97a348 | 234 | if (hi > 0) { |
96ae923a GR |
235 | s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL); |
236 | if (!s) | |
237 | goto out; | |
238 | s->hi = hi; | |
239 | s->lo = le64_to_cpu(ri.lo); | |
240 | } | |
241 | dlm_unlock_sync(lockres); | |
242 | out: | |
243 | return s; | |
244 | } | |
245 | ||
6dc69c9c | 246 | static void recover_bitmaps(struct md_thread *thread) |
e94987db GR |
247 | { |
248 | struct mddev *mddev = thread->mddev; | |
249 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
250 | struct dlm_lock_resource *bm_lockres; | |
251 | char str[64]; | |
252 | int slot, ret; | |
253 | struct suspend_info *s, *tmp; | |
254 | sector_t lo, hi; | |
255 | ||
256 | while (cinfo->recovery_map) { | |
257 | slot = fls64((u64)cinfo->recovery_map) - 1; | |
258 | ||
259 | /* Clear suspend_area associated with the bitmap */ | |
260 | spin_lock_irq(&cinfo->suspend_lock); | |
261 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
262 | if (slot == s->slot) { | |
263 | list_del(&s->list); | |
264 | kfree(s); | |
265 | } | |
266 | spin_unlock_irq(&cinfo->suspend_lock); | |
267 | ||
268 | snprintf(str, 64, "bitmap%04d", slot); | |
269 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
270 | if (!bm_lockres) { | |
271 | pr_err("md-cluster: Cannot initialize bitmaps\n"); | |
272 | goto clear_bit; | |
273 | } | |
274 | ||
275 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
276 | if (ret) { | |
277 | pr_err("md-cluster: Could not DLM lock %s: %d\n", | |
278 | str, ret); | |
279 | goto clear_bit; | |
280 | } | |
97f6cd39 | 281 | ret = bitmap_copy_from_slot(mddev, slot, &lo, &hi, true); |
4b26a08a | 282 | if (ret) { |
e94987db | 283 | pr_err("md-cluster: Could not copy data from bitmap %d\n", slot); |
4b26a08a GR |
284 | goto dlm_unlock; |
285 | } | |
286 | if (hi > 0) { | |
287 | /* TODO:Wait for current resync to get over */ | |
288 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
289 | if (lo < mddev->recovery_cp) | |
290 | mddev->recovery_cp = lo; | |
291 | md_check_recovery(mddev); | |
292 | } | |
293 | dlm_unlock: | |
e94987db GR |
294 | dlm_unlock_sync(bm_lockres); |
295 | clear_bit: | |
296 | clear_bit(slot, &cinfo->recovery_map); | |
297 | } | |
298 | } | |
299 | ||
cf921cc1 GR |
300 | static void recover_prep(void *arg) |
301 | { | |
90382ed9 GR |
302 | struct mddev *mddev = arg; |
303 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
304 | set_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); | |
cf921cc1 GR |
305 | } |
306 | ||
05cd0e51 | 307 | static void __recover_slot(struct mddev *mddev, int slot) |
cf921cc1 | 308 | { |
cf921cc1 GR |
309 | struct md_cluster_info *cinfo = mddev->cluster_info; |
310 | ||
05cd0e51 | 311 | set_bit(slot, &cinfo->recovery_map); |
e94987db GR |
312 | if (!cinfo->recovery_thread) { |
313 | cinfo->recovery_thread = md_register_thread(recover_bitmaps, | |
314 | mddev, "recover"); | |
315 | if (!cinfo->recovery_thread) { | |
316 | pr_warn("md-cluster: Could not create recovery thread\n"); | |
317 | return; | |
318 | } | |
319 | } | |
320 | md_wakeup_thread(cinfo->recovery_thread); | |
cf921cc1 GR |
321 | } |
322 | ||
05cd0e51 GJ |
323 | static void recover_slot(void *arg, struct dlm_slot *slot) |
324 | { | |
325 | struct mddev *mddev = arg; | |
326 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
327 | ||
328 | pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n", | |
329 | mddev->bitmap_info.cluster_name, | |
330 | slot->nodeid, slot->slot, | |
331 | cinfo->slot_number); | |
332 | /* deduct one since dlm slot starts from one while the num of | |
333 | * cluster-md begins with 0 */ | |
334 | __recover_slot(mddev, slot->slot - 1); | |
335 | } | |
336 | ||
cf921cc1 GR |
337 | static void recover_done(void *arg, struct dlm_slot *slots, |
338 | int num_slots, int our_slot, | |
339 | uint32_t generation) | |
340 | { | |
341 | struct mddev *mddev = arg; | |
342 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
343 | ||
344 | cinfo->slot_number = our_slot; | |
eece075c GJ |
345 | /* completion is only need to be complete when node join cluster, |
346 | * it doesn't need to run during another node's failure */ | |
347 | if (test_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state)) { | |
348 | complete(&cinfo->completion); | |
349 | clear_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state); | |
350 | } | |
90382ed9 | 351 | clear_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); |
cf921cc1 GR |
352 | } |
353 | ||
eece075c GJ |
354 | /* the ops is called when node join the cluster, and do lock recovery |
355 | * if node failure occurs */ | |
cf921cc1 GR |
356 | static const struct dlm_lockspace_ops md_ls_ops = { |
357 | .recover_prep = recover_prep, | |
358 | .recover_slot = recover_slot, | |
