]>
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 | |
0ca69886 N |
260 | if (mddev == NULL || mddev->pers == NULL |
261 | || !mddev->ready) { | |
409c57f3 | 262 | bio_io_error(bio); |
5a7bbad2 | 263 | return; |
409c57f3 | 264 | } |
bbfa57c0 SR |
265 | if (mddev->ro == 1 && unlikely(rw == WRITE)) { |
266 | bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS); | |
267 | return; | |
268 | } | |
0ca69886 | 269 | smp_rmb(); /* Ensure implications of 'active' are visible */ |
409c57f3 | 270 | rcu_read_lock(); |
e9c7469b | 271 | if (mddev->suspended) { |
409c57f3 N |
272 | DEFINE_WAIT(__wait); |
273 | for (;;) { | |
274 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
275 | TASK_UNINTERRUPTIBLE); | |
e9c7469b | 276 | if (!mddev->suspended) |
409c57f3 N |
277 | break; |
278 | rcu_read_unlock(); | |
279 | schedule(); | |
280 | rcu_read_lock(); | |
281 | } | |
282 | finish_wait(&mddev->sb_wait, &__wait); | |
283 | } | |
284 | atomic_inc(&mddev->active_io); | |
285 | rcu_read_unlock(); | |
49077326 | 286 | |
e91ece55 CM |
287 | /* |
288 | * save the sectors now since our bio can | |
289 | * go away inside make_request | |
290 | */ | |
291 | sectors = bio_sectors(bio); | |
5a7bbad2 | 292 | mddev->pers->make_request(mddev, bio); |
49077326 | 293 | |
74672d06 GZ |
294 | cpu = part_stat_lock(); |
295 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
296 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); | |
297 | part_stat_unlock(); | |
49077326 | 298 | |
409c57f3 N |
299 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
300 | wake_up(&mddev->sb_wait); | |
409c57f3 N |
301 | } |
302 | ||
9e35b99c N |
303 | /* mddev_suspend makes sure no new requests are submitted |
304 | * to the device, and that any requests that have been submitted | |
305 | * are completely handled. | |
afa0f557 N |
306 | * Once mddev_detach() is called and completes, the module will be |
307 | * completely unused. | |
9e35b99c | 308 | */ |
fd01b88c | 309 | void mddev_suspend(struct mddev *mddev) |
409c57f3 N |
310 | { |
311 | BUG_ON(mddev->suspended); | |
312 | mddev->suspended = 1; | |
313 | synchronize_rcu(); | |
314 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
315 | mddev->pers->quiesce(mddev, 1); | |
0d9f4f13 JB |
316 | |
317 | del_timer_sync(&mddev->safemode_timer); | |
409c57f3 | 318 | } |
390ee602 | 319 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 320 | |
fd01b88c | 321 | void mddev_resume(struct mddev *mddev) |
409c57f3 N |
322 | { |
323 | mddev->suspended = 0; | |
324 | wake_up(&mddev->sb_wait); | |
325 | mddev->pers->quiesce(mddev, 0); | |
0fd018af | 326 | |
47525e59 | 327 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
0fd018af JB |
328 | md_wakeup_thread(mddev->thread); |
329 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 330 | } |
390ee602 | 331 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 332 | |
fd01b88c | 333 | int mddev_congested(struct mddev *mddev, int bits) |
3fa841d7 | 334 | { |
5c675f83 N |
335 | struct md_personality *pers = mddev->pers; |
336 | int ret = 0; | |
337 | ||
338 | rcu_read_lock(); | |
339 | if (mddev->suspended) | |
340 | ret = 1; | |
341 | else if (pers && pers->congested) | |
342 | ret = pers->congested(mddev, bits); | |
343 | rcu_read_unlock(); | |
344 | return ret; | |
345 | } | |
346 | EXPORT_SYMBOL_GPL(mddev_congested); | |
347 | static int md_congested(void *data, int bits) | |
348 | { | |
349 | struct mddev *mddev = data; | |
350 | return mddev_congested(mddev, bits); | |
3fa841d7 | 351 | } |
3fa841d7 | 352 | |
64590f45 N |
353 | static int md_mergeable_bvec(struct request_queue *q, |
354 | struct bvec_merge_data *bvm, | |
355 | struct bio_vec *biovec) | |
356 | { | |
357 | struct mddev *mddev = q->queuedata; | |
358 | int ret; | |
359 | rcu_read_lock(); | |
360 | if (mddev->suspended) { | |
361 | /* Must always allow one vec */ | |
362 | if (bvm->bi_size == 0) | |
363 | ret = biovec->bv_len; | |
364 | else | |
365 | ret = 0; | |
366 | } else { | |
367 | struct md_personality *pers = mddev->pers; | |
368 | if (pers && pers->mergeable_bvec) | |
369 | ret = pers->mergeable_bvec(mddev, bvm, biovec); | |
370 | else | |
371 | ret = biovec->bv_len; | |
372 | } | |
373 | rcu_read_unlock(); | |
374 | return ret; | |
375 | } | |
a2826aa9 | 376 | /* |
e9c7469b | 377 | * Generic flush handling for md |
a2826aa9 N |
378 | */ |
379 | ||
e9c7469b | 380 | static void md_end_flush(struct bio *bio, int err) |
a2826aa9 | 381 | { |
3cb03002 | 382 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 383 | struct mddev *mddev = rdev->mddev; |
a2826aa9 N |
384 | |
385 | rdev_dec_pending(rdev, mddev); | |
386 | ||
387 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 388 | /* The pre-request flush has finished */ |
e804ac78 | 389 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
390 | } |
391 | bio_put(bio); | |
392 | } | |
393 | ||
a7a07e69 N |
394 | static void md_submit_flush_data(struct work_struct *ws); |
395 | ||
a035fc3e | 396 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 397 | { |
fd01b88c | 398 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
3cb03002 | 399 | struct md_rdev *rdev; |
a2826aa9 | 400 | |
a7a07e69 N |
401 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
402 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 | 403 | rcu_read_lock(); |
dafb20fa | 404 | rdev_for_each_rcu(rdev, mddev) |
a2826aa9 N |
405 | if (rdev->raid_disk >= 0 && |
406 | !test_bit(Faulty, &rdev->flags)) { | |
407 | /* Take two references, one is dropped | |
408 | * when request finishes, one after | |
409 | * we reclaim rcu_read_lock | |
410 | */ | |
411 | struct bio *bi; | |
412 | atomic_inc(&rdev->nr_pending); | |
413 | atomic_inc(&rdev->nr_pending); | |
414 | rcu_read_unlock(); | |
b5e1b8ce | 415 | bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); |
e9c7469b | 416 | bi->bi_end_io = md_end_flush; |
a2826aa9 N |
417 | bi->bi_private = rdev; |
418 | bi->bi_bdev = rdev->bdev; | |
419 | atomic_inc(&mddev->flush_pending); | |
e9c7469b | 420 | submit_bio(WRITE_FLUSH, bi); |
a2826aa9 N |
421 | rcu_read_lock(); |
422 | rdev_dec_pending(rdev, mddev); | |
423 | } | |
424 | rcu_read_unlock(); | |
a7a07e69 N |
425 | if (atomic_dec_and_test(&mddev->flush_pending)) |
426 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
427 | } |
428 | ||
e9c7469b | 429 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 430 | { |
fd01b88c | 431 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
e9c7469b | 432 | struct bio *bio = mddev->flush_bio; |
a2826aa9 | 433 | |
4f024f37 | 434 | if (bio->bi_iter.bi_size == 0) |
a2826aa9 N |
435 | /* an empty barrier - all done */ |
436 | bio_endio(bio, 0); | |
437 | else { | |
e9c7469b | 438 | bio->bi_rw &= ~REQ_FLUSH; |
5a7bbad2 | 439 | mddev->pers->make_request(mddev, bio); |
a2826aa9 | 440 | } |
2b74e12e N |
441 | |
442 | mddev->flush_bio = NULL; | |
443 | wake_up(&mddev->sb_wait); | |
a2826aa9 N |
444 | } |
445 | ||
fd01b88c | 446 | void md_flush_request(struct mddev *mddev, struct bio *bio) |
a2826aa9 | 447 | { |
85572d7c | 448 | spin_lock_irq(&mddev->lock); |
a2826aa9 | 449 | wait_event_lock_irq(mddev->sb_wait, |
e9c7469b | 450 | !mddev->flush_bio, |
85572d7c | 451 | mddev->lock); |
e9c7469b | 452 | mddev->flush_bio = bio; |
85572d7c | 453 | spin_unlock_irq(&mddev->lock); |
a2826aa9 | 454 | |
a035fc3e N |
455 | INIT_WORK(&mddev->flush_work, submit_flushes); |
456 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 457 | } |
e9c7469b | 458 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 459 | |
74018dc3 | 460 | void md_unplug(struct blk_plug_cb *cb, bool from_schedule) |
97658cdd | 461 | { |
9cbb1750 N |
462 | struct mddev *mddev = cb->data; |
463 | md_wakeup_thread(mddev->thread); | |
464 | kfree(cb); | |
97658cdd | 465 | } |
9cbb1750 | 466 | EXPORT_SYMBOL(md_unplug); |
2ac87401 | 467 | |
fd01b88c | 468 | static inline struct mddev *mddev_get(struct mddev *mddev) |
1da177e4 LT |
469 | { |
470 | atomic_inc(&mddev->active); | |
471 | return mddev; | |
472 | } | |
473 | ||
5fd3a17e | 474 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 475 | |
fd01b88c | 476 | static void mddev_put(struct mddev *mddev) |
1da177e4 | 477 | { |
a167f663 N |
478 | struct bio_set *bs = NULL; |
479 | ||
1da177e4 LT |
480 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
481 | return; | |
d3374825 | 482 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
483 | mddev->ctime == 0 && !mddev->hold_active) { |
484 | /* Array is not configured at all, and not held active, | |
485 | * so destroy it */ | |
af8a2434 | 486 | list_del_init(&mddev->all_mddevs); |
a167f663 N |
487 | bs = mddev->bio_set; |
488 | mddev->bio_set = NULL; | |
d3374825 | 489 | if (mddev->gendisk) { |
e804ac78 TH |
490 | /* We did a probe so need to clean up. Call |
491 | * queue_work inside the spinlock so that | |
492 | * flush_workqueue() after mddev_find will | |
493 | * succeed in waiting for the work to be done. | |
d3374825 N |
494 | */ |
495 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 496 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
497 | } else |
498 | kfree(mddev); | |
499 | } | |
500 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
501 | if (bs) |
502 | bioset_free(bs); | |
1da177e4 LT |
503 | } |
504 | ||
fd01b88c | 505 | void mddev_init(struct mddev *mddev) |
fafd7fb0 N |
506 | { |
507 | mutex_init(&mddev->open_mutex); | |
508 | mutex_init(&mddev->reconfig_mutex); | |
509 | mutex_init(&mddev->bitmap_info.mutex); | |
510 | INIT_LIST_HEAD(&mddev->disks); | |
511 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
512 | init_timer(&mddev->safemode_timer); | |
513 | atomic_set(&mddev->active, 1); | |
514 | atomic_set(&mddev->openers, 0); | |
515 | atomic_set(&mddev->active_io, 0); | |
85572d7c | 516 | spin_lock_init(&mddev->lock); |
fafd7fb0 N |
517 | atomic_set(&mddev->flush_pending, 0); |
518 | init_waitqueue_head(&mddev->sb_wait); | |
519 | init_waitqueue_head(&mddev->recovery_wait); | |
520 | mddev->reshape_position = MaxSector; | |
2c810cdd | 521 | mddev->reshape_backwards = 0; |
c4a39551 | 522 | mddev->last_sync_action = "none"; |
fafd7fb0 N |
523 | mddev->resync_min = 0; |
524 | mddev->resync_max = MaxSector; | |
525 | mddev->level = LEVEL_NONE; | |
526 | } | |
390ee602 | 527 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 528 | |
f72ffdd6 | 529 | static struct mddev *mddev_find(dev_t unit) |
1da177e4 | 530 | { |
fd01b88c | 531 | struct mddev *mddev, *new = NULL; |
1da177e4 | 532 | |
8f5f02c4 N |
533 | if (unit && MAJOR(unit) != MD_MAJOR) |
534 | unit &= ~((1<<MdpMinorShift)-1); | |
535 | ||
1da177e4 LT |
536 | retry: |
537 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
538 | |
539 | if (unit) { | |
540 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
541 | if (mddev->unit == unit) { | |
542 | mddev_get(mddev); | |
543 | spin_unlock(&all_mddevs_lock); | |
544 | kfree(new); | |
545 | return mddev; | |
546 | } | |
547 | ||
548 | if (new) { | |
549 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 550 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
551 | new->hold_active = UNTIL_IOCTL; |
552 | return new; | |
1da177e4 | 553 | } |
efeb53c0 N |
554 | } else if (new) { |
555 | /* find an unused unit number */ | |
556 | static int next_minor = 512; | |
557 | int start = next_minor; | |
558 | int is_free = 0; | |
559 | int dev = 0; | |
560 | while (!is_free) { | |
561 | dev = MKDEV(MD_MAJOR, next_minor); | |
562 | next_minor++; | |
563 | if (next_minor > MINORMASK) | |
564 | next_minor = 0; | |
565 | if (next_minor == start) { | |
566 | /* Oh dear, all in use. */ | |
567 | spin_unlock(&all_mddevs_lock); | |
568 | kfree(new); | |
569 | return NULL; | |
570 | } | |
f72ffdd6 | 571 | |
efeb53c0 N |
572 | is_free = 1; |
573 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
574 | if (mddev->unit == dev) { | |
575 | is_free = 0; | |
576 | break; | |
577 | } | |
578 | } | |
579 | new->unit = dev; | |
580 | new->md_minor = MINOR(dev); | |
581 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
582 | list_add(&new->all_mddevs, &all_mddevs); |
583 | spin_unlock(&all_mddevs_lock); | |
584 | return new; | |
585 | } | |
586 | spin_unlock(&all_mddevs_lock); | |
587 | ||
9ffae0cf | 588 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
589 | if (!new) |
590 | return NULL; | |
591 | ||
1da177e4 LT |
592 | new->unit = unit; |
593 | if (MAJOR(unit) == MD_MAJOR) | |
594 | new->md_minor = MINOR(unit); | |
595 | else | |
596 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
597 | ||
fafd7fb0 | 598 | mddev_init(new); |
1da177e4 | 599 | |
1da177e4 LT |
600 | goto retry; |
601 | } | |
602 | ||
b6eb127d N |
603 | static struct attribute_group md_redundancy_group; |
604 | ||
5c47daf6 | 605 | void mddev_unlock(struct mddev *mddev) |
1da177e4 | 606 | { |
a64c876f | 607 | if (mddev->to_remove) { |
b6eb127d N |
608 | /* These cannot be removed under reconfig_mutex as |
609 | * an access to the files will try to take reconfig_mutex | |
610 | * while holding the file unremovable, which leads to | |
611 | * a deadlock. | |
bb4f1e9d N |
612 | * So hold set sysfs_active while the remove in happeing, |
613 | * and anything else which might set ->to_remove or my | |
614 | * otherwise change the sysfs namespace will fail with | |
615 | * -EBUSY if sysfs_active is still set. | |
616 | * We set sysfs_active under reconfig_mutex and elsewhere | |
617 | * test it under the same mutex to ensure its correct value | |
618 | * is seen. | |
b6eb127d | 619 | */ |
a64c876f N |
620 | struct attribute_group *to_remove = mddev->to_remove; |
621 | mddev->to_remove = NULL; | |
bb4f1e9d | 622 | mddev->sysfs_active = 1; |
b6eb127d N |
623 | mutex_unlock(&mddev->reconfig_mutex); |
624 | ||
00bcb4ac N |
625 | if (mddev->kobj.sd) { |
626 | if (to_remove != &md_redundancy_group) | |
627 | sysfs_remove_group(&mddev->kobj, to_remove); | |
628 | if (mddev->pers == NULL || | |
629 | mddev->pers->sync_request == NULL) { | |
630 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
631 | if (mddev->sysfs_action) | |
632 | sysfs_put(mddev->sysfs_action); | |
633 | mddev->sysfs_action = NULL; | |
634 | } | |
a64c876f | 635 | } |
bb4f1e9d | 636 | mddev->sysfs_active = 0; |
b6eb127d N |
637 | } else |
638 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 639 | |
751e67ca CD |
640 | /* As we've dropped the mutex we need a spinlock to |
641 | * make sure the thread doesn't disappear | |
01f96c0a N |
642 | */ |
643 | spin_lock(&pers_lock); | |
005eca5e | 644 | md_wakeup_thread(mddev->thread); |
01f96c0a | 645 | spin_unlock(&pers_lock); |
1da177e4 | 646 | } |
5c47daf6 | 647 | EXPORT_SYMBOL_GPL(mddev_unlock); |
1da177e4 | 648 | |
57d051dc | 649 | struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr) |
1ca69c4b N |
650 | { |
651 | struct md_rdev *rdev; | |
652 | ||
653 | rdev_for_each_rcu(rdev, mddev) | |
654 | if (rdev->desc_nr == nr) | |
655 | return rdev; | |
656 | ||
657 | return NULL; | |
658 | } | |
57d051dc | 659 | EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu); |
1ca69c4b N |
660 | |
661 | static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev) | |
1da177e4 | 662 | { |
3cb03002 | 663 | struct md_rdev *rdev; |
1da177e4 | 664 | |
dafb20fa | 665 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
666 | if (rdev->bdev->bd_dev == dev) |
667 | return rdev; | |
159ec1fc | 668 | |
1da177e4 LT |
669 | return NULL; |
670 | } | |
671 | ||
1ca69c4b N |
672 | static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev) |
673 | { | |
674 | struct md_rdev *rdev; | |
675 | ||
676 | rdev_for_each_rcu(rdev, mddev) | |
677 | if (rdev->bdev->bd_dev == dev) | |
678 | return rdev; | |
679 | ||
680 | return NULL; | |
681 | } | |
682 | ||
84fc4b56 | 683 | static struct md_personality *find_pers(int level, char *clevel) |
2604b703 | 684 | { |
84fc4b56 | 685 | struct md_personality *pers; |
d9d166c2 N |
686 | list_for_each_entry(pers, &pers_list, list) { |
687 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 688 | return pers; |
d9d166c2 N |
689 | if (strcmp(pers->name, clevel)==0) |
690 | return pers; | |
691 | } | |
2604b703 N |
692 | return NULL; |
693 | } | |
694 | ||
b73df2d3 | 695 | /* return the offset of the super block in 512byte sectors */ |
3cb03002 | 696 | static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) |
1da177e4 | 697 | { |
57b2caa3 | 698 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 699 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
700 | } |
701 | ||
f72ffdd6 | 702 | static int alloc_disk_sb(struct md_rdev *rdev) |
1da177e4 | 703 | { |
1da177e4 LT |
704 | rdev->sb_page = alloc_page(GFP_KERNEL); |
705 | if (!rdev->sb_page) { | |
706 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 707 | return -ENOMEM; |
1da177e4 LT |
708 | } |
709 | ||
710 | return 0; | |
711 | } | |
712 | ||
545c8795 | 713 | void md_rdev_clear(struct md_rdev *rdev) |
1da177e4 LT |
714 | { |
715 | if (rdev->sb_page) { | |
2d1f3b5d | 716 | put_page(rdev->sb_page); |
1da177e4 LT |
717 | rdev->sb_loaded = 0; |
718 | rdev->sb_page = NULL; | |
0f420358 | 719 | rdev->sb_start = 0; |
dd8ac336 | 720 | rdev->sectors = 0; |
1da177e4 | 721 | } |
2699b672 N |
722 | if (rdev->bb_page) { |
723 | put_page(rdev->bb_page); | |
724 | rdev->bb_page = NULL; | |
725 | } | |
4fa2f327 N |
726 | kfree(rdev->badblocks.page); |
727 | rdev->badblocks.page = NULL; | |
1da177e4 | 728 | } |
545c8795 | 729 | EXPORT_SYMBOL_GPL(md_rdev_clear); |
1da177e4 | 730 | |
6712ecf8 | 731 | static void super_written(struct bio *bio, int error) |
7bfa19f2 | 732 | { |
3cb03002 | 733 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 734 | struct mddev *mddev = rdev->mddev; |
7bfa19f2 | 735 | |
3a0f5bbb N |
736 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
737 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
738 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
739 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 740 | md_error(mddev, rdev); |
3a0f5bbb | 741 | } |
7bfa19f2 | 742 | |
a9701a30 N |
743 | if (atomic_dec_and_test(&mddev->pending_writes)) |
744 | wake_up(&mddev->sb_wait); | |
f8b58edf | 745 | bio_put(bio); |
7bfa19f2 N |
746 | } |
747 | ||
fd01b88c | 748 | void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
7bfa19f2 N |
749 | sector_t sector, int size, struct page *page) |
750 | { | |
751 | /* write first size bytes of page to sector of rdev | |
752 | * Increment mddev->pending_writes before returning | |
753 | * and decrement it on completion, waking up sb_wait | |
754 | * if zero is reached. | |
755 | * If an error occurred, call md_error | |
756 | */ | |
a167f663 | 757 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); |
7bfa19f2 | 758 | |
a6ff7e08 | 759 | bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; |
4f024f37 | 760 | bio->bi_iter.bi_sector = sector; |
7bfa19f2 N |
761 | bio_add_page(bio, page, size, 0); |
762 | bio->bi_private = rdev; | |
763 | bio->bi_end_io = super_written; | |
a9701a30 | 764 | |
7bfa19f2 | 765 | atomic_inc(&mddev->pending_writes); |
a5bf4df0 | 766 | submit_bio(WRITE_FLUSH_FUA, bio); |
a9701a30 N |
767 | } |
768 | ||
fd01b88c | 769 | void md_super_wait(struct mddev *mddev) |
a9701a30 | 770 | { |
e9c7469b | 771 | /* wait for all superblock writes that were scheduled to complete */ |
1967cd56 | 772 | wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); |
7bfa19f2 N |
773 | } |
774 | ||
3cb03002 | 775 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
ccebd4c4 | 776 | struct page *page, int rw, bool metadata_op) |
1da177e4 | 777 | { |
a167f663 | 778 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
779 | int ret; |
780 | ||
a6ff7e08 JB |
781 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
782 | rdev->meta_bdev : rdev->bdev; | |
ccebd4c4 | 783 | if (metadata_op) |
4f024f37 | 784 | bio->bi_iter.bi_sector = sector + rdev->sb_start; |
1fdd6fc9 N |
785 | else if (rdev->mddev->reshape_position != MaxSector && |
786 | (rdev->mddev->reshape_backwards == | |
787 | (sector >= rdev->mddev->reshape_position))) | |
4f024f37 | 788 | bio->bi_iter.bi_sector = sector + rdev->new_data_offset; |
ccebd4c4 | 789 | else |
4f024f37 | 790 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
1da177e4 | 791 | bio_add_page(bio, page, size, 0); |
c170bbb4 | 792 | submit_bio_wait(rw, bio); |
1da177e4 LT |
793 | |
794 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
795 | bio_put(bio); | |
796 | return ret; | |
797 | } | |
a8745db2 | 798 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 799 | |
f72ffdd6 | 800 | static int read_disk_sb(struct md_rdev *rdev, int size) |
1da177e4 LT |
801 | { |
802 | char b[BDEVNAME_SIZE]; | |
403df478 | 803 | |
1da177e4 LT |
804 | if (rdev->sb_loaded) |
805 | return 0; | |
806 | ||
ccebd4c4 | 807 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) |
1da177e4 LT |
808 | goto fail; |
809 | rdev->sb_loaded = 1; | |
810 | return 0; | |
811 | ||
812 | fail: | |
813 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
814 | bdevname(rdev->bdev,b)); | |
815 | return -EINVAL; | |
816 | } | |
817 | ||
818 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
819 | { | |
f72ffdd6 | 820 | return sb1->set_uuid0 == sb2->set_uuid0 && |
05710466 AN |
821 | sb1->set_uuid1 == sb2->set_uuid1 && |
822 | sb1->set_uuid2 == sb2->set_uuid2 && | |
823 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
824 | } |
825 | ||
1da177e4 LT |
826 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
827 | { | |
828 | int ret; | |
829 | mdp_super_t *tmp1, *tmp2; | |
830 | ||
831 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
832 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
833 | ||
834 | if (!tmp1 || !tmp2) { | |
835 | ret = 0; | |
35020f1a | 836 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
837 | goto abort; |
838 | } | |
839 | ||
840 | *tmp1 = *sb1; | |
841 | *tmp2 = *sb2; | |
842 | ||
843 | /* | |
844 | * nr_disks is not constant | |
845 | */ | |
846 | tmp1->nr_disks = 0; | |
847 | tmp2->nr_disks = 0; | |
848 | ||
ce0c8e05 | 849 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 850 | abort: |
990a8baf JJ |
851 | kfree(tmp1); |
852 | kfree(tmp2); | |
1da177e4 LT |
853 | return ret; |
854 | } | |
855 | ||
4d167f09 N |
856 | static u32 md_csum_fold(u32 csum) |
857 | { | |
858 | csum = (csum & 0xffff) + (csum >> 16); | |
859 | return (csum & 0xffff) + (csum >> 16); | |
860 | } | |
861 | ||
f72ffdd6 | 862 | static unsigned int calc_sb_csum(mdp_super_t *sb) |
1da177e4 | 863 | { |
4d167f09 N |
864 | u64 newcsum = 0; |
865 | u32 *sb32 = (u32*)sb; | |
866 | int i; | |
1da177e4 LT |
867 | unsigned int disk_csum, csum; |
868 | ||
869 | disk_csum = sb->sb_csum; | |
870 | sb->sb_csum = 0; | |
4d167f09 N |
871 | |
872 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
873 | newcsum += sb32[i]; | |
874 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
875 | ||
4d167f09 N |
876 | #ifdef CONFIG_ALPHA |
877 | /* This used to use csum_partial, which was wrong for several | |
878 | * reasons including that different results are returned on | |
879 | * different architectures. It isn't critical that we get exactly | |
880 | * the same return value as before (we always csum_fold before | |
881 | * testing, and that removes any differences). However as we | |
882 | * know that csum_partial always returned a 16bit value on | |
883 | * alphas, do a fold to maximise conformity to previous behaviour. | |
884 | */ | |
885 | sb->sb_csum = md_csum_fold(disk_csum); | |
886 | #else | |
1da177e4 | 887 | sb->sb_csum = disk_csum; |
4d167f09 | 888 | #endif |
1da177e4 LT |
889 | return csum; |
890 | } | |
891 | ||
1da177e4 LT |
892 | /* |
893 | * Handle superblock details. | |
894 | * We want to be able to handle multiple superblock formats | |
895 | * so we have a common interface to them all, and an array of | |
896 | * different handlers. | |
897 | * We rely on user-space to write the initial superblock, and support | |
898 | * reading and updating of superblocks. | |
899 | * Interface methods are: | |
3cb03002 | 900 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
901 | * loads and validates a superblock on dev. |
902 | * if refdev != NULL, compare superblocks on both devices | |
903 | * Return: | |
904 | * 0 - dev has a superblock that is compatible with refdev | |
905 | * 1 - dev has a superblock that is compatible and newer than refdev | |
906 | * so dev should be used as the refdev in future | |
907 | * -EINVAL superblock incompatible or invalid | |
908 | * -othererror e.g. -EIO | |
909 | * | |
fd01b88c | 910 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
911 | * Verify that dev is acceptable into mddev. |
912 | * The first time, mddev->raid_disks will be 0, and data from | |
913 | * dev should be merged in. Subsequent calls check that dev | |
914 | * is new enough. Return 0 or -EINVAL | |
915 | * | |
fd01b88c | 916 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
917 | * Update the superblock for rdev with data in mddev |
918 | * This does not write to disc. | |
919 | * | |
920 | */ | |
921 | ||
922 | struct super_type { | |
0cd17fec CW |
923 | char *name; |
924 | struct module *owner; | |
c6563a8c N |
925 | int (*load_super)(struct md_rdev *rdev, |
926 | struct md_rdev *refdev, | |
0cd17fec | 927 | int minor_version); |
c6563a8c N |
928 | int (*validate_super)(struct mddev *mddev, |
929 | struct md_rdev *rdev); | |
930 | void (*sync_super)(struct mddev *mddev, | |
931 | struct md_rdev *rdev); | |
3cb03002 | 932 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 933 | sector_t num_sectors); |
c6563a8c N |
934 | int (*allow_new_offset)(struct md_rdev *rdev, |
935 | unsigned long long new_offset); | |
1da177e4 LT |
936 | }; |
937 | ||
0894cc30 AN |
938 | /* |
939 | * Check that the given mddev has no bitmap. | |
940 | * | |
941 | * This function is called from the run method of all personalities that do not | |
942 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
943 | * has a bitmap. Otherwise, it returns 0. | |
944 | * | |
945 | */ | |
fd01b88c | 946 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 947 | { |
c3d9714e | 948 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
949 | return 0; |
950 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
951 | mdname(mddev), mddev->pers->name); | |
952 | return 1; | |
953 | } | |
954 | EXPORT_SYMBOL(md_check_no_bitmap); | |
955 | ||
1da177e4 | 956 | /* |
f72ffdd6 | 957 | * load_super for 0.90.0 |
1da177e4 | 958 | */ |
3cb03002 | 959 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
960 | { |
961 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
962 | mdp_super_t *sb; | |
963 | int ret; | |
1da177e4 LT |
964 | |
965 | /* | |
0f420358 | 966 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
967 | * it's at the end of the disk. |
968 | * | |
969 | * It also happens to be a multiple of 4Kb. | |
970 | */ | |
57b2caa3 | 971 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 972 | |
0002b271 | 973 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
974 | if (ret) return ret; |
975 | ||
976 | ret = -EINVAL; | |
977 | ||
978 | bdevname(rdev->bdev, b); | |
65a06f06 | 979 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
980 | |
981 | if (sb->md_magic != MD_SB_MAGIC) { | |
982 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
983 | b); | |
984 | goto abort; | |
985 | } | |
986 | ||
987 | if (sb->major_version != 0 || | |
f6705578 N |
988 | sb->minor_version < 90 || |
989 | sb->minor_version > 91) { | |
1da177e4 LT |
990 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
991 | sb->major_version, sb->minor_version, | |
992 | b); | |
993 | goto abort; | |
994 | } | |
995 | ||
996 | if (sb->raid_disks <= 0) | |
997 | goto abort; | |
998 | ||
4d167f09 | 999 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
1000 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
1001 | b); | |
1002 | goto abort; | |
1003 | } | |
1004 | ||
1005 | rdev->preferred_minor = sb->md_minor; | |
1006 | rdev->data_offset = 0; | |
c6563a8c | 1007 | rdev->new_data_offset = 0; |
0002b271 | 1008 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 1009 | rdev->badblocks.shift = -1; |
1da177e4 LT |
1010 | |
1011 | if (sb->level == LEVEL_MULTIPATH) | |
1012 | rdev->desc_nr = -1; | |
1013 | else | |
1014 | rdev->desc_nr = sb->this_disk.number; | |
1015 | ||
9a7b2b0f | 1016 | if (!refdev) { |
1da177e4 | 1017 | ret = 1; |
9a7b2b0f | 1018 | } else { |
1da177e4 | 1019 | __u64 ev1, ev2; |
65a06f06 | 1020 | mdp_super_t *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1021 | if (!uuid_equal(refsb, sb)) { |
1022 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
1023 | b, bdevname(refdev->bdev,b2)); | |
1024 | goto abort; | |
1025 | } | |
1026 | if (!sb_equal(refsb, sb)) { | |
1027 | printk(KERN_WARNING "md: %s has same UUID" | |
1028 | " but different superblock to %s\n", | |
1029 | b, bdevname(refdev->bdev, b2)); | |
1030 | goto abort; | |
1031 | } | |
1032 | ev1 = md_event(sb); | |
1033 | ev2 = md_event(refsb); | |
1034 | if (ev1 > ev2) | |
1035 | ret = 1; | |
f72ffdd6 | 1036 | else |
1da177e4 LT |
1037 | ret = 0; |
1038 | } | |
8190e754 | 1039 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1040 | /* Limit to 4TB as metadata cannot record more than that. |
1041 | * (not needed for Linear and RAID0 as metadata doesn't | |
1042 | * record this size) | |
1043 | */ | |
1044 | if (rdev->sectors >= (2ULL << 32) && sb->level >= 1) | |
27a7b260 | 1045 | rdev->sectors = (2ULL << 32) - 2; |
1da177e4 | 1046 | |
27a7b260 | 1047 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1048 | /* "this cannot possibly happen" ... */ |
1049 | ret = -EINVAL; | |
1050 | ||
1da177e4 LT |
1051 | abort: |
1052 | return ret; | |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * validate_super for 0.90.0 | |
1057 | */ | |
fd01b88c | 1058 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1059 | { |
1060 | mdp_disk_t *desc; | |
65a06f06 | 1061 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1062 | __u64 ev1 = md_event(sb); |
1da177e4 | 1063 | |
41158c7e | 1064 | rdev->raid_disk = -1; |
c5d79adb N |
1065 | clear_bit(Faulty, &rdev->flags); |
1066 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1067 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1068 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1069 | |
1da177e4 LT |
1070 | if (mddev->raid_disks == 0) { |
1071 | mddev->major_version = 0; | |
1072 | mddev->minor_version = sb->minor_version; | |
1073 | mddev->patch_version = sb->patch_version; | |
e691063a | 1074 | mddev->external = 0; |
9d8f0363 | 1075 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1076 | mddev->ctime = sb->ctime; |
1077 | mddev->utime = sb->utime; | |
1078 | mddev->level = sb->level; | |
d9d166c2 | 1079 | mddev->clevel[0] = 0; |
1da177e4 LT |
1080 | mddev->layout = sb->layout; |
1081 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1082 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1083 | mddev->events = ev1; |
c3d9714e | 1084 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1085 | mddev->bitmap_info.space = 0; |
1086 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1087 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1088 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1089 | mddev->reshape_backwards = 0; |
1da177e4 | 1090 | |
f6705578 N |
1091 | if (mddev->minor_version >= 91) { |
1092 | mddev->reshape_position = sb->reshape_position; | |
1093 | mddev->delta_disks = sb->delta_disks; | |
1094 | mddev->new_level = sb->new_level; | |
1095 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1096 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1097 | if (mddev->delta_disks < 0) |
1098 | mddev->reshape_backwards = 1; | |
f6705578 N |
1099 | } else { |
1100 | mddev->reshape_position = MaxSector; | |
1101 | mddev->delta_disks = 0; | |
1102 | mddev->new_level = mddev->level; | |
1103 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1104 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1105 | } |
1106 | ||
1da177e4 LT |
1107 | if (sb->state & (1<<MD_SB_CLEAN)) |
1108 | mddev->recovery_cp = MaxSector; | |
1109 | else { | |
f72ffdd6 | 1110 | if (sb->events_hi == sb->cp_events_hi && |
1da177e4 LT |
1111 | sb->events_lo == sb->cp_events_lo) { |
1112 | mddev->recovery_cp = sb->recovery_cp; | |
1113 | } else | |
1114 | mddev->recovery_cp = 0; | |
1115 | } | |
1116 | ||
1117 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1118 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1119 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1120 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1121 | ||
1122 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1123 | |
1124 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1125 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1126 | mddev->bitmap_info.offset = |
1127 | mddev->bitmap_info.default_offset; | |
6409bb05 | 1128 | mddev->bitmap_info.space = |
c9ad020f | 1129 | mddev->bitmap_info.default_space; |
6409bb05 | 1130 | } |
a654b9d8 | 1131 | |
41158c7e | 1132 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1133 | /* Insist on good event counter while assembling, except |
1134 | * for spares (which don't need an event count) */ | |
1da177e4 | 1135 | ++ev1; |
be6800a7 N |
1136 | if (sb->disks[rdev->desc_nr].state & ( |
1137 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
f72ffdd6 | 1138 | if (ev1 < mddev->events) |
be6800a7 | 1139 | return -EINVAL; |
41158c7e N |
1140 | } else if (mddev->bitmap) { |
1141 | /* if adding to array with a bitmap, then we can accept an | |
1142 | * older device ... but not too old. | |
1143 | */ | |
41158c7e N |
1144 | if (ev1 < mddev->bitmap->events_cleared) |
1145 | return 0; | |
8313b8e5 N |
1146 | if (ev1 < mddev->events) |
1147 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1148 | } else { |
1149 | if (ev1 < mddev->events) | |
1150 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1151 | return 0; | |
1152 | } | |
41158c7e | 1153 | |
1da177e4 | 1154 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1155 | desc = sb->disks + rdev->desc_nr; |
1156 | ||
1157 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1158 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1159 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1160 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1161 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1162 | rdev->raid_disk = desc->raid_disk; |
f466722c | 1163 | rdev->saved_raid_disk = desc->raid_disk; |
0261cd9f N |
1164 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1165 | /* active but not in sync implies recovery up to | |
1166 | * reshape position. We don't know exactly where | |
1167 | * that is, so set to zero for now */ | |
1168 | if (mddev->minor_version >= 91) { | |
1169 | rdev->recovery_offset = 0; | |
1170 | rdev->raid_disk = desc->raid_disk; | |
1171 | } | |
1da177e4 | 1172 | } |
8ddf9efe N |
1173 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1174 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1175 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1176 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1177 | return 0; |
1178 | } | |
1179 | ||
1180 | /* | |
1181 | * sync_super for 0.90.0 | |
1182 | */ | |
fd01b88c | 1183 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1184 | { |
1185 | mdp_super_t *sb; | |
3cb03002 | 1186 | struct md_rdev *rdev2; |
1da177e4 | 1187 | int next_spare = mddev->raid_disks; |
19133a42 | 1188 | |
1da177e4 LT |
1189 | /* make rdev->sb match mddev data.. |
1190 | * | |
1191 | * 1/ zero out disks | |
1192 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1193 | * 3/ any empty disks < next_spare become removed | |
1194 | * | |
1195 | * disks[0] gets initialised to REMOVED because | |
1196 | * we cannot be sure from other fields if it has | |
1197 | * been initialised or not. | |
1198 | */ | |
1199 | int i; | |
1200 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1201 | ||
61181565 N |
1202 | rdev->sb_size = MD_SB_BYTES; |
1203 | ||
65a06f06 | 1204 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1205 | |
1206 | memset(sb, 0, sizeof(*sb)); | |
1207 | ||
1208 | sb->md_magic = MD_SB_MAGIC; | |
1209 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1210 | sb->patch_version = mddev->patch_version; |
1211 | sb->gvalid_words = 0; /* ignored */ | |
1212 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1213 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1214 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1215 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1216 | ||
1217 | sb->ctime = mddev->ctime; | |
1218 | sb->level = mddev->level; | |
58c0fed4 | 1219 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1220 | sb->raid_disks = mddev->raid_disks; |
1221 | sb->md_minor = mddev->md_minor; | |
e691063a | 1222 | sb->not_persistent = 0; |
1da177e4 LT |
1223 | sb->utime = mddev->utime; |
1224 | sb->state = 0; | |
1225 | sb->events_hi = (mddev->events>>32); | |
1226 | sb->events_lo = (u32)mddev->events; | |
1227 | ||
f6705578 N |
1228 | if (mddev->reshape_position == MaxSector) |
1229 | sb->minor_version = 90; | |
1230 | else { | |
1231 | sb->minor_version = 91; | |
1232 | sb->reshape_position = mddev->reshape_position; | |
1233 | sb->new_level = mddev->new_level; | |
1234 | sb->delta_disks = mddev->delta_disks; | |
1235 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1236 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1237 | } |
1238 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1239 | if (mddev->in_sync) |
1240 | { | |
1241 | sb->recovery_cp = mddev->recovery_cp; | |
1242 | sb->cp_events_hi = (mddev->events>>32); | |
1243 | sb->cp_events_lo = (u32)mddev->events; | |
1244 | if (mddev->recovery_cp == MaxSector) | |
1245 | sb->state = (1<< MD_SB_CLEAN); | |
1246 | } else | |
1247 | sb->recovery_cp = 0; | |
1248 | ||
1249 | sb->layout = mddev->layout; | |
9d8f0363 | 1250 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1251 | |
c3d9714e | 1252 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1253 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1254 | ||
1da177e4 | 1255 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1256 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1257 | mdp_disk_t *d; |
86e6ffdd | 1258 | int desc_nr; |
0261cd9f N |
1259 | int is_active = test_bit(In_sync, &rdev2->flags); |
1260 | ||
1261 | if (rdev2->raid_disk >= 0 && | |
1262 | sb->minor_version >= 91) | |
1263 | /* we have nowhere to store the recovery_offset, | |
1264 | * but if it is not below the reshape_position, | |
1265 | * we can piggy-back on that. | |
1266 | */ | |
1267 | is_active = 1; | |
1268 | if (rdev2->raid_disk < 0 || | |
1269 | test_bit(Faulty, &rdev2->flags)) | |
1270 | is_active = 0; | |
1271 | if (is_active) | |
86e6ffdd | 1272 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1273 | else |
86e6ffdd | 1274 | desc_nr = next_spare++; |
19133a42 | 1275 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1276 | d = &sb->disks[rdev2->desc_nr]; |
1277 | nr_disks++; | |
1278 | d->number = rdev2->desc_nr; | |
1279 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1280 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1281 | if (is_active) |
1da177e4 LT |
1282 | d->raid_disk = rdev2->raid_disk; |
1283 | else | |
1284 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1285 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1286 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1287 | else if (is_active) { |
1da177e4 | 1288 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1289 | if (test_bit(In_sync, &rdev2->flags)) |
1290 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1291 | active++; |
1292 | working++; | |
1293 | } else { | |
1294 | d->state = 0; | |
1295 | spare++; | |
1296 | working++; | |
1297 | } | |
8ddf9efe N |
1298 | if (test_bit(WriteMostly, &rdev2->flags)) |
1299 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1300 | } |
1da177e4 LT |
1301 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1302 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1303 | mdp_disk_t *d = &sb->disks[i]; | |
1304 | if (d->state == 0 && d->number == 0) { | |
1305 | d->number = i; | |
1306 | d->raid_disk = i; | |
1307 | d->state = (1<<MD_DISK_REMOVED); | |
1308 | d->state |= (1<<MD_DISK_FAULTY); | |
1309 | failed++; | |
1310 | } | |
1311 | } | |
1312 | sb->nr_disks = nr_disks; | |
1313 | sb->active_disks = active; | |
1314 | sb->working_disks = working; | |
1315 | sb->failed_disks = failed; | |
1316 | sb->spare_disks = spare; | |
1317 | ||
1318 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1319 | sb->sb_csum = calc_sb_csum(sb); | |
1320 | } | |
1321 | ||
0cd17fec CW |
1322 | /* |
1323 | * rdev_size_change for 0.90.0 | |
1324 | */ | |
1325 | static unsigned long long | |
3cb03002 | 1326 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1327 | { |
58c0fed4 | 1328 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1329 | return 0; /* component must fit device */ |
c3d9714e | 1330 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1331 | return 0; /* can't move bitmap */ |
57b2caa3 | 1332 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1333 | if (!num_sectors || num_sectors > rdev->sb_start) |
1334 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1335 | /* Limit to 4TB as metadata cannot record more than that. |
1336 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1337 | */ | |
667a5313 | 1338 | if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1) |
27a7b260 | 1339 | num_sectors = (2ULL << 32) - 2; |
0f420358 | 1340 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1341 | rdev->sb_page); |
1342 | md_super_wait(rdev->mddev); | |
c26a44ed | 1343 | return num_sectors; |
0cd17fec CW |
1344 | } |
1345 | ||
c6563a8c N |
1346 | static int |
1347 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1348 | { | |
1349 | /* non-zero offset changes not possible with v0.90 */ | |
1350 | return new_offset == 0; | |
1351 | } | |
0cd17fec | 1352 | |
1da177e4 LT |
1353 | /* |
1354 | * version 1 superblock | |
1355 | */ | |
1356 | ||
f72ffdd6 | 1357 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb) |
1da177e4 | 1358 | { |
1c05b4bc N |
1359 | __le32 disk_csum; |
1360 | u32 csum; | |
1da177e4 LT |
1361 | unsigned long long newcsum; |
1362 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1363 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1364 | |
1365 | disk_csum = sb->sb_csum; | |
1366 | sb->sb_csum = 0; | |
1367 | newcsum = 0; | |
1f3c9907 | 1368 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1369 | newcsum += le32_to_cpu(*isuper++); |
1370 | ||
1371 | if (size == 2) | |
1c05b4bc | 1372 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1373 | |
1374 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1375 | sb->sb_csum = disk_csum; | |
1376 | return cpu_to_le32(csum); | |
1377 | } | |
1378 | ||
2699b672 N |
1379 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, |
1380 | int acknowledged); | |
3cb03002 | 1381 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1382 | { |
1383 | struct mdp_superblock_1 *sb; | |
1384 | int ret; | |
0f420358 | 1385 | sector_t sb_start; |
c6563a8c | 1386 | sector_t sectors; |
1da177e4 | 1387 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1388 | int bmask; |
1da177e4 LT |
1389 | |
1390 | /* | |
0f420358 | 1391 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1392 | * It is always aligned to a 4K boundary and |
1393 | * depeding on minor_version, it can be: | |
1394 | * 0: At least 8K, but less than 12K, from end of device | |
1395 | * 1: At start of device | |
1396 | * 2: 4K from start of device. | |
1397 | */ | |
1398 | switch(minor_version) { | |
1399 | case 0: | |
77304d2a | 1400 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1401 | sb_start -= 8*2; |
1402 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1403 | break; |
1404 | case 1: | |
0f420358 | 1405 | sb_start = 0; |
1da177e4 LT |
1406 | break; |
1407 | case 2: | |
0f420358 | 1408 | sb_start = 8; |
1da177e4 LT |
1409 | break; |
1410 | default: | |
1411 | return -EINVAL; | |
1412 | } | |
0f420358 | 1413 | rdev->sb_start = sb_start; |
1da177e4 | 1414 | |
0002b271 N |
1415 | /* superblock is rarely larger than 1K, but it can be larger, |
1416 | * and it is safe to read 4k, so we do that | |
1417 | */ | |
1418 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1419 | if (ret) return ret; |
1420 | ||
65a06f06 | 1421 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1422 | |
1423 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1424 | sb->major_version != cpu_to_le32(1) || | |
1425 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1426 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1427 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1428 | return -EINVAL; |
1429 | ||
1430 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1431 | printk("md: invalid superblock checksum on %s\n", | |
1432 | bdevname(rdev->bdev,b)); | |
1433 | return -EINVAL; | |
1434 | } | |
1435 | if (le64_to_cpu(sb->data_size) < 10) { | |
1436 | printk("md: data_size too small on %s\n", | |
1437 | bdevname(rdev->bdev,b)); | |
1438 | return -EINVAL; | |
1439 | } | |
c6563a8c N |
1440 | if (sb->pad0 || |
1441 | sb->pad3[0] || | |
1442 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1443 | /* Some padding is non-zero, might be a new feature */ | |
1444 | return -EINVAL; | |
e11e93fa | 1445 | |
1da177e4 LT |
1446 | rdev->preferred_minor = 0xffff; |
1447 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1448 | rdev->new_data_offset = rdev->data_offset; |
1449 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1450 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1451 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1452 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1453 | |
0002b271 | 1454 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1455 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1456 | if (rdev->sb_size & bmask) |
a1801f85 N |
1457 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1458 | ||
1459 | if (minor_version | |
0f420358 | 1460 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1461 | return -EINVAL; |
c6563a8c N |
1462 | if (minor_version |
1463 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1464 | return -EINVAL; | |
0002b271 | 1465 | |
31b65a0d N |
1466 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1467 | rdev->desc_nr = -1; | |
1468 | else | |
1469 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1470 | ||
2699b672 N |
1471 | if (!rdev->bb_page) { |
1472 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1473 | if (!rdev->bb_page) | |
1474 | return -ENOMEM; | |
1475 | } | |
1476 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1477 | rdev->badblocks.count == 0) { | |
1478 | /* need to load the bad block list. | |
1479 | * Currently we limit it to one page. | |
1480 | */ | |
1481 | s32 offset; | |
1482 | sector_t bb_sector; | |
1483 | u64 *bbp; | |
1484 | int i; | |
1485 | int sectors = le16_to_cpu(sb->bblog_size); | |
1486 | if (sectors > (PAGE_SIZE / 512)) | |
1487 | return -EINVAL; | |
1488 | offset = le32_to_cpu(sb->bblog_offset); | |
1489 | if (offset == 0) | |
1490 | return -EINVAL; | |
1491 | bb_sector = (long long)offset; | |
1492 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
1493 | rdev->bb_page, READ, true)) | |
1494 | return -EIO; | |
1495 | bbp = (u64 *)page_address(rdev->bb_page); | |
1496 | rdev->badblocks.shift = sb->bblog_shift; | |
1497 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1498 | u64 bb = le64_to_cpu(*bbp); | |
1499 | int count = bb & (0x3ff); | |
1500 | u64 sector = bb >> 10; | |
1501 | sector <<= sb->bblog_shift; | |
1502 | count <<= sb->bblog_shift; | |
1503 | if (bb + 1 == 0) | |
1504 | break; | |
1505 | if (md_set_badblocks(&rdev->badblocks, | |
1506 | sector, count, 1) == 0) | |
1507 | return -EINVAL; | |
1508 | } | |
486adf72 N |
1509 | } else if (sb->bblog_offset != 0) |
1510 | rdev->badblocks.shift = 0; | |
2699b672 | 1511 | |
9a7b2b0f | 1512 | if (!refdev) { |
8ed75463 | 1513 | ret = 1; |
9a7b2b0f | 1514 | } else { |
1da177e4 | 1515 | __u64 ev1, ev2; |
65a06f06 | 1516 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1517 | |
1518 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1519 | sb->level != refsb->level || | |
1520 | sb->layout != refsb->layout || | |
1521 | sb->chunksize != refsb->chunksize) { | |
1522 | printk(KERN_WARNING "md: %s has strangely different" | |
1523 | " superblock to %s\n", | |
1524 | bdevname(rdev->bdev,b), | |
1525 | bdevname(refdev->bdev,b2)); | |
1526 | return -EINVAL; | |
1527 | } | |
1528 | ev1 = le64_to_cpu(sb->events); | |
1529 | ev2 = le64_to_cpu(refsb->events); | |
1530 | ||
1531 | if (ev1 > ev2) | |
8ed75463 N |
1532 | ret = 1; |
1533 | else | |
1534 | ret = 0; | |
1da177e4 | 1535 | } |
c6563a8c N |
1536 | if (minor_version) { |
1537 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1538 | sectors -= rdev->data_offset; | |
1539 | } else | |
1540 | sectors = rdev->sb_start; | |
1541 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1542 | return -EINVAL; |
dd8ac336 | 1543 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1544 | return ret; |
1da177e4 LT |
1545 | } |
1546 | ||
fd01b88c | 1547 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1548 | { |
65a06f06 | 1549 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1550 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1551 | |
41158c7e | 1552 | rdev->raid_disk = -1; |
c5d79adb N |
1553 | clear_bit(Faulty, &rdev->flags); |
1554 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1555 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1556 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1557 | |
1da177e4 LT |
1558 | if (mddev->raid_disks == 0) { |
1559 | mddev->major_version = 1; | |
1560 | mddev->patch_version = 0; | |
e691063a | 1561 | mddev->external = 0; |
9d8f0363 | 1562 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1563 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1564 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1565 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1566 | mddev->clevel[0] = 0; |
1da177e4 LT |
1567 | mddev->layout = le32_to_cpu(sb->layout); |
1568 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1569 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1570 | mddev->events = ev1; |
c3d9714e | 1571 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1572 | mddev->bitmap_info.space = 0; |
1573 | /* Default location for bitmap is 1K after superblock | |
1574 | * using 3K - total of 4K | |
1575 | */ | |
c3d9714e | 1576 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1577 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1578 | mddev->reshape_backwards = 0; |
1579 | ||
1da177e4 LT |
1580 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1581 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1582 | ||
1583 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1584 | |
71c0805c | 1585 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1586 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1587 | mddev->bitmap_info.offset = |
1588 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1589 | /* Metadata doesn't record how much space is available. |
1590 | * For 1.0, we assume we can use up to the superblock | |
1591 | * if before, else to 4K beyond superblock. | |
1592 | * For others, assume no change is possible. | |
1593 | */ | |
1594 | if (mddev->minor_version > 0) | |
1595 | mddev->bitmap_info.space = 0; | |
1596 | else if (mddev->bitmap_info.offset > 0) | |
1597 | mddev->bitmap_info.space = | |
1598 | 8 - mddev->bitmap_info.offset; | |
1599 | else | |
1600 | mddev->bitmap_info.space = | |
1601 | -mddev->bitmap_info.offset; | |
1602 | } | |
e11e93fa | 1603 | |
f6705578 N |
1604 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1605 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1606 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1607 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1608 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1609 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1610 | if (mddev->delta_disks < 0 || |
1611 | (mddev->delta_disks == 0 && | |
1612 | (le32_to_cpu(sb->feature_map) | |
1613 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1614 | mddev->reshape_backwards = 1; | |
f6705578 N |
1615 | } else { |
1616 | mddev->reshape_position = MaxSector; | |
1617 | mddev->delta_disks = 0; | |
1618 | mddev->new_level = mddev->level; | |
1619 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1620 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1621 | } |
1622 | ||
41158c7e | 1623 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1624 | /* Insist of good event counter while assembling, except for |
1625 | * spares (which don't need an event count) */ | |
1da177e4 | 1626 | ++ev1; |
be6800a7 N |
1627 | if (rdev->desc_nr >= 0 && |
1628 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
1629 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) | |
1630 | if (ev1 < mddev->events) | |
1631 | return -EINVAL; | |
41158c7e N |
1632 | } else if (mddev->bitmap) { |
1633 | /* If adding to array with a bitmap, then we can accept an | |
1634 | * older device, but not too old. | |
1635 | */ | |
41158c7e N |
1636 | if (ev1 < mddev->bitmap->events_cleared) |
1637 | return 0; | |
8313b8e5 N |
1638 | if (ev1 < mddev->events) |
1639 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1640 | } else { |
1641 | if (ev1 < mddev->events) | |
1642 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1643 | return 0; | |
1644 | } | |
1da177e4 LT |
1645 | if (mddev->level != LEVEL_MULTIPATH) { |
1646 | int role; | |
3673f305 N |
1647 | if (rdev->desc_nr < 0 || |
1648 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1649 | role = 0xffff; | |
1650 | rdev->desc_nr = -1; | |
1651 | } else | |
1652 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1653 | switch(role) { |
1654 | case 0xffff: /* spare */ | |
1da177e4 LT |
1655 | break; |
1656 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1657 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1658 | break; |
1659 | default: | |
f466722c | 1660 | rdev->saved_raid_disk = role; |
5fd6c1dc | 1661 | if ((le32_to_cpu(sb->feature_map) & |
f466722c | 1662 | MD_FEATURE_RECOVERY_OFFSET)) { |
5fd6c1dc | 1663 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); |
f466722c N |
1664 | if (!(le32_to_cpu(sb->feature_map) & |
1665 | MD_FEATURE_RECOVERY_BITMAP)) | |
1666 | rdev->saved_raid_disk = -1; | |
1667 | } else | |
5fd6c1dc | 1668 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1669 | rdev->raid_disk = role; |
1670 | break; | |
1671 | } | |
8ddf9efe N |
1672 | if (sb->devflags & WriteMostly1) |
1673 | set_bit(WriteMostly, &rdev->flags); | |
2d78f8c4 N |
1674 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1675 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1676 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1677 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1678 | |
1da177e4 LT |
1679 | return 0; |
1680 | } | |
1681 | ||
fd01b88c | 1682 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1683 | { |
1684 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1685 | struct md_rdev *rdev2; |
1da177e4 LT |
1686 | int max_dev, i; |
1687 | /* make rdev->sb match mddev and rdev data. */ | |
1688 | ||
65a06f06 | 1689 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1690 | |
1691 | sb->feature_map = 0; | |
1692 | sb->pad0 = 0; | |
5fd6c1dc | 1693 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1694 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1695 | ||
1696 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1697 | sb->events = cpu_to_le64(mddev->events); | |
1698 | if (mddev->in_sync) | |
1699 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1700 | else | |
1701 | sb->resync_offset = cpu_to_le64(0); | |
1702 | ||
1c05b4bc | 1703 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1704 | |
f0ca340c | 1705 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1706 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1707 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1708 | sb->level = cpu_to_le32(mddev->level); |
1709 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1710 | |
aeb9b211 N |
1711 | if (test_bit(WriteMostly, &rdev->flags)) |
1712 | sb->devflags |= WriteMostly1; | |
1713 | else | |
1714 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1715 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1716 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1717 | |
c3d9714e N |
1718 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1719 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1720 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1721 | } |
5fd6c1dc N |
1722 | |
1723 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1724 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1725 | sb->feature_map |= |
1726 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1727 | sb->recovery_offset = | |
1728 | cpu_to_le64(rdev->recovery_offset); | |
f466722c N |
1729 | if (rdev->saved_raid_disk >= 0 && mddev->bitmap) |
1730 | sb->feature_map |= | |
1731 | cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP); | |
5fd6c1dc | 1732 | } |
2d78f8c4 N |
1733 | if (test_bit(Replacement, &rdev->flags)) |
1734 | sb->feature_map |= | |
1735 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1736 | |
f6705578 N |
1737 | if (mddev->reshape_position != MaxSector) { |
1738 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1739 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1740 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1741 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1742 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1743 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1744 | if (mddev->delta_disks == 0 && |
1745 | mddev->reshape_backwards) | |
1746 | sb->feature_map | |
1747 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1748 | if (rdev->new_data_offset != rdev->data_offset) { |
1749 | sb->feature_map | |
1750 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1751 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1752 | - rdev->data_offset)); | |
1753 | } | |
f6705578 | 1754 | } |
a654b9d8 | 1755 | |
2699b672 N |
1756 | if (rdev->badblocks.count == 0) |
1757 | /* Nothing to do for bad blocks*/ ; | |
1758 | else if (sb->bblog_offset == 0) | |
1759 | /* Cannot record bad blocks on this device */ | |
1760 | md_error(mddev, rdev); | |
1761 | else { | |
1762 | struct badblocks *bb = &rdev->badblocks; | |
1763 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1764 | u64 *p = bb->page; | |
1765 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1766 | if (bb->changed) { | |
1767 | unsigned seq; | |
1768 | ||
1769 | retry: | |
1770 | seq = read_seqbegin(&bb->lock); | |
1771 | ||
1772 | memset(bbp, 0xff, PAGE_SIZE); | |
1773 | ||
1774 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1775 | u64 internal_bb = p[i]; |
2699b672 N |
1776 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1777 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1778 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1779 | } |
d0962936 | 1780 | bb->changed = 0; |
2699b672 N |
1781 | if (read_seqretry(&bb->lock, seq)) |
1782 | goto retry; | |
1783 | ||
1784 | bb->sector = (rdev->sb_start + | |
1785 | (int)le32_to_cpu(sb->bblog_offset)); | |
1786 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1787 | } |
1788 | } | |
1789 | ||
1da177e4 | 1790 | max_dev = 0; |
dafb20fa | 1791 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1792 | if (rdev2->desc_nr+1 > max_dev) |
1793 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1794 | |
70471daf N |
1795 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1796 | int bmask; | |
a778b73f | 1797 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1798 | rdev->sb_size = max_dev * 2 + 256; |
1799 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1800 | if (rdev->sb_size & bmask) | |
1801 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1802 | } else |
1803 | max_dev = le32_to_cpu(sb->max_dev); | |
1804 | ||
1da177e4 LT |
1805 | for (i=0; i<max_dev;i++) |
1806 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
f72ffdd6 | 1807 | |
dafb20fa | 1808 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1809 | i = rdev2->desc_nr; |
b2d444d7 | 1810 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1811 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1812 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1813 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1814 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1815 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1816 | else |
1817 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1818 | } | |
1819 | ||
1da177e4 LT |
1820 | sb->sb_csum = calc_sb_1_csum(sb); |
1821 | } | |
1822 | ||
0cd17fec | 1823 | static unsigned long long |
3cb03002 | 1824 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1825 | { |
1826 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1827 | sector_t max_sectors; |
58c0fed4 | 1828 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1829 | return 0; /* component must fit device */ |
c6563a8c N |
1830 | if (rdev->data_offset != rdev->new_data_offset) |
1831 | return 0; /* too confusing */ | |
0f420358 | 1832 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1833 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1834 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1835 | max_sectors -= rdev->data_offset; |
1836 | if (!num_sectors || num_sectors > max_sectors) | |
1837 | num_sectors = max_sectors; | |
c3d9714e | 1838 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1839 | /* minor version 0 with bitmap we can't move */ |
1840 | return 0; | |
1841 | } else { | |
1842 | /* minor version 0; superblock after data */ | |
0f420358 | 1843 | sector_t sb_start; |
77304d2a | 1844 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1845 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1846 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1847 | if (!num_sectors || num_sectors > max_sectors) |
1848 | num_sectors = max_sectors; | |
0f420358 | 1849 | rdev->sb_start = sb_start; |
0cd17fec | 1850 | } |
65a06f06 | 1851 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1852 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1853 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1854 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1855 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1856 | rdev->sb_page); |
1857 | md_super_wait(rdev->mddev); | |
c26a44ed | 1858 | return num_sectors; |
c6563a8c N |
1859 | |
1860 | } | |
1861 | ||
1862 | static int | |
1863 | super_1_allow_new_offset(struct md_rdev *rdev, | |
1864 | unsigned long long new_offset) | |
1865 | { | |
1866 | /* All necessary checks on new >= old have been done */ | |
1867 | struct bitmap *bitmap; | |
1868 | if (new_offset >= rdev->data_offset) | |
1869 | return 1; | |
1870 | ||
1871 | /* with 1.0 metadata, there is no metadata to tread on | |
1872 | * so we can always move back */ | |
1873 | if (rdev->mddev->minor_version == 0) | |
1874 | return 1; | |
1875 | ||
1876 | /* otherwise we must be sure not to step on | |
1877 | * any metadata, so stay: | |
1878 | * 36K beyond start of superblock | |
1879 | * beyond end of badblocks | |
1880 | * beyond write-intent bitmap | |
1881 | */ | |
1882 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
1883 | return 0; | |
1884 | bitmap = rdev->mddev->bitmap; | |
1885 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
1886 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 1887 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
1888 | return 0; |
1889 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
1890 | return 0; | |
1891 | ||
1892 | return 1; | |
0cd17fec | 1893 | } |
1da177e4 | 1894 | |
75c96f85 | 1895 | static struct super_type super_types[] = { |
1da177e4 LT |
1896 | [0] = { |
1897 | .name = "0.90.0", | |
1898 | .owner = THIS_MODULE, | |
0cd17fec CW |
1899 | .load_super = super_90_load, |
1900 | .validate_super = super_90_validate, | |
1901 | .sync_super = super_90_sync, | |
1902 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 1903 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
1904 | }, |
1905 | [1] = { | |
1906 | .name = "md-1", | |
1907 | .owner = THIS_MODULE, | |
0cd17fec CW |
1908 | .load_super = super_1_load, |
1909 | .validate_super = super_1_validate, | |
1910 | .sync_super = super_1_sync, | |
1911 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 1912 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
1913 | }, |
1914 | }; | |
1da177e4 | 1915 | |
fd01b88c | 1916 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
1917 | { |
1918 | if (mddev->sync_super) { | |
1919 | mddev->sync_super(mddev, rdev); | |
1920 | return; | |
1921 | } | |
1922 | ||
1923 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
1924 | ||
1925 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
1926 | } | |
1927 | ||
fd01b88c | 1928 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 1929 | { |
3cb03002 | 1930 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 1931 | |
4b80991c N |
1932 | rcu_read_lock(); |
1933 | rdev_for_each_rcu(rdev, mddev1) | |
1934 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1935 | if (rdev->bdev->bd_contains == |
4b80991c N |
1936 | rdev2->bdev->bd_contains) { |
1937 | rcu_read_unlock(); | |
7dd5e7c3 | 1938 | return 1; |
4b80991c N |
1939 | } |
1940 | rcu_read_unlock(); | |
1da177e4 LT |
1941 | return 0; |
1942 | } | |
1943 | ||
1944 | static LIST_HEAD(pending_raid_disks); | |
1945 | ||
ac5e7113 AN |
1946 | /* |
1947 | * Try to register data integrity profile for an mddev | |
1948 | * | |
1949 | * This is called when an array is started and after a disk has been kicked | |
1950 | * from the array. It only succeeds if all working and active component devices | |
1951 | * are integrity capable with matching profiles. | |
1952 | */ | |
fd01b88c | 1953 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 1954 | { |
3cb03002 | 1955 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
1956 | |
1957 | if (list_empty(&mddev->disks)) | |
1958 | return 0; /* nothing to do */ | |
629acb6a JB |
1959 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
1960 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 1961 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
1962 | /* skip spares and non-functional disks */ |
1963 | if (test_bit(Faulty, &rdev->flags)) | |
1964 | continue; | |
1965 | if (rdev->raid_disk < 0) | |
1966 | continue; | |
ac5e7113 AN |
1967 | if (!reference) { |
1968 | /* Use the first rdev as the reference */ | |
1969 | reference = rdev; | |
1970 | continue; | |
1971 | } | |
1972 | /* does this rdev's profile match the reference profile? */ | |
1973 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1974 | rdev->bdev->bd_disk) < 0) | |
1975 | return -EINVAL; | |
1976 | } | |
89078d57 MP |
1977 | if (!reference || !bdev_get_integrity(reference->bdev)) |
1978 | return 0; | |
ac5e7113 AN |
1979 | /* |
1980 | * All component devices are integrity capable and have matching | |
1981 | * profiles, register the common profile for the md device. | |
1982 | */ | |
1983 | if (blk_integrity_register(mddev->gendisk, | |
1984 | bdev_get_integrity(reference->bdev)) != 0) { | |
1985 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1986 | mdname(mddev)); | |
1987 | return -EINVAL; | |
1988 | } | |
a91a2785 MP |
1989 | printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); |
1990 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { | |
1991 | printk(KERN_ERR "md: failed to create integrity pool for %s\n", | |
1992 | mdname(mddev)); | |
1993 | return -EINVAL; | |
1994 | } | |
ac5e7113 AN |
1995 | return 0; |
1996 | } | |
1997 | EXPORT_SYMBOL(md_integrity_register); | |
1998 | ||
1999 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
fd01b88c | 2000 | void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) |
3f9d99c1 | 2001 | { |
2863b9eb JB |
2002 | struct blk_integrity *bi_rdev; |
2003 | struct blk_integrity *bi_mddev; | |
2004 | ||
2005 | if (!mddev->gendisk) | |
2006 | return; | |
2007 | ||
2008 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
2009 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 2010 | |
ac5e7113 | 2011 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 2012 | return; |
ac5e7113 | 2013 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 2014 | return; |
ac5e7113 AN |
2015 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
2016 | rdev->bdev->bd_disk) >= 0) | |
2017 | return; | |
2018 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
2019 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 2020 | } |
ac5e7113 | 2021 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 2022 | |
f72ffdd6 | 2023 | static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev) |
1da177e4 | 2024 | { |
7dd5e7c3 | 2025 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2026 | struct kobject *ko; |
5e55e2f5 | 2027 | int err; |
1da177e4 | 2028 | |
11e2ede0 DW |
2029 | /* prevent duplicates */ |
2030 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2031 | return -EEXIST; | |
2032 | ||
dd8ac336 AN |
2033 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
2034 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
2035 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2036 | if (mddev->pers) { |
2037 | /* Cannot change size, so fail | |
2038 | * If mddev->level <= 0, then we don't care | |
2039 | * about aligning sizes (e.g. linear) | |
2040 | */ | |
2041 | if (mddev->level > 0) | |
2042 | return -ENOSPC; | |
2043 | } else | |
dd8ac336 | 2044 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2045 | } |
1da177e4 LT |
2046 | |
2047 | /* Verify rdev->desc_nr is unique. | |
2048 | * If it is -1, assign a free number, else | |
2049 | * check number is not in use | |
2050 | */ | |
4878e9eb | 2051 | rcu_read_lock(); |
1da177e4 LT |
2052 | if (rdev->desc_nr < 0) { |
2053 | int choice = 0; | |
4878e9eb N |
2054 | if (mddev->pers) |
2055 | choice = mddev->raid_disks; | |
57d051dc | 2056 | while (md_find_rdev_nr_rcu(mddev, choice)) |
1da177e4 LT |
2057 | choice++; |
2058 | rdev->desc_nr = choice; | |
2059 | } else { | |
57d051dc | 2060 | if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) { |
4878e9eb | 2061 | rcu_read_unlock(); |
1da177e4 | 2062 | return -EBUSY; |
4878e9eb | 2063 | } |
1da177e4 | 2064 | } |
4878e9eb | 2065 | rcu_read_unlock(); |
de01dfad N |
2066 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
2067 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
2068 | mdname(mddev), mddev->max_disks); | |
2069 | return -EBUSY; | |
2070 | } | |
19133a42 | 2071 | bdevname(rdev->bdev,b); |
90a9befb | 2072 | strreplace(b, '/', '!'); |
649316b2 | 2073 | |
1da177e4 | 2074 | rdev->mddev = mddev; |
19133a42 | 2075 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 2076 | |
b2d6db58 | 2077 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2078 | goto fail; |
86e6ffdd | 2079 | |
0762b8bd | 2080 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2081 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2082 | /* failure here is OK */; | |
2083 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2084 | |
4b80991c | 2085 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2086 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2087 | |
2088 | /* May as well allow recovery to be retried once */ | |
5389042f | 2089 | mddev->recovery_disabled++; |
3f9d99c1 | 2090 | |
1da177e4 | 2091 | return 0; |
5e55e2f5 N |
2092 | |
2093 | fail: | |
2094 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
2095 | b, mdname(mddev)); | |
2096 | return err; | |
1da177e4 LT |
2097 | } |
2098 | ||
177a99b2 | 2099 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2100 | { |
3cb03002 | 2101 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2102 | kobject_del(&rdev->kobj); |
177a99b2 | 2103 | kobject_put(&rdev->kobj); |
5792a285 N |
2104 | } |
2105 | ||
f72ffdd6 | 2106 | static void unbind_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2107 | { |
2108 | char b[BDEVNAME_SIZE]; | |
403df478 | 2109 | |
49731baa | 2110 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2111 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
2112 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
2113 | rdev->mddev = NULL; | |
86e6ffdd | 2114 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2115 | sysfs_put(rdev->sysfs_state); |
2116 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2117 | rdev->badblocks.count = 0; |
5792a285 | 2118 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2119 | * writing to 'remove' to "dev/state". We also need |
2120 | * to delay it due to rcu usage. | |
5792a285 | 2121 | */ |
4b80991c | 2122 | synchronize_rcu(); |
177a99b2 N |
2123 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2124 | kobject_get(&rdev->kobj); | |
e804ac78 | 2125 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2126 | } |
2127 | ||
2128 | /* | |
2129 | * prevent the device from being mounted, repartitioned or | |
2130 | * otherwise reused by a RAID array (or any other kernel | |
2131 | * subsystem), by bd_claiming the device. | |
2132 | */ | |
3cb03002 | 2133 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2134 | { |
2135 | int err = 0; | |
2136 | struct block_device *bdev; | |
2137 | char b[BDEVNAME_SIZE]; | |
2138 | ||
d4d77629 | 2139 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2140 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 LT |
2141 | if (IS_ERR(bdev)) { |
2142 | printk(KERN_ERR "md: could not open %s.\n", | |
2143 | __bdevname(dev, b)); | |
2144 | return PTR_ERR(bdev); | |
2145 | } | |
1da177e4 LT |
2146 | rdev->bdev = bdev; |
2147 | return err; | |
2148 | } | |
2149 | ||
3cb03002 | 2150 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2151 | { |
2152 | struct block_device *bdev = rdev->bdev; | |
2153 | rdev->bdev = NULL; | |
e525fd89 | 2154 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2155 | } |
2156 | ||
2157 | void md_autodetect_dev(dev_t dev); | |
2158 | ||
f72ffdd6 | 2159 | static void export_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2160 | { |
2161 | char b[BDEVNAME_SIZE]; | |
403df478 | 2162 | |
1da177e4 LT |
2163 | printk(KERN_INFO "md: export_rdev(%s)\n", |
2164 | bdevname(rdev->bdev,b)); | |
545c8795 | 2165 | md_rdev_clear(rdev); |
1da177e4 | 2166 | #ifndef MODULE |
d0fae18f N |
2167 | if (test_bit(AutoDetected, &rdev->flags)) |
2168 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2169 | #endif |
2170 | unlock_rdev(rdev); | |
86e6ffdd | 2171 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2172 | } |
2173 | ||
fb56dfef | 2174 | void md_kick_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2175 | { |
2176 | unbind_rdev_from_array(rdev); | |
2177 | export_rdev(rdev); | |
2178 | } | |
fb56dfef | 2179 | EXPORT_SYMBOL_GPL(md_kick_rdev_from_array); |
1da177e4 | 2180 | |
fd01b88c | 2181 | static void export_array(struct mddev *mddev) |
1da177e4 | 2182 | { |
0638bb0e | 2183 | struct md_rdev *rdev; |
1da177e4 | 2184 | |
0638bb0e N |
2185 | while (!list_empty(&mddev->disks)) { |
2186 | rdev = list_first_entry(&mddev->disks, struct md_rdev, | |
2187 | same_set); | |
fb56dfef | 2188 | md_kick_rdev_from_array(rdev); |
1da177e4 | 2189 | } |
1da177e4 LT |
2190 | mddev->raid_disks = 0; |
2191 | mddev->major_version = 0; | |
2192 | } | |
2193 | ||
f72ffdd6 | 2194 | static void sync_sbs(struct mddev *mddev, int nospares) |
1da177e4 | 2195 | { |
42543769 N |
2196 | /* Update each superblock (in-memory image), but |
2197 | * if we are allowed to, skip spares which already | |
2198 | * have the right event counter, or have one earlier | |
2199 | * (which would mean they aren't being marked as dirty | |
2200 | * with the rest of the array) | |
2201 | */ | |
3cb03002 | 2202 | struct md_rdev *rdev; |
dafb20fa | 2203 | rdev_for_each(rdev, mddev) { |
42543769 N |
2204 | if (rdev->sb_events == mddev->events || |
2205 | (nospares && | |
2206 | rdev->raid_disk < 0 && | |
42543769 N |
2207 | rdev->sb_events+1 == mddev->events)) { |
2208 | /* Don't update this superblock */ | |
2209 | rdev->sb_loaded = 2; | |
2210 | } else { | |
076f968b | 2211 | sync_super(mddev, rdev); |
42543769 N |
2212 | rdev->sb_loaded = 1; |
2213 | } | |
1da177e4 LT |
2214 | } |
2215 | } | |
2216 | ||
1aee41f6 | 2217 | void md_update_sb(struct mddev *mddev, int force_change) |
1da177e4 | 2218 | { |
3cb03002 | 2219 | struct md_rdev *rdev; |
06d91a5f | 2220 | int sync_req; |
42543769 | 2221 | int nospares = 0; |
2699b672 | 2222 | int any_badblocks_changed = 0; |
1da177e4 | 2223 | |
d87f064f N |
2224 | if (mddev->ro) { |
2225 | if (force_change) | |
2226 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2227 | return; | |
2228 | } | |
1da177e4 | 2229 | repeat: |
3a3a5ddb | 2230 | /* First make sure individual recovery_offsets are correct */ |
dafb20fa | 2231 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2232 | if (rdev->raid_disk >= 0 && |
2233 | mddev->delta_disks >= 0 && | |
2234 | !test_bit(In_sync, &rdev->flags) && | |
2235 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2236 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2237 | ||
f72ffdd6 | 2238 | } |
bd52b746 | 2239 | if (!mddev->persistent) { |
070dc6dd | 2240 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2241 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
de393cde | 2242 | if (!mddev->external) { |
d97a41dc | 2243 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
dafb20fa | 2244 | rdev_for_each(rdev, mddev) { |
de393cde | 2245 | if (rdev->badblocks.changed) { |
d0962936 | 2246 | rdev->badblocks.changed = 0; |
de393cde N |
2247 | md_ack_all_badblocks(&rdev->badblocks); |
2248 | md_error(mddev, rdev); | |
2249 | } | |
2250 | clear_bit(Blocked, &rdev->flags); | |
2251 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2252 | wake_up(&rdev->blocked_wait); | |
2253 | } | |
2254 | } | |
3a3a5ddb N |
2255 | wake_up(&mddev->sb_wait); |
2256 | return; | |
2257 | } | |
2258 | ||
85572d7c | 2259 | spin_lock(&mddev->lock); |
84692195 | 2260 | |
3a3a5ddb N |
2261 | mddev->utime = get_seconds(); |
2262 | ||
850b2b42 N |
2263 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2264 | force_change = 1; | |
2265 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2266 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2267 | * though if events isn't the right even/odd, we will have to do | |
2268 | * spares after all | |
2269 | */ | |
2270 | nospares = 1; | |
2271 | if (force_change) | |
2272 | nospares = 0; | |
2273 | if (mddev->degraded) | |
84692195 N |
2274 | /* If the array is degraded, then skipping spares is both |
2275 | * dangerous and fairly pointless. | |
2276 | * Dangerous because a device that was removed from the array | |
2277 | * might have a event_count that still looks up-to-date, | |
2278 | * so it can be re-added without a resync. | |
2279 | * Pointless because if there are any spares to skip, | |
2280 | * then a recovery will happen and soon that array won't | |
2281 | * be degraded any more and the spare can go back to sleep then. | |
2282 | */ | |
850b2b42 | 2283 | nospares = 0; |
84692195 | 2284 | |
06d91a5f | 2285 | sync_req = mddev->in_sync; |
42543769 N |
2286 | |
2287 | /* If this is just a dirty<->clean transition, and the array is clean | |
2288 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2289 | if (nospares |
42543769 | 2290 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2291 | && mddev->can_decrease_events |
2292 | && mddev->events != 1) { | |
42543769 | 2293 | mddev->events--; |
a8707c08 N |
2294 | mddev->can_decrease_events = 0; |
2295 | } else { | |
42543769 N |
2296 | /* otherwise we have to go forward and ... */ |
2297 | mddev->events ++; | |
a8707c08 | 2298 | mddev->can_decrease_events = nospares; |
42543769 | 2299 | } |
1da177e4 | 2300 | |
403df478 N |
2301 | /* |
2302 | * This 64-bit counter should never wrap. | |
2303 | * Either we are in around ~1 trillion A.C., assuming | |
2304 | * 1 reboot per second, or we have a bug... | |
2305 | */ | |
2306 | WARN_ON(mddev->events == 0); | |
2699b672 | 2307 | |
dafb20fa | 2308 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2309 | if (rdev->badblocks.changed) |
2310 | any_badblocks_changed++; | |
de393cde N |
2311 | if (test_bit(Faulty, &rdev->flags)) |
2312 | set_bit(FaultRecorded, &rdev->flags); | |
2313 | } | |
2699b672 | 2314 | |
e691063a | 2315 | sync_sbs(mddev, nospares); |
85572d7c | 2316 | spin_unlock(&mddev->lock); |
1da177e4 | 2317 | |
36a4e1fe N |
2318 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2319 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2320 | |
4ad13663 | 2321 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2322 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2323 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2324 | |
42543769 N |
2325 | if (rdev->sb_loaded != 1) |
2326 | continue; /* no noise on spare devices */ | |
1da177e4 | 2327 | |
f466722c | 2328 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2329 | md_super_write(mddev,rdev, |
0f420358 | 2330 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2331 | rdev->sb_page); |
36a4e1fe N |
2332 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2333 | bdevname(rdev->bdev, b), | |
2334 | (unsigned long long)rdev->sb_start); | |
42543769 | 2335 | rdev->sb_events = mddev->events; |
2699b672 N |
2336 | if (rdev->badblocks.size) { |
2337 | md_super_write(mddev, rdev, | |
2338 | rdev->badblocks.sector, | |
2339 | rdev->badblocks.size << 9, | |
2340 | rdev->bb_page); | |
2341 | rdev->badblocks.size = 0; | |
2342 | } | |
7bfa19f2 | 2343 | |
f466722c | 2344 | } else |
36a4e1fe N |
2345 | pr_debug("md: %s (skipping faulty)\n", |
2346 | bdevname(rdev->bdev, b)); | |
d70ed2e4 | 2347 | |
7bfa19f2 | 2348 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2349 | /* only need to write one superblock... */ |
2350 | break; | |
2351 | } | |
a9701a30 | 2352 | md_super_wait(mddev); |
850b2b42 | 2353 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2354 | |
85572d7c | 2355 | spin_lock(&mddev->lock); |
850b2b42 N |
2356 | if (mddev->in_sync != sync_req || |
2357 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2358 | /* have to write it out again */ |
85572d7c | 2359 | spin_unlock(&mddev->lock); |
06d91a5f N |
2360 | goto repeat; |
2361 | } | |
850b2b42 | 2362 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
85572d7c | 2363 | spin_unlock(&mddev->lock); |
3d310eb7 | 2364 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2365 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2366 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2367 | |
dafb20fa | 2368 | rdev_for_each(rdev, mddev) { |
de393cde N |
2369 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2370 | clear_bit(Blocked, &rdev->flags); | |
2371 | ||
2372 | if (any_badblocks_changed) | |
2699b672 | 2373 | md_ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2374 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2375 | wake_up(&rdev->blocked_wait); | |
2376 | } | |
1da177e4 | 2377 | } |
1aee41f6 | 2378 | EXPORT_SYMBOL(md_update_sb); |
1da177e4 | 2379 | |
a6da4ef8 GR |
2380 | static int add_bound_rdev(struct md_rdev *rdev) |
2381 | { | |
2382 | struct mddev *mddev = rdev->mddev; | |
2383 | int err = 0; | |
2384 | ||
2385 | if (!mddev->pers->hot_remove_disk) { | |
2386 | /* If there is hot_add_disk but no hot_remove_disk | |
2387 | * then added disks for geometry changes, | |
2388 | * and should be added immediately. | |
2389 | */ | |
2390 | super_types[mddev->major_version]. | |
2391 | validate_super(mddev, rdev); | |
2392 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
2393 | if (err) { | |
2394 | unbind_rdev_from_array(rdev); | |
2395 | export_rdev(rdev); | |
2396 | return err; | |
2397 | } | |
2398 | } | |
2399 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2400 | ||
2401 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2402 | if (mddev->degraded) | |
2403 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
2404 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2405 | md_new_event(mddev); | |
2406 | md_wakeup_thread(mddev->thread); | |
2407 | return 0; | |
2408 | } | |
1da177e4 | 2409 | |
7f6ce769 | 2410 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2411 | * We want to accept with case. For this we use cmd_match. |
2412 | */ | |
2413 | static int cmd_match(const char *cmd, const char *str) | |
2414 | { | |
2415 | /* See if cmd, written into a sysfs file, matches | |
2416 | * str. They must either be the same, or cmd can | |
2417 | * have a trailing newline | |
2418 | */ | |
2419 | while (*cmd && *str && *cmd == *str) { | |
2420 | cmd++; | |
2421 | str++; | |
2422 | } | |
2423 | if (*cmd == '\n') | |
2424 | cmd++; | |
2425 | if (*str || *cmd) | |
2426 | return 0; | |
2427 | return 1; | |
2428 | } | |
2429 | ||
86e6ffdd N |
2430 | struct rdev_sysfs_entry { |
2431 | struct attribute attr; | |
3cb03002 N |
2432 | ssize_t (*show)(struct md_rdev *, char *); |
2433 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2434 | }; |
2435 | ||
2436 | static ssize_t | |
3cb03002 | 2437 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd N |
2438 | { |
2439 | char *sep = ""; | |
20a49ff6 | 2440 | size_t len = 0; |
758bfc8a | 2441 | unsigned long flags = ACCESS_ONCE(rdev->flags); |
86e6ffdd | 2442 | |
758bfc8a | 2443 | if (test_bit(Faulty, &flags) || |
de393cde | 2444 | rdev->badblocks.unacked_exist) { |
86e6ffdd N |
2445 | len+= sprintf(page+len, "%sfaulty",sep); |
2446 | sep = ","; | |
2447 | } | |
758bfc8a | 2448 | if (test_bit(In_sync, &flags)) { |
86e6ffdd N |
2449 | len += sprintf(page+len, "%sin_sync",sep); |
2450 | sep = ","; | |
2451 | } | |
758bfc8a | 2452 | if (test_bit(WriteMostly, &flags)) { |
f655675b N |
2453 | len += sprintf(page+len, "%swrite_mostly",sep); |
2454 | sep = ","; | |
2455 | } | |
758bfc8a | 2456 | if (test_bit(Blocked, &flags) || |
52c64152 | 2457 | (rdev->badblocks.unacked_exist |
758bfc8a | 2458 | && !test_bit(Faulty, &flags))) { |
6bfe0b49 DW |
2459 | len += sprintf(page+len, "%sblocked", sep); |
2460 | sep = ","; | |
2461 | } | |
758bfc8a N |
2462 | if (!test_bit(Faulty, &flags) && |
2463 | !test_bit(In_sync, &flags)) { | |
86e6ffdd N |
2464 | len += sprintf(page+len, "%sspare", sep); |
2465 | sep = ","; | |
2466 | } | |
758bfc8a | 2467 | if (test_bit(WriteErrorSeen, &flags)) { |
d7a9d443 N |
2468 | len += sprintf(page+len, "%swrite_error", sep); |
2469 | sep = ","; | |
2470 | } | |
758bfc8a | 2471 | if (test_bit(WantReplacement, &flags)) { |
2d78f8c4 N |
2472 | len += sprintf(page+len, "%swant_replacement", sep); |
2473 | sep = ","; | |
2474 | } | |
758bfc8a | 2475 | if (test_bit(Replacement, &flags)) { |
2d78f8c4 N |
2476 | len += sprintf(page+len, "%sreplacement", sep); |
2477 | sep = ","; | |
2478 | } | |
2479 | ||
86e6ffdd N |
2480 | return len+sprintf(page+len, "\n"); |
2481 | } | |
2482 | ||
45dc2de1 | 2483 | static ssize_t |
3cb03002 | 2484 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2485 | { |
2486 | /* can write | |
de393cde | 2487 | * faulty - simulates an error |
45dc2de1 | 2488 | * remove - disconnects the device |
f655675b N |
2489 | * writemostly - sets write_mostly |
2490 | * -writemostly - clears write_mostly | |
de393cde N |
2491 | * blocked - sets the Blocked flags |
2492 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2493 | * insync - sets Insync providing device isn't active |
f466722c N |
2494 | * -insync - clear Insync for a device with a slot assigned, |
2495 | * so that it gets rebuilt based on bitmap | |
d7a9d443 N |
2496 | * write_error - sets WriteErrorSeen |
2497 | * -write_error - clears WriteErrorSeen | |
45dc2de1 N |
2498 | */ |
2499 | int err = -EINVAL; | |
2500 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2501 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2502 | if (test_bit(Faulty, &rdev->flags)) |
2503 | err = 0; | |
2504 | else | |
2505 | err = -EBUSY; | |
45dc2de1 N |
2506 | } else if (cmd_match(buf, "remove")) { |
2507 | if (rdev->raid_disk >= 0) | |
2508 | err = -EBUSY; | |
2509 | else { | |
fd01b88c | 2510 | struct mddev *mddev = rdev->mddev; |
293467aa | 2511 | if (mddev_is_clustered(mddev)) |
88bcfef7 | 2512 | md_cluster_ops->remove_disk(mddev, rdev); |
fb56dfef | 2513 | md_kick_rdev_from_array(rdev); |
88bcfef7 GR |
2514 | if (mddev_is_clustered(mddev)) |
2515 | md_cluster_ops->metadata_update_start(mddev); | |
3f9d7b0d N |
2516 | if (mddev->pers) |
2517 | md_update_sb(mddev, 1); | |
45dc2de1 | 2518 | md_new_event(mddev); |
293467aa GR |
2519 | if (mddev_is_clustered(mddev)) |
2520 | md_cluster_ops->metadata_update_finish(mddev); | |
45dc2de1 N |
2521 | err = 0; |
2522 | } | |
f655675b N |
2523 | } else if (cmd_match(buf, "writemostly")) { |
2524 | set_bit(WriteMostly, &rdev->flags); | |
2525 | err = 0; | |
2526 | } else if (cmd_match(buf, "-writemostly")) { | |
2527 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2528 | err = 0; |
2529 | } else if (cmd_match(buf, "blocked")) { | |
2530 | set_bit(Blocked, &rdev->flags); | |
2531 | err = 0; | |
2532 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2533 | if (!test_bit(Faulty, &rdev->flags) && |
7da64a0a | 2534 | rdev->badblocks.unacked_exist) { |
de393cde N |
2535 | /* metadata handler doesn't understand badblocks, |
2536 | * so we need to fail the device | |
2537 | */ | |
2538 | md_error(rdev->mddev, rdev); | |
2539 | } | |
6bfe0b49 | 2540 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2541 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2542 | wake_up(&rdev->blocked_wait); |
2543 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2544 | md_wakeup_thread(rdev->mddev->thread); | |
2545 | ||
6d56e278 N |
2546 | err = 0; |
2547 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2548 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2549 | err = 0; |
f466722c | 2550 | } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) { |
e1960f8c N |
2551 | if (rdev->mddev->pers == NULL) { |
2552 | clear_bit(In_sync, &rdev->flags); | |
2553 | rdev->saved_raid_disk = rdev->raid_disk; | |
2554 | rdev->raid_disk = -1; | |
2555 | err = 0; | |
2556 | } | |
d7a9d443 N |
2557 | } else if (cmd_match(buf, "write_error")) { |
2558 | set_bit(WriteErrorSeen, &rdev->flags); | |
2559 | err = 0; | |
2560 | } else if (cmd_match(buf, "-write_error")) { | |
2561 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2562 | err = 0; | |
2d78f8c4 N |
2563 | } else if (cmd_match(buf, "want_replacement")) { |
2564 | /* Any non-spare device that is not a replacement can | |
2565 | * become want_replacement at any time, but we then need to | |
2566 | * check if recovery is needed. | |
2567 | */ | |
2568 | if (rdev->raid_disk >= 0 && | |
2569 | !test_bit(Replacement, &rdev->flags)) | |
2570 | set_bit(WantReplacement, &rdev->flags); | |
2571 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2572 | md_wakeup_thread(rdev->mddev->thread); | |
2573 | err = 0; | |
2574 | } else if (cmd_match(buf, "-want_replacement")) { | |
2575 | /* Clearing 'want_replacement' is always allowed. | |
2576 | * Once replacements starts it is too late though. | |
2577 | */ | |
2578 | err = 0; | |
2579 | clear_bit(WantReplacement, &rdev->flags); | |
2580 | } else if (cmd_match(buf, "replacement")) { | |
2581 | /* Can only set a device as a replacement when array has not | |
2582 | * yet been started. Once running, replacement is automatic | |
2583 | * from spares, or by assigning 'slot'. | |
2584 | */ | |
2585 | if (rdev->mddev->pers) | |
2586 | err = -EBUSY; | |
2587 | else { | |
2588 | set_bit(Replacement, &rdev->flags); | |
2589 | err = 0; | |
2590 | } | |
2591 | } else if (cmd_match(buf, "-replacement")) { | |
2592 | /* Similarly, can only clear Replacement before start */ | |
2593 | if (rdev->mddev->pers) | |
2594 | err = -EBUSY; | |
2595 | else { | |
2596 | clear_bit(Replacement, &rdev->flags); | |
2597 | err = 0; | |
2598 | } | |
a6da4ef8 GR |
2599 | } else if (cmd_match(buf, "re-add")) { |
2600 | if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) { | |
97f6cd39 GR |
2601 | /* clear_bit is performed _after_ all the devices |
2602 | * have their local Faulty bit cleared. If any writes | |
2603 | * happen in the meantime in the local node, they | |
2604 | * will land in the local bitmap, which will be synced | |
2605 | * by this node eventually | |
2606 | */ | |
2607 | if (!mddev_is_clustered(rdev->mddev) || | |
2608 | (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) { | |
2609 | clear_bit(Faulty, &rdev->flags); | |
2610 | err = add_bound_rdev(rdev); | |
2611 | } | |
a6da4ef8 GR |
2612 | } else |
2613 | err = -EBUSY; | |
45dc2de1 | 2614 | } |
00bcb4ac N |
2615 | if (!err) |
2616 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2617 | return err ? err : len; |
2618 | } | |
80ca3a44 | 2619 | static struct rdev_sysfs_entry rdev_state = |
750f199e | 2620 | __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store); |
86e6ffdd | 2621 | |
4dbcdc75 | 2622 | static ssize_t |
3cb03002 | 2623 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2624 | { |
2625 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2626 | } | |
2627 | ||
2628 | static ssize_t | |
3cb03002 | 2629 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 | 2630 | { |
4c9309c0 AD |
2631 | unsigned int n; |
2632 | int rv; | |
2633 | ||
2634 | rv = kstrtouint(buf, 10, &n); | |
2635 | if (rv < 0) | |
2636 | return rv; | |
2637 | atomic_set(&rdev->corrected_errors, n); | |
2638 | return len; | |
4dbcdc75 N |
2639 | } |
2640 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2641 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2642 | |
014236d2 | 2643 | static ssize_t |
3cb03002 | 2644 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 N |
2645 | { |
2646 | if (rdev->raid_disk < 0) | |
2647 | return sprintf(page, "none\n"); | |
2648 | else | |
2649 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2650 | } | |
2651 | ||
2652 | static ssize_t | |
3cb03002 | 2653 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 | 2654 | { |
4c9309c0 | 2655 | int slot; |
c303da6d | 2656 | int err; |
4c9309c0 | 2657 | |
014236d2 N |
2658 | if (strncmp(buf, "none", 4)==0) |
2659 | slot = -1; | |
4c9309c0 AD |
2660 | else { |
2661 | err = kstrtouint(buf, 10, (unsigned int *)&slot); | |
2662 | if (err < 0) | |
2663 | return err; | |
2664 | } | |
6c2fce2e | 2665 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2666 | /* Setting 'slot' on an active array requires also |
2667 | * updating the 'rd%d' link, and communicating | |
2668 | * with the personality with ->hot_*_disk. | |
2669 | * For now we only support removing | |
2670 | * failed/spare devices. This normally happens automatically, | |
2671 | * but not when the metadata is externally managed. | |
2672 | */ | |
c303da6d N |
2673 | if (rdev->raid_disk == -1) |
2674 | return -EEXIST; | |
2675 | /* personality does all needed checks */ | |
01393f3d | 2676 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2677 | return -EINVAL; |
746d3207 N |
2678 | clear_bit(Blocked, &rdev->flags); |
2679 | remove_and_add_spares(rdev->mddev, rdev); | |
2680 | if (rdev->raid_disk >= 0) | |
2681 | return -EBUSY; | |
c303da6d N |
2682 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2683 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2684 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2685 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2686 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2687 | */ |
2688 | ||
2689 | if (rdev->raid_disk != -1) | |
2690 | return -EBUSY; | |
2691 | ||
c6751b2b N |
2692 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2693 | return -EBUSY; | |
2694 | ||
6c2fce2e NB |
2695 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2696 | return -EINVAL; | |
2697 | ||
ba1b41b6 N |
2698 | if (slot >= rdev->mddev->raid_disks && |
2699 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2700 | return -ENOSPC; | |
2701 | ||
6c2fce2e NB |
2702 | rdev->raid_disk = slot; |
2703 | if (test_bit(In_sync, &rdev->flags)) | |
2704 | rdev->saved_raid_disk = slot; | |
2705 | else | |
2706 | rdev->saved_raid_disk = -1; | |
d30519fc | 2707 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2708 | clear_bit(Bitmap_sync, &rdev->flags); |
6c2fce2e NB |
2709 | err = rdev->mddev->pers-> |
2710 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2711 | if (err) { |
6c2fce2e | 2712 | rdev->raid_disk = -1; |
6c2fce2e | 2713 | return err; |
52664732 | 2714 | } else |
00bcb4ac | 2715 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
36fad858 | 2716 | if (sysfs_link_rdev(rdev->mddev, rdev)) |
00bcb4ac | 2717 | /* failure here is OK */; |
6c2fce2e | 2718 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2719 | } else { |
ba1b41b6 N |
2720 | if (slot >= rdev->mddev->raid_disks && |
2721 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2722 | return -ENOSPC; |
2723 | rdev->raid_disk = slot; | |
2724 | /* assume it is working */ | |
c5d79adb N |
2725 | clear_bit(Faulty, &rdev->flags); |
2726 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2727 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2728 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2729 | } |
014236d2 N |
2730 | return len; |
2731 | } | |
2732 | ||
014236d2 | 2733 | static struct rdev_sysfs_entry rdev_slot = |
80ca3a44 | 2734 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2735 | |
93c8cad0 | 2736 | static ssize_t |
3cb03002 | 2737 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 2738 | { |
6961ece4 | 2739 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2740 | } |
2741 | ||
2742 | static ssize_t | |
3cb03002 | 2743 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 2744 | { |
c6563a8c | 2745 | unsigned long long offset; |
b29bebd6 | 2746 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 2747 | return -EINVAL; |
8ed0a521 | 2748 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2749 | return -EBUSY; |
dd8ac336 | 2750 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2751 | /* Must set offset before size, so overlap checks |
2752 | * can be sane */ | |
2753 | return -EBUSY; | |
93c8cad0 | 2754 | rdev->data_offset = offset; |
25f7fd47 | 2755 | rdev->new_data_offset = offset; |
93c8cad0 N |
2756 | return len; |
2757 | } | |
2758 | ||
2759 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2760 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2761 | |
c6563a8c N |
2762 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
2763 | { | |
2764 | return sprintf(page, "%llu\n", | |
2765 | (unsigned long long)rdev->new_data_offset); | |
2766 | } | |
2767 | ||
2768 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
2769 | const char *buf, size_t len) | |
2770 | { | |
2771 | unsigned long long new_offset; | |
2772 | struct mddev *mddev = rdev->mddev; | |
2773 | ||
b29bebd6 | 2774 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
2775 | return -EINVAL; |
2776 | ||
f851b60d N |
2777 | if (mddev->sync_thread || |
2778 | test_bit(MD_RECOVERY_RUNNING,&mddev->recovery)) | |
c6563a8c N |
2779 | return -EBUSY; |
2780 | if (new_offset == rdev->data_offset) | |
2781 | /* reset is always permitted */ | |
2782 | ; | |
2783 | else if (new_offset > rdev->data_offset) { | |
2784 | /* must not push array size beyond rdev_sectors */ | |
2785 | if (new_offset - rdev->data_offset | |
2786 | + mddev->dev_sectors > rdev->sectors) | |
2787 | return -E2BIG; | |
2788 | } | |
2789 | /* Metadata worries about other space details. */ | |
2790 | ||
2791 | /* decreasing the offset is inconsistent with a backwards | |
2792 | * reshape. | |
2793 | */ | |
2794 | if (new_offset < rdev->data_offset && | |
2795 | mddev->reshape_backwards) | |
2796 | return -EINVAL; | |
2797 | /* Increasing offset is inconsistent with forwards | |
2798 | * reshape. reshape_direction should be set to | |
2799 | * 'backwards' first. | |
2800 | */ | |
2801 | if (new_offset > rdev->data_offset && | |
2802 | !mddev->reshape_backwards) | |
2803 | return -EINVAL; | |
2804 | ||
2805 | if (mddev->pers && mddev->persistent && | |
2806 | !super_types[mddev->major_version] | |
2807 | .allow_new_offset(rdev, new_offset)) | |
2808 | return -E2BIG; | |
2809 | rdev->new_data_offset = new_offset; | |
2810 | if (new_offset > rdev->data_offset) | |
2811 | mddev->reshape_backwards = 1; | |
2812 | else if (new_offset < rdev->data_offset) | |
2813 | mddev->reshape_backwards = 0; | |
2814 | ||
2815 | return len; | |
2816 | } | |
2817 | static struct rdev_sysfs_entry rdev_new_offset = | |
2818 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
2819 | ||
83303b61 | 2820 | static ssize_t |
3cb03002 | 2821 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 2822 | { |
dd8ac336 | 2823 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2824 | } |
2825 | ||
c5d79adb N |
2826 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2827 | { | |
2828 | /* check if two start/length pairs overlap */ | |
2829 | if (s1+l1 <= s2) | |
2830 | return 0; | |
2831 | if (s2+l2 <= s1) | |
2832 | return 0; | |
2833 | return 1; | |
2834 | } | |
2835 | ||
b522adcd DW |
2836 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2837 | { | |
2838 | unsigned long long blocks; | |
2839 | sector_t new; | |
2840 | ||
b29bebd6 | 2841 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
2842 | return -EINVAL; |
2843 | ||
2844 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2845 | return -EINVAL; /* sector conversion overflow */ | |
2846 | ||
2847 | new = blocks * 2; | |
2848 | if (new != blocks * 2) | |
2849 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2850 | ||
2851 | *sectors = new; | |
2852 | return 0; | |
2853 | } | |
2854 | ||
83303b61 | 2855 | static ssize_t |
3cb03002 | 2856 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 2857 | { |
fd01b88c | 2858 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 2859 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2860 | sector_t sectors; |
27c529bb | 2861 | |
b522adcd | 2862 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2863 | return -EINVAL; |
c6563a8c N |
2864 | if (rdev->data_offset != rdev->new_data_offset) |
2865 | return -EINVAL; /* too confusing */ | |
0cd17fec | 2866 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2867 | if (my_mddev->persistent) { |
dd8ac336 AN |
2868 | sectors = super_types[my_mddev->major_version]. |
2869 | rdev_size_change(rdev, sectors); | |
2870 | if (!sectors) | |
0cd17fec | 2871 | return -EBUSY; |
dd8ac336 | 2872 | } else if (!sectors) |
77304d2a | 2873 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 2874 | rdev->data_offset; |
a6468539 N |
2875 | if (!my_mddev->pers->resize) |
2876 | /* Cannot change size for RAID0 or Linear etc */ | |
2877 | return -EINVAL; | |
0cd17fec | 2878 | } |
dd8ac336 | 2879 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2880 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2881 | |
dd8ac336 AN |
2882 | rdev->sectors = sectors; |
2883 | if (sectors > oldsectors && my_mddev->external) { | |
8b1afc3d N |
2884 | /* Need to check that all other rdevs with the same |
2885 | * ->bdev do not overlap. 'rcu' is sufficient to walk | |
2886 | * the rdev lists safely. | |
2887 | * This check does not provide a hard guarantee, it | |
2888 | * just helps avoid dangerous mistakes. | |
c5d79adb | 2889 | */ |
fd01b88c | 2890 | struct mddev *mddev; |
c5d79adb | 2891 | int overlap = 0; |
159ec1fc | 2892 | struct list_head *tmp; |
c5d79adb | 2893 | |
8b1afc3d | 2894 | rcu_read_lock(); |
29ac4aa3 | 2895 | for_each_mddev(mddev, tmp) { |
3cb03002 | 2896 | struct md_rdev *rdev2; |
c5d79adb | 2897 | |
dafb20fa | 2898 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
2899 | if (rdev->bdev == rdev2->bdev && |
2900 | rdev != rdev2 && | |
2901 | overlaps(rdev->data_offset, rdev->sectors, | |
2902 | rdev2->data_offset, | |
2903 | rdev2->sectors)) { | |
c5d79adb N |
2904 | overlap = 1; |
2905 | break; | |
2906 | } | |
c5d79adb N |
2907 | if (overlap) { |
2908 | mddev_put(mddev); | |
2909 | break; | |
2910 | } | |
2911 | } | |
8b1afc3d | 2912 | rcu_read_unlock(); |
c5d79adb N |
2913 | if (overlap) { |
2914 | /* Someone else could have slipped in a size | |
2915 | * change here, but doing so is just silly. | |
dd8ac336 | 2916 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2917 | * safe, and trust userspace not to race with |
2918 | * itself | |
2919 | */ | |
dd8ac336 | 2920 | rdev->sectors = oldsectors; |
c5d79adb N |
2921 | return -EBUSY; |
2922 | } | |
2923 | } | |
83303b61 N |
2924 | return len; |
2925 | } | |
2926 | ||
2927 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2928 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2929 | |
3cb03002 | 2930 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
2931 | { |
2932 | unsigned long long recovery_start = rdev->recovery_offset; | |
2933 | ||
2934 | if (test_bit(In_sync, &rdev->flags) || | |
2935 | recovery_start == MaxSector) | |
2936 | return sprintf(page, "none\n"); | |
2937 | ||
2938 | return sprintf(page, "%llu\n", recovery_start); | |
2939 | } | |
2940 | ||
3cb03002 | 2941 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
2942 | { |
2943 | unsigned long long recovery_start; | |
2944 | ||
2945 | if (cmd_match(buf, "none")) | |
2946 | recovery_start = MaxSector; | |
b29bebd6 | 2947 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
2948 | return -EINVAL; |
2949 | ||
2950 | if (rdev->mddev->pers && | |
2951 | rdev->raid_disk >= 0) | |
2952 | return -EBUSY; | |
2953 | ||
2954 | rdev->recovery_offset = recovery_start; | |
2955 | if (recovery_start == MaxSector) | |
2956 | set_bit(In_sync, &rdev->flags); | |
2957 | else | |
2958 | clear_bit(In_sync, &rdev->flags); | |
2959 | return len; | |
2960 | } | |
2961 | ||
2962 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2963 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2964 | ||
16c791a5 N |
2965 | static ssize_t |
2966 | badblocks_show(struct badblocks *bb, char *page, int unack); | |
2967 | static ssize_t | |
2968 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack); | |
2969 | ||
3cb03002 | 2970 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
2971 | { |
2972 | return badblocks_show(&rdev->badblocks, page, 0); | |
2973 | } | |
3cb03002 | 2974 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 2975 | { |
de393cde N |
2976 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
2977 | /* Maybe that ack was all we needed */ | |
2978 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
2979 | wake_up(&rdev->blocked_wait); | |
2980 | return rv; | |
16c791a5 N |
2981 | } |
2982 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
2983 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
2984 | ||
3cb03002 | 2985 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
2986 | { |
2987 | return badblocks_show(&rdev->badblocks, page, 1); | |
2988 | } | |
3cb03002 | 2989 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
2990 | { |
2991 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
2992 | } | |
2993 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
2994 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
2995 | ||
86e6ffdd N |
2996 | static struct attribute *rdev_default_attrs[] = { |
2997 | &rdev_state.attr, | |
4dbcdc75 | 2998 | &rdev_errors.attr, |
014236d2 | 2999 | &rdev_slot.attr, |
93c8cad0 | 3000 | &rdev_offset.attr, |
c6563a8c | 3001 | &rdev_new_offset.attr, |
83303b61 | 3002 | &rdev_size.attr, |
06e3c817 | 3003 | &rdev_recovery_start.attr, |
16c791a5 N |
3004 | &rdev_bad_blocks.attr, |
3005 | &rdev_unack_bad_blocks.attr, | |
86e6ffdd N |
3006 | NULL, |
3007 | }; | |
3008 | static ssize_t | |
3009 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3010 | { | |
3011 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3012 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
86e6ffdd N |
3013 | |
3014 | if (!entry->show) | |
3015 | return -EIO; | |
758bfc8a N |
3016 | if (!rdev->mddev) |
3017 | return -EBUSY; | |
3018 | return entry->show(rdev, page); | |
86e6ffdd N |
3019 | } |
3020 | ||
3021 | static ssize_t | |
3022 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
3023 | const char *page, size_t length) | |
3024 | { | |
3025 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3026 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 3027 | ssize_t rv; |
fd01b88c | 3028 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
3029 | |
3030 | if (!entry->store) | |
3031 | return -EIO; | |
67463acb N |
3032 | if (!capable(CAP_SYS_ADMIN)) |
3033 | return -EACCES; | |
27c529bb | 3034 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 3035 | if (!rv) { |
27c529bb N |
3036 | if (rdev->mddev == NULL) |
3037 | rv = -EBUSY; | |
3038 | else | |
3039 | rv = entry->store(rdev, page, length); | |
6a51830e | 3040 | mddev_unlock(mddev); |
ca388059 N |
3041 | } |
3042 | return rv; | |
86e6ffdd N |
3043 | } |
3044 | ||
3045 | static void rdev_free(struct kobject *ko) | |
3046 | { | |
3cb03002 | 3047 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
3048 | kfree(rdev); |
3049 | } | |
52cf25d0 | 3050 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
3051 | .show = rdev_attr_show, |
3052 | .store = rdev_attr_store, | |
3053 | }; | |
3054 | static struct kobj_type rdev_ktype = { | |
3055 | .release = rdev_free, | |
3056 | .sysfs_ops = &rdev_sysfs_ops, | |
3057 | .default_attrs = rdev_default_attrs, | |
3058 | }; | |
3059 | ||
3cb03002 | 3060 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
3061 | { |
3062 | rdev->desc_nr = -1; | |
3063 | rdev->saved_raid_disk = -1; | |
3064 | rdev->raid_disk = -1; | |
3065 | rdev->flags = 0; | |
3066 | rdev->data_offset = 0; | |
c6563a8c | 3067 | rdev->new_data_offset = 0; |
e8bb9a83 N |
3068 | rdev->sb_events = 0; |
3069 | rdev->last_read_error.tv_sec = 0; | |
3070 | rdev->last_read_error.tv_nsec = 0; | |
2699b672 N |
3071 | rdev->sb_loaded = 0; |
3072 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3073 | atomic_set(&rdev->nr_pending, 0); |
3074 | atomic_set(&rdev->read_errors, 0); | |
3075 | atomic_set(&rdev->corrected_errors, 0); | |
3076 | ||
3077 | INIT_LIST_HEAD(&rdev->same_set); | |
3078 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3079 | |
3080 | /* Add space to store bad block list. | |
3081 | * This reserves the space even on arrays where it cannot | |
3082 | * be used - I wonder if that matters | |
3083 | */ | |
3084 | rdev->badblocks.count = 0; | |
486adf72 | 3085 | rdev->badblocks.shift = -1; /* disabled until explicitly enabled */ |
2230dfe4 N |
3086 | rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3087 | seqlock_init(&rdev->badblocks.lock); | |
3088 | if (rdev->badblocks.page == NULL) | |
3089 | return -ENOMEM; | |
3090 | ||
3091 | return 0; | |
e8bb9a83 N |
3092 | } |
3093 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3094 | /* |
3095 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3096 | * | |
3097 | * mark the device faulty if: | |
3098 | * | |
3099 | * - the device is nonexistent (zero size) | |
3100 | * - the device has no valid superblock | |
3101 | * | |
3102 | * a faulty rdev _never_ has rdev->sb set. | |
3103 | */ | |
3cb03002 | 3104 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3105 | { |
3106 | char b[BDEVNAME_SIZE]; | |
3107 | int err; | |
3cb03002 | 3108 | struct md_rdev *rdev; |
1da177e4 LT |
3109 | sector_t size; |
3110 | ||
9ffae0cf | 3111 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
3112 | if (!rdev) { |
3113 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
3114 | return ERR_PTR(-ENOMEM); | |
3115 | } | |
1da177e4 | 3116 | |
2230dfe4 N |
3117 | err = md_rdev_init(rdev); |
3118 | if (err) | |
3119 | goto abort_free; | |
3120 | err = alloc_disk_sb(rdev); | |
3121 | if (err) | |
1da177e4 LT |
3122 | goto abort_free; |
3123 | ||
c5d79adb | 3124 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3125 | if (err) |
3126 | goto abort_free; | |
3127 | ||
f9cb074b | 3128 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3129 | |
77304d2a | 3130 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 | 3131 | if (!size) { |
f72ffdd6 | 3132 | printk(KERN_WARNING |
1da177e4 LT |
3133 | "md: %s has zero or unknown size, marking faulty!\n", |
3134 | bdevname(rdev->bdev,b)); | |
3135 | err = -EINVAL; | |
3136 | goto abort_free; | |
3137 | } | |
3138 | ||
3139 | if (super_format >= 0) { | |
3140 | err = super_types[super_format]. | |
3141 | load_super(rdev, NULL, super_minor); | |
3142 | if (err == -EINVAL) { | |
df968c4e N |
3143 | printk(KERN_WARNING |
3144 | "md: %s does not have a valid v%d.%d " | |
3145 | "superblock, not importing!\n", | |
3146 | bdevname(rdev->bdev,b), | |
3147 | super_format, super_minor); | |
1da177e4 LT |
3148 | goto abort_free; |
3149 | } | |
3150 | if (err < 0) { | |
f72ffdd6 | 3151 | printk(KERN_WARNING |
1da177e4 LT |
3152 | "md: could not read %s's sb, not importing!\n", |
3153 | bdevname(rdev->bdev,b)); | |
3154 | goto abort_free; | |
3155 | } | |
3156 | } | |
6bfe0b49 | 3157 | |
1da177e4 LT |
3158 | return rdev; |
3159 | ||
3160 | abort_free: | |
2699b672 N |
3161 | if (rdev->bdev) |
3162 | unlock_rdev(rdev); | |
545c8795 | 3163 | md_rdev_clear(rdev); |
1da177e4 LT |
3164 | kfree(rdev); |
3165 | return ERR_PTR(err); | |
3166 | } | |
3167 | ||
3168 | /* | |
3169 | * Check a full RAID array for plausibility | |
3170 | */ | |
3171 | ||
f72ffdd6 | 3172 | static void analyze_sbs(struct mddev *mddev) |
1da177e4 LT |
3173 | { |
3174 | int i; | |
3cb03002 | 3175 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3176 | char b[BDEVNAME_SIZE]; |
3177 | ||
3178 | freshest = NULL; | |
dafb20fa | 3179 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3180 | switch (super_types[mddev->major_version]. |
3181 | load_super(rdev, freshest, mddev->minor_version)) { | |
3182 | case 1: | |
3183 | freshest = rdev; | |
3184 | break; | |
3185 | case 0: | |
3186 | break; | |
3187 | default: | |
3188 | printk( KERN_ERR \ | |
3189 | "md: fatal superblock inconsistency in %s" | |
f72ffdd6 | 3190 | " -- removing from array\n", |
1da177e4 | 3191 | bdevname(rdev->bdev,b)); |
fb56dfef | 3192 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3193 | } |
3194 | ||
1da177e4 LT |
3195 | super_types[mddev->major_version]. |
3196 | validate_super(mddev, freshest); | |
3197 | ||
3198 | i = 0; | |
dafb20fa | 3199 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3200 | if (mddev->max_disks && |
3201 | (rdev->desc_nr >= mddev->max_disks || | |
3202 | i > mddev->max_disks)) { | |
de01dfad N |
3203 | printk(KERN_WARNING |
3204 | "md: %s: %s: only %d devices permitted\n", | |
3205 | mdname(mddev), bdevname(rdev->bdev, b), | |
3206 | mddev->max_disks); | |
fb56dfef | 3207 | md_kick_rdev_from_array(rdev); |
de01dfad N |
3208 | continue; |
3209 | } | |
1aee41f6 | 3210 | if (rdev != freshest) { |
1da177e4 LT |
3211 | if (super_types[mddev->major_version]. |
3212 | validate_super(mddev, rdev)) { | |
3213 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
3214 | " from array!\n", | |
3215 | bdevname(rdev->bdev,b)); | |
fb56dfef | 3216 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3217 | continue; |
3218 | } | |
1aee41f6 GR |
3219 | /* No device should have a Candidate flag |
3220 | * when reading devices | |
3221 | */ | |
3222 | if (test_bit(Candidate, &rdev->flags)) { | |
3223 | pr_info("md: kicking Cluster Candidate %s from array!\n", | |
3224 | bdevname(rdev->bdev, b)); | |
fb56dfef | 3225 | md_kick_rdev_from_array(rdev); |
1aee41f6 GR |
3226 | } |
3227 | } | |
1da177e4 LT |
3228 | if (mddev->level == LEVEL_MULTIPATH) { |
3229 | rdev->desc_nr = i++; | |
3230 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3231 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 3232 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
3233 | rdev->raid_disk = -1; |
3234 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3235 | } |
3236 | } | |
1da177e4 LT |
3237 | } |
3238 | ||
72e02075 N |
3239 | /* Read a fixed-point number. |
3240 | * Numbers in sysfs attributes should be in "standard" units where | |
3241 | * possible, so time should be in seconds. | |
f72ffdd6 | 3242 | * However we internally use a a much smaller unit such as |
72e02075 N |
3243 | * milliseconds or jiffies. |
3244 | * This function takes a decimal number with a possible fractional | |
3245 | * component, and produces an integer which is the result of | |
3246 | * multiplying that number by 10^'scale'. | |
3247 | * all without any floating-point arithmetic. | |
3248 | */ | |
3249 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3250 | { | |
3251 | unsigned long result = 0; | |
3252 | long decimals = -1; | |
3253 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3254 | if (*cp == '.') | |
3255 | decimals = 0; | |
3256 | else if (decimals < scale) { | |
3257 | unsigned int value; | |
3258 | value = *cp - '0'; | |
3259 | result = result * 10 + value; | |
3260 | if (decimals >= 0) | |
3261 | decimals++; | |
3262 | } | |
3263 | cp++; | |
3264 | } | |
3265 | if (*cp == '\n') | |
3266 | cp++; | |
3267 | if (*cp) | |
3268 | return -EINVAL; | |
3269 | if (decimals < 0) | |
3270 | decimals = 0; | |
3271 | while (decimals < scale) { | |
3272 | result *= 10; | |
3273 | decimals ++; | |
3274 | } | |
3275 | *res = result; | |
3276 | return 0; | |
3277 | } | |
3278 | ||
19052c0e N |
3279 | static void md_safemode_timeout(unsigned long data); |
3280 | ||
16f17b39 | 3281 | static ssize_t |
fd01b88c | 3282 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3283 | { |
3284 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3285 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3286 | } | |
3287 | static ssize_t | |
fd01b88c | 3288 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3289 | { |
16f17b39 | 3290 | unsigned long msec; |
97ce0a7f | 3291 | |
72e02075 | 3292 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3293 | return -EINVAL; |
16f17b39 N |
3294 | if (msec == 0) |
3295 | mddev->safemode_delay = 0; | |
3296 | else { | |
19052c0e | 3297 | unsigned long old_delay = mddev->safemode_delay; |
1b30e66f N |
3298 | unsigned long new_delay = (msec*HZ)/1000; |
3299 | ||
3300 | if (new_delay == 0) | |
3301 | new_delay = 1; | |
3302 | mddev->safemode_delay = new_delay; | |
3303 | if (new_delay < old_delay || old_delay == 0) | |
3304 | mod_timer(&mddev->safemode_timer, jiffies+1); | |
16f17b39 N |
3305 | } |
3306 | return len; | |
3307 | } | |
3308 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3309 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3310 | |
eae1701f | 3311 | static ssize_t |
fd01b88c | 3312 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3313 | { |
36d091f4 N |
3314 | struct md_personality *p; |
3315 | int ret; | |
3316 | spin_lock(&mddev->lock); | |
3317 | p = mddev->pers; | |
d9d166c2 | 3318 | if (p) |
36d091f4 | 3319 | ret = sprintf(page, "%s\n", p->name); |
d9d166c2 | 3320 | else if (mddev->clevel[0]) |
36d091f4 | 3321 | ret = sprintf(page, "%s\n", mddev->clevel); |
d9d166c2 | 3322 | else if (mddev->level != LEVEL_NONE) |
36d091f4 | 3323 | ret = sprintf(page, "%d\n", mddev->level); |
d9d166c2 | 3324 | else |
36d091f4 N |
3325 | ret = 0; |
3326 | spin_unlock(&mddev->lock); | |
3327 | return ret; | |
eae1701f N |
3328 | } |
3329 | ||
d9d166c2 | 3330 | static ssize_t |
fd01b88c | 3331 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3332 | { |
f2859af6 | 3333 | char clevel[16]; |
6791875e N |
3334 | ssize_t rv; |
3335 | size_t slen = len; | |
db721d32 | 3336 | struct md_personality *pers, *oldpers; |
f2859af6 | 3337 | long level; |
db721d32 | 3338 | void *priv, *oldpriv; |
3cb03002 | 3339 | struct md_rdev *rdev; |
245f46c2 | 3340 | |
6791875e N |
3341 | if (slen == 0 || slen >= sizeof(clevel)) |
3342 | return -EINVAL; | |
3343 | ||
3344 | rv = mddev_lock(mddev); | |
3345 | if (rv) | |
3346 | return rv; | |
3347 | ||
245f46c2 | 3348 | if (mddev->pers == NULL) { |
6791875e N |
3349 | strncpy(mddev->clevel, buf, slen); |
3350 | if (mddev->clevel[slen-1] == '\n') | |
3351 | slen--; | |
3352 | mddev->clevel[slen] = 0; | |
245f46c2 | 3353 | mddev->level = LEVEL_NONE; |
6791875e N |
3354 | rv = len; |
3355 | goto out_unlock; | |
245f46c2 | 3356 | } |
6791875e | 3357 | rv = -EROFS; |
bd8839e0 | 3358 | if (mddev->ro) |
6791875e | 3359 | goto out_unlock; |
245f46c2 N |
3360 | |
3361 | /* request to change the personality. Need to ensure: | |
3362 | * - array is not engaged in resync/recovery/reshape | |
3363 | * - old personality can be suspended | |
3364 | * - new personality will access other array. | |
3365 | */ | |
3366 | ||
6791875e | 3367 | rv = -EBUSY; |
bb4f1e9d | 3368 | if (mddev->sync_thread || |
f851b60d | 3369 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
bb4f1e9d N |
3370 | mddev->reshape_position != MaxSector || |
3371 | mddev->sysfs_active) | |
6791875e | 3372 | goto out_unlock; |
245f46c2 | 3373 | |
6791875e | 3374 | rv = -EINVAL; |
245f46c2 N |
3375 | if (!mddev->pers->quiesce) { |
3376 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3377 | mdname(mddev), mddev->pers->name); | |
6791875e | 3378 | goto out_unlock; |
245f46c2 N |
3379 | } |
3380 | ||
3381 | /* Now find the new personality */ | |
6791875e N |
3382 | strncpy(clevel, buf, slen); |
3383 | if (clevel[slen-1] == '\n') | |
3384 | slen--; | |
3385 | clevel[slen] = 0; | |
b29bebd6 | 3386 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3387 | level = LEVEL_NONE; |
245f46c2 | 3388 | |
f2859af6 DW |
3389 | if (request_module("md-%s", clevel) != 0) |
3390 | request_module("md-level-%s", clevel); | |
245f46c2 | 3391 | spin_lock(&pers_lock); |
f2859af6 | 3392 | pers = find_pers(level, clevel); |
245f46c2 N |
3393 | if (!pers || !try_module_get(pers->owner)) { |
3394 | spin_unlock(&pers_lock); | |
f2859af6 | 3395 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
6791875e N |
3396 | rv = -EINVAL; |
3397 | goto out_unlock; | |
245f46c2 N |
3398 | } |
3399 | spin_unlock(&pers_lock); | |
3400 | ||
3401 | if (pers == mddev->pers) { | |
3402 | /* Nothing to do! */ | |
3403 | module_put(pers->owner); | |
6791875e N |
3404 | rv = len; |
3405 | goto out_unlock; | |
245f46c2 N |
3406 | } |
3407 | if (!pers->takeover) { | |
3408 | module_put(pers->owner); | |
3409 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3410 | mdname(mddev), clevel); |
6791875e N |
3411 | rv = -EINVAL; |
3412 | goto out_unlock; | |
245f46c2 N |
3413 | } |
3414 | ||
dafb20fa | 3415 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3416 | rdev->new_raid_disk = rdev->raid_disk; |
3417 | ||
245f46c2 N |
3418 | /* ->takeover must set new_* and/or delta_disks |
3419 | * if it succeeds, and may set them when it fails. | |
3420 | */ | |
3421 | priv = pers->takeover(mddev); | |
3422 | if (IS_ERR(priv)) { | |
3423 | mddev->new_level = mddev->level; | |
3424 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3425 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3426 | mddev->raid_disks -= mddev->delta_disks; |
3427 | mddev->delta_disks = 0; | |
2c810cdd | 3428 | mddev->reshape_backwards = 0; |
245f46c2 N |
3429 | module_put(pers->owner); |
3430 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3431 | mdname(mddev), clevel); |
6791875e N |
3432 | rv = PTR_ERR(priv); |
3433 | goto out_unlock; | |
245f46c2 N |
3434 | } |
3435 | ||
3436 | /* Looks like we have a winner */ | |
3437 | mddev_suspend(mddev); | |
5aa61f42 | 3438 | mddev_detach(mddev); |
36d091f4 N |
3439 | |
3440 | spin_lock(&mddev->lock); | |
db721d32 N |
3441 | oldpers = mddev->pers; |
3442 | oldpriv = mddev->private; | |
3443 | mddev->pers = pers; | |
3444 | mddev->private = priv; | |
3445 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3446 | mddev->level = mddev->new_level; | |
3447 | mddev->layout = mddev->new_layout; | |
3448 | mddev->chunk_sectors = mddev->new_chunk_sectors; | |
3449 | mddev->delta_disks = 0; | |
3450 | mddev->reshape_backwards = 0; | |
3451 | mddev->degraded = 0; | |
36d091f4 | 3452 | spin_unlock(&mddev->lock); |
db721d32 N |
3453 | |
3454 | if (oldpers->sync_request == NULL && | |
3455 | mddev->external) { | |
3456 | /* We are converting from a no-redundancy array | |
3457 | * to a redundancy array and metadata is managed | |
3458 | * externally so we need to be sure that writes | |
3459 | * won't block due to a need to transition | |
3460 | * clean->dirty | |
3461 | * until external management is started. | |
3462 | */ | |
3463 | mddev->in_sync = 0; | |
3464 | mddev->safemode_delay = 0; | |
3465 | mddev->safemode = 0; | |
3466 | } | |
f72ffdd6 | 3467 | |
db721d32 N |
3468 | oldpers->free(mddev, oldpriv); |
3469 | ||
3470 | if (oldpers->sync_request == NULL && | |
a64c876f N |
3471 | pers->sync_request != NULL) { |
3472 | /* need to add the md_redundancy_group */ | |
3473 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3474 | printk(KERN_WARNING | |
3475 | "md: cannot register extra attributes for %s\n", | |
3476 | mdname(mddev)); | |
388975cc | 3477 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
f72ffdd6 | 3478 | } |
db721d32 | 3479 | if (oldpers->sync_request != NULL && |
a64c876f N |
3480 | pers->sync_request == NULL) { |
3481 | /* need to remove the md_redundancy_group */ | |
3482 | if (mddev->to_remove == NULL) | |
3483 | mddev->to_remove = &md_redundancy_group; | |
3484 | } | |
3485 | ||
dafb20fa | 3486 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3487 | if (rdev->raid_disk < 0) |
3488 | continue; | |
bf2cb0da | 3489 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3490 | rdev->new_raid_disk = -1; |
3491 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3492 | continue; | |
36fad858 | 3493 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3494 | } |
dafb20fa | 3495 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3496 | if (rdev->raid_disk < 0) |
3497 | continue; | |
3498 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3499 | continue; | |
3500 | rdev->raid_disk = rdev->new_raid_disk; | |
3501 | if (rdev->raid_disk < 0) | |
3a981b03 | 3502 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3503 | else { |
36fad858 NK |
3504 | if (sysfs_link_rdev(mddev, rdev)) |
3505 | printk(KERN_WARNING "md: cannot register rd%d" | |
3506 | " for %s after level change\n", | |
3507 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3508 | } |
e93f68a1 N |
3509 | } |
3510 | ||
db721d32 | 3511 | if (pers->sync_request == NULL) { |
9af204cf TM |
3512 | /* this is now an array without redundancy, so |
3513 | * it must always be in_sync | |
3514 | */ | |
3515 | mddev->in_sync = 1; | |
3516 | del_timer_sync(&mddev->safemode_timer); | |
3517 | } | |
02e5f5c0 | 3518 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3519 | pers->run(mddev); |
245f46c2 | 3520 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
47525e59 | 3521 | mddev_resume(mddev); |
830778a1 N |
3522 | if (!mddev->thread) |
3523 | md_update_sb(mddev, 1); | |
5cac7861 | 3524 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3525 | md_new_event(mddev); |
6791875e N |
3526 | rv = len; |
3527 | out_unlock: | |
3528 | mddev_unlock(mddev); | |
d9d166c2 N |
3529 | return rv; |
3530 | } | |
3531 | ||
3532 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3533 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3534 | |
d4dbd025 | 3535 | static ssize_t |
fd01b88c | 3536 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3537 | { |
3538 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3539 | if (mddev->reshape_position != MaxSector && |
3540 | mddev->layout != mddev->new_layout) | |
3541 | return sprintf(page, "%d (%d)\n", | |
3542 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3543 | return sprintf(page, "%d\n", mddev->layout); |
3544 | } | |
3545 | ||
3546 | static ssize_t | |
fd01b88c | 3547 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 | 3548 | { |
4c9309c0 | 3549 | unsigned int n; |
6791875e | 3550 | int err; |
d4dbd025 | 3551 | |
4c9309c0 AD |
3552 | err = kstrtouint(buf, 10, &n); |
3553 | if (err < 0) | |
3554 | return err; | |
6791875e N |
3555 | err = mddev_lock(mddev); |
3556 | if (err) | |
3557 | return err; | |
d4dbd025 | 3558 | |
b3546035 | 3559 | if (mddev->pers) { |
50ac168a | 3560 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3561 | err = -EBUSY; |
3562 | else if (mddev->ro) | |
3563 | err = -EROFS; | |
3564 | else { | |
3565 | mddev->new_layout = n; | |
3566 | err = mddev->pers->check_reshape(mddev); | |
3567 | if (err) | |
3568 | mddev->new_layout = mddev->layout; | |
597a711b | 3569 | } |
b3546035 | 3570 | } else { |
08a02ecd | 3571 | mddev->new_layout = n; |
b3546035 N |
3572 | if (mddev->reshape_position == MaxSector) |
3573 | mddev->layout = n; | |
3574 | } | |
6791875e N |
3575 | mddev_unlock(mddev); |
3576 | return err ?: len; | |
d4dbd025 N |
3577 | } |
3578 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3579 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 | 3580 | |
eae1701f | 3581 | static ssize_t |
fd01b88c | 3582 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3583 | { |
bb636547 N |
3584 | if (mddev->raid_disks == 0) |
3585 | return 0; | |
08a02ecd N |
3586 | if (mddev->reshape_position != MaxSector && |
3587 | mddev->delta_disks != 0) | |
3588 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3589 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3590 | return sprintf(page, "%d\n", mddev->raid_disks); |
3591 | } | |
3592 | ||
fd01b88c | 3593 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3594 | |
3595 | static ssize_t | |
fd01b88c | 3596 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3597 | { |
4c9309c0 | 3598 | unsigned int n; |
6791875e | 3599 | int err; |
da943b99 | 3600 | |
4c9309c0 AD |
3601 | err = kstrtouint(buf, 10, &n); |
3602 | if (err < 0) | |
3603 | return err; | |
da943b99 | 3604 | |
6791875e N |
3605 | err = mddev_lock(mddev); |
3606 | if (err) | |
3607 | return err; | |
da943b99 | 3608 | if (mddev->pers) |
6791875e | 3609 | err = update_raid_disks(mddev, n); |
08a02ecd | 3610 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3611 | struct md_rdev *rdev; |
08a02ecd | 3612 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c | 3613 | |
6791875e | 3614 | err = -EINVAL; |
c6563a8c N |
3615 | rdev_for_each(rdev, mddev) { |
3616 | if (olddisks < n && | |
3617 | rdev->data_offset < rdev->new_data_offset) | |
6791875e | 3618 | goto out_unlock; |
c6563a8c N |
3619 | if (olddisks > n && |
3620 | rdev->data_offset > rdev->new_data_offset) | |
6791875e | 3621 | goto out_unlock; |
c6563a8c | 3622 | } |
6791875e | 3623 | err = 0; |
08a02ecd N |
3624 | mddev->delta_disks = n - olddisks; |
3625 | mddev->raid_disks = n; | |
2c810cdd | 3626 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3627 | } else |
da943b99 | 3628 | mddev->raid_disks = n; |
6791875e N |
3629 | out_unlock: |
3630 | mddev_unlock(mddev); | |
3631 | return err ? err : len; | |
da943b99 N |
3632 | } |
3633 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3634 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3635 | |
3b34380a | 3636 | static ssize_t |
fd01b88c | 3637 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3638 | { |
08a02ecd | 3639 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3640 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3641 | return sprintf(page, "%d (%d)\n", | |
3642 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3643 | mddev->chunk_sectors << 9); |
3644 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3645 | } |
3646 | ||
3647 | static ssize_t | |
fd01b88c | 3648 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3649 | { |
4c9309c0 | 3650 | unsigned long n; |
6791875e | 3651 | int err; |
3b34380a | 3652 | |
4c9309c0 AD |
3653 | err = kstrtoul(buf, 10, &n); |
3654 | if (err < 0) | |
3655 | return err; | |
3b34380a | 3656 | |
6791875e N |
3657 | err = mddev_lock(mddev); |
3658 | if (err) | |
3659 | return err; | |
b3546035 | 3660 | if (mddev->pers) { |
50ac168a | 3661 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3662 | err = -EBUSY; |
3663 | else if (mddev->ro) | |
3664 | err = -EROFS; | |
3665 | else { | |
3666 | mddev->new_chunk_sectors = n >> 9; | |
3667 | err = mddev->pers->check_reshape(mddev); | |
3668 | if (err) | |
3669 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
597a711b | 3670 | } |
b3546035 | 3671 | } else { |
664e7c41 | 3672 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3673 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3674 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3675 | } |
6791875e N |
3676 | mddev_unlock(mddev); |
3677 | return err ?: len; | |
3b34380a N |
3678 | } |
3679 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3680 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3681 | |
a94213b1 | 3682 | static ssize_t |
fd01b88c | 3683 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 3684 | { |
d1a7c503 N |
3685 | if (mddev->recovery_cp == MaxSector) |
3686 | return sprintf(page, "none\n"); | |
a94213b1 N |
3687 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3688 | } | |
3689 | ||
3690 | static ssize_t | |
fd01b88c | 3691 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 3692 | { |
4c9309c0 | 3693 | unsigned long long n; |
6791875e | 3694 | int err; |
4c9309c0 AD |
3695 | |
3696 | if (cmd_match(buf, "none")) | |
3697 | n = MaxSector; | |
3698 | else { | |
3699 | err = kstrtoull(buf, 10, &n); | |
3700 | if (err < 0) | |
3701 | return err; | |
3702 | if (n != (sector_t)n) | |
3703 | return -EINVAL; | |
3704 | } | |
a94213b1 | 3705 | |
6791875e N |
3706 | err = mddev_lock(mddev); |
3707 | if (err) | |
3708 | return err; | |
b098636c | 3709 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6791875e | 3710 | err = -EBUSY; |
a94213b1 | 3711 | |
6791875e N |
3712 | if (!err) { |
3713 | mddev->recovery_cp = n; | |
3714 | if (mddev->pers) | |
3715 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
3716 | } | |
3717 | mddev_unlock(mddev); | |
3718 | return err ?: len; | |
a94213b1 N |
3719 | } |
3720 | static struct md_sysfs_entry md_resync_start = | |
750f199e N |
3721 | __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR, |
3722 | resync_start_show, resync_start_store); | |
a94213b1 | 3723 | |
9e653b63 N |
3724 | /* |
3725 | * The array state can be: | |
3726 | * | |
3727 | * clear | |
3728 | * No devices, no size, no level | |
3729 | * Equivalent to STOP_ARRAY ioctl | |
3730 | * inactive | |
3731 | * May have some settings, but array is not active | |
3732 | * all IO results in error | |
3733 | * When written, doesn't tear down array, but just stops it | |
3734 | * suspended (not supported yet) | |
3735 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3736 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3737 | * readonly |
3738 | * no resync can happen. no superblocks get written. | |
3739 | * write requests fail | |
3740 | * read-auto | |
3741 | * like readonly, but behaves like 'clean' on a write request. | |
3742 | * | |
3743 | * clean - no pending writes, but otherwise active. | |
3744 | * When written to inactive array, starts without resync | |
3745 | * If a write request arrives then | |
3746 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3747 | * if not known, block and switch to write-pending | |
3748 | * If written to an active array that has pending writes, then fails. | |
3749 | * active | |
3750 | * fully active: IO and resync can be happening. | |
3751 | * When written to inactive array, starts with resync | |
3752 | * | |
3753 | * write-pending | |
3754 | * clean, but writes are blocked waiting for 'active' to be written. | |
3755 | * | |
3756 | * active-idle | |
3757 | * like active, but no writes have been seen for a while (100msec). | |
3758 | * | |
3759 | */ | |
3760 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3761 | write_pending, active_idle, bad_word}; | |
05381954 | 3762 | static char *array_states[] = { |
9e653b63 N |
3763 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3764 | "write-pending", "active-idle", NULL }; | |
3765 | ||
3766 | static int match_word(const char *word, char **list) | |
3767 | { | |
3768 | int n; | |
3769 | for (n=0; list[n]; n++) | |
3770 | if (cmd_match(word, list[n])) | |
3771 | break; | |
3772 | return n; | |
3773 | } | |
3774 | ||
3775 | static ssize_t | |
fd01b88c | 3776 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
3777 | { |
3778 | enum array_state st = inactive; | |
3779 | ||
3780 | if (mddev->pers) | |
3781 | switch(mddev->ro) { | |
3782 | case 1: | |
3783 | st = readonly; | |
3784 | break; | |
3785 | case 2: | |
3786 | st = read_auto; | |
3787 | break; | |
3788 | case 0: | |
3789 | if (mddev->in_sync) | |
3790 | st = clean; | |
070dc6dd | 3791 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3792 | st = write_pending; |
9e653b63 N |
3793 | else if (mddev->safemode) |
3794 | st = active_idle; | |
3795 | else | |
3796 | st = active; | |
3797 | } | |
3798 | else { | |
3799 | if (list_empty(&mddev->disks) && | |
3800 | mddev->raid_disks == 0 && | |
58c0fed4 | 3801 | mddev->dev_sectors == 0) |
9e653b63 N |
3802 | st = clear; |
3803 | else | |
3804 | st = inactive; | |
3805 | } | |
3806 | return sprintf(page, "%s\n", array_states[st]); | |
3807 | } | |
3808 | ||
f72ffdd6 N |
3809 | static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev); |
3810 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev); | |
3811 | static int do_md_run(struct mddev *mddev); | |
fd01b88c | 3812 | static int restart_array(struct mddev *mddev); |
9e653b63 N |
3813 | |
3814 | static ssize_t | |
fd01b88c | 3815 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 | 3816 | { |
6791875e | 3817 | int err; |
9e653b63 | 3818 | enum array_state st = match_word(buf, array_states); |
6791875e N |
3819 | |
3820 | if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) { | |
3821 | /* don't take reconfig_mutex when toggling between | |
3822 | * clean and active | |
3823 | */ | |
3824 | spin_lock(&mddev->lock); | |
3825 | if (st == active) { | |
3826 | restart_array(mddev); | |
3827 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
3828 | wake_up(&mddev->sb_wait); | |
3829 | err = 0; | |
3830 | } else /* st == clean */ { | |
3831 | restart_array(mddev); | |
3832 | if (atomic_read(&mddev->writes_pending) == 0) { | |
3833 | if (mddev->in_sync == 0) { | |
3834 | mddev->in_sync = 1; | |
3835 | if (mddev->safemode == 1) | |
3836 | mddev->safemode = 0; | |
3837 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
3838 | } | |
3839 | err = 0; | |
3840 | } else | |
3841 | err = -EBUSY; | |
3842 | } | |
3843 | spin_unlock(&mddev->lock); | |
c008f1d3 | 3844 | return err ?: len; |
6791875e N |
3845 | } |
3846 | err = mddev_lock(mddev); | |
3847 | if (err) | |
3848 | return err; | |
3849 | err = -EINVAL; | |
9e653b63 N |
3850 | switch(st) { |
3851 | case bad_word: | |
3852 | break; | |
3853 | case clear: | |
3854 | /* stopping an active array */ | |
a05b7ea0 | 3855 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
3856 | break; |
3857 | case inactive: | |
3858 | /* stopping an active array */ | |
90cf195d | 3859 | if (mddev->pers) |
a05b7ea0 | 3860 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 3861 | else |
e691063a | 3862 | err = 0; /* already inactive */ |
9e653b63 N |
3863 | break; |
3864 | case suspended: | |
3865 | break; /* not supported yet */ | |
3866 | case readonly: | |
3867 | if (mddev->pers) | |
a05b7ea0 | 3868 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
3869 | else { |
3870 | mddev->ro = 1; | |
648b629e | 3871 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3872 | err = do_md_run(mddev); |
3873 | } | |
3874 | break; | |
3875 | case read_auto: | |
9e653b63 | 3876 | if (mddev->pers) { |
80268ee9 | 3877 | if (mddev->ro == 0) |
a05b7ea0 | 3878 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 3879 | else if (mddev->ro == 1) |
648b629e N |
3880 | err = restart_array(mddev); |
3881 | if (err == 0) { | |
3882 | mddev->ro = 2; | |
3883 | set_disk_ro(mddev->gendisk, 0); | |
3884 | } | |
9e653b63 N |
3885 | } else { |
3886 | mddev->ro = 2; | |
3887 | err = do_md_run(mddev); | |
3888 | } | |
3889 | break; | |
3890 | case clean: | |
3891 | if (mddev->pers) { | |
3892 | restart_array(mddev); | |
85572d7c | 3893 | spin_lock(&mddev->lock); |
9e653b63 | 3894 | if (atomic_read(&mddev->writes_pending) == 0) { |
e691063a N |
3895 | if (mddev->in_sync == 0) { |
3896 | mddev->in_sync = 1; | |
31a59e34 N |
3897 | if (mddev->safemode == 1) |
3898 | mddev->safemode = 0; | |
070dc6dd | 3899 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3900 | } |
3901 | err = 0; | |
3902 | } else | |
3903 | err = -EBUSY; | |
85572d7c | 3904 | spin_unlock(&mddev->lock); |
5bf29597 N |
3905 | } else |
3906 | err = -EINVAL; | |
9e653b63 N |
3907 | break; |
3908 | case active: | |
3909 | if (mddev->pers) { | |
3910 | restart_array(mddev); | |
070dc6dd | 3911 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3912 | wake_up(&mddev->sb_wait); |
3913 | err = 0; | |
3914 | } else { | |
3915 | mddev->ro = 0; | |
648b629e | 3916 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3917 | err = do_md_run(mddev); |
3918 | } | |
3919 | break; | |
3920 | case write_pending: | |
3921 | case active_idle: | |
3922 | /* these cannot be set */ | |
3923 | break; | |
3924 | } | |
6791875e N |
3925 | |
3926 | if (!err) { | |
1d23f178 N |
3927 | if (mddev->hold_active == UNTIL_IOCTL) |
3928 | mddev->hold_active = 0; | |
00bcb4ac | 3929 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 3930 | } |
6791875e N |
3931 | mddev_unlock(mddev); |
3932 | return err ?: len; | |
9e653b63 | 3933 | } |
80ca3a44 | 3934 | static struct md_sysfs_entry md_array_state = |
750f199e | 3935 | __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); |
9e653b63 | 3936 | |
1e50915f | 3937 | static ssize_t |
fd01b88c | 3938 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
3939 | return sprintf(page, "%d\n", |
3940 | atomic_read(&mddev->max_corr_read_errors)); | |
3941 | } | |
3942 | ||
3943 | static ssize_t | |
fd01b88c | 3944 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f | 3945 | { |
4c9309c0 AD |
3946 | unsigned int n; |
3947 | int rv; | |
1e50915f | 3948 | |
4c9309c0 AD |
3949 | rv = kstrtouint(buf, 10, &n); |
3950 | if (rv < 0) | |
3951 | return rv; | |
3952 | atomic_set(&mddev->max_corr_read_errors, n); | |
3953 | return len; | |
1e50915f RB |
3954 | } |
3955 | ||
3956 | static struct md_sysfs_entry max_corr_read_errors = | |
3957 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3958 | max_corrected_read_errors_store); | |
3959 | ||
6d7ff738 | 3960 | static ssize_t |
fd01b88c | 3961 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
3962 | { |
3963 | return -EINVAL; | |
3964 | } | |
3965 | ||
3966 | static ssize_t | |
fd01b88c | 3967 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
3968 | { |
3969 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3970 | /* The new device is added to the array. | |
3971 | * If the array has a persistent superblock, we read the | |
3972 | * superblock to initialise info and check validity. | |
3973 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3974 | * which mainly checks size. | |
3975 | */ | |
3976 | char *e; | |
3977 | int major = simple_strtoul(buf, &e, 10); | |
3978 | int minor; | |
3979 | dev_t dev; | |
3cb03002 | 3980 | struct md_rdev *rdev; |
6d7ff738 N |
3981 | int err; |
3982 | ||
3983 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3984 | return -EINVAL; | |
3985 | minor = simple_strtoul(e+1, &e, 10); | |
3986 | if (*e && *e != '\n') | |
3987 | return -EINVAL; | |
3988 | dev = MKDEV(major, minor); | |
3989 | if (major != MAJOR(dev) || | |
3990 | minor != MINOR(dev)) | |
3991 | return -EOVERFLOW; | |
3992 | ||
6791875e N |
3993 | flush_workqueue(md_misc_wq); |
3994 | ||
3995 | err = mddev_lock(mddev); | |
3996 | if (err) | |
3997 | return err; | |
6d7ff738 N |
3998 | if (mddev->persistent) { |
3999 | rdev = md_import_device(dev, mddev->major_version, | |
4000 | mddev->minor_version); | |
4001 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
4002 | struct md_rdev *rdev0 |
4003 | = list_entry(mddev->disks.next, | |
4004 | struct md_rdev, same_set); | |
6d7ff738 N |
4005 | err = super_types[mddev->major_version] |
4006 | .load_super(rdev, rdev0, mddev->minor_version); | |
4007 | if (err < 0) | |
4008 | goto out; | |
4009 | } | |
c5d79adb N |
4010 | } else if (mddev->external) |
4011 | rdev = md_import_device(dev, -2, -1); | |
4012 | else | |
6d7ff738 N |
4013 | rdev = md_import_device(dev, -1, -1); |
4014 | ||
9a8c0fa8 N |
4015 | if (IS_ERR(rdev)) { |
4016 | mddev_unlock(mddev); | |
6d7ff738 | 4017 | return PTR_ERR(rdev); |
9a8c0fa8 | 4018 | } |
6d7ff738 N |
4019 | err = bind_rdev_to_array(rdev, mddev); |
4020 | out: | |
4021 | if (err) | |
4022 | export_rdev(rdev); | |
6791875e | 4023 | mddev_unlock(mddev); |
6d7ff738 N |
4024 | return err ? err : len; |
4025 | } | |
4026 | ||
4027 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 4028 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 4029 | |
9b1d1dac | 4030 | static ssize_t |
fd01b88c | 4031 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
4032 | { |
4033 | char *end; | |
4034 | unsigned long chunk, end_chunk; | |
6791875e | 4035 | int err; |
9b1d1dac | 4036 | |
6791875e N |
4037 | err = mddev_lock(mddev); |
4038 | if (err) | |
4039 | return err; | |
9b1d1dac PC |
4040 | if (!mddev->bitmap) |
4041 | goto out; | |
4042 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
4043 | while (*buf) { | |
4044 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
4045 | if (buf == end) break; | |
4046 | if (*end == '-') { /* range */ | |
4047 | buf = end + 1; | |
4048 | end_chunk = simple_strtoul(buf, &end, 0); | |
4049 | if (buf == end) break; | |
4050 | } | |
4051 | if (*end && !isspace(*end)) break; | |
4052 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 4053 | buf = skip_spaces(end); |
9b1d1dac PC |
4054 | } |
4055 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
4056 | out: | |
6791875e | 4057 | mddev_unlock(mddev); |
9b1d1dac PC |
4058 | return len; |
4059 | } | |
4060 | ||
4061 | static struct md_sysfs_entry md_bitmap = | |
4062 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
4063 | ||
a35b0d69 | 4064 | static ssize_t |
fd01b88c | 4065 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 4066 | { |
58c0fed4 AN |
4067 | return sprintf(page, "%llu\n", |
4068 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
4069 | } |
4070 | ||
fd01b88c | 4071 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
4072 | |
4073 | static ssize_t | |
fd01b88c | 4074 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
4075 | { |
4076 | /* If array is inactive, we can reduce the component size, but | |
4077 | * not increase it (except from 0). | |
4078 | * If array is active, we can try an on-line resize | |
4079 | */ | |
b522adcd DW |
4080 | sector_t sectors; |
4081 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 4082 | |
58c0fed4 AN |
4083 | if (err < 0) |
4084 | return err; | |
6791875e N |
4085 | err = mddev_lock(mddev); |
4086 | if (err) | |
4087 | return err; | |
a35b0d69 | 4088 | if (mddev->pers) { |
293467aa GR |
4089 | if (mddev_is_clustered(mddev)) |
4090 | md_cluster_ops->metadata_update_start(mddev); | |
58c0fed4 | 4091 | err = update_size(mddev, sectors); |
850b2b42 | 4092 | md_update_sb(mddev, 1); |
293467aa GR |
4093 | if (mddev_is_clustered(mddev)) |
4094 | md_cluster_ops->metadata_update_finish(mddev); | |
a35b0d69 | 4095 | } else { |
58c0fed4 AN |
4096 | if (mddev->dev_sectors == 0 || |
4097 | mddev->dev_sectors > sectors) | |
4098 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
4099 | else |
4100 | err = -ENOSPC; | |
4101 | } | |
6791875e | 4102 | mddev_unlock(mddev); |
a35b0d69 N |
4103 | return err ? err : len; |
4104 | } | |
4105 | ||
4106 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 4107 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 4108 | |
83f0d77a | 4109 | /* Metadata version. |
e691063a N |
4110 | * This is one of |
4111 | * 'none' for arrays with no metadata (good luck...) | |
4112 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
4113 | * or N.M for internally known formats |
4114 | */ | |
4115 | static ssize_t | |
fd01b88c | 4116 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
4117 | { |
4118 | if (mddev->persistent) | |
4119 | return sprintf(page, "%d.%d\n", | |
4120 | mddev->major_version, mddev->minor_version); | |
e691063a N |
4121 | else if (mddev->external) |
4122 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
4123 | else |
4124 | return sprintf(page, "none\n"); | |
4125 | } | |
4126 | ||
4127 | static ssize_t | |
fd01b88c | 4128 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
4129 | { |
4130 | int major, minor; | |
4131 | char *e; | |
6791875e | 4132 | int err; |
ea43ddd8 N |
4133 | /* Changing the details of 'external' metadata is |
4134 | * always permitted. Otherwise there must be | |
4135 | * no devices attached to the array. | |
4136 | */ | |
6791875e N |
4137 | |
4138 | err = mddev_lock(mddev); | |
4139 | if (err) | |
4140 | return err; | |
4141 | err = -EBUSY; | |
ea43ddd8 N |
4142 | if (mddev->external && strncmp(buf, "external:", 9) == 0) |
4143 | ; | |
4144 | else if (!list_empty(&mddev->disks)) | |
6791875e | 4145 | goto out_unlock; |
8bb93aac | 4146 | |
6791875e | 4147 | err = 0; |
8bb93aac N |
4148 | if (cmd_match(buf, "none")) { |
4149 | mddev->persistent = 0; | |
e691063a N |
4150 | mddev->external = 0; |
4151 | mddev->major_version = 0; | |
4152 | mddev->minor_version = 90; | |
6791875e | 4153 | goto out_unlock; |
e691063a N |
4154 | } |
4155 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 4156 | size_t namelen = len-9; |
e691063a N |
4157 | if (namelen >= sizeof(mddev->metadata_type)) |
4158 | namelen = sizeof(mddev->metadata_type)-1; | |
4159 | strncpy(mddev->metadata_type, buf+9, namelen); | |
4160 | mddev->metadata_type[namelen] = 0; | |
4161 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
4162 | mddev->metadata_type[--namelen] = 0; | |
4163 | mddev->persistent = 0; | |
4164 | mddev->external = 1; | |
8bb93aac N |
4165 | mddev->major_version = 0; |
4166 | mddev->minor_version = 90; | |
6791875e | 4167 | goto out_unlock; |
8bb93aac N |
4168 | } |
4169 | major = simple_strtoul(buf, &e, 10); | |
6791875e | 4170 | err = -EINVAL; |
8bb93aac | 4171 | if (e==buf || *e != '.') |
6791875e | 4172 | goto out_unlock; |
8bb93aac N |
4173 | buf = e+1; |
4174 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 4175 | if (e==buf || (*e && *e != '\n') ) |
6791875e N |
4176 | goto out_unlock; |
4177 | err = -ENOENT; | |
50511da3 | 4178 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
6791875e | 4179 | goto out_unlock; |
8bb93aac N |
4180 | mddev->major_version = major; |
4181 | mddev->minor_version = minor; | |
4182 | mddev->persistent = 1; | |
e691063a | 4183 | mddev->external = 0; |
6791875e N |
4184 | err = 0; |
4185 | out_unlock: | |
4186 | mddev_unlock(mddev); | |
4187 | return err ?: len; | |
8bb93aac N |
4188 | } |
4189 | ||
4190 | static struct md_sysfs_entry md_metadata = | |
750f199e | 4191 | __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 4192 | |
24dd469d | 4193 | static ssize_t |
fd01b88c | 4194 | action_show(struct mddev *mddev, char *page) |
24dd469d | 4195 | { |
7eec314d | 4196 | char *type = "idle"; |
b7b17c9b N |
4197 | unsigned long recovery = mddev->recovery; |
4198 | if (test_bit(MD_RECOVERY_FROZEN, &recovery)) | |
b6a9ce68 | 4199 | type = "frozen"; |
b7b17c9b N |
4200 | else if (test_bit(MD_RECOVERY_RUNNING, &recovery) || |
4201 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) { | |
4202 | if (test_bit(MD_RECOVERY_RESHAPE, &recovery)) | |
ccfcc3c1 | 4203 | type = "reshape"; |
b7b17c9b N |
4204 | else if (test_bit(MD_RECOVERY_SYNC, &recovery)) { |
4205 | if (!test_bit(MD_RECOVERY_REQUESTED, &recovery)) | |
24dd469d | 4206 | type = "resync"; |
b7b17c9b | 4207 | else if (test_bit(MD_RECOVERY_CHECK, &recovery)) |
24dd469d N |
4208 | type = "check"; |
4209 | else | |
4210 | type = "repair"; | |
b7b17c9b | 4211 | } else if (test_bit(MD_RECOVERY_RECOVER, &recovery)) |
24dd469d | 4212 | type = "recover"; |
985ca973 N |
4213 | else if (mddev->reshape_position != MaxSector) |
4214 | type = "reshape"; | |
24dd469d N |
4215 | } |
4216 | return sprintf(page, "%s\n", type); | |
4217 | } | |
4218 | ||
4219 | static ssize_t | |
fd01b88c | 4220 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4221 | { |
7eec314d N |
4222 | if (!mddev->pers || !mddev->pers->sync_request) |
4223 | return -EINVAL; | |
4224 | ||
b6a9ce68 N |
4225 | |
4226 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
56ccc112 N |
4227 | if (cmd_match(page, "frozen")) |
4228 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4229 | else | |
4230 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
8e8e2518 N |
4231 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
4232 | mddev_lock(mddev) == 0) { | |
4233 | flush_workqueue(md_misc_wq); | |
4234 | if (mddev->sync_thread) { | |
4235 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6791875e | 4236 | md_reap_sync_thread(mddev); |
6791875e | 4237 | } |
8e8e2518 | 4238 | mddev_unlock(mddev); |
7eec314d | 4239 | } |
03c902e1 N |
4240 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
4241 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 4242 | return -EBUSY; |
72a23c21 | 4243 | else if (cmd_match(page, "resync")) |
56ccc112 | 4244 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4245 | else if (cmd_match(page, "recover")) { |
56ccc112 | 4246 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4247 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
72a23c21 | 4248 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4249 | int err; |
4250 | if (mddev->pers->start_reshape == NULL) | |
4251 | return -EINVAL; | |
6791875e N |
4252 | err = mddev_lock(mddev); |
4253 | if (!err) { | |
56ccc112 | 4254 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6791875e N |
4255 | err = mddev->pers->start_reshape(mddev); |
4256 | mddev_unlock(mddev); | |
4257 | } | |
16484bf5 N |
4258 | if (err) |
4259 | return err; | |
a99ac971 | 4260 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4261 | } else { |
bce74dac | 4262 | if (cmd_match(page, "check")) |
7eec314d | 4263 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4264 | else if (!cmd_match(page, "repair")) |
7eec314d | 4265 | return -EINVAL; |
56ccc112 | 4266 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
7eec314d N |
4267 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
4268 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4269 | } |
48c26ddc N |
4270 | if (mddev->ro == 2) { |
4271 | /* A write to sync_action is enough to justify | |
4272 | * canceling read-auto mode | |
4273 | */ | |
4274 | mddev->ro = 0; | |
4275 | md_wakeup_thread(mddev->sync_thread); | |
4276 | } | |
03c902e1 | 4277 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4278 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4279 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4280 | return len; |
4281 | } | |
4282 | ||
c4a39551 | 4283 | static struct md_sysfs_entry md_scan_mode = |
750f199e | 4284 | __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); |
c4a39551 JB |
4285 | |
4286 | static ssize_t | |
4287 | last_sync_action_show(struct mddev *mddev, char *page) | |
4288 | { | |
4289 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4290 | } | |
4291 | ||
4292 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4293 | ||
9d88883e | 4294 | static ssize_t |
fd01b88c | 4295 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4296 | { |
4297 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4298 | (unsigned long long) |
4299 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4300 | } |
4301 | ||
80ca3a44 | 4302 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4303 | |
88202a0c | 4304 | static ssize_t |
fd01b88c | 4305 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4306 | { |
4307 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4308 | mddev->sync_speed_min ? "local": "system"); | |
4309 | } | |
4310 | ||
4311 | static ssize_t | |
fd01b88c | 4312 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4313 | { |
4c9309c0 AD |
4314 | unsigned int min; |
4315 | int rv; | |
4316 | ||
88202a0c | 4317 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4318 | min = 0; |
4319 | } else { | |
4320 | rv = kstrtouint(buf, 10, &min); | |
4321 | if (rv < 0) | |
4322 | return rv; | |
4323 | if (min == 0) | |
4324 | return -EINVAL; | |
88202a0c | 4325 | } |
88202a0c N |
4326 | mddev->sync_speed_min = min; |
4327 | return len; | |
4328 | } | |
4329 | ||
4330 | static struct md_sysfs_entry md_sync_min = | |
4331 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4332 | ||
4333 | static ssize_t | |
fd01b88c | 4334 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4335 | { |
4336 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4337 | mddev->sync_speed_max ? "local": "system"); | |
4338 | } | |
4339 | ||
4340 | static ssize_t | |
fd01b88c | 4341 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4342 | { |
4c9309c0 AD |
4343 | unsigned int max; |
4344 | int rv; | |
4345 | ||
88202a0c | 4346 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4347 | max = 0; |
4348 | } else { | |
4349 | rv = kstrtouint(buf, 10, &max); | |
4350 | if (rv < 0) | |
4351 | return rv; | |
4352 | if (max == 0) | |
4353 | return -EINVAL; | |
88202a0c | 4354 | } |
88202a0c N |
4355 | mddev->sync_speed_max = max; |
4356 | return len; | |
4357 | } | |
4358 | ||
4359 | static struct md_sysfs_entry md_sync_max = | |
4360 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4361 | ||
d7f3d291 | 4362 | static ssize_t |
fd01b88c | 4363 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4364 | { |
4365 | return sprintf(page, "%d\n", mddev->degraded); | |
4366 | } | |
4367 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4368 | |
90b08710 | 4369 | static ssize_t |
fd01b88c | 4370 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4371 | { |
4372 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4373 | } | |
4374 | ||
4375 | static ssize_t | |
fd01b88c | 4376 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4377 | { |
4378 | long n; | |
4379 | ||
b29bebd6 | 4380 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4381 | return -EINVAL; |
4382 | ||
4383 | if (n != 0 && n != 1) | |
4384 | return -EINVAL; | |
4385 | ||
4386 | mddev->parallel_resync = n; | |
4387 | ||
4388 | if (mddev->sync_thread) | |
4389 | wake_up(&resync_wait); | |
4390 | ||
4391 | return len; | |
4392 | } | |
4393 | ||
4394 | /* force parallel resync, even with shared block devices */ | |
4395 | static struct md_sysfs_entry md_sync_force_parallel = | |
4396 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4397 | sync_force_parallel_show, sync_force_parallel_store); | |
4398 | ||
88202a0c | 4399 | static ssize_t |
fd01b88c | 4400 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4401 | { |
4402 | unsigned long resync, dt, db; | |
d1a7c503 N |
4403 | if (mddev->curr_resync == 0) |
4404 | return sprintf(page, "none\n"); | |
9687a60c AN |
4405 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4406 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4407 | if (!dt) dt++; |
9687a60c AN |
4408 | db = resync - mddev->resync_mark_cnt; |
4409 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4410 | } |
4411 | ||
80ca3a44 | 4412 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4413 | |
4414 | static ssize_t | |
fd01b88c | 4415 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4416 | { |
13ae864b | 4417 | unsigned long long max_sectors, resync; |
88202a0c | 4418 | |
acb180b0 N |
4419 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4420 | return sprintf(page, "none\n"); | |
4421 | ||
72f36d59 N |
4422 | if (mddev->curr_resync == 1 || |
4423 | mddev->curr_resync == 2) | |
4424 | return sprintf(page, "delayed\n"); | |
4425 | ||
c804cdec N |
4426 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4427 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4428 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4429 | else |
58c0fed4 | 4430 | max_sectors = mddev->dev_sectors; |
88202a0c | 4431 | |
acb180b0 | 4432 | resync = mddev->curr_resync_completed; |
13ae864b | 4433 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4434 | } |
4435 | ||
750f199e N |
4436 | static struct md_sysfs_entry md_sync_completed = |
4437 | __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL); | |
88202a0c | 4438 | |
5e96ee65 | 4439 | static ssize_t |
fd01b88c | 4440 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4441 | { |
4442 | return sprintf(page, "%llu\n", | |
4443 | (unsigned long long)mddev->resync_min); | |
4444 | } | |
4445 | static ssize_t | |
fd01b88c | 4446 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4447 | { |
4448 | unsigned long long min; | |
23da422b | 4449 | int err; |
23da422b | 4450 | |
b29bebd6 | 4451 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 | 4452 | return -EINVAL; |
23da422b N |
4453 | |
4454 | spin_lock(&mddev->lock); | |
4455 | err = -EINVAL; | |
5e96ee65 | 4456 | if (min > mddev->resync_max) |
23da422b N |
4457 | goto out_unlock; |
4458 | ||
4459 | err = -EBUSY; | |
5e96ee65 | 4460 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4461 | goto out_unlock; |
5e96ee65 | 4462 | |
50c37b13 N |
4463 | /* Round down to multiple of 4K for safety */ |
4464 | mddev->resync_min = round_down(min, 8); | |
23da422b | 4465 | err = 0; |
5e96ee65 | 4466 | |
23da422b N |
4467 | out_unlock: |
4468 | spin_unlock(&mddev->lock); | |
4469 | return err ?: len; | |
5e96ee65 NB |
4470 | } |
4471 | ||
4472 | static struct md_sysfs_entry md_min_sync = | |
4473 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4474 | ||
c6207277 | 4475 | static ssize_t |
fd01b88c | 4476 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4477 | { |
4478 | if (mddev->resync_max == MaxSector) | |
4479 | return sprintf(page, "max\n"); | |
4480 | else | |
4481 | return sprintf(page, "%llu\n", | |
4482 | (unsigned long long)mddev->resync_max); | |
4483 | } | |
4484 | static ssize_t | |
fd01b88c | 4485 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 | 4486 | { |
23da422b N |
4487 | int err; |
4488 | spin_lock(&mddev->lock); | |
c6207277 N |
4489 | if (strncmp(buf, "max", 3) == 0) |
4490 | mddev->resync_max = MaxSector; | |
4491 | else { | |
5e96ee65 | 4492 | unsigned long long max; |
23da422b N |
4493 | int chunk; |
4494 | ||
4495 | err = -EINVAL; | |
b29bebd6 | 4496 | if (kstrtoull(buf, 10, &max)) |
23da422b | 4497 | goto out_unlock; |
5e96ee65 | 4498 | if (max < mddev->resync_min) |
23da422b N |
4499 | goto out_unlock; |
4500 | ||
4501 | err = -EBUSY; | |
c6207277 | 4502 | if (max < mddev->resync_max && |
4d484a4a | 4503 | mddev->ro == 0 && |
c6207277 | 4504 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4505 | goto out_unlock; |
c6207277 N |
4506 | |
4507 | /* Must be a multiple of chunk_size */ | |
23da422b N |
4508 | chunk = mddev->chunk_sectors; |
4509 | if (chunk) { | |
2ac06c33 | 4510 | sector_t temp = max; |
23da422b N |
4511 | |
4512 | err = -EINVAL; | |
4513 | if (sector_div(temp, chunk)) | |
4514 | goto out_unlock; | |
c6207277 N |
4515 | } |
4516 | mddev->resync_max = max; | |
4517 | } | |
4518 | wake_up(&mddev->recovery_wait); | |
23da422b N |
4519 | err = 0; |
4520 | out_unlock: | |
4521 | spin_unlock(&mddev->lock); | |
4522 | return err ?: len; | |
c6207277 N |
4523 | } |
4524 | ||
4525 | static struct md_sysfs_entry md_max_sync = | |
4526 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4527 | ||
e464eafd | 4528 | static ssize_t |
fd01b88c | 4529 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4530 | { |
4531 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4532 | } | |
4533 | ||
4534 | static ssize_t | |
fd01b88c | 4535 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4536 | { |
4c9309c0 | 4537 | unsigned long long old, new; |
6791875e | 4538 | int err; |
e464eafd | 4539 | |
4c9309c0 AD |
4540 | err = kstrtoull(buf, 10, &new); |
4541 | if (err < 0) | |
4542 | return err; | |
4543 | if (new != (sector_t)new) | |
e464eafd | 4544 | return -EINVAL; |
23ddff37 | 4545 | |
6791875e N |
4546 | err = mddev_lock(mddev); |
4547 | if (err) | |
4548 | return err; | |
4549 | err = -EINVAL; | |
4550 | if (mddev->pers == NULL || | |
4551 | mddev->pers->quiesce == NULL) | |
4552 | goto unlock; | |
4553 | old = mddev->suspend_lo; | |
23ddff37 N |
4554 | mddev->suspend_lo = new; |
4555 | if (new >= old) | |
4556 | /* Shrinking suspended region */ | |
e464eafd | 4557 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4558 | else { |
4559 | /* Expanding suspended region - need to wait */ | |
4560 | mddev->pers->quiesce(mddev, 1); | |
4561 | mddev->pers->quiesce(mddev, 0); | |
4562 | } | |
6791875e N |
4563 | err = 0; |
4564 | unlock: | |
4565 | mddev_unlock(mddev); | |
4566 | return err ?: len; | |
e464eafd N |
4567 | } |
4568 | static struct md_sysfs_entry md_suspend_lo = | |
4569 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4570 | ||
e464eafd | 4571 | static ssize_t |
fd01b88c | 4572 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4573 | { |
4574 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4575 | } | |
4576 | ||
4577 | static ssize_t | |
fd01b88c | 4578 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4579 | { |
4c9309c0 | 4580 | unsigned long long old, new; |
6791875e | 4581 | int err; |
e464eafd | 4582 | |
4c9309c0 AD |
4583 | err = kstrtoull(buf, 10, &new); |
4584 | if (err < 0) | |
4585 | return err; | |
4586 | if (new != (sector_t)new) | |
e464eafd | 4587 | return -EINVAL; |
23ddff37 | 4588 | |
6791875e N |
4589 | err = mddev_lock(mddev); |
4590 | if (err) | |
4591 | return err; | |
4592 | err = -EINVAL; | |
4593 | if (mddev->pers == NULL || | |
4594 | mddev->pers->quiesce == NULL) | |
4595 | goto unlock; | |
4596 | old = mddev->suspend_hi; | |
23ddff37 N |
4597 | mddev->suspend_hi = new; |
4598 | if (new <= old) | |
4599 | /* Shrinking suspended region */ | |
4600 | mddev->pers->quiesce(mddev, 2); | |
4601 | else { | |
4602 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4603 | mddev->pers->quiesce(mddev, 1); |
4604 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 | 4605 | } |
6791875e N |
4606 | err = 0; |
4607 | unlock: | |
4608 | mddev_unlock(mddev); | |
4609 | return err ?: len; | |
e464eafd N |
4610 | } |
4611 | static struct md_sysfs_entry md_suspend_hi = | |
4612 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4613 | ||
08a02ecd | 4614 | static ssize_t |
fd01b88c | 4615 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4616 | { |
4617 | if (mddev->reshape_position != MaxSector) | |
4618 | return sprintf(page, "%llu\n", | |
4619 | (unsigned long long)mddev->reshape_position); | |
4620 | strcpy(page, "none\n"); | |
4621 | return 5; | |
4622 | } | |
4623 | ||
4624 | static ssize_t | |
fd01b88c | 4625 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4626 | { |
c6563a8c | 4627 | struct md_rdev *rdev; |
4c9309c0 | 4628 | unsigned long long new; |
6791875e | 4629 | int err; |
6791875e | 4630 | |
4c9309c0 AD |
4631 | err = kstrtoull(buf, 10, &new); |
4632 | if (err < 0) | |
4633 | return err; | |
4634 | if (new != (sector_t)new) | |
08a02ecd | 4635 | return -EINVAL; |
6791875e N |
4636 | err = mddev_lock(mddev); |
4637 | if (err) | |
4638 | return err; | |
4639 | err = -EBUSY; | |
4640 | if (mddev->pers) | |
4641 | goto unlock; | |
08a02ecd N |
4642 | mddev->reshape_position = new; |
4643 | mddev->delta_disks = 0; | |
2c810cdd | 4644 | mddev->reshape_backwards = 0; |
08a02ecd N |
4645 | mddev->new_level = mddev->level; |
4646 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4647 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4648 | rdev_for_each(rdev, mddev) |
4649 | rdev->new_data_offset = rdev->data_offset; | |
6791875e N |
4650 | err = 0; |
4651 | unlock: | |
4652 | mddev_unlock(mddev); | |
4653 | return err ?: len; | |
08a02ecd N |
4654 | } |
4655 | ||
4656 | static struct md_sysfs_entry md_reshape_position = | |
4657 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4658 | reshape_position_store); | |
4659 | ||
2c810cdd N |
4660 | static ssize_t |
4661 | reshape_direction_show(struct mddev *mddev, char *page) | |
4662 | { | |
4663 | return sprintf(page, "%s\n", | |
4664 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4665 | } | |
4666 | ||
4667 | static ssize_t | |
4668 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4669 | { | |
4670 | int backwards = 0; | |
6791875e N |
4671 | int err; |
4672 | ||
2c810cdd N |
4673 | if (cmd_match(buf, "forwards")) |
4674 | backwards = 0; | |
4675 | else if (cmd_match(buf, "backwards")) | |
4676 | backwards = 1; | |
4677 | else | |
4678 | return -EINVAL; | |
4679 | if (mddev->reshape_backwards == backwards) | |
4680 | return len; | |
4681 | ||
6791875e N |
4682 | err = mddev_lock(mddev); |
4683 | if (err) | |
4684 | return err; | |
2c810cdd N |
4685 | /* check if we are allowed to change */ |
4686 | if (mddev->delta_disks) | |
6791875e N |
4687 | err = -EBUSY; |
4688 | else if (mddev->persistent && | |
2c810cdd | 4689 | mddev->major_version == 0) |
6791875e N |
4690 | err = -EINVAL; |
4691 | else | |
4692 | mddev->reshape_backwards = backwards; | |
4693 | mddev_unlock(mddev); | |
4694 | return err ?: len; | |
2c810cdd N |
4695 | } |
4696 | ||
4697 | static struct md_sysfs_entry md_reshape_direction = | |
4698 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
4699 | reshape_direction_store); | |
4700 | ||
b522adcd | 4701 | static ssize_t |
fd01b88c | 4702 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
4703 | { |
4704 | if (mddev->external_size) | |
4705 | return sprintf(page, "%llu\n", | |
4706 | (unsigned long long)mddev->array_sectors/2); | |
4707 | else | |
4708 | return sprintf(page, "default\n"); | |
4709 | } | |
4710 | ||
4711 | static ssize_t | |
fd01b88c | 4712 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
4713 | { |
4714 | sector_t sectors; | |
6791875e N |
4715 | int err; |
4716 | ||
4717 | err = mddev_lock(mddev); | |
4718 | if (err) | |
4719 | return err; | |
b522adcd DW |
4720 | |
4721 | if (strncmp(buf, "default", 7) == 0) { | |
4722 | if (mddev->pers) | |
4723 | sectors = mddev->pers->size(mddev, 0, 0); | |
4724 | else | |
4725 | sectors = mddev->array_sectors; | |
4726 | ||
4727 | mddev->external_size = 0; | |
4728 | } else { | |
4729 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
6791875e N |
4730 | err = -EINVAL; |
4731 | else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
4732 | err = -E2BIG; | |
4733 | else | |
4734 | mddev->external_size = 1; | |
b522adcd DW |
4735 | } |
4736 | ||
6791875e N |
4737 | if (!err) { |
4738 | mddev->array_sectors = sectors; | |
4739 | if (mddev->pers) { | |
4740 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4741 | revalidate_disk(mddev->gendisk); | |
4742 | } | |
cbe6ef1d | 4743 | } |
6791875e N |
4744 | mddev_unlock(mddev); |
4745 | return err ?: len; | |
b522adcd DW |
4746 | } |
4747 | ||
4748 | static struct md_sysfs_entry md_array_size = | |
4749 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4750 | array_size_store); | |
e464eafd | 4751 | |
eae1701f N |
4752 | static struct attribute *md_default_attrs[] = { |
4753 | &md_level.attr, | |
d4dbd025 | 4754 | &md_layout.attr, |
eae1701f | 4755 | &md_raid_disks.attr, |
3b34380a | 4756 | &md_chunk_size.attr, |
a35b0d69 | 4757 | &md_size.attr, |
a94213b1 | 4758 | &md_resync_start.attr, |
8bb93aac | 4759 | &md_metadata.attr, |
6d7ff738 | 4760 | &md_new_device.attr, |
16f17b39 | 4761 | &md_safe_delay.attr, |
9e653b63 | 4762 | &md_array_state.attr, |
08a02ecd | 4763 | &md_reshape_position.attr, |
2c810cdd | 4764 | &md_reshape_direction.attr, |
b522adcd | 4765 | &md_array_size.attr, |
1e50915f | 4766 | &max_corr_read_errors.attr, |
411036fa N |
4767 | NULL, |
4768 | }; | |
4769 | ||
4770 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4771 | &md_scan_mode.attr, |
c4a39551 | 4772 | &md_last_scan_mode.attr, |
9d88883e | 4773 | &md_mismatches.attr, |
88202a0c N |
4774 | &md_sync_min.attr, |
4775 | &md_sync_max.attr, | |
4776 | &md_sync_speed.attr, | |
90b08710 | 4777 | &md_sync_force_parallel.attr, |
88202a0c | 4778 | &md_sync_completed.attr, |
5e96ee65 | 4779 | &md_min_sync.attr, |
c6207277 | 4780 | &md_max_sync.attr, |
e464eafd N |
4781 | &md_suspend_lo.attr, |
4782 | &md_suspend_hi.attr, | |
9b1d1dac | 4783 | &md_bitmap.attr, |
d7f3d291 | 4784 | &md_degraded.attr, |
eae1701f N |
4785 | NULL, |
4786 | }; | |
411036fa N |
4787 | static struct attribute_group md_redundancy_group = { |
4788 | .name = NULL, | |
4789 | .attrs = md_redundancy_attrs, | |
4790 | }; | |
4791 | ||
eae1701f N |
4792 | static ssize_t |
4793 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4794 | { | |
4795 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4796 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4797 | ssize_t rv; |
eae1701f N |
4798 | |
4799 | if (!entry->show) | |
4800 | return -EIO; | |
af8a2434 N |
4801 | spin_lock(&all_mddevs_lock); |
4802 | if (list_empty(&mddev->all_mddevs)) { | |
4803 | spin_unlock(&all_mddevs_lock); | |
4804 | return -EBUSY; | |
4805 | } | |
4806 | mddev_get(mddev); | |
4807 | spin_unlock(&all_mddevs_lock); | |
4808 | ||
b7b17c9b | 4809 | rv = entry->show(mddev, page); |
af8a2434 | 4810 | mddev_put(mddev); |
96de1e66 | 4811 | return rv; |
eae1701f N |
4812 | } |
4813 | ||
4814 | static ssize_t | |
4815 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4816 | const char *page, size_t length) | |
4817 | { | |
4818 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4819 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4820 | ssize_t rv; |
eae1701f N |
4821 | |
4822 | if (!entry->store) | |
4823 | return -EIO; | |
67463acb N |
4824 | if (!capable(CAP_SYS_ADMIN)) |
4825 | return -EACCES; | |
af8a2434 N |
4826 | spin_lock(&all_mddevs_lock); |
4827 | if (list_empty(&mddev->all_mddevs)) { | |
4828 | spin_unlock(&all_mddevs_lock); | |
4829 | return -EBUSY; | |
4830 | } | |
4831 | mddev_get(mddev); | |
4832 | spin_unlock(&all_mddevs_lock); | |
6791875e | 4833 | rv = entry->store(mddev, page, length); |
af8a2434 | 4834 | mddev_put(mddev); |
96de1e66 | 4835 | return rv; |
eae1701f N |
4836 | } |
4837 | ||
4838 | static void md_free(struct kobject *ko) | |
4839 | { | |
fd01b88c | 4840 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
4841 | |
4842 | if (mddev->sysfs_state) | |
4843 | sysfs_put(mddev->sysfs_state); | |
4844 | ||
6cd18e71 N |
4845 | if (mddev->queue) |
4846 | blk_cleanup_queue(mddev->queue); | |
a21d1504 N |
4847 | if (mddev->gendisk) { |
4848 | del_gendisk(mddev->gendisk); | |
4849 | put_disk(mddev->gendisk); | |
4850 | } | |
a21d1504 | 4851 | |
eae1701f N |
4852 | kfree(mddev); |
4853 | } | |
4854 | ||
52cf25d0 | 4855 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4856 | .show = md_attr_show, |
4857 | .store = md_attr_store, | |
4858 | }; | |
4859 | static struct kobj_type md_ktype = { | |
4860 | .release = md_free, | |
4861 | .sysfs_ops = &md_sysfs_ops, | |
4862 | .default_attrs = md_default_attrs, | |
4863 | }; | |
4864 | ||
1da177e4 LT |
4865 | int mdp_major = 0; |
4866 | ||
5fd3a17e DW |
4867 | static void mddev_delayed_delete(struct work_struct *ws) |
4868 | { | |
fd01b88c | 4869 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 4870 | |
43a70507 | 4871 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4872 | kobject_del(&mddev->kobj); |
4873 | kobject_put(&mddev->kobj); | |
4874 | } | |
4875 | ||
efeb53c0 | 4876 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4877 | { |
48c9c27b | 4878 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 4879 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 4880 | struct gendisk *disk; |
efeb53c0 N |
4881 | int partitioned; |
4882 | int shift; | |
4883 | int unit; | |
3830c62f | 4884 | int error; |
1da177e4 LT |
4885 | |
4886 | if (!mddev) | |
efeb53c0 N |
4887 | return -ENODEV; |
4888 | ||
4889 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4890 | shift = partitioned ? MdpMinorShift : 0; | |
4891 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4892 | |
e804ac78 TH |
4893 | /* wait for any previous instance of this device to be |
4894 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4895 | */ |
e804ac78 | 4896 | flush_workqueue(md_misc_wq); |
d3374825 | 4897 | |
48c9c27b | 4898 | mutex_lock(&disks_mutex); |
0909dc44 N |
4899 | error = -EEXIST; |
4900 | if (mddev->gendisk) | |
4901 | goto abort; | |
efeb53c0 N |
4902 | |
4903 | if (name) { | |
4904 | /* Need to ensure that 'name' is not a duplicate. | |
4905 | */ | |
fd01b88c | 4906 | struct mddev *mddev2; |
efeb53c0 N |
4907 | spin_lock(&all_mddevs_lock); |
4908 | ||
4909 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4910 | if (mddev2->gendisk && | |
4911 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4912 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4913 | goto abort; |
efeb53c0 N |
4914 | } |
4915 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4916 | } |
8b765398 | 4917 | |
0909dc44 | 4918 | error = -ENOMEM; |
8b765398 | 4919 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4920 | if (!mddev->queue) |
4921 | goto abort; | |
409c57f3 N |
4922 | mddev->queue->queuedata = mddev; |
4923 | ||
409c57f3 | 4924 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 4925 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 4926 | |
1da177e4 LT |
4927 | disk = alloc_disk(1 << shift); |
4928 | if (!disk) { | |
8b765398 N |
4929 | blk_cleanup_queue(mddev->queue); |
4930 | mddev->queue = NULL; | |
0909dc44 | 4931 | goto abort; |
1da177e4 | 4932 | } |
efeb53c0 | 4933 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4934 | disk->first_minor = unit << shift; |
efeb53c0 N |
4935 | if (name) |
4936 | strcpy(disk->disk_name, name); | |
4937 | else if (partitioned) | |
1da177e4 | 4938 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4939 | else |
1da177e4 | 4940 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4941 | disk->fops = &md_fops; |
4942 | disk->private_data = mddev; | |
4943 | disk->queue = mddev->queue; | |
b0140891 | 4944 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); |
92850bbd | 4945 | /* Allow extended partitions. This makes the |
d3374825 | 4946 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4947 | * remove it now. |
4948 | */ | |
4949 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 4950 | mddev->gendisk = disk; |
b0140891 N |
4951 | /* As soon as we call add_disk(), another thread could get |
4952 | * through to md_open, so make sure it doesn't get too far | |
4953 | */ | |
4954 | mutex_lock(&mddev->open_mutex); | |
4955 | add_disk(disk); | |
4956 | ||
ed9e1982 TH |
4957 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4958 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4959 | if (error) { |
4960 | /* This isn't possible, but as kobject_init_and_add is marked | |
4961 | * __must_check, we must do something with the result | |
4962 | */ | |
5e55e2f5 N |
4963 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4964 | disk->disk_name); | |
0909dc44 N |
4965 | error = 0; |
4966 | } | |
00bcb4ac N |
4967 | if (mddev->kobj.sd && |
4968 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4969 | printk(KERN_DEBUG "pointless warning\n"); |
b0140891 | 4970 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
4971 | abort: |
4972 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4973 | if (!error && mddev->kobj.sd) { |
3830c62f | 4974 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4975 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4976 | } |
d3374825 | 4977 | mddev_put(mddev); |
0909dc44 | 4978 | return error; |
efeb53c0 N |
4979 | } |
4980 | ||
4981 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4982 | { | |
4983 | md_alloc(dev, NULL); | |
1da177e4 LT |
4984 | return NULL; |
4985 | } | |
4986 | ||
efeb53c0 N |
4987 | static int add_named_array(const char *val, struct kernel_param *kp) |
4988 | { | |
4989 | /* val must be "md_*" where * is not all digits. | |
4990 | * We allocate an array with a large free minor number, and | |
4991 | * set the name to val. val must not already be an active name. | |
4992 | */ | |
4993 | int len = strlen(val); | |
4994 | char buf[DISK_NAME_LEN]; | |
4995 | ||
4996 | while (len && val[len-1] == '\n') | |
4997 | len--; | |
4998 | if (len >= DISK_NAME_LEN) | |
4999 | return -E2BIG; | |
5000 | strlcpy(buf, val, len+1); | |
5001 | if (strncmp(buf, "md_", 3) != 0) | |
5002 | return -EINVAL; | |
5003 | return md_alloc(0, buf); | |
5004 | } | |
5005 | ||
1da177e4 LT |
5006 | static void md_safemode_timeout(unsigned long data) |
5007 | { | |
fd01b88c | 5008 | struct mddev *mddev = (struct mddev *) data; |
1da177e4 | 5009 | |
0fd62b86 NB |
5010 | if (!atomic_read(&mddev->writes_pending)) { |
5011 | mddev->safemode = 1; | |
5012 | if (mddev->external) | |
00bcb4ac | 5013 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 5014 | } |
1da177e4 LT |
5015 | md_wakeup_thread(mddev->thread); |
5016 | } | |
5017 | ||
6ff8d8ec | 5018 | static int start_dirty_degraded; |
1da177e4 | 5019 | |
fd01b88c | 5020 | int md_run(struct mddev *mddev) |
1da177e4 | 5021 | { |
2604b703 | 5022 | int err; |
3cb03002 | 5023 | struct md_rdev *rdev; |
84fc4b56 | 5024 | struct md_personality *pers; |
1da177e4 | 5025 | |
a757e64c N |
5026 | if (list_empty(&mddev->disks)) |
5027 | /* cannot run an array with no devices.. */ | |
1da177e4 | 5028 | return -EINVAL; |
1da177e4 LT |
5029 | |
5030 | if (mddev->pers) | |
5031 | return -EBUSY; | |
bb4f1e9d N |
5032 | /* Cannot run until previous stop completes properly */ |
5033 | if (mddev->sysfs_active) | |
5034 | return -EBUSY; | |
b6eb127d | 5035 | |
1da177e4 LT |
5036 | /* |
5037 | * Analyze all RAID superblock(s) | |
5038 | */ | |
1ec4a939 N |
5039 | if (!mddev->raid_disks) { |
5040 | if (!mddev->persistent) | |
5041 | return -EINVAL; | |
a757e64c | 5042 | analyze_sbs(mddev); |
1ec4a939 | 5043 | } |
1da177e4 | 5044 | |
d9d166c2 N |
5045 | if (mddev->level != LEVEL_NONE) |
5046 | request_module("md-level-%d", mddev->level); | |
5047 | else if (mddev->clevel[0]) | |
5048 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
5049 | |
5050 | /* | |
5051 | * Drop all container device buffers, from now on | |
5052 | * the only valid external interface is through the md | |
5053 | * device. | |
1da177e4 | 5054 | */ |
dafb20fa | 5055 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 5056 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5057 | continue; |
5058 | sync_blockdev(rdev->bdev); | |
f98393a6 | 5059 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
5060 | |
5061 | /* perform some consistency tests on the device. | |
5062 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 5063 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 5064 | */ |
a6ff7e08 JB |
5065 | if (rdev->meta_bdev) { |
5066 | /* Nothing to check */; | |
5067 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
5068 | if (mddev->dev_sectors && |
5069 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 5070 | > rdev->sb_start) { |
f0d76d70 N |
5071 | printk("md: %s: data overlaps metadata\n", |
5072 | mdname(mddev)); | |
5073 | return -EINVAL; | |
5074 | } | |
5075 | } else { | |
0f420358 | 5076 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
5077 | > rdev->data_offset) { |
5078 | printk("md: %s: metadata overlaps data\n", | |
5079 | mdname(mddev)); | |
5080 | return -EINVAL; | |
5081 | } | |
5082 | } | |
00bcb4ac | 5083 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
5084 | } |
5085 | ||
a167f663 | 5086 | if (mddev->bio_set == NULL) |
395c72a7 | 5087 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0); |
a167f663 | 5088 | |
1da177e4 | 5089 | spin_lock(&pers_lock); |
d9d166c2 | 5090 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 5091 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 5092 | spin_unlock(&pers_lock); |
d9d166c2 N |
5093 | if (mddev->level != LEVEL_NONE) |
5094 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
5095 | mddev->level); | |
5096 | else | |
5097 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
5098 | mddev->clevel); | |
1da177e4 LT |
5099 | return -EINVAL; |
5100 | } | |
1da177e4 | 5101 | spin_unlock(&pers_lock); |
34817e8c N |
5102 | if (mddev->level != pers->level) { |
5103 | mddev->level = pers->level; | |
5104 | mddev->new_level = pers->level; | |
5105 | } | |
d9d166c2 | 5106 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5107 | |
f6705578 | 5108 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5109 | pers->start_reshape == NULL) { |
f6705578 | 5110 | /* This personality cannot handle reshaping... */ |
f6705578 N |
5111 | module_put(pers->owner); |
5112 | return -EINVAL; | |
5113 | } | |
5114 | ||
7dd5e7c3 N |
5115 | if (pers->sync_request) { |
5116 | /* Warn if this is a potentially silly | |
5117 | * configuration. | |
5118 | */ | |
5119 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5120 | struct md_rdev *rdev2; |
7dd5e7c3 | 5121 | int warned = 0; |
159ec1fc | 5122 | |
dafb20fa N |
5123 | rdev_for_each(rdev, mddev) |
5124 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5125 | if (rdev < rdev2 && |
5126 | rdev->bdev->bd_contains == | |
5127 | rdev2->bdev->bd_contains) { | |
5128 | printk(KERN_WARNING | |
5129 | "%s: WARNING: %s appears to be" | |
5130 | " on the same physical disk as" | |
5131 | " %s.\n", | |
5132 | mdname(mddev), | |
5133 | bdevname(rdev->bdev,b), | |
5134 | bdevname(rdev2->bdev,b2)); | |
5135 | warned = 1; | |
5136 | } | |
5137 | } | |
159ec1fc | 5138 | |
7dd5e7c3 N |
5139 | if (warned) |
5140 | printk(KERN_WARNING | |
5141 | "True protection against single-disk" | |
5142 | " failure might be compromised.\n"); | |
5143 | } | |
5144 | ||
657390d2 | 5145 | mddev->recovery = 0; |
58c0fed4 AN |
5146 | /* may be over-ridden by personality */ |
5147 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5148 | ||
6ff8d8ec | 5149 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5150 | |
0f9552b5 | 5151 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5152 | mddev->ro = 2; /* read-only, but switch on first write */ |
5153 | ||
36d091f4 | 5154 | err = pers->run(mddev); |
13e53df3 AN |
5155 | if (err) |
5156 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
36d091f4 | 5157 | else if (pers->size(mddev, 0, 0) < mddev->array_sectors) { |
b522adcd DW |
5158 | WARN_ONCE(!mddev->external_size, "%s: default size too small," |
5159 | " but 'external_size' not in effect?\n", __func__); | |
5160 | printk(KERN_ERR | |
5161 | "md: invalid array_size %llu > default size %llu\n", | |
5162 | (unsigned long long)mddev->array_sectors / 2, | |
36d091f4 | 5163 | (unsigned long long)pers->size(mddev, 0, 0) / 2); |
b522adcd | 5164 | err = -EINVAL; |
b522adcd | 5165 | } |
36d091f4 | 5166 | if (err == 0 && pers->sync_request && |
ef99bf48 | 5167 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { |
f9209a32 GR |
5168 | struct bitmap *bitmap; |
5169 | ||
5170 | bitmap = bitmap_create(mddev, -1); | |
5171 | if (IS_ERR(bitmap)) { | |
5172 | err = PTR_ERR(bitmap); | |
b15c2e57 N |
5173 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", |
5174 | mdname(mddev), err); | |
f9209a32 GR |
5175 | } else |
5176 | mddev->bitmap = bitmap; | |
5177 | ||
b15c2e57 | 5178 | } |
1da177e4 | 5179 | if (err) { |
5aa61f42 | 5180 | mddev_detach(mddev); |
0c35bd47 N |
5181 | if (mddev->private) |
5182 | pers->free(mddev, mddev->private); | |
bd691922 | 5183 | mddev->private = NULL; |
36d091f4 | 5184 | module_put(pers->owner); |
32a7627c N |
5185 | bitmap_destroy(mddev); |
5186 | return err; | |
1da177e4 | 5187 | } |
5c675f83 N |
5188 | if (mddev->queue) { |
5189 | mddev->queue->backing_dev_info.congested_data = mddev; | |
5190 | mddev->queue->backing_dev_info.congested_fn = md_congested; | |
64590f45 | 5191 | blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec); |
5c675f83 | 5192 | } |
36d091f4 | 5193 | if (pers->sync_request) { |
00bcb4ac N |
5194 | if (mddev->kobj.sd && |
5195 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
5196 | printk(KERN_WARNING |
5197 | "md: cannot register extra attributes for %s\n", | |
5198 | mdname(mddev)); | |
00bcb4ac | 5199 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5200 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5201 | mddev->ro = 0; |
5202 | ||
f72ffdd6 | 5203 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
5204 | atomic_set(&mddev->max_corr_read_errors, |
5205 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
5206 | mddev->safemode = 0; |
5207 | mddev->safemode_timer.function = md_safemode_timeout; | |
5208 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 5209 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 5210 | mddev->in_sync = 1; |
0ca69886 | 5211 | smp_wmb(); |
36d091f4 N |
5212 | spin_lock(&mddev->lock); |
5213 | mddev->pers = pers; | |
0ca69886 | 5214 | mddev->ready = 1; |
36d091f4 | 5215 | spin_unlock(&mddev->lock); |
dafb20fa | 5216 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5217 | if (rdev->raid_disk >= 0) |
5218 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5219 | /* failure here is OK */; |
f72ffdd6 | 5220 | |
a4a3d26d N |
5221 | if (mddev->degraded && !mddev->ro) |
5222 | /* This ensures that recovering status is reported immediately | |
5223 | * via sysfs - until a lack of spares is confirmed. | |
5224 | */ | |
5225 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
1da177e4 | 5226 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f72ffdd6 | 5227 | |
7a0a5355 | 5228 | if (mddev->flags & MD_UPDATE_SB_FLAGS) |
850b2b42 | 5229 | md_update_sb(mddev, 0); |
1da177e4 | 5230 | |
d7603b7e | 5231 | md_new_event(mddev); |
00bcb4ac N |
5232 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5233 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5234 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
5235 | return 0; |
5236 | } | |
390ee602 | 5237 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5238 | |
fd01b88c | 5239 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5240 | { |
5241 | int err; | |
5242 | ||
5243 | err = md_run(mddev); | |
5244 | if (err) | |
5245 | goto out; | |
69e51b44 N |
5246 | err = bitmap_load(mddev); |
5247 | if (err) { | |
5248 | bitmap_destroy(mddev); | |
5249 | goto out; | |
5250 | } | |
0fd018af JB |
5251 | |
5252 | md_wakeup_thread(mddev->thread); | |
5253 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5254 | ||
fe60b014 N |
5255 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5256 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5257 | mddev->changed = 1; |
fe60b014 N |
5258 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5259 | out: | |
5260 | return err; | |
5261 | } | |
5262 | ||
fd01b88c | 5263 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5264 | { |
5265 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 5266 | |
80fab1d7 | 5267 | /* Complain if it has no devices */ |
1da177e4 | 5268 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5269 | return -ENXIO; |
5270 | if (!mddev->pers) | |
5271 | return -EINVAL; | |
5272 | if (!mddev->ro) | |
5273 | return -EBUSY; | |
5274 | mddev->safemode = 0; | |
5275 | mddev->ro = 0; | |
5276 | set_disk_ro(disk, 0); | |
5277 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
5278 | mdname(mddev)); | |
5279 | /* Kick recovery or resync if necessary */ | |
5280 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5281 | md_wakeup_thread(mddev->thread); | |
5282 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5283 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5284 | return 0; |
1da177e4 LT |
5285 | } |
5286 | ||
fd01b88c | 5287 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5288 | { |
5289 | mddev->array_sectors = 0; | |
5290 | mddev->external_size = 0; | |
5291 | mddev->dev_sectors = 0; | |
5292 | mddev->raid_disks = 0; | |
5293 | mddev->recovery_cp = 0; | |
5294 | mddev->resync_min = 0; | |
5295 | mddev->resync_max = MaxSector; | |
5296 | mddev->reshape_position = MaxSector; | |
5297 | mddev->external = 0; | |
5298 | mddev->persistent = 0; | |
5299 | mddev->level = LEVEL_NONE; | |
5300 | mddev->clevel[0] = 0; | |
5301 | mddev->flags = 0; | |
5302 | mddev->ro = 0; | |
5303 | mddev->metadata_type[0] = 0; | |
5304 | mddev->chunk_sectors = 0; | |
5305 | mddev->ctime = mddev->utime = 0; | |
5306 | mddev->layout = 0; | |
5307 | mddev->max_disks = 0; | |
5308 | mddev->events = 0; | |
a8707c08 | 5309 | mddev->can_decrease_events = 0; |
6177b472 | 5310 | mddev->delta_disks = 0; |
2c810cdd | 5311 | mddev->reshape_backwards = 0; |
6177b472 N |
5312 | mddev->new_level = LEVEL_NONE; |
5313 | mddev->new_layout = 0; | |
5314 | mddev->new_chunk_sectors = 0; | |
5315 | mddev->curr_resync = 0; | |
7f7583d4 | 5316 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5317 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5318 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5319 | mddev->recovery = 0; | |
5320 | mddev->in_sync = 0; | |
f0b4f7e2 | 5321 | mddev->changed = 0; |
6177b472 | 5322 | mddev->degraded = 0; |
6177b472 | 5323 | mddev->safemode = 0; |
bd691922 | 5324 | mddev->private = NULL; |
050b6615 | 5325 | mddev->merge_check_needed = 0; |
6177b472 N |
5326 | mddev->bitmap_info.offset = 0; |
5327 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5328 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5329 | mddev->bitmap_info.chunksize = 0; |
5330 | mddev->bitmap_info.daemon_sleep = 0; | |
5331 | mddev->bitmap_info.max_write_behind = 0; | |
5332 | } | |
5333 | ||
fd01b88c | 5334 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5335 | { |
293467aa GR |
5336 | if (mddev_is_clustered(mddev)) |
5337 | md_cluster_ops->metadata_update_start(mddev); | |
6b6204ee | 5338 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
f851b60d | 5339 | flush_workqueue(md_misc_wq); |
a047e125 | 5340 | if (mddev->sync_thread) { |
a047e125 | 5341 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5342 | md_reap_sync_thread(mddev); |
a047e125 N |
5343 | } |
5344 | ||
5345 | del_timer_sync(&mddev->safemode_timer); | |
5346 | ||
5347 | bitmap_flush(mddev); | |
5348 | md_super_wait(mddev); | |
5349 | ||
b6d428c6 | 5350 | if (mddev->ro == 0 && |
7a0a5355 | 5351 | (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) { |
a047e125 N |
5352 | /* mark array as shutdown cleanly */ |
5353 | mddev->in_sync = 1; | |
5354 | md_update_sb(mddev, 1); | |
5355 | } | |
293467aa GR |
5356 | if (mddev_is_clustered(mddev)) |
5357 | md_cluster_ops->metadata_update_finish(mddev); | |
a047e125 | 5358 | } |
defad61a | 5359 | |
fd01b88c | 5360 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5361 | { |
29f097c4 | 5362 | mddev_lock_nointr(mddev); |
defad61a N |
5363 | __md_stop_writes(mddev); |
5364 | mddev_unlock(mddev); | |
5365 | } | |
390ee602 | 5366 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5367 | |
5aa61f42 N |
5368 | static void mddev_detach(struct mddev *mddev) |
5369 | { | |
5370 | struct bitmap *bitmap = mddev->bitmap; | |
5371 | /* wait for behind writes to complete */ | |
5372 | if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { | |
5373 | printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n", | |
5374 | mdname(mddev)); | |
5375 | /* need to kick something here to make sure I/O goes? */ | |
5376 | wait_event(bitmap->behind_wait, | |
5377 | atomic_read(&bitmap->behind_writes) == 0); | |
5378 | } | |
36d091f4 | 5379 | if (mddev->pers && mddev->pers->quiesce) { |
5aa61f42 N |
5380 | mddev->pers->quiesce(mddev, 1); |
5381 | mddev->pers->quiesce(mddev, 0); | |
5382 | } | |
5383 | md_unregister_thread(&mddev->thread); | |
5384 | if (mddev->queue) | |
5385 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
5386 | } | |
5387 | ||
5eff3c43 | 5388 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5389 | { |
36d091f4 | 5390 | struct md_personality *pers = mddev->pers; |
5aa61f42 | 5391 | mddev_detach(mddev); |
ee5d004f N |
5392 | /* Ensure ->event_work is done */ |
5393 | flush_workqueue(md_misc_wq); | |
36d091f4 N |
5394 | spin_lock(&mddev->lock); |
5395 | mddev->ready = 0; | |
6177b472 | 5396 | mddev->pers = NULL; |
36d091f4 N |
5397 | spin_unlock(&mddev->lock); |
5398 | pers->free(mddev, mddev->private); | |
bd691922 | 5399 | mddev->private = NULL; |
36d091f4 N |
5400 | if (pers->sync_request && mddev->to_remove == NULL) |
5401 | mddev->to_remove = &md_redundancy_group; | |
5402 | module_put(pers->owner); | |
cca9cf90 | 5403 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5404 | } |
5eff3c43 N |
5405 | |
5406 | void md_stop(struct mddev *mddev) | |
5407 | { | |
5408 | /* stop the array and free an attached data structures. | |
5409 | * This is called from dm-raid | |
5410 | */ | |
5411 | __md_stop(mddev); | |
5412 | bitmap_destroy(mddev); | |
5413 | if (mddev->bio_set) | |
5414 | bioset_free(mddev->bio_set); | |
5415 | } | |
5416 | ||
390ee602 | 5417 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5418 | |
a05b7ea0 | 5419 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5420 | { |
5421 | int err = 0; | |
30b8feb7 N |
5422 | int did_freeze = 0; |
5423 | ||
5424 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5425 | did_freeze = 1; | |
5426 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5427 | md_wakeup_thread(mddev->thread); | |
5428 | } | |
f851b60d | 5429 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5430 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5431 | if (mddev->sync_thread) |
30b8feb7 N |
5432 | /* Thread might be blocked waiting for metadata update |
5433 | * which will now never happen */ | |
5434 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5435 | |
30b8feb7 | 5436 | mddev_unlock(mddev); |
f851b60d N |
5437 | wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING, |
5438 | &mddev->recovery)); | |
30b8feb7 N |
5439 | mddev_lock_nointr(mddev); |
5440 | ||
a4bd82d0 | 5441 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5442 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 | 5443 | mddev->sync_thread || |
f851b60d | 5444 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
30b8feb7 | 5445 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { |
a4bd82d0 | 5446 | printk("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5447 | if (did_freeze) { |
5448 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5449 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5450 | md_wakeup_thread(mddev->thread); |
5451 | } | |
a4bd82d0 N |
5452 | err = -EBUSY; |
5453 | goto out; | |
5454 | } | |
5455 | if (mddev->pers) { | |
defad61a | 5456 | __md_stop_writes(mddev); |
a4bd82d0 N |
5457 | |
5458 | err = -ENXIO; | |
5459 | if (mddev->ro==1) | |
5460 | goto out; | |
5461 | mddev->ro = 1; | |
5462 | set_disk_ro(mddev->gendisk, 1); | |
5463 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d N |
5464 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5465 | md_wakeup_thread(mddev->thread); | |
00bcb4ac | 5466 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5467 | err = 0; |
a4bd82d0 N |
5468 | } |
5469 | out: | |
5470 | mutex_unlock(&mddev->open_mutex); | |
5471 | return err; | |
5472 | } | |
5473 | ||
9e653b63 N |
5474 | /* mode: |
5475 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5476 | * 2 - stop but do not disassemble array |
5477 | */ | |
f72ffdd6 | 5478 | static int do_md_stop(struct mddev *mddev, int mode, |
a05b7ea0 | 5479 | struct block_device *bdev) |
1da177e4 | 5480 | { |
1da177e4 | 5481 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5482 | struct md_rdev *rdev; |
30b8feb7 N |
5483 | int did_freeze = 0; |
5484 | ||
5485 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5486 | did_freeze = 1; | |
5487 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5488 | md_wakeup_thread(mddev->thread); | |
5489 | } | |
f851b60d | 5490 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5491 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5492 | if (mddev->sync_thread) |
30b8feb7 N |
5493 | /* Thread might be blocked waiting for metadata update |
5494 | * which will now never happen */ | |
5495 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5496 | |
30b8feb7 | 5497 | mddev_unlock(mddev); |
f851b60d N |
5498 | wait_event(resync_wait, (mddev->sync_thread == NULL && |
5499 | !test_bit(MD_RECOVERY_RUNNING, | |
5500 | &mddev->recovery))); | |
30b8feb7 | 5501 | mddev_lock_nointr(mddev); |
1da177e4 | 5502 | |
c8c00a69 | 5503 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5504 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5505 | mddev->sysfs_active || |
5506 | mddev->sync_thread || | |
f851b60d | 5507 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
30b8feb7 | 5508 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { |
df5b20cf | 5509 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 5510 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
5511 | if (did_freeze) { |
5512 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5513 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5514 | md_wakeup_thread(mddev->thread); |
5515 | } | |
260fa034 N |
5516 | return -EBUSY; |
5517 | } | |
6e17b027 | 5518 | if (mddev->pers) { |
a4bd82d0 N |
5519 | if (mddev->ro) |
5520 | set_disk_ro(disk, 0); | |
409c57f3 | 5521 | |
defad61a | 5522 | __md_stop_writes(mddev); |
5eff3c43 | 5523 | __md_stop(mddev); |
a4bd82d0 | 5524 | mddev->queue->merge_bvec_fn = NULL; |
a4bd82d0 | 5525 | mddev->queue->backing_dev_info.congested_fn = NULL; |
6177b472 | 5526 | |
a4bd82d0 | 5527 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5528 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5529 | |
dafb20fa | 5530 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5531 | if (rdev->raid_disk >= 0) |
5532 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5533 | |
a4bd82d0 | 5534 | set_capacity(disk, 0); |
6e17b027 | 5535 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5536 | mddev->changed = 1; |
a4bd82d0 | 5537 | revalidate_disk(disk); |
0d4ca600 | 5538 | |
a4bd82d0 N |
5539 | if (mddev->ro) |
5540 | mddev->ro = 0; | |
6e17b027 N |
5541 | } else |
5542 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5543 | /* |
5544 | * Free resources if final stop | |
5545 | */ | |
9e653b63 | 5546 | if (mode == 0) { |
1da177e4 LT |
5547 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
5548 | ||
978f946b | 5549 | bitmap_destroy(mddev); |
c3d9714e | 5550 | if (mddev->bitmap_info.file) { |
4af1a041 N |
5551 | struct file *f = mddev->bitmap_info.file; |
5552 | spin_lock(&mddev->lock); | |
c3d9714e | 5553 | mddev->bitmap_info.file = NULL; |
4af1a041 N |
5554 | spin_unlock(&mddev->lock); |
5555 | fput(f); | |
978f946b | 5556 | } |
c3d9714e | 5557 | mddev->bitmap_info.offset = 0; |
978f946b | 5558 | |
1da177e4 LT |
5559 | export_array(mddev); |
5560 | ||
6177b472 | 5561 | md_clean(mddev); |
934d9c23 | 5562 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
5563 | if (mddev->hold_active == UNTIL_STOP) |
5564 | mddev->hold_active = 0; | |
a4bd82d0 | 5565 | } |
3f9d99c1 | 5566 | blk_integrity_unregister(disk); |
d7603b7e | 5567 | md_new_event(mddev); |
00bcb4ac | 5568 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 5569 | return 0; |
1da177e4 LT |
5570 | } |
5571 | ||
fdee8ae4 | 5572 | #ifndef MODULE |
fd01b88c | 5573 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 5574 | { |
3cb03002 | 5575 | struct md_rdev *rdev; |
1da177e4 LT |
5576 | int err; |
5577 | ||
a757e64c | 5578 | if (list_empty(&mddev->disks)) |
1da177e4 | 5579 | return; |
1da177e4 LT |
5580 | |
5581 | printk(KERN_INFO "md: running: "); | |
5582 | ||
dafb20fa | 5583 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
5584 | char b[BDEVNAME_SIZE]; |
5585 | printk("<%s>", bdevname(rdev->bdev,b)); | |
5586 | } | |
5587 | printk("\n"); | |
5588 | ||
d710e138 | 5589 | err = do_md_run(mddev); |
1da177e4 LT |
5590 | if (err) { |
5591 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
a05b7ea0 | 5592 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
5593 | } |
5594 | } | |
5595 | ||
5596 | /* | |
5597 | * lets try to run arrays based on all disks that have arrived | |
5598 | * until now. (those are in pending_raid_disks) | |
5599 | * | |
5600 | * the method: pick the first pending disk, collect all disks with | |
5601 | * the same UUID, remove all from the pending list and put them into | |
5602 | * the 'same_array' list. Then order this list based on superblock | |
5603 | * update time (freshest comes first), kick out 'old' disks and | |
5604 | * compare superblocks. If everything's fine then run it. | |
5605 | * | |
5606 | * If "unit" is allocated, then bump its reference count | |
5607 | */ | |
5608 | static void autorun_devices(int part) | |
5609 | { | |
3cb03002 | 5610 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 5611 | struct mddev *mddev; |
1da177e4 LT |
5612 | char b[BDEVNAME_SIZE]; |
5613 | ||
5614 | printk(KERN_INFO "md: autorun ...\n"); | |
5615 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 5616 | int unit; |
1da177e4 | 5617 | dev_t dev; |
ad01c9e3 | 5618 | LIST_HEAD(candidates); |
1da177e4 | 5619 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 5620 | struct md_rdev, same_set); |
1da177e4 LT |
5621 | |
5622 | printk(KERN_INFO "md: considering %s ...\n", | |
5623 | bdevname(rdev0->bdev,b)); | |
5624 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 5625 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
5626 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
5627 | printk(KERN_INFO "md: adding %s ...\n", | |
5628 | bdevname(rdev->bdev,b)); | |
5629 | list_move(&rdev->same_set, &candidates); | |
5630 | } | |
5631 | /* | |
5632 | * now we have a set of devices, with all of them having | |
5633 | * mostly sane superblocks. It's time to allocate the | |
5634 | * mddev. | |
5635 | */ | |
e8703fe1 N |
5636 | if (part) { |
5637 | dev = MKDEV(mdp_major, | |
5638 | rdev0->preferred_minor << MdpMinorShift); | |
5639 | unit = MINOR(dev) >> MdpMinorShift; | |
5640 | } else { | |
5641 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
5642 | unit = MINOR(dev); | |
5643 | } | |
5644 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
5645 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
5646 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
5647 | break; | |
5648 | } | |
1da177e4 LT |
5649 | |
5650 | md_probe(dev, NULL, NULL); | |
5651 | mddev = mddev_find(dev); | |
9bbbca3a NB |
5652 | if (!mddev || !mddev->gendisk) { |
5653 | if (mddev) | |
5654 | mddev_put(mddev); | |
5655 | printk(KERN_ERR | |
1da177e4 LT |
5656 | "md: cannot allocate memory for md drive.\n"); |
5657 | break; | |
5658 | } | |
f72ffdd6 | 5659 | if (mddev_lock(mddev)) |
1da177e4 LT |
5660 | printk(KERN_WARNING "md: %s locked, cannot run\n", |
5661 | mdname(mddev)); | |
5662 | else if (mddev->raid_disks || mddev->major_version | |
5663 | || !list_empty(&mddev->disks)) { | |
f72ffdd6 | 5664 | printk(KERN_WARNING |
1da177e4 LT |
5665 | "md: %s already running, cannot run %s\n", |
5666 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
5667 | mddev_unlock(mddev); | |
5668 | } else { | |
5669 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 5670 | mddev->persistent = 1; |
159ec1fc | 5671 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
5672 | list_del_init(&rdev->same_set); |
5673 | if (bind_rdev_to_array(rdev, mddev)) | |
5674 | export_rdev(rdev); | |
5675 | } | |
5676 | autorun_array(mddev); | |
5677 | mddev_unlock(mddev); | |
5678 | } | |
5679 | /* on success, candidates will be empty, on error | |
5680 | * it won't... | |
5681 | */ | |
159ec1fc | 5682 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 5683 | list_del_init(&rdev->same_set); |
1da177e4 | 5684 | export_rdev(rdev); |
4b80991c | 5685 | } |
1da177e4 LT |
5686 | mddev_put(mddev); |
5687 | } | |
5688 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
5689 | } | |
fdee8ae4 | 5690 | #endif /* !MODULE */ |
1da177e4 | 5691 | |
f72ffdd6 | 5692 | static int get_version(void __user *arg) |
1da177e4 LT |
5693 | { |
5694 | mdu_version_t ver; | |
5695 | ||
5696 | ver.major = MD_MAJOR_VERSION; | |
5697 | ver.minor = MD_MINOR_VERSION; | |
5698 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
5699 | ||
5700 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
5701 | return -EFAULT; | |
5702 | ||
5703 | return 0; | |
5704 | } | |
5705 | ||
f72ffdd6 | 5706 | static int get_array_info(struct mddev *mddev, void __user *arg) |
1da177e4 LT |
5707 | { |
5708 | mdu_array_info_t info; | |
a9f326eb | 5709 | int nr,working,insync,failed,spare; |
3cb03002 | 5710 | struct md_rdev *rdev; |
1da177e4 | 5711 | |
1ca69c4b N |
5712 | nr = working = insync = failed = spare = 0; |
5713 | rcu_read_lock(); | |
5714 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 5715 | nr++; |
b2d444d7 | 5716 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5717 | failed++; |
5718 | else { | |
5719 | working++; | |
b2d444d7 | 5720 | if (test_bit(In_sync, &rdev->flags)) |
f72ffdd6 | 5721 | insync++; |
1da177e4 LT |
5722 | else |
5723 | spare++; | |
5724 | } | |
5725 | } | |
1ca69c4b | 5726 | rcu_read_unlock(); |
1da177e4 LT |
5727 | |
5728 | info.major_version = mddev->major_version; | |
5729 | info.minor_version = mddev->minor_version; | |
5730 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5731 | info.ctime = mddev->ctime; | |
5732 | info.level = mddev->level; | |
58c0fed4 AN |
5733 | info.size = mddev->dev_sectors / 2; |
5734 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5735 | info.size = -1; |
1da177e4 LT |
5736 | info.nr_disks = nr; |
5737 | info.raid_disks = mddev->raid_disks; | |
5738 | info.md_minor = mddev->md_minor; | |
5739 | info.not_persistent= !mddev->persistent; | |
5740 | ||
5741 | info.utime = mddev->utime; | |
5742 | info.state = 0; | |
5743 | if (mddev->in_sync) | |
5744 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5745 | if (mddev->bitmap && mddev->bitmap_info.offset) |
9bd35920 | 5746 | info.state |= (1<<MD_SB_BITMAP_PRESENT); |
ca8895d9 GR |
5747 | if (mddev_is_clustered(mddev)) |
5748 | info.state |= (1<<MD_SB_CLUSTERED); | |
a9f326eb | 5749 | info.active_disks = insync; |
1da177e4 LT |
5750 | info.working_disks = working; |
5751 | info.failed_disks = failed; | |
5752 | info.spare_disks = spare; | |
5753 | ||
5754 | info.layout = mddev->layout; | |
9d8f0363 | 5755 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5756 | |
5757 | if (copy_to_user(arg, &info, sizeof(info))) | |
5758 | return -EFAULT; | |
5759 | ||
5760 | return 0; | |
5761 | } | |
5762 | ||
f72ffdd6 | 5763 | static int get_bitmap_file(struct mddev *mddev, void __user * arg) |
32a7627c N |
5764 | { |
5765 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
f4ad3d38 | 5766 | char *ptr; |
4af1a041 | 5767 | int err; |
32a7627c | 5768 | |
b6878d9e | 5769 | file = kzalloc(sizeof(*file), GFP_NOIO); |
32a7627c | 5770 | if (!file) |
4af1a041 | 5771 | return -ENOMEM; |
32a7627c | 5772 | |
4af1a041 N |
5773 | err = 0; |
5774 | spin_lock(&mddev->lock); | |
25eafe1a BR |
5775 | /* bitmap enabled */ |
5776 | if (mddev->bitmap_info.file) { | |
5777 | ptr = file_path(mddev->bitmap_info.file, file->pathname, | |
5778 | sizeof(file->pathname)); | |
5779 | if (IS_ERR(ptr)) | |
5780 | err = PTR_ERR(ptr); | |
5781 | else | |
5782 | memmove(file->pathname, ptr, | |
5783 | sizeof(file->pathname)-(ptr-file->pathname)); | |
5784 | } | |
4af1a041 | 5785 | spin_unlock(&mddev->lock); |
32a7627c | 5786 | |
4af1a041 N |
5787 | if (err == 0 && |
5788 | copy_to_user(arg, file, sizeof(*file))) | |
32a7627c | 5789 | err = -EFAULT; |
4af1a041 | 5790 | |
32a7627c N |
5791 | kfree(file); |
5792 | return err; | |
5793 | } | |
5794 | ||
f72ffdd6 | 5795 | static int get_disk_info(struct mddev *mddev, void __user * arg) |
1da177e4 LT |
5796 | { |
5797 | mdu_disk_info_t info; | |
3cb03002 | 5798 | struct md_rdev *rdev; |
1da177e4 LT |
5799 | |
5800 | if (copy_from_user(&info, arg, sizeof(info))) | |
5801 | return -EFAULT; | |
5802 | ||
1ca69c4b | 5803 | rcu_read_lock(); |
57d051dc | 5804 | rdev = md_find_rdev_nr_rcu(mddev, info.number); |
1da177e4 LT |
5805 | if (rdev) { |
5806 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5807 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5808 | info.raid_disk = rdev->raid_disk; | |
5809 | info.state = 0; | |
b2d444d7 | 5810 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5811 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5812 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5813 | info.state |= (1<<MD_DISK_ACTIVE); |
5814 | info.state |= (1<<MD_DISK_SYNC); | |
5815 | } | |
8ddf9efe N |
5816 | if (test_bit(WriteMostly, &rdev->flags)) |
5817 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5818 | } else { |
5819 | info.major = info.minor = 0; | |
5820 | info.raid_disk = -1; | |
5821 | info.state = (1<<MD_DISK_REMOVED); | |
5822 | } | |
1ca69c4b | 5823 | rcu_read_unlock(); |
1da177e4 LT |
5824 | |
5825 | if (copy_to_user(arg, &info, sizeof(info))) | |
5826 | return -EFAULT; | |
5827 | ||
5828 | return 0; | |
5829 | } | |
5830 | ||
f72ffdd6 | 5831 | static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info) |
1da177e4 LT |
5832 | { |
5833 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5834 | struct md_rdev *rdev; |
1da177e4 LT |
5835 | dev_t dev = MKDEV(info->major,info->minor); |
5836 | ||
1aee41f6 GR |
5837 | if (mddev_is_clustered(mddev) && |
5838 | !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) { | |
fa8259da | 5839 | pr_err("%s: Cannot add to clustered mddev.\n", |
1aee41f6 GR |
5840 | mdname(mddev)); |
5841 | return -EINVAL; | |
5842 | } | |
5843 | ||
1da177e4 LT |
5844 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) |
5845 | return -EOVERFLOW; | |
5846 | ||
5847 | if (!mddev->raid_disks) { | |
5848 | int err; | |
5849 | /* expecting a device which has a superblock */ | |
5850 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5851 | if (IS_ERR(rdev)) { | |
f72ffdd6 | 5852 | printk(KERN_WARNING |
1da177e4 LT |
5853 | "md: md_import_device returned %ld\n", |
5854 | PTR_ERR(rdev)); | |
5855 | return PTR_ERR(rdev); | |
5856 | } | |
5857 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
5858 | struct md_rdev *rdev0 |
5859 | = list_entry(mddev->disks.next, | |
5860 | struct md_rdev, same_set); | |
a9f326eb | 5861 | err = super_types[mddev->major_version] |
1da177e4 LT |
5862 | .load_super(rdev, rdev0, mddev->minor_version); |
5863 | if (err < 0) { | |
f72ffdd6 | 5864 | printk(KERN_WARNING |
1da177e4 | 5865 | "md: %s has different UUID to %s\n", |
f72ffdd6 | 5866 | bdevname(rdev->bdev,b), |
1da177e4 LT |
5867 | bdevname(rdev0->bdev,b2)); |
5868 | export_rdev(rdev); | |
5869 | return -EINVAL; | |
5870 | } | |
5871 | } | |
5872 | err = bind_rdev_to_array(rdev, mddev); | |
5873 | if (err) | |
5874 | export_rdev(rdev); | |
5875 | return err; | |
5876 | } | |
5877 | ||
5878 | /* | |
5879 | * add_new_disk can be used once the array is assembled | |
5880 | * to add "hot spares". They must already have a superblock | |
5881 | * written | |
5882 | */ | |
5883 | if (mddev->pers) { | |
5884 | int err; | |
5885 | if (!mddev->pers->hot_add_disk) { | |
f72ffdd6 | 5886 | printk(KERN_WARNING |
1da177e4 LT |
5887 | "%s: personality does not support diskops!\n", |
5888 | mdname(mddev)); | |
5889 | return -EINVAL; | |
5890 | } | |
7b1e35f6 N |
5891 | if (mddev->persistent) |
5892 | rdev = md_import_device(dev, mddev->major_version, | |
5893 | mddev->minor_version); | |
5894 | else | |
5895 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 | 5896 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5897 | printk(KERN_WARNING |
1da177e4 LT |
5898 | "md: md_import_device returned %ld\n", |
5899 | PTR_ERR(rdev)); | |
5900 | return PTR_ERR(rdev); | |
5901 | } | |
1a855a06 | 5902 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5903 | if (!mddev->persistent) { |
5904 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5905 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5906 | rdev->raid_disk = info->raid_disk; |
bf572541 | 5907 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 5908 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 5909 | } else |
41158c7e | 5910 | rdev->raid_disk = -1; |
f466722c | 5911 | rdev->saved_raid_disk = rdev->raid_disk; |
41158c7e N |
5912 | } else |
5913 | super_types[mddev->major_version]. | |
5914 | validate_super(mddev, rdev); | |
bedd86b7 | 5915 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 5916 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
5917 | /* This was a hot-add request, but events doesn't |
5918 | * match, so reject it. | |
5919 | */ | |
5920 | export_rdev(rdev); | |
5921 | return -EINVAL; | |
5922 | } | |
5923 | ||
b2d444d7 | 5924 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5925 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5926 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5927 | else |
5928 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5929 | |
1aee41f6 GR |
5930 | /* |
5931 | * check whether the device shows up in other nodes | |
5932 | */ | |
5933 | if (mddev_is_clustered(mddev)) { | |
5934 | if (info->state & (1 << MD_DISK_CANDIDATE)) { | |
5935 | /* Through --cluster-confirm */ | |
5936 | set_bit(Candidate, &rdev->flags); | |
fa8259da GR |
5937 | err = md_cluster_ops->new_disk_ack(mddev, true); |
5938 | if (err) { | |
5939 | export_rdev(rdev); | |
5940 | return err; | |
5941 | } | |
1aee41f6 GR |
5942 | } else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) { |
5943 | /* --add initiated by this node */ | |
5944 | err = md_cluster_ops->add_new_disk_start(mddev, rdev); | |
5945 | if (err) { | |
5946 | md_cluster_ops->add_new_disk_finish(mddev); | |
5947 | export_rdev(rdev); | |
5948 | return err; | |
5949 | } | |
5950 | } | |
5951 | } | |
5952 | ||
1da177e4 LT |
5953 | rdev->raid_disk = -1; |
5954 | err = bind_rdev_to_array(rdev, mddev); | |
5955 | if (err) | |
5956 | export_rdev(rdev); | |
52664732 | 5957 | else |
a6da4ef8 | 5958 | err = add_bound_rdev(rdev); |
1aee41f6 GR |
5959 | if (mddev_is_clustered(mddev) && |
5960 | (info->state & (1 << MD_DISK_CLUSTER_ADD))) | |
5961 | md_cluster_ops->add_new_disk_finish(mddev); | |
1da177e4 LT |
5962 | return err; |
5963 | } | |
5964 | ||
5965 | /* otherwise, add_new_disk is only allowed | |
5966 | * for major_version==0 superblocks | |
5967 | */ | |
5968 | if (mddev->major_version != 0) { | |
5969 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5970 | mdname(mddev)); | |
5971 | return -EINVAL; | |
5972 | } | |
5973 | ||
5974 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5975 | int err; | |
d710e138 | 5976 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 5977 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5978 | printk(KERN_WARNING |
1da177e4 LT |
5979 | "md: error, md_import_device() returned %ld\n", |
5980 | PTR_ERR(rdev)); | |
5981 | return PTR_ERR(rdev); | |
5982 | } | |
5983 | rdev->desc_nr = info->number; | |
5984 | if (info->raid_disk < mddev->raid_disks) | |
5985 | rdev->raid_disk = info->raid_disk; | |
5986 | else | |
5987 | rdev->raid_disk = -1; | |
5988 | ||
1da177e4 | 5989 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5990 | if (info->state & (1<<MD_DISK_SYNC)) |
5991 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5992 | |
8ddf9efe N |
5993 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5994 | set_bit(WriteMostly, &rdev->flags); | |
5995 | ||
1da177e4 LT |
5996 | if (!mddev->persistent) { |
5997 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5998 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5999 | } else | |
57b2caa3 | 6000 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 6001 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6002 | |
2bf071bf N |
6003 | err = bind_rdev_to_array(rdev, mddev); |
6004 | if (err) { | |
6005 | export_rdev(rdev); | |
6006 | return err; | |
6007 | } | |
1da177e4 LT |
6008 | } |
6009 | ||
6010 | return 0; | |
6011 | } | |
6012 | ||
f72ffdd6 | 6013 | static int hot_remove_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6014 | { |
6015 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 6016 | struct md_rdev *rdev; |
1da177e4 | 6017 | |
1da177e4 LT |
6018 | rdev = find_rdev(mddev, dev); |
6019 | if (!rdev) | |
6020 | return -ENXIO; | |
6021 | ||
293467aa GR |
6022 | if (mddev_is_clustered(mddev)) |
6023 | md_cluster_ops->metadata_update_start(mddev); | |
6024 | ||
3ea8929d N |
6025 | clear_bit(Blocked, &rdev->flags); |
6026 | remove_and_add_spares(mddev, rdev); | |
6027 | ||
1da177e4 LT |
6028 | if (rdev->raid_disk >= 0) |
6029 | goto busy; | |
6030 | ||
88bcfef7 GR |
6031 | if (mddev_is_clustered(mddev)) |
6032 | md_cluster_ops->remove_disk(mddev, rdev); | |
6033 | ||
fb56dfef | 6034 | md_kick_rdev_from_array(rdev); |
850b2b42 | 6035 | md_update_sb(mddev, 1); |
d7603b7e | 6036 | md_new_event(mddev); |
1da177e4 | 6037 | |
293467aa GR |
6038 | if (mddev_is_clustered(mddev)) |
6039 | md_cluster_ops->metadata_update_finish(mddev); | |
6040 | ||
1da177e4 LT |
6041 | return 0; |
6042 | busy: | |
293467aa GR |
6043 | if (mddev_is_clustered(mddev)) |
6044 | md_cluster_ops->metadata_update_cancel(mddev); | |
fdefa4d8 | 6045 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
6046 | bdevname(rdev->bdev,b), mdname(mddev)); |
6047 | return -EBUSY; | |
6048 | } | |
6049 | ||
f72ffdd6 | 6050 | static int hot_add_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6051 | { |
6052 | char b[BDEVNAME_SIZE]; | |
6053 | int err; | |
3cb03002 | 6054 | struct md_rdev *rdev; |
1da177e4 LT |
6055 | |
6056 | if (!mddev->pers) | |
6057 | return -ENODEV; | |
6058 | ||
6059 | if (mddev->major_version != 0) { | |
6060 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
6061 | " version-0 superblocks.\n", | |
6062 | mdname(mddev)); | |
6063 | return -EINVAL; | |
6064 | } | |
6065 | if (!mddev->pers->hot_add_disk) { | |
f72ffdd6 | 6066 | printk(KERN_WARNING |
1da177e4 LT |
6067 | "%s: personality does not support diskops!\n", |
6068 | mdname(mddev)); | |
6069 | return -EINVAL; | |
6070 | } | |
6071 | ||
d710e138 | 6072 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6073 | if (IS_ERR(rdev)) { |
f72ffdd6 | 6074 | printk(KERN_WARNING |
1da177e4 LT |
6075 | "md: error, md_import_device() returned %ld\n", |
6076 | PTR_ERR(rdev)); | |
6077 | return -EINVAL; | |
6078 | } | |
6079 | ||
6080 | if (mddev->persistent) | |
57b2caa3 | 6081 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 6082 | else |
77304d2a | 6083 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 6084 | |
8190e754 | 6085 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6086 | |
b2d444d7 | 6087 | if (test_bit(Faulty, &rdev->flags)) { |
f72ffdd6 | 6088 | printk(KERN_WARNING |
1da177e4 LT |
6089 | "md: can not hot-add faulty %s disk to %s!\n", |
6090 | bdevname(rdev->bdev,b), mdname(mddev)); | |
6091 | err = -EINVAL; | |
6092 | goto abort_export; | |
6093 | } | |
293467aa GR |
6094 | |
6095 | if (mddev_is_clustered(mddev)) | |
6096 | md_cluster_ops->metadata_update_start(mddev); | |
b2d444d7 | 6097 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 6098 | rdev->desc_nr = -1; |
5842730d | 6099 | rdev->saved_raid_disk = -1; |
2bf071bf N |
6100 | err = bind_rdev_to_array(rdev, mddev); |
6101 | if (err) | |
293467aa | 6102 | goto abort_clustered; |
1da177e4 LT |
6103 | |
6104 | /* | |
6105 | * The rest should better be atomic, we can have disk failures | |
6106 | * noticed in interrupt contexts ... | |
6107 | */ | |
6108 | ||
1da177e4 LT |
6109 | rdev->raid_disk = -1; |
6110 | ||
850b2b42 | 6111 | md_update_sb(mddev, 1); |
1da177e4 | 6112 | |
293467aa GR |
6113 | if (mddev_is_clustered(mddev)) |
6114 | md_cluster_ops->metadata_update_finish(mddev); | |
1da177e4 LT |
6115 | /* |
6116 | * Kick recovery, maybe this spare has to be added to the | |
6117 | * array immediately. | |
6118 | */ | |
6119 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6120 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 6121 | md_new_event(mddev); |
1da177e4 LT |
6122 | return 0; |
6123 | ||
293467aa GR |
6124 | abort_clustered: |
6125 | if (mddev_is_clustered(mddev)) | |
6126 | md_cluster_ops->metadata_update_cancel(mddev); | |
1da177e4 LT |
6127 | abort_export: |
6128 | export_rdev(rdev); | |
6129 | return err; | |
6130 | } | |
6131 | ||
fd01b88c | 6132 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c | 6133 | { |
035328c2 | 6134 | int err = 0; |
32a7627c | 6135 | |
36fa3063 | 6136 | if (mddev->pers) { |
d66b1b39 | 6137 | if (!mddev->pers->quiesce || !mddev->thread) |
36fa3063 N |
6138 | return -EBUSY; |
6139 | if (mddev->recovery || mddev->sync_thread) | |
6140 | return -EBUSY; | |
6141 | /* we should be able to change the bitmap.. */ | |
6142 | } | |
32a7627c | 6143 | |
36fa3063 | 6144 | if (fd >= 0) { |
035328c2 | 6145 | struct inode *inode; |
1e594bb2 N |
6146 | struct file *f; |
6147 | ||
6148 | if (mddev->bitmap || mddev->bitmap_info.file) | |
36fa3063 | 6149 | return -EEXIST; /* cannot add when bitmap is present */ |
1e594bb2 | 6150 | f = fget(fd); |
32a7627c | 6151 | |
1e594bb2 | 6152 | if (f == NULL) { |
36fa3063 N |
6153 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
6154 | mdname(mddev)); | |
6155 | return -EBADF; | |
6156 | } | |
6157 | ||
1e594bb2 | 6158 | inode = f->f_mapping->host; |
035328c2 N |
6159 | if (!S_ISREG(inode->i_mode)) { |
6160 | printk(KERN_ERR "%s: error: bitmap file must be a regular file\n", | |
6161 | mdname(mddev)); | |
6162 | err = -EBADF; | |
1e594bb2 | 6163 | } else if (!(f->f_mode & FMODE_WRITE)) { |
035328c2 N |
6164 | printk(KERN_ERR "%s: error: bitmap file must open for write\n", |
6165 | mdname(mddev)); | |
6166 | err = -EBADF; | |
6167 | } else if (atomic_read(&inode->i_writecount) != 1) { | |
36fa3063 N |
6168 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", |
6169 | mdname(mddev)); | |
035328c2 N |
6170 | err = -EBUSY; |
6171 | } | |
6172 | if (err) { | |
1e594bb2 | 6173 | fput(f); |
36fa3063 N |
6174 | return err; |
6175 | } | |
1e594bb2 | 6176 | mddev->bitmap_info.file = f; |
c3d9714e | 6177 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
6178 | } else if (mddev->bitmap == NULL) |
6179 | return -ENOENT; /* cannot remove what isn't there */ | |
6180 | err = 0; | |
6181 | if (mddev->pers) { | |
6182 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 6183 | if (fd >= 0) { |
f9209a32 GR |
6184 | struct bitmap *bitmap; |
6185 | ||
6186 | bitmap = bitmap_create(mddev, -1); | |
6187 | if (!IS_ERR(bitmap)) { | |
6188 | mddev->bitmap = bitmap; | |
69e51b44 | 6189 | err = bitmap_load(mddev); |
ba599aca N |
6190 | } else |
6191 | err = PTR_ERR(bitmap); | |
69e51b44 | 6192 | } |
d7375ab3 | 6193 | if (fd < 0 || err) { |
36fa3063 | 6194 | bitmap_destroy(mddev); |
d7375ab3 N |
6195 | fd = -1; /* make sure to put the file */ |
6196 | } | |
36fa3063 | 6197 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
6198 | } |
6199 | if (fd < 0) { | |
4af1a041 N |
6200 | struct file *f = mddev->bitmap_info.file; |
6201 | if (f) { | |
6202 | spin_lock(&mddev->lock); | |
6203 | mddev->bitmap_info.file = NULL; | |
6204 | spin_unlock(&mddev->lock); | |
6205 | fput(f); | |
6206 | } | |
36fa3063 N |
6207 | } |
6208 | ||
32a7627c N |
6209 | return err; |
6210 | } | |
6211 | ||
1da177e4 LT |
6212 | /* |
6213 | * set_array_info is used two different ways | |
6214 | * The original usage is when creating a new array. | |
6215 | * In this usage, raid_disks is > 0 and it together with | |
6216 | * level, size, not_persistent,layout,chunksize determine the | |
6217 | * shape of the array. | |
6218 | * This will always create an array with a type-0.90.0 superblock. | |
6219 | * The newer usage is when assembling an array. | |
6220 | * In this case raid_disks will be 0, and the major_version field is | |
6221 | * use to determine which style super-blocks are to be found on the devices. | |
6222 | * The minor and patch _version numbers are also kept incase the | |
6223 | * super_block handler wishes to interpret them. | |
6224 | */ | |
f72ffdd6 | 6225 | static int set_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6226 | { |
6227 | ||
6228 | if (info->raid_disks == 0) { | |
6229 | /* just setting version number for superblock loading */ | |
6230 | if (info->major_version < 0 || | |
50511da3 | 6231 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6232 | super_types[info->major_version].name == NULL) { |
6233 | /* maybe try to auto-load a module? */ | |
f72ffdd6 | 6234 | printk(KERN_INFO |
1da177e4 LT |
6235 | "md: superblock version %d not known\n", |
6236 | info->major_version); | |
6237 | return -EINVAL; | |
6238 | } | |
6239 | mddev->major_version = info->major_version; | |
6240 | mddev->minor_version = info->minor_version; | |
6241 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6242 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6243 | /* ensure mddev_put doesn't delete this now that there |
6244 | * is some minimal configuration. | |
6245 | */ | |
6246 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
6247 | return 0; |
6248 | } | |
6249 | mddev->major_version = MD_MAJOR_VERSION; | |
6250 | mddev->minor_version = MD_MINOR_VERSION; | |
6251 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
6252 | mddev->ctime = get_seconds(); | |
6253 | ||
6254 | mddev->level = info->level; | |
17115e03 | 6255 | mddev->clevel[0] = 0; |
58c0fed4 | 6256 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6257 | mddev->raid_disks = info->raid_disks; |
6258 | /* don't set md_minor, it is determined by which /dev/md* was | |
6259 | * openned | |
6260 | */ | |
6261 | if (info->state & (1<<MD_SB_CLEAN)) | |
6262 | mddev->recovery_cp = MaxSector; | |
6263 | else | |
6264 | mddev->recovery_cp = 0; | |
6265 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6266 | mddev->external = 0; |
1da177e4 LT |
6267 | |
6268 | mddev->layout = info->layout; | |
9d8f0363 | 6269 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
6270 | |
6271 | mddev->max_disks = MD_SB_DISKS; | |
6272 | ||
e691063a N |
6273 | if (mddev->persistent) |
6274 | mddev->flags = 0; | |
850b2b42 | 6275 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6276 | |
c3d9714e | 6277 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6278 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6279 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6280 | |
f6705578 N |
6281 | mddev->reshape_position = MaxSector; |
6282 | ||
1da177e4 LT |
6283 | /* |
6284 | * Generate a 128 bit UUID | |
6285 | */ | |
6286 | get_random_bytes(mddev->uuid, 16); | |
6287 | ||
f6705578 | 6288 | mddev->new_level = mddev->level; |
664e7c41 | 6289 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6290 | mddev->new_layout = mddev->layout; |
6291 | mddev->delta_disks = 0; | |
2c810cdd | 6292 | mddev->reshape_backwards = 0; |
f6705578 | 6293 | |
1da177e4 LT |
6294 | return 0; |
6295 | } | |
6296 | ||
fd01b88c | 6297 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6298 | { |
b522adcd DW |
6299 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
6300 | ||
6301 | if (mddev->external_size) | |
6302 | return; | |
6303 | ||
1f403624 DW |
6304 | mddev->array_sectors = array_sectors; |
6305 | } | |
6306 | EXPORT_SYMBOL(md_set_array_sectors); | |
6307 | ||
fd01b88c | 6308 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6309 | { |
3cb03002 | 6310 | struct md_rdev *rdev; |
a35b0d69 | 6311 | int rv; |
d71f9f88 | 6312 | int fit = (num_sectors == 0); |
a35b0d69 N |
6313 | |
6314 | if (mddev->pers->resize == NULL) | |
6315 | return -EINVAL; | |
d71f9f88 AN |
6316 | /* The "num_sectors" is the number of sectors of each device that |
6317 | * is used. This can only make sense for arrays with redundancy. | |
6318 | * linear and raid0 always use whatever space is available. We can only | |
6319 | * consider changing this number if no resync or reconstruction is | |
6320 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6321 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6322 | * of each device. If num_sectors is zero, we find the largest size |
6323 | * that fits. | |
a35b0d69 | 6324 | */ |
f851b60d N |
6325 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
6326 | mddev->sync_thread) | |
a35b0d69 | 6327 | return -EBUSY; |
bd8839e0 N |
6328 | if (mddev->ro) |
6329 | return -EROFS; | |
a4a6125a | 6330 | |
dafb20fa | 6331 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6332 | sector_t avail = rdev->sectors; |
01ab5662 | 6333 | |
d71f9f88 AN |
6334 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6335 | num_sectors = avail; | |
6336 | if (avail < num_sectors) | |
a35b0d69 N |
6337 | return -ENOSPC; |
6338 | } | |
d71f9f88 | 6339 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
6340 | if (!rv) |
6341 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
6342 | return rv; |
6343 | } | |
6344 | ||
fd01b88c | 6345 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6346 | { |
6347 | int rv; | |
c6563a8c | 6348 | struct md_rdev *rdev; |
da943b99 | 6349 | /* change the number of raid disks */ |
63c70c4f | 6350 | if (mddev->pers->check_reshape == NULL) |
da943b99 | 6351 | return -EINVAL; |
bd8839e0 N |
6352 | if (mddev->ro) |
6353 | return -EROFS; | |
da943b99 | 6354 | if (raid_disks <= 0 || |
233fca36 | 6355 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6356 | return -EINVAL; |
f851b60d N |
6357 | if (mddev->sync_thread || |
6358 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
6359 | mddev->reshape_position != MaxSector) | |
da943b99 | 6360 | return -EBUSY; |
c6563a8c N |
6361 | |
6362 | rdev_for_each(rdev, mddev) { | |
6363 | if (mddev->raid_disks < raid_disks && | |
6364 | rdev->data_offset < rdev->new_data_offset) | |
6365 | return -EINVAL; | |
6366 | if (mddev->raid_disks > raid_disks && | |
6367 | rdev->data_offset > rdev->new_data_offset) | |
6368 | return -EINVAL; | |
6369 | } | |
6370 | ||
63c70c4f | 6371 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6372 | if (mddev->delta_disks < 0) |
6373 | mddev->reshape_backwards = 1; | |
6374 | else if (mddev->delta_disks > 0) | |
6375 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6376 | |
6377 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6378 | if (rv < 0) { |
de171cb9 | 6379 | mddev->delta_disks = 0; |
2c810cdd N |
6380 | mddev->reshape_backwards = 0; |
6381 | } | |
da943b99 N |
6382 | return rv; |
6383 | } | |
6384 | ||
1da177e4 LT |
6385 | /* |
6386 | * update_array_info is used to change the configuration of an | |
6387 | * on-line array. | |
6388 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6389 | * fields in the info are checked against the array. | |
6390 | * Any differences that cannot be handled will cause an error. | |
6391 | * Normally, only one change can be managed at a time. | |
6392 | */ | |
fd01b88c | 6393 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6394 | { |
6395 | int rv = 0; | |
6396 | int cnt = 0; | |
36fa3063 N |
6397 | int state = 0; |
6398 | ||
6399 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6400 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6401 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6402 | |
6403 | if (mddev->major_version != info->major_version || | |
6404 | mddev->minor_version != info->minor_version || | |
6405 | /* mddev->patch_version != info->patch_version || */ | |
6406 | mddev->ctime != info->ctime || | |
6407 | mddev->level != info->level || | |
6408 | /* mddev->layout != info->layout || */ | |
4e023612 | 6409 | mddev->persistent != !info->not_persistent || |
9d8f0363 | 6410 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6411 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6412 | ((state^info->state) & 0xfffffe00) | |
6413 | ) | |
1da177e4 LT |
6414 | return -EINVAL; |
6415 | /* Check there is only one change */ | |
58c0fed4 AN |
6416 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6417 | cnt++; | |
6418 | if (mddev->raid_disks != info->raid_disks) | |
6419 | cnt++; | |
6420 | if (mddev->layout != info->layout) | |
6421 | cnt++; | |
6422 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6423 | cnt++; | |
6424 | if (cnt == 0) | |
6425 | return 0; | |
6426 | if (cnt > 1) | |
6427 | return -EINVAL; | |
1da177e4 LT |
6428 | |
6429 | if (mddev->layout != info->layout) { | |
6430 | /* Change layout | |
6431 | * we don't need to do anything at the md level, the | |
6432 | * personality will take care of it all. | |
6433 | */ | |
50ac168a | 6434 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6435 | return -EINVAL; |
597a711b N |
6436 | else { |
6437 | mddev->new_layout = info->layout; | |
50ac168a | 6438 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6439 | if (rv) |
6440 | mddev->new_layout = mddev->layout; | |
6441 | return rv; | |
6442 | } | |
1da177e4 | 6443 | } |
293467aa GR |
6444 | if (mddev_is_clustered(mddev)) |
6445 | md_cluster_ops->metadata_update_start(mddev); | |
58c0fed4 | 6446 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6447 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6448 | |
da943b99 N |
6449 | if (mddev->raid_disks != info->raid_disks) |
6450 | rv = update_raid_disks(mddev, info->raid_disks); | |
6451 | ||
36fa3063 | 6452 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
293467aa GR |
6453 | if (mddev->pers->quiesce == NULL || mddev->thread == NULL) { |
6454 | rv = -EINVAL; | |
6455 | goto err; | |
6456 | } | |
6457 | if (mddev->recovery || mddev->sync_thread) { | |
6458 | rv = -EBUSY; | |
6459 | goto err; | |
6460 | } | |
36fa3063 | 6461 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { |
f9209a32 | 6462 | struct bitmap *bitmap; |
36fa3063 | 6463 | /* add the bitmap */ |
293467aa GR |
6464 | if (mddev->bitmap) { |
6465 | rv = -EEXIST; | |
6466 | goto err; | |
6467 | } | |
6468 | if (mddev->bitmap_info.default_offset == 0) { | |
6469 | rv = -EINVAL; | |
6470 | goto err; | |
6471 | } | |
c3d9714e N |
6472 | mddev->bitmap_info.offset = |
6473 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6474 | mddev->bitmap_info.space = |
6475 | mddev->bitmap_info.default_space; | |
36fa3063 | 6476 | mddev->pers->quiesce(mddev, 1); |
f9209a32 GR |
6477 | bitmap = bitmap_create(mddev, -1); |
6478 | if (!IS_ERR(bitmap)) { | |
6479 | mddev->bitmap = bitmap; | |
69e51b44 | 6480 | rv = bitmap_load(mddev); |
ba599aca N |
6481 | } else |
6482 | rv = PTR_ERR(bitmap); | |
36fa3063 N |
6483 | if (rv) |
6484 | bitmap_destroy(mddev); | |
6485 | mddev->pers->quiesce(mddev, 0); | |
6486 | } else { | |
6487 | /* remove the bitmap */ | |
293467aa GR |
6488 | if (!mddev->bitmap) { |
6489 | rv = -ENOENT; | |
6490 | goto err; | |
6491 | } | |
6492 | if (mddev->bitmap->storage.file) { | |
6493 | rv = -EINVAL; | |
6494 | goto err; | |
6495 | } | |
36fa3063 N |
6496 | mddev->pers->quiesce(mddev, 1); |
6497 | bitmap_destroy(mddev); | |
6498 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 6499 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6500 | } |
6501 | } | |
850b2b42 | 6502 | md_update_sb(mddev, 1); |
293467aa GR |
6503 | if (mddev_is_clustered(mddev)) |
6504 | md_cluster_ops->metadata_update_finish(mddev); | |
6505 | return rv; | |
6506 | err: | |
6507 | if (mddev_is_clustered(mddev)) | |
6508 | md_cluster_ops->metadata_update_cancel(mddev); | |
1da177e4 LT |
6509 | return rv; |
6510 | } | |
6511 | ||
fd01b88c | 6512 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6513 | { |
3cb03002 | 6514 | struct md_rdev *rdev; |
1ca69c4b | 6515 | int err = 0; |
1da177e4 LT |
6516 | |
6517 | if (mddev->pers == NULL) | |
6518 | return -ENODEV; | |
6519 | ||
1ca69c4b N |
6520 | rcu_read_lock(); |
6521 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 6522 | if (!rdev) |
1ca69c4b N |
6523 | err = -ENODEV; |
6524 | else { | |
6525 | md_error(mddev, rdev); | |
6526 | if (!test_bit(Faulty, &rdev->flags)) | |
6527 | err = -EBUSY; | |
6528 | } | |
6529 | rcu_read_unlock(); | |
6530 | return err; | |
1da177e4 LT |
6531 | } |
6532 | ||
2f9618ce AN |
6533 | /* |
6534 | * We have a problem here : there is no easy way to give a CHS | |
6535 | * virtual geometry. We currently pretend that we have a 2 heads | |
6536 | * 4 sectors (with a BIG number of cylinders...). This drives | |
6537 | * dosfs just mad... ;-) | |
6538 | */ | |
a885c8c4 CH |
6539 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
6540 | { | |
fd01b88c | 6541 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
6542 | |
6543 | geo->heads = 2; | |
6544 | geo->sectors = 4; | |
49ce6cea | 6545 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
6546 | return 0; |
6547 | } | |
6548 | ||
cb335f88 NS |
6549 | static inline bool md_ioctl_valid(unsigned int cmd) |
6550 | { | |
6551 | switch (cmd) { | |
6552 | case ADD_NEW_DISK: | |
6553 | case BLKROSET: | |
6554 | case GET_ARRAY_INFO: | |
6555 | case GET_BITMAP_FILE: | |
6556 | case GET_DISK_INFO: | |
6557 | case HOT_ADD_DISK: | |
6558 | case HOT_REMOVE_DISK: | |
cb335f88 NS |
6559 | case RAID_AUTORUN: |
6560 | case RAID_VERSION: | |
6561 | case RESTART_ARRAY_RW: | |
6562 | case RUN_ARRAY: | |
6563 | case SET_ARRAY_INFO: | |
6564 | case SET_BITMAP_FILE: | |
6565 | case SET_DISK_FAULTY: | |
6566 | case STOP_ARRAY: | |
6567 | case STOP_ARRAY_RO: | |
1aee41f6 | 6568 | case CLUSTERED_DISK_NACK: |
cb335f88 NS |
6569 | return true; |
6570 | default: | |
6571 | return false; | |
6572 | } | |
6573 | } | |
6574 | ||
a39907fa | 6575 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
6576 | unsigned int cmd, unsigned long arg) |
6577 | { | |
6578 | int err = 0; | |
6579 | void __user *argp = (void __user *)arg; | |
fd01b88c | 6580 | struct mddev *mddev = NULL; |
e2218350 | 6581 | int ro; |
1da177e4 | 6582 | |
cb335f88 NS |
6583 | if (!md_ioctl_valid(cmd)) |
6584 | return -ENOTTY; | |
6585 | ||
506c9e44 N |
6586 | switch (cmd) { |
6587 | case RAID_VERSION: | |
6588 | case GET_ARRAY_INFO: | |
6589 | case GET_DISK_INFO: | |
6590 | break; | |
6591 | default: | |
6592 | if (!capable(CAP_SYS_ADMIN)) | |
6593 | return -EACCES; | |
6594 | } | |
1da177e4 LT |
6595 | |
6596 | /* | |
6597 | * Commands dealing with the RAID driver but not any | |
6598 | * particular array: | |
6599 | */ | |
c02c0aeb N |
6600 | switch (cmd) { |
6601 | case RAID_VERSION: | |
6602 | err = get_version(argp); | |
3adc28d8 | 6603 | goto out; |
1da177e4 | 6604 | |
1da177e4 | 6605 | #ifndef MODULE |
c02c0aeb N |
6606 | case RAID_AUTORUN: |
6607 | err = 0; | |
6608 | autostart_arrays(arg); | |
3adc28d8 | 6609 | goto out; |
1da177e4 | 6610 | #endif |
c02c0aeb | 6611 | default:; |
1da177e4 LT |
6612 | } |
6613 | ||
6614 | /* | |
6615 | * Commands creating/starting a new array: | |
6616 | */ | |
6617 | ||
a39907fa | 6618 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
6619 | |
6620 | if (!mddev) { | |
6621 | BUG(); | |
3adc28d8 | 6622 | goto out; |
1da177e4 LT |
6623 | } |
6624 | ||
1ca69c4b N |
6625 | /* Some actions do not requires the mutex */ |
6626 | switch (cmd) { | |
6627 | case GET_ARRAY_INFO: | |
6628 | if (!mddev->raid_disks && !mddev->external) | |
6629 | err = -ENODEV; | |
6630 | else | |
6631 | err = get_array_info(mddev, argp); | |
3adc28d8 | 6632 | goto out; |
1ca69c4b N |
6633 | |
6634 | case GET_DISK_INFO: | |
6635 | if (!mddev->raid_disks && !mddev->external) | |
6636 | err = -ENODEV; | |
6637 | else | |
6638 | err = get_disk_info(mddev, argp); | |
3adc28d8 | 6639 | goto out; |
1ca69c4b N |
6640 | |
6641 | case SET_DISK_FAULTY: | |
6642 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6643 | goto out; |
4af1a041 N |
6644 | |
6645 | case GET_BITMAP_FILE: | |
6646 | err = get_bitmap_file(mddev, argp); | |
6647 | goto out; | |
6648 | ||
1ca69c4b N |
6649 | } |
6650 | ||
a7a3f08d N |
6651 | if (cmd == ADD_NEW_DISK) |
6652 | /* need to ensure md_delayed_delete() has completed */ | |
6653 | flush_workqueue(md_misc_wq); | |
6654 | ||
90f5f7ad HR |
6655 | if (cmd == HOT_REMOVE_DISK) |
6656 | /* need to ensure recovery thread has run */ | |
6657 | wait_event_interruptible_timeout(mddev->sb_wait, | |
6658 | !test_bit(MD_RECOVERY_NEEDED, | |
6659 | &mddev->flags), | |
6660 | msecs_to_jiffies(5000)); | |
260fa034 N |
6661 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
6662 | /* Need to flush page cache, and ensure no-one else opens | |
6663 | * and writes | |
6664 | */ | |
6665 | mutex_lock(&mddev->open_mutex); | |
9ba3b7f5 | 6666 | if (mddev->pers && atomic_read(&mddev->openers) > 1) { |
260fa034 N |
6667 | mutex_unlock(&mddev->open_mutex); |
6668 | err = -EBUSY; | |
3adc28d8 | 6669 | goto out; |
260fa034 N |
6670 | } |
6671 | set_bit(MD_STILL_CLOSED, &mddev->flags); | |
6672 | mutex_unlock(&mddev->open_mutex); | |
6673 | sync_blockdev(bdev); | |
6674 | } | |
1da177e4 LT |
6675 | err = mddev_lock(mddev); |
6676 | if (err) { | |
f72ffdd6 | 6677 | printk(KERN_INFO |
1da177e4 LT |
6678 | "md: ioctl lock interrupted, reason %d, cmd %d\n", |
6679 | err, cmd); | |
3adc28d8 | 6680 | goto out; |
1da177e4 LT |
6681 | } |
6682 | ||
c02c0aeb N |
6683 | if (cmd == SET_ARRAY_INFO) { |
6684 | mdu_array_info_t info; | |
6685 | if (!arg) | |
6686 | memset(&info, 0, sizeof(info)); | |
6687 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
6688 | err = -EFAULT; | |
3adc28d8 | 6689 | goto unlock; |
c02c0aeb N |
6690 | } |
6691 | if (mddev->pers) { | |
6692 | err = update_array_info(mddev, &info); | |
6693 | if (err) { | |
6694 | printk(KERN_WARNING "md: couldn't update" | |
6695 | " array info. %d\n", err); | |
3adc28d8 | 6696 | goto unlock; |
1da177e4 | 6697 | } |
3adc28d8 | 6698 | goto unlock; |
c02c0aeb N |
6699 | } |
6700 | if (!list_empty(&mddev->disks)) { | |
6701 | printk(KERN_WARNING | |
6702 | "md: array %s already has disks!\n", | |
6703 | mdname(mddev)); | |
6704 | err = -EBUSY; | |
3adc28d8 | 6705 | goto unlock; |
c02c0aeb N |
6706 | } |
6707 | if (mddev->raid_disks) { | |
6708 | printk(KERN_WARNING | |
6709 | "md: array %s already initialised!\n", | |
6710 | mdname(mddev)); | |
6711 | err = -EBUSY; | |
3adc28d8 | 6712 | goto unlock; |
c02c0aeb N |
6713 | } |
6714 | err = set_array_info(mddev, &info); | |
6715 | if (err) { | |
6716 | printk(KERN_WARNING "md: couldn't set" | |
6717 | " array info. %d\n", err); | |
3adc28d8 | 6718 | goto unlock; |
c02c0aeb | 6719 | } |
3adc28d8 | 6720 | goto unlock; |
1da177e4 LT |
6721 | } |
6722 | ||
6723 | /* | |
6724 | * Commands querying/configuring an existing array: | |
6725 | */ | |
32a7627c | 6726 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 6727 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
6728 | if ((!mddev->raid_disks && !mddev->external) |
6729 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
6730 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
6731 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 | 6732 | err = -ENODEV; |
3adc28d8 | 6733 | goto unlock; |
1da177e4 LT |
6734 | } |
6735 | ||
6736 | /* | |
6737 | * Commands even a read-only array can execute: | |
6738 | */ | |
c02c0aeb | 6739 | switch (cmd) { |
c02c0aeb N |
6740 | case RESTART_ARRAY_RW: |
6741 | err = restart_array(mddev); | |
3adc28d8 | 6742 | goto unlock; |
1da177e4 | 6743 | |
c02c0aeb N |
6744 | case STOP_ARRAY: |
6745 | err = do_md_stop(mddev, 0, bdev); | |
3adc28d8 | 6746 | goto unlock; |
1da177e4 | 6747 | |
c02c0aeb N |
6748 | case STOP_ARRAY_RO: |
6749 | err = md_set_readonly(mddev, bdev); | |
3adc28d8 | 6750 | goto unlock; |
1da177e4 | 6751 | |
3ea8929d N |
6752 | case HOT_REMOVE_DISK: |
6753 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6754 | goto unlock; |
3ea8929d | 6755 | |
7ceb17e8 N |
6756 | case ADD_NEW_DISK: |
6757 | /* We can support ADD_NEW_DISK on read-only arrays | |
6758 | * on if we are re-adding a preexisting device. | |
6759 | * So require mddev->pers and MD_DISK_SYNC. | |
6760 | */ | |
6761 | if (mddev->pers) { | |
6762 | mdu_disk_info_t info; | |
6763 | if (copy_from_user(&info, argp, sizeof(info))) | |
6764 | err = -EFAULT; | |
6765 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
6766 | /* Need to clear read-only for this */ | |
6767 | break; | |
6768 | else | |
6769 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 6770 | goto unlock; |
7ceb17e8 N |
6771 | } |
6772 | break; | |
6773 | ||
c02c0aeb N |
6774 | case BLKROSET: |
6775 | if (get_user(ro, (int __user *)(arg))) { | |
6776 | err = -EFAULT; | |
3adc28d8 | 6777 | goto unlock; |
c02c0aeb N |
6778 | } |
6779 | err = -EINVAL; | |
e2218350 | 6780 | |
c02c0aeb N |
6781 | /* if the bdev is going readonly the value of mddev->ro |
6782 | * does not matter, no writes are coming | |
6783 | */ | |
6784 | if (ro) | |
3adc28d8 | 6785 | goto unlock; |
e2218350 | 6786 | |
c02c0aeb N |
6787 | /* are we are already prepared for writes? */ |
6788 | if (mddev->ro != 1) | |
3adc28d8 | 6789 | goto unlock; |
e2218350 | 6790 | |
c02c0aeb N |
6791 | /* transitioning to readauto need only happen for |
6792 | * arrays that call md_write_start | |
6793 | */ | |
6794 | if (mddev->pers) { | |
6795 | err = restart_array(mddev); | |
6796 | if (err == 0) { | |
6797 | mddev->ro = 2; | |
6798 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 6799 | } |
c02c0aeb | 6800 | } |
3adc28d8 | 6801 | goto unlock; |
1da177e4 LT |
6802 | } |
6803 | ||
6804 | /* | |
6805 | * The remaining ioctls are changing the state of the | |
f91de92e | 6806 | * superblock, so we do not allow them on read-only arrays. |
1da177e4 | 6807 | */ |
326eb17d | 6808 | if (mddev->ro && mddev->pers) { |
f91de92e N |
6809 | if (mddev->ro == 2) { |
6810 | mddev->ro = 0; | |
00bcb4ac | 6811 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 6812 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
6813 | /* mddev_unlock will wake thread */ |
6814 | /* If a device failed while we were read-only, we | |
6815 | * need to make sure the metadata is updated now. | |
6816 | */ | |
6817 | if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
6818 | mddev_unlock(mddev); | |
6819 | wait_event(mddev->sb_wait, | |
6820 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) && | |
6821 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
29f097c4 | 6822 | mddev_lock_nointr(mddev); |
f3378b48 | 6823 | } |
f91de92e N |
6824 | } else { |
6825 | err = -EROFS; | |
3adc28d8 | 6826 | goto unlock; |
f91de92e | 6827 | } |
1da177e4 LT |
6828 | } |
6829 | ||
c02c0aeb N |
6830 | switch (cmd) { |
6831 | case ADD_NEW_DISK: | |
1da177e4 | 6832 | { |
c02c0aeb N |
6833 | mdu_disk_info_t info; |
6834 | if (copy_from_user(&info, argp, sizeof(info))) | |
6835 | err = -EFAULT; | |
6836 | else | |
6837 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 6838 | goto unlock; |
c02c0aeb | 6839 | } |
1da177e4 | 6840 | |
1aee41f6 GR |
6841 | case CLUSTERED_DISK_NACK: |
6842 | if (mddev_is_clustered(mddev)) | |
6843 | md_cluster_ops->new_disk_ack(mddev, false); | |
6844 | else | |
6845 | err = -EINVAL; | |
6846 | goto unlock; | |
6847 | ||
c02c0aeb N |
6848 | case HOT_ADD_DISK: |
6849 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6850 | goto unlock; |
1da177e4 | 6851 | |
c02c0aeb N |
6852 | case RUN_ARRAY: |
6853 | err = do_md_run(mddev); | |
3adc28d8 | 6854 | goto unlock; |
1da177e4 | 6855 | |
c02c0aeb N |
6856 | case SET_BITMAP_FILE: |
6857 | err = set_bitmap_file(mddev, (int)arg); | |
3adc28d8 | 6858 | goto unlock; |
32a7627c | 6859 | |
c02c0aeb N |
6860 | default: |
6861 | err = -EINVAL; | |
3adc28d8 | 6862 | goto unlock; |
1da177e4 LT |
6863 | } |
6864 | ||
3adc28d8 | 6865 | unlock: |
d3374825 N |
6866 | if (mddev->hold_active == UNTIL_IOCTL && |
6867 | err != -EINVAL) | |
6868 | mddev->hold_active = 0; | |
1da177e4 | 6869 | mddev_unlock(mddev); |
3adc28d8 | 6870 | out: |
1da177e4 LT |
6871 | return err; |
6872 | } | |
aa98aa31 AB |
6873 | #ifdef CONFIG_COMPAT |
6874 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
6875 | unsigned int cmd, unsigned long arg) | |
6876 | { | |
6877 | switch (cmd) { | |
6878 | case HOT_REMOVE_DISK: | |
6879 | case HOT_ADD_DISK: | |
6880 | case SET_DISK_FAULTY: | |
6881 | case SET_BITMAP_FILE: | |
6882 | /* These take in integer arg, do not convert */ | |
6883 | break; | |
6884 | default: | |
6885 | arg = (unsigned long)compat_ptr(arg); | |
6886 | break; | |
6887 | } | |
6888 | ||
6889 | return md_ioctl(bdev, mode, cmd, arg); | |
6890 | } | |
6891 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 6892 | |
a39907fa | 6893 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
6894 | { |
6895 | /* | |
6896 | * Succeed if we can lock the mddev, which confirms that | |
6897 | * it isn't being stopped right now. | |
6898 | */ | |
fd01b88c | 6899 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
6900 | int err; |
6901 | ||
0c098220 YL |
6902 | if (!mddev) |
6903 | return -ENODEV; | |
6904 | ||
d3374825 N |
6905 | if (mddev->gendisk != bdev->bd_disk) { |
6906 | /* we are racing with mddev_put which is discarding this | |
6907 | * bd_disk. | |
6908 | */ | |
6909 | mddev_put(mddev); | |
6910 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 6911 | flush_workqueue(md_misc_wq); |
d3374825 N |
6912 | /* Then retry the open from the top */ |
6913 | return -ERESTARTSYS; | |
6914 | } | |
6915 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
6916 | ||
c8c00a69 | 6917 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6918 | goto out; |
6919 | ||
6920 | err = 0; | |
f2ea68cf | 6921 | atomic_inc(&mddev->openers); |
260fa034 | 6922 | clear_bit(MD_STILL_CLOSED, &mddev->flags); |
c8c00a69 | 6923 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6924 | |
f0b4f7e2 | 6925 | check_disk_change(bdev); |
1da177e4 LT |
6926 | out: |
6927 | return err; | |
6928 | } | |
6929 | ||
db2a144b | 6930 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6931 | { |
f72ffdd6 | 6932 | struct mddev *mddev = disk->private_data; |
1da177e4 | 6933 | |
52e5f9d1 | 6934 | BUG_ON(!mddev); |
f2ea68cf | 6935 | atomic_dec(&mddev->openers); |
1da177e4 | 6936 | mddev_put(mddev); |
1da177e4 | 6937 | } |
f0b4f7e2 N |
6938 | |
6939 | static int md_media_changed(struct gendisk *disk) | |
6940 | { | |
fd01b88c | 6941 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6942 | |
6943 | return mddev->changed; | |
6944 | } | |
6945 | ||
6946 | static int md_revalidate(struct gendisk *disk) | |
6947 | { | |
fd01b88c | 6948 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6949 | |
6950 | mddev->changed = 0; | |
6951 | return 0; | |
6952 | } | |
83d5cde4 | 6953 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6954 | { |
6955 | .owner = THIS_MODULE, | |
a39907fa AV |
6956 | .open = md_open, |
6957 | .release = md_release, | |
b492b852 | 6958 | .ioctl = md_ioctl, |
aa98aa31 AB |
6959 | #ifdef CONFIG_COMPAT |
6960 | .compat_ioctl = md_compat_ioctl, | |
6961 | #endif | |
a885c8c4 | 6962 | .getgeo = md_getgeo, |
f0b4f7e2 N |
6963 | .media_changed = md_media_changed, |
6964 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
6965 | }; |
6966 | ||
f72ffdd6 | 6967 | static int md_thread(void *arg) |
1da177e4 | 6968 | { |
2b8bf345 | 6969 | struct md_thread *thread = arg; |
1da177e4 | 6970 | |
1da177e4 LT |
6971 | /* |
6972 | * md_thread is a 'system-thread', it's priority should be very | |
6973 | * high. We avoid resource deadlocks individually in each | |
6974 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6975 | * the very same RT priority as kswapd, thus we will never get | |
6976 | * into a priority inversion deadlock. | |
6977 | * | |
6978 | * we definitely have to have equal or higher priority than | |
6979 | * bdflush, otherwise bdflush will deadlock if there are too | |
6980 | * many dirty RAID5 blocks. | |
6981 | */ | |
1da177e4 | 6982 | |
6985c43f | 6983 | allow_signal(SIGKILL); |
a6fb0934 | 6984 | while (!kthread_should_stop()) { |
1da177e4 | 6985 | |
93588e22 N |
6986 | /* We need to wait INTERRUPTIBLE so that |
6987 | * we don't add to the load-average. | |
6988 | * That means we need to be sure no signals are | |
6989 | * pending | |
6990 | */ | |
6991 | if (signal_pending(current)) | |
6992 | flush_signals(current); | |
6993 | ||
6994 | wait_event_interruptible_timeout | |
6995 | (thread->wqueue, | |
6996 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6997 | || kthread_should_stop(), | |
6998 | thread->timeout); | |
1da177e4 | 6999 | |
6c987910 N |
7000 | clear_bit(THREAD_WAKEUP, &thread->flags); |
7001 | if (!kthread_should_stop()) | |
4ed8731d | 7002 | thread->run(thread); |
1da177e4 | 7003 | } |
a6fb0934 | 7004 | |
1da177e4 LT |
7005 | return 0; |
7006 | } | |
7007 | ||
2b8bf345 | 7008 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
7009 | { |
7010 | if (thread) { | |
36a4e1fe | 7011 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
1da177e4 LT |
7012 | set_bit(THREAD_WAKEUP, &thread->flags); |
7013 | wake_up(&thread->wqueue); | |
7014 | } | |
7015 | } | |
6c144d31 | 7016 | EXPORT_SYMBOL(md_wakeup_thread); |
1da177e4 | 7017 | |
4ed8731d SL |
7018 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
7019 | struct mddev *mddev, const char *name) | |
1da177e4 | 7020 | { |
2b8bf345 | 7021 | struct md_thread *thread; |
1da177e4 | 7022 | |
2b8bf345 | 7023 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
7024 | if (!thread) |
7025 | return NULL; | |
7026 | ||
1da177e4 LT |
7027 | init_waitqueue_head(&thread->wqueue); |
7028 | ||
1da177e4 LT |
7029 | thread->run = run; |
7030 | thread->mddev = mddev; | |
32a7627c | 7031 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
7032 | thread->tsk = kthread_run(md_thread, thread, |
7033 | "%s_%s", | |
7034 | mdname(thread->mddev), | |
0232605d | 7035 | name); |
a6fb0934 | 7036 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
7037 | kfree(thread); |
7038 | return NULL; | |
7039 | } | |
1da177e4 LT |
7040 | return thread; |
7041 | } | |
6c144d31 | 7042 | EXPORT_SYMBOL(md_register_thread); |
1da177e4 | 7043 | |
2b8bf345 | 7044 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 7045 | { |
2b8bf345 | 7046 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
7047 | if (!thread) |
7048 | return; | |
36a4e1fe | 7049 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
7050 | /* Locking ensures that mddev_unlock does not wake_up a |
7051 | * non-existent thread | |
7052 | */ | |
7053 | spin_lock(&pers_lock); | |
7054 | *threadp = NULL; | |
7055 | spin_unlock(&pers_lock); | |
a6fb0934 N |
7056 | |
7057 | kthread_stop(thread->tsk); | |
1da177e4 LT |
7058 | kfree(thread); |
7059 | } | |
6c144d31 | 7060 | EXPORT_SYMBOL(md_unregister_thread); |
1da177e4 | 7061 | |
fd01b88c | 7062 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7063 | { |
b2d444d7 | 7064 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 7065 | return; |
6bfe0b49 | 7066 | |
de393cde | 7067 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
7068 | return; |
7069 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
7070 | if (mddev->degraded) |
7071 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 7072 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
7073 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7074 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7075 | md_wakeup_thread(mddev->thread); | |
768a418d | 7076 | if (mddev->event_work.func) |
e804ac78 | 7077 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 7078 | md_new_event_inintr(mddev); |
1da177e4 | 7079 | } |
6c144d31 | 7080 | EXPORT_SYMBOL(md_error); |
1da177e4 LT |
7081 | |
7082 | /* seq_file implementation /proc/mdstat */ | |
7083 | ||
7084 | static void status_unused(struct seq_file *seq) | |
7085 | { | |
7086 | int i = 0; | |
3cb03002 | 7087 | struct md_rdev *rdev; |
1da177e4 LT |
7088 | |
7089 | seq_printf(seq, "unused devices: "); | |
7090 | ||
159ec1fc | 7091 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
7092 | char b[BDEVNAME_SIZE]; |
7093 | i++; | |
7094 | seq_printf(seq, "%s ", | |
7095 | bdevname(rdev->bdev,b)); | |
7096 | } | |
7097 | if (!i) | |
7098 | seq_printf(seq, "<none>"); | |
7099 | ||
7100 | seq_printf(seq, "\n"); | |
7101 | } | |
7102 | ||
f7851be7 | 7103 | static int status_resync(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7104 | { |
dd71cf6b N |
7105 | sector_t max_sectors, resync, res; |
7106 | unsigned long dt, db; | |
7107 | sector_t rt; | |
4588b42e N |
7108 | int scale; |
7109 | unsigned int per_milli; | |
1da177e4 | 7110 | |
c804cdec N |
7111 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
7112 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 7113 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 7114 | else |
dd71cf6b | 7115 | max_sectors = mddev->dev_sectors; |
1da177e4 | 7116 | |
f7851be7 N |
7117 | resync = mddev->curr_resync; |
7118 | if (resync <= 3) { | |
7119 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7120 | /* Still cleaning up */ | |
7121 | resync = max_sectors; | |
7122 | } else | |
7123 | resync -= atomic_read(&mddev->recovery_active); | |
7124 | ||
7125 | if (resync == 0) { | |
7126 | if (mddev->recovery_cp < MaxSector) { | |
7127 | seq_printf(seq, "\tresync=PENDING"); | |
7128 | return 1; | |
7129 | } | |
7130 | return 0; | |
7131 | } | |
7132 | if (resync < 3) { | |
7133 | seq_printf(seq, "\tresync=DELAYED"); | |
7134 | return 1; | |
7135 | } | |
7136 | ||
403df478 | 7137 | WARN_ON(max_sectors == 0); |
4588b42e | 7138 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 7139 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
7140 | * u32, as those are the requirements for sector_div. |
7141 | * Thus 'scale' must be at least 10 | |
7142 | */ | |
7143 | scale = 10; | |
7144 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 7145 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
7146 | scale++; |
7147 | } | |
7148 | res = (resync>>scale)*1000; | |
dd71cf6b | 7149 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
7150 | |
7151 | per_milli = res; | |
1da177e4 | 7152 | { |
4588b42e | 7153 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
7154 | seq_printf(seq, "["); |
7155 | for (i = 0; i < x; i++) | |
7156 | seq_printf(seq, "="); | |
7157 | seq_printf(seq, ">"); | |
7158 | for (i = 0; i < y; i++) | |
7159 | seq_printf(seq, "."); | |
7160 | seq_printf(seq, "] "); | |
7161 | } | |
4588b42e | 7162 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
7163 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
7164 | "reshape" : | |
61df9d91 N |
7165 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
7166 | "check" : | |
7167 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
7168 | "resync" : "recovery"))), | |
7169 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
7170 | (unsigned long long) resync/2, |
7171 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
7172 | |
7173 | /* | |
1da177e4 LT |
7174 | * dt: time from mark until now |
7175 | * db: blocks written from mark until now | |
7176 | * rt: remaining time | |
dd71cf6b N |
7177 | * |
7178 | * rt is a sector_t, so could be 32bit or 64bit. | |
7179 | * So we divide before multiply in case it is 32bit and close | |
7180 | * to the limit. | |
25985edc | 7181 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
7182 | * near the end of resync when the number of remaining sectors |
7183 | * is close to 'db'. | |
7184 | * We then divide rt by 32 after multiplying by db to compensate. | |
7185 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
7186 | */ |
7187 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
7188 | if (!dt) dt++; | |
ff4e8d9a N |
7189 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
7190 | - mddev->resync_mark_cnt; | |
1da177e4 | 7191 | |
dd71cf6b N |
7192 | rt = max_sectors - resync; /* number of remaining sectors */ |
7193 | sector_div(rt, db/32+1); | |
7194 | rt *= dt; | |
7195 | rt >>= 5; | |
7196 | ||
7197 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
7198 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 7199 | |
ff4e8d9a | 7200 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
f7851be7 | 7201 | return 1; |
1da177e4 LT |
7202 | } |
7203 | ||
7204 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
7205 | { | |
7206 | struct list_head *tmp; | |
7207 | loff_t l = *pos; | |
fd01b88c | 7208 | struct mddev *mddev; |
1da177e4 LT |
7209 | |
7210 | if (l >= 0x10000) | |
7211 | return NULL; | |
7212 | if (!l--) | |
7213 | /* header */ | |
7214 | return (void*)1; | |
7215 | ||
7216 | spin_lock(&all_mddevs_lock); | |
7217 | list_for_each(tmp,&all_mddevs) | |
7218 | if (!l--) { | |
fd01b88c | 7219 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
7220 | mddev_get(mddev); |
7221 | spin_unlock(&all_mddevs_lock); | |
7222 | return mddev; | |
7223 | } | |
7224 | spin_unlock(&all_mddevs_lock); | |
7225 | if (!l--) | |
7226 | return (void*)2;/* tail */ | |
7227 | return NULL; | |
7228 | } | |
7229 | ||
7230 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
7231 | { | |
7232 | struct list_head *tmp; | |
fd01b88c | 7233 | struct mddev *next_mddev, *mddev = v; |
f72ffdd6 | 7234 | |
1da177e4 LT |
7235 | ++*pos; |
7236 | if (v == (void*)2) | |
7237 | return NULL; | |
7238 | ||
7239 | spin_lock(&all_mddevs_lock); | |
7240 | if (v == (void*)1) | |
7241 | tmp = all_mddevs.next; | |
7242 | else | |
7243 | tmp = mddev->all_mddevs.next; | |
7244 | if (tmp != &all_mddevs) | |
fd01b88c | 7245 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
7246 | else { |
7247 | next_mddev = (void*)2; | |
7248 | *pos = 0x10000; | |
f72ffdd6 | 7249 | } |
1da177e4 LT |
7250 | spin_unlock(&all_mddevs_lock); |
7251 | ||
7252 | if (v != (void*)1) | |
7253 | mddev_put(mddev); | |
7254 | return next_mddev; | |
7255 | ||
7256 | } | |
7257 | ||
7258 | static void md_seq_stop(struct seq_file *seq, void *v) | |
7259 | { | |
fd01b88c | 7260 | struct mddev *mddev = v; |
1da177e4 LT |
7261 | |
7262 | if (mddev && v != (void*)1 && v != (void*)2) | |
7263 | mddev_put(mddev); | |
7264 | } | |
7265 | ||
7266 | static int md_seq_show(struct seq_file *seq, void *v) | |
7267 | { | |
fd01b88c | 7268 | struct mddev *mddev = v; |
dd8ac336 | 7269 | sector_t sectors; |
3cb03002 | 7270 | struct md_rdev *rdev; |
1da177e4 LT |
7271 | |
7272 | if (v == (void*)1) { | |
84fc4b56 | 7273 | struct md_personality *pers; |
1da177e4 LT |
7274 | seq_printf(seq, "Personalities : "); |
7275 | spin_lock(&pers_lock); | |
2604b703 N |
7276 | list_for_each_entry(pers, &pers_list, list) |
7277 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
7278 | |
7279 | spin_unlock(&pers_lock); | |
7280 | seq_printf(seq, "\n"); | |
f1514638 | 7281 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
7282 | return 0; |
7283 | } | |
7284 | if (v == (void*)2) { | |
7285 | status_unused(seq); | |
7286 | return 0; | |
7287 | } | |
7288 | ||
36d091f4 | 7289 | spin_lock(&mddev->lock); |
1da177e4 LT |
7290 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7291 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7292 | mddev->pers ? "" : "in"); | |
7293 | if (mddev->pers) { | |
f91de92e | 7294 | if (mddev->ro==1) |
1da177e4 | 7295 | seq_printf(seq, " (read-only)"); |
f91de92e | 7296 | if (mddev->ro==2) |
52720ae7 | 7297 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7298 | seq_printf(seq, " %s", mddev->pers->name); |
7299 | } | |
7300 | ||
dd8ac336 | 7301 | sectors = 0; |
f97fcad3 N |
7302 | rcu_read_lock(); |
7303 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 LT |
7304 | char b[BDEVNAME_SIZE]; |
7305 | seq_printf(seq, " %s[%d]", | |
7306 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7307 | if (test_bit(WriteMostly, &rdev->flags)) |
7308 | seq_printf(seq, "(W)"); | |
b2d444d7 | 7309 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7310 | seq_printf(seq, "(F)"); |
7311 | continue; | |
2d78f8c4 N |
7312 | } |
7313 | if (rdev->raid_disk < 0) | |
b325a32e | 7314 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7315 | if (test_bit(Replacement, &rdev->flags)) |
7316 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7317 | sectors += rdev->sectors; |
1da177e4 | 7318 | } |
f97fcad3 | 7319 | rcu_read_unlock(); |
1da177e4 LT |
7320 | |
7321 | if (!list_empty(&mddev->disks)) { | |
7322 | if (mddev->pers) | |
7323 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7324 | (unsigned long long) |
7325 | mddev->array_sectors / 2); | |
1da177e4 LT |
7326 | else |
7327 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7328 | (unsigned long long)sectors / 2); |
1da177e4 | 7329 | } |
1cd6bf19 N |
7330 | if (mddev->persistent) { |
7331 | if (mddev->major_version != 0 || | |
7332 | mddev->minor_version != 90) { | |
7333 | seq_printf(seq," super %d.%d", | |
7334 | mddev->major_version, | |
7335 | mddev->minor_version); | |
7336 | } | |
e691063a N |
7337 | } else if (mddev->external) |
7338 | seq_printf(seq, " super external:%s", | |
7339 | mddev->metadata_type); | |
7340 | else | |
1cd6bf19 | 7341 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7342 | |
7343 | if (mddev->pers) { | |
d710e138 | 7344 | mddev->pers->status(seq, mddev); |
f72ffdd6 | 7345 | seq_printf(seq, "\n "); |
8e1b39d6 | 7346 | if (mddev->pers->sync_request) { |
f7851be7 | 7347 | if (status_resync(seq, mddev)) |
8e1b39d6 | 7348 | seq_printf(seq, "\n "); |
8e1b39d6 | 7349 | } |
32a7627c N |
7350 | } else |
7351 | seq_printf(seq, "\n "); | |
7352 | ||
57148964 | 7353 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7354 | |
7355 | seq_printf(seq, "\n"); | |
7356 | } | |
36d091f4 | 7357 | spin_unlock(&mddev->lock); |
f72ffdd6 | 7358 | |
1da177e4 LT |
7359 | return 0; |
7360 | } | |
7361 | ||
110518bc | 7362 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7363 | .start = md_seq_start, |
7364 | .next = md_seq_next, | |
7365 | .stop = md_seq_stop, | |
7366 | .show = md_seq_show, | |
7367 | }; | |
7368 | ||
7369 | static int md_seq_open(struct inode *inode, struct file *file) | |
7370 | { | |
f1514638 | 7371 | struct seq_file *seq; |
1da177e4 LT |
7372 | int error; |
7373 | ||
7374 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7375 | if (error) |
f1514638 KS |
7376 | return error; |
7377 | ||
7378 | seq = file->private_data; | |
7379 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7380 | return error; |
7381 | } | |
7382 | ||
e2f23b60 | 7383 | static int md_unloading; |
d7603b7e N |
7384 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
7385 | { | |
f1514638 | 7386 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
7387 | int mask; |
7388 | ||
e2f23b60 | 7389 | if (md_unloading) |
7d7e64f2 | 7390 | return POLLIN|POLLRDNORM|POLLERR|POLLPRI; |
d7603b7e N |
7391 | poll_wait(filp, &md_event_waiters, wait); |
7392 | ||
7393 | /* always allow read */ | |
7394 | mask = POLLIN | POLLRDNORM; | |
7395 | ||
f1514638 | 7396 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
7397 | mask |= POLLERR | POLLPRI; |
7398 | return mask; | |
7399 | } | |
7400 | ||
fa027c2a | 7401 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7402 | .owner = THIS_MODULE, |
1da177e4 LT |
7403 | .open = md_seq_open, |
7404 | .read = seq_read, | |
7405 | .llseek = seq_lseek, | |
c3f94b40 | 7406 | .release = seq_release_private, |
d7603b7e | 7407 | .poll = mdstat_poll, |
1da177e4 LT |
7408 | }; |
7409 | ||
84fc4b56 | 7410 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7411 | { |
50bd3774 CY |
7412 | printk(KERN_INFO "md: %s personality registered for level %d\n", |
7413 | p->name, p->level); | |
1da177e4 | 7414 | spin_lock(&pers_lock); |
2604b703 | 7415 | list_add_tail(&p->list, &pers_list); |
1da177e4 LT |
7416 | spin_unlock(&pers_lock); |
7417 | return 0; | |
7418 | } | |
6c144d31 | 7419 | EXPORT_SYMBOL(register_md_personality); |
1da177e4 | 7420 | |
84fc4b56 | 7421 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7422 | { |
2604b703 | 7423 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 7424 | spin_lock(&pers_lock); |
2604b703 | 7425 | list_del_init(&p->list); |
1da177e4 LT |
7426 | spin_unlock(&pers_lock); |
7427 | return 0; | |
7428 | } | |
6c144d31 | 7429 | EXPORT_SYMBOL(unregister_md_personality); |
1da177e4 | 7430 | |
edb39c9d GR |
7431 | int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module) |
7432 | { | |
7433 | if (md_cluster_ops != NULL) | |
7434 | return -EALREADY; | |
7435 | spin_lock(&pers_lock); | |
7436 | md_cluster_ops = ops; | |
7437 | md_cluster_mod = module; | |
7438 | spin_unlock(&pers_lock); | |
7439 | return 0; | |
7440 | } | |
7441 | EXPORT_SYMBOL(register_md_cluster_operations); | |
7442 | ||
7443 | int unregister_md_cluster_operations(void) | |
7444 | { | |
7445 | spin_lock(&pers_lock); | |
7446 | md_cluster_ops = NULL; | |
7447 | spin_unlock(&pers_lock); | |
7448 | return 0; | |
7449 | } | |
7450 | EXPORT_SYMBOL(unregister_md_cluster_operations); | |
7451 | ||
7452 | int md_setup_cluster(struct mddev *mddev, int nodes) | |
7453 | { | |
7454 | int err; | |
7455 | ||
7456 | err = request_module("md-cluster"); | |
7457 | if (err) { | |
7458 | pr_err("md-cluster module not found.\n"); | |
b0c26a79 | 7459 | return -ENOENT; |
edb39c9d GR |
7460 | } |
7461 | ||
7462 | spin_lock(&pers_lock); | |
7463 | if (!md_cluster_ops || !try_module_get(md_cluster_mod)) { | |
7464 | spin_unlock(&pers_lock); | |
7465 | return -ENOENT; | |
7466 | } | |
7467 | spin_unlock(&pers_lock); | |
7468 | ||
cf921cc1 | 7469 | return md_cluster_ops->join(mddev, nodes); |
edb39c9d GR |
7470 | } |
7471 | ||
7472 | void md_cluster_stop(struct mddev *mddev) | |
7473 | { | |
c4ce867f GR |
7474 | if (!md_cluster_ops) |
7475 | return; | |
edb39c9d GR |
7476 | md_cluster_ops->leave(mddev); |
7477 | module_put(md_cluster_mod); | |
7478 | } | |
7479 | ||
fd01b88c | 7480 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7481 | { |
f72ffdd6 | 7482 | struct md_rdev *rdev; |
1da177e4 | 7483 | int idle; |
eea1bf38 | 7484 | int curr_events; |
1da177e4 LT |
7485 | |
7486 | idle = 1; | |
4b80991c N |
7487 | rcu_read_lock(); |
7488 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7489 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7490 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7491 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7492 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7493 | /* sync IO will cause sync_io to increase before the disk_stats |
7494 | * as sync_io is counted when a request starts, and | |
7495 | * disk_stats is counted when it completes. | |
7496 | * So resync activity will cause curr_events to be smaller than | |
7497 | * when there was no such activity. | |
7498 | * non-sync IO will cause disk_stat to increase without | |
7499 | * increasing sync_io so curr_events will (eventually) | |
7500 | * be larger than it was before. Once it becomes | |
7501 | * substantially larger, the test below will cause | |
7502 | * the array to appear non-idle, and resync will slow | |
7503 | * down. | |
7504 | * If there is a lot of outstanding resync activity when | |
7505 | * we set last_event to curr_events, then all that activity | |
7506 | * completing might cause the array to appear non-idle | |
7507 | * and resync will be slowed down even though there might | |
7508 | * not have been non-resync activity. This will only | |
7509 | * happen once though. 'last_events' will soon reflect | |
7510 | * the state where there is little or no outstanding | |
7511 | * resync requests, and further resync activity will | |
7512 | * always make curr_events less than last_events. | |
c0e48521 | 7513 | * |
1da177e4 | 7514 | */ |
eea1bf38 | 7515 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
7516 | rdev->last_events = curr_events; |
7517 | idle = 0; | |
7518 | } | |
7519 | } | |
4b80991c | 7520 | rcu_read_unlock(); |
1da177e4 LT |
7521 | return idle; |
7522 | } | |
7523 | ||
fd01b88c | 7524 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
7525 | { |
7526 | /* another "blocks" (512byte) blocks have been synced */ | |
7527 | atomic_sub(blocks, &mddev->recovery_active); | |
7528 | wake_up(&mddev->recovery_wait); | |
7529 | if (!ok) { | |
dfc70645 | 7530 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 7531 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
7532 | md_wakeup_thread(mddev->thread); |
7533 | // stop recovery, signal do_sync .... | |
7534 | } | |
7535 | } | |
6c144d31 | 7536 | EXPORT_SYMBOL(md_done_sync); |
1da177e4 | 7537 | |
06d91a5f N |
7538 | /* md_write_start(mddev, bi) |
7539 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
7540 | * in superblock) before writing, schedule a superblock update |
7541 | * and wait for it to complete. | |
06d91a5f | 7542 | */ |
fd01b88c | 7543 | void md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 7544 | { |
0fd62b86 | 7545 | int did_change = 0; |
06d91a5f | 7546 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 7547 | return; |
06d91a5f | 7548 | |
f91de92e N |
7549 | BUG_ON(mddev->ro == 1); |
7550 | if (mddev->ro == 2) { | |
7551 | /* need to switch to read/write */ | |
7552 | mddev->ro = 0; | |
7553 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7554 | md_wakeup_thread(mddev->thread); | |
25156198 | 7555 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 7556 | did_change = 1; |
f91de92e | 7557 | } |
06d91a5f | 7558 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
7559 | if (mddev->safemode == 1) |
7560 | mddev->safemode = 0; | |
06d91a5f | 7561 | if (mddev->in_sync) { |
85572d7c | 7562 | spin_lock(&mddev->lock); |
3d310eb7 N |
7563 | if (mddev->in_sync) { |
7564 | mddev->in_sync = 0; | |
850b2b42 | 7565 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 7566 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 7567 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 7568 | did_change = 1; |
3d310eb7 | 7569 | } |
85572d7c | 7570 | spin_unlock(&mddev->lock); |
06d91a5f | 7571 | } |
0fd62b86 | 7572 | if (did_change) |
00bcb4ac | 7573 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 7574 | wait_event(mddev->sb_wait, |
09a44cc1 | 7575 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 | 7576 | } |
6c144d31 | 7577 | EXPORT_SYMBOL(md_write_start); |
1da177e4 | 7578 | |
fd01b88c | 7579 | void md_write_end(struct mddev *mddev) |
1da177e4 LT |
7580 | { |
7581 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
7582 | if (mddev->safemode == 2) | |
7583 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 7584 | else if (mddev->safemode_delay) |
1da177e4 LT |
7585 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
7586 | } | |
7587 | } | |
6c144d31 | 7588 | EXPORT_SYMBOL(md_write_end); |
1da177e4 | 7589 | |
2a2275d6 N |
7590 | /* md_allow_write(mddev) |
7591 | * Calling this ensures that the array is marked 'active' so that writes | |
7592 | * may proceed without blocking. It is important to call this before | |
7593 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
7594 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
7595 | * |
7596 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
7597 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 7598 | */ |
fd01b88c | 7599 | int md_allow_write(struct mddev *mddev) |
2a2275d6 N |
7600 | { |
7601 | if (!mddev->pers) | |
b5470dc5 | 7602 | return 0; |
2a2275d6 | 7603 | if (mddev->ro) |
b5470dc5 | 7604 | return 0; |
1a0fd497 | 7605 | if (!mddev->pers->sync_request) |
b5470dc5 | 7606 | return 0; |
2a2275d6 | 7607 | |
85572d7c | 7608 | spin_lock(&mddev->lock); |
2a2275d6 N |
7609 | if (mddev->in_sync) { |
7610 | mddev->in_sync = 0; | |
7611 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 7612 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
7613 | if (mddev->safemode_delay && |
7614 | mddev->safemode == 0) | |
7615 | mddev->safemode = 1; | |
85572d7c | 7616 | spin_unlock(&mddev->lock); |
293467aa GR |
7617 | if (mddev_is_clustered(mddev)) |
7618 | md_cluster_ops->metadata_update_start(mddev); | |
2a2275d6 | 7619 | md_update_sb(mddev, 0); |
293467aa GR |
7620 | if (mddev_is_clustered(mddev)) |
7621 | md_cluster_ops->metadata_update_finish(mddev); | |
00bcb4ac | 7622 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 | 7623 | } else |
85572d7c | 7624 | spin_unlock(&mddev->lock); |
b5470dc5 | 7625 | |
070dc6dd | 7626 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
7627 | return -EAGAIN; |
7628 | else | |
7629 | return 0; | |
2a2275d6 N |
7630 | } |
7631 | EXPORT_SYMBOL_GPL(md_allow_write); | |
7632 | ||
1da177e4 LT |
7633 | #define SYNC_MARKS 10 |
7634 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 7635 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 7636 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 7637 | { |
4ed8731d | 7638 | struct mddev *mddev = thread->mddev; |
fd01b88c | 7639 | struct mddev *mddev2; |
1da177e4 LT |
7640 | unsigned int currspeed = 0, |
7641 | window; | |
ac7e50a3 | 7642 | sector_t max_sectors,j, io_sectors, recovery_done; |
1da177e4 | 7643 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 7644 | unsigned long update_time; |
1da177e4 LT |
7645 | sector_t mark_cnt[SYNC_MARKS]; |
7646 | int last_mark,m; | |
7647 | struct list_head *tmp; | |
7648 | sector_t last_check; | |
57afd89f | 7649 | int skipped = 0; |
3cb03002 | 7650 | struct md_rdev *rdev; |
c4a39551 | 7651 | char *desc, *action = NULL; |
7c2c57c9 | 7652 | struct blk_plug plug; |
1da177e4 LT |
7653 | |
7654 | /* just incase thread restarts... */ | |
7655 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7656 | return; | |
3991b31e N |
7657 | if (mddev->ro) {/* never try to sync a read-only array */ |
7658 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5fd6c1dc | 7659 | return; |
3991b31e | 7660 | } |
1da177e4 | 7661 | |
61df9d91 | 7662 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 7663 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 7664 | desc = "data-check"; |
c4a39551 JB |
7665 | action = "check"; |
7666 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 7667 | desc = "requested-resync"; |
c4a39551 JB |
7668 | action = "repair"; |
7669 | } else | |
61df9d91 N |
7670 | desc = "resync"; |
7671 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
7672 | desc = "reshape"; | |
7673 | else | |
7674 | desc = "recovery"; | |
7675 | ||
c4a39551 JB |
7676 | mddev->last_sync_action = action ?: desc; |
7677 | ||
1da177e4 LT |
7678 | /* we overload curr_resync somewhat here. |
7679 | * 0 == not engaged in resync at all | |
7680 | * 2 == checking that there is no conflict with another sync | |
7681 | * 1 == like 2, but have yielded to allow conflicting resync to | |
7682 | * commense | |
7683 | * other == active in resync - this many blocks | |
7684 | * | |
7685 | * Before starting a resync we must have set curr_resync to | |
7686 | * 2, and then checked that every "conflicting" array has curr_resync | |
7687 | * less than ours. When we find one that is the same or higher | |
7688 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
7689 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
7690 | * This will mean we have to start checking from the beginning again. | |
7691 | * | |
7692 | */ | |
7693 | ||
7694 | do { | |
7695 | mddev->curr_resync = 2; | |
7696 | ||
7697 | try_again: | |
404e4b43 | 7698 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 7699 | goto skip; |
29ac4aa3 | 7700 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
7701 | if (mddev2 == mddev) |
7702 | continue; | |
90b08710 BS |
7703 | if (!mddev->parallel_resync |
7704 | && mddev2->curr_resync | |
7705 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
7706 | DEFINE_WAIT(wq); |
7707 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
7708 | /* arbitrarily yield */ | |
7709 | mddev->curr_resync = 1; | |
7710 | wake_up(&resync_wait); | |
7711 | } | |
7712 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
7713 | /* no need to wait here, we can wait the next | |
7714 | * time 'round when curr_resync == 2 | |
7715 | */ | |
7716 | continue; | |
9744197c N |
7717 | /* We need to wait 'interruptible' so as not to |
7718 | * contribute to the load average, and not to | |
7719 | * be caught by 'softlockup' | |
7720 | */ | |
7721 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 7722 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 7723 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
7724 | printk(KERN_INFO "md: delaying %s of %s" |
7725 | " until %s has finished (they" | |
1da177e4 | 7726 | " share one or more physical units)\n", |
61df9d91 | 7727 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 7728 | mddev_put(mddev2); |
9744197c N |
7729 | if (signal_pending(current)) |
7730 | flush_signals(current); | |
1da177e4 LT |
7731 | schedule(); |
7732 | finish_wait(&resync_wait, &wq); | |
7733 | goto try_again; | |
7734 | } | |
7735 | finish_wait(&resync_wait, &wq); | |
7736 | } | |
7737 | } | |
7738 | } while (mddev->curr_resync < 2); | |
7739 | ||
5fd6c1dc | 7740 | j = 0; |
9d88883e | 7741 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 7742 | /* resync follows the size requested by the personality, |
57afd89f | 7743 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
7744 | */ |
7745 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 7746 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 7747 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
7748 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
7749 | j = mddev->resync_min; | |
7750 | else if (!mddev->bitmap) | |
5fd6c1dc | 7751 | j = mddev->recovery_cp; |
5e96ee65 | 7752 | |
ccfcc3c1 | 7753 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 7754 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 7755 | else { |
1da177e4 | 7756 | /* recovery follows the physical size of devices */ |
58c0fed4 | 7757 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 7758 | j = MaxSector; |
4e59ca7d | 7759 | rcu_read_lock(); |
dafb20fa | 7760 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc N |
7761 | if (rdev->raid_disk >= 0 && |
7762 | !test_bit(Faulty, &rdev->flags) && | |
7763 | !test_bit(In_sync, &rdev->flags) && | |
7764 | rdev->recovery_offset < j) | |
7765 | j = rdev->recovery_offset; | |
4e59ca7d | 7766 | rcu_read_unlock(); |
133d4527 N |
7767 | |
7768 | /* If there is a bitmap, we need to make sure all | |
7769 | * writes that started before we added a spare | |
7770 | * complete before we start doing a recovery. | |
7771 | * Otherwise the write might complete and (via | |
7772 | * bitmap_endwrite) set a bit in the bitmap after the | |
7773 | * recovery has checked that bit and skipped that | |
7774 | * region. | |
7775 | */ | |
7776 | if (mddev->bitmap) { | |
7777 | mddev->pers->quiesce(mddev, 1); | |
7778 | mddev->pers->quiesce(mddev, 0); | |
7779 | } | |
5fd6c1dc | 7780 | } |
1da177e4 | 7781 | |
61df9d91 N |
7782 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
7783 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
7784 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 7785 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
7786 | "(but not more than %d KB/sec) for %s.\n", |
7787 | speed_max(mddev), desc); | |
1da177e4 | 7788 | |
eea1bf38 | 7789 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 7790 | |
57afd89f | 7791 | io_sectors = 0; |
1da177e4 LT |
7792 | for (m = 0; m < SYNC_MARKS; m++) { |
7793 | mark[m] = jiffies; | |
57afd89f | 7794 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
7795 | } |
7796 | last_mark = 0; | |
7797 | mddev->resync_mark = mark[last_mark]; | |
7798 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
7799 | ||
7800 | /* | |
7801 | * Tune reconstruction: | |
7802 | */ | |
7803 | window = 32*(PAGE_SIZE/512); | |
ac42450c JB |
7804 | printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", |
7805 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
7806 | |
7807 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
7808 | last_check = 0; |
7809 | ||
7810 | if (j>2) { | |
c91abf5a | 7811 | printk(KERN_INFO |
61df9d91 N |
7812 | "md: resuming %s of %s from checkpoint.\n", |
7813 | desc, mdname(mddev)); | |
1da177e4 | 7814 | mddev->curr_resync = j; |
72f36d59 N |
7815 | } else |
7816 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 7817 | mddev->curr_resync_completed = j; |
72f36d59 N |
7818 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
7819 | md_new_event(mddev); | |
54f89341 | 7820 | update_time = jiffies; |
1da177e4 | 7821 | |
96ae923a | 7822 | if (mddev_is_clustered(mddev)) |
965400eb | 7823 | md_cluster_ops->resync_start(mddev, j, max_sectors); |
96ae923a | 7824 | |
7c2c57c9 | 7825 | blk_start_plug(&plug); |
1da177e4 | 7826 | while (j < max_sectors) { |
57afd89f | 7827 | sector_t sectors; |
1da177e4 | 7828 | |
57afd89f | 7829 | skipped = 0; |
97e4f42d | 7830 | |
7a91ee1f N |
7831 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7832 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
7833 | (mddev->curr_resync - mddev->curr_resync_completed) | |
7834 | > (max_sectors >> 4)) || | |
54f89341 | 7835 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f | 7836 | (j - mddev->curr_resync_completed)*2 |
c5e19d90 N |
7837 | >= mddev->resync_max - mddev->curr_resync_completed || |
7838 | mddev->curr_resync_completed > mddev->resync_max | |
7a91ee1f | 7839 | )) { |
97e4f42d | 7840 | /* time to update curr_resync_completed */ |
97e4f42d N |
7841 | wait_event(mddev->recovery_wait, |
7842 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 7843 | mddev->curr_resync_completed = j; |
35d78c66 | 7844 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
7845 | j > mddev->recovery_cp) | |
7846 | mddev->recovery_cp = j; | |
54f89341 | 7847 | update_time = jiffies; |
070dc6dd | 7848 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 7849 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 7850 | } |
acb180b0 | 7851 | |
c91abf5a N |
7852 | while (j >= mddev->resync_max && |
7853 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
7854 | /* As this condition is controlled by user-space, |
7855 | * we can block indefinitely, so use '_interruptible' | |
7856 | * to avoid triggering warnings. | |
7857 | */ | |
7858 | flush_signals(current); /* just in case */ | |
7859 | wait_event_interruptible(mddev->recovery_wait, | |
7860 | mddev->resync_max > j | |
c91abf5a N |
7861 | || test_bit(MD_RECOVERY_INTR, |
7862 | &mddev->recovery)); | |
e62e58a5 | 7863 | } |
acb180b0 | 7864 | |
c91abf5a N |
7865 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7866 | break; | |
acb180b0 | 7867 | |
09314799 | 7868 | sectors = mddev->pers->sync_request(mddev, j, &skipped); |
57afd89f | 7869 | if (sectors == 0) { |
dfc70645 | 7870 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 7871 | break; |
1da177e4 | 7872 | } |
57afd89f N |
7873 | |
7874 | if (!skipped) { /* actual IO requested */ | |
7875 | io_sectors += sectors; | |
7876 | atomic_add(sectors, &mddev->recovery_active); | |
7877 | } | |
7878 | ||
e875ecea N |
7879 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7880 | break; | |
7881 | ||
1da177e4 | 7882 | j += sectors; |
72f36d59 N |
7883 | if (j > 2) |
7884 | mddev->curr_resync = j; | |
96ae923a GR |
7885 | if (mddev_is_clustered(mddev)) |
7886 | md_cluster_ops->resync_info_update(mddev, j, max_sectors); | |
ff4e8d9a | 7887 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 7888 | if (last_check == 0) |
e875ecea | 7889 | /* this is the earliest that rebuild will be |
d7603b7e N |
7890 | * visible in /proc/mdstat |
7891 | */ | |
7892 | md_new_event(mddev); | |
57afd89f N |
7893 | |
7894 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
7895 | continue; |
7896 | ||
57afd89f | 7897 | last_check = io_sectors; |
1da177e4 LT |
7898 | repeat: |
7899 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
7900 | /* step marks */ | |
7901 | int next = (last_mark+1) % SYNC_MARKS; | |
7902 | ||
7903 | mddev->resync_mark = mark[next]; | |
7904 | mddev->resync_mark_cnt = mark_cnt[next]; | |
7905 | mark[next] = jiffies; | |
57afd89f | 7906 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
7907 | last_mark = next; |
7908 | } | |
7909 | ||
c91abf5a N |
7910 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7911 | break; | |
1da177e4 LT |
7912 | |
7913 | /* | |
7914 | * this loop exits only if either when we are slower than | |
7915 | * the 'hard' speed limit, or the system was IO-idle for | |
7916 | * a jiffy. | |
7917 | * the system might be non-idle CPU-wise, but we only care | |
7918 | * about not overloading the IO subsystem. (things like an | |
7919 | * e2fsck being done on the RAID array should execute fast) | |
7920 | */ | |
1da177e4 LT |
7921 | cond_resched(); |
7922 | ||
ac7e50a3 XN |
7923 | recovery_done = io_sectors - atomic_read(&mddev->recovery_active); |
7924 | currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2 | |
57afd89f | 7925 | /((jiffies-mddev->resync_mark)/HZ +1) +1; |
1da177e4 | 7926 | |
88202a0c | 7927 | if (currspeed > speed_min(mddev)) { |
ac8fa419 | 7928 | if (currspeed > speed_max(mddev)) { |
c0e48521 | 7929 | msleep(500); |
1da177e4 LT |
7930 | goto repeat; |
7931 | } | |
ac8fa419 N |
7932 | if (!is_mddev_idle(mddev, 0)) { |
7933 | /* | |
7934 | * Give other IO more of a chance. | |
7935 | * The faster the devices, the less we wait. | |
7936 | */ | |
7937 | wait_event(mddev->recovery_wait, | |
7938 | !atomic_read(&mddev->recovery_active)); | |
7939 | } | |
1da177e4 LT |
7940 | } |
7941 | } | |
c91abf5a N |
7942 | printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc, |
7943 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
7944 | ? "interrupted" : "done"); | |
1da177e4 LT |
7945 | /* |
7946 | * this also signals 'finished resyncing' to md_stop | |
7947 | */ | |
7c2c57c9 | 7948 | blk_finish_plug(&plug); |
1da177e4 LT |
7949 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
7950 | ||
7951 | /* tell personality that we are finished */ | |
09314799 | 7952 | mddev->pers->sync_request(mddev, max_sectors, &skipped); |
1da177e4 | 7953 | |
96ae923a | 7954 | if (mddev_is_clustered(mddev)) |
965400eb | 7955 | md_cluster_ops->resync_finish(mddev); |
96ae923a | 7956 | |
dfc70645 | 7957 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
7958 | mddev->curr_resync > 2) { |
7959 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
7960 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7961 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
7962 | printk(KERN_INFO | |
61df9d91 N |
7963 | "md: checkpointing %s of %s.\n", |
7964 | desc, mdname(mddev)); | |
0a19caab | 7965 | if (test_bit(MD_RECOVERY_ERROR, |
7966 | &mddev->recovery)) | |
7967 | mddev->recovery_cp = | |
7968 | mddev->curr_resync_completed; | |
7969 | else | |
7970 | mddev->recovery_cp = | |
7971 | mddev->curr_resync; | |
5fd6c1dc N |
7972 | } |
7973 | } else | |
7974 | mddev->recovery_cp = MaxSector; | |
7975 | } else { | |
7976 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
7977 | mddev->curr_resync = MaxSector; | |
4e59ca7d | 7978 | rcu_read_lock(); |
dafb20fa | 7979 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 7980 | if (rdev->raid_disk >= 0 && |
70fffd0b | 7981 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
7982 | !test_bit(Faulty, &rdev->flags) && |
7983 | !test_bit(In_sync, &rdev->flags) && | |
7984 | rdev->recovery_offset < mddev->curr_resync) | |
7985 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 7986 | rcu_read_unlock(); |
5fd6c1dc | 7987 | } |
1da177e4 | 7988 | } |
db91ff55 | 7989 | skip: |
17571284 | 7990 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 7991 | |
23da422b | 7992 | spin_lock(&mddev->lock); |
c07b70ad N |
7993 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
7994 | /* We completed so min/max setting can be forgotten if used. */ | |
7995 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7996 | mddev->resync_min = 0; | |
7997 | mddev->resync_max = MaxSector; | |
7998 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7999 | mddev->resync_min = mddev->curr_resync_completed; | |
f7851be7 | 8000 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); |
1da177e4 | 8001 | mddev->curr_resync = 0; |
23da422b N |
8002 | spin_unlock(&mddev->lock); |
8003 | ||
1da177e4 | 8004 | wake_up(&resync_wait); |
1da177e4 | 8005 | md_wakeup_thread(mddev->thread); |
c6207277 | 8006 | return; |
1da177e4 | 8007 | } |
29269553 | 8008 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 8009 | |
746d3207 N |
8010 | static int remove_and_add_spares(struct mddev *mddev, |
8011 | struct md_rdev *this) | |
b4c4c7b8 | 8012 | { |
3cb03002 | 8013 | struct md_rdev *rdev; |
b4c4c7b8 | 8014 | int spares = 0; |
f2a371c5 | 8015 | int removed = 0; |
b4c4c7b8 | 8016 | |
dafb20fa | 8017 | rdev_for_each(rdev, mddev) |
746d3207 N |
8018 | if ((this == NULL || rdev == this) && |
8019 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 8020 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
8021 | (test_bit(Faulty, &rdev->flags) || |
8022 | ! test_bit(In_sync, &rdev->flags)) && | |
8023 | atomic_read(&rdev->nr_pending)==0) { | |
8024 | if (mddev->pers->hot_remove_disk( | |
b8321b68 | 8025 | mddev, rdev) == 0) { |
36fad858 | 8026 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 8027 | rdev->raid_disk = -1; |
f2a371c5 | 8028 | removed++; |
b4c4c7b8 N |
8029 | } |
8030 | } | |
90584fc9 JB |
8031 | if (removed && mddev->kobj.sd) |
8032 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 8033 | |
746d3207 N |
8034 | if (this) |
8035 | goto no_add; | |
8036 | ||
dafb20fa | 8037 | rdev_for_each(rdev, mddev) { |
7bfec5f3 N |
8038 | if (rdev->raid_disk >= 0 && |
8039 | !test_bit(In_sync, &rdev->flags) && | |
8040 | !test_bit(Faulty, &rdev->flags)) | |
8041 | spares++; | |
7ceb17e8 N |
8042 | if (rdev->raid_disk >= 0) |
8043 | continue; | |
8044 | if (test_bit(Faulty, &rdev->flags)) | |
8045 | continue; | |
8046 | if (mddev->ro && | |
8313b8e5 N |
8047 | ! (rdev->saved_raid_disk >= 0 && |
8048 | !test_bit(Bitmap_sync, &rdev->flags))) | |
7ceb17e8 N |
8049 | continue; |
8050 | ||
7eb41885 N |
8051 | if (rdev->saved_raid_disk < 0) |
8052 | rdev->recovery_offset = 0; | |
7ceb17e8 N |
8053 | if (mddev->pers-> |
8054 | hot_add_disk(mddev, rdev) == 0) { | |
8055 | if (sysfs_link_rdev(mddev, rdev)) | |
8056 | /* failure here is OK */; | |
8057 | spares++; | |
8058 | md_new_event(mddev); | |
8059 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
dfc70645 | 8060 | } |
b4c4c7b8 | 8061 | } |
746d3207 | 8062 | no_add: |
6dafab6b N |
8063 | if (removed) |
8064 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
b4c4c7b8 N |
8065 | return spares; |
8066 | } | |
7ebc0be7 | 8067 | |
ac05f256 N |
8068 | static void md_start_sync(struct work_struct *ws) |
8069 | { | |
8070 | struct mddev *mddev = container_of(ws, struct mddev, del_work); | |
8071 | ||
8072 | mddev->sync_thread = md_register_thread(md_do_sync, | |
8073 | mddev, | |
8074 | "resync"); | |
8075 | if (!mddev->sync_thread) { | |
8076 | printk(KERN_ERR "%s: could not start resync" | |
8077 | " thread...\n", | |
8078 | mdname(mddev)); | |
8079 | /* leave the spares where they are, it shouldn't hurt */ | |
8080 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
8081 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8082 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8083 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
8084 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8085 | wake_up(&resync_wait); |
ac05f256 N |
8086 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8087 | &mddev->recovery)) | |
8088 | if (mddev->sysfs_action) | |
8089 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8090 | } else | |
8091 | md_wakeup_thread(mddev->sync_thread); | |
8092 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8093 | md_new_event(mddev); | |
8094 | } | |
8095 | ||
1da177e4 LT |
8096 | /* |
8097 | * This routine is regularly called by all per-raid-array threads to | |
8098 | * deal with generic issues like resync and super-block update. | |
8099 | * Raid personalities that don't have a thread (linear/raid0) do not | |
8100 | * need this as they never do any recovery or update the superblock. | |
8101 | * | |
8102 | * It does not do any resync itself, but rather "forks" off other threads | |
8103 | * to do that as needed. | |
8104 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
8105 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 8106 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
8107 | * and wakeups up this thread which will reap the thread and finish up. |
8108 | * This thread also removes any faulty devices (with nr_pending == 0). | |
8109 | * | |
8110 | * The overall approach is: | |
8111 | * 1/ if the superblock needs updating, update it. | |
8112 | * 2/ If a recovery thread is running, don't do anything else. | |
8113 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
8114 | * 4/ If there are any faulty devices, remove them. | |
8115 | * 5/ If array is degraded, try to add spares devices | |
8116 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
8117 | */ | |
fd01b88c | 8118 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 8119 | { |
68866e42 JB |
8120 | if (mddev->suspended) |
8121 | return; | |
8122 | ||
5f40402d | 8123 | if (mddev->bitmap) |
aa5cbd10 | 8124 | bitmap_daemon_work(mddev); |
1da177e4 | 8125 | |
fca4d848 | 8126 | if (signal_pending(current)) { |
31a59e34 | 8127 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
8128 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
8129 | mdname(mddev)); | |
8130 | mddev->safemode = 2; | |
8131 | } | |
8132 | flush_signals(current); | |
8133 | } | |
8134 | ||
c89a8eee N |
8135 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
8136 | return; | |
1da177e4 | 8137 | if ( ! ( |
142d44c3 | 8138 | (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 8139 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 8140 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 8141 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
8142 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
8143 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
8144 | )) |
8145 | return; | |
fca4d848 | 8146 | |
df5b89b3 | 8147 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 8148 | int spares = 0; |
fca4d848 | 8149 | |
c89a8eee | 8150 | if (mddev->ro) { |
ab16bfc7 NB |
8151 | struct md_rdev *rdev; |
8152 | if (!mddev->external && mddev->in_sync) | |
8153 | /* 'Blocked' flag not needed as failed devices | |
8154 | * will be recorded if array switched to read/write. | |
8155 | * Leaving it set will prevent the device | |
8156 | * from being removed. | |
8157 | */ | |
8158 | rdev_for_each(rdev, mddev) | |
8159 | clear_bit(Blocked, &rdev->flags); | |
7ceb17e8 N |
8160 | /* On a read-only array we can: |
8161 | * - remove failed devices | |
8162 | * - add already-in_sync devices if the array itself | |
8163 | * is in-sync. | |
8164 | * As we only add devices that are already in-sync, | |
8165 | * we can activate the spares immediately. | |
c89a8eee | 8166 | */ |
7ceb17e8 | 8167 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
8168 | /* There is no thread, but we need to call |
8169 | * ->spare_active and clear saved_raid_disk | |
8170 | */ | |
2ac295a5 | 8171 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8313b8e5 | 8172 | md_reap_sync_thread(mddev); |
a4a3d26d | 8173 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
8313b8e5 | 8174 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
c89a8eee N |
8175 | goto unlock; |
8176 | } | |
8177 | ||
31a59e34 | 8178 | if (!mddev->external) { |
0fd62b86 | 8179 | int did_change = 0; |
85572d7c | 8180 | spin_lock(&mddev->lock); |
31a59e34 N |
8181 | if (mddev->safemode && |
8182 | !atomic_read(&mddev->writes_pending) && | |
8183 | !mddev->in_sync && | |
8184 | mddev->recovery_cp == MaxSector) { | |
8185 | mddev->in_sync = 1; | |
0fd62b86 | 8186 | did_change = 1; |
070dc6dd | 8187 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
8188 | } |
8189 | if (mddev->safemode == 1) | |
8190 | mddev->safemode = 0; | |
85572d7c | 8191 | spin_unlock(&mddev->lock); |
0fd62b86 | 8192 | if (did_change) |
00bcb4ac | 8193 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 8194 | } |
fca4d848 | 8195 | |
293467aa GR |
8196 | if (mddev->flags & MD_UPDATE_SB_FLAGS) { |
8197 | if (mddev_is_clustered(mddev)) | |
8198 | md_cluster_ops->metadata_update_start(mddev); | |
850b2b42 | 8199 | md_update_sb(mddev, 0); |
293467aa GR |
8200 | if (mddev_is_clustered(mddev)) |
8201 | md_cluster_ops->metadata_update_finish(mddev); | |
8202 | } | |
06d91a5f | 8203 | |
1da177e4 LT |
8204 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
8205 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
8206 | /* resync/recovery still happening */ | |
8207 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8208 | goto unlock; | |
8209 | } | |
8210 | if (mddev->sync_thread) { | |
a91d5ac0 | 8211 | md_reap_sync_thread(mddev); |
1da177e4 LT |
8212 | goto unlock; |
8213 | } | |
72a23c21 NB |
8214 | /* Set RUNNING before clearing NEEDED to avoid |
8215 | * any transients in the value of "sync_action". | |
8216 | */ | |
72f36d59 | 8217 | mddev->curr_resync_completed = 0; |
23da422b | 8218 | spin_lock(&mddev->lock); |
72a23c21 | 8219 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
23da422b | 8220 | spin_unlock(&mddev->lock); |
24dd469d N |
8221 | /* Clear some bits that don't mean anything, but |
8222 | * might be left set | |
8223 | */ | |
24dd469d N |
8224 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8225 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 8226 | |
ed209584 N |
8227 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
8228 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
ac05f256 | 8229 | goto not_running; |
1da177e4 LT |
8230 | /* no recovery is running. |
8231 | * remove any failed drives, then | |
8232 | * add spares if possible. | |
72f36d59 | 8233 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
8234 | * the personality to fail the re-add. |
8235 | */ | |
1da177e4 | 8236 | |
b4c4c7b8 | 8237 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
8238 | if (mddev->pers->check_reshape == NULL || |
8239 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 | 8240 | /* Cannot proceed */ |
ac05f256 | 8241 | goto not_running; |
b4c4c7b8 | 8242 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
72a23c21 | 8243 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 8244 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
8245 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8246 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 8247 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 8248 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8249 | } else if (mddev->recovery_cp < MaxSector) { |
8250 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 8251 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8252 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
8253 | /* nothing to be done ... */ | |
ac05f256 | 8254 | goto not_running; |
24dd469d | 8255 | |
1da177e4 | 8256 | if (mddev->pers->sync_request) { |
ef99bf48 | 8257 | if (spares) { |
a654b9d8 N |
8258 | /* We are adding a device or devices to an array |
8259 | * which has the bitmap stored on all devices. | |
8260 | * So make sure all bitmap pages get written | |
8261 | */ | |
8262 | bitmap_write_all(mddev->bitmap); | |
8263 | } | |
ac05f256 N |
8264 | INIT_WORK(&mddev->del_work, md_start_sync); |
8265 | queue_work(md_misc_wq, &mddev->del_work); | |
8266 | goto unlock; | |
1da177e4 | 8267 | } |
ac05f256 | 8268 | not_running: |
72a23c21 NB |
8269 | if (!mddev->sync_thread) { |
8270 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8271 | wake_up(&resync_wait); |
72a23c21 NB |
8272 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8273 | &mddev->recovery)) | |
0c3573f1 | 8274 | if (mddev->sysfs_action) |
00bcb4ac | 8275 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 8276 | } |
ac05f256 N |
8277 | unlock: |
8278 | wake_up(&mddev->sb_wait); | |
1da177e4 LT |
8279 | mddev_unlock(mddev); |
8280 | } | |
8281 | } | |
6c144d31 | 8282 | EXPORT_SYMBOL(md_check_recovery); |
1da177e4 | 8283 | |
a91d5ac0 JB |
8284 | void md_reap_sync_thread(struct mddev *mddev) |
8285 | { | |
8286 | struct md_rdev *rdev; | |
8287 | ||
8288 | /* resync has finished, collect result */ | |
8289 | md_unregister_thread(&mddev->sync_thread); | |
8290 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
8291 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
8292 | /* success...*/ | |
8293 | /* activate any spares */ | |
8294 | if (mddev->pers->spare_active(mddev)) { | |
8295 | sysfs_notify(&mddev->kobj, NULL, | |
8296 | "degraded"); | |
8297 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
8298 | } | |
8299 | } | |
293467aa GR |
8300 | if (mddev_is_clustered(mddev)) |
8301 | md_cluster_ops->metadata_update_start(mddev); | |
a91d5ac0 JB |
8302 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8303 | mddev->pers->finish_reshape) | |
8304 | mddev->pers->finish_reshape(mddev); | |
8305 | ||
8306 | /* If array is no-longer degraded, then any saved_raid_disk | |
f466722c | 8307 | * information must be scrapped. |
a91d5ac0 | 8308 | */ |
f466722c N |
8309 | if (!mddev->degraded) |
8310 | rdev_for_each(rdev, mddev) | |
a91d5ac0 JB |
8311 | rdev->saved_raid_disk = -1; |
8312 | ||
8313 | md_update_sb(mddev, 1); | |
293467aa GR |
8314 | if (mddev_is_clustered(mddev)) |
8315 | md_cluster_ops->metadata_update_finish(mddev); | |
a91d5ac0 | 8316 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
ea358cd0 | 8317 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
a91d5ac0 JB |
8318 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8319 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8320 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8321 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
f851b60d | 8322 | wake_up(&resync_wait); |
a91d5ac0 JB |
8323 | /* flag recovery needed just to double check */ |
8324 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8325 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8326 | md_new_event(mddev); | |
8327 | if (mddev->event_work.func) | |
8328 | queue_work(md_misc_wq, &mddev->event_work); | |
8329 | } | |
6c144d31 | 8330 | EXPORT_SYMBOL(md_reap_sync_thread); |
a91d5ac0 | 8331 | |
fd01b88c | 8332 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 8333 | { |
00bcb4ac | 8334 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 8335 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
8336 | !test_bit(Blocked, &rdev->flags) && |
8337 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
8338 | msecs_to_jiffies(5000)); |
8339 | rdev_dec_pending(rdev, mddev); | |
8340 | } | |
8341 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
8342 | ||
c6563a8c N |
8343 | void md_finish_reshape(struct mddev *mddev) |
8344 | { | |
8345 | /* called be personality module when reshape completes. */ | |
8346 | struct md_rdev *rdev; | |
8347 | ||
8348 | rdev_for_each(rdev, mddev) { | |
8349 | if (rdev->data_offset > rdev->new_data_offset) | |
8350 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
8351 | else | |
8352 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
8353 | rdev->data_offset = rdev->new_data_offset; | |
8354 | } | |
8355 | } | |
8356 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 N |
8357 | |
8358 | /* Bad block management. | |
8359 | * We can record which blocks on each device are 'bad' and so just | |
8360 | * fail those blocks, or that stripe, rather than the whole device. | |
8361 | * Entries in the bad-block table are 64bits wide. This comprises: | |
8362 | * Length of bad-range, in sectors: 0-511 for lengths 1-512 | |
8363 | * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes) | |
8364 | * A 'shift' can be set so that larger blocks are tracked and | |
8365 | * consequently larger devices can be covered. | |
8366 | * 'Acknowledged' flag - 1 bit. - the most significant bit. | |
8367 | * | |
8368 | * Locking of the bad-block table uses a seqlock so md_is_badblock | |
8369 | * might need to retry if it is very unlucky. | |
8370 | * We will sometimes want to check for bad blocks in a bi_end_io function, | |
8371 | * so we use the write_seqlock_irq variant. | |
8372 | * | |
8373 | * When looking for a bad block we specify a range and want to | |
8374 | * know if any block in the range is bad. So we binary-search | |
8375 | * to the last range that starts at-or-before the given endpoint, | |
8376 | * (or "before the sector after the target range") | |
8377 | * then see if it ends after the given start. | |
8378 | * We return | |
8379 | * 0 if there are no known bad blocks in the range | |
8380 | * 1 if there are known bad block which are all acknowledged | |
8381 | * -1 if there are bad blocks which have not yet been acknowledged in metadata. | |
8382 | * plus the start/length of the first bad section we overlap. | |
8383 | */ | |
8384 | int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, | |
8385 | sector_t *first_bad, int *bad_sectors) | |
8386 | { | |
8387 | int hi; | |
ab05613a | 8388 | int lo; |
2230dfe4 | 8389 | u64 *p = bb->page; |
ab05613a | 8390 | int rv; |
2230dfe4 N |
8391 | sector_t target = s + sectors; |
8392 | unsigned seq; | |
8393 | ||
8394 | if (bb->shift > 0) { | |
8395 | /* round the start down, and the end up */ | |
8396 | s >>= bb->shift; | |
8397 | target += (1<<bb->shift) - 1; | |
8398 | target >>= bb->shift; | |
8399 | sectors = target - s; | |
8400 | } | |
8401 | /* 'target' is now the first block after the bad range */ | |
8402 | ||
8403 | retry: | |
8404 | seq = read_seqbegin(&bb->lock); | |
ab05613a | 8405 | lo = 0; |
8406 | rv = 0; | |
2230dfe4 N |
8407 | hi = bb->count; |
8408 | ||
8409 | /* Binary search between lo and hi for 'target' | |
8410 | * i.e. for the last range that starts before 'target' | |
8411 | */ | |
8412 | /* INVARIANT: ranges before 'lo' and at-or-after 'hi' | |
8413 | * are known not to be the last range before target. | |
8414 | * VARIANT: hi-lo is the number of possible | |
8415 | * ranges, and decreases until it reaches 1 | |
8416 | */ | |
8417 | while (hi - lo > 1) { | |
8418 | int mid = (lo + hi) / 2; | |
8419 | sector_t a = BB_OFFSET(p[mid]); | |
8420 | if (a < target) | |
8421 | /* This could still be the one, earlier ranges | |
8422 | * could not. */ | |
8423 | lo = mid; | |
8424 | else | |
8425 | /* This and later ranges are definitely out. */ | |
8426 | hi = mid; | |
8427 | } | |
8428 | /* 'lo' might be the last that started before target, but 'hi' isn't */ | |
8429 | if (hi > lo) { | |
8430 | /* need to check all range that end after 's' to see if | |
8431 | * any are unacknowledged. | |
8432 | */ | |
8433 | while (lo >= 0 && | |
8434 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8435 | if (BB_OFFSET(p[lo]) < target) { | |
8436 | /* starts before the end, and finishes after | |
8437 | * the start, so they must overlap | |
8438 | */ | |
8439 | if (rv != -1 && BB_ACK(p[lo])) | |
8440 | rv = 1; | |
8441 | else | |
8442 | rv = -1; | |
8443 | *first_bad = BB_OFFSET(p[lo]); | |
8444 | *bad_sectors = BB_LEN(p[lo]); | |
8445 | } | |
8446 | lo--; | |
8447 | } | |
8448 | } | |
8449 | ||
8450 | if (read_seqretry(&bb->lock, seq)) | |
8451 | goto retry; | |
8452 | ||
8453 | return rv; | |
8454 | } | |
8455 | EXPORT_SYMBOL_GPL(md_is_badblock); | |
8456 | ||
8457 | /* | |
8458 | * Add a range of bad blocks to the table. | |
8459 | * This might extend the table, or might contract it | |
8460 | * if two adjacent ranges can be merged. | |
8461 | * We binary-search to find the 'insertion' point, then | |
8462 | * decide how best to handle it. | |
8463 | */ | |
8464 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, | |
8465 | int acknowledged) | |
8466 | { | |
8467 | u64 *p; | |
8468 | int lo, hi; | |
8469 | int rv = 1; | |
905b0297 | 8470 | unsigned long flags; |
2230dfe4 N |
8471 | |
8472 | if (bb->shift < 0) | |
8473 | /* badblocks are disabled */ | |
8474 | return 0; | |
8475 | ||
8476 | if (bb->shift) { | |
8477 | /* round the start down, and the end up */ | |
8478 | sector_t next = s + sectors; | |
8479 | s >>= bb->shift; | |
8480 | next += (1<<bb->shift) - 1; | |
8481 | next >>= bb->shift; | |
8482 | sectors = next - s; | |
8483 | } | |
8484 | ||
905b0297 | 8485 | write_seqlock_irqsave(&bb->lock, flags); |
2230dfe4 N |
8486 | |
8487 | p = bb->page; | |
8488 | lo = 0; | |
8489 | hi = bb->count; | |
8490 | /* Find the last range that starts at-or-before 's' */ | |
8491 | while (hi - lo > 1) { | |
8492 | int mid = (lo + hi) / 2; | |
8493 | sector_t a = BB_OFFSET(p[mid]); | |
8494 | if (a <= s) | |
8495 | lo = mid; | |
8496 | else | |
8497 | hi = mid; | |
8498 | } | |
8499 | if (hi > lo && BB_OFFSET(p[lo]) > s) | |
8500 | hi = lo; | |
8501 | ||
8502 | if (hi > lo) { | |
8503 | /* we found a range that might merge with the start | |
8504 | * of our new range | |
8505 | */ | |
8506 | sector_t a = BB_OFFSET(p[lo]); | |
8507 | sector_t e = a + BB_LEN(p[lo]); | |
8508 | int ack = BB_ACK(p[lo]); | |
8509 | if (e >= s) { | |
8510 | /* Yes, we can merge with a previous range */ | |
8511 | if (s == a && s + sectors >= e) | |
8512 | /* new range covers old */ | |
8513 | ack = acknowledged; | |
8514 | else | |
8515 | ack = ack && acknowledged; | |
8516 | ||
8517 | if (e < s + sectors) | |
8518 | e = s + sectors; | |
8519 | if (e - a <= BB_MAX_LEN) { | |
8520 | p[lo] = BB_MAKE(a, e-a, ack); | |
8521 | s = e; | |
8522 | } else { | |
8523 | /* does not all fit in one range, | |
8524 | * make p[lo] maximal | |
8525 | */ | |
8526 | if (BB_LEN(p[lo]) != BB_MAX_LEN) | |
8527 | p[lo] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8528 | s = a + BB_MAX_LEN; | |
8529 | } | |
8530 | sectors = e - s; | |
8531 | } | |
8532 | } | |
8533 | if (sectors && hi < bb->count) { | |
8534 | /* 'hi' points to the first range that starts after 's'. | |
8535 | * Maybe we can merge with the start of that range */ | |
8536 | sector_t a = BB_OFFSET(p[hi]); | |
8537 | sector_t e = a + BB_LEN(p[hi]); | |
8538 | int ack = BB_ACK(p[hi]); | |
8539 | if (a <= s + sectors) { | |
8540 | /* merging is possible */ | |
8541 | if (e <= s + sectors) { | |
8542 | /* full overlap */ | |
8543 | e = s + sectors; | |
8544 | ack = acknowledged; | |
8545 | } else | |
8546 | ack = ack && acknowledged; | |
8547 | ||
8548 | a = s; | |
8549 | if (e - a <= BB_MAX_LEN) { | |
8550 | p[hi] = BB_MAKE(a, e-a, ack); | |
8551 | s = e; | |
8552 | } else { | |
8553 | p[hi] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8554 | s = a + BB_MAX_LEN; | |
8555 | } | |
8556 | sectors = e - s; | |
8557 | lo = hi; | |
8558 | hi++; | |
8559 | } | |
8560 | } | |
8561 | if (sectors == 0 && hi < bb->count) { | |
8562 | /* we might be able to combine lo and hi */ | |
8563 | /* Note: 's' is at the end of 'lo' */ | |
8564 | sector_t a = BB_OFFSET(p[hi]); | |
8565 | int lolen = BB_LEN(p[lo]); | |
8566 | int hilen = BB_LEN(p[hi]); | |
8567 | int newlen = lolen + hilen - (s - a); | |
8568 | if (s >= a && newlen < BB_MAX_LEN) { | |
8569 | /* yes, we can combine them */ | |
8570 | int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]); | |
8571 | p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack); | |
8572 | memmove(p + hi, p + hi + 1, | |
8573 | (bb->count - hi - 1) * 8); | |
8574 | bb->count--; | |
8575 | } | |
8576 | } | |
8577 | while (sectors) { | |
8578 | /* didn't merge (it all). | |
8579 | * Need to add a range just before 'hi' */ | |
8580 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8581 | /* No room for more */ | |
8582 | rv = 0; | |
8583 | break; | |
8584 | } else { | |
8585 | int this_sectors = sectors; | |
8586 | memmove(p + hi + 1, p + hi, | |
8587 | (bb->count - hi) * 8); | |
8588 | bb->count++; | |
8589 | ||
8590 | if (this_sectors > BB_MAX_LEN) | |
8591 | this_sectors = BB_MAX_LEN; | |
8592 | p[hi] = BB_MAKE(s, this_sectors, acknowledged); | |
8593 | sectors -= this_sectors; | |
8594 | s += this_sectors; | |
8595 | } | |
8596 | } | |
8597 | ||
8598 | bb->changed = 1; | |
de393cde N |
8599 | if (!acknowledged) |
8600 | bb->unacked_exist = 1; | |
905b0297 | 8601 | write_sequnlock_irqrestore(&bb->lock, flags); |
2230dfe4 N |
8602 | |
8603 | return rv; | |
8604 | } | |
8605 | ||
3cb03002 | 8606 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8607 | int is_new) |
2230dfe4 | 8608 | { |
c6563a8c N |
8609 | int rv; |
8610 | if (is_new) | |
8611 | s += rdev->new_data_offset; | |
8612 | else | |
8613 | s += rdev->data_offset; | |
8614 | rv = md_set_badblocks(&rdev->badblocks, | |
8615 | s, sectors, 0); | |
2230dfe4 N |
8616 | if (rv) { |
8617 | /* Make sure they get written out promptly */ | |
8bd2f0a0 | 8618 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2230dfe4 N |
8619 | set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags); |
8620 | md_wakeup_thread(rdev->mddev->thread); | |
8621 | } | |
8622 | return rv; | |
8623 | } | |
8624 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8625 | ||
8626 | /* | |
8627 | * Remove a range of bad blocks from the table. | |
8628 | * This may involve extending the table if we spilt a region, | |
8629 | * but it must not fail. So if the table becomes full, we just | |
8630 | * drop the remove request. | |
8631 | */ | |
8632 | static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors) | |
8633 | { | |
8634 | u64 *p; | |
8635 | int lo, hi; | |
8636 | sector_t target = s + sectors; | |
8637 | int rv = 0; | |
8638 | ||
8639 | if (bb->shift > 0) { | |
8640 | /* When clearing we round the start up and the end down. | |
8641 | * This should not matter as the shift should align with | |
8642 | * the block size and no rounding should ever be needed. | |
8643 | * However it is better the think a block is bad when it | |
8644 | * isn't than to think a block is not bad when it is. | |
8645 | */ | |
8646 | s += (1<<bb->shift) - 1; | |
8647 | s >>= bb->shift; | |
8648 | target >>= bb->shift; | |
8649 | sectors = target - s; | |
8650 | } | |
8651 | ||
8652 | write_seqlock_irq(&bb->lock); | |
8653 | ||
8654 | p = bb->page; | |
8655 | lo = 0; | |
8656 | hi = bb->count; | |
8657 | /* Find the last range that starts before 'target' */ | |
8658 | while (hi - lo > 1) { | |
8659 | int mid = (lo + hi) / 2; | |
8660 | sector_t a = BB_OFFSET(p[mid]); | |
8661 | if (a < target) | |
8662 | lo = mid; | |
8663 | else | |
8664 | hi = mid; | |
8665 | } | |
8666 | if (hi > lo) { | |
8667 | /* p[lo] is the last range that could overlap the | |
8668 | * current range. Earlier ranges could also overlap, | |
8669 | * but only this one can overlap the end of the range. | |
8670 | */ | |
8671 | if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) { | |
8672 | /* Partial overlap, leave the tail of this range */ | |
8673 | int ack = BB_ACK(p[lo]); | |
8674 | sector_t a = BB_OFFSET(p[lo]); | |
8675 | sector_t end = a + BB_LEN(p[lo]); | |
8676 | ||
8677 | if (a < s) { | |
8678 | /* we need to split this range */ | |
8679 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8b32bf5e | 8680 | rv = -ENOSPC; |
2230dfe4 N |
8681 | goto out; |
8682 | } | |
8683 | memmove(p+lo+1, p+lo, (bb->count - lo) * 8); | |
8684 | bb->count++; | |
8685 | p[lo] = BB_MAKE(a, s-a, ack); | |
8686 | lo++; | |
8687 | } | |
8688 | p[lo] = BB_MAKE(target, end - target, ack); | |
8689 | /* there is no longer an overlap */ | |
8690 | hi = lo; | |
8691 | lo--; | |
8692 | } | |
8693 | while (lo >= 0 && | |
8694 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8695 | /* This range does overlap */ | |
8696 | if (BB_OFFSET(p[lo]) < s) { | |
8697 | /* Keep the early parts of this range. */ | |
8698 | int ack = BB_ACK(p[lo]); | |
8699 | sector_t start = BB_OFFSET(p[lo]); | |
8700 | p[lo] = BB_MAKE(start, s - start, ack); | |
8701 | /* now low doesn't overlap, so.. */ | |
8702 | break; | |
8703 | } | |
8704 | lo--; | |
8705 | } | |
8706 | /* 'lo' is strictly before, 'hi' is strictly after, | |
8707 | * anything between needs to be discarded | |
8708 | */ | |
8709 | if (hi - lo > 1) { | |
8710 | memmove(p+lo+1, p+hi, (bb->count - hi) * 8); | |
8711 | bb->count -= (hi - lo - 1); | |
8712 | } | |
8713 | } | |
8714 | ||
8715 | bb->changed = 1; | |
8716 | out: | |
8717 | write_sequnlock_irq(&bb->lock); | |
8718 | return rv; | |
8719 | } | |
8720 | ||
c6563a8c N |
8721 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8722 | int is_new) | |
2230dfe4 | 8723 | { |
c6563a8c N |
8724 | if (is_new) |
8725 | s += rdev->new_data_offset; | |
8726 | else | |
8727 | s += rdev->data_offset; | |
2230dfe4 | 8728 | return md_clear_badblocks(&rdev->badblocks, |
c6563a8c | 8729 | s, sectors); |
2230dfe4 N |
8730 | } |
8731 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8732 | ||
8733 | /* | |
8734 | * Acknowledge all bad blocks in a list. | |
8735 | * This only succeeds if ->changed is clear. It is used by | |
8736 | * in-kernel metadata updates | |
8737 | */ | |
8738 | void md_ack_all_badblocks(struct badblocks *bb) | |
8739 | { | |
8740 | if (bb->page == NULL || bb->changed) | |
8741 | /* no point even trying */ | |
8742 | return; | |
8743 | write_seqlock_irq(&bb->lock); | |
8744 | ||
ecb178bb | 8745 | if (bb->changed == 0 && bb->unacked_exist) { |
2230dfe4 N |
8746 | u64 *p = bb->page; |
8747 | int i; | |
8748 | for (i = 0; i < bb->count ; i++) { | |
8749 | if (!BB_ACK(p[i])) { | |
8750 | sector_t start = BB_OFFSET(p[i]); | |
8751 | int len = BB_LEN(p[i]); | |
8752 | p[i] = BB_MAKE(start, len, 1); | |
8753 | } | |
8754 | } | |
de393cde | 8755 | bb->unacked_exist = 0; |
2230dfe4 N |
8756 | } |
8757 | write_sequnlock_irq(&bb->lock); | |
8758 | } | |
8759 | EXPORT_SYMBOL_GPL(md_ack_all_badblocks); | |
8760 | ||
16c791a5 N |
8761 | /* sysfs access to bad-blocks list. |
8762 | * We present two files. | |
8763 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
8764 | * are recorded as bad. The list is truncated to fit within | |
8765 | * the one-page limit of sysfs. | |
8766 | * Writing "sector length" to this file adds an acknowledged | |
8767 | * bad block list. | |
8768 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
8769 | * been acknowledged. Writing to this file adds bad blocks | |
8770 | * without acknowledging them. This is largely for testing. | |
8771 | */ | |
8772 | ||
8773 | static ssize_t | |
8774 | badblocks_show(struct badblocks *bb, char *page, int unack) | |
8775 | { | |
8776 | size_t len; | |
8777 | int i; | |
8778 | u64 *p = bb->page; | |
8779 | unsigned seq; | |
8780 | ||
8781 | if (bb->shift < 0) | |
8782 | return 0; | |
8783 | ||
8784 | retry: | |
8785 | seq = read_seqbegin(&bb->lock); | |
8786 | ||
8787 | len = 0; | |
8788 | i = 0; | |
8789 | ||
8790 | while (len < PAGE_SIZE && i < bb->count) { | |
8791 | sector_t s = BB_OFFSET(p[i]); | |
8792 | unsigned int length = BB_LEN(p[i]); | |
8793 | int ack = BB_ACK(p[i]); | |
8794 | i++; | |
8795 | ||
8796 | if (unack && ack) | |
8797 | continue; | |
8798 | ||
8799 | len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n", | |
8800 | (unsigned long long)s << bb->shift, | |
8801 | length << bb->shift); | |
8802 | } | |
de393cde N |
8803 | if (unack && len == 0) |
8804 | bb->unacked_exist = 0; | |
16c791a5 N |
8805 | |
8806 | if (read_seqretry(&bb->lock, seq)) | |
8807 | goto retry; | |
8808 | ||
8809 | return len; | |
8810 | } | |
8811 | ||
8812 | #define DO_DEBUG 1 | |
8813 | ||
8814 | static ssize_t | |
8815 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack) | |
8816 | { | |
8817 | unsigned long long sector; | |
8818 | int length; | |
8819 | char newline; | |
8820 | #ifdef DO_DEBUG | |
8821 | /* Allow clearing via sysfs *only* for testing/debugging. | |
8822 | * Normally only a successful write may clear a badblock | |
8823 | */ | |
8824 | int clear = 0; | |
8825 | if (page[0] == '-') { | |
8826 | clear = 1; | |
8827 | page++; | |
8828 | } | |
8829 | #endif /* DO_DEBUG */ | |
8830 | ||
8831 | switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) { | |
8832 | case 3: | |
8833 | if (newline != '\n') | |
8834 | return -EINVAL; | |
8835 | case 2: | |
8836 | if (length <= 0) | |
8837 | return -EINVAL; | |
8838 | break; | |
8839 | default: | |
8840 | return -EINVAL; | |
8841 | } | |
8842 | ||
8843 | #ifdef DO_DEBUG | |
8844 | if (clear) { | |
8845 | md_clear_badblocks(bb, sector, length); | |
8846 | return len; | |
8847 | } | |
8848 | #endif /* DO_DEBUG */ | |
8849 | if (md_set_badblocks(bb, sector, length, !unack)) | |
8850 | return len; | |
8851 | else | |
8852 | return -ENOSPC; | |
8853 | } | |
8854 | ||
75c96f85 AB |
8855 | static int md_notify_reboot(struct notifier_block *this, |
8856 | unsigned long code, void *x) | |
1da177e4 LT |
8857 | { |
8858 | struct list_head *tmp; | |
fd01b88c | 8859 | struct mddev *mddev; |
2dba6a91 | 8860 | int need_delay = 0; |
1da177e4 | 8861 | |
c744a65c N |
8862 | for_each_mddev(mddev, tmp) { |
8863 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8864 | if (mddev->pers) |
8865 | __md_stop_writes(mddev); | |
0f62fb22 N |
8866 | if (mddev->persistent) |
8867 | mddev->safemode = 2; | |
c744a65c | 8868 | mddev_unlock(mddev); |
2dba6a91 | 8869 | } |
c744a65c | 8870 | need_delay = 1; |
1da177e4 | 8871 | } |
c744a65c N |
8872 | /* |
8873 | * certain more exotic SCSI devices are known to be | |
8874 | * volatile wrt too early system reboots. While the | |
8875 | * right place to handle this issue is the given | |
8876 | * driver, we do want to have a safe RAID driver ... | |
8877 | */ | |
8878 | if (need_delay) | |
8879 | mdelay(1000*1); | |
8880 | ||
1da177e4 LT |
8881 | return NOTIFY_DONE; |
8882 | } | |
8883 | ||
75c96f85 | 8884 | static struct notifier_block md_notifier = { |
1da177e4 LT |
8885 | .notifier_call = md_notify_reboot, |
8886 | .next = NULL, | |
8887 | .priority = INT_MAX, /* before any real devices */ | |
8888 | }; | |
8889 | ||
8890 | static void md_geninit(void) | |
8891 | { | |
36a4e1fe | 8892 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 8893 | |
c7705f34 | 8894 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
8895 | } |
8896 | ||
75c96f85 | 8897 | static int __init md_init(void) |
1da177e4 | 8898 | { |
e804ac78 TH |
8899 | int ret = -ENOMEM; |
8900 | ||
ada609ee | 8901 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
8902 | if (!md_wq) |
8903 | goto err_wq; | |
8904 | ||
8905 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
8906 | if (!md_misc_wq) | |
8907 | goto err_misc_wq; | |
8908 | ||
8909 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
8910 | goto err_md; | |
8911 | ||
8912 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
8913 | goto err_mdp; | |
8914 | mdp_major = ret; | |
8915 | ||
af5628f0 | 8916 | blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE, |
e8703fe1 N |
8917 | md_probe, NULL, NULL); |
8918 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
8919 | md_probe, NULL, NULL); |
8920 | ||
1da177e4 | 8921 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 8922 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
8923 | |
8924 | md_geninit(); | |
d710e138 | 8925 | return 0; |
1da177e4 | 8926 | |
e804ac78 TH |
8927 | err_mdp: |
8928 | unregister_blkdev(MD_MAJOR, "md"); | |
8929 | err_md: | |
8930 | destroy_workqueue(md_misc_wq); | |
8931 | err_misc_wq: | |
8932 | destroy_workqueue(md_wq); | |
8933 | err_wq: | |
8934 | return ret; | |
8935 | } | |
1da177e4 | 8936 | |
1d7e3e96 GR |
8937 | void md_reload_sb(struct mddev *mddev) |
8938 | { | |
8939 | struct md_rdev *rdev, *tmp; | |
8940 | ||
8941 | rdev_for_each_safe(rdev, tmp, mddev) { | |
8942 | rdev->sb_loaded = 0; | |
8943 | ClearPageUptodate(rdev->sb_page); | |
8944 | } | |
8945 | mddev->raid_disks = 0; | |
8946 | analyze_sbs(mddev); | |
8947 | rdev_for_each_safe(rdev, tmp, mddev) { | |
8948 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); | |
8949 | /* since we don't write to faulty devices, we figure out if the | |
8950 | * disk is faulty by comparing events | |
8951 | */ | |
8952 | if (mddev->events > sb->events) | |
8953 | set_bit(Faulty, &rdev->flags); | |
8954 | } | |
8955 | ||
8956 | } | |
8957 | EXPORT_SYMBOL(md_reload_sb); | |
8958 | ||
1da177e4 LT |
8959 | #ifndef MODULE |
8960 | ||
8961 | /* | |
8962 | * Searches all registered partitions for autorun RAID arrays | |
8963 | * at boot time. | |
8964 | */ | |
4d936ec1 ME |
8965 | |
8966 | static LIST_HEAD(all_detected_devices); | |
8967 | struct detected_devices_node { | |
8968 | struct list_head list; | |
8969 | dev_t dev; | |
8970 | }; | |
1da177e4 LT |
8971 | |
8972 | void md_autodetect_dev(dev_t dev) | |
8973 | { | |
4d936ec1 ME |
8974 | struct detected_devices_node *node_detected_dev; |
8975 | ||
8976 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
8977 | if (node_detected_dev) { | |
8978 | node_detected_dev->dev = dev; | |
8979 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
8980 | } else { | |
8981 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
8982 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
8983 | } | |
1da177e4 LT |
8984 | } |
8985 | ||
1da177e4 LT |
8986 | static void autostart_arrays(int part) |
8987 | { | |
3cb03002 | 8988 | struct md_rdev *rdev; |
4d936ec1 ME |
8989 | struct detected_devices_node *node_detected_dev; |
8990 | dev_t dev; | |
8991 | int i_scanned, i_passed; | |
1da177e4 | 8992 | |
4d936ec1 ME |
8993 | i_scanned = 0; |
8994 | i_passed = 0; | |
1da177e4 | 8995 | |
4d936ec1 | 8996 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 8997 | |
4d936ec1 ME |
8998 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
8999 | i_scanned++; | |
9000 | node_detected_dev = list_entry(all_detected_devices.next, | |
9001 | struct detected_devices_node, list); | |
9002 | list_del(&node_detected_dev->list); | |
9003 | dev = node_detected_dev->dev; | |
9004 | kfree(node_detected_dev); | |
df968c4e | 9005 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
9006 | if (IS_ERR(rdev)) |
9007 | continue; | |
9008 | ||
403df478 | 9009 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 9010 | continue; |
403df478 | 9011 | |
d0fae18f | 9012 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 9013 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 9014 | i_passed++; |
1da177e4 | 9015 | } |
4d936ec1 ME |
9016 | |
9017 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
9018 | i_scanned, i_passed); | |
1da177e4 LT |
9019 | |
9020 | autorun_devices(part); | |
9021 | } | |
9022 | ||
fdee8ae4 | 9023 | #endif /* !MODULE */ |
1da177e4 LT |
9024 | |
9025 | static __exit void md_exit(void) | |
9026 | { | |
fd01b88c | 9027 | struct mddev *mddev; |
1da177e4 | 9028 | struct list_head *tmp; |
e2f23b60 | 9029 | int delay = 1; |
8ab5e4c1 | 9030 | |
af5628f0 | 9031 | blk_unregister_region(MKDEV(MD_MAJOR,0), 512); |
e8703fe1 | 9032 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 9033 | |
3dbd8c2e | 9034 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
9035 | unregister_blkdev(mdp_major, "mdp"); |
9036 | unregister_reboot_notifier(&md_notifier); | |
9037 | unregister_sysctl_table(raid_table_header); | |
e2f23b60 N |
9038 | |
9039 | /* We cannot unload the modules while some process is | |
9040 | * waiting for us in select() or poll() - wake them up | |
9041 | */ | |
9042 | md_unloading = 1; | |
9043 | while (waitqueue_active(&md_event_waiters)) { | |
9044 | /* not safe to leave yet */ | |
9045 | wake_up(&md_event_waiters); | |
9046 | msleep(delay); | |
9047 | delay += delay; | |
9048 | } | |
1da177e4 | 9049 | remove_proc_entry("mdstat", NULL); |
e2f23b60 | 9050 | |
29ac4aa3 | 9051 | for_each_mddev(mddev, tmp) { |
1da177e4 | 9052 | export_array(mddev); |
d3374825 | 9053 | mddev->hold_active = 0; |
1da177e4 | 9054 | } |
e804ac78 TH |
9055 | destroy_workqueue(md_misc_wq); |
9056 | destroy_workqueue(md_wq); | |
1da177e4 LT |
9057 | } |
9058 | ||
685784aa | 9059 | subsys_initcall(md_init); |
1da177e4 LT |
9060 | module_exit(md_exit) |
9061 | ||
f91de92e N |
9062 | static int get_ro(char *buffer, struct kernel_param *kp) |
9063 | { | |
9064 | return sprintf(buffer, "%d", start_readonly); | |
9065 | } | |
9066 | static int set_ro(const char *val, struct kernel_param *kp) | |
9067 | { | |
4c9309c0 | 9068 | return kstrtouint(val, 10, (unsigned int *)&start_readonly); |
f91de92e N |
9069 | } |
9070 | ||
80ca3a44 N |
9071 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
9072 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
efeb53c0 | 9073 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 9074 | |
1da177e4 | 9075 | MODULE_LICENSE("GPL"); |
0efb9e61 | 9076 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 9077 | MODULE_ALIAS("md"); |
72008652 | 9078 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |