]>
Commit | Line | Data |
---|---|---|
8e854e9c GR |
1 | /* |
2 | * Copyright (C) 2015, SUSE | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2, or (at your option) | |
7 | * any later version. | |
8 | * | |
9 | */ | |
10 | ||
11 | ||
12 | #include <linux/module.h> | |
47741b7c GR |
13 | #include <linux/dlm.h> |
14 | #include <linux/sched.h> | |
1aee41f6 | 15 | #include <linux/raid/md_p.h> |
47741b7c | 16 | #include "md.h" |
e94987db | 17 | #include "bitmap.h" |
edb39c9d | 18 | #include "md-cluster.h" |
47741b7c GR |
19 | |
20 | #define LVB_SIZE 64 | |
1aee41f6 | 21 | #define NEW_DEV_TIMEOUT 5000 |
47741b7c GR |
22 | |
23 | struct dlm_lock_resource { | |
24 | dlm_lockspace_t *ls; | |
25 | struct dlm_lksb lksb; | |
26 | char *name; /* lock name. */ | |
27 | uint32_t flags; /* flags to pass to dlm_lock() */ | |
47741b7c | 28 | struct completion completion; /* completion for synchronized locking */ |
c4ce867f GR |
29 | void (*bast)(void *arg, int mode); /* blocking AST function pointer*/ |
30 | struct mddev *mddev; /* pointing back to mddev. */ | |
dbb64f86 | 31 | int mode; |
c4ce867f GR |
32 | }; |
33 | ||
96ae923a GR |
34 | struct suspend_info { |
35 | int slot; | |
36 | sector_t lo; | |
37 | sector_t hi; | |
38 | struct list_head list; | |
39 | }; | |
40 | ||
41 | struct resync_info { | |
42 | __le64 lo; | |
43 | __le64 hi; | |
44 | }; | |
45 | ||
fa8259da GR |
46 | /* md_cluster_info flags */ |
47 | #define MD_CLUSTER_WAITING_FOR_NEWDISK 1 | |
90382ed9 | 48 | #define MD_CLUSTER_SUSPEND_READ_BALANCING 2 |
eece075c | 49 | #define MD_CLUSTER_BEGIN_JOIN_CLUSTER 3 |
fa8259da | 50 | |
8b9277c8 GJ |
51 | /* Lock the send communication. This is done through |
52 | * bit manipulation as opposed to a mutex in order to | |
53 | * accomodate lock and hold. See next comment. | |
54 | */ | |
55 | #define MD_CLUSTER_SEND_LOCK 4 | |
e19508fa GJ |
56 | /* If cluster operations (such as adding a disk) must lock the |
57 | * communication channel, so as to perform extra operations | |
58 | * (update metadata) and no other operation is allowed on the | |
59 | * MD. Token needs to be locked and held until the operation | |
60 | * completes witha md_update_sb(), which would eventually release | |
61 | * the lock. | |
8b9277c8 GJ |
62 | */ |
63 | #define MD_CLUSTER_SEND_LOCKED_ALREADY 5 | |
64 | ||
fa8259da | 65 | |
c4ce867f GR |
66 | struct md_cluster_info { |
67 | /* dlm lock space and resources for clustered raid. */ | |
68 | dlm_lockspace_t *lockspace; | |
cf921cc1 GR |
69 | int slot_number; |
70 | struct completion completion; | |
8b9277c8 | 71 | struct mutex recv_mutex; |
54519c5f | 72 | struct dlm_lock_resource *bitmap_lockres; |
f6a2dc64 | 73 | struct dlm_lock_resource **other_bitmap_lockres; |
c186b128 | 74 | struct dlm_lock_resource *resync_lockres; |
96ae923a GR |
75 | struct list_head suspend_list; |
76 | spinlock_t suspend_lock; | |
e94987db GR |
77 | struct md_thread *recovery_thread; |
78 | unsigned long recovery_map; | |
4664680c GR |
79 | /* communication loc resources */ |
80 | struct dlm_lock_resource *ack_lockres; | |
81 | struct dlm_lock_resource *message_lockres; | |
82 | struct dlm_lock_resource *token_lockres; | |
1aee41f6 | 83 | struct dlm_lock_resource *no_new_dev_lockres; |
4664680c | 84 | struct md_thread *recv_thread; |
1aee41f6 | 85 | struct completion newdisk_completion; |
8b9277c8 | 86 | wait_queue_head_t wait; |
fa8259da | 87 | unsigned long state; |
4664680c GR |
88 | }; |
89 | ||
90 | enum msg_type { | |
91 | METADATA_UPDATED = 0, | |
92 | RESYNCING, | |
1aee41f6 | 93 | NEWDISK, |
88bcfef7 | 94 | REMOVE, |
97f6cd39 | 95 | RE_ADD, |
dc737d7c | 96 | BITMAP_NEEDS_SYNC, |
4664680c GR |
97 | }; |
98 | ||
99 | struct cluster_msg { | |
cf97a348 GJ |
100 | __le32 type; |
101 | __le32 slot; | |
1aee41f6 | 102 | /* TODO: Unionize this for smaller footprint */ |
cf97a348 GJ |
103 | __le64 low; |
104 | __le64 high; | |
1aee41f6 | 105 | char uuid[16]; |
cf97a348 | 106 | __le32 raid_slot; |
47741b7c GR |
107 | }; |
108 | ||
109 | static void sync_ast(void *arg) | |
110 | { | |
111 | struct dlm_lock_resource *res; | |
112 | ||
2e2a7cd9 | 113 | res = arg; |
47741b7c GR |
114 | complete(&res->completion); |
115 | } | |
116 | ||
117 | static int dlm_lock_sync(struct dlm_lock_resource *res, int mode) | |
118 | { | |
119 | int ret = 0; | |
120 | ||
47741b7c GR |
121 | ret = dlm_lock(res->ls, mode, &res->lksb, |
122 | res->flags, res->name, strlen(res->name), | |
123 | 0, sync_ast, res, res->bast); | |
124 | if (ret) | |
125 | return ret; | |
126 | wait_for_completion(&res->completion); | |
dbb64f86 GR |
127 | if (res->lksb.sb_status == 0) |
128 | res->mode = mode; | |
47741b7c GR |
129 | return res->lksb.sb_status; |
130 | } | |
131 | ||
132 | static int dlm_unlock_sync(struct dlm_lock_resource *res) | |
133 | { | |
134 | return dlm_lock_sync(res, DLM_LOCK_NL); | |
135 | } | |
136 | ||
c4ce867f | 137 | static struct dlm_lock_resource *lockres_init(struct mddev *mddev, |
47741b7c GR |
138 | char *name, void (*bastfn)(void *arg, int mode), int with_lvb) |
139 | { | |
140 | struct dlm_lock_resource *res = NULL; | |
141 | int ret, namelen; | |
c4ce867f | 142 | struct md_cluster_info *cinfo = mddev->cluster_info; |
47741b7c GR |
143 | |
144 | res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL); | |
145 | if (!res) | |
146 | return NULL; | |
b83d51c0 | 147 | init_completion(&res->completion); |
c4ce867f GR |
148 | res->ls = cinfo->lockspace; |
149 | res->mddev = mddev; | |
dbb64f86 | 150 | res->mode = DLM_LOCK_IV; |
47741b7c GR |
151 | namelen = strlen(name); |
152 | res->name = kzalloc(namelen + 1, GFP_KERNEL); | |
153 | if (!res->name) { | |
154 | pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name); | |
155 | goto out_err; | |
156 | } | |
157 | strlcpy(res->name, name, namelen + 1); | |
158 | if (with_lvb) { | |
159 | res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL); | |
160 | if (!res->lksb.sb_lvbptr) { | |
161 | pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name); | |
162 | goto out_err; | |
163 | } | |
164 | res->flags = DLM_LKF_VALBLK; | |
165 | } | |
166 | ||
167 | if (bastfn) | |
168 | res->bast = bastfn; | |
169 | ||
170 | res->flags |= DLM_LKF_EXPEDITE; | |
171 | ||
172 | ret = dlm_lock_sync(res, DLM_LOCK_NL); | |
173 | if (ret) { | |
174 | pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name); | |
175 | goto out_err; | |
176 | } | |
177 | res->flags &= ~DLM_LKF_EXPEDITE; | |
178 | res->flags |= DLM_LKF_CONVERT; | |
179 | ||
180 | return res; | |
181 | out_err: | |
182 | kfree(res->lksb.sb_lvbptr); | |
183 | kfree(res->name); | |
184 | kfree(res); | |
185 | return NULL; | |
186 | } | |
187 | ||
188 | static void lockres_free(struct dlm_lock_resource *res) | |
189 | { | |
b5ef5678 GJ |
190 | int ret; |
191 | ||
47741b7c GR |
192 | if (!res) |
193 | return; | |
194 | ||
b5ef5678 GJ |
195 | /* cancel a lock request or a conversion request that is blocked */ |
196 | res->flags |= DLM_LKF_CANCEL; | |
197 | retry: | |
198 | ret = dlm_unlock(res->ls, res->lksb.sb_lkid, 0, &res->lksb, res); | |
199 | if (unlikely(ret != 0)) { | |
200 | pr_info("%s: failed to unlock %s return %d\n", __func__, res->name, ret); | |
201 | ||
202 | /* if a lock conversion is cancelled, then the lock is put | |
203 | * back to grant queue, need to ensure it is unlocked */ | |
204 | if (ret == -DLM_ECANCEL) | |
205 | goto retry; | |
206 | } | |
207 | res->flags &= ~DLM_LKF_CANCEL; | |
47741b7c GR |
208 | wait_for_completion(&res->completion); |
209 | ||
210 | kfree(res->name); | |
211 | kfree(res->lksb.sb_lvbptr); | |
212 | kfree(res); | |
213 | } | |
8e854e9c | 214 | |
30661b49 N |
215 | static void add_resync_info(struct dlm_lock_resource *lockres, |
216 | sector_t lo, sector_t hi) | |
96ae923a GR |
217 | { |
218 | struct resync_info *ri; | |
219 | ||
220 | ri = (struct resync_info *)lockres->lksb.sb_lvbptr; | |
221 | ri->lo = cpu_to_le64(lo); | |
222 | ri->hi = cpu_to_le64(hi); | |
223 | } | |
224 | ||
225 | static struct suspend_info *read_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres) | |
226 | { | |
227 | struct resync_info ri; | |
228 | struct suspend_info *s = NULL; | |
229 | sector_t hi = 0; | |
230 | ||
231 | dlm_lock_sync(lockres, DLM_LOCK_CR); | |
232 | memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info)); | |
233 | hi = le64_to_cpu(ri.hi); | |
cf97a348 | 234 | if (hi > 0) { |
96ae923a GR |
235 | s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL); |
236 | if (!s) | |
237 | goto out; | |
238 | s->hi = hi; | |
239 | s->lo = le64_to_cpu(ri.lo); | |
240 | } | |
241 | dlm_unlock_sync(lockres); | |
242 | out: | |
243 | return s; | |
244 | } | |
245 | ||
6dc69c9c | 246 | static void recover_bitmaps(struct md_thread *thread) |
e94987db GR |
247 | { |
248 | struct mddev *mddev = thread->mddev; | |
249 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
250 | struct dlm_lock_resource *bm_lockres; | |
251 | char str[64]; | |
252 | int slot, ret; | |
253 | struct suspend_info *s, *tmp; | |
254 | sector_t lo, hi; | |
255 | ||
256 | while (cinfo->recovery_map) { | |
257 | slot = fls64((u64)cinfo->recovery_map) - 1; | |
258 | ||
259 | /* Clear suspend_area associated with the bitmap */ | |
260 | spin_lock_irq(&cinfo->suspend_lock); | |
261 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
262 | if (slot == s->slot) { | |
263 | list_del(&s->list); | |
264 | kfree(s); | |
265 | } | |
266 | spin_unlock_irq(&cinfo->suspend_lock); | |
267 | ||
268 | snprintf(str, 64, "bitmap%04d", slot); | |
269 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
270 | if (!bm_lockres) { | |
271 | pr_err("md-cluster: Cannot initialize bitmaps\n"); | |
272 | goto clear_bit; | |
273 | } | |
274 | ||
275 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
276 | if (ret) { | |
277 | pr_err("md-cluster: Could not DLM lock %s: %d\n", | |
278 | str, ret); | |
279 | goto clear_bit; | |
280 | } | |
97f6cd39 | 281 | ret = bitmap_copy_from_slot(mddev, slot, &lo, &hi, true); |
4b26a08a | 282 | if (ret) { |
e94987db | 283 | pr_err("md-cluster: Could not copy data from bitmap %d\n", slot); |
4b26a08a GR |
284 | goto dlm_unlock; |
285 | } | |
286 | if (hi > 0) { | |
4b26a08a GR |
287 | if (lo < mddev->recovery_cp) |
288 | mddev->recovery_cp = lo; | |
eb315cd0 GJ |
289 | /* wake up thread to continue resync in case resync |
290 | * is not finished */ | |
291 | if (mddev->recovery_cp != MaxSector) { | |
292 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
293 | md_wakeup_thread(mddev->thread); | |
294 | } | |
4b26a08a GR |
295 | } |
296 | dlm_unlock: | |
e94987db GR |
297 | dlm_unlock_sync(bm_lockres); |
298 | clear_bit: | |
4ac7a65f | 299 | lockres_free(bm_lockres); |
e94987db GR |
300 | clear_bit(slot, &cinfo->recovery_map); |
301 | } | |
302 | } | |
303 | ||
cf921cc1 GR |
304 | static void recover_prep(void *arg) |
305 | { | |
90382ed9 GR |
306 | struct mddev *mddev = arg; |
307 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
308 | set_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); | |
cf921cc1 GR |
309 | } |
310 | ||
05cd0e51 | 311 | static void __recover_slot(struct mddev *mddev, int slot) |
cf921cc1 | 312 | { |
cf921cc1 GR |
313 | struct md_cluster_info *cinfo = mddev->cluster_info; |
314 | ||
05cd0e51 | 315 | set_bit(slot, &cinfo->recovery_map); |
e94987db GR |
316 | if (!cinfo->recovery_thread) { |
317 | cinfo->recovery_thread = md_register_thread(recover_bitmaps, | |
318 | mddev, "recover"); | |
319 | if (!cinfo->recovery_thread) { | |
320 | pr_warn("md-cluster: Could not create recovery thread\n"); | |
321 | return; | |
322 | } | |
323 | } | |
324 | md_wakeup_thread(cinfo->recovery_thread); | |
cf921cc1 GR |
325 | } |
326 | ||
05cd0e51 GJ |
327 | static void recover_slot(void *arg, struct dlm_slot *slot) |
328 | { | |
329 | struct mddev *mddev = arg; | |