359 | .recover_done = recover_done, | |
360 | }; | |
361 | ||
4664680c GR |
362 | /* |
363 | * The BAST function for the ack lock resource | |
364 | * This function wakes up the receive thread in | |
365 | * order to receive and process the message. | |
366 | */ | |
367 | static void ack_bast(void *arg, int mode) | |
368 | { | |
2e2a7cd9 | 369 | struct dlm_lock_resource *res = arg; |
4664680c GR |
370 | struct md_cluster_info *cinfo = res->mddev->cluster_info; |
371 | ||
372 | if (mode == DLM_LOCK_EX) | |
373 | md_wakeup_thread(cinfo->recv_thread); | |
374 | } | |
375 | ||
e59721cc GR |
376 | static void __remove_suspend_info(struct md_cluster_info *cinfo, int slot) |
377 | { | |
378 | struct suspend_info *s, *tmp; | |
379 | ||
380 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
381 | if (slot == s->slot) { | |
e59721cc GR |
382 | list_del(&s->list); |
383 | kfree(s); | |
384 | break; | |
385 | } | |
386 | } | |
387 | ||
b8ca846e | 388 | static void remove_suspend_info(struct mddev *mddev, int slot) |
e59721cc | 389 | { |
b8ca846e | 390 | struct md_cluster_info *cinfo = mddev->cluster_info; |
e59721cc GR |
391 | spin_lock_irq(&cinfo->suspend_lock); |
392 | __remove_suspend_info(cinfo, slot); | |
393 | spin_unlock_irq(&cinfo->suspend_lock); | |
b8ca846e | 394 | mddev->pers->quiesce(mddev, 2); |
e59721cc GR |
395 | } |
396 | ||
397 | ||
9ed38ff5 | 398 | static void process_suspend_info(struct mddev *mddev, |
e59721cc GR |
399 | int slot, sector_t lo, sector_t hi) |
400 | { | |
9ed38ff5 | 401 | struct md_cluster_info *cinfo = mddev->cluster_info; |
e59721cc GR |
402 | struct suspend_info *s; |
403 | ||
404 | if (!hi) { | |
b8ca846e | 405 | remove_suspend_info(mddev, slot); |
c186b128 GR |
406 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
407 | md_wakeup_thread(mddev->thread); | |
e59721cc GR |
408 | return; |
409 | } | |
410 | s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL); | |
411 | if (!s) | |
412 | return; | |
413 | s->slot = slot; | |
414 | s->lo = lo; | |
415 | s->hi = hi; | |
9ed38ff5 GR |
416 | mddev->pers->quiesce(mddev, 1); |
417 | mddev->pers->quiesce(mddev, 0); | |
e59721cc GR |
418 | spin_lock_irq(&cinfo->suspend_lock); |
419 | /* Remove existing entry (if exists) before adding */ | |
420 | __remove_suspend_info(cinfo, slot); | |
421 | list_add(&s->list, &cinfo->suspend_list); | |
422 | spin_unlock_irq(&cinfo->suspend_lock); | |
b8ca846e | 423 | mddev->pers->quiesce(mddev, 2); |
e59721cc GR |
424 | } |
425 | ||
1aee41f6 GR |
426 | static void process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg) |
427 | { | |
428 | char disk_uuid[64]; | |
429 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
430 | char event_name[] = "EVENT=ADD_DEVICE"; | |
431 | char raid_slot[16]; | |
432 | char *envp[] = {event_name, disk_uuid, raid_slot, NULL}; | |
433 | int len; | |
434 | ||
435 | len = snprintf(disk_uuid, 64, "DEVICE_UUID="); | |
b89f704a | 436 | sprintf(disk_uuid + len, "%pU", cmsg->uuid); |
faeff83f | 437 | snprintf(raid_slot, 16, "RAID_DISK=%d", le32_to_cpu(cmsg->raid_slot)); |
1aee41f6 GR |
438 | pr_info("%s:%d Sending kobject change with %s and %s\n", __func__, __LINE__, disk_uuid, raid_slot); |
439 | init_completion(&cinfo->newdisk_completion); | |
fa8259da | 440 | set_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
441 | kobject_uevent_env(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE, envp); |
442 | wait_for_completion_timeout(&cinfo->newdisk_completion, | |
443 | NEW_DEV_TIMEOUT); | |
fa8259da | 444 | clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
445 | } |
446 | ||
447 | ||
448 | static void process_metadata_update(struct mddev *mddev, struct cluster_msg *msg) | |
449 | { | |
450 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
15858fa5 GJ |
451 | mddev->good_device_nr = le32_to_cpu(msg->raid_slot); |
452 | set_bit(MD_RELOAD_SB, &mddev->flags); | |
1aee41f6 | 453 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); |
15858fa5 | 454 | md_wakeup_thread(mddev->thread); |
1aee41f6 GR |
455 | } |
456 | ||
88bcfef7 GR |
457 | static void process_remove_disk(struct mddev *mddev, struct cluster_msg *msg) |
458 | { | |
faeff83f GJ |
459 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, |
460 | le32_to_cpu(msg->raid_slot)); | |
88bcfef7 | 461 | |
659b254f GJ |
462 | if (rdev) { |
463 | set_bit(ClusterRemove, &rdev->flags); | |
464 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
465 | md_wakeup_thread(mddev->thread); | |
466 | } | |
88bcfef7 | 467 | else |
faeff83f GJ |
468 | pr_warn("%s: %d Could not find disk(%d) to REMOVE\n", |
469 | __func__, __LINE__, le32_to_cpu(msg->raid_slot)); | |
88bcfef7 GR |
470 | } |
471 | ||
97f6cd39 GR |
472 | static void process_readd_disk(struct mddev *mddev, struct cluster_msg *msg) |
