Commit | Line | Data |
---|---|---|
8d7c56d0 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
8e854e9c GR |
2 | /* |
3 | * Copyright (C) 2015, SUSE | |
8e854e9c GR |
4 | */ |
5 | ||
6 | ||
7 | #include <linux/module.h> | |
7bcda714 | 8 | #include <linux/kthread.h> |
47741b7c GR |
9 | #include <linux/dlm.h> |
10 | #include <linux/sched.h> | |
1aee41f6 | 11 | #include <linux/raid/md_p.h> |
47741b7c | 12 | #include "md.h" |
935fe098 | 13 | #include "md-bitmap.h" |
edb39c9d | 14 | #include "md-cluster.h" |
47741b7c GR |
15 | |
16 | #define LVB_SIZE 64 | |
1aee41f6 | 17 | #define NEW_DEV_TIMEOUT 5000 |
47741b7c GR |
18 | |
19 | struct dlm_lock_resource { | |
20 | dlm_lockspace_t *ls; | |
21 | struct dlm_lksb lksb; | |
22 | char *name; /* lock name. */ | |
23 | uint32_t flags; /* flags to pass to dlm_lock() */ | |
fccb60a4 GJ |
24 | wait_queue_head_t sync_locking; /* wait queue for synchronized locking */ |
25 | bool sync_locking_done; | |
c4ce867f GR |
26 | void (*bast)(void *arg, int mode); /* blocking AST function pointer*/ |
27 | struct mddev *mddev; /* pointing back to mddev. */ | |
dbb64f86 | 28 | int mode; |
c4ce867f GR |
29 | }; |
30 | ||
96ae923a GR |
31 | struct resync_info { |
32 | __le64 lo; | |
33 | __le64 hi; | |
34 | }; | |
35 | ||
fa8259da GR |
36 | /* md_cluster_info flags */ |
37 | #define MD_CLUSTER_WAITING_FOR_NEWDISK 1 | |
90382ed9 | 38 | #define MD_CLUSTER_SUSPEND_READ_BALANCING 2 |
eece075c | 39 | #define MD_CLUSTER_BEGIN_JOIN_CLUSTER 3 |
fa8259da | 40 | |
8b9277c8 GJ |
41 | /* Lock the send communication. This is done through |
42 | * bit manipulation as opposed to a mutex in order to | |
43 | * accomodate lock and hold. See next comment. | |
44 | */ | |
45 | #define MD_CLUSTER_SEND_LOCK 4 | |
e19508fa GJ |
46 | /* If cluster operations (such as adding a disk) must lock the |
47 | * communication channel, so as to perform extra operations | |
48 | * (update metadata) and no other operation is allowed on the | |
49 | * MD. Token needs to be locked and held until the operation | |
50 | * completes witha md_update_sb(), which would eventually release | |
51 | * the lock. | |
8b9277c8 GJ |
52 | */ |
53 | #define MD_CLUSTER_SEND_LOCKED_ALREADY 5 | |
51e453ae GJ |
54 | /* We should receive message after node joined cluster and |
55 | * set up all the related infos such as bitmap and personality */ | |
56 | #define MD_CLUSTER_ALREADY_IN_CLUSTER 6 | |
57 | #define MD_CLUSTER_PENDING_RECV_EVENT 7 | |
0ba95977 | 58 | #define MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD 8 |
fa8259da | 59 | |
c4ce867f | 60 | struct md_cluster_info { |
0ba95977 | 61 | struct mddev *mddev; /* the md device which md_cluster_info belongs to */ |
c4ce867f GR |
62 | /* dlm lock space and resources for clustered raid. */ |
63 | dlm_lockspace_t *lockspace; | |
cf921cc1 GR |
64 | int slot_number; |
65 | struct completion completion; | |
8b9277c8 | 66 | struct mutex recv_mutex; |
54519c5f | 67 | struct dlm_lock_resource *bitmap_lockres; |
f6a2dc64 | 68 | struct dlm_lock_resource **other_bitmap_lockres; |
c186b128 | 69 | struct dlm_lock_resource *resync_lockres; |
96ae923a | 70 | struct list_head suspend_list; |
ea89238c | 71 | |
96ae923a | 72 | spinlock_t suspend_lock; |
ea89238c GJ |
73 | /* record the region which write should be suspended */ |
74 | sector_t suspend_lo; | |
75 | sector_t suspend_hi; | |
76 | int suspend_from; /* the slot which broadcast suspend_lo/hi */ | |
77 | ||
e94987db GR |
78 | struct md_thread *recovery_thread; |
79 | unsigned long recovery_map; | |
4664680c GR |
80 | /* communication loc resources */ |
81 | struct dlm_lock_resource *ack_lockres; | |
82 | struct dlm_lock_resource *message_lockres; | |
83 | struct dlm_lock_resource *token_lockres; | |
1aee41f6 | 84 | struct dlm_lock_resource *no_new_dev_lockres; |
4664680c | 85 | struct md_thread *recv_thread; |
1aee41f6 | 86 | struct completion newdisk_completion; |
8b9277c8 | 87 | wait_queue_head_t wait; |
fa8259da | 88 | unsigned long state; |
18c9ff7f GJ |
89 | /* record the region in RESYNCING message */ |
90 | sector_t sync_low; | |
91 | sector_t sync_hi; | |
4664680c GR |
92 | }; |
93 | ||
94 | enum msg_type { | |
95 | METADATA_UPDATED = 0, | |
96 | RESYNCING, | |
1aee41f6 | 97 | NEWDISK, |
88bcfef7 | 98 | REMOVE, |
97f6cd39 | 99 | RE_ADD, |
dc737d7c | 100 | BITMAP_NEEDS_SYNC, |
7da3d203 | 101 | CHANGE_CAPACITY, |
afd75628 | 102 | BITMAP_RESIZE, |
4664680c GR |
103 | }; |
104 | ||
105 | struct cluster_msg { | |
cf97a348 GJ |
106 | __le32 type; |
107 | __le32 slot; | |
1aee41f6 | 108 | /* TODO: Unionize this for smaller footprint */ |
cf97a348 GJ |
109 | __le64 low; |
110 | __le64 high; | |
1aee41f6 | 111 | char uuid[16]; |
cf97a348 | 112 | __le32 raid_slot; |
47741b7c GR |
113 | }; |
114 | ||
115 | static void sync_ast(void *arg) | |
116 | { | |
117 | struct dlm_lock_resource *res; | |
118 | ||
2e2a7cd9 | 119 | res = arg; |
fccb60a4 GJ |
120 | res->sync_locking_done = true; |
121 | wake_up(&res->sync_locking); | |
47741b7c GR |
122 | } |
123 | ||
124 | static int dlm_lock_sync(struct dlm_lock_resource *res, int mode) | |
125 | { | |
126 | int ret = 0; | |
127 | ||
47741b7c GR |
128 | ret = dlm_lock(res->ls, mode, &res->lksb, |
129 | res->flags, res->name, strlen(res->name), | |
130 | 0, sync_ast, res, res->bast); | |
131 | if (ret) | |
132 | return ret; | |
fccb60a4 GJ |
133 | wait_event(res->sync_locking, res->sync_locking_done); |
134 | res->sync_locking_done = false; | |
dbb64f86 GR |
135 | if (res->lksb.sb_status == 0) |
136 | res->mode = mode; | |
47741b7c GR |
137 | return res->lksb.sb_status; |
138 | } | |
139 | ||
140 | static int dlm_unlock_sync(struct dlm_lock_resource *res) | |
141 | { | |
142 | return dlm_lock_sync(res, DLM_LOCK_NL); | |
143 | } | |
144 | ||
7bcda714 GJ |
145 | /* |
146 | * An variation of dlm_lock_sync, which make lock request could | |
147 | * be interrupted | |
148 | */ | |
149 | static int dlm_lock_sync_interruptible(struct dlm_lock_resource *res, int mode, | |
150 | struct mddev *mddev) | |
151 | { | |
152 | int ret = 0; | |
153 | ||
154 | ret = dlm_lock(res->ls, mode, &res->lksb, | |
155 | res->flags, res->name, strlen(res->name), | |
156 | 0, sync_ast, res, res->bast); | |
157 | if (ret) | |
158 | return ret; | |
159 | ||
160 | wait_event(res->sync_locking, res->sync_locking_done | |
d6385db9 GJ |
161 | || kthread_should_stop() |
162 | || test_bit(MD_CLOSING, &mddev->flags)); | |
7bcda714 GJ |
163 | if (!res->sync_locking_done) { |
164 | /* | |
165 | * the convert queue contains the lock request when request is | |
166 | * interrupted, and sync_ast could still be run, so need to | |
167 | * cancel the request and reset completion | |
168 | */ | |
169 | ret = dlm_unlock(res->ls, res->lksb.sb_lkid, DLM_LKF_CANCEL, | |
170 | &res->lksb, res); | |
171 | res->sync_locking_done = false; | |
172 | if (unlikely(ret != 0)) | |
173 | pr_info("failed to cancel previous lock request " | |
174 | "%s return %d\n", res->name, ret); | |
175 | return -EPERM; | |
176 | } else | |
177 | res->sync_locking_done = false; | |
178 | if (res->lksb.sb_status == 0) | |
179 | res->mode = mode; | |
180 | return res->lksb.sb_status; | |
181 | } | |
182 | ||
c4ce867f | 183 | static struct dlm_lock_resource *lockres_init(struct mddev *mddev, |
47741b7c GR |
184 | char *name, void (*bastfn)(void *arg, int mode), int with_lvb) |
185 | { | |
186 | struct dlm_lock_resource *res = NULL; | |
187 | int ret, namelen; | |
c4ce867f | 188 | struct md_cluster_info *cinfo = mddev->cluster_info; |
47741b7c GR |
189 | |
190 | res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL); | |
191 | if (!res) | |
192 | return NULL; | |
fccb60a4 GJ |
193 | init_waitqueue_head(&res->sync_locking); |
194 | res->sync_locking_done = false; | |
c4ce867f GR |
195 | res->ls = cinfo->lockspace; |
196 | res->mddev = mddev; | |
dbb64f86 | 197 | res->mode = DLM_LOCK_IV; |
47741b7c GR |
198 | namelen = strlen(name); |
199 | res->name = kzalloc(namelen + 1, GFP_KERNEL); | |
200 | if (!res->name) { | |
201 | pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name); | |
202 | goto out_err; | |
203 | } | |
92d9aac9 | 204 | strscpy(res->name, name, namelen + 1); |
47741b7c GR |
205 | if (with_lvb) { |
206 | res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL); | |
207 | if (!res->lksb.sb_lvbptr) { | |
208 | pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name); | |
209 | goto out_err; | |
210 | } | |
211 | res->flags = DLM_LKF_VALBLK; | |
212 | } | |
213 | ||
214 | if (bastfn) | |
215 | res->bast = bastfn; | |
216 | ||
217 | res->flags |= DLM_LKF_EXPEDITE; | |
218 | ||
219 | ret = dlm_lock_sync(res, DLM_LOCK_NL); | |
220 | if (ret) { | |
221 | pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name); | |
222 | goto out_err; | |
223 | } | |
224 | res->flags &= ~DLM_LKF_EXPEDITE; | |
225 | res->flags |= DLM_LKF_CONVERT; | |
226 | ||
227 | return res; | |
228 | out_err: | |
229 | kfree(res->lksb.sb_lvbptr); | |
230 | kfree(res->name); | |
231 | kfree(res); | |
232 | return NULL; | |
233 | } | |
234 | ||
235 | static void lockres_free(struct dlm_lock_resource *res) | |
236 | { | |
400cb454 | 237 | int ret = 0; |
b5ef5678 | 238 | |
47741b7c GR |
239 | if (!res) |
240 | return; | |
241 | ||
400cb454 GJ |
242 | /* |
243 | * use FORCEUNLOCK flag, so we can unlock even the lock is on the | |
244 | * waiting or convert queue | |
245 | */ | |
246 | ret = dlm_unlock(res->ls, res->lksb.sb_lkid, DLM_LKF_FORCEUNLOCK, | |
247 | &res->lksb, res); | |
248 | if (unlikely(ret != 0)) | |
249 | pr_err("failed to unlock %s return %d\n", res->name, ret); | |
250 | else | |
fccb60a4 | 251 | wait_event(res->sync_locking, res->sync_locking_done); |
47741b7c GR |
252 | |
253 | kfree(res->name); | |
254 | kfree(res->lksb.sb_lvbptr); | |
255 | kfree(res); | |
256 | } | |
8e854e9c | 257 | |
30661b49 N |
258 | static void add_resync_info(struct dlm_lock_resource *lockres, |
259 | sector_t lo, sector_t hi) | |
96ae923a GR |
260 | { |
261 | struct resync_info *ri; | |
262 | ||
263 | ri = (struct resync_info *)lockres->lksb.sb_lvbptr; | |
264 | ri->lo = cpu_to_le64(lo); | |
265 | ri->hi = cpu_to_le64(hi); | |
266 | } | |
267 | ||
ea89238c GJ |
268 | static int read_resync_info(struct mddev *mddev, |
269 | struct dlm_lock_resource *lockres) | |
96ae923a GR |
270 | { |
271 | struct resync_info ri; | |
ea89238c GJ |
272 | struct md_cluster_info *cinfo = mddev->cluster_info; |
273 | int ret = 0; | |
96ae923a GR |
274 | |
275 | dlm_lock_sync(lockres, DLM_LOCK_CR); | |
276 | memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info)); | |
ea89238c GJ |
277 | if (le64_to_cpu(ri.hi) > 0) { |
278 | cinfo->suspend_hi = le64_to_cpu(ri.hi); | |
279 | cinfo->suspend_lo = le64_to_cpu(ri.lo); | |
280 | ret = 1; | |
96ae923a GR |
281 | } |
282 | dlm_unlock_sync(lockres); | |
ea89238c | 283 | return ret; |
96ae923a GR |
284 | } |
285 | ||
6dc69c9c | 286 | static void recover_bitmaps(struct md_thread *thread) |
e94987db GR |
287 | { |
288 | struct mddev *mddev = thread->mddev; | |
289 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
290 | struct dlm_lock_resource *bm_lockres; | |
291 | char str[64]; | |
292 | int slot, ret; | |
e94987db GR |
293 | sector_t lo, hi; |
294 | ||
295 | while (cinfo->recovery_map) { | |
296 | slot = fls64((u64)cinfo->recovery_map) - 1; | |
297 | ||
e94987db GR |
298 | snprintf(str, 64, "bitmap%04d", slot); |
299 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
300 | if (!bm_lockres) { | |
301 | pr_err("md-cluster: Cannot initialize bitmaps\n"); | |
302 | goto clear_bit; | |
303 | } | |
304 | ||
7bcda714 | 305 | ret = dlm_lock_sync_interruptible(bm_lockres, DLM_LOCK_PW, mddev); |
e94987db GR |
306 | if (ret) { |
307 | pr_err("md-cluster: Could not DLM lock %s: %d\n", | |
308 | str, ret); | |
309 | goto clear_bit; | |
310 | } | |
e64e4018 | 311 | ret = md_bitmap_copy_from_slot(mddev, slot, &lo, &hi, true); |
4b26a08a | 312 | if (ret) { |
e94987db | 313 | pr_err("md-cluster: Could not copy data from bitmap %d\n", slot); |
e3f924d3 | 314 | goto clear_bit; |
4b26a08a | 315 | } |
010228e4 GJ |
316 | |
317 | /* Clear suspend_area associated with the bitmap */ | |
318 | spin_lock_irq(&cinfo->suspend_lock); | |
ea89238c GJ |
319 | cinfo->suspend_hi = 0; |
320 | cinfo->suspend_lo = 0; | |
321 | cinfo->suspend_from = -1; | |
010228e4 GJ |
322 | spin_unlock_irq(&cinfo->suspend_lock); |
323 | ||
cb9ee154 GJ |
324 | /* Kick off a reshape if needed */ |
325 | if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) && | |
326 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
327 | mddev->reshape_position != MaxSector) | |
328 | md_wakeup_thread(mddev->sync_thread); | |
329 | ||
4b26a08a | 330 | if (hi > 0) { |
4b26a08a GR |
331 | if (lo < mddev->recovery_cp) |
332 | mddev->recovery_cp = lo; | |
eb315cd0 GJ |
333 | /* wake up thread to continue resync in case resync |
334 | * is not finished */ | |
335 | if (mddev->recovery_cp != MaxSector) { | |
0357ba27 GJ |
336 | /* |
337 | * clear the REMOTE flag since we will launch | |
338 | * resync thread in current node. | |
339 | */ | |
340 | clear_bit(MD_RESYNCING_REMOTE, | |
341 | &mddev->recovery); | |
342 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
343 | md_wakeup_thread(mddev->thread); | |
eb315cd0 | 344 | } |
4b26a08a | 345 | } |
e94987db | 346 | clear_bit: |
4ac7a65f | 347 | lockres_free(bm_lockres); |
e94987db GR |
348 | clear_bit(slot, &cinfo->recovery_map); |
349 | } | |
350 | } | |
351 | ||
cf921cc1 GR |
352 | static void recover_prep(void *arg) |
353 | { | |
90382ed9 GR |
354 | struct mddev *mddev = arg; |
355 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
356 | set_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); | |
cf921cc1 GR |
357 | } |
358 | ||
05cd0e51 | 359 | static void __recover_slot(struct mddev *mddev, int slot) |
cf921cc1 | 360 | { |
cf921cc1 GR |
361 | struct md_cluster_info *cinfo = mddev->cluster_info; |
362 | ||
05cd0e51 | 363 | set_bit(slot, &cinfo->recovery_map); |
e94987db GR |
364 | if (!cinfo->recovery_thread) { |
365 | cinfo->recovery_thread = md_register_thread(recover_bitmaps, | |
366 | mddev, "recover"); | |
367 | if (!cinfo->recovery_thread) { | |
368 | pr_warn("md-cluster: Could not create recovery thread\n"); | |
369 | return; | |
370 | } | |
371 | } | |
372 | md_wakeup_thread(cinfo->recovery_thread); | |
cf921cc1 GR |
373 | } |
374 | ||
05cd0e51 GJ |
375 | static void recover_slot(void *arg, struct dlm_slot *slot) |
376 | { | |
377 | struct mddev *mddev = arg; | |
378 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
379 | ||
380 | pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n", | |
381 | mddev->bitmap_info.cluster_name, | |
382 | slot->nodeid, slot->slot, | |
383 | cinfo->slot_number); | |
384 | /* deduct one since dlm slot starts from one while the num of | |
385 | * cluster-md begins with 0 */ | |
386 | __recover_slot(mddev, slot->slot - 1); | |
387 | } | |
388 | ||
cf921cc1 GR |
389 | static void recover_done(void *arg, struct dlm_slot *slots, |
390 | int num_slots, int our_slot, | |
391 | uint32_t generation) | |
392 | { | |
393 | struct mddev *mddev = arg; | |
394 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
395 | ||
396 | cinfo->slot_number = our_slot; | |
eece075c GJ |
397 | /* completion is only need to be complete when node join cluster, |
398 | * it doesn't need to run during another node's failure */ | |
399 | if (test_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state)) { | |
400 | complete(&cinfo->completion); | |
401 | clear_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state); | |
402 | } | |
90382ed9 | 403 | clear_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state); |
cf921cc1 GR |
404 | } |
405 | ||
eece075c GJ |
406 | /* the ops is called when node join the cluster, and do lock recovery |
407 | * if node failure occurs */ | |
cf921cc1 GR |
408 | static const struct dlm_lockspace_ops md_ls_ops = { |
409 | .recover_prep = recover_prep, | |
410 | .recover_slot = recover_slot, | |
411 | .recover_done = recover_done, | |
412 | }; | |
413 | ||
4664680c GR |
414 | /* |
415 | * The BAST function for the ack lock resource | |
416 | * This function wakes up the receive thread in | |
417 | * order to receive and process the message. | |
418 | */ | |
419 | static void ack_bast(void *arg, int mode) | |
420 | { | |
2e2a7cd9 | 421 | struct dlm_lock_resource *res = arg; |
4664680c GR |
422 | struct md_cluster_info *cinfo = res->mddev->cluster_info; |
423 | ||
51e453ae GJ |
424 | if (mode == DLM_LOCK_EX) { |
425 | if (test_bit(MD_CLUSTER_ALREADY_IN_CLUSTER, &cinfo->state)) | |
426 | md_wakeup_thread(cinfo->recv_thread); | |
427 | else | |
428 | set_bit(MD_CLUSTER_PENDING_RECV_EVENT, &cinfo->state); | |
429 | } | |
4664680c GR |
430 | } |
431 | ||
b8ca846e | 432 | static void remove_suspend_info(struct mddev *mddev, int slot) |
e59721cc | 433 | { |
b8ca846e | 434 | struct md_cluster_info *cinfo = mddev->cluster_info; |
b03e0ccb | 435 | mddev->pers->quiesce(mddev, 1); |
e59721cc | 436 | spin_lock_irq(&cinfo->suspend_lock); |
ea89238c GJ |
437 | cinfo->suspend_hi = 0; |
438 | cinfo->suspend_lo = 0; | |
e59721cc | 439 | spin_unlock_irq(&cinfo->suspend_lock); |
b03e0ccb | 440 | mddev->pers->quiesce(mddev, 0); |
e59721cc GR |
441 | } |
442 | ||
9ed38ff5 | 443 | static void process_suspend_info(struct mddev *mddev, |
e59721cc GR |
444 | int slot, sector_t lo, sector_t hi) |
445 | { | |
9ed38ff5 | 446 | struct md_cluster_info *cinfo = mddev->cluster_info; |
cbce6863 GJ |
447 | struct mdp_superblock_1 *sb = NULL; |
448 | struct md_rdev *rdev; | |
e59721cc GR |
449 | |
450 | if (!hi) { | |
0357ba27 GJ |
451 | /* |
452 | * clear the REMOTE flag since resync or recovery is finished | |
453 | * in remote node. | |
454 | */ | |
455 | clear_bit(MD_RESYNCING_REMOTE, &mddev->recovery); | |
b8ca846e | 456 | remove_suspend_info(mddev, slot); |
c186b128 GR |
457 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
458 | md_wakeup_thread(mddev->thread); | |
e59721cc GR |
459 | return; |
460 | } | |
18c9ff7f | 461 | |
cbce6863 GJ |
462 | rdev_for_each(rdev, mddev) |
463 | if (rdev->raid_disk > -1 && !test_bit(Faulty, &rdev->flags)) { | |
464 | sb = page_address(rdev->sb_page); | |
465 | break; | |
466 | } | |
467 | ||
18c9ff7f GJ |
468 | /* |
469 | * The bitmaps are not same for different nodes | |
470 | * if RESYNCING is happening in one node, then | |
471 | * the node which received the RESYNCING message | |
472 | * probably will perform resync with the region | |
473 | * [lo, hi] again, so we could reduce resync time | |
474 | * a lot if we can ensure that the bitmaps among | |
475 | * different nodes are match up well. | |
476 | * | |
477 | * sync_low/hi is used to record the region which | |
478 | * arrived in the previous RESYNCING message, | |
479 | * | |
cbce6863 GJ |
480 | * Call md_bitmap_sync_with_cluster to clear NEEDED_MASK |
481 | * and set RESYNC_MASK since resync thread is running | |
482 | * in another node, so we don't need to do the resync | |
483 | * again with the same section. | |
484 | * | |
485 | * Skip md_bitmap_sync_with_cluster in case reshape | |
486 | * happening, because reshaping region is small and | |
487 | * we don't want to trigger lots of WARN. | |
488 | */ | |
489 | if (sb && !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) | |
490 | md_bitmap_sync_with_cluster(mddev, cinfo->sync_low, | |
491 | cinfo->sync_hi, lo, hi); | |
18c9ff7f GJ |
492 | cinfo->sync_low = lo; |
493 | cinfo->sync_hi = hi; | |
494 | ||
9ed38ff5 | 495 | mddev->pers->quiesce(mddev, 1); |
e59721cc | 496 | spin_lock_irq(&cinfo->suspend_lock); |
ea89238c GJ |
497 | cinfo->suspend_from = slot; |
498 | cinfo->suspend_lo = lo; | |
499 | cinfo->suspend_hi = hi; | |
e59721cc | 500 | spin_unlock_irq(&cinfo->suspend_lock); |
b03e0ccb | 501 | mddev->pers->quiesce(mddev, 0); |
e59721cc GR |
502 | } |
503 | ||
1aee41f6 GR |
504 | static void process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg) |
505 | { | |
506 | char disk_uuid[64]; | |
507 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
508 | char event_name[] = "EVENT=ADD_DEVICE"; | |
509 | char raid_slot[16]; | |
510 | char *envp[] = {event_name, disk_uuid, raid_slot, NULL}; | |
511 | int len; | |
512 | ||
513 | len = snprintf(disk_uuid, 64, "DEVICE_UUID="); | |
b89f704a | 514 | sprintf(disk_uuid + len, "%pU", cmsg->uuid); |
faeff83f | 515 | snprintf(raid_slot, 16, "RAID_DISK=%d", le32_to_cpu(cmsg->raid_slot)); |
1aee41f6 GR |
516 | pr_info("%s:%d Sending kobject change with %s and %s\n", __func__, __LINE__, disk_uuid, raid_slot); |
517 | init_completion(&cinfo->newdisk_completion); | |
fa8259da | 518 | set_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
519 | kobject_uevent_env(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE, envp); |
520 | wait_for_completion_timeout(&cinfo->newdisk_completion, | |
521 | NEW_DEV_TIMEOUT); | |
fa8259da | 522 | clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); |
1aee41f6 GR |
523 | } |
524 | ||
525 | ||
526 | static void process_metadata_update(struct mddev *mddev, struct cluster_msg *msg) | |
527 | { | |
0ba95977 | 528 | int got_lock = 0; |
1aee41f6 | 529 | struct md_cluster_info *cinfo = mddev->cluster_info; |
15858fa5 | 530 | mddev->good_device_nr = le32_to_cpu(msg->raid_slot); |
0ba95977 | 531 | |
1aee41f6 | 532 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); |
0ba95977 GJ |
533 | wait_event(mddev->thread->wqueue, |
534 | (got_lock = mddev_trylock(mddev)) || | |
535 | test_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state)); | |
536 | md_reload_sb(mddev, mddev->good_device_nr); | |
537 | if (got_lock) | |
538 | mddev_unlock(mddev); | |
1aee41f6 GR |
539 | } |
540 | ||
88bcfef7 GR |
541 | static void process_remove_disk(struct mddev *mddev, struct cluster_msg *msg) |
542 | { | |
5f0aa21d | 543 | struct md_rdev *rdev; |
88bcfef7 | 544 | |
5f0aa21d GJ |
545 | rcu_read_lock(); |
546 | rdev = md_find_rdev_nr_rcu(mddev, le32_to_cpu(msg->raid_slot)); | |
659b254f GJ |
547 | if (rdev) { |
548 | set_bit(ClusterRemove, &rdev->flags); | |
549 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
550 | md_wakeup_thread(mddev->thread); | |
551 | } | |
88bcfef7 | 552 | else |
faeff83f GJ |
553 | pr_warn("%s: %d Could not find disk(%d) to REMOVE\n", |
554 | __func__, __LINE__, le32_to_cpu(msg->raid_slot)); | |
5f0aa21d | 555 | rcu_read_unlock(); |
88bcfef7 GR |
556 | } |
557 | ||
97f6cd39 GR |
558 | static void process_readd_disk(struct mddev *mddev, struct cluster_msg *msg) |
559 | { | |
5f0aa21d | 560 | struct md_rdev *rdev; |
97f6cd39 | 561 | |
5f0aa21d GJ |
562 | rcu_read_lock(); |
563 | rdev = md_find_rdev_nr_rcu(mddev, le32_to_cpu(msg->raid_slot)); | |
97f6cd39 GR |
564 | if (rdev && test_bit(Faulty, &rdev->flags)) |
565 | clear_bit(Faulty, &rdev->flags); | |
566 | else | |
faeff83f GJ |
567 | pr_warn("%s: %d Could not find disk(%d) which is faulty", |
568 | __func__, __LINE__, le32_to_cpu(msg->raid_slot)); | |
5f0aa21d | 569 | rcu_read_unlock(); |
97f6cd39 GR |
570 | } |
571 | ||
1fa9a1ad | 572 | static int process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg) |
4664680c | 573 | { |
1fa9a1ad GJ |
574 | int ret = 0; |
575 | ||
256f5b24 | 576 | if (WARN(mddev->cluster_info->slot_number - 1 == le32_to_cpu(msg->slot), |
dd3dc5f4 | 577 | "node %d received its own msg\n", le32_to_cpu(msg->slot))) |
1fa9a1ad | 578 | return -1; |
cf97a348 | 579 | switch (le32_to_cpu(msg->type)) { |
4664680c | 580 | case METADATA_UPDATED: |
1aee41f6 | 581 | process_metadata_update(mddev, msg); |
4664680c | 582 | break; |
7da3d203 | 583 | case CHANGE_CAPACITY: |
2c247c51 | 584 | set_capacity_and_notify(mddev->gendisk, mddev->array_sectors); |
7da3d203 | 585 | break; |
4664680c | 586 | case RESYNCING: |
0357ba27 | 587 | set_bit(MD_RESYNCING_REMOTE, &mddev->recovery); |
cf97a348 GJ |
588 | process_suspend_info(mddev, le32_to_cpu(msg->slot), |
589 | le64_to_cpu(msg->low), | |
590 | le64_to_cpu(msg->high)); | |
4664680c | 591 | break; |
1aee41f6 | 592 | case NEWDISK: |
1aee41f6 | 593 | process_add_new_disk(mddev, msg); |
88bcfef7 GR |
594 | break; |
595 | case REMOVE: | |
88bcfef7 GR |
596 | process_remove_disk(mddev, msg); |
597 | break; | |
97f6cd39 | 598 | case RE_ADD: |
97f6cd39 GR |
599 | process_readd_disk(mddev, msg); |
600 | break; | |
dc737d7c | 601 | case BITMAP_NEEDS_SYNC: |
cf97a348 | 602 | __recover_slot(mddev, le32_to_cpu(msg->slot)); |
dc737d7c | 603 | break; |
afd75628 GJ |
604 | case BITMAP_RESIZE: |
605 | if (le64_to_cpu(msg->high) != mddev->pers->size(mddev, 0, 0)) | |
606 | ret = md_bitmap_resize(mddev->bitmap, | |
607 | le64_to_cpu(msg->high), 0, 0); | |
608 | break; | |
88bcfef7 | 609 | default: |
1fa9a1ad | 610 | ret = -1; |
88bcfef7 GR |
611 | pr_warn("%s:%d Received unknown message from %d\n", |
612 | __func__, __LINE__, msg->slot); | |
09dd1af2 | 613 | } |
1fa9a1ad | 614 | return ret; |
4664680c GR |
615 | } |
616 | ||
617 | /* | |
618 | * thread for receiving message | |
619 | */ | |
620 | static void recv_daemon(struct md_thread *thread) | |
621 | { | |
622 | struct md_cluster_info *cinfo = thread->mddev->cluster_info; | |
623 | struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres; | |
624 | struct dlm_lock_resource *message_lockres = cinfo->message_lockres; | |
625 | struct cluster_msg msg; | |
b5ef5678 | 626 | int ret; |
4664680c | 627 | |
8b9277c8 | 628 | mutex_lock(&cinfo->recv_mutex); |
4664680c GR |
629 | /*get CR on Message*/ |
630 | if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) { | |
631 | pr_err("md/raid1:failed to get CR on MESSAGE\n"); | |
8b9277c8 | 632 | mutex_unlock(&cinfo->recv_mutex); |
4664680c GR |
633 | return; |
634 | } | |
635 | ||
636 | /* read lvb and wake up thread to process this message_lockres */ | |
637 | memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg)); | |
1fa9a1ad GJ |
638 | ret = process_recvd_msg(thread->mddev, &msg); |
639 | if (ret) | |
640 | goto out; | |
4664680c GR |
641 | |
642 | /*release CR on ack_lockres*/ | |
b5ef5678 GJ |
643 | ret = dlm_unlock_sync(ack_lockres); |
644 | if (unlikely(ret != 0)) | |
645 | pr_info("unlock ack failed return %d\n", ret); | |
66099bb0 | 646 | /*up-convert to PR on message_lockres*/ |
b5ef5678 GJ |
647 | ret = dlm_lock_sync(message_lockres, DLM_LOCK_PR); |
648 | if (unlikely(ret != 0)) | |
649 | pr_info("lock PR on msg failed return %d\n", ret); | |
4664680c | 650 | /*get CR on ack_lockres again*/ |
b5ef5678 GJ |
651 | ret = dlm_lock_sync(ack_lockres, DLM_LOCK_CR); |
652 | if (unlikely(ret != 0)) | |
653 | pr_info("lock CR on ack failed return %d\n", ret); | |
1fa9a1ad | 654 | out: |
4664680c | 655 | /*release CR on message_lockres*/ |
b5ef5678 GJ |
656 | ret = dlm_unlock_sync(message_lockres); |
657 | if (unlikely(ret != 0)) | |
658 | pr_info("unlock msg failed return %d\n", ret); | |
8b9277c8 | 659 | mutex_unlock(&cinfo->recv_mutex); |
4664680c GR |
660 | } |
661 | ||
8b9277c8 | 662 | /* lock_token() |
601b515c GR |
663 | * Takes the lock on the TOKEN lock resource so no other |
664 | * node can communicate while the operation is underway. | |
665 | */ | |
bca5b065 | 666 | static int lock_token(struct md_cluster_info *cinfo) |
601b515c | 667 | { |
bca5b065 ZH |
668 | int error; |
669 | ||
670 | error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX); | |
671 | if (error) { | |
672 | pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n", | |
673 | __func__, __LINE__, error); | |
674 | } else { | |
675 | /* Lock the receive sequence */ | |
676 | mutex_lock(&cinfo->recv_mutex); | |
677 | } | |
678 | return error; | |
679 | } | |
680 | ||
681 | /* lock_comm() | |
682 | * Sets the MD_CLUSTER_SEND_LOCK bit to lock the send channel. | |
683 | */ | |
684 | static int lock_comm(struct md_cluster_info *cinfo, bool mddev_locked) | |
685 | { | |
686 | int rv, set_bit = 0; | |
0ba95977 | 687 | struct mddev *mddev = cinfo->mddev; |
601b515c | 688 | |
0ba95977 GJ |
689 | /* |
690 | * If resync thread run after raid1d thread, then process_metadata_update | |
691 | * could not continue if raid1d held reconfig_mutex (and raid1d is blocked | |
692 | * since another node already got EX on Token and waitting the EX of Ack), | |
693 | * so let resync wake up thread in case flag is set. | |
694 | */ | |
695 | if (mddev_locked && !test_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, | |
696 | &cinfo->state)) { | |
bca5b065 | 697 | rv = test_and_set_bit_lock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, |
0ba95977 | 698 | &cinfo->state); |
bca5b065 | 699 | WARN_ON_ONCE(rv); |
0ba95977 GJ |
700 | md_wakeup_thread(mddev->thread); |
701 | set_bit = 1; | |
702 | } | |
0ba95977 | 703 | |
8b9277c8 GJ |
704 | wait_event(cinfo->wait, |
705 | !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state)); | |
bca5b065 ZH |
706 | rv = lock_token(cinfo); |
707 | if (set_bit) | |
708 | clear_bit_unlock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state); | |
709 | return rv; | |
8b9277c8 GJ |
710 | } |
711 | ||
601b515c GR |
712 | static void unlock_comm(struct md_cluster_info *cinfo) |
713 | { | |
dbb64f86 | 714 | WARN_ON(cinfo->token_lockres->mode != DLM_LOCK_EX); |
8b9277c8 | 715 | mutex_unlock(&cinfo->recv_mutex); |
601b515c | 716 | dlm_unlock_sync(cinfo->token_lockres); |
8b9277c8 GJ |
717 | clear_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state); |
718 | wake_up(&cinfo->wait); | |
601b515c GR |
719 | } |
720 | ||
721 | /* __sendmsg() | |
722 | * This function performs the actual sending of the message. This function is | |
723 | * usually called after performing the encompassing operation | |
724 | * The function: | |
725 | * 1. Grabs the message lockresource in EX mode | |
726 | * 2. Copies the message to the message LVB | |
66099bb0 | 727 | * 3. Downconverts message lockresource to CW |
601b515c GR |
728 | * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes |
729 | * and the other nodes read the message. The thread will wait here until all other | |
730 | * nodes have released ack lock resource. | |
731 | * 5. Downconvert ack lockresource to CR | |
732 | */ | |
733 | static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg) | |
734 | { | |
735 | int error; | |
736 | int slot = cinfo->slot_number - 1; | |
737 | ||
738 | cmsg->slot = cpu_to_le32(slot); | |
739 | /*get EX on Message*/ | |
740 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX); | |
741 | if (error) { | |
742 | pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error); | |
743 | goto failed_message; | |
744 | } | |
745 | ||
746 | memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg, | |
747 | sizeof(struct cluster_msg)); | |
66099bb0 GJ |
748 | /*down-convert EX to CW on Message*/ |
749 | error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CW); | |
601b515c | 750 | if (error) { |
66099bb0 | 751 | pr_err("md-cluster: failed to convert EX to CW on MESSAGE(%d)\n", |
601b515c | 752 | error); |
66099bb0 | 753 | goto failed_ack; |
601b515c GR |
754 | } |
755 | ||
756 | /*up-convert CR to EX on Ack*/ | |
757 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX); | |
758 | if (error) { | |
759 | pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n", | |
760 | error); | |
761 | goto failed_ack; | |
762 | } | |
763 | ||
764 | /*down-convert EX to CR on Ack*/ | |
765 | error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR); | |
766 | if (error) { | |
767 | pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n", | |
768 | error); | |
769 | goto failed_ack; | |
770 | } | |
771 | ||
772 | failed_ack: | |
b5ef5678 GJ |
773 | error = dlm_unlock_sync(cinfo->message_lockres); |
774 | if (unlikely(error != 0)) { | |
775 | pr_err("md-cluster: failed convert to NL on MESSAGE(%d)\n", | |
776 | error); | |
777 | /* in case the message can't be released due to some reason */ | |
778 | goto failed_ack; | |
779 | } | |
601b515c GR |
780 | failed_message: |
781 | return error; | |
782 | } | |
783 | ||
0ba95977 GJ |
784 | static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg, |
785 | bool mddev_locked) | |
601b515c GR |
786 | { |
787 | int ret; | |
788 | ||
bca5b065 ZH |
789 | ret = lock_comm(cinfo, mddev_locked); |
790 | if (!ret) { | |
791 | ret = __sendmsg(cinfo, cmsg); | |
792 | unlock_comm(cinfo); | |
793 | } | |
601b515c GR |
794 | return ret; |
795 | } | |
796 | ||
96ae923a GR |
797 | static int gather_all_resync_info(struct mddev *mddev, int total_slots) |
798 | { | |
799 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
800 | int i, ret = 0; | |
801 | struct dlm_lock_resource *bm_lockres; | |
96ae923a | 802 | char str[64]; |
abb9b22a | 803 | sector_t lo, hi; |
96ae923a GR |
804 | |
805 | ||
806 | for (i = 0; i < total_slots; i++) { | |
807 | memset(str, '\0', 64); | |
808 | snprintf(str, 64, "bitmap%04d", i); | |
809 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
810 | if (!bm_lockres) | |
811 | return -ENOMEM; | |
4ac7a65f SL |
812 | if (i == (cinfo->slot_number - 1)) { |
813 | lockres_free(bm_lockres); | |
96ae923a | 814 | continue; |
4ac7a65f | 815 | } |
96ae923a GR |
816 | |
817 | bm_lockres->flags |= DLM_LKF_NOQUEUE; | |
818 | ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
819 | if (ret == -EAGAIN) { | |
ea89238c | 820 | if (read_resync_info(mddev, bm_lockres)) { |
96ae923a GR |
821 | pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n", |
822 | __func__, __LINE__, | |
ea89238c GJ |
823 | (unsigned long long) cinfo->suspend_lo, |
824 | (unsigned long long) cinfo->suspend_hi, | |
825 | i); | |
826 | cinfo->suspend_from = i; | |
96ae923a GR |
827 | } |
828 | ret = 0; | |
829 | lockres_free(bm_lockres); | |
830 | continue; | |
831 | } | |
6e6d9f2c GJ |
832 | if (ret) { |
833 | lockres_free(bm_lockres); | |
96ae923a | 834 | goto out; |
6e6d9f2c | 835 | } |
abb9b22a GJ |
836 | |
837 | /* Read the disk bitmap sb and check if it needs recovery */ | |
e64e4018 | 838 | ret = md_bitmap_copy_from_slot(mddev, i, &lo, &hi, false); |
abb9b22a GJ |
839 | if (ret) { |
840 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", i); | |
841 | lockres_free(bm_lockres); | |
842 | continue; | |
843 | } | |
844 | if ((hi > 0) && (lo < mddev->recovery_cp)) { | |
845 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
846 | mddev->recovery_cp = lo; | |
847 | md_check_recovery(mddev); | |
848 | } | |
849 | ||
96ae923a GR |
850 | lockres_free(bm_lockres); |
851 | } | |
852 | out: | |
853 | return ret; | |
854 | } | |
855 | ||
edb39c9d GR |
856 | static int join(struct mddev *mddev, int nodes) |
857 | { | |
c4ce867f | 858 | struct md_cluster_info *cinfo; |
cf921cc1 | 859 | int ret, ops_rv; |
c4ce867f GR |
860 | char str[64]; |
861 | ||
c4ce867f GR |
862 | cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL); |
863 | if (!cinfo) | |
864 | return -ENOMEM; | |
865 | ||
9e3072e3 GJ |
866 | INIT_LIST_HEAD(&cinfo->suspend_list); |
867 | spin_lock_init(&cinfo->suspend_lock); | |
cf921cc1 | 868 | init_completion(&cinfo->completion); |
eece075c | 869 | set_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state); |
8b9277c8 GJ |
870 | init_waitqueue_head(&cinfo->wait); |
871 | mutex_init(&cinfo->recv_mutex); | |
cf921cc1 | 872 | |
cf921cc1 | 873 | mddev->cluster_info = cinfo; |
0ba95977 | 874 | cinfo->mddev = mddev; |
cf921cc1 | 875 | |
c4ce867f | 876 | memset(str, 0, 64); |
b89f704a | 877 | sprintf(str, "%pU", mddev->uuid); |
cf921cc1 GR |
878 | ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name, |
879 | DLM_LSFL_FS, LVB_SIZE, | |
880 | &md_ls_ops, mddev, &ops_rv, &cinfo->lockspace); | |
c4ce867f GR |
881 | if (ret) |
882 | goto err; | |
cf921cc1 | 883 | wait_for_completion(&cinfo->completion); |
8c58f02e GJ |
884 | if (nodes < cinfo->slot_number) { |
885 | pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).", | |
886 | cinfo->slot_number, nodes); | |
b97e9257 GR |
887 | ret = -ERANGE; |
888 | goto err; | |
889 | } | |
4664680c GR |
890 | /* Initiate the communication resources */ |
891 | ret = -ENOMEM; | |
892 | cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv"); | |
893 | if (!cinfo->recv_thread) { | |
894 | pr_err("md-cluster: cannot allocate memory for recv_thread!\n"); | |
895 | goto err; | |
896 | } | |
897 | cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1); | |
898 | if (!cinfo->message_lockres) | |
899 | goto err; | |
900 | cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0); | |
901 | if (!cinfo->token_lockres) | |
902 | goto err; | |
1aee41f6 GR |
903 | cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0); |
904 | if (!cinfo->no_new_dev_lockres) | |
905 | goto err; | |
906 | ||
1535212c GJ |
907 | ret = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX); |
908 | if (ret) { | |
909 | ret = -EAGAIN; | |
910 | pr_err("md-cluster: can't join cluster to avoid lock issue\n"); | |
911 | goto err; | |
912 | } | |
913 | cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0); | |
0f6187db WY |
914 | if (!cinfo->ack_lockres) { |
915 | ret = -ENOMEM; | |
1535212c | 916 | goto err; |
0f6187db | 917 | } |
4664680c GR |
918 | /* get sync CR lock on ACK. */ |
919 | if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR)) | |
920 | pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n", | |
921 | ret); | |
1535212c | 922 | dlm_unlock_sync(cinfo->token_lockres); |
1aee41f6 GR |
923 | /* get sync CR lock on no-new-dev. */ |
924 | if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR)) | |
925 | pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret); | |
926 | ||
54519c5f GR |
927 | |
928 | pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number); | |
929 | snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1); | |
930 | cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1); | |
0f6187db WY |
931 | if (!cinfo->bitmap_lockres) { |
932 | ret = -ENOMEM; | |
54519c5f | 933 | goto err; |
0f6187db | 934 | } |
54519c5f GR |
935 | if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) { |
936 | pr_err("Failed to get bitmap lock\n"); | |
937 | ret = -EINVAL; | |
938 | goto err; | |
939 | } | |
940 | ||
c186b128 | 941 | cinfo->resync_lockres = lockres_init(mddev, "resync", NULL, 0); |
0f6187db WY |
942 | if (!cinfo->resync_lockres) { |
943 | ret = -ENOMEM; | |
c186b128 | 944 | goto err; |
0f6187db | 945 | } |
c186b128 | 946 | |
edb39c9d | 947 | return 0; |
c4ce867f | 948 | err: |
0ba95977 | 949 | set_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state); |
5b0fb33e GJ |
950 | md_unregister_thread(&cinfo->recovery_thread); |
951 | md_unregister_thread(&cinfo->recv_thread); | |
4664680c GR |
952 | lockres_free(cinfo->message_lockres); |
953 | lockres_free(cinfo->token_lockres); | |
954 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 955 | lockres_free(cinfo->no_new_dev_lockres); |
c186b128 | 956 | lockres_free(cinfo->resync_lockres); |
96ae923a | 957 | lockres_free(cinfo->bitmap_lockres); |
c4ce867f GR |
958 | if (cinfo->lockspace) |
959 | dlm_release_lockspace(cinfo->lockspace, 2); | |
cf921cc1 | 960 | mddev->cluster_info = NULL; |
c4ce867f | 961 | kfree(cinfo); |
c4ce867f | 962 | return ret; |
edb39c9d GR |
963 | } |
964 | ||
51e453ae GJ |
965 | static void load_bitmaps(struct mddev *mddev, int total_slots) |
966 | { | |
967 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
968 | ||
969 | /* load all the node's bitmap info for resync */ | |
970 | if (gather_all_resync_info(mddev, total_slots)) | |
971 | pr_err("md-cluster: failed to gather all resyn infos\n"); | |
972 | set_bit(MD_CLUSTER_ALREADY_IN_CLUSTER, &cinfo->state); | |
973 | /* wake up recv thread in case something need to be handled */ | |
974 | if (test_and_clear_bit(MD_CLUSTER_PENDING_RECV_EVENT, &cinfo->state)) | |
975 | md_wakeup_thread(cinfo->recv_thread); | |
976 | } | |
977 | ||
09995411 GJ |
978 | static void resync_bitmap(struct mddev *mddev) |
979 | { | |
980 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
981 | struct cluster_msg cmsg = {0}; | |
982 | int err; | |
983 | ||
984 | cmsg.type = cpu_to_le32(BITMAP_NEEDS_SYNC); | |
0ba95977 | 985 | err = sendmsg(cinfo, &cmsg, 1); |
09995411 GJ |
986 | if (err) |
987 | pr_err("%s:%d: failed to send BITMAP_NEEDS_SYNC message (%d)\n", | |
988 | __func__, __LINE__, err); | |
989 | } | |
990 | ||
f6a2dc64 | 991 | static void unlock_all_bitmaps(struct mddev *mddev); |
edb39c9d GR |
992 | static int leave(struct mddev *mddev) |
993 | { | |
c4ce867f GR |
994 | struct md_cluster_info *cinfo = mddev->cluster_info; |
995 | ||
996 | if (!cinfo) | |
997 | return 0; | |
09995411 | 998 | |
cb9ee154 GJ |
999 | /* |
1000 | * BITMAP_NEEDS_SYNC message should be sent when node | |
09995411 | 1001 | * is leaving the cluster with dirty bitmap, also we |
cb9ee154 GJ |
1002 | * can only deliver it when dlm connection is available. |
1003 | * | |
1004 | * Also, we should send BITMAP_NEEDS_SYNC message in | |
1005 | * case reshaping is interrupted. | |
1006 | */ | |
1007 | if ((cinfo->slot_number > 0 && mddev->recovery_cp != MaxSector) || | |
1008 | (mddev->reshape_position != MaxSector && | |
1009 | test_bit(MD_CLOSING, &mddev->flags))) | |
09995411 GJ |
1010 | resync_bitmap(mddev); |
1011 | ||
0ba95977 | 1012 | set_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state); |
e94987db | 1013 | md_unregister_thread(&cinfo->recovery_thread); |
4664680c GR |
1014 | md_unregister_thread(&cinfo->recv_thread); |
1015 | lockres_free(cinfo->message_lockres); | |
1016 | lockres_free(cinfo->token_lockres); | |
1017 | lockres_free(cinfo->ack_lockres); | |
1aee41f6 | 1018 | lockres_free(cinfo->no_new_dev_lockres); |
4ac7a65f | 1019 | lockres_free(cinfo->resync_lockres); |
54519c5f | 1020 | lockres_free(cinfo->bitmap_lockres); |
f6a2dc64 | 1021 | unlock_all_bitmaps(mddev); |
c4ce867f | 1022 | dlm_release_lockspace(cinfo->lockspace, 2); |
9c8043f3 | 1023 | kfree(cinfo); |
edb39c9d GR |
1024 | return 0; |
1025 | } | |
1026 | ||
cf921cc1 GR |
1027 | /* slot_number(): Returns the MD slot number to use |
1028 | * DLM starts the slot numbers from 1, wheras cluster-md | |
1029 | * wants the number to be from zero, so we deduct one | |
1030 | */ | |
1031 | static int slot_number(struct mddev *mddev) | |
1032 | { | |
1033 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1034 | ||
1035 | return cinfo->slot_number - 1; | |
1036 | } | |
1037 | ||
8b9277c8 GJ |
1038 | /* |
1039 | * Check if the communication is already locked, else lock the communication | |
1040 | * channel. | |
1041 | * If it is already locked, token is in EX mode, and hence lock_token() | |
1042 | * should not be called. | |
1043 | */ | |
293467aa GR |
1044 | static int metadata_update_start(struct mddev *mddev) |
1045 | { | |
8b9277c8 | 1046 | struct md_cluster_info *cinfo = mddev->cluster_info; |
0ba95977 GJ |
1047 | int ret; |
1048 | ||
1049 | /* | |
1050 | * metadata_update_start is always called with the protection of | |
1051 | * reconfig_mutex, so set WAITING_FOR_TOKEN here. | |
1052 | */ | |
1053 | ret = test_and_set_bit_lock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, | |
1054 | &cinfo->state); | |
1055 | WARN_ON_ONCE(ret); | |
1056 | md_wakeup_thread(mddev->thread); | |
8b9277c8 GJ |
1057 | |
1058 | wait_event(cinfo->wait, | |
1059 | !test_and_set_bit(MD_CLUSTER_SEND_LOCK, &cinfo->state) || | |
1060 | test_and_clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state)); | |
1061 | ||
1062 | /* If token is already locked, return 0 */ | |
0ba95977 GJ |
1063 | if (cinfo->token_lockres->mode == DLM_LOCK_EX) { |
1064 | clear_bit_unlock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state); | |
8b9277c8 | 1065 | return 0; |
0ba95977 | 1066 | } |
8b9277c8 | 1067 | |
bca5b065 | 1068 | ret = lock_token(cinfo); |
0ba95977 GJ |
1069 | clear_bit_unlock(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state); |
1070 | return ret; | |
293467aa GR |
1071 | } |
1072 | ||
1073 | static int metadata_update_finish(struct mddev *mddev) | |
1074 | { | |
1075 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1076 | struct cluster_msg cmsg; | |
70bcecdb GR |
1077 | struct md_rdev *rdev; |
1078 | int ret = 0; | |
ba2746b0 | 1079 | int raid_slot = -1; |
293467aa GR |
1080 | |
1081 | memset(&cmsg, 0, sizeof(cmsg)); | |
1082 | cmsg.type = cpu_to_le32(METADATA_UPDATED); | |
70bcecdb GR |
1083 | /* Pick up a good active device number to send. |
1084 | */ | |
1085 | rdev_for_each(rdev, mddev) | |
1086 | if (rdev->raid_disk > -1 && !test_bit(Faulty, &rdev->flags)) { | |
ba2746b0 | 1087 | raid_slot = rdev->desc_nr; |
70bcecdb GR |
1088 | break; |
1089 | } | |
ba2746b0 N |
1090 | if (raid_slot >= 0) { |
1091 | cmsg.raid_slot = cpu_to_le32(raid_slot); | |
70bcecdb | 1092 | ret = __sendmsg(cinfo, &cmsg); |
ba2746b0 | 1093 | } else |
70bcecdb | 1094 | pr_warn("md-cluster: No good device id found to send\n"); |
8b9277c8 | 1095 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
293467aa GR |
1096 | unlock_comm(cinfo); |
1097 | return ret; | |
1098 | } | |
1099 | ||
dbb64f86 | 1100 | static void metadata_update_cancel(struct mddev *mddev) |
293467aa GR |
1101 | { |
1102 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
8b9277c8 | 1103 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
dbb64f86 | 1104 | unlock_comm(cinfo); |
293467aa GR |
1105 | } |
1106 | ||
afd75628 GJ |
1107 | static int update_bitmap_size(struct mddev *mddev, sector_t size) |
1108 | { | |
1109 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1110 | struct cluster_msg cmsg = {0}; | |
1111 | int ret; | |
1112 | ||
1113 | cmsg.type = cpu_to_le32(BITMAP_RESIZE); | |
1114 | cmsg.high = cpu_to_le64(size); | |
1115 | ret = sendmsg(cinfo, &cmsg, 0); | |
1116 | if (ret) | |
1117 | pr_err("%s:%d: failed to send BITMAP_RESIZE message (%d)\n", | |
1118 | __func__, __LINE__, ret); | |
1119 | return ret; | |
1120 | } | |
1121 | ||
1122 | static int resize_bitmaps(struct mddev *mddev, sector_t newsize, sector_t oldsize) | |
1123 | { | |
1124 | struct bitmap_counts *counts; | |
1125 | char str[64]; | |
1126 | struct dlm_lock_resource *bm_lockres; | |
1127 | struct bitmap *bitmap = mddev->bitmap; | |
1128 | unsigned long my_pages = bitmap->counts.pages; | |
1129 | int i, rv; | |
1130 | ||
1131 | /* | |
1132 | * We need to ensure all the nodes can grow to a larger | |
1133 | * bitmap size before make the reshaping. | |
1134 | */ | |
1135 | rv = update_bitmap_size(mddev, newsize); | |
1136 | if (rv) | |
1137 | return rv; | |
1138 | ||
1139 | for (i = 0; i < mddev->bitmap_info.nodes; i++) { | |
1140 | if (i == md_cluster_ops->slot_number(mddev)) | |
1141 | continue; | |
1142 | ||
1143 | bitmap = get_bitmap_from_slot(mddev, i); | |
1144 | if (IS_ERR(bitmap)) { | |
1145 | pr_err("can't get bitmap from slot %d\n", i); | |
e8abe1de | 1146 | bitmap = NULL; |
afd75628 GJ |
1147 | goto out; |
1148 | } | |
1149 | counts = &bitmap->counts; | |
1150 | ||
1151 | /* | |
1152 | * If we can hold the bitmap lock of one node then | |
1153 | * the slot is not occupied, update the pages. | |
1154 | */ | |
1155 | snprintf(str, 64, "bitmap%04d", i); | |
1156 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
1157 | if (!bm_lockres) { | |
1158 | pr_err("Cannot initialize %s lock\n", str); | |
1159 | goto out; | |
1160 | } | |
1161 | bm_lockres->flags |= DLM_LKF_NOQUEUE; | |
1162 | rv = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
1163 | if (!rv) | |
1164 | counts->pages = my_pages; | |
1165 | lockres_free(bm_lockres); | |
1166 | ||
1167 | if (my_pages != counts->pages) | |
1168 | /* | |
1169 | * Let's revert the bitmap size if one node | |
1170 | * can't resize bitmap | |
1171 | */ | |
1172 | goto out; | |
1383b347 | 1173 | md_bitmap_free(bitmap); |
afd75628 GJ |
1174 | } |
1175 | ||
1176 | return 0; | |
1177 | out: | |
1178 | md_bitmap_free(bitmap); | |
1179 | update_bitmap_size(mddev, oldsize); | |
1180 | return -1; | |
1181 | } | |
1182 | ||
b98938d1 GJ |
1183 | /* |
1184 | * return 0 if all the bitmaps have the same sync_size | |
1185 | */ | |
7a57157a | 1186 | static int cluster_check_sync_size(struct mddev *mddev) |
b98938d1 GJ |
1187 | { |
1188 | int i, rv; | |
1189 | bitmap_super_t *sb; | |
1190 | unsigned long my_sync_size, sync_size = 0; | |
1191 | int node_num = mddev->bitmap_info.nodes; | |
1192 | int current_slot = md_cluster_ops->slot_number(mddev); | |
1193 | struct bitmap *bitmap = mddev->bitmap; | |
1194 | char str[64]; | |
1195 | struct dlm_lock_resource *bm_lockres; | |
1196 | ||
1197 | sb = kmap_atomic(bitmap->storage.sb_page); | |
1198 | my_sync_size = sb->sync_size; | |
1199 | kunmap_atomic(sb); | |
1200 | ||
1201 | for (i = 0; i < node_num; i++) { | |
1202 | if (i == current_slot) | |
1203 | continue; | |
1204 | ||
1205 | bitmap = get_bitmap_from_slot(mddev, i); | |
1206 | if (IS_ERR(bitmap)) { | |
1207 | pr_err("can't get bitmap from slot %d\n", i); | |
1208 | return -1; | |
1209 | } | |
1210 | ||
1211 | /* | |
1212 | * If we can hold the bitmap lock of one node then | |
1213 | * the slot is not occupied, update the sb. | |
1214 | */ | |
1215 | snprintf(str, 64, "bitmap%04d", i); | |
1216 | bm_lockres = lockres_init(mddev, str, NULL, 1); | |
1217 | if (!bm_lockres) { | |
1218 | pr_err("md-cluster: Cannot initialize %s\n", str); | |
e64e4018 | 1219 | md_bitmap_free(bitmap); |
b98938d1 GJ |
1220 | return -1; |
1221 | } | |
1222 | bm_lockres->flags |= DLM_LKF_NOQUEUE; | |
1223 | rv = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); | |
1224 | if (!rv) | |
e64e4018 | 1225 | md_bitmap_update_sb(bitmap); |
b98938d1 GJ |
1226 | lockres_free(bm_lockres); |
1227 | ||
1228 | sb = kmap_atomic(bitmap->storage.sb_page); | |
1229 | if (sync_size == 0) | |
1230 | sync_size = sb->sync_size; | |
1231 | else if (sync_size != sb->sync_size) { | |
1232 | kunmap_atomic(sb); | |
e64e4018 | 1233 | md_bitmap_free(bitmap); |
b98938d1 GJ |
1234 | return -1; |
1235 | } | |
1236 | kunmap_atomic(sb); | |
e64e4018 | 1237 | md_bitmap_free(bitmap); |
b98938d1 GJ |
1238 | } |
1239 | ||
1240 | return (my_sync_size == sync_size) ? 0 : -1; | |
1241 | } | |
1242 | ||
818da59f GJ |
1243 | /* |
1244 | * Update the size for cluster raid is a little more complex, we perform it | |
1245 | * by the steps: | |
1246 | * 1. hold token lock and update superblock in initiator node. | |
1247 | * 2. send METADATA_UPDATED msg to other nodes. | |
1248 | * 3. The initiator node continues to check each bitmap's sync_size, if all | |
1249 | * bitmaps have the same value of sync_size, then we can set capacity and | |
1250 | * let other nodes to perform it. If one node can't update sync_size | |
1251 | * accordingly, we need to revert to previous value. | |
1252 | */ | |
1253 | static void update_size(struct mddev *mddev, sector_t old_dev_sectors) | |
1254 | { | |
1255 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1256 | struct cluster_msg cmsg; | |
1257 | struct md_rdev *rdev; | |
1258 | int ret = 0; | |
1259 | int raid_slot = -1; | |
1260 | ||
1261 | md_update_sb(mddev, 1); | |
bca5b065 ZH |
1262 | if (lock_comm(cinfo, 1)) { |
1263 | pr_err("%s: lock_comm failed\n", __func__); | |
1264 | return; | |
1265 | } | |
818da59f GJ |
1266 | |
1267 | memset(&cmsg, 0, sizeof(cmsg)); | |
1268 | cmsg.type = cpu_to_le32(METADATA_UPDATED); | |
1269 | rdev_for_each(rdev, mddev) | |
1270 | if (rdev->raid_disk >= 0 && !test_bit(Faulty, &rdev->flags)) { | |
1271 | raid_slot = rdev->desc_nr; | |
1272 | break; | |
1273 | } | |
1274 | if (raid_slot >= 0) { | |
1275 | cmsg.raid_slot = cpu_to_le32(raid_slot); | |
1276 | /* | |
1277 | * We can only change capiticy after all the nodes can do it, | |
1278 | * so need to wait after other nodes already received the msg | |
1279 | * and handled the change | |
1280 | */ | |
1281 | ret = __sendmsg(cinfo, &cmsg); | |
1282 | if (ret) { | |
1283 | pr_err("%s:%d: failed to send METADATA_UPDATED msg\n", | |
1284 | __func__, __LINE__); | |
1285 | unlock_comm(cinfo); | |
1286 | return; | |
1287 | } | |
1288 | } else { | |
1289 | pr_err("md-cluster: No good device id found to send\n"); | |
1290 | unlock_comm(cinfo); | |
1291 | return; | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * check the sync_size from other node's bitmap, if sync_size | |
1296 | * have already updated in other nodes as expected, send an | |
1297 | * empty metadata msg to permit the change of capacity | |
1298 | */ | |
1299 | if (cluster_check_sync_size(mddev) == 0) { | |
1300 | memset(&cmsg, 0, sizeof(cmsg)); | |
1301 | cmsg.type = cpu_to_le32(CHANGE_CAPACITY); | |
1302 | ret = __sendmsg(cinfo, &cmsg); | |
1303 | if (ret) | |
1304 | pr_err("%s:%d: failed to send CHANGE_CAPACITY msg\n", | |
1305 | __func__, __LINE__); | |
2c247c51 | 1306 | set_capacity_and_notify(mddev->gendisk, mddev->array_sectors); |
818da59f GJ |
1307 | } else { |
1308 | /* revert to previous sectors */ | |
1309 | ret = mddev->pers->resize(mddev, old_dev_sectors); | |
818da59f GJ |
1310 | ret = __sendmsg(cinfo, &cmsg); |
1311 | if (ret) | |
1312 | pr_err("%s:%d: failed to send METADATA_UPDATED msg\n", | |
1313 | __func__, __LINE__); | |
1314 | } | |
1315 | unlock_comm(cinfo); | |
1316 | } | |
1317 | ||
c186b128 GR |
1318 | static int resync_start(struct mddev *mddev) |
1319 | { | |
1320 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
d6385db9 | 1321 | return dlm_lock_sync_interruptible(cinfo->resync_lockres, DLM_LOCK_EX, mddev); |
c186b128 GR |
1322 | } |
1323 | ||
5ebaf80b GJ |
1324 | static void resync_info_get(struct mddev *mddev, sector_t *lo, sector_t *hi) |
1325 | { | |
1326 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
5ebaf80b GJ |
1327 | |
1328 | spin_lock_irq(&cinfo->suspend_lock); | |
ea89238c GJ |
1329 | *lo = cinfo->suspend_lo; |
1330 | *hi = cinfo->suspend_hi; | |
5ebaf80b GJ |
1331 | spin_unlock_irq(&cinfo->suspend_lock); |
1332 | } | |
1333 | ||
c40f341f | 1334 | static int resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi) |
965400eb GR |
1335 | { |
1336 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
ac277c6a | 1337 | struct resync_info ri; |
aee177ac | 1338 | struct cluster_msg cmsg = {0}; |
965400eb | 1339 | |
ac277c6a GR |
1340 | /* do not send zero again, if we have sent before */ |
1341 | if (hi == 0) { | |
1342 | memcpy(&ri, cinfo->bitmap_lockres->lksb.sb_lvbptr, sizeof(struct resync_info)); | |
1343 | if (le64_to_cpu(ri.hi) == 0) | |
1344 | return 0; | |
1345 | } | |
1346 | ||
30661b49 | 1347 | add_resync_info(cinfo->bitmap_lockres, lo, hi); |
c40f341f GR |
1348 | /* Re-acquire the lock to refresh LVB */ |
1349 | dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW); | |
c40f341f | 1350 | cmsg.type = cpu_to_le32(RESYNCING); |
965400eb GR |
1351 | cmsg.low = cpu_to_le64(lo); |
1352 | cmsg.high = cpu_to_le64(hi); | |
c186b128 | 1353 | |
0ba95977 GJ |
1354 | /* |
1355 | * mddev_lock is held if resync_info_update is called from | |
1356 | * resync_finish (md_reap_sync_thread -> resync_finish) | |
1357 | */ | |
1358 | if (lo == 0 && hi == 0) | |
1359 | return sendmsg(cinfo, &cmsg, 1); | |
1360 | else | |
1361 | return sendmsg(cinfo, &cmsg, 0); | |
965400eb GR |
1362 | } |
1363 | ||
c186b128 GR |
1364 | static int resync_finish(struct mddev *mddev) |
1365 | { | |
1366 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
41a95041 | 1367 | int ret = 0; |
0357ba27 GJ |
1368 | |
1369 | clear_bit(MD_RESYNCING_REMOTE, &mddev->recovery); | |
df8c6764 GJ |
1370 | |
1371 | /* | |
1372 | * If resync thread is interrupted so we can't say resync is finished, | |
1373 | * another node will launch resync thread to continue. | |
1374 | */ | |
41a95041 GJ |
1375 | if (!test_bit(MD_CLOSING, &mddev->flags)) |
1376 | ret = resync_info_update(mddev, 0, 0); | |
1377 | dlm_unlock_sync(cinfo->resync_lockres); | |
1378 | return ret; | |
c186b128 GR |
1379 | } |
1380 | ||
90382ed9 GR |
1381 | static int area_resyncing(struct mddev *mddev, int direction, |
1382 | sector_t lo, sector_t hi) | |
589a1c49 GR |
1383 | { |
1384 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1385 | int ret = 0; | |
589a1c49 | 1386 | |
90382ed9 GR |
1387 | if ((direction == READ) && |
1388 | test_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state)) | |
1389 | return 1; | |
1390 | ||
589a1c49 | 1391 | spin_lock_irq(&cinfo->suspend_lock); |
ea89238c GJ |
1392 | if (hi > cinfo->suspend_lo && lo < cinfo->suspend_hi) |
1393 | ret = 1; | |
589a1c49 GR |
1394 | spin_unlock_irq(&cinfo->suspend_lock); |
1395 | return ret; | |
1396 | } | |
1397 | ||
dbb64f86 GR |
1398 | /* add_new_disk() - initiates a disk add |
1399 | * However, if this fails before writing md_update_sb(), | |
1400 | * add_new_disk_cancel() must be called to release token lock | |
1401 | */ | |
1402 | static int add_new_disk(struct mddev *mddev, struct md_rdev *rdev) | |
1aee41f6 GR |
1403 | { |
1404 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1405 | struct cluster_msg cmsg; | |
1406 | int ret = 0; | |
1407 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); | |
1408 | char *uuid = sb->device_uuid; | |
1409 | ||
1410 | memset(&cmsg, 0, sizeof(cmsg)); | |
1411 | cmsg.type = cpu_to_le32(NEWDISK); | |
1412 | memcpy(cmsg.uuid, uuid, 16); | |
faeff83f | 1413 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); |
bca5b065 ZH |
1414 | if (lock_comm(cinfo, 1)) |
1415 | return -EAGAIN; | |
1aee41f6 | 1416 | ret = __sendmsg(cinfo, &cmsg); |
2dffdc07 GJ |
1417 | if (ret) { |
1418 | unlock_comm(cinfo); | |
1aee41f6 | 1419 | return ret; |
2dffdc07 | 1420 | } |
1aee41f6 GR |
1421 | cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE; |
1422 | ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX); | |
1423 | cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE; | |
1424 | /* Some node does not "see" the device */ | |
1425 | if (ret == -EAGAIN) | |
1426 | ret = -ENOENT; | |
dbb64f86 GR |
1427 | if (ret) |
1428 | unlock_comm(cinfo); | |
8b9277c8 | 1429 | else { |
1aee41f6 | 1430 | dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR); |
e19508fa GJ |
1431 | /* Since MD_CHANGE_DEVS will be set in add_bound_rdev which |
1432 | * will run soon after add_new_disk, the below path will be | |
1433 | * invoked: | |
1434 | * md_wakeup_thread(mddev->thread) | |
1435 | * -> conf->thread (raid1d) | |
1436 | * -> md_check_recovery -> md_update_sb | |
1437 | * -> metadata_update_start/finish | |
1438 | * MD_CLUSTER_SEND_LOCKED_ALREADY will be cleared eventually. | |
1439 | * | |
1440 | * For other failure cases, metadata_update_cancel and | |
1441 | * add_new_disk_cancel also clear below bit as well. | |
1442 | * */ | |
8b9277c8 GJ |
1443 | set_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
1444 | wake_up(&cinfo->wait); | |
1445 | } | |
1aee41f6 GR |
1446 | return ret; |
1447 | } | |
1448 | ||
dbb64f86 | 1449 | static void add_new_disk_cancel(struct mddev *mddev) |
1aee41f6 | 1450 | { |
dbb64f86 | 1451 | struct md_cluster_info *cinfo = mddev->cluster_info; |
8b9277c8 | 1452 | clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state); |
dbb64f86 | 1453 | unlock_comm(cinfo); |
1aee41f6 GR |
1454 | } |
1455 | ||
fa8259da | 1456 | static int new_disk_ack(struct mddev *mddev, bool ack) |
1aee41f6 GR |
1457 | { |
1458 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1459 | ||
fa8259da GR |
1460 | if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) { |
1461 | pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev)); | |
1462 | return -EINVAL; | |
1463 | } | |
1464 | ||
1aee41f6 GR |
1465 | if (ack) |
1466 | dlm_unlock_sync(cinfo->no_new_dev_lockres); | |
1467 | complete(&cinfo->newdisk_completion); | |
fa8259da | 1468 | return 0; |
1aee41f6 GR |
1469 | } |
1470 | ||
88bcfef7 GR |
1471 | static int remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1472 | { | |
aee177ac | 1473 | struct cluster_msg cmsg = {0}; |
88bcfef7 | 1474 | struct md_cluster_info *cinfo = mddev->cluster_info; |
faeff83f GJ |
1475 | cmsg.type = cpu_to_le32(REMOVE); |
1476 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); | |
0ba95977 | 1477 | return sendmsg(cinfo, &cmsg, 1); |
88bcfef7 GR |
1478 | } |
1479 | ||
f6a2dc64 GJ |
1480 | static int lock_all_bitmaps(struct mddev *mddev) |
1481 | { | |
1482 | int slot, my_slot, ret, held = 1, i = 0; | |
1483 | char str[64]; | |
1484 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1485 | ||
6396bb22 KC |
1486 | cinfo->other_bitmap_lockres = |
1487 | kcalloc(mddev->bitmap_info.nodes - 1, | |
1488 | sizeof(struct dlm_lock_resource *), GFP_KERNEL); | |
f6a2dc64 GJ |
1489 | if (!cinfo->other_bitmap_lockres) { |
1490 | pr_err("md: can't alloc mem for other bitmap locks\n"); | |
1491 | return 0; | |
1492 | } | |
1493 | ||
1494 | my_slot = slot_number(mddev); | |
1495 | for (slot = 0; slot < mddev->bitmap_info.nodes; slot++) { | |
1496 | if (slot == my_slot) | |
1497 | continue; | |
1498 | ||
1499 | memset(str, '\0', 64); | |
1500 | snprintf(str, 64, "bitmap%04d", slot); | |
1501 | cinfo->other_bitmap_lockres[i] = lockres_init(mddev, str, NULL, 1); | |
1502 | if (!cinfo->other_bitmap_lockres[i]) | |
1503 | return -ENOMEM; | |
1504 | ||
1505 | cinfo->other_bitmap_lockres[i]->flags |= DLM_LKF_NOQUEUE; | |
1506 | ret = dlm_lock_sync(cinfo->other_bitmap_lockres[i], DLM_LOCK_PW); | |
1507 | if (ret) | |
1508 | held = -1; | |
1509 | i++; | |
1510 | } | |
1511 | ||
1512 | return held; | |
1513 | } | |
1514 | ||
1515 | static void unlock_all_bitmaps(struct mddev *mddev) | |
1516 | { | |
1517 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1518 | int i; | |
1519 | ||
1520 | /* release other node's bitmap lock if they are existed */ | |
1521 | if (cinfo->other_bitmap_lockres) { | |
1522 | for (i = 0; i < mddev->bitmap_info.nodes - 1; i++) { | |
1523 | if (cinfo->other_bitmap_lockres[i]) { | |
f6a2dc64 GJ |
1524 | lockres_free(cinfo->other_bitmap_lockres[i]); |
1525 | } | |
1526 | } | |
1527 | kfree(cinfo->other_bitmap_lockres); | |
60f80d6f | 1528 | cinfo->other_bitmap_lockres = NULL; |
f6a2dc64 GJ |
1529 | } |
1530 | } | |
1531 | ||
97f6cd39 GR |
1532 | static int gather_bitmaps(struct md_rdev *rdev) |
1533 | { | |
1534 | int sn, err; | |
1535 | sector_t lo, hi; | |
aee177ac | 1536 | struct cluster_msg cmsg = {0}; |
97f6cd39 GR |
1537 | struct mddev *mddev = rdev->mddev; |
1538 | struct md_cluster_info *cinfo = mddev->cluster_info; | |
1539 | ||
faeff83f GJ |
1540 | cmsg.type = cpu_to_le32(RE_ADD); |
1541 | cmsg.raid_slot = cpu_to_le32(rdev->desc_nr); | |
0ba95977 | 1542 | err = sendmsg(cinfo, &cmsg, 1); |
97f6cd39 GR |
1543 | if (err) |
1544 | goto out; | |
1545 | ||
1546 | for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) { | |
1547 | if (sn == (cinfo->slot_number - 1)) | |
1548 | continue; | |
e64e4018 | 1549 | err = md_bitmap_copy_from_slot(mddev, sn, &lo, &hi, false); |
97f6cd39 GR |
1550 | if (err) { |
1551 | pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn); | |
1552 | goto out; | |
1553 | } | |
1554 | if ((hi > 0) && (lo < mddev->recovery_cp)) | |
1555 | mddev->recovery_cp = lo; | |
1556 | } | |
1557 | out: | |
1558 | return err; | |
1559 | } | |
1560 | ||
edb39c9d GR |
1561 | static struct md_cluster_operations cluster_ops = { |
1562 | .join = join, | |
1563 | .leave = leave, | |
cf921cc1 | 1564 | .slot_number = slot_number, |
c186b128 GR |
1565 | .resync_start = resync_start, |
1566 | .resync_finish = resync_finish, | |
96ae923a | 1567 | .resync_info_update = resync_info_update, |
5ebaf80b | 1568 | .resync_info_get = resync_info_get, |
293467aa GR |
1569 | .metadata_update_start = metadata_update_start, |
1570 | .metadata_update_finish = metadata_update_finish, | |
1571 | .metadata_update_cancel = metadata_update_cancel, | |
589a1c49 | 1572 | .area_resyncing = area_resyncing, |
dbb64f86 GR |
1573 | .add_new_disk = add_new_disk, |
1574 | .add_new_disk_cancel = add_new_disk_cancel, | |
1aee41f6 | 1575 | .new_disk_ack = new_disk_ack, |
88bcfef7 | 1576 | .remove_disk = remove_disk, |
51e453ae | 1577 | .load_bitmaps = load_bitmaps, |
97f6cd39 | 1578 | .gather_bitmaps = gather_bitmaps, |
afd75628 | 1579 | .resize_bitmaps = resize_bitmaps, |
f6a2dc64 GJ |
1580 | .lock_all_bitmaps = lock_all_bitmaps, |
1581 | .unlock_all_bitmaps = unlock_all_bitmaps, | |
818da59f | 1582 | .update_size = update_size, |
edb39c9d GR |
1583 | }; |
1584 | ||
8e854e9c GR |
1585 | static int __init cluster_init(void) |
1586 | { | |
f0e230ad | 1587 | pr_warn("md-cluster: support raid1 and raid10 (limited support)\n"); |
8e854e9c | 1588 | pr_info("Registering Cluster MD functions\n"); |
edb39c9d | 1589 | register_md_cluster_operations(&cluster_ops, THIS_MODULE); |
8e854e9c GR |
1590 | return 0; |
1591 | } | |
1592 | ||
1593 | static void cluster_exit(void) | |
1594 | { | |
edb39c9d | 1595 | unregister_md_cluster_operations(); |
8e854e9c GR |
1596 | } |
1597 | ||
1598 | module_init(cluster_init); | |
1599 | module_exit(cluster_exit); | |
86b57277 | 1600 | MODULE_AUTHOR("SUSE"); |
8e854e9c GR |
1601 | MODULE_LICENSE("GPL"); |
1602 | MODULE_DESCRIPTION("Clustering support for MD"); |