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 | |
51e453ae GJ |
64 | /* We should receive message after node joined cluster and |
65 | * set up all the related infos such as bitmap and personality */ | |
66 | #define MD_CLUSTER_ALREADY_IN_CLUSTER 6 | |
67 | #define MD_CLUSTER_PENDING_RECV_EVENT 7 | |
8b9277c8 | 68 | |
fa8259da | 69 | |
c4ce867f GR |
70 | struct md_cluster_info { |
71 | /* dlm lock space and resources for clustered raid. */ | |
72 | dlm_lockspace_t *lockspace; | |
cf921cc1 GR |
73 | int slot_number; |
74 | struct completion completion; | |
8b9277c8 | 75 | struct mutex recv_mutex; |
54519c5f | 76 | struct dlm_lock_resource *bitmap_lockres; |
f6a2dc64 | 77 | struct dlm_lock_resource **other_bitmap_lockres; |
c186b128 | 78 | struct dlm_lock_resource *resync_lockres; |
96ae923a GR |
79 | struct list_head suspend_list; |
80 | spinlock_t suspend_lock; | |
e94987db GR |
81 | struct md_thread *recovery_thread; |
82 | unsigned long recovery_map; | |
4664680c GR |
83 | /* communication loc resources */ |
84 | struct dlm_lock_resource *ack_lockres; | |
85 | struct dlm_lock_resource *message_lockres; | |
86 | struct dlm_lock_resource *token_lockres; | |
1aee41f6 | 87 | struct dlm_lock_resource *no_new_dev_lockres; |
4664680c | 88 | struct md_thread *recv_thread; |
1aee41f6 | 89 | struct completion newdisk_completion; |
8b9277c8 | 90 | wait_queue_head_t wait; |
fa8259da | 91 | unsigned long state; |
18c9ff7f GJ |
92 | /* record the region in RESYNCING message */ |
93 | sector_t sync_low; | |
94 | sector_t sync_hi; | |
4664680c GR |
95 | }; |
96 | ||
97 | enum msg_type { | |
98 | METADATA_UPDATED = 0, | |
99 | RESYNCING, | |
1aee41f6 | 100 | NEWDISK, |
88bcfef7 | 101 | REMOVE, |
97f6cd39 | 102 | RE_ADD, |
dc737d7c | 103 | BITMAP_NEEDS_SYNC, |
4664680c GR |
104 | }; |
105 | ||
106 | struct cluster_msg { | |
cf97a348 GJ |
107 | __le32 type; |
108 | __le32 slot; | |
1aee41f6 | 109 | /* TODO: Unionize this for smaller footprint */ |
cf97a348 GJ |
110 | __le64 low; |
111 | __le64 high; | |
1aee41f6 | 112 | char uuid[16]; |
cf97a348 | 113 | __le32 raid_slot; |
47741b7c GR |
114 | }; |
115 | ||
116 | static void sync_ast(void *arg) | |
117 | { | |
118 | struct dlm_lock_resource *res; | |
119 | ||
2e2a7cd9 | 120 | res = arg; |
47741b7c GR |
121 | complete(&res->completion); |
122 | } | |
123 | ||
124 | static int dlm_lock_sync(struct dlm_lock_resource *res, int mode) | |
125 | { | |
126 | int ret = 0; | |
127 | ||
47741b7c GR |
128 | ret = dlm_lock(res->ls, mode, &res->lksb, |
129 | res->flags, res->name, strlen(res->name), | |
130 | 0, sync_ast, res, res->bast); | |
131 | if (ret) | |
132 | return ret; | |
133 | wait_for_completion(&res->completion); | |
dbb64f86 GR |
134 | if (res->lksb.sb_status == 0) |
135 | res->mode = mode; | |
47741b7c GR |
136 | return res->lksb.sb_status; |
137 | } | |
138 | ||
139 | static int dlm_unlock_sync(struct dlm_lock_resource *res) | |
140 | { | |
141 | return dlm_lock_sync(res, DLM_LOCK_NL); | |
142 | } | |
143 | ||
c4ce867f | 144 | static struct dlm_lock_resource *lockres_init(struct mddev *mddev, |
47741b7c GR |
145 | char *name, void (*bastfn)(void *arg, int mode), int with_lvb) |
146 | { | |
147 | struct dlm_lock_resource *res = NULL; | |
148 | int ret, namelen; | |
c4ce867f | 149 | struct md_cluster_info *cinfo = mddev->cluster_info; |
47741b7c GR |
150 | |
151 | res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL); | |
152 | if (!res) | |
153 | return NULL; | |
b83d51c0 | 154 | init_completion(&res->completion); |
c4ce867f GR |
155 | res->ls = cinfo->lockspace; |
156 | res->mddev = mddev; | |
dbb64f86 | 157 | res->mode = DLM_LOCK_IV; |
47741b7c GR |
158 | namelen = strlen(name); |
159 | res->name = kzalloc(namelen + 1, GFP_KERNEL); | |
160 | if (!res->name) { | |
161 | pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name); | |
162 | goto out_err; | |
163 | } | |
164 | strlcpy(res->name, name, namelen + 1); | |
165 | if (with_lvb) { | |
166 | res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL); | |
167 | if (!res->lksb.sb_lvbptr) { | |
168 | pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name); | |
169 | goto out_err; | |
170 | } | |
171 | res->flags = DLM_LKF_VALBLK; | |
172 | } | |
173 | ||
174 | if (bastfn) | |
175 | res->bast = bastfn; | |
176 | ||
177 | res->flags |= DLM_LKF_EXPEDITE; | |
178 | ||
179 | ret = dlm_lock_sync(res, DLM_LOCK_NL); | |
180 | if (ret) { | |
181 | pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name); | |
182 | goto out_err; | |
183 | } | |
184 | res->flags &= ~DLM_LKF_EXPEDITE; | |
185 | res->flags |= DLM_LKF_CONVERT; | |
186 | ||
187 | return res; | |
188 | out_err: | |
189 | kfree(res->lksb.sb_lvbptr); | |
190 | kfree(res->name); | |
191 | kfree(res); | |
192 | return NULL; | |
193 | } | |
194 | ||
195 | static void lockres_free(struct dlm_lock_resource *res) | |
196 | { | |
b5ef5678 GJ |
197 | int ret; |
198 | ||
47741b7c GR |
199 | if (!res) |
200 | return; | |
201 | ||
b5ef5678 GJ |
202 | /* cancel a lock request or a conversion request that is blocked */ |
203 | res->flags |= DLM_LKF_CANCEL; | |
204 | retry: | |
205 | ret = dlm_unlock(res->ls, res->lksb.sb_lkid, 0, &res->lksb, res); | |
206 | if (unlikely(ret != 0)) { | |
207 | pr_info("%s: failed to unlock %s return %d\n", __func__, res->name, ret); | |
208 | ||
209 | /* if a lock conversion is cancelled, then the lock is put | |
210 | * back to grant queue, need to ensure it is unlocked */ | |
211 | if (ret == -DLM_ECANCEL) | |
212 | goto retry; | |
213 | } | |
214 | res->flags &= ~DLM_LKF_CANCEL; | |
47741b7c GR |
215 | wait_for_completion(&res->completion); |
216 | ||
217 | kfree(res->name); | |
218 | kfree(res->lksb.sb_lvbptr); | |
219 | kfree(res); | |
220 | } | |
8e854e9c | 221 | |
30661b49 N |
222 | static void add_resync_info(struct dlm_lock_resource *lockres, |
223 | sector_t lo, sector_t hi) | |
96ae923a GR |
224 | { |
225 | struct resync_info *ri; | |
226 | ||
227 | ri = (struct resync_info *)lockres->lksb.sb_lvbptr; | |
228 | ri->lo = cpu_to_le64(lo); | |
229 | ri->hi = cpu_to_le64(hi); | |
230 | } | |
231 | ||
232 | static struct suspend_info *read_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres) | |
233 | { | |
234 | struct resync_info ri; | |
235 | struct suspend_info *s = NULL; | |
236 | sector_t hi = 0; | |
237 | ||
238 | dlm_lock_sync(lockres, DLM_LOCK_CR); | |
239 | memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info)); | |
240 | hi = le64_to_cpu(ri.hi); | |
cf97a348 | 241 | if (hi > 0) { |
96ae923a GR |
242 | s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL); |
243 | if (!s) | |
244 | goto out; | |
245 | s->hi = hi; | |
246 | s->lo = le64_to_cpu(ri.lo); | |
247 | } | |
248 | dlm_unlock_sync(lockres); | |
249 | out: | |
250 | return s; | |
251 | } | |
252 | ||
6dc69c9c | 253 | static void recover_bitmaps(struct md_thread *thread) |
e94987db GR |
254 | { |
255 | struct mddev *mddev = thread->mddev; | |
256 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
257 | struct dlm_lock_resource *bm_lockres; | |
258 | char str[64]; | |
259 | int slot, ret; | |
260 | struct suspend_info *s, *tmp; | |
261 | sector_t lo, hi; | |
262 | ||
263 | while (cinfo->recovery_map) { | |
264 | slot = fls64((u64)cinfo->recovery_map) - 1; | |
265 | ||
266 | /* Clear suspend_area associated with the bitmap */ | |
267 | spin_lock_irq(&cinfo->suspend_lock); | |
268 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
269 | if (slot == s->slot) { | |
270 | list_del(&s->list); | |
271 | kfree(s); | |
272 | } | |
273 | spin_unlock_irq(&cinfo->suspend_lock); | |
274 | ||
275 | snprintf(str, 64, "bitmap%04d", slot); | |
276 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
277 | if (!bm_lockres) { | |
278 | pr_err("md-cluster: Cannot initialize bitmaps\n"); | |
279 | goto clear_bit; | |
280 | } | |
281 | ||
282 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
283 | if (ret) { | |
284 | pr_err("md-cluster: Could not DLM lock %s: %d\n", | |
285 | str, ret); | |
286 | goto clear_bit; | |
287 | } | |
97f6cd39 | 288 | ret = bitmap_copy_from_slot(mddev, slot, &lo, &hi, true); |
4b26a08a | 289 | if (ret) { |
e94987db | 290 | pr_err("md-cluster: Could not copy data from bitmap %d\n", slot); |
4b26a08a GR |
291 | goto dlm_unlock; |
292 | } | |
293 | if (hi > 0) { | |
4b26a08a GR |
294 | if (lo < mddev->recovery_cp) |
295 | mddev->recovery_cp = lo; | |
eb315cd0 GJ |
296 | /* wake up thread to continue resync in case resync |
297 | * is not finished */ | |
298 | if (mddev->recovery_cp != MaxSector) { | |
299 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
300 | md_wakeup_thread(mddev->thread); | |
301 | } | |
4b26a08a GR |
302 | } |
303 | dlm_unlock: | |
e94987db GR |
304 | dlm_unlock_sync(bm_lockres); |
305 | clear_bit: | |
4ac7a65f | 306 | lockres_free(bm_lockres); |
e94987db GR |
307 | clear_bit(slot, &cinfo->recovery_map); |
308 | } | |
309 | } | |
310 | ||
cf921cc1 GR |
311 | static void recover_prep(void *arg) |
312 | { | |
90382ed9 GR |
313 | struct mddev *mddev = arg; |
314 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
315 | set_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); | |
cf921cc1 GR |
316 | } |
317 | ||
05cd0e51 | 318 | static void __recover_slot(struct mddev *mddev, int slot) |
cf921cc1 | 319 | { |
cf921cc1 GR |
320 | struct md_cluster_info *cinfo = mddev->cluster_info; |
321 | ||
05cd0e51 | 322 | set_bit(slot, &cinfo->recovery_map); |
e94987db GR |
323 | if (!cinfo->recovery_thread) { |
324 | cinfo->recovery_thread = md_register_thread(recover_bitmaps, | |
325 | mddev, "recover"); | |
326 | if (!cinfo->recovery_thread) { | |
327 | pr_warn("md-cluster: Could not create recovery thread\n"); | |
328 | return; | |
329 | } | |
330 | } | |
331 | md_wakeup_thread(cinfo->recovery_thread); | |
cf921cc1 GR |
332 | } |
333 | ||
05cd0e51 GJ |
334 | static void recover_slot(void *arg, struct dlm_slot *slot) |
335 | { | |
336 | struct mddev *mddev = arg; | |
337 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
338 | ||
339 | pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n", | |
340 | mddev->bitmap_info.cluster_name, | |
341 | slot->nodeid, slot->slot, | |
342 | cinfo->slot_number); | |
343 | /* deduct one since dlm slot starts from one while the num of | |
344 | * cluster-md begins with 0 */ | |
345 | __recover_slot(mddev, slot->slot - 1); | |
346 | } | |
347 | ||
cf921cc1 GR |
348 | static void recover_done(void *arg, struct dlm_slot *slots, |
349 | int num_slots, int our_slot, | |
350 | uint32_t generation) | |
351 | { | |
352 | struct mddev *mddev = arg; | |
353 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
354 | ||
355 | cinfo->slot_number = our_slot; | |
eece075c GJ |
356 | /* completion is only need to be complete when node join cluster, |
357 | * it doesn't need to run during another node's failure */ | |
358 | if (test_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state)) { | |
359 | complete(&cinfo->completion); | |
360 | clear_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state); | |
361 | } | |
90382ed9 | 362 | clear_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); |
cf921cc1 GR |
363 | } |
364 | ||
eece075c GJ |
365 | /* the ops is called when node join the cluster, and do lock recovery |
366 | * if node failure occurs */ | |
cf921cc1 GR |
367 | static const struct dlm_lockspace_ops md_ls_ops = { |
368 | .recover_prep = recover_prep, | |
369 | .recover_slot = recover_slot, | |
370 | .recover_done = recover_done, | |
371 | }; | |
372 | ||
4664680c GR |
373 | /* |
374 | * The BAST function for the ack lock resource | |
375 | * This function wakes up the receive thread in | |
376 | * order to receive and process the message. | |
377 | */ | |
378 | static void ack_bast(void *arg, int mode) | |
379 | { | |
2e2a7cd9 | 380 | struct dlm_lock_resource *res = arg; |
4664680c GR |
381 | struct md_cluster_info *cinfo = res->mddev->cluster_info; |
382 | ||
51e453ae GJ |
383 | if (mode == DLM_LOCK_EX) { |
384 | if (test_bit(MD_CLUSTER_ALREADY_IN_CLUSTER, &cinfo->state)) | |
385 | md_wakeup_thread(cinfo->recv_thread); | |
386 | else | |
387 | set_bit(MD_CLUSTER_PENDING_RECV_EVENT, &cinfo->state); | |
388 | } | |
4664680c GR |
389 | } |
390 | ||
e59721cc GR |
391 | static void __remove_suspend_info(struct md_cluster_info *cinfo, int slot) |
392 | { | |
393 | struct suspend_info *s, *tmp; | |
394 | ||
395 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
396 | if (slot == s->slot) { | |
e59721cc GR |
397 | list_del(&s->list); |
398 | kfree(s); | |
399 | break; | |
400 | } | |
401 | } | |
402 | ||
b8ca846e | 403 | static void remove_suspend_info(struct mddev *mddev, int slot) |
e59721cc | 404 | { |
b8ca846e | 405 | struct md_cluster_info *cinfo = mddev->cluster_info; |
e59721cc GR |
406 | spin_lock_irq(&cinfo->suspend_lock); |
407 | __remove_suspend_info(cinfo, slot); | |
408 | spin_unlock_irq(&cinfo->suspend_lock); | |
b8ca846e | 409 | mddev->pers->quiesce(mddev, 2); |
e59721cc GR |
410 | } |
411 | ||
412 | ||
9ed38ff5 | 413 | static void process_suspend_info(struct mddev *mddev, |
e59721cc GR |
414 | int slot, sector_t lo, sector_t hi) |
415 | { | |
9ed38ff5 | 416 | struct md_cluster_info *cinfo = mddev->cluster_info; |
e59721cc GR |
417 | struct suspend_info *s; |
418 | ||
419 | if (!hi) { | |
b8ca846e | 420 | remove_suspend_info(mddev, slot); |
c186b128 GR |
421 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
422 | md_wakeup_thread(mddev->thread); | |
e59721cc GR |
423 | return; |
424 | } | |
18c9ff7f GJ |
425 | |
426 | /* | |
427 | * The bitmaps are not same for different nodes | |
428 | * if RESYNCING is happening in one node, then | |
429 | * the node which received the RESYNCING message | |
430 | * probably will perform resync with the region | |
431 | * [lo, hi] again, so we could reduce resync time | |
432 | * a lot if we can ensure that the bitmaps among | |
433 | * different nodes are match up well. | |
434 | * | |
435 | * sync_low/hi is used to record the region which | |
436 | * arrived in the previous RESYNCING message, | |
437 | * | |
438 | * Call bitmap_sync_with_cluster to clear | |
439 | * NEEDED_MASK and set RESYNC_MASK since | |
440 | * resync thread is running in another node, | |
441 | * so we don't need to do the resync again | |
442 | * with the same section */ | |
443 | bitmap_sync_with_cluster(mddev, cinfo->sync_low, | |
444 | cinfo->sync_hi, | |
445 | lo, hi); | |
446 | cinfo->sync_low = lo; | |
447 | cinfo->sync_hi = hi; | |
448 | ||
e59721cc GR |
449 | s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL); |
450 | if (!s) | |
451 | return; | |
452 | s->slot = slot; | |
453 | s->lo = lo; | |
454 | s->hi = hi; | |
9ed38ff5 GR |
455 | mddev->pers->quiesce(mddev, 1); |
456 | mddev->pers->quiesce(mddev, 0); | |
e59721cc GR |
457 | spin_lock_irq(&cinfo->suspend_lock); |
458 | /* Remove existing entry (if exists) before adding */ | |
459 | __remove_suspend_info(cinfo, slot); | |
460 | list_add(&s->list, &cinfo->suspend_list); | |
461 | spin_unlock_irq(&cinfo->suspend_lock); | |
b8ca846e | 462 | mddev->pers->quiesce(mddev, 2); |
e59721cc GR |
463 | } |
464 | ||
1aee41f6 GR |
465 | static void process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg) |
466 | { | |
467 | char disk_uuid[64]; | |
468 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
469 | char event_name[] = "EVENT=ADD_DEVICE"; | |
470 | char raid_slot[16]; | |
471 | char *envp[] = {event_name, disk_uuid, raid_slot, NULL}; | |
472 | int len; | |
473 | ||
474 | len = snprintf(disk_uuid, 64, "DEVICE_UUID="); | |
b89f704a | 475 | sprintf(disk_uuid + len, "%pU", cmsg->uuid); |
faeff83f | 476 | snprintf(raid_slot, 16, "RAID_DISK=%d", le32_to_cpu(cmsg->raid_slot)); |
1aee41f6 GR |
477 | pr_info("%s:%d Sending kobject change with %s and %s\n", __func__, __LINE__, disk_uuid, raid_slot); |
478 | init_completion(&cinfo->newdisk_completion); | |
fa8259da | 479 | set_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
480 | kobject_uevent_env(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE, envp); |
481 | wait_for_completion_timeout(&cinfo->newdisk_completion, | |
482 | NEW_DEV_TIMEOUT); | |
fa8259da | 483 | clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
484 | } |
485 | ||
486 | ||
487 | static void process_metadata_update(struct mddev *mddev, struct cluster_msg *msg) | |
488 | { | |
489 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
15858fa5 GJ |
490 | mddev->good_device_nr = le32_to_cpu(msg->raid_slot); |
491 | set_bit(MD_RELOAD_SB, &mddev->flags); | |
1aee41f6 | 492 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); |
15858fa5 | 493 | md_wakeup_thread(mddev->thread); |
1aee41f6 GR |
494 | } |
495 | ||
88bcfef7 GR |
496 | static void process_remove_disk(struct mddev *mddev, struct cluster_msg *msg) |
497 | { | |
faeff83f GJ |
498 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, |
499 | le32_to_cpu(msg->raid_slot)); | |
88bcfef7 | 500 | |
659b254f GJ |
501 | if (rdev) { |
502 | set_bit(ClusterRemove, &rdev->flags); | |
503 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
504 | md_wakeup_thread(mddev->thread); | |
505 | } | |
88bcfef7 | 506 | else |
faeff83f GJ |
507 | pr_warn("%s: %d Could not find disk(%d) to REMOVE\n", |
508 | __func__, __LINE__, le32_to_cpu(msg->raid_slot)); | |
88bcfef7 GR |
509 | } |
510 | ||
97f6cd39 GR |
511 | static void process_readd_disk(struct mddev *mddev, struct cluster_msg *msg) |
512 | { | |
faeff83f GJ |
513 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, |
514 | le32_to_cpu(msg->raid_slot)); | |
97f6cd39 GR |
515 | |
516 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
517 | clear_bit(Faulty, &rdev->flags); | |
518 | else | |
faeff83f GJ |
519 | pr_warn("%s: %d Could not find disk(%d) which is faulty", |
520 | __func__, __LINE__, le32_to_cpu(msg->raid_slot)); | |
97f6cd39 GR |
521 | } |
522 | ||
1fa9a1ad | 523 | static int process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg) |
4664680c | 524 | { |
1fa9a1ad GJ |
525 | int ret = 0; |
526 | ||
256f5b24 GJ |
527 | if (WARN(mddev->cluster_info->slot_number - 1 == le32_to_cpu(msg->slot), |
528 | "node %d received it's own msg\n", le32_to_cpu(msg->slot))) | |
1fa9a1ad | 529 | return -1; |
cf97a348 | 530 | switch (le32_to_cpu(msg->type)) { |
4664680c | 531 | case METADATA_UPDATED: |
1aee41f6 | 532 | process_metadata_update(mddev, msg); |
4664680c GR |
533 | break; |
534 | case RESYNCING: | |
cf97a348 GJ |
535 | process_suspend_info(mddev, le32_to_cpu(msg->slot), |
536 | le64_to_cpu(msg->low), | |
537 | le64_to_cpu(msg->high)); | |
4664680c | 538 | break; |
1aee41f6 | 539 | case NEWDISK: |
1aee41f6 | 540 | process_add_new_disk(mddev, msg); |
88bcfef7 GR |
541 | break; |
542 | case REMOVE: | |
88bcfef7 GR |
543 | process_remove_disk(mddev, msg); |
544 | break; | |
97f6cd39 | 545 | case RE_ADD: |
97f6cd39 GR |
546 | process_readd_disk(mddev, msg); |
547 | break; | |
dc737d7c | 548 | case BITMAP_NEEDS_SYNC: |
cf97a348 | 549 | __recover_slot(mddev, le32_to_cpu(msg->slot)); |
dc737d7c | 550 | break; |
88bcfef7 | 551 | default: |
1fa9a1ad | 552 | ret = -1; |
88bcfef7 GR |
553 | pr_warn("%s:%d Received unknown message from %d\n", |
554 | __func__, __LINE__, msg->slot); | |
09dd1af2 | 555 | } |
1fa9a1ad | 556 | return ret; |
4664680c GR |
557 | } |
558 | ||
559 | /* | |
560 | * thread for receiving message | |
561 | */ | |
562 | static void recv_daemon(struct md_thread *thread) | |
563 | { | |
564 | struct md_cluster_info *cinfo = thread->mddev->cluster_info; | |
565 | struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres; | |
566 | struct dlm_lock_resource *message_lockres = cinfo->message_lockres; | |
567 | struct cluster_msg msg; | |
b5ef5678 | 568 | int ret; |
4664680c | 569 | |
8b9277c8 | 570 | mutex_lock(&cinfo->recv_mutex); |
4664680c GR |
571 | /*get CR on Message*/ |
572 | if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) { | |
573 | pr_err("md/raid1:failed to get CR on MESSAGE\n"); | |
8b9277c8 | 574 | mutex_unlock(&cinfo->recv_mutex); |
4664680c GR |
575 | return; |
576 | } | |
577 | ||
578 | /* read lvb and wake up thread to process this message_lockres */ | |
579 | memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg)); | |
1fa9a1ad GJ |
580 | ret = process_recvd_msg(thread->mddev, &msg); |
581 | if (ret) | |
582 | goto out; | |
4664680c GR |
583 | |
584 | /*release CR on ack_lockres*/ | |
b5ef5678 GJ |
585 | ret = dlm_unlock_sync(ack_lockres); |
586 | if (unlikely(ret != 0)) | |
587 | pr_info("unlock ack failed return %d\n", ret); | |
66099bb0 | 588 | /*up-convert to PR on message_lockres*/ |
b5ef5678 GJ |
589 | ret = dlm_lock_sync(message_lockres, DLM_LOCK_PR); |
590 | if (unlikely(ret != 0)) | |
591 | pr_info("lock PR on msg failed return %d\n", ret); | |
4664680c | 592 | /*get CR on ack_lockres again*/ |
b5ef5678 GJ |
593 | ret = dlm_lock_sync(ack_lockres, DLM_LOCK_CR); |
594 | if (unlikely(ret != 0)) | |
595 | pr_info("lock CR on ack failed return %d\n", ret); | |
1fa9a1ad | 596 | out: |
4664680c | 597 | /*release CR on message_lockres*/ |
b5ef5678 GJ |
598 | ret = dlm_unlock_sync(message_lockres); |
599 | if (unlikely(ret != 0)) | |
600 | pr_info("unlock msg failed return %d\n", ret); | |
8b9277c8 | 601 | mutex_unlock(&cinfo->recv_mutex); |
4664680c GR |
602 | } |
603 | ||
8b9277c8 | 604 | /* lock_token() |
601b515c GR |
605 | * Takes the lock on the TOKEN lock resource so no other |
606 | * node can communicate while the operation is underway. | |
607 | */ | |
8b9277c8 | 608 | static int lock_token(struct md_cluster_info *cinfo) |
601b515c GR |
609 | { |
610 | int error; | |
611 | ||
612 | error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX); | |
613 | if (error) | |
614 | pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n", | |
615 | __func__, __LINE__, error); | |
8b9277c8 GJ |
616 | |
617 | /* Lock the receive sequence */ | |
618 | mutex_lock(&cinfo->recv_mutex); | |
601b515c GR |
619 | return error; |
620 | } | |
621 | ||
8b9277c8 GJ |
622 | /* lock_comm() |
623 | * Sets the MD_CLUSTER_SEND_LOCK bit to lock the send channel. | |
624 | */ | |
625 | static int lock_comm(struct md_cluster_info *cinfo) | |
626 | { | |
627 | wait_event(cinfo->wait, | |
628 | !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state)); | |
629 | ||
630 | return lock_token(cinfo); | |
631 | } | |
632 | ||
601b515c GR |
633 | static void unlock_comm(struct md_cluster_info *cinfo) |
634 | { | |
dbb64f86 | 635 | WARN_ON(cinfo->token_lockres->mode != DLM_LOCK_EX); |
8b9277c8 | 636 | mutex_unlock(&cinfo->recv_mutex); |
601b515c | 637 | dlm_unlock_sync(cinfo->token_lockres); |
8b9277c8 GJ |
638 | clear_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state); |
639 | wake_up(&cinfo->wait); | |
601b515c GR |
640 | } |
641 | ||
642 | /* __sendmsg() | |
643 | * This function performs the actual sending of the message. This function is | |
644 | * usually called after performing the encompassing operation | |
645 | * The function: | |
646 | * 1. Grabs the message lockresource in EX mode | |
647 | * 2. Copies the message to the message LVB | |
66099bb0 | 648 | * 3. Downconverts message lockresource to CW |
601b515c GR |
649 | * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes |
650 | * and the other nodes read the message. The thread will wait here until all other | |
651 | * nodes have released ack lock resource. | |
652 | * 5. Downconvert ack lockresource to CR | |
653 | */ | |
654 | static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
655 | { | |
656 | int error; | |
657 | int slot = cinfo->slot_number - 1; | |
658 | ||
659 | cmsg->slot = cpu_to_le32(slot); | |
660 | /*get EX on Message*/ | |
661 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX); | |
662 | if (error) { | |
663 | pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error); | |
664 | goto failed_message; | |
665 | } | |
666 | ||
667 | memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg, | |
668 | sizeof(struct cluster_msg)); | |
66099bb0 GJ |
669 | /*down-convert EX to CW on Message*/ |
670 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CW); | |
601b515c | 671 | if (error) { |
66099bb0 | 672 | pr_err("md-cluster: failed to convert EX to CW on MESSAGE(%d)\n", |
601b515c | 673 | error); |
66099bb0 | 674 | goto failed_ack; |
601b515c GR |
675 | } |
676 | ||
677 | /*up-convert CR to EX on Ack*/ | |
678 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX); | |
679 | if (error) { | |
680 | pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n", | |
681 | error); | |
682 | goto failed_ack; | |
683 | } | |
684 | ||
685 | /*down-convert EX to CR on Ack*/ | |
686 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR); | |
687 | if (error) { | |
688 | pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n", | |
689 | error); | |
690 | goto failed_ack; | |
691 | } | |
692 | ||
693 | failed_ack: | |
b5ef5678 GJ |
694 | error = dlm_unlock_sync(cinfo->message_lockres); |
695 | if (unlikely(error != 0)) { | |
696 | pr_err("md-cluster: failed convert to NL on MESSAGE(%d)\n", | |
697 | error); | |
698 | /* in case the message can't be released due to some reason */ | |
699 | goto failed_ack; | |
700 | } | |
601b515c GR |
701 | failed_message: |
702 | return error; | |
703 | } | |
704 | ||
705 | static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
706 | { | |
707 | int ret; | |
708 | ||
709 | lock_comm(cinfo); | |
710 | ret = __sendmsg(cinfo, cmsg); | |
711 | unlock_comm(cinfo); | |
712 | return ret; | |
713 | } | |
714 | ||
96ae923a GR |
715 | static int gather_all_resync_info(struct mddev *mddev, int total_slots) |
716 | { | |
717 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
718 | int i, ret = 0; | |
719 | struct dlm_lock_resource *bm_lockres; | |
720 | struct suspend_info *s; | |
721 | char str[64]; | |
abb9b22a | 722 | sector_t lo, hi; |
96ae923a GR |
723 | |
724 | ||
725 | for (i = 0; i < total_slots; i++) { | |
726 | memset(str, '\0', 64); | |
727 | snprintf(str, 64, "bitmap%04d", i); | |
728 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
729 | if (!bm_lockres) | |
730 | return -ENOMEM; | |
4ac7a65f SL |
731 | if (i == (cinfo->slot_number - 1)) { |
732 | lockres_free(bm_lockres); | |
96ae923a | 733 | continue; |
4ac7a65f | 734 | } |
96ae923a GR |
735 | |
736 | bm_lockres->flags |= DLM_LKF_NOQUEUE; | |
737 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
738 | if (ret == -EAGAIN) { | |
739 | memset(bm_lockres->lksb.sb_lvbptr, '\0', LVB_SIZE); | |
740 | s = read_resync_info(mddev, bm_lockres); | |
741 | if (s) { | |
742 | pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n", | |
743 | __func__, __LINE__, | |
744 | (unsigned long long) s->lo, | |
745 | (unsigned long long) s->hi, i); | |
746 | spin_lock_irq(&cinfo->suspend_lock); | |
747 | s->slot = i; | |
748 | list_add(&s->list, &cinfo->suspend_list); | |
749 | spin_unlock_irq(&cinfo->suspend_lock); | |
750 | } | |
751 | ret = 0; | |
752 | lockres_free(bm_lockres); | |
753 | continue; | |
754 | } | |
6e6d9f2c GJ |
755 | if (ret) { |
756 | lockres_free(bm_lockres); | |
96ae923a | 757 | goto out; |
6e6d9f2c | 758 | } |
abb9b22a GJ |
759 | |
760 | /* Read the disk bitmap sb and check if it needs recovery */ | |
761 | ret = bitmap_copy_from_slot(mddev, i, &lo, &hi, false); | |
762 | if (ret) { | |
763 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", i); | |
764 | lockres_free(bm_lockres); | |
765 | continue; | |
766 | } | |
767 | if ((hi > 0) && (lo < mddev->recovery_cp)) { | |
768 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
769 | mddev->recovery_cp = lo; | |
770 | md_check_recovery(mddev); | |
771 | } | |
772 | ||
96ae923a GR |
773 | dlm_unlock_sync(bm_lockres); |
774 | lockres_free(bm_lockres); | |
775 | } | |
776 | out: | |
777 | return ret; | |
778 | } | |
779 | ||
edb39c9d GR |
780 | static int join(struct mddev *mddev, int nodes) |
781 | { | |
c4ce867f | 782 | struct md_cluster_info *cinfo; |
cf921cc1 | 783 | int ret, ops_rv; |
c4ce867f GR |
784 | char str[64]; |
785 | ||
c4ce867f GR |
786 | cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL); |
787 | if (!cinfo) | |
788 | return -ENOMEM; | |
789 | ||
9e3072e3 GJ |
790 | INIT_LIST_HEAD(&cinfo->suspend_list); |
791 | spin_lock_init(&cinfo->suspend_lock); | |
cf921cc1 | 792 | init_completion(&cinfo->completion); |
eece075c | 793 | set_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state); |
8b9277c8 GJ |
794 | init_waitqueue_head(&cinfo->wait); |
795 | mutex_init(&cinfo->recv_mutex); | |
cf921cc1 | 796 | |
cf921cc1 GR |
797 | mddev->cluster_info = cinfo; |
798 | ||
c4ce867f | 799 | memset(str, 0, 64); |
b89f704a | 800 | sprintf(str, "%pU", mddev->uuid); |
cf921cc1 GR |
801 | ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name, |
802 | DLM_LSFL_FS, LVB_SIZE, | |
803 | &md_ls_ops, mddev, &ops_rv, &cinfo->lockspace); | |
c4ce867f GR |
804 | if (ret) |
805 | goto err; | |
cf921cc1 | 806 | wait_for_completion(&cinfo->completion); |
8c58f02e GJ |
807 | if (nodes < cinfo->slot_number) { |
808 | pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).", | |
809 | cinfo->slot_number, nodes); | |
b97e9257 GR |
810 | ret = -ERANGE; |
811 | goto err; | |
812 | } | |
4664680c GR |
813 | /* Initiate the communication resources */ |
814 | ret = -ENOMEM; | |
815 | cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv"); | |
816 | if (!cinfo->recv_thread) { | |
817 | pr_err("md-cluster: cannot allocate memory for recv_thread!\n"); | |
818 | goto err; | |
819 | } | |
820 | cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1); | |
821 | if (!cinfo->message_lockres) | |
822 | goto err; | |
823 | cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0); | |
824 | if (!cinfo->token_lockres) | |
825 | goto err; | |
1aee41f6 GR |
826 | cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0); |
827 | if (!cinfo->no_new_dev_lockres) | |
828 | goto err; | |
829 | ||
1535212c GJ |
830 | ret = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX); |
831 | if (ret) { | |
832 | ret = -EAGAIN; | |
833 | pr_err("md-cluster: can't join cluster to avoid lock issue\n"); | |
834 | goto err; | |
835 | } | |
836 | cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0); | |
837 | if (!cinfo->ack_lockres) | |
838 | goto err; | |
4664680c GR |
839 | /* get sync CR lock on ACK. */ |
840 | if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR)) | |
841 | pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n", | |
842 | ret); | |
1535212c | 843 | dlm_unlock_sync(cinfo->token_lockres); |
1aee41f6 GR |
844 | /* get sync CR lock on no-new-dev. */ |
845 | if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR)) | |
846 | pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret); | |
847 | ||
54519c5f GR |
848 | |
849 | pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number); | |
850 | snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1); | |
851 | cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1); | |
852 | if (!cinfo->bitmap_lockres) | |
853 | goto err; | |
854 | if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) { | |
855 | pr_err("Failed to get bitmap lock\n"); | |
856 | ret = -EINVAL; | |
857 | goto err; | |
858 | } | |
859 | ||
c186b128 GR |
860 | cinfo->resync_lockres = lockres_init(mddev, "resync", NULL, 0); |
861 | if (!cinfo->resync_lockres) | |
862 | goto err; | |
863 | ||
edb39c9d | 864 | return 0; |
c4ce867f | 865 | err: |
5b0fb33e GJ |
866 | md_unregister_thread(&cinfo->recovery_thread); |
867 | md_unregister_thread(&cinfo->recv_thread); | |
4664680c GR |
868 | lockres_free(cinfo->message_lockres); |
869 | lockres_free(cinfo->token_lockres); | |
870 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 871 | lockres_free(cinfo->no_new_dev_lockres); |
c186b128 | 872 | lockres_free(cinfo->resync_lockres); |
96ae923a | 873 | lockres_free(cinfo->bitmap_lockres); |
c4ce867f GR |
874 | if (cinfo->lockspace) |
875 | dlm_release_lockspace(cinfo->lockspace, 2); | |
cf921cc1 | 876 | mddev->cluster_info = NULL; |
c4ce867f | 877 | kfree(cinfo); |
c4ce867f | 878 | return ret; |
edb39c9d GR |
879 | } |
880 | ||
51e453ae GJ |
881 | static void load_bitmaps(struct mddev *mddev, int total_slots) |
882 | { | |
883 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
884 | ||
885 | /* load all the node's bitmap info for resync */ | |
886 | if (gather_all_resync_info(mddev, total_slots)) | |
887 | pr_err("md-cluster: failed to gather all resyn infos\n"); | |
888 | set_bit(MD_CLUSTER_ALREADY_IN_CLUSTER, &cinfo->state); | |
889 | /* wake up recv thread in case something need to be handled */ | |
890 | if (test_and_clear_bit(MD_CLUSTER_PENDING_RECV_EVENT, &cinfo->state)) | |
891 | md_wakeup_thread(cinfo->recv_thread); | |
892 | } | |
893 | ||
09995411 GJ |
894 | static void resync_bitmap(struct mddev *mddev) |
895 | { | |
896 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
897 | struct cluster_msg cmsg = {0}; | |
898 | int err; | |
899 | ||
900 | cmsg.type = cpu_to_le32(BITMAP_NEEDS_SYNC); | |
901 | err = sendmsg(cinfo, &cmsg); | |
902 | if (err) | |
903 | pr_err("%s:%d: failed to send BITMAP_NEEDS_SYNC message (%d)\n", | |
904 | __func__, __LINE__, err); | |
905 | } | |
906 | ||
f6a2dc64 | 907 | static void unlock_all_bitmaps(struct mddev *mddev); |
edb39c9d GR |
908 | static int leave(struct mddev *mddev) |
909 | { | |
c4ce867f GR |
910 | struct md_cluster_info *cinfo = mddev->cluster_info; |
911 | ||
912 | if (!cinfo) | |
913 | return 0; | |
09995411 GJ |
914 | |
915 | /* BITMAP_NEEDS_SYNC message should be sent when node | |
916 | * is leaving the cluster with dirty bitmap, also we | |
917 | * can only deliver it when dlm connection is available */ | |
918 | if (cinfo->slot_number > 0 && mddev->recovery_cp != MaxSector) | |
919 | resync_bitmap(mddev); | |
920 | ||
e94987db | 921 | md_unregister_thread(&cinfo->recovery_thread); |
4664680c GR |
922 | md_unregister_thread(&cinfo->recv_thread); |
923 | lockres_free(cinfo->message_lockres); | |
924 | lockres_free(cinfo->token_lockres); | |
925 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 926 | lockres_free(cinfo->no_new_dev_lockres); |
4ac7a65f | 927 | lockres_free(cinfo->resync_lockres); |
54519c5f | 928 | lockres_free(cinfo->bitmap_lockres); |
f6a2dc64 | 929 | unlock_all_bitmaps(mddev); |
c4ce867f | 930 | dlm_release_lockspace(cinfo->lockspace, 2); |
edb39c9d GR |
931 | return 0; |
932 | } | |
933 | ||
cf921cc1 GR |
934 | /* slot_number(): Returns the MD slot number to use |
935 | * DLM starts the slot numbers from 1, wheras cluster-md | |
936 | * wants the number to be from zero, so we deduct one | |
937 | */ | |
938 | static int slot_number(struct mddev *mddev) | |
939 | { | |
940 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
941 | ||
942 | return cinfo->slot_number - 1; | |
943 | } | |
944 | ||
8b9277c8 GJ |
945 | /* |
946 | * Check if the communication is already locked, else lock the communication | |
947 | * channel. | |
948 | * If it is already locked, token is in EX mode, and hence lock_token() | |
949 | * should not be called. | |
950 | */ | |
293467aa GR |
951 | static int metadata_update_start(struct mddev *mddev) |
952 | { | |
8b9277c8 GJ |
953 | struct md_cluster_info *cinfo = mddev->cluster_info; |
954 | ||
955 | wait_event(cinfo->wait, | |
956 | !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state) || | |
957 | test_and_clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state)); | |
958 | ||
959 | /* If token is already locked, return 0 */ | |
960 | if (cinfo->token_lockres->mode == DLM_LOCK_EX) | |
961 | return 0; | |
962 | ||
963 | return lock_token(cinfo); | |
293467aa GR |
964 | } |
965 | ||
966 | static int metadata_update_finish(struct mddev *mddev) | |
967 | { | |
968 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
969 | struct cluster_msg cmsg; | |
70bcecdb GR |
970 | struct md_rdev *rdev; |
971 | int ret = 0; | |
ba2746b0 | 972 | int raid_slot = -1; |
293467aa GR |
973 | |
974 | memset(&cmsg, 0, sizeof(cmsg)); | |
975 | cmsg.type = cpu_to_le32(METADATA_UPDATED); | |
70bcecdb GR |
976 | /* Pick up a good active device number to send. |
977 | */ | |
978 | rdev_for_each(rdev, mddev) | |
979 | if (rdev->raid_disk > -1 && !test_bit(Faulty, &rdev->flags)) { | |
ba2746b0 | 980 | raid_slot = rdev->desc_nr; |
70bcecdb GR |
981 | break; |
982 | } | |
ba2746b0 N |
983 | if (raid_slot >= 0) { |
984 | cmsg.raid_slot = cpu_to_le32(raid_slot); | |
70bcecdb | 985 | ret = __sendmsg(cinfo, &cmsg); |
ba2746b0 | 986 | } else |
70bcecdb | 987 | pr_warn("md-cluster: No good device id found to send\n"); |
8b9277c8 | 988 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
293467aa GR |
989 | unlock_comm(cinfo); |
990 | return ret; | |
991 | } | |
992 | ||
dbb64f86 | 993 | static void metadata_update_cancel(struct mddev *mddev) |
293467aa GR |
994 | { |
995 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
8b9277c8 | 996 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
dbb64f86 | 997 | unlock_comm(cinfo); |
293467aa GR |
998 | } |
999 | ||
c186b128 GR |
1000 | static int resync_start(struct mddev *mddev) |
1001 | { | |
1002 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
c186b128 GR |
1003 | return dlm_lock_sync(cinfo->resync_lockres, DLM_LOCK_EX); |
1004 | } | |
1005 | ||
c40f341f | 1006 | static int resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi) |
965400eb GR |
1007 | { |
1008 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
ac277c6a | 1009 | struct resync_info ri; |
aee177ac | 1010 | struct cluster_msg cmsg = {0}; |
965400eb | 1011 | |
ac277c6a GR |
1012 | /* do not send zero again, if we have sent before */ |
1013 | if (hi == 0) { | |
1014 | memcpy(&ri, cinfo->bitmap_lockres->lksb.sb_lvbptr, sizeof(struct resync_info)); | |
1015 | if (le64_to_cpu(ri.hi) == 0) | |
1016 | return 0; | |
1017 | } | |
1018 | ||
30661b49 | 1019 | add_resync_info(cinfo->bitmap_lockres, lo, hi); |
c40f341f GR |
1020 | /* Re-acquire the lock to refresh LVB */ |
1021 | dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW); | |
c40f341f | 1022 | cmsg.type = cpu_to_le32(RESYNCING); |
965400eb GR |
1023 | cmsg.low = cpu_to_le64(lo); |
1024 | cmsg.high = cpu_to_le64(hi); | |
c186b128 | 1025 | |
965400eb GR |
1026 | return sendmsg(cinfo, &cmsg); |
1027 | } | |
1028 | ||
c186b128 GR |
1029 | static int resync_finish(struct mddev *mddev) |
1030 | { | |
1031 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
c186b128 GR |
1032 | dlm_unlock_sync(cinfo->resync_lockres); |
1033 | return resync_info_update(mddev, 0, 0); | |
1034 | } | |
1035 | ||
90382ed9 GR |
1036 | static int area_resyncing(struct mddev *mddev, int direction, |
1037 | sector_t lo, sector_t hi) | |
589a1c49 GR |
1038 | { |
1039 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1040 | int ret = 0; | |
1041 | struct suspend_info *s; | |
1042 | ||
90382ed9 GR |
1043 | if ((direction == READ) && |
1044 | test_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state)) | |
1045 | return 1; | |
1046 | ||
589a1c49 GR |
1047 | spin_lock_irq(&cinfo->suspend_lock); |
1048 | if (list_empty(&cinfo->suspend_list)) | |
1049 | goto out; | |
1050 | list_for_each_entry(s, &cinfo->suspend_list, list) | |
1051 | if (hi > s->lo && lo < s->hi) { | |
1052 | ret = 1; | |
1053 | break; | |
1054 | } | |
1055 | out: | |
1056 | spin_unlock_irq(&cinfo->suspend_lock); | |
1057 | return ret; | |
1058 | } | |
1059 | ||
dbb64f86 GR |
1060 | /* add_new_disk() - initiates a disk add |
1061 | * However, if this fails before writing md_update_sb(), | |
1062 | * add_new_disk_cancel() must be called to release token lock | |
1063 | */ | |
1064 | static int add_new_disk(struct mddev *mddev, struct md_rdev *rdev) | |
1aee41f6 GR |
1065 | { |
1066 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1067 | struct cluster_msg cmsg; | |
1068 | int ret = 0; | |
1069 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); | |
1070 | char *uuid = sb->device_uuid; | |
1071 | ||
1072 | memset(&cmsg, 0, sizeof(cmsg)); | |
1073 | cmsg.type = cpu_to_le32(NEWDISK); | |
1074 | memcpy(cmsg.uuid, uuid, 16); | |
faeff83f | 1075 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); |
1aee41f6 GR |
1076 | lock_comm(cinfo); |
1077 | ret = __sendmsg(cinfo, &cmsg); | |
1078 | if (ret) | |
1079 | return ret; | |
1080 | cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE; | |
1081 | ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX); | |
1082 | cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE; | |
1083 | /* Some node does not "see" the device */ | |
1084 | if (ret == -EAGAIN) | |
1085 | ret = -ENOENT; | |
dbb64f86 GR |
1086 | if (ret) |
1087 | unlock_comm(cinfo); | |
8b9277c8 | 1088 | else { |
1aee41f6 | 1089 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); |
e19508fa GJ |
1090 | /* Since MD_CHANGE_DEVS will be set in add_bound_rdev which |
1091 | * will run soon after add_new_disk, the below path will be | |
1092 | * invoked: | |
1093 | * md_wakeup_thread(mddev->thread) | |
1094 | * -> conf->thread (raid1d) | |
1095 | * -> md_check_recovery -> md_update_sb | |
1096 | * -> metadata_update_start/finish | |
1097 | * MD_CLUSTER_SEND_LOCKED_ALREADY will be cleared eventually. | |
1098 | * | |
1099 | * For other failure cases, metadata_update_cancel and | |
1100 | * add_new_disk_cancel also clear below bit as well. | |
1101 | * */ | |
8b9277c8 GJ |
1102 | set_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
1103 | wake_up(&cinfo->wait); | |
1104 | } | |
1aee41f6 GR |
1105 | return ret; |
1106 | } | |
1107 | ||
dbb64f86 | 1108 | static void add_new_disk_cancel(struct mddev *mddev) |
1aee41f6 | 1109 | { |
dbb64f86 | 1110 | struct md_cluster_info *cinfo = mddev->cluster_info; |
8b9277c8 | 1111 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
dbb64f86 | 1112 | unlock_comm(cinfo); |
1aee41f6 GR |
1113 | } |
1114 | ||
fa8259da | 1115 | static int new_disk_ack(struct mddev *mddev, bool ack) |
1aee41f6 GR |
1116 | { |
1117 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1118 | ||
fa8259da GR |
1119 | if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) { |
1120 | pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev)); | |
1121 | return -EINVAL; | |
1122 | } | |
1123 | ||
1aee41f6 GR |
1124 | if (ack) |
1125 | dlm_unlock_sync(cinfo->no_new_dev_lockres); | |
1126 | complete(&cinfo->newdisk_completion); | |
fa8259da | 1127 | return 0; |
1aee41f6 GR |
1128 | } |
1129 | ||
88bcfef7 GR |
1130 | static int remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1131 | { | |
aee177ac | 1132 | struct cluster_msg cmsg = {0}; |
88bcfef7 | 1133 | struct md_cluster_info *cinfo = mddev->cluster_info; |
faeff83f GJ |
1134 | cmsg.type = cpu_to_le32(REMOVE); |
1135 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); | |
54a88392 | 1136 | return sendmsg(cinfo, &cmsg); |
88bcfef7 GR |
1137 | } |
1138 | ||
f6a2dc64 GJ |
1139 | static int lock_all_bitmaps(struct mddev *mddev) |
1140 | { | |
1141 | int slot, my_slot, ret, held = 1, i = 0; | |
1142 | char str[64]; | |
1143 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1144 | ||
1145 | cinfo->other_bitmap_lockres = kzalloc((mddev->bitmap_info.nodes - 1) * | |
1146 | sizeof(struct dlm_lock_resource *), | |
1147 | GFP_KERNEL); | |
1148 | if (!cinfo->other_bitmap_lockres) { | |
1149 | pr_err("md: can't alloc mem for other bitmap locks\n"); | |
1150 | return 0; | |
1151 | } | |
1152 | ||
1153 | my_slot = slot_number(mddev); | |
1154 | for (slot = 0; slot < mddev->bitmap_info.nodes; slot++) { | |
1155 | if (slot == my_slot) | |
1156 | continue; | |
1157 | ||
1158 | memset(str, '\0', 64); | |
1159 | snprintf(str, 64, "bitmap%04d", slot); | |
1160 | cinfo->other_bitmap_lockres[i] = lockres_init(mddev, str, NULL, 1); | |
1161 | if (!cinfo->other_bitmap_lockres[i]) | |
1162 | return -ENOMEM; | |
1163 | ||
1164 | cinfo->other_bitmap_lockres[i]->flags |= DLM_LKF_NOQUEUE; | |
1165 | ret = dlm_lock_sync(cinfo->other_bitmap_lockres[i], DLM_LOCK_PW); | |
1166 | if (ret) | |
1167 | held = -1; | |
1168 | i++; | |
1169 | } | |
1170 | ||
1171 | return held; | |
1172 | } | |
1173 | ||
1174 | static void unlock_all_bitmaps(struct mddev *mddev) | |
1175 | { | |
1176 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1177 | int i; | |
1178 | ||
1179 | /* release other node's bitmap lock if they are existed */ | |
1180 | if (cinfo->other_bitmap_lockres) { | |
1181 | for (i = 0; i < mddev->bitmap_info.nodes - 1; i++) { | |
1182 | if (cinfo->other_bitmap_lockres[i]) { | |
1183 | dlm_unlock_sync(cinfo->other_bitmap_lockres[i]); | |
1184 | lockres_free(cinfo->other_bitmap_lockres[i]); | |
1185 | } | |
1186 | } | |
1187 | kfree(cinfo->other_bitmap_lockres); | |
1188 | } | |
1189 | } | |
1190 | ||
97f6cd39 GR |
1191 | static int gather_bitmaps(struct md_rdev *rdev) |
1192 | { | |
1193 | int sn, err; | |
1194 | sector_t lo, hi; | |
aee177ac | 1195 | struct cluster_msg cmsg = {0}; |
97f6cd39 GR |
1196 | struct mddev *mddev = rdev->mddev; |
1197 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1198 | ||
faeff83f GJ |
1199 | cmsg.type = cpu_to_le32(RE_ADD); |
1200 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); | |
97f6cd39 GR |
1201 | err = sendmsg(cinfo, &cmsg); |
1202 | if (err) | |
1203 | goto out; | |
1204 | ||
1205 | for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) { | |
1206 | if (sn == (cinfo->slot_number - 1)) | |
1207 | continue; | |
1208 | err = bitmap_copy_from_slot(mddev, sn, &lo, &hi, false); | |
1209 | if (err) { | |
1210 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn); | |
1211 | goto out; | |
1212 | } | |
1213 | if ((hi > 0) && (lo < mddev->recovery_cp)) | |
1214 | mddev->recovery_cp = lo; | |
1215 | } | |
1216 | out: | |
1217 | return err; | |
1218 | } | |
1219 | ||
edb39c9d GR |
1220 | static struct md_cluster_operations cluster_ops = { |
1221 | .join = join, | |
1222 | .leave = leave, | |
cf921cc1 | 1223 | .slot_number = slot_number, |
c186b128 GR |
1224 | .resync_start = resync_start, |
1225 | .resync_finish = resync_finish, | |
96ae923a | 1226 | .resync_info_update = resync_info_update, |
293467aa GR |
1227 | .metadata_update_start = metadata_update_start, |
1228 | .metadata_update_finish = metadata_update_finish, | |
1229 | .metadata_update_cancel = metadata_update_cancel, | |
589a1c49 | 1230 | .area_resyncing = area_resyncing, |
dbb64f86 GR |
1231 | .add_new_disk = add_new_disk, |
1232 | .add_new_disk_cancel = add_new_disk_cancel, | |
1aee41f6 | 1233 | .new_disk_ack = new_disk_ack, |
88bcfef7 | 1234 | .remove_disk = remove_disk, |
51e453ae | 1235 | .load_bitmaps = load_bitmaps, |
97f6cd39 | 1236 | .gather_bitmaps = gather_bitmaps, |
f6a2dc64 GJ |
1237 | .lock_all_bitmaps = lock_all_bitmaps, |
1238 | .unlock_all_bitmaps = unlock_all_bitmaps, | |
edb39c9d GR |
1239 | }; |
1240 | ||
8e854e9c GR |
1241 | static int __init cluster_init(void) |
1242 | { | |
1243 | pr_warn("md-cluster: EXPERIMENTAL. Use with caution\n"); | |
1244 | pr_info("Registering Cluster MD functions\n"); | |
edb39c9d | 1245 | register_md_cluster_operations(&cluster_ops, THIS_MODULE); |
8e854e9c GR |
1246 | return 0; |
1247 | } | |
1248 | ||
1249 | static void cluster_exit(void) | |
1250 | { | |
edb39c9d | 1251 | unregister_md_cluster_operations(); |
8e854e9c GR |
1252 | } |
1253 | ||
1254 | module_init(cluster_init); | |
1255 | module_exit(cluster_exit); | |
86b57277 | 1256 | MODULE_AUTHOR("SUSE"); |
8e854e9c GR |
1257 | MODULE_LICENSE("GPL"); |
1258 | MODULE_DESCRIPTION("Clustering support for MD"); |