473 | { | |
faeff83f GJ |
474 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, |
475 | le32_to_cpu(msg->raid_slot)); | |
97f6cd39 GR |
476 | |
477 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
478 | clear_bit(Faulty, &rdev->flags); | |
479 | else | |
faeff83f GJ |
480 | pr_warn("%s: %d Could not find disk(%d) which is faulty", |
481 | __func__, __LINE__, le32_to_cpu(msg->raid_slot)); | |
97f6cd39 GR |
482 | } |
483 | ||
4664680c GR |
484 | static void process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg) |
485 | { | |
256f5b24 GJ |
486 | if (WARN(mddev->cluster_info->slot_number - 1 == le32_to_cpu(msg->slot), |
487 | "node %d received it's own msg\n", le32_to_cpu(msg->slot))) | |
488 | return; | |
cf97a348 | 489 | switch (le32_to_cpu(msg->type)) { |
4664680c | 490 | case METADATA_UPDATED: |
1aee41f6 | 491 | process_metadata_update(mddev, msg); |
4664680c GR |
492 | break; |
493 | case RESYNCING: | |
cf97a348 GJ |
494 | process_suspend_info(mddev, le32_to_cpu(msg->slot), |
495 | le64_to_cpu(msg->low), | |
496 | le64_to_cpu(msg->high)); | |
4664680c | 497 | break; |
1aee41f6 | 498 | case NEWDISK: |
1aee41f6 | 499 | process_add_new_disk(mddev, msg); |
88bcfef7 GR |
500 | break; |
501 | case REMOVE: | |
88bcfef7 GR |
502 | process_remove_disk(mddev, msg); |
503 | break; | |
97f6cd39 | 504 | case RE_ADD: |
97f6cd39 GR |
505 | process_readd_disk(mddev, msg); |
506 | break; | |
dc737d7c | 507 | case BITMAP_NEEDS_SYNC: |
cf97a348 | 508 | __recover_slot(mddev, le32_to_cpu(msg->slot)); |
dc737d7c | 509 | break; |
88bcfef7 GR |
510 | default: |
511 | pr_warn("%s:%d Received unknown message from %d\n", | |
512 | __func__, __LINE__, msg->slot); | |
09dd1af2 | 513 | } |
4664680c GR |
514 | } |
515 | ||
516 | /* | |
517 | * thread for receiving message | |
518 | */ | |
519 | static void recv_daemon(struct md_thread *thread) | |
520 | { | |
521 | struct md_cluster_info *cinfo = thread->mddev->cluster_info; | |
522 | struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres; | |
523 | struct dlm_lock_resource *message_lockres = cinfo->message_lockres; | |
524 | struct cluster_msg msg; | |
b5ef5678 | 525 | int ret; |
4664680c | 526 | |
8b9277c8 | 527 | mutex_lock(&cinfo->recv_mutex); |
4664680c GR |
528 | /*get CR on Message*/ |
529 | if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) { | |
530 | pr_err("md/raid1:failed to get CR on MESSAGE\n"); | |
8b9277c8 | 531 | mutex_unlock(&cinfo->recv_mutex); |
4664680c GR |
532 | return; |
533 | } | |
534 | ||
535 | /* read lvb and wake up thread to process this message_lockres */ | |
536 | memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg)); | |
537 | process_recvd_msg(thread->mddev, &msg); | |
538 | ||
539 | /*release CR on ack_lockres*/ | |
b5ef5678 GJ |
540 | ret = dlm_unlock_sync(ack_lockres); |
541 | if (unlikely(ret != 0)) | |
542 | pr_info("unlock ack failed return %d\n", ret); | |
66099bb0 | 543 | /*up-convert to PR on message_lockres*/ |
b5ef5678 GJ |
544 | ret = dlm_lock_sync(message_lockres, DLM_LOCK_PR); |
545 | if (unlikely(ret != 0)) | |
546 | pr_info("lock PR on msg failed return %d\n", ret); | |
4664680c | 547 | /*get CR on ack_lockres again*/ |
b5ef5678 GJ |
548 | ret = dlm_lock_sync(ack_lockres, DLM_LOCK_CR); |
549 | if (unlikely(ret != 0)) | |
550 | pr_info("lock CR on ack failed return %d\n", ret); | |
4664680c | 551 | /*release CR on message_lockres*/ |
b5ef5678 GJ |
552 | ret = dlm_unlock_sync(message_lockres); |
553 | if (unlikely(ret != 0)) | |
554 | pr_info("unlock msg failed return %d\n", ret); | |
8b9277c8 | 555 | mutex_unlock(&cinfo->recv_mutex); |
4664680c GR |
556 | } |
557 | ||
8b9277c8 | 558 | /* lock_token() |
601b515c GR |
559 | * Takes the lock on the TOKEN lock resource so no other |
560 | * node can communicate while the operation is underway. | |
561 | */ | |
8b9277c8 | 562 | static int lock_token(struct md_cluster_info *cinfo) |
601b515c GR |
563 | { |
564 | int error; | |
565 | ||
566 | error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX); | |
567 | if (error) | |
568 | pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n", | |
569 | __func__, __LINE__, error); | |
8b9277c8 GJ |
570 | |
571 | /* Lock the receive sequence */ | |
572 | mutex_lock(&cinfo->recv_mutex); | |
601b515c GR |
573 | return error; |
574 | } | |
575 | ||
8b9277c8 GJ |
576 | /* lock_comm() |
577 | * Sets the MD_CLUSTER_SEND_LOCK bit to lock the send channel. | |
578 | */ | |
579 | static int lock_comm(struct md_cluster_info *cinfo) | |
580 | { | |
581 | wait_event(cinfo->wait, | |
582 | !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state)); | |
583 | ||
584 | return lock_token(cinfo); | |
585 | } | |
586 | ||
601b515c GR |
587 | static void unlock_comm(struct md_cluster_info *cinfo) |
588 | { | |
dbb64f86 | 589 | WARN_ON(cinfo->token_lockres->mode != DLM_LOCK_EX); |
8b9277c8 | 590 | mutex_unlock(&cinfo->recv_mutex); |
601b515c | 591 | dlm_unlock_sync(cinfo->token_lockres); |
8b9277c8 GJ |
592 | clear_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state); |
593 | wake_up(&cinfo->wait); | |
601b515c GR |
594 | } |
595 | ||
596 | /* __sendmsg() | |
597 | * This function performs the actual sending of the message. This function is | |
598 | * usually called after performing the encompassing operation | |
599 | * The function: | |
600 | * 1. Grabs the message lockresource in EX mode | |
601 | * 2. Copies the message to the message LVB | |
66099bb0 | 602 | * 3. Downconverts message lockresource to CW |
601b515c GR |
603 | * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes |
604 | * and the other nodes read the message. The thread will wait here until all other | |
605 | * nodes have released ack lock resource. | |
606 | * 5. Downconvert ack lockresource to CR | |
607 | */ | |
608 | static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
609 | { | |
610 | int error; | |
611 | int slot = cinfo->slot_number - 1; | |
612 | ||
613 | cmsg->slot = cpu_to_le32(slot); | |
614 | /*get EX on Message*/ | |
615 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX); | |
616 | if (error) { | |
617 | pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error); | |
618 | goto failed_message; | |
619 | } | |
620 | ||
621 | memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg, | |
622 | sizeof(struct cluster_msg)); | |
66099bb0 GJ |
623 | /*down-convert EX to CW on Message*/ |
624 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CW); | |
601b515c | 625 | if (error) { |
66099bb0 | 626 | pr_err("md-cluster: failed to convert EX to CW on MESSAGE(%d)\n", |
601b515c | 627 | error); |
66099bb0 | 628 | goto failed_ack; |
601b515c GR |
629 | } |
630 | ||
631 | /*up-convert CR to EX on Ack*/ | |
632 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX); | |
633 | if (error) { | |
634 | pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n", | |
635 | error); | |
636 | goto failed_ack; | |
637 | } | |
638 | ||
639 | /*down-convert EX to CR on Ack*/ | |
640 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR); | |
641 | if (error) { | |
642 | pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n", | |
643 | error); | |
644 | goto failed_ack; | |
645 | } | |
646 | ||
647 | failed_ack: | |
b5ef5678 GJ |
648 | error = dlm_unlock_sync(cinfo->message_lockres); |
649 | if (unlikely(error != 0)) { | |
650 | pr_err("md-cluster: failed convert to NL on MESSAGE(%d)\n", | |
651 | error); | |
652 | /* in case the message can't be released due to some reason */ | |
653 | goto failed_ack; | |
654 | } | |
601b515c GR |
655 | failed_message: |
656 | return error; | |
657 | } | |
658 | ||
659 | static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
660 | { | |
661 | int ret; | |
662 | ||
663 | lock_comm(cinfo); | |
664 | ret = __sendmsg(cinfo, cmsg); | |
665 | unlock_comm(cinfo); | |
666 | return ret; | |
667 | } | |
668 | ||
96ae923a GR |
669 | static int gather_all_resync_info(struct mddev *mddev, int total_slots) |
670 | { | |
671 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
672 | int i, ret = 0; | |
673 | struct dlm_lock_resource *bm_lockres; | |
674 | struct suspend_info *s; | |
675 | char str[64]; | |
abb9b22a | 676 | sector_t lo, hi; |
96ae923a GR |
677 | |
678 | ||
679 | for (i = 0; i < total_slots; i++) { | |
680 | memset(str, '\0', 64); | |
681 | snprintf(str, 64, "bitmap%04d", i); | |
682 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
683 | if (!bm_lockres) | |
684 | return -ENOMEM; | |
685 | if (i == (cinfo->slot_number - 1)) | |
686 | continue; | |
687 | ||
688 | bm_lockres->flags |= DLM_LKF_NOQUEUE; | |
689 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
690 | if (ret == -EAGAIN) { | |
691 | memset(bm_lockres->lksb.sb_lvbptr, '\0', LVB_SIZE); | |
692 | s = read_resync_info(mddev, bm_lockres); | |
693 | if (s) { | |
694 | pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n", | |
695 | __func__, __LINE__, | |
696 | (unsigned long long) s->lo, | |
697 | (unsigned long long) s->hi, i); | |
698 | spin_lock_irq(&cinfo->suspend_lock); | |
699 | s->slot = i; | |
700 | list_add(&s->list, &cinfo->suspend_list); | |
701 | spin_unlock_irq(&cinfo->suspend_lock); | |
702 | } | |
703 | ret = 0; | |
704 | lockres_free(bm_lockres); | |
705 | continue; | |
706 | } | |
6e6d9f2c GJ |
707 | if (ret) { |
708 | lockres_free(bm_lockres); | |
96ae923a | 709 | goto out; |
6e6d9f2c | 710 | } |
abb9b22a GJ |
711 | |
712 | /* Read the disk bitmap sb and check if it needs recovery */ | |
713 | ret = bitmap_copy_from_slot(mddev, i, &lo, &hi, false); | |
714 | if (ret) { | |
715 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", i); | |
716 | lockres_free(bm_lockres); | |
717 | continue; | |
718 | } | |
719 | if ((hi > 0) && (lo < mddev->recovery_cp)) { | |
720 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
721 | mddev->recovery_cp = lo; | |
722 | md_check_recovery(mddev); | |
723 | } | |
724 | ||
96ae923a GR |
725 | dlm_unlock_sync(bm_lockres); |
726 | lockres_free(bm_lockres); | |
727 | } | |
728 | out: | |
729 | return ret; | |
730 | } | |
731 | ||
edb39c9d GR |
732 | static int join(struct mddev *mddev, int nodes) |
733 | { | |
c4ce867f | 734 | struct md_cluster_info *cinfo; |
cf921cc1 | 735 | int ret, ops_rv; |
c4ce867f GR |
736 | char str[64]; |
737 | ||
c4ce867f GR |
738 | cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL); |
739 | if (!cinfo) | |
740 | return -ENOMEM; | |
741 | ||
9e3072e3 GJ |
742 | INIT_LIST_HEAD(&cinfo->suspend_list); |
743 | spin_lock_init(&cinfo->suspend_lock); | |
cf921cc1 | 744 | init_completion(&cinfo->completion); |
eece075c | 745 | set_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state); |
8b9277c8 GJ |
746 | init_waitqueue_head(&cinfo->wait); |
747 | mutex_init(&cinfo->recv_mutex); | |
cf921cc1 | 748 | |
cf921cc1 GR |
749 | mddev->cluster_info = cinfo; |
750 | ||
c4ce867f | 751 | memset(str, 0, 64); |
b89f704a | 752 | sprintf(str, "%pU", mddev->uuid); |
cf921cc1 GR |
753 | ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name, |
754 | DLM_LSFL_FS, LVB_SIZE, | |
755 | &md_ls_ops, mddev, &ops_rv, &cinfo->lockspace); | |
c4ce867f GR |
756 | if (ret) |
757 | goto err; | |
cf921cc1 | 758 | wait_for_completion(&cinfo->completion); |
8c58f02e GJ |
759 | if (nodes < cinfo->slot_number) { |
760 | pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).", | |
761 | cinfo->slot_number, nodes); | |
b97e9257 GR |
762 | ret = -ERANGE; |
763 | goto err; | |
764 | } | |
4664680c GR |
765 | /* Initiate the communication resources */ |
766 | ret = -ENOMEM; | |
767 | cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv"); | |
768 | if (!cinfo->recv_thread) { | |
769 | pr_err("md-cluster: cannot allocate memory for recv_thread!\n"); | |
770 | goto err; | |
771 | } | |
772 | cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1); | |
773 | if (!cinfo->message_lockres) | |
774 | goto err; | |
775 | cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0); | |
776 | if (!cinfo->token_lockres) | |
777 | goto err; | |
778 | cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0); | |
779 | if (!cinfo->ack_lockres) | |
780 | goto err; | |
1aee41f6 GR |
781 | cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0); |
782 | if (!cinfo->no_new_dev_lockres) | |
783 | goto err; | |
784 | ||
4664680c GR |
785 | /* get sync CR lock on ACK. */ |
786 | if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR)) | |
787 | pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n", | |
788 | ret); | |
1aee41f6 GR |
789 | /* get sync CR lock on no-new-dev. */ |
790 | if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR)) | |
791 | pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret); | |
792 | ||
54519c5f GR |
793 | |
794 | pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number); | |
795 | snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1); | |
796 | cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1); | |
797 | if (!cinfo->bitmap_lockres) | |
798 | goto err; | |
799 | if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) { | |
800 | pr_err("Failed to get bitmap lock\n"); | |
801 | ret = -EINVAL; | |
802 | goto err; | |
803 | } | |
804 | ||
c186b128 GR |
805 | cinfo->resync_lockres = lockres_init(mddev, "resync", NULL, 0); |
806 | if (!cinfo->resync_lockres) | |
807 | goto err; | |
808 | ||
96ae923a GR |
809 | ret = gather_all_resync_info(mddev, nodes); |
810 | if (ret) | |
811 | goto err; | |
812 | ||
edb39c9d | 813 | return 0; |
c4ce867f | 814 | err: |
4664680c GR |
815 | lockres_free(cinfo->message_lockres); |
816 | lockres_free(cinfo->token_lockres); | |
817 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 818 | lockres_free(cinfo->no_new_dev_lockres); |
c186b128 | 819 | lockres_free(cinfo->resync_lockres); |
96ae923a | 820 | lockres_free(cinfo->bitmap_lockres); |
c4ce867f GR |
821 | if (cinfo->lockspace) |
822 | dlm_release_lockspace(cinfo->lockspace, 2); | |
cf921cc1 | 823 | mddev->cluster_info = NULL; |
c4ce867f | 824 | kfree(cinfo); |
c4ce867f | 825 | return ret; |
edb39c9d GR |
826 | } |
827 | ||
09995411 GJ |
828 | static void resync_bitmap(struct mddev *mddev) |
829 | { | |
830 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
831 | struct cluster_msg cmsg = {0}; | |
832 | int err; | |
833 | ||
834 | cmsg.type = cpu_to_le32(BITMAP_NEEDS_SYNC); | |
835 | err = sendmsg(cinfo, &cmsg); | |
836 | if (err) | |
837 | pr_err("%s:%d: failed to send BITMAP_NEEDS_SYNC message (%d)\n", | |
838 | __func__, __LINE__, err); | |
839 | } | |
840 | ||
f6a2dc64 | 841 | static void unlock_all_bitmaps(struct mddev *mddev); |
edb39c9d GR |
842 | static int leave(struct mddev *mddev) |
843 | { | |
c4ce867f GR |
844 | struct md_cluster_info *cinfo = mddev->cluster_info; |
845 | ||
846 | if (!cinfo) | |
847 | return 0; | |
09995411 GJ |
848 | |
849 | /* BITMAP_NEEDS_SYNC message should be sent when node | |
850 | * is leaving the cluster with dirty bitmap, also we | |
851 | * can only deliver it when dlm connection is available */ | |
852 | if (cinfo->slot_number > 0 && mddev->recovery_cp != MaxSector) | |
853 | resync_bitmap(mddev); | |
854 | ||
e94987db | 855 | md_unregister_thread(&cinfo->recovery_thread); |
4664680c GR |
856 | md_unregister_thread(&cinfo->recv_thread); |
857 | lockres_free(cinfo->message_lockres); | |
858 | lockres_free(cinfo->token_lockres); | |
859 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 860 | lockres_free(cinfo->no_new_dev_lockres); |
54519c5f | 861 | lockres_free(cinfo->bitmap_lockres); |
f6a2dc64 | 862 | unlock_all_bitmaps(mddev); |
c4ce867f | 863 | dlm_release_lockspace(cinfo->lockspace, 2); |
edb39c9d GR |
864 | return 0; |
865 | } | |
866 | ||
cf921cc1 GR |
867 | /* slot_number(): Returns the MD slot number to use |
868 | * DLM starts the slot numbers from 1, wheras cluster-md | |
869 | * wants the number to be from zero, so we deduct one | |
870 | */ | |
871 | static int slot_number(struct mddev *mddev) | |
872 | { | |
873 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
874 | ||
875 | return cinfo->slot_number - 1; | |
876 | } | |
877 | ||
8b9277c8 GJ |
878 | /* |
879 | * Check if the communication is already locked, else lock the communication | |
880 | * channel. | |
881 | * If it is already locked, token is in EX mode, and hence lock_token() | |
882 | * should not be called. | |
883 | */ | |
293467aa GR |
884 | static int metadata_update_start(struct mddev *mddev) |
885 | { | |
8b9277c8 GJ |
886 | struct md_cluster_info *cinfo = mddev->cluster_info; |
887 | ||
888 | wait_event(cinfo->wait, | |
889 | !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state) || | |
890 | test_and_clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state)); | |
891 | ||
892 | /* If token is already locked, return 0 */ | |
893 | if (cinfo->token_lockres->mode == DLM_LOCK_EX) | |
894 | return 0; | |
895 | ||
896 | return lock_token(cinfo); | |
293467aa GR |
897 | } |
898 | ||
899 | static int metadata_update_finish(struct mddev *mddev) | |
900 | { | |
901 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
902 | struct cluster_msg cmsg; | |
70bcecdb GR |
903 | struct md_rdev *rdev; |
904 | int ret = 0; | |
ba2746b0 | 905 | int raid_slot = -1; |
293467aa GR |
906 | |
907 | memset(&cmsg, 0, sizeof(cmsg)); | |
908 | cmsg.type = cpu_to_le32(METADATA_UPDATED); | |
70bcecdb GR |
909 | /* Pick up a good active device number to send. |
910 | */ | |
911 | rdev_for_each(rdev, mddev) | |
912 | if (rdev->raid_disk > -1 && !test_bit(Faulty, &rdev->flags)) { | |
ba2746b0 | 913 | raid_slot = rdev->desc_nr; |
70bcecdb GR |
914 | break; |
915 | } | |
ba2746b0 N |
916 | if (raid_slot >= 0) { |
917 | cmsg.raid_slot = cpu_to_le32(raid_slot); | |
70bcecdb | 918 | ret = __sendmsg(cinfo, &cmsg); |
ba2746b0 | 919 | } else |
70bcecdb | 920 | pr_warn("md-cluster: No good device id found to send\n"); |
8b9277c8 | 921 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
293467aa GR |
922 | unlock_comm(cinfo); |
923 | return ret; | |
924 | } | |
925 | ||
dbb64f86 | 926 | static void metadata_update_cancel(struct mddev *mddev) |
293467aa GR |
927 | { |
928 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
8b9277c8 | 929 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
dbb64f86 | 930 | unlock_comm(cinfo); |
293467aa GR |
931 | } |
932 | ||
c186b128 GR |
933 | static int resync_start(struct mddev *mddev) |
934 | { | |
935 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
936 | cinfo->resync_lockres->flags |= DLM_LKF_NOQUEUE; | |
937 | return dlm_lock_sync(cinfo->resync_lockres, DLM_LOCK_EX); | |
938 | } | |
939 | ||
c40f341f | 940 | static int resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi) |
965400eb GR |
941 | { |
942 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
ac277c6a | 943 | struct resync_info ri; |
aee177ac | 944 | struct cluster_msg cmsg = {0}; |
965400eb | 945 | |
ac277c6a GR |
946 | /* do not send zero again, if we have sent before */ |
947 | if (hi == 0) { | |
948 | memcpy(&ri, cinfo->bitmap_lockres->lksb.sb_lvbptr, sizeof(struct resync_info)); | |
949 | if (le64_to_cpu(ri.hi) == 0) | |
950 | return 0; | |
951 | } | |
952 | ||
30661b49 | 953 | add_resync_info(cinfo->bitmap_lockres, lo, hi); |
c40f341f GR |
954 | /* Re-acquire the lock to refresh LVB */ |
955 | dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW); | |
c40f341f | 956 | cmsg.type = cpu_to_le32(RESYNCING); |
965400eb GR |
957 | cmsg.low = cpu_to_le64(lo); |
958 | cmsg.high = cpu_to_le64(hi); | |
c186b128 | 959 | |
965400eb GR |
960 | return sendmsg(cinfo, &cmsg); |
961 | } | |
962 | ||
c186b128 GR |
963 | static int resync_finish(struct mddev *mddev) |
964 | { | |
965 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
966 | cinfo->resync_lockres->flags &= ~DLM_LKF_NOQUEUE; | |
967 | dlm_unlock_sync(cinfo->resync_lockres); | |
968 | return resync_info_update(mddev, 0, 0); | |
969 | } | |
970 | ||
90382ed9 GR |
971 | static int area_resyncing(struct mddev *mddev, int direction, |
972 | sector_t lo, sector_t hi) | |
589a1c49 GR |
973 | { |
974 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
975 | int ret = 0; | |
976 | struct suspend_info *s; | |
977 | ||
90382ed9 GR |
978 | if ((direction == READ) && |
979 | test_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state)) | |
980 | return 1; | |
981 | ||
589a1c49 GR |
982 | spin_lock_irq(&cinfo->suspend_lock); |
983 | if (list_empty(&cinfo->suspend_list)) | |
984 | goto out; | |
985 | list_for_each_entry(s, &cinfo->suspend_list, list) | |
986 | if (hi > s->lo && lo < s->hi) { | |
987 | ret = 1; | |
988 | break; | |
989 | } | |
990 | out: | |
991 | spin_unlock_irq(&cinfo->suspend_lock); | |
992 | return ret; | |
993 | } | |
994 | ||
dbb64f86 GR |
995 | /* add_new_disk() - initiates a disk add |
996 | * However, if this fails before writing md_update_sb(), | |
997 | * add_new_disk_cancel() must be called to release token lock | |
998 | */ | |
999 | static int add_new_disk(struct mddev *mddev, struct md_rdev *rdev) | |
1aee41f6 GR |
1000 | { |
1001 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1002 | struct cluster_msg cmsg; | |
1003 | int ret = 0; | |
1004 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); | |
1005 | char *uuid = sb->device_uuid; | |
1006 | ||
1007 | memset(&cmsg, 0, sizeof(cmsg)); | |
1008 | cmsg.type = cpu_to_le32(NEWDISK); | |
1009 | memcpy(cmsg.uuid, uuid, 16); | |
faeff83f | 1010 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); |
1aee41f6 GR |
1011 | lock_comm(cinfo); |
1012 | ret = __sendmsg(cinfo, &cmsg); | |
1013 | if (ret) | |
1014 | return ret; | |
1015 | cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE; | |
1016 | ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX); | |
1017 | cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE; | |
1018 | /* Some node does not "see" the device */ | |
1019 | if (ret == -EAGAIN) | |
1020 | ret = -ENOENT; | |
dbb64f86 GR |
1021 | if (ret) |
1022 | unlock_comm(cinfo); | |
8b9277c8 | 1023 | else { |
1aee41f6 | 1024 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); |
e19508fa GJ |
1025 | /* Since MD_CHANGE_DEVS will be set in add_bound_rdev which |
1026 | * will run soon after add_new_disk, the below path will be | |
1027 | * invoked: | |
1028 | * md_wakeup_thread(mddev->thread) | |
1029 | * -> conf->thread (raid1d) | |
1030 | * -> md_check_recovery -> md_update_sb | |
1031 | * -> metadata_update_start/finish | |
1032 | * MD_CLUSTER_SEND_LOCKED_ALREADY will be cleared eventually. | |
1033 | * | |
1034 | * For other failure cases, metadata_update_cancel and | |
1035 | * add_new_disk_cancel also clear below bit as well. | |
1036 | * */ | |
8b9277c8 GJ |
1037 | set_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
1038 | wake_up(&cinfo->wait); | |
1039 | } | |
1aee41f6 GR |
1040 | return ret; |
1041 | } | |
1042 | ||
dbb64f86 | 1043 | static void add_new_disk_cancel(struct mddev *mddev) |
1aee41f6 | 1044 | { |
dbb64f86 | 1045 | struct md_cluster_info *cinfo = mddev->cluster_info; |
8b9277c8 | 1046 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
dbb64f86 | 1047 | unlock_comm(cinfo); |
1aee41f6 GR |
1048 | } |
1049 | ||
fa8259da | 1050 | static int new_disk_ack(struct mddev *mddev, bool ack) |
1aee41f6 GR |
1051 | { |
1052 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1053 | ||
fa8259da GR |
1054 | if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) { |
1055 | pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev)); | |
1056 | return -EINVAL; | |
1057 | } | |
1058 | ||
1aee41f6 GR |
1059 | if (ack) |
1060 | dlm_unlock_sync(cinfo->no_new_dev_lockres); | |
1061 | complete(&cinfo->newdisk_completion); | |
fa8259da | 1062 | return 0; |
1aee41f6 GR |
1063 | } |
1064 | ||
88bcfef7 GR |
1065 | static int remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1066 | { | |
aee177ac | 1067 | struct cluster_msg cmsg = {0}; |
88bcfef7 | 1068 | struct md_cluster_info *cinfo = mddev->cluster_info; |
