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