]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | md.c : Multiple Devices driver for Linux | |
f72ffdd6 | 3 | Copyright (C) 1998, 1999, 2000 Ingo Molnar |
1da177e4 LT |
4 | |
5 | completely rewritten, based on the MD driver code from Marc Zyngier | |
6 | ||
7 | Changes: | |
8 | ||
9 | - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar | |
10 | - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com> | |
11 | - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net> | |
12 | - kerneld support by Boris Tobotras <boris@xtalk.msk.su> | |
13 | - kmod support by: Cyrus Durgin | |
14 | - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com> | |
15 | - Devfs support by Richard Gooch <rgooch@atnf.csiro.au> | |
16 | ||
17 | - lots of fixes and improvements to the RAID1/RAID5 and generic | |
18 | RAID code (such as request based resynchronization): | |
19 | ||
20 | Neil Brown <neilb@cse.unsw.edu.au>. | |
21 | ||
32a7627c N |
22 | - persistent bitmap code |
23 | Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc. | |
24 | ||
1da177e4 LT |
25 | This program is free software; you can redistribute it and/or modify |
26 | it under the terms of the GNU General Public License as published by | |
27 | the Free Software Foundation; either version 2, or (at your option) | |
28 | any later version. | |
29 | ||
30 | You should have received a copy of the GNU General Public License | |
31 | (for example /usr/src/linux/COPYING); if not, write to the Free | |
32 | Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
33 | */ | |
34 | ||
a6fb0934 | 35 | #include <linux/kthread.h> |
bff61975 | 36 | #include <linux/blkdev.h> |
1da177e4 | 37 | #include <linux/sysctl.h> |
bff61975 | 38 | #include <linux/seq_file.h> |
ff01bb48 | 39 | #include <linux/fs.h> |
d7603b7e | 40 | #include <linux/poll.h> |
16f17b39 | 41 | #include <linux/ctype.h> |
e7d2860b | 42 | #include <linux/string.h> |
fb4d8c76 N |
43 | #include <linux/hdreg.h> |
44 | #include <linux/proc_fs.h> | |
45 | #include <linux/random.h> | |
056075c7 | 46 | #include <linux/module.h> |
fb4d8c76 | 47 | #include <linux/reboot.h> |
32a7627c | 48 | #include <linux/file.h> |
aa98aa31 | 49 | #include <linux/compat.h> |
25570727 | 50 | #include <linux/delay.h> |
bff61975 N |
51 | #include <linux/raid/md_p.h> |
52 | #include <linux/raid/md_u.h> | |
5a0e3ad6 | 53 | #include <linux/slab.h> |
43b2e5d8 | 54 | #include "md.h" |
ef740c37 | 55 | #include "bitmap.h" |
edb39c9d | 56 | #include "md-cluster.h" |
1da177e4 | 57 | |
1da177e4 | 58 | #ifndef MODULE |
d710e138 | 59 | static void autostart_arrays(int part); |
1da177e4 LT |
60 | #endif |
61 | ||
01f96c0a N |
62 | /* pers_list is a list of registered personalities protected |
63 | * by pers_lock. | |
64 | * pers_lock does extra service to protect accesses to | |
65 | * mddev->thread when the mutex cannot be held. | |
66 | */ | |
2604b703 | 67 | static LIST_HEAD(pers_list); |
1da177e4 LT |
68 | static DEFINE_SPINLOCK(pers_lock); |
69 | ||
edb39c9d | 70 | struct md_cluster_operations *md_cluster_ops; |
589a1c49 | 71 | EXPORT_SYMBOL(md_cluster_ops); |
edb39c9d GR |
72 | struct module *md_cluster_mod; |
73 | EXPORT_SYMBOL(md_cluster_mod); | |
74 | ||
90b08710 | 75 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
e804ac78 TH |
76 | static struct workqueue_struct *md_wq; |
77 | static struct workqueue_struct *md_misc_wq; | |
90b08710 | 78 | |
746d3207 N |
79 | static int remove_and_add_spares(struct mddev *mddev, |
80 | struct md_rdev *this); | |
5aa61f42 | 81 | static void mddev_detach(struct mddev *mddev); |
746d3207 | 82 | |
1e50915f RB |
83 | /* |
84 | * Default number of read corrections we'll attempt on an rdev | |
85 | * before ejecting it from the array. We divide the read error | |
86 | * count by 2 for every hour elapsed between read errors. | |
87 | */ | |
88 | #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 | |
1da177e4 LT |
89 | /* |
90 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
91 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
92 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 93 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
94 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
95 | * speed limit - in case reconstruction slows down your system despite | |
96 | * idle IO detection. | |
97 | * | |
98 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 99 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
100 | */ |
101 | ||
102 | static int sysctl_speed_limit_min = 1000; | |
103 | static int sysctl_speed_limit_max = 200000; | |
fd01b88c | 104 | static inline int speed_min(struct mddev *mddev) |
88202a0c N |
105 | { |
106 | return mddev->sync_speed_min ? | |
107 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
108 | } | |
109 | ||
fd01b88c | 110 | static inline int speed_max(struct mddev *mddev) |
88202a0c N |
111 | { |
112 | return mddev->sync_speed_max ? | |
113 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
114 | } | |
1da177e4 LT |
115 | |
116 | static struct ctl_table_header *raid_table_header; | |
117 | ||
82592c38 | 118 | static struct ctl_table raid_table[] = { |
1da177e4 | 119 | { |
1da177e4 LT |
120 | .procname = "speed_limit_min", |
121 | .data = &sysctl_speed_limit_min, | |
122 | .maxlen = sizeof(int), | |
80ca3a44 | 123 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 124 | .proc_handler = proc_dointvec, |
1da177e4 LT |
125 | }, |
126 | { | |
1da177e4 LT |
127 | .procname = "speed_limit_max", |
128 | .data = &sysctl_speed_limit_max, | |
129 | .maxlen = sizeof(int), | |
80ca3a44 | 130 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 131 | .proc_handler = proc_dointvec, |
1da177e4 | 132 | }, |
894d2491 | 133 | { } |
1da177e4 LT |
134 | }; |
135 | ||
82592c38 | 136 | static struct ctl_table raid_dir_table[] = { |
1da177e4 | 137 | { |
1da177e4 LT |
138 | .procname = "raid", |
139 | .maxlen = 0, | |
80ca3a44 | 140 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
141 | .child = raid_table, |
142 | }, | |
894d2491 | 143 | { } |
1da177e4 LT |
144 | }; |
145 | ||
82592c38 | 146 | static struct ctl_table raid_root_table[] = { |
1da177e4 | 147 | { |
1da177e4 LT |
148 | .procname = "dev", |
149 | .maxlen = 0, | |
150 | .mode = 0555, | |
151 | .child = raid_dir_table, | |
152 | }, | |
894d2491 | 153 | { } |
1da177e4 LT |
154 | }; |
155 | ||
83d5cde4 | 156 | static const struct block_device_operations md_fops; |
1da177e4 | 157 | |
f91de92e N |
158 | static int start_readonly; |
159 | ||
a167f663 N |
160 | /* bio_clone_mddev |
161 | * like bio_clone, but with a local bio set | |
162 | */ | |
163 | ||
a167f663 | 164 | struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, |
fd01b88c | 165 | struct mddev *mddev) |
a167f663 N |
166 | { |
167 | struct bio *b; | |
a167f663 N |
168 | |
169 | if (!mddev || !mddev->bio_set) | |
170 | return bio_alloc(gfp_mask, nr_iovecs); | |
171 | ||
395c72a7 | 172 | b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set); |
a167f663 N |
173 | if (!b) |
174 | return NULL; | |
a167f663 N |
175 | return b; |
176 | } | |
177 | EXPORT_SYMBOL_GPL(bio_alloc_mddev); | |
178 | ||
179 | struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask, | |
fd01b88c | 180 | struct mddev *mddev) |
a167f663 | 181 | { |
a167f663 N |
182 | if (!mddev || !mddev->bio_set) |
183 | return bio_clone(bio, gfp_mask); | |
184 | ||
bf800ef1 | 185 | return bio_clone_bioset(bio, gfp_mask, mddev->bio_set); |
a167f663 N |
186 | } |
187 | EXPORT_SYMBOL_GPL(bio_clone_mddev); | |
188 | ||
d7603b7e N |
189 | /* |
190 | * We have a system wide 'event count' that is incremented | |
191 | * on any 'interesting' event, and readers of /proc/mdstat | |
192 | * can use 'poll' or 'select' to find out when the event | |
193 | * count increases. | |
194 | * | |
195 | * Events are: | |
196 | * start array, stop array, error, add device, remove device, | |
197 | * start build, activate spare | |
198 | */ | |
2989ddbd | 199 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 200 | static atomic_t md_event_count; |
fd01b88c | 201 | void md_new_event(struct mddev *mddev) |
d7603b7e N |
202 | { |
203 | atomic_inc(&md_event_count); | |
204 | wake_up(&md_event_waiters); | |
205 | } | |
29269553 | 206 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 207 | |
c331eb04 N |
208 | /* Alternate version that can be called from interrupts |
209 | * when calling sysfs_notify isn't needed. | |
210 | */ | |
fd01b88c | 211 | static void md_new_event_inintr(struct mddev *mddev) |
c331eb04 N |
212 | { |
213 | atomic_inc(&md_event_count); | |
214 | wake_up(&md_event_waiters); | |
215 | } | |
216 | ||
1da177e4 LT |
217 | /* |
218 | * Enables to iterate over all existing md arrays | |
219 | * all_mddevs_lock protects this list. | |
220 | */ | |
221 | static LIST_HEAD(all_mddevs); | |
222 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
223 | ||
1da177e4 LT |
224 | /* |
225 | * iterates through all used mddevs in the system. | |
226 | * We take care to grab the all_mddevs_lock whenever navigating | |
227 | * the list, and to always hold a refcount when unlocked. | |
228 | * Any code which breaks out of this loop while own | |
229 | * a reference to the current mddev and must mddev_put it. | |
230 | */ | |
fd01b88c | 231 | #define for_each_mddev(_mddev,_tmp) \ |
1da177e4 | 232 | \ |
f72ffdd6 | 233 | for (({ spin_lock(&all_mddevs_lock); \ |
fd01b88c N |
234 | _tmp = all_mddevs.next; \ |
235 | _mddev = NULL;}); \ | |
236 | ({ if (_tmp != &all_mddevs) \ | |
237 | mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\ | |
1da177e4 | 238 | spin_unlock(&all_mddevs_lock); \ |
fd01b88c N |
239 | if (_mddev) mddev_put(_mddev); \ |
240 | _mddev = list_entry(_tmp, struct mddev, all_mddevs); \ | |
241 | _tmp != &all_mddevs;}); \ | |
1da177e4 | 242 | ({ spin_lock(&all_mddevs_lock); \ |
fd01b88c | 243 | _tmp = _tmp->next;}) \ |
1da177e4 LT |
244 | ) |
245 | ||
409c57f3 N |
246 | /* Rather than calling directly into the personality make_request function, |
247 | * IO requests come here first so that we can check if the device is | |
248 | * being suspended pending a reconfiguration. | |
249 | * We hold a refcount over the call to ->make_request. By the time that | |
250 | * call has finished, the bio has been linked into some internal structure | |
251 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
252 | */ | |
5a7bbad2 | 253 | static void md_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 254 | { |
49077326 | 255 | const int rw = bio_data_dir(bio); |
fd01b88c | 256 | struct mddev *mddev = q->queuedata; |
e91ece55 | 257 | unsigned int sectors; |
74672d06 | 258 | int cpu; |
49077326 | 259 | |
54efd50b KO |
260 | blk_queue_split(q, &bio, q->bio_split); |
261 | ||
0ca69886 N |
262 | if (mddev == NULL || mddev->pers == NULL |
263 | || !mddev->ready) { | |
409c57f3 | 264 | bio_io_error(bio); |
5a7bbad2 | 265 | return; |
409c57f3 | 266 | } |
bbfa57c0 | 267 | if (mddev->ro == 1 && unlikely(rw == WRITE)) { |
4246a0b6 CH |
268 | if (bio_sectors(bio) != 0) |
269 | bio->bi_error = -EROFS; | |
270 | bio_endio(bio); | |
bbfa57c0 SR |
271 | return; |
272 | } | |
0ca69886 | 273 | smp_rmb(); /* Ensure implications of 'active' are visible */ |
409c57f3 | 274 | rcu_read_lock(); |
e9c7469b | 275 | if (mddev->suspended) { |
409c57f3 N |
276 | DEFINE_WAIT(__wait); |
277 | for (;;) { | |
278 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
279 | TASK_UNINTERRUPTIBLE); | |
e9c7469b | 280 | if (!mddev->suspended) |
409c57f3 N |
281 | break; |
282 | rcu_read_unlock(); | |
283 | schedule(); | |
284 | rcu_read_lock(); | |
285 | } | |
286 | finish_wait(&mddev->sb_wait, &__wait); | |
287 | } | |
288 | atomic_inc(&mddev->active_io); | |
289 | rcu_read_unlock(); | |
49077326 | 290 | |
e91ece55 CM |
291 | /* |
292 | * save the sectors now since our bio can | |
293 | * go away inside make_request | |
294 | */ | |
295 | sectors = bio_sectors(bio); | |
5a7bbad2 | 296 | mddev->pers->make_request(mddev, bio); |
49077326 | 297 | |
74672d06 GZ |
298 | cpu = part_stat_lock(); |
299 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
300 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); | |
301 | part_stat_unlock(); | |
49077326 | 302 | |
409c57f3 N |
303 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
304 | wake_up(&mddev->sb_wait); | |
409c57f3 N |
305 | } |
306 | ||
9e35b99c N |
307 | /* mddev_suspend makes sure no new requests are submitted |
308 | * to the device, and that any requests that have been submitted | |
309 | * are completely handled. | |
afa0f557 N |
310 | * Once mddev_detach() is called and completes, the module will be |
311 | * completely unused. | |
9e35b99c | 312 | */ |
fd01b88c | 313 | void mddev_suspend(struct mddev *mddev) |
409c57f3 N |
314 | { |
315 | BUG_ON(mddev->suspended); | |
316 | mddev->suspended = 1; | |
317 | synchronize_rcu(); | |
318 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
319 | mddev->pers->quiesce(mddev, 1); | |
0d9f4f13 JB |
320 | |
321 | del_timer_sync(&mddev->safemode_timer); | |
409c57f3 | 322 | } |
390ee602 | 323 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 324 | |
fd01b88c | 325 | void mddev_resume(struct mddev *mddev) |
409c57f3 N |
326 | { |
327 | mddev->suspended = 0; | |
328 | wake_up(&mddev->sb_wait); | |
329 | mddev->pers->quiesce(mddev, 0); | |
0fd018af | 330 | |
47525e59 | 331 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
0fd018af JB |
332 | md_wakeup_thread(mddev->thread); |
333 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 334 | } |
390ee602 | 335 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 336 | |
fd01b88c | 337 | int mddev_congested(struct mddev *mddev, int bits) |
3fa841d7 | 338 | { |
5c675f83 N |
339 | struct md_personality *pers = mddev->pers; |
340 | int ret = 0; | |
341 | ||
342 | rcu_read_lock(); | |
343 | if (mddev->suspended) | |
344 | ret = 1; | |
345 | else if (pers && pers->congested) | |
346 | ret = pers->congested(mddev, bits); | |
347 | rcu_read_unlock(); | |
348 | return ret; | |
349 | } | |
350 | EXPORT_SYMBOL_GPL(mddev_congested); | |
351 | static int md_congested(void *data, int bits) | |
352 | { | |
353 | struct mddev *mddev = data; | |
354 | return mddev_congested(mddev, bits); | |
3fa841d7 | 355 | } |
3fa841d7 | 356 | |
a2826aa9 | 357 | /* |
e9c7469b | 358 | * Generic flush handling for md |
a2826aa9 N |
359 | */ |
360 | ||
4246a0b6 | 361 | static void md_end_flush(struct bio *bio) |
a2826aa9 | 362 | { |
3cb03002 | 363 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 364 | struct mddev *mddev = rdev->mddev; |
a2826aa9 N |
365 | |
366 | rdev_dec_pending(rdev, mddev); | |
367 | ||
368 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 369 | /* The pre-request flush has finished */ |
e804ac78 | 370 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
371 | } |
372 | bio_put(bio); | |
373 | } | |
374 | ||
a7a07e69 N |
375 | static void md_submit_flush_data(struct work_struct *ws); |
376 | ||
a035fc3e | 377 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 378 | { |
fd01b88c | 379 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
3cb03002 | 380 | struct md_rdev *rdev; |
a2826aa9 | 381 | |
a7a07e69 N |
382 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
383 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 | 384 | rcu_read_lock(); |
dafb20fa | 385 | rdev_for_each_rcu(rdev, mddev) |
a2826aa9 N |
386 | if (rdev->raid_disk >= 0 && |
387 | !test_bit(Faulty, &rdev->flags)) { | |
388 | /* Take two references, one is dropped | |
389 | * when request finishes, one after | |
390 | * we reclaim rcu_read_lock | |
391 | */ | |
392 | struct bio *bi; | |
393 | atomic_inc(&rdev->nr_pending); | |
394 | atomic_inc(&rdev->nr_pending); | |
395 | rcu_read_unlock(); | |
b5e1b8ce | 396 | bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); |
e9c7469b | 397 | bi->bi_end_io = md_end_flush; |
a2826aa9 N |
398 | bi->bi_private = rdev; |
399 | bi->bi_bdev = rdev->bdev; | |
400 | atomic_inc(&mddev->flush_pending); | |
e9c7469b | 401 | submit_bio(WRITE_FLUSH, bi); |
a2826aa9 N |
402 | rcu_read_lock(); |
403 | rdev_dec_pending(rdev, mddev); | |
404 | } | |
405 | rcu_read_unlock(); | |
a7a07e69 N |
406 | if (atomic_dec_and_test(&mddev->flush_pending)) |
407 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
408 | } |
409 | ||
e9c7469b | 410 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 411 | { |
fd01b88c | 412 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
e9c7469b | 413 | struct bio *bio = mddev->flush_bio; |
a2826aa9 | 414 | |
4f024f37 | 415 | if (bio->bi_iter.bi_size == 0) |
a2826aa9 | 416 | /* an empty barrier - all done */ |
4246a0b6 | 417 | bio_endio(bio); |
a2826aa9 | 418 | else { |
e9c7469b | 419 | bio->bi_rw &= ~REQ_FLUSH; |
5a7bbad2 | 420 | mddev->pers->make_request(mddev, bio); |
a2826aa9 | 421 | } |
2b74e12e N |
422 | |
423 | mddev->flush_bio = NULL; | |
424 | wake_up(&mddev->sb_wait); | |
a2826aa9 N |
425 | } |
426 | ||
fd01b88c | 427 | void md_flush_request(struct mddev *mddev, struct bio *bio) |
a2826aa9 | 428 | { |
85572d7c | 429 | spin_lock_irq(&mddev->lock); |
a2826aa9 | 430 | wait_event_lock_irq(mddev->sb_wait, |
e9c7469b | 431 | !mddev->flush_bio, |
85572d7c | 432 | mddev->lock); |
e9c7469b | 433 | mddev->flush_bio = bio; |
85572d7c | 434 | spin_unlock_irq(&mddev->lock); |
a2826aa9 | 435 | |
a035fc3e N |
436 | INIT_WORK(&mddev->flush_work, submit_flushes); |
437 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 438 | } |
e9c7469b | 439 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 440 | |
74018dc3 | 441 | void md_unplug(struct blk_plug_cb *cb, bool from_schedule) |
97658cdd | 442 | { |
9cbb1750 N |
443 | struct mddev *mddev = cb->data; |
444 | md_wakeup_thread(mddev->thread); | |
445 | kfree(cb); | |
97658cdd | 446 | } |
9cbb1750 | 447 | EXPORT_SYMBOL(md_unplug); |
2ac87401 | 448 | |
fd01b88c | 449 | static inline struct mddev *mddev_get(struct mddev *mddev) |
1da177e4 LT |
450 | { |
451 | atomic_inc(&mddev->active); | |
452 | return mddev; | |
453 | } | |
454 | ||
5fd3a17e | 455 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 456 | |
fd01b88c | 457 | static void mddev_put(struct mddev *mddev) |
1da177e4 | 458 | { |
a167f663 N |
459 | struct bio_set *bs = NULL; |
460 | ||
1da177e4 LT |
461 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
462 | return; | |
d3374825 | 463 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
464 | mddev->ctime == 0 && !mddev->hold_active) { |
465 | /* Array is not configured at all, and not held active, | |
466 | * so destroy it */ | |
af8a2434 | 467 | list_del_init(&mddev->all_mddevs); |
a167f663 N |
468 | bs = mddev->bio_set; |
469 | mddev->bio_set = NULL; | |
d3374825 | 470 | if (mddev->gendisk) { |
e804ac78 TH |
471 | /* We did a probe so need to clean up. Call |
472 | * queue_work inside the spinlock so that | |
473 | * flush_workqueue() after mddev_find will | |
474 | * succeed in waiting for the work to be done. | |
d3374825 N |
475 | */ |
476 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 477 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
478 | } else |
479 | kfree(mddev); | |
480 | } | |
481 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
482 | if (bs) |
483 | bioset_free(bs); | |
1da177e4 LT |
484 | } |
485 | ||
25b2edfa SL |
486 | static void md_safemode_timeout(unsigned long data); |
487 | ||
fd01b88c | 488 | void mddev_init(struct mddev *mddev) |
fafd7fb0 N |
489 | { |
490 | mutex_init(&mddev->open_mutex); | |
491 | mutex_init(&mddev->reconfig_mutex); | |
492 | mutex_init(&mddev->bitmap_info.mutex); | |
493 | INIT_LIST_HEAD(&mddev->disks); | |
494 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
25b2edfa SL |
495 | setup_timer(&mddev->safemode_timer, md_safemode_timeout, |
496 | (unsigned long) mddev); | |
fafd7fb0 N |
497 | atomic_set(&mddev->active, 1); |
498 | atomic_set(&mddev->openers, 0); | |
499 | atomic_set(&mddev->active_io, 0); | |
85572d7c | 500 | spin_lock_init(&mddev->lock); |
fafd7fb0 N |
501 | atomic_set(&mddev->flush_pending, 0); |
502 | init_waitqueue_head(&mddev->sb_wait); | |
503 | init_waitqueue_head(&mddev->recovery_wait); | |
504 | mddev->reshape_position = MaxSector; | |
2c810cdd | 505 | mddev->reshape_backwards = 0; |
c4a39551 | 506 | mddev->last_sync_action = "none"; |
fafd7fb0 N |
507 | mddev->resync_min = 0; |
508 | mddev->resync_max = MaxSector; | |
509 | mddev->level = LEVEL_NONE; | |
510 | } | |
390ee602 | 511 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 512 | |
f72ffdd6 | 513 | static struct mddev *mddev_find(dev_t unit) |
1da177e4 | 514 | { |
fd01b88c | 515 | struct mddev *mddev, *new = NULL; |
1da177e4 | 516 | |
8f5f02c4 N |
517 | if (unit && MAJOR(unit) != MD_MAJOR) |
518 | unit &= ~((1<<MdpMinorShift)-1); | |
519 | ||
1da177e4 LT |
520 | retry: |
521 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
522 | |
523 | if (unit) { | |
524 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
525 | if (mddev->unit == unit) { | |
526 | mddev_get(mddev); | |
527 | spin_unlock(&all_mddevs_lock); | |
528 | kfree(new); | |
529 | return mddev; | |
530 | } | |
531 | ||
532 | if (new) { | |
533 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 534 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
535 | new->hold_active = UNTIL_IOCTL; |
536 | return new; | |
1da177e4 | 537 | } |
efeb53c0 N |
538 | } else if (new) { |
539 | /* find an unused unit number */ | |
540 | static int next_minor = 512; | |
541 | int start = next_minor; | |
542 | int is_free = 0; | |
543 | int dev = 0; | |
544 | while (!is_free) { | |
545 | dev = MKDEV(MD_MAJOR, next_minor); | |
546 | next_minor++; | |
547 | if (next_minor > MINORMASK) | |
548 | next_minor = 0; | |
549 | if (next_minor == start) { | |
550 | /* Oh dear, all in use. */ | |
551 | spin_unlock(&all_mddevs_lock); | |
552 | kfree(new); | |
553 | return NULL; | |
554 | } | |
f72ffdd6 | 555 | |
efeb53c0 N |
556 | is_free = 1; |
557 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
558 | if (mddev->unit == dev) { | |
559 | is_free = 0; | |
560 | break; | |
561 | } | |
562 | } | |
563 | new->unit = dev; | |
564 | new->md_minor = MINOR(dev); | |
565 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
566 | list_add(&new->all_mddevs, &all_mddevs); |
567 | spin_unlock(&all_mddevs_lock); | |
568 | return new; | |
569 | } | |
570 | spin_unlock(&all_mddevs_lock); | |
571 | ||
9ffae0cf | 572 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
573 | if (!new) |
574 | return NULL; | |
575 | ||
1da177e4 LT |
576 | new->unit = unit; |
577 | if (MAJOR(unit) == MD_MAJOR) | |
578 | new->md_minor = MINOR(unit); | |
579 | else | |
580 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
581 | ||
fafd7fb0 | 582 | mddev_init(new); |
1da177e4 | 583 | |
1da177e4 LT |
584 | goto retry; |
585 | } | |
586 | ||
b6eb127d N |
587 | static struct attribute_group md_redundancy_group; |
588 | ||
5c47daf6 | 589 | void mddev_unlock(struct mddev *mddev) |
1da177e4 | 590 | { |
a64c876f | 591 | if (mddev->to_remove) { |
b6eb127d N |
592 | /* These cannot be removed under reconfig_mutex as |
593 | * an access to the files will try to take reconfig_mutex | |
594 | * while holding the file unremovable, which leads to | |
595 | * a deadlock. | |
bb4f1e9d N |
596 | * So hold set sysfs_active while the remove in happeing, |
597 | * and anything else which might set ->to_remove or my | |
598 | * otherwise change the sysfs namespace will fail with | |
599 | * -EBUSY if sysfs_active is still set. | |
600 | * We set sysfs_active under reconfig_mutex and elsewhere | |
601 | * test it under the same mutex to ensure its correct value | |
602 | * is seen. | |
b6eb127d | 603 | */ |
a64c876f N |
604 | struct attribute_group *to_remove = mddev->to_remove; |
605 | mddev->to_remove = NULL; | |
bb4f1e9d | 606 | mddev->sysfs_active = 1; |
b6eb127d N |
607 | mutex_unlock(&mddev->reconfig_mutex); |
608 | ||
00bcb4ac N |
609 | if (mddev->kobj.sd) { |
610 | if (to_remove != &md_redundancy_group) | |
611 | sysfs_remove_group(&mddev->kobj, to_remove); | |
612 | if (mddev->pers == NULL || | |
613 | mddev->pers->sync_request == NULL) { | |
614 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
615 | if (mddev->sysfs_action) | |
616 | sysfs_put(mddev->sysfs_action); | |
617 | mddev->sysfs_action = NULL; | |
618 | } | |
a64c876f | 619 | } |
bb4f1e9d | 620 | mddev->sysfs_active = 0; |
b6eb127d N |
621 | } else |
622 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 623 | |
751e67ca CD |
624 | /* As we've dropped the mutex we need a spinlock to |
625 | * make sure the thread doesn't disappear | |
01f96c0a N |
626 | */ |
627 | spin_lock(&pers_lock); | |
005eca5e | 628 | md_wakeup_thread(mddev->thread); |
01f96c0a | 629 | spin_unlock(&pers_lock); |
1da177e4 | 630 | } |
5c47daf6 | 631 | EXPORT_SYMBOL_GPL(mddev_unlock); |
1da177e4 | 632 | |
57d051dc | 633 | struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr) |
1ca69c4b N |
634 | { |
635 | struct md_rdev *rdev; | |
636 | ||
637 | rdev_for_each_rcu(rdev, mddev) | |
638 | if (rdev->desc_nr == nr) | |
639 | return rdev; | |
640 | ||
641 | return NULL; | |
642 | } | |
57d051dc | 643 | EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu); |
1ca69c4b N |
644 | |
645 | static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev) | |
1da177e4 | 646 | { |
3cb03002 | 647 | struct md_rdev *rdev; |
1da177e4 | 648 | |
dafb20fa | 649 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
650 | if (rdev->bdev->bd_dev == dev) |
651 | return rdev; | |
159ec1fc | 652 | |
1da177e4 LT |
653 | return NULL; |
654 | } | |
655 | ||
1ca69c4b N |
656 | static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev) |
657 | { | |
658 | struct md_rdev *rdev; | |
659 | ||
660 | rdev_for_each_rcu(rdev, mddev) | |
661 | if (rdev->bdev->bd_dev == dev) | |
662 | return rdev; | |
663 | ||
664 | return NULL; | |
665 | } | |
666 | ||
84fc4b56 | 667 | static struct md_personality *find_pers(int level, char *clevel) |
2604b703 | 668 | { |
84fc4b56 | 669 | struct md_personality *pers; |
d9d166c2 N |
670 | list_for_each_entry(pers, &pers_list, list) { |
671 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 672 | return pers; |
d9d166c2 N |
673 | if (strcmp(pers->name, clevel)==0) |
674 | return pers; | |
675 | } | |
2604b703 N |
676 | return NULL; |
677 | } | |
678 | ||
b73df2d3 | 679 | /* return the offset of the super block in 512byte sectors */ |
3cb03002 | 680 | static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) |
1da177e4 | 681 | { |
57b2caa3 | 682 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 683 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
684 | } |
685 | ||
f72ffdd6 | 686 | static int alloc_disk_sb(struct md_rdev *rdev) |
1da177e4 | 687 | { |
1da177e4 LT |
688 | rdev->sb_page = alloc_page(GFP_KERNEL); |
689 | if (!rdev->sb_page) { | |
690 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 691 | return -ENOMEM; |
1da177e4 LT |
692 | } |
693 | ||
694 | return 0; | |
695 | } | |
696 | ||
545c8795 | 697 | void md_rdev_clear(struct md_rdev *rdev) |
1da177e4 LT |
698 | { |
699 | if (rdev->sb_page) { | |
2d1f3b5d | 700 | put_page(rdev->sb_page); |
1da177e4 LT |
701 | rdev->sb_loaded = 0; |
702 | rdev->sb_page = NULL; | |
0f420358 | 703 | rdev->sb_start = 0; |
dd8ac336 | 704 | rdev->sectors = 0; |
1da177e4 | 705 | } |
2699b672 N |
706 | if (rdev->bb_page) { |
707 | put_page(rdev->bb_page); | |
708 | rdev->bb_page = NULL; | |
709 | } | |
4fa2f327 N |
710 | kfree(rdev->badblocks.page); |
711 | rdev->badblocks.page = NULL; | |
1da177e4 | 712 | } |
545c8795 | 713 | EXPORT_SYMBOL_GPL(md_rdev_clear); |
1da177e4 | 714 | |
4246a0b6 | 715 | static void super_written(struct bio *bio) |
7bfa19f2 | 716 | { |
3cb03002 | 717 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 718 | struct mddev *mddev = rdev->mddev; |
7bfa19f2 | 719 | |
4246a0b6 CH |
720 | if (bio->bi_error) { |
721 | printk("md: super_written gets error=%d\n", bio->bi_error); | |
a9701a30 | 722 | md_error(mddev, rdev); |
3a0f5bbb | 723 | } |
7bfa19f2 | 724 | |
a9701a30 N |
725 | if (atomic_dec_and_test(&mddev->pending_writes)) |
726 | wake_up(&mddev->sb_wait); | |
f8b58edf | 727 | bio_put(bio); |
7bfa19f2 N |
728 | } |
729 | ||
fd01b88c | 730 | void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
7bfa19f2 N |
731 | sector_t sector, int size, struct page *page) |
732 | { | |
733 | /* write first size bytes of page to sector of rdev | |
734 | * Increment mddev->pending_writes before returning | |
735 | * and decrement it on completion, waking up sb_wait | |
736 | * if zero is reached. | |
737 | * If an error occurred, call md_error | |
738 | */ | |
a167f663 | 739 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); |
7bfa19f2 | 740 | |
a6ff7e08 | 741 | bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; |
4f024f37 | 742 | bio->bi_iter.bi_sector = sector; |
7bfa19f2 N |
743 | bio_add_page(bio, page, size, 0); |
744 | bio->bi_private = rdev; | |
745 | bio->bi_end_io = super_written; | |
a9701a30 | 746 | |
7bfa19f2 | 747 | atomic_inc(&mddev->pending_writes); |
a5bf4df0 | 748 | submit_bio(WRITE_FLUSH_FUA, bio); |
a9701a30 N |
749 | } |
750 | ||
fd01b88c | 751 | void md_super_wait(struct mddev *mddev) |
a9701a30 | 752 | { |
e9c7469b | 753 | /* wait for all superblock writes that were scheduled to complete */ |
1967cd56 | 754 | wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); |
7bfa19f2 N |
755 | } |
756 | ||
3cb03002 | 757 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
ccebd4c4 | 758 | struct page *page, int rw, bool metadata_op) |
1da177e4 | 759 | { |
a167f663 | 760 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
761 | int ret; |
762 | ||
a6ff7e08 JB |
763 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
764 | rdev->meta_bdev : rdev->bdev; | |
ccebd4c4 | 765 | if (metadata_op) |
4f024f37 | 766 | bio->bi_iter.bi_sector = sector + rdev->sb_start; |
1fdd6fc9 N |
767 | else if (rdev->mddev->reshape_position != MaxSector && |
768 | (rdev->mddev->reshape_backwards == | |
769 | (sector >= rdev->mddev->reshape_position))) | |
4f024f37 | 770 | bio->bi_iter.bi_sector = sector + rdev->new_data_offset; |
ccebd4c4 | 771 | else |
4f024f37 | 772 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
1da177e4 | 773 | bio_add_page(bio, page, size, 0); |
c170bbb4 | 774 | submit_bio_wait(rw, bio); |
1da177e4 | 775 | |
4246a0b6 | 776 | ret = !bio->bi_error; |
1da177e4 LT |
777 | bio_put(bio); |
778 | return ret; | |
779 | } | |
a8745db2 | 780 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 781 | |
f72ffdd6 | 782 | static int read_disk_sb(struct md_rdev *rdev, int size) |
1da177e4 LT |
783 | { |
784 | char b[BDEVNAME_SIZE]; | |
403df478 | 785 | |
1da177e4 LT |
786 | if (rdev->sb_loaded) |
787 | return 0; | |
788 | ||
ccebd4c4 | 789 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) |
1da177e4 LT |
790 | goto fail; |
791 | rdev->sb_loaded = 1; | |
792 | return 0; | |
793 | ||
794 | fail: | |
795 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
796 | bdevname(rdev->bdev,b)); | |
797 | return -EINVAL; | |
798 | } | |
799 | ||
800 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
801 | { | |
f72ffdd6 | 802 | return sb1->set_uuid0 == sb2->set_uuid0 && |
05710466 AN |
803 | sb1->set_uuid1 == sb2->set_uuid1 && |
804 | sb1->set_uuid2 == sb2->set_uuid2 && | |
805 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
806 | } |
807 | ||
1da177e4 LT |
808 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
809 | { | |
810 | int ret; | |
811 | mdp_super_t *tmp1, *tmp2; | |
812 | ||
813 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
814 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
815 | ||
816 | if (!tmp1 || !tmp2) { | |
817 | ret = 0; | |
35020f1a | 818 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
819 | goto abort; |
820 | } | |
821 | ||
822 | *tmp1 = *sb1; | |
823 | *tmp2 = *sb2; | |
824 | ||
825 | /* | |
826 | * nr_disks is not constant | |
827 | */ | |
828 | tmp1->nr_disks = 0; | |
829 | tmp2->nr_disks = 0; | |
830 | ||
ce0c8e05 | 831 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 832 | abort: |
990a8baf JJ |
833 | kfree(tmp1); |
834 | kfree(tmp2); | |
1da177e4 LT |
835 | return ret; |
836 | } | |
837 | ||
4d167f09 N |
838 | static u32 md_csum_fold(u32 csum) |
839 | { | |
840 | csum = (csum & 0xffff) + (csum >> 16); | |
841 | return (csum & 0xffff) + (csum >> 16); | |
842 | } | |
843 | ||
f72ffdd6 | 844 | static unsigned int calc_sb_csum(mdp_super_t *sb) |
1da177e4 | 845 | { |
4d167f09 N |
846 | u64 newcsum = 0; |
847 | u32 *sb32 = (u32*)sb; | |
848 | int i; | |
1da177e4 LT |
849 | unsigned int disk_csum, csum; |
850 | ||
851 | disk_csum = sb->sb_csum; | |
852 | sb->sb_csum = 0; | |
4d167f09 N |
853 | |
854 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
855 | newcsum += sb32[i]; | |
856 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
857 | ||
4d167f09 N |
858 | #ifdef CONFIG_ALPHA |
859 | /* This used to use csum_partial, which was wrong for several | |
860 | * reasons including that different results are returned on | |
861 | * different architectures. It isn't critical that we get exactly | |
862 | * the same return value as before (we always csum_fold before | |
863 | * testing, and that removes any differences). However as we | |
864 | * know that csum_partial always returned a 16bit value on | |
865 | * alphas, do a fold to maximise conformity to previous behaviour. | |
866 | */ | |
867 | sb->sb_csum = md_csum_fold(disk_csum); | |
868 | #else | |
1da177e4 | 869 | sb->sb_csum = disk_csum; |
4d167f09 | 870 | #endif |
1da177e4 LT |
871 | return csum; |
872 | } | |
873 | ||
1da177e4 LT |
874 | /* |
875 | * Handle superblock details. | |
876 | * We want to be able to handle multiple superblock formats | |
877 | * so we have a common interface to them all, and an array of | |
878 | * different handlers. | |
879 | * We rely on user-space to write the initial superblock, and support | |
880 | * reading and updating of superblocks. | |
881 | * Interface methods are: | |
3cb03002 | 882 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
883 | * loads and validates a superblock on dev. |
884 | * if refdev != NULL, compare superblocks on both devices | |
885 | * Return: | |
886 | * 0 - dev has a superblock that is compatible with refdev | |
887 | * 1 - dev has a superblock that is compatible and newer than refdev | |
888 | * so dev should be used as the refdev in future | |
889 | * -EINVAL superblock incompatible or invalid | |
890 | * -othererror e.g. -EIO | |
891 | * | |
fd01b88c | 892 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
893 | * Verify that dev is acceptable into mddev. |
894 | * The first time, mddev->raid_disks will be 0, and data from | |
895 | * dev should be merged in. Subsequent calls check that dev | |
896 | * is new enough. Return 0 or -EINVAL | |
897 | * | |
fd01b88c | 898 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
899 | * Update the superblock for rdev with data in mddev |
900 | * This does not write to disc. | |
901 | * | |
902 | */ | |
903 | ||
904 | struct super_type { | |
0cd17fec CW |
905 | char *name; |
906 | struct module *owner; | |
c6563a8c N |
907 | int (*load_super)(struct md_rdev *rdev, |
908 | struct md_rdev *refdev, | |
0cd17fec | 909 | int minor_version); |
c6563a8c N |
910 | int (*validate_super)(struct mddev *mddev, |
911 | struct md_rdev *rdev); | |
912 | void (*sync_super)(struct mddev *mddev, | |
913 | struct md_rdev *rdev); | |
3cb03002 | 914 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 915 | sector_t num_sectors); |
c6563a8c N |
916 | int (*allow_new_offset)(struct md_rdev *rdev, |
917 | unsigned long long new_offset); | |
1da177e4 LT |
918 | }; |
919 | ||
0894cc30 AN |
920 | /* |
921 | * Check that the given mddev has no bitmap. | |
922 | * | |
923 | * This function is called from the run method of all personalities that do not | |
924 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
925 | * has a bitmap. Otherwise, it returns 0. | |
926 | * | |
927 | */ | |
fd01b88c | 928 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 929 | { |
c3d9714e | 930 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
931 | return 0; |
932 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
933 | mdname(mddev), mddev->pers->name); | |
934 | return 1; | |
935 | } | |
936 | EXPORT_SYMBOL(md_check_no_bitmap); | |
937 | ||
1da177e4 | 938 | /* |
f72ffdd6 | 939 | * load_super for 0.90.0 |
1da177e4 | 940 | */ |
3cb03002 | 941 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
942 | { |
943 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
944 | mdp_super_t *sb; | |
945 | int ret; | |
1da177e4 LT |
946 | |
947 | /* | |
0f420358 | 948 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
949 | * it's at the end of the disk. |
950 | * | |
951 | * It also happens to be a multiple of 4Kb. | |
952 | */ | |
57b2caa3 | 953 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 954 | |
0002b271 | 955 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
956 | if (ret) return ret; |
957 | ||
958 | ret = -EINVAL; | |
959 | ||
960 | bdevname(rdev->bdev, b); | |
65a06f06 | 961 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
962 | |
963 | if (sb->md_magic != MD_SB_MAGIC) { | |
964 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
965 | b); | |
966 | goto abort; | |
967 | } | |
968 | ||
969 | if (sb->major_version != 0 || | |
f6705578 N |
970 | sb->minor_version < 90 || |
971 | sb->minor_version > 91) { | |
1da177e4 LT |
972 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
973 | sb->major_version, sb->minor_version, | |
974 | b); | |
975 | goto abort; | |
976 | } | |
977 | ||
978 | if (sb->raid_disks <= 0) | |
979 | goto abort; | |
980 | ||
4d167f09 | 981 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
982 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
983 | b); | |
984 | goto abort; | |
985 | } | |
986 | ||
987 | rdev->preferred_minor = sb->md_minor; | |
988 | rdev->data_offset = 0; | |
c6563a8c | 989 | rdev->new_data_offset = 0; |
0002b271 | 990 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 991 | rdev->badblocks.shift = -1; |
1da177e4 LT |
992 | |
993 | if (sb->level == LEVEL_MULTIPATH) | |
994 | rdev->desc_nr = -1; | |
995 | else | |
996 | rdev->desc_nr = sb->this_disk.number; | |
997 | ||
9a7b2b0f | 998 | if (!refdev) { |
1da177e4 | 999 | ret = 1; |
9a7b2b0f | 1000 | } else { |
1da177e4 | 1001 | __u64 ev1, ev2; |
65a06f06 | 1002 | mdp_super_t *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1003 | if (!uuid_equal(refsb, sb)) { |
1004 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
1005 | b, bdevname(refdev->bdev,b2)); | |
1006 | goto abort; | |
1007 | } | |
1008 | if (!sb_equal(refsb, sb)) { | |
1009 | printk(KERN_WARNING "md: %s has same UUID" | |
1010 | " but different superblock to %s\n", | |
1011 | b, bdevname(refdev->bdev, b2)); | |
1012 | goto abort; | |
1013 | } | |
1014 | ev1 = md_event(sb); | |
1015 | ev2 = md_event(refsb); | |
1016 | if (ev1 > ev2) | |
1017 | ret = 1; | |
f72ffdd6 | 1018 | else |
1da177e4 LT |
1019 | ret = 0; |
1020 | } | |
8190e754 | 1021 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1022 | /* Limit to 4TB as metadata cannot record more than that. |
1023 | * (not needed for Linear and RAID0 as metadata doesn't | |
1024 | * record this size) | |
1025 | */ | |
1026 | if (rdev->sectors >= (2ULL << 32) && sb->level >= 1) | |
27a7b260 | 1027 | rdev->sectors = (2ULL << 32) - 2; |
1da177e4 | 1028 | |
27a7b260 | 1029 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1030 | /* "this cannot possibly happen" ... */ |
1031 | ret = -EINVAL; | |
1032 | ||
1da177e4 LT |
1033 | abort: |
1034 | return ret; | |
1035 | } | |
1036 | ||
1037 | /* | |
1038 | * validate_super for 0.90.0 | |
1039 | */ | |
fd01b88c | 1040 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1041 | { |
1042 | mdp_disk_t *desc; | |
65a06f06 | 1043 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1044 | __u64 ev1 = md_event(sb); |
1da177e4 | 1045 | |
41158c7e | 1046 | rdev->raid_disk = -1; |
c5d79adb N |
1047 | clear_bit(Faulty, &rdev->flags); |
1048 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1049 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1050 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1051 | |
1da177e4 LT |
1052 | if (mddev->raid_disks == 0) { |
1053 | mddev->major_version = 0; | |
1054 | mddev->minor_version = sb->minor_version; | |
1055 | mddev->patch_version = sb->patch_version; | |
e691063a | 1056 | mddev->external = 0; |
9d8f0363 | 1057 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1058 | mddev->ctime = sb->ctime; |
1059 | mddev->utime = sb->utime; | |
1060 | mddev->level = sb->level; | |
d9d166c2 | 1061 | mddev->clevel[0] = 0; |
1da177e4 LT |
1062 | mddev->layout = sb->layout; |
1063 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1064 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1065 | mddev->events = ev1; |
c3d9714e | 1066 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1067 | mddev->bitmap_info.space = 0; |
1068 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1069 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1070 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1071 | mddev->reshape_backwards = 0; |
1da177e4 | 1072 | |
f6705578 N |
1073 | if (mddev->minor_version >= 91) { |
1074 | mddev->reshape_position = sb->reshape_position; | |
1075 | mddev->delta_disks = sb->delta_disks; | |
1076 | mddev->new_level = sb->new_level; | |
1077 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1078 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1079 | if (mddev->delta_disks < 0) |
1080 | mddev->reshape_backwards = 1; | |
f6705578 N |
1081 | } else { |
1082 | mddev->reshape_position = MaxSector; | |
1083 | mddev->delta_disks = 0; | |
1084 | mddev->new_level = mddev->level; | |
1085 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1086 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1087 | } |
1088 | ||
1da177e4 LT |
1089 | if (sb->state & (1<<MD_SB_CLEAN)) |
1090 | mddev->recovery_cp = MaxSector; | |
1091 | else { | |
f72ffdd6 | 1092 | if (sb->events_hi == sb->cp_events_hi && |
1da177e4 LT |
1093 | sb->events_lo == sb->cp_events_lo) { |
1094 | mddev->recovery_cp = sb->recovery_cp; | |
1095 | } else | |
1096 | mddev->recovery_cp = 0; | |
1097 | } | |
1098 | ||
1099 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1100 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1101 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1102 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1103 | ||
1104 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1105 | |
1106 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1107 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1108 | mddev->bitmap_info.offset = |
1109 | mddev->bitmap_info.default_offset; | |
6409bb05 | 1110 | mddev->bitmap_info.space = |
c9ad020f | 1111 | mddev->bitmap_info.default_space; |
6409bb05 | 1112 | } |
a654b9d8 | 1113 | |
41158c7e | 1114 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1115 | /* Insist on good event counter while assembling, except |
1116 | * for spares (which don't need an event count) */ | |
1da177e4 | 1117 | ++ev1; |
be6800a7 N |
1118 | if (sb->disks[rdev->desc_nr].state & ( |
1119 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
f72ffdd6 | 1120 | if (ev1 < mddev->events) |