faeff83f GJ |
1069 | cmsg.type = cpu_to_le32(REMOVE); |
1070 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); | |
54a88392 | 1071 | return sendmsg(cinfo, &cmsg); |
88bcfef7 GR |
1072 | } |
1073 | ||
f6a2dc64 GJ |
1074 | static int lock_all_bitmaps(struct mddev *mddev) |
1075 | { | |
1076 | int slot, my_slot, ret, held = 1, i = 0; | |
1077 | char str[64]; | |
1078 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1079 | ||
1080 | cinfo->other_bitmap_lockres = kzalloc((mddev->bitmap_info.nodes - 1) * | |
1081 | sizeof(struct dlm_lock_resource *), | |
1082 | GFP_KERNEL); | |
1083 | if (!cinfo->other_bitmap_lockres) { | |
1084 | pr_err("md: can't alloc mem for other bitmap locks\n"); | |
1085 | return 0; | |
1086 | } | |
1087 | ||
1088 | my_slot = slot_number(mddev); | |
1089 | for (slot = 0; slot < mddev->bitmap_info.nodes; slot++) { | |
1090 | if (slot == my_slot) | |
1091 | continue; | |
1092 | ||
1093 | memset(str, '\0', 64); | |
1094 | snprintf(str, 64, "bitmap%04d", slot); | |
1095 | cinfo->other_bitmap_lockres[i] = lockres_init(mddev, str, NULL, 1); | |
1096 | if (!cinfo->other_bitmap_lockres[i]) | |
1097 | return -ENOMEM; | |
1098 | ||
1099 | cinfo->other_bitmap_lockres[i]->flags |= DLM_LKF_NOQUEUE; | |
1100 | ret = dlm_lock_sync(cinfo->other_bitmap_lockres[i], DLM_LOCK_PW); | |
1101 | if (ret) | |
1102 | held = -1; | |
1103 | i++; | |
1104 | } | |
1105 | ||
1106 | return held; | |
1107 | } | |
1108 | ||
1109 | static void unlock_all_bitmaps(struct mddev *mddev) | |
1110 | { | |
1111 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1112 | int i; | |
1113 | ||
1114 | /* release other node's bitmap lock if they are existed */ | |
1115 | if (cinfo->other_bitmap_lockres) { | |
1116 | for (i = 0; i < mddev->bitmap_info.nodes - 1; i++) { | |
1117 | if (cinfo->other_bitmap_lockres[i]) { | |
1118 | dlm_unlock_sync(cinfo->other_bitmap_lockres[i]); | |
1119 | lockres_free(cinfo->other_bitmap_lockres[i]); | |
1120 | } | |
1121 | } | |
1122 | kfree(cinfo->other_bitmap_lockres); | |
1123 | } | |
1124 | } | |
1125 | ||
97f6cd39 GR |
1126 | static int gather_bitmaps(struct md_rdev *rdev) |
1127 | { | |
1128 | int sn, err; | |
1129 | sector_t lo, hi; | |
aee177ac | 1130 | struct cluster_msg cmsg = {0}; |
97f6cd39 GR |
1131 | struct mddev *mddev = rdev->mddev; |
1132 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1133 | ||
faeff83f GJ |
1134 | cmsg.type = cpu_to_le32(RE_ADD); |
1135 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); | |
97f6cd39 GR |
1136 | err = sendmsg(cinfo, &cmsg); |
1137 | if (err) | |
1138 | goto out; | |
1139 | ||
1140 | for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) { | |
1141 | if (sn == (cinfo->slot_number - 1)) | |
1142 | continue; | |
1143 | err = bitmap_copy_from_slot(mddev, sn, &lo, &hi, false); | |
1144 | if (err) { | |
1145 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn); | |
1146 | goto out; | |
1147 | } | |
1148 | if ((hi > 0) && (lo < mddev->recovery_cp)) | |
1149 | mddev->recovery_cp = lo; | |
1150 | } | |
1151 | out: | |
1152 | return err; | |
1153 | } | |
1154 | ||
edb39c9d GR |
1155 | static struct md_cluster_operations cluster_ops = { |
1156 | .join = join, | |
1157 | .leave = leave, | |
cf921cc1 | 1158 | .slot_number = slot_number, |
c186b128 GR |
1159 | .resync_start = resync_start, |
1160 | .resync_finish = resync_finish, | |
96ae923a | 1161 | .resync_info_update = resync_info_update, |
293467aa GR |
1162 | .metadata_update_start = metadata_update_start, |
1163 | .metadata_update_finish = metadata_update_finish, | |
1164 | .metadata_update_cancel = metadata_update_cancel, | |
589a1c49 | 1165 | .area_resyncing = area_resyncing, |
dbb64f86 GR |
1166 | .add_new_disk = add_new_disk, |
1167 | .add_new_disk_cancel = add_new_disk_cancel, | |
1aee41f6 | 1168 | .new_disk_ack = new_disk_ack, |
88bcfef7 | 1169 | .remove_disk = remove_disk, |
97f6cd39 | 1170 | .gather_bitmaps = gather_bitmaps, |
f6a2dc64 GJ |
1171 | .lock_all_bitmaps = lock_all_bitmaps, |
1172 | .unlock_all_bitmaps = unlock_all_bitmaps, | |
edb39c9d GR |
1173 | }; |
1174 | ||
8e854e9c GR |
1175 | static int __init cluster_init(void) |
1176 | { | |
1177 | pr_warn("md-cluster: EXPERIMENTAL. Use with caution\n"); | |
1178 | pr_info("Registering Cluster MD functions\n"); | |
edb39c9d | 1179 | register_md_cluster_operations(&cluster_ops, THIS_MODULE); |
8e854e9c GR |
1180 | return 0; |
1181 | } | |
1182 | ||
1183 | static void cluster_exit(void) | |
1184 | { | |
edb39c9d | 1185 | unregister_md_cluster_operations(); |
8e854e9c GR |
1186 | } |
1187 | ||
1188 | module_init(cluster_init); | |
1189 | module_exit(cluster_exit); | |
86b57277 | 1190 | MODULE_AUTHOR("SUSE"); |
8e854e9c GR |
1191 | MODULE_LICENSE("GPL"); |
1192 | MODULE_DESCRIPTION("Clustering support for MD"); |