330 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
331 | ||
332 | pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n", | |
333 | mddev->bitmap_info.cluster_name, | |
334 | slot->nodeid, slot->slot, | |
335 | cinfo->slot_number); | |
336 | /* deduct one since dlm slot starts from one while the num of | |
337 | * cluster-md begins with 0 */ | |
338 | __recover_slot(mddev, slot->slot - 1); | |
339 | } | |
340 | ||
cf921cc1 GR |
341 | static void recover_done(void *arg, struct dlm_slot *slots, |
342 | int num_slots, int our_slot, | |
343 | uint32_t generation) | |
344 | { | |
345 | struct mddev *mddev = arg; | |
346 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
347 | ||
348 | cinfo->slot_number = our_slot; | |
eece075c GJ |
349 | /* completion is only need to be complete when node join cluster, |
350 | * it doesn't need to run during another node's failure */ | |
351 | if (test_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state)) { | |
352 | complete(&cinfo->completion); | |
353 | clear_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state); | |
354 | } | |
90382ed9 | 355 | clear_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); |
cf921cc1 GR |
356 | } |
357 | ||
eece075c GJ |
358 | /* the ops is called when node join the cluster, and do lock recovery |
359 | * if node failure occurs */ | |
cf921cc1 GR |
360 | static const struct dlm_lockspace_ops md_ls_ops = { |
361 | .recover_prep = recover_prep, | |
362 | .recover_slot = recover_slot, | |
363 | .recover_done = recover_done, | |
364 | }; | |
365 | ||
4664680c GR |
366 | /* |
367 | * The BAST function for the ack lock resource | |
368 | * This function wakes up the receive thread in | |
369 | * order to receive and process the message. | |
370 | */ | |
371 | static void ack_bast(void *arg, int mode) | |
372 | { | |
2e2a7cd9 | 373 | struct dlm_lock_resource *res = arg; |
4664680c GR |
374 | struct md_cluster_info *cinfo = res->mddev->cluster_info; |
375 | ||
376 | if (mode == DLM_LOCK_EX) | |
377 | md_wakeup_thread(cinfo->recv_thread); | |
378 | } | |
379 | ||
e59721cc GR |
380 | static void __remove_suspend_info(struct md_cluster_info *cinfo, int slot) |
381 | { | |
382 | struct suspend_info *s, *tmp; | |
383 | ||
384 | list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list) | |
385 | if (slot == s->slot) { | |
e59721cc GR |
386 | list_del(&s->list); |
387 | kfree(s); | |
388 | break; | |
389 | } | |
390 | } | |
391 | ||
b8ca846e | 392 | static void remove_suspend_info(struct mddev *mddev, int slot) |
e59721cc | 393 | { |
b8ca846e | 394 | struct md_cluster_info *cinfo = mddev->cluster_info; |
e59721cc GR |
395 | spin_lock_irq(&cinfo->suspend_lock); |
396 | __remove_suspend_info(cinfo, slot); | |
397 | spin_unlock_irq(&cinfo->suspend_lock); | |
b8ca846e | 398 | mddev->pers->quiesce(mddev, 2); |
e59721cc GR |
399 | } |
400 | ||
401 | ||
9ed38ff5 | 402 | static void process_suspend_info(struct mddev *mddev, |
e59721cc GR |
403 | int slot, sector_t lo, sector_t hi) |
404 | { | |
9ed38ff5 | 405 | struct md_cluster_info *cinfo = mddev->cluster_info; |
e59721cc GR |
406 | struct suspend_info *s; |
407 | ||
408 | if (!hi) { | |
b8ca846e | 409 | remove_suspend_info(mddev, slot); |
c186b128 GR |
410 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
411 | md_wakeup_thread(mddev->thread); | |
e59721cc GR |
412 | return; |
413 | } | |
414 | s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL); | |
415 | if (!s) | |
416 | return; | |
417 | s->slot = slot; | |
418 | s->lo = lo; | |
419 | s->hi = hi; | |
9ed38ff5 GR |
420 | mddev->pers->quiesce(mddev, 1); |
421 | mddev->pers->quiesce(mddev, 0); | |
e59721cc GR |
422 | spin_lock_irq(&cinfo->suspend_lock); |
423 | /* Remove existing entry (if exists) before adding */ | |
424 | __remove_suspend_info(cinfo, slot); | |
425 | list_add(&s->list, &cinfo->suspend_list); | |
426 | spin_unlock_irq(&cinfo->suspend_lock); | |
b8ca846e | 427 | mddev->pers->quiesce(mddev, 2); |
e59721cc GR |
428 | } |
429 | ||
1aee41f6 GR |
430 | static void process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg) |
431 | { | |
432 | char disk_uuid[64]; | |
433 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
434 | char event_name[] = "EVENT=ADD_DEVICE"; | |
435 | char raid_slot[16]; | |
436 | char *envp[] = {event_name, disk_uuid, raid_slot, NULL}; | |
437 | int len; | |
438 | ||
439 | len = snprintf(disk_uuid, 64, "DEVICE_UUID="); | |
b89f704a | 440 | sprintf(disk_uuid + len, "%pU", cmsg->uuid); |
faeff83f | 441 | snprintf(raid_slot, 16, "RAID_DISK=%d", le32_to_cpu(cmsg->raid_slot)); |
1aee41f6 GR |
442 | pr_info("%s:%d Sending kobject change with %s and %s\n", __func__, __LINE__, disk_uuid, raid_slot); |
443 | init_completion(&cinfo->newdisk_completion); | |
fa8259da | 444 | set_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
445 | kobject_uevent_env(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE, envp); |
446 | wait_for_completion_timeout(&cinfo->newdisk_completion, | |
447 | NEW_DEV_TIMEOUT); | |
fa8259da | 448 | clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
449 | } |
450 | ||
451 | ||
452 | static void process_metadata_update(struct mddev *mddev, struct cluster_msg *msg) | |
453 | { | |
454 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
15858fa5 GJ |
455 | mddev->good_device_nr = le32_to_cpu(msg->raid_slot); |
456 | set_bit(MD_RELOAD_SB, &mddev->flags); | |
1aee41f6 | 457 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); |
15858fa5 | 458 | md_wakeup_thread(mddev->thread); |
1aee41f6 GR |
459 | } |
460 | ||
88bcfef7 GR |
461 | static void process_remove_disk(struct mddev *mddev, struct cluster_msg *msg) |
462 | { | |
faeff83f GJ |
463 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, |
464 | le32_to_cpu(msg->raid_slot)); | |
88bcfef7 | 465 | |
659b254f GJ |
466 | if (rdev) { |
467 | set_bit(ClusterRemove, &rdev->flags); | |
468 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
469 | md_wakeup_thread(mddev->thread); | |
470 | } | |
88bcfef7 | 471 | else |
faeff83f GJ |
472 | pr_warn("%s: %d Could not find disk(%d) to REMOVE\n", |
473 | __func__, __LINE__, le32_to_cpu(msg->raid_slot)); | |
88bcfef7 GR |
474 | } |
475 | ||
97f6cd39 GR |
476 | static void process_readd_disk(struct mddev *mddev, struct cluster_msg *msg) |
477 | { | |
faeff83f GJ |
478 | struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, |
479 | le32_to_cpu(msg->raid_slot)); | |
97f6cd39 GR |
480 | |
481 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
482 | clear_bit(Faulty, &rdev->flags); | |
483 | else | |
faeff83f GJ |
484 | pr_warn("%s: %d Could not find disk(%d) which is faulty", |
485 | __func__, __LINE__, le32_to_cpu(msg->raid_slot)); | |
97f6cd39 GR |
486 | } |
487 | ||
4664680c GR |
488 | static void process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg) |
489 | { | |
256f5b24 GJ |
490 | if (WARN(mddev->cluster_info->slot_number - 1 == le32_to_cpu(msg->slot), |
491 | "node %d received it's own msg\n", le32_to_cpu(msg->slot))) | |
492 | return; | |
cf97a348 | 493 | switch (le32_to_cpu(msg->type)) { |
4664680c | 494 | case METADATA_UPDATED: |
1aee41f6 | 495 | process_metadata_update(mddev, msg); |
4664680c GR |
496 | break; |
497 | case RESYNCING: | |
cf97a348 GJ |
498 | process_suspend_info(mddev, le32_to_cpu(msg->slot), |
499 | le64_to_cpu(msg->low), | |
500 | le64_to_cpu(msg->high)); | |
4664680c | 501 | break; |
1aee41f6 | 502 | case NEWDISK: |
1aee41f6 | 503 | process_add_new_disk(mddev, msg); |
88bcfef7 GR |
504 | break; |
505 | case REMOVE: | |
88bcfef7 GR |
506 | process_remove_disk(mddev, msg); |
507 | break; | |
97f6cd39 | 508 | case RE_ADD: |
97f6cd39 GR |
509 | process_readd_disk(mddev, msg); |
510 | break; | |
dc737d7c | 511 | case BITMAP_NEEDS_SYNC: |
cf97a348 | 512 | __recover_slot(mddev, le32_to_cpu(msg->slot)); |
dc737d7c | 513 | break; |
88bcfef7 GR |
514 | default: |
515 | pr_warn("%s:%d Received unknown message from %d\n", | |
516 | __func__, __LINE__, msg->slot); | |
09dd1af2 | 517 | } |
4664680c GR |
518 | } |
519 | ||
520 | /* | |
521 | * thread for receiving message | |
522 | */ | |
523 | static void recv_daemon(struct md_thread *thread) | |
524 | { | |
525 | struct md_cluster_info *cinfo = thread->mddev->cluster_info; | |
526 | struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres; | |
527 | struct dlm_lock_resource *message_lockres = cinfo->message_lockres; | |
528 | struct cluster_msg msg; | |
b5ef5678 | 529 | int ret; |
4664680c | 530 | |
8b9277c8 | 531 | mutex_lock(&cinfo->recv_mutex); |
4664680c GR |
532 | /*get CR on Message*/ |
533 | if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) { | |
534 | pr_err("md/raid1:failed to get CR on MESSAGE\n"); | |
8b9277c8 | 535 | mutex_unlock(&cinfo->recv_mutex); |
4664680c GR |
536 | return; |
537 | } | |
538 | ||
539 | /* read lvb and wake up thread to process this message_lockres */ | |
540 | memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg)); | |
541 | process_recvd_msg(thread->mddev, &msg); | |
542 | ||
543 | /*release CR on ack_lockres*/ | |
b5ef5678 GJ |
544 | ret = dlm_unlock_sync(ack_lockres); |
545 | if (unlikely(ret != 0)) | |
546 | pr_info("unlock ack failed return %d\n", ret); | |
66099bb0 | 547 | /*up-convert to PR on message_lockres*/ |
b5ef5678 GJ |
548 | ret = dlm_lock_sync(message_lockres, DLM_LOCK_PR); |
549 | if (unlikely(ret != 0)) | |
550 | pr_info("lock PR on msg failed return %d\n", ret); | |
4664680c | 551 | /*get CR on ack_lockres again*/ |
b5ef5678 GJ |
552 | ret = dlm_lock_sync(ack_lockres, DLM_LOCK_CR); |
553 | if (unlikely(ret != 0)) | |
554 | pr_info("lock CR on ack failed return %d\n", ret); | |
4664680c | 555 | /*release CR on message_lockres*/ |
b5ef5678 GJ |
556 | ret = dlm_unlock_sync(message_lockres); |
557 | if (unlikely(ret != 0)) | |
558 | pr_info("unlock msg failed return %d\n", ret); | |
8b9277c8 | 559 | mutex_unlock(&cinfo->recv_mutex); |
4664680c GR |
560 | } |
561 | ||
8b9277c8 | 562 | /* lock_token() |
601b515c GR |
563 | * Takes the lock on the TOKEN lock resource so no other |
564 | * node can communicate while the operation is underway. | |
565 | */ | |
8b9277c8 | 566 | static int lock_token(struct md_cluster_info *cinfo) |
601b515c GR |
567 | { |
568 | int error; | |
569 | ||
570 | error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX); | |
571 | if (error) | |
572 | pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n", | |
573 | __func__, __LINE__, error); | |
8b9277c8 GJ |
574 | |
575 | /* Lock the receive sequence */ | |
576 | mutex_lock(&cinfo->recv_mutex); | |
601b515c GR |
577 | return error; |
578 | } | |
579 | ||
8b9277c8 GJ |
580 | /* lock_comm() |
581 | * Sets the MD_CLUSTER_SEND_LOCK bit to lock the send channel. | |
582 | */ | |
583 | static int lock_comm(struct md_cluster_info *cinfo) | |
584 | { | |
585 | wait_event(cinfo->wait, | |
586 | !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state)); | |
587 | ||
588 | return lock_token(cinfo); | |
589 | } | |
590 | ||
601b515c GR |
591 | static void unlock_comm(struct md_cluster_info *cinfo) |
592 | { | |
dbb64f86 | 593 | WARN_ON(cinfo->token_lockres->mode != DLM_LOCK_EX); |
8b9277c8 | 594 | mutex_unlock(&cinfo->recv_mutex); |
601b515c | 595 | dlm_unlock_sync(cinfo->token_lockres); |
8b9277c8 GJ |
596 | clear_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state); |
597 | wake_up(&cinfo->wait); | |
601b515c GR |
598 | } |
599 | ||
600 | /* __sendmsg() | |
601 | * This function performs the actual sending of the message. This function is | |
602 | * usually called after performing the encompassing operation | |
603 | * The function: | |
604 | * 1. Grabs the message lockresource in EX mode | |
605 | * 2. Copies the message to the message LVB | |
66099bb0 | 606 | * 3. Downconverts message lockresource to CW |
601b515c GR |
607 | * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes |
608 | * and the other nodes read the message. The thread will wait here until all other | |
609 | * nodes have released ack lock resource. | |
610 | * 5. Downconvert ack lockresource to CR | |
611 | */ | |
612 | static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
613 | { | |
614 | int error; | |
615 | int slot = cinfo->slot_number - 1; | |
616 | ||
617 | cmsg->slot = cpu_to_le32(slot); | |
618 | /*get EX on Message*/ | |
619 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX); | |
620 | if (error) { | |
621 | pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error); | |
622 | goto failed_message; | |
623 | } | |
624 | ||
625 | memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg, | |
626 | sizeof(struct cluster_msg)); | |
66099bb0 GJ |
627 | /*down-convert EX to CW on Message*/ |
628 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CW); | |
601b515c | 629 | if (error) { |
66099bb0 | 630 | pr_err("md-cluster: failed to convert EX to CW on MESSAGE(%d)\n", |
601b515c | 631 | error); |
66099bb0 | 632 | goto failed_ack; |
601b515c GR |
633 | } |
634 | ||
635 | /*up-convert CR to EX on Ack*/ | |
636 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX); | |
637 | if (error) { | |
638 | pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n", | |
639 | error); | |
640 | goto failed_ack; | |
641 | } | |
642 | ||
643 | /*down-convert EX to CR on Ack*/ | |
644 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR); | |
645 | if (error) { | |
646 | pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n", | |
647 | error); | |
648 | goto failed_ack; | |
649 | } | |
650 | ||
651 | failed_ack: | |
b5ef5678 GJ |
652 | error = dlm_unlock_sync(cinfo->message_lockres); |
653 | if (unlikely(error != 0)) { | |
654 | pr_err("md-cluster: failed convert to NL on MESSAGE(%d)\n", | |
655 | error); | |
656 | /* in case the message can't be released due to some reason */ | |
657 | goto failed_ack; | |
658 | } | |
601b515c GR |
659 | failed_message: |
660 | return error; | |
661 | } | |
662 | ||
663 | static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
664 | { | |
665 | int ret; | |
666 | ||
667 | lock_comm(cinfo); | |
668 | ret = __sendmsg(cinfo, cmsg); | |
669 | unlock_comm(cinfo); | |
670 | return ret; | |
671 | } | |
672 | ||
96ae923a GR |
673 | static int gather_all_resync_info(struct mddev *mddev, int total_slots) |
674 | { | |
675 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
676 | int i, ret = 0; | |
677 | struct dlm_lock_resource *bm_lockres; | |
678 | struct suspend_info *s; | |
679 | char str[64]; | |
abb9b22a | 680 | sector_t lo, hi; |
96ae923a GR |
681 | |
682 | ||
683 | for (i = 0; i < total_slots; i++) { | |
684 | memset(str, '\0', 64); | |
685 | snprintf(str, 64, "bitmap%04d", i); | |
686 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
687 | if (!bm_lockres) | |
688 | return -ENOMEM; | |
4ac7a65f SL |
689 | if (i == (cinfo->slot_number - 1)) { |
690 | lockres_free(bm_lockres); | |
96ae923a | 691 | continue; |
4ac7a65f | 692 | } |
96ae923a GR |
693 | |
694 | bm_lockres->flags |= DLM_LKF_NOQUEUE; | |
695 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
696 | if (ret == -EAGAIN) { | |
697 | memset(bm_lockres->lksb.sb_lvbptr, '\0', LVB_SIZE); | |
698 | s = read_resync_info(mddev, bm_lockres); | |
699 | if (s) { | |
700 | pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n", | |
701 | __func__, __LINE__, | |
702 | (unsigned long long) s->lo, | |
703 | (unsigned long long) s->hi, i); | |
704 | spin_lock_irq(&cinfo->suspend_lock); | |
705 | s->slot = i; | |
706 | list_add(&s->list, &cinfo->suspend_list); | |
707 | spin_unlock_irq(&cinfo->suspend_lock); | |
708 | } | |
709 | ret = 0; | |
710 | lockres_free(bm_lockres); | |
711 | continue; | |
712 | } | |
6e6d9f2c GJ |
713 | if (ret) { |
714 | lockres_free(bm_lockres); | |
96ae923a | 715 | goto out; |
6e6d9f2c | 716 | } |
abb9b22a GJ |
717 | |
718 | /* Read the disk bitmap sb and check if it needs recovery */ | |
719 | ret = bitmap_copy_from_slot(mddev, i, &lo, &hi, false); | |
720 | if (ret) { | |
721 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", i); | |
722 | lockres_free(bm_lockres); | |
723 | continue; | |
724 | } | |
725 | if ((hi > 0) && (lo < mddev->recovery_cp)) { | |
726 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
727 | mddev->recovery_cp = lo; | |
728 | md_check_recovery(mddev); | |
729 | } | |
730 | ||
96ae923a GR |
731 | dlm_unlock_sync(bm_lockres); |
732 | lockres_free(bm_lockres); | |
733 | } | |
734 | out: | |
735 | return ret; | |
736 | } | |
737 | ||
edb39c9d GR |
738 | static int join(struct mddev *mddev, int nodes) |
739 | { | |
c4ce867f | 740 | struct md_cluster_info *cinfo; |
cf921cc1 | 741 | int ret, ops_rv; |
c4ce867f GR |
742 | char str[64]; |
743 | ||
c4ce867f GR |
744 | cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL); |
745 | if (!cinfo) | |
746 | return -ENOMEM; | |
747 | ||
9e3072e3 GJ |
748 | INIT_LIST_HEAD(&cinfo->suspend_list); |
749 | spin_lock_init(&cinfo->suspend_lock); | |
cf921cc1 | 750 | init_completion(&cinfo->completion); |
eece075c | 751 | set_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state); |
8b9277c8 GJ |
752 | init_waitqueue_head(&cinfo->wait); |
753 | mutex_init(&cinfo->recv_mutex); | |
cf921cc1 | 754 | |
cf921cc1 GR |
755 | mddev->cluster_info = cinfo; |
756 | ||
c4ce867f | 757 | memset(str, 0, 64); |
b89f704a | 758 | sprintf(str, "%pU", mddev->uuid); |
cf921cc1 GR |
759 | ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name, |
760 | DLM_LSFL_FS, LVB_SIZE, | |
761 | &md_ls_ops, mddev, &ops_rv, &cinfo->lockspace); | |
c4ce867f GR |
762 | if (ret) |
763 | goto err; | |
cf921cc1 | 764 | wait_for_completion(&cinfo->completion); |
8c58f02e GJ |
765 | if (nodes < cinfo->slot_number) { |
766 | pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).", | |
767 | cinfo->slot_number, nodes); | |
b97e9257 GR |
768 | ret = -ERANGE; |
769 | goto err; | |
770 | } | |
4664680c GR |
771 | /* Initiate the communication resources */ |
772 | ret = -ENOMEM; | |
773 | cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv"); | |
774 | if (!cinfo->recv_thread) { | |
775 | pr_err("md-cluster: cannot allocate memory for recv_thread!\n"); | |
776 | goto err; | |
777 | } | |
778 | cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1); | |
779 | if (!cinfo->message_lockres) | |
780 | goto err; | |
781 | cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0); | |
782 | if (!cinfo->token_lockres) | |
783 | goto err; | |
1aee41f6 GR |
784 | cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0); |
785 | if (!cinfo->no_new_dev_lockres) | |
786 | goto err; | |
787 | ||
1535212c GJ |
788 | ret = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX); |
789 | if (ret) { | |
790 | ret = -EAGAIN; | |
791 | pr_err("md-cluster: can't join cluster to avoid lock issue\n"); | |
792 | goto err; | |
793 | } | |
794 | cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0); | |
795 | if (!cinfo->ack_lockres) | |
796 | goto err; | |
4664680c GR |
797 | /* get sync CR lock on ACK. */ |
798 | if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR)) | |
799 | pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n", | |
800 | ret); | |
1535212c | 801 | dlm_unlock_sync(cinfo->token_lockres); |
1aee41f6 GR |
802 | /* get sync CR lock on no-new-dev. */ |
803 | if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR)) | |
804 | pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret); | |
805 | ||
54519c5f GR |
806 | |
807 | pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number); | |
808 | snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1); | |
809 | cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1); | |
810 | if (!cinfo->bitmap_lockres) | |
811 | goto err; | |
812 | if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) { | |
813 | pr_err("Failed to get bitmap lock\n"); | |
814 | ret = -EINVAL; | |
815 | goto err; | |
816 | } | |
817 | ||
c186b128 GR |
818 | cinfo->resync_lockres = lockres_init(mddev, "resync", NULL, 0); |
819 | if (!cinfo->resync_lockres) | |
820 | goto err; | |
821 | ||
96ae923a GR |
822 | ret = gather_all_resync_info(mddev, nodes); |
823 | if (ret) | |
824 | goto err; | |
825 | ||
edb39c9d | 826 | return 0; |
c4ce867f | 827 | err: |
5b0fb33e GJ |
828 | md_unregister_thread(&cinfo->recovery_thread); |
829 | md_unregister_thread(&cinfo->recv_thread); | |
4664680c GR |
830 | lockres_free(cinfo->message_lockres); |
831 | lockres_free(cinfo->token_lockres); | |
832 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 833 | lockres_free(cinfo->no_new_dev_lockres); |
c186b128 | 834 | lockres_free(cinfo->resync_lockres); |
96ae923a | 835 | lockres_free(cinfo->bitmap_lockres); |
c4ce867f GR |
836 | if (cinfo->lockspace) |
837 | dlm_release_lockspace(cinfo->lockspace, 2); | |
cf921cc1 | 838 | mddev->cluster_info = NULL; |
c4ce867f | 839 | kfree(cinfo); |
c4ce867f | 840 | return ret; |
edb39c9d GR |
841 | } |
842 | ||
09995411 GJ |
843 | static void resync_bitmap(struct mddev *mddev) |
844 | { | |
845 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
846 | struct cluster_msg cmsg = {0}; | |
847 | int err; | |
848 | ||
849 | cmsg.type = cpu_to_le32(BITMAP_NEEDS_SYNC); | |
850 | err = sendmsg(cinfo, &cmsg); | |
851 | if (err) | |
852 | pr_err("%s:%d: failed to send BITMAP_NEEDS_SYNC message (%d)\n", | |
853 | __func__, __LINE__, err); | |
854 | } | |
855 | ||
f6a2dc64 | 856 | static void unlock_all_bitmaps(struct mddev *mddev); |
edb39c9d GR |
857 | static int leave(struct mddev *mddev) |
858 | { | |
c4ce867f GR |
859 | struct md_cluster_info *cinfo = mddev->cluster_info; |
860 | ||
861 | if (!cinfo) | |
862 | return 0; | |
09995411 GJ |
863 | |
864 | /* BITMAP_NEEDS_SYNC message should be sent when node | |
865 | * is leaving the cluster with dirty bitmap, also we | |
866 | * can only deliver it when dlm connection is available */ | |
867 | if (cinfo->slot_number > 0 && mddev->recovery_cp != MaxSector) | |
868 | resync_bitmap(mddev); | |
869 | ||
e94987db | 870 | md_unregister_thread(&cinfo->recovery_thread); |
4664680c GR |
871 | md_unregister_thread(&cinfo->recv_thread); |
872 | lockres_free(cinfo->message_lockres); | |
873 | lockres_free(cinfo->token_lockres); | |
874 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 875 | lockres_free(cinfo->no_new_dev_lockres); |
4ac7a65f | 876 | lockres_free(cinfo->resync_lockres); |
54519c5f | 877 | lockres_free(cinfo->bitmap_lockres); |
f6a2dc64 | 878 | unlock_all_bitmaps(mddev); |
c4ce867f | 879 | dlm_release_lockspace(cinfo->lockspace, 2); |
edb39c9d GR |
880 | return 0; |
881 | } | |
882 | ||
cf921cc1 GR |
883 | /* slot_number(): Returns the MD slot number to use |
884 | * DLM starts the slot numbers from 1, wheras cluster-md | |
885 | * wants the number to be from zero, so we deduct one | |
886 | */ | |
887 | static int slot_number(struct mddev *mddev) | |
888 | { | |
889 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
890 | ||
891 | return cinfo->slot_number - 1; | |
892 | } | |
893 | ||
8b9277c8 GJ |
894 | /* |
895 | * Check if the communication is already locked, else lock the communication | |
896 | * channel. | |
897 | * If it is already locked, token is in EX mode, and hence lock_token() | |
898 | * should not be called. | |
899 | */ | |
293467aa GR |
900 | static int metadata_update_start(struct mddev *mddev) |
901 | { | |
8b9277c8 GJ |
902 | struct md_cluster_info *cinfo = mddev->cluster_info; |
903 | ||
904 | wait_event(cinfo->wait, | |
905 | !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state) || | |
906 | test_and_clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state)); | |
907 | ||
908 | /* If token is already locked, return 0 */ | |
909 | if (cinfo->token_lockres->mode == DLM_LOCK_EX) | |
910 | return 0; | |
911 | ||
912 | return lock_token(cinfo); | |
293467aa GR |
913 | } |
914 | ||
915 | static int metadata_update_finish(struct mddev *mddev) | |
916 | { | |
917 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
918 | struct cluster_msg cmsg; | |
70bcecdb GR |
919 | struct md_rdev *rdev; |
920 | int ret = 0; | |
ba2746b0 | 921 | int raid_slot = -1; |
293467aa GR |
922 | |
923 | memset(&cmsg, 0, sizeof(cmsg)); | |
924 | cmsg.type = cpu_to_le32(METADATA_UPDATED); | |
70bcecdb GR |
925 | /* Pick up a good active device number to send. |
926 | */ | |
927 | rdev_for_each(rdev, mddev) | |
928 | if (rdev->raid_disk > -1 && !test_bit(Faulty, &rdev->flags)) { | |
ba2746b0 | 929 | raid_slot = rdev->desc_nr; |
70bcecdb GR |
930 | break; |
931 | } | |
ba2746b0 N |
932 | if (raid_slot >= 0) { |
933 | cmsg.raid_slot = cpu_to_le32(raid_slot); | |
70bcecdb | 934 | ret = __sendmsg(cinfo, &cmsg); |
ba2746b0 | 935 | } else |
70bcecdb | 936 | pr_warn("md-cluster: No good device id found to send\n"); |
8b9277c8 | 937 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
293467aa GR |
938 | unlock_comm(cinfo); |
939 | return ret; | |
940 | } | |
941 | ||
dbb64f86 | 942 | static void metadata_update_cancel(struct mddev *mddev) |
293467aa GR |
943 | { |
944 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
8b9277c8 | 945 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
dbb64f86 | 946 | unlock_comm(cinfo); |
293467aa GR |
947 | } |
948 | ||
c186b128 GR |
949 | static int resync_start(struct mddev *mddev) |
950 | { | |
951 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
c186b128 GR |
952 | return dlm_lock_sync(cinfo->resync_lockres, DLM_LOCK_EX); |
953 | } | |
954 | ||
c40f341f | 955 | static int resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi) |
965400eb GR |
956 | { |
957 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
ac277c6a | 958 | struct resync_info ri; |
aee177ac | 959 | struct cluster_msg cmsg = {0}; |
965400eb | 960 | |
ac277c6a GR |
961 | /* do not send zero again, if we have sent before */ |
962 | if (hi == 0) { | |
963 | memcpy(&ri, cinfo->bitmap_lockres->lksb.sb_lvbptr, sizeof(struct resync_info)); | |
964 | if (le64_to_cpu(ri.hi) == 0) | |
965 | return 0; | |
966 | } | |
967 | ||
30661b49 | 968 | add_resync_info(cinfo->bitmap_lockres, lo, hi); |
c40f341f GR |
969 | /* Re-acquire the lock to refresh LVB */ |
970 | dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW); | |
c40f341f | 971 | cmsg.type = cpu_to_le32(RESYNCING); |
965400eb GR |
972 | cmsg.low = cpu_to_le64(lo); |
973 | cmsg.high = cpu_to_le64(hi); | |
c186b128 | 974 | |
965400eb GR |
975 | return sendmsg(cinfo, &cmsg); |
976 | } | |
977 | ||
c186b128 GR |
978 | static int resync_finish(struct mddev *mddev) |
979 | { | |
980 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
c186b128 GR |
981 | dlm_unlock_sync(cinfo->resync_lockres); |
982 | return resync_info_update(mddev, 0, 0); | |
983 | } | |
984 | ||
90382ed9 GR |
985 | static int area_resyncing(struct mddev *mddev, int direction, |
986 | sector_t lo, sector_t hi) | |
589a1c49 GR |
987 | { |
988 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
989 | int ret = 0; | |
990 | struct suspend_info *s; | |
991 | ||
90382ed9 GR |
992 | if ((direction == READ) && |
993 | test_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state)) | |
994 | return 1; | |
995 | ||
589a1c49 GR |
996 | spin_lock_irq(&cinfo->suspend_lock); |
997 | if (list_empty(&cinfo->suspend_list)) | |
998 | goto out; | |
999 | list_for_each_entry(s, &cinfo->suspend_list, list) | |
1000 | if (hi > s->lo && lo < s->hi) { | |
1001 | ret = 1; | |
1002 | break; | |
1003 | } | |
1004 | out: | |
1005 | spin_unlock_irq(&cinfo->suspend_lock); | |
1006 | return ret; | |
1007 | } | |
1008 | ||
dbb64f86 GR |
1009 | /* add_new_disk() - initiates a disk add |
1010 | * However, if this fails before writing md_update_sb(), | |
1011 | * add_new_disk_cancel() must be called to release token lock | |
1012 | */ | |
1013 | static int add_new_disk(struct mddev *mddev, struct md_rdev *rdev) | |
1aee41f6 GR |
1014 | { |
1015 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1016 | struct cluster_msg cmsg; | |
1017 | int ret = 0; | |
1018 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); | |
1019 | char *uuid = sb->device_uuid; | |
1020 | ||
1021 | memset(&cmsg, 0, sizeof(cmsg)); | |
1022 | cmsg.type = cpu_to_le32(NEWDISK); | |
1023 | memcpy(cmsg.uuid, uuid, 16); | |
faeff83f | 1024 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); |
1aee41f6 GR |
1025 | lock_comm(cinfo); |
1026 | ret = __sendmsg(cinfo, &cmsg); | |
1027 | if (ret) | |
1028 | return ret; | |
1029 | cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE; | |
1030 | ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX); | |
1031 | cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE; | |
1032 | /* Some node does not "see" the device */ | |
1033 | if (ret == -EAGAIN) | |
1034 | ret = -ENOENT; | |
dbb64f86 GR |
1035 | if (ret) |
1036 | unlock_comm(cinfo); | |
8b9277c8 | 1037 | else { |
1aee41f6 | 1038 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); |
e19508fa GJ |
1039 | /* Since MD_CHANGE_DEVS will be set in add_bound_rdev which |
1040 | * will run soon after add_new_disk, the below path will be | |
1041 | * invoked: | |
1042 | * md_wakeup_thread(mddev->thread) | |
1043 | * -> conf->thread (raid1d) | |
1044 | * -> md_check_recovery -> md_update_sb | |
1045 | * -> metadata_update_start/finish | |
1046 | * MD_CLUSTER_SEND_LOCKED_ALREADY will be cleared eventually. | |
1047 | * | |
1048 | * For other failure cases, metadata_update_cancel and | |
1049 | * add_new_disk_cancel also clear below bit as well. | |
1050 | * */ | |
8b9277c8 GJ |
1051 | set_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
1052 | wake_up(&cinfo->wait); | |
1053 | } | |
1aee41f6 GR |
1054 | return ret; |
1055 | } | |
1056 | ||
dbb64f86 | 1057 | static void add_new_disk_cancel(struct mddev *mddev) |
1aee41f6 | 1058 | { |
dbb64f86 | 1059 | struct md_cluster_info *cinfo = mddev->cluster_info; |
8b9277c8 | 1060 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
dbb64f86 | 1061 | unlock_comm(cinfo); |
1aee41f6 GR |
1062 | } |
1063 | ||
fa8259da | 1064 | static int new_disk_ack(struct mddev *mddev, bool ack) |
1aee41f6 GR |
1065 | { |
1066 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1067 | ||
fa8259da GR |
1068 | if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) { |
1069 | pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev)); | |
1070 | return -EINVAL; | |
1071 | } | |
1072 | ||
1aee41f6 GR |
1073 | if (ack) |
1074 | dlm_unlock_sync(cinfo->no_new_dev_lockres); | |
1075 | complete(&cinfo->newdisk_completion); | |
fa8259da | 1076 | return 0; |
1aee41f6 GR |
1077 | } |
1078 | ||
88bcfef7 GR |
1079 | static int remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1080 | { | |
aee177ac | 1081 | struct cluster_msg cmsg = {0}; |
88bcfef7 | 1082 | struct md_cluster_info *cinfo = mddev->cluster_info; |
faeff83f GJ |
1083 | cmsg.type = cpu_to_le32(REMOVE); |
1084 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); | |
54a88392 | 1085 | return sendmsg(cinfo, &cmsg); |
88bcfef7 GR |
1086 | } |
1087 | ||
f6a2dc64 GJ |
1088 | static int lock_all_bitmaps(struct mddev *mddev) |
1089 | { | |
1090 | int slot, my_slot, ret, held = 1, i = 0; | |
1091 | char str[64]; | |
1092 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1093 | ||
1094 | cinfo->other_bitmap_lockres = kzalloc((mddev->bitmap_info.nodes - 1) * | |
1095 | sizeof(struct dlm_lock_resource *), | |
1096 | GFP_KERNEL); | |
1097 | if (!cinfo->other_bitmap_lockres) { | |
1098 | pr_err("md: can't alloc mem for other bitmap locks\n"); | |
1099 | return 0; | |
1100 | } | |
1101 | ||
1102 | my_slot = slot_number(mddev); | |
1103 | for (slot = 0; slot < mddev->bitmap_info.nodes; slot++) { | |
1104 | if (slot == my_slot) | |
1105 | continue; | |
1106 | ||
1107 | memset(str, '\0', 64); | |
1108 | snprintf(str, 64, "bitmap%04d", slot); | |
1109 | cinfo->other_bitmap_lockres[i] = lockres_init(mddev, str, NULL, 1); | |
1110 | if (!cinfo->other_bitmap_lockres[i]) | |
1111 | return -ENOMEM; | |
1112 | ||
1113 | cinfo->other_bitmap_lockres[i]->flags |= DLM_LKF_NOQUEUE; | |
1114 | ret = dlm_lock_sync(cinfo->other_bitmap_lockres[i], DLM_LOCK_PW); | |
1115 | if (ret) | |
1116 | held = -1; | |
1117 | i++; | |
1118 | } | |
1119 | ||
1120 | return held; | |
1121 | } | |
1122 | ||
1123 | static void unlock_all_bitmaps(struct mddev *mddev) | |
1124 | { | |
1125 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1126 | int i; | |
1127 | ||
1128 | /* release other node's bitmap lock if they are existed */ | |
1129 | if (cinfo->other_bitmap_lockres) { | |
1130 | for (i = 0; i < mddev->bitmap_info.nodes - 1; i++) { | |
1131 | if (cinfo->other_bitmap_lockres[i]) { | |
1132 | dlm_unlock_sync(cinfo->other_bitmap_lockres[i]); | |
1133 | lockres_free(cinfo->other_bitmap_lockres[i]); | |
1134 | } | |
1135 | } | |
1136 | kfree(cinfo->other_bitmap_lockres); | |
1137 | } | |
1138 | } | |
1139 | ||
97f6cd39 GR |
1140 | static int gather_bitmaps(struct md_rdev *rdev) |
1141 | { | |
1142 | int sn, err; | |
1143 | sector_t lo, hi; | |
aee177ac | 1144 | struct cluster_msg cmsg = {0}; |
97f6cd39 GR |
1145 | struct mddev *mddev = rdev->mddev; |
1146 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1147 | ||
faeff83f GJ |
1148 | cmsg.type = cpu_to_le32(RE_ADD); |
1149 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); | |
97f6cd39 GR |
1150 | err = sendmsg(cinfo, &cmsg); |
1151 | if (err) | |
1152 | goto out; | |
1153 | ||
1154 | for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) { | |
1155 | if (sn == (cinfo->slot_number - 1)) | |
1156 | continue; | |
1157 | err = bitmap_copy_from_slot(mddev, sn, &lo, &hi, false); | |
1158 | if (err) { | |
1159 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn); | |
1160 | goto out; | |
1161 | } | |
1162 | if ((hi > 0) && (lo < mddev->recovery_cp)) | |
1163 | mddev->recovery_cp = lo; | |
1164 | } | |
1165 | out: | |
1166 | return err; | |
1167 | } | |
1168 | ||
edb39c9d GR |
1169 | static struct md_cluster_operations cluster_ops = { |
1170 | .join = join, | |
1171 | .leave = leave, | |
cf921cc1 | 1172 | .slot_number = slot_number, |
c186b128 GR |
1173 | .resync_start = resync_start, |
1174 | .resync_finish = resync_finish, | |
96ae923a | 1175 | .resync_info_update = resync_info_update, |
293467aa GR |
1176 | .metadata_update_start = metadata_update_start, |
1177 | .metadata_update_finish = metadata_update_finish, | |
1178 | .metadata_update_cancel = metadata_update_cancel, | |
589a1c49 | 1179 | .area_resyncing = area_resyncing, |
dbb64f86 GR |
1180 | .add_new_disk = add_new_disk, |
1181 | .add_new_disk_cancel = add_new_disk_cancel, | |
1aee41f6 | 1182 | .new_disk_ack = new_disk_ack, |
88bcfef7 | 1183 | .remove_disk = remove_disk, |
97f6cd39 | 1184 | .gather_bitmaps = gather_bitmaps, |
f6a2dc64 GJ |
1185 | .lock_all_bitmaps = lock_all_bitmaps, |
1186 | .unlock_all_bitmaps = unlock_all_bitmaps, | |
edb39c9d GR |
1187 | }; |
1188 | ||
8e854e9c GR |
1189 | static int __init cluster_init(void) |
1190 | { | |
1191 | pr_warn("md-cluster: EXPERIMENTAL. Use with caution\n"); | |
1192 | pr_info("Registering Cluster MD functions\n"); | |
edb39c9d | 1193 | register_md_cluster_operations(&cluster_ops, THIS_MODULE); |
8e854e9c GR |
1194 | return 0; |
1195 | } | |
1196 | ||
1197 | static void cluster_exit(void) | |
1198 | { | |
edb39c9d | 1199 | unregister_md_cluster_operations(); |
8e854e9c GR |
1200 | } |
1201 | ||
1202 | module_init(cluster_init); | |
1203 | module_exit(cluster_exit); | |
86b57277 | 1204 | MODULE_AUTHOR("SUSE"); |
8e854e9c GR |
1205 | MODULE_LICENSE("GPL"); |
1206 | MODULE_DESCRIPTION("Clustering support for MD"); |