be6800a7 | 1121 | return -EINVAL; |
41158c7e N |
1122 | } else if (mddev->bitmap) { |
1123 | /* if adding to array with a bitmap, then we can accept an | |
1124 | * older device ... but not too old. | |
1125 | */ | |
41158c7e N |
1126 | if (ev1 < mddev->bitmap->events_cleared) |
1127 | return 0; | |
8313b8e5 N |
1128 | if (ev1 < mddev->events) |
1129 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1130 | } else { |
1131 | if (ev1 < mddev->events) | |
1132 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1133 | return 0; | |
1134 | } | |
41158c7e | 1135 | |
1da177e4 | 1136 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1137 | desc = sb->disks + rdev->desc_nr; |
1138 | ||
1139 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1140 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1141 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1142 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1143 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1144 | rdev->raid_disk = desc->raid_disk; |
f466722c | 1145 | rdev->saved_raid_disk = desc->raid_disk; |
0261cd9f N |
1146 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1147 | /* active but not in sync implies recovery up to | |
1148 | * reshape position. We don't know exactly where | |
1149 | * that is, so set to zero for now */ | |
1150 | if (mddev->minor_version >= 91) { | |
1151 | rdev->recovery_offset = 0; | |
1152 | rdev->raid_disk = desc->raid_disk; | |
1153 | } | |
1da177e4 | 1154 | } |
8ddf9efe N |
1155 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1156 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1157 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1158 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1159 | return 0; |
1160 | } | |
1161 | ||
1162 | /* | |
1163 | * sync_super for 0.90.0 | |
1164 | */ | |
fd01b88c | 1165 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1166 | { |
1167 | mdp_super_t *sb; | |
3cb03002 | 1168 | struct md_rdev *rdev2; |
1da177e4 | 1169 | int next_spare = mddev->raid_disks; |
19133a42 | 1170 | |
1da177e4 LT |
1171 | /* make rdev->sb match mddev data.. |
1172 | * | |
1173 | * 1/ zero out disks | |
1174 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1175 | * 3/ any empty disks < next_spare become removed | |
1176 | * | |
1177 | * disks[0] gets initialised to REMOVED because | |
1178 | * we cannot be sure from other fields if it has | |
1179 | * been initialised or not. | |
1180 | */ | |
1181 | int i; | |
1182 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1183 | ||
61181565 N |
1184 | rdev->sb_size = MD_SB_BYTES; |
1185 | ||
65a06f06 | 1186 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1187 | |
1188 | memset(sb, 0, sizeof(*sb)); | |
1189 | ||
1190 | sb->md_magic = MD_SB_MAGIC; | |
1191 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1192 | sb->patch_version = mddev->patch_version; |
1193 | sb->gvalid_words = 0; /* ignored */ | |
1194 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1195 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1196 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1197 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1198 | ||
1199 | sb->ctime = mddev->ctime; | |
1200 | sb->level = mddev->level; | |
58c0fed4 | 1201 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1202 | sb->raid_disks = mddev->raid_disks; |
1203 | sb->md_minor = mddev->md_minor; | |
e691063a | 1204 | sb->not_persistent = 0; |
1da177e4 LT |
1205 | sb->utime = mddev->utime; |
1206 | sb->state = 0; | |
1207 | sb->events_hi = (mddev->events>>32); | |
1208 | sb->events_lo = (u32)mddev->events; | |
1209 | ||
f6705578 N |
1210 | if (mddev->reshape_position == MaxSector) |
1211 | sb->minor_version = 90; | |
1212 | else { | |
1213 | sb->minor_version = 91; | |
1214 | sb->reshape_position = mddev->reshape_position; | |
1215 | sb->new_level = mddev->new_level; | |
1216 | sb->delta_disks = mddev->delta_disks; | |
1217 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1218 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1219 | } |
1220 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1221 | if (mddev->in_sync) |
1222 | { | |
1223 | sb->recovery_cp = mddev->recovery_cp; | |
1224 | sb->cp_events_hi = (mddev->events>>32); | |
1225 | sb->cp_events_lo = (u32)mddev->events; | |
1226 | if (mddev->recovery_cp == MaxSector) | |
1227 | sb->state = (1<< MD_SB_CLEAN); | |
1228 | } else | |
1229 | sb->recovery_cp = 0; | |
1230 | ||
1231 | sb->layout = mddev->layout; | |
9d8f0363 | 1232 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1233 | |
c3d9714e | 1234 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1235 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1236 | ||
1da177e4 | 1237 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1238 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1239 | mdp_disk_t *d; |
86e6ffdd | 1240 | int desc_nr; |
0261cd9f N |
1241 | int is_active = test_bit(In_sync, &rdev2->flags); |
1242 | ||
1243 | if (rdev2->raid_disk >= 0 && | |
1244 | sb->minor_version >= 91) | |
1245 | /* we have nowhere to store the recovery_offset, | |
1246 | * but if it is not below the reshape_position, | |
1247 | * we can piggy-back on that. | |
1248 | */ | |
1249 | is_active = 1; | |
1250 | if (rdev2->raid_disk < 0 || | |
1251 | test_bit(Faulty, &rdev2->flags)) | |
1252 | is_active = 0; | |
1253 | if (is_active) | |
86e6ffdd | 1254 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1255 | else |
86e6ffdd | 1256 | desc_nr = next_spare++; |
19133a42 | 1257 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1258 | d = &sb->disks[rdev2->desc_nr]; |
1259 | nr_disks++; | |
1260 | d->number = rdev2->desc_nr; | |
1261 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1262 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1263 | if (is_active) |
1da177e4 LT |
1264 | d->raid_disk = rdev2->raid_disk; |
1265 | else | |
1266 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1267 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1268 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1269 | else if (is_active) { |
1da177e4 | 1270 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1271 | if (test_bit(In_sync, &rdev2->flags)) |
1272 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1273 | active++; |
1274 | working++; | |
1275 | } else { | |
1276 | d->state = 0; | |
1277 | spare++; | |
1278 | working++; | |
1279 | } | |
8ddf9efe N |
1280 | if (test_bit(WriteMostly, &rdev2->flags)) |
1281 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1282 | } |
1da177e4 LT |
1283 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1284 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1285 | mdp_disk_t *d = &sb->disks[i]; | |
1286 | if (d->state == 0 && d->number == 0) { | |
1287 | d->number = i; | |
1288 | d->raid_disk = i; | |
1289 | d->state = (1<<MD_DISK_REMOVED); | |
1290 | d->state |= (1<<MD_DISK_FAULTY); | |
1291 | failed++; | |
1292 | } | |
1293 | } | |
1294 | sb->nr_disks = nr_disks; | |
1295 | sb->active_disks = active; | |
1296 | sb->working_disks = working; | |
1297 | sb->failed_disks = failed; | |
1298 | sb->spare_disks = spare; | |
1299 | ||
1300 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1301 | sb->sb_csum = calc_sb_csum(sb); | |
1302 | } | |
1303 | ||
0cd17fec CW |
1304 | /* |
1305 | * rdev_size_change for 0.90.0 | |
1306 | */ | |
1307 | static unsigned long long | |
3cb03002 | 1308 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1309 | { |
58c0fed4 | 1310 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1311 | return 0; /* component must fit device */ |
c3d9714e | 1312 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1313 | return 0; /* can't move bitmap */ |
57b2caa3 | 1314 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1315 | if (!num_sectors || num_sectors > rdev->sb_start) |
1316 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1317 | /* Limit to 4TB as metadata cannot record more than that. |
1318 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1319 | */ | |
667a5313 | 1320 | if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1) |
27a7b260 | 1321 | num_sectors = (2ULL << 32) - 2; |
0f420358 | 1322 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1323 | rdev->sb_page); |
1324 | md_super_wait(rdev->mddev); | |
c26a44ed | 1325 | return num_sectors; |
0cd17fec CW |
1326 | } |
1327 | ||
c6563a8c N |
1328 | static int |
1329 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1330 | { | |
1331 | /* non-zero offset changes not possible with v0.90 */ | |
1332 | return new_offset == 0; | |
1333 | } | |
0cd17fec | 1334 | |
1da177e4 LT |
1335 | /* |
1336 | * version 1 superblock | |
1337 | */ | |
1338 | ||
f72ffdd6 | 1339 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb) |
1da177e4 | 1340 | { |
1c05b4bc N |
1341 | __le32 disk_csum; |
1342 | u32 csum; | |
1da177e4 LT |
1343 | unsigned long long newcsum; |
1344 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1345 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1346 | |
1347 | disk_csum = sb->sb_csum; | |
1348 | sb->sb_csum = 0; | |
1349 | newcsum = 0; | |
1f3c9907 | 1350 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1351 | newcsum += le32_to_cpu(*isuper++); |
1352 | ||
1353 | if (size == 2) | |
1c05b4bc | 1354 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1355 | |
1356 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1357 | sb->sb_csum = disk_csum; | |
1358 | return cpu_to_le32(csum); | |
1359 | } | |
1360 | ||
2699b672 N |
1361 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, |
1362 | int acknowledged); | |
3cb03002 | 1363 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1364 | { |
1365 | struct mdp_superblock_1 *sb; | |
1366 | int ret; | |
0f420358 | 1367 | sector_t sb_start; |
c6563a8c | 1368 | sector_t sectors; |
1da177e4 | 1369 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1370 | int bmask; |
1da177e4 LT |
1371 | |
1372 | /* | |
0f420358 | 1373 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1374 | * It is always aligned to a 4K boundary and |
1375 | * depeding on minor_version, it can be: | |
1376 | * 0: At least 8K, but less than 12K, from end of device | |
1377 | * 1: At start of device | |
1378 | * 2: 4K from start of device. | |
1379 | */ | |
1380 | switch(minor_version) { | |
1381 | case 0: | |
77304d2a | 1382 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1383 | sb_start -= 8*2; |
1384 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1385 | break; |
1386 | case 1: | |
0f420358 | 1387 | sb_start = 0; |
1da177e4 LT |
1388 | break; |
1389 | case 2: | |
0f420358 | 1390 | sb_start = 8; |
1da177e4 LT |
1391 | break; |
1392 | default: | |
1393 | return -EINVAL; | |
1394 | } | |
0f420358 | 1395 | rdev->sb_start = sb_start; |
1da177e4 | 1396 | |
0002b271 N |
1397 | /* superblock is rarely larger than 1K, but it can be larger, |
1398 | * and it is safe to read 4k, so we do that | |
1399 | */ | |
1400 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1401 | if (ret) return ret; |
1402 | ||
65a06f06 | 1403 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1404 | |
1405 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1406 | sb->major_version != cpu_to_le32(1) || | |
1407 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1408 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1409 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1410 | return -EINVAL; |
1411 | ||
1412 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1413 | printk("md: invalid superblock checksum on %s\n", | |
1414 | bdevname(rdev->bdev,b)); | |
1415 | return -EINVAL; | |
1416 | } | |
1417 | if (le64_to_cpu(sb->data_size) < 10) { | |
1418 | printk("md: data_size too small on %s\n", | |
1419 | bdevname(rdev->bdev,b)); | |
1420 | return -EINVAL; | |
1421 | } | |
c6563a8c N |
1422 | if (sb->pad0 || |
1423 | sb->pad3[0] || | |
1424 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1425 | /* Some padding is non-zero, might be a new feature */ | |
1426 | return -EINVAL; | |
e11e93fa | 1427 | |
1da177e4 LT |
1428 | rdev->preferred_minor = 0xffff; |
1429 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1430 | rdev->new_data_offset = rdev->data_offset; |
1431 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1432 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1433 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1434 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1435 | |
0002b271 | 1436 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1437 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1438 | if (rdev->sb_size & bmask) |
a1801f85 N |
1439 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1440 | ||
1441 | if (minor_version | |
0f420358 | 1442 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1443 | return -EINVAL; |
c6563a8c N |
1444 | if (minor_version |
1445 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1446 | return -EINVAL; | |
0002b271 | 1447 | |
31b65a0d N |
1448 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1449 | rdev->desc_nr = -1; | |
1450 | else | |
1451 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1452 | ||
2699b672 N |
1453 | if (!rdev->bb_page) { |
1454 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1455 | if (!rdev->bb_page) | |
1456 | return -ENOMEM; | |
1457 | } | |
1458 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1459 | rdev->badblocks.count == 0) { | |
1460 | /* need to load the bad block list. | |
1461 | * Currently we limit it to one page. | |
1462 | */ | |
1463 | s32 offset; | |
1464 | sector_t bb_sector; | |
1465 | u64 *bbp; | |
1466 | int i; | |
1467 | int sectors = le16_to_cpu(sb->bblog_size); | |
1468 | if (sectors > (PAGE_SIZE / 512)) | |
1469 | return -EINVAL; | |
1470 | offset = le32_to_cpu(sb->bblog_offset); | |
1471 | if (offset == 0) | |
1472 | return -EINVAL; | |
1473 | bb_sector = (long long)offset; | |
1474 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
1475 | rdev->bb_page, READ, true)) | |
1476 | return -EIO; | |
1477 | bbp = (u64 *)page_address(rdev->bb_page); | |
1478 | rdev->badblocks.shift = sb->bblog_shift; | |
1479 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1480 | u64 bb = le64_to_cpu(*bbp); | |
1481 | int count = bb & (0x3ff); | |
1482 | u64 sector = bb >> 10; | |
1483 | sector <<= sb->bblog_shift; | |
1484 | count <<= sb->bblog_shift; | |
1485 | if (bb + 1 == 0) | |
1486 | break; | |
1487 | if (md_set_badblocks(&rdev->badblocks, | |
1488 | sector, count, 1) == 0) | |
1489 | return -EINVAL; | |
1490 | } | |
486adf72 N |
1491 | } else if (sb->bblog_offset != 0) |
1492 | rdev->badblocks.shift = 0; | |
2699b672 | 1493 | |
9a7b2b0f | 1494 | if (!refdev) { |
8ed75463 | 1495 | ret = 1; |
9a7b2b0f | 1496 | } else { |
1da177e4 | 1497 | __u64 ev1, ev2; |
65a06f06 | 1498 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1499 | |
1500 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1501 | sb->level != refsb->level || | |
1502 | sb->layout != refsb->layout || | |
1503 | sb->chunksize != refsb->chunksize) { | |
1504 | printk(KERN_WARNING "md: %s has strangely different" | |
1505 | " superblock to %s\n", | |
1506 | bdevname(rdev->bdev,b), | |
1507 | bdevname(refdev->bdev,b2)); | |
1508 | return -EINVAL; | |
1509 | } | |
1510 | ev1 = le64_to_cpu(sb->events); | |
1511 | ev2 = le64_to_cpu(refsb->events); | |
1512 | ||
1513 | if (ev1 > ev2) | |
8ed75463 N |
1514 | ret = 1; |
1515 | else | |
1516 | ret = 0; | |
1da177e4 | 1517 | } |
c6563a8c N |
1518 | if (minor_version) { |
1519 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1520 | sectors -= rdev->data_offset; | |
1521 | } else | |
1522 | sectors = rdev->sb_start; | |
1523 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1524 | return -EINVAL; |
dd8ac336 | 1525 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1526 | return ret; |
1da177e4 LT |
1527 | } |
1528 | ||
fd01b88c | 1529 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1530 | { |
65a06f06 | 1531 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1532 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1533 | |
41158c7e | 1534 | rdev->raid_disk = -1; |
c5d79adb N |
1535 | clear_bit(Faulty, &rdev->flags); |
1536 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1537 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1538 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1539 | |
1da177e4 LT |
1540 | if (mddev->raid_disks == 0) { |
1541 | mddev->major_version = 1; | |
1542 | mddev->patch_version = 0; | |
e691063a | 1543 | mddev->external = 0; |
9d8f0363 | 1544 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1545 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1546 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1547 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1548 | mddev->clevel[0] = 0; |
1da177e4 LT |
1549 | mddev->layout = le32_to_cpu(sb->layout); |
1550 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1551 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1552 | mddev->events = ev1; |
c3d9714e | 1553 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1554 | mddev->bitmap_info.space = 0; |
1555 | /* Default location for bitmap is 1K after superblock | |
1556 | * using 3K - total of 4K | |
1557 | */ | |
c3d9714e | 1558 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1559 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1560 | mddev->reshape_backwards = 0; |
1561 | ||
1da177e4 LT |
1562 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1563 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1564 | ||
1565 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1566 | |
71c0805c | 1567 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1568 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1569 | mddev->bitmap_info.offset = |
1570 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1571 | /* Metadata doesn't record how much space is available. |
1572 | * For 1.0, we assume we can use up to the superblock | |
1573 | * if before, else to 4K beyond superblock. | |
1574 | * For others, assume no change is possible. | |
1575 | */ | |
1576 | if (mddev->minor_version > 0) | |
1577 | mddev->bitmap_info.space = 0; | |
1578 | else if (mddev->bitmap_info.offset > 0) | |
1579 | mddev->bitmap_info.space = | |
1580 | 8 - mddev->bitmap_info.offset; | |
1581 | else | |
1582 | mddev->bitmap_info.space = | |
1583 | -mddev->bitmap_info.offset; | |
1584 | } | |
e11e93fa | 1585 | |
f6705578 N |
1586 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1587 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1588 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1589 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1590 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1591 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1592 | if (mddev->delta_disks < 0 || |
1593 | (mddev->delta_disks == 0 && | |
1594 | (le32_to_cpu(sb->feature_map) | |
1595 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1596 | mddev->reshape_backwards = 1; | |
f6705578 N |
1597 | } else { |
1598 | mddev->reshape_position = MaxSector; | |
1599 | mddev->delta_disks = 0; | |
1600 | mddev->new_level = mddev->level; | |
1601 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1602 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1603 | } |
1604 | ||
41158c7e | 1605 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1606 | /* Insist of good event counter while assembling, except for |
1607 | * spares (which don't need an event count) */ | |
1da177e4 | 1608 | ++ev1; |
be6800a7 N |
1609 | if (rdev->desc_nr >= 0 && |
1610 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
1611 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) | |
1612 | if (ev1 < mddev->events) | |
1613 | return -EINVAL; | |
41158c7e N |
1614 | } else if (mddev->bitmap) { |
1615 | /* If adding to array with a bitmap, then we can accept an | |
1616 | * older device, but not too old. | |
1617 | */ | |
41158c7e N |
1618 | if (ev1 < mddev->bitmap->events_cleared) |
1619 | return 0; | |
8313b8e5 N |
1620 | if (ev1 < mddev->events) |
1621 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1622 | } else { |
1623 | if (ev1 < mddev->events) | |
1624 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1625 | return 0; | |
1626 | } | |
1da177e4 LT |
1627 | if (mddev->level != LEVEL_MULTIPATH) { |
1628 | int role; | |
3673f305 N |
1629 | if (rdev->desc_nr < 0 || |
1630 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1631 | role = 0xffff; | |
1632 | rdev->desc_nr = -1; | |
1633 | } else | |
1634 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1635 | switch(role) { |
1636 | case 0xffff: /* spare */ | |
1da177e4 LT |
1637 | break; |
1638 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1639 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1640 | break; |
1641 | default: | |
f466722c | 1642 | rdev->saved_raid_disk = role; |
5fd6c1dc | 1643 | if ((le32_to_cpu(sb->feature_map) & |
f466722c | 1644 | MD_FEATURE_RECOVERY_OFFSET)) { |
5fd6c1dc | 1645 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); |
f466722c N |
1646 | if (!(le32_to_cpu(sb->feature_map) & |
1647 | MD_FEATURE_RECOVERY_BITMAP)) | |
1648 | rdev->saved_raid_disk = -1; | |
1649 | } else | |
5fd6c1dc | 1650 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1651 | rdev->raid_disk = role; |
1652 | break; | |
1653 | } | |
8ddf9efe N |
1654 | if (sb->devflags & WriteMostly1) |
1655 | set_bit(WriteMostly, &rdev->flags); | |
2d78f8c4 N |
1656 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1657 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1658 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1659 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1660 | |
1da177e4 LT |
1661 | return 0; |
1662 | } | |
1663 | ||
fd01b88c | 1664 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1665 | { |
1666 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1667 | struct md_rdev *rdev2; |
1da177e4 LT |
1668 | int max_dev, i; |
1669 | /* make rdev->sb match mddev and rdev data. */ | |
1670 | ||
65a06f06 | 1671 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1672 | |
1673 | sb->feature_map = 0; | |
1674 | sb->pad0 = 0; | |
5fd6c1dc | 1675 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1676 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1677 | ||
1678 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1679 | sb->events = cpu_to_le64(mddev->events); | |
1680 | if (mddev->in_sync) | |
1681 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1682 | else | |
1683 | sb->resync_offset = cpu_to_le64(0); | |
1684 | ||
1c05b4bc | 1685 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1686 | |
f0ca340c | 1687 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1688 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1689 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1690 | sb->level = cpu_to_le32(mddev->level); |
1691 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1692 | |
aeb9b211 N |
1693 | if (test_bit(WriteMostly, &rdev->flags)) |
1694 | sb->devflags |= WriteMostly1; | |
1695 | else | |
1696 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1697 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1698 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1699 | |
c3d9714e N |
1700 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1701 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1702 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1703 | } |
5fd6c1dc N |
1704 | |
1705 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1706 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1707 | sb->feature_map |= |
1708 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1709 | sb->recovery_offset = | |
1710 | cpu_to_le64(rdev->recovery_offset); | |
f466722c N |
1711 | if (rdev->saved_raid_disk >= 0 && mddev->bitmap) |
1712 | sb->feature_map |= | |
1713 | cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP); | |
5fd6c1dc | 1714 | } |
2d78f8c4 N |
1715 | if (test_bit(Replacement, &rdev->flags)) |
1716 | sb->feature_map |= | |
1717 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1718 | |
f6705578 N |
1719 | if (mddev->reshape_position != MaxSector) { |
1720 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1721 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1722 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1723 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1724 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1725 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1726 | if (mddev->delta_disks == 0 && |
1727 | mddev->reshape_backwards) | |
1728 | sb->feature_map | |
1729 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1730 | if (rdev->new_data_offset != rdev->data_offset) { |
1731 | sb->feature_map | |
1732 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1733 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1734 | - rdev->data_offset)); | |
1735 | } | |
f6705578 | 1736 | } |
a654b9d8 | 1737 | |
2699b672 N |
1738 | if (rdev->badblocks.count == 0) |
1739 | /* Nothing to do for bad blocks*/ ; | |
1740 | else if (sb->bblog_offset == 0) | |
1741 | /* Cannot record bad blocks on this device */ | |
1742 | md_error(mddev, rdev); | |
1743 | else { | |
1744 | struct badblocks *bb = &rdev->badblocks; | |
1745 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1746 | u64 *p = bb->page; | |
1747 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1748 | if (bb->changed) { | |
1749 | unsigned seq; | |
1750 | ||
1751 | retry: | |
1752 | seq = read_seqbegin(&bb->lock); | |
1753 | ||
1754 | memset(bbp, 0xff, PAGE_SIZE); | |
1755 | ||
1756 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1757 | u64 internal_bb = p[i]; |
2699b672 N |
1758 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1759 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1760 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1761 | } |
d0962936 | 1762 | bb->changed = 0; |
2699b672 N |
1763 | if (read_seqretry(&bb->lock, seq)) |
1764 | goto retry; | |
1765 | ||
1766 | bb->sector = (rdev->sb_start + | |
1767 | (int)le32_to_cpu(sb->bblog_offset)); | |
1768 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1769 | } |
1770 | } | |
1771 | ||
1da177e4 | 1772 | max_dev = 0; |
dafb20fa | 1773 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1774 | if (rdev2->desc_nr+1 > max_dev) |
1775 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1776 | |
70471daf N |
1777 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1778 | int bmask; | |
a778b73f | 1779 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1780 | rdev->sb_size = max_dev * 2 + 256; |
1781 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1782 | if (rdev->sb_size & bmask) | |
1783 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1784 | } else |
1785 | max_dev = le32_to_cpu(sb->max_dev); | |
1786 | ||
1da177e4 LT |
1787 | for (i=0; i<max_dev;i++) |
1788 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
f72ffdd6 | 1789 | |
dafb20fa | 1790 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1791 | i = rdev2->desc_nr; |
b2d444d7 | 1792 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1793 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1794 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1795 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1796 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1797 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1798 | else |
1799 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1800 | } | |
1801 | ||
1da177e4 LT |
1802 | sb->sb_csum = calc_sb_1_csum(sb); |
1803 | } | |
1804 | ||
0cd17fec | 1805 | static unsigned long long |
3cb03002 | 1806 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1807 | { |
1808 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1809 | sector_t max_sectors; |
58c0fed4 | 1810 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1811 | return 0; /* component must fit device */ |
c6563a8c N |
1812 | if (rdev->data_offset != rdev->new_data_offset) |
1813 | return 0; /* too confusing */ | |
0f420358 | 1814 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1815 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1816 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1817 | max_sectors -= rdev->data_offset; |
1818 | if (!num_sectors || num_sectors > max_sectors) | |
1819 | num_sectors = max_sectors; | |
c3d9714e | 1820 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1821 | /* minor version 0 with bitmap we can't move */ |
1822 | return 0; | |
1823 | } else { | |
1824 | /* minor version 0; superblock after data */ | |
0f420358 | 1825 | sector_t sb_start; |
77304d2a | 1826 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1827 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1828 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1829 | if (!num_sectors || num_sectors > max_sectors) |
1830 | num_sectors = max_sectors; | |
0f420358 | 1831 | rdev->sb_start = sb_start; |
0cd17fec | 1832 | } |
65a06f06 | 1833 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1834 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1835 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1836 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1837 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1838 | rdev->sb_page); |
1839 | md_super_wait(rdev->mddev); | |
c26a44ed | 1840 | return num_sectors; |
c6563a8c N |
1841 | |
1842 | } | |
1843 | ||
1844 | static int | |
1845 | super_1_allow_new_offset(struct md_rdev *rdev, | |
1846 | unsigned long long new_offset) | |
1847 | { | |
1848 | /* All necessary checks on new >= old have been done */ | |
1849 | struct bitmap *bitmap; | |
1850 | if (new_offset >= rdev->data_offset) | |
1851 | return 1; | |
1852 | ||
1853 | /* with 1.0 metadata, there is no metadata to tread on | |
1854 | * so we can always move back */ | |
1855 | if (rdev->mddev->minor_version == 0) | |
1856 | return 1; | |
1857 | ||
1858 | /* otherwise we must be sure not to step on | |
1859 | * any metadata, so stay: | |
1860 | * 36K beyond start of superblock | |
1861 | * beyond end of badblocks | |
1862 | * beyond write-intent bitmap | |
1863 | */ | |
1864 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
1865 | return 0; | |
1866 | bitmap = rdev->mddev->bitmap; | |
1867 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
1868 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 1869 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
1870 | return 0; |
1871 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
1872 | return 0; | |
1873 | ||
1874 | return 1; | |
0cd17fec | 1875 | } |
1da177e4 | 1876 | |
75c96f85 | 1877 | static struct super_type super_types[] = { |
1da177e4 LT |
1878 | [0] = { |
1879 | .name = "0.90.0", | |
1880 | .owner = THIS_MODULE, | |
0cd17fec CW |
1881 | .load_super = super_90_load, |
1882 | .validate_super = super_90_validate, | |
1883 | .sync_super = super_90_sync, | |
1884 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 1885 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
1886 | }, |
1887 | [1] = { | |
1888 | .name = "md-1", | |
1889 | .owner = THIS_MODULE, | |
0cd17fec CW |
1890 | .load_super = super_1_load, |
1891 | .validate_super = super_1_validate, | |
1892 | .sync_super = super_1_sync, | |
1893 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 1894 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
1895 | }, |
1896 | }; | |
1da177e4 | 1897 | |
fd01b88c | 1898 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
1899 | { |
1900 | if (mddev->sync_super) { | |
1901 | mddev->sync_super(mddev, rdev); | |
1902 | return; | |
1903 | } | |
1904 | ||
1905 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
1906 | ||
1907 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
1908 | } | |
1909 | ||
fd01b88c | 1910 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 1911 | { |
3cb03002 | 1912 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 1913 | |
4b80991c N |
1914 | rcu_read_lock(); |
1915 | rdev_for_each_rcu(rdev, mddev1) | |
1916 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1917 | if (rdev->bdev->bd_contains == |
4b80991c N |
1918 | rdev2->bdev->bd_contains) { |
1919 | rcu_read_unlock(); | |
7dd5e7c3 | 1920 | return 1; |
4b80991c N |
1921 | } |
1922 | rcu_read_unlock(); | |
1da177e4 LT |
1923 | return 0; |
1924 | } | |
1925 | ||
1926 | static LIST_HEAD(pending_raid_disks); | |
1927 | ||
ac5e7113 AN |
1928 | /* |
1929 | * Try to register data integrity profile for an mddev | |
1930 | * | |
1931 | * This is called when an array is started and after a disk has been kicked | |
1932 | * from the array. It only succeeds if all working and active component devices | |
1933 | * are integrity capable with matching profiles. | |
1934 | */ | |
fd01b88c | 1935 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 1936 | { |
3cb03002 | 1937 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
1938 | |
1939 | if (list_empty(&mddev->disks)) | |
1940 | return 0; /* nothing to do */ | |
629acb6a JB |
1941 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
1942 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 1943 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
1944 | /* skip spares and non-functional disks */ |
1945 | if (test_bit(Faulty, &rdev->flags)) | |
1946 | continue; | |
1947 | if (rdev->raid_disk < 0) | |
1948 | continue; | |
ac5e7113 AN |
1949 | if (!reference) { |
1950 | /* Use the first rdev as the reference */ | |
1951 | reference = rdev; | |
1952 | continue; | |
1953 | } | |
1954 | /* does this rdev's profile match the reference profile? */ | |
1955 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1956 | rdev->bdev->bd_disk) < 0) | |
1957 | return -EINVAL; | |
1958 | } | |
89078d57 MP |
1959 | if (!reference || !bdev_get_integrity(reference->bdev)) |
1960 | return 0; | |
ac5e7113 AN |
1961 | /* |
1962 | * All component devices are integrity capable and have matching | |
1963 | * profiles, register the common profile for the md device. | |
1964 | */ | |
1965 | if (blk_integrity_register(mddev->gendisk, | |
1966 | bdev_get_integrity(reference->bdev)) != 0) { | |
1967 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1968 | mdname(mddev)); | |
1969 | return -EINVAL; | |
1970 | } | |
a91a2785 MP |
1971 | printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); |
1972 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { | |
1973 | printk(KERN_ERR "md: failed to create integrity pool for %s\n", | |
1974 | mdname(mddev)); | |
1975 | return -EINVAL; | |
1976 | } | |
ac5e7113 AN |
1977 | return 0; |
1978 | } | |
1979 | EXPORT_SYMBOL(md_integrity_register); | |
1980 | ||
1981 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
fd01b88c | 1982 | void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) |
3f9d99c1 | 1983 | { |
2863b9eb JB |
1984 | struct blk_integrity *bi_rdev; |
1985 | struct blk_integrity *bi_mddev; | |
1986 | ||
1987 | if (!mddev->gendisk) | |
1988 | return; | |
1989 | ||
1990 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
1991 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 1992 | |
ac5e7113 | 1993 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 1994 | return; |
ac5e7113 | 1995 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 1996 | return; |
ac5e7113 AN |
1997 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
1998 | rdev->bdev->bd_disk) >= 0) | |
1999 | return; | |
2000 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
2001 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 2002 | } |
ac5e7113 | 2003 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 2004 | |
f72ffdd6 | 2005 | static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev) |
1da177e4 | 2006 | { |
7dd5e7c3 | 2007 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2008 | struct kobject *ko; |
5e55e2f5 | 2009 | int err; |
1da177e4 | 2010 | |
11e2ede0 DW |
2011 | /* prevent duplicates */ |
2012 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2013 | return -EEXIST; | |
2014 | ||
dd8ac336 AN |
2015 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
2016 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
2017 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2018 | if (mddev->pers) { |
2019 | /* Cannot change size, so fail | |
2020 | * If mddev->level <= 0, then we don't care | |
2021 | * about aligning sizes (e.g. linear) | |
2022 | */ | |
2023 | if (mddev->level > 0) | |
2024 | return -ENOSPC; | |
2025 | } else | |
dd8ac336 | 2026 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2027 | } |
1da177e4 LT |
2028 | |
2029 | /* Verify rdev->desc_nr is unique. | |
2030 | * If it is -1, assign a free number, else | |
2031 | * check number is not in use | |
2032 | */ | |
4878e9eb | 2033 | rcu_read_lock(); |
1da177e4 LT |
2034 | if (rdev->desc_nr < 0) { |
2035 | int choice = 0; | |
4878e9eb N |
2036 | if (mddev->pers) |
2037 | choice = mddev->raid_disks; | |
57d051dc | 2038 | while (md_find_rdev_nr_rcu(mddev, choice)) |
1da177e4 LT |
2039 | choice++; |
2040 | rdev->desc_nr = choice; | |
2041 | } else { | |
57d051dc | 2042 | if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) { |
4878e9eb | 2043 | rcu_read_unlock(); |
1da177e4 | 2044 | return -EBUSY; |
4878e9eb | 2045 | } |
1da177e4 | 2046 | } |
4878e9eb | 2047 | rcu_read_unlock(); |
de01dfad N |
2048 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
2049 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
2050 | mdname(mddev), mddev->max_disks); | |
2051 | return -EBUSY; | |
2052 | } | |
19133a42 | 2053 | bdevname(rdev->bdev,b); |
90a9befb | 2054 | strreplace(b, '/', '!'); |
649316b2 | 2055 | |
1da177e4 | 2056 | rdev->mddev = mddev; |
19133a42 | 2057 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 2058 | |
b2d6db58 | 2059 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2060 | goto fail; |
86e6ffdd | 2061 | |
0762b8bd | 2062 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2063 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2064 | /* failure here is OK */; | |
2065 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2066 | |
4b80991c | 2067 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2068 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2069 | |
2070 | /* May as well allow recovery to be retried once */ | |
5389042f | 2071 | mddev->recovery_disabled++; |
3f9d99c1 | 2072 | |
1da177e4 | 2073 | return 0; |
5e55e2f5 N |
2074 | |
2075 | fail: | |
2076 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
2077 | b, mdname(mddev)); | |
2078 | return err; | |
1da177e4 LT |
2079 | } |
2080 | ||
177a99b2 | 2081 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2082 | { |
3cb03002 | 2083 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2084 | kobject_del(&rdev->kobj); |
177a99b2 | 2085 | kobject_put(&rdev->kobj); |
5792a285 N |
2086 | } |
2087 | ||
f72ffdd6 | 2088 | static void unbind_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2089 | { |
2090 | char b[BDEVNAME_SIZE]; | |
403df478 | 2091 | |
49731baa | 2092 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2093 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
2094 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
2095 | rdev->mddev = NULL; | |
86e6ffdd | 2096 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2097 | sysfs_put(rdev->sysfs_state); |
2098 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2099 | rdev->badblocks.count = 0; |
5792a285 | 2100 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2101 | * writing to 'remove' to "dev/state". We also need |
2102 | * to delay it due to rcu usage. | |
5792a285 | 2103 | */ |
4b80991c | 2104 | synchronize_rcu(); |
177a99b2 N |
2105 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2106 | kobject_get(&rdev->kobj); | |
e804ac78 | 2107 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2108 | } |
2109 | ||
2110 | /* | |
2111 | * prevent the device from being mounted, repartitioned or | |
2112 | * otherwise reused by a RAID array (or any other kernel | |
2113 | * subsystem), by bd_claiming the device. | |
2114 | */ | |
3cb03002 | 2115 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2116 | { |
2117 | int err = 0; | |
2118 | struct block_device *bdev; | |
2119 | char b[BDEVNAME_SIZE]; | |
2120 | ||
d4d77629 | 2121 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2122 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 LT |
2123 | if (IS_ERR(bdev)) { |
2124 | printk(KERN_ERR "md: could not open %s.\n", | |
2125 | __bdevname(dev, b)); | |
2126 | return PTR_ERR(bdev); | |
2127 | } | |
1da177e4 LT |
2128 | rdev->bdev = bdev; |
2129 | return err; | |
2130 | } | |
2131 | ||
3cb03002 | 2132 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2133 | { |
2134 | struct block_device *bdev = rdev->bdev; | |
2135 | rdev->bdev = NULL; | |
e525fd89 | 2136 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2137 | } |
2138 | ||
2139 | void md_autodetect_dev(dev_t dev); | |
2140 | ||
f72ffdd6 | 2141 | static void export_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2142 | { |
2143 | char b[BDEVNAME_SIZE]; | |
403df478 | 2144 | |
1da177e4 LT |
2145 | printk(KERN_INFO "md: export_rdev(%s)\n", |
2146 | bdevname(rdev->bdev,b)); | |
545c8795 | 2147 | md_rdev_clear(rdev); |
1da177e4 | 2148 | #ifndef MODULE |
d0fae18f N |
2149 | if (test_bit(AutoDetected, &rdev->flags)) |
2150 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2151 | #endif |
2152 | unlock_rdev(rdev); | |
86e6ffdd | 2153 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2154 | } |
2155 | ||
fb56dfef | 2156 | void md_kick_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2157 | { |
2158 | unbind_rdev_from_array(rdev); | |
2159 | export_rdev(rdev); | |
2160 | } | |
fb56dfef | 2161 | EXPORT_SYMBOL_GPL(md_kick_rdev_from_array); |
1da177e4 | 2162 | |
fd01b88c | 2163 | static void export_array(struct mddev *mddev) |
1da177e4 | 2164 | { |
0638bb0e | 2165 | struct md_rdev *rdev; |
1da177e4 | 2166 | |
0638bb0e N |
2167 | while (!list_empty(&mddev->disks)) { |
2168 | rdev = list_first_entry(&mddev->disks, struct md_rdev, | |
2169 | same_set); | |
fb56dfef | 2170 | md_kick_rdev_from_array(rdev); |
1da177e4 | 2171 | } |
1da177e4 LT |
2172 | mddev->raid_disks = 0; |
2173 | mddev->major_version = 0; | |
2174 | } | |
2175 | ||
f72ffdd6 | 2176 | static void sync_sbs(struct mddev *mddev, int nospares) |
1da177e4 | 2177 | { |
42543769 N |
2178 | /* Update each superblock (in-memory image), but |
2179 | * if we are allowed to, skip spares which already | |
2180 | * have the right event counter, or have one earlier | |
2181 | * (which would mean they aren't being marked as dirty | |
2182 | * with the rest of the array) | |
2183 | */ | |
3cb03002 | 2184 | struct md_rdev *rdev; |
dafb20fa | 2185 | rdev_for_each(rdev, mddev) { |
42543769 N |
2186 | if (rdev->sb_events == mddev->events || |
2187 | (nospares && | |
2188 | rdev->raid_disk < 0 && | |
42543769 N |
2189 | rdev->sb_events+1 == mddev->events)) { |
2190 | /* Don't update this superblock */ | |
2191 | rdev->sb_loaded = 2; | |
2192 | } else { | |
076f968b | 2193 | sync_super(mddev, rdev); |
42543769 N |
2194 | rdev->sb_loaded = 1; |
2195 | } | |
1da177e4 LT |
2196 | } |
2197 | } | |
2198 | ||
1aee41f6 | 2199 | void md_update_sb(struct mddev *mddev, int force_change) |
1da177e4 | 2200 | { |
3cb03002 | 2201 | struct md_rdev *rdev; |
06d91a5f | 2202 | int sync_req; |
42543769 | 2203 | int nospares = 0; |
2699b672 | 2204 | int any_badblocks_changed = 0; |
1da177e4 | 2205 | |
d87f064f N |
2206 | if (mddev->ro) { |
2207 | if (force_change) | |
2208 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2209 | return; | |
2210 | } | |
1da177e4 | 2211 | repeat: |
3a3a5ddb | 2212 | /* First make sure individual recovery_offsets are correct */ |
dafb20fa | 2213 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2214 | if (rdev->raid_disk >= 0 && |
2215 | mddev->delta_disks >= 0 && | |
2216 | !test_bit(In_sync, &rdev->flags) && | |
2217 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2218 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2219 | ||
f72ffdd6 | 2220 | } |
bd52b746 | 2221 | if (!mddev->persistent) { |
070dc6dd | 2222 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2223 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
de393cde | 2224 | if (!mddev->external) { |
d97a41dc | 2225 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
dafb20fa | 2226 | rdev_for_each(rdev, mddev) { |
de393cde | 2227 | if (rdev->badblocks.changed) { |
d0962936 | 2228 | rdev->badblocks.changed = 0; |
de393cde N |
2229 | md_ack_all_badblocks(&rdev->badblocks); |
2230 | md_error(mddev, rdev); | |
2231 | } | |
2232 | clear_bit(Blocked, &rdev->flags); | |
2233 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2234 | wake_up(&rdev->blocked_wait); | |
2235 | } | |
2236 | } | |
3a3a5ddb N |
2237 | wake_up(&mddev->sb_wait); |
2238 | return; | |
2239 | } | |
2240 | ||
85572d7c | 2241 | spin_lock(&mddev->lock); |
84692195 | 2242 | |
3a3a5ddb N |
2243 | mddev->utime = get_seconds(); |
2244 | ||
850b2b42 N |
2245 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2246 | force_change = 1; | |
2247 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2248 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2249 | * though if events isn't the right even/odd, we will have to do | |
2250 | * spares after all | |
2251 | */ | |
2252 | nospares = 1; | |
2253 | if (force_change) | |
2254 | nospares = 0; | |
2255 | if (mddev->degraded) | |
84692195 N |
2256 | /* If the array is degraded, then skipping spares is both |
2257 | * dangerous and fairly pointless. | |
2258 | * Dangerous because a device that was removed from the array | |
2259 | * might have a event_count that still looks up-to-date, | |
2260 | * so it can be re-added without a resync. | |
2261 | * Pointless because if there are any spares to skip, | |
2262 | * then a recovery will happen and soon that array won't | |
2263 | * be degraded any more and the spare can go back to sleep then. | |
2264 | */ | |
850b2b42 | 2265 | nospares = 0; |
84692195 | 2266 | |
06d91a5f | 2267 | sync_req = mddev->in_sync; |
42543769 N |
2268 | |
2269 | /* If this is just a dirty<->clean transition, and the array is clean | |
2270 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2271 | if (nospares |
42543769 | 2272 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2273 | && mddev->can_decrease_events |
2274 | && mddev->events != 1) { | |
42543769 | 2275 | mddev->events--; |
a8707c08 N |
2276 | mddev->can_decrease_events = 0; |
2277 | } else { | |
42543769 N |
2278 | /* otherwise we have to go forward and ... */ |
2279 | mddev->events ++; | |
a8707c08 | 2280 | mddev->can_decrease_events = nospares; |
42543769 | 2281 | } |
1da177e4 | 2282 | |
403df478 N |
2283 | /* |
2284 | * This 64-bit counter should never wrap. | |
2285 | * Either we are in around ~1 trillion A.C., assuming | |
2286 | * 1 reboot per second, or we have a bug... | |
2287 | */ | |
2288 | WARN_ON(mddev->events == 0); | |
2699b672 | 2289 | |
dafb20fa | 2290 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2291 | if (rdev->badblocks.changed) |
2292 | any_badblocks_changed++; | |
de393cde N |
2293 | if (test_bit(Faulty, &rdev->flags)) |
2294 | set_bit(FaultRecorded, &rdev->flags); | |
2295 | } | |
2699b672 | 2296 | |
e691063a | 2297 | sync_sbs(mddev, nospares); |
85572d7c | 2298 | spin_unlock(&mddev->lock); |
1da177e4 | 2299 | |
36a4e1fe N |
2300 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2301 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2302 | |
4ad13663 | 2303 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2304 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2305 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2306 | |
42543769 N |
2307 | if (rdev->sb_loaded != 1) |
2308 | continue; /* no noise on spare devices */ | |
1da177e4 | 2309 | |
f466722c | 2310 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2311 | md_super_write(mddev,rdev, |
0f420358 | 2312 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2313 | rdev->sb_page); |
36a4e1fe N |
2314 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2315 | bdevname(rdev->bdev, b), | |
2316 | (unsigned long long)rdev->sb_start); | |
42543769 | 2317 | rdev->sb_events = mddev->events; |
2699b672 N |
2318 | if (rdev->badblocks.size) { |
2319 | md_super_write(mddev, rdev, | |
2320 | rdev->badblocks.sector, | |
2321 | rdev->badblocks.size << 9, | |
2322 | rdev->bb_page); | |
2323 | rdev->badblocks.size = 0; | |
2324 | } | |
7bfa19f2 | 2325 | |
f466722c | 2326 | } else |
36a4e1fe N |
2327 | pr_debug("md: %s (skipping faulty)\n", |
2328 | bdevname(rdev->bdev, b)); | |
d70ed2e4 | 2329 | |
7bfa19f2 | 2330 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2331 | /* only need to write one superblock... */ |
2332 | break; | |
2333 | } | |
a9701a30 | 2334 | md_super_wait(mddev); |
850b2b42 | 2335 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2336 | |
85572d7c | 2337 | spin_lock(&mddev->lock); |
850b2b42 N |
2338 | if (mddev->in_sync != sync_req || |
2339 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2340 | /* have to write it out again */ |
85572d7c | 2341 | spin_unlock(&mddev->lock); |
06d91a5f N |
2342 | goto repeat; |
2343 | } | |
850b2b42 | 2344 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
85572d7c | 2345 | spin_unlock(&mddev->lock); |
3d310eb7 | 2346 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2347 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2348 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2349 | |
dafb20fa | 2350 | rdev_for_each(rdev, mddev) { |
de393cde N |
2351 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2352 | clear_bit(Blocked, &rdev->flags); | |
2353 | ||
2354 | if (any_badblocks_changed) | |
2699b672 | 2355 | md_ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2356 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2357 | wake_up(&rdev->blocked_wait); | |
2358 | } | |
1da177e4 | 2359 | } |
1aee41f6 | 2360 | EXPORT_SYMBOL(md_update_sb); |
1da177e4 | 2361 | |
a6da4ef8 GR |
2362 | static int add_bound_rdev(struct md_rdev *rdev) |
2363 | { | |
2364 | struct mddev *mddev = rdev->mddev; | |
2365 | int err = 0; | |
2366 | ||
2367 | if (!mddev->pers->hot_remove_disk) { | |
2368 | /* If there is hot_add_disk but no hot_remove_disk | |
2369 | * then added disks for geometry changes, | |
2370 | * and should be added immediately. | |
2371 | */ | |
2372 | super_types[mddev->major_version]. | |
2373 | validate_super(mddev, rdev); | |
2374 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
2375 | if (err) { | |
2376 | unbind_rdev_from_array(rdev); | |
2377 | export_rdev(rdev); | |
2378 | return err; | |
2379 | } | |
2380 | } | |
2381 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2382 | ||
2383 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2384 | if (mddev->degraded) | |
2385 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
2386 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2387 | md_new_event(mddev); | |
2388 | md_wakeup_thread(mddev->thread); | |
2389 | return 0; | |
2390 | } | |
1da177e4 | 2391 | |
7f6ce769 | 2392 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2393 | * We want to accept with case. For this we use cmd_match. |
2394 | */ | |
2395 | static int cmd_match(const char *cmd, const char *str) | |
2396 | { | |
2397 | /* See if cmd, written into a sysfs file, matches | |
2398 | * str. They must either be the same, or cmd can | |
2399 | * have a trailing newline | |
2400 | */ | |
2401 | while (*cmd && *str && *cmd == *str) { | |
2402 | cmd++; | |
2403 | str++; | |
2404 | } | |
2405 | if (*cmd == '\n') | |
2406 | cmd++; | |
2407 | if (*str || *cmd) | |
2408 | return 0; | |
2409 | return 1; | |
2410 | } | |
2411 | ||
86e6ffdd N |
2412 | struct rdev_sysfs_entry { |
2413 | struct attribute attr; | |
3cb03002 N |
2414 | ssize_t (*show)(struct md_rdev *, char *); |
2415 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2416 | }; |
2417 | ||
2418 | static ssize_t | |
3cb03002 | 2419 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd N |
2420 | { |
2421 | char *sep = ""; | |
20a49ff6 | 2422 | size_t len = 0; |
758bfc8a | 2423 | unsigned long flags = ACCESS_ONCE(rdev->flags); |
86e6ffdd | 2424 | |
758bfc8a | 2425 | if (test_bit(Faulty, &flags) || |
de393cde | 2426 | rdev->badblocks.unacked_exist) { |
86e6ffdd N |
2427 | len+= sprintf(page+len, "%sfaulty",sep); |
2428 | sep = ","; | |
2429 | } | |
758bfc8a | 2430 | if (test_bit(In_sync, &flags)) { |
86e6ffdd N |
2431 | len += sprintf(page+len, "%sin_sync",sep); |
2432 | sep = ","; | |
2433 | } | |
758bfc8a | 2434 | if (test_bit(WriteMostly, &flags)) { |
f655675b N |
2435 | len += sprintf(page+len, "%swrite_mostly",sep); |
2436 | sep = ","; | |
2437 | } | |
758bfc8a | 2438 | if (test_bit(Blocked, &flags) || |
52c64152 | 2439 | (rdev->badblocks.unacked_exist |
758bfc8a | 2440 | && !test_bit(Faulty, &flags))) { |
6bfe0b49 DW |
2441 | len += sprintf(page+len, "%sblocked", sep); |
2442 | sep = ","; | |
2443 | } | |
758bfc8a N |
2444 | if (!test_bit(Faulty, &flags) && |
2445 | !test_bit(In_sync, &flags)) { | |
86e6ffdd N |
2446 | len += sprintf(page+len, "%sspare", sep); |
2447 | sep = ","; | |
2448 | } | |
758bfc8a | 2449 | if (test_bit(WriteErrorSeen, &flags)) { |
d7a9d443 N |
2450 | len += sprintf(page+len, "%swrite_error", sep); |
2451 | sep = ","; | |
2452 | } | |
758bfc8a | 2453 | if (test_bit(WantReplacement, &flags)) { |
2d78f8c4 N |
2454 | len += sprintf(page+len, "%swant_replacement", sep); |
2455 | sep = ","; | |
2456 | } | |
758bfc8a | 2457 | if (test_bit(Replacement, &flags)) { |
2d78f8c4 N |
2458 | len += sprintf(page+len, "%sreplacement", sep); |
2459 | sep = ","; | |
2460 | } | |
2461 | ||
86e6ffdd N |
2462 | return len+sprintf(page+len, "\n"); |
2463 | } | |
2464 | ||
45dc2de1 | 2465 | static ssize_t |
3cb03002 | 2466 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2467 | { |
2468 | /* can write | |
de393cde | 2469 | * faulty - simulates an error |
45dc2de1 | 2470 | * remove - disconnects the device |
f655675b N |
2471 | * writemostly - sets write_mostly |
2472 | * -writemostly - clears write_mostly | |
de393cde N |
2473 | * blocked - sets the Blocked flags |
2474 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2475 | * insync - sets Insync providing device isn't active |
f466722c N |
2476 | * -insync - clear Insync for a device with a slot assigned, |
2477 | * so that it gets rebuilt based on bitmap | |
d7a9d443 N |
2478 | * write_error - sets WriteErrorSeen |
2479 | * -write_error - clears WriteErrorSeen | |
45dc2de1 N |
2480 | */ |
2481 | int err = -EINVAL; | |
2482 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2483 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2484 | if (test_bit(Faulty, &rdev->flags)) |
2485 | err = 0; | |
2486 | else | |
2487 | err = -EBUSY; | |
45dc2de1 N |
2488 | } else if (cmd_match(buf, "remove")) { |
2489 | if (rdev->raid_disk >= 0) | |
2490 | err = -EBUSY; | |
2491 | else { | |
fd01b88c | 2492 | struct mddev *mddev = rdev->mddev; |
293467aa | 2493 | if (mddev_is_clustered(mddev)) |
88bcfef7 | 2494 | md_cluster_ops->remove_disk(mddev, rdev); |
fb56dfef | 2495 | md_kick_rdev_from_array(rdev); |
88bcfef7 GR |
2496 | if (mddev_is_clustered(mddev)) |
2497 | md_cluster_ops->metadata_update_start(mddev); | |
3f9d7b0d N |
2498 | if (mddev->pers) |
2499 | md_update_sb(mddev, 1); | |
45dc2de1 | 2500 | md_new_event(mddev); |
293467aa GR |
2501 | if (mddev_is_clustered(mddev)) |
2502 | md_cluster_ops->metadata_update_finish(mddev); | |
45dc2de1 N |
2503 | err = 0; |
2504 | } | |
f655675b N |
2505 | } else if (cmd_match(buf, "writemostly")) { |
2506 | set_bit(WriteMostly, &rdev->flags); | |
2507 | err = 0; | |
2508 | } else if (cmd_match(buf, "-writemostly")) { | |
2509 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2510 | err = 0; |
2511 | } else if (cmd_match(buf, "blocked")) { | |
2512 | set_bit(Blocked, &rdev->flags); | |
2513 | err = 0; | |
2514 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2515 | if (!test_bit(Faulty, &rdev->flags) && |
7da64a0a | 2516 | rdev->badblocks.unacked_exist) { |
de393cde N |
2517 | /* metadata handler doesn't understand badblocks, |
2518 | * so we need to fail the device | |
2519 | */ | |
2520 | md_error(rdev->mddev, rdev); | |
2521 | } | |
6bfe0b49 | 2522 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2523 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2524 | wake_up(&rdev->blocked_wait); |
2525 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2526 | md_wakeup_thread(rdev->mddev->thread); | |
2527 | ||
6d56e278 N |
2528 | err = 0; |
2529 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2530 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2531 | err = 0; |
f466722c | 2532 | } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) { |
e1960f8c N |
2533 | if (rdev->mddev->pers == NULL) { |
2534 | clear_bit(In_sync, &rdev->flags); | |
2535 | rdev->saved_raid_disk = rdev->raid_disk; | |
2536 | rdev->raid_disk = -1; | |
2537 | err = 0; | |
2538 | } | |
d7a9d443 N |
2539 | } else if (cmd_match(buf, "write_error")) { |
2540 | set_bit(WriteErrorSeen, &rdev->flags); | |
2541 | err = 0; | |
2542 | } else if (cmd_match(buf, "-write_error")) { | |
2543 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2544 | err = 0; | |
2d78f8c4 N |
2545 | } else if (cmd_match(buf, "want_replacement")) { |
2546 | /* Any non-spare device that is not a replacement can | |
2547 | * become want_replacement at any time, but we then need to | |
2548 | * check if recovery is needed. | |
2549 | */ | |
2550 | if (rdev->raid_disk >= 0 && | |
2551 | !test_bit(Replacement, &rdev->flags)) | |
2552 | set_bit(WantReplacement, &rdev->flags); | |
2553 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2554 | md_wakeup_thread(rdev->mddev->thread); | |
2555 | err = 0; | |
2556 | } else if (cmd_match(buf, "-want_replacement")) { | |
2557 | /* Clearing 'want_replacement' is always allowed. | |
2558 | * Once replacements starts it is too late though. | |
2559 | */ | |
2560 | err = 0; | |
2561 | clear_bit(WantReplacement, &rdev->flags); | |
2562 | } else if (cmd_match(buf, "replacement")) { | |
2563 | /* Can only set a device as a replacement when array has not | |
2564 | * yet been started. Once running, replacement is automatic | |
2565 | * from spares, or by assigning 'slot'. | |
2566 | */ | |
2567 | if (rdev->mddev->pers) | |
2568 | err = -EBUSY; | |
2569 | else { | |
2570 | set_bit(Replacement, &rdev->flags); | |
2571 | err = 0; | |
2572 | } | |
2573 | } else if (cmd_match(buf, "-replacement")) { | |
2574 | /* Similarly, can only clear Replacement before start */ | |
2575 | if (rdev->mddev->pers) | |
2576 | err = -EBUSY; | |
2577 | else { | |
2578 | clear_bit(Replacement, &rdev->flags); | |
2579 | err = 0; | |
2580 | } | |
a6da4ef8 GR |
2581 | } else if (cmd_match(buf, "re-add")) { |
2582 | if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) { | |
97f6cd39 GR |
2583 | /* clear_bit is performed _after_ all the devices |
2584 | * have their local Faulty bit cleared. If any writes | |
2585 | * happen in the meantime in the local node, they | |
2586 | * will land in the local bitmap, which will be synced | |
2587 | * by this node eventually | |
2588 | */ | |
2589 | if (!mddev_is_clustered(rdev->mddev) || | |
2590 | (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) { | |
2591 | clear_bit(Faulty, &rdev->flags); | |
2592 | err = add_bound_rdev(rdev); | |
2593 | } | |
a6da4ef8 GR |
2594 | } else |
2595 | err = -EBUSY; | |
45dc2de1 | 2596 | } |
00bcb4ac N |
2597 | if (!err) |
2598 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2599 | return err ? err : len; |
2600 | } | |
80ca3a44 | 2601 | static struct rdev_sysfs_entry rdev_state = |
750f199e | 2602 | __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store); |
86e6ffdd | 2603 | |
4dbcdc75 | 2604 | static ssize_t |
3cb03002 | 2605 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2606 | { |
2607 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2608 | } | |
2609 | ||
2610 | static ssize_t | |
3cb03002 | 2611 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 | 2612 | { |
4c9309c0 AD |
2613 | unsigned int n; |
2614 | int rv; | |
2615 | ||
2616 | rv = kstrtouint(buf, 10, &n); | |
2617 | if (rv < 0) | |
2618 | return rv; | |
2619 | atomic_set(&rdev->corrected_errors, n); | |
2620 | return len; | |
4dbcdc75 N |
2621 | } |
2622 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2623 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2624 | |
014236d2 | 2625 | static ssize_t |
3cb03002 | 2626 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 N |
2627 | { |
2628 | if (rdev->raid_disk < 0) | |
2629 | return sprintf(page, "none\n"); | |
2630 | else | |
2631 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2632 | } | |
2633 | ||
2634 | static ssize_t | |
3cb03002 | 2635 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 | 2636 | { |
4c9309c0 | 2637 | int slot; |
c303da6d | 2638 | int err; |
4c9309c0 | 2639 | |
014236d2 N |
2640 | if (strncmp(buf, "none", 4)==0) |
2641 | slot = -1; | |
4c9309c0 AD |
2642 | else { |
2643 | err = kstrtouint(buf, 10, (unsigned int *)&slot); | |
2644 | if (err < 0) | |
2645 | return err; | |
2646 | } | |
6c2fce2e | 2647 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2648 | /* Setting 'slot' on an active array requires also |
2649 | * updating the 'rd%d' link, and communicating | |
2650 | * with the personality with ->hot_*_disk. | |
2651 | * For now we only support removing | |
2652 | * failed/spare devices. This normally happens automatically, | |
2653 | * but not when the metadata is externally managed. | |
2654 | */ | |
c303da6d N |
2655 | if (rdev->raid_disk == -1) |
2656 | return -EEXIST; | |
2657 | /* personality does all needed checks */ | |
01393f3d | 2658 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2659 | return -EINVAL; |
746d3207 N |
2660 | clear_bit(Blocked, &rdev->flags); |
2661 | remove_and_add_spares(rdev->mddev, rdev); | |
2662 | if (rdev->raid_disk >= 0) | |
2663 | return -EBUSY; | |
c303da6d N |
2664 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2665 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2666 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2667 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2668 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2669 | */ |
2670 | ||
2671 | if (rdev->raid_disk != -1) | |
2672 | return -EBUSY; | |
2673 | ||
c6751b2b N |
2674 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2675 | return -EBUSY; | |
2676 | ||
6c2fce2e NB |
2677 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2678 | return -EINVAL; | |
2679 | ||
ba1b41b6 N |
2680 | if (slot >= rdev->mddev->raid_disks && |
2681 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2682 | return -ENOSPC; | |
2683 | ||
6c2fce2e NB |
2684 | rdev->raid_disk = slot; |
2685 | if (test_bit(In_sync, &rdev->flags)) | |
2686 | rdev->saved_raid_disk = slot; | |
2687 | else | |
2688 | rdev->saved_raid_disk = -1; | |
d30519fc | 2689 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2690 | clear_bit(Bitmap_sync, &rdev->flags); |
6c2fce2e NB |
2691 | err = rdev->mddev->pers-> |
2692 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2693 | if (err) { |
6c2fce2e | 2694 | rdev->raid_disk = -1; |
6c2fce2e | 2695 | return err; |
52664732 | 2696 | } else |
00bcb4ac | 2697 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
36fad858 | 2698 | if (sysfs_link_rdev(rdev->mddev, rdev)) |
00bcb4ac | 2699 | /* failure here is OK */; |
6c2fce2e | 2700 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2701 | } else { |
ba1b41b6 N |
2702 | if (slot >= rdev->mddev->raid_disks && |
2703 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2704 | return -ENOSPC; |
2705 | rdev->raid_disk = slot; | |
2706 | /* assume it is working */ | |
c5d79adb N |
2707 | clear_bit(Faulty, &rdev->flags); |
2708 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2709 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2710 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2711 | } |
014236d2 N |
2712 | return len; |
2713 | } | |
2714 | ||
014236d2 | 2715 | static struct rdev_sysfs_entry rdev_slot = |
80ca3a44 | 2716 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2717 | |
93c8cad0 | 2718 | static ssize_t |
3cb03002 | 2719 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 2720 | { |
6961ece4 | 2721 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2722 | } |
2723 | ||
2724 | static ssize_t | |
3cb03002 | 2725 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 2726 | { |
c6563a8c | 2727 | unsigned long long offset; |
b29bebd6 | 2728 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 2729 | return -EINVAL; |
8ed0a521 | 2730 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2731 | return -EBUSY; |
dd8ac336 | 2732 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2733 | /* Must set offset before size, so overlap checks |
2734 | * can be sane */ | |
2735 | return -EBUSY; | |
93c8cad0 | 2736 | rdev->data_offset = offset; |
25f7fd47 | 2737 | rdev->new_data_offset = offset; |
93c8cad0 N |
2738 | return len; |
2739 | } | |
2740 | ||
2741 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2742 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2743 | |
c6563a8c N |
2744 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
2745 | { | |
2746 | return sprintf(page, "%llu\n", | |
2747 | (unsigned long long)rdev->new_data_offset); | |
2748 | } | |
2749 | ||
2750 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
2751 | const char *buf, size_t len) | |
2752 | { | |
2753 | unsigned long long new_offset; | |
2754 | struct mddev *mddev = rdev->mddev; | |
2755 | ||
b29bebd6 | 2756 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
2757 | return -EINVAL; |
2758 | ||
f851b60d N |
2759 | if (mddev->sync_thread || |
2760 | test_bit(MD_RECOVERY_RUNNING,&mddev->recovery)) | |
c6563a8c N |
2761 | return -EBUSY; |
2762 | if (new_offset == rdev->data_offset) | |
2763 | /* reset is always permitted */ | |
2764 | ; | |
2765 | else if (new_offset > rdev->data_offset) { | |
2766 | /* must not push array size beyond rdev_sectors */ | |
2767 | if (new_offset - rdev->data_offset | |
2768 | + mddev->dev_sectors > rdev->sectors) | |
2769 | return -E2BIG; | |
2770 | } | |
2771 | /* Metadata worries about other space details. */ | |
2772 | ||
2773 | /* decreasing the offset is inconsistent with a backwards | |
2774 | * reshape. | |
2775 | */ | |
2776 | if (new_offset < rdev->data_offset && | |
2777 | mddev->reshape_backwards) | |
2778 | return -EINVAL; | |
2779 | /* Increasing offset is inconsistent with forwards | |
2780 | * reshape. reshape_direction should be set to | |
2781 | * 'backwards' first. | |
2782 | */ | |
2783 | if (new_offset > rdev->data_offset && | |
2784 | !mddev->reshape_backwards) | |
2785 | return -EINVAL; | |
2786 | ||
2787 | if (mddev->pers && mddev->persistent && | |
2788 | !super_types[mddev->major_version] | |
2789 | .allow_new_offset(rdev, new_offset)) | |
2790 | return -E2BIG; | |
2791 | rdev->new_data_offset = new_offset; | |
2792 | if (new_offset > rdev->data_offset) | |
2793 | mddev->reshape_backwards = 1; | |
2794 | else if (new_offset < rdev->data_offset) | |
2795 | mddev->reshape_backwards = 0; | |
2796 | ||
2797 | return len; | |
2798 | } | |
2799 | static struct rdev_sysfs_entry rdev_new_offset = | |
2800 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
2801 | ||
83303b61 | 2802 | static ssize_t |
3cb03002 | 2803 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 2804 | { |
dd8ac336 | 2805 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2806 | } |
2807 | ||
c5d79adb N |
2808 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2809 | { | |
2810 | /* check if two start/length pairs overlap */ | |
2811 | if (s1+l1 <= s2) | |
2812 | return 0; | |
2813 | if (s2+l2 <= s1) | |
2814 | return 0; | |
2815 | return 1; | |
2816 | } | |
2817 | ||
b522adcd DW |
2818 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2819 | { | |
2820 | unsigned long long blocks; | |
2821 | sector_t new; | |
2822 | ||
b29bebd6 | 2823 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
2824 | return -EINVAL; |
2825 | ||
2826 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2827 | return -EINVAL; /* sector conversion overflow */ | |
2828 | ||
2829 | new = blocks * 2; | |
2830 | if (new != blocks * 2) | |
2831 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2832 | ||
2833 | *sectors = new; | |
2834 | return 0; | |
2835 | } | |
2836 | ||
83303b61 | 2837 | static ssize_t |
3cb03002 | 2838 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 2839 | { |
fd01b88c | 2840 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 2841 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2842 | sector_t sectors; |
27c529bb | 2843 | |
b522adcd | 2844 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2845 | return -EINVAL; |
c6563a8c N |
2846 | if (rdev->data_offset != rdev->new_data_offset) |
2847 | return -EINVAL; /* too confusing */ | |
0cd17fec | 2848 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2849 | if (my_mddev->persistent) { |
dd8ac336 AN |
2850 | sectors = super_types[my_mddev->major_version]. |
2851 | rdev_size_change(rdev, sectors); | |
2852 | if (!sectors) | |
0cd17fec | 2853 | return -EBUSY; |
dd8ac336 | 2854 | } else if (!sectors) |
77304d2a | 2855 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 2856 | rdev->data_offset; |
a6468539 N |
2857 | if (!my_mddev->pers->resize) |
2858 | /* Cannot change size for RAID0 or Linear etc */ | |
2859 | return -EINVAL; | |
0cd17fec | 2860 | } |
dd8ac336 | 2861 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2862 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2863 | |
dd8ac336 AN |
2864 | rdev->sectors = sectors; |
2865 | if (sectors > oldsectors && my_mddev->external) { | |
8b1afc3d N |
2866 | /* Need to check that all other rdevs with the same |
2867 | * ->bdev do not overlap. 'rcu' is sufficient to walk | |
2868 | * the rdev lists safely. | |
2869 | * This check does not provide a hard guarantee, it | |
2870 | * just helps avoid dangerous mistakes. | |
c5d79adb | 2871 | */ |
fd01b88c | 2872 | struct mddev *mddev; |
c5d79adb | 2873 | int overlap = 0; |
159ec1fc | 2874 | struct list_head *tmp; |
c5d79adb | 2875 | |
8b1afc3d | 2876 | rcu_read_lock(); |
29ac4aa3 | 2877 | for_each_mddev(mddev, tmp) { |
3cb03002 | 2878 | struct md_rdev *rdev2; |
c5d79adb | 2879 | |
dafb20fa | 2880 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
2881 | if (rdev->bdev == rdev2->bdev && |
2882 | rdev != rdev2 && | |
2883 | overlaps(rdev->data_offset, rdev->sectors, | |
2884 | rdev2->data_offset, | |
2885 | rdev2->sectors)) { | |
c5d79adb N |
2886 | overlap = 1; |
2887 | break; | |
2888 | } | |
c5d79adb N |
2889 | if (overlap) { |
2890 | mddev_put(mddev); | |
2891 | break; | |
2892 | } | |
2893 | } | |
8b1afc3d | 2894 | rcu_read_unlock(); |
c5d79adb N |
2895 | if (overlap) { |
2896 | /* Someone else could have slipped in a size | |
2897 | * change here, but doing so is just silly. | |
dd8ac336 | 2898 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2899 | * safe, and trust userspace not to race with |
2900 | * itself | |
2901 | */ | |
dd8ac336 | 2902 | rdev->sectors = oldsectors; |
c5d79adb N |
2903 | return -EBUSY; |
2904 | } | |
2905 | } | |
83303b61 N |
2906 | return len; |
2907 | } | |
2908 | ||
2909 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2910 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2911 | |
3cb03002 | 2912 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
2913 | { |
2914 | unsigned long long recovery_start = rdev->recovery_offset; | |
2915 | ||
2916 | if (test_bit(In_sync, &rdev->flags) || | |
2917 | recovery_start == MaxSector) | |
2918 | return sprintf(page, "none\n"); | |
2919 | ||
2920 | return sprintf(page, "%llu\n", recovery_start); | |
2921 | } | |
2922 | ||
3cb03002 | 2923 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
2924 | { |
2925 | unsigned long long recovery_start; | |
2926 | ||
2927 | if (cmd_match(buf, "none")) | |
2928 | recovery_start = MaxSector; | |
b29bebd6 | 2929 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
2930 | return -EINVAL; |
2931 | ||
2932 | if (rdev->mddev->pers && | |
2933 | rdev->raid_disk >= 0) | |
2934 | return -EBUSY; | |
2935 | ||
2936 | rdev->recovery_offset = recovery_start; | |
2937 | if (recovery_start == MaxSector) | |
2938 | set_bit(In_sync, &rdev->flags); | |
2939 | else | |
2940 | clear_bit(In_sync, &rdev->flags); | |
2941 | return len; | |
2942 | } | |
2943 | ||
2944 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2945 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2946 | ||
16c791a5 N |
2947 | static ssize_t |
2948 | badblocks_show(struct badblocks *bb, char *page, int unack); | |
2949 | static ssize_t | |
2950 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack); | |
2951 | ||
3cb03002 | 2952 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
2953 | { |
2954 | return badblocks_show(&rdev->badblocks, page, 0); | |
2955 | } | |
3cb03002 | 2956 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 2957 | { |
de393cde N |
2958 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
2959 | /* Maybe that ack was all we needed */ | |
2960 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
2961 | wake_up(&rdev->blocked_wait); | |
2962 | return rv; | |
16c791a5 N |
2963 | } |
2964 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
2965 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
2966 | ||
3cb03002 | 2967 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
2968 | { |
2969 | return badblocks_show(&rdev->badblocks, page, 1); | |
2970 | } | |
3cb03002 | 2971 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
2972 | { |
2973 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
2974 | } | |
2975 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
2976 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
2977 | ||
86e6ffdd N |
2978 | static struct attribute *rdev_default_attrs[] = { |
2979 | &rdev_state.attr, | |
4dbcdc75 | 2980 | &rdev_errors.attr, |
014236d2 | 2981 | &rdev_slot.attr, |
93c8cad0 | 2982 | &rdev_offset.attr, |
c6563a8c | 2983 | &rdev_new_offset.attr, |
83303b61 | 2984 | &rdev_size.attr, |
06e3c817 | 2985 | &rdev_recovery_start.attr, |
16c791a5 N |
2986 | &rdev_bad_blocks.attr, |
2987 | &rdev_unack_bad_blocks.attr, | |
86e6ffdd N |
2988 | NULL, |
2989 | }; | |
2990 | static ssize_t | |
2991 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2992 | { | |
2993 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 2994 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
86e6ffdd N |
2995 | |
2996 | if (!entry->show) | |
2997 | return -EIO; | |
758bfc8a N |
2998 | if (!rdev->mddev) |
2999 | return -EBUSY; | |
3000 | return entry->show(rdev, page); | |
86e6ffdd N |
3001 | } |
3002 | ||
3003 | static ssize_t | |
3004 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
3005 | const char *page, size_t length) | |
3006 | { | |
3007 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3008 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 3009 | ssize_t rv; |
fd01b88c | 3010 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
3011 | |
3012 | if (!entry->store) | |
3013 | return -EIO; | |
67463acb N |
3014 | if (!capable(CAP_SYS_ADMIN)) |
3015 | return -EACCES; | |
27c529bb | 3016 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 3017 | if (!rv) { |
27c529bb N |
3018 | if (rdev->mddev == NULL) |
3019 | rv = -EBUSY; | |
3020 | else | |
3021 | rv = entry->store(rdev, page, length); | |
6a51830e | 3022 | mddev_unlock(mddev); |
ca388059 N |
3023 | } |
3024 | return rv; | |
86e6ffdd N |
3025 | } |
3026 | ||
3027 | static void rdev_free(struct kobject *ko) | |
3028 | { | |
3cb03002 | 3029 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
3030 | kfree(rdev); |
3031 | } | |
52cf25d0 | 3032 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
3033 | .show = rdev_attr_show, |
3034 | .store = rdev_attr_store, | |
3035 | }; | |
3036 | static struct kobj_type rdev_ktype = { | |
3037 | .release = rdev_free, | |
3038 | .sysfs_ops = &rdev_sysfs_ops, | |
3039 | .default_attrs = rdev_default_attrs, | |
3040 | }; | |
3041 | ||
3cb03002 | 3042 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
3043 | { |
3044 | rdev->desc_nr = -1; | |
3045 | rdev->saved_raid_disk = -1; | |
3046 | rdev->raid_disk = -1; | |
3047 | rdev->flags = 0; | |
3048 | rdev->data_offset = 0; | |
c6563a8c | 3049 | rdev->new_data_offset = 0; |
e8bb9a83 N |
3050 | rdev->sb_events = 0; |
3051 | rdev->last_read_error.tv_sec = 0; | |
3052 | rdev->last_read_error.tv_nsec = 0; | |
2699b672 N |
3053 | rdev->sb_loaded = 0; |
3054 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3055 | atomic_set(&rdev->nr_pending, 0); |
3056 | atomic_set(&rdev->read_errors, 0); | |
3057 | atomic_set(&rdev->corrected_errors, 0); | |
3058 | ||
3059 | INIT_LIST_HEAD(&rdev->same_set); | |
3060 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3061 | |
3062 | /* Add space to store bad block list. | |
3063 | * This reserves the space even on arrays where it cannot | |
3064 | * be used - I wonder if that matters | |
3065 | */ | |
3066 | rdev->badblocks.count = 0; | |
486adf72 | 3067 | rdev->badblocks.shift = -1; /* disabled until explicitly enabled */ |
2230dfe4 N |
3068 | rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3069 | seqlock_init(&rdev->badblocks.lock); | |
3070 | if (rdev->badblocks.page == NULL) | |
3071 | return -ENOMEM; | |
3072 | ||
3073 | return 0; | |
e8bb9a83 N |
3074 | } |
3075 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3076 | /* |
3077 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3078 | * | |
3079 | * mark the device faulty if: | |
3080 | * | |
3081 | * - the device is nonexistent (zero size) | |
3082 | * - the device has no valid superblock | |
3083 | * | |
3084 | * a faulty rdev _never_ has rdev->sb set. | |
3085 | */ | |
3cb03002 | 3086 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3087 | { |
3088 | char b[BDEVNAME_SIZE]; | |
3089 | int err; | |
3cb03002 | 3090 | struct md_rdev *rdev; |
1da177e4 LT |
3091 | sector_t size; |
3092 | ||
9ffae0cf | 3093 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
3094 | if (!rdev) { |
3095 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
3096 | return ERR_PTR(-ENOMEM); | |
3097 | } | |
1da177e4 | 3098 | |
2230dfe4 N |
3099 | err = md_rdev_init(rdev); |
3100 | if (err) | |
3101 | goto abort_free; | |
3102 | err = alloc_disk_sb(rdev); | |
3103 | if (err) | |
1da177e4 LT |
3104 | goto abort_free; |
3105 | ||
c5d79adb | 3106 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3107 | if (err) |
3108 | goto abort_free; | |
3109 | ||
f9cb074b | 3110 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3111 | |
77304d2a | 3112 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 | 3113 | if (!size) { |
f72ffdd6 | 3114 | printk(KERN_WARNING |
1da177e4 LT |
3115 | "md: %s has zero or unknown size, marking faulty!\n", |
3116 | bdevname(rdev->bdev,b)); | |
3117 | err = -EINVAL; | |
3118 | goto abort_free; | |
3119 | } | |
3120 | ||
3121 | if (super_format >= 0) { | |
3122 | err = super_types[super_format]. | |
3123 | load_super(rdev, NULL, super_minor); | |
3124 | if (err == -EINVAL) { | |
df968c4e N |
3125 | printk(KERN_WARNING |
3126 | "md: %s does not have a valid v%d.%d " | |
3127 | "superblock, not importing!\n", | |
3128 | bdevname(rdev->bdev,b), | |
3129 | super_format, super_minor); | |
1da177e4 LT |
3130 | goto abort_free; |
3131 | } | |
3132 | if (err < 0) { | |
f72ffdd6 | 3133 | printk(KERN_WARNING |
1da177e4 LT |
3134 | "md: could not read %s's sb, not importing!\n", |
3135 | bdevname(rdev->bdev,b)); | |
3136 | goto abort_free; | |
3137 | } | |
3138 | } | |
6bfe0b49 | 3139 | |
1da177e4 LT |
3140 | return rdev; |
3141 | ||
3142 | abort_free: | |
2699b672 N |
3143 | if (rdev->bdev) |
3144 | unlock_rdev(rdev); | |
545c8795 | 3145 | md_rdev_clear(rdev); |
1da177e4 LT |
3146 | kfree(rdev); |
3147 | return ERR_PTR(err); | |
3148 | } | |
3149 | ||
3150 | /* | |
3151 | * Check a full RAID array for plausibility | |
3152 | */ | |
3153 | ||
f72ffdd6 | 3154 | static void analyze_sbs(struct mddev *mddev) |
1da177e4 LT |
3155 | { |
3156 | int i; | |
3cb03002 | 3157 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3158 | char b[BDEVNAME_SIZE]; |
3159 | ||
3160 | freshest = NULL; | |
dafb20fa | 3161 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3162 | switch (super_types[mddev->major_version]. |
3163 | load_super(rdev, freshest, mddev->minor_version)) { | |
3164 | case 1: | |
3165 | freshest = rdev; | |
3166 | break; | |
3167 | case 0: | |
3168 | break; | |
3169 | default: | |
3170 | printk( KERN_ERR \ | |
3171 | "md: fatal superblock inconsistency in %s" | |
f72ffdd6 | 3172 | " -- removing from array\n", |
1da177e4 | 3173 | bdevname(rdev->bdev,b)); |
fb56dfef | 3174 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3175 | } |
3176 | ||
1da177e4 LT |
3177 | super_types[mddev->major_version]. |
3178 | validate_super(mddev, freshest); | |
3179 | ||
3180 | i = 0; | |
dafb20fa | 3181 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3182 | if (mddev->max_disks && |
3183 | (rdev->desc_nr >= mddev->max_disks || | |
3184 | i > mddev->max_disks)) { | |
de01dfad N |
3185 | printk(KERN_WARNING |
3186 | "md: %s: %s: only %d devices permitted\n", | |
3187 | mdname(mddev), bdevname(rdev->bdev, b), | |
3188 | mddev->max_disks); | |
fb56dfef | 3189 | md_kick_rdev_from_array(rdev); |
de01dfad N |
3190 | continue; |
3191 | } | |
1aee41f6 | 3192 | if (rdev != freshest) { |
1da177e4 LT |
3193 | if (super_types[mddev->major_version]. |
3194 | validate_super(mddev, rdev)) { | |
3195 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
3196 | " from array!\n", | |
3197 | bdevname(rdev->bdev,b)); | |
fb56dfef | 3198 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3199 | continue; |
3200 | } | |
1aee41f6 GR |
3201 | /* No device should have a Candidate flag |
3202 | * when reading devices | |
3203 | */ | |
3204 | if (test_bit(Candidate, &rdev->flags)) { | |
3205 | pr_info("md: kicking Cluster Candidate %s from array!\n", | |
3206 | bdevname(rdev->bdev, b)); | |
fb56dfef | 3207 | md_kick_rdev_from_array(rdev); |
1aee41f6 GR |
3208 | } |
3209 | } | |
1da177e4 LT |
3210 | if (mddev->level == LEVEL_MULTIPATH) { |
3211 | rdev->desc_nr = i++; | |
3212 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3213 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 3214 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
3215 | rdev->raid_disk = -1; |
3216 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3217 | } |
3218 | } | |
1da177e4 LT |
3219 | } |
3220 | ||
72e02075 N |
3221 | /* Read a fixed-point number. |
3222 | * Numbers in sysfs attributes should be in "standard" units where | |
3223 | * possible, so time should be in seconds. | |
f72ffdd6 | 3224 | * However we internally use a a much smaller unit such as |
72e02075 N |
3225 | * milliseconds or jiffies. |
3226 | * This function takes a decimal number with a possible fractional | |
3227 | * component, and produces an integer which is the result of | |
3228 | * multiplying that number by 10^'scale'. | |
3229 | * all without any floating-point arithmetic. | |
3230 | */ | |
3231 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3232 | { | |
3233 | unsigned long result = 0; | |
3234 | long decimals = -1; | |
3235 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3236 | if (*cp == '.') | |
3237 | decimals = 0; | |
3238 | else if (decimals < scale) { | |
3239 | unsigned int value; | |
3240 | value = *cp - '0'; | |
3241 | result = result * 10 + value; | |
3242 | if (decimals >= 0) | |
3243 | decimals++; | |
3244 | } | |
3245 | cp++; | |
3246 | } | |
3247 | if (*cp == '\n') | |
3248 | cp++; | |
3249 | if (*cp) | |
3250 | return -EINVAL; | |
3251 | if (decimals < 0) | |
3252 | decimals = 0; | |
3253 | while (decimals < scale) { | |
3254 | result *= 10; | |
3255 | decimals ++; | |
3256 | } | |
3257 | *res = result; | |
3258 | return 0; | |
3259 | } | |
3260 | ||
16f17b39 | 3261 | static ssize_t |
fd01b88c | 3262 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3263 | { |
3264 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3265 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3266 | } | |
3267 | static ssize_t | |
fd01b88c | 3268 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3269 | { |
16f17b39 | 3270 | unsigned long msec; |
97ce0a7f | 3271 | |
72e02075 | 3272 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3273 | return -EINVAL; |
16f17b39 N |
3274 | if (msec == 0) |
3275 | mddev->safemode_delay = 0; | |
3276 | else { | |
19052c0e | 3277 | unsigned long old_delay = mddev->safemode_delay; |
1b30e66f N |
3278 | unsigned long new_delay = (msec*HZ)/1000; |
3279 | ||
3280 | if (new_delay == 0) | |
3281 | new_delay = 1; | |
3282 | mddev->safemode_delay = new_delay; | |
3283 | if (new_delay < old_delay || old_delay == 0) | |
3284 | mod_timer(&mddev->safemode_timer, jiffies+1); | |
16f17b39 N |
3285 | } |
3286 | return len; | |
3287 | } | |
3288 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3289 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3290 | |
eae1701f | 3291 | static ssize_t |
fd01b88c | 3292 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3293 | { |
36d091f4 N |
3294 | struct md_personality *p; |
3295 | int ret; | |
3296 | spin_lock(&mddev->lock); | |
3297 | p = mddev->pers; | |
d9d166c2 | 3298 | if (p) |
36d091f4 | 3299 | ret = sprintf(page, "%s\n", p->name); |
d9d166c2 | 3300 | else if (mddev->clevel[0]) |
36d091f4 | 3301 | ret = sprintf(page, "%s\n", mddev->clevel); |
d9d166c2 | 3302 | else if (mddev->level != LEVEL_NONE) |
36d091f4 | 3303 | ret = sprintf(page, "%d\n", mddev->level); |
d9d166c2 | 3304 | else |
36d091f4 N |
3305 | ret = 0; |
3306 | spin_unlock(&mddev->lock); | |
3307 | return ret; | |
eae1701f N |
3308 | } |
3309 | ||
d9d166c2 | 3310 | static ssize_t |
fd01b88c | 3311 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3312 | { |
f2859af6 | 3313 | char clevel[16]; |
6791875e N |
3314 | ssize_t rv; |
3315 | size_t slen = len; | |
db721d32 | 3316 | struct md_personality *pers, *oldpers; |
f2859af6 | 3317 | long level; |
db721d32 | 3318 | void *priv, *oldpriv; |
3cb03002 | 3319 | struct md_rdev *rdev; |
245f46c2 | 3320 | |
6791875e N |
3321 | if (slen == 0 || slen >= sizeof(clevel)) |
3322 | return -EINVAL; | |
3323 | ||
3324 | rv = mddev_lock(mddev); | |
3325 | if (rv) | |
3326 | return rv; | |
3327 | ||
245f46c2 | 3328 | if (mddev->pers == NULL) { |
6791875e N |
3329 | strncpy(mddev->clevel, buf, slen); |
3330 | if (mddev->clevel[slen-1] == '\n') | |
3331 | slen--; | |
3332 | mddev->clevel[slen] = 0; | |
245f46c2 | 3333 | mddev->level = LEVEL_NONE; |
6791875e N |
3334 | rv = len; |
3335 | goto out_unlock; | |
245f46c2 | 3336 | } |
6791875e | 3337 | rv = -EROFS; |
bd8839e0 | 3338 | if (mddev->ro) |
6791875e | 3339 | goto out_unlock; |
245f46c2 N |
3340 | |
3341 | /* request to change the personality. Need to ensure: | |
3342 | * - array is not engaged in resync/recovery/reshape | |
3343 | * - old personality can be suspended | |
3344 | * - new personality will access other array. | |
3345 | */ | |
3346 | ||
6791875e | 3347 | rv = -EBUSY; |
bb4f1e9d | 3348 | if (mddev->sync_thread || |
f851b60d | 3349 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
bb4f1e9d N |
3350 | mddev->reshape_position != MaxSector || |
3351 | mddev->sysfs_active) | |
6791875e | 3352 | goto out_unlock; |
245f46c2 | 3353 | |
6791875e | 3354 | rv = -EINVAL; |
245f46c2 N |
3355 | if (!mddev->pers->quiesce) { |
3356 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3357 | mdname(mddev), mddev->pers->name); | |
6791875e | 3358 | goto out_unlock; |
245f46c2 N |
3359 | } |
3360 | ||
3361 | /* Now find the new personality */ | |
6791875e N |
3362 | strncpy(clevel, buf, slen); |
3363 | if (clevel[slen-1] == '\n') | |
3364 | slen--; | |
3365 | clevel[slen] = 0; | |
b29bebd6 | 3366 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3367 | level = LEVEL_NONE; |
245f46c2 | 3368 | |
f2859af6 DW |
3369 | if (request_module("md-%s", clevel) != 0) |
3370 | request_module("md-level-%s", clevel); | |
245f46c2 | 3371 | spin_lock(&pers_lock); |
f2859af6 | 3372 | pers = find_pers(level, clevel); |
245f46c2 N |
3373 | if (!pers || !try_module_get(pers->owner)) { |
3374 | spin_unlock(&pers_lock); | |
f2859af6 | 3375 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
6791875e N |
3376 | rv = -EINVAL; |
3377 | goto out_unlock; | |
245f46c2 N |
3378 | } |
3379 | spin_unlock(&pers_lock); | |
3380 | ||
3381 | if (pers == mddev->pers) { | |
3382 | /* Nothing to do! */ | |
3383 | module_put(pers->owner); | |
6791875e N |
3384 | rv = len; |
3385 | goto out_unlock; | |
245f46c2 N |
3386 | } |
3387 | if (!pers->takeover) { | |
3388 | module_put(pers->owner); | |
3389 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3390 | mdname(mddev), clevel); |
6791875e N |
3391 | rv = -EINVAL; |
3392 | goto out_unlock; | |
245f46c2 N |
3393 | } |
3394 | ||
dafb20fa | 3395 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3396 | rdev->new_raid_disk = rdev->raid_disk; |
3397 | ||
245f46c2 N |
3398 | /* ->takeover must set new_* and/or delta_disks |
3399 | * if it succeeds, and may set them when it fails. | |
3400 | */ | |
3401 | priv = pers->takeover(mddev); | |
3402 | if (IS_ERR(priv)) { | |
3403 | mddev->new_level = mddev->level; | |
3404 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3405 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3406 | mddev->raid_disks -= mddev->delta_disks; |
3407 | mddev->delta_disks = 0; | |
2c810cdd | 3408 | mddev->reshape_backwards = 0; |
245f46c2 N |
3409 | module_put(pers->owner); |
3410 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3411 | mdname(mddev), clevel); |
6791875e N |
3412 | rv = PTR_ERR(priv); |
3413 | goto out_unlock; | |
245f46c2 N |
3414 | } |
3415 | ||
3416 | /* Looks like we have a winner */ | |
3417 | mddev_suspend(mddev); | |
5aa61f42 | 3418 | mddev_detach(mddev); |
36d091f4 N |
3419 | |
3420 | spin_lock(&mddev->lock); | |
db721d32 N |
3421 | oldpers = mddev->pers; |
3422 | oldpriv = mddev->private; | |
3423 | mddev->pers = pers; | |
3424 | mddev->private = priv; | |
3425 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3426 | mddev->level = mddev->new_level; | |
3427 | mddev->layout = mddev->new_layout; | |
3428 | mddev->chunk_sectors = mddev->new_chunk_sectors; | |
3429 | mddev->delta_disks = 0; | |
3430 | mddev->reshape_backwards = 0; | |
3431 | mddev->degraded = 0; | |
36d091f4 | 3432 | spin_unlock(&mddev->lock); |
db721d32 N |
3433 | |
3434 | if (oldpers->sync_request == NULL && | |
3435 | mddev->external) { | |
3436 | /* We are converting from a no-redundancy array | |
3437 | * to a redundancy array and metadata is managed | |
3438 | * externally so we need to be sure that writes | |
3439 | * won't block due to a need to transition | |
3440 | * clean->dirty | |
3441 | * until external management is started. | |
3442 | */ | |
3443 | mddev->in_sync = 0; | |
3444 | mddev->safemode_delay = 0; | |
3445 | mddev->safemode = 0; | |
3446 | } | |
f72ffdd6 | 3447 | |
db721d32 N |
3448 | oldpers->free(mddev, oldpriv); |
3449 | ||
3450 | if (oldpers->sync_request == NULL && | |
a64c876f N |
3451 | pers->sync_request != NULL) { |
3452 | /* need to add the md_redundancy_group */ | |
3453 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3454 | printk(KERN_WARNING | |
3455 | "md: cannot register extra attributes for %s\n", | |
3456 | mdname(mddev)); | |
388975cc | 3457 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
f72ffdd6 | 3458 | } |
db721d32 | 3459 | if (oldpers->sync_request != NULL && |
a64c876f N |
3460 | pers->sync_request == NULL) { |
3461 | /* need to remove the md_redundancy_group */ | |
3462 | if (mddev->to_remove == NULL) | |
3463 | mddev->to_remove = &md_redundancy_group; | |
3464 | } | |
3465 | ||
dafb20fa | 3466 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3467 | if (rdev->raid_disk < 0) |
3468 | continue; | |
bf2cb0da | 3469 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3470 | rdev->new_raid_disk = -1; |
3471 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3472 | continue; | |
36fad858 | 3473 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3474 | } |
dafb20fa | 3475 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3476 | if (rdev->raid_disk < 0) |
3477 | continue; | |
3478 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3479 | continue; | |
3480 | rdev->raid_disk = rdev->new_raid_disk; | |
3481 | if (rdev->raid_disk < 0) | |
3a981b03 | 3482 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3483 | else { |
36fad858 NK |
3484 | if (sysfs_link_rdev(mddev, rdev)) |
3485 | printk(KERN_WARNING "md: cannot register rd%d" | |
3486 | " for %s after level change\n", | |
3487 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3488 | } |
e93f68a1 N |
3489 | } |
3490 | ||
db721d32 | 3491 | if (pers->sync_request == NULL) { |
9af204cf TM |
3492 | /* this is now an array without redundancy, so |
3493 | * it must always be in_sync | |
3494 | */ | |
3495 | mddev->in_sync = 1; | |
3496 | del_timer_sync(&mddev->safemode_timer); | |
3497 | } | |
02e5f5c0 | 3498 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3499 | pers->run(mddev); |
245f46c2 | 3500 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
47525e59 | 3501 | mddev_resume(mddev); |
830778a1 N |
3502 | if (!mddev->thread) |
3503 | md_update_sb(mddev, 1); | |
5cac7861 | 3504 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3505 | md_new_event(mddev); |
6791875e N |
3506 | rv = len; |
3507 | out_unlock: | |
3508 | mddev_unlock(mddev); | |
d9d166c2 N |
3509 | return rv; |
3510 | } | |
3511 | ||
3512 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3513 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3514 | |
d4dbd025 | 3515 | static ssize_t |
fd01b88c | 3516 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3517 | { |
3518 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3519 | if (mddev->reshape_position != MaxSector && |
3520 | mddev->layout != mddev->new_layout) | |
3521 | return sprintf(page, "%d (%d)\n", | |
3522 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3523 | return sprintf(page, "%d\n", mddev->layout); |
3524 | } | |
3525 | ||
3526 | static ssize_t | |
fd01b88c | 3527 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 | 3528 | { |
4c9309c0 | 3529 | unsigned int n; |
6791875e | 3530 | int err; |
d4dbd025 | 3531 | |
4c9309c0 AD |
3532 | err = kstrtouint(buf, 10, &n); |
3533 | if (err < 0) | |
3534 | return err; | |
6791875e N |
3535 | err = mddev_lock(mddev); |
3536 | if (err) | |
3537 | return err; | |
d4dbd025 | 3538 | |
b3546035 | 3539 | if (mddev->pers) { |
50ac168a | 3540 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3541 | err = -EBUSY; |
3542 | else if (mddev->ro) | |
3543 | err = -EROFS; | |
3544 | else { | |
3545 | mddev->new_layout = n; | |
3546 | err = mddev->pers->check_reshape(mddev); | |
3547 | if (err) | |
3548 | mddev->new_layout = mddev->layout; | |
597a711b | 3549 | } |
b3546035 | 3550 | } else { |
08a02ecd | 3551 | mddev->new_layout = n; |
b3546035 N |
3552 | if (mddev->reshape_position == MaxSector) |
3553 | mddev->layout = n; | |
3554 | } | |
6791875e N |
3555 | mddev_unlock(mddev); |
3556 | return err ?: len; | |
d4dbd025 N |
3557 | } |
3558 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3559 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 | 3560 | |
eae1701f | 3561 | static ssize_t |
fd01b88c | 3562 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3563 | { |
bb636547 N |
3564 | if (mddev->raid_disks == 0) |
3565 | return 0; | |
08a02ecd N |
3566 | if (mddev->reshape_position != MaxSector && |
3567 | mddev->delta_disks != 0) | |
3568 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3569 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3570 | return sprintf(page, "%d\n", mddev->raid_disks); |
3571 | } | |
3572 | ||
fd01b88c | 3573 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3574 | |
3575 | static ssize_t | |
fd01b88c | 3576 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3577 | { |
4c9309c0 | 3578 | unsigned int n; |
6791875e | 3579 | int err; |
da943b99 | 3580 | |
4c9309c0 AD |
3581 | err = kstrtouint(buf, 10, &n); |
3582 | if (err < 0) | |
3583 | return err; | |
da943b99 | 3584 | |
6791875e N |
3585 | err = mddev_lock(mddev); |
3586 | if (err) | |
3587 | return err; | |
da943b99 | 3588 | if (mddev->pers) |
6791875e | 3589 | err = update_raid_disks(mddev, n); |
08a02ecd | 3590 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3591 | struct md_rdev *rdev; |
08a02ecd | 3592 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c | 3593 | |
6791875e | 3594 | err = -EINVAL; |
c6563a8c N |
3595 | rdev_for_each(rdev, mddev) { |
3596 | if (olddisks < n && | |
3597 | rdev->data_offset < rdev->new_data_offset) | |
6791875e | 3598 | goto out_unlock; |
c6563a8c N |
3599 | if (olddisks > n && |
3600 | rdev->data_offset > rdev->new_data_offset) | |
6791875e | 3601 | goto out_unlock; |
c6563a8c | 3602 | } |
6791875e | 3603 | err = 0; |
08a02ecd N |
3604 | mddev->delta_disks = n - olddisks; |
3605 | mddev->raid_disks = n; | |
2c810cdd | 3606 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3607 | } else |
da943b99 | 3608 | mddev->raid_disks = n; |
6791875e N |
3609 | out_unlock: |
3610 | mddev_unlock(mddev); | |
3611 | return err ? err : len; | |
da943b99 N |
3612 | } |
3613 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3614 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3615 | |
3b34380a | 3616 | static ssize_t |
fd01b88c | 3617 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3618 | { |
08a02ecd | 3619 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3620 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3621 | return sprintf(page, "%d (%d)\n", | |
3622 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3623 | mddev->chunk_sectors << 9); |
3624 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3625 | } |
3626 | ||
3627 | static ssize_t | |
fd01b88c | 3628 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3629 | { |
4c9309c0 | 3630 | unsigned long n; |
6791875e | 3631 | int err; |
3b34380a | 3632 | |
4c9309c0 AD |
3633 | err = kstrtoul(buf, 10, &n); |
3634 | if (err < 0) | |
3635 | return err; | |
3b34380a | 3636 | |
6791875e N |
3637 | err = mddev_lock(mddev); |
3638 | if (err) | |
3639 | return err; | |
b3546035 | 3640 | if (mddev->pers) { |
50ac168a | 3641 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3642 | err = -EBUSY; |
3643 | else if (mddev->ro) | |
3644 | err = -EROFS; | |
3645 | else { | |
3646 | mddev->new_chunk_sectors = n >> 9; | |
3647 | err = mddev->pers->check_reshape(mddev); | |
3648 | if (err) | |
3649 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
597a711b | 3650 | } |
b3546035 | 3651 | } else { |
664e7c41 | 3652 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3653 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3654 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3655 | } |
6791875e N |
3656 | mddev_unlock(mddev); |
3657 | return err ?: len; | |
3b34380a N |
3658 | } |
3659 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3660 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3661 | |
a94213b1 | 3662 | static ssize_t |
fd01b88c | 3663 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 3664 | { |
d1a7c503 N |
3665 | if (mddev->recovery_cp == MaxSector) |
3666 | return sprintf(page, "none\n"); | |
a94213b1 N |
3667 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3668 | } | |
3669 | ||
3670 | static ssize_t | |
fd01b88c | 3671 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 3672 | { |
4c9309c0 | 3673 | unsigned long long n; |
6791875e | 3674 | int err; |
4c9309c0 AD |
3675 | |
3676 | if (cmd_match(buf, "none")) | |
3677 | n = MaxSector; | |
3678 | else { | |
3679 | err = kstrtoull(buf, 10, &n); | |
3680 | if (err < 0) | |
3681 | return err; | |
3682 | if (n != (sector_t)n) | |
3683 | return -EINVAL; | |
3684 | } | |
a94213b1 | 3685 | |
6791875e N |
3686 | err = mddev_lock(mddev); |
3687 | if (err) | |
3688 | return err; | |
b098636c | 3689 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6791875e | 3690 | err = -EBUSY; |
a94213b1 | 3691 | |
6791875e N |
3692 | if (!err) { |
3693 | mddev->recovery_cp = n; | |
3694 | if (mddev->pers) | |
3695 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
3696 | } | |
3697 | mddev_unlock(mddev); | |
3698 | return err ?: len; | |
a94213b1 N |
3699 | } |
3700 | static struct md_sysfs_entry md_resync_start = | |
750f199e N |
3701 | __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR, |
3702 | resync_start_show, resync_start_store); | |
a94213b1 | 3703 | |
9e653b63 N |
3704 | /* |
3705 | * The array state can be: | |
3706 | * | |
3707 | * clear | |
3708 | * No devices, no size, no level | |
3709 | * Equivalent to STOP_ARRAY ioctl | |
3710 | * inactive | |
3711 | * May have some settings, but array is not active | |
3712 | * all IO results in error | |
3713 | * When written, doesn't tear down array, but just stops it | |
3714 | * suspended (not supported yet) | |
3715 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3716 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3717 | * readonly |
3718 | * no resync can happen. no superblocks get written. | |
3719 | * write requests fail | |
3720 | * read-auto | |
3721 | * like readonly, but behaves like 'clean' on a write request. | |
3722 | * | |
3723 | * clean - no pending writes, but otherwise active. | |
3724 | * When written to inactive array, starts without resync | |
3725 | * If a write request arrives then | |
3726 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3727 | * if not known, block and switch to write-pending | |
3728 | * If written to an active array that has pending writes, then fails. | |
3729 | * active | |
3730 | * fully active: IO and resync can be happening. | |
3731 | * When written to inactive array, starts with resync | |
3732 | * | |
3733 | * write-pending | |
3734 | * clean, but writes are blocked waiting for 'active' to be written. | |
3735 | * | |
3736 | * active-idle | |
3737 | * like active, but no writes have been seen for a while (100msec). | |
3738 | * | |
3739 | */ | |
3740 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3741 | write_pending, active_idle, bad_word}; | |
05381954 | 3742 | static char *array_states[] = { |
9e653b63 N |
3743 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3744 | "write-pending", "active-idle", NULL }; | |
3745 | ||
3746 | static int match_word(const char *word, char **list) | |
3747 | { | |
3748 | int n; | |
3749 | for (n=0; list[n]; n++) | |
3750 | if (cmd_match(word, list[n])) | |
3751 | break; | |
3752 | return n; | |
3753 | } | |
3754 | ||
3755 | static ssize_t | |
fd01b88c | 3756 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
3757 | { |
3758 | enum array_state st = inactive; | |
3759 | ||
3760 | if (mddev->pers) | |
3761 | switch(mddev->ro) { | |
3762 | case 1: | |
3763 | st = readonly; | |
3764 | break; | |
3765 | case 2: | |
3766 | st = read_auto; | |
3767 | break; | |
3768 | case 0: | |
3769 | if (mddev->in_sync) | |
3770 | st = clean; | |
070dc6dd | 3771 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3772 | st = write_pending; |
9e653b63 N |
3773 | else if (mddev->safemode) |
3774 | st = active_idle; | |
3775 | else | |
3776 | st = active; | |
3777 | } | |
3778 | else { | |
3779 | if (list_empty(&mddev->disks) && | |
3780 | mddev->raid_disks == 0 && | |
58c0fed4 | 3781 | mddev->dev_sectors == 0) |
9e653b63 N |
3782 | st = clear; |
3783 | else | |
3784 | st = inactive; | |
3785 | } | |
3786 | return sprintf(page, "%s\n", array_states[st]); | |
3787 | } | |
3788 | ||
f72ffdd6 N |
3789 | static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev); |
3790 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev); | |
3791 | static int do_md_run(struct mddev *mddev); | |
fd01b88c | 3792 | static int restart_array(struct mddev *mddev); |
9e653b63 N |
3793 | |
3794 | static ssize_t | |
fd01b88c | 3795 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 | 3796 | { |
6791875e | 3797 | int err; |
9e653b63 | 3798 | enum array_state st = match_word(buf, array_states); |
6791875e N |
3799 | |
3800 | if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) { | |
3801 | /* don't take reconfig_mutex when toggling between | |
3802 | * clean and active | |
3803 | */ | |
3804 | spin_lock(&mddev->lock); | |
3805 | if (st == active) { | |
3806 | restart_array(mddev); | |
3807 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
3808 | wake_up(&mddev->sb_wait); | |
3809 | err = 0; | |
3810 | } else /* st == clean */ { | |
3811 | restart_array(mddev); | |
3812 | if (atomic_read(&mddev->writes_pending) == 0) { | |
3813 | if (mddev->in_sync == 0) { | |
3814 | mddev->in_sync = 1; | |
3815 | if (mddev->safemode == 1) | |
3816 | mddev->safemode = 0; | |
3817 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
3818 | } | |
3819 | err = 0; | |
3820 | } else | |
3821 | err = -EBUSY; | |
3822 | } | |
3823 | spin_unlock(&mddev->lock); | |
c008f1d3 | 3824 | return err ?: len; |
6791875e N |
3825 | } |
3826 | err = mddev_lock(mddev); | |
3827 | if (err) | |
3828 | return err; | |
3829 | err = -EINVAL; | |
9e653b63 N |
3830 | switch(st) { |
3831 | case bad_word: | |
3832 | break; | |
3833 | case clear: | |
3834 | /* stopping an active array */ | |
a05b7ea0 | 3835 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
3836 | break; |
3837 | case inactive: | |
3838 | /* stopping an active array */ | |
90cf195d | 3839 | if (mddev->pers) |
a05b7ea0 | 3840 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 3841 | else |
e691063a | 3842 | err = 0; /* already inactive */ |
9e653b63 N |
3843 | break; |
3844 | case suspended: | |
3845 | break; /* not supported yet */ | |
3846 | case readonly: | |
3847 | if (mddev->pers) | |
a05b7ea0 | 3848 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
3849 | else { |
3850 | mddev->ro = 1; | |
648b629e | 3851 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3852 | err = do_md_run(mddev); |
3853 | } | |
3854 | break; | |
3855 | case read_auto: | |
9e653b63 | 3856 | if (mddev->pers) { |
80268ee9 | 3857 | if (mddev->ro == 0) |
a05b7ea0 | 3858 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 3859 | else if (mddev->ro == 1) |
648b629e N |
3860 | err = restart_array(mddev); |
3861 | if (err == 0) { | |
3862 | mddev->ro = 2; | |
3863 | set_disk_ro(mddev->gendisk, 0); | |
3864 | } | |
9e653b63 N |
3865 | } else { |
3866 | mddev->ro = 2; | |
3867 | err = do_md_run(mddev); | |
3868 | } | |
3869 | break; | |
3870 | case clean: | |
3871 | if (mddev->pers) { | |
3872 | restart_array(mddev); | |
85572d7c | 3873 | spin_lock(&mddev->lock); |
9e653b63 | 3874 | if (atomic_read(&mddev->writes_pending) == 0) { |
e691063a N |
3875 | if (mddev->in_sync == 0) { |
3876 | mddev->in_sync = 1; | |
31a59e34 N |
3877 | if (mddev->safemode == 1) |
3878 | mddev->safemode = 0; | |
070dc6dd | 3879 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3880 | } |
3881 | err = 0; | |
3882 | } else | |
3883 | err = -EBUSY; | |
85572d7c | 3884 | spin_unlock(&mddev->lock); |
5bf29597 N |
3885 | } else |
3886 | err = -EINVAL; | |
9e653b63 N |
3887 | break; |
3888 | case active: | |
3889 | if (mddev->pers) { | |
3890 | restart_array(mddev); | |
070dc6dd | 3891 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3892 | wake_up(&mddev->sb_wait); |
3893 | err = 0; | |
3894 | } else { | |
3895 | mddev->ro = 0; | |
648b629e | 3896 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3897 | err = do_md_run(mddev); |
3898 | } | |
3899 | break; | |
3900 | case write_pending: | |
3901 | case active_idle: | |
3902 | /* these cannot be set */ | |
3903 | break; | |
3904 | } | |
6791875e N |
3905 | |
3906 | if (!err) { | |
1d23f178 N |
3907 | if (mddev->hold_active == UNTIL_IOCTL) |
3908 | mddev->hold_active = 0; | |
00bcb4ac | 3909 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 3910 | } |
6791875e N |
3911 | mddev_unlock(mddev); |
3912 | return err ?: len; | |
9e653b63 | 3913 | } |
80ca3a44 | 3914 | static struct md_sysfs_entry md_array_state = |
750f199e | 3915 | __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); |
9e653b63 | 3916 | |
1e50915f | 3917 | static ssize_t |
fd01b88c | 3918 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
3919 | return sprintf(page, "%d\n", |
3920 | atomic_read(&mddev->max_corr_read_errors)); | |
3921 | } | |
3922 | ||
3923 | static ssize_t | |
fd01b88c | 3924 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f | 3925 | { |
4c9309c0 AD |
3926 | unsigned int n; |
3927 | int rv; | |
1e50915f | 3928 | |
4c9309c0 AD |
3929 | rv = kstrtouint(buf, 10, &n); |
3930 | if (rv < 0) | |
3931 | return rv; | |
3932 | atomic_set(&mddev->max_corr_read_errors, n); | |
3933 | return len; | |
1e50915f RB |
3934 | } |
3935 | ||
3936 | static struct md_sysfs_entry max_corr_read_errors = | |
3937 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3938 | max_corrected_read_errors_store); | |
3939 | ||
6d7ff738 | 3940 | static ssize_t |
fd01b88c | 3941 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
3942 | { |
3943 | return -EINVAL; | |
3944 | } | |
3945 | ||
3946 | static ssize_t | |
fd01b88c | 3947 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
3948 | { |
3949 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3950 | /* The new device is added to the array. | |
3951 | * If the array has a persistent superblock, we read the | |
3952 | * superblock to initialise info and check validity. | |
3953 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3954 | * which mainly checks size. | |
3955 | */ | |
3956 | char *e; | |
3957 | int major = simple_strtoul(buf, &e, 10); | |
3958 | int minor; | |
3959 | dev_t dev; | |
3cb03002 | 3960 | struct md_rdev *rdev; |
6d7ff738 N |
3961 | int err; |
3962 | ||
3963 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3964 | return -EINVAL; | |
3965 | minor = simple_strtoul(e+1, &e, 10); | |
3966 | if (*e && *e != '\n') | |
3967 | return -EINVAL; | |
3968 | dev = MKDEV(major, minor); | |
3969 | if (major != MAJOR(dev) || | |
3970 | minor != MINOR(dev)) | |
3971 | return -EOVERFLOW; | |
3972 | ||
6791875e N |
3973 | flush_workqueue(md_misc_wq); |
3974 | ||
3975 | err = mddev_lock(mddev); | |
3976 | if (err) | |
3977 | return err; | |
6d7ff738 N |
3978 | if (mddev->persistent) { |
3979 | rdev = md_import_device(dev, mddev->major_version, | |
3980 | mddev->minor_version); | |
3981 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
3982 | struct md_rdev *rdev0 |
3983 | = list_entry(mddev->disks.next, | |
3984 | struct md_rdev, same_set); | |
6d7ff738 N |
3985 | err = super_types[mddev->major_version] |
3986 | .load_super(rdev, rdev0, mddev->minor_version); | |
3987 | if (err < 0) | |
3988 | goto out; | |
3989 | } | |
c5d79adb N |
3990 | } else if (mddev->external) |
3991 | rdev = md_import_device(dev, -2, -1); | |
3992 | else | |
6d7ff738 N |
3993 | rdev = md_import_device(dev, -1, -1); |
3994 | ||
9a8c0fa8 N |
3995 | if (IS_ERR(rdev)) { |
3996 | mddev_unlock(mddev); | |
6d7ff738 | 3997 | return PTR_ERR(rdev); |
9a8c0fa8 | 3998 | } |
6d7ff738 N |
3999 | err = bind_rdev_to_array(rdev, mddev); |
4000 | out: | |
4001 | if (err) | |
4002 | export_rdev(rdev); | |
6791875e | 4003 | mddev_unlock(mddev); |
6d7ff738 N |
4004 | return err ? err : len; |
4005 | } | |
4006 | ||
4007 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 4008 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 4009 | |
9b1d1dac | 4010 | static ssize_t |
fd01b88c | 4011 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
4012 | { |
4013 | char *end; | |
4014 | unsigned long chunk, end_chunk; | |
6791875e | 4015 | int err; |
9b1d1dac | 4016 | |
6791875e N |
4017 | err = mddev_lock(mddev); |
4018 | if (err) | |
4019 | return err; | |
9b1d1dac PC |
4020 | if (!mddev->bitmap) |
4021 | goto out; | |
4022 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
4023 | while (*buf) { | |
4024 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
4025 | if (buf == end) break; | |
4026 | if (*end == '-') { /* range */ | |
4027 | buf = end + 1; | |
4028 | end_chunk = simple_strtoul(buf, &end, 0); | |
4029 | if (buf == end) break; | |
4030 | } | |
4031 | if (*end && !isspace(*end)) break; | |
4032 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 4033 | buf = skip_spaces(end); |
9b1d1dac PC |
4034 | } |
4035 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
4036 | out: | |
6791875e | 4037 | mddev_unlock(mddev); |
9b1d1dac PC |
4038 | return len; |
4039 | } | |
4040 | ||
4041 | static struct md_sysfs_entry md_bitmap = | |
4042 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
4043 | ||
a35b0d69 | 4044 | static ssize_t |
fd01b88c | 4045 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 4046 | { |
58c0fed4 AN |
4047 | return sprintf(page, "%llu\n", |
4048 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
4049 | } |
4050 | ||
fd01b88c | 4051 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
4052 | |
4053 | static ssize_t | |
fd01b88c | 4054 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
4055 | { |
4056 | /* If array is inactive, we can reduce the component size, but | |
4057 | * not increase it (except from 0). | |
4058 | * If array is active, we can try an on-line resize | |
4059 | */ | |
b522adcd DW |
4060 | sector_t sectors; |
4061 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 4062 | |
58c0fed4 AN |
4063 | if (err < 0) |
4064 | return err; | |
6791875e N |
4065 | err = mddev_lock(mddev); |
4066 | if (err) | |
4067 | return err; | |
a35b0d69 | 4068 | if (mddev->pers) { |
293467aa GR |
4069 | if (mddev_is_clustered(mddev)) |
4070 | md_cluster_ops->metadata_update_start(mddev); | |
58c0fed4 | 4071 | err = update_size(mddev, sectors); |
850b2b42 | 4072 | md_update_sb(mddev, 1); |
293467aa GR |
4073 | if (mddev_is_clustered(mddev)) |
4074 | md_cluster_ops->metadata_update_finish(mddev); | |
a35b0d69 | 4075 | } else { |
58c0fed4 AN |
4076 | if (mddev->dev_sectors == 0 || |
4077 | mddev->dev_sectors > sectors) | |
4078 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
4079 | else |
4080 | err = -ENOSPC; | |
4081 | } | |
6791875e | 4082 | mddev_unlock(mddev); |
a35b0d69 N |
4083 | return err ? err : len; |
4084 | } | |
4085 | ||
4086 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 4087 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 4088 | |
83f0d77a | 4089 | /* Metadata version. |
e691063a N |
4090 | * This is one of |
4091 | * 'none' for arrays with no metadata (good luck...) | |
4092 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
4093 | * or N.M for internally known formats |
4094 | */ | |
4095 | static ssize_t | |
fd01b88c | 4096 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
4097 | { |
4098 | if (mddev->persistent) | |
4099 | return sprintf(page, "%d.%d\n", | |
4100 | mddev->major_version, mddev->minor_version); | |
e691063a N |
4101 | else if (mddev->external) |
4102 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
4103 | else |
4104 | return sprintf(page, "none\n"); | |
4105 | } | |
4106 | ||
4107 | static ssize_t | |
fd01b88c | 4108 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
4109 | { |
4110 | int major, minor; | |
4111 | char *e; | |
6791875e | 4112 | int err; |
ea43ddd8 N |
4113 | /* Changing the details of 'external' metadata is |
4114 | * always permitted. Otherwise there must be | |
4115 | * no devices attached to the array. | |
4116 | */ | |
6791875e N |
4117 | |
4118 | err = mddev_lock(mddev); | |
4119 | if (err) | |
4120 | return err; | |
4121 | err = -EBUSY; | |
ea43ddd8 N |
4122 | if (mddev->external && strncmp(buf, "external:", 9) == 0) |
4123 | ; | |
4124 | else if (!list_empty(&mddev->disks)) | |
6791875e | 4125 | goto out_unlock; |
8bb93aac | 4126 | |
6791875e | 4127 | err = 0; |
8bb93aac N |
4128 | if (cmd_match(buf, "none")) { |
4129 | mddev->persistent = 0; | |
e691063a N |
4130 | mddev->external = 0; |
4131 | mddev->major_version = 0; | |
4132 | mddev->minor_version = 90; | |
6791875e | 4133 | goto out_unlock; |
e691063a N |
4134 | } |
4135 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 4136 | size_t namelen = len-9; |
e691063a N |
4137 | if (namelen >= sizeof(mddev->metadata_type)) |
4138 | namelen = sizeof(mddev->metadata_type)-1; | |
4139 | strncpy(mddev->metadata_type, buf+9, namelen); | |
4140 | mddev->metadata_type[namelen] = 0; | |
4141 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
4142 | mddev->metadata_type[--namelen] = 0; | |
4143 | mddev->persistent = 0; | |
4144 | mddev->external = 1; | |
8bb93aac N |
4145 | mddev->major_version = 0; |
4146 | mddev->minor_version = 90; | |
6791875e | 4147 | goto out_unlock; |
8bb93aac N |
4148 | } |
4149 | major = simple_strtoul(buf, &e, 10); | |
6791875e | 4150 | err = -EINVAL; |
8bb93aac | 4151 | if (e==buf || *e != '.') |
6791875e | 4152 | goto out_unlock; |
8bb93aac N |
4153 | buf = e+1; |
4154 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 4155 | if (e==buf || (*e && *e != '\n') ) |
6791875e N |
4156 | goto out_unlock; |
4157 | err = -ENOENT; | |
50511da3 | 4158 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
6791875e | 4159 | goto out_unlock; |
8bb93aac N |
4160 | mddev->major_version = major; |
4161 | mddev->minor_version = minor; | |
4162 | mddev->persistent = 1; | |
e691063a | 4163 | mddev->external = 0; |
6791875e N |
4164 | err = 0; |
4165 | out_unlock: | |
4166 | mddev_unlock(mddev); | |
4167 | return err ?: len; | |
8bb93aac N |
4168 | } |
4169 | ||
4170 | static struct md_sysfs_entry md_metadata = | |
750f199e | 4171 | __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 4172 | |
24dd469d | 4173 | static ssize_t |
fd01b88c | 4174 | action_show(struct mddev *mddev, char *page) |
24dd469d | 4175 | { |
7eec314d | 4176 | char *type = "idle"; |
b7b17c9b N |
4177 | unsigned long recovery = mddev->recovery; |
4178 | if (test_bit(MD_RECOVERY_FROZEN, &recovery)) | |
b6a9ce68 | 4179 | type = "frozen"; |
b7b17c9b N |
4180 | else if (test_bit(MD_RECOVERY_RUNNING, &recovery) || |
4181 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) { | |
4182 | if (test_bit(MD_RECOVERY_RESHAPE, &recovery)) | |
ccfcc3c1 | 4183 | type = "reshape"; |
b7b17c9b N |
4184 | else if (test_bit(MD_RECOVERY_SYNC, &recovery)) { |
4185 | if (!test_bit(MD_RECOVERY_REQUESTED, &recovery)) | |
24dd469d | 4186 | type = "resync"; |
b7b17c9b | 4187 | else if (test_bit(MD_RECOVERY_CHECK, &recovery)) |
24dd469d N |
4188 | type = "check"; |
4189 | else | |
4190 | type = "repair"; | |
b7b17c9b | 4191 | } else if (test_bit(MD_RECOVERY_RECOVER, &recovery)) |
24dd469d | 4192 | type = "recover"; |
985ca973 N |
4193 | else if (mddev->reshape_position != MaxSector) |
4194 | type = "reshape"; | |
24dd469d N |
4195 | } |
4196 | return sprintf(page, "%s\n", type); | |
4197 | } | |
4198 | ||
4199 | static ssize_t | |
fd01b88c | 4200 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4201 | { |
7eec314d N |
4202 | if (!mddev->pers || !mddev->pers->sync_request) |
4203 | return -EINVAL; | |
4204 | ||
b6a9ce68 N |
4205 | |
4206 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
56ccc112 N |
4207 | if (cmd_match(page, "frozen")) |
4208 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4209 | else | |
4210 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
8e8e2518 N |
4211 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
4212 | mddev_lock(mddev) == 0) { | |
4213 | flush_workqueue(md_misc_wq); | |
4214 | if (mddev->sync_thread) { | |
4215 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6791875e | 4216 | md_reap_sync_thread(mddev); |
6791875e | 4217 | } |
8e8e2518 | 4218 | mddev_unlock(mddev); |
7eec314d | 4219 | } |
03c902e1 N |
4220 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
4221 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 4222 | return -EBUSY; |
72a23c21 | 4223 | else if (cmd_match(page, "resync")) |
56ccc112 | 4224 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4225 | else if (cmd_match(page, "recover")) { |
56ccc112 | 4226 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4227 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
72a23c21 | 4228 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4229 | int err; |
4230 | if (mddev->pers->start_reshape == NULL) | |
4231 | return -EINVAL; | |
6791875e N |
4232 | err = mddev_lock(mddev); |
4233 | if (!err) { | |
56ccc112 | 4234 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6791875e N |
4235 | err = mddev->pers->start_reshape(mddev); |
4236 | mddev_unlock(mddev); | |
4237 | } | |
16484bf5 N |
4238 | if (err) |
4239 | return err; | |
a99ac971 | 4240 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4241 | } else { |
bce74dac | 4242 | if (cmd_match(page, "check")) |
7eec314d | 4243 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4244 | else if (!cmd_match(page, "repair")) |
7eec314d | 4245 | return -EINVAL; |
56ccc112 | 4246 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
7eec314d N |
4247 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
4248 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4249 | } |
48c26ddc N |
4250 | if (mddev->ro == 2) { |
4251 | /* A write to sync_action is enough to justify | |
4252 | * canceling read-auto mode | |
4253 | */ | |
4254 | mddev->ro = 0; | |
4255 | md_wakeup_thread(mddev->sync_thread); | |
4256 | } | |
03c902e1 | 4257 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4258 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4259 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4260 | return len; |
4261 | } | |
4262 | ||
c4a39551 | 4263 | static struct md_sysfs_entry md_scan_mode = |
750f199e | 4264 | __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); |
c4a39551 JB |
4265 | |
4266 | static ssize_t | |
4267 | last_sync_action_show(struct mddev *mddev, char *page) | |
4268 | { | |
4269 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4270 | } | |
4271 | ||
4272 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4273 | ||
9d88883e | 4274 | static ssize_t |
fd01b88c | 4275 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4276 | { |
4277 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4278 | (unsigned long long) |
4279 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4280 | } |
4281 | ||
80ca3a44 | 4282 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4283 | |
88202a0c | 4284 | static ssize_t |
fd01b88c | 4285 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4286 | { |
4287 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4288 | mddev->sync_speed_min ? "local": "system"); | |
4289 | } | |
4290 | ||
4291 | static ssize_t | |
fd01b88c | 4292 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4293 | { |
4c9309c0 AD |
4294 | unsigned int min; |
4295 | int rv; | |
4296 | ||
88202a0c | 4297 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4298 | min = 0; |
4299 | } else { | |
4300 | rv = kstrtouint(buf, 10, &min); | |
4301 | if (rv < 0) | |
4302 | return rv; | |
4303 | if (min == 0) | |
4304 | return -EINVAL; | |
88202a0c | 4305 | } |
88202a0c N |
4306 | mddev->sync_speed_min = min; |
4307 | return len; | |
4308 | } | |
4309 | ||
4310 | static struct md_sysfs_entry md_sync_min = | |
4311 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4312 | ||
4313 | static ssize_t | |
fd01b88c | 4314 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4315 | { |
4316 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4317 | mddev->sync_speed_max ? "local": "system"); | |
4318 | } | |
4319 | ||
4320 | static ssize_t | |
fd01b88c | 4321 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4322 | { |
4c9309c0 AD |
4323 | unsigned int max; |
4324 | int rv; | |
4325 | ||
88202a0c | 4326 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4327 | max = 0; |
4328 | } else { | |
4329 | rv = kstrtouint(buf, 10, &max); | |
4330 | if (rv < 0) | |
4331 | return rv; | |
4332 | if (max == 0) | |
4333 | return -EINVAL; | |
88202a0c | 4334 | } |
88202a0c N |
4335 | mddev->sync_speed_max = max; |
4336 | return len; | |
4337 | } | |
4338 | ||
4339 | static struct md_sysfs_entry md_sync_max = | |
4340 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4341 | ||
d7f3d291 | 4342 | static ssize_t |
fd01b88c | 4343 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4344 | { |
4345 | return sprintf(page, "%d\n", mddev->degraded); | |
4346 | } | |
4347 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4348 | |
90b08710 | 4349 | static ssize_t |
fd01b88c | 4350 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4351 | { |
4352 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4353 | } | |
4354 | ||
4355 | static ssize_t | |
fd01b88c | 4356 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4357 | { |
4358 | long n; | |
4359 | ||
b29bebd6 | 4360 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4361 | return -EINVAL; |
4362 | ||
4363 | if (n != 0 && n != 1) | |
4364 | return -EINVAL; | |
4365 | ||
4366 | mddev->parallel_resync = n; | |
4367 | ||
4368 | if (mddev->sync_thread) | |
4369 | wake_up(&resync_wait); | |
4370 | ||
4371 | return len; | |
4372 | } | |
4373 | ||
4374 | /* force parallel resync, even with shared block devices */ | |
4375 | static struct md_sysfs_entry md_sync_force_parallel = | |
4376 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4377 | sync_force_parallel_show, sync_force_parallel_store); | |
4378 | ||
88202a0c | 4379 | static ssize_t |
fd01b88c | 4380 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4381 | { |
4382 | unsigned long resync, dt, db; | |
d1a7c503 N |
4383 | if (mddev->curr_resync == 0) |
4384 | return sprintf(page, "none\n"); | |
9687a60c AN |
4385 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4386 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4387 | if (!dt) dt++; |
9687a60c AN |
4388 | db = resync - mddev->resync_mark_cnt; |
4389 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4390 | } |
4391 | ||
80ca3a44 | 4392 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4393 | |
4394 | static ssize_t | |
fd01b88c | 4395 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4396 | { |
13ae864b | 4397 | unsigned long long max_sectors, resync; |
88202a0c | 4398 | |
acb180b0 N |
4399 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4400 | return sprintf(page, "none\n"); | |
4401 | ||
72f36d59 N |
4402 | if (mddev->curr_resync == 1 || |
4403 | mddev->curr_resync == 2) | |
4404 | return sprintf(page, "delayed\n"); | |
4405 | ||
c804cdec N |
4406 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4407 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4408 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4409 | else |
58c0fed4 | 4410 | max_sectors = mddev->dev_sectors; |
88202a0c | 4411 | |
acb180b0 | 4412 | resync = mddev->curr_resync_completed; |
13ae864b | 4413 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4414 | } |
4415 | ||
750f199e N |
4416 | static struct md_sysfs_entry md_sync_completed = |
4417 | __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL); | |
88202a0c | 4418 | |
5e96ee65 | 4419 | static ssize_t |
fd01b88c | 4420 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4421 | { |
4422 | return sprintf(page, "%llu\n", | |
4423 | (unsigned long long)mddev->resync_min); | |
4424 | } | |
4425 | static ssize_t | |
fd01b88c | 4426 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4427 | { |
4428 | unsigned long long min; | |
23da422b | 4429 | int err; |
23da422b | 4430 | |
b29bebd6 | 4431 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 | 4432 | return -EINVAL; |
23da422b N |
4433 | |
4434 | spin_lock(&mddev->lock); | |
4435 | err = -EINVAL; | |
5e96ee65 | 4436 | if (min > mddev->resync_max) |
23da422b N |
4437 | goto out_unlock; |
4438 | ||
4439 | err = -EBUSY; | |
5e96ee65 | 4440 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4441 | goto out_unlock; |
5e96ee65 | 4442 | |
50c37b13 N |
4443 | /* Round down to multiple of 4K for safety */ |
4444 | mddev->resync_min = round_down(min, 8); | |
23da422b | 4445 | err = 0; |
5e96ee65 | 4446 | |
23da422b N |
4447 | out_unlock: |
4448 | spin_unlock(&mddev->lock); | |
4449 | return err ?: len; | |
5e96ee65 NB |
4450 | } |
4451 | ||
4452 | static struct md_sysfs_entry md_min_sync = | |
4453 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4454 | ||
c6207277 | 4455 | static ssize_t |
fd01b88c | 4456 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4457 | { |
4458 | if (mddev->resync_max == MaxSector) | |
4459 | return sprintf(page, "max\n"); | |
4460 | else | |
4461 | return sprintf(page, "%llu\n", | |
4462 | (unsigned long long)mddev->resync_max); | |
4463 | } | |
4464 | static ssize_t | |
fd01b88c | 4465 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 | 4466 | { |
23da422b N |
4467 | int err; |
4468 | spin_lock(&mddev->lock); | |
c6207277 N |
4469 | if (strncmp(buf, "max", 3) == 0) |
4470 | mddev->resync_max = MaxSector; | |
4471 | else { | |
5e96ee65 | 4472 | unsigned long long max; |
23da422b N |
4473 | int chunk; |
4474 | ||
4475 | err = -EINVAL; | |
b29bebd6 | 4476 | if (kstrtoull(buf, 10, &max)) |
23da422b | 4477 | goto out_unlock; |
5e96ee65 | 4478 | if (max < mddev->resync_min) |
23da422b N |
4479 | goto out_unlock; |
4480 | ||
4481 | err = -EBUSY; | |
c6207277 | 4482 | if (max < mddev->resync_max && |
4d484a4a | 4483 | mddev->ro == 0 && |
c6207277 | 4484 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4485 | goto out_unlock; |
c6207277 N |
4486 | |
4487 | /* Must be a multiple of chunk_size */ | |
23da422b N |
4488 | chunk = mddev->chunk_sectors; |
4489 | if (chunk) { | |
2ac06c33 | 4490 | sector_t temp = max; |
23da422b N |
4491 | |
4492 | err = -EINVAL; | |
4493 | if (sector_div(temp, chunk)) | |
4494 | goto out_unlock; | |
c6207277 N |
4495 | } |
4496 | mddev->resync_max = max; | |
4497 | } | |
4498 | wake_up(&mddev->recovery_wait); | |
23da422b N |
4499 | err = 0; |
4500 | out_unlock: | |
4501 | spin_unlock(&mddev->lock); | |
4502 | return err ?: len; | |
c6207277 N |
4503 | } |
4504 | ||
4505 | static struct md_sysfs_entry md_max_sync = | |
4506 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4507 | ||
e464eafd | 4508 | static ssize_t |
fd01b88c | 4509 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4510 | { |
4511 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4512 | } | |
4513 | ||
4514 | static ssize_t | |
fd01b88c | 4515 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4516 | { |
4c9309c0 | 4517 | unsigned long long old, new; |
6791875e | 4518 | int err; |
e464eafd | 4519 | |
4c9309c0 AD |
4520 | err = kstrtoull(buf, 10, &new); |
4521 | if (err < 0) | |
4522 | return err; | |
4523 | if (new != (sector_t)new) | |
e464eafd | 4524 | return -EINVAL; |
23ddff37 | 4525 | |
6791875e N |
4526 | err = mddev_lock(mddev); |
4527 | if (err) | |
4528 | return err; | |
4529 | err = -EINVAL; | |
4530 | if (mddev->pers == NULL || | |
4531 | mddev->pers->quiesce == NULL) | |
4532 | goto unlock; | |
4533 | old = mddev->suspend_lo; | |
23ddff37 N |
4534 | mddev->suspend_lo = new; |
4535 | if (new >= old) | |
4536 | /* Shrinking suspended region */ | |
e464eafd | 4537 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4538 | else { |
4539 | /* Expanding suspended region - need to wait */ | |
4540 | mddev->pers->quiesce(mddev, 1); | |
4541 | mddev->pers->quiesce(mddev, 0); | |
4542 | } | |
6791875e N |
4543 | err = 0; |
4544 | unlock: | |
4545 | mddev_unlock(mddev); | |
4546 | return err ?: len; | |
e464eafd N |
4547 | } |
4548 | static struct md_sysfs_entry md_suspend_lo = | |
4549 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4550 | ||
e464eafd | 4551 | static ssize_t |
fd01b88c | 4552 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4553 | { |
4554 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4555 | } | |
4556 | ||
4557 | static ssize_t | |
fd01b88c | 4558 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4559 | { |
4c9309c0 | 4560 | unsigned long long old, new; |
6791875e | 4561 | int err; |
e464eafd | 4562 | |
4c9309c0 AD |
4563 | err = kstrtoull(buf, 10, &new); |
4564 | if (err < 0) | |
4565 | return err; | |
4566 | if (new != (sector_t)new) | |
e464eafd | 4567 | return -EINVAL; |
23ddff37 | 4568 | |
6791875e N |
4569 | err = mddev_lock(mddev); |
4570 | if (err) | |
4571 | return err; | |
4572 | err = -EINVAL; | |
4573 | if (mddev->pers == NULL || | |
4574 | mddev->pers->quiesce == NULL) | |
4575 | goto unlock; | |
4576 | old = mddev->suspend_hi; | |
23ddff37 N |
4577 | mddev->suspend_hi = new; |
4578 | if (new <= old) | |
4579 | /* Shrinking suspended region */ | |
4580 | mddev->pers->quiesce(mddev, 2); | |
4581 | else { | |
4582 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4583 | mddev->pers->quiesce(mddev, 1); |
4584 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 | 4585 | } |
6791875e N |
4586 | err = 0; |
4587 | unlock: | |
4588 | mddev_unlock(mddev); | |
4589 | return err ?: len; | |
e464eafd N |
4590 | } |
4591 | static struct md_sysfs_entry md_suspend_hi = | |
4592 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4593 | ||
08a02ecd | 4594 | static ssize_t |
fd01b88c | 4595 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4596 | { |
4597 | if (mddev->reshape_position != MaxSector) | |
4598 | return sprintf(page, "%llu\n", | |
4599 | (unsigned long long)mddev->reshape_position); | |
4600 | strcpy(page, "none\n"); | |
4601 | return 5; | |
4602 | } | |
4603 | ||
4604 | static ssize_t | |
fd01b88c | 4605 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4606 | { |
c6563a8c | 4607 | struct md_rdev *rdev; |
4c9309c0 | 4608 | unsigned long long new; |
6791875e | 4609 | int err; |
6791875e | 4610 | |
4c9309c0 AD |
4611 | err = kstrtoull(buf, 10, &new); |
4612 | if (err < 0) | |
4613 | return err; | |
4614 | if (new != (sector_t)new) | |
08a02ecd | 4615 | return -EINVAL; |
6791875e N |
4616 | err = mddev_lock(mddev); |
4617 | if (err) | |
4618 | return err; | |
4619 | err = -EBUSY; | |
4620 | if (mddev->pers) | |
4621 | goto unlock; | |
08a02ecd N |
4622 | mddev->reshape_position = new; |
4623 | mddev->delta_disks = 0; | |
2c810cdd | 4624 | mddev->reshape_backwards = 0; |
08a02ecd N |
4625 | mddev->new_level = mddev->level; |
4626 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4627 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4628 | rdev_for_each(rdev, mddev) |
4629 | rdev->new_data_offset = rdev->data_offset; | |
6791875e N |
4630 | err = 0; |
4631 | unlock: | |
4632 | mddev_unlock(mddev); | |
4633 | return err ?: len; | |
08a02ecd N |
4634 | } |
4635 | ||
4636 | static struct md_sysfs_entry md_reshape_position = | |
4637 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4638 | reshape_position_store); | |
4639 | ||
2c810cdd N |
4640 | static ssize_t |
4641 | reshape_direction_show(struct mddev *mddev, char *page) | |
4642 | { | |
4643 | return sprintf(page, "%s\n", | |
4644 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4645 | } | |
4646 | ||
4647 | static ssize_t | |
4648 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4649 | { | |
4650 | int backwards = 0; | |
6791875e N |
4651 | int err; |
4652 | ||
2c810cdd N |
4653 | if (cmd_match(buf, "forwards")) |
4654 | backwards = 0; | |
4655 | else if (cmd_match(buf, "backwards")) | |
4656 | backwards = 1; | |
4657 | else | |
4658 | return -EINVAL; | |
4659 | if (mddev->reshape_backwards == backwards) | |
4660 | return len; | |
4661 | ||
6791875e N |
4662 | err = mddev_lock(mddev); |
4663 | if (err) | |
4664 | return err; | |
2c810cdd N |
4665 | /* check if we are allowed to change */ |
4666 | if (mddev->delta_disks) | |
6791875e N |
4667 | err = -EBUSY; |
4668 | else if (mddev->persistent && | |
2c810cdd | 4669 | mddev->major_version == 0) |
6791875e N |
4670 | err = -EINVAL; |
4671 | else | |
4672 | mddev->reshape_backwards = backwards; | |
4673 | mddev_unlock(mddev); | |
4674 | return err ?: len; | |
2c810cdd N |
4675 | } |
4676 | ||
4677 | static struct md_sysfs_entry md_reshape_direction = | |
4678 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
4679 | reshape_direction_store); | |
4680 | ||
b522adcd | 4681 | static ssize_t |
fd01b88c | 4682 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
4683 | { |
4684 | if (mddev->external_size) | |
4685 | return sprintf(page, "%llu\n", | |
4686 | (unsigned long long)mddev->array_sectors/2); | |
4687 | else | |
4688 | return sprintf(page, "default\n"); | |
4689 | } | |
4690 | ||
4691 | static ssize_t | |
fd01b88c | 4692 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
4693 | { |
4694 | sector_t sectors; | |
6791875e N |
4695 | int err; |
4696 | ||
4697 | err = mddev_lock(mddev); | |
4698 | if (err) | |
4699 | return err; | |
b522adcd DW |
4700 | |
4701 | if (strncmp(buf, "default", 7) == 0) { | |
4702 | if (mddev->pers) | |
4703 | sectors = mddev->pers->size(mddev, 0, 0); | |
4704 | else | |
4705 | sectors = mddev->array_sectors; | |
4706 | ||
4707 | mddev->external_size = 0; | |
4708 | } else { | |
4709 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
6791875e N |
4710 | err = -EINVAL; |
4711 | else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
4712 | err = -E2BIG; | |
4713 | else | |
4714 | mddev->external_size = 1; | |
b522adcd DW |
4715 | } |
4716 | ||
6791875e N |
4717 | if (!err) { |
4718 | mddev->array_sectors = sectors; | |
4719 | if (mddev->pers) { | |
4720 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4721 | revalidate_disk(mddev->gendisk); | |
4722 | } | |
cbe6ef1d | 4723 | } |
6791875e N |
4724 | mddev_unlock(mddev); |
4725 | return err ?: len; | |
b522adcd DW |
4726 | } |
4727 | ||
4728 | static struct md_sysfs_entry md_array_size = | |
4729 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4730 | array_size_store); | |
e464eafd | 4731 | |
eae1701f N |
4732 | static struct attribute *md_default_attrs[] = { |
4733 | &md_level.attr, | |
d4dbd025 | 4734 | &md_layout.attr, |
eae1701f | 4735 | &md_raid_disks.attr, |
3b34380a | 4736 | &md_chunk_size.attr, |
a35b0d69 | 4737 | &md_size.attr, |
a94213b1 | 4738 | &md_resync_start.attr, |
8bb93aac | 4739 | &md_metadata.attr, |
6d7ff738 | 4740 | &md_new_device.attr, |
16f17b39 | 4741 | &md_safe_delay.attr, |
9e653b63 | 4742 | &md_array_state.attr, |
08a02ecd | 4743 | &md_reshape_position.attr, |
2c810cdd | 4744 | &md_reshape_direction.attr, |
b522adcd | 4745 | &md_array_size.attr, |
1e50915f | 4746 | &max_corr_read_errors.attr, |
411036fa N |
4747 | NULL, |
4748 | }; | |
4749 | ||
4750 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4751 | &md_scan_mode.attr, |
c4a39551 | 4752 | &md_last_scan_mode.attr, |
9d88883e | 4753 | &md_mismatches.attr, |
88202a0c N |
4754 | &md_sync_min.attr, |
4755 | &md_sync_max.attr, | |
4756 | &md_sync_speed.attr, | |
90b08710 | 4757 | &md_sync_force_parallel.attr, |
88202a0c | 4758 | &md_sync_completed.attr, |
5e96ee65 | 4759 | &md_min_sync.attr, |
c6207277 | 4760 | &md_max_sync.attr, |
e464eafd N |
4761 | &md_suspend_lo.attr, |
4762 | &md_suspend_hi.attr, | |
9b1d1dac | 4763 | &md_bitmap.attr, |
d7f3d291 | 4764 | &md_degraded.attr, |
eae1701f N |
4765 | NULL, |
4766 | }; | |
411036fa N |
4767 | static struct attribute_group md_redundancy_group = { |
4768 | .name = NULL, | |
4769 | .attrs = md_redundancy_attrs, | |
4770 | }; | |
4771 | ||
eae1701f N |
4772 | static ssize_t |
4773 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4774 | { | |
4775 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4776 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4777 | ssize_t rv; |
eae1701f N |
4778 | |
4779 | if (!entry->show) | |
4780 | return -EIO; | |
af8a2434 N |
4781 | spin_lock(&all_mddevs_lock); |
4782 | if (list_empty(&mddev->all_mddevs)) { | |
4783 | spin_unlock(&all_mddevs_lock); | |
4784 | return -EBUSY; | |
4785 | } | |
4786 | mddev_get(mddev); | |
4787 | spin_unlock(&all_mddevs_lock); | |
4788 | ||
b7b17c9b | 4789 | rv = entry->show(mddev, page); |
af8a2434 | 4790 | mddev_put(mddev); |
96de1e66 | 4791 | return rv; |
eae1701f N |
4792 | } |
4793 | ||
4794 | static ssize_t | |
4795 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4796 | const char *page, size_t length) | |
4797 | { | |
4798 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4799 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4800 | ssize_t rv; |
eae1701f N |
4801 | |
4802 | if (!entry->store) | |
4803 | return -EIO; | |
67463acb N |
4804 | if (!capable(CAP_SYS_ADMIN)) |
4805 | return -EACCES; | |
af8a2434 N |
4806 | spin_lock(&all_mddevs_lock); |
4807 | if (list_empty(&mddev->all_mddevs)) { | |
4808 | spin_unlock(&all_mddevs_lock); | |
4809 | return -EBUSY; | |
4810 | } | |
4811 | mddev_get(mddev); | |
4812 | spin_unlock(&all_mddevs_lock); | |
6791875e | 4813 | rv = entry->store(mddev, page, length); |
af8a2434 | 4814 | mddev_put(mddev); |
96de1e66 | 4815 | return rv; |
eae1701f N |
4816 | } |
4817 | ||
4818 | static void md_free(struct kobject *ko) | |
4819 | { | |
fd01b88c | 4820 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
4821 | |
4822 | if (mddev->sysfs_state) | |
4823 | sysfs_put(mddev->sysfs_state); | |
4824 | ||
6cd18e71 N |
4825 | if (mddev->queue) |
4826 | blk_cleanup_queue(mddev->queue); | |
a21d1504 N |
4827 | if (mddev->gendisk) { |
4828 | del_gendisk(mddev->gendisk); | |
4829 | put_disk(mddev->gendisk); | |
4830 | } | |
a21d1504 | 4831 | |
eae1701f N |
4832 | kfree(mddev); |
4833 | } | |
4834 | ||
52cf25d0 | 4835 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4836 | .show = md_attr_show, |
4837 | .store = md_attr_store, | |
4838 | }; | |
4839 | static struct kobj_type md_ktype = { | |
4840 | .release = md_free, | |
4841 | .sysfs_ops = &md_sysfs_ops, | |
4842 | .default_attrs = md_default_attrs, | |
4843 | }; | |
4844 | ||
1da177e4 LT |
4845 | int mdp_major = 0; |
4846 | ||
5fd3a17e DW |
4847 | static void mddev_delayed_delete(struct work_struct *ws) |
4848 | { | |
fd01b88c | 4849 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 4850 | |
43a70507 | 4851 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4852 | kobject_del(&mddev->kobj); |
4853 | kobject_put(&mddev->kobj); | |
4854 | } | |
4855 | ||
efeb53c0 | 4856 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4857 | { |
48c9c27b | 4858 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 4859 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 4860 | struct gendisk *disk; |
efeb53c0 N |
4861 | int partitioned; |
4862 | int shift; | |
4863 | int unit; | |
3830c62f | 4864 | int error; |
1da177e4 LT |
4865 | |
4866 | if (!mddev) | |
efeb53c0 N |
4867 | return -ENODEV; |
4868 | ||
4869 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4870 | shift = partitioned ? MdpMinorShift : 0; | |
4871 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4872 | |
e804ac78 TH |
4873 | /* wait for any previous instance of this device to be |
4874 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4875 | */ |
e804ac78 | 4876 | flush_workqueue(md_misc_wq); |
d3374825 | 4877 | |
48c9c27b | 4878 | mutex_lock(&disks_mutex); |
0909dc44 N |
4879 | error = -EEXIST; |
4880 | if (mddev->gendisk) | |
4881 | goto abort; | |
efeb53c0 N |
4882 | |
4883 | if (name) { | |
4884 | /* Need to ensure that 'name' is not a duplicate. | |
4885 | */ | |
fd01b88c | 4886 | struct mddev *mddev2; |
efeb53c0 N |
4887 | spin_lock(&all_mddevs_lock); |
4888 | ||
4889 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4890 | if (mddev2->gendisk && | |
4891 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4892 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4893 | goto abort; |
efeb53c0 N |
4894 | } |
4895 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4896 | } |
8b765398 | 4897 | |
0909dc44 | 4898 | error = -ENOMEM; |
8b765398 | 4899 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4900 | if (!mddev->queue) |
4901 | goto abort; | |
409c57f3 N |
4902 | mddev->queue->queuedata = mddev; |
4903 | ||
409c57f3 | 4904 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 4905 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 4906 | |
1da177e4 LT |
4907 | disk = alloc_disk(1 << shift); |
4908 | if (!disk) { | |
8b765398 N |
4909 | blk_cleanup_queue(mddev->queue); |
4910 | mddev->queue = NULL; | |
0909dc44 | 4911 | goto abort; |
1da177e4 | 4912 | } |
efeb53c0 | 4913 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4914 | disk->first_minor = unit << shift; |
efeb53c0 N |
4915 | if (name) |
4916 | strcpy(disk->disk_name, name); | |
4917 | else if (partitioned) | |
1da177e4 | 4918 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4919 | else |
1da177e4 | 4920 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4921 | disk->fops = &md_fops; |
4922 | disk->private_data = mddev; | |
4923 | disk->queue = mddev->queue; | |
b0140891 | 4924 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); |
92850bbd | 4925 | /* Allow extended partitions. This makes the |
d3374825 | 4926 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4927 | * remove it now. |
4928 | */ | |
4929 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 4930 | mddev->gendisk = disk; |
b0140891 N |
4931 | /* As soon as we call add_disk(), another thread could get |
4932 | * through to md_open, so make sure it doesn't get too far | |
4933 | */ | |
4934 | mutex_lock(&mddev->open_mutex); | |
4935 | add_disk(disk); | |
4936 | ||
ed9e1982 TH |
4937 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4938 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4939 | if (error) { |
4940 | /* This isn't possible, but as kobject_init_and_add is marked | |
4941 | * __must_check, we must do something with the result | |
4942 | */ | |
5e55e2f5 N |
4943 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4944 | disk->disk_name); | |
0909dc44 N |
4945 | error = 0; |
4946 | } | |
00bcb4ac N |
4947 | if (mddev->kobj.sd && |
4948 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4949 | printk(KERN_DEBUG "pointless warning\n"); |
b0140891 | 4950 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
4951 | abort: |
4952 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4953 | if (!error && mddev->kobj.sd) { |
3830c62f | 4954 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4955 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4956 | } |
d3374825 | 4957 | mddev_put(mddev); |
0909dc44 | 4958 | return error; |
efeb53c0 N |
4959 | } |
4960 | ||
4961 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4962 | { | |
4963 | md_alloc(dev, NULL); | |
1da177e4 LT |
4964 | return NULL; |
4965 | } | |
4966 | ||
efeb53c0 N |
4967 | static int add_named_array(const char *val, struct kernel_param *kp) |
4968 | { | |
4969 | /* val must be "md_*" where * is not all digits. | |
4970 | * We allocate an array with a large free minor number, and | |
4971 | * set the name to val. val must not already be an active name. | |
4972 | */ | |
4973 | int len = strlen(val); | |
4974 | char buf[DISK_NAME_LEN]; | |
4975 | ||
4976 | while (len && val[len-1] == '\n') | |
4977 | len--; | |
4978 | if (len >= DISK_NAME_LEN) | |
4979 | return -E2BIG; | |
4980 | strlcpy(buf, val, len+1); | |
4981 | if (strncmp(buf, "md_", 3) != 0) | |
4982 | return -EINVAL; | |
4983 | return md_alloc(0, buf); | |
4984 | } | |
4985 | ||
1da177e4 LT |
4986 | static void md_safemode_timeout(unsigned long data) |
4987 | { | |
fd01b88c | 4988 | struct mddev *mddev = (struct mddev *) data; |
1da177e4 | 4989 | |
0fd62b86 NB |
4990 | if (!atomic_read(&mddev->writes_pending)) { |
4991 | mddev->safemode = 1; | |
4992 | if (mddev->external) | |
00bcb4ac | 4993 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4994 | } |
1da177e4 LT |
4995 | md_wakeup_thread(mddev->thread); |
4996 | } | |
4997 | ||
6ff8d8ec | 4998 | static int start_dirty_degraded; |
1da177e4 | 4999 | |
fd01b88c | 5000 | int md_run(struct mddev *mddev) |
1da177e4 | 5001 | { |
2604b703 | 5002 | int err; |
3cb03002 | 5003 | struct md_rdev *rdev; |
84fc4b56 | 5004 | struct md_personality *pers; |
1da177e4 | 5005 | |
a757e64c N |
5006 | if (list_empty(&mddev->disks)) |
5007 | /* cannot run an array with no devices.. */ | |
1da177e4 | 5008 | return -EINVAL; |
1da177e4 LT |
5009 | |
5010 | if (mddev->pers) | |
5011 | return -EBUSY; | |
bb4f1e9d N |
5012 | /* Cannot run until previous stop completes properly */ |
5013 | if (mddev->sysfs_active) | |
5014 | return -EBUSY; | |
b6eb127d | 5015 | |
1da177e4 LT |
5016 | /* |
5017 | * Analyze all RAID superblock(s) | |
5018 | */ | |
1ec4a939 N |
5019 | if (!mddev->raid_disks) { |
5020 | if (!mddev->persistent) | |
5021 | return -EINVAL; | |
a757e64c | 5022 | analyze_sbs(mddev); |
1ec4a939 | 5023 | } |
1da177e4 | 5024 | |
d9d166c2 N |
5025 | if (mddev->level != LEVEL_NONE) |
5026 | request_module("md-level-%d", mddev->level); | |
5027 | else if (mddev->clevel[0]) | |
5028 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
5029 | |
5030 | /* | |
5031 | * Drop all container device buffers, from now on | |
5032 | * the only valid external interface is through the md | |
5033 | * device. | |
1da177e4 | 5034 | */ |
dafb20fa | 5035 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 5036 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5037 | continue; |
5038 | sync_blockdev(rdev->bdev); | |
f98393a6 | 5039 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
5040 | |
5041 | /* perform some consistency tests on the device. | |
5042 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 5043 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 5044 | */ |
a6ff7e08 JB |
5045 | if (rdev->meta_bdev) { |
5046 | /* Nothing to check */; | |
5047 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
5048 | if (mddev->dev_sectors && |
5049 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 5050 | > rdev->sb_start) { |
f0d76d70 N |
5051 | printk("md: %s: data overlaps metadata\n", |
5052 | mdname(mddev)); | |
5053 | return -EINVAL; | |
5054 | } | |
5055 | } else { | |
0f420358 | 5056 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
5057 | > rdev->data_offset) { |
5058 | printk("md: %s: metadata overlaps data\n", | |
5059 | mdname(mddev)); | |
5060 | return -EINVAL; | |
5061 | } | |
5062 | } | |
00bcb4ac | 5063 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
5064 | } |
5065 | ||
a167f663 | 5066 | if (mddev->bio_set == NULL) |
395c72a7 | 5067 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0); |
a167f663 | 5068 | |
1da177e4 | 5069 | spin_lock(&pers_lock); |
d9d166c2 | 5070 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 5071 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 5072 | spin_unlock(&pers_lock); |
d9d166c2 N |
5073 | if (mddev->level != LEVEL_NONE) |
5074 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
5075 | mddev->level); | |
5076 | else | |
5077 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
5078 | mddev->clevel); | |
1da177e4 LT |
5079 | return -EINVAL; |
5080 | } | |
1da177e4 | 5081 | spin_unlock(&pers_lock); |
34817e8c N |
5082 | if (mddev->level != pers->level) { |
5083 | mddev->level = pers->level; | |
5084 | mddev->new_level = pers->level; | |
5085 | } | |
d9d166c2 | 5086 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5087 | |
f6705578 | 5088 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5089 | pers->start_reshape == NULL) { |
f6705578 | 5090 | /* This personality cannot handle reshaping... */ |
f6705578 N |
5091 | module_put(pers->owner); |
5092 | return -EINVAL; | |
5093 | } | |
5094 | ||
7dd5e7c3 N |
5095 | if (pers->sync_request) { |
5096 | /* Warn if this is a potentially silly | |
5097 | * configuration. | |
5098 | */ | |
5099 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5100 | struct md_rdev *rdev2; |
7dd5e7c3 | 5101 | int warned = 0; |
159ec1fc | 5102 | |
dafb20fa N |
5103 | rdev_for_each(rdev, mddev) |
5104 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5105 | if (rdev < rdev2 && |
5106 | rdev->bdev->bd_contains == | |
5107 | rdev2->bdev->bd_contains) { | |
5108 | printk(KERN_WARNING | |
5109 | "%s: WARNING: %s appears to be" | |
5110 | " on the same physical disk as" | |
5111 | " %s.\n", | |
5112 | mdname(mddev), | |
5113 | bdevname(rdev->bdev,b), | |
5114 | bdevname(rdev2->bdev,b2)); | |
5115 | warned = 1; | |
5116 | } | |
5117 | } | |
159ec1fc | 5118 | |
7dd5e7c3 N |
5119 | if (warned) |
5120 | printk(KERN_WARNING | |
5121 | "True protection against single-disk" | |
5122 | " failure might be compromised.\n"); | |
5123 | } | |
5124 | ||
657390d2 | 5125 | mddev->recovery = 0; |
58c0fed4 AN |
5126 | /* may be over-ridden by personality */ |
5127 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5128 | ||
6ff8d8ec | 5129 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5130 | |
0f9552b5 | 5131 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5132 | mddev->ro = 2; /* read-only, but switch on first write */ |
5133 | ||
36d091f4 | 5134 | err = pers->run(mddev); |
13e53df3 AN |
5135 | if (err) |
5136 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
36d091f4 | 5137 | else if (pers->size(mddev, 0, 0) < mddev->array_sectors) { |
b522adcd DW |
5138 | WARN_ONCE(!mddev->external_size, "%s: default size too small," |
5139 | " but 'external_size' not in effect?\n", __func__); | |
5140 | printk(KERN_ERR | |
5141 | "md: invalid array_size %llu > default size %llu\n", | |
5142 | (unsigned long long)mddev->array_sectors / 2, | |
36d091f4 | 5143 | (unsigned long long)pers->size(mddev, 0, 0) / 2); |
b522adcd | 5144 | err = -EINVAL; |
b522adcd | 5145 | } |
36d091f4 | 5146 | if (err == 0 && pers->sync_request && |
ef99bf48 | 5147 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { |
f9209a32 GR |
5148 | struct bitmap *bitmap; |
5149 | ||
5150 | bitmap = bitmap_create(mddev, -1); | |
5151 | if (IS_ERR(bitmap)) { | |
5152 | err = PTR_ERR(bitmap); | |
b15c2e57 N |
5153 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", |
5154 | mdname(mddev), err); | |
f9209a32 GR |
5155 | } else |
5156 | mddev->bitmap = bitmap; | |
5157 | ||
b15c2e57 | 5158 | } |
1da177e4 | 5159 | if (err) { |
5aa61f42 | 5160 | mddev_detach(mddev); |
0c35bd47 N |
5161 | if (mddev->private) |
5162 | pers->free(mddev, mddev->private); | |
bd691922 | 5163 | mddev->private = NULL; |
36d091f4 | 5164 | module_put(pers->owner); |
32a7627c N |
5165 | bitmap_destroy(mddev); |
5166 | return err; | |
1da177e4 | 5167 | } |
5c675f83 N |
5168 | if (mddev->queue) { |
5169 | mddev->queue->backing_dev_info.congested_data = mddev; | |
5170 | mddev->queue->backing_dev_info.congested_fn = md_congested; | |
5171 | } | |
36d091f4 | 5172 | if (pers->sync_request) { |
00bcb4ac N |
5173 | if (mddev->kobj.sd && |
5174 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
5175 | printk(KERN_WARNING |
5176 | "md: cannot register extra attributes for %s\n", | |
5177 | mdname(mddev)); | |
00bcb4ac | 5178 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5179 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5180 | mddev->ro = 0; |
5181 | ||
f72ffdd6 | 5182 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
5183 | atomic_set(&mddev->max_corr_read_errors, |
5184 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 | 5185 | mddev->safemode = 0; |
16f17b39 | 5186 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 5187 | mddev->in_sync = 1; |
0ca69886 | 5188 | smp_wmb(); |
36d091f4 N |
5189 | spin_lock(&mddev->lock); |
5190 | mddev->pers = pers; | |
0ca69886 | 5191 | mddev->ready = 1; |
36d091f4 | 5192 | spin_unlock(&mddev->lock); |
dafb20fa | 5193 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5194 | if (rdev->raid_disk >= 0) |
5195 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5196 | /* failure here is OK */; |
f72ffdd6 | 5197 | |
a4a3d26d N |
5198 | if (mddev->degraded && !mddev->ro) |
5199 | /* This ensures that recovering status is reported immediately | |
5200 | * via sysfs - until a lack of spares is confirmed. | |
5201 | */ | |
5202 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
1da177e4 | 5203 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f72ffdd6 | 5204 | |
7a0a5355 | 5205 | if (mddev->flags & MD_UPDATE_SB_FLAGS) |
850b2b42 | 5206 | md_update_sb(mddev, 0); |
1da177e4 | 5207 | |
d7603b7e | 5208 | md_new_event(mddev); |
00bcb4ac N |
5209 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5210 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5211 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
5212 | return 0; |
5213 | } | |
390ee602 | 5214 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5215 | |
fd01b88c | 5216 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5217 | { |
5218 | int err; | |
5219 | ||
5220 | err = md_run(mddev); | |
5221 | if (err) | |
5222 | goto out; | |
69e51b44 N |
5223 | err = bitmap_load(mddev); |
5224 | if (err) { | |
5225 | bitmap_destroy(mddev); | |
5226 | goto out; | |
5227 | } | |
0fd018af JB |
5228 | |
5229 | md_wakeup_thread(mddev->thread); | |
5230 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5231 | ||
fe60b014 N |
5232 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5233 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5234 | mddev->changed = 1; |
fe60b014 N |
5235 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5236 | out: | |
5237 | return err; | |
5238 | } | |
5239 | ||
fd01b88c | 5240 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5241 | { |
5242 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 5243 | |
80fab1d7 | 5244 | /* Complain if it has no devices */ |
1da177e4 | 5245 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5246 | return -ENXIO; |
5247 | if (!mddev->pers) | |
5248 | return -EINVAL; | |
5249 | if (!mddev->ro) | |
5250 | return -EBUSY; | |
5251 | mddev->safemode = 0; | |
5252 | mddev->ro = 0; | |
5253 | set_disk_ro(disk, 0); | |
5254 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
5255 | mdname(mddev)); | |
5256 | /* Kick recovery or resync if necessary */ | |
5257 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5258 | md_wakeup_thread(mddev->thread); | |
5259 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5260 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5261 | return 0; |
1da177e4 LT |
5262 | } |
5263 | ||
fd01b88c | 5264 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5265 | { |
5266 | mddev->array_sectors = 0; | |
5267 | mddev->external_size = 0; | |
5268 | mddev->dev_sectors = 0; | |
5269 | mddev->raid_disks = 0; | |
5270 | mddev->recovery_cp = 0; | |
5271 | mddev->resync_min = 0; | |
5272 | mddev->resync_max = MaxSector; | |
5273 | mddev->reshape_position = MaxSector; | |
5274 | mddev->external = 0; | |
5275 | mddev->persistent = 0; | |
5276 | mddev->level = LEVEL_NONE; | |
5277 | mddev->clevel[0] = 0; | |
5278 | mddev->flags = 0; | |
5279 | mddev->ro = 0; | |
5280 | mddev->metadata_type[0] = 0; | |
5281 | mddev->chunk_sectors = 0; | |
5282 | mddev->ctime = mddev->utime = 0; | |
5283 | mddev->layout = 0; | |
5284 | mddev->max_disks = 0; | |
5285 | mddev->events = 0; | |
a8707c08 | 5286 | mddev->can_decrease_events = 0; |
6177b472 | 5287 | mddev->delta_disks = 0; |
2c810cdd | 5288 | mddev->reshape_backwards = 0; |
6177b472 N |
5289 | mddev->new_level = LEVEL_NONE; |
5290 | mddev->new_layout = 0; | |
5291 | mddev->new_chunk_sectors = 0; | |
5292 | mddev->curr_resync = 0; | |
7f7583d4 | 5293 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5294 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5295 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5296 | mddev->recovery = 0; | |
5297 | mddev->in_sync = 0; | |
f0b4f7e2 | 5298 | mddev->changed = 0; |
6177b472 | 5299 | mddev->degraded = 0; |
6177b472 | 5300 | mddev->safemode = 0; |
bd691922 | 5301 | mddev->private = NULL; |
6177b472 N |
5302 | mddev->bitmap_info.offset = 0; |
5303 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5304 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5305 | mddev->bitmap_info.chunksize = 0; |
5306 | mddev->bitmap_info.daemon_sleep = 0; | |
5307 | mddev->bitmap_info.max_write_behind = 0; | |
5308 | } | |
5309 | ||
fd01b88c | 5310 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5311 | { |
293467aa GR |
5312 | if (mddev_is_clustered(mddev)) |
5313 | md_cluster_ops->metadata_update_start(mddev); | |
6b6204ee | 5314 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
f851b60d | 5315 | flush_workqueue(md_misc_wq); |
a047e125 | 5316 | if (mddev->sync_thread) { |
a047e125 | 5317 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5318 | md_reap_sync_thread(mddev); |
a047e125 N |
5319 | } |
5320 | ||
5321 | del_timer_sync(&mddev->safemode_timer); | |
5322 | ||
5323 | bitmap_flush(mddev); | |
5324 | md_super_wait(mddev); | |
5325 | ||
b6d428c6 | 5326 | if (mddev->ro == 0 && |
7a0a5355 | 5327 | (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) { |
a047e125 N |
5328 | /* mark array as shutdown cleanly */ |
5329 | mddev->in_sync = 1; | |
5330 | md_update_sb(mddev, 1); | |
5331 | } | |
293467aa GR |
5332 | if (mddev_is_clustered(mddev)) |
5333 | md_cluster_ops->metadata_update_finish(mddev); | |
a047e125 | 5334 | } |
defad61a | 5335 | |
fd01b88c | 5336 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5337 | { |
29f097c4 | 5338 | mddev_lock_nointr(mddev); |
defad61a N |
5339 | __md_stop_writes(mddev); |
5340 | mddev_unlock(mddev); | |
5341 | } | |
390ee602 | 5342 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5343 | |
5aa61f42 N |
5344 | static void mddev_detach(struct mddev *mddev) |
5345 | { | |
5346 | struct bitmap *bitmap = mddev->bitmap; | |
5347 | /* wait for behind writes to complete */ | |
5348 | if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { | |
5349 | printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n", | |
5350 | mdname(mddev)); | |
5351 | /* need to kick something here to make sure I/O goes? */ | |
5352 | wait_event(bitmap->behind_wait, | |
5353 | atomic_read(&bitmap->behind_writes) == 0); | |
5354 | } | |
36d091f4 | 5355 | if (mddev->pers && mddev->pers->quiesce) { |
5aa61f42 N |
5356 | mddev->pers->quiesce(mddev, 1); |
5357 | mddev->pers->quiesce(mddev, 0); | |
5358 | } | |
5359 | md_unregister_thread(&mddev->thread); | |
5360 | if (mddev->queue) | |
5361 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
5362 | } | |
5363 | ||
5eff3c43 | 5364 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5365 | { |
36d091f4 | 5366 | struct md_personality *pers = mddev->pers; |
5aa61f42 | 5367 | mddev_detach(mddev); |
ee5d004f N |
5368 | /* Ensure ->event_work is done */ |
5369 | flush_workqueue(md_misc_wq); | |
36d091f4 N |
5370 | spin_lock(&mddev->lock); |
5371 | mddev->ready = 0; | |
6177b472 | 5372 | mddev->pers = NULL; |
36d091f4 N |
5373 | spin_unlock(&mddev->lock); |
5374 | pers->free(mddev, mddev->private); | |
bd691922 | 5375 | mddev->private = NULL; |
36d091f4 N |
5376 | if (pers->sync_request && mddev->to_remove == NULL) |
5377 | mddev->to_remove = &md_redundancy_group; | |
5378 | module_put(pers->owner); | |
cca9cf90 | 5379 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5380 | } |
5eff3c43 N |
5381 | |
5382 | void md_stop(struct mddev *mddev) | |
5383 | { | |
5384 | /* stop the array and free an attached data structures. | |
5385 | * This is called from dm-raid | |
5386 | */ | |
5387 | __md_stop(mddev); | |
5388 | bitmap_destroy(mddev); | |
5389 | if (mddev->bio_set) | |
5390 | bioset_free(mddev->bio_set); | |
5391 | } | |
5392 | ||
390ee602 | 5393 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5394 | |
a05b7ea0 | 5395 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5396 | { |
5397 | int err = 0; | |
30b8feb7 N |
5398 | int did_freeze = 0; |
5399 | ||
5400 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5401 | did_freeze = 1; | |
5402 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5403 | md_wakeup_thread(mddev->thread); | |
5404 | } | |
f851b60d | 5405 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5406 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5407 | if (mddev->sync_thread) |
30b8feb7 N |
5408 | /* Thread might be blocked waiting for metadata update |
5409 | * which will now never happen */ | |
5410 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5411 | |
88724bfa N |
5412 | if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
5413 | return -EBUSY; | |
30b8feb7 | 5414 | mddev_unlock(mddev); |
f851b60d N |
5415 | wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING, |
5416 | &mddev->recovery)); | |
88724bfa N |
5417 | wait_event(mddev->sb_wait, |
5418 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
30b8feb7 N |
5419 | mddev_lock_nointr(mddev); |
5420 | ||
a4bd82d0 | 5421 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5422 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 | 5423 | mddev->sync_thread || |
f851b60d | 5424 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
30b8feb7 | 5425 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { |
a4bd82d0 | 5426 | printk("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5427 | if (did_freeze) { |
5428 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5429 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5430 | md_wakeup_thread(mddev->thread); |
5431 | } | |
a4bd82d0 N |
5432 | err = -EBUSY; |
5433 | goto out; | |
5434 | } | |
5435 | if (mddev->pers) { | |
defad61a | 5436 | __md_stop_writes(mddev); |
a4bd82d0 N |
5437 | |
5438 | err = -ENXIO; | |
5439 | if (mddev->ro==1) | |
5440 | goto out; | |
5441 | mddev->ro = 1; | |
5442 | set_disk_ro(mddev->gendisk, 1); | |
5443 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d N |
5444 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5445 | md_wakeup_thread(mddev->thread); | |
00bcb4ac | 5446 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5447 | err = 0; |
a4bd82d0 N |
5448 | } |
5449 | out: | |
5450 | mutex_unlock(&mddev->open_mutex); | |
5451 | return err; | |
5452 | } | |
5453 | ||
9e653b63 N |
5454 | /* mode: |
5455 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5456 | * 2 - stop but do not disassemble array |
5457 | */ | |
f72ffdd6 | 5458 | static int do_md_stop(struct mddev *mddev, int mode, |
a05b7ea0 | 5459 | struct block_device *bdev) |
1da177e4 | 5460 | { |
1da177e4 | 5461 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5462 | struct md_rdev *rdev; |
30b8feb7 N |
5463 | int did_freeze = 0; |
5464 | ||
5465 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5466 | did_freeze = 1; | |
5467 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5468 | md_wakeup_thread(mddev->thread); | |
5469 | } | |
f851b60d | 5470 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5471 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5472 | if (mddev->sync_thread) |
30b8feb7 N |
5473 | /* Thread might be blocked waiting for metadata update |
5474 | * which will now never happen */ | |
5475 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5476 | |
30b8feb7 | 5477 | mddev_unlock(mddev); |
f851b60d N |
5478 | wait_event(resync_wait, (mddev->sync_thread == NULL && |
5479 | !test_bit(MD_RECOVERY_RUNNING, | |
5480 | &mddev->recovery))); | |
30b8feb7 | 5481 | mddev_lock_nointr(mddev); |
1da177e4 | 5482 | |
c8c00a69 | 5483 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5484 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5485 | mddev->sysfs_active || |
5486 | mddev->sync_thread || | |
f851b60d | 5487 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
30b8feb7 | 5488 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { |
df5b20cf | 5489 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 5490 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
5491 | if (did_freeze) { |
5492 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5493 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5494 | md_wakeup_thread(mddev->thread); |
5495 | } | |
260fa034 N |
5496 | return -EBUSY; |
5497 | } | |
6e17b027 | 5498 | if (mddev->pers) { |
a4bd82d0 N |
5499 | if (mddev->ro) |
5500 | set_disk_ro(disk, 0); | |
409c57f3 | 5501 | |
defad61a | 5502 | __md_stop_writes(mddev); |
5eff3c43 | 5503 | __md_stop(mddev); |
a4bd82d0 | 5504 | mddev->queue->backing_dev_info.congested_fn = NULL; |
6177b472 | 5505 | |
a4bd82d0 | 5506 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5507 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5508 | |
dafb20fa | 5509 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5510 | if (rdev->raid_disk >= 0) |
5511 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5512 | |
a4bd82d0 | 5513 | set_capacity(disk, 0); |
6e17b027 | 5514 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5515 | mddev->changed = 1; |
a4bd82d0 | 5516 | revalidate_disk(disk); |
0d4ca600 | 5517 | |
a4bd82d0 N |
5518 | if (mddev->ro) |
5519 | mddev->ro = 0; | |
6e17b027 N |
5520 | } else |
5521 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5522 | /* |
5523 | * Free resources if final stop | |
5524 | */ | |
9e653b63 | 5525 | if (mode == 0) { |
1da177e4 LT |
5526 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
5527 | ||
978f946b | 5528 | bitmap_destroy(mddev); |
c3d9714e | 5529 | if (mddev->bitmap_info.file) { |
4af1a041 N |
5530 | struct file *f = mddev->bitmap_info.file; |
5531 | spin_lock(&mddev->lock); | |
c3d9714e | 5532 | mddev->bitmap_info.file = NULL; |
4af1a041 N |
5533 | spin_unlock(&mddev->lock); |
5534 | fput(f); | |
978f946b | 5535 | } |
c3d9714e | 5536 | mddev->bitmap_info.offset = 0; |
978f946b | 5537 | |
1da177e4 LT |
5538 | export_array(mddev); |
5539 | ||
6177b472 | 5540 | md_clean(mddev); |
934d9c23 | 5541 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
5542 | if (mddev->hold_active == UNTIL_STOP) |
5543 | mddev->hold_active = 0; | |
a4bd82d0 | 5544 | } |
3f9d99c1 | 5545 | blk_integrity_unregister(disk); |
d7603b7e | 5546 | md_new_event(mddev); |
00bcb4ac | 5547 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 5548 | return 0; |
1da177e4 LT |
5549 | } |
5550 | ||
fdee8ae4 | 5551 | #ifndef MODULE |
fd01b88c | 5552 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 5553 | { |
3cb03002 | 5554 | struct md_rdev *rdev; |
1da177e4 LT |
5555 | int err; |
5556 | ||
a757e64c | 5557 | if (list_empty(&mddev->disks)) |
1da177e4 | 5558 | return; |
1da177e4 LT |
5559 | |
5560 | printk(KERN_INFO "md: running: "); | |
5561 | ||
dafb20fa | 5562 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
5563 | char b[BDEVNAME_SIZE]; |
5564 | printk("<%s>", bdevname(rdev->bdev,b)); | |
5565 | } | |
5566 | printk("\n"); | |
5567 | ||
d710e138 | 5568 | err = do_md_run(mddev); |
1da177e4 LT |
5569 | if (err) { |
5570 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
a05b7ea0 | 5571 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
5572 | } |
5573 | } | |
5574 | ||
5575 | /* | |
5576 | * lets try to run arrays based on all disks that have arrived | |
5577 | * until now. (those are in pending_raid_disks) | |
5578 | * | |
5579 | * the method: pick the first pending disk, collect all disks with | |
5580 | * the same UUID, remove all from the pending list and put them into | |
5581 | * the 'same_array' list. Then order this list based on superblock | |
5582 | * update time (freshest comes first), kick out 'old' disks and | |
5583 | * compare superblocks. If everything's fine then run it. | |
5584 | * | |
5585 | * If "unit" is allocated, then bump its reference count | |
5586 | */ | |
5587 | static void autorun_devices(int part) | |
5588 | { | |
3cb03002 | 5589 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 5590 | struct mddev *mddev; |
1da177e4 LT |
5591 | char b[BDEVNAME_SIZE]; |
5592 | ||
5593 | printk(KERN_INFO "md: autorun ...\n"); | |
5594 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 5595 | int unit; |
1da177e4 | 5596 | dev_t dev; |
ad01c9e3 | 5597 | LIST_HEAD(candidates); |
1da177e4 | 5598 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 5599 | struct md_rdev, same_set); |
1da177e4 LT |
5600 | |
5601 | printk(KERN_INFO "md: considering %s ...\n", | |
5602 | bdevname(rdev0->bdev,b)); | |
5603 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 5604 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
5605 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
5606 | printk(KERN_INFO "md: adding %s ...\n", | |
5607 | bdevname(rdev->bdev,b)); | |
5608 | list_move(&rdev->same_set, &candidates); | |
5609 | } | |
5610 | /* | |
5611 | * now we have a set of devices, with all of them having | |
5612 | * mostly sane superblocks. It's time to allocate the | |
5613 | * mddev. | |
5614 | */ | |
e8703fe1 N |
5615 | if (part) { |
5616 | dev = MKDEV(mdp_major, | |
5617 | rdev0->preferred_minor << MdpMinorShift); | |
5618 | unit = MINOR(dev) >> MdpMinorShift; | |
5619 | } else { | |
5620 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
5621 | unit = MINOR(dev); | |
5622 | } | |
5623 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
5624 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
5625 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
5626 | break; | |
5627 | } | |
1da177e4 LT |
5628 | |
5629 | md_probe(dev, NULL, NULL); | |
5630 | mddev = mddev_find(dev); | |
9bbbca3a NB |
5631 | if (!mddev || !mddev->gendisk) { |
5632 | if (mddev) | |
5633 | mddev_put(mddev); | |
5634 | printk(KERN_ERR | |
1da177e4 LT |
5635 | "md: cannot allocate memory for md drive.\n"); |
5636 | break; | |
5637 | } | |
f72ffdd6 | 5638 | if (mddev_lock(mddev)) |
1da177e4 LT |
5639 | printk(KERN_WARNING "md: %s locked, cannot run\n", |
5640 | mdname(mddev)); | |
5641 | else if (mddev->raid_disks || mddev->major_version | |
5642 | || !list_empty(&mddev->disks)) { | |
f72ffdd6 | 5643 | printk(KERN_WARNING |
1da177e4 LT |
5644 | "md: %s already running, cannot run %s\n", |
5645 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
5646 | mddev_unlock(mddev); | |
5647 | } else { | |
5648 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 5649 | mddev->persistent = 1; |
159ec1fc | 5650 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
5651 | list_del_init(&rdev->same_set); |
5652 | if (bind_rdev_to_array(rdev, mddev)) | |
5653 | export_rdev(rdev); | |
5654 | } | |
5655 | autorun_array(mddev); | |
5656 | mddev_unlock(mddev); | |
5657 | } | |
5658 | /* on success, candidates will be empty, on error | |
5659 | * it won't... | |
5660 | */ | |
159ec1fc | 5661 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 5662 | list_del_init(&rdev->same_set); |
1da177e4 | 5663 | export_rdev(rdev); |
4b80991c | 5664 | } |
1da177e4 LT |
5665 | mddev_put(mddev); |
5666 | } | |
5667 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
5668 | } | |
fdee8ae4 | 5669 | #endif /* !MODULE */ |
1da177e4 | 5670 | |
f72ffdd6 | 5671 | static int get_version(void __user *arg) |
1da177e4 LT |
5672 | { |
5673 | mdu_version_t ver; | |
5674 | ||
5675 | ver.major = MD_MAJOR_VERSION; | |
5676 | ver.minor = MD_MINOR_VERSION; | |
5677 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
5678 | ||
5679 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
5680 | return -EFAULT; | |
5681 | ||
5682 | return 0; | |
5683 | } | |
5684 | ||
f72ffdd6 | 5685 | static int get_array_info(struct mddev *mddev, void __user *arg) |
1da177e4 LT |
5686 | { |
5687 | mdu_array_info_t info; | |
a9f326eb | 5688 | int nr,working,insync,failed,spare; |
3cb03002 | 5689 | struct md_rdev *rdev; |
1da177e4 | 5690 | |
1ca69c4b N |
5691 | nr = working = insync = failed = spare = 0; |
5692 | rcu_read_lock(); | |
5693 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 5694 | nr++; |
b2d444d7 | 5695 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5696 | failed++; |
5697 | else { | |
5698 | working++; | |
b2d444d7 | 5699 | if (test_bit(In_sync, &rdev->flags)) |
f72ffdd6 | 5700 | insync++; |
1da177e4 LT |
5701 | else |
5702 | spare++; | |
5703 | } | |
5704 | } | |
1ca69c4b | 5705 | rcu_read_unlock(); |
1da177e4 LT |
5706 | |
5707 | info.major_version = mddev->major_version; | |
5708 | info.minor_version = mddev->minor_version; | |
5709 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5710 | info.ctime = mddev->ctime; | |
5711 | info.level = mddev->level; | |
58c0fed4 AN |
5712 | info.size = mddev->dev_sectors / 2; |
5713 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5714 | info.size = -1; |
1da177e4 LT |
5715 | info.nr_disks = nr; |
5716 | info.raid_disks = mddev->raid_disks; | |
5717 | info.md_minor = mddev->md_minor; | |
5718 | info.not_persistent= !mddev->persistent; | |
5719 | ||
5720 | info.utime = mddev->utime; | |
5721 | info.state = 0; | |
5722 | if (mddev->in_sync) | |
5723 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5724 | if (mddev->bitmap && mddev->bitmap_info.offset) |
9bd35920 | 5725 | info.state |= (1<<MD_SB_BITMAP_PRESENT); |
ca8895d9 GR |
5726 | if (mddev_is_clustered(mddev)) |
5727 | info.state |= (1<<MD_SB_CLUSTERED); | |
a9f326eb | 5728 | info.active_disks = insync; |
1da177e4 LT |
5729 | info.working_disks = working; |
5730 | info.failed_disks = failed; | |
5731 | info.spare_disks = spare; | |
5732 | ||
5733 | info.layout = mddev->layout; | |
9d8f0363 | 5734 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5735 | |
5736 | if (copy_to_user(arg, &info, sizeof(info))) | |
5737 | return -EFAULT; | |
5738 | ||
5739 | return 0; | |
5740 | } | |
5741 | ||
f72ffdd6 | 5742 | static int get_bitmap_file(struct mddev *mddev, void __user * arg) |
32a7627c N |
5743 | { |
5744 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
f4ad3d38 | 5745 | char *ptr; |
4af1a041 | 5746 | int err; |
32a7627c | 5747 | |
b6878d9e | 5748 | file = kzalloc(sizeof(*file), GFP_NOIO); |
32a7627c | 5749 | if (!file) |
4af1a041 | 5750 | return -ENOMEM; |
32a7627c | 5751 | |
4af1a041 N |
5752 | err = 0; |
5753 | spin_lock(&mddev->lock); | |
25eafe1a BR |
5754 | /* bitmap enabled */ |
5755 | if (mddev->bitmap_info.file) { | |
5756 | ptr = file_path(mddev->bitmap_info.file, file->pathname, | |
5757 | sizeof(file->pathname)); | |
5758 | if (IS_ERR(ptr)) | |
5759 | err = PTR_ERR(ptr); | |
5760 | else | |
5761 | memmove(file->pathname, ptr, | |
5762 | sizeof(file->pathname)-(ptr-file->pathname)); | |
5763 | } | |
4af1a041 | 5764 | spin_unlock(&mddev->lock); |
32a7627c | 5765 | |
4af1a041 N |
5766 | if (err == 0 && |
5767 | copy_to_user(arg, file, sizeof(*file))) | |
32a7627c | 5768 | err = -EFAULT; |
4af1a041 | 5769 | |
32a7627c N |
5770 | kfree(file); |
5771 | return err; | |
5772 | } | |
5773 | ||
f72ffdd6 | 5774 | static int get_disk_info(struct mddev *mddev, void __user * arg) |
1da177e4 LT |
5775 | { |
5776 | mdu_disk_info_t info; | |
3cb03002 | 5777 | struct md_rdev *rdev; |
1da177e4 LT |
5778 | |
5779 | if (copy_from_user(&info, arg, sizeof(info))) | |
5780 | return -EFAULT; | |
5781 | ||
1ca69c4b | 5782 | rcu_read_lock(); |
57d051dc | 5783 | rdev = md_find_rdev_nr_rcu(mddev, info.number); |
1da177e4 LT |
5784 | if (rdev) { |
5785 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5786 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5787 | info.raid_disk = rdev->raid_disk; | |
5788 | info.state = 0; | |
b2d444d7 | 5789 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5790 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5791 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5792 | info.state |= (1<<MD_DISK_ACTIVE); |
5793 | info.state |= (1<<MD_DISK_SYNC); | |
5794 | } | |
8ddf9efe N |
5795 | if (test_bit(WriteMostly, &rdev->flags)) |
5796 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5797 | } else { |
5798 | info.major = info.minor = 0; | |
5799 | info.raid_disk = -1; | |
5800 | info.state = (1<<MD_DISK_REMOVED); | |
5801 | } | |
1ca69c4b | 5802 | rcu_read_unlock(); |
1da177e4 LT |
5803 | |
5804 | if (copy_to_user(arg, &info, sizeof(info))) | |
5805 | return -EFAULT; | |
5806 | ||
5807 | return 0; | |
5808 | } | |
5809 | ||
f72ffdd6 | 5810 | static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info) |
1da177e4 LT |
5811 | { |
5812 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5813 | struct md_rdev *rdev; |
1da177e4 LT |
5814 | dev_t dev = MKDEV(info->major,info->minor); |
5815 | ||
1aee41f6 GR |
5816 | if (mddev_is_clustered(mddev) && |
5817 | !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) { | |
fa8259da | 5818 | pr_err("%s: Cannot add to clustered mddev.\n", |
1aee41f6 GR |
5819 | mdname(mddev)); |
5820 | return -EINVAL; | |
5821 | } | |
5822 | ||
1da177e4 LT |
5823 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) |
5824 | return -EOVERFLOW; | |
5825 | ||
5826 | if (!mddev->raid_disks) { | |
5827 | int err; | |
5828 | /* expecting a device which has a superblock */ | |
5829 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5830 | if (IS_ERR(rdev)) { | |
f72ffdd6 | 5831 | printk(KERN_WARNING |
1da177e4 LT |
5832 | "md: md_import_device returned %ld\n", |
5833 | PTR_ERR(rdev)); | |
5834 | return PTR_ERR(rdev); | |
5835 | } | |
5836 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
5837 | struct md_rdev *rdev0 |
5838 | = list_entry(mddev->disks.next, | |
5839 | struct md_rdev, same_set); | |
a9f326eb | 5840 | err = super_types[mddev->major_version] |
1da177e4 LT |
5841 | .load_super(rdev, rdev0, mddev->minor_version); |
5842 | if (err < 0) { | |
f72ffdd6 | 5843 | printk(KERN_WARNING |
1da177e4 | 5844 | "md: %s has different UUID to %s\n", |
f72ffdd6 | 5845 | bdevname(rdev->bdev,b), |
1da177e4 LT |
5846 | bdevname(rdev0->bdev,b2)); |
5847 | export_rdev(rdev); | |
5848 | return -EINVAL; | |
5849 | } | |
5850 | } | |
5851 | err = bind_rdev_to_array(rdev, mddev); | |
5852 | if (err) | |
5853 | export_rdev(rdev); | |
5854 | return err; | |
5855 | } | |
5856 | ||
5857 | /* | |
5858 | * add_new_disk can be used once the array is assembled | |
5859 | * to add "hot spares". They must already have a superblock | |
5860 | * written | |
5861 | */ | |
5862 | if (mddev->pers) { | |
5863 | int err; | |
5864 | if (!mddev->pers->hot_add_disk) { | |
f72ffdd6 | 5865 | printk(KERN_WARNING |
1da177e4 LT |
5866 | "%s: personality does not support diskops!\n", |
5867 | mdname(mddev)); | |
5868 | return -EINVAL; | |
5869 | } | |
7b1e35f6 N |
5870 | if (mddev->persistent) |
5871 | rdev = md_import_device(dev, mddev->major_version, | |
5872 | mddev->minor_version); | |
5873 | else | |
5874 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 | 5875 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5876 | printk(KERN_WARNING |
1da177e4 LT |
5877 | "md: md_import_device returned %ld\n", |
5878 | PTR_ERR(rdev)); | |
5879 | return PTR_ERR(rdev); | |
5880 | } | |
1a855a06 | 5881 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5882 | if (!mddev->persistent) { |
5883 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5884 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5885 | rdev->raid_disk = info->raid_disk; |
bf572541 | 5886 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 5887 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 5888 | } else |
41158c7e | 5889 | rdev->raid_disk = -1; |
f466722c | 5890 | rdev->saved_raid_disk = rdev->raid_disk; |
41158c7e N |
5891 | } else |
5892 | super_types[mddev->major_version]. | |
5893 | validate_super(mddev, rdev); | |
bedd86b7 | 5894 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 5895 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
5896 | /* This was a hot-add request, but events doesn't |
5897 | * match, so reject it. | |
5898 | */ | |
5899 | export_rdev(rdev); | |
5900 | return -EINVAL; | |
5901 | } | |
5902 | ||
b2d444d7 | 5903 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5904 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5905 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5906 | else |
5907 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5908 | |
1aee41f6 GR |
5909 | /* |
5910 | * check whether the device shows up in other nodes | |
5911 | */ | |
5912 | if (mddev_is_clustered(mddev)) { | |
5913 | if (info->state & (1 << MD_DISK_CANDIDATE)) { | |
5914 | /* Through --cluster-confirm */ | |
5915 | set_bit(Candidate, &rdev->flags); | |
fa8259da GR |
5916 | err = md_cluster_ops->new_disk_ack(mddev, true); |
5917 | if (err) { | |
5918 | export_rdev(rdev); | |
5919 | return err; | |
5920 | } | |
1aee41f6 GR |
5921 | } else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) { |
5922 | /* --add initiated by this node */ | |
5923 | err = md_cluster_ops->add_new_disk_start(mddev, rdev); | |
5924 | if (err) { | |
5925 | md_cluster_ops->add_new_disk_finish(mddev); | |
5926 | export_rdev(rdev); | |
5927 | return err; | |
5928 | } | |
5929 | } | |
5930 | } | |
5931 | ||
1da177e4 LT |
5932 | rdev->raid_disk = -1; |
5933 | err = bind_rdev_to_array(rdev, mddev); | |
5934 | if (err) | |
5935 | export_rdev(rdev); | |
52664732 | 5936 | else |
a6da4ef8 | 5937 | err = add_bound_rdev(rdev); |
1aee41f6 GR |
5938 | if (mddev_is_clustered(mddev) && |
5939 | (info->state & (1 << MD_DISK_CLUSTER_ADD))) | |
5940 | md_cluster_ops->add_new_disk_finish(mddev); | |
1da177e4 LT |
5941 | return err; |
5942 | } | |
5943 | ||
5944 | /* otherwise, add_new_disk is only allowed | |
5945 | * for major_version==0 superblocks | |
5946 | */ | |
5947 | if (mddev->major_version != 0) { | |
5948 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5949 | mdname(mddev)); | |
5950 | return -EINVAL; | |
5951 | } | |
5952 | ||
5953 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5954 | int err; | |
d710e138 | 5955 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 5956 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5957 | printk(KERN_WARNING |
1da177e4 LT |
5958 | "md: error, md_import_device() returned %ld\n", |
5959 | PTR_ERR(rdev)); | |
5960 | return PTR_ERR(rdev); | |
5961 | } | |
5962 | rdev->desc_nr = info->number; | |
5963 | if (info->raid_disk < mddev->raid_disks) | |
5964 | rdev->raid_disk = info->raid_disk; | |
5965 | else | |
5966 | rdev->raid_disk = -1; | |
5967 | ||
1da177e4 | 5968 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5969 | if (info->state & (1<<MD_DISK_SYNC)) |
5970 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5971 | |
8ddf9efe N |
5972 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5973 | set_bit(WriteMostly, &rdev->flags); | |
5974 | ||
1da177e4 LT |
5975 | if (!mddev->persistent) { |
5976 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5977 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5978 | } else | |
57b2caa3 | 5979 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 5980 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5981 | |
2bf071bf N |
5982 | err = bind_rdev_to_array(rdev, mddev); |
5983 | if (err) { | |
5984 | export_rdev(rdev); | |
5985 | return err; | |
5986 | } | |
1da177e4 LT |
5987 | } |
5988 | ||
5989 | return 0; | |
5990 | } | |
5991 | ||
f72ffdd6 | 5992 | static int hot_remove_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
5993 | { |
5994 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 5995 | struct md_rdev *rdev; |
1da177e4 | 5996 | |
1da177e4 LT |
5997 | rdev = find_rdev(mddev, dev); |
5998 | if (!rdev) | |
5999 | return -ENXIO; | |
6000 | ||
293467aa GR |
6001 | if (mddev_is_clustered(mddev)) |
6002 | md_cluster_ops->metadata_update_start(mddev); | |
6003 | ||
3ea8929d N |
6004 | clear_bit(Blocked, &rdev->flags); |
6005 | remove_and_add_spares(mddev, rdev); | |
6006 | ||
1da177e4 LT |
6007 | if (rdev->raid_disk >= 0) |
6008 | goto busy; | |
6009 | ||
88bcfef7 GR |
6010 | if (mddev_is_clustered(mddev)) |
6011 | md_cluster_ops->remove_disk(mddev, rdev); | |
6012 | ||
fb56dfef | 6013 | md_kick_rdev_from_array(rdev); |
850b2b42 | 6014 | md_update_sb(mddev, 1); |
d7603b7e | 6015 | md_new_event(mddev); |
1da177e4 | 6016 | |
293467aa GR |
6017 | if (mddev_is_clustered(mddev)) |
6018 | md_cluster_ops->metadata_update_finish(mddev); | |
6019 | ||
1da177e4 LT |
6020 | return 0; |
6021 | busy: | |
293467aa GR |
6022 | if (mddev_is_clustered(mddev)) |
6023 | md_cluster_ops->metadata_update_cancel(mddev); | |
fdefa4d8 | 6024 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
6025 | bdevname(rdev->bdev,b), mdname(mddev)); |
6026 | return -EBUSY; | |
6027 | } | |
6028 | ||
f72ffdd6 | 6029 | static int hot_add_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6030 | { |
6031 | char b[BDEVNAME_SIZE]; | |
6032 | int err; | |
3cb03002 | 6033 | struct md_rdev *rdev; |
1da177e4 LT |
6034 | |
6035 | if (!mddev->pers) | |
6036 | return -ENODEV; | |
6037 | ||
6038 | if (mddev->major_version != 0) { | |
6039 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
6040 | " version-0 superblocks.\n", | |
6041 | mdname(mddev)); | |
6042 | return -EINVAL; | |
6043 | } | |
6044 | if (!mddev->pers->hot_add_disk) { | |
f72ffdd6 | 6045 | printk(KERN_WARNING |
1da177e4 LT |
6046 | "%s: personality does not support diskops!\n", |
6047 | mdname(mddev)); | |
6048 | return -EINVAL; | |
6049 | } | |
6050 | ||
d710e138 | 6051 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6052 | if (IS_ERR(rdev)) { |
f72ffdd6 | 6053 | printk(KERN_WARNING |
1da177e4 LT |
6054 | "md: error, md_import_device() returned %ld\n", |
6055 | PTR_ERR(rdev)); | |
6056 | return -EINVAL; | |
6057 | } | |
6058 | ||
6059 | if (mddev->persistent) | |
57b2caa3 | 6060 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 6061 | else |
77304d2a | 6062 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 6063 | |
8190e754 | 6064 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6065 | |
b2d444d7 | 6066 | if (test_bit(Faulty, &rdev->flags)) { |
f72ffdd6 | 6067 | printk(KERN_WARNING |
1da177e4 LT |
6068 | "md: can not hot-add faulty %s disk to %s!\n", |
6069 | bdevname(rdev->bdev,b), mdname(mddev)); | |
6070 | err = -EINVAL; | |
6071 | goto abort_export; | |
6072 | } | |
293467aa GR |
6073 | |
6074 | if (mddev_is_clustered(mddev)) | |
6075 | md_cluster_ops->metadata_update_start(mddev); | |
b2d444d7 | 6076 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 6077 | rdev->desc_nr = -1; |
5842730d | 6078 | rdev->saved_raid_disk = -1; |
2bf071bf N |
6079 | err = bind_rdev_to_array(rdev, mddev); |
6080 | if (err) | |
293467aa | 6081 | goto abort_clustered; |
1da177e4 LT |
6082 | |
6083 | /* | |
6084 | * The rest should better be atomic, we can have disk failures | |
6085 | * noticed in interrupt contexts ... | |
6086 | */ | |
6087 | ||
1da177e4 LT |
6088 | rdev->raid_disk = -1; |
6089 | ||
850b2b42 | 6090 | md_update_sb(mddev, 1); |
1da177e4 | 6091 | |
293467aa GR |
6092 | if (mddev_is_clustered(mddev)) |
6093 | md_cluster_ops->metadata_update_finish(mddev); | |
1da177e4 LT |
6094 | /* |
6095 | * Kick recovery, maybe this spare has to be added to the | |
6096 | * array immediately. | |
6097 | */ | |
6098 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6099 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 6100 | md_new_event(mddev); |
1da177e4 LT |
6101 | return 0; |
6102 | ||
293467aa GR |
6103 | abort_clustered: |
6104 | if (mddev_is_clustered(mddev)) | |
6105 | md_cluster_ops->metadata_update_cancel(mddev); | |
1da177e4 LT |
6106 | abort_export: |
6107 | export_rdev(rdev); | |
6108 | return err; | |
6109 | } | |
6110 | ||
fd01b88c | 6111 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c | 6112 | { |
035328c2 | 6113 | int err = 0; |
32a7627c | 6114 | |
36fa3063 | 6115 | if (mddev->pers) { |
d66b1b39 | 6116 | if (!mddev->pers->quiesce || !mddev->thread) |
36fa3063 N |
6117 | return -EBUSY; |
6118 | if (mddev->recovery || mddev->sync_thread) | |
6119 | return -EBUSY; | |
6120 | /* we should be able to change the bitmap.. */ | |
6121 | } | |
32a7627c | 6122 | |
36fa3063 | 6123 | if (fd >= 0) { |
035328c2 | 6124 | struct inode *inode; |
1e594bb2 N |
6125 | struct file *f; |
6126 | ||
6127 | if (mddev->bitmap || mddev->bitmap_info.file) | |
36fa3063 | 6128 | return -EEXIST; /* cannot add when bitmap is present */ |
1e594bb2 | 6129 | f = fget(fd); |
32a7627c | 6130 | |
1e594bb2 | 6131 | if (f == NULL) { |
36fa3063 N |
6132 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
6133 | mdname(mddev)); | |
6134 | return -EBADF; | |
6135 | } | |
6136 | ||
1e594bb2 | 6137 | inode = f->f_mapping->host; |
035328c2 N |
6138 | if (!S_ISREG(inode->i_mode)) { |
6139 | printk(KERN_ERR "%s: error: bitmap file must be a regular file\n", | |
6140 | mdname(mddev)); | |
6141 | err = -EBADF; | |
1e594bb2 | 6142 | } else if (!(f->f_mode & FMODE_WRITE)) { |
035328c2 N |
6143 | printk(KERN_ERR "%s: error: bitmap file must open for write\n", |
6144 | mdname(mddev)); | |
6145 | err = -EBADF; | |
6146 | } else if (atomic_read(&inode->i_writecount) != 1) { | |
36fa3063 N |
6147 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", |
6148 | mdname(mddev)); | |
035328c2 N |
6149 | err = -EBUSY; |
6150 | } | |
6151 | if (err) { | |
1e594bb2 | 6152 | fput(f); |
36fa3063 N |
6153 | return err; |
6154 | } | |
1e594bb2 | 6155 | mddev->bitmap_info.file = f; |
c3d9714e | 6156 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
6157 | } else if (mddev->bitmap == NULL) |
6158 | return -ENOENT; /* cannot remove what isn't there */ | |
6159 | err = 0; | |
6160 | if (mddev->pers) { | |
6161 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 6162 | if (fd >= 0) { |
f9209a32 GR |
6163 | struct bitmap *bitmap; |
6164 | ||
6165 | bitmap = bitmap_create(mddev, -1); | |
6166 | if (!IS_ERR(bitmap)) { | |
6167 | mddev->bitmap = bitmap; | |
69e51b44 | 6168 | err = bitmap_load(mddev); |
ba599aca N |
6169 | } else |
6170 | err = PTR_ERR(bitmap); | |
69e51b44 | 6171 | } |
d7375ab3 | 6172 | if (fd < 0 || err) { |
36fa3063 | 6173 | bitmap_destroy(mddev); |
d7375ab3 N |
6174 | fd = -1; /* make sure to put the file */ |
6175 | } | |
36fa3063 | 6176 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
6177 | } |
6178 | if (fd < 0) { | |
4af1a041 N |
6179 | struct file *f = mddev->bitmap_info.file; |
6180 | if (f) { | |
6181 | spin_lock(&mddev->lock); | |
6182 | mddev->bitmap_info.file = NULL; | |
6183 | spin_unlock(&mddev->lock); | |
6184 | fput(f); | |
6185 | } | |
36fa3063 N |
6186 | } |
6187 | ||
32a7627c N |
6188 | return err; |
6189 | } | |
6190 | ||
1da177e4 LT |
6191 | /* |
6192 | * set_array_info is used two different ways | |
6193 | * The original usage is when creating a new array. | |
6194 | * In this usage, raid_disks is > 0 and it together with | |
6195 | * level, size, not_persistent,layout,chunksize determine the | |
6196 | * shape of the array. | |
6197 | * This will always create an array with a type-0.90.0 superblock. | |
6198 | * The newer usage is when assembling an array. | |
6199 | * In this case raid_disks will be 0, and the major_version field is | |
6200 | * use to determine which style super-blocks are to be found on the devices. | |
6201 | * The minor and patch _version numbers are also kept incase the | |
6202 | * super_block handler wishes to interpret them. | |
6203 | */ | |
f72ffdd6 | 6204 | static int set_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6205 | { |
6206 | ||
6207 | if (info->raid_disks == 0) { | |
6208 | /* just setting version number for superblock loading */ | |
6209 | if (info->major_version < 0 || | |
50511da3 | 6210 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6211 | super_types[info->major_version].name == NULL) { |
6212 | /* maybe try to auto-load a module? */ | |
f72ffdd6 | 6213 | printk(KERN_INFO |
1da177e4 LT |
6214 | "md: superblock version %d not known\n", |
6215 | info->major_version); | |
6216 | return -EINVAL; | |
6217 | } | |
6218 | mddev->major_version = info->major_version; | |
6219 | mddev->minor_version = info->minor_version; | |
6220 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6221 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6222 | /* ensure mddev_put doesn't delete this now that there |
6223 | * is some minimal configuration. | |
6224 | */ | |
6225 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
6226 | return 0; |
6227 | } | |
6228 | mddev->major_version = MD_MAJOR_VERSION; | |
6229 | mddev->minor_version = MD_MINOR_VERSION; | |
6230 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
6231 | mddev->ctime = get_seconds(); | |
6232 | ||
6233 | mddev->level = info->level; | |
17115e03 | 6234 | mddev->clevel[0] = 0; |
58c0fed4 | 6235 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6236 | mddev->raid_disks = info->raid_disks; |
6237 | /* don't set md_minor, it is determined by which /dev/md* was | |
6238 | * openned | |
6239 | */ | |
6240 | if (info->state & (1<<MD_SB_CLEAN)) | |
6241 | mddev->recovery_cp = MaxSector; | |
6242 | else | |
6243 | mddev->recovery_cp = 0; | |
6244 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6245 | mddev->external = 0; |
1da177e4 LT |
6246 | |
6247 | mddev->layout = info->layout; | |
9d8f0363 | 6248 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
6249 | |
6250 | mddev->max_disks = MD_SB_DISKS; | |
6251 | ||
e691063a N |
6252 | if (mddev->persistent) |
6253 | mddev->flags = 0; | |
850b2b42 | 6254 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6255 | |
c3d9714e | 6256 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6257 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6258 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6259 | |
f6705578 N |
6260 | mddev->reshape_position = MaxSector; |
6261 | ||
1da177e4 LT |
6262 | /* |
6263 | * Generate a 128 bit UUID | |
6264 | */ | |
6265 | get_random_bytes(mddev->uuid, 16); | |
6266 | ||
f6705578 | 6267 | mddev->new_level = mddev->level; |
664e7c41 | 6268 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6269 | mddev->new_layout = mddev->layout; |
6270 | mddev->delta_disks = 0; | |
2c810cdd | 6271 | mddev->reshape_backwards = 0; |
f6705578 | 6272 | |
1da177e4 LT |
6273 | return 0; |
6274 | } | |
6275 | ||
fd01b88c | 6276 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6277 | { |
b522adcd DW |
6278 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
6279 | ||
6280 | if (mddev->external_size) | |
6281 | return; | |
6282 | ||
1f403624 DW |
6283 | mddev->array_sectors = array_sectors; |
6284 | } | |
6285 | EXPORT_SYMBOL(md_set_array_sectors); | |
6286 | ||
fd01b88c | 6287 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6288 | { |
3cb03002 | 6289 | struct md_rdev *rdev; |
a35b0d69 | 6290 | int rv; |
d71f9f88 | 6291 | int fit = (num_sectors == 0); |
a35b0d69 N |
6292 | |
6293 | if (mddev->pers->resize == NULL) | |
6294 | return -EINVAL; | |
d71f9f88 AN |
6295 | /* The "num_sectors" is the number of sectors of each device that |
6296 | * is used. This can only make sense for arrays with redundancy. | |
6297 | * linear and raid0 always use whatever space is available. We can only | |
6298 | * consider changing this number if no resync or reconstruction is | |
6299 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6300 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6301 | * of each device. If num_sectors is zero, we find the largest size |
6302 | * that fits. | |
a35b0d69 | 6303 | */ |
f851b60d N |
6304 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
6305 | mddev->sync_thread) | |
a35b0d69 | 6306 | return -EBUSY; |
bd8839e0 N |
6307 | if (mddev->ro) |
6308 | return -EROFS; | |
a4a6125a | 6309 | |
dafb20fa | 6310 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6311 | sector_t avail = rdev->sectors; |
01ab5662 | 6312 | |
d71f9f88 AN |
6313 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6314 | num_sectors = avail; | |
6315 | if (avail < num_sectors) | |
a35b0d69 N |
6316 | return -ENOSPC; |
6317 | } | |
d71f9f88 | 6318 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
6319 | if (!rv) |
6320 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
6321 | return rv; |
6322 | } | |
6323 | ||
fd01b88c | 6324 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6325 | { |
6326 | int rv; | |
c6563a8c | 6327 | struct md_rdev *rdev; |
da943b99 | 6328 | /* change the number of raid disks */ |
63c70c4f | 6329 | if (mddev->pers->check_reshape == NULL) |
da943b99 | 6330 | return -EINVAL; |
bd8839e0 N |
6331 | if (mddev->ro) |
6332 | return -EROFS; | |
da943b99 | 6333 | if (raid_disks <= 0 || |
233fca36 | 6334 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6335 | return -EINVAL; |
f851b60d N |
6336 | if (mddev->sync_thread || |
6337 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
6338 | mddev->reshape_position != MaxSector) | |
da943b99 | 6339 | return -EBUSY; |
c6563a8c N |
6340 | |
6341 | rdev_for_each(rdev, mddev) { | |
6342 | if (mddev->raid_disks < raid_disks && | |
6343 | rdev->data_offset < rdev->new_data_offset) | |
6344 | return -EINVAL; | |
6345 | if (mddev->raid_disks > raid_disks && | |
6346 | rdev->data_offset > rdev->new_data_offset) | |
6347 | return -EINVAL; | |
6348 | } | |
6349 | ||
63c70c4f | 6350 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6351 | if (mddev->delta_disks < 0) |
6352 | mddev->reshape_backwards = 1; | |
6353 | else if (mddev->delta_disks > 0) | |
6354 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6355 | |
6356 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6357 | if (rv < 0) { |
de171cb9 | 6358 | mddev->delta_disks = 0; |
2c810cdd N |
6359 | mddev->reshape_backwards = 0; |
6360 | } | |
da943b99 N |
6361 | return rv; |
6362 | } | |
6363 | ||
1da177e4 LT |
6364 | /* |
6365 | * update_array_info is used to change the configuration of an | |
6366 | * on-line array. | |
6367 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6368 | * fields in the info are checked against the array. | |
6369 | * Any differences that cannot be handled will cause an error. | |
6370 | * Normally, only one change can be managed at a time. | |
6371 | */ | |
fd01b88c | 6372 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6373 | { |
6374 | int rv = 0; | |
6375 | int cnt = 0; | |
36fa3063 N |
6376 | int state = 0; |
6377 | ||
6378 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6379 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6380 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6381 | |
6382 | if (mddev->major_version != info->major_version || | |
6383 | mddev->minor_version != info->minor_version || | |
6384 | /* mddev->patch_version != info->patch_version || */ | |
6385 | mddev->ctime != info->ctime || | |
6386 | mddev->level != info->level || | |
6387 | /* mddev->layout != info->layout || */ | |
4e023612 | 6388 | mddev->persistent != !info->not_persistent || |
9d8f0363 | 6389 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6390 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6391 | ((state^info->state) & 0xfffffe00) | |
6392 | ) | |
1da177e4 LT |
6393 | return -EINVAL; |
6394 | /* Check there is only one change */ | |
58c0fed4 AN |
6395 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6396 | cnt++; | |
6397 | if (mddev->raid_disks != info->raid_disks) | |
6398 | cnt++; | |
6399 | if (mddev->layout != info->layout) | |
6400 | cnt++; | |
6401 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6402 | cnt++; | |
6403 | if (cnt == 0) | |
6404 | return 0; | |
6405 | if (cnt > 1) | |
6406 | return -EINVAL; | |
1da177e4 LT |
6407 | |
6408 | if (mddev->layout != info->layout) { | |
6409 | /* Change layout | |
6410 | * we don't need to do anything at the md level, the | |
6411 | * personality will take care of it all. | |
6412 | */ | |
50ac168a | 6413 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6414 | return -EINVAL; |
597a711b N |
6415 | else { |
6416 | mddev->new_layout = info->layout; | |
50ac168a | 6417 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6418 | if (rv) |
6419 | mddev->new_layout = mddev->layout; | |
6420 | return rv; | |
6421 | } | |
1da177e4 | 6422 | } |
293467aa GR |
6423 | if (mddev_is_clustered(mddev)) |
6424 | md_cluster_ops->metadata_update_start(mddev); | |
58c0fed4 | 6425 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6426 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6427 | |
da943b99 N |
6428 | if (mddev->raid_disks != info->raid_disks) |
6429 | rv = update_raid_disks(mddev, info->raid_disks); | |
6430 | ||
36fa3063 | 6431 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
293467aa GR |
6432 | if (mddev->pers->quiesce == NULL || mddev->thread == NULL) { |
6433 | rv = -EINVAL; | |
6434 | goto err; | |
6435 | } | |
6436 | if (mddev->recovery || mddev->sync_thread) { | |
6437 | rv = -EBUSY; | |
6438 | goto err; | |
6439 | } | |
36fa3063 | 6440 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { |
f9209a32 | 6441 | struct bitmap *bitmap; |
36fa3063 | 6442 | /* add the bitmap */ |
293467aa GR |
6443 | if (mddev->bitmap) { |
6444 | rv = -EEXIST; | |
6445 | goto err; | |
6446 | } | |
6447 | if (mddev->bitmap_info.default_offset == 0) { | |
6448 | rv = -EINVAL; | |
6449 | goto err; | |
6450 | } | |
c3d9714e N |
6451 | mddev->bitmap_info.offset = |
6452 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6453 | mddev->bitmap_info.space = |
6454 | mddev->bitmap_info.default_space; | |
36fa3063 | 6455 | mddev->pers->quiesce(mddev, 1); |
f9209a32 GR |
6456 | bitmap = bitmap_create(mddev, -1); |
6457 | if (!IS_ERR(bitmap)) { | |
6458 | mddev->bitmap = bitmap; | |
69e51b44 | 6459 | rv = bitmap_load(mddev); |
ba599aca N |
6460 | } else |
6461 | rv = PTR_ERR(bitmap); | |
36fa3063 N |
6462 | if (rv) |
6463 | bitmap_destroy(mddev); | |
6464 | mddev->pers->quiesce(mddev, 0); | |
6465 | } else { | |
6466 | /* remove the bitmap */ | |
293467aa GR |
6467 | if (!mddev->bitmap) { |
6468 | rv = -ENOENT; | |
6469 | goto err; | |
6470 | } | |
6471 | if (mddev->bitmap->storage.file) { | |
6472 | rv = -EINVAL; | |
6473 | goto err; | |
6474 | } | |
36fa3063 N |
6475 | mddev->pers->quiesce(mddev, 1); |
6476 | bitmap_destroy(mddev); | |
6477 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 6478 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6479 | } |
6480 | } | |
850b2b42 | 6481 | md_update_sb(mddev, 1); |
293467aa GR |
6482 | if (mddev_is_clustered(mddev)) |
6483 | md_cluster_ops->metadata_update_finish(mddev); | |
6484 | return rv; | |
6485 | err: | |
6486 | if (mddev_is_clustered(mddev)) | |
6487 | md_cluster_ops->metadata_update_cancel(mddev); | |
1da177e4 LT |
6488 | return rv; |
6489 | } | |
6490 | ||
fd01b88c | 6491 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6492 | { |
3cb03002 | 6493 | struct md_rdev *rdev; |
1ca69c4b | 6494 | int err = 0; |
1da177e4 LT |
6495 | |
6496 | if (mddev->pers == NULL) | |
6497 | return -ENODEV; | |
6498 | ||
1ca69c4b N |
6499 | rcu_read_lock(); |
6500 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 6501 | if (!rdev) |
1ca69c4b N |
6502 | err = -ENODEV; |
6503 | else { | |
6504 | md_error(mddev, rdev); | |
6505 | if (!test_bit(Faulty, &rdev->flags)) | |
6506 | err = -EBUSY; | |
6507 | } | |
6508 | rcu_read_unlock(); | |
6509 | return err; | |
1da177e4 LT |
6510 | } |
6511 | ||
2f9618ce AN |
6512 | /* |
6513 | * We have a problem here : there is no easy way to give a CHS | |
6514 | * virtual geometry. We currently pretend that we have a 2 heads | |
6515 | * 4 sectors (with a BIG number of cylinders...). This drives | |
6516 | * dosfs just mad... ;-) | |
6517 | */ | |
a885c8c4 CH |
6518 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
6519 | { | |
fd01b88c | 6520 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
6521 | |
6522 | geo->heads = 2; | |
6523 | geo->sectors = 4; | |
49ce6cea | 6524 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
6525 | return 0; |
6526 | } | |
6527 | ||
cb335f88 NS |
6528 | static inline bool md_ioctl_valid(unsigned int cmd) |
6529 | { | |
6530 | switch (cmd) { | |
6531 | case ADD_NEW_DISK: | |
6532 | case BLKROSET: | |
6533 | case GET_ARRAY_INFO: | |
6534 | case GET_BITMAP_FILE: | |
6535 | case GET_DISK_INFO: | |
6536 | case HOT_ADD_DISK: | |
6537 | case HOT_REMOVE_DISK: | |
cb335f88 NS |
6538 | case RAID_AUTORUN: |
6539 | case RAID_VERSION: | |
6540 | case RESTART_ARRAY_RW: | |
6541 | case RUN_ARRAY: | |
6542 | case SET_ARRAY_INFO: | |
6543 | case SET_BITMAP_FILE: | |
6544 | case SET_DISK_FAULTY: | |
6545 | case STOP_ARRAY: | |
6546 | case STOP_ARRAY_RO: | |
1aee41f6 | 6547 | case CLUSTERED_DISK_NACK: |
cb335f88 NS |
6548 | return true; |
6549 | default: | |
6550 | return false; | |
6551 | } | |
6552 | } | |
6553 | ||
a39907fa | 6554 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
6555 | unsigned int cmd, unsigned long arg) |
6556 | { | |
6557 | int err = 0; | |
6558 | void __user *argp = (void __user *)arg; | |
fd01b88c | 6559 | struct mddev *mddev = NULL; |
e2218350 | 6560 | int ro; |
1da177e4 | 6561 | |
cb335f88 NS |
6562 | if (!md_ioctl_valid(cmd)) |
6563 | return -ENOTTY; | |
6564 | ||
506c9e44 N |
6565 | switch (cmd) { |
6566 | case RAID_VERSION: | |
6567 | case GET_ARRAY_INFO: | |
6568 | case GET_DISK_INFO: | |
6569 | break; | |
6570 | default: | |
6571 | if (!capable(CAP_SYS_ADMIN)) | |
6572 | return -EACCES; | |
6573 | } | |
1da177e4 LT |
6574 | |
6575 | /* | |
6576 | * Commands dealing with the RAID driver but not any | |
6577 | * particular array: | |
6578 | */ | |
c02c0aeb N |
6579 | switch (cmd) { |
6580 | case RAID_VERSION: | |
6581 | err = get_version(argp); | |
3adc28d8 | 6582 | goto out; |
1da177e4 | 6583 | |
1da177e4 | 6584 | #ifndef MODULE |
c02c0aeb N |
6585 | case RAID_AUTORUN: |
6586 | err = 0; | |
6587 | autostart_arrays(arg); | |
3adc28d8 | 6588 | goto out; |
1da177e4 | 6589 | #endif |
c02c0aeb | 6590 | default:; |
1da177e4 LT |
6591 | } |
6592 | ||
6593 | /* | |
6594 | * Commands creating/starting a new array: | |
6595 | */ | |
6596 | ||
a39907fa | 6597 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
6598 | |
6599 | if (!mddev) { | |
6600 | BUG(); | |
3adc28d8 | 6601 | goto out; |
1da177e4 LT |
6602 | } |
6603 | ||
1ca69c4b N |
6604 | /* Some actions do not requires the mutex */ |
6605 | switch (cmd) { | |
6606 | case GET_ARRAY_INFO: | |
6607 | if (!mddev->raid_disks && !mddev->external) | |
6608 | err = -ENODEV; | |
6609 | else | |
6610 | err = get_array_info(mddev, argp); | |
3adc28d8 | 6611 | goto out; |
1ca69c4b N |
6612 | |
6613 | case GET_DISK_INFO: | |
6614 | if (!mddev->raid_disks && !mddev->external) | |
6615 | err = -ENODEV; | |
6616 | else | |
6617 | err = get_disk_info(mddev, argp); | |
3adc28d8 | 6618 | goto out; |
1ca69c4b N |
6619 | |
6620 | case SET_DISK_FAULTY: | |
6621 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6622 | goto out; |
4af1a041 N |
6623 | |
6624 | case GET_BITMAP_FILE: | |
6625 | err = get_bitmap_file(mddev, argp); | |
6626 | goto out; | |
6627 | ||
1ca69c4b N |
6628 | } |
6629 | ||
a7a3f08d N |
6630 | if (cmd == ADD_NEW_DISK) |
6631 | /* need to ensure md_delayed_delete() has completed */ | |
6632 | flush_workqueue(md_misc_wq); | |
6633 | ||
90f5f7ad HR |
6634 | if (cmd == HOT_REMOVE_DISK) |
6635 | /* need to ensure recovery thread has run */ | |
6636 | wait_event_interruptible_timeout(mddev->sb_wait, | |
6637 | !test_bit(MD_RECOVERY_NEEDED, | |
6638 | &mddev->flags), | |
6639 | msecs_to_jiffies(5000)); | |
260fa034 N |
6640 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
6641 | /* Need to flush page cache, and ensure no-one else opens | |
6642 | * and writes | |
6643 | */ | |
6644 | mutex_lock(&mddev->open_mutex); | |
9ba3b7f5 | 6645 | if (mddev->pers && atomic_read(&mddev->openers) > 1) { |
260fa034 N |
6646 | mutex_unlock(&mddev->open_mutex); |
6647 | err = -EBUSY; | |
3adc28d8 | 6648 | goto out; |
260fa034 N |
6649 | } |
6650 | set_bit(MD_STILL_CLOSED, &mddev->flags); | |
6651 | mutex_unlock(&mddev->open_mutex); | |
6652 | sync_blockdev(bdev); | |
6653 | } | |
1da177e4 LT |
6654 | err = mddev_lock(mddev); |
6655 | if (err) { | |
f72ffdd6 | 6656 | printk(KERN_INFO |
1da177e4 LT |
6657 | "md: ioctl lock interrupted, reason %d, cmd %d\n", |
6658 | err, cmd); | |
3adc28d8 | 6659 | goto out; |
1da177e4 LT |
6660 | } |
6661 | ||
c02c0aeb N |
6662 | if (cmd == SET_ARRAY_INFO) { |
6663 | mdu_array_info_t info; | |
6664 | if (!arg) | |
6665 | memset(&info, 0, sizeof(info)); | |
6666 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
6667 | err = -EFAULT; | |
3adc28d8 | 6668 | goto unlock; |
c02c0aeb N |
6669 | } |
6670 | if (mddev->pers) { | |
6671 | err = update_array_info(mddev, &info); | |
6672 | if (err) { | |
6673 | printk(KERN_WARNING "md: couldn't update" | |
6674 | " array info. %d\n", err); | |
3adc28d8 | 6675 | goto unlock; |
1da177e4 | 6676 | } |
3adc28d8 | 6677 | goto unlock; |
c02c0aeb N |
6678 | } |
6679 | if (!list_empty(&mddev->disks)) { | |
6680 | printk(KERN_WARNING | |
6681 | "md: array %s already has disks!\n", | |
6682 | mdname(mddev)); | |
6683 | err = -EBUSY; | |
3adc28d8 | 6684 | goto unlock; |
c02c0aeb N |
6685 | } |
6686 | if (mddev->raid_disks) { | |
6687 | printk(KERN_WARNING | |
6688 | "md: array %s already initialised!\n", | |
6689 | mdname(mddev)); | |
6690 | err = -EBUSY; | |
3adc28d8 | 6691 | goto unlock; |
c02c0aeb N |
6692 | } |
6693 | err = set_array_info(mddev, &info); | |
6694 | if (err) { | |
6695 | printk(KERN_WARNING "md: couldn't set" | |
6696 | " array info. %d\n", err); | |
3adc28d8 | 6697 | goto unlock; |
c02c0aeb | 6698 | } |
3adc28d8 | 6699 | goto unlock; |
1da177e4 LT |
6700 | } |
6701 | ||
6702 | /* | |
6703 | * Commands querying/configuring an existing array: | |
6704 | */ | |
32a7627c | 6705 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 6706 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
6707 | if ((!mddev->raid_disks && !mddev->external) |
6708 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
6709 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
6710 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 | 6711 | err = -ENODEV; |
3adc28d8 | 6712 | goto unlock; |
1da177e4 LT |
6713 | } |
6714 | ||
6715 | /* | |
6716 | * Commands even a read-only array can execute: | |
6717 | */ | |
c02c0aeb | 6718 | switch (cmd) { |
c02c0aeb N |
6719 | case RESTART_ARRAY_RW: |
6720 | err = restart_array(mddev); | |
3adc28d8 | 6721 | goto unlock; |
1da177e4 | 6722 | |
c02c0aeb N |
6723 | case STOP_ARRAY: |
6724 | err = do_md_stop(mddev, 0, bdev); | |
3adc28d8 | 6725 | goto unlock; |
1da177e4 | 6726 | |
c02c0aeb N |
6727 | case STOP_ARRAY_RO: |
6728 | err = md_set_readonly(mddev, bdev); | |
3adc28d8 | 6729 | goto unlock; |
1da177e4 | 6730 | |
3ea8929d N |
6731 | case HOT_REMOVE_DISK: |
6732 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6733 | goto unlock; |
3ea8929d | 6734 | |
7ceb17e8 N |
6735 | case ADD_NEW_DISK: |
6736 | /* We can support ADD_NEW_DISK on read-only arrays | |
6737 | * on if we are re-adding a preexisting device. | |
6738 | * So require mddev->pers and MD_DISK_SYNC. | |
6739 | */ | |
6740 | if (mddev->pers) { | |
6741 | mdu_disk_info_t info; | |
6742 | if (copy_from_user(&info, argp, sizeof(info))) | |
6743 | err = -EFAULT; | |
6744 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
6745 | /* Need to clear read-only for this */ | |
6746 | break; | |
6747 | else | |
6748 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 6749 | goto unlock; |
7ceb17e8 N |
6750 | } |
6751 | break; | |
6752 | ||
c02c0aeb N |
6753 | case BLKROSET: |
6754 | if (get_user(ro, (int __user *)(arg))) { | |
6755 | err = -EFAULT; | |
3adc28d8 | 6756 | goto unlock; |
c02c0aeb N |
6757 | } |
6758 | err = -EINVAL; | |
e2218350 | 6759 | |
c02c0aeb N |
6760 | /* if the bdev is going readonly the value of mddev->ro |
6761 | * does not matter, no writes are coming | |
6762 | */ | |
6763 | if (ro) | |
3adc28d8 | 6764 | goto unlock; |
e2218350 | 6765 | |
c02c0aeb N |
6766 | /* are we are already prepared for writes? */ |
6767 | if (mddev->ro != 1) | |
3adc28d8 | 6768 | goto unlock; |
e2218350 | 6769 | |
c02c0aeb N |
6770 | /* transitioning to readauto need only happen for |
6771 | * arrays that call md_write_start | |
6772 | */ | |
6773 | if (mddev->pers) { | |
6774 | err = restart_array(mddev); | |
6775 | if (err == 0) { | |
6776 | mddev->ro = 2; | |
6777 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 6778 | } |
c02c0aeb | 6779 | } |
3adc28d8 | 6780 | goto unlock; |
1da177e4 LT |
6781 | } |
6782 | ||
6783 | /* | |
6784 | * The remaining ioctls are changing the state of the | |
f91de92e | 6785 | * superblock, so we do not allow them on read-only arrays. |
1da177e4 | 6786 | */ |
326eb17d | 6787 | if (mddev->ro && mddev->pers) { |
f91de92e N |
6788 | if (mddev->ro == 2) { |
6789 | mddev->ro = 0; | |
00bcb4ac | 6790 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 6791 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
6792 | /* mddev_unlock will wake thread */ |
6793 | /* If a device failed while we were read-only, we | |
6794 | * need to make sure the metadata is updated now. | |
6795 | */ | |
6796 | if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
6797 | mddev_unlock(mddev); | |
6798 | wait_event(mddev->sb_wait, | |
6799 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) && | |
6800 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
29f097c4 | 6801 | mddev_lock_nointr(mddev); |
f3378b48 | 6802 | } |
f91de92e N |
6803 | } else { |
6804 | err = -EROFS; | |
3adc28d8 | 6805 | goto unlock; |
f91de92e | 6806 | } |
1da177e4 LT |
6807 | } |
6808 | ||
c02c0aeb N |
6809 | switch (cmd) { |
6810 | case ADD_NEW_DISK: | |
1da177e4 | 6811 | { |
c02c0aeb N |
6812 | mdu_disk_info_t info; |
6813 | if (copy_from_user(&info, argp, sizeof(info))) | |
6814 | err = -EFAULT; | |
6815 | else | |
6816 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 6817 | goto unlock; |
c02c0aeb | 6818 | } |
1da177e4 | 6819 | |
1aee41f6 GR |
6820 | case CLUSTERED_DISK_NACK: |
6821 | if (mddev_is_clustered(mddev)) | |
6822 | md_cluster_ops->new_disk_ack(mddev, false); | |
6823 | else | |
6824 | err = -EINVAL; | |
6825 | goto unlock; | |
6826 | ||
c02c0aeb N |
6827 | case HOT_ADD_DISK: |
6828 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6829 | goto unlock; |
1da177e4 | 6830 | |
c02c0aeb N |
6831 | case RUN_ARRAY: |
6832 | err = do_md_run(mddev); | |
3adc28d8 | 6833 | goto unlock; |
1da177e4 | 6834 | |
c02c0aeb N |
6835 | case SET_BITMAP_FILE: |
6836 | err = set_bitmap_file(mddev, (int)arg); | |
3adc28d8 | 6837 | goto unlock; |
32a7627c | 6838 | |
c02c0aeb N |
6839 | default: |
6840 | err = -EINVAL; | |
3adc28d8 | 6841 | goto unlock; |
1da177e4 LT |
6842 | } |
6843 | ||
3adc28d8 | 6844 | unlock: |
d3374825 N |
6845 | if (mddev->hold_active == UNTIL_IOCTL && |
6846 | err != -EINVAL) | |
6847 | mddev->hold_active = 0; | |
1da177e4 | 6848 | mddev_unlock(mddev); |
3adc28d8 | 6849 | out: |
1da177e4 LT |
6850 | return err; |
6851 | } | |
aa98aa31 AB |
6852 | #ifdef CONFIG_COMPAT |
6853 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
6854 | unsigned int cmd, unsigned long arg) | |
6855 | { | |
6856 | switch (cmd) { | |
6857 | case HOT_REMOVE_DISK: | |
6858 | case HOT_ADD_DISK: | |
6859 | case SET_DISK_FAULTY: | |
6860 | case SET_BITMAP_FILE: | |
6861 | /* These take in integer arg, do not convert */ | |
6862 | break; | |
6863 | default: | |
6864 | arg = (unsigned long)compat_ptr(arg); | |
6865 | break; | |
6866 | } | |
6867 | ||
6868 | return md_ioctl(bdev, mode, cmd, arg); | |
6869 | } | |
6870 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 6871 | |
a39907fa | 6872 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
6873 | { |
6874 | /* | |
6875 | * Succeed if we can lock the mddev, which confirms that | |
6876 | * it isn't being stopped right now. | |
6877 | */ | |
fd01b88c | 6878 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
6879 | int err; |
6880 | ||
0c098220 YL |
6881 | if (!mddev) |
6882 | return -ENODEV; | |
6883 | ||
d3374825 N |
6884 | if (mddev->gendisk != bdev->bd_disk) { |
6885 | /* we are racing with mddev_put which is discarding this | |
6886 | * bd_disk. | |
6887 | */ | |
6888 | mddev_put(mddev); | |
6889 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 6890 | flush_workqueue(md_misc_wq); |
d3374825 N |
6891 | /* Then retry the open from the top */ |
6892 | return -ERESTARTSYS; | |
6893 | } | |
6894 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
6895 | ||
c8c00a69 | 6896 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6897 | goto out; |
6898 | ||
6899 | err = 0; | |
f2ea68cf | 6900 | atomic_inc(&mddev->openers); |
260fa034 | 6901 | clear_bit(MD_STILL_CLOSED, &mddev->flags); |
c8c00a69 | 6902 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6903 | |
f0b4f7e2 | 6904 | check_disk_change(bdev); |
1da177e4 LT |
6905 | out: |
6906 | return err; | |
6907 | } | |
6908 | ||
db2a144b | 6909 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6910 | { |
f72ffdd6 | 6911 | struct mddev *mddev = disk->private_data; |
1da177e4 | 6912 | |
52e5f9d1 | 6913 | BUG_ON(!mddev); |
f2ea68cf | 6914 | atomic_dec(&mddev->openers); |
1da177e4 | 6915 | mddev_put(mddev); |
1da177e4 | 6916 | } |
f0b4f7e2 N |
6917 | |
6918 | static int md_media_changed(struct gendisk *disk) | |
6919 | { | |
fd01b88c | 6920 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6921 | |
6922 | return mddev->changed; | |
6923 | } | |
6924 | ||
6925 | static int md_revalidate(struct gendisk *disk) | |
6926 | { | |
fd01b88c | 6927 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6928 | |
6929 | mddev->changed = 0; | |
6930 | return 0; | |
6931 | } | |
83d5cde4 | 6932 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6933 | { |
6934 | .owner = THIS_MODULE, | |
a39907fa AV |
6935 | .open = md_open, |
6936 | .release = md_release, | |
b492b852 | 6937 | .ioctl = md_ioctl, |
aa98aa31 AB |
6938 | #ifdef CONFIG_COMPAT |
6939 | .compat_ioctl = md_compat_ioctl, | |
6940 | #endif | |
a885c8c4 | 6941 | .getgeo = md_getgeo, |
f0b4f7e2 N |
6942 | .media_changed = md_media_changed, |
6943 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
6944 | }; |
6945 | ||
f72ffdd6 | 6946 | static int md_thread(void *arg) |
1da177e4 | 6947 | { |
2b8bf345 | 6948 | struct md_thread *thread = arg; |
1da177e4 | 6949 | |
1da177e4 LT |
6950 | /* |
6951 | * md_thread is a 'system-thread', it's priority should be very | |
6952 | * high. We avoid resource deadlocks individually in each | |
6953 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6954 | * the very same RT priority as kswapd, thus we will never get | |
6955 | * into a priority inversion deadlock. | |
6956 | * | |
6957 | * we definitely have to have equal or higher priority than | |
6958 | * bdflush, otherwise bdflush will deadlock if there are too | |
6959 | * many dirty RAID5 blocks. | |
6960 | */ | |
1da177e4 | 6961 | |
6985c43f | 6962 | allow_signal(SIGKILL); |
a6fb0934 | 6963 | while (!kthread_should_stop()) { |
1da177e4 | 6964 | |
93588e22 N |
6965 | /* We need to wait INTERRUPTIBLE so that |
6966 | * we don't add to the load-average. | |
6967 | * That means we need to be sure no signals are | |
6968 | * pending | |
6969 | */ | |
6970 | if (signal_pending(current)) | |
6971 | flush_signals(current); | |
6972 | ||
6973 | wait_event_interruptible_timeout | |
6974 | (thread->wqueue, | |
6975 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6976 | || kthread_should_stop(), | |
6977 | thread->timeout); | |
1da177e4 | 6978 | |
6c987910 N |
6979 | clear_bit(THREAD_WAKEUP, &thread->flags); |
6980 | if (!kthread_should_stop()) | |
4ed8731d | 6981 | thread->run(thread); |
1da177e4 | 6982 | } |
a6fb0934 | 6983 | |
1da177e4 LT |
6984 | return 0; |
6985 | } | |
6986 | ||
2b8bf345 | 6987 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
6988 | { |
6989 | if (thread) { | |
36a4e1fe | 6990 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
1da177e4 LT |
6991 | set_bit(THREAD_WAKEUP, &thread->flags); |
6992 | wake_up(&thread->wqueue); | |
6993 | } | |
6994 | } | |
6c144d31 | 6995 | EXPORT_SYMBOL(md_wakeup_thread); |
1da177e4 | 6996 | |
4ed8731d SL |
6997 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
6998 | struct mddev *mddev, const char *name) | |
1da177e4 | 6999 | { |
2b8bf345 | 7000 | struct md_thread *thread; |
1da177e4 | 7001 | |
2b8bf345 | 7002 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
7003 | if (!thread) |
7004 | return NULL; | |
7005 | ||
1da177e4 LT |
7006 | init_waitqueue_head(&thread->wqueue); |
7007 | ||
1da177e4 LT |
7008 | thread->run = run; |
7009 | thread->mddev = mddev; | |
32a7627c | 7010 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
7011 | thread->tsk = kthread_run(md_thread, thread, |
7012 | "%s_%s", | |
7013 | mdname(thread->mddev), | |
0232605d | 7014 | name); |
a6fb0934 | 7015 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
7016 | kfree(thread); |
7017 | return NULL; | |
7018 | } | |
1da177e4 LT |
7019 | return thread; |
7020 | } | |
6c144d31 | 7021 | EXPORT_SYMBOL(md_register_thread); |
1da177e4 | 7022 | |
2b8bf345 | 7023 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 7024 | { |
2b8bf345 | 7025 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
7026 | if (!thread) |
7027 | return; | |
36a4e1fe | 7028 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
7029 | /* Locking ensures that mddev_unlock does not wake_up a |
7030 | * non-existent thread | |
7031 | */ | |
7032 | spin_lock(&pers_lock); | |
7033 | *threadp = NULL; | |
7034 | spin_unlock(&pers_lock); | |
a6fb0934 N |
7035 | |
7036 | kthread_stop(thread->tsk); | |
1da177e4 LT |
7037 | kfree(thread); |
7038 | } | |
6c144d31 | 7039 | EXPORT_SYMBOL(md_unregister_thread); |
1da177e4 | 7040 | |
fd01b88c | 7041 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7042 | { |
b2d444d7 | 7043 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 7044 | return; |
6bfe0b49 | 7045 | |
de393cde | 7046 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
7047 | return; |
7048 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
7049 | if (mddev->degraded) |
7050 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 7051 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
7052 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7053 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7054 | md_wakeup_thread(mddev->thread); | |
768a418d | 7055 | if (mddev->event_work.func) |
e804ac78 | 7056 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 7057 | md_new_event_inintr(mddev); |
1da177e4 | 7058 | } |
6c144d31 | 7059 | EXPORT_SYMBOL(md_error); |
1da177e4 LT |
7060 | |
7061 | /* seq_file implementation /proc/mdstat */ | |
7062 | ||
7063 | static void status_unused(struct seq_file *seq) | |
7064 | { | |
7065 | int i = 0; | |
3cb03002 | 7066 | struct md_rdev *rdev; |
1da177e4 LT |
7067 | |
7068 | seq_printf(seq, "unused devices: "); | |
7069 | ||
159ec1fc | 7070 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
7071 | char b[BDEVNAME_SIZE]; |
7072 | i++; | |
7073 | seq_printf(seq, "%s ", | |
7074 | bdevname(rdev->bdev,b)); | |
7075 | } | |
7076 | if (!i) | |
7077 | seq_printf(seq, "<none>"); | |
7078 | ||
7079 | seq_printf(seq, "\n"); | |
7080 | } | |
7081 | ||
f7851be7 | 7082 | static int status_resync(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7083 | { |
dd71cf6b N |
7084 | sector_t max_sectors, resync, res; |
7085 | unsigned long dt, db; | |
7086 | sector_t rt; | |
4588b42e N |
7087 | int scale; |
7088 | unsigned int per_milli; | |
1da177e4 | 7089 | |
c804cdec N |
7090 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
7091 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 7092 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 7093 | else |
dd71cf6b | 7094 | max_sectors = mddev->dev_sectors; |
1da177e4 | 7095 | |
f7851be7 N |
7096 | resync = mddev->curr_resync; |
7097 | if (resync <= 3) { | |
7098 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7099 | /* Still cleaning up */ | |
7100 | resync = max_sectors; | |
7101 | } else | |
7102 | resync -= atomic_read(&mddev->recovery_active); | |
7103 | ||
7104 | if (resync == 0) { | |
7105 | if (mddev->recovery_cp < MaxSector) { | |
7106 | seq_printf(seq, "\tresync=PENDING"); | |
7107 | return 1; | |
7108 | } | |
7109 | return 0; | |
7110 | } | |
7111 | if (resync < 3) { | |
7112 | seq_printf(seq, "\tresync=DELAYED"); | |
7113 | return 1; | |
7114 | } | |
7115 | ||
403df478 | 7116 | WARN_ON(max_sectors == 0); |
4588b42e | 7117 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 7118 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
7119 | * u32, as those are the requirements for sector_div. |
7120 | * Thus 'scale' must be at least 10 | |
7121 | */ | |
7122 | scale = 10; | |
7123 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 7124 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
7125 | scale++; |
7126 | } | |
7127 | res = (resync>>scale)*1000; | |
dd71cf6b | 7128 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
7129 | |
7130 | per_milli = res; | |
1da177e4 | 7131 | { |
4588b42e | 7132 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
7133 | seq_printf(seq, "["); |
7134 | for (i = 0; i < x; i++) | |
7135 | seq_printf(seq, "="); | |
7136 | seq_printf(seq, ">"); | |
7137 | for (i = 0; i < y; i++) | |
7138 | seq_printf(seq, "."); | |
7139 | seq_printf(seq, "] "); | |
7140 | } | |
4588b42e | 7141 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
7142 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
7143 | "reshape" : | |
61df9d91 N |
7144 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
7145 | "check" : | |
7146 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
7147 | "resync" : "recovery"))), | |
7148 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
7149 | (unsigned long long) resync/2, |
7150 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
7151 | |
7152 | /* | |
1da177e4 LT |
7153 | * dt: time from mark until now |
7154 | * db: blocks written from mark until now | |
7155 | * rt: remaining time | |
dd71cf6b N |
7156 | * |
7157 | * rt is a sector_t, so could be 32bit or 64bit. | |
7158 | * So we divide before multiply in case it is 32bit and close | |
7159 | * to the limit. | |
25985edc | 7160 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
7161 | * near the end of resync when the number of remaining sectors |
7162 | * is close to 'db'. | |
7163 | * We then divide rt by 32 after multiplying by db to compensate. | |
7164 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
7165 | */ |
7166 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
7167 | if (!dt) dt++; | |
ff4e8d9a N |
7168 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
7169 | - mddev->resync_mark_cnt; | |
1da177e4 | 7170 | |
dd71cf6b N |
7171 | rt = max_sectors - resync; /* number of remaining sectors */ |
7172 | sector_div(rt, db/32+1); | |
7173 | rt *= dt; | |
7174 | rt >>= 5; | |
7175 | ||
7176 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
7177 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 7178 | |
ff4e8d9a | 7179 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
f7851be7 | 7180 | return 1; |
1da177e4 LT |
7181 | } |
7182 | ||
7183 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
7184 | { | |
7185 | struct list_head *tmp; | |
7186 | loff_t l = *pos; | |
fd01b88c | 7187 | struct mddev *mddev; |
1da177e4 LT |
7188 | |
7189 | if (l >= 0x10000) | |
7190 | return NULL; | |
7191 | if (!l--) | |
7192 | /* header */ | |
7193 | return (void*)1; | |
7194 | ||
7195 | spin_lock(&all_mddevs_lock); | |
7196 | list_for_each(tmp,&all_mddevs) | |
7197 | if (!l--) { | |
fd01b88c | 7198 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
7199 | mddev_get(mddev); |
7200 | spin_unlock(&all_mddevs_lock); | |
7201 | return mddev; | |
7202 | } | |
7203 | spin_unlock(&all_mddevs_lock); | |
7204 | if (!l--) | |
7205 | return (void*)2;/* tail */ | |
7206 | return NULL; | |
7207 | } | |
7208 | ||
7209 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
7210 | { | |
7211 | struct list_head *tmp; | |
fd01b88c | 7212 | struct mddev *next_mddev, *mddev = v; |
f72ffdd6 | 7213 | |
1da177e4 LT |
7214 | ++*pos; |
7215 | if (v == (void*)2) | |
7216 | return NULL; | |
7217 | ||
7218 | spin_lock(&all_mddevs_lock); | |
7219 | if (v == (void*)1) | |
7220 | tmp = all_mddevs.next; | |
7221 | else | |
7222 | tmp = mddev->all_mddevs.next; | |
7223 | if (tmp != &all_mddevs) | |
fd01b88c | 7224 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
7225 | else { |
7226 | next_mddev = (void*)2; | |
7227 | *pos = 0x10000; | |
f72ffdd6 | 7228 | } |
1da177e4 LT |
7229 | spin_unlock(&all_mddevs_lock); |
7230 | ||
7231 | if (v != (void*)1) | |
7232 | mddev_put(mddev); | |
7233 | return next_mddev; | |
7234 | ||
7235 | } | |
7236 | ||
7237 | static void md_seq_stop(struct seq_file *seq, void *v) | |
7238 | { | |
fd01b88c | 7239 | struct mddev *mddev = v; |
1da177e4 LT |
7240 | |
7241 | if (mddev && v != (void*)1 && v != (void*)2) | |
7242 | mddev_put(mddev); | |
7243 | } | |
7244 | ||
7245 | static int md_seq_show(struct seq_file *seq, void *v) | |
7246 | { | |
fd01b88c | 7247 | struct mddev *mddev = v; |
dd8ac336 | 7248 | sector_t sectors; |
3cb03002 | 7249 | struct md_rdev *rdev; |
1da177e4 LT |
7250 | |
7251 | if (v == (void*)1) { | |
84fc4b56 | 7252 | struct md_personality *pers; |
1da177e4 LT |
7253 | seq_printf(seq, "Personalities : "); |
7254 | spin_lock(&pers_lock); | |
2604b703 N |
7255 | list_for_each_entry(pers, &pers_list, list) |
7256 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
7257 | |
7258 | spin_unlock(&pers_lock); | |
7259 | seq_printf(seq, "\n"); | |
f1514638 | 7260 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
7261 | return 0; |
7262 | } | |
7263 | if (v == (void*)2) { | |
7264 | status_unused(seq); | |
7265 | return 0; | |
7266 | } | |
7267 | ||
36d091f4 | 7268 | spin_lock(&mddev->lock); |
1da177e4 LT |
7269 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7270 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7271 | mddev->pers ? "" : "in"); | |
7272 | if (mddev->pers) { | |
f91de92e | 7273 | if (mddev->ro==1) |
1da177e4 | 7274 | seq_printf(seq, " (read-only)"); |
f91de92e | 7275 | if (mddev->ro==2) |
52720ae7 | 7276 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7277 | seq_printf(seq, " %s", mddev->pers->name); |
7278 | } | |
7279 | ||
dd8ac336 | 7280 | sectors = 0; |
f97fcad3 N |
7281 | rcu_read_lock(); |
7282 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 LT |
7283 | char b[BDEVNAME_SIZE]; |
7284 | seq_printf(seq, " %s[%d]", | |
7285 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7286 | if (test_bit(WriteMostly, &rdev->flags)) |
7287 | seq_printf(seq, "(W)"); | |
b2d444d7 | 7288 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7289 | seq_printf(seq, "(F)"); |
7290 | continue; | |
2d78f8c4 N |
7291 | } |
7292 | if (rdev->raid_disk < 0) | |
b325a32e | 7293 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7294 | if (test_bit(Replacement, &rdev->flags)) |
7295 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7296 | sectors += rdev->sectors; |
1da177e4 | 7297 | } |
f97fcad3 | 7298 | rcu_read_unlock(); |
1da177e4 LT |
7299 | |
7300 | if (!list_empty(&mddev->disks)) { | |
7301 | if (mddev->pers) | |
7302 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7303 | (unsigned long long) |
7304 | mddev->array_sectors / 2); | |
1da177e4 LT |
7305 | else |
7306 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7307 | (unsigned long long)sectors / 2); |
1da177e4 | 7308 | } |
1cd6bf19 N |
7309 | if (mddev->persistent) { |
7310 | if (mddev->major_version != 0 || | |
7311 | mddev->minor_version != 90) { | |
7312 | seq_printf(seq," super %d.%d", | |
7313 | mddev->major_version, | |
7314 | mddev->minor_version); | |
7315 | } | |
e691063a N |
7316 | } else if (mddev->external) |
7317 | seq_printf(seq, " super external:%s", | |
7318 | mddev->metadata_type); | |
7319 | else | |
1cd6bf19 | 7320 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7321 | |
7322 | if (mddev->pers) { | |
d710e138 | 7323 | mddev->pers->status(seq, mddev); |
f72ffdd6 | 7324 | seq_printf(seq, "\n "); |
8e1b39d6 | 7325 | if (mddev->pers->sync_request) { |
f7851be7 | 7326 | if (status_resync(seq, mddev)) |
8e1b39d6 | 7327 | seq_printf(seq, "\n "); |
8e1b39d6 | 7328 | } |
32a7627c N |
7329 | } else |
7330 | seq_printf(seq, "\n "); | |
7331 | ||
57148964 | 7332 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7333 | |
7334 | seq_printf(seq, "\n"); | |
7335 | } | |
36d091f4 | 7336 | spin_unlock(&mddev->lock); |
f72ffdd6 | 7337 | |
1da177e4 LT |
7338 | return 0; |
7339 | } | |
7340 | ||
110518bc | 7341 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7342 | .start = md_seq_start, |
7343 | .next = md_seq_next, | |
7344 | .stop = md_seq_stop, | |
7345 | .show = md_seq_show, | |
7346 | }; | |
7347 | ||
7348 | static int md_seq_open(struct inode *inode, struct file *file) | |
7349 | { | |
f1514638 | 7350 | struct seq_file *seq; |
1da177e4 LT |
7351 | int error; |
7352 | ||
7353 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7354 | if (error) |
f1514638 KS |
7355 | return error; |
7356 | ||
7357 | seq = file->private_data; | |
7358 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7359 | return error; |
7360 | } | |
7361 | ||
e2f23b60 | 7362 | static int md_unloading; |
d7603b7e N |
7363 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
7364 | { | |
f1514638 | 7365 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
7366 | int mask; |
7367 | ||
e2f23b60 | 7368 | if (md_unloading) |
7d7e64f2 | 7369 | return POLLIN|POLLRDNORM|POLLERR|POLLPRI; |
d7603b7e N |
7370 | poll_wait(filp, &md_event_waiters, wait); |
7371 | ||
7372 | /* always allow read */ | |
7373 | mask = POLLIN | POLLRDNORM; | |
7374 | ||
f1514638 | 7375 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
7376 | mask |= POLLERR | POLLPRI; |
7377 | return mask; | |
7378 | } | |
7379 | ||
fa027c2a | 7380 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7381 | .owner = THIS_MODULE, |
1da177e4 LT |
7382 | .open = md_seq_open, |
7383 | .read = seq_read, | |
7384 | .llseek = seq_lseek, | |
c3f94b40 | 7385 | .release = seq_release_private, |
d7603b7e | 7386 | .poll = mdstat_poll, |
1da177e4 LT |
7387 | }; |
7388 | ||
84fc4b56 | 7389 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7390 | { |
50bd3774 CY |
7391 | printk(KERN_INFO "md: %s personality registered for level %d\n", |
7392 | p->name, p->level); | |
1da177e4 | 7393 | spin_lock(&pers_lock); |
2604b703 | 7394 | list_add_tail(&p->list, &pers_list); |
1da177e4 LT |
7395 | spin_unlock(&pers_lock); |
7396 | return 0; | |
7397 | } | |
6c144d31 | 7398 | EXPORT_SYMBOL(register_md_personality); |
1da177e4 | 7399 | |
84fc4b56 | 7400 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7401 | { |
2604b703 | 7402 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 7403 | spin_lock(&pers_lock); |
2604b703 | 7404 | list_del_init(&p->list); |
1da177e4 LT |
7405 | spin_unlock(&pers_lock); |
7406 | return 0; | |
7407 | } | |
6c144d31 | 7408 | EXPORT_SYMBOL(unregister_md_personality); |
1da177e4 | 7409 | |
6022e75b N |
7410 | int register_md_cluster_operations(struct md_cluster_operations *ops, |
7411 | struct module *module) | |
edb39c9d | 7412 | { |
6022e75b | 7413 | int ret = 0; |
edb39c9d | 7414 | spin_lock(&pers_lock); |
6022e75b N |
7415 | if (md_cluster_ops != NULL) |
7416 | ret = -EALREADY; | |
7417 | else { | |
7418 | md_cluster_ops = ops; | |
7419 | md_cluster_mod = module; | |
7420 | } | |
edb39c9d | 7421 | spin_unlock(&pers_lock); |
6022e75b | 7422 | return ret; |
edb39c9d GR |
7423 | } |
7424 | EXPORT_SYMBOL(register_md_cluster_operations); | |
7425 | ||
7426 | int unregister_md_cluster_operations(void) | |
7427 | { | |
7428 | spin_lock(&pers_lock); | |
7429 | md_cluster_ops = NULL; | |
7430 | spin_unlock(&pers_lock); | |
7431 | return 0; | |
7432 | } | |
7433 | EXPORT_SYMBOL(unregister_md_cluster_operations); | |
7434 | ||
7435 | int md_setup_cluster(struct mddev *mddev, int nodes) | |
7436 | { | |
7437 | int err; | |
7438 | ||
7439 | err = request_module("md-cluster"); | |
7440 | if (err) { | |
7441 | pr_err("md-cluster module not found.\n"); | |
b0c26a79 | 7442 | return -ENOENT; |
edb39c9d GR |
7443 | } |
7444 | ||
7445 | spin_lock(&pers_lock); | |
7446 | if (!md_cluster_ops || !try_module_get(md_cluster_mod)) { | |
7447 | spin_unlock(&pers_lock); | |
7448 | return -ENOENT; | |
7449 | } | |
7450 | spin_unlock(&pers_lock); | |
7451 | ||
cf921cc1 | 7452 | return md_cluster_ops->join(mddev, nodes); |
edb39c9d GR |
7453 | } |
7454 | ||
7455 | void md_cluster_stop(struct mddev *mddev) | |
7456 | { | |
c4ce867f GR |
7457 | if (!md_cluster_ops) |
7458 | return; | |
edb39c9d GR |
7459 | md_cluster_ops->leave(mddev); |
7460 | module_put(md_cluster_mod); | |
7461 | } | |
7462 | ||
fd01b88c | 7463 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7464 | { |
f72ffdd6 | 7465 | struct md_rdev *rdev; |
1da177e4 | 7466 | int idle; |
eea1bf38 | 7467 | int curr_events; |
1da177e4 LT |
7468 | |
7469 | idle = 1; | |
4b80991c N |
7470 | rcu_read_lock(); |
7471 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7472 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7473 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7474 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7475 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7476 | /* sync IO will cause sync_io to increase before the disk_stats |
7477 | * as sync_io is counted when a request starts, and | |
7478 | * disk_stats is counted when it completes. | |
7479 | * So resync activity will cause curr_events to be smaller than | |
7480 | * when there was no such activity. | |
7481 | * non-sync IO will cause disk_stat to increase without | |
7482 | * increasing sync_io so curr_events will (eventually) | |
7483 | * be larger than it was before. Once it becomes | |
7484 | * substantially larger, the test below will cause | |
7485 | * the array to appear non-idle, and resync will slow | |
7486 | * down. | |
7487 | * If there is a lot of outstanding resync activity when | |
7488 | * we set last_event to curr_events, then all that activity | |
7489 | * completing might cause the array to appear non-idle | |
7490 | * and resync will be slowed down even though there might | |
7491 | * not have been non-resync activity. This will only | |
7492 | * happen once though. 'last_events' will soon reflect | |
7493 | * the state where there is little or no outstanding | |
7494 | * resync requests, and further resync activity will | |
7495 | * always make curr_events less than last_events. | |
c0e48521 | 7496 | * |
1da177e4 | 7497 | */ |
eea1bf38 | 7498 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
7499 | rdev->last_events = curr_events; |
7500 | idle = 0; | |
7501 | } | |
7502 | } | |
4b80991c | 7503 | rcu_read_unlock(); |
1da177e4 LT |
7504 | return idle; |
7505 | } | |
7506 | ||
fd01b88c | 7507 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
7508 | { |
7509 | /* another "blocks" (512byte) blocks have been synced */ | |
7510 | atomic_sub(blocks, &mddev->recovery_active); | |
7511 | wake_up(&mddev->recovery_wait); | |
7512 | if (!ok) { | |
dfc70645 | 7513 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 7514 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
7515 | md_wakeup_thread(mddev->thread); |
7516 | // stop recovery, signal do_sync .... | |
7517 | } | |
7518 | } | |
6c144d31 | 7519 | EXPORT_SYMBOL(md_done_sync); |
1da177e4 | 7520 | |
06d91a5f N |
7521 | /* md_write_start(mddev, bi) |
7522 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
7523 | * in superblock) before writing, schedule a superblock update |
7524 | * and wait for it to complete. | |
06d91a5f | 7525 | */ |
fd01b88c | 7526 | void md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 7527 | { |
0fd62b86 | 7528 | int did_change = 0; |
06d91a5f | 7529 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 7530 | return; |
06d91a5f | 7531 | |
f91de92e N |
7532 | BUG_ON(mddev->ro == 1); |
7533 | if (mddev->ro == 2) { | |
7534 | /* need to switch to read/write */ | |
7535 | mddev->ro = 0; | |
7536 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7537 | md_wakeup_thread(mddev->thread); | |
25156198 | 7538 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 7539 | did_change = 1; |
f91de92e | 7540 | } |
06d91a5f | 7541 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
7542 | if (mddev->safemode == 1) |
7543 | mddev->safemode = 0; | |
06d91a5f | 7544 | if (mddev->in_sync) { |
85572d7c | 7545 | spin_lock(&mddev->lock); |
3d310eb7 N |
7546 | if (mddev->in_sync) { |
7547 | mddev->in_sync = 0; | |
850b2b42 | 7548 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 7549 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 7550 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 7551 | did_change = 1; |
3d310eb7 | 7552 | } |
85572d7c | 7553 | spin_unlock(&mddev->lock); |
06d91a5f | 7554 | } |
0fd62b86 | 7555 | if (did_change) |
00bcb4ac | 7556 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 7557 | wait_event(mddev->sb_wait, |
09a44cc1 | 7558 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 | 7559 | } |
6c144d31 | 7560 | EXPORT_SYMBOL(md_write_start); |
1da177e4 | 7561 | |
fd01b88c | 7562 | void md_write_end(struct mddev *mddev) |
1da177e4 LT |
7563 | { |
7564 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
7565 | if (mddev->safemode == 2) | |
7566 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 7567 | else if (mddev->safemode_delay) |
1da177e4 LT |
7568 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
7569 | } | |
7570 | } | |
6c144d31 | 7571 | EXPORT_SYMBOL(md_write_end); |
1da177e4 | 7572 | |
2a2275d6 N |
7573 | /* md_allow_write(mddev) |
7574 | * Calling this ensures that the array is marked 'active' so that writes | |
7575 | * may proceed without blocking. It is important to call this before | |
7576 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
7577 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
7578 | * |
7579 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
7580 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 7581 | */ |
fd01b88c | 7582 | int md_allow_write(struct mddev *mddev) |
2a2275d6 N |
7583 | { |
7584 | if (!mddev->pers) | |
b5470dc5 | 7585 | return 0; |
2a2275d6 | 7586 | if (mddev->ro) |
b5470dc5 | 7587 | return 0; |
1a0fd497 | 7588 | if (!mddev->pers->sync_request) |
b5470dc5 | 7589 | return 0; |
2a2275d6 | 7590 | |
85572d7c | 7591 | spin_lock(&mddev->lock); |
2a2275d6 N |
7592 | if (mddev->in_sync) { |
7593 | mddev->in_sync = 0; | |
7594 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 7595 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
7596 | if (mddev->safemode_delay && |
7597 | mddev->safemode == 0) | |
7598 | mddev->safemode = 1; | |
85572d7c | 7599 | spin_unlock(&mddev->lock); |
293467aa GR |
7600 | if (mddev_is_clustered(mddev)) |
7601 | md_cluster_ops->metadata_update_start(mddev); | |
2a2275d6 | 7602 | md_update_sb(mddev, 0); |
293467aa GR |
7603 | if (mddev_is_clustered(mddev)) |
7604 | md_cluster_ops->metadata_update_finish(mddev); | |
00bcb4ac | 7605 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 | 7606 | } else |
85572d7c | 7607 | spin_unlock(&mddev->lock); |
b5470dc5 | 7608 | |
070dc6dd | 7609 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
7610 | return -EAGAIN; |
7611 | else | |
7612 | return 0; | |
2a2275d6 N |
7613 | } |
7614 | EXPORT_SYMBOL_GPL(md_allow_write); | |
7615 | ||
1da177e4 LT |
7616 | #define SYNC_MARKS 10 |
7617 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 7618 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 7619 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 7620 | { |
4ed8731d | 7621 | struct mddev *mddev = thread->mddev; |
fd01b88c | 7622 | struct mddev *mddev2; |
1da177e4 LT |
7623 | unsigned int currspeed = 0, |
7624 | window; | |
ac7e50a3 | 7625 | sector_t max_sectors,j, io_sectors, recovery_done; |
1da177e4 | 7626 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 7627 | unsigned long update_time; |
1da177e4 LT |
7628 | sector_t mark_cnt[SYNC_MARKS]; |
7629 | int last_mark,m; | |
7630 | struct list_head *tmp; | |
7631 | sector_t last_check; | |
57afd89f | 7632 | int skipped = 0; |
3cb03002 | 7633 | struct md_rdev *rdev; |
c4a39551 | 7634 | char *desc, *action = NULL; |
7c2c57c9 | 7635 | struct blk_plug plug; |
1da177e4 LT |
7636 | |
7637 | /* just incase thread restarts... */ | |
7638 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7639 | return; | |
3991b31e N |
7640 | if (mddev->ro) {/* never try to sync a read-only array */ |
7641 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5fd6c1dc | 7642 | return; |
3991b31e | 7643 | } |
1da177e4 | 7644 | |
61df9d91 | 7645 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 7646 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 7647 | desc = "data-check"; |
c4a39551 JB |
7648 | action = "check"; |
7649 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 7650 | desc = "requested-resync"; |
c4a39551 JB |
7651 | action = "repair"; |
7652 | } else | |
61df9d91 N |
7653 | desc = "resync"; |
7654 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
7655 | desc = "reshape"; | |
7656 | else | |
7657 | desc = "recovery"; | |
7658 | ||
c4a39551 JB |
7659 | mddev->last_sync_action = action ?: desc; |
7660 | ||
1da177e4 LT |
7661 | /* we overload curr_resync somewhat here. |
7662 | * 0 == not engaged in resync at all | |
7663 | * 2 == checking that there is no conflict with another sync | |
7664 | * 1 == like 2, but have yielded to allow conflicting resync to | |
7665 | * commense | |
7666 | * other == active in resync - this many blocks | |
7667 | * | |
7668 | * Before starting a resync we must have set curr_resync to | |
7669 | * 2, and then checked that every "conflicting" array has curr_resync | |
7670 | * less than ours. When we find one that is the same or higher | |
7671 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
7672 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
7673 | * This will mean we have to start checking from the beginning again. | |
7674 | * | |
7675 | */ | |
7676 | ||
7677 | do { | |
7678 | mddev->curr_resync = 2; | |
7679 | ||
7680 | try_again: | |
404e4b43 | 7681 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 7682 | goto skip; |
29ac4aa3 | 7683 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
7684 | if (mddev2 == mddev) |
7685 | continue; | |
90b08710 BS |
7686 | if (!mddev->parallel_resync |
7687 | && mddev2->curr_resync | |
7688 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
7689 | DEFINE_WAIT(wq); |
7690 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
7691 | /* arbitrarily yield */ | |
7692 | mddev->curr_resync = 1; | |
7693 | wake_up(&resync_wait); | |
7694 | } | |
7695 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
7696 | /* no need to wait here, we can wait the next | |
7697 | * time 'round when curr_resync == 2 | |
7698 | */ | |
7699 | continue; | |
9744197c N |
7700 | /* We need to wait 'interruptible' so as not to |
7701 | * contribute to the load average, and not to | |
7702 | * be caught by 'softlockup' | |
7703 | */ | |
7704 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 7705 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 7706 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
7707 | printk(KERN_INFO "md: delaying %s of %s" |
7708 | " until %s has finished (they" | |
1da177e4 | 7709 | " share one or more physical units)\n", |
61df9d91 | 7710 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 7711 | mddev_put(mddev2); |
9744197c N |
7712 | if (signal_pending(current)) |
7713 | flush_signals(current); | |
1da177e4 LT |
7714 | schedule(); |
7715 | finish_wait(&resync_wait, &wq); | |
7716 | goto try_again; | |
7717 | } | |
7718 | finish_wait(&resync_wait, &wq); | |
7719 | } | |
7720 | } | |
7721 | } while (mddev->curr_resync < 2); | |
7722 | ||
5fd6c1dc | 7723 | j = 0; |
9d88883e | 7724 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 7725 | /* resync follows the size requested by the personality, |
57afd89f | 7726 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
7727 | */ |
7728 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 7729 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 7730 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
7731 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
7732 | j = mddev->resync_min; | |
7733 | else if (!mddev->bitmap) | |
5fd6c1dc | 7734 | j = mddev->recovery_cp; |
5e96ee65 | 7735 | |
ccfcc3c1 | 7736 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 7737 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 7738 | else { |
1da177e4 | 7739 | /* recovery follows the physical size of devices */ |
58c0fed4 | 7740 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 7741 | j = MaxSector; |
4e59ca7d | 7742 | rcu_read_lock(); |
dafb20fa | 7743 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc N |
7744 | if (rdev->raid_disk >= 0 && |
7745 | !test_bit(Faulty, &rdev->flags) && | |
7746 | !test_bit(In_sync, &rdev->flags) && | |
7747 | rdev->recovery_offset < j) | |
7748 | j = rdev->recovery_offset; | |
4e59ca7d | 7749 | rcu_read_unlock(); |
133d4527 N |
7750 | |
7751 | /* If there is a bitmap, we need to make sure all | |
7752 | * writes that started before we added a spare | |
7753 | * complete before we start doing a recovery. | |
7754 | * Otherwise the write might complete and (via | |
7755 | * bitmap_endwrite) set a bit in the bitmap after the | |
7756 | * recovery has checked that bit and skipped that | |
7757 | * region. | |
7758 | */ | |
7759 | if (mddev->bitmap) { | |
7760 | mddev->pers->quiesce(mddev, 1); | |
7761 | mddev->pers->quiesce(mddev, 0); | |
7762 | } | |
5fd6c1dc | 7763 | } |
1da177e4 | 7764 | |
61df9d91 N |
7765 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
7766 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
7767 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 7768 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
7769 | "(but not more than %d KB/sec) for %s.\n", |
7770 | speed_max(mddev), desc); | |
1da177e4 | 7771 | |
eea1bf38 | 7772 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 7773 | |
57afd89f | 7774 | io_sectors = 0; |
1da177e4 LT |
7775 | for (m = 0; m < SYNC_MARKS; m++) { |
7776 | mark[m] = jiffies; | |
57afd89f | 7777 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
7778 | } |
7779 | last_mark = 0; | |
7780 | mddev->resync_mark = mark[last_mark]; | |
7781 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
7782 | ||
7783 | /* | |
7784 | * Tune reconstruction: | |
7785 | */ | |
7786 | window = 32*(PAGE_SIZE/512); | |
ac42450c JB |
7787 | printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", |
7788 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
7789 | |
7790 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
7791 | last_check = 0; |
7792 | ||
7793 | if (j>2) { | |
c91abf5a | 7794 | printk(KERN_INFO |
61df9d91 N |
7795 | "md: resuming %s of %s from checkpoint.\n", |
7796 | desc, mdname(mddev)); | |
1da177e4 | 7797 | mddev->curr_resync = j; |
72f36d59 N |
7798 | } else |
7799 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 7800 | mddev->curr_resync_completed = j; |
72f36d59 N |
7801 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
7802 | md_new_event(mddev); | |
54f89341 | 7803 | update_time = jiffies; |
1da177e4 | 7804 | |
96ae923a | 7805 | if (mddev_is_clustered(mddev)) |
965400eb | 7806 | md_cluster_ops->resync_start(mddev, j, max_sectors); |
96ae923a | 7807 | |
7c2c57c9 | 7808 | blk_start_plug(&plug); |
1da177e4 | 7809 | while (j < max_sectors) { |
57afd89f | 7810 | sector_t sectors; |
1da177e4 | 7811 | |
57afd89f | 7812 | skipped = 0; |
97e4f42d | 7813 | |
7a91ee1f N |
7814 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7815 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
7816 | (mddev->curr_resync - mddev->curr_resync_completed) | |
7817 | > (max_sectors >> 4)) || | |
54f89341 | 7818 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f | 7819 | (j - mddev->curr_resync_completed)*2 |
c5e19d90 N |
7820 | >= mddev->resync_max - mddev->curr_resync_completed || |
7821 | mddev->curr_resync_completed > mddev->resync_max | |
7a91ee1f | 7822 | )) { |
97e4f42d | 7823 | /* time to update curr_resync_completed */ |
97e4f42d N |
7824 | wait_event(mddev->recovery_wait, |
7825 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 7826 | mddev->curr_resync_completed = j; |
35d78c66 | 7827 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
7828 | j > mddev->recovery_cp) | |
7829 | mddev->recovery_cp = j; | |
54f89341 | 7830 | update_time = jiffies; |
070dc6dd | 7831 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 7832 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 7833 | } |
acb180b0 | 7834 | |
c91abf5a N |
7835 | while (j >= mddev->resync_max && |
7836 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
7837 | /* As this condition is controlled by user-space, |
7838 | * we can block indefinitely, so use '_interruptible' | |
7839 | * to avoid triggering warnings. | |
7840 | */ | |
7841 | flush_signals(current); /* just in case */ | |
7842 | wait_event_interruptible(mddev->recovery_wait, | |
7843 | mddev->resync_max > j | |
c91abf5a N |
7844 | || test_bit(MD_RECOVERY_INTR, |
7845 | &mddev->recovery)); | |
e62e58a5 | 7846 | } |
acb180b0 | 7847 | |
c91abf5a N |
7848 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7849 | break; | |
acb180b0 | 7850 | |
09314799 | 7851 | sectors = mddev->pers->sync_request(mddev, j, &skipped); |
57afd89f | 7852 | if (sectors == 0) { |
dfc70645 | 7853 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 7854 | break; |
1da177e4 | 7855 | } |
57afd89f N |
7856 | |
7857 | if (!skipped) { /* actual IO requested */ | |
7858 | io_sectors += sectors; | |
7859 | atomic_add(sectors, &mddev->recovery_active); | |
7860 | } | |
7861 | ||
e875ecea N |
7862 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7863 | break; | |
7864 | ||
1da177e4 | 7865 | j += sectors; |
5ed1df2e N |
7866 | if (j > max_sectors) |
7867 | /* when skipping, extra large numbers can be returned. */ | |
7868 | j = max_sectors; | |
72f36d59 N |
7869 | if (j > 2) |
7870 | mddev->curr_resync = j; | |
96ae923a GR |
7871 | if (mddev_is_clustered(mddev)) |
7872 | md_cluster_ops->resync_info_update(mddev, j, max_sectors); | |
ff4e8d9a | 7873 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 7874 | if (last_check == 0) |
e875ecea | 7875 | /* this is the earliest that rebuild will be |
d7603b7e N |
7876 | * visible in /proc/mdstat |
7877 | */ | |
7878 | md_new_event(mddev); | |
57afd89f N |
7879 | |
7880 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
7881 | continue; |
7882 | ||
57afd89f | 7883 | last_check = io_sectors; |
1da177e4 LT |
7884 | repeat: |
7885 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
7886 | /* step marks */ | |
7887 | int next = (last_mark+1) % SYNC_MARKS; | |
7888 | ||
7889 | mddev->resync_mark = mark[next]; | |
7890 | mddev->resync_mark_cnt = mark_cnt[next]; | |
7891 | mark[next] = jiffies; | |
57afd89f | 7892 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
7893 | last_mark = next; |
7894 | } | |
7895 | ||
c91abf5a N |
7896 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7897 | break; | |
1da177e4 LT |
7898 | |
7899 | /* | |
7900 | * this loop exits only if either when we are slower than | |
7901 | * the 'hard' speed limit, or the system was IO-idle for | |
7902 | * a jiffy. | |
7903 | * the system might be non-idle CPU-wise, but we only care | |
7904 | * about not overloading the IO subsystem. (things like an | |
7905 | * e2fsck being done on the RAID array should execute fast) | |
7906 | */ | |
1da177e4 LT |
7907 | cond_resched(); |
7908 | ||
ac7e50a3 XN |
7909 | recovery_done = io_sectors - atomic_read(&mddev->recovery_active); |
7910 | currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2 | |
57afd89f | 7911 | /((jiffies-mddev->resync_mark)/HZ +1) +1; |
1da177e4 | 7912 | |
88202a0c | 7913 | if (currspeed > speed_min(mddev)) { |
ac8fa419 | 7914 | if (currspeed > speed_max(mddev)) { |
c0e48521 | 7915 | msleep(500); |
1da177e4 LT |
7916 | goto repeat; |
7917 | } | |
ac8fa419 N |
7918 | if (!is_mddev_idle(mddev, 0)) { |
7919 | /* | |
7920 | * Give other IO more of a chance. | |
7921 | * The faster the devices, the less we wait. | |
7922 | */ | |
7923 | wait_event(mddev->recovery_wait, | |
7924 | !atomic_read(&mddev->recovery_active)); | |
7925 | } | |
1da177e4 LT |
7926 | } |
7927 | } | |
c91abf5a N |
7928 | printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc, |
7929 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
7930 | ? "interrupted" : "done"); | |
1da177e4 LT |
7931 | /* |
7932 | * this also signals 'finished resyncing' to md_stop | |
7933 | */ | |
7c2c57c9 | 7934 | blk_finish_plug(&plug); |
1da177e4 LT |
7935 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
7936 | ||
5ed1df2e N |
7937 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7938 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
7939 | mddev->curr_resync > 2) { | |
7940 | mddev->curr_resync_completed = mddev->curr_resync; | |
7941 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
7942 | } | |
1da177e4 | 7943 | /* tell personality that we are finished */ |
09314799 | 7944 | mddev->pers->sync_request(mddev, max_sectors, &skipped); |
1da177e4 | 7945 | |
dfc70645 | 7946 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
7947 | mddev->curr_resync > 2) { |
7948 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
7949 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7950 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
7951 | printk(KERN_INFO | |
61df9d91 N |
7952 | "md: checkpointing %s of %s.\n", |
7953 | desc, mdname(mddev)); | |
0a19caab | 7954 | if (test_bit(MD_RECOVERY_ERROR, |
7955 | &mddev->recovery)) | |
7956 | mddev->recovery_cp = | |
7957 | mddev->curr_resync_completed; | |
7958 | else | |
7959 | mddev->recovery_cp = | |
7960 | mddev->curr_resync; | |
5fd6c1dc N |
7961 | } |
7962 | } else | |
7963 | mddev->recovery_cp = MaxSector; | |
7964 | } else { | |
7965 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
7966 | mddev->curr_resync = MaxSector; | |
4e59ca7d | 7967 | rcu_read_lock(); |
dafb20fa | 7968 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 7969 | if (rdev->raid_disk >= 0 && |
70fffd0b | 7970 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
7971 | !test_bit(Faulty, &rdev->flags) && |
7972 | !test_bit(In_sync, &rdev->flags) && | |
7973 | rdev->recovery_offset < mddev->curr_resync) | |
7974 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 7975 | rcu_read_unlock(); |
5fd6c1dc | 7976 | } |
1da177e4 | 7977 | } |
db91ff55 | 7978 | skip: |
dc737d7c GJ |
7979 | if (mddev_is_clustered(mddev)) |
7980 | md_cluster_ops->resync_finish(mddev); | |
7981 | ||
17571284 | 7982 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 7983 | |
23da422b | 7984 | spin_lock(&mddev->lock); |
c07b70ad N |
7985 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
7986 | /* We completed so min/max setting can be forgotten if used. */ | |
7987 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7988 | mddev->resync_min = 0; | |
7989 | mddev->resync_max = MaxSector; | |
7990 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7991 | mddev->resync_min = mddev->curr_resync_completed; | |
f7851be7 | 7992 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); |
1da177e4 | 7993 | mddev->curr_resync = 0; |
23da422b N |
7994 | spin_unlock(&mddev->lock); |
7995 | ||
1da177e4 | 7996 | wake_up(&resync_wait); |
1da177e4 | 7997 | md_wakeup_thread(mddev->thread); |
c6207277 | 7998 | return; |
1da177e4 | 7999 | } |
29269553 | 8000 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 8001 | |
746d3207 N |
8002 | static int remove_and_add_spares(struct mddev *mddev, |
8003 | struct md_rdev *this) | |
b4c4c7b8 | 8004 | { |
3cb03002 | 8005 | struct md_rdev *rdev; |
b4c4c7b8 | 8006 | int spares = 0; |
f2a371c5 | 8007 | int removed = 0; |
b4c4c7b8 | 8008 | |
dafb20fa | 8009 | rdev_for_each(rdev, mddev) |
746d3207 N |
8010 | if ((this == NULL || rdev == this) && |
8011 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 8012 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
8013 | (test_bit(Faulty, &rdev->flags) || |
8014 | ! test_bit(In_sync, &rdev->flags)) && | |
8015 | atomic_read(&rdev->nr_pending)==0) { | |
8016 | if (mddev->pers->hot_remove_disk( | |
b8321b68 | 8017 | mddev, rdev) == 0) { |
36fad858 | 8018 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 8019 | rdev->raid_disk = -1; |
f2a371c5 | 8020 | removed++; |
b4c4c7b8 N |
8021 | } |
8022 | } | |
90584fc9 JB |
8023 | if (removed && mddev->kobj.sd) |
8024 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 8025 | |
746d3207 N |
8026 | if (this) |
8027 | goto no_add; | |
8028 | ||
dafb20fa | 8029 | rdev_for_each(rdev, mddev) { |
7bfec5f3 N |
8030 | if (rdev->raid_disk >= 0 && |
8031 | !test_bit(In_sync, &rdev->flags) && | |
8032 | !test_bit(Faulty, &rdev->flags)) | |
8033 | spares++; | |
7ceb17e8 N |
8034 | if (rdev->raid_disk >= 0) |
8035 | continue; | |
8036 | if (test_bit(Faulty, &rdev->flags)) | |
8037 | continue; | |
8038 | if (mddev->ro && | |
8313b8e5 N |
8039 | ! (rdev->saved_raid_disk >= 0 && |
8040 | !test_bit(Bitmap_sync, &rdev->flags))) | |
7ceb17e8 N |
8041 | continue; |
8042 | ||
7eb41885 N |
8043 | if (rdev->saved_raid_disk < 0) |
8044 | rdev->recovery_offset = 0; | |
7ceb17e8 N |
8045 | if (mddev->pers-> |
8046 | hot_add_disk(mddev, rdev) == 0) { | |
8047 | if (sysfs_link_rdev(mddev, rdev)) | |
8048 | /* failure here is OK */; | |
8049 | spares++; | |
8050 | md_new_event(mddev); | |
8051 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
dfc70645 | 8052 | } |
b4c4c7b8 | 8053 | } |
746d3207 | 8054 | no_add: |
6dafab6b N |
8055 | if (removed) |
8056 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
b4c4c7b8 N |
8057 | return spares; |
8058 | } | |
7ebc0be7 | 8059 | |
ac05f256 N |
8060 | static void md_start_sync(struct work_struct *ws) |
8061 | { | |
8062 | struct mddev *mddev = container_of(ws, struct mddev, del_work); | |
8063 | ||
8064 | mddev->sync_thread = md_register_thread(md_do_sync, | |
8065 | mddev, | |
8066 | "resync"); | |
8067 | if (!mddev->sync_thread) { | |
8068 | printk(KERN_ERR "%s: could not start resync" | |
8069 | " thread...\n", | |
8070 | mdname(mddev)); | |
8071 | /* leave the spares where they are, it shouldn't hurt */ | |
8072 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
8073 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8074 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8075 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
8076 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8077 | wake_up(&resync_wait); |
ac05f256 N |
8078 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8079 | &mddev->recovery)) | |
8080 | if (mddev->sysfs_action) | |
8081 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8082 | } else | |
8083 | md_wakeup_thread(mddev->sync_thread); | |
8084 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8085 | md_new_event(mddev); | |
8086 | } | |
8087 | ||
1da177e4 LT |
8088 | /* |
8089 | * This routine is regularly called by all per-raid-array threads to | |
8090 | * deal with generic issues like resync and super-block update. | |
8091 | * Raid personalities that don't have a thread (linear/raid0) do not | |
8092 | * need this as they never do any recovery or update the superblock. | |
8093 | * | |
8094 | * It does not do any resync itself, but rather "forks" off other threads | |
8095 | * to do that as needed. | |
8096 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
8097 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 8098 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
8099 | * and wakeups up this thread which will reap the thread and finish up. |
8100 | * This thread also removes any faulty devices (with nr_pending == 0). | |
8101 | * | |
8102 | * The overall approach is: | |
8103 | * 1/ if the superblock needs updating, update it. | |
8104 | * 2/ If a recovery thread is running, don't do anything else. | |
8105 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
8106 | * 4/ If there are any faulty devices, remove them. | |
8107 | * 5/ If array is degraded, try to add spares devices | |
8108 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
8109 | */ | |
fd01b88c | 8110 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 8111 | { |
68866e42 JB |
8112 | if (mddev->suspended) |
8113 | return; | |
8114 | ||
5f40402d | 8115 | if (mddev->bitmap) |
aa5cbd10 | 8116 | bitmap_daemon_work(mddev); |
1da177e4 | 8117 | |
fca4d848 | 8118 | if (signal_pending(current)) { |
31a59e34 | 8119 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
8120 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
8121 | mdname(mddev)); | |
8122 | mddev->safemode = 2; | |
8123 | } | |
8124 | flush_signals(current); | |
8125 | } | |
8126 | ||
c89a8eee N |
8127 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
8128 | return; | |
1da177e4 | 8129 | if ( ! ( |
142d44c3 | 8130 | (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 8131 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 8132 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 8133 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
8134 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
8135 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
8136 | )) |
8137 | return; | |
fca4d848 | 8138 | |
df5b89b3 | 8139 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 8140 | int spares = 0; |
fca4d848 | 8141 | |
c89a8eee | 8142 | if (mddev->ro) { |
ab16bfc7 NB |
8143 | struct md_rdev *rdev; |
8144 | if (!mddev->external && mddev->in_sync) | |
8145 | /* 'Blocked' flag not needed as failed devices | |
8146 | * will be recorded if array switched to read/write. | |
8147 | * Leaving it set will prevent the device | |
8148 | * from being removed. | |
8149 | */ | |
8150 | rdev_for_each(rdev, mddev) | |
8151 | clear_bit(Blocked, &rdev->flags); | |
7ceb17e8 N |
8152 | /* On a read-only array we can: |
8153 | * - remove failed devices | |
8154 | * - add already-in_sync devices if the array itself | |
8155 | * is in-sync. | |
8156 | * As we only add devices that are already in-sync, | |
8157 | * we can activate the spares immediately. | |
c89a8eee | 8158 | */ |
7ceb17e8 | 8159 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
8160 | /* There is no thread, but we need to call |
8161 | * ->spare_active and clear saved_raid_disk | |
8162 | */ | |
2ac295a5 | 8163 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8313b8e5 | 8164 | md_reap_sync_thread(mddev); |
a4a3d26d | 8165 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
8313b8e5 | 8166 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
d4929add | 8167 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
c89a8eee N |
8168 | goto unlock; |
8169 | } | |
8170 | ||
31a59e34 | 8171 | if (!mddev->external) { |
0fd62b86 | 8172 | int did_change = 0; |
85572d7c | 8173 | spin_lock(&mddev->lock); |
31a59e34 N |
8174 | if (mddev->safemode && |
8175 | !atomic_read(&mddev->writes_pending) && | |
8176 | !mddev->in_sync && | |
8177 | mddev->recovery_cp == MaxSector) { | |
8178 | mddev->in_sync = 1; | |
0fd62b86 | 8179 | did_change = 1; |
070dc6dd | 8180 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
8181 | } |
8182 | if (mddev->safemode == 1) | |
8183 | mddev->safemode = 0; | |
85572d7c | 8184 | spin_unlock(&mddev->lock); |
0fd62b86 | 8185 | if (did_change) |
00bcb4ac | 8186 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 8187 | } |
fca4d848 | 8188 | |
293467aa GR |
8189 | if (mddev->flags & MD_UPDATE_SB_FLAGS) { |
8190 | if (mddev_is_clustered(mddev)) | |
8191 | md_cluster_ops->metadata_update_start(mddev); | |
850b2b42 | 8192 | md_update_sb(mddev, 0); |
293467aa GR |
8193 | if (mddev_is_clustered(mddev)) |
8194 | md_cluster_ops->metadata_update_finish(mddev); | |
8195 | } | |
06d91a5f | 8196 | |
1da177e4 LT |
8197 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
8198 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
8199 | /* resync/recovery still happening */ | |
8200 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8201 | goto unlock; | |
8202 | } | |
8203 | if (mddev->sync_thread) { | |
a91d5ac0 | 8204 | md_reap_sync_thread(mddev); |
1da177e4 LT |
8205 | goto unlock; |
8206 | } | |
72a23c21 NB |
8207 | /* Set RUNNING before clearing NEEDED to avoid |
8208 | * any transients in the value of "sync_action". | |
8209 | */ | |
72f36d59 | 8210 | mddev->curr_resync_completed = 0; |
23da422b | 8211 | spin_lock(&mddev->lock); |
72a23c21 | 8212 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
23da422b | 8213 | spin_unlock(&mddev->lock); |
24dd469d N |
8214 | /* Clear some bits that don't mean anything, but |
8215 | * might be left set | |
8216 | */ | |
24dd469d N |
8217 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8218 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 8219 | |
ed209584 N |
8220 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
8221 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
ac05f256 | 8222 | goto not_running; |
1da177e4 LT |
8223 | /* no recovery is running. |
8224 | * remove any failed drives, then | |
8225 | * add spares if possible. | |
72f36d59 | 8226 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
8227 | * the personality to fail the re-add. |
8228 | */ | |
1da177e4 | 8229 | |
b4c4c7b8 | 8230 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
8231 | if (mddev->pers->check_reshape == NULL || |
8232 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 | 8233 | /* Cannot proceed */ |
ac05f256 | 8234 | goto not_running; |
b4c4c7b8 | 8235 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
72a23c21 | 8236 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 8237 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
8238 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8239 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 8240 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 8241 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8242 | } else if (mddev->recovery_cp < MaxSector) { |
8243 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 8244 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8245 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
8246 | /* nothing to be done ... */ | |
ac05f256 | 8247 | goto not_running; |
24dd469d | 8248 | |
1da177e4 | 8249 | if (mddev->pers->sync_request) { |
ef99bf48 | 8250 | if (spares) { |
a654b9d8 N |
8251 | /* We are adding a device or devices to an array |
8252 | * which has the bitmap stored on all devices. | |
8253 | * So make sure all bitmap pages get written | |
8254 | */ | |
8255 | bitmap_write_all(mddev->bitmap); | |
8256 | } | |
ac05f256 N |
8257 | INIT_WORK(&mddev->del_work, md_start_sync); |
8258 | queue_work(md_misc_wq, &mddev->del_work); | |
8259 | goto unlock; | |
1da177e4 | 8260 | } |
ac05f256 | 8261 | not_running: |
72a23c21 NB |
8262 | if (!mddev->sync_thread) { |
8263 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8264 | wake_up(&resync_wait); |
72a23c21 NB |
8265 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8266 | &mddev->recovery)) | |
0c3573f1 | 8267 | if (mddev->sysfs_action) |
00bcb4ac | 8268 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 8269 | } |
ac05f256 N |
8270 | unlock: |
8271 | wake_up(&mddev->sb_wait); | |
1da177e4 LT |
8272 | mddev_unlock(mddev); |
8273 | } | |
8274 | } | |
6c144d31 | 8275 | EXPORT_SYMBOL(md_check_recovery); |
1da177e4 | 8276 | |
a91d5ac0 JB |
8277 | void md_reap_sync_thread(struct mddev *mddev) |
8278 | { | |
8279 | struct md_rdev *rdev; | |
8280 | ||
8281 | /* resync has finished, collect result */ | |
8282 | md_unregister_thread(&mddev->sync_thread); | |
8283 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
8284 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
8285 | /* success...*/ | |
8286 | /* activate any spares */ | |
8287 | if (mddev->pers->spare_active(mddev)) { | |
8288 | sysfs_notify(&mddev->kobj, NULL, | |
8289 | "degraded"); | |
8290 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
8291 | } | |
8292 | } | |
293467aa GR |
8293 | if (mddev_is_clustered(mddev)) |
8294 | md_cluster_ops->metadata_update_start(mddev); | |
a91d5ac0 JB |
8295 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8296 | mddev->pers->finish_reshape) | |
8297 | mddev->pers->finish_reshape(mddev); | |
8298 | ||
8299 | /* If array is no-longer degraded, then any saved_raid_disk | |
f466722c | 8300 | * information must be scrapped. |
a91d5ac0 | 8301 | */ |
f466722c N |
8302 | if (!mddev->degraded) |
8303 | rdev_for_each(rdev, mddev) | |
a91d5ac0 JB |
8304 | rdev->saved_raid_disk = -1; |
8305 | ||
8306 | md_update_sb(mddev, 1); | |
293467aa GR |
8307 | if (mddev_is_clustered(mddev)) |
8308 | md_cluster_ops->metadata_update_finish(mddev); | |
a91d5ac0 | 8309 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
ea358cd0 | 8310 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
a91d5ac0 JB |
8311 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8312 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8313 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8314 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
f851b60d | 8315 | wake_up(&resync_wait); |
a91d5ac0 JB |
8316 | /* flag recovery needed just to double check */ |
8317 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8318 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8319 | md_new_event(mddev); | |
8320 | if (mddev->event_work.func) | |
8321 | queue_work(md_misc_wq, &mddev->event_work); | |
8322 | } | |
6c144d31 | 8323 | EXPORT_SYMBOL(md_reap_sync_thread); |
a91d5ac0 | 8324 | |
fd01b88c | 8325 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 8326 | { |
00bcb4ac | 8327 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 8328 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
8329 | !test_bit(Blocked, &rdev->flags) && |
8330 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
8331 | msecs_to_jiffies(5000)); |
8332 | rdev_dec_pending(rdev, mddev); | |
8333 | } | |
8334 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
8335 | ||
c6563a8c N |
8336 | void md_finish_reshape(struct mddev *mddev) |
8337 | { | |
8338 | /* called be personality module when reshape completes. */ | |
8339 | struct md_rdev *rdev; | |
8340 | ||
8341 | rdev_for_each(rdev, mddev) { | |
8342 | if (rdev->data_offset > rdev->new_data_offset) | |
8343 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
8344 | else | |
8345 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
8346 | rdev->data_offset = rdev->new_data_offset; | |
8347 | } | |
8348 | } | |
8349 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 N |
8350 | |
8351 | /* Bad block management. | |
8352 | * We can record which blocks on each device are 'bad' and so just | |
8353 | * fail those blocks, or that stripe, rather than the whole device. | |
8354 | * Entries in the bad-block table are 64bits wide. This comprises: | |
8355 | * Length of bad-range, in sectors: 0-511 for lengths 1-512 | |
8356 | * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes) | |
8357 | * A 'shift' can be set so that larger blocks are tracked and | |
8358 | * consequently larger devices can be covered. | |
8359 | * 'Acknowledged' flag - 1 bit. - the most significant bit. | |
8360 | * | |
8361 | * Locking of the bad-block table uses a seqlock so md_is_badblock | |
8362 | * might need to retry if it is very unlucky. | |
8363 | * We will sometimes want to check for bad blocks in a bi_end_io function, | |
8364 | * so we use the write_seqlock_irq variant. | |
8365 | * | |
8366 | * When looking for a bad block we specify a range and want to | |
8367 | * know if any block in the range is bad. So we binary-search | |
8368 | * to the last range that starts at-or-before the given endpoint, | |
8369 | * (or "before the sector after the target range") | |
8370 | * then see if it ends after the given start. | |
8371 | * We return | |
8372 | * 0 if there are no known bad blocks in the range | |
8373 | * 1 if there are known bad block which are all acknowledged | |
8374 | * -1 if there are bad blocks which have not yet been acknowledged in metadata. | |
8375 | * plus the start/length of the first bad section we overlap. | |
8376 | */ | |
8377 | int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, | |
8378 | sector_t *first_bad, int *bad_sectors) | |
8379 | { | |
8380 | int hi; | |
ab05613a | 8381 | int lo; |
2230dfe4 | 8382 | u64 *p = bb->page; |
ab05613a | 8383 | int rv; |
2230dfe4 N |
8384 | sector_t target = s + sectors; |
8385 | unsigned seq; | |
8386 | ||
8387 | if (bb->shift > 0) { | |
8388 | /* round the start down, and the end up */ | |
8389 | s >>= bb->shift; | |
8390 | target += (1<<bb->shift) - 1; | |
8391 | target >>= bb->shift; | |
8392 | sectors = target - s; | |
8393 | } | |
8394 | /* 'target' is now the first block after the bad range */ | |
8395 | ||
8396 | retry: | |
8397 | seq = read_seqbegin(&bb->lock); | |
ab05613a | 8398 | lo = 0; |
8399 | rv = 0; | |
2230dfe4 N |
8400 | hi = bb->count; |
8401 | ||
8402 | /* Binary search between lo and hi for 'target' | |
8403 | * i.e. for the last range that starts before 'target' | |
8404 | */ | |
8405 | /* INVARIANT: ranges before 'lo' and at-or-after 'hi' | |
8406 | * are known not to be the last range before target. | |
8407 | * VARIANT: hi-lo is the number of possible | |
8408 | * ranges, and decreases until it reaches 1 | |
8409 | */ | |
8410 | while (hi - lo > 1) { | |
8411 | int mid = (lo + hi) / 2; | |
8412 | sector_t a = BB_OFFSET(p[mid]); | |
8413 | if (a < target) | |
8414 | /* This could still be the one, earlier ranges | |
8415 | * could not. */ | |
8416 | lo = mid; | |
8417 | else | |
8418 | /* This and later ranges are definitely out. */ | |
8419 | hi = mid; | |
8420 | } | |
8421 | /* 'lo' might be the last that started before target, but 'hi' isn't */ | |
8422 | if (hi > lo) { | |
8423 | /* need to check all range that end after 's' to see if | |
8424 | * any are unacknowledged. | |
8425 | */ | |
8426 | while (lo >= 0 && | |
8427 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8428 | if (BB_OFFSET(p[lo]) < target) { | |
8429 | /* starts before the end, and finishes after | |
8430 | * the start, so they must overlap | |
8431 | */ | |
8432 | if (rv != -1 && BB_ACK(p[lo])) | |
8433 | rv = 1; | |
8434 | else | |
8435 | rv = -1; | |
8436 | *first_bad = BB_OFFSET(p[lo]); | |
8437 | *bad_sectors = BB_LEN(p[lo]); | |
8438 | } | |
8439 | lo--; | |
8440 | } | |
8441 | } | |
8442 | ||
8443 | if (read_seqretry(&bb->lock, seq)) | |
8444 | goto retry; | |
8445 | ||
8446 | return rv; | |
8447 | } | |
8448 | EXPORT_SYMBOL_GPL(md_is_badblock); | |
8449 | ||
8450 | /* | |
8451 | * Add a range of bad blocks to the table. | |
8452 | * This might extend the table, or might contract it | |
8453 | * if two adjacent ranges can be merged. | |
8454 | * We binary-search to find the 'insertion' point, then | |
8455 | * decide how best to handle it. | |
8456 | */ | |
8457 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, | |
8458 | int acknowledged) | |
8459 | { | |
8460 | u64 *p; | |
8461 | int lo, hi; | |
8462 | int rv = 1; | |
905b0297 | 8463 | unsigned long flags; |
2230dfe4 N |
8464 | |
8465 | if (bb->shift < 0) | |
8466 | /* badblocks are disabled */ | |
8467 | return 0; | |
8468 | ||
8469 | if (bb->shift) { | |
8470 | /* round the start down, and the end up */ | |
8471 | sector_t next = s + sectors; | |
8472 | s >>= bb->shift; | |
8473 | next += (1<<bb->shift) - 1; | |
8474 | next >>= bb->shift; | |
8475 | sectors = next - s; | |
8476 | } | |
8477 | ||
905b0297 | 8478 | write_seqlock_irqsave(&bb->lock, flags); |
2230dfe4 N |
8479 | |
8480 | p = bb->page; | |
8481 | lo = 0; | |
8482 | hi = bb->count; | |
8483 | /* Find the last range that starts at-or-before 's' */ | |
8484 | while (hi - lo > 1) { | |
8485 | int mid = (lo + hi) / 2; | |
8486 | sector_t a = BB_OFFSET(p[mid]); | |
8487 | if (a <= s) | |
8488 | lo = mid; | |
8489 | else | |
8490 | hi = mid; | |
8491 | } | |
8492 | if (hi > lo && BB_OFFSET(p[lo]) > s) | |
8493 | hi = lo; | |
8494 | ||
8495 | if (hi > lo) { | |
8496 | /* we found a range that might merge with the start | |
8497 | * of our new range | |
8498 | */ | |
8499 | sector_t a = BB_OFFSET(p[lo]); | |
8500 | sector_t e = a + BB_LEN(p[lo]); | |
8501 | int ack = BB_ACK(p[lo]); | |
8502 | if (e >= s) { | |
8503 | /* Yes, we can merge with a previous range */ | |
8504 | if (s == a && s + sectors >= e) | |
8505 | /* new range covers old */ | |
8506 | ack = acknowledged; | |
8507 | else | |
8508 | ack = ack && acknowledged; | |
8509 | ||
8510 | if (e < s + sectors) | |
8511 | e = s + sectors; | |
8512 | if (e - a <= BB_MAX_LEN) { | |
8513 | p[lo] = BB_MAKE(a, e-a, ack); | |
8514 | s = e; | |
8515 | } else { | |
8516 | /* does not all fit in one range, | |
8517 | * make p[lo] maximal | |
8518 | */ | |
8519 | if (BB_LEN(p[lo]) != BB_MAX_LEN) | |
8520 | p[lo] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8521 | s = a + BB_MAX_LEN; | |
8522 | } | |
8523 | sectors = e - s; | |
8524 | } | |
8525 | } | |
8526 | if (sectors && hi < bb->count) { | |
8527 | /* 'hi' points to the first range that starts after 's'. | |
8528 | * Maybe we can merge with the start of that range */ | |
8529 | sector_t a = BB_OFFSET(p[hi]); | |
8530 | sector_t e = a + BB_LEN(p[hi]); | |
8531 | int ack = BB_ACK(p[hi]); | |
8532 | if (a <= s + sectors) { | |
8533 | /* merging is possible */ | |
8534 | if (e <= s + sectors) { | |
8535 | /* full overlap */ | |
8536 | e = s + sectors; | |
8537 | ack = acknowledged; | |
8538 | } else | |
8539 | ack = ack && acknowledged; | |
8540 | ||
8541 | a = s; | |
8542 | if (e - a <= BB_MAX_LEN) { | |
8543 | p[hi] = BB_MAKE(a, e-a, ack); | |
8544 | s = e; | |
8545 | } else { | |
8546 | p[hi] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8547 | s = a + BB_MAX_LEN; | |
8548 | } | |
8549 | sectors = e - s; | |
8550 | lo = hi; | |
8551 | hi++; | |
8552 | } | |
8553 | } | |
8554 | if (sectors == 0 && hi < bb->count) { | |
8555 | /* we might be able to combine lo and hi */ | |
8556 | /* Note: 's' is at the end of 'lo' */ | |
8557 | sector_t a = BB_OFFSET(p[hi]); | |
8558 | int lolen = BB_LEN(p[lo]); | |
8559 | int hilen = BB_LEN(p[hi]); | |
8560 | int newlen = lolen + hilen - (s - a); | |
8561 | if (s >= a && newlen < BB_MAX_LEN) { | |
8562 | /* yes, we can combine them */ | |
8563 | int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]); | |
8564 | p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack); | |
8565 | memmove(p + hi, p + hi + 1, | |
8566 | (bb->count - hi - 1) * 8); | |
8567 | bb->count--; | |
8568 | } | |
8569 | } | |
8570 | while (sectors) { | |
8571 | /* didn't merge (it all). | |
8572 | * Need to add a range just before 'hi' */ | |
8573 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8574 | /* No room for more */ | |
8575 | rv = 0; | |
8576 | break; | |
8577 | } else { | |
8578 | int this_sectors = sectors; | |
8579 | memmove(p + hi + 1, p + hi, | |
8580 | (bb->count - hi) * 8); | |
8581 | bb->count++; | |
8582 | ||
8583 | if (this_sectors > BB_MAX_LEN) | |
8584 | this_sectors = BB_MAX_LEN; | |
8585 | p[hi] = BB_MAKE(s, this_sectors, acknowledged); | |
8586 | sectors -= this_sectors; | |
8587 | s += this_sectors; | |
8588 | } | |
8589 | } | |
8590 | ||
8591 | bb->changed = 1; | |
de393cde N |
8592 | if (!acknowledged) |
8593 | bb->unacked_exist = 1; | |
905b0297 | 8594 | write_sequnlock_irqrestore(&bb->lock, flags); |
2230dfe4 N |
8595 | |
8596 | return rv; | |
8597 | } | |
8598 | ||
3cb03002 | 8599 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8600 | int is_new) |
2230dfe4 | 8601 | { |
c6563a8c N |
8602 | int rv; |
8603 | if (is_new) | |
8604 | s += rdev->new_data_offset; | |
8605 | else | |
8606 | s += rdev->data_offset; | |
8607 | rv = md_set_badblocks(&rdev->badblocks, | |
8608 | s, sectors, 0); | |
2230dfe4 N |
8609 | if (rv) { |
8610 | /* Make sure they get written out promptly */ | |
8bd2f0a0 | 8611 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2230dfe4 | 8612 | set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags); |
55ce74d4 | 8613 | set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags); |
2230dfe4 N |
8614 | md_wakeup_thread(rdev->mddev->thread); |
8615 | } | |
8616 | return rv; | |
8617 | } | |
8618 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8619 | ||
8620 | /* | |
8621 | * Remove a range of bad blocks from the table. | |
8622 | * This may involve extending the table if we spilt a region, | |
8623 | * but it must not fail. So if the table becomes full, we just | |
8624 | * drop the remove request. | |
8625 | */ | |
8626 | static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors) | |
8627 | { | |
8628 | u64 *p; | |
8629 | int lo, hi; | |
8630 | sector_t target = s + sectors; | |
8631 | int rv = 0; | |
8632 | ||
8633 | if (bb->shift > 0) { | |
8634 | /* When clearing we round the start up and the end down. | |
8635 | * This should not matter as the shift should align with | |
8636 | * the block size and no rounding should ever be needed. | |
8637 | * However it is better the think a block is bad when it | |
8638 | * isn't than to think a block is not bad when it is. | |
8639 | */ | |
8640 | s += (1<<bb->shift) - 1; | |
8641 | s >>= bb->shift; | |
8642 | target >>= bb->shift; | |
8643 | sectors = target - s; | |
8644 | } | |
8645 | ||
8646 | write_seqlock_irq(&bb->lock); | |
8647 | ||
8648 | p = bb->page; | |
8649 | lo = 0; | |
8650 | hi = bb->count; | |
8651 | /* Find the last range that starts before 'target' */ | |
8652 | while (hi - lo > 1) { | |
8653 | int mid = (lo + hi) / 2; | |
8654 | sector_t a = BB_OFFSET(p[mid]); | |
8655 | if (a < target) | |
8656 | lo = mid; | |
8657 | else | |
8658 | hi = mid; | |
8659 | } | |
8660 | if (hi > lo) { | |
8661 | /* p[lo] is the last range that could overlap the | |
8662 | * current range. Earlier ranges could also overlap, | |
8663 | * but only this one can overlap the end of the range. | |
8664 | */ | |
8665 | if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) { | |
8666 | /* Partial overlap, leave the tail of this range */ | |
8667 | int ack = BB_ACK(p[lo]); | |
8668 | sector_t a = BB_OFFSET(p[lo]); | |
8669 | sector_t end = a + BB_LEN(p[lo]); | |
8670 | ||
8671 | if (a < s) { | |
8672 | /* we need to split this range */ | |
8673 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8b32bf5e | 8674 | rv = -ENOSPC; |
2230dfe4 N |
8675 | goto out; |
8676 | } | |
8677 | memmove(p+lo+1, p+lo, (bb->count - lo) * 8); | |
8678 | bb->count++; | |
8679 | p[lo] = BB_MAKE(a, s-a, ack); | |
8680 | lo++; | |
8681 | } | |
8682 | p[lo] = BB_MAKE(target, end - target, ack); | |
8683 | /* there is no longer an overlap */ | |
8684 | hi = lo; | |
8685 | lo--; | |
8686 | } | |
8687 | while (lo >= 0 && | |
8688 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8689 | /* This range does overlap */ | |
8690 | if (BB_OFFSET(p[lo]) < s) { | |
8691 | /* Keep the early parts of this range. */ | |
8692 | int ack = BB_ACK(p[lo]); | |
8693 | sector_t start = BB_OFFSET(p[lo]); | |
8694 | p[lo] = BB_MAKE(start, s - start, ack); | |
8695 | /* now low doesn't overlap, so.. */ | |
8696 | break; | |
8697 | } | |
8698 | lo--; | |
8699 | } | |
8700 | /* 'lo' is strictly before, 'hi' is strictly after, | |
8701 | * anything between needs to be discarded | |
8702 | */ | |
8703 | if (hi - lo > 1) { | |
8704 | memmove(p+lo+1, p+hi, (bb->count - hi) * 8); | |
8705 | bb->count -= (hi - lo - 1); | |
8706 | } | |
8707 | } | |
8708 | ||
8709 | bb->changed = 1; | |
8710 | out: | |
8711 | write_sequnlock_irq(&bb->lock); | |
8712 | return rv; | |
8713 | } | |
8714 | ||
c6563a8c N |
8715 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8716 | int is_new) | |
2230dfe4 | 8717 | { |
c6563a8c N |
8718 | if (is_new) |
8719 | s += rdev->new_data_offset; | |
8720 | else | |
8721 | s += rdev->data_offset; | |
2230dfe4 | 8722 | return md_clear_badblocks(&rdev->badblocks, |
c6563a8c | 8723 | s, sectors); |
2230dfe4 N |
8724 | } |
8725 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8726 | ||
8727 | /* | |
8728 | * Acknowledge all bad blocks in a list. | |
8729 | * This only succeeds if ->changed is clear. It is used by | |
8730 | * in-kernel metadata updates | |
8731 | */ | |
8732 | void md_ack_all_badblocks(struct badblocks *bb) | |
8733 | { | |
8734 | if (bb->page == NULL || bb->changed) | |
8735 | /* no point even trying */ | |
8736 | return; | |
8737 | write_seqlock_irq(&bb->lock); | |
8738 | ||
ecb178bb | 8739 | if (bb->changed == 0 && bb->unacked_exist) { |
2230dfe4 N |
8740 | u64 *p = bb->page; |
8741 | int i; | |
8742 | for (i = 0; i < bb->count ; i++) { | |
8743 | if (!BB_ACK(p[i])) { | |
8744 | sector_t start = BB_OFFSET(p[i]); | |
8745 | int len = BB_LEN(p[i]); | |
8746 | p[i] = BB_MAKE(start, len, 1); | |
8747 | } | |
8748 | } | |
de393cde | 8749 | bb->unacked_exist = 0; |
2230dfe4 N |
8750 | } |
8751 | write_sequnlock_irq(&bb->lock); | |
8752 | } | |
8753 | EXPORT_SYMBOL_GPL(md_ack_all_badblocks); | |
8754 | ||
16c791a5 N |
8755 | /* sysfs access to bad-blocks list. |
8756 | * We present two files. | |
8757 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
8758 | * are recorded as bad. The list is truncated to fit within | |
8759 | * the one-page limit of sysfs. | |
8760 | * Writing "sector length" to this file adds an acknowledged | |
8761 | * bad block list. | |
8762 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
8763 | * been acknowledged. Writing to this file adds bad blocks | |
8764 | * without acknowledging them. This is largely for testing. | |
8765 | */ | |
8766 | ||
8767 | static ssize_t | |
8768 | badblocks_show(struct badblocks *bb, char *page, int unack) | |
8769 | { | |
8770 | size_t len; | |
8771 | int i; | |
8772 | u64 *p = bb->page; | |
8773 | unsigned seq; | |
8774 | ||
8775 | if (bb->shift < 0) | |
8776 | return 0; | |
8777 | ||
8778 | retry: | |
8779 | seq = read_seqbegin(&bb->lock); | |
8780 | ||
8781 | len = 0; | |
8782 | i = 0; | |
8783 | ||
8784 | while (len < PAGE_SIZE && i < bb->count) { | |
8785 | sector_t s = BB_OFFSET(p[i]); | |
8786 | unsigned int length = BB_LEN(p[i]); | |
8787 | int ack = BB_ACK(p[i]); | |
8788 | i++; | |
8789 | ||
8790 | if (unack && ack) | |
8791 | continue; | |
8792 | ||
8793 | len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n", | |
8794 | (unsigned long long)s << bb->shift, | |
8795 | length << bb->shift); | |
8796 | } | |
de393cde N |
8797 | if (unack && len == 0) |
8798 | bb->unacked_exist = 0; | |
16c791a5 N |
8799 | |
8800 | if (read_seqretry(&bb->lock, seq)) | |
8801 | goto retry; | |
8802 | ||
8803 | return len; | |
8804 | } | |
8805 | ||
8806 | #define DO_DEBUG 1 | |
8807 | ||
8808 | static ssize_t | |
8809 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack) | |
8810 | { | |
8811 | unsigned long long sector; | |
8812 | int length; | |
8813 | char newline; | |
8814 | #ifdef DO_DEBUG | |
8815 | /* Allow clearing via sysfs *only* for testing/debugging. | |
8816 | * Normally only a successful write may clear a badblock | |
8817 | */ | |
8818 | int clear = 0; | |
8819 | if (page[0] == '-') { | |
8820 | clear = 1; | |
8821 | page++; | |
8822 | } | |
8823 | #endif /* DO_DEBUG */ | |
8824 | ||
8825 | switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) { | |
8826 | case 3: | |
8827 | if (newline != '\n') | |
8828 | return -EINVAL; | |
8829 | case 2: | |
8830 | if (length <= 0) | |
8831 | return -EINVAL; | |
8832 | break; | |
8833 | default: | |
8834 | return -EINVAL; | |
8835 | } | |
8836 | ||
8837 | #ifdef DO_DEBUG | |
8838 | if (clear) { | |
8839 | md_clear_badblocks(bb, sector, length); | |
8840 | return len; | |
8841 | } | |
8842 | #endif /* DO_DEBUG */ | |
8843 | if (md_set_badblocks(bb, sector, length, !unack)) | |
8844 | return len; | |
8845 | else | |
8846 | return -ENOSPC; | |
8847 | } | |
8848 | ||
75c96f85 AB |
8849 | static int md_notify_reboot(struct notifier_block *this, |
8850 | unsigned long code, void *x) | |
1da177e4 LT |
8851 | { |
8852 | struct list_head *tmp; | |
fd01b88c | 8853 | struct mddev *mddev; |
2dba6a91 | 8854 | int need_delay = 0; |
1da177e4 | 8855 | |
c744a65c N |
8856 | for_each_mddev(mddev, tmp) { |
8857 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8858 | if (mddev->pers) |
8859 | __md_stop_writes(mddev); | |
0f62fb22 N |
8860 | if (mddev->persistent) |
8861 | mddev->safemode = 2; | |
c744a65c | 8862 | mddev_unlock(mddev); |
2dba6a91 | 8863 | } |
c744a65c | 8864 | need_delay = 1; |
1da177e4 | 8865 | } |
c744a65c N |
8866 | /* |
8867 | * certain more exotic SCSI devices are known to be | |
8868 | * volatile wrt too early system reboots. While the | |
8869 | * right place to handle this issue is the given | |
8870 | * driver, we do want to have a safe RAID driver ... | |
8871 | */ | |
8872 | if (need_delay) | |
8873 | mdelay(1000*1); | |
8874 | ||
1da177e4 LT |
8875 | return NOTIFY_DONE; |
8876 | } | |
8877 | ||
75c96f85 | 8878 | static struct notifier_block md_notifier = { |
1da177e4 LT |
8879 | .notifier_call = md_notify_reboot, |
8880 | .next = NULL, | |
8881 | .priority = INT_MAX, /* before any real devices */ | |
8882 | }; | |
8883 | ||
8884 | static void md_geninit(void) | |
8885 | { | |
36a4e1fe | 8886 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 8887 | |
c7705f34 | 8888 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
8889 | } |
8890 | ||
75c96f85 | 8891 | static int __init md_init(void) |
1da177e4 | 8892 | { |
e804ac78 TH |
8893 | int ret = -ENOMEM; |
8894 | ||
ada609ee | 8895 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
8896 | if (!md_wq) |
8897 | goto err_wq; | |
8898 | ||
8899 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
8900 | if (!md_misc_wq) | |
8901 | goto err_misc_wq; | |
8902 | ||
8903 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
8904 | goto err_md; | |
8905 | ||
8906 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
8907 | goto err_mdp; | |
8908 | mdp_major = ret; | |
8909 | ||
af5628f0 | 8910 | blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE, |
e8703fe1 N |
8911 | md_probe, NULL, NULL); |
8912 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
8913 | md_probe, NULL, NULL); |
8914 | ||
1da177e4 | 8915 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 8916 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
8917 | |
8918 | md_geninit(); | |
d710e138 | 8919 | return 0; |
1da177e4 | 8920 | |
e804ac78 TH |
8921 | err_mdp: |
8922 | unregister_blkdev(MD_MAJOR, "md"); | |
8923 | err_md: | |
8924 | destroy_workqueue(md_misc_wq); | |
8925 | err_misc_wq: | |
8926 | destroy_workqueue(md_wq); | |
8927 | err_wq: | |
8928 | return ret; | |
8929 | } | |
1da177e4 | 8930 | |
1d7e3e96 GR |
8931 | void md_reload_sb(struct mddev *mddev) |
8932 | { | |
8933 | struct md_rdev *rdev, *tmp; | |
8934 | ||
8935 | rdev_for_each_safe(rdev, tmp, mddev) { | |
8936 | rdev->sb_loaded = 0; | |
8937 | ClearPageUptodate(rdev->sb_page); | |
8938 | } | |
8939 | mddev->raid_disks = 0; | |
8940 | analyze_sbs(mddev); | |
8941 | rdev_for_each_safe(rdev, tmp, mddev) { | |
8942 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); | |
8943 | /* since we don't write to faulty devices, we figure out if the | |
8944 | * disk is faulty by comparing events | |
8945 | */ | |
8946 | if (mddev->events > sb->events) | |
8947 | set_bit(Faulty, &rdev->flags); | |
8948 | } | |
8949 | ||
8950 | } | |
8951 | EXPORT_SYMBOL(md_reload_sb); | |
8952 | ||
1da177e4 LT |
8953 | #ifndef MODULE |
8954 | ||
8955 | /* | |
8956 | * Searches all registered partitions for autorun RAID arrays | |
8957 | * at boot time. | |
8958 | */ | |
4d936ec1 ME |
8959 | |
8960 | static LIST_HEAD(all_detected_devices); | |
8961 | struct detected_devices_node { | |
8962 | struct list_head list; | |
8963 | dev_t dev; | |
8964 | }; | |
1da177e4 LT |
8965 | |
8966 | void md_autodetect_dev(dev_t dev) | |
8967 | { | |
4d936ec1 ME |
8968 | struct detected_devices_node *node_detected_dev; |
8969 | ||
8970 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
8971 | if (node_detected_dev) { | |
8972 | node_detected_dev->dev = dev; | |
8973 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
8974 | } else { | |
8975 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
8976 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
8977 | } | |
1da177e4 LT |
8978 | } |
8979 | ||
1da177e4 LT |
8980 | static void autostart_arrays(int part) |
8981 | { | |
3cb03002 | 8982 | struct md_rdev *rdev; |
4d936ec1 ME |
8983 | struct detected_devices_node *node_detected_dev; |
8984 | dev_t dev; | |
8985 | int i_scanned, i_passed; | |
1da177e4 | 8986 | |
4d936ec1 ME |
8987 | i_scanned = 0; |
8988 | i_passed = 0; | |
1da177e4 | 8989 | |
4d936ec1 | 8990 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 8991 | |
4d936ec1 ME |
8992 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
8993 | i_scanned++; | |
8994 | node_detected_dev = list_entry(all_detected_devices.next, | |
8995 | struct detected_devices_node, list); | |
8996 | list_del(&node_detected_dev->list); | |
8997 | dev = node_detected_dev->dev; | |
8998 | kfree(node_detected_dev); | |
df968c4e | 8999 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
9000 | if (IS_ERR(rdev)) |
9001 | continue; | |
9002 | ||
403df478 | 9003 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 9004 | continue; |
403df478 | 9005 | |
d0fae18f | 9006 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 9007 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 9008 | i_passed++; |
1da177e4 | 9009 | } |
4d936ec1 ME |
9010 | |
9011 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
9012 | i_scanned, i_passed); | |
1da177e4 LT |
9013 | |
9014 | autorun_devices(part); | |
9015 | } | |
9016 | ||
fdee8ae4 | 9017 | #endif /* !MODULE */ |
1da177e4 LT |
9018 | |
9019 | static __exit void md_exit(void) | |
9020 | { | |
fd01b88c | 9021 | struct mddev *mddev; |
1da177e4 | 9022 | struct list_head *tmp; |
e2f23b60 | 9023 | int delay = 1; |
8ab5e4c1 | 9024 | |
af5628f0 | 9025 | blk_unregister_region(MKDEV(MD_MAJOR,0), 512); |
e8703fe1 | 9026 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 9027 | |
3dbd8c2e | 9028 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
9029 | unregister_blkdev(mdp_major, "mdp"); |
9030 | unregister_reboot_notifier(&md_notifier); | |
9031 | unregister_sysctl_table(raid_table_header); | |
e2f23b60 N |
9032 | |
9033 | /* We cannot unload the modules while some process is | |
9034 | * waiting for us in select() or poll() - wake them up | |
9035 | */ | |
9036 | md_unloading = 1; | |
9037 | while (waitqueue_active(&md_event_waiters)) { | |
9038 | /* not safe to leave yet */ | |
9039 | wake_up(&md_event_waiters); | |
9040 | msleep(delay); | |
9041 | delay += delay; | |
9042 | } | |
1da177e4 | 9043 | remove_proc_entry("mdstat", NULL); |
e2f23b60 | 9044 | |
29ac4aa3 | 9045 | for_each_mddev(mddev, tmp) { |
1da177e4 | 9046 | export_array(mddev); |
d3374825 | 9047 | mddev->hold_active = 0; |
1da177e4 | 9048 | } |
e804ac78 TH |
9049 | destroy_workqueue(md_misc_wq); |
9050 | destroy_workqueue(md_wq); | |
1da177e4 LT |
9051 | } |
9052 | ||
685784aa | 9053 | subsys_initcall(md_init); |
1da177e4 LT |
9054 | module_exit(md_exit) |
9055 | ||
f91de92e N |
9056 | static int get_ro(char *buffer, struct kernel_param *kp) |
9057 | { | |
9058 | return sprintf(buffer, "%d", start_readonly); | |
9059 | } | |
9060 | static int set_ro(const char *val, struct kernel_param *kp) | |
9061 | { | |
4c9309c0 | 9062 | return kstrtouint(val, 10, (unsigned int *)&start_readonly); |
f91de92e N |
9063 | } |
9064 | ||
80ca3a44 N |
9065 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
9066 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
efeb53c0 | 9067 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 9068 | |
1da177e4 | 9069 | MODULE_LICENSE("GPL"); |
0efb9e61 | 9070 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 9071 | MODULE_ALIAS("md"); |
72008652 | 9072 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |