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; |
1edf80d3 | 2027 | char *s; |
5e55e2f5 | 2028 | int err; |
1da177e4 | 2029 | |
11e2ede0 DW |
2030 | /* prevent duplicates */ |
2031 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2032 | return -EEXIST; | |
2033 | ||
dd8ac336 AN |
2034 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
2035 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
2036 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2037 | if (mddev->pers) { |
2038 | /* Cannot change size, so fail | |
2039 | * If mddev->level <= 0, then we don't care | |
2040 | * about aligning sizes (e.g. linear) | |
2041 | */ | |
2042 | if (mddev->level > 0) | |
2043 | return -ENOSPC; | |
2044 | } else | |
dd8ac336 | 2045 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2046 | } |
1da177e4 LT |
2047 | |
2048 | /* Verify rdev->desc_nr is unique. | |
2049 | * If it is -1, assign a free number, else | |
2050 | * check number is not in use | |
2051 | */ | |
4878e9eb | 2052 | rcu_read_lock(); |
1da177e4 LT |
2053 | if (rdev->desc_nr < 0) { |
2054 | int choice = 0; | |
4878e9eb N |
2055 | if (mddev->pers) |
2056 | choice = mddev->raid_disks; | |
57d051dc | 2057 | while (md_find_rdev_nr_rcu(mddev, choice)) |
1da177e4 LT |
2058 | choice++; |
2059 | rdev->desc_nr = choice; | |
2060 | } else { | |
57d051dc | 2061 | if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) { |
4878e9eb | 2062 | rcu_read_unlock(); |
1da177e4 | 2063 | return -EBUSY; |
4878e9eb | 2064 | } |
1da177e4 | 2065 | } |
4878e9eb | 2066 | rcu_read_unlock(); |
de01dfad N |
2067 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
2068 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
2069 | mdname(mddev), mddev->max_disks); | |
2070 | return -EBUSY; | |
2071 | } | |
19133a42 | 2072 | bdevname(rdev->bdev,b); |
649316b2 | 2073 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 2074 | *s = '!'; |
649316b2 | 2075 | |
1da177e4 | 2076 | rdev->mddev = mddev; |
19133a42 | 2077 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 2078 | |
b2d6db58 | 2079 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2080 | goto fail; |
86e6ffdd | 2081 | |
0762b8bd | 2082 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2083 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2084 | /* failure here is OK */; | |
2085 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2086 | |
4b80991c | 2087 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2088 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2089 | |
2090 | /* May as well allow recovery to be retried once */ | |
5389042f | 2091 | mddev->recovery_disabled++; |
3f9d99c1 | 2092 | |
1da177e4 | 2093 | return 0; |
5e55e2f5 N |
2094 | |
2095 | fail: | |
2096 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
2097 | b, mdname(mddev)); | |
2098 | return err; | |
1da177e4 LT |
2099 | } |
2100 | ||
177a99b2 | 2101 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2102 | { |
3cb03002 | 2103 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2104 | kobject_del(&rdev->kobj); |
177a99b2 | 2105 | kobject_put(&rdev->kobj); |
5792a285 N |
2106 | } |
2107 | ||
f72ffdd6 | 2108 | static void unbind_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2109 | { |
2110 | char b[BDEVNAME_SIZE]; | |
403df478 | 2111 | |
49731baa | 2112 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2113 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
2114 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
2115 | rdev->mddev = NULL; | |
86e6ffdd | 2116 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2117 | sysfs_put(rdev->sysfs_state); |
2118 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2119 | rdev->badblocks.count = 0; |
5792a285 | 2120 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2121 | * writing to 'remove' to "dev/state". We also need |
2122 | * to delay it due to rcu usage. | |
5792a285 | 2123 | */ |
4b80991c | 2124 | synchronize_rcu(); |
177a99b2 N |
2125 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2126 | kobject_get(&rdev->kobj); | |
e804ac78 | 2127 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2128 | } |
2129 | ||
2130 | /* | |
2131 | * prevent the device from being mounted, repartitioned or | |
2132 | * otherwise reused by a RAID array (or any other kernel | |
2133 | * subsystem), by bd_claiming the device. | |
2134 | */ | |
3cb03002 | 2135 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2136 | { |
2137 | int err = 0; | |
2138 | struct block_device *bdev; | |
2139 | char b[BDEVNAME_SIZE]; | |
2140 | ||
d4d77629 | 2141 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2142 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 LT |
2143 | if (IS_ERR(bdev)) { |
2144 | printk(KERN_ERR "md: could not open %s.\n", | |
2145 | __bdevname(dev, b)); | |
2146 | return PTR_ERR(bdev); | |
2147 | } | |
1da177e4 LT |
2148 | rdev->bdev = bdev; |
2149 | return err; | |
2150 | } | |
2151 | ||
3cb03002 | 2152 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2153 | { |
2154 | struct block_device *bdev = rdev->bdev; | |
2155 | rdev->bdev = NULL; | |
e525fd89 | 2156 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2157 | } |
2158 | ||
2159 | void md_autodetect_dev(dev_t dev); | |
2160 | ||
f72ffdd6 | 2161 | static void export_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2162 | { |
2163 | char b[BDEVNAME_SIZE]; | |
403df478 | 2164 | |
1da177e4 LT |
2165 | printk(KERN_INFO "md: export_rdev(%s)\n", |
2166 | bdevname(rdev->bdev,b)); | |
545c8795 | 2167 | md_rdev_clear(rdev); |
1da177e4 | 2168 | #ifndef MODULE |
d0fae18f N |
2169 | if (test_bit(AutoDetected, &rdev->flags)) |
2170 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2171 | #endif |
2172 | unlock_rdev(rdev); | |
86e6ffdd | 2173 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2174 | } |
2175 | ||
fb56dfef | 2176 | void md_kick_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2177 | { |
2178 | unbind_rdev_from_array(rdev); | |
2179 | export_rdev(rdev); | |
2180 | } | |
fb56dfef | 2181 | EXPORT_SYMBOL_GPL(md_kick_rdev_from_array); |
1da177e4 | 2182 | |
fd01b88c | 2183 | static void export_array(struct mddev *mddev) |
1da177e4 | 2184 | { |
0638bb0e | 2185 | struct md_rdev *rdev; |
1da177e4 | 2186 | |
0638bb0e N |
2187 | while (!list_empty(&mddev->disks)) { |
2188 | rdev = list_first_entry(&mddev->disks, struct md_rdev, | |
2189 | same_set); | |
fb56dfef | 2190 | md_kick_rdev_from_array(rdev); |
1da177e4 | 2191 | } |
1da177e4 LT |
2192 | mddev->raid_disks = 0; |
2193 | mddev->major_version = 0; | |
2194 | } | |
2195 | ||
f72ffdd6 | 2196 | static void sync_sbs(struct mddev *mddev, int nospares) |
1da177e4 | 2197 | { |
42543769 N |
2198 | /* Update each superblock (in-memory image), but |
2199 | * if we are allowed to, skip spares which already | |
2200 | * have the right event counter, or have one earlier | |
2201 | * (which would mean they aren't being marked as dirty | |
2202 | * with the rest of the array) | |
2203 | */ | |
3cb03002 | 2204 | struct md_rdev *rdev; |
dafb20fa | 2205 | rdev_for_each(rdev, mddev) { |
42543769 N |
2206 | if (rdev->sb_events == mddev->events || |
2207 | (nospares && | |
2208 | rdev->raid_disk < 0 && | |
42543769 N |
2209 | rdev->sb_events+1 == mddev->events)) { |
2210 | /* Don't update this superblock */ | |
2211 | rdev->sb_loaded = 2; | |
2212 | } else { | |
076f968b | 2213 | sync_super(mddev, rdev); |
42543769 N |
2214 | rdev->sb_loaded = 1; |
2215 | } | |
1da177e4 LT |
2216 | } |
2217 | } | |
2218 | ||
1aee41f6 | 2219 | void md_update_sb(struct mddev *mddev, int force_change) |
1da177e4 | 2220 | { |
3cb03002 | 2221 | struct md_rdev *rdev; |
06d91a5f | 2222 | int sync_req; |
42543769 | 2223 | int nospares = 0; |
2699b672 | 2224 | int any_badblocks_changed = 0; |
1da177e4 | 2225 | |
d87f064f N |
2226 | if (mddev->ro) { |
2227 | if (force_change) | |
2228 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2229 | return; | |
2230 | } | |
1da177e4 | 2231 | repeat: |
3a3a5ddb | 2232 | /* First make sure individual recovery_offsets are correct */ |
dafb20fa | 2233 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2234 | if (rdev->raid_disk >= 0 && |
2235 | mddev->delta_disks >= 0 && | |
2236 | !test_bit(In_sync, &rdev->flags) && | |
2237 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2238 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2239 | ||
f72ffdd6 | 2240 | } |
bd52b746 | 2241 | if (!mddev->persistent) { |
070dc6dd | 2242 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2243 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
de393cde | 2244 | if (!mddev->external) { |
d97a41dc | 2245 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
dafb20fa | 2246 | rdev_for_each(rdev, mddev) { |
de393cde | 2247 | if (rdev->badblocks.changed) { |
d0962936 | 2248 | rdev->badblocks.changed = 0; |
de393cde N |
2249 | md_ack_all_badblocks(&rdev->badblocks); |
2250 | md_error(mddev, rdev); | |
2251 | } | |
2252 | clear_bit(Blocked, &rdev->flags); | |
2253 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2254 | wake_up(&rdev->blocked_wait); | |
2255 | } | |
2256 | } | |
3a3a5ddb N |
2257 | wake_up(&mddev->sb_wait); |
2258 | return; | |
2259 | } | |
2260 | ||
85572d7c | 2261 | spin_lock(&mddev->lock); |
84692195 | 2262 | |
3a3a5ddb N |
2263 | mddev->utime = get_seconds(); |
2264 | ||
850b2b42 N |
2265 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2266 | force_change = 1; | |
2267 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2268 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2269 | * though if events isn't the right even/odd, we will have to do | |
2270 | * spares after all | |
2271 | */ | |
2272 | nospares = 1; | |
2273 | if (force_change) | |
2274 | nospares = 0; | |
2275 | if (mddev->degraded) | |
84692195 N |
2276 | /* If the array is degraded, then skipping spares is both |
2277 | * dangerous and fairly pointless. | |
2278 | * Dangerous because a device that was removed from the array | |
2279 | * might have a event_count that still looks up-to-date, | |
2280 | * so it can be re-added without a resync. | |
2281 | * Pointless because if there are any spares to skip, | |
2282 | * then a recovery will happen and soon that array won't | |
2283 | * be degraded any more and the spare can go back to sleep then. | |
2284 | */ | |
850b2b42 | 2285 | nospares = 0; |
84692195 | 2286 | |
06d91a5f | 2287 | sync_req = mddev->in_sync; |
42543769 N |
2288 | |
2289 | /* If this is just a dirty<->clean transition, and the array is clean | |
2290 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2291 | if (nospares |
42543769 | 2292 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2293 | && mddev->can_decrease_events |
2294 | && mddev->events != 1) { | |
42543769 | 2295 | mddev->events--; |
a8707c08 N |
2296 | mddev->can_decrease_events = 0; |
2297 | } else { | |
42543769 N |
2298 | /* otherwise we have to go forward and ... */ |
2299 | mddev->events ++; | |
a8707c08 | 2300 | mddev->can_decrease_events = nospares; |
42543769 | 2301 | } |
1da177e4 | 2302 | |
403df478 N |
2303 | /* |
2304 | * This 64-bit counter should never wrap. | |
2305 | * Either we are in around ~1 trillion A.C., assuming | |
2306 | * 1 reboot per second, or we have a bug... | |
2307 | */ | |
2308 | WARN_ON(mddev->events == 0); | |
2699b672 | 2309 | |
dafb20fa | 2310 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2311 | if (rdev->badblocks.changed) |
2312 | any_badblocks_changed++; | |
de393cde N |
2313 | if (test_bit(Faulty, &rdev->flags)) |
2314 | set_bit(FaultRecorded, &rdev->flags); | |
2315 | } | |
2699b672 | 2316 | |
e691063a | 2317 | sync_sbs(mddev, nospares); |
85572d7c | 2318 | spin_unlock(&mddev->lock); |
1da177e4 | 2319 | |
36a4e1fe N |
2320 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2321 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2322 | |
4ad13663 | 2323 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2324 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2325 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2326 | |
42543769 N |
2327 | if (rdev->sb_loaded != 1) |
2328 | continue; /* no noise on spare devices */ | |
1da177e4 | 2329 | |
f466722c | 2330 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2331 | md_super_write(mddev,rdev, |
0f420358 | 2332 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2333 | rdev->sb_page); |
36a4e1fe N |
2334 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2335 | bdevname(rdev->bdev, b), | |
2336 | (unsigned long long)rdev->sb_start); | |
42543769 | 2337 | rdev->sb_events = mddev->events; |
2699b672 N |
2338 | if (rdev->badblocks.size) { |
2339 | md_super_write(mddev, rdev, | |
2340 | rdev->badblocks.sector, | |
2341 | rdev->badblocks.size << 9, | |
2342 | rdev->bb_page); | |
2343 | rdev->badblocks.size = 0; | |
2344 | } | |
7bfa19f2 | 2345 | |
f466722c | 2346 | } else |
36a4e1fe N |
2347 | pr_debug("md: %s (skipping faulty)\n", |
2348 | bdevname(rdev->bdev, b)); | |
d70ed2e4 | 2349 | |
7bfa19f2 | 2350 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2351 | /* only need to write one superblock... */ |
2352 | break; | |
2353 | } | |
a9701a30 | 2354 | md_super_wait(mddev); |
850b2b42 | 2355 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2356 | |
85572d7c | 2357 | spin_lock(&mddev->lock); |
850b2b42 N |
2358 | if (mddev->in_sync != sync_req || |
2359 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2360 | /* have to write it out again */ |
85572d7c | 2361 | spin_unlock(&mddev->lock); |
06d91a5f N |
2362 | goto repeat; |
2363 | } | |
850b2b42 | 2364 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
85572d7c | 2365 | spin_unlock(&mddev->lock); |
3d310eb7 | 2366 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2367 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2368 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2369 | |
dafb20fa | 2370 | rdev_for_each(rdev, mddev) { |
de393cde N |
2371 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2372 | clear_bit(Blocked, &rdev->flags); | |
2373 | ||
2374 | if (any_badblocks_changed) | |
2699b672 | 2375 | md_ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2376 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2377 | wake_up(&rdev->blocked_wait); | |
2378 | } | |
1da177e4 | 2379 | } |
1aee41f6 | 2380 | EXPORT_SYMBOL(md_update_sb); |
1da177e4 | 2381 | |
a6da4ef8 GR |
2382 | static int add_bound_rdev(struct md_rdev *rdev) |
2383 | { | |
2384 | struct mddev *mddev = rdev->mddev; | |
2385 | int err = 0; | |
2386 | ||
2387 | if (!mddev->pers->hot_remove_disk) { | |
2388 | /* If there is hot_add_disk but no hot_remove_disk | |
2389 | * then added disks for geometry changes, | |
2390 | * and should be added immediately. | |
2391 | */ | |
2392 | super_types[mddev->major_version]. | |
2393 | validate_super(mddev, rdev); | |
2394 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
2395 | if (err) { | |
2396 | unbind_rdev_from_array(rdev); | |
2397 | export_rdev(rdev); | |
2398 | return err; | |
2399 | } | |
2400 | } | |
2401 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
2402 | ||
2403 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2404 | if (mddev->degraded) | |
2405 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
2406 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2407 | md_new_event(mddev); | |
2408 | md_wakeup_thread(mddev->thread); | |
2409 | return 0; | |
2410 | } | |
1da177e4 | 2411 | |
7f6ce769 | 2412 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2413 | * We want to accept with case. For this we use cmd_match. |
2414 | */ | |
2415 | static int cmd_match(const char *cmd, const char *str) | |
2416 | { | |
2417 | /* See if cmd, written into a sysfs file, matches | |
2418 | * str. They must either be the same, or cmd can | |
2419 | * have a trailing newline | |
2420 | */ | |
2421 | while (*cmd && *str && *cmd == *str) { | |
2422 | cmd++; | |
2423 | str++; | |
2424 | } | |
2425 | if (*cmd == '\n') | |
2426 | cmd++; | |
2427 | if (*str || *cmd) | |
2428 | return 0; | |
2429 | return 1; | |
2430 | } | |
2431 | ||
86e6ffdd N |
2432 | struct rdev_sysfs_entry { |
2433 | struct attribute attr; | |
3cb03002 N |
2434 | ssize_t (*show)(struct md_rdev *, char *); |
2435 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2436 | }; |
2437 | ||
2438 | static ssize_t | |
3cb03002 | 2439 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd N |
2440 | { |
2441 | char *sep = ""; | |
20a49ff6 | 2442 | size_t len = 0; |
758bfc8a | 2443 | unsigned long flags = ACCESS_ONCE(rdev->flags); |
86e6ffdd | 2444 | |
758bfc8a | 2445 | if (test_bit(Faulty, &flags) || |
de393cde | 2446 | rdev->badblocks.unacked_exist) { |
86e6ffdd N |
2447 | len+= sprintf(page+len, "%sfaulty",sep); |
2448 | sep = ","; | |
2449 | } | |
758bfc8a | 2450 | if (test_bit(In_sync, &flags)) { |
86e6ffdd N |
2451 | len += sprintf(page+len, "%sin_sync",sep); |
2452 | sep = ","; | |
2453 | } | |
758bfc8a | 2454 | if (test_bit(WriteMostly, &flags)) { |
f655675b N |
2455 | len += sprintf(page+len, "%swrite_mostly",sep); |
2456 | sep = ","; | |
2457 | } | |
758bfc8a | 2458 | if (test_bit(Blocked, &flags) || |
52c64152 | 2459 | (rdev->badblocks.unacked_exist |
758bfc8a | 2460 | && !test_bit(Faulty, &flags))) { |
6bfe0b49 DW |
2461 | len += sprintf(page+len, "%sblocked", sep); |
2462 | sep = ","; | |
2463 | } | |
758bfc8a N |
2464 | if (!test_bit(Faulty, &flags) && |
2465 | !test_bit(In_sync, &flags)) { | |
86e6ffdd N |
2466 | len += sprintf(page+len, "%sspare", sep); |
2467 | sep = ","; | |
2468 | } | |
758bfc8a | 2469 | if (test_bit(WriteErrorSeen, &flags)) { |
d7a9d443 N |
2470 | len += sprintf(page+len, "%swrite_error", sep); |
2471 | sep = ","; | |
2472 | } | |
758bfc8a | 2473 | if (test_bit(WantReplacement, &flags)) { |
2d78f8c4 N |
2474 | len += sprintf(page+len, "%swant_replacement", sep); |
2475 | sep = ","; | |
2476 | } | |
758bfc8a | 2477 | if (test_bit(Replacement, &flags)) { |
2d78f8c4 N |
2478 | len += sprintf(page+len, "%sreplacement", sep); |
2479 | sep = ","; | |
2480 | } | |
2481 | ||
86e6ffdd N |
2482 | return len+sprintf(page+len, "\n"); |
2483 | } | |
2484 | ||
45dc2de1 | 2485 | static ssize_t |
3cb03002 | 2486 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2487 | { |
2488 | /* can write | |
de393cde | 2489 | * faulty - simulates an error |
45dc2de1 | 2490 | * remove - disconnects the device |
f655675b N |
2491 | * writemostly - sets write_mostly |
2492 | * -writemostly - clears write_mostly | |
de393cde N |
2493 | * blocked - sets the Blocked flags |
2494 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2495 | * insync - sets Insync providing device isn't active |
f466722c N |
2496 | * -insync - clear Insync for a device with a slot assigned, |
2497 | * so that it gets rebuilt based on bitmap | |
d7a9d443 N |
2498 | * write_error - sets WriteErrorSeen |
2499 | * -write_error - clears WriteErrorSeen | |
45dc2de1 N |
2500 | */ |
2501 | int err = -EINVAL; | |
2502 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2503 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2504 | if (test_bit(Faulty, &rdev->flags)) |
2505 | err = 0; | |
2506 | else | |
2507 | err = -EBUSY; | |
45dc2de1 N |
2508 | } else if (cmd_match(buf, "remove")) { |
2509 | if (rdev->raid_disk >= 0) | |
2510 | err = -EBUSY; | |
2511 | else { | |
fd01b88c | 2512 | struct mddev *mddev = rdev->mddev; |
293467aa | 2513 | if (mddev_is_clustered(mddev)) |
88bcfef7 | 2514 | md_cluster_ops->remove_disk(mddev, rdev); |
fb56dfef | 2515 | md_kick_rdev_from_array(rdev); |
88bcfef7 GR |
2516 | if (mddev_is_clustered(mddev)) |
2517 | md_cluster_ops->metadata_update_start(mddev); | |
3f9d7b0d N |
2518 | if (mddev->pers) |
2519 | md_update_sb(mddev, 1); | |
45dc2de1 | 2520 | md_new_event(mddev); |
293467aa GR |
2521 | if (mddev_is_clustered(mddev)) |
2522 | md_cluster_ops->metadata_update_finish(mddev); | |
45dc2de1 N |
2523 | err = 0; |
2524 | } | |
f655675b N |
2525 | } else if (cmd_match(buf, "writemostly")) { |
2526 | set_bit(WriteMostly, &rdev->flags); | |
2527 | err = 0; | |
2528 | } else if (cmd_match(buf, "-writemostly")) { | |
2529 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2530 | err = 0; |
2531 | } else if (cmd_match(buf, "blocked")) { | |
2532 | set_bit(Blocked, &rdev->flags); | |
2533 | err = 0; | |
2534 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2535 | if (!test_bit(Faulty, &rdev->flags) && |
7da64a0a | 2536 | rdev->badblocks.unacked_exist) { |
de393cde N |
2537 | /* metadata handler doesn't understand badblocks, |
2538 | * so we need to fail the device | |
2539 | */ | |
2540 | md_error(rdev->mddev, rdev); | |
2541 | } | |
6bfe0b49 | 2542 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2543 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2544 | wake_up(&rdev->blocked_wait); |
2545 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2546 | md_wakeup_thread(rdev->mddev->thread); | |
2547 | ||
6d56e278 N |
2548 | err = 0; |
2549 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2550 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2551 | err = 0; |
f466722c | 2552 | } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) { |
e1960f8c N |
2553 | if (rdev->mddev->pers == NULL) { |
2554 | clear_bit(In_sync, &rdev->flags); | |
2555 | rdev->saved_raid_disk = rdev->raid_disk; | |
2556 | rdev->raid_disk = -1; | |
2557 | err = 0; | |
2558 | } | |
d7a9d443 N |
2559 | } else if (cmd_match(buf, "write_error")) { |
2560 | set_bit(WriteErrorSeen, &rdev->flags); | |
2561 | err = 0; | |
2562 | } else if (cmd_match(buf, "-write_error")) { | |
2563 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2564 | err = 0; | |
2d78f8c4 N |
2565 | } else if (cmd_match(buf, "want_replacement")) { |
2566 | /* Any non-spare device that is not a replacement can | |
2567 | * become want_replacement at any time, but we then need to | |
2568 | * check if recovery is needed. | |
2569 | */ | |
2570 | if (rdev->raid_disk >= 0 && | |
2571 | !test_bit(Replacement, &rdev->flags)) | |
2572 | set_bit(WantReplacement, &rdev->flags); | |
2573 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2574 | md_wakeup_thread(rdev->mddev->thread); | |
2575 | err = 0; | |
2576 | } else if (cmd_match(buf, "-want_replacement")) { | |
2577 | /* Clearing 'want_replacement' is always allowed. | |
2578 | * Once replacements starts it is too late though. | |
2579 | */ | |
2580 | err = 0; | |
2581 | clear_bit(WantReplacement, &rdev->flags); | |
2582 | } else if (cmd_match(buf, "replacement")) { | |
2583 | /* Can only set a device as a replacement when array has not | |
2584 | * yet been started. Once running, replacement is automatic | |
2585 | * from spares, or by assigning 'slot'. | |
2586 | */ | |
2587 | if (rdev->mddev->pers) | |
2588 | err = -EBUSY; | |
2589 | else { | |
2590 | set_bit(Replacement, &rdev->flags); | |
2591 | err = 0; | |
2592 | } | |
2593 | } else if (cmd_match(buf, "-replacement")) { | |
2594 | /* Similarly, can only clear Replacement before start */ | |
2595 | if (rdev->mddev->pers) | |
2596 | err = -EBUSY; | |
2597 | else { | |
2598 | clear_bit(Replacement, &rdev->flags); | |
2599 | err = 0; | |
2600 | } | |
a6da4ef8 GR |
2601 | } else if (cmd_match(buf, "re-add")) { |
2602 | if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) { | |
97f6cd39 GR |
2603 | /* clear_bit is performed _after_ all the devices |
2604 | * have their local Faulty bit cleared. If any writes | |
2605 | * happen in the meantime in the local node, they | |
2606 | * will land in the local bitmap, which will be synced | |
2607 | * by this node eventually | |
2608 | */ | |
2609 | if (!mddev_is_clustered(rdev->mddev) || | |
2610 | (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) { | |
2611 | clear_bit(Faulty, &rdev->flags); | |
2612 | err = add_bound_rdev(rdev); | |
2613 | } | |
a6da4ef8 GR |
2614 | } else |
2615 | err = -EBUSY; | |
45dc2de1 | 2616 | } |
00bcb4ac N |
2617 | if (!err) |
2618 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2619 | return err ? err : len; |
2620 | } | |
80ca3a44 | 2621 | static struct rdev_sysfs_entry rdev_state = |
750f199e | 2622 | __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store); |
86e6ffdd | 2623 | |
4dbcdc75 | 2624 | static ssize_t |
3cb03002 | 2625 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2626 | { |
2627 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2628 | } | |
2629 | ||
2630 | static ssize_t | |
3cb03002 | 2631 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 | 2632 | { |
4c9309c0 AD |
2633 | unsigned int n; |
2634 | int rv; | |
2635 | ||
2636 | rv = kstrtouint(buf, 10, &n); | |
2637 | if (rv < 0) | |
2638 | return rv; | |
2639 | atomic_set(&rdev->corrected_errors, n); | |
2640 | return len; | |
4dbcdc75 N |
2641 | } |
2642 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2643 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2644 | |
014236d2 | 2645 | static ssize_t |
3cb03002 | 2646 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 N |
2647 | { |
2648 | if (rdev->raid_disk < 0) | |
2649 | return sprintf(page, "none\n"); | |
2650 | else | |
2651 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2652 | } | |
2653 | ||
2654 | static ssize_t | |
3cb03002 | 2655 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 | 2656 | { |
4c9309c0 | 2657 | int slot; |
c303da6d | 2658 | int err; |
4c9309c0 | 2659 | |
014236d2 N |
2660 | if (strncmp(buf, "none", 4)==0) |
2661 | slot = -1; | |
4c9309c0 AD |
2662 | else { |
2663 | err = kstrtouint(buf, 10, (unsigned int *)&slot); | |
2664 | if (err < 0) | |
2665 | return err; | |
2666 | } | |
6c2fce2e | 2667 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2668 | /* Setting 'slot' on an active array requires also |
2669 | * updating the 'rd%d' link, and communicating | |
2670 | * with the personality with ->hot_*_disk. | |
2671 | * For now we only support removing | |
2672 | * failed/spare devices. This normally happens automatically, | |
2673 | * but not when the metadata is externally managed. | |
2674 | */ | |
c303da6d N |
2675 | if (rdev->raid_disk == -1) |
2676 | return -EEXIST; | |
2677 | /* personality does all needed checks */ | |
01393f3d | 2678 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2679 | return -EINVAL; |
746d3207 N |
2680 | clear_bit(Blocked, &rdev->flags); |
2681 | remove_and_add_spares(rdev->mddev, rdev); | |
2682 | if (rdev->raid_disk >= 0) | |
2683 | return -EBUSY; | |
c303da6d N |
2684 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2685 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2686 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2687 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2688 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2689 | */ |
2690 | ||
2691 | if (rdev->raid_disk != -1) | |
2692 | return -EBUSY; | |
2693 | ||
c6751b2b N |
2694 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2695 | return -EBUSY; | |
2696 | ||
6c2fce2e NB |
2697 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2698 | return -EINVAL; | |
2699 | ||
ba1b41b6 N |
2700 | if (slot >= rdev->mddev->raid_disks && |
2701 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2702 | return -ENOSPC; | |
2703 | ||
6c2fce2e NB |
2704 | rdev->raid_disk = slot; |
2705 | if (test_bit(In_sync, &rdev->flags)) | |
2706 | rdev->saved_raid_disk = slot; | |
2707 | else | |
2708 | rdev->saved_raid_disk = -1; | |
d30519fc | 2709 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2710 | clear_bit(Bitmap_sync, &rdev->flags); |
6c2fce2e NB |
2711 | err = rdev->mddev->pers-> |
2712 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2713 | if (err) { |
6c2fce2e | 2714 | rdev->raid_disk = -1; |
6c2fce2e | 2715 | return err; |
52664732 | 2716 | } else |
00bcb4ac | 2717 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
36fad858 | 2718 | if (sysfs_link_rdev(rdev->mddev, rdev)) |
00bcb4ac | 2719 | /* failure here is OK */; |
6c2fce2e | 2720 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2721 | } else { |
ba1b41b6 N |
2722 | if (slot >= rdev->mddev->raid_disks && |
2723 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2724 | return -ENOSPC; |
2725 | rdev->raid_disk = slot; | |
2726 | /* assume it is working */ | |
c5d79adb N |
2727 | clear_bit(Faulty, &rdev->flags); |
2728 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2729 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2730 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2731 | } |
014236d2 N |
2732 | return len; |
2733 | } | |
2734 | ||
014236d2 | 2735 | static struct rdev_sysfs_entry rdev_slot = |
80ca3a44 | 2736 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2737 | |
93c8cad0 | 2738 | static ssize_t |
3cb03002 | 2739 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 2740 | { |
6961ece4 | 2741 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2742 | } |
2743 | ||
2744 | static ssize_t | |
3cb03002 | 2745 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 2746 | { |
c6563a8c | 2747 | unsigned long long offset; |
b29bebd6 | 2748 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 2749 | return -EINVAL; |
8ed0a521 | 2750 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2751 | return -EBUSY; |
dd8ac336 | 2752 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2753 | /* Must set offset before size, so overlap checks |
2754 | * can be sane */ | |
2755 | return -EBUSY; | |
93c8cad0 | 2756 | rdev->data_offset = offset; |
25f7fd47 | 2757 | rdev->new_data_offset = offset; |
93c8cad0 N |
2758 | return len; |
2759 | } | |
2760 | ||
2761 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2762 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2763 | |
c6563a8c N |
2764 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
2765 | { | |
2766 | return sprintf(page, "%llu\n", | |
2767 | (unsigned long long)rdev->new_data_offset); | |
2768 | } | |
2769 | ||
2770 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
2771 | const char *buf, size_t len) | |
2772 | { | |
2773 | unsigned long long new_offset; | |
2774 | struct mddev *mddev = rdev->mddev; | |
2775 | ||
b29bebd6 | 2776 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
2777 | return -EINVAL; |
2778 | ||
f851b60d N |
2779 | if (mddev->sync_thread || |
2780 | test_bit(MD_RECOVERY_RUNNING,&mddev->recovery)) | |
c6563a8c N |
2781 | return -EBUSY; |
2782 | if (new_offset == rdev->data_offset) | |
2783 | /* reset is always permitted */ | |
2784 | ; | |
2785 | else if (new_offset > rdev->data_offset) { | |
2786 | /* must not push array size beyond rdev_sectors */ | |
2787 | if (new_offset - rdev->data_offset | |
2788 | + mddev->dev_sectors > rdev->sectors) | |
2789 | return -E2BIG; | |
2790 | } | |
2791 | /* Metadata worries about other space details. */ | |
2792 | ||
2793 | /* decreasing the offset is inconsistent with a backwards | |
2794 | * reshape. | |
2795 | */ | |
2796 | if (new_offset < rdev->data_offset && | |
2797 | mddev->reshape_backwards) | |
2798 | return -EINVAL; | |
2799 | /* Increasing offset is inconsistent with forwards | |
2800 | * reshape. reshape_direction should be set to | |
2801 | * 'backwards' first. | |
2802 | */ | |
2803 | if (new_offset > rdev->data_offset && | |
2804 | !mddev->reshape_backwards) | |
2805 | return -EINVAL; | |
2806 | ||
2807 | if (mddev->pers && mddev->persistent && | |
2808 | !super_types[mddev->major_version] | |
2809 | .allow_new_offset(rdev, new_offset)) | |
2810 | return -E2BIG; | |
2811 | rdev->new_data_offset = new_offset; | |
2812 | if (new_offset > rdev->data_offset) | |
2813 | mddev->reshape_backwards = 1; | |
2814 | else if (new_offset < rdev->data_offset) | |
2815 | mddev->reshape_backwards = 0; | |
2816 | ||
2817 | return len; | |
2818 | } | |
2819 | static struct rdev_sysfs_entry rdev_new_offset = | |
2820 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
2821 | ||
83303b61 | 2822 | static ssize_t |
3cb03002 | 2823 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 2824 | { |
dd8ac336 | 2825 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2826 | } |
2827 | ||
c5d79adb N |
2828 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2829 | { | |
2830 | /* check if two start/length pairs overlap */ | |
2831 | if (s1+l1 <= s2) | |
2832 | return 0; | |
2833 | if (s2+l2 <= s1) | |
2834 | return 0; | |
2835 | return 1; | |
2836 | } | |
2837 | ||
b522adcd DW |
2838 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2839 | { | |
2840 | unsigned long long blocks; | |
2841 | sector_t new; | |
2842 | ||
b29bebd6 | 2843 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
2844 | return -EINVAL; |
2845 | ||
2846 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2847 | return -EINVAL; /* sector conversion overflow */ | |
2848 | ||
2849 | new = blocks * 2; | |
2850 | if (new != blocks * 2) | |
2851 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2852 | ||
2853 | *sectors = new; | |
2854 | return 0; | |
2855 | } | |
2856 | ||
83303b61 | 2857 | static ssize_t |
3cb03002 | 2858 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 2859 | { |
fd01b88c | 2860 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 2861 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2862 | sector_t sectors; |
27c529bb | 2863 | |
b522adcd | 2864 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2865 | return -EINVAL; |
c6563a8c N |
2866 | if (rdev->data_offset != rdev->new_data_offset) |
2867 | return -EINVAL; /* too confusing */ | |
0cd17fec | 2868 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2869 | if (my_mddev->persistent) { |
dd8ac336 AN |
2870 | sectors = super_types[my_mddev->major_version]. |
2871 | rdev_size_change(rdev, sectors); | |
2872 | if (!sectors) | |
0cd17fec | 2873 | return -EBUSY; |
dd8ac336 | 2874 | } else if (!sectors) |
77304d2a | 2875 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 2876 | rdev->data_offset; |
a6468539 N |
2877 | if (!my_mddev->pers->resize) |
2878 | /* Cannot change size for RAID0 or Linear etc */ | |
2879 | return -EINVAL; | |
0cd17fec | 2880 | } |
dd8ac336 | 2881 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2882 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2883 | |
dd8ac336 AN |
2884 | rdev->sectors = sectors; |
2885 | if (sectors > oldsectors && my_mddev->external) { | |
8b1afc3d N |
2886 | /* Need to check that all other rdevs with the same |
2887 | * ->bdev do not overlap. 'rcu' is sufficient to walk | |
2888 | * the rdev lists safely. | |
2889 | * This check does not provide a hard guarantee, it | |
2890 | * just helps avoid dangerous mistakes. | |
c5d79adb | 2891 | */ |
fd01b88c | 2892 | struct mddev *mddev; |
c5d79adb | 2893 | int overlap = 0; |
159ec1fc | 2894 | struct list_head *tmp; |
c5d79adb | 2895 | |
8b1afc3d | 2896 | rcu_read_lock(); |
29ac4aa3 | 2897 | for_each_mddev(mddev, tmp) { |
3cb03002 | 2898 | struct md_rdev *rdev2; |
c5d79adb | 2899 | |
dafb20fa | 2900 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
2901 | if (rdev->bdev == rdev2->bdev && |
2902 | rdev != rdev2 && | |
2903 | overlaps(rdev->data_offset, rdev->sectors, | |
2904 | rdev2->data_offset, | |
2905 | rdev2->sectors)) { | |
c5d79adb N |
2906 | overlap = 1; |
2907 | break; | |
2908 | } | |
c5d79adb N |
2909 | if (overlap) { |
2910 | mddev_put(mddev); | |
2911 | break; | |
2912 | } | |
2913 | } | |
8b1afc3d | 2914 | rcu_read_unlock(); |
c5d79adb N |
2915 | if (overlap) { |
2916 | /* Someone else could have slipped in a size | |
2917 | * change here, but doing so is just silly. | |
dd8ac336 | 2918 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2919 | * safe, and trust userspace not to race with |
2920 | * itself | |
2921 | */ | |
dd8ac336 | 2922 | rdev->sectors = oldsectors; |
c5d79adb N |
2923 | return -EBUSY; |
2924 | } | |
2925 | } | |
83303b61 N |
2926 | return len; |
2927 | } | |
2928 | ||
2929 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2930 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2931 | |
3cb03002 | 2932 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
2933 | { |
2934 | unsigned long long recovery_start = rdev->recovery_offset; | |
2935 | ||
2936 | if (test_bit(In_sync, &rdev->flags) || | |
2937 | recovery_start == MaxSector) | |
2938 | return sprintf(page, "none\n"); | |
2939 | ||
2940 | return sprintf(page, "%llu\n", recovery_start); | |
2941 | } | |
2942 | ||
3cb03002 | 2943 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
2944 | { |
2945 | unsigned long long recovery_start; | |
2946 | ||
2947 | if (cmd_match(buf, "none")) | |
2948 | recovery_start = MaxSector; | |
b29bebd6 | 2949 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
2950 | return -EINVAL; |
2951 | ||
2952 | if (rdev->mddev->pers && | |
2953 | rdev->raid_disk >= 0) | |
2954 | return -EBUSY; | |
2955 | ||
2956 | rdev->recovery_offset = recovery_start; | |
2957 | if (recovery_start == MaxSector) | |
2958 | set_bit(In_sync, &rdev->flags); | |
2959 | else | |
2960 | clear_bit(In_sync, &rdev->flags); | |
2961 | return len; | |
2962 | } | |
2963 | ||
2964 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2965 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2966 | ||
16c791a5 N |
2967 | static ssize_t |
2968 | badblocks_show(struct badblocks *bb, char *page, int unack); | |
2969 | static ssize_t | |
2970 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack); | |
2971 | ||
3cb03002 | 2972 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
2973 | { |
2974 | return badblocks_show(&rdev->badblocks, page, 0); | |
2975 | } | |
3cb03002 | 2976 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 2977 | { |
de393cde N |
2978 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
2979 | /* Maybe that ack was all we needed */ | |
2980 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
2981 | wake_up(&rdev->blocked_wait); | |
2982 | return rv; | |
16c791a5 N |
2983 | } |
2984 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
2985 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
2986 | ||
3cb03002 | 2987 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
2988 | { |
2989 | return badblocks_show(&rdev->badblocks, page, 1); | |
2990 | } | |
3cb03002 | 2991 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
2992 | { |
2993 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
2994 | } | |
2995 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
2996 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
2997 | ||
86e6ffdd N |
2998 | static struct attribute *rdev_default_attrs[] = { |
2999 | &rdev_state.attr, | |
4dbcdc75 | 3000 | &rdev_errors.attr, |
014236d2 | 3001 | &rdev_slot.attr, |
93c8cad0 | 3002 | &rdev_offset.attr, |
c6563a8c | 3003 | &rdev_new_offset.attr, |
83303b61 | 3004 | &rdev_size.attr, |
06e3c817 | 3005 | &rdev_recovery_start.attr, |
16c791a5 N |
3006 | &rdev_bad_blocks.attr, |
3007 | &rdev_unack_bad_blocks.attr, | |
86e6ffdd N |
3008 | NULL, |
3009 | }; | |
3010 | static ssize_t | |
3011 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3012 | { | |
3013 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3014 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
86e6ffdd N |
3015 | |
3016 | if (!entry->show) | |
3017 | return -EIO; | |
758bfc8a N |
3018 | if (!rdev->mddev) |
3019 | return -EBUSY; | |
3020 | return entry->show(rdev, page); | |
86e6ffdd N |
3021 | } |
3022 | ||
3023 | static ssize_t | |
3024 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
3025 | const char *page, size_t length) | |
3026 | { | |
3027 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 3028 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 3029 | ssize_t rv; |
fd01b88c | 3030 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
3031 | |
3032 | if (!entry->store) | |
3033 | return -EIO; | |
67463acb N |
3034 | if (!capable(CAP_SYS_ADMIN)) |
3035 | return -EACCES; | |
27c529bb | 3036 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 3037 | if (!rv) { |
27c529bb N |
3038 | if (rdev->mddev == NULL) |
3039 | rv = -EBUSY; | |
3040 | else | |
3041 | rv = entry->store(rdev, page, length); | |
6a51830e | 3042 | mddev_unlock(mddev); |
ca388059 N |
3043 | } |
3044 | return rv; | |
86e6ffdd N |
3045 | } |
3046 | ||
3047 | static void rdev_free(struct kobject *ko) | |
3048 | { | |
3cb03002 | 3049 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
3050 | kfree(rdev); |
3051 | } | |
52cf25d0 | 3052 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
3053 | .show = rdev_attr_show, |
3054 | .store = rdev_attr_store, | |
3055 | }; | |
3056 | static struct kobj_type rdev_ktype = { | |
3057 | .release = rdev_free, | |
3058 | .sysfs_ops = &rdev_sysfs_ops, | |
3059 | .default_attrs = rdev_default_attrs, | |
3060 | }; | |
3061 | ||
3cb03002 | 3062 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
3063 | { |
3064 | rdev->desc_nr = -1; | |
3065 | rdev->saved_raid_disk = -1; | |
3066 | rdev->raid_disk = -1; | |
3067 | rdev->flags = 0; | |
3068 | rdev->data_offset = 0; | |
c6563a8c | 3069 | rdev->new_data_offset = 0; |
e8bb9a83 N |
3070 | rdev->sb_events = 0; |
3071 | rdev->last_read_error.tv_sec = 0; | |
3072 | rdev->last_read_error.tv_nsec = 0; | |
2699b672 N |
3073 | rdev->sb_loaded = 0; |
3074 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3075 | atomic_set(&rdev->nr_pending, 0); |
3076 | atomic_set(&rdev->read_errors, 0); | |
3077 | atomic_set(&rdev->corrected_errors, 0); | |
3078 | ||
3079 | INIT_LIST_HEAD(&rdev->same_set); | |
3080 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3081 | |
3082 | /* Add space to store bad block list. | |
3083 | * This reserves the space even on arrays where it cannot | |
3084 | * be used - I wonder if that matters | |
3085 | */ | |
3086 | rdev->badblocks.count = 0; | |
486adf72 | 3087 | rdev->badblocks.shift = -1; /* disabled until explicitly enabled */ |
2230dfe4 N |
3088 | rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3089 | seqlock_init(&rdev->badblocks.lock); | |
3090 | if (rdev->badblocks.page == NULL) | |
3091 | return -ENOMEM; | |
3092 | ||
3093 | return 0; | |
e8bb9a83 N |
3094 | } |
3095 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3096 | /* |
3097 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3098 | * | |
3099 | * mark the device faulty if: | |
3100 | * | |
3101 | * - the device is nonexistent (zero size) | |
3102 | * - the device has no valid superblock | |
3103 | * | |
3104 | * a faulty rdev _never_ has rdev->sb set. | |
3105 | */ | |
3cb03002 | 3106 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3107 | { |
3108 | char b[BDEVNAME_SIZE]; | |
3109 | int err; | |
3cb03002 | 3110 | struct md_rdev *rdev; |
1da177e4 LT |
3111 | sector_t size; |
3112 | ||
9ffae0cf | 3113 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
3114 | if (!rdev) { |
3115 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
3116 | return ERR_PTR(-ENOMEM); | |
3117 | } | |
1da177e4 | 3118 | |
2230dfe4 N |
3119 | err = md_rdev_init(rdev); |
3120 | if (err) | |
3121 | goto abort_free; | |
3122 | err = alloc_disk_sb(rdev); | |
3123 | if (err) | |
1da177e4 LT |
3124 | goto abort_free; |
3125 | ||
c5d79adb | 3126 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3127 | if (err) |
3128 | goto abort_free; | |
3129 | ||
f9cb074b | 3130 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3131 | |
77304d2a | 3132 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 | 3133 | if (!size) { |
f72ffdd6 | 3134 | printk(KERN_WARNING |
1da177e4 LT |
3135 | "md: %s has zero or unknown size, marking faulty!\n", |
3136 | bdevname(rdev->bdev,b)); | |
3137 | err = -EINVAL; | |
3138 | goto abort_free; | |
3139 | } | |
3140 | ||
3141 | if (super_format >= 0) { | |
3142 | err = super_types[super_format]. | |
3143 | load_super(rdev, NULL, super_minor); | |
3144 | if (err == -EINVAL) { | |
df968c4e N |
3145 | printk(KERN_WARNING |
3146 | "md: %s does not have a valid v%d.%d " | |
3147 | "superblock, not importing!\n", | |
3148 | bdevname(rdev->bdev,b), | |
3149 | super_format, super_minor); | |
1da177e4 LT |
3150 | goto abort_free; |
3151 | } | |
3152 | if (err < 0) { | |
f72ffdd6 | 3153 | printk(KERN_WARNING |
1da177e4 LT |
3154 | "md: could not read %s's sb, not importing!\n", |
3155 | bdevname(rdev->bdev,b)); | |
3156 | goto abort_free; | |
3157 | } | |
3158 | } | |
6bfe0b49 | 3159 | |
1da177e4 LT |
3160 | return rdev; |
3161 | ||
3162 | abort_free: | |
2699b672 N |
3163 | if (rdev->bdev) |
3164 | unlock_rdev(rdev); | |
545c8795 | 3165 | md_rdev_clear(rdev); |
1da177e4 LT |
3166 | kfree(rdev); |
3167 | return ERR_PTR(err); | |
3168 | } | |
3169 | ||
3170 | /* | |
3171 | * Check a full RAID array for plausibility | |
3172 | */ | |
3173 | ||
f72ffdd6 | 3174 | static void analyze_sbs(struct mddev *mddev) |
1da177e4 LT |
3175 | { |
3176 | int i; | |
3cb03002 | 3177 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3178 | char b[BDEVNAME_SIZE]; |
3179 | ||
3180 | freshest = NULL; | |
dafb20fa | 3181 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3182 | switch (super_types[mddev->major_version]. |
3183 | load_super(rdev, freshest, mddev->minor_version)) { | |
3184 | case 1: | |
3185 | freshest = rdev; | |
3186 | break; | |
3187 | case 0: | |
3188 | break; | |
3189 | default: | |
3190 | printk( KERN_ERR \ | |
3191 | "md: fatal superblock inconsistency in %s" | |
f72ffdd6 | 3192 | " -- removing from array\n", |
1da177e4 | 3193 | bdevname(rdev->bdev,b)); |
fb56dfef | 3194 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3195 | } |
3196 | ||
1da177e4 LT |
3197 | super_types[mddev->major_version]. |
3198 | validate_super(mddev, freshest); | |
3199 | ||
3200 | i = 0; | |
dafb20fa | 3201 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3202 | if (mddev->max_disks && |
3203 | (rdev->desc_nr >= mddev->max_disks || | |
3204 | i > mddev->max_disks)) { | |
de01dfad N |
3205 | printk(KERN_WARNING |
3206 | "md: %s: %s: only %d devices permitted\n", | |
3207 | mdname(mddev), bdevname(rdev->bdev, b), | |
3208 | mddev->max_disks); | |
fb56dfef | 3209 | md_kick_rdev_from_array(rdev); |
de01dfad N |
3210 | continue; |
3211 | } | |
1aee41f6 | 3212 | if (rdev != freshest) { |
1da177e4 LT |
3213 | if (super_types[mddev->major_version]. |
3214 | validate_super(mddev, rdev)) { | |
3215 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
3216 | " from array!\n", | |
3217 | bdevname(rdev->bdev,b)); | |
fb56dfef | 3218 | md_kick_rdev_from_array(rdev); |
1da177e4 LT |
3219 | continue; |
3220 | } | |
1aee41f6 GR |
3221 | /* No device should have a Candidate flag |
3222 | * when reading devices | |
3223 | */ | |
3224 | if (test_bit(Candidate, &rdev->flags)) { | |
3225 | pr_info("md: kicking Cluster Candidate %s from array!\n", | |
3226 | bdevname(rdev->bdev, b)); | |
fb56dfef | 3227 | md_kick_rdev_from_array(rdev); |
1aee41f6 GR |
3228 | } |
3229 | } | |
1da177e4 LT |
3230 | if (mddev->level == LEVEL_MULTIPATH) { |
3231 | rdev->desc_nr = i++; | |
3232 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3233 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 3234 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
3235 | rdev->raid_disk = -1; |
3236 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3237 | } |
3238 | } | |
1da177e4 LT |
3239 | } |
3240 | ||
72e02075 N |
3241 | /* Read a fixed-point number. |
3242 | * Numbers in sysfs attributes should be in "standard" units where | |
3243 | * possible, so time should be in seconds. | |
f72ffdd6 | 3244 | * However we internally use a a much smaller unit such as |
72e02075 N |
3245 | * milliseconds or jiffies. |
3246 | * This function takes a decimal number with a possible fractional | |
3247 | * component, and produces an integer which is the result of | |
3248 | * multiplying that number by 10^'scale'. | |
3249 | * all without any floating-point arithmetic. | |
3250 | */ | |
3251 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3252 | { | |
3253 | unsigned long result = 0; | |
3254 | long decimals = -1; | |
3255 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3256 | if (*cp == '.') | |
3257 | decimals = 0; | |
3258 | else if (decimals < scale) { | |
3259 | unsigned int value; | |
3260 | value = *cp - '0'; | |
3261 | result = result * 10 + value; | |
3262 | if (decimals >= 0) | |
3263 | decimals++; | |
3264 | } | |
3265 | cp++; | |
3266 | } | |
3267 | if (*cp == '\n') | |
3268 | cp++; | |
3269 | if (*cp) | |
3270 | return -EINVAL; | |
3271 | if (decimals < 0) | |
3272 | decimals = 0; | |
3273 | while (decimals < scale) { | |
3274 | result *= 10; | |
3275 | decimals ++; | |
3276 | } | |
3277 | *res = result; | |
3278 | return 0; | |
3279 | } | |
3280 | ||
19052c0e N |
3281 | static void md_safemode_timeout(unsigned long data); |
3282 | ||
16f17b39 | 3283 | static ssize_t |
fd01b88c | 3284 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3285 | { |
3286 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3287 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3288 | } | |
3289 | static ssize_t | |
fd01b88c | 3290 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3291 | { |
16f17b39 | 3292 | unsigned long msec; |
97ce0a7f | 3293 | |
72e02075 | 3294 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3295 | return -EINVAL; |
16f17b39 N |
3296 | if (msec == 0) |
3297 | mddev->safemode_delay = 0; | |
3298 | else { | |
19052c0e | 3299 | unsigned long old_delay = mddev->safemode_delay; |
1b30e66f N |
3300 | unsigned long new_delay = (msec*HZ)/1000; |
3301 | ||
3302 | if (new_delay == 0) | |
3303 | new_delay = 1; | |
3304 | mddev->safemode_delay = new_delay; | |
3305 | if (new_delay < old_delay || old_delay == 0) | |
3306 | mod_timer(&mddev->safemode_timer, jiffies+1); | |
16f17b39 N |
3307 | } |
3308 | return len; | |
3309 | } | |
3310 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3311 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3312 | |
eae1701f | 3313 | static ssize_t |
fd01b88c | 3314 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3315 | { |
36d091f4 N |
3316 | struct md_personality *p; |
3317 | int ret; | |
3318 | spin_lock(&mddev->lock); | |
3319 | p = mddev->pers; | |
d9d166c2 | 3320 | if (p) |
36d091f4 | 3321 | ret = sprintf(page, "%s\n", p->name); |
d9d166c2 | 3322 | else if (mddev->clevel[0]) |
36d091f4 | 3323 | ret = sprintf(page, "%s\n", mddev->clevel); |
d9d166c2 | 3324 | else if (mddev->level != LEVEL_NONE) |
36d091f4 | 3325 | ret = sprintf(page, "%d\n", mddev->level); |
d9d166c2 | 3326 | else |
36d091f4 N |
3327 | ret = 0; |
3328 | spin_unlock(&mddev->lock); | |
3329 | return ret; | |
eae1701f N |
3330 | } |
3331 | ||
d9d166c2 | 3332 | static ssize_t |
fd01b88c | 3333 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3334 | { |
f2859af6 | 3335 | char clevel[16]; |
6791875e N |
3336 | ssize_t rv; |
3337 | size_t slen = len; | |
db721d32 | 3338 | struct md_personality *pers, *oldpers; |
f2859af6 | 3339 | long level; |
db721d32 | 3340 | void *priv, *oldpriv; |
3cb03002 | 3341 | struct md_rdev *rdev; |
245f46c2 | 3342 | |
6791875e N |
3343 | if (slen == 0 || slen >= sizeof(clevel)) |
3344 | return -EINVAL; | |
3345 | ||
3346 | rv = mddev_lock(mddev); | |
3347 | if (rv) | |
3348 | return rv; | |
3349 | ||
245f46c2 | 3350 | if (mddev->pers == NULL) { |
6791875e N |
3351 | strncpy(mddev->clevel, buf, slen); |
3352 | if (mddev->clevel[slen-1] == '\n') | |
3353 | slen--; | |
3354 | mddev->clevel[slen] = 0; | |
245f46c2 | 3355 | mddev->level = LEVEL_NONE; |
6791875e N |
3356 | rv = len; |
3357 | goto out_unlock; | |
245f46c2 | 3358 | } |
6791875e | 3359 | rv = -EROFS; |
bd8839e0 | 3360 | if (mddev->ro) |
6791875e | 3361 | goto out_unlock; |
245f46c2 N |
3362 | |
3363 | /* request to change the personality. Need to ensure: | |
3364 | * - array is not engaged in resync/recovery/reshape | |
3365 | * - old personality can be suspended | |
3366 | * - new personality will access other array. | |
3367 | */ | |
3368 | ||
6791875e | 3369 | rv = -EBUSY; |
bb4f1e9d | 3370 | if (mddev->sync_thread || |
f851b60d | 3371 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
bb4f1e9d N |
3372 | mddev->reshape_position != MaxSector || |
3373 | mddev->sysfs_active) | |
6791875e | 3374 | goto out_unlock; |
245f46c2 | 3375 | |
6791875e | 3376 | rv = -EINVAL; |
245f46c2 N |
3377 | if (!mddev->pers->quiesce) { |
3378 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3379 | mdname(mddev), mddev->pers->name); | |
6791875e | 3380 | goto out_unlock; |
245f46c2 N |
3381 | } |
3382 | ||
3383 | /* Now find the new personality */ | |
6791875e N |
3384 | strncpy(clevel, buf, slen); |
3385 | if (clevel[slen-1] == '\n') | |
3386 | slen--; | |
3387 | clevel[slen] = 0; | |
b29bebd6 | 3388 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3389 | level = LEVEL_NONE; |
245f46c2 | 3390 | |
f2859af6 DW |
3391 | if (request_module("md-%s", clevel) != 0) |
3392 | request_module("md-level-%s", clevel); | |
245f46c2 | 3393 | spin_lock(&pers_lock); |
f2859af6 | 3394 | pers = find_pers(level, clevel); |
245f46c2 N |
3395 | if (!pers || !try_module_get(pers->owner)) { |
3396 | spin_unlock(&pers_lock); | |
f2859af6 | 3397 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
6791875e N |
3398 | rv = -EINVAL; |
3399 | goto out_unlock; | |
245f46c2 N |
3400 | } |
3401 | spin_unlock(&pers_lock); | |
3402 | ||
3403 | if (pers == mddev->pers) { | |
3404 | /* Nothing to do! */ | |
3405 | module_put(pers->owner); | |
6791875e N |
3406 | rv = len; |
3407 | goto out_unlock; | |
245f46c2 N |
3408 | } |
3409 | if (!pers->takeover) { | |
3410 | module_put(pers->owner); | |
3411 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3412 | mdname(mddev), clevel); |
6791875e N |
3413 | rv = -EINVAL; |
3414 | goto out_unlock; | |
245f46c2 N |
3415 | } |
3416 | ||
dafb20fa | 3417 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3418 | rdev->new_raid_disk = rdev->raid_disk; |
3419 | ||
245f46c2 N |
3420 | /* ->takeover must set new_* and/or delta_disks |
3421 | * if it succeeds, and may set them when it fails. | |
3422 | */ | |
3423 | priv = pers->takeover(mddev); | |
3424 | if (IS_ERR(priv)) { | |
3425 | mddev->new_level = mddev->level; | |
3426 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3427 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3428 | mddev->raid_disks -= mddev->delta_disks; |
3429 | mddev->delta_disks = 0; | |
2c810cdd | 3430 | mddev->reshape_backwards = 0; |
245f46c2 N |
3431 | module_put(pers->owner); |
3432 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3433 | mdname(mddev), clevel); |
6791875e N |
3434 | rv = PTR_ERR(priv); |
3435 | goto out_unlock; | |
245f46c2 N |
3436 | } |
3437 | ||
3438 | /* Looks like we have a winner */ | |
3439 | mddev_suspend(mddev); | |
5aa61f42 | 3440 | mddev_detach(mddev); |
36d091f4 N |
3441 | |
3442 | spin_lock(&mddev->lock); | |
db721d32 N |
3443 | oldpers = mddev->pers; |
3444 | oldpriv = mddev->private; | |
3445 | mddev->pers = pers; | |
3446 | mddev->private = priv; | |
3447 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3448 | mddev->level = mddev->new_level; | |
3449 | mddev->layout = mddev->new_layout; | |
3450 | mddev->chunk_sectors = mddev->new_chunk_sectors; | |
3451 | mddev->delta_disks = 0; | |
3452 | mddev->reshape_backwards = 0; | |
3453 | mddev->degraded = 0; | |
36d091f4 | 3454 | spin_unlock(&mddev->lock); |
db721d32 N |
3455 | |
3456 | if (oldpers->sync_request == NULL && | |
3457 | mddev->external) { | |
3458 | /* We are converting from a no-redundancy array | |
3459 | * to a redundancy array and metadata is managed | |
3460 | * externally so we need to be sure that writes | |
3461 | * won't block due to a need to transition | |
3462 | * clean->dirty | |
3463 | * until external management is started. | |
3464 | */ | |
3465 | mddev->in_sync = 0; | |
3466 | mddev->safemode_delay = 0; | |
3467 | mddev->safemode = 0; | |
3468 | } | |
f72ffdd6 | 3469 | |
db721d32 N |
3470 | oldpers->free(mddev, oldpriv); |
3471 | ||
3472 | if (oldpers->sync_request == NULL && | |
a64c876f N |
3473 | pers->sync_request != NULL) { |
3474 | /* need to add the md_redundancy_group */ | |
3475 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3476 | printk(KERN_WARNING | |
3477 | "md: cannot register extra attributes for %s\n", | |
3478 | mdname(mddev)); | |
388975cc | 3479 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
f72ffdd6 | 3480 | } |
db721d32 | 3481 | if (oldpers->sync_request != NULL && |
a64c876f N |
3482 | pers->sync_request == NULL) { |
3483 | /* need to remove the md_redundancy_group */ | |
3484 | if (mddev->to_remove == NULL) | |
3485 | mddev->to_remove = &md_redundancy_group; | |
3486 | } | |
3487 | ||
dafb20fa | 3488 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3489 | if (rdev->raid_disk < 0) |
3490 | continue; | |
bf2cb0da | 3491 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3492 | rdev->new_raid_disk = -1; |
3493 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3494 | continue; | |
36fad858 | 3495 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3496 | } |
dafb20fa | 3497 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3498 | if (rdev->raid_disk < 0) |
3499 | continue; | |
3500 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3501 | continue; | |
3502 | rdev->raid_disk = rdev->new_raid_disk; | |
3503 | if (rdev->raid_disk < 0) | |
3a981b03 | 3504 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3505 | else { |
36fad858 NK |
3506 | if (sysfs_link_rdev(mddev, rdev)) |
3507 | printk(KERN_WARNING "md: cannot register rd%d" | |
3508 | " for %s after level change\n", | |
3509 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3510 | } |
e93f68a1 N |
3511 | } |
3512 | ||
db721d32 | 3513 | if (pers->sync_request == NULL) { |
9af204cf TM |
3514 | /* this is now an array without redundancy, so |
3515 | * it must always be in_sync | |
3516 | */ | |
3517 | mddev->in_sync = 1; | |
3518 | del_timer_sync(&mddev->safemode_timer); | |
3519 | } | |
02e5f5c0 | 3520 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3521 | pers->run(mddev); |
245f46c2 | 3522 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
47525e59 | 3523 | mddev_resume(mddev); |
830778a1 N |
3524 | if (!mddev->thread) |
3525 | md_update_sb(mddev, 1); | |
5cac7861 | 3526 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3527 | md_new_event(mddev); |
6791875e N |
3528 | rv = len; |
3529 | out_unlock: | |
3530 | mddev_unlock(mddev); | |
d9d166c2 N |
3531 | return rv; |
3532 | } | |
3533 | ||
3534 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3535 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3536 | |
d4dbd025 | 3537 | static ssize_t |
fd01b88c | 3538 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3539 | { |
3540 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3541 | if (mddev->reshape_position != MaxSector && |
3542 | mddev->layout != mddev->new_layout) | |
3543 | return sprintf(page, "%d (%d)\n", | |
3544 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3545 | return sprintf(page, "%d\n", mddev->layout); |
3546 | } | |
3547 | ||
3548 | static ssize_t | |
fd01b88c | 3549 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 | 3550 | { |
4c9309c0 | 3551 | unsigned int n; |
6791875e | 3552 | int err; |
d4dbd025 | 3553 | |
4c9309c0 AD |
3554 | err = kstrtouint(buf, 10, &n); |
3555 | if (err < 0) | |
3556 | return err; | |
6791875e N |
3557 | err = mddev_lock(mddev); |
3558 | if (err) | |
3559 | return err; | |
d4dbd025 | 3560 | |
b3546035 | 3561 | if (mddev->pers) { |
50ac168a | 3562 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3563 | err = -EBUSY; |
3564 | else if (mddev->ro) | |
3565 | err = -EROFS; | |
3566 | else { | |
3567 | mddev->new_layout = n; | |
3568 | err = mddev->pers->check_reshape(mddev); | |
3569 | if (err) | |
3570 | mddev->new_layout = mddev->layout; | |
597a711b | 3571 | } |
b3546035 | 3572 | } else { |
08a02ecd | 3573 | mddev->new_layout = n; |
b3546035 N |
3574 | if (mddev->reshape_position == MaxSector) |
3575 | mddev->layout = n; | |
3576 | } | |
6791875e N |
3577 | mddev_unlock(mddev); |
3578 | return err ?: len; | |
d4dbd025 N |
3579 | } |
3580 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3581 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 | 3582 | |
eae1701f | 3583 | static ssize_t |
fd01b88c | 3584 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3585 | { |
bb636547 N |
3586 | if (mddev->raid_disks == 0) |
3587 | return 0; | |
08a02ecd N |
3588 | if (mddev->reshape_position != MaxSector && |
3589 | mddev->delta_disks != 0) | |
3590 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3591 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3592 | return sprintf(page, "%d\n", mddev->raid_disks); |
3593 | } | |
3594 | ||
fd01b88c | 3595 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3596 | |
3597 | static ssize_t | |
fd01b88c | 3598 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3599 | { |
4c9309c0 | 3600 | unsigned int n; |
6791875e | 3601 | int err; |
da943b99 | 3602 | |
4c9309c0 AD |
3603 | err = kstrtouint(buf, 10, &n); |
3604 | if (err < 0) | |
3605 | return err; | |
da943b99 | 3606 | |
6791875e N |
3607 | err = mddev_lock(mddev); |
3608 | if (err) | |
3609 | return err; | |
da943b99 | 3610 | if (mddev->pers) |
6791875e | 3611 | err = update_raid_disks(mddev, n); |
08a02ecd | 3612 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3613 | struct md_rdev *rdev; |
08a02ecd | 3614 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c | 3615 | |
6791875e | 3616 | err = -EINVAL; |
c6563a8c N |
3617 | rdev_for_each(rdev, mddev) { |
3618 | if (olddisks < n && | |
3619 | rdev->data_offset < rdev->new_data_offset) | |
6791875e | 3620 | goto out_unlock; |
c6563a8c N |
3621 | if (olddisks > n && |
3622 | rdev->data_offset > rdev->new_data_offset) | |
6791875e | 3623 | goto out_unlock; |
c6563a8c | 3624 | } |
6791875e | 3625 | err = 0; |
08a02ecd N |
3626 | mddev->delta_disks = n - olddisks; |
3627 | mddev->raid_disks = n; | |
2c810cdd | 3628 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3629 | } else |
da943b99 | 3630 | mddev->raid_disks = n; |
6791875e N |
3631 | out_unlock: |
3632 | mddev_unlock(mddev); | |
3633 | return err ? err : len; | |
da943b99 N |
3634 | } |
3635 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3636 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3637 | |
3b34380a | 3638 | static ssize_t |
fd01b88c | 3639 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3640 | { |
08a02ecd | 3641 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3642 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3643 | return sprintf(page, "%d (%d)\n", | |
3644 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3645 | mddev->chunk_sectors << 9); |
3646 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3647 | } |
3648 | ||
3649 | static ssize_t | |
fd01b88c | 3650 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3651 | { |
4c9309c0 | 3652 | unsigned long n; |
6791875e | 3653 | int err; |
3b34380a | 3654 | |
4c9309c0 AD |
3655 | err = kstrtoul(buf, 10, &n); |
3656 | if (err < 0) | |
3657 | return err; | |
3b34380a | 3658 | |
6791875e N |
3659 | err = mddev_lock(mddev); |
3660 | if (err) | |
3661 | return err; | |
b3546035 | 3662 | if (mddev->pers) { |
50ac168a | 3663 | if (mddev->pers->check_reshape == NULL) |
6791875e N |
3664 | err = -EBUSY; |
3665 | else if (mddev->ro) | |
3666 | err = -EROFS; | |
3667 | else { | |
3668 | mddev->new_chunk_sectors = n >> 9; | |
3669 | err = mddev->pers->check_reshape(mddev); | |
3670 | if (err) | |
3671 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
597a711b | 3672 | } |
b3546035 | 3673 | } else { |
664e7c41 | 3674 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3675 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3676 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3677 | } |
6791875e N |
3678 | mddev_unlock(mddev); |
3679 | return err ?: len; | |
3b34380a N |
3680 | } |
3681 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3682 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3683 | |
a94213b1 | 3684 | static ssize_t |
fd01b88c | 3685 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 3686 | { |
d1a7c503 N |
3687 | if (mddev->recovery_cp == MaxSector) |
3688 | return sprintf(page, "none\n"); | |
a94213b1 N |
3689 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3690 | } | |
3691 | ||
3692 | static ssize_t | |
fd01b88c | 3693 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 3694 | { |
4c9309c0 | 3695 | unsigned long long n; |
6791875e | 3696 | int err; |
4c9309c0 AD |
3697 | |
3698 | if (cmd_match(buf, "none")) | |
3699 | n = MaxSector; | |
3700 | else { | |
3701 | err = kstrtoull(buf, 10, &n); | |
3702 | if (err < 0) | |
3703 | return err; | |
3704 | if (n != (sector_t)n) | |
3705 | return -EINVAL; | |
3706 | } | |
a94213b1 | 3707 | |
6791875e N |
3708 | err = mddev_lock(mddev); |
3709 | if (err) | |
3710 | return err; | |
b098636c | 3711 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
6791875e | 3712 | err = -EBUSY; |
a94213b1 | 3713 | |
6791875e N |
3714 | if (!err) { |
3715 | mddev->recovery_cp = n; | |
3716 | if (mddev->pers) | |
3717 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
3718 | } | |
3719 | mddev_unlock(mddev); | |
3720 | return err ?: len; | |
a94213b1 N |
3721 | } |
3722 | static struct md_sysfs_entry md_resync_start = | |
750f199e N |
3723 | __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR, |
3724 | resync_start_show, resync_start_store); | |
a94213b1 | 3725 | |
9e653b63 N |
3726 | /* |
3727 | * The array state can be: | |
3728 | * | |
3729 | * clear | |
3730 | * No devices, no size, no level | |
3731 | * Equivalent to STOP_ARRAY ioctl | |
3732 | * inactive | |
3733 | * May have some settings, but array is not active | |
3734 | * all IO results in error | |
3735 | * When written, doesn't tear down array, but just stops it | |
3736 | * suspended (not supported yet) | |
3737 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3738 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3739 | * readonly |
3740 | * no resync can happen. no superblocks get written. | |
3741 | * write requests fail | |
3742 | * read-auto | |
3743 | * like readonly, but behaves like 'clean' on a write request. | |
3744 | * | |
3745 | * clean - no pending writes, but otherwise active. | |
3746 | * When written to inactive array, starts without resync | |
3747 | * If a write request arrives then | |
3748 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3749 | * if not known, block and switch to write-pending | |
3750 | * If written to an active array that has pending writes, then fails. | |
3751 | * active | |
3752 | * fully active: IO and resync can be happening. | |
3753 | * When written to inactive array, starts with resync | |
3754 | * | |
3755 | * write-pending | |
3756 | * clean, but writes are blocked waiting for 'active' to be written. | |
3757 | * | |
3758 | * active-idle | |
3759 | * like active, but no writes have been seen for a while (100msec). | |
3760 | * | |
3761 | */ | |
3762 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3763 | write_pending, active_idle, bad_word}; | |
05381954 | 3764 | static char *array_states[] = { |
9e653b63 N |
3765 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3766 | "write-pending", "active-idle", NULL }; | |
3767 | ||
3768 | static int match_word(const char *word, char **list) | |
3769 | { | |
3770 | int n; | |
3771 | for (n=0; list[n]; n++) | |
3772 | if (cmd_match(word, list[n])) | |
3773 | break; | |
3774 | return n; | |
3775 | } | |
3776 | ||
3777 | static ssize_t | |
fd01b88c | 3778 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
3779 | { |
3780 | enum array_state st = inactive; | |
3781 | ||
3782 | if (mddev->pers) | |
3783 | switch(mddev->ro) { | |
3784 | case 1: | |
3785 | st = readonly; | |
3786 | break; | |
3787 | case 2: | |
3788 | st = read_auto; | |
3789 | break; | |
3790 | case 0: | |
3791 | if (mddev->in_sync) | |
3792 | st = clean; | |
070dc6dd | 3793 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3794 | st = write_pending; |
9e653b63 N |
3795 | else if (mddev->safemode) |
3796 | st = active_idle; | |
3797 | else | |
3798 | st = active; | |
3799 | } | |
3800 | else { | |
3801 | if (list_empty(&mddev->disks) && | |
3802 | mddev->raid_disks == 0 && | |
58c0fed4 | 3803 | mddev->dev_sectors == 0) |
9e653b63 N |
3804 | st = clear; |
3805 | else | |
3806 | st = inactive; | |
3807 | } | |
3808 | return sprintf(page, "%s\n", array_states[st]); | |
3809 | } | |
3810 | ||
f72ffdd6 N |
3811 | static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev); |
3812 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev); | |
3813 | static int do_md_run(struct mddev *mddev); | |
fd01b88c | 3814 | static int restart_array(struct mddev *mddev); |
9e653b63 N |
3815 | |
3816 | static ssize_t | |
fd01b88c | 3817 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 | 3818 | { |
6791875e | 3819 | int err; |
9e653b63 | 3820 | enum array_state st = match_word(buf, array_states); |
6791875e N |
3821 | |
3822 | if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) { | |
3823 | /* don't take reconfig_mutex when toggling between | |
3824 | * clean and active | |
3825 | */ | |
3826 | spin_lock(&mddev->lock); | |
3827 | if (st == active) { | |
3828 | restart_array(mddev); | |
3829 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); | |
3830 | wake_up(&mddev->sb_wait); | |
3831 | err = 0; | |
3832 | } else /* st == clean */ { | |
3833 | restart_array(mddev); | |
3834 | if (atomic_read(&mddev->writes_pending) == 0) { | |
3835 | if (mddev->in_sync == 0) { | |
3836 | mddev->in_sync = 1; | |
3837 | if (mddev->safemode == 1) | |
3838 | mddev->safemode = 0; | |
3839 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
3840 | } | |
3841 | err = 0; | |
3842 | } else | |
3843 | err = -EBUSY; | |
3844 | } | |
3845 | spin_unlock(&mddev->lock); | |
c008f1d3 | 3846 | return err ?: len; |
6791875e N |
3847 | } |
3848 | err = mddev_lock(mddev); | |
3849 | if (err) | |
3850 | return err; | |
3851 | err = -EINVAL; | |
9e653b63 N |
3852 | switch(st) { |
3853 | case bad_word: | |
3854 | break; | |
3855 | case clear: | |
3856 | /* stopping an active array */ | |
a05b7ea0 | 3857 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
3858 | break; |
3859 | case inactive: | |
3860 | /* stopping an active array */ | |
90cf195d | 3861 | if (mddev->pers) |
a05b7ea0 | 3862 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 3863 | else |
e691063a | 3864 | err = 0; /* already inactive */ |
9e653b63 N |
3865 | break; |
3866 | case suspended: | |
3867 | break; /* not supported yet */ | |
3868 | case readonly: | |
3869 | if (mddev->pers) | |
a05b7ea0 | 3870 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
3871 | else { |
3872 | mddev->ro = 1; | |
648b629e | 3873 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3874 | err = do_md_run(mddev); |
3875 | } | |
3876 | break; | |
3877 | case read_auto: | |
9e653b63 | 3878 | if (mddev->pers) { |
80268ee9 | 3879 | if (mddev->ro == 0) |
a05b7ea0 | 3880 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 3881 | else if (mddev->ro == 1) |
648b629e N |
3882 | err = restart_array(mddev); |
3883 | if (err == 0) { | |
3884 | mddev->ro = 2; | |
3885 | set_disk_ro(mddev->gendisk, 0); | |
3886 | } | |
9e653b63 N |
3887 | } else { |
3888 | mddev->ro = 2; | |
3889 | err = do_md_run(mddev); | |
3890 | } | |
3891 | break; | |
3892 | case clean: | |
3893 | if (mddev->pers) { | |
3894 | restart_array(mddev); | |
85572d7c | 3895 | spin_lock(&mddev->lock); |
9e653b63 | 3896 | if (atomic_read(&mddev->writes_pending) == 0) { |
e691063a N |
3897 | if (mddev->in_sync == 0) { |
3898 | mddev->in_sync = 1; | |
31a59e34 N |
3899 | if (mddev->safemode == 1) |
3900 | mddev->safemode = 0; | |
070dc6dd | 3901 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3902 | } |
3903 | err = 0; | |
3904 | } else | |
3905 | err = -EBUSY; | |
85572d7c | 3906 | spin_unlock(&mddev->lock); |
5bf29597 N |
3907 | } else |
3908 | err = -EINVAL; | |
9e653b63 N |
3909 | break; |
3910 | case active: | |
3911 | if (mddev->pers) { | |
3912 | restart_array(mddev); | |
070dc6dd | 3913 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3914 | wake_up(&mddev->sb_wait); |
3915 | err = 0; | |
3916 | } else { | |
3917 | mddev->ro = 0; | |
648b629e | 3918 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3919 | err = do_md_run(mddev); |
3920 | } | |
3921 | break; | |
3922 | case write_pending: | |
3923 | case active_idle: | |
3924 | /* these cannot be set */ | |
3925 | break; | |
3926 | } | |
6791875e N |
3927 | |
3928 | if (!err) { | |
1d23f178 N |
3929 | if (mddev->hold_active == UNTIL_IOCTL) |
3930 | mddev->hold_active = 0; | |
00bcb4ac | 3931 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 3932 | } |
6791875e N |
3933 | mddev_unlock(mddev); |
3934 | return err ?: len; | |
9e653b63 | 3935 | } |
80ca3a44 | 3936 | static struct md_sysfs_entry md_array_state = |
750f199e | 3937 | __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); |
9e653b63 | 3938 | |
1e50915f | 3939 | static ssize_t |
fd01b88c | 3940 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
3941 | return sprintf(page, "%d\n", |
3942 | atomic_read(&mddev->max_corr_read_errors)); | |
3943 | } | |
3944 | ||
3945 | static ssize_t | |
fd01b88c | 3946 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f | 3947 | { |
4c9309c0 AD |
3948 | unsigned int n; |
3949 | int rv; | |
1e50915f | 3950 | |
4c9309c0 AD |
3951 | rv = kstrtouint(buf, 10, &n); |
3952 | if (rv < 0) | |
3953 | return rv; | |
3954 | atomic_set(&mddev->max_corr_read_errors, n); | |
3955 | return len; | |
1e50915f RB |
3956 | } |
3957 | ||
3958 | static struct md_sysfs_entry max_corr_read_errors = | |
3959 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3960 | max_corrected_read_errors_store); | |
3961 | ||
6d7ff738 | 3962 | static ssize_t |
fd01b88c | 3963 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
3964 | { |
3965 | return -EINVAL; | |
3966 | } | |
3967 | ||
3968 | static ssize_t | |
fd01b88c | 3969 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
3970 | { |
3971 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3972 | /* The new device is added to the array. | |
3973 | * If the array has a persistent superblock, we read the | |
3974 | * superblock to initialise info and check validity. | |
3975 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3976 | * which mainly checks size. | |
3977 | */ | |
3978 | char *e; | |
3979 | int major = simple_strtoul(buf, &e, 10); | |
3980 | int minor; | |
3981 | dev_t dev; | |
3cb03002 | 3982 | struct md_rdev *rdev; |
6d7ff738 N |
3983 | int err; |
3984 | ||
3985 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3986 | return -EINVAL; | |
3987 | minor = simple_strtoul(e+1, &e, 10); | |
3988 | if (*e && *e != '\n') | |
3989 | return -EINVAL; | |
3990 | dev = MKDEV(major, minor); | |
3991 | if (major != MAJOR(dev) || | |
3992 | minor != MINOR(dev)) | |
3993 | return -EOVERFLOW; | |
3994 | ||
6791875e N |
3995 | flush_workqueue(md_misc_wq); |
3996 | ||
3997 | err = mddev_lock(mddev); | |
3998 | if (err) | |
3999 | return err; | |
6d7ff738 N |
4000 | if (mddev->persistent) { |
4001 | rdev = md_import_device(dev, mddev->major_version, | |
4002 | mddev->minor_version); | |
4003 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
4004 | struct md_rdev *rdev0 |
4005 | = list_entry(mddev->disks.next, | |
4006 | struct md_rdev, same_set); | |
6d7ff738 N |
4007 | err = super_types[mddev->major_version] |
4008 | .load_super(rdev, rdev0, mddev->minor_version); | |
4009 | if (err < 0) | |
4010 | goto out; | |
4011 | } | |
c5d79adb N |
4012 | } else if (mddev->external) |
4013 | rdev = md_import_device(dev, -2, -1); | |
4014 | else | |
6d7ff738 N |
4015 | rdev = md_import_device(dev, -1, -1); |
4016 | ||
4017 | if (IS_ERR(rdev)) | |
4018 | return PTR_ERR(rdev); | |
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 N |
4212 | type = "recover"; |
4213 | } | |
4214 | return sprintf(page, "%s\n", type); | |
4215 | } | |
4216 | ||
4217 | static ssize_t | |
fd01b88c | 4218 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4219 | { |
7eec314d N |
4220 | if (!mddev->pers || !mddev->pers->sync_request) |
4221 | return -EINVAL; | |
4222 | ||
b6a9ce68 N |
4223 | |
4224 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
56ccc112 N |
4225 | if (cmd_match(page, "frozen")) |
4226 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4227 | else | |
4228 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
8e8e2518 N |
4229 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
4230 | mddev_lock(mddev) == 0) { | |
4231 | flush_workqueue(md_misc_wq); | |
4232 | if (mddev->sync_thread) { | |
4233 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
6791875e | 4234 | md_reap_sync_thread(mddev); |
6791875e | 4235 | } |
8e8e2518 | 4236 | mddev_unlock(mddev); |
7eec314d | 4237 | } |
03c902e1 N |
4238 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
4239 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 4240 | return -EBUSY; |
72a23c21 | 4241 | else if (cmd_match(page, "resync")) |
56ccc112 | 4242 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4243 | else if (cmd_match(page, "recover")) { |
56ccc112 | 4244 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
72a23c21 | 4245 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
72a23c21 | 4246 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4247 | int err; |
4248 | if (mddev->pers->start_reshape == NULL) | |
4249 | return -EINVAL; | |
6791875e N |
4250 | err = mddev_lock(mddev); |
4251 | if (!err) { | |
56ccc112 | 4252 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6791875e N |
4253 | err = mddev->pers->start_reshape(mddev); |
4254 | mddev_unlock(mddev); | |
4255 | } | |
16484bf5 N |
4256 | if (err) |
4257 | return err; | |
a99ac971 | 4258 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4259 | } else { |
bce74dac | 4260 | if (cmd_match(page, "check")) |
7eec314d | 4261 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4262 | else if (!cmd_match(page, "repair")) |
7eec314d | 4263 | return -EINVAL; |
56ccc112 | 4264 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
7eec314d N |
4265 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
4266 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4267 | } |
48c26ddc N |
4268 | if (mddev->ro == 2) { |
4269 | /* A write to sync_action is enough to justify | |
4270 | * canceling read-auto mode | |
4271 | */ | |
4272 | mddev->ro = 0; | |
4273 | md_wakeup_thread(mddev->sync_thread); | |
4274 | } | |
03c902e1 | 4275 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4276 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4277 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4278 | return len; |
4279 | } | |
4280 | ||
c4a39551 | 4281 | static struct md_sysfs_entry md_scan_mode = |
750f199e | 4282 | __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); |
c4a39551 JB |
4283 | |
4284 | static ssize_t | |
4285 | last_sync_action_show(struct mddev *mddev, char *page) | |
4286 | { | |
4287 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4288 | } | |
4289 | ||
4290 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4291 | ||
9d88883e | 4292 | static ssize_t |
fd01b88c | 4293 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4294 | { |
4295 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4296 | (unsigned long long) |
4297 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4298 | } |
4299 | ||
80ca3a44 | 4300 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4301 | |
88202a0c | 4302 | static ssize_t |
fd01b88c | 4303 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4304 | { |
4305 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4306 | mddev->sync_speed_min ? "local": "system"); | |
4307 | } | |
4308 | ||
4309 | static ssize_t | |
fd01b88c | 4310 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4311 | { |
4c9309c0 AD |
4312 | unsigned int min; |
4313 | int rv; | |
4314 | ||
88202a0c | 4315 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4316 | min = 0; |
4317 | } else { | |
4318 | rv = kstrtouint(buf, 10, &min); | |
4319 | if (rv < 0) | |
4320 | return rv; | |
4321 | if (min == 0) | |
4322 | return -EINVAL; | |
88202a0c | 4323 | } |
88202a0c N |
4324 | mddev->sync_speed_min = min; |
4325 | return len; | |
4326 | } | |
4327 | ||
4328 | static struct md_sysfs_entry md_sync_min = | |
4329 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4330 | ||
4331 | static ssize_t | |
fd01b88c | 4332 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4333 | { |
4334 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4335 | mddev->sync_speed_max ? "local": "system"); | |
4336 | } | |
4337 | ||
4338 | static ssize_t | |
fd01b88c | 4339 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c | 4340 | { |
4c9309c0 AD |
4341 | unsigned int max; |
4342 | int rv; | |
4343 | ||
88202a0c | 4344 | if (strncmp(buf, "system", 6)==0) { |
4c9309c0 AD |
4345 | max = 0; |
4346 | } else { | |
4347 | rv = kstrtouint(buf, 10, &max); | |
4348 | if (rv < 0) | |
4349 | return rv; | |
4350 | if (max == 0) | |
4351 | return -EINVAL; | |
88202a0c | 4352 | } |
88202a0c N |
4353 | mddev->sync_speed_max = max; |
4354 | return len; | |
4355 | } | |
4356 | ||
4357 | static struct md_sysfs_entry md_sync_max = | |
4358 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4359 | ||
d7f3d291 | 4360 | static ssize_t |
fd01b88c | 4361 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4362 | { |
4363 | return sprintf(page, "%d\n", mddev->degraded); | |
4364 | } | |
4365 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4366 | |
90b08710 | 4367 | static ssize_t |
fd01b88c | 4368 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4369 | { |
4370 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4371 | } | |
4372 | ||
4373 | static ssize_t | |
fd01b88c | 4374 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4375 | { |
4376 | long n; | |
4377 | ||
b29bebd6 | 4378 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4379 | return -EINVAL; |
4380 | ||
4381 | if (n != 0 && n != 1) | |
4382 | return -EINVAL; | |
4383 | ||
4384 | mddev->parallel_resync = n; | |
4385 | ||
4386 | if (mddev->sync_thread) | |
4387 | wake_up(&resync_wait); | |
4388 | ||
4389 | return len; | |
4390 | } | |
4391 | ||
4392 | /* force parallel resync, even with shared block devices */ | |
4393 | static struct md_sysfs_entry md_sync_force_parallel = | |
4394 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4395 | sync_force_parallel_show, sync_force_parallel_store); | |
4396 | ||
88202a0c | 4397 | static ssize_t |
fd01b88c | 4398 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4399 | { |
4400 | unsigned long resync, dt, db; | |
d1a7c503 N |
4401 | if (mddev->curr_resync == 0) |
4402 | return sprintf(page, "none\n"); | |
9687a60c AN |
4403 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4404 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4405 | if (!dt) dt++; |
9687a60c AN |
4406 | db = resync - mddev->resync_mark_cnt; |
4407 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4408 | } |
4409 | ||
80ca3a44 | 4410 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4411 | |
4412 | static ssize_t | |
fd01b88c | 4413 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4414 | { |
13ae864b | 4415 | unsigned long long max_sectors, resync; |
88202a0c | 4416 | |
acb180b0 N |
4417 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4418 | return sprintf(page, "none\n"); | |
4419 | ||
72f36d59 N |
4420 | if (mddev->curr_resync == 1 || |
4421 | mddev->curr_resync == 2) | |
4422 | return sprintf(page, "delayed\n"); | |
4423 | ||
c804cdec N |
4424 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4425 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4426 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4427 | else |
58c0fed4 | 4428 | max_sectors = mddev->dev_sectors; |
88202a0c | 4429 | |
acb180b0 | 4430 | resync = mddev->curr_resync_completed; |
13ae864b | 4431 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4432 | } |
4433 | ||
750f199e N |
4434 | static struct md_sysfs_entry md_sync_completed = |
4435 | __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL); | |
88202a0c | 4436 | |
5e96ee65 | 4437 | static ssize_t |
fd01b88c | 4438 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4439 | { |
4440 | return sprintf(page, "%llu\n", | |
4441 | (unsigned long long)mddev->resync_min); | |
4442 | } | |
4443 | static ssize_t | |
fd01b88c | 4444 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4445 | { |
4446 | unsigned long long min; | |
23da422b | 4447 | int err; |
23da422b | 4448 | |
b29bebd6 | 4449 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 | 4450 | return -EINVAL; |
23da422b N |
4451 | |
4452 | spin_lock(&mddev->lock); | |
4453 | err = -EINVAL; | |
5e96ee65 | 4454 | if (min > mddev->resync_max) |
23da422b N |
4455 | goto out_unlock; |
4456 | ||
4457 | err = -EBUSY; | |
5e96ee65 | 4458 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4459 | goto out_unlock; |
5e96ee65 | 4460 | |
50c37b13 N |
4461 | /* Round down to multiple of 4K for safety */ |
4462 | mddev->resync_min = round_down(min, 8); | |
23da422b | 4463 | err = 0; |
5e96ee65 | 4464 | |
23da422b N |
4465 | out_unlock: |
4466 | spin_unlock(&mddev->lock); | |
4467 | return err ?: len; | |
5e96ee65 NB |
4468 | } |
4469 | ||
4470 | static struct md_sysfs_entry md_min_sync = | |
4471 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4472 | ||
c6207277 | 4473 | static ssize_t |
fd01b88c | 4474 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4475 | { |
4476 | if (mddev->resync_max == MaxSector) | |
4477 | return sprintf(page, "max\n"); | |
4478 | else | |
4479 | return sprintf(page, "%llu\n", | |
4480 | (unsigned long long)mddev->resync_max); | |
4481 | } | |
4482 | static ssize_t | |
fd01b88c | 4483 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 | 4484 | { |
23da422b N |
4485 | int err; |
4486 | spin_lock(&mddev->lock); | |
c6207277 N |
4487 | if (strncmp(buf, "max", 3) == 0) |
4488 | mddev->resync_max = MaxSector; | |
4489 | else { | |
5e96ee65 | 4490 | unsigned long long max; |
23da422b N |
4491 | int chunk; |
4492 | ||
4493 | err = -EINVAL; | |
b29bebd6 | 4494 | if (kstrtoull(buf, 10, &max)) |
23da422b | 4495 | goto out_unlock; |
5e96ee65 | 4496 | if (max < mddev->resync_min) |
23da422b N |
4497 | goto out_unlock; |
4498 | ||
4499 | err = -EBUSY; | |
c6207277 | 4500 | if (max < mddev->resync_max && |
4d484a4a | 4501 | mddev->ro == 0 && |
c6207277 | 4502 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
23da422b | 4503 | goto out_unlock; |
c6207277 N |
4504 | |
4505 | /* Must be a multiple of chunk_size */ | |
23da422b N |
4506 | chunk = mddev->chunk_sectors; |
4507 | if (chunk) { | |
2ac06c33 | 4508 | sector_t temp = max; |
23da422b N |
4509 | |
4510 | err = -EINVAL; | |
4511 | if (sector_div(temp, chunk)) | |
4512 | goto out_unlock; | |
c6207277 N |
4513 | } |
4514 | mddev->resync_max = max; | |
4515 | } | |
4516 | wake_up(&mddev->recovery_wait); | |
23da422b N |
4517 | err = 0; |
4518 | out_unlock: | |
4519 | spin_unlock(&mddev->lock); | |
4520 | return err ?: len; | |
c6207277 N |
4521 | } |
4522 | ||
4523 | static struct md_sysfs_entry md_max_sync = | |
4524 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4525 | ||
e464eafd | 4526 | static ssize_t |
fd01b88c | 4527 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4528 | { |
4529 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4530 | } | |
4531 | ||
4532 | static ssize_t | |
fd01b88c | 4533 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4534 | { |
4c9309c0 | 4535 | unsigned long long old, new; |
6791875e | 4536 | int err; |
e464eafd | 4537 | |
4c9309c0 AD |
4538 | err = kstrtoull(buf, 10, &new); |
4539 | if (err < 0) | |
4540 | return err; | |
4541 | if (new != (sector_t)new) | |
e464eafd | 4542 | return -EINVAL; |
23ddff37 | 4543 | |
6791875e N |
4544 | err = mddev_lock(mddev); |
4545 | if (err) | |
4546 | return err; | |
4547 | err = -EINVAL; | |
4548 | if (mddev->pers == NULL || | |
4549 | mddev->pers->quiesce == NULL) | |
4550 | goto unlock; | |
4551 | old = mddev->suspend_lo; | |
23ddff37 N |
4552 | mddev->suspend_lo = new; |
4553 | if (new >= old) | |
4554 | /* Shrinking suspended region */ | |
e464eafd | 4555 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4556 | else { |
4557 | /* Expanding suspended region - need to wait */ | |
4558 | mddev->pers->quiesce(mddev, 1); | |
4559 | mddev->pers->quiesce(mddev, 0); | |
4560 | } | |
6791875e N |
4561 | err = 0; |
4562 | unlock: | |
4563 | mddev_unlock(mddev); | |
4564 | return err ?: len; | |
e464eafd N |
4565 | } |
4566 | static struct md_sysfs_entry md_suspend_lo = | |
4567 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4568 | ||
e464eafd | 4569 | static ssize_t |
fd01b88c | 4570 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4571 | { |
4572 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4573 | } | |
4574 | ||
4575 | static ssize_t | |
fd01b88c | 4576 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd | 4577 | { |
4c9309c0 | 4578 | unsigned long long old, new; |
6791875e | 4579 | int err; |
e464eafd | 4580 | |
4c9309c0 AD |
4581 | err = kstrtoull(buf, 10, &new); |
4582 | if (err < 0) | |
4583 | return err; | |
4584 | if (new != (sector_t)new) | |
e464eafd | 4585 | return -EINVAL; |
23ddff37 | 4586 | |
6791875e N |
4587 | err = mddev_lock(mddev); |
4588 | if (err) | |
4589 | return err; | |
4590 | err = -EINVAL; | |
4591 | if (mddev->pers == NULL || | |
4592 | mddev->pers->quiesce == NULL) | |
4593 | goto unlock; | |
4594 | old = mddev->suspend_hi; | |
23ddff37 N |
4595 | mddev->suspend_hi = new; |
4596 | if (new <= old) | |
4597 | /* Shrinking suspended region */ | |
4598 | mddev->pers->quiesce(mddev, 2); | |
4599 | else { | |
4600 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4601 | mddev->pers->quiesce(mddev, 1); |
4602 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 | 4603 | } |
6791875e N |
4604 | err = 0; |
4605 | unlock: | |
4606 | mddev_unlock(mddev); | |
4607 | return err ?: len; | |
e464eafd N |
4608 | } |
4609 | static struct md_sysfs_entry md_suspend_hi = | |
4610 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4611 | ||
08a02ecd | 4612 | static ssize_t |
fd01b88c | 4613 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4614 | { |
4615 | if (mddev->reshape_position != MaxSector) | |
4616 | return sprintf(page, "%llu\n", | |
4617 | (unsigned long long)mddev->reshape_position); | |
4618 | strcpy(page, "none\n"); | |
4619 | return 5; | |
4620 | } | |
4621 | ||
4622 | static ssize_t | |
fd01b88c | 4623 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4624 | { |
c6563a8c | 4625 | struct md_rdev *rdev; |
4c9309c0 | 4626 | unsigned long long new; |
6791875e | 4627 | int err; |
6791875e | 4628 | |
4c9309c0 AD |
4629 | err = kstrtoull(buf, 10, &new); |
4630 | if (err < 0) | |
4631 | return err; | |
4632 | if (new != (sector_t)new) | |
08a02ecd | 4633 | return -EINVAL; |
6791875e N |
4634 | err = mddev_lock(mddev); |
4635 | if (err) | |
4636 | return err; | |
4637 | err = -EBUSY; | |
4638 | if (mddev->pers) | |
4639 | goto unlock; | |
08a02ecd N |
4640 | mddev->reshape_position = new; |
4641 | mddev->delta_disks = 0; | |
2c810cdd | 4642 | mddev->reshape_backwards = 0; |
08a02ecd N |
4643 | mddev->new_level = mddev->level; |
4644 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4645 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4646 | rdev_for_each(rdev, mddev) |
4647 | rdev->new_data_offset = rdev->data_offset; | |
6791875e N |
4648 | err = 0; |
4649 | unlock: | |
4650 | mddev_unlock(mddev); | |
4651 | return err ?: len; | |
08a02ecd N |
4652 | } |
4653 | ||
4654 | static struct md_sysfs_entry md_reshape_position = | |
4655 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4656 | reshape_position_store); | |
4657 | ||
2c810cdd N |
4658 | static ssize_t |
4659 | reshape_direction_show(struct mddev *mddev, char *page) | |
4660 | { | |
4661 | return sprintf(page, "%s\n", | |
4662 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4663 | } | |
4664 | ||
4665 | static ssize_t | |
4666 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4667 | { | |
4668 | int backwards = 0; | |
6791875e N |
4669 | int err; |
4670 | ||
2c810cdd N |
4671 | if (cmd_match(buf, "forwards")) |
4672 | backwards = 0; | |
4673 | else if (cmd_match(buf, "backwards")) | |
4674 | backwards = 1; | |
4675 | else | |
4676 | return -EINVAL; | |
4677 | if (mddev->reshape_backwards == backwards) | |
4678 | return len; | |
4679 | ||
6791875e N |
4680 | err = mddev_lock(mddev); |
4681 | if (err) | |
4682 | return err; | |
2c810cdd N |
4683 | /* check if we are allowed to change */ |
4684 | if (mddev->delta_disks) | |
6791875e N |
4685 | err = -EBUSY; |
4686 | else if (mddev->persistent && | |
2c810cdd | 4687 | mddev->major_version == 0) |
6791875e N |
4688 | err = -EINVAL; |
4689 | else | |
4690 | mddev->reshape_backwards = backwards; | |
4691 | mddev_unlock(mddev); | |
4692 | return err ?: len; | |
2c810cdd N |
4693 | } |
4694 | ||
4695 | static struct md_sysfs_entry md_reshape_direction = | |
4696 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
4697 | reshape_direction_store); | |
4698 | ||
b522adcd | 4699 | static ssize_t |
fd01b88c | 4700 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
4701 | { |
4702 | if (mddev->external_size) | |
4703 | return sprintf(page, "%llu\n", | |
4704 | (unsigned long long)mddev->array_sectors/2); | |
4705 | else | |
4706 | return sprintf(page, "default\n"); | |
4707 | } | |
4708 | ||
4709 | static ssize_t | |
fd01b88c | 4710 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
4711 | { |
4712 | sector_t sectors; | |
6791875e N |
4713 | int err; |
4714 | ||
4715 | err = mddev_lock(mddev); | |
4716 | if (err) | |
4717 | return err; | |
b522adcd DW |
4718 | |
4719 | if (strncmp(buf, "default", 7) == 0) { | |
4720 | if (mddev->pers) | |
4721 | sectors = mddev->pers->size(mddev, 0, 0); | |
4722 | else | |
4723 | sectors = mddev->array_sectors; | |
4724 | ||
4725 | mddev->external_size = 0; | |
4726 | } else { | |
4727 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
6791875e N |
4728 | err = -EINVAL; |
4729 | else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
4730 | err = -E2BIG; | |
4731 | else | |
4732 | mddev->external_size = 1; | |
b522adcd DW |
4733 | } |
4734 | ||
6791875e N |
4735 | if (!err) { |
4736 | mddev->array_sectors = sectors; | |
4737 | if (mddev->pers) { | |
4738 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4739 | revalidate_disk(mddev->gendisk); | |
4740 | } | |
cbe6ef1d | 4741 | } |
6791875e N |
4742 | mddev_unlock(mddev); |
4743 | return err ?: len; | |
b522adcd DW |
4744 | } |
4745 | ||
4746 | static struct md_sysfs_entry md_array_size = | |
4747 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4748 | array_size_store); | |
e464eafd | 4749 | |
eae1701f N |
4750 | static struct attribute *md_default_attrs[] = { |
4751 | &md_level.attr, | |
d4dbd025 | 4752 | &md_layout.attr, |
eae1701f | 4753 | &md_raid_disks.attr, |
3b34380a | 4754 | &md_chunk_size.attr, |
a35b0d69 | 4755 | &md_size.attr, |
a94213b1 | 4756 | &md_resync_start.attr, |
8bb93aac | 4757 | &md_metadata.attr, |
6d7ff738 | 4758 | &md_new_device.attr, |
16f17b39 | 4759 | &md_safe_delay.attr, |
9e653b63 | 4760 | &md_array_state.attr, |
08a02ecd | 4761 | &md_reshape_position.attr, |
2c810cdd | 4762 | &md_reshape_direction.attr, |
b522adcd | 4763 | &md_array_size.attr, |
1e50915f | 4764 | &max_corr_read_errors.attr, |
411036fa N |
4765 | NULL, |
4766 | }; | |
4767 | ||
4768 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4769 | &md_scan_mode.attr, |
c4a39551 | 4770 | &md_last_scan_mode.attr, |
9d88883e | 4771 | &md_mismatches.attr, |
88202a0c N |
4772 | &md_sync_min.attr, |
4773 | &md_sync_max.attr, | |
4774 | &md_sync_speed.attr, | |
90b08710 | 4775 | &md_sync_force_parallel.attr, |
88202a0c | 4776 | &md_sync_completed.attr, |
5e96ee65 | 4777 | &md_min_sync.attr, |
c6207277 | 4778 | &md_max_sync.attr, |
e464eafd N |
4779 | &md_suspend_lo.attr, |
4780 | &md_suspend_hi.attr, | |
9b1d1dac | 4781 | &md_bitmap.attr, |
d7f3d291 | 4782 | &md_degraded.attr, |
eae1701f N |
4783 | NULL, |
4784 | }; | |
411036fa N |
4785 | static struct attribute_group md_redundancy_group = { |
4786 | .name = NULL, | |
4787 | .attrs = md_redundancy_attrs, | |
4788 | }; | |
4789 | ||
eae1701f N |
4790 | static ssize_t |
4791 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4792 | { | |
4793 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4794 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4795 | ssize_t rv; |
eae1701f N |
4796 | |
4797 | if (!entry->show) | |
4798 | return -EIO; | |
af8a2434 N |
4799 | spin_lock(&all_mddevs_lock); |
4800 | if (list_empty(&mddev->all_mddevs)) { | |
4801 | spin_unlock(&all_mddevs_lock); | |
4802 | return -EBUSY; | |
4803 | } | |
4804 | mddev_get(mddev); | |
4805 | spin_unlock(&all_mddevs_lock); | |
4806 | ||
b7b17c9b | 4807 | rv = entry->show(mddev, page); |
af8a2434 | 4808 | mddev_put(mddev); |
96de1e66 | 4809 | return rv; |
eae1701f N |
4810 | } |
4811 | ||
4812 | static ssize_t | |
4813 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4814 | const char *page, size_t length) | |
4815 | { | |
4816 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4817 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4818 | ssize_t rv; |
eae1701f N |
4819 | |
4820 | if (!entry->store) | |
4821 | return -EIO; | |
67463acb N |
4822 | if (!capable(CAP_SYS_ADMIN)) |
4823 | return -EACCES; | |
af8a2434 N |
4824 | spin_lock(&all_mddevs_lock); |
4825 | if (list_empty(&mddev->all_mddevs)) { | |
4826 | spin_unlock(&all_mddevs_lock); | |
4827 | return -EBUSY; | |
4828 | } | |
4829 | mddev_get(mddev); | |
4830 | spin_unlock(&all_mddevs_lock); | |
6791875e | 4831 | rv = entry->store(mddev, page, length); |
af8a2434 | 4832 | mddev_put(mddev); |
96de1e66 | 4833 | return rv; |
eae1701f N |
4834 | } |
4835 | ||
4836 | static void md_free(struct kobject *ko) | |
4837 | { | |
fd01b88c | 4838 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
4839 | |
4840 | if (mddev->sysfs_state) | |
4841 | sysfs_put(mddev->sysfs_state); | |
4842 | ||
6cd18e71 N |
4843 | if (mddev->queue) |
4844 | blk_cleanup_queue(mddev->queue); | |
a21d1504 N |
4845 | if (mddev->gendisk) { |
4846 | del_gendisk(mddev->gendisk); | |
4847 | put_disk(mddev->gendisk); | |
4848 | } | |
a21d1504 | 4849 | |
eae1701f N |
4850 | kfree(mddev); |
4851 | } | |
4852 | ||
52cf25d0 | 4853 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4854 | .show = md_attr_show, |
4855 | .store = md_attr_store, | |
4856 | }; | |
4857 | static struct kobj_type md_ktype = { | |
4858 | .release = md_free, | |
4859 | .sysfs_ops = &md_sysfs_ops, | |
4860 | .default_attrs = md_default_attrs, | |
4861 | }; | |
4862 | ||
1da177e4 LT |
4863 | int mdp_major = 0; |
4864 | ||
5fd3a17e DW |
4865 | static void mddev_delayed_delete(struct work_struct *ws) |
4866 | { | |
fd01b88c | 4867 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 4868 | |
43a70507 | 4869 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4870 | kobject_del(&mddev->kobj); |
4871 | kobject_put(&mddev->kobj); | |
4872 | } | |
4873 | ||
efeb53c0 | 4874 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4875 | { |
48c9c27b | 4876 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 4877 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 4878 | struct gendisk *disk; |
efeb53c0 N |
4879 | int partitioned; |
4880 | int shift; | |
4881 | int unit; | |
3830c62f | 4882 | int error; |
1da177e4 LT |
4883 | |
4884 | if (!mddev) | |
efeb53c0 N |
4885 | return -ENODEV; |
4886 | ||
4887 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4888 | shift = partitioned ? MdpMinorShift : 0; | |
4889 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4890 | |
e804ac78 TH |
4891 | /* wait for any previous instance of this device to be |
4892 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4893 | */ |
e804ac78 | 4894 | flush_workqueue(md_misc_wq); |
d3374825 | 4895 | |
48c9c27b | 4896 | mutex_lock(&disks_mutex); |
0909dc44 N |
4897 | error = -EEXIST; |
4898 | if (mddev->gendisk) | |
4899 | goto abort; | |
efeb53c0 N |
4900 | |
4901 | if (name) { | |
4902 | /* Need to ensure that 'name' is not a duplicate. | |
4903 | */ | |
fd01b88c | 4904 | struct mddev *mddev2; |
efeb53c0 N |
4905 | spin_lock(&all_mddevs_lock); |
4906 | ||
4907 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4908 | if (mddev2->gendisk && | |
4909 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4910 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4911 | goto abort; |
efeb53c0 N |
4912 | } |
4913 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4914 | } |
8b765398 | 4915 | |
0909dc44 | 4916 | error = -ENOMEM; |
8b765398 | 4917 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4918 | if (!mddev->queue) |
4919 | goto abort; | |
409c57f3 N |
4920 | mddev->queue->queuedata = mddev; |
4921 | ||
409c57f3 | 4922 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 4923 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 4924 | |
1da177e4 LT |
4925 | disk = alloc_disk(1 << shift); |
4926 | if (!disk) { | |
8b765398 N |
4927 | blk_cleanup_queue(mddev->queue); |
4928 | mddev->queue = NULL; | |
0909dc44 | 4929 | goto abort; |
1da177e4 | 4930 | } |
efeb53c0 | 4931 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4932 | disk->first_minor = unit << shift; |
efeb53c0 N |
4933 | if (name) |
4934 | strcpy(disk->disk_name, name); | |
4935 | else if (partitioned) | |
1da177e4 | 4936 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4937 | else |
1da177e4 | 4938 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4939 | disk->fops = &md_fops; |
4940 | disk->private_data = mddev; | |
4941 | disk->queue = mddev->queue; | |
b0140891 | 4942 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); |
92850bbd | 4943 | /* Allow extended partitions. This makes the |
d3374825 | 4944 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4945 | * remove it now. |
4946 | */ | |
4947 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 4948 | mddev->gendisk = disk; |
b0140891 N |
4949 | /* As soon as we call add_disk(), another thread could get |
4950 | * through to md_open, so make sure it doesn't get too far | |
4951 | */ | |
4952 | mutex_lock(&mddev->open_mutex); | |
4953 | add_disk(disk); | |
4954 | ||
ed9e1982 TH |
4955 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4956 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4957 | if (error) { |
4958 | /* This isn't possible, but as kobject_init_and_add is marked | |
4959 | * __must_check, we must do something with the result | |
4960 | */ | |
5e55e2f5 N |
4961 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4962 | disk->disk_name); | |
0909dc44 N |
4963 | error = 0; |
4964 | } | |
00bcb4ac N |
4965 | if (mddev->kobj.sd && |
4966 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4967 | printk(KERN_DEBUG "pointless warning\n"); |
b0140891 | 4968 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
4969 | abort: |
4970 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4971 | if (!error && mddev->kobj.sd) { |
3830c62f | 4972 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4973 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4974 | } |
d3374825 | 4975 | mddev_put(mddev); |
0909dc44 | 4976 | return error; |
efeb53c0 N |
4977 | } |
4978 | ||
4979 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4980 | { | |
4981 | md_alloc(dev, NULL); | |
1da177e4 LT |
4982 | return NULL; |
4983 | } | |
4984 | ||
efeb53c0 N |
4985 | static int add_named_array(const char *val, struct kernel_param *kp) |
4986 | { | |
4987 | /* val must be "md_*" where * is not all digits. | |
4988 | * We allocate an array with a large free minor number, and | |
4989 | * set the name to val. val must not already be an active name. | |
4990 | */ | |
4991 | int len = strlen(val); | |
4992 | char buf[DISK_NAME_LEN]; | |
4993 | ||
4994 | while (len && val[len-1] == '\n') | |
4995 | len--; | |
4996 | if (len >= DISK_NAME_LEN) | |
4997 | return -E2BIG; | |
4998 | strlcpy(buf, val, len+1); | |
4999 | if (strncmp(buf, "md_", 3) != 0) | |
5000 | return -EINVAL; | |
5001 | return md_alloc(0, buf); | |
5002 | } | |
5003 | ||
1da177e4 LT |
5004 | static void md_safemode_timeout(unsigned long data) |
5005 | { | |
fd01b88c | 5006 | struct mddev *mddev = (struct mddev *) data; |
1da177e4 | 5007 | |
0fd62b86 NB |
5008 | if (!atomic_read(&mddev->writes_pending)) { |
5009 | mddev->safemode = 1; | |
5010 | if (mddev->external) | |
00bcb4ac | 5011 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 5012 | } |
1da177e4 LT |
5013 | md_wakeup_thread(mddev->thread); |
5014 | } | |
5015 | ||
6ff8d8ec | 5016 | static int start_dirty_degraded; |
1da177e4 | 5017 | |
fd01b88c | 5018 | int md_run(struct mddev *mddev) |
1da177e4 | 5019 | { |
2604b703 | 5020 | int err; |
3cb03002 | 5021 | struct md_rdev *rdev; |
84fc4b56 | 5022 | struct md_personality *pers; |
1da177e4 | 5023 | |
a757e64c N |
5024 | if (list_empty(&mddev->disks)) |
5025 | /* cannot run an array with no devices.. */ | |
1da177e4 | 5026 | return -EINVAL; |
1da177e4 LT |
5027 | |
5028 | if (mddev->pers) | |
5029 | return -EBUSY; | |
bb4f1e9d N |
5030 | /* Cannot run until previous stop completes properly */ |
5031 | if (mddev->sysfs_active) | |
5032 | return -EBUSY; | |
b6eb127d | 5033 | |
1da177e4 LT |
5034 | /* |
5035 | * Analyze all RAID superblock(s) | |
5036 | */ | |
1ec4a939 N |
5037 | if (!mddev->raid_disks) { |
5038 | if (!mddev->persistent) | |
5039 | return -EINVAL; | |
a757e64c | 5040 | analyze_sbs(mddev); |
1ec4a939 | 5041 | } |
1da177e4 | 5042 | |
d9d166c2 N |
5043 | if (mddev->level != LEVEL_NONE) |
5044 | request_module("md-level-%d", mddev->level); | |
5045 | else if (mddev->clevel[0]) | |
5046 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
5047 | |
5048 | /* | |
5049 | * Drop all container device buffers, from now on | |
5050 | * the only valid external interface is through the md | |
5051 | * device. | |
1da177e4 | 5052 | */ |
dafb20fa | 5053 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 5054 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5055 | continue; |
5056 | sync_blockdev(rdev->bdev); | |
f98393a6 | 5057 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
5058 | |
5059 | /* perform some consistency tests on the device. | |
5060 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 5061 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 5062 | */ |
a6ff7e08 JB |
5063 | if (rdev->meta_bdev) { |
5064 | /* Nothing to check */; | |
5065 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
5066 | if (mddev->dev_sectors && |
5067 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 5068 | > rdev->sb_start) { |
f0d76d70 N |
5069 | printk("md: %s: data overlaps metadata\n", |
5070 | mdname(mddev)); | |
5071 | return -EINVAL; | |
5072 | } | |
5073 | } else { | |
0f420358 | 5074 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
5075 | > rdev->data_offset) { |
5076 | printk("md: %s: metadata overlaps data\n", | |
5077 | mdname(mddev)); | |
5078 | return -EINVAL; | |
5079 | } | |
5080 | } | |
00bcb4ac | 5081 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
5082 | } |
5083 | ||
a167f663 | 5084 | if (mddev->bio_set == NULL) |
395c72a7 | 5085 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0); |
a167f663 | 5086 | |
1da177e4 | 5087 | spin_lock(&pers_lock); |
d9d166c2 | 5088 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 5089 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 5090 | spin_unlock(&pers_lock); |
d9d166c2 N |
5091 | if (mddev->level != LEVEL_NONE) |
5092 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
5093 | mddev->level); | |
5094 | else | |
5095 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
5096 | mddev->clevel); | |
1da177e4 LT |
5097 | return -EINVAL; |
5098 | } | |
1da177e4 | 5099 | spin_unlock(&pers_lock); |
34817e8c N |
5100 | if (mddev->level != pers->level) { |
5101 | mddev->level = pers->level; | |
5102 | mddev->new_level = pers->level; | |
5103 | } | |
d9d166c2 | 5104 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 5105 | |
f6705578 | 5106 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 5107 | pers->start_reshape == NULL) { |
f6705578 | 5108 | /* This personality cannot handle reshaping... */ |
f6705578 N |
5109 | module_put(pers->owner); |
5110 | return -EINVAL; | |
5111 | } | |
5112 | ||
7dd5e7c3 N |
5113 | if (pers->sync_request) { |
5114 | /* Warn if this is a potentially silly | |
5115 | * configuration. | |
5116 | */ | |
5117 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5118 | struct md_rdev *rdev2; |
7dd5e7c3 | 5119 | int warned = 0; |
159ec1fc | 5120 | |
dafb20fa N |
5121 | rdev_for_each(rdev, mddev) |
5122 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
5123 | if (rdev < rdev2 && |
5124 | rdev->bdev->bd_contains == | |
5125 | rdev2->bdev->bd_contains) { | |
5126 | printk(KERN_WARNING | |
5127 | "%s: WARNING: %s appears to be" | |
5128 | " on the same physical disk as" | |
5129 | " %s.\n", | |
5130 | mdname(mddev), | |
5131 | bdevname(rdev->bdev,b), | |
5132 | bdevname(rdev2->bdev,b2)); | |
5133 | warned = 1; | |
5134 | } | |
5135 | } | |
159ec1fc | 5136 | |
7dd5e7c3 N |
5137 | if (warned) |
5138 | printk(KERN_WARNING | |
5139 | "True protection against single-disk" | |
5140 | " failure might be compromised.\n"); | |
5141 | } | |
5142 | ||
657390d2 | 5143 | mddev->recovery = 0; |
58c0fed4 AN |
5144 | /* may be over-ridden by personality */ |
5145 | mddev->resync_max_sectors = mddev->dev_sectors; | |
5146 | ||
6ff8d8ec | 5147 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 5148 | |
0f9552b5 | 5149 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
5150 | mddev->ro = 2; /* read-only, but switch on first write */ |
5151 | ||
36d091f4 | 5152 | err = pers->run(mddev); |
13e53df3 AN |
5153 | if (err) |
5154 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
36d091f4 | 5155 | else if (pers->size(mddev, 0, 0) < mddev->array_sectors) { |
b522adcd DW |
5156 | WARN_ONCE(!mddev->external_size, "%s: default size too small," |
5157 | " but 'external_size' not in effect?\n", __func__); | |
5158 | printk(KERN_ERR | |
5159 | "md: invalid array_size %llu > default size %llu\n", | |
5160 | (unsigned long long)mddev->array_sectors / 2, | |
36d091f4 | 5161 | (unsigned long long)pers->size(mddev, 0, 0) / 2); |
b522adcd | 5162 | err = -EINVAL; |
b522adcd | 5163 | } |
36d091f4 | 5164 | if (err == 0 && pers->sync_request && |
ef99bf48 | 5165 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { |
f9209a32 GR |
5166 | struct bitmap *bitmap; |
5167 | ||
5168 | bitmap = bitmap_create(mddev, -1); | |
5169 | if (IS_ERR(bitmap)) { | |
5170 | err = PTR_ERR(bitmap); | |
b15c2e57 N |
5171 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", |
5172 | mdname(mddev), err); | |
f9209a32 GR |
5173 | } else |
5174 | mddev->bitmap = bitmap; | |
5175 | ||
b15c2e57 | 5176 | } |
1da177e4 | 5177 | if (err) { |
5aa61f42 | 5178 | mddev_detach(mddev); |
0c35bd47 N |
5179 | if (mddev->private) |
5180 | pers->free(mddev, mddev->private); | |
36d091f4 | 5181 | module_put(pers->owner); |
32a7627c N |
5182 | bitmap_destroy(mddev); |
5183 | return err; | |
1da177e4 | 5184 | } |
5c675f83 N |
5185 | if (mddev->queue) { |
5186 | mddev->queue->backing_dev_info.congested_data = mddev; | |
5187 | mddev->queue->backing_dev_info.congested_fn = md_congested; | |
64590f45 | 5188 | blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec); |
5c675f83 | 5189 | } |
36d091f4 | 5190 | if (pers->sync_request) { |
00bcb4ac N |
5191 | if (mddev->kobj.sd && |
5192 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
5193 | printk(KERN_WARNING |
5194 | "md: cannot register extra attributes for %s\n", | |
5195 | mdname(mddev)); | |
00bcb4ac | 5196 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 5197 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
5198 | mddev->ro = 0; |
5199 | ||
f72ffdd6 | 5200 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
5201 | atomic_set(&mddev->max_corr_read_errors, |
5202 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
5203 | mddev->safemode = 0; |
5204 | mddev->safemode_timer.function = md_safemode_timeout; | |
5205 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 5206 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 5207 | mddev->in_sync = 1; |
0ca69886 | 5208 | smp_wmb(); |
36d091f4 N |
5209 | spin_lock(&mddev->lock); |
5210 | mddev->pers = pers; | |
0ca69886 | 5211 | mddev->ready = 1; |
36d091f4 | 5212 | spin_unlock(&mddev->lock); |
dafb20fa | 5213 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5214 | if (rdev->raid_disk >= 0) |
5215 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 5216 | /* failure here is OK */; |
f72ffdd6 | 5217 | |
1da177e4 | 5218 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f72ffdd6 | 5219 | |
7a0a5355 | 5220 | if (mddev->flags & MD_UPDATE_SB_FLAGS) |
850b2b42 | 5221 | md_update_sb(mddev, 0); |
1da177e4 | 5222 | |
d7603b7e | 5223 | md_new_event(mddev); |
00bcb4ac N |
5224 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
5225 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 5226 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
5227 | return 0; |
5228 | } | |
390ee602 | 5229 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 5230 | |
fd01b88c | 5231 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
5232 | { |
5233 | int err; | |
5234 | ||
5235 | err = md_run(mddev); | |
5236 | if (err) | |
5237 | goto out; | |
69e51b44 N |
5238 | err = bitmap_load(mddev); |
5239 | if (err) { | |
5240 | bitmap_destroy(mddev); | |
5241 | goto out; | |
5242 | } | |
0fd018af JB |
5243 | |
5244 | md_wakeup_thread(mddev->thread); | |
5245 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
5246 | ||
fe60b014 N |
5247 | set_capacity(mddev->gendisk, mddev->array_sectors); |
5248 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 5249 | mddev->changed = 1; |
fe60b014 N |
5250 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
5251 | out: | |
5252 | return err; | |
5253 | } | |
5254 | ||
fd01b88c | 5255 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
5256 | { |
5257 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 5258 | |
80fab1d7 | 5259 | /* Complain if it has no devices */ |
1da177e4 | 5260 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
5261 | return -ENXIO; |
5262 | if (!mddev->pers) | |
5263 | return -EINVAL; | |
5264 | if (!mddev->ro) | |
5265 | return -EBUSY; | |
5266 | mddev->safemode = 0; | |
5267 | mddev->ro = 0; | |
5268 | set_disk_ro(disk, 0); | |
5269 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
5270 | mdname(mddev)); | |
5271 | /* Kick recovery or resync if necessary */ | |
5272 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5273 | md_wakeup_thread(mddev->thread); | |
5274 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 5275 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 5276 | return 0; |
1da177e4 LT |
5277 | } |
5278 | ||
fd01b88c | 5279 | static void md_clean(struct mddev *mddev) |
6177b472 N |
5280 | { |
5281 | mddev->array_sectors = 0; | |
5282 | mddev->external_size = 0; | |
5283 | mddev->dev_sectors = 0; | |
5284 | mddev->raid_disks = 0; | |
5285 | mddev->recovery_cp = 0; | |
5286 | mddev->resync_min = 0; | |
5287 | mddev->resync_max = MaxSector; | |
5288 | mddev->reshape_position = MaxSector; | |
5289 | mddev->external = 0; | |
5290 | mddev->persistent = 0; | |
5291 | mddev->level = LEVEL_NONE; | |
5292 | mddev->clevel[0] = 0; | |
5293 | mddev->flags = 0; | |
5294 | mddev->ro = 0; | |
5295 | mddev->metadata_type[0] = 0; | |
5296 | mddev->chunk_sectors = 0; | |
5297 | mddev->ctime = mddev->utime = 0; | |
5298 | mddev->layout = 0; | |
5299 | mddev->max_disks = 0; | |
5300 | mddev->events = 0; | |
a8707c08 | 5301 | mddev->can_decrease_events = 0; |
6177b472 | 5302 | mddev->delta_disks = 0; |
2c810cdd | 5303 | mddev->reshape_backwards = 0; |
6177b472 N |
5304 | mddev->new_level = LEVEL_NONE; |
5305 | mddev->new_layout = 0; | |
5306 | mddev->new_chunk_sectors = 0; | |
5307 | mddev->curr_resync = 0; | |
7f7583d4 | 5308 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5309 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5310 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5311 | mddev->recovery = 0; | |
5312 | mddev->in_sync = 0; | |
f0b4f7e2 | 5313 | mddev->changed = 0; |
6177b472 | 5314 | mddev->degraded = 0; |
6177b472 | 5315 | mddev->safemode = 0; |
050b6615 | 5316 | mddev->merge_check_needed = 0; |
6177b472 N |
5317 | mddev->bitmap_info.offset = 0; |
5318 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5319 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5320 | mddev->bitmap_info.chunksize = 0; |
5321 | mddev->bitmap_info.daemon_sleep = 0; | |
5322 | mddev->bitmap_info.max_write_behind = 0; | |
5323 | } | |
5324 | ||
fd01b88c | 5325 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5326 | { |
293467aa GR |
5327 | if (mddev_is_clustered(mddev)) |
5328 | md_cluster_ops->metadata_update_start(mddev); | |
6b6204ee | 5329 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
f851b60d | 5330 | flush_workqueue(md_misc_wq); |
a047e125 | 5331 | if (mddev->sync_thread) { |
a047e125 | 5332 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5333 | md_reap_sync_thread(mddev); |
a047e125 N |
5334 | } |
5335 | ||
5336 | del_timer_sync(&mddev->safemode_timer); | |
5337 | ||
5338 | bitmap_flush(mddev); | |
5339 | md_super_wait(mddev); | |
5340 | ||
b6d428c6 | 5341 | if (mddev->ro == 0 && |
7a0a5355 | 5342 | (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) { |
a047e125 N |
5343 | /* mark array as shutdown cleanly */ |
5344 | mddev->in_sync = 1; | |
5345 | md_update_sb(mddev, 1); | |
5346 | } | |
293467aa GR |
5347 | if (mddev_is_clustered(mddev)) |
5348 | md_cluster_ops->metadata_update_finish(mddev); | |
a047e125 | 5349 | } |
defad61a | 5350 | |
fd01b88c | 5351 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5352 | { |
29f097c4 | 5353 | mddev_lock_nointr(mddev); |
defad61a N |
5354 | __md_stop_writes(mddev); |
5355 | mddev_unlock(mddev); | |
5356 | } | |
390ee602 | 5357 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5358 | |
5aa61f42 N |
5359 | static void mddev_detach(struct mddev *mddev) |
5360 | { | |
5361 | struct bitmap *bitmap = mddev->bitmap; | |
5362 | /* wait for behind writes to complete */ | |
5363 | if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { | |
5364 | printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n", | |
5365 | mdname(mddev)); | |
5366 | /* need to kick something here to make sure I/O goes? */ | |
5367 | wait_event(bitmap->behind_wait, | |
5368 | atomic_read(&bitmap->behind_writes) == 0); | |
5369 | } | |
36d091f4 | 5370 | if (mddev->pers && mddev->pers->quiesce) { |
5aa61f42 N |
5371 | mddev->pers->quiesce(mddev, 1); |
5372 | mddev->pers->quiesce(mddev, 0); | |
5373 | } | |
5374 | md_unregister_thread(&mddev->thread); | |
5375 | if (mddev->queue) | |
5376 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
5377 | } | |
5378 | ||
5eff3c43 | 5379 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5380 | { |
36d091f4 | 5381 | struct md_personality *pers = mddev->pers; |
5aa61f42 | 5382 | mddev_detach(mddev); |
36d091f4 N |
5383 | spin_lock(&mddev->lock); |
5384 | mddev->ready = 0; | |
6177b472 | 5385 | mddev->pers = NULL; |
36d091f4 N |
5386 | spin_unlock(&mddev->lock); |
5387 | pers->free(mddev, mddev->private); | |
5388 | if (pers->sync_request && mddev->to_remove == NULL) | |
5389 | mddev->to_remove = &md_redundancy_group; | |
5390 | module_put(pers->owner); | |
cca9cf90 | 5391 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5392 | } |
5eff3c43 N |
5393 | |
5394 | void md_stop(struct mddev *mddev) | |
5395 | { | |
5396 | /* stop the array and free an attached data structures. | |
5397 | * This is called from dm-raid | |
5398 | */ | |
5399 | __md_stop(mddev); | |
5400 | bitmap_destroy(mddev); | |
5401 | if (mddev->bio_set) | |
5402 | bioset_free(mddev->bio_set); | |
5403 | } | |
5404 | ||
390ee602 | 5405 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5406 | |
a05b7ea0 | 5407 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5408 | { |
5409 | int err = 0; | |
30b8feb7 N |
5410 | int did_freeze = 0; |
5411 | ||
5412 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5413 | did_freeze = 1; | |
5414 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5415 | md_wakeup_thread(mddev->thread); | |
5416 | } | |
f851b60d | 5417 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5418 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5419 | if (mddev->sync_thread) |
30b8feb7 N |
5420 | /* Thread might be blocked waiting for metadata update |
5421 | * which will now never happen */ | |
5422 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5423 | |
30b8feb7 | 5424 | mddev_unlock(mddev); |
f851b60d N |
5425 | wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING, |
5426 | &mddev->recovery)); | |
30b8feb7 N |
5427 | mddev_lock_nointr(mddev); |
5428 | ||
a4bd82d0 | 5429 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5430 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 | 5431 | mddev->sync_thread || |
f851b60d | 5432 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
30b8feb7 | 5433 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { |
a4bd82d0 | 5434 | printk("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5435 | if (did_freeze) { |
5436 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5437 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5438 | md_wakeup_thread(mddev->thread); |
5439 | } | |
a4bd82d0 N |
5440 | err = -EBUSY; |
5441 | goto out; | |
5442 | } | |
5443 | if (mddev->pers) { | |
defad61a | 5444 | __md_stop_writes(mddev); |
a4bd82d0 N |
5445 | |
5446 | err = -ENXIO; | |
5447 | if (mddev->ro==1) | |
5448 | goto out; | |
5449 | mddev->ro = 1; | |
5450 | set_disk_ro(mddev->gendisk, 1); | |
5451 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d N |
5452 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5453 | md_wakeup_thread(mddev->thread); | |
00bcb4ac | 5454 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5455 | err = 0; |
a4bd82d0 N |
5456 | } |
5457 | out: | |
5458 | mutex_unlock(&mddev->open_mutex); | |
5459 | return err; | |
5460 | } | |
5461 | ||
9e653b63 N |
5462 | /* mode: |
5463 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5464 | * 2 - stop but do not disassemble array |
5465 | */ | |
f72ffdd6 | 5466 | static int do_md_stop(struct mddev *mddev, int mode, |
a05b7ea0 | 5467 | struct block_device *bdev) |
1da177e4 | 5468 | { |
1da177e4 | 5469 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5470 | struct md_rdev *rdev; |
30b8feb7 N |
5471 | int did_freeze = 0; |
5472 | ||
5473 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5474 | did_freeze = 1; | |
5475 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5476 | md_wakeup_thread(mddev->thread); | |
5477 | } | |
f851b60d | 5478 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
30b8feb7 | 5479 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
f851b60d | 5480 | if (mddev->sync_thread) |
30b8feb7 N |
5481 | /* Thread might be blocked waiting for metadata update |
5482 | * which will now never happen */ | |
5483 | wake_up_process(mddev->sync_thread->tsk); | |
f851b60d | 5484 | |
30b8feb7 | 5485 | mddev_unlock(mddev); |
f851b60d N |
5486 | wait_event(resync_wait, (mddev->sync_thread == NULL && |
5487 | !test_bit(MD_RECOVERY_RUNNING, | |
5488 | &mddev->recovery))); | |
30b8feb7 | 5489 | mddev_lock_nointr(mddev); |
1da177e4 | 5490 | |
c8c00a69 | 5491 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5492 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5493 | mddev->sysfs_active || |
5494 | mddev->sync_thread || | |
f851b60d | 5495 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
30b8feb7 | 5496 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { |
df5b20cf | 5497 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 5498 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
5499 | if (did_freeze) { |
5500 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
45eaf45d | 5501 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
30b8feb7 N |
5502 | md_wakeup_thread(mddev->thread); |
5503 | } | |
260fa034 N |
5504 | return -EBUSY; |
5505 | } | |
6e17b027 | 5506 | if (mddev->pers) { |
a4bd82d0 N |
5507 | if (mddev->ro) |
5508 | set_disk_ro(disk, 0); | |
409c57f3 | 5509 | |
defad61a | 5510 | __md_stop_writes(mddev); |
5eff3c43 | 5511 | __md_stop(mddev); |
a4bd82d0 | 5512 | mddev->queue->merge_bvec_fn = NULL; |
a4bd82d0 | 5513 | mddev->queue->backing_dev_info.congested_fn = NULL; |
6177b472 | 5514 | |
a4bd82d0 | 5515 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5516 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5517 | |
dafb20fa | 5518 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5519 | if (rdev->raid_disk >= 0) |
5520 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5521 | |
a4bd82d0 | 5522 | set_capacity(disk, 0); |
6e17b027 | 5523 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5524 | mddev->changed = 1; |
a4bd82d0 | 5525 | revalidate_disk(disk); |
0d4ca600 | 5526 | |
a4bd82d0 N |
5527 | if (mddev->ro) |
5528 | mddev->ro = 0; | |
6e17b027 N |
5529 | } else |
5530 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5531 | /* |
5532 | * Free resources if final stop | |
5533 | */ | |
9e653b63 | 5534 | if (mode == 0) { |
1da177e4 LT |
5535 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
5536 | ||
978f946b | 5537 | bitmap_destroy(mddev); |
c3d9714e | 5538 | if (mddev->bitmap_info.file) { |
4af1a041 N |
5539 | struct file *f = mddev->bitmap_info.file; |
5540 | spin_lock(&mddev->lock); | |
c3d9714e | 5541 | mddev->bitmap_info.file = NULL; |
4af1a041 N |
5542 | spin_unlock(&mddev->lock); |
5543 | fput(f); | |
978f946b | 5544 | } |
c3d9714e | 5545 | mddev->bitmap_info.offset = 0; |
978f946b | 5546 | |
1da177e4 LT |
5547 | export_array(mddev); |
5548 | ||
6177b472 | 5549 | md_clean(mddev); |
934d9c23 | 5550 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
5551 | if (mddev->hold_active == UNTIL_STOP) |
5552 | mddev->hold_active = 0; | |
a4bd82d0 | 5553 | } |
3f9d99c1 | 5554 | blk_integrity_unregister(disk); |
d7603b7e | 5555 | md_new_event(mddev); |
00bcb4ac | 5556 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 5557 | return 0; |
1da177e4 LT |
5558 | } |
5559 | ||
fdee8ae4 | 5560 | #ifndef MODULE |
fd01b88c | 5561 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 5562 | { |
3cb03002 | 5563 | struct md_rdev *rdev; |
1da177e4 LT |
5564 | int err; |
5565 | ||
a757e64c | 5566 | if (list_empty(&mddev->disks)) |
1da177e4 | 5567 | return; |
1da177e4 LT |
5568 | |
5569 | printk(KERN_INFO "md: running: "); | |
5570 | ||
dafb20fa | 5571 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
5572 | char b[BDEVNAME_SIZE]; |
5573 | printk("<%s>", bdevname(rdev->bdev,b)); | |
5574 | } | |
5575 | printk("\n"); | |
5576 | ||
d710e138 | 5577 | err = do_md_run(mddev); |
1da177e4 LT |
5578 | if (err) { |
5579 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
a05b7ea0 | 5580 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
5581 | } |
5582 | } | |
5583 | ||
5584 | /* | |
5585 | * lets try to run arrays based on all disks that have arrived | |
5586 | * until now. (those are in pending_raid_disks) | |
5587 | * | |
5588 | * the method: pick the first pending disk, collect all disks with | |
5589 | * the same UUID, remove all from the pending list and put them into | |
5590 | * the 'same_array' list. Then order this list based on superblock | |
5591 | * update time (freshest comes first), kick out 'old' disks and | |
5592 | * compare superblocks. If everything's fine then run it. | |
5593 | * | |
5594 | * If "unit" is allocated, then bump its reference count | |
5595 | */ | |
5596 | static void autorun_devices(int part) | |
5597 | { | |
3cb03002 | 5598 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 5599 | struct mddev *mddev; |
1da177e4 LT |
5600 | char b[BDEVNAME_SIZE]; |
5601 | ||
5602 | printk(KERN_INFO "md: autorun ...\n"); | |
5603 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 5604 | int unit; |
1da177e4 | 5605 | dev_t dev; |
ad01c9e3 | 5606 | LIST_HEAD(candidates); |
1da177e4 | 5607 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 5608 | struct md_rdev, same_set); |
1da177e4 LT |
5609 | |
5610 | printk(KERN_INFO "md: considering %s ...\n", | |
5611 | bdevname(rdev0->bdev,b)); | |
5612 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 5613 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
5614 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
5615 | printk(KERN_INFO "md: adding %s ...\n", | |
5616 | bdevname(rdev->bdev,b)); | |
5617 | list_move(&rdev->same_set, &candidates); | |
5618 | } | |
5619 | /* | |
5620 | * now we have a set of devices, with all of them having | |
5621 | * mostly sane superblocks. It's time to allocate the | |
5622 | * mddev. | |
5623 | */ | |
e8703fe1 N |
5624 | if (part) { |
5625 | dev = MKDEV(mdp_major, | |
5626 | rdev0->preferred_minor << MdpMinorShift); | |
5627 | unit = MINOR(dev) >> MdpMinorShift; | |
5628 | } else { | |
5629 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
5630 | unit = MINOR(dev); | |
5631 | } | |
5632 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
5633 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
5634 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
5635 | break; | |
5636 | } | |
1da177e4 LT |
5637 | |
5638 | md_probe(dev, NULL, NULL); | |
5639 | mddev = mddev_find(dev); | |
9bbbca3a NB |
5640 | if (!mddev || !mddev->gendisk) { |
5641 | if (mddev) | |
5642 | mddev_put(mddev); | |
5643 | printk(KERN_ERR | |
1da177e4 LT |
5644 | "md: cannot allocate memory for md drive.\n"); |
5645 | break; | |
5646 | } | |
f72ffdd6 | 5647 | if (mddev_lock(mddev)) |
1da177e4 LT |
5648 | printk(KERN_WARNING "md: %s locked, cannot run\n", |
5649 | mdname(mddev)); | |
5650 | else if (mddev->raid_disks || mddev->major_version | |
5651 | || !list_empty(&mddev->disks)) { | |
f72ffdd6 | 5652 | printk(KERN_WARNING |
1da177e4 LT |
5653 | "md: %s already running, cannot run %s\n", |
5654 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
5655 | mddev_unlock(mddev); | |
5656 | } else { | |
5657 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 5658 | mddev->persistent = 1; |
159ec1fc | 5659 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
5660 | list_del_init(&rdev->same_set); |
5661 | if (bind_rdev_to_array(rdev, mddev)) | |
5662 | export_rdev(rdev); | |
5663 | } | |
5664 | autorun_array(mddev); | |
5665 | mddev_unlock(mddev); | |
5666 | } | |
5667 | /* on success, candidates will be empty, on error | |
5668 | * it won't... | |
5669 | */ | |
159ec1fc | 5670 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 5671 | list_del_init(&rdev->same_set); |
1da177e4 | 5672 | export_rdev(rdev); |
4b80991c | 5673 | } |
1da177e4 LT |
5674 | mddev_put(mddev); |
5675 | } | |
5676 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
5677 | } | |
fdee8ae4 | 5678 | #endif /* !MODULE */ |
1da177e4 | 5679 | |
f72ffdd6 | 5680 | static int get_version(void __user *arg) |
1da177e4 LT |
5681 | { |
5682 | mdu_version_t ver; | |
5683 | ||
5684 | ver.major = MD_MAJOR_VERSION; | |
5685 | ver.minor = MD_MINOR_VERSION; | |
5686 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
5687 | ||
5688 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
5689 | return -EFAULT; | |
5690 | ||
5691 | return 0; | |
5692 | } | |
5693 | ||
f72ffdd6 | 5694 | static int get_array_info(struct mddev *mddev, void __user *arg) |
1da177e4 LT |
5695 | { |
5696 | mdu_array_info_t info; | |
a9f326eb | 5697 | int nr,working,insync,failed,spare; |
3cb03002 | 5698 | struct md_rdev *rdev; |
1da177e4 | 5699 | |
1ca69c4b N |
5700 | nr = working = insync = failed = spare = 0; |
5701 | rcu_read_lock(); | |
5702 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 5703 | nr++; |
b2d444d7 | 5704 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5705 | failed++; |
5706 | else { | |
5707 | working++; | |
b2d444d7 | 5708 | if (test_bit(In_sync, &rdev->flags)) |
f72ffdd6 | 5709 | insync++; |
1da177e4 LT |
5710 | else |
5711 | spare++; | |
5712 | } | |
5713 | } | |
1ca69c4b | 5714 | rcu_read_unlock(); |
1da177e4 LT |
5715 | |
5716 | info.major_version = mddev->major_version; | |
5717 | info.minor_version = mddev->minor_version; | |
5718 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5719 | info.ctime = mddev->ctime; | |
5720 | info.level = mddev->level; | |
58c0fed4 AN |
5721 | info.size = mddev->dev_sectors / 2; |
5722 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5723 | info.size = -1; |
1da177e4 LT |
5724 | info.nr_disks = nr; |
5725 | info.raid_disks = mddev->raid_disks; | |
5726 | info.md_minor = mddev->md_minor; | |
5727 | info.not_persistent= !mddev->persistent; | |
5728 | ||
5729 | info.utime = mddev->utime; | |
5730 | info.state = 0; | |
5731 | if (mddev->in_sync) | |
5732 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5733 | if (mddev->bitmap && mddev->bitmap_info.offset) |
9bd35920 | 5734 | info.state |= (1<<MD_SB_BITMAP_PRESENT); |
ca8895d9 GR |
5735 | if (mddev_is_clustered(mddev)) |
5736 | info.state |= (1<<MD_SB_CLUSTERED); | |
a9f326eb | 5737 | info.active_disks = insync; |
1da177e4 LT |
5738 | info.working_disks = working; |
5739 | info.failed_disks = failed; | |
5740 | info.spare_disks = spare; | |
5741 | ||
5742 | info.layout = mddev->layout; | |
9d8f0363 | 5743 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5744 | |
5745 | if (copy_to_user(arg, &info, sizeof(info))) | |
5746 | return -EFAULT; | |
5747 | ||
5748 | return 0; | |
5749 | } | |
5750 | ||
f72ffdd6 | 5751 | static int get_bitmap_file(struct mddev *mddev, void __user * arg) |
32a7627c N |
5752 | { |
5753 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
f4ad3d38 | 5754 | char *ptr; |
4af1a041 | 5755 | int err; |
32a7627c | 5756 | |
60559da4 | 5757 | file = kmalloc(sizeof(*file), GFP_NOIO); |
32a7627c | 5758 | if (!file) |
4af1a041 | 5759 | return -ENOMEM; |
32a7627c | 5760 | |
4af1a041 N |
5761 | err = 0; |
5762 | spin_lock(&mddev->lock); | |
32a7627c | 5763 | /* bitmap disabled, zero the first byte and copy out */ |
4af1a041 | 5764 | if (!mddev->bitmap_info.file) |
32a7627c | 5765 | file->pathname[0] = '\0'; |
4af1a041 N |
5766 | else if ((ptr = d_path(&mddev->bitmap_info.file->f_path, |
5767 | file->pathname, sizeof(file->pathname))), | |
5768 | IS_ERR(ptr)) | |
5769 | err = PTR_ERR(ptr); | |
5770 | else | |
5771 | memmove(file->pathname, ptr, | |
5772 | sizeof(file->pathname)-(ptr-file->pathname)); | |
5773 | spin_unlock(&mddev->lock); | |
32a7627c | 5774 | |
4af1a041 N |
5775 | if (err == 0 && |
5776 | copy_to_user(arg, file, sizeof(*file))) | |
32a7627c | 5777 | err = -EFAULT; |
4af1a041 | 5778 | |
32a7627c N |
5779 | kfree(file); |
5780 | return err; | |
5781 | } | |
5782 | ||
f72ffdd6 | 5783 | static int get_disk_info(struct mddev *mddev, void __user * arg) |
1da177e4 LT |
5784 | { |
5785 | mdu_disk_info_t info; | |
3cb03002 | 5786 | struct md_rdev *rdev; |
1da177e4 LT |
5787 | |
5788 | if (copy_from_user(&info, arg, sizeof(info))) | |
5789 | return -EFAULT; | |
5790 | ||
1ca69c4b | 5791 | rcu_read_lock(); |
57d051dc | 5792 | rdev = md_find_rdev_nr_rcu(mddev, info.number); |
1da177e4 LT |
5793 | if (rdev) { |
5794 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5795 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5796 | info.raid_disk = rdev->raid_disk; | |
5797 | info.state = 0; | |
b2d444d7 | 5798 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5799 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5800 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5801 | info.state |= (1<<MD_DISK_ACTIVE); |
5802 | info.state |= (1<<MD_DISK_SYNC); | |
5803 | } | |
8ddf9efe N |
5804 | if (test_bit(WriteMostly, &rdev->flags)) |
5805 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5806 | } else { |
5807 | info.major = info.minor = 0; | |
5808 | info.raid_disk = -1; | |
5809 | info.state = (1<<MD_DISK_REMOVED); | |
5810 | } | |
1ca69c4b | 5811 | rcu_read_unlock(); |
1da177e4 LT |
5812 | |
5813 | if (copy_to_user(arg, &info, sizeof(info))) | |
5814 | return -EFAULT; | |
5815 | ||
5816 | return 0; | |
5817 | } | |
5818 | ||
f72ffdd6 | 5819 | static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info) |
1da177e4 LT |
5820 | { |
5821 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5822 | struct md_rdev *rdev; |
1da177e4 LT |
5823 | dev_t dev = MKDEV(info->major,info->minor); |
5824 | ||
1aee41f6 GR |
5825 | if (mddev_is_clustered(mddev) && |
5826 | !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) { | |
fa8259da | 5827 | pr_err("%s: Cannot add to clustered mddev.\n", |
1aee41f6 GR |
5828 | mdname(mddev)); |
5829 | return -EINVAL; | |
5830 | } | |
5831 | ||
1da177e4 LT |
5832 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) |
5833 | return -EOVERFLOW; | |
5834 | ||
5835 | if (!mddev->raid_disks) { | |
5836 | int err; | |
5837 | /* expecting a device which has a superblock */ | |
5838 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5839 | if (IS_ERR(rdev)) { | |
f72ffdd6 | 5840 | printk(KERN_WARNING |
1da177e4 LT |
5841 | "md: md_import_device returned %ld\n", |
5842 | PTR_ERR(rdev)); | |
5843 | return PTR_ERR(rdev); | |
5844 | } | |
5845 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
5846 | struct md_rdev *rdev0 |
5847 | = list_entry(mddev->disks.next, | |
5848 | struct md_rdev, same_set); | |
a9f326eb | 5849 | err = super_types[mddev->major_version] |
1da177e4 LT |
5850 | .load_super(rdev, rdev0, mddev->minor_version); |
5851 | if (err < 0) { | |
f72ffdd6 | 5852 | printk(KERN_WARNING |
1da177e4 | 5853 | "md: %s has different UUID to %s\n", |
f72ffdd6 | 5854 | bdevname(rdev->bdev,b), |
1da177e4 LT |
5855 | bdevname(rdev0->bdev,b2)); |
5856 | export_rdev(rdev); | |
5857 | return -EINVAL; | |
5858 | } | |
5859 | } | |
5860 | err = bind_rdev_to_array(rdev, mddev); | |
5861 | if (err) | |
5862 | export_rdev(rdev); | |
5863 | return err; | |
5864 | } | |
5865 | ||
5866 | /* | |
5867 | * add_new_disk can be used once the array is assembled | |
5868 | * to add "hot spares". They must already have a superblock | |
5869 | * written | |
5870 | */ | |
5871 | if (mddev->pers) { | |
5872 | int err; | |
5873 | if (!mddev->pers->hot_add_disk) { | |
f72ffdd6 | 5874 | printk(KERN_WARNING |
1da177e4 LT |
5875 | "%s: personality does not support diskops!\n", |
5876 | mdname(mddev)); | |
5877 | return -EINVAL; | |
5878 | } | |
7b1e35f6 N |
5879 | if (mddev->persistent) |
5880 | rdev = md_import_device(dev, mddev->major_version, | |
5881 | mddev->minor_version); | |
5882 | else | |
5883 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 | 5884 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5885 | printk(KERN_WARNING |
1da177e4 LT |
5886 | "md: md_import_device returned %ld\n", |
5887 | PTR_ERR(rdev)); | |
5888 | return PTR_ERR(rdev); | |
5889 | } | |
1a855a06 | 5890 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5891 | if (!mddev->persistent) { |
5892 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5893 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5894 | rdev->raid_disk = info->raid_disk; |
bf572541 | 5895 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 5896 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 5897 | } else |
41158c7e | 5898 | rdev->raid_disk = -1; |
f466722c | 5899 | rdev->saved_raid_disk = rdev->raid_disk; |
41158c7e N |
5900 | } else |
5901 | super_types[mddev->major_version]. | |
5902 | validate_super(mddev, rdev); | |
bedd86b7 | 5903 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 5904 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
5905 | /* This was a hot-add request, but events doesn't |
5906 | * match, so reject it. | |
5907 | */ | |
5908 | export_rdev(rdev); | |
5909 | return -EINVAL; | |
5910 | } | |
5911 | ||
b2d444d7 | 5912 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5913 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5914 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5915 | else |
5916 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5917 | |
1aee41f6 GR |
5918 | /* |
5919 | * check whether the device shows up in other nodes | |
5920 | */ | |
5921 | if (mddev_is_clustered(mddev)) { | |
5922 | if (info->state & (1 << MD_DISK_CANDIDATE)) { | |
5923 | /* Through --cluster-confirm */ | |
5924 | set_bit(Candidate, &rdev->flags); | |
fa8259da GR |
5925 | err = md_cluster_ops->new_disk_ack(mddev, true); |
5926 | if (err) { | |
5927 | export_rdev(rdev); | |
5928 | return err; | |
5929 | } | |
1aee41f6 GR |
5930 | } else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) { |
5931 | /* --add initiated by this node */ | |
5932 | err = md_cluster_ops->add_new_disk_start(mddev, rdev); | |
5933 | if (err) { | |
5934 | md_cluster_ops->add_new_disk_finish(mddev); | |
5935 | export_rdev(rdev); | |
5936 | return err; | |
5937 | } | |
5938 | } | |
5939 | } | |
5940 | ||
1da177e4 LT |
5941 | rdev->raid_disk = -1; |
5942 | err = bind_rdev_to_array(rdev, mddev); | |
5943 | if (err) | |
5944 | export_rdev(rdev); | |
52664732 | 5945 | else |
a6da4ef8 | 5946 | err = add_bound_rdev(rdev); |
1aee41f6 GR |
5947 | if (mddev_is_clustered(mddev) && |
5948 | (info->state & (1 << MD_DISK_CLUSTER_ADD))) | |
5949 | md_cluster_ops->add_new_disk_finish(mddev); | |
1da177e4 LT |
5950 | return err; |
5951 | } | |
5952 | ||
5953 | /* otherwise, add_new_disk is only allowed | |
5954 | * for major_version==0 superblocks | |
5955 | */ | |
5956 | if (mddev->major_version != 0) { | |
5957 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5958 | mdname(mddev)); | |
5959 | return -EINVAL; | |
5960 | } | |
5961 | ||
5962 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5963 | int err; | |
d710e138 | 5964 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 5965 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5966 | printk(KERN_WARNING |
1da177e4 LT |
5967 | "md: error, md_import_device() returned %ld\n", |
5968 | PTR_ERR(rdev)); | |
5969 | return PTR_ERR(rdev); | |
5970 | } | |
5971 | rdev->desc_nr = info->number; | |
5972 | if (info->raid_disk < mddev->raid_disks) | |
5973 | rdev->raid_disk = info->raid_disk; | |
5974 | else | |
5975 | rdev->raid_disk = -1; | |
5976 | ||
1da177e4 | 5977 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5978 | if (info->state & (1<<MD_DISK_SYNC)) |
5979 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5980 | |
8ddf9efe N |
5981 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5982 | set_bit(WriteMostly, &rdev->flags); | |
5983 | ||
1da177e4 LT |
5984 | if (!mddev->persistent) { |
5985 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5986 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5987 | } else | |
57b2caa3 | 5988 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 5989 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5990 | |
2bf071bf N |
5991 | err = bind_rdev_to_array(rdev, mddev); |
5992 | if (err) { | |
5993 | export_rdev(rdev); | |
5994 | return err; | |
5995 | } | |
1da177e4 LT |
5996 | } |
5997 | ||
5998 | return 0; | |
5999 | } | |
6000 | ||
f72ffdd6 | 6001 | static int hot_remove_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6002 | { |
6003 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 6004 | struct md_rdev *rdev; |
1da177e4 | 6005 | |
1da177e4 LT |
6006 | rdev = find_rdev(mddev, dev); |
6007 | if (!rdev) | |
6008 | return -ENXIO; | |
6009 | ||
293467aa GR |
6010 | if (mddev_is_clustered(mddev)) |
6011 | md_cluster_ops->metadata_update_start(mddev); | |
6012 | ||
3ea8929d N |
6013 | clear_bit(Blocked, &rdev->flags); |
6014 | remove_and_add_spares(mddev, rdev); | |
6015 | ||
1da177e4 LT |
6016 | if (rdev->raid_disk >= 0) |
6017 | goto busy; | |
6018 | ||
88bcfef7 GR |
6019 | if (mddev_is_clustered(mddev)) |
6020 | md_cluster_ops->remove_disk(mddev, rdev); | |
6021 | ||
fb56dfef | 6022 | md_kick_rdev_from_array(rdev); |
850b2b42 | 6023 | md_update_sb(mddev, 1); |
d7603b7e | 6024 | md_new_event(mddev); |
1da177e4 | 6025 | |
293467aa GR |
6026 | if (mddev_is_clustered(mddev)) |
6027 | md_cluster_ops->metadata_update_finish(mddev); | |
6028 | ||
1da177e4 LT |
6029 | return 0; |
6030 | busy: | |
293467aa GR |
6031 | if (mddev_is_clustered(mddev)) |
6032 | md_cluster_ops->metadata_update_cancel(mddev); | |
fdefa4d8 | 6033 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
6034 | bdevname(rdev->bdev,b), mdname(mddev)); |
6035 | return -EBUSY; | |
6036 | } | |
6037 | ||
f72ffdd6 | 6038 | static int hot_add_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
6039 | { |
6040 | char b[BDEVNAME_SIZE]; | |
6041 | int err; | |
3cb03002 | 6042 | struct md_rdev *rdev; |
1da177e4 LT |
6043 | |
6044 | if (!mddev->pers) | |
6045 | return -ENODEV; | |
6046 | ||
6047 | if (mddev->major_version != 0) { | |
6048 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
6049 | " version-0 superblocks.\n", | |
6050 | mdname(mddev)); | |
6051 | return -EINVAL; | |
6052 | } | |
6053 | if (!mddev->pers->hot_add_disk) { | |
f72ffdd6 | 6054 | printk(KERN_WARNING |
1da177e4 LT |
6055 | "%s: personality does not support diskops!\n", |
6056 | mdname(mddev)); | |
6057 | return -EINVAL; | |
6058 | } | |
6059 | ||
d710e138 | 6060 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 6061 | if (IS_ERR(rdev)) { |
f72ffdd6 | 6062 | printk(KERN_WARNING |
1da177e4 LT |
6063 | "md: error, md_import_device() returned %ld\n", |
6064 | PTR_ERR(rdev)); | |
6065 | return -EINVAL; | |
6066 | } | |
6067 | ||
6068 | if (mddev->persistent) | |
57b2caa3 | 6069 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 6070 | else |
77304d2a | 6071 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 6072 | |
8190e754 | 6073 | rdev->sectors = rdev->sb_start; |
1da177e4 | 6074 | |
b2d444d7 | 6075 | if (test_bit(Faulty, &rdev->flags)) { |
f72ffdd6 | 6076 | printk(KERN_WARNING |
1da177e4 LT |
6077 | "md: can not hot-add faulty %s disk to %s!\n", |
6078 | bdevname(rdev->bdev,b), mdname(mddev)); | |
6079 | err = -EINVAL; | |
6080 | goto abort_export; | |
6081 | } | |
293467aa GR |
6082 | |
6083 | if (mddev_is_clustered(mddev)) | |
6084 | md_cluster_ops->metadata_update_start(mddev); | |
b2d444d7 | 6085 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 6086 | rdev->desc_nr = -1; |
5842730d | 6087 | rdev->saved_raid_disk = -1; |
2bf071bf N |
6088 | err = bind_rdev_to_array(rdev, mddev); |
6089 | if (err) | |
293467aa | 6090 | goto abort_clustered; |
1da177e4 LT |
6091 | |
6092 | /* | |
6093 | * The rest should better be atomic, we can have disk failures | |
6094 | * noticed in interrupt contexts ... | |
6095 | */ | |
6096 | ||
1da177e4 LT |
6097 | rdev->raid_disk = -1; |
6098 | ||
850b2b42 | 6099 | md_update_sb(mddev, 1); |
1da177e4 | 6100 | |
293467aa GR |
6101 | if (mddev_is_clustered(mddev)) |
6102 | md_cluster_ops->metadata_update_finish(mddev); | |
1da177e4 LT |
6103 | /* |
6104 | * Kick recovery, maybe this spare has to be added to the | |
6105 | * array immediately. | |
6106 | */ | |
6107 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6108 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 6109 | md_new_event(mddev); |
1da177e4 LT |
6110 | return 0; |
6111 | ||
293467aa GR |
6112 | abort_clustered: |
6113 | if (mddev_is_clustered(mddev)) | |
6114 | md_cluster_ops->metadata_update_cancel(mddev); | |
1da177e4 LT |
6115 | abort_export: |
6116 | export_rdev(rdev); | |
6117 | return err; | |
6118 | } | |
6119 | ||
fd01b88c | 6120 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c | 6121 | { |
035328c2 | 6122 | int err = 0; |
32a7627c | 6123 | |
36fa3063 | 6124 | if (mddev->pers) { |
d66b1b39 | 6125 | if (!mddev->pers->quiesce || !mddev->thread) |
36fa3063 N |
6126 | return -EBUSY; |
6127 | if (mddev->recovery || mddev->sync_thread) | |
6128 | return -EBUSY; | |
6129 | /* we should be able to change the bitmap.. */ | |
6130 | } | |
32a7627c | 6131 | |
36fa3063 | 6132 | if (fd >= 0) { |
035328c2 | 6133 | struct inode *inode; |
1e594bb2 N |
6134 | struct file *f; |
6135 | ||
6136 | if (mddev->bitmap || mddev->bitmap_info.file) | |
36fa3063 | 6137 | return -EEXIST; /* cannot add when bitmap is present */ |
1e594bb2 | 6138 | f = fget(fd); |
32a7627c | 6139 | |
1e594bb2 | 6140 | if (f == NULL) { |
36fa3063 N |
6141 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
6142 | mdname(mddev)); | |
6143 | return -EBADF; | |
6144 | } | |
6145 | ||
1e594bb2 | 6146 | inode = f->f_mapping->host; |
035328c2 N |
6147 | if (!S_ISREG(inode->i_mode)) { |
6148 | printk(KERN_ERR "%s: error: bitmap file must be a regular file\n", | |
6149 | mdname(mddev)); | |
6150 | err = -EBADF; | |
1e594bb2 | 6151 | } else if (!(f->f_mode & FMODE_WRITE)) { |
035328c2 N |
6152 | printk(KERN_ERR "%s: error: bitmap file must open for write\n", |
6153 | mdname(mddev)); | |
6154 | err = -EBADF; | |
6155 | } else if (atomic_read(&inode->i_writecount) != 1) { | |
36fa3063 N |
6156 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", |
6157 | mdname(mddev)); | |
035328c2 N |
6158 | err = -EBUSY; |
6159 | } | |
6160 | if (err) { | |
1e594bb2 | 6161 | fput(f); |
36fa3063 N |
6162 | return err; |
6163 | } | |
1e594bb2 | 6164 | mddev->bitmap_info.file = f; |
c3d9714e | 6165 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
6166 | } else if (mddev->bitmap == NULL) |
6167 | return -ENOENT; /* cannot remove what isn't there */ | |
6168 | err = 0; | |
6169 | if (mddev->pers) { | |
6170 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 6171 | if (fd >= 0) { |
f9209a32 GR |
6172 | struct bitmap *bitmap; |
6173 | ||
6174 | bitmap = bitmap_create(mddev, -1); | |
6175 | if (!IS_ERR(bitmap)) { | |
6176 | mddev->bitmap = bitmap; | |
69e51b44 | 6177 | err = bitmap_load(mddev); |
ba599aca N |
6178 | } else |
6179 | err = PTR_ERR(bitmap); | |
69e51b44 | 6180 | } |
d7375ab3 | 6181 | if (fd < 0 || err) { |
36fa3063 | 6182 | bitmap_destroy(mddev); |
d7375ab3 N |
6183 | fd = -1; /* make sure to put the file */ |
6184 | } | |
36fa3063 | 6185 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
6186 | } |
6187 | if (fd < 0) { | |
4af1a041 N |
6188 | struct file *f = mddev->bitmap_info.file; |
6189 | if (f) { | |
6190 | spin_lock(&mddev->lock); | |
6191 | mddev->bitmap_info.file = NULL; | |
6192 | spin_unlock(&mddev->lock); | |
6193 | fput(f); | |
6194 | } | |
36fa3063 N |
6195 | } |
6196 | ||
32a7627c N |
6197 | return err; |
6198 | } | |
6199 | ||
1da177e4 LT |
6200 | /* |
6201 | * set_array_info is used two different ways | |
6202 | * The original usage is when creating a new array. | |
6203 | * In this usage, raid_disks is > 0 and it together with | |
6204 | * level, size, not_persistent,layout,chunksize determine the | |
6205 | * shape of the array. | |
6206 | * This will always create an array with a type-0.90.0 superblock. | |
6207 | * The newer usage is when assembling an array. | |
6208 | * In this case raid_disks will be 0, and the major_version field is | |
6209 | * use to determine which style super-blocks are to be found on the devices. | |
6210 | * The minor and patch _version numbers are also kept incase the | |
6211 | * super_block handler wishes to interpret them. | |
6212 | */ | |
f72ffdd6 | 6213 | static int set_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6214 | { |
6215 | ||
6216 | if (info->raid_disks == 0) { | |
6217 | /* just setting version number for superblock loading */ | |
6218 | if (info->major_version < 0 || | |
50511da3 | 6219 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
6220 | super_types[info->major_version].name == NULL) { |
6221 | /* maybe try to auto-load a module? */ | |
f72ffdd6 | 6222 | printk(KERN_INFO |
1da177e4 LT |
6223 | "md: superblock version %d not known\n", |
6224 | info->major_version); | |
6225 | return -EINVAL; | |
6226 | } | |
6227 | mddev->major_version = info->major_version; | |
6228 | mddev->minor_version = info->minor_version; | |
6229 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 6230 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
6231 | /* ensure mddev_put doesn't delete this now that there |
6232 | * is some minimal configuration. | |
6233 | */ | |
6234 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
6235 | return 0; |
6236 | } | |
6237 | mddev->major_version = MD_MAJOR_VERSION; | |
6238 | mddev->minor_version = MD_MINOR_VERSION; | |
6239 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
6240 | mddev->ctime = get_seconds(); | |
6241 | ||
6242 | mddev->level = info->level; | |
17115e03 | 6243 | mddev->clevel[0] = 0; |
58c0fed4 | 6244 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
6245 | mddev->raid_disks = info->raid_disks; |
6246 | /* don't set md_minor, it is determined by which /dev/md* was | |
6247 | * openned | |
6248 | */ | |
6249 | if (info->state & (1<<MD_SB_CLEAN)) | |
6250 | mddev->recovery_cp = MaxSector; | |
6251 | else | |
6252 | mddev->recovery_cp = 0; | |
6253 | mddev->persistent = ! info->not_persistent; | |
e691063a | 6254 | mddev->external = 0; |
1da177e4 LT |
6255 | |
6256 | mddev->layout = info->layout; | |
9d8f0363 | 6257 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
6258 | |
6259 | mddev->max_disks = MD_SB_DISKS; | |
6260 | ||
e691063a N |
6261 | if (mddev->persistent) |
6262 | mddev->flags = 0; | |
850b2b42 | 6263 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 6264 | |
c3d9714e | 6265 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 6266 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 6267 | mddev->bitmap_info.offset = 0; |
b2a2703c | 6268 | |
f6705578 N |
6269 | mddev->reshape_position = MaxSector; |
6270 | ||
1da177e4 LT |
6271 | /* |
6272 | * Generate a 128 bit UUID | |
6273 | */ | |
6274 | get_random_bytes(mddev->uuid, 16); | |
6275 | ||
f6705578 | 6276 | mddev->new_level = mddev->level; |
664e7c41 | 6277 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
6278 | mddev->new_layout = mddev->layout; |
6279 | mddev->delta_disks = 0; | |
2c810cdd | 6280 | mddev->reshape_backwards = 0; |
f6705578 | 6281 | |
1da177e4 LT |
6282 | return 0; |
6283 | } | |
6284 | ||
fd01b88c | 6285 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 6286 | { |
b522adcd DW |
6287 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
6288 | ||
6289 | if (mddev->external_size) | |
6290 | return; | |
6291 | ||
1f403624 DW |
6292 | mddev->array_sectors = array_sectors; |
6293 | } | |
6294 | EXPORT_SYMBOL(md_set_array_sectors); | |
6295 | ||
fd01b88c | 6296 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 6297 | { |
3cb03002 | 6298 | struct md_rdev *rdev; |
a35b0d69 | 6299 | int rv; |
d71f9f88 | 6300 | int fit = (num_sectors == 0); |
a35b0d69 N |
6301 | |
6302 | if (mddev->pers->resize == NULL) | |
6303 | return -EINVAL; | |
d71f9f88 AN |
6304 | /* The "num_sectors" is the number of sectors of each device that |
6305 | * is used. This can only make sense for arrays with redundancy. | |
6306 | * linear and raid0 always use whatever space is available. We can only | |
6307 | * consider changing this number if no resync or reconstruction is | |
6308 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 6309 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
6310 | * of each device. If num_sectors is zero, we find the largest size |
6311 | * that fits. | |
a35b0d69 | 6312 | */ |
f851b60d N |
6313 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
6314 | mddev->sync_thread) | |
a35b0d69 | 6315 | return -EBUSY; |
bd8839e0 N |
6316 | if (mddev->ro) |
6317 | return -EROFS; | |
a4a6125a | 6318 | |
dafb20fa | 6319 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 6320 | sector_t avail = rdev->sectors; |
01ab5662 | 6321 | |
d71f9f88 AN |
6322 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
6323 | num_sectors = avail; | |
6324 | if (avail < num_sectors) | |
a35b0d69 N |
6325 | return -ENOSPC; |
6326 | } | |
d71f9f88 | 6327 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
6328 | if (!rv) |
6329 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
6330 | return rv; |
6331 | } | |
6332 | ||
fd01b88c | 6333 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
6334 | { |
6335 | int rv; | |
c6563a8c | 6336 | struct md_rdev *rdev; |
da943b99 | 6337 | /* change the number of raid disks */ |
63c70c4f | 6338 | if (mddev->pers->check_reshape == NULL) |
da943b99 | 6339 | return -EINVAL; |
bd8839e0 N |
6340 | if (mddev->ro) |
6341 | return -EROFS; | |
da943b99 | 6342 | if (raid_disks <= 0 || |
233fca36 | 6343 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 6344 | return -EINVAL; |
f851b60d N |
6345 | if (mddev->sync_thread || |
6346 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
6347 | mddev->reshape_position != MaxSector) | |
da943b99 | 6348 | return -EBUSY; |
c6563a8c N |
6349 | |
6350 | rdev_for_each(rdev, mddev) { | |
6351 | if (mddev->raid_disks < raid_disks && | |
6352 | rdev->data_offset < rdev->new_data_offset) | |
6353 | return -EINVAL; | |
6354 | if (mddev->raid_disks > raid_disks && | |
6355 | rdev->data_offset > rdev->new_data_offset) | |
6356 | return -EINVAL; | |
6357 | } | |
6358 | ||
63c70c4f | 6359 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
6360 | if (mddev->delta_disks < 0) |
6361 | mddev->reshape_backwards = 1; | |
6362 | else if (mddev->delta_disks > 0) | |
6363 | mddev->reshape_backwards = 0; | |
63c70c4f N |
6364 | |
6365 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 6366 | if (rv < 0) { |
de171cb9 | 6367 | mddev->delta_disks = 0; |
2c810cdd N |
6368 | mddev->reshape_backwards = 0; |
6369 | } | |
da943b99 N |
6370 | return rv; |
6371 | } | |
6372 | ||
1da177e4 LT |
6373 | /* |
6374 | * update_array_info is used to change the configuration of an | |
6375 | * on-line array. | |
6376 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6377 | * fields in the info are checked against the array. | |
6378 | * Any differences that cannot be handled will cause an error. | |
6379 | * Normally, only one change can be managed at a time. | |
6380 | */ | |
fd01b88c | 6381 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6382 | { |
6383 | int rv = 0; | |
6384 | int cnt = 0; | |
36fa3063 N |
6385 | int state = 0; |
6386 | ||
6387 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6388 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6389 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6390 | |
6391 | if (mddev->major_version != info->major_version || | |
6392 | mddev->minor_version != info->minor_version || | |
6393 | /* mddev->patch_version != info->patch_version || */ | |
6394 | mddev->ctime != info->ctime || | |
6395 | mddev->level != info->level || | |
6396 | /* mddev->layout != info->layout || */ | |
6397 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 6398 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6399 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6400 | ((state^info->state) & 0xfffffe00) | |
6401 | ) | |
1da177e4 LT |
6402 | return -EINVAL; |
6403 | /* Check there is only one change */ | |
58c0fed4 AN |
6404 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6405 | cnt++; | |
6406 | if (mddev->raid_disks != info->raid_disks) | |
6407 | cnt++; | |
6408 | if (mddev->layout != info->layout) | |
6409 | cnt++; | |
6410 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6411 | cnt++; | |
6412 | if (cnt == 0) | |
6413 | return 0; | |
6414 | if (cnt > 1) | |
6415 | return -EINVAL; | |
1da177e4 LT |
6416 | |
6417 | if (mddev->layout != info->layout) { | |
6418 | /* Change layout | |
6419 | * we don't need to do anything at the md level, the | |
6420 | * personality will take care of it all. | |
6421 | */ | |
50ac168a | 6422 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6423 | return -EINVAL; |
597a711b N |
6424 | else { |
6425 | mddev->new_layout = info->layout; | |
50ac168a | 6426 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6427 | if (rv) |
6428 | mddev->new_layout = mddev->layout; | |
6429 | return rv; | |
6430 | } | |
1da177e4 | 6431 | } |
293467aa GR |
6432 | if (mddev_is_clustered(mddev)) |
6433 | md_cluster_ops->metadata_update_start(mddev); | |
58c0fed4 | 6434 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6435 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6436 | |
da943b99 N |
6437 | if (mddev->raid_disks != info->raid_disks) |
6438 | rv = update_raid_disks(mddev, info->raid_disks); | |
6439 | ||
36fa3063 | 6440 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
293467aa GR |
6441 | if (mddev->pers->quiesce == NULL || mddev->thread == NULL) { |
6442 | rv = -EINVAL; | |
6443 | goto err; | |
6444 | } | |
6445 | if (mddev->recovery || mddev->sync_thread) { | |
6446 | rv = -EBUSY; | |
6447 | goto err; | |
6448 | } | |
36fa3063 | 6449 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { |
f9209a32 | 6450 | struct bitmap *bitmap; |
36fa3063 | 6451 | /* add the bitmap */ |
293467aa GR |
6452 | if (mddev->bitmap) { |
6453 | rv = -EEXIST; | |
6454 | goto err; | |
6455 | } | |
6456 | if (mddev->bitmap_info.default_offset == 0) { | |
6457 | rv = -EINVAL; | |
6458 | goto err; | |
6459 | } | |
c3d9714e N |
6460 | mddev->bitmap_info.offset = |
6461 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6462 | mddev->bitmap_info.space = |
6463 | mddev->bitmap_info.default_space; | |
36fa3063 | 6464 | mddev->pers->quiesce(mddev, 1); |
f9209a32 GR |
6465 | bitmap = bitmap_create(mddev, -1); |
6466 | if (!IS_ERR(bitmap)) { | |
6467 | mddev->bitmap = bitmap; | |
69e51b44 | 6468 | rv = bitmap_load(mddev); |
ba599aca N |
6469 | } else |
6470 | rv = PTR_ERR(bitmap); | |
36fa3063 N |
6471 | if (rv) |
6472 | bitmap_destroy(mddev); | |
6473 | mddev->pers->quiesce(mddev, 0); | |
6474 | } else { | |
6475 | /* remove the bitmap */ | |
293467aa GR |
6476 | if (!mddev->bitmap) { |
6477 | rv = -ENOENT; | |
6478 | goto err; | |
6479 | } | |
6480 | if (mddev->bitmap->storage.file) { | |
6481 | rv = -EINVAL; | |
6482 | goto err; | |
6483 | } | |
36fa3063 N |
6484 | mddev->pers->quiesce(mddev, 1); |
6485 | bitmap_destroy(mddev); | |
6486 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 6487 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6488 | } |
6489 | } | |
850b2b42 | 6490 | md_update_sb(mddev, 1); |
293467aa GR |
6491 | if (mddev_is_clustered(mddev)) |
6492 | md_cluster_ops->metadata_update_finish(mddev); | |
6493 | return rv; | |
6494 | err: | |
6495 | if (mddev_is_clustered(mddev)) | |
6496 | md_cluster_ops->metadata_update_cancel(mddev); | |
1da177e4 LT |
6497 | return rv; |
6498 | } | |
6499 | ||
fd01b88c | 6500 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6501 | { |
3cb03002 | 6502 | struct md_rdev *rdev; |
1ca69c4b | 6503 | int err = 0; |
1da177e4 LT |
6504 | |
6505 | if (mddev->pers == NULL) | |
6506 | return -ENODEV; | |
6507 | ||
1ca69c4b N |
6508 | rcu_read_lock(); |
6509 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 6510 | if (!rdev) |
1ca69c4b N |
6511 | err = -ENODEV; |
6512 | else { | |
6513 | md_error(mddev, rdev); | |
6514 | if (!test_bit(Faulty, &rdev->flags)) | |
6515 | err = -EBUSY; | |
6516 | } | |
6517 | rcu_read_unlock(); | |
6518 | return err; | |
1da177e4 LT |
6519 | } |
6520 | ||
2f9618ce AN |
6521 | /* |
6522 | * We have a problem here : there is no easy way to give a CHS | |
6523 | * virtual geometry. We currently pretend that we have a 2 heads | |
6524 | * 4 sectors (with a BIG number of cylinders...). This drives | |
6525 | * dosfs just mad... ;-) | |
6526 | */ | |
a885c8c4 CH |
6527 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
6528 | { | |
fd01b88c | 6529 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
6530 | |
6531 | geo->heads = 2; | |
6532 | geo->sectors = 4; | |
49ce6cea | 6533 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
6534 | return 0; |
6535 | } | |
6536 | ||
cb335f88 NS |
6537 | static inline bool md_ioctl_valid(unsigned int cmd) |
6538 | { | |
6539 | switch (cmd) { | |
6540 | case ADD_NEW_DISK: | |
6541 | case BLKROSET: | |
6542 | case GET_ARRAY_INFO: | |
6543 | case GET_BITMAP_FILE: | |
6544 | case GET_DISK_INFO: | |
6545 | case HOT_ADD_DISK: | |
6546 | case HOT_REMOVE_DISK: | |
cb335f88 NS |
6547 | case RAID_AUTORUN: |
6548 | case RAID_VERSION: | |
6549 | case RESTART_ARRAY_RW: | |
6550 | case RUN_ARRAY: | |
6551 | case SET_ARRAY_INFO: | |
6552 | case SET_BITMAP_FILE: | |
6553 | case SET_DISK_FAULTY: | |
6554 | case STOP_ARRAY: | |
6555 | case STOP_ARRAY_RO: | |
1aee41f6 | 6556 | case CLUSTERED_DISK_NACK: |
cb335f88 NS |
6557 | return true; |
6558 | default: | |
6559 | return false; | |
6560 | } | |
6561 | } | |
6562 | ||
a39907fa | 6563 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
6564 | unsigned int cmd, unsigned long arg) |
6565 | { | |
6566 | int err = 0; | |
6567 | void __user *argp = (void __user *)arg; | |
fd01b88c | 6568 | struct mddev *mddev = NULL; |
e2218350 | 6569 | int ro; |
1da177e4 | 6570 | |
cb335f88 NS |
6571 | if (!md_ioctl_valid(cmd)) |
6572 | return -ENOTTY; | |
6573 | ||
506c9e44 N |
6574 | switch (cmd) { |
6575 | case RAID_VERSION: | |
6576 | case GET_ARRAY_INFO: | |
6577 | case GET_DISK_INFO: | |
6578 | break; | |
6579 | default: | |
6580 | if (!capable(CAP_SYS_ADMIN)) | |
6581 | return -EACCES; | |
6582 | } | |
1da177e4 LT |
6583 | |
6584 | /* | |
6585 | * Commands dealing with the RAID driver but not any | |
6586 | * particular array: | |
6587 | */ | |
c02c0aeb N |
6588 | switch (cmd) { |
6589 | case RAID_VERSION: | |
6590 | err = get_version(argp); | |
3adc28d8 | 6591 | goto out; |
1da177e4 | 6592 | |
1da177e4 | 6593 | #ifndef MODULE |
c02c0aeb N |
6594 | case RAID_AUTORUN: |
6595 | err = 0; | |
6596 | autostart_arrays(arg); | |
3adc28d8 | 6597 | goto out; |
1da177e4 | 6598 | #endif |
c02c0aeb | 6599 | default:; |
1da177e4 LT |
6600 | } |
6601 | ||
6602 | /* | |
6603 | * Commands creating/starting a new array: | |
6604 | */ | |
6605 | ||
a39907fa | 6606 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
6607 | |
6608 | if (!mddev) { | |
6609 | BUG(); | |
3adc28d8 | 6610 | goto out; |
1da177e4 LT |
6611 | } |
6612 | ||
1ca69c4b N |
6613 | /* Some actions do not requires the mutex */ |
6614 | switch (cmd) { | |
6615 | case GET_ARRAY_INFO: | |
6616 | if (!mddev->raid_disks && !mddev->external) | |
6617 | err = -ENODEV; | |
6618 | else | |
6619 | err = get_array_info(mddev, argp); | |
3adc28d8 | 6620 | goto out; |
1ca69c4b N |
6621 | |
6622 | case GET_DISK_INFO: | |
6623 | if (!mddev->raid_disks && !mddev->external) | |
6624 | err = -ENODEV; | |
6625 | else | |
6626 | err = get_disk_info(mddev, argp); | |
3adc28d8 | 6627 | goto out; |
1ca69c4b N |
6628 | |
6629 | case SET_DISK_FAULTY: | |
6630 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6631 | goto out; |
4af1a041 N |
6632 | |
6633 | case GET_BITMAP_FILE: | |
6634 | err = get_bitmap_file(mddev, argp); | |
6635 | goto out; | |
6636 | ||
1ca69c4b N |
6637 | } |
6638 | ||
a7a3f08d N |
6639 | if (cmd == ADD_NEW_DISK) |
6640 | /* need to ensure md_delayed_delete() has completed */ | |
6641 | flush_workqueue(md_misc_wq); | |
6642 | ||
90f5f7ad HR |
6643 | if (cmd == HOT_REMOVE_DISK) |
6644 | /* need to ensure recovery thread has run */ | |
6645 | wait_event_interruptible_timeout(mddev->sb_wait, | |
6646 | !test_bit(MD_RECOVERY_NEEDED, | |
6647 | &mddev->flags), | |
6648 | msecs_to_jiffies(5000)); | |
260fa034 N |
6649 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
6650 | /* Need to flush page cache, and ensure no-one else opens | |
6651 | * and writes | |
6652 | */ | |
6653 | mutex_lock(&mddev->open_mutex); | |
9ba3b7f5 | 6654 | if (mddev->pers && atomic_read(&mddev->openers) > 1) { |
260fa034 N |
6655 | mutex_unlock(&mddev->open_mutex); |
6656 | err = -EBUSY; | |
3adc28d8 | 6657 | goto out; |
260fa034 N |
6658 | } |
6659 | set_bit(MD_STILL_CLOSED, &mddev->flags); | |
6660 | mutex_unlock(&mddev->open_mutex); | |
6661 | sync_blockdev(bdev); | |
6662 | } | |
1da177e4 LT |
6663 | err = mddev_lock(mddev); |
6664 | if (err) { | |
f72ffdd6 | 6665 | printk(KERN_INFO |
1da177e4 LT |
6666 | "md: ioctl lock interrupted, reason %d, cmd %d\n", |
6667 | err, cmd); | |
3adc28d8 | 6668 | goto out; |
1da177e4 LT |
6669 | } |
6670 | ||
c02c0aeb N |
6671 | if (cmd == SET_ARRAY_INFO) { |
6672 | mdu_array_info_t info; | |
6673 | if (!arg) | |
6674 | memset(&info, 0, sizeof(info)); | |
6675 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
6676 | err = -EFAULT; | |
3adc28d8 | 6677 | goto unlock; |
c02c0aeb N |
6678 | } |
6679 | if (mddev->pers) { | |
6680 | err = update_array_info(mddev, &info); | |
6681 | if (err) { | |
6682 | printk(KERN_WARNING "md: couldn't update" | |
6683 | " array info. %d\n", err); | |
3adc28d8 | 6684 | goto unlock; |
1da177e4 | 6685 | } |
3adc28d8 | 6686 | goto unlock; |
c02c0aeb N |
6687 | } |
6688 | if (!list_empty(&mddev->disks)) { | |
6689 | printk(KERN_WARNING | |
6690 | "md: array %s already has disks!\n", | |
6691 | mdname(mddev)); | |
6692 | err = -EBUSY; | |
3adc28d8 | 6693 | goto unlock; |
c02c0aeb N |
6694 | } |
6695 | if (mddev->raid_disks) { | |
6696 | printk(KERN_WARNING | |
6697 | "md: array %s already initialised!\n", | |
6698 | mdname(mddev)); | |
6699 | err = -EBUSY; | |
3adc28d8 | 6700 | goto unlock; |
c02c0aeb N |
6701 | } |
6702 | err = set_array_info(mddev, &info); | |
6703 | if (err) { | |
6704 | printk(KERN_WARNING "md: couldn't set" | |
6705 | " array info. %d\n", err); | |
3adc28d8 | 6706 | goto unlock; |
c02c0aeb | 6707 | } |
3adc28d8 | 6708 | goto unlock; |
1da177e4 LT |
6709 | } |
6710 | ||
6711 | /* | |
6712 | * Commands querying/configuring an existing array: | |
6713 | */ | |
32a7627c | 6714 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 6715 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
6716 | if ((!mddev->raid_disks && !mddev->external) |
6717 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
6718 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
6719 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 | 6720 | err = -ENODEV; |
3adc28d8 | 6721 | goto unlock; |
1da177e4 LT |
6722 | } |
6723 | ||
6724 | /* | |
6725 | * Commands even a read-only array can execute: | |
6726 | */ | |
c02c0aeb | 6727 | switch (cmd) { |
c02c0aeb N |
6728 | case RESTART_ARRAY_RW: |
6729 | err = restart_array(mddev); | |
3adc28d8 | 6730 | goto unlock; |
1da177e4 | 6731 | |
c02c0aeb N |
6732 | case STOP_ARRAY: |
6733 | err = do_md_stop(mddev, 0, bdev); | |
3adc28d8 | 6734 | goto unlock; |
1da177e4 | 6735 | |
c02c0aeb N |
6736 | case STOP_ARRAY_RO: |
6737 | err = md_set_readonly(mddev, bdev); | |
3adc28d8 | 6738 | goto unlock; |
1da177e4 | 6739 | |
3ea8929d N |
6740 | case HOT_REMOVE_DISK: |
6741 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6742 | goto unlock; |
3ea8929d | 6743 | |
7ceb17e8 N |
6744 | case ADD_NEW_DISK: |
6745 | /* We can support ADD_NEW_DISK on read-only arrays | |
6746 | * on if we are re-adding a preexisting device. | |
6747 | * So require mddev->pers and MD_DISK_SYNC. | |
6748 | */ | |
6749 | if (mddev->pers) { | |
6750 | mdu_disk_info_t info; | |
6751 | if (copy_from_user(&info, argp, sizeof(info))) | |
6752 | err = -EFAULT; | |
6753 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
6754 | /* Need to clear read-only for this */ | |
6755 | break; | |
6756 | else | |
6757 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 6758 | goto unlock; |
7ceb17e8 N |
6759 | } |
6760 | break; | |
6761 | ||
c02c0aeb N |
6762 | case BLKROSET: |
6763 | if (get_user(ro, (int __user *)(arg))) { | |
6764 | err = -EFAULT; | |
3adc28d8 | 6765 | goto unlock; |
c02c0aeb N |
6766 | } |
6767 | err = -EINVAL; | |
e2218350 | 6768 | |
c02c0aeb N |
6769 | /* if the bdev is going readonly the value of mddev->ro |
6770 | * does not matter, no writes are coming | |
6771 | */ | |
6772 | if (ro) | |
3adc28d8 | 6773 | goto unlock; |
e2218350 | 6774 | |
c02c0aeb N |
6775 | /* are we are already prepared for writes? */ |
6776 | if (mddev->ro != 1) | |
3adc28d8 | 6777 | goto unlock; |
e2218350 | 6778 | |
c02c0aeb N |
6779 | /* transitioning to readauto need only happen for |
6780 | * arrays that call md_write_start | |
6781 | */ | |
6782 | if (mddev->pers) { | |
6783 | err = restart_array(mddev); | |
6784 | if (err == 0) { | |
6785 | mddev->ro = 2; | |
6786 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 6787 | } |
c02c0aeb | 6788 | } |
3adc28d8 | 6789 | goto unlock; |
1da177e4 LT |
6790 | } |
6791 | ||
6792 | /* | |
6793 | * The remaining ioctls are changing the state of the | |
f91de92e | 6794 | * superblock, so we do not allow them on read-only arrays. |
1da177e4 | 6795 | */ |
326eb17d | 6796 | if (mddev->ro && mddev->pers) { |
f91de92e N |
6797 | if (mddev->ro == 2) { |
6798 | mddev->ro = 0; | |
00bcb4ac | 6799 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 6800 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
6801 | /* mddev_unlock will wake thread */ |
6802 | /* If a device failed while we were read-only, we | |
6803 | * need to make sure the metadata is updated now. | |
6804 | */ | |
6805 | if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
6806 | mddev_unlock(mddev); | |
6807 | wait_event(mddev->sb_wait, | |
6808 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) && | |
6809 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
29f097c4 | 6810 | mddev_lock_nointr(mddev); |
f3378b48 | 6811 | } |
f91de92e N |
6812 | } else { |
6813 | err = -EROFS; | |
3adc28d8 | 6814 | goto unlock; |
f91de92e | 6815 | } |
1da177e4 LT |
6816 | } |
6817 | ||
c02c0aeb N |
6818 | switch (cmd) { |
6819 | case ADD_NEW_DISK: | |
1da177e4 | 6820 | { |
c02c0aeb N |
6821 | mdu_disk_info_t info; |
6822 | if (copy_from_user(&info, argp, sizeof(info))) | |
6823 | err = -EFAULT; | |
6824 | else | |
6825 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 6826 | goto unlock; |
c02c0aeb | 6827 | } |
1da177e4 | 6828 | |
1aee41f6 GR |
6829 | case CLUSTERED_DISK_NACK: |
6830 | if (mddev_is_clustered(mddev)) | |
6831 | md_cluster_ops->new_disk_ack(mddev, false); | |
6832 | else | |
6833 | err = -EINVAL; | |
6834 | goto unlock; | |
6835 | ||
c02c0aeb N |
6836 | case HOT_ADD_DISK: |
6837 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6838 | goto unlock; |
1da177e4 | 6839 | |
c02c0aeb N |
6840 | case RUN_ARRAY: |
6841 | err = do_md_run(mddev); | |
3adc28d8 | 6842 | goto unlock; |
1da177e4 | 6843 | |
c02c0aeb N |
6844 | case SET_BITMAP_FILE: |
6845 | err = set_bitmap_file(mddev, (int)arg); | |
3adc28d8 | 6846 | goto unlock; |
32a7627c | 6847 | |
c02c0aeb N |
6848 | default: |
6849 | err = -EINVAL; | |
3adc28d8 | 6850 | goto unlock; |
1da177e4 LT |
6851 | } |
6852 | ||
3adc28d8 | 6853 | unlock: |
d3374825 N |
6854 | if (mddev->hold_active == UNTIL_IOCTL && |
6855 | err != -EINVAL) | |
6856 | mddev->hold_active = 0; | |
1da177e4 | 6857 | mddev_unlock(mddev); |
3adc28d8 | 6858 | out: |
1da177e4 LT |
6859 | return err; |
6860 | } | |
aa98aa31 AB |
6861 | #ifdef CONFIG_COMPAT |
6862 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
6863 | unsigned int cmd, unsigned long arg) | |
6864 | { | |
6865 | switch (cmd) { | |
6866 | case HOT_REMOVE_DISK: | |
6867 | case HOT_ADD_DISK: | |
6868 | case SET_DISK_FAULTY: | |
6869 | case SET_BITMAP_FILE: | |
6870 | /* These take in integer arg, do not convert */ | |
6871 | break; | |
6872 | default: | |
6873 | arg = (unsigned long)compat_ptr(arg); | |
6874 | break; | |
6875 | } | |
6876 | ||
6877 | return md_ioctl(bdev, mode, cmd, arg); | |
6878 | } | |
6879 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 6880 | |
a39907fa | 6881 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
6882 | { |
6883 | /* | |
6884 | * Succeed if we can lock the mddev, which confirms that | |
6885 | * it isn't being stopped right now. | |
6886 | */ | |
fd01b88c | 6887 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
6888 | int err; |
6889 | ||
0c098220 YL |
6890 | if (!mddev) |
6891 | return -ENODEV; | |
6892 | ||
d3374825 N |
6893 | if (mddev->gendisk != bdev->bd_disk) { |
6894 | /* we are racing with mddev_put which is discarding this | |
6895 | * bd_disk. | |
6896 | */ | |
6897 | mddev_put(mddev); | |
6898 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 6899 | flush_workqueue(md_misc_wq); |
d3374825 N |
6900 | /* Then retry the open from the top */ |
6901 | return -ERESTARTSYS; | |
6902 | } | |
6903 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
6904 | ||
c8c00a69 | 6905 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6906 | goto out; |
6907 | ||
6908 | err = 0; | |
f2ea68cf | 6909 | atomic_inc(&mddev->openers); |
260fa034 | 6910 | clear_bit(MD_STILL_CLOSED, &mddev->flags); |
c8c00a69 | 6911 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6912 | |
f0b4f7e2 | 6913 | check_disk_change(bdev); |
1da177e4 LT |
6914 | out: |
6915 | return err; | |
6916 | } | |
6917 | ||
db2a144b | 6918 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6919 | { |
f72ffdd6 | 6920 | struct mddev *mddev = disk->private_data; |
1da177e4 | 6921 | |
52e5f9d1 | 6922 | BUG_ON(!mddev); |
f2ea68cf | 6923 | atomic_dec(&mddev->openers); |
1da177e4 | 6924 | mddev_put(mddev); |
1da177e4 | 6925 | } |
f0b4f7e2 N |
6926 | |
6927 | static int md_media_changed(struct gendisk *disk) | |
6928 | { | |
fd01b88c | 6929 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6930 | |
6931 | return mddev->changed; | |
6932 | } | |
6933 | ||
6934 | static int md_revalidate(struct gendisk *disk) | |
6935 | { | |
fd01b88c | 6936 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6937 | |
6938 | mddev->changed = 0; | |
6939 | return 0; | |
6940 | } | |
83d5cde4 | 6941 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6942 | { |
6943 | .owner = THIS_MODULE, | |
a39907fa AV |
6944 | .open = md_open, |
6945 | .release = md_release, | |
b492b852 | 6946 | .ioctl = md_ioctl, |
aa98aa31 AB |
6947 | #ifdef CONFIG_COMPAT |
6948 | .compat_ioctl = md_compat_ioctl, | |
6949 | #endif | |
a885c8c4 | 6950 | .getgeo = md_getgeo, |
f0b4f7e2 N |
6951 | .media_changed = md_media_changed, |
6952 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
6953 | }; |
6954 | ||
f72ffdd6 | 6955 | static int md_thread(void *arg) |
1da177e4 | 6956 | { |
2b8bf345 | 6957 | struct md_thread *thread = arg; |
1da177e4 | 6958 | |
1da177e4 LT |
6959 | /* |
6960 | * md_thread is a 'system-thread', it's priority should be very | |
6961 | * high. We avoid resource deadlocks individually in each | |
6962 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6963 | * the very same RT priority as kswapd, thus we will never get | |
6964 | * into a priority inversion deadlock. | |
6965 | * | |
6966 | * we definitely have to have equal or higher priority than | |
6967 | * bdflush, otherwise bdflush will deadlock if there are too | |
6968 | * many dirty RAID5 blocks. | |
6969 | */ | |
1da177e4 | 6970 | |
6985c43f | 6971 | allow_signal(SIGKILL); |
a6fb0934 | 6972 | while (!kthread_should_stop()) { |
1da177e4 | 6973 | |
93588e22 N |
6974 | /* We need to wait INTERRUPTIBLE so that |
6975 | * we don't add to the load-average. | |
6976 | * That means we need to be sure no signals are | |
6977 | * pending | |
6978 | */ | |
6979 | if (signal_pending(current)) | |
6980 | flush_signals(current); | |
6981 | ||
6982 | wait_event_interruptible_timeout | |
6983 | (thread->wqueue, | |
6984 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6985 | || kthread_should_stop(), | |
6986 | thread->timeout); | |
1da177e4 | 6987 | |
6c987910 N |
6988 | clear_bit(THREAD_WAKEUP, &thread->flags); |
6989 | if (!kthread_should_stop()) | |
4ed8731d | 6990 | thread->run(thread); |
1da177e4 | 6991 | } |
a6fb0934 | 6992 | |
1da177e4 LT |
6993 | return 0; |
6994 | } | |
6995 | ||
2b8bf345 | 6996 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
6997 | { |
6998 | if (thread) { | |
36a4e1fe | 6999 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
1da177e4 LT |
7000 | set_bit(THREAD_WAKEUP, &thread->flags); |
7001 | wake_up(&thread->wqueue); | |
7002 | } | |
7003 | } | |
6c144d31 | 7004 | EXPORT_SYMBOL(md_wakeup_thread); |
1da177e4 | 7005 | |
4ed8731d SL |
7006 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
7007 | struct mddev *mddev, const char *name) | |
1da177e4 | 7008 | { |
2b8bf345 | 7009 | struct md_thread *thread; |
1da177e4 | 7010 | |
2b8bf345 | 7011 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
7012 | if (!thread) |
7013 | return NULL; | |
7014 | ||
1da177e4 LT |
7015 | init_waitqueue_head(&thread->wqueue); |
7016 | ||
1da177e4 LT |
7017 | thread->run = run; |
7018 | thread->mddev = mddev; | |
32a7627c | 7019 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
7020 | thread->tsk = kthread_run(md_thread, thread, |
7021 | "%s_%s", | |
7022 | mdname(thread->mddev), | |
0232605d | 7023 | name); |
a6fb0934 | 7024 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
7025 | kfree(thread); |
7026 | return NULL; | |
7027 | } | |
1da177e4 LT |
7028 | return thread; |
7029 | } | |
6c144d31 | 7030 | EXPORT_SYMBOL(md_register_thread); |
1da177e4 | 7031 | |
2b8bf345 | 7032 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 7033 | { |
2b8bf345 | 7034 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
7035 | if (!thread) |
7036 | return; | |
36a4e1fe | 7037 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
7038 | /* Locking ensures that mddev_unlock does not wake_up a |
7039 | * non-existent thread | |
7040 | */ | |
7041 | spin_lock(&pers_lock); | |
7042 | *threadp = NULL; | |
7043 | spin_unlock(&pers_lock); | |
a6fb0934 N |
7044 | |
7045 | kthread_stop(thread->tsk); | |
1da177e4 LT |
7046 | kfree(thread); |
7047 | } | |
6c144d31 | 7048 | EXPORT_SYMBOL(md_unregister_thread); |
1da177e4 | 7049 | |
fd01b88c | 7050 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7051 | { |
b2d444d7 | 7052 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 7053 | return; |
6bfe0b49 | 7054 | |
de393cde | 7055 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
7056 | return; |
7057 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
7058 | if (mddev->degraded) |
7059 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 7060 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
7061 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7062 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7063 | md_wakeup_thread(mddev->thread); | |
768a418d | 7064 | if (mddev->event_work.func) |
e804ac78 | 7065 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 7066 | md_new_event_inintr(mddev); |
1da177e4 | 7067 | } |
6c144d31 | 7068 | EXPORT_SYMBOL(md_error); |
1da177e4 LT |
7069 | |
7070 | /* seq_file implementation /proc/mdstat */ | |
7071 | ||
7072 | static void status_unused(struct seq_file *seq) | |
7073 | { | |
7074 | int i = 0; | |
3cb03002 | 7075 | struct md_rdev *rdev; |
1da177e4 LT |
7076 | |
7077 | seq_printf(seq, "unused devices: "); | |
7078 | ||
159ec1fc | 7079 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
7080 | char b[BDEVNAME_SIZE]; |
7081 | i++; | |
7082 | seq_printf(seq, "%s ", | |
7083 | bdevname(rdev->bdev,b)); | |
7084 | } | |
7085 | if (!i) | |
7086 | seq_printf(seq, "<none>"); | |
7087 | ||
7088 | seq_printf(seq, "\n"); | |
7089 | } | |
7090 | ||
f72ffdd6 | 7091 | static void status_resync(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7092 | { |
dd71cf6b N |
7093 | sector_t max_sectors, resync, res; |
7094 | unsigned long dt, db; | |
7095 | sector_t rt; | |
4588b42e N |
7096 | int scale; |
7097 | unsigned int per_milli; | |
1da177e4 | 7098 | |
72f36d59 N |
7099 | if (mddev->curr_resync <= 3) |
7100 | resync = 0; | |
7101 | else | |
7102 | resync = mddev->curr_resync | |
7103 | - atomic_read(&mddev->recovery_active); | |
1da177e4 | 7104 | |
c804cdec N |
7105 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
7106 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 7107 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 7108 | else |
dd71cf6b | 7109 | max_sectors = mddev->dev_sectors; |
1da177e4 | 7110 | |
403df478 | 7111 | WARN_ON(max_sectors == 0); |
4588b42e | 7112 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 7113 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
7114 | * u32, as those are the requirements for sector_div. |
7115 | * Thus 'scale' must be at least 10 | |
7116 | */ | |
7117 | scale = 10; | |
7118 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 7119 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
7120 | scale++; |
7121 | } | |
7122 | res = (resync>>scale)*1000; | |
dd71cf6b | 7123 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
7124 | |
7125 | per_milli = res; | |
1da177e4 | 7126 | { |
4588b42e | 7127 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
7128 | seq_printf(seq, "["); |
7129 | for (i = 0; i < x; i++) | |
7130 | seq_printf(seq, "="); | |
7131 | seq_printf(seq, ">"); | |
7132 | for (i = 0; i < y; i++) | |
7133 | seq_printf(seq, "."); | |
7134 | seq_printf(seq, "] "); | |
7135 | } | |
4588b42e | 7136 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
7137 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
7138 | "reshape" : | |
61df9d91 N |
7139 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
7140 | "check" : | |
7141 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
7142 | "resync" : "recovery"))), | |
7143 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
7144 | (unsigned long long) resync/2, |
7145 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
7146 | |
7147 | /* | |
1da177e4 LT |
7148 | * dt: time from mark until now |
7149 | * db: blocks written from mark until now | |
7150 | * rt: remaining time | |
dd71cf6b N |
7151 | * |
7152 | * rt is a sector_t, so could be 32bit or 64bit. | |
7153 | * So we divide before multiply in case it is 32bit and close | |
7154 | * to the limit. | |
25985edc | 7155 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
7156 | * near the end of resync when the number of remaining sectors |
7157 | * is close to 'db'. | |
7158 | * We then divide rt by 32 after multiplying by db to compensate. | |
7159 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
7160 | */ |
7161 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
7162 | if (!dt) dt++; | |
ff4e8d9a N |
7163 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
7164 | - mddev->resync_mark_cnt; | |
1da177e4 | 7165 | |
dd71cf6b N |
7166 | rt = max_sectors - resync; /* number of remaining sectors */ |
7167 | sector_div(rt, db/32+1); | |
7168 | rt *= dt; | |
7169 | rt >>= 5; | |
7170 | ||
7171 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
7172 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 7173 | |
ff4e8d9a | 7174 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
7175 | } |
7176 | ||
7177 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
7178 | { | |
7179 | struct list_head *tmp; | |
7180 | loff_t l = *pos; | |
fd01b88c | 7181 | struct mddev *mddev; |
1da177e4 LT |
7182 | |
7183 | if (l >= 0x10000) | |
7184 | return NULL; | |
7185 | if (!l--) | |
7186 | /* header */ | |
7187 | return (void*)1; | |
7188 | ||
7189 | spin_lock(&all_mddevs_lock); | |
7190 | list_for_each(tmp,&all_mddevs) | |
7191 | if (!l--) { | |
fd01b88c | 7192 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
7193 | mddev_get(mddev); |
7194 | spin_unlock(&all_mddevs_lock); | |
7195 | return mddev; | |
7196 | } | |
7197 | spin_unlock(&all_mddevs_lock); | |
7198 | if (!l--) | |
7199 | return (void*)2;/* tail */ | |
7200 | return NULL; | |
7201 | } | |
7202 | ||
7203 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
7204 | { | |
7205 | struct list_head *tmp; | |
fd01b88c | 7206 | struct mddev *next_mddev, *mddev = v; |
f72ffdd6 | 7207 | |
1da177e4 LT |
7208 | ++*pos; |
7209 | if (v == (void*)2) | |
7210 | return NULL; | |
7211 | ||
7212 | spin_lock(&all_mddevs_lock); | |
7213 | if (v == (void*)1) | |
7214 | tmp = all_mddevs.next; | |
7215 | else | |
7216 | tmp = mddev->all_mddevs.next; | |
7217 | if (tmp != &all_mddevs) | |
fd01b88c | 7218 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
7219 | else { |
7220 | next_mddev = (void*)2; | |
7221 | *pos = 0x10000; | |
f72ffdd6 | 7222 | } |
1da177e4 LT |
7223 | spin_unlock(&all_mddevs_lock); |
7224 | ||
7225 | if (v != (void*)1) | |
7226 | mddev_put(mddev); | |
7227 | return next_mddev; | |
7228 | ||
7229 | } | |
7230 | ||
7231 | static void md_seq_stop(struct seq_file *seq, void *v) | |
7232 | { | |
fd01b88c | 7233 | struct mddev *mddev = v; |
1da177e4 LT |
7234 | |
7235 | if (mddev && v != (void*)1 && v != (void*)2) | |
7236 | mddev_put(mddev); | |
7237 | } | |
7238 | ||
7239 | static int md_seq_show(struct seq_file *seq, void *v) | |
7240 | { | |
fd01b88c | 7241 | struct mddev *mddev = v; |
dd8ac336 | 7242 | sector_t sectors; |
3cb03002 | 7243 | struct md_rdev *rdev; |
1da177e4 LT |
7244 | |
7245 | if (v == (void*)1) { | |
84fc4b56 | 7246 | struct md_personality *pers; |
1da177e4 LT |
7247 | seq_printf(seq, "Personalities : "); |
7248 | spin_lock(&pers_lock); | |
2604b703 N |
7249 | list_for_each_entry(pers, &pers_list, list) |
7250 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
7251 | |
7252 | spin_unlock(&pers_lock); | |
7253 | seq_printf(seq, "\n"); | |
f1514638 | 7254 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
7255 | return 0; |
7256 | } | |
7257 | if (v == (void*)2) { | |
7258 | status_unused(seq); | |
7259 | return 0; | |
7260 | } | |
7261 | ||
36d091f4 | 7262 | spin_lock(&mddev->lock); |
1da177e4 LT |
7263 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
7264 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
7265 | mddev->pers ? "" : "in"); | |
7266 | if (mddev->pers) { | |
f91de92e | 7267 | if (mddev->ro==1) |
1da177e4 | 7268 | seq_printf(seq, " (read-only)"); |
f91de92e | 7269 | if (mddev->ro==2) |
52720ae7 | 7270 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
7271 | seq_printf(seq, " %s", mddev->pers->name); |
7272 | } | |
7273 | ||
dd8ac336 | 7274 | sectors = 0; |
f97fcad3 N |
7275 | rcu_read_lock(); |
7276 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 LT |
7277 | char b[BDEVNAME_SIZE]; |
7278 | seq_printf(seq, " %s[%d]", | |
7279 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
7280 | if (test_bit(WriteMostly, &rdev->flags)) |
7281 | seq_printf(seq, "(W)"); | |
b2d444d7 | 7282 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
7283 | seq_printf(seq, "(F)"); |
7284 | continue; | |
2d78f8c4 N |
7285 | } |
7286 | if (rdev->raid_disk < 0) | |
b325a32e | 7287 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
7288 | if (test_bit(Replacement, &rdev->flags)) |
7289 | seq_printf(seq, "(R)"); | |
dd8ac336 | 7290 | sectors += rdev->sectors; |
1da177e4 | 7291 | } |
f97fcad3 | 7292 | rcu_read_unlock(); |
1da177e4 LT |
7293 | |
7294 | if (!list_empty(&mddev->disks)) { | |
7295 | if (mddev->pers) | |
7296 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
7297 | (unsigned long long) |
7298 | mddev->array_sectors / 2); | |
1da177e4 LT |
7299 | else |
7300 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 7301 | (unsigned long long)sectors / 2); |
1da177e4 | 7302 | } |
1cd6bf19 N |
7303 | if (mddev->persistent) { |
7304 | if (mddev->major_version != 0 || | |
7305 | mddev->minor_version != 90) { | |
7306 | seq_printf(seq," super %d.%d", | |
7307 | mddev->major_version, | |
7308 | mddev->minor_version); | |
7309 | } | |
e691063a N |
7310 | } else if (mddev->external) |
7311 | seq_printf(seq, " super external:%s", | |
7312 | mddev->metadata_type); | |
7313 | else | |
1cd6bf19 | 7314 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
7315 | |
7316 | if (mddev->pers) { | |
d710e138 | 7317 | mddev->pers->status(seq, mddev); |
f72ffdd6 | 7318 | seq_printf(seq, "\n "); |
8e1b39d6 N |
7319 | if (mddev->pers->sync_request) { |
7320 | if (mddev->curr_resync > 2) { | |
d710e138 | 7321 | status_resync(seq, mddev); |
8e1b39d6 | 7322 | seq_printf(seq, "\n "); |
72f36d59 | 7323 | } else if (mddev->curr_resync >= 1) |
8e1b39d6 N |
7324 | seq_printf(seq, "\tresync=DELAYED\n "); |
7325 | else if (mddev->recovery_cp < MaxSector) | |
7326 | seq_printf(seq, "\tresync=PENDING\n "); | |
7327 | } | |
32a7627c N |
7328 | } else |
7329 | seq_printf(seq, "\n "); | |
7330 | ||
57148964 | 7331 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
7332 | |
7333 | seq_printf(seq, "\n"); | |
7334 | } | |
36d091f4 | 7335 | spin_unlock(&mddev->lock); |
f72ffdd6 | 7336 | |
1da177e4 LT |
7337 | return 0; |
7338 | } | |
7339 | ||
110518bc | 7340 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
7341 | .start = md_seq_start, |
7342 | .next = md_seq_next, | |
7343 | .stop = md_seq_stop, | |
7344 | .show = md_seq_show, | |
7345 | }; | |
7346 | ||
7347 | static int md_seq_open(struct inode *inode, struct file *file) | |
7348 | { | |
f1514638 | 7349 | struct seq_file *seq; |
1da177e4 LT |
7350 | int error; |
7351 | ||
7352 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 7353 | if (error) |
f1514638 KS |
7354 | return error; |
7355 | ||
7356 | seq = file->private_data; | |
7357 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
7358 | return error; |
7359 | } | |
7360 | ||
e2f23b60 | 7361 | static int md_unloading; |
d7603b7e N |
7362 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
7363 | { | |
f1514638 | 7364 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
7365 | int mask; |
7366 | ||
e2f23b60 | 7367 | if (md_unloading) |
7d7e64f2 | 7368 | return POLLIN|POLLRDNORM|POLLERR|POLLPRI; |
d7603b7e N |
7369 | poll_wait(filp, &md_event_waiters, wait); |
7370 | ||
7371 | /* always allow read */ | |
7372 | mask = POLLIN | POLLRDNORM; | |
7373 | ||
f1514638 | 7374 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
7375 | mask |= POLLERR | POLLPRI; |
7376 | return mask; | |
7377 | } | |
7378 | ||
fa027c2a | 7379 | static const struct file_operations md_seq_fops = { |
e24650c2 | 7380 | .owner = THIS_MODULE, |
1da177e4 LT |
7381 | .open = md_seq_open, |
7382 | .read = seq_read, | |
7383 | .llseek = seq_lseek, | |
c3f94b40 | 7384 | .release = seq_release_private, |
d7603b7e | 7385 | .poll = mdstat_poll, |
1da177e4 LT |
7386 | }; |
7387 | ||
84fc4b56 | 7388 | int register_md_personality(struct md_personality *p) |
1da177e4 | 7389 | { |
50bd3774 CY |
7390 | printk(KERN_INFO "md: %s personality registered for level %d\n", |
7391 | p->name, p->level); | |
1da177e4 | 7392 | spin_lock(&pers_lock); |
2604b703 | 7393 | list_add_tail(&p->list, &pers_list); |
1da177e4 LT |
7394 | spin_unlock(&pers_lock); |
7395 | return 0; | |
7396 | } | |
6c144d31 | 7397 | EXPORT_SYMBOL(register_md_personality); |
1da177e4 | 7398 | |
84fc4b56 | 7399 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7400 | { |
2604b703 | 7401 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 7402 | spin_lock(&pers_lock); |
2604b703 | 7403 | list_del_init(&p->list); |
1da177e4 LT |
7404 | spin_unlock(&pers_lock); |
7405 | return 0; | |
7406 | } | |
6c144d31 | 7407 | EXPORT_SYMBOL(unregister_md_personality); |
1da177e4 | 7408 | |
edb39c9d GR |
7409 | int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module) |
7410 | { | |
7411 | if (md_cluster_ops != NULL) | |
7412 | return -EALREADY; | |
7413 | spin_lock(&pers_lock); | |
7414 | md_cluster_ops = ops; | |
7415 | md_cluster_mod = module; | |
7416 | spin_unlock(&pers_lock); | |
7417 | return 0; | |
7418 | } | |
7419 | EXPORT_SYMBOL(register_md_cluster_operations); | |
7420 | ||
7421 | int unregister_md_cluster_operations(void) | |
7422 | { | |
7423 | spin_lock(&pers_lock); | |
7424 | md_cluster_ops = NULL; | |
7425 | spin_unlock(&pers_lock); | |
7426 | return 0; | |
7427 | } | |
7428 | EXPORT_SYMBOL(unregister_md_cluster_operations); | |
7429 | ||
7430 | int md_setup_cluster(struct mddev *mddev, int nodes) | |
7431 | { | |
7432 | int err; | |
7433 | ||
7434 | err = request_module("md-cluster"); | |
7435 | if (err) { | |
7436 | pr_err("md-cluster module not found.\n"); | |
7437 | return err; | |
7438 | } | |
7439 | ||
7440 | spin_lock(&pers_lock); | |
7441 | if (!md_cluster_ops || !try_module_get(md_cluster_mod)) { | |
7442 | spin_unlock(&pers_lock); | |
7443 | return -ENOENT; | |
7444 | } | |
7445 | spin_unlock(&pers_lock); | |
7446 | ||
cf921cc1 | 7447 | return md_cluster_ops->join(mddev, nodes); |
edb39c9d GR |
7448 | } |
7449 | ||
7450 | void md_cluster_stop(struct mddev *mddev) | |
7451 | { | |
c4ce867f GR |
7452 | if (!md_cluster_ops) |
7453 | return; | |
edb39c9d GR |
7454 | md_cluster_ops->leave(mddev); |
7455 | module_put(md_cluster_mod); | |
7456 | } | |
7457 | ||
fd01b88c | 7458 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7459 | { |
f72ffdd6 | 7460 | struct md_rdev *rdev; |
1da177e4 | 7461 | int idle; |
eea1bf38 | 7462 | int curr_events; |
1da177e4 LT |
7463 | |
7464 | idle = 1; | |
4b80991c N |
7465 | rcu_read_lock(); |
7466 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7467 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7468 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7469 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7470 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7471 | /* sync IO will cause sync_io to increase before the disk_stats |
7472 | * as sync_io is counted when a request starts, and | |
7473 | * disk_stats is counted when it completes. | |
7474 | * So resync activity will cause curr_events to be smaller than | |
7475 | * when there was no such activity. | |
7476 | * non-sync IO will cause disk_stat to increase without | |
7477 | * increasing sync_io so curr_events will (eventually) | |
7478 | * be larger than it was before. Once it becomes | |
7479 | * substantially larger, the test below will cause | |
7480 | * the array to appear non-idle, and resync will slow | |
7481 | * down. | |
7482 | * If there is a lot of outstanding resync activity when | |
7483 | * we set last_event to curr_events, then all that activity | |
7484 | * completing might cause the array to appear non-idle | |
7485 | * and resync will be slowed down even though there might | |
7486 | * not have been non-resync activity. This will only | |
7487 | * happen once though. 'last_events' will soon reflect | |
7488 | * the state where there is little or no outstanding | |
7489 | * resync requests, and further resync activity will | |
7490 | * always make curr_events less than last_events. | |
c0e48521 | 7491 | * |
1da177e4 | 7492 | */ |
eea1bf38 | 7493 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
7494 | rdev->last_events = curr_events; |
7495 | idle = 0; | |
7496 | } | |
7497 | } | |
4b80991c | 7498 | rcu_read_unlock(); |
1da177e4 LT |
7499 | return idle; |
7500 | } | |
7501 | ||
fd01b88c | 7502 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
7503 | { |
7504 | /* another "blocks" (512byte) blocks have been synced */ | |
7505 | atomic_sub(blocks, &mddev->recovery_active); | |
7506 | wake_up(&mddev->recovery_wait); | |
7507 | if (!ok) { | |
dfc70645 | 7508 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 7509 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
7510 | md_wakeup_thread(mddev->thread); |
7511 | // stop recovery, signal do_sync .... | |
7512 | } | |
7513 | } | |
6c144d31 | 7514 | EXPORT_SYMBOL(md_done_sync); |
1da177e4 | 7515 | |
06d91a5f N |
7516 | /* md_write_start(mddev, bi) |
7517 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
7518 | * in superblock) before writing, schedule a superblock update |
7519 | * and wait for it to complete. | |
06d91a5f | 7520 | */ |
fd01b88c | 7521 | void md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 7522 | { |
0fd62b86 | 7523 | int did_change = 0; |
06d91a5f | 7524 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 7525 | return; |
06d91a5f | 7526 | |
f91de92e N |
7527 | BUG_ON(mddev->ro == 1); |
7528 | if (mddev->ro == 2) { | |
7529 | /* need to switch to read/write */ | |
7530 | mddev->ro = 0; | |
7531 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7532 | md_wakeup_thread(mddev->thread); | |
25156198 | 7533 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 7534 | did_change = 1; |
f91de92e | 7535 | } |
06d91a5f | 7536 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
7537 | if (mddev->safemode == 1) |
7538 | mddev->safemode = 0; | |
06d91a5f | 7539 | if (mddev->in_sync) { |
85572d7c | 7540 | spin_lock(&mddev->lock); |
3d310eb7 N |
7541 | if (mddev->in_sync) { |
7542 | mddev->in_sync = 0; | |
850b2b42 | 7543 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 7544 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 7545 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 7546 | did_change = 1; |
3d310eb7 | 7547 | } |
85572d7c | 7548 | spin_unlock(&mddev->lock); |
06d91a5f | 7549 | } |
0fd62b86 | 7550 | if (did_change) |
00bcb4ac | 7551 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 7552 | wait_event(mddev->sb_wait, |
09a44cc1 | 7553 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 | 7554 | } |
6c144d31 | 7555 | EXPORT_SYMBOL(md_write_start); |
1da177e4 | 7556 | |
fd01b88c | 7557 | void md_write_end(struct mddev *mddev) |
1da177e4 LT |
7558 | { |
7559 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
7560 | if (mddev->safemode == 2) | |
7561 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 7562 | else if (mddev->safemode_delay) |
1da177e4 LT |
7563 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
7564 | } | |
7565 | } | |
6c144d31 | 7566 | EXPORT_SYMBOL(md_write_end); |
1da177e4 | 7567 | |
2a2275d6 N |
7568 | /* md_allow_write(mddev) |
7569 | * Calling this ensures that the array is marked 'active' so that writes | |
7570 | * may proceed without blocking. It is important to call this before | |
7571 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
7572 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
7573 | * |
7574 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
7575 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 7576 | */ |
fd01b88c | 7577 | int md_allow_write(struct mddev *mddev) |
2a2275d6 N |
7578 | { |
7579 | if (!mddev->pers) | |
b5470dc5 | 7580 | return 0; |
2a2275d6 | 7581 | if (mddev->ro) |
b5470dc5 | 7582 | return 0; |
1a0fd497 | 7583 | if (!mddev->pers->sync_request) |
b5470dc5 | 7584 | return 0; |
2a2275d6 | 7585 | |
85572d7c | 7586 | spin_lock(&mddev->lock); |
2a2275d6 N |
7587 | if (mddev->in_sync) { |
7588 | mddev->in_sync = 0; | |
7589 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 7590 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
7591 | if (mddev->safemode_delay && |
7592 | mddev->safemode == 0) | |
7593 | mddev->safemode = 1; | |
85572d7c | 7594 | spin_unlock(&mddev->lock); |
293467aa GR |
7595 | if (mddev_is_clustered(mddev)) |
7596 | md_cluster_ops->metadata_update_start(mddev); | |
2a2275d6 | 7597 | md_update_sb(mddev, 0); |
293467aa GR |
7598 | if (mddev_is_clustered(mddev)) |
7599 | md_cluster_ops->metadata_update_finish(mddev); | |
00bcb4ac | 7600 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 | 7601 | } else |
85572d7c | 7602 | spin_unlock(&mddev->lock); |
b5470dc5 | 7603 | |
070dc6dd | 7604 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
7605 | return -EAGAIN; |
7606 | else | |
7607 | return 0; | |
2a2275d6 N |
7608 | } |
7609 | EXPORT_SYMBOL_GPL(md_allow_write); | |
7610 | ||
1da177e4 LT |
7611 | #define SYNC_MARKS 10 |
7612 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 7613 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 7614 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 7615 | { |
4ed8731d | 7616 | struct mddev *mddev = thread->mddev; |
fd01b88c | 7617 | struct mddev *mddev2; |
1da177e4 LT |
7618 | unsigned int currspeed = 0, |
7619 | window; | |
ac7e50a3 | 7620 | sector_t max_sectors,j, io_sectors, recovery_done; |
1da177e4 | 7621 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 7622 | unsigned long update_time; |
1da177e4 LT |
7623 | sector_t mark_cnt[SYNC_MARKS]; |
7624 | int last_mark,m; | |
7625 | struct list_head *tmp; | |
7626 | sector_t last_check; | |
57afd89f | 7627 | int skipped = 0; |
3cb03002 | 7628 | struct md_rdev *rdev; |
c4a39551 | 7629 | char *desc, *action = NULL; |
7c2c57c9 | 7630 | struct blk_plug plug; |
1da177e4 LT |
7631 | |
7632 | /* just incase thread restarts... */ | |
7633 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7634 | return; | |
3991b31e N |
7635 | if (mddev->ro) {/* never try to sync a read-only array */ |
7636 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5fd6c1dc | 7637 | return; |
3991b31e | 7638 | } |
1da177e4 | 7639 | |
61df9d91 | 7640 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 7641 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 7642 | desc = "data-check"; |
c4a39551 JB |
7643 | action = "check"; |
7644 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 7645 | desc = "requested-resync"; |
c4a39551 JB |
7646 | action = "repair"; |
7647 | } else | |
61df9d91 N |
7648 | desc = "resync"; |
7649 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
7650 | desc = "reshape"; | |
7651 | else | |
7652 | desc = "recovery"; | |
7653 | ||
c4a39551 JB |
7654 | mddev->last_sync_action = action ?: desc; |
7655 | ||
1da177e4 LT |
7656 | /* we overload curr_resync somewhat here. |
7657 | * 0 == not engaged in resync at all | |
7658 | * 2 == checking that there is no conflict with another sync | |
7659 | * 1 == like 2, but have yielded to allow conflicting resync to | |
7660 | * commense | |
7661 | * other == active in resync - this many blocks | |
7662 | * | |
7663 | * Before starting a resync we must have set curr_resync to | |
7664 | * 2, and then checked that every "conflicting" array has curr_resync | |
7665 | * less than ours. When we find one that is the same or higher | |
7666 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
7667 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
7668 | * This will mean we have to start checking from the beginning again. | |
7669 | * | |
7670 | */ | |
7671 | ||
7672 | do { | |
7673 | mddev->curr_resync = 2; | |
7674 | ||
7675 | try_again: | |
404e4b43 | 7676 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 7677 | goto skip; |
29ac4aa3 | 7678 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
7679 | if (mddev2 == mddev) |
7680 | continue; | |
90b08710 BS |
7681 | if (!mddev->parallel_resync |
7682 | && mddev2->curr_resync | |
7683 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
7684 | DEFINE_WAIT(wq); |
7685 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
7686 | /* arbitrarily yield */ | |
7687 | mddev->curr_resync = 1; | |
7688 | wake_up(&resync_wait); | |
7689 | } | |
7690 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
7691 | /* no need to wait here, we can wait the next | |
7692 | * time 'round when curr_resync == 2 | |
7693 | */ | |
7694 | continue; | |
9744197c N |
7695 | /* We need to wait 'interruptible' so as not to |
7696 | * contribute to the load average, and not to | |
7697 | * be caught by 'softlockup' | |
7698 | */ | |
7699 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 7700 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 7701 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
7702 | printk(KERN_INFO "md: delaying %s of %s" |
7703 | " until %s has finished (they" | |
1da177e4 | 7704 | " share one or more physical units)\n", |
61df9d91 | 7705 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 7706 | mddev_put(mddev2); |
9744197c N |
7707 | if (signal_pending(current)) |
7708 | flush_signals(current); | |
1da177e4 LT |
7709 | schedule(); |
7710 | finish_wait(&resync_wait, &wq); | |
7711 | goto try_again; | |
7712 | } | |
7713 | finish_wait(&resync_wait, &wq); | |
7714 | } | |
7715 | } | |
7716 | } while (mddev->curr_resync < 2); | |
7717 | ||
5fd6c1dc | 7718 | j = 0; |
9d88883e | 7719 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 7720 | /* resync follows the size requested by the personality, |
57afd89f | 7721 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
7722 | */ |
7723 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 7724 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 7725 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
7726 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
7727 | j = mddev->resync_min; | |
7728 | else if (!mddev->bitmap) | |
5fd6c1dc | 7729 | j = mddev->recovery_cp; |
5e96ee65 | 7730 | |
ccfcc3c1 | 7731 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 7732 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 7733 | else { |
1da177e4 | 7734 | /* recovery follows the physical size of devices */ |
58c0fed4 | 7735 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 7736 | j = MaxSector; |
4e59ca7d | 7737 | rcu_read_lock(); |
dafb20fa | 7738 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc N |
7739 | if (rdev->raid_disk >= 0 && |
7740 | !test_bit(Faulty, &rdev->flags) && | |
7741 | !test_bit(In_sync, &rdev->flags) && | |
7742 | rdev->recovery_offset < j) | |
7743 | j = rdev->recovery_offset; | |
4e59ca7d | 7744 | rcu_read_unlock(); |
133d4527 N |
7745 | |
7746 | /* If there is a bitmap, we need to make sure all | |
7747 | * writes that started before we added a spare | |
7748 | * complete before we start doing a recovery. | |
7749 | * Otherwise the write might complete and (via | |
7750 | * bitmap_endwrite) set a bit in the bitmap after the | |
7751 | * recovery has checked that bit and skipped that | |
7752 | * region. | |
7753 | */ | |
7754 | if (mddev->bitmap) { | |
7755 | mddev->pers->quiesce(mddev, 1); | |
7756 | mddev->pers->quiesce(mddev, 0); | |
7757 | } | |
5fd6c1dc | 7758 | } |
1da177e4 | 7759 | |
61df9d91 N |
7760 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
7761 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
7762 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 7763 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
7764 | "(but not more than %d KB/sec) for %s.\n", |
7765 | speed_max(mddev), desc); | |
1da177e4 | 7766 | |
eea1bf38 | 7767 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 7768 | |
57afd89f | 7769 | io_sectors = 0; |
1da177e4 LT |
7770 | for (m = 0; m < SYNC_MARKS; m++) { |
7771 | mark[m] = jiffies; | |
57afd89f | 7772 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
7773 | } |
7774 | last_mark = 0; | |
7775 | mddev->resync_mark = mark[last_mark]; | |
7776 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
7777 | ||
7778 | /* | |
7779 | * Tune reconstruction: | |
7780 | */ | |
7781 | window = 32*(PAGE_SIZE/512); | |
ac42450c JB |
7782 | printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", |
7783 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
7784 | |
7785 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
7786 | last_check = 0; |
7787 | ||
7788 | if (j>2) { | |
c91abf5a | 7789 | printk(KERN_INFO |
61df9d91 N |
7790 | "md: resuming %s of %s from checkpoint.\n", |
7791 | desc, mdname(mddev)); | |
1da177e4 | 7792 | mddev->curr_resync = j; |
72f36d59 N |
7793 | } else |
7794 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 7795 | mddev->curr_resync_completed = j; |
72f36d59 N |
7796 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
7797 | md_new_event(mddev); | |
54f89341 | 7798 | update_time = jiffies; |
1da177e4 | 7799 | |
96ae923a | 7800 | if (mddev_is_clustered(mddev)) |
965400eb | 7801 | md_cluster_ops->resync_start(mddev, j, max_sectors); |
96ae923a | 7802 | |
7c2c57c9 | 7803 | blk_start_plug(&plug); |
1da177e4 | 7804 | while (j < max_sectors) { |
57afd89f | 7805 | sector_t sectors; |
1da177e4 | 7806 | |
57afd89f | 7807 | skipped = 0; |
97e4f42d | 7808 | |
7a91ee1f N |
7809 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7810 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
7811 | (mddev->curr_resync - mddev->curr_resync_completed) | |
7812 | > (max_sectors >> 4)) || | |
54f89341 | 7813 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f N |
7814 | (j - mddev->curr_resync_completed)*2 |
7815 | >= mddev->resync_max - mddev->curr_resync_completed | |
7816 | )) { | |
97e4f42d | 7817 | /* time to update curr_resync_completed */ |
97e4f42d N |
7818 | wait_event(mddev->recovery_wait, |
7819 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 7820 | mddev->curr_resync_completed = j; |
35d78c66 | 7821 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
7822 | j > mddev->recovery_cp) | |
7823 | mddev->recovery_cp = j; | |
54f89341 | 7824 | update_time = jiffies; |
070dc6dd | 7825 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 7826 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 7827 | } |
acb180b0 | 7828 | |
c91abf5a N |
7829 | while (j >= mddev->resync_max && |
7830 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
7831 | /* As this condition is controlled by user-space, |
7832 | * we can block indefinitely, so use '_interruptible' | |
7833 | * to avoid triggering warnings. | |
7834 | */ | |
7835 | flush_signals(current); /* just in case */ | |
7836 | wait_event_interruptible(mddev->recovery_wait, | |
7837 | mddev->resync_max > j | |
c91abf5a N |
7838 | || test_bit(MD_RECOVERY_INTR, |
7839 | &mddev->recovery)); | |
e62e58a5 | 7840 | } |
acb180b0 | 7841 | |
c91abf5a N |
7842 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7843 | break; | |
acb180b0 | 7844 | |
09314799 | 7845 | sectors = mddev->pers->sync_request(mddev, j, &skipped); |
57afd89f | 7846 | if (sectors == 0) { |
dfc70645 | 7847 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 7848 | break; |
1da177e4 | 7849 | } |
57afd89f N |
7850 | |
7851 | if (!skipped) { /* actual IO requested */ | |
7852 | io_sectors += sectors; | |
7853 | atomic_add(sectors, &mddev->recovery_active); | |
7854 | } | |
7855 | ||
e875ecea N |
7856 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7857 | break; | |
7858 | ||
1da177e4 | 7859 | j += sectors; |
72f36d59 N |
7860 | if (j > 2) |
7861 | mddev->curr_resync = j; | |
96ae923a GR |
7862 | if (mddev_is_clustered(mddev)) |
7863 | md_cluster_ops->resync_info_update(mddev, j, max_sectors); | |
ff4e8d9a | 7864 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 7865 | if (last_check == 0) |
e875ecea | 7866 | /* this is the earliest that rebuild will be |
d7603b7e N |
7867 | * visible in /proc/mdstat |
7868 | */ | |
7869 | md_new_event(mddev); | |
57afd89f N |
7870 | |
7871 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
7872 | continue; |
7873 | ||
57afd89f | 7874 | last_check = io_sectors; |
1da177e4 LT |
7875 | repeat: |
7876 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
7877 | /* step marks */ | |
7878 | int next = (last_mark+1) % SYNC_MARKS; | |
7879 | ||
7880 | mddev->resync_mark = mark[next]; | |
7881 | mddev->resync_mark_cnt = mark_cnt[next]; | |
7882 | mark[next] = jiffies; | |
57afd89f | 7883 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
7884 | last_mark = next; |
7885 | } | |
7886 | ||
c91abf5a N |
7887 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7888 | break; | |
1da177e4 LT |
7889 | |
7890 | /* | |
7891 | * this loop exits only if either when we are slower than | |
7892 | * the 'hard' speed limit, or the system was IO-idle for | |
7893 | * a jiffy. | |
7894 | * the system might be non-idle CPU-wise, but we only care | |
7895 | * about not overloading the IO subsystem. (things like an | |
7896 | * e2fsck being done on the RAID array should execute fast) | |
7897 | */ | |
1da177e4 LT |
7898 | cond_resched(); |
7899 | ||
ac7e50a3 XN |
7900 | recovery_done = io_sectors - atomic_read(&mddev->recovery_active); |
7901 | currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2 | |
57afd89f | 7902 | /((jiffies-mddev->resync_mark)/HZ +1) +1; |
1da177e4 | 7903 | |
88202a0c | 7904 | if (currspeed > speed_min(mddev)) { |
ac8fa419 | 7905 | if (currspeed > speed_max(mddev)) { |
c0e48521 | 7906 | msleep(500); |
1da177e4 LT |
7907 | goto repeat; |
7908 | } | |
ac8fa419 N |
7909 | if (!is_mddev_idle(mddev, 0)) { |
7910 | /* | |
7911 | * Give other IO more of a chance. | |
7912 | * The faster the devices, the less we wait. | |
7913 | */ | |
7914 | wait_event(mddev->recovery_wait, | |
7915 | !atomic_read(&mddev->recovery_active)); | |
7916 | } | |
1da177e4 LT |
7917 | } |
7918 | } | |
c91abf5a N |
7919 | printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc, |
7920 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
7921 | ? "interrupted" : "done"); | |
1da177e4 LT |
7922 | /* |
7923 | * this also signals 'finished resyncing' to md_stop | |
7924 | */ | |
7c2c57c9 | 7925 | blk_finish_plug(&plug); |
1da177e4 LT |
7926 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
7927 | ||
7928 | /* tell personality that we are finished */ | |
09314799 | 7929 | mddev->pers->sync_request(mddev, max_sectors, &skipped); |
1da177e4 | 7930 | |
96ae923a | 7931 | if (mddev_is_clustered(mddev)) |
965400eb | 7932 | md_cluster_ops->resync_finish(mddev); |
96ae923a | 7933 | |
dfc70645 | 7934 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
7935 | mddev->curr_resync > 2) { |
7936 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
7937 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7938 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
7939 | printk(KERN_INFO | |
61df9d91 N |
7940 | "md: checkpointing %s of %s.\n", |
7941 | desc, mdname(mddev)); | |
0a19caab | 7942 | if (test_bit(MD_RECOVERY_ERROR, |
7943 | &mddev->recovery)) | |
7944 | mddev->recovery_cp = | |
7945 | mddev->curr_resync_completed; | |
7946 | else | |
7947 | mddev->recovery_cp = | |
7948 | mddev->curr_resync; | |
5fd6c1dc N |
7949 | } |
7950 | } else | |
7951 | mddev->recovery_cp = MaxSector; | |
7952 | } else { | |
7953 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
7954 | mddev->curr_resync = MaxSector; | |
4e59ca7d | 7955 | rcu_read_lock(); |
dafb20fa | 7956 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 7957 | if (rdev->raid_disk >= 0 && |
70fffd0b | 7958 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
7959 | !test_bit(Faulty, &rdev->flags) && |
7960 | !test_bit(In_sync, &rdev->flags) && | |
7961 | rdev->recovery_offset < mddev->curr_resync) | |
7962 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 7963 | rcu_read_unlock(); |
5fd6c1dc | 7964 | } |
1da177e4 | 7965 | } |
db91ff55 | 7966 | skip: |
17571284 | 7967 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 7968 | |
23da422b | 7969 | spin_lock(&mddev->lock); |
c07b70ad N |
7970 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
7971 | /* We completed so min/max setting can be forgotten if used. */ | |
7972 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7973 | mddev->resync_min = 0; | |
7974 | mddev->resync_max = MaxSector; | |
7975 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7976 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 | 7977 | mddev->curr_resync = 0; |
23da422b N |
7978 | spin_unlock(&mddev->lock); |
7979 | ||
1da177e4 LT |
7980 | wake_up(&resync_wait); |
7981 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
7982 | md_wakeup_thread(mddev->thread); | |
c6207277 | 7983 | return; |
1da177e4 | 7984 | } |
29269553 | 7985 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 7986 | |
746d3207 N |
7987 | static int remove_and_add_spares(struct mddev *mddev, |
7988 | struct md_rdev *this) | |
b4c4c7b8 | 7989 | { |
3cb03002 | 7990 | struct md_rdev *rdev; |
b4c4c7b8 | 7991 | int spares = 0; |
f2a371c5 | 7992 | int removed = 0; |
b4c4c7b8 | 7993 | |
dafb20fa | 7994 | rdev_for_each(rdev, mddev) |
746d3207 N |
7995 | if ((this == NULL || rdev == this) && |
7996 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 7997 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
7998 | (test_bit(Faulty, &rdev->flags) || |
7999 | ! test_bit(In_sync, &rdev->flags)) && | |
8000 | atomic_read(&rdev->nr_pending)==0) { | |
8001 | if (mddev->pers->hot_remove_disk( | |
b8321b68 | 8002 | mddev, rdev) == 0) { |
36fad858 | 8003 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 8004 | rdev->raid_disk = -1; |
f2a371c5 | 8005 | removed++; |
b4c4c7b8 N |
8006 | } |
8007 | } | |
90584fc9 JB |
8008 | if (removed && mddev->kobj.sd) |
8009 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 8010 | |
746d3207 N |
8011 | if (this) |
8012 | goto no_add; | |
8013 | ||
dafb20fa | 8014 | rdev_for_each(rdev, mddev) { |
7bfec5f3 N |
8015 | if (rdev->raid_disk >= 0 && |
8016 | !test_bit(In_sync, &rdev->flags) && | |
8017 | !test_bit(Faulty, &rdev->flags)) | |
8018 | spares++; | |
7ceb17e8 N |
8019 | if (rdev->raid_disk >= 0) |
8020 | continue; | |
8021 | if (test_bit(Faulty, &rdev->flags)) | |
8022 | continue; | |
8023 | if (mddev->ro && | |
8313b8e5 N |
8024 | ! (rdev->saved_raid_disk >= 0 && |
8025 | !test_bit(Bitmap_sync, &rdev->flags))) | |
7ceb17e8 N |
8026 | continue; |
8027 | ||
7eb41885 N |
8028 | if (rdev->saved_raid_disk < 0) |
8029 | rdev->recovery_offset = 0; | |
7ceb17e8 N |
8030 | if (mddev->pers-> |
8031 | hot_add_disk(mddev, rdev) == 0) { | |
8032 | if (sysfs_link_rdev(mddev, rdev)) | |
8033 | /* failure here is OK */; | |
8034 | spares++; | |
8035 | md_new_event(mddev); | |
8036 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
dfc70645 | 8037 | } |
b4c4c7b8 | 8038 | } |
746d3207 | 8039 | no_add: |
6dafab6b N |
8040 | if (removed) |
8041 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
b4c4c7b8 N |
8042 | return spares; |
8043 | } | |
7ebc0be7 | 8044 | |
ac05f256 N |
8045 | static void md_start_sync(struct work_struct *ws) |
8046 | { | |
8047 | struct mddev *mddev = container_of(ws, struct mddev, del_work); | |
8048 | ||
8049 | mddev->sync_thread = md_register_thread(md_do_sync, | |
8050 | mddev, | |
8051 | "resync"); | |
8052 | if (!mddev->sync_thread) { | |
8053 | printk(KERN_ERR "%s: could not start resync" | |
8054 | " thread...\n", | |
8055 | mdname(mddev)); | |
8056 | /* leave the spares where they are, it shouldn't hurt */ | |
8057 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
8058 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8059 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8060 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
8061 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8062 | wake_up(&resync_wait); |
ac05f256 N |
8063 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8064 | &mddev->recovery)) | |
8065 | if (mddev->sysfs_action) | |
8066 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8067 | } else | |
8068 | md_wakeup_thread(mddev->sync_thread); | |
8069 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8070 | md_new_event(mddev); | |
8071 | } | |
8072 | ||
1da177e4 LT |
8073 | /* |
8074 | * This routine is regularly called by all per-raid-array threads to | |
8075 | * deal with generic issues like resync and super-block update. | |
8076 | * Raid personalities that don't have a thread (linear/raid0) do not | |
8077 | * need this as they never do any recovery or update the superblock. | |
8078 | * | |
8079 | * It does not do any resync itself, but rather "forks" off other threads | |
8080 | * to do that as needed. | |
8081 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
8082 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 8083 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
8084 | * and wakeups up this thread which will reap the thread and finish up. |
8085 | * This thread also removes any faulty devices (with nr_pending == 0). | |
8086 | * | |
8087 | * The overall approach is: | |
8088 | * 1/ if the superblock needs updating, update it. | |
8089 | * 2/ If a recovery thread is running, don't do anything else. | |
8090 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
8091 | * 4/ If there are any faulty devices, remove them. | |
8092 | * 5/ If array is degraded, try to add spares devices | |
8093 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
8094 | */ | |
fd01b88c | 8095 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 8096 | { |
68866e42 JB |
8097 | if (mddev->suspended) |
8098 | return; | |
8099 | ||
5f40402d | 8100 | if (mddev->bitmap) |
aa5cbd10 | 8101 | bitmap_daemon_work(mddev); |
1da177e4 | 8102 | |
fca4d848 | 8103 | if (signal_pending(current)) { |
31a59e34 | 8104 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
8105 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
8106 | mdname(mddev)); | |
8107 | mddev->safemode = 2; | |
8108 | } | |
8109 | flush_signals(current); | |
8110 | } | |
8111 | ||
c89a8eee N |
8112 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
8113 | return; | |
1da177e4 | 8114 | if ( ! ( |
142d44c3 | 8115 | (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 8116 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 8117 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 8118 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
8119 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
8120 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
8121 | )) |
8122 | return; | |
fca4d848 | 8123 | |
df5b89b3 | 8124 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 8125 | int spares = 0; |
fca4d848 | 8126 | |
c89a8eee | 8127 | if (mddev->ro) { |
7ceb17e8 N |
8128 | /* On a read-only array we can: |
8129 | * - remove failed devices | |
8130 | * - add already-in_sync devices if the array itself | |
8131 | * is in-sync. | |
8132 | * As we only add devices that are already in-sync, | |
8133 | * we can activate the spares immediately. | |
c89a8eee | 8134 | */ |
7ceb17e8 | 8135 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
8136 | /* There is no thread, but we need to call |
8137 | * ->spare_active and clear saved_raid_disk | |
8138 | */ | |
2ac295a5 | 8139 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8313b8e5 N |
8140 | md_reap_sync_thread(mddev); |
8141 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
c89a8eee N |
8142 | goto unlock; |
8143 | } | |
8144 | ||
31a59e34 | 8145 | if (!mddev->external) { |
0fd62b86 | 8146 | int did_change = 0; |
85572d7c | 8147 | spin_lock(&mddev->lock); |
31a59e34 N |
8148 | if (mddev->safemode && |
8149 | !atomic_read(&mddev->writes_pending) && | |
8150 | !mddev->in_sync && | |
8151 | mddev->recovery_cp == MaxSector) { | |
8152 | mddev->in_sync = 1; | |
0fd62b86 | 8153 | did_change = 1; |
070dc6dd | 8154 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
8155 | } |
8156 | if (mddev->safemode == 1) | |
8157 | mddev->safemode = 0; | |
85572d7c | 8158 | spin_unlock(&mddev->lock); |
0fd62b86 | 8159 | if (did_change) |
00bcb4ac | 8160 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 8161 | } |
fca4d848 | 8162 | |
293467aa GR |
8163 | if (mddev->flags & MD_UPDATE_SB_FLAGS) { |
8164 | if (mddev_is_clustered(mddev)) | |
8165 | md_cluster_ops->metadata_update_start(mddev); | |
850b2b42 | 8166 | md_update_sb(mddev, 0); |
293467aa GR |
8167 | if (mddev_is_clustered(mddev)) |
8168 | md_cluster_ops->metadata_update_finish(mddev); | |
8169 | } | |
06d91a5f | 8170 | |
1da177e4 LT |
8171 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
8172 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
8173 | /* resync/recovery still happening */ | |
8174 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8175 | goto unlock; | |
8176 | } | |
8177 | if (mddev->sync_thread) { | |
a91d5ac0 | 8178 | md_reap_sync_thread(mddev); |
1da177e4 LT |
8179 | goto unlock; |
8180 | } | |
72a23c21 NB |
8181 | /* Set RUNNING before clearing NEEDED to avoid |
8182 | * any transients in the value of "sync_action". | |
8183 | */ | |
72f36d59 | 8184 | mddev->curr_resync_completed = 0; |
23da422b | 8185 | spin_lock(&mddev->lock); |
72a23c21 | 8186 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
23da422b | 8187 | spin_unlock(&mddev->lock); |
24dd469d N |
8188 | /* Clear some bits that don't mean anything, but |
8189 | * might be left set | |
8190 | */ | |
24dd469d N |
8191 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8192 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 8193 | |
ed209584 N |
8194 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
8195 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
ac05f256 | 8196 | goto not_running; |
1da177e4 LT |
8197 | /* no recovery is running. |
8198 | * remove any failed drives, then | |
8199 | * add spares if possible. | |
72f36d59 | 8200 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
8201 | * the personality to fail the re-add. |
8202 | */ | |
1da177e4 | 8203 | |
b4c4c7b8 | 8204 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
8205 | if (mddev->pers->check_reshape == NULL || |
8206 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 | 8207 | /* Cannot proceed */ |
ac05f256 | 8208 | goto not_running; |
b4c4c7b8 | 8209 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
72a23c21 | 8210 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 8211 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
8212 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8213 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 8214 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 8215 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8216 | } else if (mddev->recovery_cp < MaxSector) { |
8217 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 8218 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
8219 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
8220 | /* nothing to be done ... */ | |
ac05f256 | 8221 | goto not_running; |
24dd469d | 8222 | |
1da177e4 | 8223 | if (mddev->pers->sync_request) { |
ef99bf48 | 8224 | if (spares) { |
a654b9d8 N |
8225 | /* We are adding a device or devices to an array |
8226 | * which has the bitmap stored on all devices. | |
8227 | * So make sure all bitmap pages get written | |
8228 | */ | |
8229 | bitmap_write_all(mddev->bitmap); | |
8230 | } | |
ac05f256 N |
8231 | INIT_WORK(&mddev->del_work, md_start_sync); |
8232 | queue_work(md_misc_wq, &mddev->del_work); | |
8233 | goto unlock; | |
1da177e4 | 8234 | } |
ac05f256 | 8235 | not_running: |
72a23c21 NB |
8236 | if (!mddev->sync_thread) { |
8237 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
f851b60d | 8238 | wake_up(&resync_wait); |
72a23c21 NB |
8239 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, |
8240 | &mddev->recovery)) | |
0c3573f1 | 8241 | if (mddev->sysfs_action) |
00bcb4ac | 8242 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 8243 | } |
ac05f256 N |
8244 | unlock: |
8245 | wake_up(&mddev->sb_wait); | |
1da177e4 LT |
8246 | mddev_unlock(mddev); |
8247 | } | |
8248 | } | |
6c144d31 | 8249 | EXPORT_SYMBOL(md_check_recovery); |
1da177e4 | 8250 | |
a91d5ac0 JB |
8251 | void md_reap_sync_thread(struct mddev *mddev) |
8252 | { | |
8253 | struct md_rdev *rdev; | |
8254 | ||
8255 | /* resync has finished, collect result */ | |
8256 | md_unregister_thread(&mddev->sync_thread); | |
8257 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && | |
8258 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
8259 | /* success...*/ | |
8260 | /* activate any spares */ | |
8261 | if (mddev->pers->spare_active(mddev)) { | |
8262 | sysfs_notify(&mddev->kobj, NULL, | |
8263 | "degraded"); | |
8264 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
8265 | } | |
8266 | } | |
293467aa GR |
8267 | if (mddev_is_clustered(mddev)) |
8268 | md_cluster_ops->metadata_update_start(mddev); | |
a91d5ac0 JB |
8269 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
8270 | mddev->pers->finish_reshape) | |
8271 | mddev->pers->finish_reshape(mddev); | |
8272 | ||
8273 | /* If array is no-longer degraded, then any saved_raid_disk | |
f466722c | 8274 | * information must be scrapped. |
a91d5ac0 | 8275 | */ |
f466722c N |
8276 | if (!mddev->degraded) |
8277 | rdev_for_each(rdev, mddev) | |
a91d5ac0 JB |
8278 | rdev->saved_raid_disk = -1; |
8279 | ||
8280 | md_update_sb(mddev, 1); | |
293467aa GR |
8281 | if (mddev_is_clustered(mddev)) |
8282 | md_cluster_ops->metadata_update_finish(mddev); | |
a91d5ac0 | 8283 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
ea358cd0 | 8284 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
a91d5ac0 JB |
8285 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8286 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
8287 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
8288 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
f851b60d | 8289 | wake_up(&resync_wait); |
a91d5ac0 JB |
8290 | /* flag recovery needed just to double check */ |
8291 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
8292 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
8293 | md_new_event(mddev); | |
8294 | if (mddev->event_work.func) | |
8295 | queue_work(md_misc_wq, &mddev->event_work); | |
8296 | } | |
6c144d31 | 8297 | EXPORT_SYMBOL(md_reap_sync_thread); |
a91d5ac0 | 8298 | |
fd01b88c | 8299 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 8300 | { |
00bcb4ac | 8301 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 8302 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
8303 | !test_bit(Blocked, &rdev->flags) && |
8304 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
8305 | msecs_to_jiffies(5000)); |
8306 | rdev_dec_pending(rdev, mddev); | |
8307 | } | |
8308 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
8309 | ||
c6563a8c N |
8310 | void md_finish_reshape(struct mddev *mddev) |
8311 | { | |
8312 | /* called be personality module when reshape completes. */ | |
8313 | struct md_rdev *rdev; | |
8314 | ||
8315 | rdev_for_each(rdev, mddev) { | |
8316 | if (rdev->data_offset > rdev->new_data_offset) | |
8317 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
8318 | else | |
8319 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
8320 | rdev->data_offset = rdev->new_data_offset; | |
8321 | } | |
8322 | } | |
8323 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 N |
8324 | |
8325 | /* Bad block management. | |
8326 | * We can record which blocks on each device are 'bad' and so just | |
8327 | * fail those blocks, or that stripe, rather than the whole device. | |
8328 | * Entries in the bad-block table are 64bits wide. This comprises: | |
8329 | * Length of bad-range, in sectors: 0-511 for lengths 1-512 | |
8330 | * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes) | |
8331 | * A 'shift' can be set so that larger blocks are tracked and | |
8332 | * consequently larger devices can be covered. | |
8333 | * 'Acknowledged' flag - 1 bit. - the most significant bit. | |
8334 | * | |
8335 | * Locking of the bad-block table uses a seqlock so md_is_badblock | |
8336 | * might need to retry if it is very unlucky. | |
8337 | * We will sometimes want to check for bad blocks in a bi_end_io function, | |
8338 | * so we use the write_seqlock_irq variant. | |
8339 | * | |
8340 | * When looking for a bad block we specify a range and want to | |
8341 | * know if any block in the range is bad. So we binary-search | |
8342 | * to the last range that starts at-or-before the given endpoint, | |
8343 | * (or "before the sector after the target range") | |
8344 | * then see if it ends after the given start. | |
8345 | * We return | |
8346 | * 0 if there are no known bad blocks in the range | |
8347 | * 1 if there are known bad block which are all acknowledged | |
8348 | * -1 if there are bad blocks which have not yet been acknowledged in metadata. | |
8349 | * plus the start/length of the first bad section we overlap. | |
8350 | */ | |
8351 | int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, | |
8352 | sector_t *first_bad, int *bad_sectors) | |
8353 | { | |
8354 | int hi; | |
ab05613a | 8355 | int lo; |
2230dfe4 | 8356 | u64 *p = bb->page; |
ab05613a | 8357 | int rv; |
2230dfe4 N |
8358 | sector_t target = s + sectors; |
8359 | unsigned seq; | |
8360 | ||
8361 | if (bb->shift > 0) { | |
8362 | /* round the start down, and the end up */ | |
8363 | s >>= bb->shift; | |
8364 | target += (1<<bb->shift) - 1; | |
8365 | target >>= bb->shift; | |
8366 | sectors = target - s; | |
8367 | } | |
8368 | /* 'target' is now the first block after the bad range */ | |
8369 | ||
8370 | retry: | |
8371 | seq = read_seqbegin(&bb->lock); | |
ab05613a | 8372 | lo = 0; |
8373 | rv = 0; | |
2230dfe4 N |
8374 | hi = bb->count; |
8375 | ||
8376 | /* Binary search between lo and hi for 'target' | |
8377 | * i.e. for the last range that starts before 'target' | |
8378 | */ | |
8379 | /* INVARIANT: ranges before 'lo' and at-or-after 'hi' | |
8380 | * are known not to be the last range before target. | |
8381 | * VARIANT: hi-lo is the number of possible | |
8382 | * ranges, and decreases until it reaches 1 | |
8383 | */ | |
8384 | while (hi - lo > 1) { | |
8385 | int mid = (lo + hi) / 2; | |
8386 | sector_t a = BB_OFFSET(p[mid]); | |
8387 | if (a < target) | |
8388 | /* This could still be the one, earlier ranges | |
8389 | * could not. */ | |
8390 | lo = mid; | |
8391 | else | |
8392 | /* This and later ranges are definitely out. */ | |
8393 | hi = mid; | |
8394 | } | |
8395 | /* 'lo' might be the last that started before target, but 'hi' isn't */ | |
8396 | if (hi > lo) { | |
8397 | /* need to check all range that end after 's' to see if | |
8398 | * any are unacknowledged. | |
8399 | */ | |
8400 | while (lo >= 0 && | |
8401 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8402 | if (BB_OFFSET(p[lo]) < target) { | |
8403 | /* starts before the end, and finishes after | |
8404 | * the start, so they must overlap | |
8405 | */ | |
8406 | if (rv != -1 && BB_ACK(p[lo])) | |
8407 | rv = 1; | |
8408 | else | |
8409 | rv = -1; | |
8410 | *first_bad = BB_OFFSET(p[lo]); | |
8411 | *bad_sectors = BB_LEN(p[lo]); | |
8412 | } | |
8413 | lo--; | |
8414 | } | |
8415 | } | |
8416 | ||
8417 | if (read_seqretry(&bb->lock, seq)) | |
8418 | goto retry; | |
8419 | ||
8420 | return rv; | |
8421 | } | |
8422 | EXPORT_SYMBOL_GPL(md_is_badblock); | |
8423 | ||
8424 | /* | |
8425 | * Add a range of bad blocks to the table. | |
8426 | * This might extend the table, or might contract it | |
8427 | * if two adjacent ranges can be merged. | |
8428 | * We binary-search to find the 'insertion' point, then | |
8429 | * decide how best to handle it. | |
8430 | */ | |
8431 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, | |
8432 | int acknowledged) | |
8433 | { | |
8434 | u64 *p; | |
8435 | int lo, hi; | |
8436 | int rv = 1; | |
905b0297 | 8437 | unsigned long flags; |
2230dfe4 N |
8438 | |
8439 | if (bb->shift < 0) | |
8440 | /* badblocks are disabled */ | |
8441 | return 0; | |
8442 | ||
8443 | if (bb->shift) { | |
8444 | /* round the start down, and the end up */ | |
8445 | sector_t next = s + sectors; | |
8446 | s >>= bb->shift; | |
8447 | next += (1<<bb->shift) - 1; | |
8448 | next >>= bb->shift; | |
8449 | sectors = next - s; | |
8450 | } | |
8451 | ||
905b0297 | 8452 | write_seqlock_irqsave(&bb->lock, flags); |
2230dfe4 N |
8453 | |
8454 | p = bb->page; | |
8455 | lo = 0; | |
8456 | hi = bb->count; | |
8457 | /* Find the last range that starts at-or-before 's' */ | |
8458 | while (hi - lo > 1) { | |
8459 | int mid = (lo + hi) / 2; | |
8460 | sector_t a = BB_OFFSET(p[mid]); | |
8461 | if (a <= s) | |
8462 | lo = mid; | |
8463 | else | |
8464 | hi = mid; | |
8465 | } | |
8466 | if (hi > lo && BB_OFFSET(p[lo]) > s) | |
8467 | hi = lo; | |
8468 | ||
8469 | if (hi > lo) { | |
8470 | /* we found a range that might merge with the start | |
8471 | * of our new range | |
8472 | */ | |
8473 | sector_t a = BB_OFFSET(p[lo]); | |
8474 | sector_t e = a + BB_LEN(p[lo]); | |
8475 | int ack = BB_ACK(p[lo]); | |
8476 | if (e >= s) { | |
8477 | /* Yes, we can merge with a previous range */ | |
8478 | if (s == a && s + sectors >= e) | |
8479 | /* new range covers old */ | |
8480 | ack = acknowledged; | |
8481 | else | |
8482 | ack = ack && acknowledged; | |
8483 | ||
8484 | if (e < s + sectors) | |
8485 | e = s + sectors; | |
8486 | if (e - a <= BB_MAX_LEN) { | |
8487 | p[lo] = BB_MAKE(a, e-a, ack); | |
8488 | s = e; | |
8489 | } else { | |
8490 | /* does not all fit in one range, | |
8491 | * make p[lo] maximal | |
8492 | */ | |
8493 | if (BB_LEN(p[lo]) != BB_MAX_LEN) | |
8494 | p[lo] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8495 | s = a + BB_MAX_LEN; | |
8496 | } | |
8497 | sectors = e - s; | |
8498 | } | |
8499 | } | |
8500 | if (sectors && hi < bb->count) { | |
8501 | /* 'hi' points to the first range that starts after 's'. | |
8502 | * Maybe we can merge with the start of that range */ | |
8503 | sector_t a = BB_OFFSET(p[hi]); | |
8504 | sector_t e = a + BB_LEN(p[hi]); | |
8505 | int ack = BB_ACK(p[hi]); | |
8506 | if (a <= s + sectors) { | |
8507 | /* merging is possible */ | |
8508 | if (e <= s + sectors) { | |
8509 | /* full overlap */ | |
8510 | e = s + sectors; | |
8511 | ack = acknowledged; | |
8512 | } else | |
8513 | ack = ack && acknowledged; | |
8514 | ||
8515 | a = s; | |
8516 | if (e - a <= BB_MAX_LEN) { | |
8517 | p[hi] = BB_MAKE(a, e-a, ack); | |
8518 | s = e; | |
8519 | } else { | |
8520 | p[hi] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8521 | s = a + BB_MAX_LEN; | |
8522 | } | |
8523 | sectors = e - s; | |
8524 | lo = hi; | |
8525 | hi++; | |
8526 | } | |
8527 | } | |
8528 | if (sectors == 0 && hi < bb->count) { | |
8529 | /* we might be able to combine lo and hi */ | |
8530 | /* Note: 's' is at the end of 'lo' */ | |
8531 | sector_t a = BB_OFFSET(p[hi]); | |
8532 | int lolen = BB_LEN(p[lo]); | |
8533 | int hilen = BB_LEN(p[hi]); | |
8534 | int newlen = lolen + hilen - (s - a); | |
8535 | if (s >= a && newlen < BB_MAX_LEN) { | |
8536 | /* yes, we can combine them */ | |
8537 | int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]); | |
8538 | p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack); | |
8539 | memmove(p + hi, p + hi + 1, | |
8540 | (bb->count - hi - 1) * 8); | |
8541 | bb->count--; | |
8542 | } | |
8543 | } | |
8544 | while (sectors) { | |
8545 | /* didn't merge (it all). | |
8546 | * Need to add a range just before 'hi' */ | |
8547 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8548 | /* No room for more */ | |
8549 | rv = 0; | |
8550 | break; | |
8551 | } else { | |
8552 | int this_sectors = sectors; | |
8553 | memmove(p + hi + 1, p + hi, | |
8554 | (bb->count - hi) * 8); | |
8555 | bb->count++; | |
8556 | ||
8557 | if (this_sectors > BB_MAX_LEN) | |
8558 | this_sectors = BB_MAX_LEN; | |
8559 | p[hi] = BB_MAKE(s, this_sectors, acknowledged); | |
8560 | sectors -= this_sectors; | |
8561 | s += this_sectors; | |
8562 | } | |
8563 | } | |
8564 | ||
8565 | bb->changed = 1; | |
de393cde N |
8566 | if (!acknowledged) |
8567 | bb->unacked_exist = 1; | |
905b0297 | 8568 | write_sequnlock_irqrestore(&bb->lock, flags); |
2230dfe4 N |
8569 | |
8570 | return rv; | |
8571 | } | |
8572 | ||
3cb03002 | 8573 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8574 | int is_new) |
2230dfe4 | 8575 | { |
c6563a8c N |
8576 | int rv; |
8577 | if (is_new) | |
8578 | s += rdev->new_data_offset; | |
8579 | else | |
8580 | s += rdev->data_offset; | |
8581 | rv = md_set_badblocks(&rdev->badblocks, | |
8582 | s, sectors, 0); | |
2230dfe4 N |
8583 | if (rv) { |
8584 | /* Make sure they get written out promptly */ | |
8bd2f0a0 | 8585 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2230dfe4 N |
8586 | set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags); |
8587 | md_wakeup_thread(rdev->mddev->thread); | |
8588 | } | |
8589 | return rv; | |
8590 | } | |
8591 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8592 | ||
8593 | /* | |
8594 | * Remove a range of bad blocks from the table. | |
8595 | * This may involve extending the table if we spilt a region, | |
8596 | * but it must not fail. So if the table becomes full, we just | |
8597 | * drop the remove request. | |
8598 | */ | |
8599 | static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors) | |
8600 | { | |
8601 | u64 *p; | |
8602 | int lo, hi; | |
8603 | sector_t target = s + sectors; | |
8604 | int rv = 0; | |
8605 | ||
8606 | if (bb->shift > 0) { | |
8607 | /* When clearing we round the start up and the end down. | |
8608 | * This should not matter as the shift should align with | |
8609 | * the block size and no rounding should ever be needed. | |
8610 | * However it is better the think a block is bad when it | |
8611 | * isn't than to think a block is not bad when it is. | |
8612 | */ | |
8613 | s += (1<<bb->shift) - 1; | |
8614 | s >>= bb->shift; | |
8615 | target >>= bb->shift; | |
8616 | sectors = target - s; | |
8617 | } | |
8618 | ||
8619 | write_seqlock_irq(&bb->lock); | |
8620 | ||
8621 | p = bb->page; | |
8622 | lo = 0; | |
8623 | hi = bb->count; | |
8624 | /* Find the last range that starts before 'target' */ | |
8625 | while (hi - lo > 1) { | |
8626 | int mid = (lo + hi) / 2; | |
8627 | sector_t a = BB_OFFSET(p[mid]); | |
8628 | if (a < target) | |
8629 | lo = mid; | |
8630 | else | |
8631 | hi = mid; | |
8632 | } | |
8633 | if (hi > lo) { | |
8634 | /* p[lo] is the last range that could overlap the | |
8635 | * current range. Earlier ranges could also overlap, | |
8636 | * but only this one can overlap the end of the range. | |
8637 | */ | |
8638 | if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) { | |
8639 | /* Partial overlap, leave the tail of this range */ | |
8640 | int ack = BB_ACK(p[lo]); | |
8641 | sector_t a = BB_OFFSET(p[lo]); | |
8642 | sector_t end = a + BB_LEN(p[lo]); | |
8643 | ||
8644 | if (a < s) { | |
8645 | /* we need to split this range */ | |
8646 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8b32bf5e | 8647 | rv = -ENOSPC; |
2230dfe4 N |
8648 | goto out; |
8649 | } | |
8650 | memmove(p+lo+1, p+lo, (bb->count - lo) * 8); | |
8651 | bb->count++; | |
8652 | p[lo] = BB_MAKE(a, s-a, ack); | |
8653 | lo++; | |
8654 | } | |
8655 | p[lo] = BB_MAKE(target, end - target, ack); | |
8656 | /* there is no longer an overlap */ | |
8657 | hi = lo; | |
8658 | lo--; | |
8659 | } | |
8660 | while (lo >= 0 && | |
8661 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8662 | /* This range does overlap */ | |
8663 | if (BB_OFFSET(p[lo]) < s) { | |
8664 | /* Keep the early parts of this range. */ | |
8665 | int ack = BB_ACK(p[lo]); | |
8666 | sector_t start = BB_OFFSET(p[lo]); | |
8667 | p[lo] = BB_MAKE(start, s - start, ack); | |
8668 | /* now low doesn't overlap, so.. */ | |
8669 | break; | |
8670 | } | |
8671 | lo--; | |
8672 | } | |
8673 | /* 'lo' is strictly before, 'hi' is strictly after, | |
8674 | * anything between needs to be discarded | |
8675 | */ | |
8676 | if (hi - lo > 1) { | |
8677 | memmove(p+lo+1, p+hi, (bb->count - hi) * 8); | |
8678 | bb->count -= (hi - lo - 1); | |
8679 | } | |
8680 | } | |
8681 | ||
8682 | bb->changed = 1; | |
8683 | out: | |
8684 | write_sequnlock_irq(&bb->lock); | |
8685 | return rv; | |
8686 | } | |
8687 | ||
c6563a8c N |
8688 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8689 | int is_new) | |
2230dfe4 | 8690 | { |
c6563a8c N |
8691 | if (is_new) |
8692 | s += rdev->new_data_offset; | |
8693 | else | |
8694 | s += rdev->data_offset; | |
2230dfe4 | 8695 | return md_clear_badblocks(&rdev->badblocks, |
c6563a8c | 8696 | s, sectors); |
2230dfe4 N |
8697 | } |
8698 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8699 | ||
8700 | /* | |
8701 | * Acknowledge all bad blocks in a list. | |
8702 | * This only succeeds if ->changed is clear. It is used by | |
8703 | * in-kernel metadata updates | |
8704 | */ | |
8705 | void md_ack_all_badblocks(struct badblocks *bb) | |
8706 | { | |
8707 | if (bb->page == NULL || bb->changed) | |
8708 | /* no point even trying */ | |
8709 | return; | |
8710 | write_seqlock_irq(&bb->lock); | |
8711 | ||
ecb178bb | 8712 | if (bb->changed == 0 && bb->unacked_exist) { |
2230dfe4 N |
8713 | u64 *p = bb->page; |
8714 | int i; | |
8715 | for (i = 0; i < bb->count ; i++) { | |
8716 | if (!BB_ACK(p[i])) { | |
8717 | sector_t start = BB_OFFSET(p[i]); | |
8718 | int len = BB_LEN(p[i]); | |
8719 | p[i] = BB_MAKE(start, len, 1); | |
8720 | } | |
8721 | } | |
de393cde | 8722 | bb->unacked_exist = 0; |
2230dfe4 N |
8723 | } |
8724 | write_sequnlock_irq(&bb->lock); | |
8725 | } | |
8726 | EXPORT_SYMBOL_GPL(md_ack_all_badblocks); | |
8727 | ||
16c791a5 N |
8728 | /* sysfs access to bad-blocks list. |
8729 | * We present two files. | |
8730 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
8731 | * are recorded as bad. The list is truncated to fit within | |
8732 | * the one-page limit of sysfs. | |
8733 | * Writing "sector length" to this file adds an acknowledged | |
8734 | * bad block list. | |
8735 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
8736 | * been acknowledged. Writing to this file adds bad blocks | |
8737 | * without acknowledging them. This is largely for testing. | |
8738 | */ | |
8739 | ||
8740 | static ssize_t | |
8741 | badblocks_show(struct badblocks *bb, char *page, int unack) | |
8742 | { | |
8743 | size_t len; | |
8744 | int i; | |
8745 | u64 *p = bb->page; | |
8746 | unsigned seq; | |
8747 | ||
8748 | if (bb->shift < 0) | |
8749 | return 0; | |
8750 | ||
8751 | retry: | |
8752 | seq = read_seqbegin(&bb->lock); | |
8753 | ||
8754 | len = 0; | |
8755 | i = 0; | |
8756 | ||
8757 | while (len < PAGE_SIZE && i < bb->count) { | |
8758 | sector_t s = BB_OFFSET(p[i]); | |
8759 | unsigned int length = BB_LEN(p[i]); | |
8760 | int ack = BB_ACK(p[i]); | |
8761 | i++; | |
8762 | ||
8763 | if (unack && ack) | |
8764 | continue; | |
8765 | ||
8766 | len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n", | |
8767 | (unsigned long long)s << bb->shift, | |
8768 | length << bb->shift); | |
8769 | } | |
de393cde N |
8770 | if (unack && len == 0) |
8771 | bb->unacked_exist = 0; | |
16c791a5 N |
8772 | |
8773 | if (read_seqretry(&bb->lock, seq)) | |
8774 | goto retry; | |
8775 | ||
8776 | return len; | |
8777 | } | |
8778 | ||
8779 | #define DO_DEBUG 1 | |
8780 | ||
8781 | static ssize_t | |
8782 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack) | |
8783 | { | |
8784 | unsigned long long sector; | |
8785 | int length; | |
8786 | char newline; | |
8787 | #ifdef DO_DEBUG | |
8788 | /* Allow clearing via sysfs *only* for testing/debugging. | |
8789 | * Normally only a successful write may clear a badblock | |
8790 | */ | |
8791 | int clear = 0; | |
8792 | if (page[0] == '-') { | |
8793 | clear = 1; | |
8794 | page++; | |
8795 | } | |
8796 | #endif /* DO_DEBUG */ | |
8797 | ||
8798 | switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) { | |
8799 | case 3: | |
8800 | if (newline != '\n') | |
8801 | return -EINVAL; | |
8802 | case 2: | |
8803 | if (length <= 0) | |
8804 | return -EINVAL; | |
8805 | break; | |
8806 | default: | |
8807 | return -EINVAL; | |
8808 | } | |
8809 | ||
8810 | #ifdef DO_DEBUG | |
8811 | if (clear) { | |
8812 | md_clear_badblocks(bb, sector, length); | |
8813 | return len; | |
8814 | } | |
8815 | #endif /* DO_DEBUG */ | |
8816 | if (md_set_badblocks(bb, sector, length, !unack)) | |
8817 | return len; | |
8818 | else | |
8819 | return -ENOSPC; | |
8820 | } | |
8821 | ||
75c96f85 AB |
8822 | static int md_notify_reboot(struct notifier_block *this, |
8823 | unsigned long code, void *x) | |
1da177e4 LT |
8824 | { |
8825 | struct list_head *tmp; | |
fd01b88c | 8826 | struct mddev *mddev; |
2dba6a91 | 8827 | int need_delay = 0; |
1da177e4 | 8828 | |
c744a65c N |
8829 | for_each_mddev(mddev, tmp) { |
8830 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8831 | if (mddev->pers) |
8832 | __md_stop_writes(mddev); | |
0f62fb22 N |
8833 | if (mddev->persistent) |
8834 | mddev->safemode = 2; | |
c744a65c | 8835 | mddev_unlock(mddev); |
2dba6a91 | 8836 | } |
c744a65c | 8837 | need_delay = 1; |
1da177e4 | 8838 | } |
c744a65c N |
8839 | /* |
8840 | * certain more exotic SCSI devices are known to be | |
8841 | * volatile wrt too early system reboots. While the | |
8842 | * right place to handle this issue is the given | |
8843 | * driver, we do want to have a safe RAID driver ... | |
8844 | */ | |
8845 | if (need_delay) | |
8846 | mdelay(1000*1); | |
8847 | ||
1da177e4 LT |
8848 | return NOTIFY_DONE; |
8849 | } | |
8850 | ||
75c96f85 | 8851 | static struct notifier_block md_notifier = { |
1da177e4 LT |
8852 | .notifier_call = md_notify_reboot, |
8853 | .next = NULL, | |
8854 | .priority = INT_MAX, /* before any real devices */ | |
8855 | }; | |
8856 | ||
8857 | static void md_geninit(void) | |
8858 | { | |
36a4e1fe | 8859 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 8860 | |
c7705f34 | 8861 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
8862 | } |
8863 | ||
75c96f85 | 8864 | static int __init md_init(void) |
1da177e4 | 8865 | { |
e804ac78 TH |
8866 | int ret = -ENOMEM; |
8867 | ||
ada609ee | 8868 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
8869 | if (!md_wq) |
8870 | goto err_wq; | |
8871 | ||
8872 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
8873 | if (!md_misc_wq) | |
8874 | goto err_misc_wq; | |
8875 | ||
8876 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
8877 | goto err_md; | |
8878 | ||
8879 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
8880 | goto err_mdp; | |
8881 | mdp_major = ret; | |
8882 | ||
af5628f0 | 8883 | blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE, |
e8703fe1 N |
8884 | md_probe, NULL, NULL); |
8885 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
8886 | md_probe, NULL, NULL); |
8887 | ||
1da177e4 | 8888 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 8889 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
8890 | |
8891 | md_geninit(); | |
d710e138 | 8892 | return 0; |
1da177e4 | 8893 | |
e804ac78 TH |
8894 | err_mdp: |
8895 | unregister_blkdev(MD_MAJOR, "md"); | |
8896 | err_md: | |
8897 | destroy_workqueue(md_misc_wq); | |
8898 | err_misc_wq: | |
8899 | destroy_workqueue(md_wq); | |
8900 | err_wq: | |
8901 | return ret; | |
8902 | } | |
1da177e4 | 8903 | |
1d7e3e96 GR |
8904 | void md_reload_sb(struct mddev *mddev) |
8905 | { | |
8906 | struct md_rdev *rdev, *tmp; | |
8907 | ||
8908 | rdev_for_each_safe(rdev, tmp, mddev) { | |
8909 | rdev->sb_loaded = 0; | |
8910 | ClearPageUptodate(rdev->sb_page); | |
8911 | } | |
8912 | mddev->raid_disks = 0; | |
8913 | analyze_sbs(mddev); | |
8914 | rdev_for_each_safe(rdev, tmp, mddev) { | |
8915 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); | |
8916 | /* since we don't write to faulty devices, we figure out if the | |
8917 | * disk is faulty by comparing events | |
8918 | */ | |
8919 | if (mddev->events > sb->events) | |
8920 | set_bit(Faulty, &rdev->flags); | |
8921 | } | |
8922 | ||
8923 | } | |
8924 | EXPORT_SYMBOL(md_reload_sb); | |
8925 | ||
1da177e4 LT |
8926 | #ifndef MODULE |
8927 | ||
8928 | /* | |
8929 | * Searches all registered partitions for autorun RAID arrays | |
8930 | * at boot time. | |
8931 | */ | |
4d936ec1 ME |
8932 | |
8933 | static LIST_HEAD(all_detected_devices); | |
8934 | struct detected_devices_node { | |
8935 | struct list_head list; | |
8936 | dev_t dev; | |
8937 | }; | |
1da177e4 LT |
8938 | |
8939 | void md_autodetect_dev(dev_t dev) | |
8940 | { | |
4d936ec1 ME |
8941 | struct detected_devices_node *node_detected_dev; |
8942 | ||
8943 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
8944 | if (node_detected_dev) { | |
8945 | node_detected_dev->dev = dev; | |
8946 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
8947 | } else { | |
8948 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
8949 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
8950 | } | |
1da177e4 LT |
8951 | } |
8952 | ||
1da177e4 LT |
8953 | static void autostart_arrays(int part) |
8954 | { | |
3cb03002 | 8955 | struct md_rdev *rdev; |
4d936ec1 ME |
8956 | struct detected_devices_node *node_detected_dev; |
8957 | dev_t dev; | |
8958 | int i_scanned, i_passed; | |
1da177e4 | 8959 | |
4d936ec1 ME |
8960 | i_scanned = 0; |
8961 | i_passed = 0; | |
1da177e4 | 8962 | |
4d936ec1 | 8963 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 8964 | |
4d936ec1 ME |
8965 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
8966 | i_scanned++; | |
8967 | node_detected_dev = list_entry(all_detected_devices.next, | |
8968 | struct detected_devices_node, list); | |
8969 | list_del(&node_detected_dev->list); | |
8970 | dev = node_detected_dev->dev; | |
8971 | kfree(node_detected_dev); | |
df968c4e | 8972 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
8973 | if (IS_ERR(rdev)) |
8974 | continue; | |
8975 | ||
403df478 | 8976 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 8977 | continue; |
403df478 | 8978 | |
d0fae18f | 8979 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 8980 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 8981 | i_passed++; |
1da177e4 | 8982 | } |
4d936ec1 ME |
8983 | |
8984 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
8985 | i_scanned, i_passed); | |
1da177e4 LT |
8986 | |
8987 | autorun_devices(part); | |
8988 | } | |
8989 | ||
fdee8ae4 | 8990 | #endif /* !MODULE */ |
1da177e4 LT |
8991 | |
8992 | static __exit void md_exit(void) | |
8993 | { | |
fd01b88c | 8994 | struct mddev *mddev; |
1da177e4 | 8995 | struct list_head *tmp; |
e2f23b60 | 8996 | int delay = 1; |
8ab5e4c1 | 8997 | |
af5628f0 | 8998 | blk_unregister_region(MKDEV(MD_MAJOR,0), 512); |
e8703fe1 | 8999 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 9000 | |
3dbd8c2e | 9001 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
9002 | unregister_blkdev(mdp_major, "mdp"); |
9003 | unregister_reboot_notifier(&md_notifier); | |
9004 | unregister_sysctl_table(raid_table_header); | |
e2f23b60 N |
9005 | |
9006 | /* We cannot unload the modules while some process is | |
9007 | * waiting for us in select() or poll() - wake them up | |
9008 | */ | |
9009 | md_unloading = 1; | |
9010 | while (waitqueue_active(&md_event_waiters)) { | |
9011 | /* not safe to leave yet */ | |
9012 | wake_up(&md_event_waiters); | |
9013 | msleep(delay); | |
9014 | delay += delay; | |
9015 | } | |
1da177e4 | 9016 | remove_proc_entry("mdstat", NULL); |
e2f23b60 | 9017 | |
29ac4aa3 | 9018 | for_each_mddev(mddev, tmp) { |
1da177e4 | 9019 | export_array(mddev); |
d3374825 | 9020 | mddev->hold_active = 0; |
1da177e4 | 9021 | } |
e804ac78 TH |
9022 | destroy_workqueue(md_misc_wq); |
9023 | destroy_workqueue(md_wq); | |
1da177e4 LT |
9024 | } |
9025 | ||
685784aa | 9026 | subsys_initcall(md_init); |
1da177e4 LT |
9027 | module_exit(md_exit) |
9028 | ||
f91de92e N |
9029 | static int get_ro(char *buffer, struct kernel_param *kp) |
9030 | { | |
9031 | return sprintf(buffer, "%d", start_readonly); | |
9032 | } | |
9033 | static int set_ro(const char *val, struct kernel_param *kp) | |
9034 | { | |
4c9309c0 | 9035 | return kstrtouint(val, 10, (unsigned int *)&start_readonly); |
f91de92e N |
9036 | } |
9037 | ||
80ca3a44 N |
9038 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
9039 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
efeb53c0 | 9040 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 9041 | |
1da177e4 | 9042 | MODULE_LICENSE("GPL"); |
0efb9e61 | 9043 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 9044 | MODULE_ALIAS("md"); |
72008652 | 9045 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |