]>
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" |
1da177e4 | 56 | |
1da177e4 | 57 | #ifndef MODULE |
d710e138 | 58 | static void autostart_arrays(int part); |
1da177e4 LT |
59 | #endif |
60 | ||
01f96c0a N |
61 | /* pers_list is a list of registered personalities protected |
62 | * by pers_lock. | |
63 | * pers_lock does extra service to protect accesses to | |
64 | * mddev->thread when the mutex cannot be held. | |
65 | */ | |
2604b703 | 66 | static LIST_HEAD(pers_list); |
1da177e4 LT |
67 | static DEFINE_SPINLOCK(pers_lock); |
68 | ||
90b08710 | 69 | static DECLARE_WAIT_QUEUE_HEAD(resync_wait); |
e804ac78 TH |
70 | static struct workqueue_struct *md_wq; |
71 | static struct workqueue_struct *md_misc_wq; | |
90b08710 | 72 | |
746d3207 N |
73 | static int remove_and_add_spares(struct mddev *mddev, |
74 | struct md_rdev *this); | |
75 | ||
1e50915f RB |
76 | /* |
77 | * Default number of read corrections we'll attempt on an rdev | |
78 | * before ejecting it from the array. We divide the read error | |
79 | * count by 2 for every hour elapsed between read errors. | |
80 | */ | |
81 | #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20 | |
1da177e4 LT |
82 | /* |
83 | * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' | |
84 | * is 1000 KB/sec, so the extra system load does not show up that much. | |
85 | * Increase it if you want to have more _guaranteed_ speed. Note that | |
338cec32 | 86 | * the RAID driver will use the maximum available bandwidth if the IO |
1da177e4 LT |
87 | * subsystem is idle. There is also an 'absolute maximum' reconstruction |
88 | * speed limit - in case reconstruction slows down your system despite | |
89 | * idle IO detection. | |
90 | * | |
91 | * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. | |
88202a0c | 92 | * or /sys/block/mdX/md/sync_speed_{min,max} |
1da177e4 LT |
93 | */ |
94 | ||
95 | static int sysctl_speed_limit_min = 1000; | |
96 | static int sysctl_speed_limit_max = 200000; | |
fd01b88c | 97 | static inline int speed_min(struct mddev *mddev) |
88202a0c N |
98 | { |
99 | return mddev->sync_speed_min ? | |
100 | mddev->sync_speed_min : sysctl_speed_limit_min; | |
101 | } | |
102 | ||
fd01b88c | 103 | static inline int speed_max(struct mddev *mddev) |
88202a0c N |
104 | { |
105 | return mddev->sync_speed_max ? | |
106 | mddev->sync_speed_max : sysctl_speed_limit_max; | |
107 | } | |
1da177e4 LT |
108 | |
109 | static struct ctl_table_header *raid_table_header; | |
110 | ||
82592c38 | 111 | static struct ctl_table raid_table[] = { |
1da177e4 | 112 | { |
1da177e4 LT |
113 | .procname = "speed_limit_min", |
114 | .data = &sysctl_speed_limit_min, | |
115 | .maxlen = sizeof(int), | |
80ca3a44 | 116 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 117 | .proc_handler = proc_dointvec, |
1da177e4 LT |
118 | }, |
119 | { | |
1da177e4 LT |
120 | .procname = "speed_limit_max", |
121 | .data = &sysctl_speed_limit_max, | |
122 | .maxlen = sizeof(int), | |
80ca3a44 | 123 | .mode = S_IRUGO|S_IWUSR, |
6d456111 | 124 | .proc_handler = proc_dointvec, |
1da177e4 | 125 | }, |
894d2491 | 126 | { } |
1da177e4 LT |
127 | }; |
128 | ||
82592c38 | 129 | static struct ctl_table raid_dir_table[] = { |
1da177e4 | 130 | { |
1da177e4 LT |
131 | .procname = "raid", |
132 | .maxlen = 0, | |
80ca3a44 | 133 | .mode = S_IRUGO|S_IXUGO, |
1da177e4 LT |
134 | .child = raid_table, |
135 | }, | |
894d2491 | 136 | { } |
1da177e4 LT |
137 | }; |
138 | ||
82592c38 | 139 | static struct ctl_table raid_root_table[] = { |
1da177e4 | 140 | { |
1da177e4 LT |
141 | .procname = "dev", |
142 | .maxlen = 0, | |
143 | .mode = 0555, | |
144 | .child = raid_dir_table, | |
145 | }, | |
894d2491 | 146 | { } |
1da177e4 LT |
147 | }; |
148 | ||
83d5cde4 | 149 | static const struct block_device_operations md_fops; |
1da177e4 | 150 | |
f91de92e N |
151 | static int start_readonly; |
152 | ||
a167f663 N |
153 | /* bio_clone_mddev |
154 | * like bio_clone, but with a local bio set | |
155 | */ | |
156 | ||
a167f663 | 157 | struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, |
fd01b88c | 158 | struct mddev *mddev) |
a167f663 N |
159 | { |
160 | struct bio *b; | |
a167f663 N |
161 | |
162 | if (!mddev || !mddev->bio_set) | |
163 | return bio_alloc(gfp_mask, nr_iovecs); | |
164 | ||
395c72a7 | 165 | b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set); |
a167f663 N |
166 | if (!b) |
167 | return NULL; | |
a167f663 N |
168 | return b; |
169 | } | |
170 | EXPORT_SYMBOL_GPL(bio_alloc_mddev); | |
171 | ||
172 | struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask, | |
fd01b88c | 173 | struct mddev *mddev) |
a167f663 | 174 | { |
a167f663 N |
175 | if (!mddev || !mddev->bio_set) |
176 | return bio_clone(bio, gfp_mask); | |
177 | ||
bf800ef1 | 178 | return bio_clone_bioset(bio, gfp_mask, mddev->bio_set); |
a167f663 N |
179 | } |
180 | EXPORT_SYMBOL_GPL(bio_clone_mddev); | |
181 | ||
d7603b7e N |
182 | /* |
183 | * We have a system wide 'event count' that is incremented | |
184 | * on any 'interesting' event, and readers of /proc/mdstat | |
185 | * can use 'poll' or 'select' to find out when the event | |
186 | * count increases. | |
187 | * | |
188 | * Events are: | |
189 | * start array, stop array, error, add device, remove device, | |
190 | * start build, activate spare | |
191 | */ | |
2989ddbd | 192 | static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters); |
d7603b7e | 193 | static atomic_t md_event_count; |
fd01b88c | 194 | void md_new_event(struct mddev *mddev) |
d7603b7e N |
195 | { |
196 | atomic_inc(&md_event_count); | |
197 | wake_up(&md_event_waiters); | |
198 | } | |
29269553 | 199 | EXPORT_SYMBOL_GPL(md_new_event); |
d7603b7e | 200 | |
c331eb04 N |
201 | /* Alternate version that can be called from interrupts |
202 | * when calling sysfs_notify isn't needed. | |
203 | */ | |
fd01b88c | 204 | static void md_new_event_inintr(struct mddev *mddev) |
c331eb04 N |
205 | { |
206 | atomic_inc(&md_event_count); | |
207 | wake_up(&md_event_waiters); | |
208 | } | |
209 | ||
1da177e4 LT |
210 | /* |
211 | * Enables to iterate over all existing md arrays | |
212 | * all_mddevs_lock protects this list. | |
213 | */ | |
214 | static LIST_HEAD(all_mddevs); | |
215 | static DEFINE_SPINLOCK(all_mddevs_lock); | |
216 | ||
1da177e4 LT |
217 | /* |
218 | * iterates through all used mddevs in the system. | |
219 | * We take care to grab the all_mddevs_lock whenever navigating | |
220 | * the list, and to always hold a refcount when unlocked. | |
221 | * Any code which breaks out of this loop while own | |
222 | * a reference to the current mddev and must mddev_put it. | |
223 | */ | |
fd01b88c | 224 | #define for_each_mddev(_mddev,_tmp) \ |
1da177e4 | 225 | \ |
f72ffdd6 | 226 | for (({ spin_lock(&all_mddevs_lock); \ |
fd01b88c N |
227 | _tmp = all_mddevs.next; \ |
228 | _mddev = NULL;}); \ | |
229 | ({ if (_tmp != &all_mddevs) \ | |
230 | mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\ | |
1da177e4 | 231 | spin_unlock(&all_mddevs_lock); \ |
fd01b88c N |
232 | if (_mddev) mddev_put(_mddev); \ |
233 | _mddev = list_entry(_tmp, struct mddev, all_mddevs); \ | |
234 | _tmp != &all_mddevs;}); \ | |
1da177e4 | 235 | ({ spin_lock(&all_mddevs_lock); \ |
fd01b88c | 236 | _tmp = _tmp->next;}) \ |
1da177e4 LT |
237 | ) |
238 | ||
409c57f3 N |
239 | /* Rather than calling directly into the personality make_request function, |
240 | * IO requests come here first so that we can check if the device is | |
241 | * being suspended pending a reconfiguration. | |
242 | * We hold a refcount over the call to ->make_request. By the time that | |
243 | * call has finished, the bio has been linked into some internal structure | |
244 | * and so is visible to ->quiesce(), so we don't need the refcount any more. | |
245 | */ | |
5a7bbad2 | 246 | static void md_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 247 | { |
49077326 | 248 | const int rw = bio_data_dir(bio); |
fd01b88c | 249 | struct mddev *mddev = q->queuedata; |
49077326 | 250 | int cpu; |
e91ece55 | 251 | unsigned int sectors; |
49077326 | 252 | |
0ca69886 N |
253 | if (mddev == NULL || mddev->pers == NULL |
254 | || !mddev->ready) { | |
409c57f3 | 255 | bio_io_error(bio); |
5a7bbad2 | 256 | return; |
409c57f3 | 257 | } |
bbfa57c0 SR |
258 | if (mddev->ro == 1 && unlikely(rw == WRITE)) { |
259 | bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS); | |
260 | return; | |
261 | } | |
0ca69886 | 262 | smp_rmb(); /* Ensure implications of 'active' are visible */ |
409c57f3 | 263 | rcu_read_lock(); |
e9c7469b | 264 | if (mddev->suspended) { |
409c57f3 N |
265 | DEFINE_WAIT(__wait); |
266 | for (;;) { | |
267 | prepare_to_wait(&mddev->sb_wait, &__wait, | |
268 | TASK_UNINTERRUPTIBLE); | |
e9c7469b | 269 | if (!mddev->suspended) |
409c57f3 N |
270 | break; |
271 | rcu_read_unlock(); | |
272 | schedule(); | |
273 | rcu_read_lock(); | |
274 | } | |
275 | finish_wait(&mddev->sb_wait, &__wait); | |
276 | } | |
277 | atomic_inc(&mddev->active_io); | |
278 | rcu_read_unlock(); | |
49077326 | 279 | |
e91ece55 CM |
280 | /* |
281 | * save the sectors now since our bio can | |
282 | * go away inside make_request | |
283 | */ | |
284 | sectors = bio_sectors(bio); | |
5a7bbad2 | 285 | mddev->pers->make_request(mddev, bio); |
49077326 N |
286 | |
287 | cpu = part_stat_lock(); | |
288 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
e91ece55 | 289 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors); |
49077326 N |
290 | part_stat_unlock(); |
291 | ||
409c57f3 N |
292 | if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended) |
293 | wake_up(&mddev->sb_wait); | |
409c57f3 N |
294 | } |
295 | ||
9e35b99c N |
296 | /* mddev_suspend makes sure no new requests are submitted |
297 | * to the device, and that any requests that have been submitted | |
298 | * are completely handled. | |
299 | * Once ->stop is called and completes, the module will be completely | |
300 | * unused. | |
301 | */ | |
fd01b88c | 302 | void mddev_suspend(struct mddev *mddev) |
409c57f3 N |
303 | { |
304 | BUG_ON(mddev->suspended); | |
305 | mddev->suspended = 1; | |
306 | synchronize_rcu(); | |
307 | wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0); | |
308 | mddev->pers->quiesce(mddev, 1); | |
0d9f4f13 JB |
309 | |
310 | del_timer_sync(&mddev->safemode_timer); | |
409c57f3 | 311 | } |
390ee602 | 312 | EXPORT_SYMBOL_GPL(mddev_suspend); |
409c57f3 | 313 | |
fd01b88c | 314 | void mddev_resume(struct mddev *mddev) |
409c57f3 N |
315 | { |
316 | mddev->suspended = 0; | |
317 | wake_up(&mddev->sb_wait); | |
318 | mddev->pers->quiesce(mddev, 0); | |
0fd018af | 319 | |
47525e59 | 320 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
0fd018af JB |
321 | md_wakeup_thread(mddev->thread); |
322 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
1da177e4 | 323 | } |
390ee602 | 324 | EXPORT_SYMBOL_GPL(mddev_resume); |
1da177e4 | 325 | |
fd01b88c | 326 | int mddev_congested(struct mddev *mddev, int bits) |
3fa841d7 N |
327 | { |
328 | return mddev->suspended; | |
329 | } | |
330 | EXPORT_SYMBOL(mddev_congested); | |
331 | ||
a2826aa9 | 332 | /* |
e9c7469b | 333 | * Generic flush handling for md |
a2826aa9 N |
334 | */ |
335 | ||
e9c7469b | 336 | static void md_end_flush(struct bio *bio, int err) |
a2826aa9 | 337 | { |
3cb03002 | 338 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 339 | struct mddev *mddev = rdev->mddev; |
a2826aa9 N |
340 | |
341 | rdev_dec_pending(rdev, mddev); | |
342 | ||
343 | if (atomic_dec_and_test(&mddev->flush_pending)) { | |
e9c7469b | 344 | /* The pre-request flush has finished */ |
e804ac78 | 345 | queue_work(md_wq, &mddev->flush_work); |
a2826aa9 N |
346 | } |
347 | bio_put(bio); | |
348 | } | |
349 | ||
a7a07e69 N |
350 | static void md_submit_flush_data(struct work_struct *ws); |
351 | ||
a035fc3e | 352 | static void submit_flushes(struct work_struct *ws) |
a2826aa9 | 353 | { |
fd01b88c | 354 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
3cb03002 | 355 | struct md_rdev *rdev; |
a2826aa9 | 356 | |
a7a07e69 N |
357 | INIT_WORK(&mddev->flush_work, md_submit_flush_data); |
358 | atomic_set(&mddev->flush_pending, 1); | |
a2826aa9 | 359 | rcu_read_lock(); |
dafb20fa | 360 | rdev_for_each_rcu(rdev, mddev) |
a2826aa9 N |
361 | if (rdev->raid_disk >= 0 && |
362 | !test_bit(Faulty, &rdev->flags)) { | |
363 | /* Take two references, one is dropped | |
364 | * when request finishes, one after | |
365 | * we reclaim rcu_read_lock | |
366 | */ | |
367 | struct bio *bi; | |
368 | atomic_inc(&rdev->nr_pending); | |
369 | atomic_inc(&rdev->nr_pending); | |
370 | rcu_read_unlock(); | |
b5e1b8ce | 371 | bi = bio_alloc_mddev(GFP_NOIO, 0, mddev); |
e9c7469b | 372 | bi->bi_end_io = md_end_flush; |
a2826aa9 N |
373 | bi->bi_private = rdev; |
374 | bi->bi_bdev = rdev->bdev; | |
375 | atomic_inc(&mddev->flush_pending); | |
e9c7469b | 376 | submit_bio(WRITE_FLUSH, bi); |
a2826aa9 N |
377 | rcu_read_lock(); |
378 | rdev_dec_pending(rdev, mddev); | |
379 | } | |
380 | rcu_read_unlock(); | |
a7a07e69 N |
381 | if (atomic_dec_and_test(&mddev->flush_pending)) |
382 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 N |
383 | } |
384 | ||
e9c7469b | 385 | static void md_submit_flush_data(struct work_struct *ws) |
a2826aa9 | 386 | { |
fd01b88c | 387 | struct mddev *mddev = container_of(ws, struct mddev, flush_work); |
e9c7469b | 388 | struct bio *bio = mddev->flush_bio; |
a2826aa9 | 389 | |
4f024f37 | 390 | if (bio->bi_iter.bi_size == 0) |
a2826aa9 N |
391 | /* an empty barrier - all done */ |
392 | bio_endio(bio, 0); | |
393 | else { | |
e9c7469b | 394 | bio->bi_rw &= ~REQ_FLUSH; |
5a7bbad2 | 395 | mddev->pers->make_request(mddev, bio); |
a2826aa9 | 396 | } |
2b74e12e N |
397 | |
398 | mddev->flush_bio = NULL; | |
399 | wake_up(&mddev->sb_wait); | |
a2826aa9 N |
400 | } |
401 | ||
fd01b88c | 402 | void md_flush_request(struct mddev *mddev, struct bio *bio) |
a2826aa9 N |
403 | { |
404 | spin_lock_irq(&mddev->write_lock); | |
405 | wait_event_lock_irq(mddev->sb_wait, | |
e9c7469b | 406 | !mddev->flush_bio, |
eed8c02e | 407 | mddev->write_lock); |
e9c7469b | 408 | mddev->flush_bio = bio; |
a2826aa9 N |
409 | spin_unlock_irq(&mddev->write_lock); |
410 | ||
a035fc3e N |
411 | INIT_WORK(&mddev->flush_work, submit_flushes); |
412 | queue_work(md_wq, &mddev->flush_work); | |
a2826aa9 | 413 | } |
e9c7469b | 414 | EXPORT_SYMBOL(md_flush_request); |
409c57f3 | 415 | |
74018dc3 | 416 | void md_unplug(struct blk_plug_cb *cb, bool from_schedule) |
97658cdd | 417 | { |
9cbb1750 N |
418 | struct mddev *mddev = cb->data; |
419 | md_wakeup_thread(mddev->thread); | |
420 | kfree(cb); | |
97658cdd | 421 | } |
9cbb1750 | 422 | EXPORT_SYMBOL(md_unplug); |
2ac87401 | 423 | |
fd01b88c | 424 | static inline struct mddev *mddev_get(struct mddev *mddev) |
1da177e4 LT |
425 | { |
426 | atomic_inc(&mddev->active); | |
427 | return mddev; | |
428 | } | |
429 | ||
5fd3a17e | 430 | static void mddev_delayed_delete(struct work_struct *ws); |
d3374825 | 431 | |
fd01b88c | 432 | static void mddev_put(struct mddev *mddev) |
1da177e4 | 433 | { |
a167f663 N |
434 | struct bio_set *bs = NULL; |
435 | ||
1da177e4 LT |
436 | if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) |
437 | return; | |
d3374825 | 438 | if (!mddev->raid_disks && list_empty(&mddev->disks) && |
cbd19983 N |
439 | mddev->ctime == 0 && !mddev->hold_active) { |
440 | /* Array is not configured at all, and not held active, | |
441 | * so destroy it */ | |
af8a2434 | 442 | list_del_init(&mddev->all_mddevs); |
a167f663 N |
443 | bs = mddev->bio_set; |
444 | mddev->bio_set = NULL; | |
d3374825 | 445 | if (mddev->gendisk) { |
e804ac78 TH |
446 | /* We did a probe so need to clean up. Call |
447 | * queue_work inside the spinlock so that | |
448 | * flush_workqueue() after mddev_find will | |
449 | * succeed in waiting for the work to be done. | |
d3374825 N |
450 | */ |
451 | INIT_WORK(&mddev->del_work, mddev_delayed_delete); | |
e804ac78 | 452 | queue_work(md_misc_wq, &mddev->del_work); |
d3374825 N |
453 | } else |
454 | kfree(mddev); | |
455 | } | |
456 | spin_unlock(&all_mddevs_lock); | |
a167f663 N |
457 | if (bs) |
458 | bioset_free(bs); | |
1da177e4 LT |
459 | } |
460 | ||
fd01b88c | 461 | void mddev_init(struct mddev *mddev) |
fafd7fb0 N |
462 | { |
463 | mutex_init(&mddev->open_mutex); | |
464 | mutex_init(&mddev->reconfig_mutex); | |
465 | mutex_init(&mddev->bitmap_info.mutex); | |
466 | INIT_LIST_HEAD(&mddev->disks); | |
467 | INIT_LIST_HEAD(&mddev->all_mddevs); | |
468 | init_timer(&mddev->safemode_timer); | |
469 | atomic_set(&mddev->active, 1); | |
470 | atomic_set(&mddev->openers, 0); | |
471 | atomic_set(&mddev->active_io, 0); | |
472 | spin_lock_init(&mddev->write_lock); | |
473 | atomic_set(&mddev->flush_pending, 0); | |
474 | init_waitqueue_head(&mddev->sb_wait); | |
475 | init_waitqueue_head(&mddev->recovery_wait); | |
476 | mddev->reshape_position = MaxSector; | |
2c810cdd | 477 | mddev->reshape_backwards = 0; |
c4a39551 | 478 | mddev->last_sync_action = "none"; |
fafd7fb0 N |
479 | mddev->resync_min = 0; |
480 | mddev->resync_max = MaxSector; | |
481 | mddev->level = LEVEL_NONE; | |
482 | } | |
390ee602 | 483 | EXPORT_SYMBOL_GPL(mddev_init); |
fafd7fb0 | 484 | |
f72ffdd6 | 485 | static struct mddev *mddev_find(dev_t unit) |
1da177e4 | 486 | { |
fd01b88c | 487 | struct mddev *mddev, *new = NULL; |
1da177e4 | 488 | |
8f5f02c4 N |
489 | if (unit && MAJOR(unit) != MD_MAJOR) |
490 | unit &= ~((1<<MdpMinorShift)-1); | |
491 | ||
1da177e4 LT |
492 | retry: |
493 | spin_lock(&all_mddevs_lock); | |
efeb53c0 N |
494 | |
495 | if (unit) { | |
496 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
497 | if (mddev->unit == unit) { | |
498 | mddev_get(mddev); | |
499 | spin_unlock(&all_mddevs_lock); | |
500 | kfree(new); | |
501 | return mddev; | |
502 | } | |
503 | ||
504 | if (new) { | |
505 | list_add(&new->all_mddevs, &all_mddevs); | |
1da177e4 | 506 | spin_unlock(&all_mddevs_lock); |
efeb53c0 N |
507 | new->hold_active = UNTIL_IOCTL; |
508 | return new; | |
1da177e4 | 509 | } |
efeb53c0 N |
510 | } else if (new) { |
511 | /* find an unused unit number */ | |
512 | static int next_minor = 512; | |
513 | int start = next_minor; | |
514 | int is_free = 0; | |
515 | int dev = 0; | |
516 | while (!is_free) { | |
517 | dev = MKDEV(MD_MAJOR, next_minor); | |
518 | next_minor++; | |
519 | if (next_minor > MINORMASK) | |
520 | next_minor = 0; | |
521 | if (next_minor == start) { | |
522 | /* Oh dear, all in use. */ | |
523 | spin_unlock(&all_mddevs_lock); | |
524 | kfree(new); | |
525 | return NULL; | |
526 | } | |
f72ffdd6 | 527 | |
efeb53c0 N |
528 | is_free = 1; |
529 | list_for_each_entry(mddev, &all_mddevs, all_mddevs) | |
530 | if (mddev->unit == dev) { | |
531 | is_free = 0; | |
532 | break; | |
533 | } | |
534 | } | |
535 | new->unit = dev; | |
536 | new->md_minor = MINOR(dev); | |
537 | new->hold_active = UNTIL_STOP; | |
1da177e4 LT |
538 | list_add(&new->all_mddevs, &all_mddevs); |
539 | spin_unlock(&all_mddevs_lock); | |
540 | return new; | |
541 | } | |
542 | spin_unlock(&all_mddevs_lock); | |
543 | ||
9ffae0cf | 544 | new = kzalloc(sizeof(*new), GFP_KERNEL); |
1da177e4 LT |
545 | if (!new) |
546 | return NULL; | |
547 | ||
1da177e4 LT |
548 | new->unit = unit; |
549 | if (MAJOR(unit) == MD_MAJOR) | |
550 | new->md_minor = MINOR(unit); | |
551 | else | |
552 | new->md_minor = MINOR(unit) >> MdpMinorShift; | |
553 | ||
fafd7fb0 | 554 | mddev_init(new); |
1da177e4 | 555 | |
1da177e4 LT |
556 | goto retry; |
557 | } | |
558 | ||
f72ffdd6 | 559 | static inline int __must_check mddev_lock(struct mddev *mddev) |
1da177e4 | 560 | { |
df5b89b3 | 561 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
1da177e4 LT |
562 | } |
563 | ||
29f097c4 N |
564 | /* Sometimes we need to take the lock in a situation where |
565 | * failure due to interrupts is not acceptable. | |
566 | */ | |
f72ffdd6 | 567 | static inline void mddev_lock_nointr(struct mddev *mddev) |
29f097c4 N |
568 | { |
569 | mutex_lock(&mddev->reconfig_mutex); | |
570 | } | |
571 | ||
fd01b88c | 572 | static inline int mddev_is_locked(struct mddev *mddev) |
b522adcd DW |
573 | { |
574 | return mutex_is_locked(&mddev->reconfig_mutex); | |
575 | } | |
576 | ||
f72ffdd6 | 577 | static inline int mddev_trylock(struct mddev *mddev) |
1da177e4 | 578 | { |
df5b89b3 | 579 | return mutex_trylock(&mddev->reconfig_mutex); |
1da177e4 LT |
580 | } |
581 | ||
b6eb127d N |
582 | static struct attribute_group md_redundancy_group; |
583 | ||
f72ffdd6 | 584 | static void mddev_unlock(struct mddev *mddev) |
1da177e4 | 585 | { |
a64c876f | 586 | if (mddev->to_remove) { |
b6eb127d N |
587 | /* These cannot be removed under reconfig_mutex as |
588 | * an access to the files will try to take reconfig_mutex | |
589 | * while holding the file unremovable, which leads to | |
590 | * a deadlock. | |
bb4f1e9d N |
591 | * So hold set sysfs_active while the remove in happeing, |
592 | * and anything else which might set ->to_remove or my | |
593 | * otherwise change the sysfs namespace will fail with | |
594 | * -EBUSY if sysfs_active is still set. | |
595 | * We set sysfs_active under reconfig_mutex and elsewhere | |
596 | * test it under the same mutex to ensure its correct value | |
597 | * is seen. | |
b6eb127d | 598 | */ |
a64c876f N |
599 | struct attribute_group *to_remove = mddev->to_remove; |
600 | mddev->to_remove = NULL; | |
bb4f1e9d | 601 | mddev->sysfs_active = 1; |
b6eb127d N |
602 | mutex_unlock(&mddev->reconfig_mutex); |
603 | ||
00bcb4ac N |
604 | if (mddev->kobj.sd) { |
605 | if (to_remove != &md_redundancy_group) | |
606 | sysfs_remove_group(&mddev->kobj, to_remove); | |
607 | if (mddev->pers == NULL || | |
608 | mddev->pers->sync_request == NULL) { | |
609 | sysfs_remove_group(&mddev->kobj, &md_redundancy_group); | |
610 | if (mddev->sysfs_action) | |
611 | sysfs_put(mddev->sysfs_action); | |
612 | mddev->sysfs_action = NULL; | |
613 | } | |
a64c876f | 614 | } |
bb4f1e9d | 615 | mddev->sysfs_active = 0; |
b6eb127d N |
616 | } else |
617 | mutex_unlock(&mddev->reconfig_mutex); | |
1da177e4 | 618 | |
751e67ca CD |
619 | /* As we've dropped the mutex we need a spinlock to |
620 | * make sure the thread doesn't disappear | |
01f96c0a N |
621 | */ |
622 | spin_lock(&pers_lock); | |
005eca5e | 623 | md_wakeup_thread(mddev->thread); |
01f96c0a | 624 | spin_unlock(&pers_lock); |
1da177e4 LT |
625 | } |
626 | ||
1ca69c4b N |
627 | static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr) |
628 | { | |
629 | struct md_rdev *rdev; | |
630 | ||
631 | rdev_for_each_rcu(rdev, mddev) | |
632 | if (rdev->desc_nr == nr) | |
633 | return rdev; | |
634 | ||
635 | return NULL; | |
636 | } | |
637 | ||
638 | static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev) | |
1da177e4 | 639 | { |
3cb03002 | 640 | struct md_rdev *rdev; |
1da177e4 | 641 | |
dafb20fa | 642 | rdev_for_each(rdev, mddev) |
1da177e4 LT |
643 | if (rdev->bdev->bd_dev == dev) |
644 | return rdev; | |
159ec1fc | 645 | |
1da177e4 LT |
646 | return NULL; |
647 | } | |
648 | ||
1ca69c4b N |
649 | static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev) |
650 | { | |
651 | struct md_rdev *rdev; | |
652 | ||
653 | rdev_for_each_rcu(rdev, mddev) | |
654 | if (rdev->bdev->bd_dev == dev) | |
655 | return rdev; | |
656 | ||
657 | return NULL; | |
658 | } | |
659 | ||
84fc4b56 | 660 | static struct md_personality *find_pers(int level, char *clevel) |
2604b703 | 661 | { |
84fc4b56 | 662 | struct md_personality *pers; |
d9d166c2 N |
663 | list_for_each_entry(pers, &pers_list, list) { |
664 | if (level != LEVEL_NONE && pers->level == level) | |
2604b703 | 665 | return pers; |
d9d166c2 N |
666 | if (strcmp(pers->name, clevel)==0) |
667 | return pers; | |
668 | } | |
2604b703 N |
669 | return NULL; |
670 | } | |
671 | ||
b73df2d3 | 672 | /* return the offset of the super block in 512byte sectors */ |
3cb03002 | 673 | static inline sector_t calc_dev_sboffset(struct md_rdev *rdev) |
1da177e4 | 674 | { |
57b2caa3 | 675 | sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512; |
b73df2d3 | 676 | return MD_NEW_SIZE_SECTORS(num_sectors); |
1da177e4 LT |
677 | } |
678 | ||
f72ffdd6 | 679 | static int alloc_disk_sb(struct md_rdev *rdev) |
1da177e4 | 680 | { |
1da177e4 LT |
681 | rdev->sb_page = alloc_page(GFP_KERNEL); |
682 | if (!rdev->sb_page) { | |
683 | printk(KERN_ALERT "md: out of memory.\n"); | |
ebc24337 | 684 | return -ENOMEM; |
1da177e4 LT |
685 | } |
686 | ||
687 | return 0; | |
688 | } | |
689 | ||
545c8795 | 690 | void md_rdev_clear(struct md_rdev *rdev) |
1da177e4 LT |
691 | { |
692 | if (rdev->sb_page) { | |
2d1f3b5d | 693 | put_page(rdev->sb_page); |
1da177e4 LT |
694 | rdev->sb_loaded = 0; |
695 | rdev->sb_page = NULL; | |
0f420358 | 696 | rdev->sb_start = 0; |
dd8ac336 | 697 | rdev->sectors = 0; |
1da177e4 | 698 | } |
2699b672 N |
699 | if (rdev->bb_page) { |
700 | put_page(rdev->bb_page); | |
701 | rdev->bb_page = NULL; | |
702 | } | |
4fa2f327 N |
703 | kfree(rdev->badblocks.page); |
704 | rdev->badblocks.page = NULL; | |
1da177e4 | 705 | } |
545c8795 | 706 | EXPORT_SYMBOL_GPL(md_rdev_clear); |
1da177e4 | 707 | |
6712ecf8 | 708 | static void super_written(struct bio *bio, int error) |
7bfa19f2 | 709 | { |
3cb03002 | 710 | struct md_rdev *rdev = bio->bi_private; |
fd01b88c | 711 | struct mddev *mddev = rdev->mddev; |
7bfa19f2 | 712 | |
3a0f5bbb N |
713 | if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) { |
714 | printk("md: super_written gets error=%d, uptodate=%d\n", | |
715 | error, test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
716 | WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags)); | |
a9701a30 | 717 | md_error(mddev, rdev); |
3a0f5bbb | 718 | } |
7bfa19f2 | 719 | |
a9701a30 N |
720 | if (atomic_dec_and_test(&mddev->pending_writes)) |
721 | wake_up(&mddev->sb_wait); | |
f8b58edf | 722 | bio_put(bio); |
7bfa19f2 N |
723 | } |
724 | ||
fd01b88c | 725 | void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
7bfa19f2 N |
726 | sector_t sector, int size, struct page *page) |
727 | { | |
728 | /* write first size bytes of page to sector of rdev | |
729 | * Increment mddev->pending_writes before returning | |
730 | * and decrement it on completion, waking up sb_wait | |
731 | * if zero is reached. | |
732 | * If an error occurred, call md_error | |
733 | */ | |
a167f663 | 734 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev); |
7bfa19f2 | 735 | |
a6ff7e08 | 736 | bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev; |
4f024f37 | 737 | bio->bi_iter.bi_sector = sector; |
7bfa19f2 N |
738 | bio_add_page(bio, page, size, 0); |
739 | bio->bi_private = rdev; | |
740 | bio->bi_end_io = super_written; | |
a9701a30 | 741 | |
7bfa19f2 | 742 | atomic_inc(&mddev->pending_writes); |
a5bf4df0 | 743 | submit_bio(WRITE_FLUSH_FUA, bio); |
a9701a30 N |
744 | } |
745 | ||
fd01b88c | 746 | void md_super_wait(struct mddev *mddev) |
a9701a30 | 747 | { |
e9c7469b | 748 | /* wait for all superblock writes that were scheduled to complete */ |
1967cd56 | 749 | wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); |
7bfa19f2 N |
750 | } |
751 | ||
3cb03002 | 752 | int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
ccebd4c4 | 753 | struct page *page, int rw, bool metadata_op) |
1da177e4 | 754 | { |
a167f663 | 755 | struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev); |
1da177e4 LT |
756 | int ret; |
757 | ||
a6ff7e08 JB |
758 | bio->bi_bdev = (metadata_op && rdev->meta_bdev) ? |
759 | rdev->meta_bdev : rdev->bdev; | |
ccebd4c4 | 760 | if (metadata_op) |
4f024f37 | 761 | bio->bi_iter.bi_sector = sector + rdev->sb_start; |
1fdd6fc9 N |
762 | else if (rdev->mddev->reshape_position != MaxSector && |
763 | (rdev->mddev->reshape_backwards == | |
764 | (sector >= rdev->mddev->reshape_position))) | |
4f024f37 | 765 | bio->bi_iter.bi_sector = sector + rdev->new_data_offset; |
ccebd4c4 | 766 | else |
4f024f37 | 767 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
1da177e4 | 768 | bio_add_page(bio, page, size, 0); |
c170bbb4 | 769 | submit_bio_wait(rw, bio); |
1da177e4 LT |
770 | |
771 | ret = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
772 | bio_put(bio); | |
773 | return ret; | |
774 | } | |
a8745db2 | 775 | EXPORT_SYMBOL_GPL(sync_page_io); |
1da177e4 | 776 | |
f72ffdd6 | 777 | static int read_disk_sb(struct md_rdev *rdev, int size) |
1da177e4 LT |
778 | { |
779 | char b[BDEVNAME_SIZE]; | |
403df478 | 780 | |
1da177e4 LT |
781 | if (rdev->sb_loaded) |
782 | return 0; | |
783 | ||
ccebd4c4 | 784 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true)) |
1da177e4 LT |
785 | goto fail; |
786 | rdev->sb_loaded = 1; | |
787 | return 0; | |
788 | ||
789 | fail: | |
790 | printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", | |
791 | bdevname(rdev->bdev,b)); | |
792 | return -EINVAL; | |
793 | } | |
794 | ||
795 | static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) | |
796 | { | |
f72ffdd6 | 797 | return sb1->set_uuid0 == sb2->set_uuid0 && |
05710466 AN |
798 | sb1->set_uuid1 == sb2->set_uuid1 && |
799 | sb1->set_uuid2 == sb2->set_uuid2 && | |
800 | sb1->set_uuid3 == sb2->set_uuid3; | |
1da177e4 LT |
801 | } |
802 | ||
1da177e4 LT |
803 | static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) |
804 | { | |
805 | int ret; | |
806 | mdp_super_t *tmp1, *tmp2; | |
807 | ||
808 | tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); | |
809 | tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); | |
810 | ||
811 | if (!tmp1 || !tmp2) { | |
812 | ret = 0; | |
35020f1a | 813 | printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n"); |
1da177e4 LT |
814 | goto abort; |
815 | } | |
816 | ||
817 | *tmp1 = *sb1; | |
818 | *tmp2 = *sb2; | |
819 | ||
820 | /* | |
821 | * nr_disks is not constant | |
822 | */ | |
823 | tmp1->nr_disks = 0; | |
824 | tmp2->nr_disks = 0; | |
825 | ||
ce0c8e05 | 826 | ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0); |
1da177e4 | 827 | abort: |
990a8baf JJ |
828 | kfree(tmp1); |
829 | kfree(tmp2); | |
1da177e4 LT |
830 | return ret; |
831 | } | |
832 | ||
4d167f09 N |
833 | static u32 md_csum_fold(u32 csum) |
834 | { | |
835 | csum = (csum & 0xffff) + (csum >> 16); | |
836 | return (csum & 0xffff) + (csum >> 16); | |
837 | } | |
838 | ||
f72ffdd6 | 839 | static unsigned int calc_sb_csum(mdp_super_t *sb) |
1da177e4 | 840 | { |
4d167f09 N |
841 | u64 newcsum = 0; |
842 | u32 *sb32 = (u32*)sb; | |
843 | int i; | |
1da177e4 LT |
844 | unsigned int disk_csum, csum; |
845 | ||
846 | disk_csum = sb->sb_csum; | |
847 | sb->sb_csum = 0; | |
4d167f09 N |
848 | |
849 | for (i = 0; i < MD_SB_BYTES/4 ; i++) | |
850 | newcsum += sb32[i]; | |
851 | csum = (newcsum & 0xffffffff) + (newcsum>>32); | |
852 | ||
4d167f09 N |
853 | #ifdef CONFIG_ALPHA |
854 | /* This used to use csum_partial, which was wrong for several | |
855 | * reasons including that different results are returned on | |
856 | * different architectures. It isn't critical that we get exactly | |
857 | * the same return value as before (we always csum_fold before | |
858 | * testing, and that removes any differences). However as we | |
859 | * know that csum_partial always returned a 16bit value on | |
860 | * alphas, do a fold to maximise conformity to previous behaviour. | |
861 | */ | |
862 | sb->sb_csum = md_csum_fold(disk_csum); | |
863 | #else | |
1da177e4 | 864 | sb->sb_csum = disk_csum; |
4d167f09 | 865 | #endif |
1da177e4 LT |
866 | return csum; |
867 | } | |
868 | ||
1da177e4 LT |
869 | /* |
870 | * Handle superblock details. | |
871 | * We want to be able to handle multiple superblock formats | |
872 | * so we have a common interface to them all, and an array of | |
873 | * different handlers. | |
874 | * We rely on user-space to write the initial superblock, and support | |
875 | * reading and updating of superblocks. | |
876 | * Interface methods are: | |
3cb03002 | 877 | * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
878 | * loads and validates a superblock on dev. |
879 | * if refdev != NULL, compare superblocks on both devices | |
880 | * Return: | |
881 | * 0 - dev has a superblock that is compatible with refdev | |
882 | * 1 - dev has a superblock that is compatible and newer than refdev | |
883 | * so dev should be used as the refdev in future | |
884 | * -EINVAL superblock incompatible or invalid | |
885 | * -othererror e.g. -EIO | |
886 | * | |
fd01b88c | 887 | * int validate_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
888 | * Verify that dev is acceptable into mddev. |
889 | * The first time, mddev->raid_disks will be 0, and data from | |
890 | * dev should be merged in. Subsequent calls check that dev | |
891 | * is new enough. Return 0 or -EINVAL | |
892 | * | |
fd01b88c | 893 | * void sync_super(struct mddev *mddev, struct md_rdev *dev) |
1da177e4 LT |
894 | * Update the superblock for rdev with data in mddev |
895 | * This does not write to disc. | |
896 | * | |
897 | */ | |
898 | ||
899 | struct super_type { | |
0cd17fec CW |
900 | char *name; |
901 | struct module *owner; | |
c6563a8c N |
902 | int (*load_super)(struct md_rdev *rdev, |
903 | struct md_rdev *refdev, | |
0cd17fec | 904 | int minor_version); |
c6563a8c N |
905 | int (*validate_super)(struct mddev *mddev, |
906 | struct md_rdev *rdev); | |
907 | void (*sync_super)(struct mddev *mddev, | |
908 | struct md_rdev *rdev); | |
3cb03002 | 909 | unsigned long long (*rdev_size_change)(struct md_rdev *rdev, |
15f4a5fd | 910 | sector_t num_sectors); |
c6563a8c N |
911 | int (*allow_new_offset)(struct md_rdev *rdev, |
912 | unsigned long long new_offset); | |
1da177e4 LT |
913 | }; |
914 | ||
0894cc30 AN |
915 | /* |
916 | * Check that the given mddev has no bitmap. | |
917 | * | |
918 | * This function is called from the run method of all personalities that do not | |
919 | * support bitmaps. It prints an error message and returns non-zero if mddev | |
920 | * has a bitmap. Otherwise, it returns 0. | |
921 | * | |
922 | */ | |
fd01b88c | 923 | int md_check_no_bitmap(struct mddev *mddev) |
0894cc30 | 924 | { |
c3d9714e | 925 | if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset) |
0894cc30 AN |
926 | return 0; |
927 | printk(KERN_ERR "%s: bitmaps are not supported for %s\n", | |
928 | mdname(mddev), mddev->pers->name); | |
929 | return 1; | |
930 | } | |
931 | EXPORT_SYMBOL(md_check_no_bitmap); | |
932 | ||
1da177e4 | 933 | /* |
f72ffdd6 | 934 | * load_super for 0.90.0 |
1da177e4 | 935 | */ |
3cb03002 | 936 | static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
937 | { |
938 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
939 | mdp_super_t *sb; | |
940 | int ret; | |
1da177e4 LT |
941 | |
942 | /* | |
0f420358 | 943 | * Calculate the position of the superblock (512byte sectors), |
1da177e4 LT |
944 | * it's at the end of the disk. |
945 | * | |
946 | * It also happens to be a multiple of 4Kb. | |
947 | */ | |
57b2caa3 | 948 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 949 | |
0002b271 | 950 | ret = read_disk_sb(rdev, MD_SB_BYTES); |
1da177e4 LT |
951 | if (ret) return ret; |
952 | ||
953 | ret = -EINVAL; | |
954 | ||
955 | bdevname(rdev->bdev, b); | |
65a06f06 | 956 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
957 | |
958 | if (sb->md_magic != MD_SB_MAGIC) { | |
959 | printk(KERN_ERR "md: invalid raid superblock magic on %s\n", | |
960 | b); | |
961 | goto abort; | |
962 | } | |
963 | ||
964 | if (sb->major_version != 0 || | |
f6705578 N |
965 | sb->minor_version < 90 || |
966 | sb->minor_version > 91) { | |
1da177e4 LT |
967 | printk(KERN_WARNING "Bad version number %d.%d on %s\n", |
968 | sb->major_version, sb->minor_version, | |
969 | b); | |
970 | goto abort; | |
971 | } | |
972 | ||
973 | if (sb->raid_disks <= 0) | |
974 | goto abort; | |
975 | ||
4d167f09 | 976 | if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) { |
1da177e4 LT |
977 | printk(KERN_WARNING "md: invalid superblock checksum on %s\n", |
978 | b); | |
979 | goto abort; | |
980 | } | |
981 | ||
982 | rdev->preferred_minor = sb->md_minor; | |
983 | rdev->data_offset = 0; | |
c6563a8c | 984 | rdev->new_data_offset = 0; |
0002b271 | 985 | rdev->sb_size = MD_SB_BYTES; |
9f2f3830 | 986 | rdev->badblocks.shift = -1; |
1da177e4 LT |
987 | |
988 | if (sb->level == LEVEL_MULTIPATH) | |
989 | rdev->desc_nr = -1; | |
990 | else | |
991 | rdev->desc_nr = sb->this_disk.number; | |
992 | ||
9a7b2b0f | 993 | if (!refdev) { |
1da177e4 | 994 | ret = 1; |
9a7b2b0f | 995 | } else { |
1da177e4 | 996 | __u64 ev1, ev2; |
65a06f06 | 997 | mdp_super_t *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
998 | if (!uuid_equal(refsb, sb)) { |
999 | printk(KERN_WARNING "md: %s has different UUID to %s\n", | |
1000 | b, bdevname(refdev->bdev,b2)); | |
1001 | goto abort; | |
1002 | } | |
1003 | if (!sb_equal(refsb, sb)) { | |
1004 | printk(KERN_WARNING "md: %s has same UUID" | |
1005 | " but different superblock to %s\n", | |
1006 | b, bdevname(refdev->bdev, b2)); | |
1007 | goto abort; | |
1008 | } | |
1009 | ev1 = md_event(sb); | |
1010 | ev2 = md_event(refsb); | |
1011 | if (ev1 > ev2) | |
1012 | ret = 1; | |
f72ffdd6 | 1013 | else |
1da177e4 LT |
1014 | ret = 0; |
1015 | } | |
8190e754 | 1016 | rdev->sectors = rdev->sb_start; |
667a5313 N |
1017 | /* Limit to 4TB as metadata cannot record more than that. |
1018 | * (not needed for Linear and RAID0 as metadata doesn't | |
1019 | * record this size) | |
1020 | */ | |
1021 | if (rdev->sectors >= (2ULL << 32) && sb->level >= 1) | |
27a7b260 | 1022 | rdev->sectors = (2ULL << 32) - 2; |
1da177e4 | 1023 | |
27a7b260 | 1024 | if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) |
2bf071bf N |
1025 | /* "this cannot possibly happen" ... */ |
1026 | ret = -EINVAL; | |
1027 | ||
1da177e4 LT |
1028 | abort: |
1029 | return ret; | |
1030 | } | |
1031 | ||
1032 | /* | |
1033 | * validate_super for 0.90.0 | |
1034 | */ | |
fd01b88c | 1035 | static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1036 | { |
1037 | mdp_disk_t *desc; | |
65a06f06 | 1038 | mdp_super_t *sb = page_address(rdev->sb_page); |
07d84d10 | 1039 | __u64 ev1 = md_event(sb); |
1da177e4 | 1040 | |
41158c7e | 1041 | rdev->raid_disk = -1; |
c5d79adb N |
1042 | clear_bit(Faulty, &rdev->flags); |
1043 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1044 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1045 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1046 | |
1da177e4 LT |
1047 | if (mddev->raid_disks == 0) { |
1048 | mddev->major_version = 0; | |
1049 | mddev->minor_version = sb->minor_version; | |
1050 | mddev->patch_version = sb->patch_version; | |
e691063a | 1051 | mddev->external = 0; |
9d8f0363 | 1052 | mddev->chunk_sectors = sb->chunk_size >> 9; |
1da177e4 LT |
1053 | mddev->ctime = sb->ctime; |
1054 | mddev->utime = sb->utime; | |
1055 | mddev->level = sb->level; | |
d9d166c2 | 1056 | mddev->clevel[0] = 0; |
1da177e4 LT |
1057 | mddev->layout = sb->layout; |
1058 | mddev->raid_disks = sb->raid_disks; | |
27a7b260 | 1059 | mddev->dev_sectors = ((sector_t)sb->size) * 2; |
07d84d10 | 1060 | mddev->events = ev1; |
c3d9714e | 1061 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1062 | mddev->bitmap_info.space = 0; |
1063 | /* bitmap can use 60 K after the 4K superblocks */ | |
c3d9714e | 1064 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 1065 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
2c810cdd | 1066 | mddev->reshape_backwards = 0; |
1da177e4 | 1067 | |
f6705578 N |
1068 | if (mddev->minor_version >= 91) { |
1069 | mddev->reshape_position = sb->reshape_position; | |
1070 | mddev->delta_disks = sb->delta_disks; | |
1071 | mddev->new_level = sb->new_level; | |
1072 | mddev->new_layout = sb->new_layout; | |
664e7c41 | 1073 | mddev->new_chunk_sectors = sb->new_chunk >> 9; |
2c810cdd N |
1074 | if (mddev->delta_disks < 0) |
1075 | mddev->reshape_backwards = 1; | |
f6705578 N |
1076 | } else { |
1077 | mddev->reshape_position = MaxSector; | |
1078 | mddev->delta_disks = 0; | |
1079 | mddev->new_level = mddev->level; | |
1080 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1081 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1082 | } |
1083 | ||
1da177e4 LT |
1084 | if (sb->state & (1<<MD_SB_CLEAN)) |
1085 | mddev->recovery_cp = MaxSector; | |
1086 | else { | |
f72ffdd6 | 1087 | if (sb->events_hi == sb->cp_events_hi && |
1da177e4 LT |
1088 | sb->events_lo == sb->cp_events_lo) { |
1089 | mddev->recovery_cp = sb->recovery_cp; | |
1090 | } else | |
1091 | mddev->recovery_cp = 0; | |
1092 | } | |
1093 | ||
1094 | memcpy(mddev->uuid+0, &sb->set_uuid0, 4); | |
1095 | memcpy(mddev->uuid+4, &sb->set_uuid1, 4); | |
1096 | memcpy(mddev->uuid+8, &sb->set_uuid2, 4); | |
1097 | memcpy(mddev->uuid+12,&sb->set_uuid3, 4); | |
1098 | ||
1099 | mddev->max_disks = MD_SB_DISKS; | |
a654b9d8 N |
1100 | |
1101 | if (sb->state & (1<<MD_SB_BITMAP_PRESENT) && | |
6409bb05 | 1102 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1103 | mddev->bitmap_info.offset = |
1104 | mddev->bitmap_info.default_offset; | |
6409bb05 | 1105 | mddev->bitmap_info.space = |
c9ad020f | 1106 | mddev->bitmap_info.default_space; |
6409bb05 | 1107 | } |
a654b9d8 | 1108 | |
41158c7e | 1109 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1110 | /* Insist on good event counter while assembling, except |
1111 | * for spares (which don't need an event count) */ | |
1da177e4 | 1112 | ++ev1; |
be6800a7 N |
1113 | if (sb->disks[rdev->desc_nr].state & ( |
1114 | (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))) | |
f72ffdd6 | 1115 | if (ev1 < mddev->events) |
be6800a7 | 1116 | return -EINVAL; |
41158c7e N |
1117 | } else if (mddev->bitmap) { |
1118 | /* if adding to array with a bitmap, then we can accept an | |
1119 | * older device ... but not too old. | |
1120 | */ | |
41158c7e N |
1121 | if (ev1 < mddev->bitmap->events_cleared) |
1122 | return 0; | |
8313b8e5 N |
1123 | if (ev1 < mddev->events) |
1124 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1125 | } else { |
1126 | if (ev1 < mddev->events) | |
1127 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1128 | return 0; | |
1129 | } | |
41158c7e | 1130 | |
1da177e4 | 1131 | if (mddev->level != LEVEL_MULTIPATH) { |
1da177e4 LT |
1132 | desc = sb->disks + rdev->desc_nr; |
1133 | ||
1134 | if (desc->state & (1<<MD_DISK_FAULTY)) | |
b2d444d7 | 1135 | set_bit(Faulty, &rdev->flags); |
7c7546cc N |
1136 | else if (desc->state & (1<<MD_DISK_SYNC) /* && |
1137 | desc->raid_disk < mddev->raid_disks */) { | |
b2d444d7 | 1138 | set_bit(In_sync, &rdev->flags); |
1da177e4 | 1139 | rdev->raid_disk = desc->raid_disk; |
f466722c | 1140 | rdev->saved_raid_disk = desc->raid_disk; |
0261cd9f N |
1141 | } else if (desc->state & (1<<MD_DISK_ACTIVE)) { |
1142 | /* active but not in sync implies recovery up to | |
1143 | * reshape position. We don't know exactly where | |
1144 | * that is, so set to zero for now */ | |
1145 | if (mddev->minor_version >= 91) { | |
1146 | rdev->recovery_offset = 0; | |
1147 | rdev->raid_disk = desc->raid_disk; | |
1148 | } | |
1da177e4 | 1149 | } |
8ddf9efe N |
1150 | if (desc->state & (1<<MD_DISK_WRITEMOSTLY)) |
1151 | set_bit(WriteMostly, &rdev->flags); | |
41158c7e | 1152 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1153 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1154 | return 0; |
1155 | } | |
1156 | ||
1157 | /* | |
1158 | * sync_super for 0.90.0 | |
1159 | */ | |
fd01b88c | 1160 | static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1161 | { |
1162 | mdp_super_t *sb; | |
3cb03002 | 1163 | struct md_rdev *rdev2; |
1da177e4 | 1164 | int next_spare = mddev->raid_disks; |
19133a42 | 1165 | |
1da177e4 LT |
1166 | /* make rdev->sb match mddev data.. |
1167 | * | |
1168 | * 1/ zero out disks | |
1169 | * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); | |
1170 | * 3/ any empty disks < next_spare become removed | |
1171 | * | |
1172 | * disks[0] gets initialised to REMOVED because | |
1173 | * we cannot be sure from other fields if it has | |
1174 | * been initialised or not. | |
1175 | */ | |
1176 | int i; | |
1177 | int active=0, working=0,failed=0,spare=0,nr_disks=0; | |
1178 | ||
61181565 N |
1179 | rdev->sb_size = MD_SB_BYTES; |
1180 | ||
65a06f06 | 1181 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1182 | |
1183 | memset(sb, 0, sizeof(*sb)); | |
1184 | ||
1185 | sb->md_magic = MD_SB_MAGIC; | |
1186 | sb->major_version = mddev->major_version; | |
1da177e4 LT |
1187 | sb->patch_version = mddev->patch_version; |
1188 | sb->gvalid_words = 0; /* ignored */ | |
1189 | memcpy(&sb->set_uuid0, mddev->uuid+0, 4); | |
1190 | memcpy(&sb->set_uuid1, mddev->uuid+4, 4); | |
1191 | memcpy(&sb->set_uuid2, mddev->uuid+8, 4); | |
1192 | memcpy(&sb->set_uuid3, mddev->uuid+12,4); | |
1193 | ||
1194 | sb->ctime = mddev->ctime; | |
1195 | sb->level = mddev->level; | |
58c0fed4 | 1196 | sb->size = mddev->dev_sectors / 2; |
1da177e4 LT |
1197 | sb->raid_disks = mddev->raid_disks; |
1198 | sb->md_minor = mddev->md_minor; | |
e691063a | 1199 | sb->not_persistent = 0; |
1da177e4 LT |
1200 | sb->utime = mddev->utime; |
1201 | sb->state = 0; | |
1202 | sb->events_hi = (mddev->events>>32); | |
1203 | sb->events_lo = (u32)mddev->events; | |
1204 | ||
f6705578 N |
1205 | if (mddev->reshape_position == MaxSector) |
1206 | sb->minor_version = 90; | |
1207 | else { | |
1208 | sb->minor_version = 91; | |
1209 | sb->reshape_position = mddev->reshape_position; | |
1210 | sb->new_level = mddev->new_level; | |
1211 | sb->delta_disks = mddev->delta_disks; | |
1212 | sb->new_layout = mddev->new_layout; | |
664e7c41 | 1213 | sb->new_chunk = mddev->new_chunk_sectors << 9; |
f6705578 N |
1214 | } |
1215 | mddev->minor_version = sb->minor_version; | |
1da177e4 LT |
1216 | if (mddev->in_sync) |
1217 | { | |
1218 | sb->recovery_cp = mddev->recovery_cp; | |
1219 | sb->cp_events_hi = (mddev->events>>32); | |
1220 | sb->cp_events_lo = (u32)mddev->events; | |
1221 | if (mddev->recovery_cp == MaxSector) | |
1222 | sb->state = (1<< MD_SB_CLEAN); | |
1223 | } else | |
1224 | sb->recovery_cp = 0; | |
1225 | ||
1226 | sb->layout = mddev->layout; | |
9d8f0363 | 1227 | sb->chunk_size = mddev->chunk_sectors << 9; |
1da177e4 | 1228 | |
c3d9714e | 1229 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) |
a654b9d8 N |
1230 | sb->state |= (1<<MD_SB_BITMAP_PRESENT); |
1231 | ||
1da177e4 | 1232 | sb->disks[0].state = (1<<MD_DISK_REMOVED); |
dafb20fa | 1233 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1234 | mdp_disk_t *d; |
86e6ffdd | 1235 | int desc_nr; |
0261cd9f N |
1236 | int is_active = test_bit(In_sync, &rdev2->flags); |
1237 | ||
1238 | if (rdev2->raid_disk >= 0 && | |
1239 | sb->minor_version >= 91) | |
1240 | /* we have nowhere to store the recovery_offset, | |
1241 | * but if it is not below the reshape_position, | |
1242 | * we can piggy-back on that. | |
1243 | */ | |
1244 | is_active = 1; | |
1245 | if (rdev2->raid_disk < 0 || | |
1246 | test_bit(Faulty, &rdev2->flags)) | |
1247 | is_active = 0; | |
1248 | if (is_active) | |
86e6ffdd | 1249 | desc_nr = rdev2->raid_disk; |
1da177e4 | 1250 | else |
86e6ffdd | 1251 | desc_nr = next_spare++; |
19133a42 | 1252 | rdev2->desc_nr = desc_nr; |
1da177e4 LT |
1253 | d = &sb->disks[rdev2->desc_nr]; |
1254 | nr_disks++; | |
1255 | d->number = rdev2->desc_nr; | |
1256 | d->major = MAJOR(rdev2->bdev->bd_dev); | |
1257 | d->minor = MINOR(rdev2->bdev->bd_dev); | |
0261cd9f | 1258 | if (is_active) |
1da177e4 LT |
1259 | d->raid_disk = rdev2->raid_disk; |
1260 | else | |
1261 | d->raid_disk = rdev2->desc_nr; /* compatibility */ | |
1be7892f | 1262 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1263 | d->state = (1<<MD_DISK_FAULTY); |
0261cd9f | 1264 | else if (is_active) { |
1da177e4 | 1265 | d->state = (1<<MD_DISK_ACTIVE); |
0261cd9f N |
1266 | if (test_bit(In_sync, &rdev2->flags)) |
1267 | d->state |= (1<<MD_DISK_SYNC); | |
1da177e4 LT |
1268 | active++; |
1269 | working++; | |
1270 | } else { | |
1271 | d->state = 0; | |
1272 | spare++; | |
1273 | working++; | |
1274 | } | |
8ddf9efe N |
1275 | if (test_bit(WriteMostly, &rdev2->flags)) |
1276 | d->state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 | 1277 | } |
1da177e4 LT |
1278 | /* now set the "removed" and "faulty" bits on any missing devices */ |
1279 | for (i=0 ; i < mddev->raid_disks ; i++) { | |
1280 | mdp_disk_t *d = &sb->disks[i]; | |
1281 | if (d->state == 0 && d->number == 0) { | |
1282 | d->number = i; | |
1283 | d->raid_disk = i; | |
1284 | d->state = (1<<MD_DISK_REMOVED); | |
1285 | d->state |= (1<<MD_DISK_FAULTY); | |
1286 | failed++; | |
1287 | } | |
1288 | } | |
1289 | sb->nr_disks = nr_disks; | |
1290 | sb->active_disks = active; | |
1291 | sb->working_disks = working; | |
1292 | sb->failed_disks = failed; | |
1293 | sb->spare_disks = spare; | |
1294 | ||
1295 | sb->this_disk = sb->disks[rdev->desc_nr]; | |
1296 | sb->sb_csum = calc_sb_csum(sb); | |
1297 | } | |
1298 | ||
0cd17fec CW |
1299 | /* |
1300 | * rdev_size_change for 0.90.0 | |
1301 | */ | |
1302 | static unsigned long long | |
3cb03002 | 1303 | super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec | 1304 | { |
58c0fed4 | 1305 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1306 | return 0; /* component must fit device */ |
c3d9714e | 1307 | if (rdev->mddev->bitmap_info.offset) |
0cd17fec | 1308 | return 0; /* can't move bitmap */ |
57b2caa3 | 1309 | rdev->sb_start = calc_dev_sboffset(rdev); |
15f4a5fd AN |
1310 | if (!num_sectors || num_sectors > rdev->sb_start) |
1311 | num_sectors = rdev->sb_start; | |
27a7b260 N |
1312 | /* Limit to 4TB as metadata cannot record more than that. |
1313 | * 4TB == 2^32 KB, or 2*2^32 sectors. | |
1314 | */ | |
667a5313 | 1315 | if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1) |
27a7b260 | 1316 | num_sectors = (2ULL << 32) - 2; |
0f420358 | 1317 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1318 | rdev->sb_page); |
1319 | md_super_wait(rdev->mddev); | |
c26a44ed | 1320 | return num_sectors; |
0cd17fec CW |
1321 | } |
1322 | ||
c6563a8c N |
1323 | static int |
1324 | super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset) | |
1325 | { | |
1326 | /* non-zero offset changes not possible with v0.90 */ | |
1327 | return new_offset == 0; | |
1328 | } | |
0cd17fec | 1329 | |
1da177e4 LT |
1330 | /* |
1331 | * version 1 superblock | |
1332 | */ | |
1333 | ||
f72ffdd6 | 1334 | static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb) |
1da177e4 | 1335 | { |
1c05b4bc N |
1336 | __le32 disk_csum; |
1337 | u32 csum; | |
1da177e4 LT |
1338 | unsigned long long newcsum; |
1339 | int size = 256 + le32_to_cpu(sb->max_dev)*2; | |
1c05b4bc | 1340 | __le32 *isuper = (__le32*)sb; |
1da177e4 LT |
1341 | |
1342 | disk_csum = sb->sb_csum; | |
1343 | sb->sb_csum = 0; | |
1344 | newcsum = 0; | |
1f3c9907 | 1345 | for (; size >= 4; size -= 4) |
1da177e4 LT |
1346 | newcsum += le32_to_cpu(*isuper++); |
1347 | ||
1348 | if (size == 2) | |
1c05b4bc | 1349 | newcsum += le16_to_cpu(*(__le16*) isuper); |
1da177e4 LT |
1350 | |
1351 | csum = (newcsum & 0xffffffff) + (newcsum >> 32); | |
1352 | sb->sb_csum = disk_csum; | |
1353 | return cpu_to_le32(csum); | |
1354 | } | |
1355 | ||
2699b672 N |
1356 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, |
1357 | int acknowledged); | |
3cb03002 | 1358 | static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version) |
1da177e4 LT |
1359 | { |
1360 | struct mdp_superblock_1 *sb; | |
1361 | int ret; | |
0f420358 | 1362 | sector_t sb_start; |
c6563a8c | 1363 | sector_t sectors; |
1da177e4 | 1364 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; |
0002b271 | 1365 | int bmask; |
1da177e4 LT |
1366 | |
1367 | /* | |
0f420358 | 1368 | * Calculate the position of the superblock in 512byte sectors. |
1da177e4 LT |
1369 | * It is always aligned to a 4K boundary and |
1370 | * depeding on minor_version, it can be: | |
1371 | * 0: At least 8K, but less than 12K, from end of device | |
1372 | * 1: At start of device | |
1373 | * 2: 4K from start of device. | |
1374 | */ | |
1375 | switch(minor_version) { | |
1376 | case 0: | |
77304d2a | 1377 | sb_start = i_size_read(rdev->bdev->bd_inode) >> 9; |
0f420358 AN |
1378 | sb_start -= 8*2; |
1379 | sb_start &= ~(sector_t)(4*2-1); | |
1da177e4 LT |
1380 | break; |
1381 | case 1: | |
0f420358 | 1382 | sb_start = 0; |
1da177e4 LT |
1383 | break; |
1384 | case 2: | |
0f420358 | 1385 | sb_start = 8; |
1da177e4 LT |
1386 | break; |
1387 | default: | |
1388 | return -EINVAL; | |
1389 | } | |
0f420358 | 1390 | rdev->sb_start = sb_start; |
1da177e4 | 1391 | |
0002b271 N |
1392 | /* superblock is rarely larger than 1K, but it can be larger, |
1393 | * and it is safe to read 4k, so we do that | |
1394 | */ | |
1395 | ret = read_disk_sb(rdev, 4096); | |
1da177e4 LT |
1396 | if (ret) return ret; |
1397 | ||
65a06f06 | 1398 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1399 | |
1400 | if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || | |
1401 | sb->major_version != cpu_to_le32(1) || | |
1402 | le32_to_cpu(sb->max_dev) > (4096-256)/2 || | |
0f420358 | 1403 | le64_to_cpu(sb->super_offset) != rdev->sb_start || |
71c0805c | 1404 | (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0) |
1da177e4 LT |
1405 | return -EINVAL; |
1406 | ||
1407 | if (calc_sb_1_csum(sb) != sb->sb_csum) { | |
1408 | printk("md: invalid superblock checksum on %s\n", | |
1409 | bdevname(rdev->bdev,b)); | |
1410 | return -EINVAL; | |
1411 | } | |
1412 | if (le64_to_cpu(sb->data_size) < 10) { | |
1413 | printk("md: data_size too small on %s\n", | |
1414 | bdevname(rdev->bdev,b)); | |
1415 | return -EINVAL; | |
1416 | } | |
c6563a8c N |
1417 | if (sb->pad0 || |
1418 | sb->pad3[0] || | |
1419 | memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1]))) | |
1420 | /* Some padding is non-zero, might be a new feature */ | |
1421 | return -EINVAL; | |
e11e93fa | 1422 | |
1da177e4 LT |
1423 | rdev->preferred_minor = 0xffff; |
1424 | rdev->data_offset = le64_to_cpu(sb->data_offset); | |
c6563a8c N |
1425 | rdev->new_data_offset = rdev->data_offset; |
1426 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) && | |
1427 | (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) | |
1428 | rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset); | |
4dbcdc75 | 1429 | atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read)); |
1da177e4 | 1430 | |
0002b271 | 1431 | rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256; |
e1defc4f | 1432 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; |
0002b271 | 1433 | if (rdev->sb_size & bmask) |
a1801f85 N |
1434 | rdev->sb_size = (rdev->sb_size | bmask) + 1; |
1435 | ||
1436 | if (minor_version | |
0f420358 | 1437 | && rdev->data_offset < sb_start + (rdev->sb_size/512)) |
a1801f85 | 1438 | return -EINVAL; |
c6563a8c N |
1439 | if (minor_version |
1440 | && rdev->new_data_offset < sb_start + (rdev->sb_size/512)) | |
1441 | return -EINVAL; | |
0002b271 | 1442 | |
31b65a0d N |
1443 | if (sb->level == cpu_to_le32(LEVEL_MULTIPATH)) |
1444 | rdev->desc_nr = -1; | |
1445 | else | |
1446 | rdev->desc_nr = le32_to_cpu(sb->dev_number); | |
1447 | ||
2699b672 N |
1448 | if (!rdev->bb_page) { |
1449 | rdev->bb_page = alloc_page(GFP_KERNEL); | |
1450 | if (!rdev->bb_page) | |
1451 | return -ENOMEM; | |
1452 | } | |
1453 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) && | |
1454 | rdev->badblocks.count == 0) { | |
1455 | /* need to load the bad block list. | |
1456 | * Currently we limit it to one page. | |
1457 | */ | |
1458 | s32 offset; | |
1459 | sector_t bb_sector; | |
1460 | u64 *bbp; | |
1461 | int i; | |
1462 | int sectors = le16_to_cpu(sb->bblog_size); | |
1463 | if (sectors > (PAGE_SIZE / 512)) | |
1464 | return -EINVAL; | |
1465 | offset = le32_to_cpu(sb->bblog_offset); | |
1466 | if (offset == 0) | |
1467 | return -EINVAL; | |
1468 | bb_sector = (long long)offset; | |
1469 | if (!sync_page_io(rdev, bb_sector, sectors << 9, | |
1470 | rdev->bb_page, READ, true)) | |
1471 | return -EIO; | |
1472 | bbp = (u64 *)page_address(rdev->bb_page); | |
1473 | rdev->badblocks.shift = sb->bblog_shift; | |
1474 | for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) { | |
1475 | u64 bb = le64_to_cpu(*bbp); | |
1476 | int count = bb & (0x3ff); | |
1477 | u64 sector = bb >> 10; | |
1478 | sector <<= sb->bblog_shift; | |
1479 | count <<= sb->bblog_shift; | |
1480 | if (bb + 1 == 0) | |
1481 | break; | |
1482 | if (md_set_badblocks(&rdev->badblocks, | |
1483 | sector, count, 1) == 0) | |
1484 | return -EINVAL; | |
1485 | } | |
486adf72 N |
1486 | } else if (sb->bblog_offset != 0) |
1487 | rdev->badblocks.shift = 0; | |
2699b672 | 1488 | |
9a7b2b0f | 1489 | if (!refdev) { |
8ed75463 | 1490 | ret = 1; |
9a7b2b0f | 1491 | } else { |
1da177e4 | 1492 | __u64 ev1, ev2; |
65a06f06 | 1493 | struct mdp_superblock_1 *refsb = page_address(refdev->sb_page); |
1da177e4 LT |
1494 | |
1495 | if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || | |
1496 | sb->level != refsb->level || | |
1497 | sb->layout != refsb->layout || | |
1498 | sb->chunksize != refsb->chunksize) { | |
1499 | printk(KERN_WARNING "md: %s has strangely different" | |
1500 | " superblock to %s\n", | |
1501 | bdevname(rdev->bdev,b), | |
1502 | bdevname(refdev->bdev,b2)); | |
1503 | return -EINVAL; | |
1504 | } | |
1505 | ev1 = le64_to_cpu(sb->events); | |
1506 | ev2 = le64_to_cpu(refsb->events); | |
1507 | ||
1508 | if (ev1 > ev2) | |
8ed75463 N |
1509 | ret = 1; |
1510 | else | |
1511 | ret = 0; | |
1da177e4 | 1512 | } |
c6563a8c N |
1513 | if (minor_version) { |
1514 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9); | |
1515 | sectors -= rdev->data_offset; | |
1516 | } else | |
1517 | sectors = rdev->sb_start; | |
1518 | if (sectors < le64_to_cpu(sb->data_size)) | |
1da177e4 | 1519 | return -EINVAL; |
dd8ac336 | 1520 | rdev->sectors = le64_to_cpu(sb->data_size); |
8ed75463 | 1521 | return ret; |
1da177e4 LT |
1522 | } |
1523 | ||
fd01b88c | 1524 | static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1525 | { |
65a06f06 | 1526 | struct mdp_superblock_1 *sb = page_address(rdev->sb_page); |
07d84d10 | 1527 | __u64 ev1 = le64_to_cpu(sb->events); |
1da177e4 | 1528 | |
41158c7e | 1529 | rdev->raid_disk = -1; |
c5d79adb N |
1530 | clear_bit(Faulty, &rdev->flags); |
1531 | clear_bit(In_sync, &rdev->flags); | |
8313b8e5 | 1532 | clear_bit(Bitmap_sync, &rdev->flags); |
c5d79adb | 1533 | clear_bit(WriteMostly, &rdev->flags); |
c5d79adb | 1534 | |
1da177e4 LT |
1535 | if (mddev->raid_disks == 0) { |
1536 | mddev->major_version = 1; | |
1537 | mddev->patch_version = 0; | |
e691063a | 1538 | mddev->external = 0; |
9d8f0363 | 1539 | mddev->chunk_sectors = le32_to_cpu(sb->chunksize); |
1da177e4 LT |
1540 | mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); |
1541 | mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); | |
1542 | mddev->level = le32_to_cpu(sb->level); | |
d9d166c2 | 1543 | mddev->clevel[0] = 0; |
1da177e4 LT |
1544 | mddev->layout = le32_to_cpu(sb->layout); |
1545 | mddev->raid_disks = le32_to_cpu(sb->raid_disks); | |
58c0fed4 | 1546 | mddev->dev_sectors = le64_to_cpu(sb->size); |
07d84d10 | 1547 | mddev->events = ev1; |
c3d9714e | 1548 | mddev->bitmap_info.offset = 0; |
6409bb05 N |
1549 | mddev->bitmap_info.space = 0; |
1550 | /* Default location for bitmap is 1K after superblock | |
1551 | * using 3K - total of 4K | |
1552 | */ | |
c3d9714e | 1553 | mddev->bitmap_info.default_offset = 1024 >> 9; |
6409bb05 | 1554 | mddev->bitmap_info.default_space = (4096-1024) >> 9; |
2c810cdd N |
1555 | mddev->reshape_backwards = 0; |
1556 | ||
1da177e4 LT |
1557 | mddev->recovery_cp = le64_to_cpu(sb->resync_offset); |
1558 | memcpy(mddev->uuid, sb->set_uuid, 16); | |
1559 | ||
1560 | mddev->max_disks = (4096-256)/2; | |
a654b9d8 | 1561 | |
71c0805c | 1562 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) && |
6409bb05 | 1563 | mddev->bitmap_info.file == NULL) { |
c3d9714e N |
1564 | mddev->bitmap_info.offset = |
1565 | (__s32)le32_to_cpu(sb->bitmap_offset); | |
6409bb05 N |
1566 | /* Metadata doesn't record how much space is available. |
1567 | * For 1.0, we assume we can use up to the superblock | |
1568 | * if before, else to 4K beyond superblock. | |
1569 | * For others, assume no change is possible. | |
1570 | */ | |
1571 | if (mddev->minor_version > 0) | |
1572 | mddev->bitmap_info.space = 0; | |
1573 | else if (mddev->bitmap_info.offset > 0) | |
1574 | mddev->bitmap_info.space = | |
1575 | 8 - mddev->bitmap_info.offset; | |
1576 | else | |
1577 | mddev->bitmap_info.space = | |
1578 | -mddev->bitmap_info.offset; | |
1579 | } | |
e11e93fa | 1580 | |
f6705578 N |
1581 | if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) { |
1582 | mddev->reshape_position = le64_to_cpu(sb->reshape_position); | |
1583 | mddev->delta_disks = le32_to_cpu(sb->delta_disks); | |
1584 | mddev->new_level = le32_to_cpu(sb->new_level); | |
1585 | mddev->new_layout = le32_to_cpu(sb->new_layout); | |
664e7c41 | 1586 | mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk); |
2c810cdd N |
1587 | if (mddev->delta_disks < 0 || |
1588 | (mddev->delta_disks == 0 && | |
1589 | (le32_to_cpu(sb->feature_map) | |
1590 | & MD_FEATURE_RESHAPE_BACKWARDS))) | |
1591 | mddev->reshape_backwards = 1; | |
f6705578 N |
1592 | } else { |
1593 | mddev->reshape_position = MaxSector; | |
1594 | mddev->delta_disks = 0; | |
1595 | mddev->new_level = mddev->level; | |
1596 | mddev->new_layout = mddev->layout; | |
664e7c41 | 1597 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
1598 | } |
1599 | ||
41158c7e | 1600 | } else if (mddev->pers == NULL) { |
be6800a7 N |
1601 | /* Insist of good event counter while assembling, except for |
1602 | * spares (which don't need an event count) */ | |
1da177e4 | 1603 | ++ev1; |
be6800a7 N |
1604 | if (rdev->desc_nr >= 0 && |
1605 | rdev->desc_nr < le32_to_cpu(sb->max_dev) && | |
1606 | le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe) | |
1607 | if (ev1 < mddev->events) | |
1608 | return -EINVAL; | |
41158c7e N |
1609 | } else if (mddev->bitmap) { |
1610 | /* If adding to array with a bitmap, then we can accept an | |
1611 | * older device, but not too old. | |
1612 | */ | |
41158c7e N |
1613 | if (ev1 < mddev->bitmap->events_cleared) |
1614 | return 0; | |
8313b8e5 N |
1615 | if (ev1 < mddev->events) |
1616 | set_bit(Bitmap_sync, &rdev->flags); | |
07d84d10 N |
1617 | } else { |
1618 | if (ev1 < mddev->events) | |
1619 | /* just a hot-add of a new device, leave raid_disk at -1 */ | |
1620 | return 0; | |
1621 | } | |
1da177e4 LT |
1622 | if (mddev->level != LEVEL_MULTIPATH) { |
1623 | int role; | |
3673f305 N |
1624 | if (rdev->desc_nr < 0 || |
1625 | rdev->desc_nr >= le32_to_cpu(sb->max_dev)) { | |
1626 | role = 0xffff; | |
1627 | rdev->desc_nr = -1; | |
1628 | } else | |
1629 | role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); | |
1da177e4 LT |
1630 | switch(role) { |
1631 | case 0xffff: /* spare */ | |
1da177e4 LT |
1632 | break; |
1633 | case 0xfffe: /* faulty */ | |
b2d444d7 | 1634 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
1635 | break; |
1636 | default: | |
f466722c | 1637 | rdev->saved_raid_disk = role; |
5fd6c1dc | 1638 | if ((le32_to_cpu(sb->feature_map) & |
f466722c | 1639 | MD_FEATURE_RECOVERY_OFFSET)) { |
5fd6c1dc | 1640 | rdev->recovery_offset = le64_to_cpu(sb->recovery_offset); |
f466722c N |
1641 | if (!(le32_to_cpu(sb->feature_map) & |
1642 | MD_FEATURE_RECOVERY_BITMAP)) | |
1643 | rdev->saved_raid_disk = -1; | |
1644 | } else | |
5fd6c1dc | 1645 | set_bit(In_sync, &rdev->flags); |
1da177e4 LT |
1646 | rdev->raid_disk = role; |
1647 | break; | |
1648 | } | |
8ddf9efe N |
1649 | if (sb->devflags & WriteMostly1) |
1650 | set_bit(WriteMostly, &rdev->flags); | |
2d78f8c4 N |
1651 | if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT) |
1652 | set_bit(Replacement, &rdev->flags); | |
41158c7e | 1653 | } else /* MULTIPATH are always insync */ |
b2d444d7 | 1654 | set_bit(In_sync, &rdev->flags); |
41158c7e | 1655 | |
1da177e4 LT |
1656 | return 0; |
1657 | } | |
1658 | ||
fd01b88c | 1659 | static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1660 | { |
1661 | struct mdp_superblock_1 *sb; | |
3cb03002 | 1662 | struct md_rdev *rdev2; |
1da177e4 LT |
1663 | int max_dev, i; |
1664 | /* make rdev->sb match mddev and rdev data. */ | |
1665 | ||
65a06f06 | 1666 | sb = page_address(rdev->sb_page); |
1da177e4 LT |
1667 | |
1668 | sb->feature_map = 0; | |
1669 | sb->pad0 = 0; | |
5fd6c1dc | 1670 | sb->recovery_offset = cpu_to_le64(0); |
1da177e4 LT |
1671 | memset(sb->pad3, 0, sizeof(sb->pad3)); |
1672 | ||
1673 | sb->utime = cpu_to_le64((__u64)mddev->utime); | |
1674 | sb->events = cpu_to_le64(mddev->events); | |
1675 | if (mddev->in_sync) | |
1676 | sb->resync_offset = cpu_to_le64(mddev->recovery_cp); | |
1677 | else | |
1678 | sb->resync_offset = cpu_to_le64(0); | |
1679 | ||
1c05b4bc | 1680 | sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors)); |
4dbcdc75 | 1681 | |
f0ca340c | 1682 | sb->raid_disks = cpu_to_le32(mddev->raid_disks); |
58c0fed4 | 1683 | sb->size = cpu_to_le64(mddev->dev_sectors); |
9d8f0363 | 1684 | sb->chunksize = cpu_to_le32(mddev->chunk_sectors); |
62e1e389 N |
1685 | sb->level = cpu_to_le32(mddev->level); |
1686 | sb->layout = cpu_to_le32(mddev->layout); | |
f0ca340c | 1687 | |
aeb9b211 N |
1688 | if (test_bit(WriteMostly, &rdev->flags)) |
1689 | sb->devflags |= WriteMostly1; | |
1690 | else | |
1691 | sb->devflags &= ~WriteMostly1; | |
c6563a8c N |
1692 | sb->data_offset = cpu_to_le64(rdev->data_offset); |
1693 | sb->data_size = cpu_to_le64(rdev->sectors); | |
aeb9b211 | 1694 | |
c3d9714e N |
1695 | if (mddev->bitmap && mddev->bitmap_info.file == NULL) { |
1696 | sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); | |
71c0805c | 1697 | sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); |
a654b9d8 | 1698 | } |
5fd6c1dc N |
1699 | |
1700 | if (rdev->raid_disk >= 0 && | |
97e4f42d | 1701 | !test_bit(In_sync, &rdev->flags)) { |
93be75ff N |
1702 | sb->feature_map |= |
1703 | cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET); | |
1704 | sb->recovery_offset = | |
1705 | cpu_to_le64(rdev->recovery_offset); | |
f466722c N |
1706 | if (rdev->saved_raid_disk >= 0 && mddev->bitmap) |
1707 | sb->feature_map |= | |
1708 | cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP); | |
5fd6c1dc | 1709 | } |
2d78f8c4 N |
1710 | if (test_bit(Replacement, &rdev->flags)) |
1711 | sb->feature_map |= | |
1712 | cpu_to_le32(MD_FEATURE_REPLACEMENT); | |
5fd6c1dc | 1713 | |
f6705578 N |
1714 | if (mddev->reshape_position != MaxSector) { |
1715 | sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE); | |
1716 | sb->reshape_position = cpu_to_le64(mddev->reshape_position); | |
1717 | sb->new_layout = cpu_to_le32(mddev->new_layout); | |
1718 | sb->delta_disks = cpu_to_le32(mddev->delta_disks); | |
1719 | sb->new_level = cpu_to_le32(mddev->new_level); | |
664e7c41 | 1720 | sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors); |
2c810cdd N |
1721 | if (mddev->delta_disks == 0 && |
1722 | mddev->reshape_backwards) | |
1723 | sb->feature_map | |
1724 | |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS); | |
c6563a8c N |
1725 | if (rdev->new_data_offset != rdev->data_offset) { |
1726 | sb->feature_map | |
1727 | |= cpu_to_le32(MD_FEATURE_NEW_OFFSET); | |
1728 | sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset | |
1729 | - rdev->data_offset)); | |
1730 | } | |
f6705578 | 1731 | } |
a654b9d8 | 1732 | |
2699b672 N |
1733 | if (rdev->badblocks.count == 0) |
1734 | /* Nothing to do for bad blocks*/ ; | |
1735 | else if (sb->bblog_offset == 0) | |
1736 | /* Cannot record bad blocks on this device */ | |
1737 | md_error(mddev, rdev); | |
1738 | else { | |
1739 | struct badblocks *bb = &rdev->badblocks; | |
1740 | u64 *bbp = (u64 *)page_address(rdev->bb_page); | |
1741 | u64 *p = bb->page; | |
1742 | sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS); | |
1743 | if (bb->changed) { | |
1744 | unsigned seq; | |
1745 | ||
1746 | retry: | |
1747 | seq = read_seqbegin(&bb->lock); | |
1748 | ||
1749 | memset(bbp, 0xff, PAGE_SIZE); | |
1750 | ||
1751 | for (i = 0 ; i < bb->count ; i++) { | |
35f9ac2d | 1752 | u64 internal_bb = p[i]; |
2699b672 N |
1753 | u64 store_bb = ((BB_OFFSET(internal_bb) << 10) |
1754 | | BB_LEN(internal_bb)); | |
35f9ac2d | 1755 | bbp[i] = cpu_to_le64(store_bb); |
2699b672 | 1756 | } |
d0962936 | 1757 | bb->changed = 0; |
2699b672 N |
1758 | if (read_seqretry(&bb->lock, seq)) |
1759 | goto retry; | |
1760 | ||
1761 | bb->sector = (rdev->sb_start + | |
1762 | (int)le32_to_cpu(sb->bblog_offset)); | |
1763 | bb->size = le16_to_cpu(sb->bblog_size); | |
2699b672 N |
1764 | } |
1765 | } | |
1766 | ||
1da177e4 | 1767 | max_dev = 0; |
dafb20fa | 1768 | rdev_for_each(rdev2, mddev) |
1da177e4 LT |
1769 | if (rdev2->desc_nr+1 > max_dev) |
1770 | max_dev = rdev2->desc_nr+1; | |
a778b73f | 1771 | |
70471daf N |
1772 | if (max_dev > le32_to_cpu(sb->max_dev)) { |
1773 | int bmask; | |
a778b73f | 1774 | sb->max_dev = cpu_to_le32(max_dev); |
70471daf N |
1775 | rdev->sb_size = max_dev * 2 + 256; |
1776 | bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1; | |
1777 | if (rdev->sb_size & bmask) | |
1778 | rdev->sb_size = (rdev->sb_size | bmask) + 1; | |
ddcf3522 N |
1779 | } else |
1780 | max_dev = le32_to_cpu(sb->max_dev); | |
1781 | ||
1da177e4 LT |
1782 | for (i=0; i<max_dev;i++) |
1783 | sb->dev_roles[i] = cpu_to_le16(0xfffe); | |
f72ffdd6 | 1784 | |
dafb20fa | 1785 | rdev_for_each(rdev2, mddev) { |
1da177e4 | 1786 | i = rdev2->desc_nr; |
b2d444d7 | 1787 | if (test_bit(Faulty, &rdev2->flags)) |
1da177e4 | 1788 | sb->dev_roles[i] = cpu_to_le16(0xfffe); |
b2d444d7 | 1789 | else if (test_bit(In_sync, &rdev2->flags)) |
1da177e4 | 1790 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
93be75ff | 1791 | else if (rdev2->raid_disk >= 0) |
5fd6c1dc | 1792 | sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); |
1da177e4 LT |
1793 | else |
1794 | sb->dev_roles[i] = cpu_to_le16(0xffff); | |
1795 | } | |
1796 | ||
1da177e4 LT |
1797 | sb->sb_csum = calc_sb_1_csum(sb); |
1798 | } | |
1799 | ||
0cd17fec | 1800 | static unsigned long long |
3cb03002 | 1801 | super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) |
0cd17fec CW |
1802 | { |
1803 | struct mdp_superblock_1 *sb; | |
15f4a5fd | 1804 | sector_t max_sectors; |
58c0fed4 | 1805 | if (num_sectors && num_sectors < rdev->mddev->dev_sectors) |
0cd17fec | 1806 | return 0; /* component must fit device */ |
c6563a8c N |
1807 | if (rdev->data_offset != rdev->new_data_offset) |
1808 | return 0; /* too confusing */ | |
0f420358 | 1809 | if (rdev->sb_start < rdev->data_offset) { |
0cd17fec | 1810 | /* minor versions 1 and 2; superblock before data */ |
77304d2a | 1811 | max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9; |
15f4a5fd AN |
1812 | max_sectors -= rdev->data_offset; |
1813 | if (!num_sectors || num_sectors > max_sectors) | |
1814 | num_sectors = max_sectors; | |
c3d9714e | 1815 | } else if (rdev->mddev->bitmap_info.offset) { |
0cd17fec CW |
1816 | /* minor version 0 with bitmap we can't move */ |
1817 | return 0; | |
1818 | } else { | |
1819 | /* minor version 0; superblock after data */ | |
0f420358 | 1820 | sector_t sb_start; |
77304d2a | 1821 | sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2; |
0f420358 | 1822 | sb_start &= ~(sector_t)(4*2 - 1); |
dd8ac336 | 1823 | max_sectors = rdev->sectors + sb_start - rdev->sb_start; |
15f4a5fd AN |
1824 | if (!num_sectors || num_sectors > max_sectors) |
1825 | num_sectors = max_sectors; | |
0f420358 | 1826 | rdev->sb_start = sb_start; |
0cd17fec | 1827 | } |
65a06f06 | 1828 | sb = page_address(rdev->sb_page); |
15f4a5fd | 1829 | sb->data_size = cpu_to_le64(num_sectors); |
0f420358 | 1830 | sb->super_offset = rdev->sb_start; |
0cd17fec | 1831 | sb->sb_csum = calc_sb_1_csum(sb); |
0f420358 | 1832 | md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, |
0cd17fec CW |
1833 | rdev->sb_page); |
1834 | md_super_wait(rdev->mddev); | |
c26a44ed | 1835 | return num_sectors; |
c6563a8c N |
1836 | |
1837 | } | |
1838 | ||
1839 | static int | |
1840 | super_1_allow_new_offset(struct md_rdev *rdev, | |
1841 | unsigned long long new_offset) | |
1842 | { | |
1843 | /* All necessary checks on new >= old have been done */ | |
1844 | struct bitmap *bitmap; | |
1845 | if (new_offset >= rdev->data_offset) | |
1846 | return 1; | |
1847 | ||
1848 | /* with 1.0 metadata, there is no metadata to tread on | |
1849 | * so we can always move back */ | |
1850 | if (rdev->mddev->minor_version == 0) | |
1851 | return 1; | |
1852 | ||
1853 | /* otherwise we must be sure not to step on | |
1854 | * any metadata, so stay: | |
1855 | * 36K beyond start of superblock | |
1856 | * beyond end of badblocks | |
1857 | * beyond write-intent bitmap | |
1858 | */ | |
1859 | if (rdev->sb_start + (32+4)*2 > new_offset) | |
1860 | return 0; | |
1861 | bitmap = rdev->mddev->bitmap; | |
1862 | if (bitmap && !rdev->mddev->bitmap_info.file && | |
1863 | rdev->sb_start + rdev->mddev->bitmap_info.offset + | |
1ec885cd | 1864 | bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset) |
c6563a8c N |
1865 | return 0; |
1866 | if (rdev->badblocks.sector + rdev->badblocks.size > new_offset) | |
1867 | return 0; | |
1868 | ||
1869 | return 1; | |
0cd17fec | 1870 | } |
1da177e4 | 1871 | |
75c96f85 | 1872 | static struct super_type super_types[] = { |
1da177e4 LT |
1873 | [0] = { |
1874 | .name = "0.90.0", | |
1875 | .owner = THIS_MODULE, | |
0cd17fec CW |
1876 | .load_super = super_90_load, |
1877 | .validate_super = super_90_validate, | |
1878 | .sync_super = super_90_sync, | |
1879 | .rdev_size_change = super_90_rdev_size_change, | |
c6563a8c | 1880 | .allow_new_offset = super_90_allow_new_offset, |
1da177e4 LT |
1881 | }, |
1882 | [1] = { | |
1883 | .name = "md-1", | |
1884 | .owner = THIS_MODULE, | |
0cd17fec CW |
1885 | .load_super = super_1_load, |
1886 | .validate_super = super_1_validate, | |
1887 | .sync_super = super_1_sync, | |
1888 | .rdev_size_change = super_1_rdev_size_change, | |
c6563a8c | 1889 | .allow_new_offset = super_1_allow_new_offset, |
1da177e4 LT |
1890 | }, |
1891 | }; | |
1da177e4 | 1892 | |
fd01b88c | 1893 | static void sync_super(struct mddev *mddev, struct md_rdev *rdev) |
076f968b JB |
1894 | { |
1895 | if (mddev->sync_super) { | |
1896 | mddev->sync_super(mddev, rdev); | |
1897 | return; | |
1898 | } | |
1899 | ||
1900 | BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types)); | |
1901 | ||
1902 | super_types[mddev->major_version].sync_super(mddev, rdev); | |
1903 | } | |
1904 | ||
fd01b88c | 1905 | static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2) |
1da177e4 | 1906 | { |
3cb03002 | 1907 | struct md_rdev *rdev, *rdev2; |
1da177e4 | 1908 | |
4b80991c N |
1909 | rcu_read_lock(); |
1910 | rdev_for_each_rcu(rdev, mddev1) | |
1911 | rdev_for_each_rcu(rdev2, mddev2) | |
7dd5e7c3 | 1912 | if (rdev->bdev->bd_contains == |
4b80991c N |
1913 | rdev2->bdev->bd_contains) { |
1914 | rcu_read_unlock(); | |
7dd5e7c3 | 1915 | return 1; |
4b80991c N |
1916 | } |
1917 | rcu_read_unlock(); | |
1da177e4 LT |
1918 | return 0; |
1919 | } | |
1920 | ||
1921 | static LIST_HEAD(pending_raid_disks); | |
1922 | ||
ac5e7113 AN |
1923 | /* |
1924 | * Try to register data integrity profile for an mddev | |
1925 | * | |
1926 | * This is called when an array is started and after a disk has been kicked | |
1927 | * from the array. It only succeeds if all working and active component devices | |
1928 | * are integrity capable with matching profiles. | |
1929 | */ | |
fd01b88c | 1930 | int md_integrity_register(struct mddev *mddev) |
ac5e7113 | 1931 | { |
3cb03002 | 1932 | struct md_rdev *rdev, *reference = NULL; |
ac5e7113 AN |
1933 | |
1934 | if (list_empty(&mddev->disks)) | |
1935 | return 0; /* nothing to do */ | |
629acb6a JB |
1936 | if (!mddev->gendisk || blk_get_integrity(mddev->gendisk)) |
1937 | return 0; /* shouldn't register, or already is */ | |
dafb20fa | 1938 | rdev_for_each(rdev, mddev) { |
ac5e7113 AN |
1939 | /* skip spares and non-functional disks */ |
1940 | if (test_bit(Faulty, &rdev->flags)) | |
1941 | continue; | |
1942 | if (rdev->raid_disk < 0) | |
1943 | continue; | |
ac5e7113 AN |
1944 | if (!reference) { |
1945 | /* Use the first rdev as the reference */ | |
1946 | reference = rdev; | |
1947 | continue; | |
1948 | } | |
1949 | /* does this rdev's profile match the reference profile? */ | |
1950 | if (blk_integrity_compare(reference->bdev->bd_disk, | |
1951 | rdev->bdev->bd_disk) < 0) | |
1952 | return -EINVAL; | |
1953 | } | |
89078d57 MP |
1954 | if (!reference || !bdev_get_integrity(reference->bdev)) |
1955 | return 0; | |
ac5e7113 AN |
1956 | /* |
1957 | * All component devices are integrity capable and have matching | |
1958 | * profiles, register the common profile for the md device. | |
1959 | */ | |
1960 | if (blk_integrity_register(mddev->gendisk, | |
1961 | bdev_get_integrity(reference->bdev)) != 0) { | |
1962 | printk(KERN_ERR "md: failed to register integrity for %s\n", | |
1963 | mdname(mddev)); | |
1964 | return -EINVAL; | |
1965 | } | |
a91a2785 MP |
1966 | printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev)); |
1967 | if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) { | |
1968 | printk(KERN_ERR "md: failed to create integrity pool for %s\n", | |
1969 | mdname(mddev)); | |
1970 | return -EINVAL; | |
1971 | } | |
ac5e7113 AN |
1972 | return 0; |
1973 | } | |
1974 | EXPORT_SYMBOL(md_integrity_register); | |
1975 | ||
1976 | /* Disable data integrity if non-capable/non-matching disk is being added */ | |
fd01b88c | 1977 | void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev) |
3f9d99c1 | 1978 | { |
2863b9eb JB |
1979 | struct blk_integrity *bi_rdev; |
1980 | struct blk_integrity *bi_mddev; | |
1981 | ||
1982 | if (!mddev->gendisk) | |
1983 | return; | |
1984 | ||
1985 | bi_rdev = bdev_get_integrity(rdev->bdev); | |
1986 | bi_mddev = blk_get_integrity(mddev->gendisk); | |
3f9d99c1 | 1987 | |
ac5e7113 | 1988 | if (!bi_mddev) /* nothing to do */ |
3f9d99c1 | 1989 | return; |
ac5e7113 | 1990 | if (rdev->raid_disk < 0) /* skip spares */ |
3f9d99c1 | 1991 | return; |
ac5e7113 AN |
1992 | if (bi_rdev && blk_integrity_compare(mddev->gendisk, |
1993 | rdev->bdev->bd_disk) >= 0) | |
1994 | return; | |
1995 | printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev)); | |
1996 | blk_integrity_unregister(mddev->gendisk); | |
3f9d99c1 | 1997 | } |
ac5e7113 | 1998 | EXPORT_SYMBOL(md_integrity_add_rdev); |
3f9d99c1 | 1999 | |
f72ffdd6 | 2000 | static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev) |
1da177e4 | 2001 | { |
7dd5e7c3 | 2002 | char b[BDEVNAME_SIZE]; |
f637b9f9 | 2003 | struct kobject *ko; |
1edf80d3 | 2004 | char *s; |
5e55e2f5 | 2005 | int err; |
1da177e4 | 2006 | |
11e2ede0 DW |
2007 | /* prevent duplicates */ |
2008 | if (find_rdev(mddev, rdev->bdev->bd_dev)) | |
2009 | return -EEXIST; | |
2010 | ||
dd8ac336 AN |
2011 | /* make sure rdev->sectors exceeds mddev->dev_sectors */ |
2012 | if (rdev->sectors && (mddev->dev_sectors == 0 || | |
2013 | rdev->sectors < mddev->dev_sectors)) { | |
a778b73f N |
2014 | if (mddev->pers) { |
2015 | /* Cannot change size, so fail | |
2016 | * If mddev->level <= 0, then we don't care | |
2017 | * about aligning sizes (e.g. linear) | |
2018 | */ | |
2019 | if (mddev->level > 0) | |
2020 | return -ENOSPC; | |
2021 | } else | |
dd8ac336 | 2022 | mddev->dev_sectors = rdev->sectors; |
2bf071bf | 2023 | } |
1da177e4 LT |
2024 | |
2025 | /* Verify rdev->desc_nr is unique. | |
2026 | * If it is -1, assign a free number, else | |
2027 | * check number is not in use | |
2028 | */ | |
4878e9eb | 2029 | rcu_read_lock(); |
1da177e4 LT |
2030 | if (rdev->desc_nr < 0) { |
2031 | int choice = 0; | |
4878e9eb N |
2032 | if (mddev->pers) |
2033 | choice = mddev->raid_disks; | |
2034 | while (find_rdev_nr_rcu(mddev, choice)) | |
1da177e4 LT |
2035 | choice++; |
2036 | rdev->desc_nr = choice; | |
2037 | } else { | |
4878e9eb N |
2038 | if (find_rdev_nr_rcu(mddev, rdev->desc_nr)) { |
2039 | rcu_read_unlock(); | |
1da177e4 | 2040 | return -EBUSY; |
4878e9eb | 2041 | } |
1da177e4 | 2042 | } |
4878e9eb | 2043 | rcu_read_unlock(); |
de01dfad N |
2044 | if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) { |
2045 | printk(KERN_WARNING "md: %s: array is limited to %d devices\n", | |
2046 | mdname(mddev), mddev->max_disks); | |
2047 | return -EBUSY; | |
2048 | } | |
19133a42 | 2049 | bdevname(rdev->bdev,b); |
649316b2 | 2050 | while ( (s=strchr(b, '/')) != NULL) |
1edf80d3 | 2051 | *s = '!'; |
649316b2 | 2052 | |
1da177e4 | 2053 | rdev->mddev = mddev; |
19133a42 | 2054 | printk(KERN_INFO "md: bind<%s>\n", b); |
86e6ffdd | 2055 | |
b2d6db58 | 2056 | if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b))) |
5e55e2f5 | 2057 | goto fail; |
86e6ffdd | 2058 | |
0762b8bd | 2059 | ko = &part_to_dev(rdev->bdev->bd_part)->kobj; |
00bcb4ac N |
2060 | if (sysfs_create_link(&rdev->kobj, ko, "block")) |
2061 | /* failure here is OK */; | |
2062 | rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state"); | |
3c0ee63a | 2063 | |
4b80991c | 2064 | list_add_rcu(&rdev->same_set, &mddev->disks); |
e09b457b | 2065 | bd_link_disk_holder(rdev->bdev, mddev->gendisk); |
4044ba58 N |
2066 | |
2067 | /* May as well allow recovery to be retried once */ | |
5389042f | 2068 | mddev->recovery_disabled++; |
3f9d99c1 | 2069 | |
1da177e4 | 2070 | return 0; |
5e55e2f5 N |
2071 | |
2072 | fail: | |
2073 | printk(KERN_WARNING "md: failed to register dev-%s for %s\n", | |
2074 | b, mdname(mddev)); | |
2075 | return err; | |
1da177e4 LT |
2076 | } |
2077 | ||
177a99b2 | 2078 | static void md_delayed_delete(struct work_struct *ws) |
5792a285 | 2079 | { |
3cb03002 | 2080 | struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work); |
5792a285 | 2081 | kobject_del(&rdev->kobj); |
177a99b2 | 2082 | kobject_put(&rdev->kobj); |
5792a285 N |
2083 | } |
2084 | ||
f72ffdd6 | 2085 | static void unbind_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2086 | { |
2087 | char b[BDEVNAME_SIZE]; | |
403df478 | 2088 | |
49731baa | 2089 | bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk); |
4b80991c | 2090 | list_del_rcu(&rdev->same_set); |
1da177e4 LT |
2091 | printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); |
2092 | rdev->mddev = NULL; | |
86e6ffdd | 2093 | sysfs_remove_link(&rdev->kobj, "block"); |
3c0ee63a N |
2094 | sysfs_put(rdev->sysfs_state); |
2095 | rdev->sysfs_state = NULL; | |
2230dfe4 | 2096 | rdev->badblocks.count = 0; |
5792a285 | 2097 | /* We need to delay this, otherwise we can deadlock when |
4b80991c N |
2098 | * writing to 'remove' to "dev/state". We also need |
2099 | * to delay it due to rcu usage. | |
5792a285 | 2100 | */ |
4b80991c | 2101 | synchronize_rcu(); |
177a99b2 N |
2102 | INIT_WORK(&rdev->del_work, md_delayed_delete); |
2103 | kobject_get(&rdev->kobj); | |
e804ac78 | 2104 | queue_work(md_misc_wq, &rdev->del_work); |
1da177e4 LT |
2105 | } |
2106 | ||
2107 | /* | |
2108 | * prevent the device from being mounted, repartitioned or | |
2109 | * otherwise reused by a RAID array (or any other kernel | |
2110 | * subsystem), by bd_claiming the device. | |
2111 | */ | |
3cb03002 | 2112 | static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared) |
1da177e4 LT |
2113 | { |
2114 | int err = 0; | |
2115 | struct block_device *bdev; | |
2116 | char b[BDEVNAME_SIZE]; | |
2117 | ||
d4d77629 | 2118 | bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, |
3cb03002 | 2119 | shared ? (struct md_rdev *)lock_rdev : rdev); |
1da177e4 LT |
2120 | if (IS_ERR(bdev)) { |
2121 | printk(KERN_ERR "md: could not open %s.\n", | |
2122 | __bdevname(dev, b)); | |
2123 | return PTR_ERR(bdev); | |
2124 | } | |
1da177e4 LT |
2125 | rdev->bdev = bdev; |
2126 | return err; | |
2127 | } | |
2128 | ||
3cb03002 | 2129 | static void unlock_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2130 | { |
2131 | struct block_device *bdev = rdev->bdev; | |
2132 | rdev->bdev = NULL; | |
e525fd89 | 2133 | blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL); |
1da177e4 LT |
2134 | } |
2135 | ||
2136 | void md_autodetect_dev(dev_t dev); | |
2137 | ||
f72ffdd6 | 2138 | static void export_rdev(struct md_rdev *rdev) |
1da177e4 LT |
2139 | { |
2140 | char b[BDEVNAME_SIZE]; | |
403df478 | 2141 | |
1da177e4 LT |
2142 | printk(KERN_INFO "md: export_rdev(%s)\n", |
2143 | bdevname(rdev->bdev,b)); | |
545c8795 | 2144 | md_rdev_clear(rdev); |
1da177e4 | 2145 | #ifndef MODULE |
d0fae18f N |
2146 | if (test_bit(AutoDetected, &rdev->flags)) |
2147 | md_autodetect_dev(rdev->bdev->bd_dev); | |
1da177e4 LT |
2148 | #endif |
2149 | unlock_rdev(rdev); | |
86e6ffdd | 2150 | kobject_put(&rdev->kobj); |
1da177e4 LT |
2151 | } |
2152 | ||
f72ffdd6 | 2153 | static void kick_rdev_from_array(struct md_rdev *rdev) |
1da177e4 LT |
2154 | { |
2155 | unbind_rdev_from_array(rdev); | |
2156 | export_rdev(rdev); | |
2157 | } | |
2158 | ||
fd01b88c | 2159 | static void export_array(struct mddev *mddev) |
1da177e4 | 2160 | { |
0638bb0e | 2161 | struct md_rdev *rdev; |
1da177e4 | 2162 | |
0638bb0e N |
2163 | while (!list_empty(&mddev->disks)) { |
2164 | rdev = list_first_entry(&mddev->disks, struct md_rdev, | |
2165 | same_set); | |
1da177e4 LT |
2166 | kick_rdev_from_array(rdev); |
2167 | } | |
1da177e4 LT |
2168 | mddev->raid_disks = 0; |
2169 | mddev->major_version = 0; | |
2170 | } | |
2171 | ||
f72ffdd6 | 2172 | static void sync_sbs(struct mddev *mddev, int nospares) |
1da177e4 | 2173 | { |
42543769 N |
2174 | /* Update each superblock (in-memory image), but |
2175 | * if we are allowed to, skip spares which already | |
2176 | * have the right event counter, or have one earlier | |
2177 | * (which would mean they aren't being marked as dirty | |
2178 | * with the rest of the array) | |
2179 | */ | |
3cb03002 | 2180 | struct md_rdev *rdev; |
dafb20fa | 2181 | rdev_for_each(rdev, mddev) { |
42543769 N |
2182 | if (rdev->sb_events == mddev->events || |
2183 | (nospares && | |
2184 | rdev->raid_disk < 0 && | |
42543769 N |
2185 | rdev->sb_events+1 == mddev->events)) { |
2186 | /* Don't update this superblock */ | |
2187 | rdev->sb_loaded = 2; | |
2188 | } else { | |
076f968b | 2189 | sync_super(mddev, rdev); |
42543769 N |
2190 | rdev->sb_loaded = 1; |
2191 | } | |
1da177e4 LT |
2192 | } |
2193 | } | |
2194 | ||
f72ffdd6 | 2195 | static void md_update_sb(struct mddev *mddev, int force_change) |
1da177e4 | 2196 | { |
3cb03002 | 2197 | struct md_rdev *rdev; |
06d91a5f | 2198 | int sync_req; |
42543769 | 2199 | int nospares = 0; |
2699b672 | 2200 | int any_badblocks_changed = 0; |
1da177e4 | 2201 | |
d87f064f N |
2202 | if (mddev->ro) { |
2203 | if (force_change) | |
2204 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2205 | return; | |
2206 | } | |
1da177e4 | 2207 | repeat: |
3a3a5ddb | 2208 | /* First make sure individual recovery_offsets are correct */ |
dafb20fa | 2209 | rdev_for_each(rdev, mddev) { |
3a3a5ddb N |
2210 | if (rdev->raid_disk >= 0 && |
2211 | mddev->delta_disks >= 0 && | |
2212 | !test_bit(In_sync, &rdev->flags) && | |
2213 | mddev->curr_resync_completed > rdev->recovery_offset) | |
2214 | rdev->recovery_offset = mddev->curr_resync_completed; | |
2215 | ||
f72ffdd6 | 2216 | } |
bd52b746 | 2217 | if (!mddev->persistent) { |
070dc6dd | 2218 | clear_bit(MD_CHANGE_CLEAN, &mddev->flags); |
3a3a5ddb | 2219 | clear_bit(MD_CHANGE_DEVS, &mddev->flags); |
de393cde | 2220 | if (!mddev->external) { |
d97a41dc | 2221 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
dafb20fa | 2222 | rdev_for_each(rdev, mddev) { |
de393cde | 2223 | if (rdev->badblocks.changed) { |
d0962936 | 2224 | rdev->badblocks.changed = 0; |
de393cde N |
2225 | md_ack_all_badblocks(&rdev->badblocks); |
2226 | md_error(mddev, rdev); | |
2227 | } | |
2228 | clear_bit(Blocked, &rdev->flags); | |
2229 | clear_bit(BlockedBadBlocks, &rdev->flags); | |
2230 | wake_up(&rdev->blocked_wait); | |
2231 | } | |
2232 | } | |
3a3a5ddb N |
2233 | wake_up(&mddev->sb_wait); |
2234 | return; | |
2235 | } | |
2236 | ||
a9701a30 | 2237 | spin_lock_irq(&mddev->write_lock); |
84692195 | 2238 | |
3a3a5ddb N |
2239 | mddev->utime = get_seconds(); |
2240 | ||
850b2b42 N |
2241 | if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags)) |
2242 | force_change = 1; | |
2243 | if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags)) | |
2244 | /* just a clean<-> dirty transition, possibly leave spares alone, | |
2245 | * though if events isn't the right even/odd, we will have to do | |
2246 | * spares after all | |
2247 | */ | |
2248 | nospares = 1; | |
2249 | if (force_change) | |
2250 | nospares = 0; | |
2251 | if (mddev->degraded) | |
84692195 N |
2252 | /* If the array is degraded, then skipping spares is both |
2253 | * dangerous and fairly pointless. | |
2254 | * Dangerous because a device that was removed from the array | |
2255 | * might have a event_count that still looks up-to-date, | |
2256 | * so it can be re-added without a resync. | |
2257 | * Pointless because if there are any spares to skip, | |
2258 | * then a recovery will happen and soon that array won't | |
2259 | * be degraded any more and the spare can go back to sleep then. | |
2260 | */ | |
850b2b42 | 2261 | nospares = 0; |
84692195 | 2262 | |
06d91a5f | 2263 | sync_req = mddev->in_sync; |
42543769 N |
2264 | |
2265 | /* If this is just a dirty<->clean transition, and the array is clean | |
2266 | * and 'events' is odd, we can roll back to the previous clean state */ | |
850b2b42 | 2267 | if (nospares |
42543769 | 2268 | && (mddev->in_sync && mddev->recovery_cp == MaxSector) |
a8707c08 N |
2269 | && mddev->can_decrease_events |
2270 | && mddev->events != 1) { | |
42543769 | 2271 | mddev->events--; |
a8707c08 N |
2272 | mddev->can_decrease_events = 0; |
2273 | } else { | |
42543769 N |
2274 | /* otherwise we have to go forward and ... */ |
2275 | mddev->events ++; | |
a8707c08 | 2276 | mddev->can_decrease_events = nospares; |
42543769 | 2277 | } |
1da177e4 | 2278 | |
403df478 N |
2279 | /* |
2280 | * This 64-bit counter should never wrap. | |
2281 | * Either we are in around ~1 trillion A.C., assuming | |
2282 | * 1 reboot per second, or we have a bug... | |
2283 | */ | |
2284 | WARN_ON(mddev->events == 0); | |
2699b672 | 2285 | |
dafb20fa | 2286 | rdev_for_each(rdev, mddev) { |
2699b672 N |
2287 | if (rdev->badblocks.changed) |
2288 | any_badblocks_changed++; | |
de393cde N |
2289 | if (test_bit(Faulty, &rdev->flags)) |
2290 | set_bit(FaultRecorded, &rdev->flags); | |
2291 | } | |
2699b672 | 2292 | |
e691063a | 2293 | sync_sbs(mddev, nospares); |
a9701a30 | 2294 | spin_unlock_irq(&mddev->write_lock); |
1da177e4 | 2295 | |
36a4e1fe N |
2296 | pr_debug("md: updating %s RAID superblock on device (in sync %d)\n", |
2297 | mdname(mddev), mddev->in_sync); | |
1da177e4 | 2298 | |
4ad13663 | 2299 | bitmap_update_sb(mddev->bitmap); |
dafb20fa | 2300 | rdev_for_each(rdev, mddev) { |
1da177e4 | 2301 | char b[BDEVNAME_SIZE]; |
36a4e1fe | 2302 | |
42543769 N |
2303 | if (rdev->sb_loaded != 1) |
2304 | continue; /* no noise on spare devices */ | |
1da177e4 | 2305 | |
f466722c | 2306 | if (!test_bit(Faulty, &rdev->flags)) { |
7bfa19f2 | 2307 | md_super_write(mddev,rdev, |
0f420358 | 2308 | rdev->sb_start, rdev->sb_size, |
7bfa19f2 | 2309 | rdev->sb_page); |
36a4e1fe N |
2310 | pr_debug("md: (write) %s's sb offset: %llu\n", |
2311 | bdevname(rdev->bdev, b), | |
2312 | (unsigned long long)rdev->sb_start); | |
42543769 | 2313 | rdev->sb_events = mddev->events; |
2699b672 N |
2314 | if (rdev->badblocks.size) { |
2315 | md_super_write(mddev, rdev, | |
2316 | rdev->badblocks.sector, | |
2317 | rdev->badblocks.size << 9, | |
2318 | rdev->bb_page); | |
2319 | rdev->badblocks.size = 0; | |
2320 | } | |
7bfa19f2 | 2321 | |
f466722c | 2322 | } else |
36a4e1fe N |
2323 | pr_debug("md: %s (skipping faulty)\n", |
2324 | bdevname(rdev->bdev, b)); | |
d70ed2e4 | 2325 | |
7bfa19f2 | 2326 | if (mddev->level == LEVEL_MULTIPATH) |
1da177e4 LT |
2327 | /* only need to write one superblock... */ |
2328 | break; | |
2329 | } | |
a9701a30 | 2330 | md_super_wait(mddev); |
850b2b42 | 2331 | /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */ |
7bfa19f2 | 2332 | |
a9701a30 | 2333 | spin_lock_irq(&mddev->write_lock); |
850b2b42 N |
2334 | if (mddev->in_sync != sync_req || |
2335 | test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
06d91a5f | 2336 | /* have to write it out again */ |
a9701a30 | 2337 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f N |
2338 | goto repeat; |
2339 | } | |
850b2b42 | 2340 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
a9701a30 | 2341 | spin_unlock_irq(&mddev->write_lock); |
3d310eb7 | 2342 | wake_up(&mddev->sb_wait); |
acb180b0 N |
2343 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
2344 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
06d91a5f | 2345 | |
dafb20fa | 2346 | rdev_for_each(rdev, mddev) { |
de393cde N |
2347 | if (test_and_clear_bit(FaultRecorded, &rdev->flags)) |
2348 | clear_bit(Blocked, &rdev->flags); | |
2349 | ||
2350 | if (any_badblocks_changed) | |
2699b672 | 2351 | md_ack_all_badblocks(&rdev->badblocks); |
de393cde N |
2352 | clear_bit(BlockedBadBlocks, &rdev->flags); |
2353 | wake_up(&rdev->blocked_wait); | |
2354 | } | |
1da177e4 LT |
2355 | } |
2356 | ||
7f6ce769 | 2357 | /* words written to sysfs files may, or may not, be \n terminated. |
bce74dac N |
2358 | * We want to accept with case. For this we use cmd_match. |
2359 | */ | |
2360 | static int cmd_match(const char *cmd, const char *str) | |
2361 | { | |
2362 | /* See if cmd, written into a sysfs file, matches | |
2363 | * str. They must either be the same, or cmd can | |
2364 | * have a trailing newline | |
2365 | */ | |
2366 | while (*cmd && *str && *cmd == *str) { | |
2367 | cmd++; | |
2368 | str++; | |
2369 | } | |
2370 | if (*cmd == '\n') | |
2371 | cmd++; | |
2372 | if (*str || *cmd) | |
2373 | return 0; | |
2374 | return 1; | |
2375 | } | |
2376 | ||
86e6ffdd N |
2377 | struct rdev_sysfs_entry { |
2378 | struct attribute attr; | |
3cb03002 N |
2379 | ssize_t (*show)(struct md_rdev *, char *); |
2380 | ssize_t (*store)(struct md_rdev *, const char *, size_t); | |
86e6ffdd N |
2381 | }; |
2382 | ||
2383 | static ssize_t | |
3cb03002 | 2384 | state_show(struct md_rdev *rdev, char *page) |
86e6ffdd N |
2385 | { |
2386 | char *sep = ""; | |
20a49ff6 | 2387 | size_t len = 0; |
86e6ffdd | 2388 | |
de393cde N |
2389 | if (test_bit(Faulty, &rdev->flags) || |
2390 | rdev->badblocks.unacked_exist) { | |
86e6ffdd N |
2391 | len+= sprintf(page+len, "%sfaulty",sep); |
2392 | sep = ","; | |
2393 | } | |
b2d444d7 | 2394 | if (test_bit(In_sync, &rdev->flags)) { |
86e6ffdd N |
2395 | len += sprintf(page+len, "%sin_sync",sep); |
2396 | sep = ","; | |
2397 | } | |
f655675b N |
2398 | if (test_bit(WriteMostly, &rdev->flags)) { |
2399 | len += sprintf(page+len, "%swrite_mostly",sep); | |
2400 | sep = ","; | |
2401 | } | |
de393cde | 2402 | if (test_bit(Blocked, &rdev->flags) || |
52c64152 N |
2403 | (rdev->badblocks.unacked_exist |
2404 | && !test_bit(Faulty, &rdev->flags))) { | |
6bfe0b49 DW |
2405 | len += sprintf(page+len, "%sblocked", sep); |
2406 | sep = ","; | |
2407 | } | |
b2d444d7 N |
2408 | if (!test_bit(Faulty, &rdev->flags) && |
2409 | !test_bit(In_sync, &rdev->flags)) { | |
86e6ffdd N |
2410 | len += sprintf(page+len, "%sspare", sep); |
2411 | sep = ","; | |
2412 | } | |
d7a9d443 N |
2413 | if (test_bit(WriteErrorSeen, &rdev->flags)) { |
2414 | len += sprintf(page+len, "%swrite_error", sep); | |
2415 | sep = ","; | |
2416 | } | |
2d78f8c4 N |
2417 | if (test_bit(WantReplacement, &rdev->flags)) { |
2418 | len += sprintf(page+len, "%swant_replacement", sep); | |
2419 | sep = ","; | |
2420 | } | |
2421 | if (test_bit(Replacement, &rdev->flags)) { | |
2422 | len += sprintf(page+len, "%sreplacement", sep); | |
2423 | sep = ","; | |
2424 | } | |
2425 | ||
86e6ffdd N |
2426 | return len+sprintf(page+len, "\n"); |
2427 | } | |
2428 | ||
45dc2de1 | 2429 | static ssize_t |
3cb03002 | 2430 | state_store(struct md_rdev *rdev, const char *buf, size_t len) |
45dc2de1 N |
2431 | { |
2432 | /* can write | |
de393cde | 2433 | * faulty - simulates an error |
45dc2de1 | 2434 | * remove - disconnects the device |
f655675b N |
2435 | * writemostly - sets write_mostly |
2436 | * -writemostly - clears write_mostly | |
de393cde N |
2437 | * blocked - sets the Blocked flags |
2438 | * -blocked - clears the Blocked and possibly simulates an error | |
6d56e278 | 2439 | * insync - sets Insync providing device isn't active |
f466722c N |
2440 | * -insync - clear Insync for a device with a slot assigned, |
2441 | * so that it gets rebuilt based on bitmap | |
d7a9d443 N |
2442 | * write_error - sets WriteErrorSeen |
2443 | * -write_error - clears WriteErrorSeen | |
45dc2de1 N |
2444 | */ |
2445 | int err = -EINVAL; | |
2446 | if (cmd_match(buf, "faulty") && rdev->mddev->pers) { | |
2447 | md_error(rdev->mddev, rdev); | |
5ef56c8f N |
2448 | if (test_bit(Faulty, &rdev->flags)) |
2449 | err = 0; | |
2450 | else | |
2451 | err = -EBUSY; | |
45dc2de1 N |
2452 | } else if (cmd_match(buf, "remove")) { |
2453 | if (rdev->raid_disk >= 0) | |
2454 | err = -EBUSY; | |
2455 | else { | |
fd01b88c | 2456 | struct mddev *mddev = rdev->mddev; |
45dc2de1 | 2457 | kick_rdev_from_array(rdev); |
3f9d7b0d N |
2458 | if (mddev->pers) |
2459 | md_update_sb(mddev, 1); | |
45dc2de1 N |
2460 | md_new_event(mddev); |
2461 | err = 0; | |
2462 | } | |
f655675b N |
2463 | } else if (cmd_match(buf, "writemostly")) { |
2464 | set_bit(WriteMostly, &rdev->flags); | |
2465 | err = 0; | |
2466 | } else if (cmd_match(buf, "-writemostly")) { | |
2467 | clear_bit(WriteMostly, &rdev->flags); | |
6bfe0b49 DW |
2468 | err = 0; |
2469 | } else if (cmd_match(buf, "blocked")) { | |
2470 | set_bit(Blocked, &rdev->flags); | |
2471 | err = 0; | |
2472 | } else if (cmd_match(buf, "-blocked")) { | |
de393cde | 2473 | if (!test_bit(Faulty, &rdev->flags) && |
7da64a0a | 2474 | rdev->badblocks.unacked_exist) { |
de393cde N |
2475 | /* metadata handler doesn't understand badblocks, |
2476 | * so we need to fail the device | |
2477 | */ | |
2478 | md_error(rdev->mddev, rdev); | |
2479 | } | |
6bfe0b49 | 2480 | clear_bit(Blocked, &rdev->flags); |
de393cde | 2481 | clear_bit(BlockedBadBlocks, &rdev->flags); |
6bfe0b49 DW |
2482 | wake_up(&rdev->blocked_wait); |
2483 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2484 | md_wakeup_thread(rdev->mddev->thread); | |
2485 | ||
6d56e278 N |
2486 | err = 0; |
2487 | } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) { | |
2488 | set_bit(In_sync, &rdev->flags); | |
f655675b | 2489 | err = 0; |
f466722c | 2490 | } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) { |
e1960f8c N |
2491 | if (rdev->mddev->pers == NULL) { |
2492 | clear_bit(In_sync, &rdev->flags); | |
2493 | rdev->saved_raid_disk = rdev->raid_disk; | |
2494 | rdev->raid_disk = -1; | |
2495 | err = 0; | |
2496 | } | |
d7a9d443 N |
2497 | } else if (cmd_match(buf, "write_error")) { |
2498 | set_bit(WriteErrorSeen, &rdev->flags); | |
2499 | err = 0; | |
2500 | } else if (cmd_match(buf, "-write_error")) { | |
2501 | clear_bit(WriteErrorSeen, &rdev->flags); | |
2502 | err = 0; | |
2d78f8c4 N |
2503 | } else if (cmd_match(buf, "want_replacement")) { |
2504 | /* Any non-spare device that is not a replacement can | |
2505 | * become want_replacement at any time, but we then need to | |
2506 | * check if recovery is needed. | |
2507 | */ | |
2508 | if (rdev->raid_disk >= 0 && | |
2509 | !test_bit(Replacement, &rdev->flags)) | |
2510 | set_bit(WantReplacement, &rdev->flags); | |
2511 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); | |
2512 | md_wakeup_thread(rdev->mddev->thread); | |
2513 | err = 0; | |
2514 | } else if (cmd_match(buf, "-want_replacement")) { | |
2515 | /* Clearing 'want_replacement' is always allowed. | |
2516 | * Once replacements starts it is too late though. | |
2517 | */ | |
2518 | err = 0; | |
2519 | clear_bit(WantReplacement, &rdev->flags); | |
2520 | } else if (cmd_match(buf, "replacement")) { | |
2521 | /* Can only set a device as a replacement when array has not | |
2522 | * yet been started. Once running, replacement is automatic | |
2523 | * from spares, or by assigning 'slot'. | |
2524 | */ | |
2525 | if (rdev->mddev->pers) | |
2526 | err = -EBUSY; | |
2527 | else { | |
2528 | set_bit(Replacement, &rdev->flags); | |
2529 | err = 0; | |
2530 | } | |
2531 | } else if (cmd_match(buf, "-replacement")) { | |
2532 | /* Similarly, can only clear Replacement before start */ | |
2533 | if (rdev->mddev->pers) | |
2534 | err = -EBUSY; | |
2535 | else { | |
2536 | clear_bit(Replacement, &rdev->flags); | |
2537 | err = 0; | |
2538 | } | |
45dc2de1 | 2539 | } |
00bcb4ac N |
2540 | if (!err) |
2541 | sysfs_notify_dirent_safe(rdev->sysfs_state); | |
45dc2de1 N |
2542 | return err ? err : len; |
2543 | } | |
80ca3a44 N |
2544 | static struct rdev_sysfs_entry rdev_state = |
2545 | __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store); | |
86e6ffdd | 2546 | |
4dbcdc75 | 2547 | static ssize_t |
3cb03002 | 2548 | errors_show(struct md_rdev *rdev, char *page) |
4dbcdc75 N |
2549 | { |
2550 | return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors)); | |
2551 | } | |
2552 | ||
2553 | static ssize_t | |
3cb03002 | 2554 | errors_store(struct md_rdev *rdev, const char *buf, size_t len) |
4dbcdc75 N |
2555 | { |
2556 | char *e; | |
2557 | unsigned long n = simple_strtoul(buf, &e, 10); | |
2558 | if (*buf && (*e == 0 || *e == '\n')) { | |
2559 | atomic_set(&rdev->corrected_errors, n); | |
2560 | return len; | |
2561 | } | |
2562 | return -EINVAL; | |
2563 | } | |
2564 | static struct rdev_sysfs_entry rdev_errors = | |
80ca3a44 | 2565 | __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store); |
4dbcdc75 | 2566 | |
014236d2 | 2567 | static ssize_t |
3cb03002 | 2568 | slot_show(struct md_rdev *rdev, char *page) |
014236d2 N |
2569 | { |
2570 | if (rdev->raid_disk < 0) | |
2571 | return sprintf(page, "none\n"); | |
2572 | else | |
2573 | return sprintf(page, "%d\n", rdev->raid_disk); | |
2574 | } | |
2575 | ||
2576 | static ssize_t | |
3cb03002 | 2577 | slot_store(struct md_rdev *rdev, const char *buf, size_t len) |
014236d2 N |
2578 | { |
2579 | char *e; | |
c303da6d | 2580 | int err; |
014236d2 N |
2581 | int slot = simple_strtoul(buf, &e, 10); |
2582 | if (strncmp(buf, "none", 4)==0) | |
2583 | slot = -1; | |
2584 | else if (e==buf || (*e && *e!= '\n')) | |
2585 | return -EINVAL; | |
6c2fce2e | 2586 | if (rdev->mddev->pers && slot == -1) { |
c303da6d N |
2587 | /* Setting 'slot' on an active array requires also |
2588 | * updating the 'rd%d' link, and communicating | |
2589 | * with the personality with ->hot_*_disk. | |
2590 | * For now we only support removing | |
2591 | * failed/spare devices. This normally happens automatically, | |
2592 | * but not when the metadata is externally managed. | |
2593 | */ | |
c303da6d N |
2594 | if (rdev->raid_disk == -1) |
2595 | return -EEXIST; | |
2596 | /* personality does all needed checks */ | |
01393f3d | 2597 | if (rdev->mddev->pers->hot_remove_disk == NULL) |
c303da6d | 2598 | return -EINVAL; |
746d3207 N |
2599 | clear_bit(Blocked, &rdev->flags); |
2600 | remove_and_add_spares(rdev->mddev, rdev); | |
2601 | if (rdev->raid_disk >= 0) | |
2602 | return -EBUSY; | |
c303da6d N |
2603 | set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery); |
2604 | md_wakeup_thread(rdev->mddev->thread); | |
6c2fce2e | 2605 | } else if (rdev->mddev->pers) { |
6c2fce2e | 2606 | /* Activating a spare .. or possibly reactivating |
6d56e278 | 2607 | * if we ever get bitmaps working here. |
6c2fce2e NB |
2608 | */ |
2609 | ||
2610 | if (rdev->raid_disk != -1) | |
2611 | return -EBUSY; | |
2612 | ||
c6751b2b N |
2613 | if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery)) |
2614 | return -EBUSY; | |
2615 | ||
6c2fce2e NB |
2616 | if (rdev->mddev->pers->hot_add_disk == NULL) |
2617 | return -EINVAL; | |
2618 | ||
ba1b41b6 N |
2619 | if (slot >= rdev->mddev->raid_disks && |
2620 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
2621 | return -ENOSPC; | |
2622 | ||
6c2fce2e NB |
2623 | rdev->raid_disk = slot; |
2624 | if (test_bit(In_sync, &rdev->flags)) | |
2625 | rdev->saved_raid_disk = slot; | |
2626 | else | |
2627 | rdev->saved_raid_disk = -1; | |
d30519fc | 2628 | clear_bit(In_sync, &rdev->flags); |
8313b8e5 | 2629 | clear_bit(Bitmap_sync, &rdev->flags); |
6c2fce2e NB |
2630 | err = rdev->mddev->pers-> |
2631 | hot_add_disk(rdev->mddev, rdev); | |
199050ea | 2632 | if (err) { |
6c2fce2e | 2633 | rdev->raid_disk = -1; |
6c2fce2e | 2634 | return err; |
52664732 | 2635 | } else |
00bcb4ac | 2636 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
36fad858 | 2637 | if (sysfs_link_rdev(rdev->mddev, rdev)) |
00bcb4ac | 2638 | /* failure here is OK */; |
6c2fce2e | 2639 | /* don't wakeup anyone, leave that to userspace. */ |
c303da6d | 2640 | } else { |
ba1b41b6 N |
2641 | if (slot >= rdev->mddev->raid_disks && |
2642 | slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks) | |
c303da6d N |
2643 | return -ENOSPC; |
2644 | rdev->raid_disk = slot; | |
2645 | /* assume it is working */ | |
c5d79adb N |
2646 | clear_bit(Faulty, &rdev->flags); |
2647 | clear_bit(WriteMostly, &rdev->flags); | |
c303da6d | 2648 | set_bit(In_sync, &rdev->flags); |
00bcb4ac | 2649 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c303da6d | 2650 | } |
014236d2 N |
2651 | return len; |
2652 | } | |
2653 | ||
014236d2 | 2654 | static struct rdev_sysfs_entry rdev_slot = |
80ca3a44 | 2655 | __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store); |
014236d2 | 2656 | |
93c8cad0 | 2657 | static ssize_t |
3cb03002 | 2658 | offset_show(struct md_rdev *rdev, char *page) |
93c8cad0 | 2659 | { |
6961ece4 | 2660 | return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset); |
93c8cad0 N |
2661 | } |
2662 | ||
2663 | static ssize_t | |
3cb03002 | 2664 | offset_store(struct md_rdev *rdev, const char *buf, size_t len) |
93c8cad0 | 2665 | { |
c6563a8c | 2666 | unsigned long long offset; |
b29bebd6 | 2667 | if (kstrtoull(buf, 10, &offset) < 0) |
93c8cad0 | 2668 | return -EINVAL; |
8ed0a521 | 2669 | if (rdev->mddev->pers && rdev->raid_disk >= 0) |
93c8cad0 | 2670 | return -EBUSY; |
dd8ac336 | 2671 | if (rdev->sectors && rdev->mddev->external) |
c5d79adb N |
2672 | /* Must set offset before size, so overlap checks |
2673 | * can be sane */ | |
2674 | return -EBUSY; | |
93c8cad0 | 2675 | rdev->data_offset = offset; |
25f7fd47 | 2676 | rdev->new_data_offset = offset; |
93c8cad0 N |
2677 | return len; |
2678 | } | |
2679 | ||
2680 | static struct rdev_sysfs_entry rdev_offset = | |
80ca3a44 | 2681 | __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store); |
93c8cad0 | 2682 | |
c6563a8c N |
2683 | static ssize_t new_offset_show(struct md_rdev *rdev, char *page) |
2684 | { | |
2685 | return sprintf(page, "%llu\n", | |
2686 | (unsigned long long)rdev->new_data_offset); | |
2687 | } | |
2688 | ||
2689 | static ssize_t new_offset_store(struct md_rdev *rdev, | |
2690 | const char *buf, size_t len) | |
2691 | { | |
2692 | unsigned long long new_offset; | |
2693 | struct mddev *mddev = rdev->mddev; | |
2694 | ||
b29bebd6 | 2695 | if (kstrtoull(buf, 10, &new_offset) < 0) |
c6563a8c N |
2696 | return -EINVAL; |
2697 | ||
2698 | if (mddev->sync_thread) | |
2699 | return -EBUSY; | |
2700 | if (new_offset == rdev->data_offset) | |
2701 | /* reset is always permitted */ | |
2702 | ; | |
2703 | else if (new_offset > rdev->data_offset) { | |
2704 | /* must not push array size beyond rdev_sectors */ | |
2705 | if (new_offset - rdev->data_offset | |
2706 | + mddev->dev_sectors > rdev->sectors) | |
2707 | return -E2BIG; | |
2708 | } | |
2709 | /* Metadata worries about other space details. */ | |
2710 | ||
2711 | /* decreasing the offset is inconsistent with a backwards | |
2712 | * reshape. | |
2713 | */ | |
2714 | if (new_offset < rdev->data_offset && | |
2715 | mddev->reshape_backwards) | |
2716 | return -EINVAL; | |
2717 | /* Increasing offset is inconsistent with forwards | |
2718 | * reshape. reshape_direction should be set to | |
2719 | * 'backwards' first. | |
2720 | */ | |
2721 | if (new_offset > rdev->data_offset && | |
2722 | !mddev->reshape_backwards) | |
2723 | return -EINVAL; | |
2724 | ||
2725 | if (mddev->pers && mddev->persistent && | |
2726 | !super_types[mddev->major_version] | |
2727 | .allow_new_offset(rdev, new_offset)) | |
2728 | return -E2BIG; | |
2729 | rdev->new_data_offset = new_offset; | |
2730 | if (new_offset > rdev->data_offset) | |
2731 | mddev->reshape_backwards = 1; | |
2732 | else if (new_offset < rdev->data_offset) | |
2733 | mddev->reshape_backwards = 0; | |
2734 | ||
2735 | return len; | |
2736 | } | |
2737 | static struct rdev_sysfs_entry rdev_new_offset = | |
2738 | __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store); | |
2739 | ||
83303b61 | 2740 | static ssize_t |
3cb03002 | 2741 | rdev_size_show(struct md_rdev *rdev, char *page) |
83303b61 | 2742 | { |
dd8ac336 | 2743 | return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2); |
83303b61 N |
2744 | } |
2745 | ||
c5d79adb N |
2746 | static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2) |
2747 | { | |
2748 | /* check if two start/length pairs overlap */ | |
2749 | if (s1+l1 <= s2) | |
2750 | return 0; | |
2751 | if (s2+l2 <= s1) | |
2752 | return 0; | |
2753 | return 1; | |
2754 | } | |
2755 | ||
b522adcd DW |
2756 | static int strict_blocks_to_sectors(const char *buf, sector_t *sectors) |
2757 | { | |
2758 | unsigned long long blocks; | |
2759 | sector_t new; | |
2760 | ||
b29bebd6 | 2761 | if (kstrtoull(buf, 10, &blocks) < 0) |
b522adcd DW |
2762 | return -EINVAL; |
2763 | ||
2764 | if (blocks & 1ULL << (8 * sizeof(blocks) - 1)) | |
2765 | return -EINVAL; /* sector conversion overflow */ | |
2766 | ||
2767 | new = blocks * 2; | |
2768 | if (new != blocks * 2) | |
2769 | return -EINVAL; /* unsigned long long to sector_t overflow */ | |
2770 | ||
2771 | *sectors = new; | |
2772 | return 0; | |
2773 | } | |
2774 | ||
83303b61 | 2775 | static ssize_t |
3cb03002 | 2776 | rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len) |
83303b61 | 2777 | { |
fd01b88c | 2778 | struct mddev *my_mddev = rdev->mddev; |
dd8ac336 | 2779 | sector_t oldsectors = rdev->sectors; |
b522adcd | 2780 | sector_t sectors; |
27c529bb | 2781 | |
b522adcd | 2782 | if (strict_blocks_to_sectors(buf, §ors) < 0) |
d7027458 | 2783 | return -EINVAL; |
c6563a8c N |
2784 | if (rdev->data_offset != rdev->new_data_offset) |
2785 | return -EINVAL; /* too confusing */ | |
0cd17fec | 2786 | if (my_mddev->pers && rdev->raid_disk >= 0) { |
d7027458 | 2787 | if (my_mddev->persistent) { |
dd8ac336 AN |
2788 | sectors = super_types[my_mddev->major_version]. |
2789 | rdev_size_change(rdev, sectors); | |
2790 | if (!sectors) | |
0cd17fec | 2791 | return -EBUSY; |
dd8ac336 | 2792 | } else if (!sectors) |
77304d2a | 2793 | sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) - |
dd8ac336 | 2794 | rdev->data_offset; |
a6468539 N |
2795 | if (!my_mddev->pers->resize) |
2796 | /* Cannot change size for RAID0 or Linear etc */ | |
2797 | return -EINVAL; | |
0cd17fec | 2798 | } |
dd8ac336 | 2799 | if (sectors < my_mddev->dev_sectors) |
7d3c6f87 | 2800 | return -EINVAL; /* component must fit device */ |
0cd17fec | 2801 | |
dd8ac336 AN |
2802 | rdev->sectors = sectors; |
2803 | if (sectors > oldsectors && my_mddev->external) { | |
8b1afc3d N |
2804 | /* Need to check that all other rdevs with the same |
2805 | * ->bdev do not overlap. 'rcu' is sufficient to walk | |
2806 | * the rdev lists safely. | |
2807 | * This check does not provide a hard guarantee, it | |
2808 | * just helps avoid dangerous mistakes. | |
c5d79adb | 2809 | */ |
fd01b88c | 2810 | struct mddev *mddev; |
c5d79adb | 2811 | int overlap = 0; |
159ec1fc | 2812 | struct list_head *tmp; |
c5d79adb | 2813 | |
8b1afc3d | 2814 | rcu_read_lock(); |
29ac4aa3 | 2815 | for_each_mddev(mddev, tmp) { |
3cb03002 | 2816 | struct md_rdev *rdev2; |
c5d79adb | 2817 | |
dafb20fa | 2818 | rdev_for_each(rdev2, mddev) |
f21e9ff7 N |
2819 | if (rdev->bdev == rdev2->bdev && |
2820 | rdev != rdev2 && | |
2821 | overlaps(rdev->data_offset, rdev->sectors, | |
2822 | rdev2->data_offset, | |
2823 | rdev2->sectors)) { | |
c5d79adb N |
2824 | overlap = 1; |
2825 | break; | |
2826 | } | |
c5d79adb N |
2827 | if (overlap) { |
2828 | mddev_put(mddev); | |
2829 | break; | |
2830 | } | |
2831 | } | |
8b1afc3d | 2832 | rcu_read_unlock(); |
c5d79adb N |
2833 | if (overlap) { |
2834 | /* Someone else could have slipped in a size | |
2835 | * change here, but doing so is just silly. | |
dd8ac336 | 2836 | * We put oldsectors back because we *know* it is |
c5d79adb N |
2837 | * safe, and trust userspace not to race with |
2838 | * itself | |
2839 | */ | |
dd8ac336 | 2840 | rdev->sectors = oldsectors; |
c5d79adb N |
2841 | return -EBUSY; |
2842 | } | |
2843 | } | |
83303b61 N |
2844 | return len; |
2845 | } | |
2846 | ||
2847 | static struct rdev_sysfs_entry rdev_size = | |
80ca3a44 | 2848 | __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store); |
83303b61 | 2849 | |
3cb03002 | 2850 | static ssize_t recovery_start_show(struct md_rdev *rdev, char *page) |
06e3c817 DW |
2851 | { |
2852 | unsigned long long recovery_start = rdev->recovery_offset; | |
2853 | ||
2854 | if (test_bit(In_sync, &rdev->flags) || | |
2855 | recovery_start == MaxSector) | |
2856 | return sprintf(page, "none\n"); | |
2857 | ||
2858 | return sprintf(page, "%llu\n", recovery_start); | |
2859 | } | |
2860 | ||
3cb03002 | 2861 | static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len) |
06e3c817 DW |
2862 | { |
2863 | unsigned long long recovery_start; | |
2864 | ||
2865 | if (cmd_match(buf, "none")) | |
2866 | recovery_start = MaxSector; | |
b29bebd6 | 2867 | else if (kstrtoull(buf, 10, &recovery_start)) |
06e3c817 DW |
2868 | return -EINVAL; |
2869 | ||
2870 | if (rdev->mddev->pers && | |
2871 | rdev->raid_disk >= 0) | |
2872 | return -EBUSY; | |
2873 | ||
2874 | rdev->recovery_offset = recovery_start; | |
2875 | if (recovery_start == MaxSector) | |
2876 | set_bit(In_sync, &rdev->flags); | |
2877 | else | |
2878 | clear_bit(In_sync, &rdev->flags); | |
2879 | return len; | |
2880 | } | |
2881 | ||
2882 | static struct rdev_sysfs_entry rdev_recovery_start = | |
2883 | __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store); | |
2884 | ||
16c791a5 N |
2885 | static ssize_t |
2886 | badblocks_show(struct badblocks *bb, char *page, int unack); | |
2887 | static ssize_t | |
2888 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack); | |
2889 | ||
3cb03002 | 2890 | static ssize_t bb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
2891 | { |
2892 | return badblocks_show(&rdev->badblocks, page, 0); | |
2893 | } | |
3cb03002 | 2894 | static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 | 2895 | { |
de393cde N |
2896 | int rv = badblocks_store(&rdev->badblocks, page, len, 0); |
2897 | /* Maybe that ack was all we needed */ | |
2898 | if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags)) | |
2899 | wake_up(&rdev->blocked_wait); | |
2900 | return rv; | |
16c791a5 N |
2901 | } |
2902 | static struct rdev_sysfs_entry rdev_bad_blocks = | |
2903 | __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store); | |
2904 | ||
3cb03002 | 2905 | static ssize_t ubb_show(struct md_rdev *rdev, char *page) |
16c791a5 N |
2906 | { |
2907 | return badblocks_show(&rdev->badblocks, page, 1); | |
2908 | } | |
3cb03002 | 2909 | static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len) |
16c791a5 N |
2910 | { |
2911 | return badblocks_store(&rdev->badblocks, page, len, 1); | |
2912 | } | |
2913 | static struct rdev_sysfs_entry rdev_unack_bad_blocks = | |
2914 | __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store); | |
2915 | ||
86e6ffdd N |
2916 | static struct attribute *rdev_default_attrs[] = { |
2917 | &rdev_state.attr, | |
4dbcdc75 | 2918 | &rdev_errors.attr, |
014236d2 | 2919 | &rdev_slot.attr, |
93c8cad0 | 2920 | &rdev_offset.attr, |
c6563a8c | 2921 | &rdev_new_offset.attr, |
83303b61 | 2922 | &rdev_size.attr, |
06e3c817 | 2923 | &rdev_recovery_start.attr, |
16c791a5 N |
2924 | &rdev_bad_blocks.attr, |
2925 | &rdev_unack_bad_blocks.attr, | |
86e6ffdd N |
2926 | NULL, |
2927 | }; | |
2928 | static ssize_t | |
2929 | rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
2930 | { | |
2931 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 2932 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
fd01b88c | 2933 | struct mddev *mddev = rdev->mddev; |
27c529bb | 2934 | ssize_t rv; |
86e6ffdd N |
2935 | |
2936 | if (!entry->show) | |
2937 | return -EIO; | |
27c529bb N |
2938 | |
2939 | rv = mddev ? mddev_lock(mddev) : -EBUSY; | |
2940 | if (!rv) { | |
2941 | if (rdev->mddev == NULL) | |
2942 | rv = -EBUSY; | |
2943 | else | |
2944 | rv = entry->show(rdev, page); | |
2945 | mddev_unlock(mddev); | |
2946 | } | |
2947 | return rv; | |
86e6ffdd N |
2948 | } |
2949 | ||
2950 | static ssize_t | |
2951 | rdev_attr_store(struct kobject *kobj, struct attribute *attr, | |
2952 | const char *page, size_t length) | |
2953 | { | |
2954 | struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr); | |
3cb03002 | 2955 | struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj); |
27c529bb | 2956 | ssize_t rv; |
fd01b88c | 2957 | struct mddev *mddev = rdev->mddev; |
86e6ffdd N |
2958 | |
2959 | if (!entry->store) | |
2960 | return -EIO; | |
67463acb N |
2961 | if (!capable(CAP_SYS_ADMIN)) |
2962 | return -EACCES; | |
27c529bb | 2963 | rv = mddev ? mddev_lock(mddev): -EBUSY; |
ca388059 | 2964 | if (!rv) { |
27c529bb N |
2965 | if (rdev->mddev == NULL) |
2966 | rv = -EBUSY; | |
2967 | else | |
2968 | rv = entry->store(rdev, page, length); | |
6a51830e | 2969 | mddev_unlock(mddev); |
ca388059 N |
2970 | } |
2971 | return rv; | |
86e6ffdd N |
2972 | } |
2973 | ||
2974 | static void rdev_free(struct kobject *ko) | |
2975 | { | |
3cb03002 | 2976 | struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj); |
86e6ffdd N |
2977 | kfree(rdev); |
2978 | } | |
52cf25d0 | 2979 | static const struct sysfs_ops rdev_sysfs_ops = { |
86e6ffdd N |
2980 | .show = rdev_attr_show, |
2981 | .store = rdev_attr_store, | |
2982 | }; | |
2983 | static struct kobj_type rdev_ktype = { | |
2984 | .release = rdev_free, | |
2985 | .sysfs_ops = &rdev_sysfs_ops, | |
2986 | .default_attrs = rdev_default_attrs, | |
2987 | }; | |
2988 | ||
3cb03002 | 2989 | int md_rdev_init(struct md_rdev *rdev) |
e8bb9a83 N |
2990 | { |
2991 | rdev->desc_nr = -1; | |
2992 | rdev->saved_raid_disk = -1; | |
2993 | rdev->raid_disk = -1; | |
2994 | rdev->flags = 0; | |
2995 | rdev->data_offset = 0; | |
c6563a8c | 2996 | rdev->new_data_offset = 0; |
e8bb9a83 N |
2997 | rdev->sb_events = 0; |
2998 | rdev->last_read_error.tv_sec = 0; | |
2999 | rdev->last_read_error.tv_nsec = 0; | |
2699b672 N |
3000 | rdev->sb_loaded = 0; |
3001 | rdev->bb_page = NULL; | |
e8bb9a83 N |
3002 | atomic_set(&rdev->nr_pending, 0); |
3003 | atomic_set(&rdev->read_errors, 0); | |
3004 | atomic_set(&rdev->corrected_errors, 0); | |
3005 | ||
3006 | INIT_LIST_HEAD(&rdev->same_set); | |
3007 | init_waitqueue_head(&rdev->blocked_wait); | |
2230dfe4 N |
3008 | |
3009 | /* Add space to store bad block list. | |
3010 | * This reserves the space even on arrays where it cannot | |
3011 | * be used - I wonder if that matters | |
3012 | */ | |
3013 | rdev->badblocks.count = 0; | |
486adf72 | 3014 | rdev->badblocks.shift = -1; /* disabled until explicitly enabled */ |
2230dfe4 N |
3015 | rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3016 | seqlock_init(&rdev->badblocks.lock); | |
3017 | if (rdev->badblocks.page == NULL) | |
3018 | return -ENOMEM; | |
3019 | ||
3020 | return 0; | |
e8bb9a83 N |
3021 | } |
3022 | EXPORT_SYMBOL_GPL(md_rdev_init); | |
1da177e4 LT |
3023 | /* |
3024 | * Import a device. If 'super_format' >= 0, then sanity check the superblock | |
3025 | * | |
3026 | * mark the device faulty if: | |
3027 | * | |
3028 | * - the device is nonexistent (zero size) | |
3029 | * - the device has no valid superblock | |
3030 | * | |
3031 | * a faulty rdev _never_ has rdev->sb set. | |
3032 | */ | |
3cb03002 | 3033 | static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor) |
1da177e4 LT |
3034 | { |
3035 | char b[BDEVNAME_SIZE]; | |
3036 | int err; | |
3cb03002 | 3037 | struct md_rdev *rdev; |
1da177e4 LT |
3038 | sector_t size; |
3039 | ||
9ffae0cf | 3040 | rdev = kzalloc(sizeof(*rdev), GFP_KERNEL); |
1da177e4 LT |
3041 | if (!rdev) { |
3042 | printk(KERN_ERR "md: could not alloc mem for new device!\n"); | |
3043 | return ERR_PTR(-ENOMEM); | |
3044 | } | |
1da177e4 | 3045 | |
2230dfe4 N |
3046 | err = md_rdev_init(rdev); |
3047 | if (err) | |
3048 | goto abort_free; | |
3049 | err = alloc_disk_sb(rdev); | |
3050 | if (err) | |
1da177e4 LT |
3051 | goto abort_free; |
3052 | ||
c5d79adb | 3053 | err = lock_rdev(rdev, newdev, super_format == -2); |
1da177e4 LT |
3054 | if (err) |
3055 | goto abort_free; | |
3056 | ||
f9cb074b | 3057 | kobject_init(&rdev->kobj, &rdev_ktype); |
86e6ffdd | 3058 | |
77304d2a | 3059 | size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS; |
1da177e4 | 3060 | if (!size) { |
f72ffdd6 | 3061 | printk(KERN_WARNING |
1da177e4 LT |
3062 | "md: %s has zero or unknown size, marking faulty!\n", |
3063 | bdevname(rdev->bdev,b)); | |
3064 | err = -EINVAL; | |
3065 | goto abort_free; | |
3066 | } | |
3067 | ||
3068 | if (super_format >= 0) { | |
3069 | err = super_types[super_format]. | |
3070 | load_super(rdev, NULL, super_minor); | |
3071 | if (err == -EINVAL) { | |
df968c4e N |
3072 | printk(KERN_WARNING |
3073 | "md: %s does not have a valid v%d.%d " | |
3074 | "superblock, not importing!\n", | |
3075 | bdevname(rdev->bdev,b), | |
3076 | super_format, super_minor); | |
1da177e4 LT |
3077 | goto abort_free; |
3078 | } | |
3079 | if (err < 0) { | |
f72ffdd6 | 3080 | printk(KERN_WARNING |
1da177e4 LT |
3081 | "md: could not read %s's sb, not importing!\n", |
3082 | bdevname(rdev->bdev,b)); | |
3083 | goto abort_free; | |
3084 | } | |
3085 | } | |
6bfe0b49 | 3086 | |
1da177e4 LT |
3087 | return rdev; |
3088 | ||
3089 | abort_free: | |
2699b672 N |
3090 | if (rdev->bdev) |
3091 | unlock_rdev(rdev); | |
545c8795 | 3092 | md_rdev_clear(rdev); |
1da177e4 LT |
3093 | kfree(rdev); |
3094 | return ERR_PTR(err); | |
3095 | } | |
3096 | ||
3097 | /* | |
3098 | * Check a full RAID array for plausibility | |
3099 | */ | |
3100 | ||
f72ffdd6 | 3101 | static void analyze_sbs(struct mddev *mddev) |
1da177e4 LT |
3102 | { |
3103 | int i; | |
3cb03002 | 3104 | struct md_rdev *rdev, *freshest, *tmp; |
1da177e4 LT |
3105 | char b[BDEVNAME_SIZE]; |
3106 | ||
3107 | freshest = NULL; | |
dafb20fa | 3108 | rdev_for_each_safe(rdev, tmp, mddev) |
1da177e4 LT |
3109 | switch (super_types[mddev->major_version]. |
3110 | load_super(rdev, freshest, mddev->minor_version)) { | |
3111 | case 1: | |
3112 | freshest = rdev; | |
3113 | break; | |
3114 | case 0: | |
3115 | break; | |
3116 | default: | |
3117 | printk( KERN_ERR \ | |
3118 | "md: fatal superblock inconsistency in %s" | |
f72ffdd6 | 3119 | " -- removing from array\n", |
1da177e4 LT |
3120 | bdevname(rdev->bdev,b)); |
3121 | kick_rdev_from_array(rdev); | |
3122 | } | |
3123 | ||
1da177e4 LT |
3124 | super_types[mddev->major_version]. |
3125 | validate_super(mddev, freshest); | |
3126 | ||
3127 | i = 0; | |
dafb20fa | 3128 | rdev_for_each_safe(rdev, tmp, mddev) { |
233fca36 N |
3129 | if (mddev->max_disks && |
3130 | (rdev->desc_nr >= mddev->max_disks || | |
3131 | i > mddev->max_disks)) { | |
de01dfad N |
3132 | printk(KERN_WARNING |
3133 | "md: %s: %s: only %d devices permitted\n", | |
3134 | mdname(mddev), bdevname(rdev->bdev, b), | |
3135 | mddev->max_disks); | |
3136 | kick_rdev_from_array(rdev); | |
3137 | continue; | |
3138 | } | |
1da177e4 LT |
3139 | if (rdev != freshest) |
3140 | if (super_types[mddev->major_version]. | |
3141 | validate_super(mddev, rdev)) { | |
3142 | printk(KERN_WARNING "md: kicking non-fresh %s" | |
3143 | " from array!\n", | |
3144 | bdevname(rdev->bdev,b)); | |
3145 | kick_rdev_from_array(rdev); | |
3146 | continue; | |
3147 | } | |
3148 | if (mddev->level == LEVEL_MULTIPATH) { | |
3149 | rdev->desc_nr = i++; | |
3150 | rdev->raid_disk = rdev->desc_nr; | |
b2d444d7 | 3151 | set_bit(In_sync, &rdev->flags); |
5e5e3e78 | 3152 | } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) { |
a778b73f N |
3153 | rdev->raid_disk = -1; |
3154 | clear_bit(In_sync, &rdev->flags); | |
1da177e4 LT |
3155 | } |
3156 | } | |
1da177e4 LT |
3157 | } |
3158 | ||
72e02075 N |
3159 | /* Read a fixed-point number. |
3160 | * Numbers in sysfs attributes should be in "standard" units where | |
3161 | * possible, so time should be in seconds. | |
f72ffdd6 | 3162 | * However we internally use a a much smaller unit such as |
72e02075 N |
3163 | * milliseconds or jiffies. |
3164 | * This function takes a decimal number with a possible fractional | |
3165 | * component, and produces an integer which is the result of | |
3166 | * multiplying that number by 10^'scale'. | |
3167 | * all without any floating-point arithmetic. | |
3168 | */ | |
3169 | int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale) | |
3170 | { | |
3171 | unsigned long result = 0; | |
3172 | long decimals = -1; | |
3173 | while (isdigit(*cp) || (*cp == '.' && decimals < 0)) { | |
3174 | if (*cp == '.') | |
3175 | decimals = 0; | |
3176 | else if (decimals < scale) { | |
3177 | unsigned int value; | |
3178 | value = *cp - '0'; | |
3179 | result = result * 10 + value; | |
3180 | if (decimals >= 0) | |
3181 | decimals++; | |
3182 | } | |
3183 | cp++; | |
3184 | } | |
3185 | if (*cp == '\n') | |
3186 | cp++; | |
3187 | if (*cp) | |
3188 | return -EINVAL; | |
3189 | if (decimals < 0) | |
3190 | decimals = 0; | |
3191 | while (decimals < scale) { | |
3192 | result *= 10; | |
3193 | decimals ++; | |
3194 | } | |
3195 | *res = result; | |
3196 | return 0; | |
3197 | } | |
3198 | ||
19052c0e N |
3199 | static void md_safemode_timeout(unsigned long data); |
3200 | ||
16f17b39 | 3201 | static ssize_t |
fd01b88c | 3202 | safe_delay_show(struct mddev *mddev, char *page) |
16f17b39 N |
3203 | { |
3204 | int msec = (mddev->safemode_delay*1000)/HZ; | |
3205 | return sprintf(page, "%d.%03d\n", msec/1000, msec%1000); | |
3206 | } | |
3207 | static ssize_t | |
fd01b88c | 3208 | safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len) |
16f17b39 | 3209 | { |
16f17b39 | 3210 | unsigned long msec; |
97ce0a7f | 3211 | |
72e02075 | 3212 | if (strict_strtoul_scaled(cbuf, &msec, 3) < 0) |
16f17b39 | 3213 | return -EINVAL; |
16f17b39 N |
3214 | if (msec == 0) |
3215 | mddev->safemode_delay = 0; | |
3216 | else { | |
19052c0e | 3217 | unsigned long old_delay = mddev->safemode_delay; |
16f17b39 N |
3218 | mddev->safemode_delay = (msec*HZ)/1000; |
3219 | if (mddev->safemode_delay == 0) | |
3220 | mddev->safemode_delay = 1; | |
275c51c4 | 3221 | if (mddev->safemode_delay < old_delay || old_delay == 0) |
19052c0e | 3222 | md_safemode_timeout((unsigned long)mddev); |
16f17b39 N |
3223 | } |
3224 | return len; | |
3225 | } | |
3226 | static struct md_sysfs_entry md_safe_delay = | |
80ca3a44 | 3227 | __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store); |
16f17b39 | 3228 | |
eae1701f | 3229 | static ssize_t |
fd01b88c | 3230 | level_show(struct mddev *mddev, char *page) |
eae1701f | 3231 | { |
84fc4b56 | 3232 | struct md_personality *p = mddev->pers; |
d9d166c2 | 3233 | if (p) |
eae1701f | 3234 | return sprintf(page, "%s\n", p->name); |
d9d166c2 N |
3235 | else if (mddev->clevel[0]) |
3236 | return sprintf(page, "%s\n", mddev->clevel); | |
3237 | else if (mddev->level != LEVEL_NONE) | |
3238 | return sprintf(page, "%d\n", mddev->level); | |
3239 | else | |
3240 | return 0; | |
eae1701f N |
3241 | } |
3242 | ||
d9d166c2 | 3243 | static ssize_t |
fd01b88c | 3244 | level_store(struct mddev *mddev, const char *buf, size_t len) |
d9d166c2 | 3245 | { |
f2859af6 | 3246 | char clevel[16]; |
20a49ff6 | 3247 | ssize_t rv = len; |
84fc4b56 | 3248 | struct md_personality *pers; |
f2859af6 | 3249 | long level; |
245f46c2 | 3250 | void *priv; |
3cb03002 | 3251 | struct md_rdev *rdev; |
245f46c2 N |
3252 | |
3253 | if (mddev->pers == NULL) { | |
3254 | if (len == 0) | |
3255 | return 0; | |
3256 | if (len >= sizeof(mddev->clevel)) | |
3257 | return -ENOSPC; | |
3258 | strncpy(mddev->clevel, buf, len); | |
3259 | if (mddev->clevel[len-1] == '\n') | |
3260 | len--; | |
3261 | mddev->clevel[len] = 0; | |
3262 | mddev->level = LEVEL_NONE; | |
3263 | return rv; | |
3264 | } | |
bd8839e0 N |
3265 | if (mddev->ro) |
3266 | return -EROFS; | |
245f46c2 N |
3267 | |
3268 | /* request to change the personality. Need to ensure: | |
3269 | * - array is not engaged in resync/recovery/reshape | |
3270 | * - old personality can be suspended | |
3271 | * - new personality will access other array. | |
3272 | */ | |
3273 | ||
bb4f1e9d N |
3274 | if (mddev->sync_thread || |
3275 | mddev->reshape_position != MaxSector || | |
3276 | mddev->sysfs_active) | |
d9d166c2 | 3277 | return -EBUSY; |
245f46c2 N |
3278 | |
3279 | if (!mddev->pers->quiesce) { | |
3280 | printk(KERN_WARNING "md: %s: %s does not support online personality change\n", | |
3281 | mdname(mddev), mddev->pers->name); | |
3282 | return -EINVAL; | |
3283 | } | |
3284 | ||
3285 | /* Now find the new personality */ | |
f2859af6 | 3286 | if (len == 0 || len >= sizeof(clevel)) |
245f46c2 | 3287 | return -EINVAL; |
f2859af6 DW |
3288 | strncpy(clevel, buf, len); |
3289 | if (clevel[len-1] == '\n') | |
d9d166c2 | 3290 | len--; |
f2859af6 | 3291 | clevel[len] = 0; |
b29bebd6 | 3292 | if (kstrtol(clevel, 10, &level)) |
f2859af6 | 3293 | level = LEVEL_NONE; |
245f46c2 | 3294 | |
f2859af6 DW |
3295 | if (request_module("md-%s", clevel) != 0) |
3296 | request_module("md-level-%s", clevel); | |
245f46c2 | 3297 | spin_lock(&pers_lock); |
f2859af6 | 3298 | pers = find_pers(level, clevel); |
245f46c2 N |
3299 | if (!pers || !try_module_get(pers->owner)) { |
3300 | spin_unlock(&pers_lock); | |
f2859af6 | 3301 | printk(KERN_WARNING "md: personality %s not loaded\n", clevel); |
245f46c2 N |
3302 | return -EINVAL; |
3303 | } | |
3304 | spin_unlock(&pers_lock); | |
3305 | ||
3306 | if (pers == mddev->pers) { | |
3307 | /* Nothing to do! */ | |
3308 | module_put(pers->owner); | |
3309 | return rv; | |
3310 | } | |
3311 | if (!pers->takeover) { | |
3312 | module_put(pers->owner); | |
3313 | printk(KERN_WARNING "md: %s: %s does not support personality takeover\n", | |
f2859af6 | 3314 | mdname(mddev), clevel); |
245f46c2 N |
3315 | return -EINVAL; |
3316 | } | |
3317 | ||
dafb20fa | 3318 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3319 | rdev->new_raid_disk = rdev->raid_disk; |
3320 | ||
245f46c2 N |
3321 | /* ->takeover must set new_* and/or delta_disks |
3322 | * if it succeeds, and may set them when it fails. | |
3323 | */ | |
3324 | priv = pers->takeover(mddev); | |
3325 | if (IS_ERR(priv)) { | |
3326 | mddev->new_level = mddev->level; | |
3327 | mddev->new_layout = mddev->layout; | |
664e7c41 | 3328 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
245f46c2 N |
3329 | mddev->raid_disks -= mddev->delta_disks; |
3330 | mddev->delta_disks = 0; | |
2c810cdd | 3331 | mddev->reshape_backwards = 0; |
245f46c2 N |
3332 | module_put(pers->owner); |
3333 | printk(KERN_WARNING "md: %s: %s would not accept array\n", | |
f2859af6 | 3334 | mdname(mddev), clevel); |
245f46c2 N |
3335 | return PTR_ERR(priv); |
3336 | } | |
3337 | ||
3338 | /* Looks like we have a winner */ | |
3339 | mddev_suspend(mddev); | |
3340 | mddev->pers->stop(mddev); | |
f72ffdd6 | 3341 | |
a64c876f N |
3342 | if (mddev->pers->sync_request == NULL && |
3343 | pers->sync_request != NULL) { | |
3344 | /* need to add the md_redundancy_group */ | |
3345 | if (sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
3346 | printk(KERN_WARNING | |
3347 | "md: cannot register extra attributes for %s\n", | |
3348 | mdname(mddev)); | |
388975cc | 3349 | mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action"); |
f72ffdd6 | 3350 | } |
a64c876f N |
3351 | if (mddev->pers->sync_request != NULL && |
3352 | pers->sync_request == NULL) { | |
3353 | /* need to remove the md_redundancy_group */ | |
3354 | if (mddev->to_remove == NULL) | |
3355 | mddev->to_remove = &md_redundancy_group; | |
3356 | } | |
3357 | ||
54071b38 TM |
3358 | if (mddev->pers->sync_request == NULL && |
3359 | mddev->external) { | |
3360 | /* We are converting from a no-redundancy array | |
3361 | * to a redundancy array and metadata is managed | |
3362 | * externally so we need to be sure that writes | |
3363 | * won't block due to a need to transition | |
3364 | * clean->dirty | |
3365 | * until external management is started. | |
3366 | */ | |
3367 | mddev->in_sync = 0; | |
3368 | mddev->safemode_delay = 0; | |
3369 | mddev->safemode = 0; | |
3370 | } | |
3371 | ||
dafb20fa | 3372 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3373 | if (rdev->raid_disk < 0) |
3374 | continue; | |
bf2cb0da | 3375 | if (rdev->new_raid_disk >= mddev->raid_disks) |
e93f68a1 N |
3376 | rdev->new_raid_disk = -1; |
3377 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3378 | continue; | |
36fad858 | 3379 | sysfs_unlink_rdev(mddev, rdev); |
e93f68a1 | 3380 | } |
dafb20fa | 3381 | rdev_for_each(rdev, mddev) { |
e93f68a1 N |
3382 | if (rdev->raid_disk < 0) |
3383 | continue; | |
3384 | if (rdev->new_raid_disk == rdev->raid_disk) | |
3385 | continue; | |
3386 | rdev->raid_disk = rdev->new_raid_disk; | |
3387 | if (rdev->raid_disk < 0) | |
3a981b03 | 3388 | clear_bit(In_sync, &rdev->flags); |
e93f68a1 | 3389 | else { |
36fad858 NK |
3390 | if (sysfs_link_rdev(mddev, rdev)) |
3391 | printk(KERN_WARNING "md: cannot register rd%d" | |
3392 | " for %s after level change\n", | |
3393 | rdev->raid_disk, mdname(mddev)); | |
3a981b03 | 3394 | } |
e93f68a1 N |
3395 | } |
3396 | ||
3397 | module_put(mddev->pers->owner); | |
245f46c2 N |
3398 | mddev->pers = pers; |
3399 | mddev->private = priv; | |
3400 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); | |
3401 | mddev->level = mddev->new_level; | |
3402 | mddev->layout = mddev->new_layout; | |
664e7c41 | 3403 | mddev->chunk_sectors = mddev->new_chunk_sectors; |
245f46c2 | 3404 | mddev->delta_disks = 0; |
2c810cdd | 3405 | mddev->reshape_backwards = 0; |
fee68723 | 3406 | mddev->degraded = 0; |
9af204cf TM |
3407 | if (mddev->pers->sync_request == NULL) { |
3408 | /* this is now an array without redundancy, so | |
3409 | * it must always be in_sync | |
3410 | */ | |
3411 | mddev->in_sync = 1; | |
3412 | del_timer_sync(&mddev->safemode_timer); | |
3413 | } | |
02e5f5c0 | 3414 | blk_set_stacking_limits(&mddev->queue->limits); |
245f46c2 | 3415 | pers->run(mddev); |
245f46c2 | 3416 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
47525e59 | 3417 | mddev_resume(mddev); |
830778a1 N |
3418 | if (!mddev->thread) |
3419 | md_update_sb(mddev, 1); | |
5cac7861 | 3420 | sysfs_notify(&mddev->kobj, NULL, "level"); |
bb7f8d22 | 3421 | md_new_event(mddev); |
d9d166c2 N |
3422 | return rv; |
3423 | } | |
3424 | ||
3425 | static struct md_sysfs_entry md_level = | |
80ca3a44 | 3426 | __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store); |
eae1701f | 3427 | |
d4dbd025 | 3428 | static ssize_t |
fd01b88c | 3429 | layout_show(struct mddev *mddev, char *page) |
d4dbd025 N |
3430 | { |
3431 | /* just a number, not meaningful for all levels */ | |
08a02ecd N |
3432 | if (mddev->reshape_position != MaxSector && |
3433 | mddev->layout != mddev->new_layout) | |
3434 | return sprintf(page, "%d (%d)\n", | |
3435 | mddev->new_layout, mddev->layout); | |
d4dbd025 N |
3436 | return sprintf(page, "%d\n", mddev->layout); |
3437 | } | |
3438 | ||
3439 | static ssize_t | |
fd01b88c | 3440 | layout_store(struct mddev *mddev, const char *buf, size_t len) |
d4dbd025 N |
3441 | { |
3442 | char *e; | |
3443 | unsigned long n = simple_strtoul(buf, &e, 10); | |
d4dbd025 N |
3444 | |
3445 | if (!*buf || (*e && *e != '\n')) | |
3446 | return -EINVAL; | |
3447 | ||
b3546035 N |
3448 | if (mddev->pers) { |
3449 | int err; | |
50ac168a | 3450 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3451 | return -EBUSY; |
bd8839e0 N |
3452 | if (mddev->ro) |
3453 | return -EROFS; | |
597a711b | 3454 | mddev->new_layout = n; |
50ac168a | 3455 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3456 | if (err) { |
3457 | mddev->new_layout = mddev->layout; | |
b3546035 | 3458 | return err; |
597a711b | 3459 | } |
b3546035 | 3460 | } else { |
08a02ecd | 3461 | mddev->new_layout = n; |
b3546035 N |
3462 | if (mddev->reshape_position == MaxSector) |
3463 | mddev->layout = n; | |
3464 | } | |
d4dbd025 N |
3465 | return len; |
3466 | } | |
3467 | static struct md_sysfs_entry md_layout = | |
80ca3a44 | 3468 | __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store); |
d4dbd025 | 3469 | |
eae1701f | 3470 | static ssize_t |
fd01b88c | 3471 | raid_disks_show(struct mddev *mddev, char *page) |
eae1701f | 3472 | { |
bb636547 N |
3473 | if (mddev->raid_disks == 0) |
3474 | return 0; | |
08a02ecd N |
3475 | if (mddev->reshape_position != MaxSector && |
3476 | mddev->delta_disks != 0) | |
3477 | return sprintf(page, "%d (%d)\n", mddev->raid_disks, | |
3478 | mddev->raid_disks - mddev->delta_disks); | |
eae1701f N |
3479 | return sprintf(page, "%d\n", mddev->raid_disks); |
3480 | } | |
3481 | ||
fd01b88c | 3482 | static int update_raid_disks(struct mddev *mddev, int raid_disks); |
da943b99 N |
3483 | |
3484 | static ssize_t | |
fd01b88c | 3485 | raid_disks_store(struct mddev *mddev, const char *buf, size_t len) |
da943b99 | 3486 | { |
da943b99 N |
3487 | char *e; |
3488 | int rv = 0; | |
3489 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3490 | ||
3491 | if (!*buf || (*e && *e != '\n')) | |
3492 | return -EINVAL; | |
3493 | ||
3494 | if (mddev->pers) | |
3495 | rv = update_raid_disks(mddev, n); | |
08a02ecd | 3496 | else if (mddev->reshape_position != MaxSector) { |
c6563a8c | 3497 | struct md_rdev *rdev; |
08a02ecd | 3498 | int olddisks = mddev->raid_disks - mddev->delta_disks; |
c6563a8c N |
3499 | |
3500 | rdev_for_each(rdev, mddev) { | |
3501 | if (olddisks < n && | |
3502 | rdev->data_offset < rdev->new_data_offset) | |
3503 | return -EINVAL; | |
3504 | if (olddisks > n && | |
3505 | rdev->data_offset > rdev->new_data_offset) | |
3506 | return -EINVAL; | |
3507 | } | |
08a02ecd N |
3508 | mddev->delta_disks = n - olddisks; |
3509 | mddev->raid_disks = n; | |
2c810cdd | 3510 | mddev->reshape_backwards = (mddev->delta_disks < 0); |
08a02ecd | 3511 | } else |
da943b99 N |
3512 | mddev->raid_disks = n; |
3513 | return rv ? rv : len; | |
3514 | } | |
3515 | static struct md_sysfs_entry md_raid_disks = | |
80ca3a44 | 3516 | __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store); |
eae1701f | 3517 | |
3b34380a | 3518 | static ssize_t |
fd01b88c | 3519 | chunk_size_show(struct mddev *mddev, char *page) |
3b34380a | 3520 | { |
08a02ecd | 3521 | if (mddev->reshape_position != MaxSector && |
664e7c41 AN |
3522 | mddev->chunk_sectors != mddev->new_chunk_sectors) |
3523 | return sprintf(page, "%d (%d)\n", | |
3524 | mddev->new_chunk_sectors << 9, | |
9d8f0363 AN |
3525 | mddev->chunk_sectors << 9); |
3526 | return sprintf(page, "%d\n", mddev->chunk_sectors << 9); | |
3b34380a N |
3527 | } |
3528 | ||
3529 | static ssize_t | |
fd01b88c | 3530 | chunk_size_store(struct mddev *mddev, const char *buf, size_t len) |
3b34380a | 3531 | { |
3b34380a N |
3532 | char *e; |
3533 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3534 | ||
3b34380a N |
3535 | if (!*buf || (*e && *e != '\n')) |
3536 | return -EINVAL; | |
3537 | ||
b3546035 N |
3538 | if (mddev->pers) { |
3539 | int err; | |
50ac168a | 3540 | if (mddev->pers->check_reshape == NULL) |
b3546035 | 3541 | return -EBUSY; |
bd8839e0 N |
3542 | if (mddev->ro) |
3543 | return -EROFS; | |
597a711b | 3544 | mddev->new_chunk_sectors = n >> 9; |
50ac168a | 3545 | err = mddev->pers->check_reshape(mddev); |
597a711b N |
3546 | if (err) { |
3547 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
b3546035 | 3548 | return err; |
597a711b | 3549 | } |
b3546035 | 3550 | } else { |
664e7c41 | 3551 | mddev->new_chunk_sectors = n >> 9; |
b3546035 | 3552 | if (mddev->reshape_position == MaxSector) |
9d8f0363 | 3553 | mddev->chunk_sectors = n >> 9; |
b3546035 | 3554 | } |
3b34380a N |
3555 | return len; |
3556 | } | |
3557 | static struct md_sysfs_entry md_chunk_size = | |
80ca3a44 | 3558 | __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store); |
3b34380a | 3559 | |
a94213b1 | 3560 | static ssize_t |
fd01b88c | 3561 | resync_start_show(struct mddev *mddev, char *page) |
a94213b1 | 3562 | { |
d1a7c503 N |
3563 | if (mddev->recovery_cp == MaxSector) |
3564 | return sprintf(page, "none\n"); | |
a94213b1 N |
3565 | return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp); |
3566 | } | |
3567 | ||
3568 | static ssize_t | |
fd01b88c | 3569 | resync_start_store(struct mddev *mddev, const char *buf, size_t len) |
a94213b1 | 3570 | { |
a94213b1 N |
3571 | char *e; |
3572 | unsigned long long n = simple_strtoull(buf, &e, 10); | |
3573 | ||
b098636c | 3574 | if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
a94213b1 | 3575 | return -EBUSY; |
06e3c817 DW |
3576 | if (cmd_match(buf, "none")) |
3577 | n = MaxSector; | |
3578 | else if (!*buf || (*e && *e != '\n')) | |
a94213b1 N |
3579 | return -EINVAL; |
3580 | ||
3581 | mddev->recovery_cp = n; | |
db07d85e N |
3582 | if (mddev->pers) |
3583 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
a94213b1 N |
3584 | return len; |
3585 | } | |
3586 | static struct md_sysfs_entry md_resync_start = | |
80ca3a44 | 3587 | __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store); |
a94213b1 | 3588 | |
9e653b63 N |
3589 | /* |
3590 | * The array state can be: | |
3591 | * | |
3592 | * clear | |
3593 | * No devices, no size, no level | |
3594 | * Equivalent to STOP_ARRAY ioctl | |
3595 | * inactive | |
3596 | * May have some settings, but array is not active | |
3597 | * all IO results in error | |
3598 | * When written, doesn't tear down array, but just stops it | |
3599 | * suspended (not supported yet) | |
3600 | * All IO requests will block. The array can be reconfigured. | |
910d8cb3 | 3601 | * Writing this, if accepted, will block until array is quiescent |
9e653b63 N |
3602 | * readonly |
3603 | * no resync can happen. no superblocks get written. | |
3604 | * write requests fail | |
3605 | * read-auto | |
3606 | * like readonly, but behaves like 'clean' on a write request. | |
3607 | * | |
3608 | * clean - no pending writes, but otherwise active. | |
3609 | * When written to inactive array, starts without resync | |
3610 | * If a write request arrives then | |
3611 | * if metadata is known, mark 'dirty' and switch to 'active'. | |
3612 | * if not known, block and switch to write-pending | |
3613 | * If written to an active array that has pending writes, then fails. | |
3614 | * active | |
3615 | * fully active: IO and resync can be happening. | |
3616 | * When written to inactive array, starts with resync | |
3617 | * | |
3618 | * write-pending | |
3619 | * clean, but writes are blocked waiting for 'active' to be written. | |
3620 | * | |
3621 | * active-idle | |
3622 | * like active, but no writes have been seen for a while (100msec). | |
3623 | * | |
3624 | */ | |
3625 | enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active, | |
3626 | write_pending, active_idle, bad_word}; | |
05381954 | 3627 | static char *array_states[] = { |
9e653b63 N |
3628 | "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active", |
3629 | "write-pending", "active-idle", NULL }; | |
3630 | ||
3631 | static int match_word(const char *word, char **list) | |
3632 | { | |
3633 | int n; | |
3634 | for (n=0; list[n]; n++) | |
3635 | if (cmd_match(word, list[n])) | |
3636 | break; | |
3637 | return n; | |
3638 | } | |
3639 | ||
3640 | static ssize_t | |
fd01b88c | 3641 | array_state_show(struct mddev *mddev, char *page) |
9e653b63 N |
3642 | { |
3643 | enum array_state st = inactive; | |
3644 | ||
3645 | if (mddev->pers) | |
3646 | switch(mddev->ro) { | |
3647 | case 1: | |
3648 | st = readonly; | |
3649 | break; | |
3650 | case 2: | |
3651 | st = read_auto; | |
3652 | break; | |
3653 | case 0: | |
3654 | if (mddev->in_sync) | |
3655 | st = clean; | |
070dc6dd | 3656 | else if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
e691063a | 3657 | st = write_pending; |
9e653b63 N |
3658 | else if (mddev->safemode) |
3659 | st = active_idle; | |
3660 | else | |
3661 | st = active; | |
3662 | } | |
3663 | else { | |
3664 | if (list_empty(&mddev->disks) && | |
3665 | mddev->raid_disks == 0 && | |
58c0fed4 | 3666 | mddev->dev_sectors == 0) |
9e653b63 N |
3667 | st = clear; |
3668 | else | |
3669 | st = inactive; | |
3670 | } | |
3671 | return sprintf(page, "%s\n", array_states[st]); | |
3672 | } | |
3673 | ||
f72ffdd6 N |
3674 | static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev); |
3675 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev); | |
3676 | static int do_md_run(struct mddev *mddev); | |
fd01b88c | 3677 | static int restart_array(struct mddev *mddev); |
9e653b63 N |
3678 | |
3679 | static ssize_t | |
fd01b88c | 3680 | array_state_store(struct mddev *mddev, const char *buf, size_t len) |
9e653b63 N |
3681 | { |
3682 | int err = -EINVAL; | |
3683 | enum array_state st = match_word(buf, array_states); | |
3684 | switch(st) { | |
3685 | case bad_word: | |
3686 | break; | |
3687 | case clear: | |
3688 | /* stopping an active array */ | |
a05b7ea0 | 3689 | err = do_md_stop(mddev, 0, NULL); |
9e653b63 N |
3690 | break; |
3691 | case inactive: | |
3692 | /* stopping an active array */ | |
90cf195d | 3693 | if (mddev->pers) |
a05b7ea0 | 3694 | err = do_md_stop(mddev, 2, NULL); |
90cf195d | 3695 | else |
e691063a | 3696 | err = 0; /* already inactive */ |
9e653b63 N |
3697 | break; |
3698 | case suspended: | |
3699 | break; /* not supported yet */ | |
3700 | case readonly: | |
3701 | if (mddev->pers) | |
a05b7ea0 | 3702 | err = md_set_readonly(mddev, NULL); |
9e653b63 N |
3703 | else { |
3704 | mddev->ro = 1; | |
648b629e | 3705 | set_disk_ro(mddev->gendisk, 1); |
9e653b63 N |
3706 | err = do_md_run(mddev); |
3707 | } | |
3708 | break; | |
3709 | case read_auto: | |
9e653b63 | 3710 | if (mddev->pers) { |
80268ee9 | 3711 | if (mddev->ro == 0) |
a05b7ea0 | 3712 | err = md_set_readonly(mddev, NULL); |
80268ee9 | 3713 | else if (mddev->ro == 1) |
648b629e N |
3714 | err = restart_array(mddev); |
3715 | if (err == 0) { | |
3716 | mddev->ro = 2; | |
3717 | set_disk_ro(mddev->gendisk, 0); | |
3718 | } | |
9e653b63 N |
3719 | } else { |
3720 | mddev->ro = 2; | |
3721 | err = do_md_run(mddev); | |
3722 | } | |
3723 | break; | |
3724 | case clean: | |
3725 | if (mddev->pers) { | |
3726 | restart_array(mddev); | |
3727 | spin_lock_irq(&mddev->write_lock); | |
3728 | if (atomic_read(&mddev->writes_pending) == 0) { | |
e691063a N |
3729 | if (mddev->in_sync == 0) { |
3730 | mddev->in_sync = 1; | |
31a59e34 N |
3731 | if (mddev->safemode == 1) |
3732 | mddev->safemode = 0; | |
070dc6dd | 3733 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
e691063a N |
3734 | } |
3735 | err = 0; | |
3736 | } else | |
3737 | err = -EBUSY; | |
9e653b63 | 3738 | spin_unlock_irq(&mddev->write_lock); |
5bf29597 N |
3739 | } else |
3740 | err = -EINVAL; | |
9e653b63 N |
3741 | break; |
3742 | case active: | |
3743 | if (mddev->pers) { | |
3744 | restart_array(mddev); | |
070dc6dd | 3745 | clear_bit(MD_CHANGE_PENDING, &mddev->flags); |
9e653b63 N |
3746 | wake_up(&mddev->sb_wait); |
3747 | err = 0; | |
3748 | } else { | |
3749 | mddev->ro = 0; | |
648b629e | 3750 | set_disk_ro(mddev->gendisk, 0); |
9e653b63 N |
3751 | err = do_md_run(mddev); |
3752 | } | |
3753 | break; | |
3754 | case write_pending: | |
3755 | case active_idle: | |
3756 | /* these cannot be set */ | |
3757 | break; | |
3758 | } | |
3759 | if (err) | |
3760 | return err; | |
0fd62b86 | 3761 | else { |
1d23f178 N |
3762 | if (mddev->hold_active == UNTIL_IOCTL) |
3763 | mddev->hold_active = 0; | |
00bcb4ac | 3764 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
9e653b63 | 3765 | return len; |
0fd62b86 | 3766 | } |
9e653b63 | 3767 | } |
80ca3a44 N |
3768 | static struct md_sysfs_entry md_array_state = |
3769 | __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store); | |
9e653b63 | 3770 | |
1e50915f | 3771 | static ssize_t |
fd01b88c | 3772 | max_corrected_read_errors_show(struct mddev *mddev, char *page) { |
1e50915f RB |
3773 | return sprintf(page, "%d\n", |
3774 | atomic_read(&mddev->max_corr_read_errors)); | |
3775 | } | |
3776 | ||
3777 | static ssize_t | |
fd01b88c | 3778 | max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len) |
1e50915f RB |
3779 | { |
3780 | char *e; | |
3781 | unsigned long n = simple_strtoul(buf, &e, 10); | |
3782 | ||
3783 | if (*buf && (*e == 0 || *e == '\n')) { | |
3784 | atomic_set(&mddev->max_corr_read_errors, n); | |
3785 | return len; | |
3786 | } | |
3787 | return -EINVAL; | |
3788 | } | |
3789 | ||
3790 | static struct md_sysfs_entry max_corr_read_errors = | |
3791 | __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show, | |
3792 | max_corrected_read_errors_store); | |
3793 | ||
6d7ff738 | 3794 | static ssize_t |
fd01b88c | 3795 | null_show(struct mddev *mddev, char *page) |
6d7ff738 N |
3796 | { |
3797 | return -EINVAL; | |
3798 | } | |
3799 | ||
3800 | static ssize_t | |
fd01b88c | 3801 | new_dev_store(struct mddev *mddev, const char *buf, size_t len) |
6d7ff738 N |
3802 | { |
3803 | /* buf must be %d:%d\n? giving major and minor numbers */ | |
3804 | /* The new device is added to the array. | |
3805 | * If the array has a persistent superblock, we read the | |
3806 | * superblock to initialise info and check validity. | |
3807 | * Otherwise, only checking done is that in bind_rdev_to_array, | |
3808 | * which mainly checks size. | |
3809 | */ | |
3810 | char *e; | |
3811 | int major = simple_strtoul(buf, &e, 10); | |
3812 | int minor; | |
3813 | dev_t dev; | |
3cb03002 | 3814 | struct md_rdev *rdev; |
6d7ff738 N |
3815 | int err; |
3816 | ||
3817 | if (!*buf || *e != ':' || !e[1] || e[1] == '\n') | |
3818 | return -EINVAL; | |
3819 | minor = simple_strtoul(e+1, &e, 10); | |
3820 | if (*e && *e != '\n') | |
3821 | return -EINVAL; | |
3822 | dev = MKDEV(major, minor); | |
3823 | if (major != MAJOR(dev) || | |
3824 | minor != MINOR(dev)) | |
3825 | return -EOVERFLOW; | |
3826 | ||
6d7ff738 N |
3827 | if (mddev->persistent) { |
3828 | rdev = md_import_device(dev, mddev->major_version, | |
3829 | mddev->minor_version); | |
3830 | if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) { | |
3cb03002 N |
3831 | struct md_rdev *rdev0 |
3832 | = list_entry(mddev->disks.next, | |
3833 | struct md_rdev, same_set); | |
6d7ff738 N |
3834 | err = super_types[mddev->major_version] |
3835 | .load_super(rdev, rdev0, mddev->minor_version); | |
3836 | if (err < 0) | |
3837 | goto out; | |
3838 | } | |
c5d79adb N |
3839 | } else if (mddev->external) |
3840 | rdev = md_import_device(dev, -2, -1); | |
3841 | else | |
6d7ff738 N |
3842 | rdev = md_import_device(dev, -1, -1); |
3843 | ||
3844 | if (IS_ERR(rdev)) | |
3845 | return PTR_ERR(rdev); | |
3846 | err = bind_rdev_to_array(rdev, mddev); | |
3847 | out: | |
3848 | if (err) | |
3849 | export_rdev(rdev); | |
3850 | return err ? err : len; | |
3851 | } | |
3852 | ||
3853 | static struct md_sysfs_entry md_new_device = | |
80ca3a44 | 3854 | __ATTR(new_dev, S_IWUSR, null_show, new_dev_store); |
3b34380a | 3855 | |
9b1d1dac | 3856 | static ssize_t |
fd01b88c | 3857 | bitmap_store(struct mddev *mddev, const char *buf, size_t len) |
9b1d1dac PC |
3858 | { |
3859 | char *end; | |
3860 | unsigned long chunk, end_chunk; | |
3861 | ||
3862 | if (!mddev->bitmap) | |
3863 | goto out; | |
3864 | /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */ | |
3865 | while (*buf) { | |
3866 | chunk = end_chunk = simple_strtoul(buf, &end, 0); | |
3867 | if (buf == end) break; | |
3868 | if (*end == '-') { /* range */ | |
3869 | buf = end + 1; | |
3870 | end_chunk = simple_strtoul(buf, &end, 0); | |
3871 | if (buf == end) break; | |
3872 | } | |
3873 | if (*end && !isspace(*end)) break; | |
3874 | bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk); | |
e7d2860b | 3875 | buf = skip_spaces(end); |
9b1d1dac PC |
3876 | } |
3877 | bitmap_unplug(mddev->bitmap); /* flush the bits to disk */ | |
3878 | out: | |
3879 | return len; | |
3880 | } | |
3881 | ||
3882 | static struct md_sysfs_entry md_bitmap = | |
3883 | __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store); | |
3884 | ||
a35b0d69 | 3885 | static ssize_t |
fd01b88c | 3886 | size_show(struct mddev *mddev, char *page) |
a35b0d69 | 3887 | { |
58c0fed4 AN |
3888 | return sprintf(page, "%llu\n", |
3889 | (unsigned long long)mddev->dev_sectors / 2); | |
a35b0d69 N |
3890 | } |
3891 | ||
fd01b88c | 3892 | static int update_size(struct mddev *mddev, sector_t num_sectors); |
a35b0d69 N |
3893 | |
3894 | static ssize_t | |
fd01b88c | 3895 | size_store(struct mddev *mddev, const char *buf, size_t len) |
a35b0d69 N |
3896 | { |
3897 | /* If array is inactive, we can reduce the component size, but | |
3898 | * not increase it (except from 0). | |
3899 | * If array is active, we can try an on-line resize | |
3900 | */ | |
b522adcd DW |
3901 | sector_t sectors; |
3902 | int err = strict_blocks_to_sectors(buf, §ors); | |
a35b0d69 | 3903 | |
58c0fed4 AN |
3904 | if (err < 0) |
3905 | return err; | |
a35b0d69 | 3906 | if (mddev->pers) { |
58c0fed4 | 3907 | err = update_size(mddev, sectors); |
850b2b42 | 3908 | md_update_sb(mddev, 1); |
a35b0d69 | 3909 | } else { |
58c0fed4 AN |
3910 | if (mddev->dev_sectors == 0 || |
3911 | mddev->dev_sectors > sectors) | |
3912 | mddev->dev_sectors = sectors; | |
a35b0d69 N |
3913 | else |
3914 | err = -ENOSPC; | |
3915 | } | |
3916 | return err ? err : len; | |
3917 | } | |
3918 | ||
3919 | static struct md_sysfs_entry md_size = | |
80ca3a44 | 3920 | __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store); |
a35b0d69 | 3921 | |
83f0d77a | 3922 | /* Metadata version. |
e691063a N |
3923 | * This is one of |
3924 | * 'none' for arrays with no metadata (good luck...) | |
3925 | * 'external' for arrays with externally managed metadata, | |
8bb93aac N |
3926 | * or N.M for internally known formats |
3927 | */ | |
3928 | static ssize_t | |
fd01b88c | 3929 | metadata_show(struct mddev *mddev, char *page) |
8bb93aac N |
3930 | { |
3931 | if (mddev->persistent) | |
3932 | return sprintf(page, "%d.%d\n", | |
3933 | mddev->major_version, mddev->minor_version); | |
e691063a N |
3934 | else if (mddev->external) |
3935 | return sprintf(page, "external:%s\n", mddev->metadata_type); | |
8bb93aac N |
3936 | else |
3937 | return sprintf(page, "none\n"); | |
3938 | } | |
3939 | ||
3940 | static ssize_t | |
fd01b88c | 3941 | metadata_store(struct mddev *mddev, const char *buf, size_t len) |
8bb93aac N |
3942 | { |
3943 | int major, minor; | |
3944 | char *e; | |
ea43ddd8 N |
3945 | /* Changing the details of 'external' metadata is |
3946 | * always permitted. Otherwise there must be | |
3947 | * no devices attached to the array. | |
3948 | */ | |
3949 | if (mddev->external && strncmp(buf, "external:", 9) == 0) | |
3950 | ; | |
3951 | else if (!list_empty(&mddev->disks)) | |
8bb93aac N |
3952 | return -EBUSY; |
3953 | ||
3954 | if (cmd_match(buf, "none")) { | |
3955 | mddev->persistent = 0; | |
e691063a N |
3956 | mddev->external = 0; |
3957 | mddev->major_version = 0; | |
3958 | mddev->minor_version = 90; | |
3959 | return len; | |
3960 | } | |
3961 | if (strncmp(buf, "external:", 9) == 0) { | |
20a49ff6 | 3962 | size_t namelen = len-9; |
e691063a N |
3963 | if (namelen >= sizeof(mddev->metadata_type)) |
3964 | namelen = sizeof(mddev->metadata_type)-1; | |
3965 | strncpy(mddev->metadata_type, buf+9, namelen); | |
3966 | mddev->metadata_type[namelen] = 0; | |
3967 | if (namelen && mddev->metadata_type[namelen-1] == '\n') | |
3968 | mddev->metadata_type[--namelen] = 0; | |
3969 | mddev->persistent = 0; | |
3970 | mddev->external = 1; | |
8bb93aac N |
3971 | mddev->major_version = 0; |
3972 | mddev->minor_version = 90; | |
3973 | return len; | |
3974 | } | |
3975 | major = simple_strtoul(buf, &e, 10); | |
3976 | if (e==buf || *e != '.') | |
3977 | return -EINVAL; | |
3978 | buf = e+1; | |
3979 | minor = simple_strtoul(buf, &e, 10); | |
3f9d7b0d | 3980 | if (e==buf || (*e && *e != '\n') ) |
8bb93aac | 3981 | return -EINVAL; |
50511da3 | 3982 | if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL) |
8bb93aac N |
3983 | return -ENOENT; |
3984 | mddev->major_version = major; | |
3985 | mddev->minor_version = minor; | |
3986 | mddev->persistent = 1; | |
e691063a | 3987 | mddev->external = 0; |
8bb93aac N |
3988 | return len; |
3989 | } | |
3990 | ||
3991 | static struct md_sysfs_entry md_metadata = | |
80ca3a44 | 3992 | __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
8bb93aac | 3993 | |
24dd469d | 3994 | static ssize_t |
fd01b88c | 3995 | action_show(struct mddev *mddev, char *page) |
24dd469d | 3996 | { |
7eec314d | 3997 | char *type = "idle"; |
b6a9ce68 N |
3998 | if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) |
3999 | type = "frozen"; | |
4000 | else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
2b12ab6d | 4001 | (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) { |
ccfcc3c1 N |
4002 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4003 | type = "reshape"; | |
4004 | else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
24dd469d N |
4005 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
4006 | type = "resync"; | |
4007 | else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) | |
4008 | type = "check"; | |
4009 | else | |
4010 | type = "repair"; | |
72a23c21 | 4011 | } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) |
24dd469d N |
4012 | type = "recover"; |
4013 | } | |
4014 | return sprintf(page, "%s\n", type); | |
4015 | } | |
4016 | ||
4017 | static ssize_t | |
fd01b88c | 4018 | action_store(struct mddev *mddev, const char *page, size_t len) |
24dd469d | 4019 | { |
7eec314d N |
4020 | if (!mddev->pers || !mddev->pers->sync_request) |
4021 | return -EINVAL; | |
4022 | ||
b6a9ce68 N |
4023 | if (cmd_match(page, "frozen")) |
4024 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4025 | else | |
4026 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
4027 | ||
4028 | if (cmd_match(page, "idle") || cmd_match(page, "frozen")) { | |
7eec314d N |
4029 | if (mddev->sync_thread) { |
4030 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
a91d5ac0 | 4031 | md_reap_sync_thread(mddev); |
7eec314d | 4032 | } |
03c902e1 N |
4033 | } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || |
4034 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) | |
24dd469d | 4035 | return -EBUSY; |
72a23c21 NB |
4036 | else if (cmd_match(page, "resync")) |
4037 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4038 | else if (cmd_match(page, "recover")) { | |
4039 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
7eec314d | 4040 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
72a23c21 | 4041 | } else if (cmd_match(page, "reshape")) { |
16484bf5 N |
4042 | int err; |
4043 | if (mddev->pers->start_reshape == NULL) | |
4044 | return -EINVAL; | |
4045 | err = mddev->pers->start_reshape(mddev); | |
4046 | if (err) | |
4047 | return err; | |
a99ac971 | 4048 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
16484bf5 | 4049 | } else { |
bce74dac | 4050 | if (cmd_match(page, "check")) |
7eec314d | 4051 | set_bit(MD_RECOVERY_CHECK, &mddev->recovery); |
2adc7d47 | 4052 | else if (!cmd_match(page, "repair")) |
7eec314d N |
4053 | return -EINVAL; |
4054 | set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
4055 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7eec314d | 4056 | } |
48c26ddc N |
4057 | if (mddev->ro == 2) { |
4058 | /* A write to sync_action is enough to justify | |
4059 | * canceling read-auto mode | |
4060 | */ | |
4061 | mddev->ro = 0; | |
4062 | md_wakeup_thread(mddev->sync_thread); | |
4063 | } | |
03c902e1 | 4064 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
24dd469d | 4065 | md_wakeup_thread(mddev->thread); |
00bcb4ac | 4066 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
24dd469d N |
4067 | return len; |
4068 | } | |
4069 | ||
c4a39551 JB |
4070 | static struct md_sysfs_entry md_scan_mode = |
4071 | __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store); | |
4072 | ||
4073 | static ssize_t | |
4074 | last_sync_action_show(struct mddev *mddev, char *page) | |
4075 | { | |
4076 | return sprintf(page, "%s\n", mddev->last_sync_action); | |
4077 | } | |
4078 | ||
4079 | static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action); | |
4080 | ||
9d88883e | 4081 | static ssize_t |
fd01b88c | 4082 | mismatch_cnt_show(struct mddev *mddev, char *page) |
9d88883e N |
4083 | { |
4084 | return sprintf(page, "%llu\n", | |
7f7583d4 JM |
4085 | (unsigned long long) |
4086 | atomic64_read(&mddev->resync_mismatches)); | |
9d88883e N |
4087 | } |
4088 | ||
80ca3a44 | 4089 | static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt); |
9d88883e | 4090 | |
88202a0c | 4091 | static ssize_t |
fd01b88c | 4092 | sync_min_show(struct mddev *mddev, char *page) |
88202a0c N |
4093 | { |
4094 | return sprintf(page, "%d (%s)\n", speed_min(mddev), | |
4095 | mddev->sync_speed_min ? "local": "system"); | |
4096 | } | |
4097 | ||
4098 | static ssize_t | |
fd01b88c | 4099 | sync_min_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c N |
4100 | { |
4101 | int min; | |
4102 | char *e; | |
4103 | if (strncmp(buf, "system", 6)==0) { | |
4104 | mddev->sync_speed_min = 0; | |
4105 | return len; | |
4106 | } | |
4107 | min = simple_strtoul(buf, &e, 10); | |
4108 | if (buf == e || (*e && *e != '\n') || min <= 0) | |
4109 | return -EINVAL; | |
4110 | mddev->sync_speed_min = min; | |
4111 | return len; | |
4112 | } | |
4113 | ||
4114 | static struct md_sysfs_entry md_sync_min = | |
4115 | __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store); | |
4116 | ||
4117 | static ssize_t | |
fd01b88c | 4118 | sync_max_show(struct mddev *mddev, char *page) |
88202a0c N |
4119 | { |
4120 | return sprintf(page, "%d (%s)\n", speed_max(mddev), | |
4121 | mddev->sync_speed_max ? "local": "system"); | |
4122 | } | |
4123 | ||
4124 | static ssize_t | |
fd01b88c | 4125 | sync_max_store(struct mddev *mddev, const char *buf, size_t len) |
88202a0c N |
4126 | { |
4127 | int max; | |
4128 | char *e; | |
4129 | if (strncmp(buf, "system", 6)==0) { | |
4130 | mddev->sync_speed_max = 0; | |
4131 | return len; | |
4132 | } | |
4133 | max = simple_strtoul(buf, &e, 10); | |
4134 | if (buf == e || (*e && *e != '\n') || max <= 0) | |
4135 | return -EINVAL; | |
4136 | mddev->sync_speed_max = max; | |
4137 | return len; | |
4138 | } | |
4139 | ||
4140 | static struct md_sysfs_entry md_sync_max = | |
4141 | __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store); | |
4142 | ||
d7f3d291 | 4143 | static ssize_t |
fd01b88c | 4144 | degraded_show(struct mddev *mddev, char *page) |
d7f3d291 IP |
4145 | { |
4146 | return sprintf(page, "%d\n", mddev->degraded); | |
4147 | } | |
4148 | static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded); | |
88202a0c | 4149 | |
90b08710 | 4150 | static ssize_t |
fd01b88c | 4151 | sync_force_parallel_show(struct mddev *mddev, char *page) |
90b08710 BS |
4152 | { |
4153 | return sprintf(page, "%d\n", mddev->parallel_resync); | |
4154 | } | |
4155 | ||
4156 | static ssize_t | |
fd01b88c | 4157 | sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len) |
90b08710 BS |
4158 | { |
4159 | long n; | |
4160 | ||
b29bebd6 | 4161 | if (kstrtol(buf, 10, &n)) |
90b08710 BS |
4162 | return -EINVAL; |
4163 | ||
4164 | if (n != 0 && n != 1) | |
4165 | return -EINVAL; | |
4166 | ||
4167 | mddev->parallel_resync = n; | |
4168 | ||
4169 | if (mddev->sync_thread) | |
4170 | wake_up(&resync_wait); | |
4171 | ||
4172 | return len; | |
4173 | } | |
4174 | ||
4175 | /* force parallel resync, even with shared block devices */ | |
4176 | static struct md_sysfs_entry md_sync_force_parallel = | |
4177 | __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR, | |
4178 | sync_force_parallel_show, sync_force_parallel_store); | |
4179 | ||
88202a0c | 4180 | static ssize_t |
fd01b88c | 4181 | sync_speed_show(struct mddev *mddev, char *page) |
88202a0c N |
4182 | { |
4183 | unsigned long resync, dt, db; | |
d1a7c503 N |
4184 | if (mddev->curr_resync == 0) |
4185 | return sprintf(page, "none\n"); | |
9687a60c AN |
4186 | resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active); |
4187 | dt = (jiffies - mddev->resync_mark) / HZ; | |
88202a0c | 4188 | if (!dt) dt++; |
9687a60c AN |
4189 | db = resync - mddev->resync_mark_cnt; |
4190 | return sprintf(page, "%lu\n", db/dt/2); /* K/sec */ | |
88202a0c N |
4191 | } |
4192 | ||
80ca3a44 | 4193 | static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed); |
88202a0c N |
4194 | |
4195 | static ssize_t | |
fd01b88c | 4196 | sync_completed_show(struct mddev *mddev, char *page) |
88202a0c | 4197 | { |
13ae864b | 4198 | unsigned long long max_sectors, resync; |
88202a0c | 4199 | |
acb180b0 N |
4200 | if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4201 | return sprintf(page, "none\n"); | |
4202 | ||
72f36d59 N |
4203 | if (mddev->curr_resync == 1 || |
4204 | mddev->curr_resync == 2) | |
4205 | return sprintf(page, "delayed\n"); | |
4206 | ||
c804cdec N |
4207 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
4208 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
58c0fed4 | 4209 | max_sectors = mddev->resync_max_sectors; |
88202a0c | 4210 | else |
58c0fed4 | 4211 | max_sectors = mddev->dev_sectors; |
88202a0c | 4212 | |
acb180b0 | 4213 | resync = mddev->curr_resync_completed; |
13ae864b | 4214 | return sprintf(page, "%llu / %llu\n", resync, max_sectors); |
88202a0c N |
4215 | } |
4216 | ||
80ca3a44 | 4217 | static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed); |
88202a0c | 4218 | |
5e96ee65 | 4219 | static ssize_t |
fd01b88c | 4220 | min_sync_show(struct mddev *mddev, char *page) |
5e96ee65 NB |
4221 | { |
4222 | return sprintf(page, "%llu\n", | |
4223 | (unsigned long long)mddev->resync_min); | |
4224 | } | |
4225 | static ssize_t | |
fd01b88c | 4226 | min_sync_store(struct mddev *mddev, const char *buf, size_t len) |
5e96ee65 NB |
4227 | { |
4228 | unsigned long long min; | |
b29bebd6 | 4229 | if (kstrtoull(buf, 10, &min)) |
5e96ee65 NB |
4230 | return -EINVAL; |
4231 | if (min > mddev->resync_max) | |
4232 | return -EINVAL; | |
4233 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
4234 | return -EBUSY; | |
4235 | ||
4236 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 4237 | if (mddev->chunk_sectors) { |
2ac06c33 | 4238 | sector_t temp = min; |
9d8f0363 | 4239 | if (sector_div(temp, mddev->chunk_sectors)) |
5e96ee65 NB |
4240 | return -EINVAL; |
4241 | } | |
4242 | mddev->resync_min = min; | |
4243 | ||
4244 | return len; | |
4245 | } | |
4246 | ||
4247 | static struct md_sysfs_entry md_min_sync = | |
4248 | __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store); | |
4249 | ||
c6207277 | 4250 | static ssize_t |
fd01b88c | 4251 | max_sync_show(struct mddev *mddev, char *page) |
c6207277 N |
4252 | { |
4253 | if (mddev->resync_max == MaxSector) | |
4254 | return sprintf(page, "max\n"); | |
4255 | else | |
4256 | return sprintf(page, "%llu\n", | |
4257 | (unsigned long long)mddev->resync_max); | |
4258 | } | |
4259 | static ssize_t | |
fd01b88c | 4260 | max_sync_store(struct mddev *mddev, const char *buf, size_t len) |
c6207277 N |
4261 | { |
4262 | if (strncmp(buf, "max", 3) == 0) | |
4263 | mddev->resync_max = MaxSector; | |
4264 | else { | |
5e96ee65 | 4265 | unsigned long long max; |
b29bebd6 | 4266 | if (kstrtoull(buf, 10, &max)) |
5e96ee65 NB |
4267 | return -EINVAL; |
4268 | if (max < mddev->resync_min) | |
c6207277 N |
4269 | return -EINVAL; |
4270 | if (max < mddev->resync_max && | |
4d484a4a | 4271 | mddev->ro == 0 && |
c6207277 N |
4272 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
4273 | return -EBUSY; | |
4274 | ||
4275 | /* Must be a multiple of chunk_size */ | |
9d8f0363 | 4276 | if (mddev->chunk_sectors) { |
2ac06c33 | 4277 | sector_t temp = max; |
9d8f0363 | 4278 | if (sector_div(temp, mddev->chunk_sectors)) |
c6207277 N |
4279 | return -EINVAL; |
4280 | } | |
4281 | mddev->resync_max = max; | |
4282 | } | |
4283 | wake_up(&mddev->recovery_wait); | |
4284 | return len; | |
4285 | } | |
4286 | ||
4287 | static struct md_sysfs_entry md_max_sync = | |
4288 | __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store); | |
4289 | ||
e464eafd | 4290 | static ssize_t |
fd01b88c | 4291 | suspend_lo_show(struct mddev *mddev, char *page) |
e464eafd N |
4292 | { |
4293 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo); | |
4294 | } | |
4295 | ||
4296 | static ssize_t | |
fd01b88c | 4297 | suspend_lo_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd N |
4298 | { |
4299 | char *e; | |
4300 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4301 | unsigned long long old = mddev->suspend_lo; |
e464eafd | 4302 | |
f72ffdd6 | 4303 | if (mddev->pers == NULL || |
b8d966ef | 4304 | mddev->pers->quiesce == NULL) |
e464eafd N |
4305 | return -EINVAL; |
4306 | if (buf == e || (*e && *e != '\n')) | |
4307 | return -EINVAL; | |
23ddff37 N |
4308 | |
4309 | mddev->suspend_lo = new; | |
4310 | if (new >= old) | |
4311 | /* Shrinking suspended region */ | |
e464eafd | 4312 | mddev->pers->quiesce(mddev, 2); |
23ddff37 N |
4313 | else { |
4314 | /* Expanding suspended region - need to wait */ | |
4315 | mddev->pers->quiesce(mddev, 1); | |
4316 | mddev->pers->quiesce(mddev, 0); | |
4317 | } | |
4318 | return len; | |
e464eafd N |
4319 | } |
4320 | static struct md_sysfs_entry md_suspend_lo = | |
4321 | __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store); | |
4322 | ||
e464eafd | 4323 | static ssize_t |
fd01b88c | 4324 | suspend_hi_show(struct mddev *mddev, char *page) |
e464eafd N |
4325 | { |
4326 | return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi); | |
4327 | } | |
4328 | ||
4329 | static ssize_t | |
fd01b88c | 4330 | suspend_hi_store(struct mddev *mddev, const char *buf, size_t len) |
e464eafd N |
4331 | { |
4332 | char *e; | |
4333 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
23ddff37 | 4334 | unsigned long long old = mddev->suspend_hi; |
e464eafd | 4335 | |
b8d966ef N |
4336 | if (mddev->pers == NULL || |
4337 | mddev->pers->quiesce == NULL) | |
e464eafd N |
4338 | return -EINVAL; |
4339 | if (buf == e || (*e && *e != '\n')) | |
4340 | return -EINVAL; | |
23ddff37 N |
4341 | |
4342 | mddev->suspend_hi = new; | |
4343 | if (new <= old) | |
4344 | /* Shrinking suspended region */ | |
4345 | mddev->pers->quiesce(mddev, 2); | |
4346 | else { | |
4347 | /* Expanding suspended region - need to wait */ | |
e464eafd N |
4348 | mddev->pers->quiesce(mddev, 1); |
4349 | mddev->pers->quiesce(mddev, 0); | |
23ddff37 N |
4350 | } |
4351 | return len; | |
e464eafd N |
4352 | } |
4353 | static struct md_sysfs_entry md_suspend_hi = | |
4354 | __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store); | |
4355 | ||
08a02ecd | 4356 | static ssize_t |
fd01b88c | 4357 | reshape_position_show(struct mddev *mddev, char *page) |
08a02ecd N |
4358 | { |
4359 | if (mddev->reshape_position != MaxSector) | |
4360 | return sprintf(page, "%llu\n", | |
4361 | (unsigned long long)mddev->reshape_position); | |
4362 | strcpy(page, "none\n"); | |
4363 | return 5; | |
4364 | } | |
4365 | ||
4366 | static ssize_t | |
fd01b88c | 4367 | reshape_position_store(struct mddev *mddev, const char *buf, size_t len) |
08a02ecd | 4368 | { |
c6563a8c | 4369 | struct md_rdev *rdev; |
08a02ecd N |
4370 | char *e; |
4371 | unsigned long long new = simple_strtoull(buf, &e, 10); | |
4372 | if (mddev->pers) | |
4373 | return -EBUSY; | |
4374 | if (buf == e || (*e && *e != '\n')) | |
4375 | return -EINVAL; | |
4376 | mddev->reshape_position = new; | |
4377 | mddev->delta_disks = 0; | |
2c810cdd | 4378 | mddev->reshape_backwards = 0; |
08a02ecd N |
4379 | mddev->new_level = mddev->level; |
4380 | mddev->new_layout = mddev->layout; | |
664e7c41 | 4381 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
c6563a8c N |
4382 | rdev_for_each(rdev, mddev) |
4383 | rdev->new_data_offset = rdev->data_offset; | |
08a02ecd N |
4384 | return len; |
4385 | } | |
4386 | ||
4387 | static struct md_sysfs_entry md_reshape_position = | |
4388 | __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show, | |
4389 | reshape_position_store); | |
4390 | ||
2c810cdd N |
4391 | static ssize_t |
4392 | reshape_direction_show(struct mddev *mddev, char *page) | |
4393 | { | |
4394 | return sprintf(page, "%s\n", | |
4395 | mddev->reshape_backwards ? "backwards" : "forwards"); | |
4396 | } | |
4397 | ||
4398 | static ssize_t | |
4399 | reshape_direction_store(struct mddev *mddev, const char *buf, size_t len) | |
4400 | { | |
4401 | int backwards = 0; | |
4402 | if (cmd_match(buf, "forwards")) | |
4403 | backwards = 0; | |
4404 | else if (cmd_match(buf, "backwards")) | |
4405 | backwards = 1; | |
4406 | else | |
4407 | return -EINVAL; | |
4408 | if (mddev->reshape_backwards == backwards) | |
4409 | return len; | |
4410 | ||
4411 | /* check if we are allowed to change */ | |
4412 | if (mddev->delta_disks) | |
4413 | return -EBUSY; | |
4414 | ||
4415 | if (mddev->persistent && | |
4416 | mddev->major_version == 0) | |
4417 | return -EINVAL; | |
4418 | ||
4419 | mddev->reshape_backwards = backwards; | |
4420 | return len; | |
4421 | } | |
4422 | ||
4423 | static struct md_sysfs_entry md_reshape_direction = | |
4424 | __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show, | |
4425 | reshape_direction_store); | |
4426 | ||
b522adcd | 4427 | static ssize_t |
fd01b88c | 4428 | array_size_show(struct mddev *mddev, char *page) |
b522adcd DW |
4429 | { |
4430 | if (mddev->external_size) | |
4431 | return sprintf(page, "%llu\n", | |
4432 | (unsigned long long)mddev->array_sectors/2); | |
4433 | else | |
4434 | return sprintf(page, "default\n"); | |
4435 | } | |
4436 | ||
4437 | static ssize_t | |
fd01b88c | 4438 | array_size_store(struct mddev *mddev, const char *buf, size_t len) |
b522adcd DW |
4439 | { |
4440 | sector_t sectors; | |
4441 | ||
4442 | if (strncmp(buf, "default", 7) == 0) { | |
4443 | if (mddev->pers) | |
4444 | sectors = mddev->pers->size(mddev, 0, 0); | |
4445 | else | |
4446 | sectors = mddev->array_sectors; | |
4447 | ||
4448 | mddev->external_size = 0; | |
4449 | } else { | |
4450 | if (strict_blocks_to_sectors(buf, §ors) < 0) | |
4451 | return -EINVAL; | |
4452 | if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors) | |
2b69c839 | 4453 | return -E2BIG; |
b522adcd DW |
4454 | |
4455 | mddev->external_size = 1; | |
4456 | } | |
4457 | ||
4458 | mddev->array_sectors = sectors; | |
cbe6ef1d N |
4459 | if (mddev->pers) { |
4460 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 4461 | revalidate_disk(mddev->gendisk); |
cbe6ef1d | 4462 | } |
b522adcd DW |
4463 | return len; |
4464 | } | |
4465 | ||
4466 | static struct md_sysfs_entry md_array_size = | |
4467 | __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show, | |
4468 | array_size_store); | |
e464eafd | 4469 | |
eae1701f N |
4470 | static struct attribute *md_default_attrs[] = { |
4471 | &md_level.attr, | |
d4dbd025 | 4472 | &md_layout.attr, |
eae1701f | 4473 | &md_raid_disks.attr, |
3b34380a | 4474 | &md_chunk_size.attr, |
a35b0d69 | 4475 | &md_size.attr, |
a94213b1 | 4476 | &md_resync_start.attr, |
8bb93aac | 4477 | &md_metadata.attr, |
6d7ff738 | 4478 | &md_new_device.attr, |
16f17b39 | 4479 | &md_safe_delay.attr, |
9e653b63 | 4480 | &md_array_state.attr, |
08a02ecd | 4481 | &md_reshape_position.attr, |
2c810cdd | 4482 | &md_reshape_direction.attr, |
b522adcd | 4483 | &md_array_size.attr, |
1e50915f | 4484 | &max_corr_read_errors.attr, |
411036fa N |
4485 | NULL, |
4486 | }; | |
4487 | ||
4488 | static struct attribute *md_redundancy_attrs[] = { | |
24dd469d | 4489 | &md_scan_mode.attr, |
c4a39551 | 4490 | &md_last_scan_mode.attr, |
9d88883e | 4491 | &md_mismatches.attr, |
88202a0c N |
4492 | &md_sync_min.attr, |
4493 | &md_sync_max.attr, | |
4494 | &md_sync_speed.attr, | |
90b08710 | 4495 | &md_sync_force_parallel.attr, |
88202a0c | 4496 | &md_sync_completed.attr, |
5e96ee65 | 4497 | &md_min_sync.attr, |
c6207277 | 4498 | &md_max_sync.attr, |
e464eafd N |
4499 | &md_suspend_lo.attr, |
4500 | &md_suspend_hi.attr, | |
9b1d1dac | 4501 | &md_bitmap.attr, |
d7f3d291 | 4502 | &md_degraded.attr, |
eae1701f N |
4503 | NULL, |
4504 | }; | |
411036fa N |
4505 | static struct attribute_group md_redundancy_group = { |
4506 | .name = NULL, | |
4507 | .attrs = md_redundancy_attrs, | |
4508 | }; | |
4509 | ||
eae1701f N |
4510 | static ssize_t |
4511 | md_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4512 | { | |
4513 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4514 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4515 | ssize_t rv; |
eae1701f N |
4516 | |
4517 | if (!entry->show) | |
4518 | return -EIO; | |
af8a2434 N |
4519 | spin_lock(&all_mddevs_lock); |
4520 | if (list_empty(&mddev->all_mddevs)) { | |
4521 | spin_unlock(&all_mddevs_lock); | |
4522 | return -EBUSY; | |
4523 | } | |
4524 | mddev_get(mddev); | |
4525 | spin_unlock(&all_mddevs_lock); | |
4526 | ||
5dc5cf7d IM |
4527 | rv = mddev_lock(mddev); |
4528 | if (!rv) { | |
4529 | rv = entry->show(mddev, page); | |
4530 | mddev_unlock(mddev); | |
4531 | } | |
af8a2434 | 4532 | mddev_put(mddev); |
96de1e66 | 4533 | return rv; |
eae1701f N |
4534 | } |
4535 | ||
4536 | static ssize_t | |
4537 | md_attr_store(struct kobject *kobj, struct attribute *attr, | |
4538 | const char *page, size_t length) | |
4539 | { | |
4540 | struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr); | |
fd01b88c | 4541 | struct mddev *mddev = container_of(kobj, struct mddev, kobj); |
96de1e66 | 4542 | ssize_t rv; |
eae1701f N |
4543 | |
4544 | if (!entry->store) | |
4545 | return -EIO; | |
67463acb N |
4546 | if (!capable(CAP_SYS_ADMIN)) |
4547 | return -EACCES; | |
af8a2434 N |
4548 | spin_lock(&all_mddevs_lock); |
4549 | if (list_empty(&mddev->all_mddevs)) { | |
4550 | spin_unlock(&all_mddevs_lock); | |
4551 | return -EBUSY; | |
4552 | } | |
4553 | mddev_get(mddev); | |
4554 | spin_unlock(&all_mddevs_lock); | |
a7a3f08d N |
4555 | if (entry->store == new_dev_store) |
4556 | flush_workqueue(md_misc_wq); | |
5dc5cf7d IM |
4557 | rv = mddev_lock(mddev); |
4558 | if (!rv) { | |
4559 | rv = entry->store(mddev, page, length); | |
4560 | mddev_unlock(mddev); | |
4561 | } | |
af8a2434 | 4562 | mddev_put(mddev); |
96de1e66 | 4563 | return rv; |
eae1701f N |
4564 | } |
4565 | ||
4566 | static void md_free(struct kobject *ko) | |
4567 | { | |
fd01b88c | 4568 | struct mddev *mddev = container_of(ko, struct mddev, kobj); |
a21d1504 N |
4569 | |
4570 | if (mddev->sysfs_state) | |
4571 | sysfs_put(mddev->sysfs_state); | |
4572 | ||
4573 | if (mddev->gendisk) { | |
4574 | del_gendisk(mddev->gendisk); | |
4575 | put_disk(mddev->gendisk); | |
4576 | } | |
4577 | if (mddev->queue) | |
4578 | blk_cleanup_queue(mddev->queue); | |
4579 | ||
eae1701f N |
4580 | kfree(mddev); |
4581 | } | |
4582 | ||
52cf25d0 | 4583 | static const struct sysfs_ops md_sysfs_ops = { |
eae1701f N |
4584 | .show = md_attr_show, |
4585 | .store = md_attr_store, | |
4586 | }; | |
4587 | static struct kobj_type md_ktype = { | |
4588 | .release = md_free, | |
4589 | .sysfs_ops = &md_sysfs_ops, | |
4590 | .default_attrs = md_default_attrs, | |
4591 | }; | |
4592 | ||
1da177e4 LT |
4593 | int mdp_major = 0; |
4594 | ||
5fd3a17e DW |
4595 | static void mddev_delayed_delete(struct work_struct *ws) |
4596 | { | |
fd01b88c | 4597 | struct mddev *mddev = container_of(ws, struct mddev, del_work); |
5fd3a17e | 4598 | |
43a70507 | 4599 | sysfs_remove_group(&mddev->kobj, &md_bitmap_group); |
5fd3a17e DW |
4600 | kobject_del(&mddev->kobj); |
4601 | kobject_put(&mddev->kobj); | |
4602 | } | |
4603 | ||
efeb53c0 | 4604 | static int md_alloc(dev_t dev, char *name) |
1da177e4 | 4605 | { |
48c9c27b | 4606 | static DEFINE_MUTEX(disks_mutex); |
fd01b88c | 4607 | struct mddev *mddev = mddev_find(dev); |
1da177e4 | 4608 | struct gendisk *disk; |
efeb53c0 N |
4609 | int partitioned; |
4610 | int shift; | |
4611 | int unit; | |
3830c62f | 4612 | int error; |
1da177e4 LT |
4613 | |
4614 | if (!mddev) | |
efeb53c0 N |
4615 | return -ENODEV; |
4616 | ||
4617 | partitioned = (MAJOR(mddev->unit) != MD_MAJOR); | |
4618 | shift = partitioned ? MdpMinorShift : 0; | |
4619 | unit = MINOR(mddev->unit) >> shift; | |
1da177e4 | 4620 | |
e804ac78 TH |
4621 | /* wait for any previous instance of this device to be |
4622 | * completely removed (mddev_delayed_delete). | |
d3374825 | 4623 | */ |
e804ac78 | 4624 | flush_workqueue(md_misc_wq); |
d3374825 | 4625 | |
48c9c27b | 4626 | mutex_lock(&disks_mutex); |
0909dc44 N |
4627 | error = -EEXIST; |
4628 | if (mddev->gendisk) | |
4629 | goto abort; | |
efeb53c0 N |
4630 | |
4631 | if (name) { | |
4632 | /* Need to ensure that 'name' is not a duplicate. | |
4633 | */ | |
fd01b88c | 4634 | struct mddev *mddev2; |
efeb53c0 N |
4635 | spin_lock(&all_mddevs_lock); |
4636 | ||
4637 | list_for_each_entry(mddev2, &all_mddevs, all_mddevs) | |
4638 | if (mddev2->gendisk && | |
4639 | strcmp(mddev2->gendisk->disk_name, name) == 0) { | |
4640 | spin_unlock(&all_mddevs_lock); | |
0909dc44 | 4641 | goto abort; |
efeb53c0 N |
4642 | } |
4643 | spin_unlock(&all_mddevs_lock); | |
1da177e4 | 4644 | } |
8b765398 | 4645 | |
0909dc44 | 4646 | error = -ENOMEM; |
8b765398 | 4647 | mddev->queue = blk_alloc_queue(GFP_KERNEL); |
0909dc44 N |
4648 | if (!mddev->queue) |
4649 | goto abort; | |
409c57f3 N |
4650 | mddev->queue->queuedata = mddev; |
4651 | ||
409c57f3 | 4652 | blk_queue_make_request(mddev->queue, md_make_request); |
b1bd055d | 4653 | blk_set_stacking_limits(&mddev->queue->limits); |
8b765398 | 4654 | |
1da177e4 LT |
4655 | disk = alloc_disk(1 << shift); |
4656 | if (!disk) { | |
8b765398 N |
4657 | blk_cleanup_queue(mddev->queue); |
4658 | mddev->queue = NULL; | |
0909dc44 | 4659 | goto abort; |
1da177e4 | 4660 | } |
efeb53c0 | 4661 | disk->major = MAJOR(mddev->unit); |
1da177e4 | 4662 | disk->first_minor = unit << shift; |
efeb53c0 N |
4663 | if (name) |
4664 | strcpy(disk->disk_name, name); | |
4665 | else if (partitioned) | |
1da177e4 | 4666 | sprintf(disk->disk_name, "md_d%d", unit); |
ce7b0f46 | 4667 | else |
1da177e4 | 4668 | sprintf(disk->disk_name, "md%d", unit); |
1da177e4 LT |
4669 | disk->fops = &md_fops; |
4670 | disk->private_data = mddev; | |
4671 | disk->queue = mddev->queue; | |
b0140891 | 4672 | blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA); |
92850bbd | 4673 | /* Allow extended partitions. This makes the |
d3374825 | 4674 | * 'mdp' device redundant, but we can't really |
92850bbd N |
4675 | * remove it now. |
4676 | */ | |
4677 | disk->flags |= GENHD_FL_EXT_DEVT; | |
1da177e4 | 4678 | mddev->gendisk = disk; |
b0140891 N |
4679 | /* As soon as we call add_disk(), another thread could get |
4680 | * through to md_open, so make sure it doesn't get too far | |
4681 | */ | |
4682 | mutex_lock(&mddev->open_mutex); | |
4683 | add_disk(disk); | |
4684 | ||
ed9e1982 TH |
4685 | error = kobject_init_and_add(&mddev->kobj, &md_ktype, |
4686 | &disk_to_dev(disk)->kobj, "%s", "md"); | |
0909dc44 N |
4687 | if (error) { |
4688 | /* This isn't possible, but as kobject_init_and_add is marked | |
4689 | * __must_check, we must do something with the result | |
4690 | */ | |
5e55e2f5 N |
4691 | printk(KERN_WARNING "md: cannot register %s/md - name in use\n", |
4692 | disk->disk_name); | |
0909dc44 N |
4693 | error = 0; |
4694 | } | |
00bcb4ac N |
4695 | if (mddev->kobj.sd && |
4696 | sysfs_create_group(&mddev->kobj, &md_bitmap_group)) | |
43a70507 | 4697 | printk(KERN_DEBUG "pointless warning\n"); |
b0140891 | 4698 | mutex_unlock(&mddev->open_mutex); |
0909dc44 N |
4699 | abort: |
4700 | mutex_unlock(&disks_mutex); | |
00bcb4ac | 4701 | if (!error && mddev->kobj.sd) { |
3830c62f | 4702 | kobject_uevent(&mddev->kobj, KOBJ_ADD); |
00bcb4ac | 4703 | mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state"); |
b62b7590 | 4704 | } |
d3374825 | 4705 | mddev_put(mddev); |
0909dc44 | 4706 | return error; |
efeb53c0 N |
4707 | } |
4708 | ||
4709 | static struct kobject *md_probe(dev_t dev, int *part, void *data) | |
4710 | { | |
4711 | md_alloc(dev, NULL); | |
1da177e4 LT |
4712 | return NULL; |
4713 | } | |
4714 | ||
efeb53c0 N |
4715 | static int add_named_array(const char *val, struct kernel_param *kp) |
4716 | { | |
4717 | /* val must be "md_*" where * is not all digits. | |
4718 | * We allocate an array with a large free minor number, and | |
4719 | * set the name to val. val must not already be an active name. | |
4720 | */ | |
4721 | int len = strlen(val); | |
4722 | char buf[DISK_NAME_LEN]; | |
4723 | ||
4724 | while (len && val[len-1] == '\n') | |
4725 | len--; | |
4726 | if (len >= DISK_NAME_LEN) | |
4727 | return -E2BIG; | |
4728 | strlcpy(buf, val, len+1); | |
4729 | if (strncmp(buf, "md_", 3) != 0) | |
4730 | return -EINVAL; | |
4731 | return md_alloc(0, buf); | |
4732 | } | |
4733 | ||
1da177e4 LT |
4734 | static void md_safemode_timeout(unsigned long data) |
4735 | { | |
fd01b88c | 4736 | struct mddev *mddev = (struct mddev *) data; |
1da177e4 | 4737 | |
0fd62b86 NB |
4738 | if (!atomic_read(&mddev->writes_pending)) { |
4739 | mddev->safemode = 1; | |
4740 | if (mddev->external) | |
00bcb4ac | 4741 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 4742 | } |
1da177e4 LT |
4743 | md_wakeup_thread(mddev->thread); |
4744 | } | |
4745 | ||
6ff8d8ec | 4746 | static int start_dirty_degraded; |
1da177e4 | 4747 | |
fd01b88c | 4748 | int md_run(struct mddev *mddev) |
1da177e4 | 4749 | { |
2604b703 | 4750 | int err; |
3cb03002 | 4751 | struct md_rdev *rdev; |
84fc4b56 | 4752 | struct md_personality *pers; |
1da177e4 | 4753 | |
a757e64c N |
4754 | if (list_empty(&mddev->disks)) |
4755 | /* cannot run an array with no devices.. */ | |
1da177e4 | 4756 | return -EINVAL; |
1da177e4 LT |
4757 | |
4758 | if (mddev->pers) | |
4759 | return -EBUSY; | |
bb4f1e9d N |
4760 | /* Cannot run until previous stop completes properly */ |
4761 | if (mddev->sysfs_active) | |
4762 | return -EBUSY; | |
b6eb127d | 4763 | |
1da177e4 LT |
4764 | /* |
4765 | * Analyze all RAID superblock(s) | |
4766 | */ | |
1ec4a939 N |
4767 | if (!mddev->raid_disks) { |
4768 | if (!mddev->persistent) | |
4769 | return -EINVAL; | |
a757e64c | 4770 | analyze_sbs(mddev); |
1ec4a939 | 4771 | } |
1da177e4 | 4772 | |
d9d166c2 N |
4773 | if (mddev->level != LEVEL_NONE) |
4774 | request_module("md-level-%d", mddev->level); | |
4775 | else if (mddev->clevel[0]) | |
4776 | request_module("md-%s", mddev->clevel); | |
1da177e4 LT |
4777 | |
4778 | /* | |
4779 | * Drop all container device buffers, from now on | |
4780 | * the only valid external interface is through the md | |
4781 | * device. | |
1da177e4 | 4782 | */ |
dafb20fa | 4783 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 4784 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
4785 | continue; |
4786 | sync_blockdev(rdev->bdev); | |
f98393a6 | 4787 | invalidate_bdev(rdev->bdev); |
f0d76d70 N |
4788 | |
4789 | /* perform some consistency tests on the device. | |
4790 | * We don't want the data to overlap the metadata, | |
58c0fed4 | 4791 | * Internal Bitmap issues have been handled elsewhere. |
f0d76d70 | 4792 | */ |
a6ff7e08 JB |
4793 | if (rdev->meta_bdev) { |
4794 | /* Nothing to check */; | |
4795 | } else if (rdev->data_offset < rdev->sb_start) { | |
58c0fed4 AN |
4796 | if (mddev->dev_sectors && |
4797 | rdev->data_offset + mddev->dev_sectors | |
0f420358 | 4798 | > rdev->sb_start) { |
f0d76d70 N |
4799 | printk("md: %s: data overlaps metadata\n", |
4800 | mdname(mddev)); | |
4801 | return -EINVAL; | |
4802 | } | |
4803 | } else { | |
0f420358 | 4804 | if (rdev->sb_start + rdev->sb_size/512 |
f0d76d70 N |
4805 | > rdev->data_offset) { |
4806 | printk("md: %s: metadata overlaps data\n", | |
4807 | mdname(mddev)); | |
4808 | return -EINVAL; | |
4809 | } | |
4810 | } | |
00bcb4ac | 4811 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
4812 | } |
4813 | ||
a167f663 | 4814 | if (mddev->bio_set == NULL) |
395c72a7 | 4815 | mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0); |
a167f663 | 4816 | |
1da177e4 | 4817 | spin_lock(&pers_lock); |
d9d166c2 | 4818 | pers = find_pers(mddev->level, mddev->clevel); |
2604b703 | 4819 | if (!pers || !try_module_get(pers->owner)) { |
1da177e4 | 4820 | spin_unlock(&pers_lock); |
d9d166c2 N |
4821 | if (mddev->level != LEVEL_NONE) |
4822 | printk(KERN_WARNING "md: personality for level %d is not loaded!\n", | |
4823 | mddev->level); | |
4824 | else | |
4825 | printk(KERN_WARNING "md: personality for level %s is not loaded!\n", | |
4826 | mddev->clevel); | |
1da177e4 LT |
4827 | return -EINVAL; |
4828 | } | |
2604b703 | 4829 | mddev->pers = pers; |
1da177e4 | 4830 | spin_unlock(&pers_lock); |
34817e8c N |
4831 | if (mddev->level != pers->level) { |
4832 | mddev->level = pers->level; | |
4833 | mddev->new_level = pers->level; | |
4834 | } | |
d9d166c2 | 4835 | strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel)); |
1da177e4 | 4836 | |
f6705578 | 4837 | if (mddev->reshape_position != MaxSector && |
63c70c4f | 4838 | pers->start_reshape == NULL) { |
f6705578 N |
4839 | /* This personality cannot handle reshaping... */ |
4840 | mddev->pers = NULL; | |
4841 | module_put(pers->owner); | |
4842 | return -EINVAL; | |
4843 | } | |
4844 | ||
7dd5e7c3 N |
4845 | if (pers->sync_request) { |
4846 | /* Warn if this is a potentially silly | |
4847 | * configuration. | |
4848 | */ | |
4849 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 4850 | struct md_rdev *rdev2; |
7dd5e7c3 | 4851 | int warned = 0; |
159ec1fc | 4852 | |
dafb20fa N |
4853 | rdev_for_each(rdev, mddev) |
4854 | rdev_for_each(rdev2, mddev) { | |
7dd5e7c3 N |
4855 | if (rdev < rdev2 && |
4856 | rdev->bdev->bd_contains == | |
4857 | rdev2->bdev->bd_contains) { | |
4858 | printk(KERN_WARNING | |
4859 | "%s: WARNING: %s appears to be" | |
4860 | " on the same physical disk as" | |
4861 | " %s.\n", | |
4862 | mdname(mddev), | |
4863 | bdevname(rdev->bdev,b), | |
4864 | bdevname(rdev2->bdev,b2)); | |
4865 | warned = 1; | |
4866 | } | |
4867 | } | |
159ec1fc | 4868 | |
7dd5e7c3 N |
4869 | if (warned) |
4870 | printk(KERN_WARNING | |
4871 | "True protection against single-disk" | |
4872 | " failure might be compromised.\n"); | |
4873 | } | |
4874 | ||
657390d2 | 4875 | mddev->recovery = 0; |
58c0fed4 AN |
4876 | /* may be over-ridden by personality */ |
4877 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4878 | ||
6ff8d8ec | 4879 | mddev->ok_start_degraded = start_dirty_degraded; |
1da177e4 | 4880 | |
0f9552b5 | 4881 | if (start_readonly && mddev->ro == 0) |
f91de92e N |
4882 | mddev->ro = 2; /* read-only, but switch on first write */ |
4883 | ||
b15c2e57 | 4884 | err = mddev->pers->run(mddev); |
13e53df3 AN |
4885 | if (err) |
4886 | printk(KERN_ERR "md: pers->run() failed ...\n"); | |
b522adcd DW |
4887 | else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) { |
4888 | WARN_ONCE(!mddev->external_size, "%s: default size too small," | |
4889 | " but 'external_size' not in effect?\n", __func__); | |
4890 | printk(KERN_ERR | |
4891 | "md: invalid array_size %llu > default size %llu\n", | |
4892 | (unsigned long long)mddev->array_sectors / 2, | |
4893 | (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2); | |
4894 | err = -EINVAL; | |
4895 | mddev->pers->stop(mddev); | |
4896 | } | |
ef99bf48 N |
4897 | if (err == 0 && mddev->pers->sync_request && |
4898 | (mddev->bitmap_info.file || mddev->bitmap_info.offset)) { | |
b15c2e57 N |
4899 | err = bitmap_create(mddev); |
4900 | if (err) { | |
4901 | printk(KERN_ERR "%s: failed to create bitmap (%d)\n", | |
4902 | mdname(mddev), err); | |
4903 | mddev->pers->stop(mddev); | |
4904 | } | |
4905 | } | |
1da177e4 | 4906 | if (err) { |
1da177e4 LT |
4907 | module_put(mddev->pers->owner); |
4908 | mddev->pers = NULL; | |
32a7627c N |
4909 | bitmap_destroy(mddev); |
4910 | return err; | |
1da177e4 | 4911 | } |
5e55e2f5 | 4912 | if (mddev->pers->sync_request) { |
00bcb4ac N |
4913 | if (mddev->kobj.sd && |
4914 | sysfs_create_group(&mddev->kobj, &md_redundancy_group)) | |
5e55e2f5 N |
4915 | printk(KERN_WARNING |
4916 | "md: cannot register extra attributes for %s\n", | |
4917 | mdname(mddev)); | |
00bcb4ac | 4918 | mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action"); |
5e55e2f5 | 4919 | } else if (mddev->ro == 2) /* auto-readonly not meaningful */ |
fd9d49ca N |
4920 | mddev->ro = 0; |
4921 | ||
f72ffdd6 | 4922 | atomic_set(&mddev->writes_pending,0); |
1e50915f RB |
4923 | atomic_set(&mddev->max_corr_read_errors, |
4924 | MD_DEFAULT_MAX_CORRECTED_READ_ERRORS); | |
1da177e4 LT |
4925 | mddev->safemode = 0; |
4926 | mddev->safemode_timer.function = md_safemode_timeout; | |
4927 | mddev->safemode_timer.data = (unsigned long) mddev; | |
16f17b39 | 4928 | mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */ |
1da177e4 | 4929 | mddev->in_sync = 1; |
0ca69886 N |
4930 | smp_wmb(); |
4931 | mddev->ready = 1; | |
dafb20fa | 4932 | rdev_for_each(rdev, mddev) |
36fad858 NK |
4933 | if (rdev->raid_disk >= 0) |
4934 | if (sysfs_link_rdev(mddev, rdev)) | |
00bcb4ac | 4935 | /* failure here is OK */; |
f72ffdd6 | 4936 | |
1da177e4 | 4937 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f72ffdd6 | 4938 | |
7a0a5355 | 4939 | if (mddev->flags & MD_UPDATE_SB_FLAGS) |
850b2b42 | 4940 | md_update_sb(mddev, 0); |
1da177e4 | 4941 | |
d7603b7e | 4942 | md_new_event(mddev); |
00bcb4ac N |
4943 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
4944 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
a99ac971 | 4945 | sysfs_notify(&mddev->kobj, NULL, "degraded"); |
1da177e4 LT |
4946 | return 0; |
4947 | } | |
390ee602 | 4948 | EXPORT_SYMBOL_GPL(md_run); |
1da177e4 | 4949 | |
fd01b88c | 4950 | static int do_md_run(struct mddev *mddev) |
fe60b014 N |
4951 | { |
4952 | int err; | |
4953 | ||
4954 | err = md_run(mddev); | |
4955 | if (err) | |
4956 | goto out; | |
69e51b44 N |
4957 | err = bitmap_load(mddev); |
4958 | if (err) { | |
4959 | bitmap_destroy(mddev); | |
4960 | goto out; | |
4961 | } | |
0fd018af JB |
4962 | |
4963 | md_wakeup_thread(mddev->thread); | |
4964 | md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */ | |
4965 | ||
fe60b014 N |
4966 | set_capacity(mddev->gendisk, mddev->array_sectors); |
4967 | revalidate_disk(mddev->gendisk); | |
f0b4f7e2 | 4968 | mddev->changed = 1; |
fe60b014 N |
4969 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
4970 | out: | |
4971 | return err; | |
4972 | } | |
4973 | ||
fd01b88c | 4974 | static int restart_array(struct mddev *mddev) |
1da177e4 LT |
4975 | { |
4976 | struct gendisk *disk = mddev->gendisk; | |
1da177e4 | 4977 | |
80fab1d7 | 4978 | /* Complain if it has no devices */ |
1da177e4 | 4979 | if (list_empty(&mddev->disks)) |
80fab1d7 AN |
4980 | return -ENXIO; |
4981 | if (!mddev->pers) | |
4982 | return -EINVAL; | |
4983 | if (!mddev->ro) | |
4984 | return -EBUSY; | |
4985 | mddev->safemode = 0; | |
4986 | mddev->ro = 0; | |
4987 | set_disk_ro(disk, 0); | |
4988 | printk(KERN_INFO "md: %s switched to read-write mode.\n", | |
4989 | mdname(mddev)); | |
4990 | /* Kick recovery or resync if necessary */ | |
4991 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4992 | md_wakeup_thread(mddev->thread); | |
4993 | md_wakeup_thread(mddev->sync_thread); | |
00bcb4ac | 4994 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
80fab1d7 | 4995 | return 0; |
1da177e4 LT |
4996 | } |
4997 | ||
fd01b88c | 4998 | static void md_clean(struct mddev *mddev) |
6177b472 N |
4999 | { |
5000 | mddev->array_sectors = 0; | |
5001 | mddev->external_size = 0; | |
5002 | mddev->dev_sectors = 0; | |
5003 | mddev->raid_disks = 0; | |
5004 | mddev->recovery_cp = 0; | |
5005 | mddev->resync_min = 0; | |
5006 | mddev->resync_max = MaxSector; | |
5007 | mddev->reshape_position = MaxSector; | |
5008 | mddev->external = 0; | |
5009 | mddev->persistent = 0; | |
5010 | mddev->level = LEVEL_NONE; | |
5011 | mddev->clevel[0] = 0; | |
5012 | mddev->flags = 0; | |
5013 | mddev->ro = 0; | |
5014 | mddev->metadata_type[0] = 0; | |
5015 | mddev->chunk_sectors = 0; | |
5016 | mddev->ctime = mddev->utime = 0; | |
5017 | mddev->layout = 0; | |
5018 | mddev->max_disks = 0; | |
5019 | mddev->events = 0; | |
a8707c08 | 5020 | mddev->can_decrease_events = 0; |
6177b472 | 5021 | mddev->delta_disks = 0; |
2c810cdd | 5022 | mddev->reshape_backwards = 0; |
6177b472 N |
5023 | mddev->new_level = LEVEL_NONE; |
5024 | mddev->new_layout = 0; | |
5025 | mddev->new_chunk_sectors = 0; | |
5026 | mddev->curr_resync = 0; | |
7f7583d4 | 5027 | atomic64_set(&mddev->resync_mismatches, 0); |
6177b472 N |
5028 | mddev->suspend_lo = mddev->suspend_hi = 0; |
5029 | mddev->sync_speed_min = mddev->sync_speed_max = 0; | |
5030 | mddev->recovery = 0; | |
5031 | mddev->in_sync = 0; | |
f0b4f7e2 | 5032 | mddev->changed = 0; |
6177b472 | 5033 | mddev->degraded = 0; |
6177b472 | 5034 | mddev->safemode = 0; |
050b6615 | 5035 | mddev->merge_check_needed = 0; |
6177b472 N |
5036 | mddev->bitmap_info.offset = 0; |
5037 | mddev->bitmap_info.default_offset = 0; | |
6409bb05 | 5038 | mddev->bitmap_info.default_space = 0; |
6177b472 N |
5039 | mddev->bitmap_info.chunksize = 0; |
5040 | mddev->bitmap_info.daemon_sleep = 0; | |
5041 | mddev->bitmap_info.max_write_behind = 0; | |
5042 | } | |
5043 | ||
fd01b88c | 5044 | static void __md_stop_writes(struct mddev *mddev) |
a047e125 | 5045 | { |
6b6204ee | 5046 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
a047e125 | 5047 | if (mddev->sync_thread) { |
a047e125 | 5048 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
a91d5ac0 | 5049 | md_reap_sync_thread(mddev); |
a047e125 N |
5050 | } |
5051 | ||
5052 | del_timer_sync(&mddev->safemode_timer); | |
5053 | ||
5054 | bitmap_flush(mddev); | |
5055 | md_super_wait(mddev); | |
5056 | ||
b6d428c6 | 5057 | if (mddev->ro == 0 && |
7a0a5355 | 5058 | (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) { |
a047e125 N |
5059 | /* mark array as shutdown cleanly */ |
5060 | mddev->in_sync = 1; | |
5061 | md_update_sb(mddev, 1); | |
5062 | } | |
5063 | } | |
defad61a | 5064 | |
fd01b88c | 5065 | void md_stop_writes(struct mddev *mddev) |
defad61a | 5066 | { |
29f097c4 | 5067 | mddev_lock_nointr(mddev); |
defad61a N |
5068 | __md_stop_writes(mddev); |
5069 | mddev_unlock(mddev); | |
5070 | } | |
390ee602 | 5071 | EXPORT_SYMBOL_GPL(md_stop_writes); |
a047e125 | 5072 | |
5eff3c43 | 5073 | static void __md_stop(struct mddev *mddev) |
6177b472 | 5074 | { |
0ca69886 | 5075 | mddev->ready = 0; |
6177b472 N |
5076 | mddev->pers->stop(mddev); |
5077 | if (mddev->pers->sync_request && mddev->to_remove == NULL) | |
5078 | mddev->to_remove = &md_redundancy_group; | |
5079 | module_put(mddev->pers->owner); | |
5080 | mddev->pers = NULL; | |
cca9cf90 | 5081 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); |
6177b472 | 5082 | } |
5eff3c43 N |
5083 | |
5084 | void md_stop(struct mddev *mddev) | |
5085 | { | |
5086 | /* stop the array and free an attached data structures. | |
5087 | * This is called from dm-raid | |
5088 | */ | |
5089 | __md_stop(mddev); | |
5090 | bitmap_destroy(mddev); | |
5091 | if (mddev->bio_set) | |
5092 | bioset_free(mddev->bio_set); | |
5093 | } | |
5094 | ||
390ee602 | 5095 | EXPORT_SYMBOL_GPL(md_stop); |
6177b472 | 5096 | |
a05b7ea0 | 5097 | static int md_set_readonly(struct mddev *mddev, struct block_device *bdev) |
a4bd82d0 N |
5098 | { |
5099 | int err = 0; | |
30b8feb7 N |
5100 | int did_freeze = 0; |
5101 | ||
5102 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5103 | did_freeze = 1; | |
5104 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5105 | md_wakeup_thread(mddev->thread); | |
5106 | } | |
5107 | if (mddev->sync_thread) { | |
5108 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5109 | /* Thread might be blocked waiting for metadata update | |
5110 | * which will now never happen */ | |
5111 | wake_up_process(mddev->sync_thread->tsk); | |
5112 | } | |
5113 | mddev_unlock(mddev); | |
5114 | wait_event(resync_wait, mddev->sync_thread == NULL); | |
5115 | mddev_lock_nointr(mddev); | |
5116 | ||
a4bd82d0 | 5117 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5118 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5119 | mddev->sync_thread || |
5120 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { | |
a4bd82d0 | 5121 | printk("md: %s still in use.\n",mdname(mddev)); |
30b8feb7 N |
5122 | if (did_freeze) { |
5123 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5124 | md_wakeup_thread(mddev->thread); | |
5125 | } | |
a4bd82d0 N |
5126 | err = -EBUSY; |
5127 | goto out; | |
5128 | } | |
5129 | if (mddev->pers) { | |
defad61a | 5130 | __md_stop_writes(mddev); |
a4bd82d0 N |
5131 | |
5132 | err = -ENXIO; | |
5133 | if (mddev->ro==1) | |
5134 | goto out; | |
5135 | mddev->ro = 1; | |
5136 | set_disk_ro(mddev->gendisk, 1); | |
5137 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
00bcb4ac | 5138 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
30b8feb7 | 5139 | err = 0; |
a4bd82d0 N |
5140 | } |
5141 | out: | |
5142 | mutex_unlock(&mddev->open_mutex); | |
5143 | return err; | |
5144 | } | |
5145 | ||
9e653b63 N |
5146 | /* mode: |
5147 | * 0 - completely stop and dis-assemble array | |
9e653b63 N |
5148 | * 2 - stop but do not disassemble array |
5149 | */ | |
f72ffdd6 | 5150 | static int do_md_stop(struct mddev *mddev, int mode, |
a05b7ea0 | 5151 | struct block_device *bdev) |
1da177e4 | 5152 | { |
1da177e4 | 5153 | struct gendisk *disk = mddev->gendisk; |
3cb03002 | 5154 | struct md_rdev *rdev; |
30b8feb7 N |
5155 | int did_freeze = 0; |
5156 | ||
5157 | if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) { | |
5158 | did_freeze = 1; | |
5159 | set_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5160 | md_wakeup_thread(mddev->thread); | |
5161 | } | |
5162 | if (mddev->sync_thread) { | |
5163 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5164 | /* Thread might be blocked waiting for metadata update | |
5165 | * which will now never happen */ | |
5166 | wake_up_process(mddev->sync_thread->tsk); | |
5167 | } | |
5168 | mddev_unlock(mddev); | |
5169 | wait_event(resync_wait, mddev->sync_thread == NULL); | |
5170 | mddev_lock_nointr(mddev); | |
1da177e4 | 5171 | |
c8c00a69 | 5172 | mutex_lock(&mddev->open_mutex); |
9ba3b7f5 | 5173 | if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) || |
30b8feb7 N |
5174 | mddev->sysfs_active || |
5175 | mddev->sync_thread || | |
5176 | (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) { | |
df5b20cf | 5177 | printk("md: %s still in use.\n",mdname(mddev)); |
6e17b027 | 5178 | mutex_unlock(&mddev->open_mutex); |
30b8feb7 N |
5179 | if (did_freeze) { |
5180 | clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery); | |
5181 | md_wakeup_thread(mddev->thread); | |
5182 | } | |
260fa034 N |
5183 | return -EBUSY; |
5184 | } | |
6e17b027 | 5185 | if (mddev->pers) { |
a4bd82d0 N |
5186 | if (mddev->ro) |
5187 | set_disk_ro(disk, 0); | |
409c57f3 | 5188 | |
defad61a | 5189 | __md_stop_writes(mddev); |
5eff3c43 | 5190 | __md_stop(mddev); |
a4bd82d0 | 5191 | mddev->queue->merge_bvec_fn = NULL; |
a4bd82d0 | 5192 | mddev->queue->backing_dev_info.congested_fn = NULL; |
6177b472 | 5193 | |
a4bd82d0 | 5194 | /* tell userspace to handle 'inactive' */ |
00bcb4ac | 5195 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0d4ca600 | 5196 | |
dafb20fa | 5197 | rdev_for_each(rdev, mddev) |
36fad858 NK |
5198 | if (rdev->raid_disk >= 0) |
5199 | sysfs_unlink_rdev(mddev, rdev); | |
c4647292 | 5200 | |
a4bd82d0 | 5201 | set_capacity(disk, 0); |
6e17b027 | 5202 | mutex_unlock(&mddev->open_mutex); |
f0b4f7e2 | 5203 | mddev->changed = 1; |
a4bd82d0 | 5204 | revalidate_disk(disk); |
0d4ca600 | 5205 | |
a4bd82d0 N |
5206 | if (mddev->ro) |
5207 | mddev->ro = 0; | |
6e17b027 N |
5208 | } else |
5209 | mutex_unlock(&mddev->open_mutex); | |
1da177e4 LT |
5210 | /* |
5211 | * Free resources if final stop | |
5212 | */ | |
9e653b63 | 5213 | if (mode == 0) { |
1da177e4 LT |
5214 | printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); |
5215 | ||
978f946b | 5216 | bitmap_destroy(mddev); |
c3d9714e | 5217 | if (mddev->bitmap_info.file) { |
c3d9714e N |
5218 | fput(mddev->bitmap_info.file); |
5219 | mddev->bitmap_info.file = NULL; | |
978f946b | 5220 | } |
c3d9714e | 5221 | mddev->bitmap_info.offset = 0; |
978f946b | 5222 | |
1da177e4 LT |
5223 | export_array(mddev); |
5224 | ||
6177b472 | 5225 | md_clean(mddev); |
934d9c23 | 5226 | kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE); |
efeb53c0 N |
5227 | if (mddev->hold_active == UNTIL_STOP) |
5228 | mddev->hold_active = 0; | |
a4bd82d0 | 5229 | } |
3f9d99c1 | 5230 | blk_integrity_unregister(disk); |
d7603b7e | 5231 | md_new_event(mddev); |
00bcb4ac | 5232 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
6e17b027 | 5233 | return 0; |
1da177e4 LT |
5234 | } |
5235 | ||
fdee8ae4 | 5236 | #ifndef MODULE |
fd01b88c | 5237 | static void autorun_array(struct mddev *mddev) |
1da177e4 | 5238 | { |
3cb03002 | 5239 | struct md_rdev *rdev; |
1da177e4 LT |
5240 | int err; |
5241 | ||
a757e64c | 5242 | if (list_empty(&mddev->disks)) |
1da177e4 | 5243 | return; |
1da177e4 LT |
5244 | |
5245 | printk(KERN_INFO "md: running: "); | |
5246 | ||
dafb20fa | 5247 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
5248 | char b[BDEVNAME_SIZE]; |
5249 | printk("<%s>", bdevname(rdev->bdev,b)); | |
5250 | } | |
5251 | printk("\n"); | |
5252 | ||
d710e138 | 5253 | err = do_md_run(mddev); |
1da177e4 LT |
5254 | if (err) { |
5255 | printk(KERN_WARNING "md: do_md_run() returned %d\n", err); | |
a05b7ea0 | 5256 | do_md_stop(mddev, 0, NULL); |
1da177e4 LT |
5257 | } |
5258 | } | |
5259 | ||
5260 | /* | |
5261 | * lets try to run arrays based on all disks that have arrived | |
5262 | * until now. (those are in pending_raid_disks) | |
5263 | * | |
5264 | * the method: pick the first pending disk, collect all disks with | |
5265 | * the same UUID, remove all from the pending list and put them into | |
5266 | * the 'same_array' list. Then order this list based on superblock | |
5267 | * update time (freshest comes first), kick out 'old' disks and | |
5268 | * compare superblocks. If everything's fine then run it. | |
5269 | * | |
5270 | * If "unit" is allocated, then bump its reference count | |
5271 | */ | |
5272 | static void autorun_devices(int part) | |
5273 | { | |
3cb03002 | 5274 | struct md_rdev *rdev0, *rdev, *tmp; |
fd01b88c | 5275 | struct mddev *mddev; |
1da177e4 LT |
5276 | char b[BDEVNAME_SIZE]; |
5277 | ||
5278 | printk(KERN_INFO "md: autorun ...\n"); | |
5279 | while (!list_empty(&pending_raid_disks)) { | |
e8703fe1 | 5280 | int unit; |
1da177e4 | 5281 | dev_t dev; |
ad01c9e3 | 5282 | LIST_HEAD(candidates); |
1da177e4 | 5283 | rdev0 = list_entry(pending_raid_disks.next, |
3cb03002 | 5284 | struct md_rdev, same_set); |
1da177e4 LT |
5285 | |
5286 | printk(KERN_INFO "md: considering %s ...\n", | |
5287 | bdevname(rdev0->bdev,b)); | |
5288 | INIT_LIST_HEAD(&candidates); | |
159ec1fc | 5289 | rdev_for_each_list(rdev, tmp, &pending_raid_disks) |
1da177e4 LT |
5290 | if (super_90_load(rdev, rdev0, 0) >= 0) { |
5291 | printk(KERN_INFO "md: adding %s ...\n", | |
5292 | bdevname(rdev->bdev,b)); | |
5293 | list_move(&rdev->same_set, &candidates); | |
5294 | } | |
5295 | /* | |
5296 | * now we have a set of devices, with all of them having | |
5297 | * mostly sane superblocks. It's time to allocate the | |
5298 | * mddev. | |
5299 | */ | |
e8703fe1 N |
5300 | if (part) { |
5301 | dev = MKDEV(mdp_major, | |
5302 | rdev0->preferred_minor << MdpMinorShift); | |
5303 | unit = MINOR(dev) >> MdpMinorShift; | |
5304 | } else { | |
5305 | dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); | |
5306 | unit = MINOR(dev); | |
5307 | } | |
5308 | if (rdev0->preferred_minor != unit) { | |
1da177e4 LT |
5309 | printk(KERN_INFO "md: unit number in %s is bad: %d\n", |
5310 | bdevname(rdev0->bdev, b), rdev0->preferred_minor); | |
5311 | break; | |
5312 | } | |
1da177e4 LT |
5313 | |
5314 | md_probe(dev, NULL, NULL); | |
5315 | mddev = mddev_find(dev); | |
9bbbca3a NB |
5316 | if (!mddev || !mddev->gendisk) { |
5317 | if (mddev) | |
5318 | mddev_put(mddev); | |
5319 | printk(KERN_ERR | |
1da177e4 LT |
5320 | "md: cannot allocate memory for md drive.\n"); |
5321 | break; | |
5322 | } | |
f72ffdd6 | 5323 | if (mddev_lock(mddev)) |
1da177e4 LT |
5324 | printk(KERN_WARNING "md: %s locked, cannot run\n", |
5325 | mdname(mddev)); | |
5326 | else if (mddev->raid_disks || mddev->major_version | |
5327 | || !list_empty(&mddev->disks)) { | |
f72ffdd6 | 5328 | printk(KERN_WARNING |
1da177e4 LT |
5329 | "md: %s already running, cannot run %s\n", |
5330 | mdname(mddev), bdevname(rdev0->bdev,b)); | |
5331 | mddev_unlock(mddev); | |
5332 | } else { | |
5333 | printk(KERN_INFO "md: created %s\n", mdname(mddev)); | |
1ec4a939 | 5334 | mddev->persistent = 1; |
159ec1fc | 5335 | rdev_for_each_list(rdev, tmp, &candidates) { |
1da177e4 LT |
5336 | list_del_init(&rdev->same_set); |
5337 | if (bind_rdev_to_array(rdev, mddev)) | |
5338 | export_rdev(rdev); | |
5339 | } | |
5340 | autorun_array(mddev); | |
5341 | mddev_unlock(mddev); | |
5342 | } | |
5343 | /* on success, candidates will be empty, on error | |
5344 | * it won't... | |
5345 | */ | |
159ec1fc | 5346 | rdev_for_each_list(rdev, tmp, &candidates) { |
4b80991c | 5347 | list_del_init(&rdev->same_set); |
1da177e4 | 5348 | export_rdev(rdev); |
4b80991c | 5349 | } |
1da177e4 LT |
5350 | mddev_put(mddev); |
5351 | } | |
5352 | printk(KERN_INFO "md: ... autorun DONE.\n"); | |
5353 | } | |
fdee8ae4 | 5354 | #endif /* !MODULE */ |
1da177e4 | 5355 | |
f72ffdd6 | 5356 | static int get_version(void __user *arg) |
1da177e4 LT |
5357 | { |
5358 | mdu_version_t ver; | |
5359 | ||
5360 | ver.major = MD_MAJOR_VERSION; | |
5361 | ver.minor = MD_MINOR_VERSION; | |
5362 | ver.patchlevel = MD_PATCHLEVEL_VERSION; | |
5363 | ||
5364 | if (copy_to_user(arg, &ver, sizeof(ver))) | |
5365 | return -EFAULT; | |
5366 | ||
5367 | return 0; | |
5368 | } | |
5369 | ||
f72ffdd6 | 5370 | static int get_array_info(struct mddev *mddev, void __user *arg) |
1da177e4 LT |
5371 | { |
5372 | mdu_array_info_t info; | |
a9f326eb | 5373 | int nr,working,insync,failed,spare; |
3cb03002 | 5374 | struct md_rdev *rdev; |
1da177e4 | 5375 | |
1ca69c4b N |
5376 | nr = working = insync = failed = spare = 0; |
5377 | rcu_read_lock(); | |
5378 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 5379 | nr++; |
b2d444d7 | 5380 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
5381 | failed++; |
5382 | else { | |
5383 | working++; | |
b2d444d7 | 5384 | if (test_bit(In_sync, &rdev->flags)) |
f72ffdd6 | 5385 | insync++; |
1da177e4 LT |
5386 | else |
5387 | spare++; | |
5388 | } | |
5389 | } | |
1ca69c4b | 5390 | rcu_read_unlock(); |
1da177e4 LT |
5391 | |
5392 | info.major_version = mddev->major_version; | |
5393 | info.minor_version = mddev->minor_version; | |
5394 | info.patch_version = MD_PATCHLEVEL_VERSION; | |
5395 | info.ctime = mddev->ctime; | |
5396 | info.level = mddev->level; | |
58c0fed4 AN |
5397 | info.size = mddev->dev_sectors / 2; |
5398 | if (info.size != mddev->dev_sectors / 2) /* overflow */ | |
284ae7ca | 5399 | info.size = -1; |
1da177e4 LT |
5400 | info.nr_disks = nr; |
5401 | info.raid_disks = mddev->raid_disks; | |
5402 | info.md_minor = mddev->md_minor; | |
5403 | info.not_persistent= !mddev->persistent; | |
5404 | ||
5405 | info.utime = mddev->utime; | |
5406 | info.state = 0; | |
5407 | if (mddev->in_sync) | |
5408 | info.state = (1<<MD_SB_CLEAN); | |
c3d9714e | 5409 | if (mddev->bitmap && mddev->bitmap_info.offset) |
9bd35920 | 5410 | info.state |= (1<<MD_SB_BITMAP_PRESENT); |
a9f326eb | 5411 | info.active_disks = insync; |
1da177e4 LT |
5412 | info.working_disks = working; |
5413 | info.failed_disks = failed; | |
5414 | info.spare_disks = spare; | |
5415 | ||
5416 | info.layout = mddev->layout; | |
9d8f0363 | 5417 | info.chunk_size = mddev->chunk_sectors << 9; |
1da177e4 LT |
5418 | |
5419 | if (copy_to_user(arg, &info, sizeof(info))) | |
5420 | return -EFAULT; | |
5421 | ||
5422 | return 0; | |
5423 | } | |
5424 | ||
f72ffdd6 | 5425 | static int get_bitmap_file(struct mddev *mddev, void __user * arg) |
32a7627c N |
5426 | { |
5427 | mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ | |
5428 | char *ptr, *buf = NULL; | |
5429 | int err = -ENOMEM; | |
5430 | ||
60559da4 | 5431 | file = kmalloc(sizeof(*file), GFP_NOIO); |
2a2275d6 | 5432 | |
32a7627c N |
5433 | if (!file) |
5434 | goto out; | |
5435 | ||
5436 | /* bitmap disabled, zero the first byte and copy out */ | |
1ec885cd | 5437 | if (!mddev->bitmap || !mddev->bitmap->storage.file) { |
32a7627c N |
5438 | file->pathname[0] = '\0'; |
5439 | goto copy_out; | |
5440 | } | |
5441 | ||
5442 | buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); | |
5443 | if (!buf) | |
5444 | goto out; | |
5445 | ||
1ec885cd N |
5446 | ptr = d_path(&mddev->bitmap->storage.file->f_path, |
5447 | buf, sizeof(file->pathname)); | |
6bcfd601 | 5448 | if (IS_ERR(ptr)) |
32a7627c N |
5449 | goto out; |
5450 | ||
5451 | strcpy(file->pathname, ptr); | |
5452 | ||
5453 | copy_out: | |
5454 | err = 0; | |
5455 | if (copy_to_user(arg, file, sizeof(*file))) | |
5456 | err = -EFAULT; | |
5457 | out: | |
5458 | kfree(buf); | |
5459 | kfree(file); | |
5460 | return err; | |
5461 | } | |
5462 | ||
f72ffdd6 | 5463 | static int get_disk_info(struct mddev *mddev, void __user * arg) |
1da177e4 LT |
5464 | { |
5465 | mdu_disk_info_t info; | |
3cb03002 | 5466 | struct md_rdev *rdev; |
1da177e4 LT |
5467 | |
5468 | if (copy_from_user(&info, arg, sizeof(info))) | |
5469 | return -EFAULT; | |
5470 | ||
1ca69c4b N |
5471 | rcu_read_lock(); |
5472 | rdev = find_rdev_nr_rcu(mddev, info.number); | |
1da177e4 LT |
5473 | if (rdev) { |
5474 | info.major = MAJOR(rdev->bdev->bd_dev); | |
5475 | info.minor = MINOR(rdev->bdev->bd_dev); | |
5476 | info.raid_disk = rdev->raid_disk; | |
5477 | info.state = 0; | |
b2d444d7 | 5478 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 5479 | info.state |= (1<<MD_DISK_FAULTY); |
b2d444d7 | 5480 | else if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
5481 | info.state |= (1<<MD_DISK_ACTIVE); |
5482 | info.state |= (1<<MD_DISK_SYNC); | |
5483 | } | |
8ddf9efe N |
5484 | if (test_bit(WriteMostly, &rdev->flags)) |
5485 | info.state |= (1<<MD_DISK_WRITEMOSTLY); | |
1da177e4 LT |
5486 | } else { |
5487 | info.major = info.minor = 0; | |
5488 | info.raid_disk = -1; | |
5489 | info.state = (1<<MD_DISK_REMOVED); | |
5490 | } | |
1ca69c4b | 5491 | rcu_read_unlock(); |
1da177e4 LT |
5492 | |
5493 | if (copy_to_user(arg, &info, sizeof(info))) | |
5494 | return -EFAULT; | |
5495 | ||
5496 | return 0; | |
5497 | } | |
5498 | ||
f72ffdd6 | 5499 | static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info) |
1da177e4 LT |
5500 | { |
5501 | char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; | |
3cb03002 | 5502 | struct md_rdev *rdev; |
1da177e4 LT |
5503 | dev_t dev = MKDEV(info->major,info->minor); |
5504 | ||
5505 | if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) | |
5506 | return -EOVERFLOW; | |
5507 | ||
5508 | if (!mddev->raid_disks) { | |
5509 | int err; | |
5510 | /* expecting a device which has a superblock */ | |
5511 | rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); | |
5512 | if (IS_ERR(rdev)) { | |
f72ffdd6 | 5513 | printk(KERN_WARNING |
1da177e4 LT |
5514 | "md: md_import_device returned %ld\n", |
5515 | PTR_ERR(rdev)); | |
5516 | return PTR_ERR(rdev); | |
5517 | } | |
5518 | if (!list_empty(&mddev->disks)) { | |
3cb03002 N |
5519 | struct md_rdev *rdev0 |
5520 | = list_entry(mddev->disks.next, | |
5521 | struct md_rdev, same_set); | |
a9f326eb | 5522 | err = super_types[mddev->major_version] |
1da177e4 LT |
5523 | .load_super(rdev, rdev0, mddev->minor_version); |
5524 | if (err < 0) { | |
f72ffdd6 | 5525 | printk(KERN_WARNING |
1da177e4 | 5526 | "md: %s has different UUID to %s\n", |
f72ffdd6 | 5527 | bdevname(rdev->bdev,b), |
1da177e4 LT |
5528 | bdevname(rdev0->bdev,b2)); |
5529 | export_rdev(rdev); | |
5530 | return -EINVAL; | |
5531 | } | |
5532 | } | |
5533 | err = bind_rdev_to_array(rdev, mddev); | |
5534 | if (err) | |
5535 | export_rdev(rdev); | |
5536 | return err; | |
5537 | } | |
5538 | ||
5539 | /* | |
5540 | * add_new_disk can be used once the array is assembled | |
5541 | * to add "hot spares". They must already have a superblock | |
5542 | * written | |
5543 | */ | |
5544 | if (mddev->pers) { | |
5545 | int err; | |
5546 | if (!mddev->pers->hot_add_disk) { | |
f72ffdd6 | 5547 | printk(KERN_WARNING |
1da177e4 LT |
5548 | "%s: personality does not support diskops!\n", |
5549 | mdname(mddev)); | |
5550 | return -EINVAL; | |
5551 | } | |
7b1e35f6 N |
5552 | if (mddev->persistent) |
5553 | rdev = md_import_device(dev, mddev->major_version, | |
5554 | mddev->minor_version); | |
5555 | else | |
5556 | rdev = md_import_device(dev, -1, -1); | |
1da177e4 | 5557 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5558 | printk(KERN_WARNING |
1da177e4 LT |
5559 | "md: md_import_device returned %ld\n", |
5560 | PTR_ERR(rdev)); | |
5561 | return PTR_ERR(rdev); | |
5562 | } | |
1a855a06 | 5563 | /* set saved_raid_disk if appropriate */ |
41158c7e N |
5564 | if (!mddev->persistent) { |
5565 | if (info->state & (1<<MD_DISK_SYNC) && | |
bf572541 | 5566 | info->raid_disk < mddev->raid_disks) { |
41158c7e | 5567 | rdev->raid_disk = info->raid_disk; |
bf572541 | 5568 | set_bit(In_sync, &rdev->flags); |
8313b8e5 | 5569 | clear_bit(Bitmap_sync, &rdev->flags); |
bf572541 | 5570 | } else |
41158c7e | 5571 | rdev->raid_disk = -1; |
f466722c | 5572 | rdev->saved_raid_disk = rdev->raid_disk; |
41158c7e N |
5573 | } else |
5574 | super_types[mddev->major_version]. | |
5575 | validate_super(mddev, rdev); | |
bedd86b7 | 5576 | if ((info->state & (1<<MD_DISK_SYNC)) && |
f4563091 | 5577 | rdev->raid_disk != info->raid_disk) { |
bedd86b7 N |
5578 | /* This was a hot-add request, but events doesn't |
5579 | * match, so reject it. | |
5580 | */ | |
5581 | export_rdev(rdev); | |
5582 | return -EINVAL; | |
5583 | } | |
5584 | ||
b2d444d7 | 5585 | clear_bit(In_sync, &rdev->flags); /* just to be sure */ |
8ddf9efe N |
5586 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5587 | set_bit(WriteMostly, &rdev->flags); | |
575a80fa N |
5588 | else |
5589 | clear_bit(WriteMostly, &rdev->flags); | |
8ddf9efe | 5590 | |
1da177e4 LT |
5591 | rdev->raid_disk = -1; |
5592 | err = bind_rdev_to_array(rdev, mddev); | |
7c7546cc N |
5593 | if (!err && !mddev->pers->hot_remove_disk) { |
5594 | /* If there is hot_add_disk but no hot_remove_disk | |
5595 | * then added disks for geometry changes, | |
5596 | * and should be added immediately. | |
5597 | */ | |
5598 | super_types[mddev->major_version]. | |
5599 | validate_super(mddev, rdev); | |
5600 | err = mddev->pers->hot_add_disk(mddev, rdev); | |
5601 | if (err) | |
5602 | unbind_rdev_from_array(rdev); | |
5603 | } | |
1da177e4 LT |
5604 | if (err) |
5605 | export_rdev(rdev); | |
52664732 | 5606 | else |
00bcb4ac | 5607 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
c361777f | 5608 | |
7ceb17e8 | 5609 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
72a23c21 NB |
5610 | if (mddev->degraded) |
5611 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
c361777f | 5612 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
9864c005 | 5613 | if (!err) |
5614 | md_new_event(mddev); | |
005eca5e | 5615 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5616 | return err; |
5617 | } | |
5618 | ||
5619 | /* otherwise, add_new_disk is only allowed | |
5620 | * for major_version==0 superblocks | |
5621 | */ | |
5622 | if (mddev->major_version != 0) { | |
5623 | printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", | |
5624 | mdname(mddev)); | |
5625 | return -EINVAL; | |
5626 | } | |
5627 | ||
5628 | if (!(info->state & (1<<MD_DISK_FAULTY))) { | |
5629 | int err; | |
d710e138 | 5630 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 5631 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5632 | printk(KERN_WARNING |
1da177e4 LT |
5633 | "md: error, md_import_device() returned %ld\n", |
5634 | PTR_ERR(rdev)); | |
5635 | return PTR_ERR(rdev); | |
5636 | } | |
5637 | rdev->desc_nr = info->number; | |
5638 | if (info->raid_disk < mddev->raid_disks) | |
5639 | rdev->raid_disk = info->raid_disk; | |
5640 | else | |
5641 | rdev->raid_disk = -1; | |
5642 | ||
1da177e4 | 5643 | if (rdev->raid_disk < mddev->raid_disks) |
b2d444d7 N |
5644 | if (info->state & (1<<MD_DISK_SYNC)) |
5645 | set_bit(In_sync, &rdev->flags); | |
1da177e4 | 5646 | |
8ddf9efe N |
5647 | if (info->state & (1<<MD_DISK_WRITEMOSTLY)) |
5648 | set_bit(WriteMostly, &rdev->flags); | |
5649 | ||
1da177e4 LT |
5650 | if (!mddev->persistent) { |
5651 | printk(KERN_INFO "md: nonpersistent superblock ...\n"); | |
77304d2a MS |
5652 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
5653 | } else | |
57b2caa3 | 5654 | rdev->sb_start = calc_dev_sboffset(rdev); |
8190e754 | 5655 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5656 | |
2bf071bf N |
5657 | err = bind_rdev_to_array(rdev, mddev); |
5658 | if (err) { | |
5659 | export_rdev(rdev); | |
5660 | return err; | |
5661 | } | |
1da177e4 LT |
5662 | } |
5663 | ||
5664 | return 0; | |
5665 | } | |
5666 | ||
f72ffdd6 | 5667 | static int hot_remove_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
5668 | { |
5669 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 5670 | struct md_rdev *rdev; |
1da177e4 | 5671 | |
1da177e4 LT |
5672 | rdev = find_rdev(mddev, dev); |
5673 | if (!rdev) | |
5674 | return -ENXIO; | |
5675 | ||
3ea8929d N |
5676 | clear_bit(Blocked, &rdev->flags); |
5677 | remove_and_add_spares(mddev, rdev); | |
5678 | ||
1da177e4 LT |
5679 | if (rdev->raid_disk >= 0) |
5680 | goto busy; | |
5681 | ||
5682 | kick_rdev_from_array(rdev); | |
850b2b42 | 5683 | md_update_sb(mddev, 1); |
d7603b7e | 5684 | md_new_event(mddev); |
1da177e4 LT |
5685 | |
5686 | return 0; | |
5687 | busy: | |
fdefa4d8 | 5688 | printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n", |
1da177e4 LT |
5689 | bdevname(rdev->bdev,b), mdname(mddev)); |
5690 | return -EBUSY; | |
5691 | } | |
5692 | ||
f72ffdd6 | 5693 | static int hot_add_disk(struct mddev *mddev, dev_t dev) |
1da177e4 LT |
5694 | { |
5695 | char b[BDEVNAME_SIZE]; | |
5696 | int err; | |
3cb03002 | 5697 | struct md_rdev *rdev; |
1da177e4 LT |
5698 | |
5699 | if (!mddev->pers) | |
5700 | return -ENODEV; | |
5701 | ||
5702 | if (mddev->major_version != 0) { | |
5703 | printk(KERN_WARNING "%s: HOT_ADD may only be used with" | |
5704 | " version-0 superblocks.\n", | |
5705 | mdname(mddev)); | |
5706 | return -EINVAL; | |
5707 | } | |
5708 | if (!mddev->pers->hot_add_disk) { | |
f72ffdd6 | 5709 | printk(KERN_WARNING |
1da177e4 LT |
5710 | "%s: personality does not support diskops!\n", |
5711 | mdname(mddev)); | |
5712 | return -EINVAL; | |
5713 | } | |
5714 | ||
d710e138 | 5715 | rdev = md_import_device(dev, -1, 0); |
1da177e4 | 5716 | if (IS_ERR(rdev)) { |
f72ffdd6 | 5717 | printk(KERN_WARNING |
1da177e4 LT |
5718 | "md: error, md_import_device() returned %ld\n", |
5719 | PTR_ERR(rdev)); | |
5720 | return -EINVAL; | |
5721 | } | |
5722 | ||
5723 | if (mddev->persistent) | |
57b2caa3 | 5724 | rdev->sb_start = calc_dev_sboffset(rdev); |
1da177e4 | 5725 | else |
77304d2a | 5726 | rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512; |
1da177e4 | 5727 | |
8190e754 | 5728 | rdev->sectors = rdev->sb_start; |
1da177e4 | 5729 | |
b2d444d7 | 5730 | if (test_bit(Faulty, &rdev->flags)) { |
f72ffdd6 | 5731 | printk(KERN_WARNING |
1da177e4 LT |
5732 | "md: can not hot-add faulty %s disk to %s!\n", |
5733 | bdevname(rdev->bdev,b), mdname(mddev)); | |
5734 | err = -EINVAL; | |
5735 | goto abort_export; | |
5736 | } | |
b2d444d7 | 5737 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5738 | rdev->desc_nr = -1; |
5842730d | 5739 | rdev->saved_raid_disk = -1; |
2bf071bf N |
5740 | err = bind_rdev_to_array(rdev, mddev); |
5741 | if (err) | |
5742 | goto abort_export; | |
1da177e4 LT |
5743 | |
5744 | /* | |
5745 | * The rest should better be atomic, we can have disk failures | |
5746 | * noticed in interrupt contexts ... | |
5747 | */ | |
5748 | ||
1da177e4 LT |
5749 | rdev->raid_disk = -1; |
5750 | ||
850b2b42 | 5751 | md_update_sb(mddev, 1); |
1da177e4 LT |
5752 | |
5753 | /* | |
5754 | * Kick recovery, maybe this spare has to be added to the | |
5755 | * array immediately. | |
5756 | */ | |
5757 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5758 | md_wakeup_thread(mddev->thread); | |
d7603b7e | 5759 | md_new_event(mddev); |
1da177e4 LT |
5760 | return 0; |
5761 | ||
1da177e4 LT |
5762 | abort_export: |
5763 | export_rdev(rdev); | |
5764 | return err; | |
5765 | } | |
5766 | ||
fd01b88c | 5767 | static int set_bitmap_file(struct mddev *mddev, int fd) |
32a7627c | 5768 | { |
035328c2 | 5769 | int err = 0; |
32a7627c | 5770 | |
36fa3063 | 5771 | if (mddev->pers) { |
d66b1b39 | 5772 | if (!mddev->pers->quiesce || !mddev->thread) |
36fa3063 N |
5773 | return -EBUSY; |
5774 | if (mddev->recovery || mddev->sync_thread) | |
5775 | return -EBUSY; | |
5776 | /* we should be able to change the bitmap.. */ | |
5777 | } | |
32a7627c | 5778 | |
36fa3063 | 5779 | if (fd >= 0) { |
035328c2 | 5780 | struct inode *inode; |
36fa3063 N |
5781 | if (mddev->bitmap) |
5782 | return -EEXIST; /* cannot add when bitmap is present */ | |
c3d9714e | 5783 | mddev->bitmap_info.file = fget(fd); |
32a7627c | 5784 | |
c3d9714e | 5785 | if (mddev->bitmap_info.file == NULL) { |
36fa3063 N |
5786 | printk(KERN_ERR "%s: error: failed to get bitmap file\n", |
5787 | mdname(mddev)); | |
5788 | return -EBADF; | |
5789 | } | |
5790 | ||
035328c2 N |
5791 | inode = mddev->bitmap_info.file->f_mapping->host; |
5792 | if (!S_ISREG(inode->i_mode)) { | |
5793 | printk(KERN_ERR "%s: error: bitmap file must be a regular file\n", | |
5794 | mdname(mddev)); | |
5795 | err = -EBADF; | |
5796 | } else if (!(mddev->bitmap_info.file->f_mode & FMODE_WRITE)) { | |
5797 | printk(KERN_ERR "%s: error: bitmap file must open for write\n", | |
5798 | mdname(mddev)); | |
5799 | err = -EBADF; | |
5800 | } else if (atomic_read(&inode->i_writecount) != 1) { | |
36fa3063 N |
5801 | printk(KERN_ERR "%s: error: bitmap file is already in use\n", |
5802 | mdname(mddev)); | |
035328c2 N |
5803 | err = -EBUSY; |
5804 | } | |
5805 | if (err) { | |
c3d9714e N |
5806 | fput(mddev->bitmap_info.file); |
5807 | mddev->bitmap_info.file = NULL; | |
36fa3063 N |
5808 | return err; |
5809 | } | |
c3d9714e | 5810 | mddev->bitmap_info.offset = 0; /* file overrides offset */ |
36fa3063 N |
5811 | } else if (mddev->bitmap == NULL) |
5812 | return -ENOENT; /* cannot remove what isn't there */ | |
5813 | err = 0; | |
5814 | if (mddev->pers) { | |
5815 | mddev->pers->quiesce(mddev, 1); | |
69e51b44 | 5816 | if (fd >= 0) { |
36fa3063 | 5817 | err = bitmap_create(mddev); |
69e51b44 N |
5818 | if (!err) |
5819 | err = bitmap_load(mddev); | |
5820 | } | |
d7375ab3 | 5821 | if (fd < 0 || err) { |
36fa3063 | 5822 | bitmap_destroy(mddev); |
d7375ab3 N |
5823 | fd = -1; /* make sure to put the file */ |
5824 | } | |
36fa3063 | 5825 | mddev->pers->quiesce(mddev, 0); |
d7375ab3 N |
5826 | } |
5827 | if (fd < 0) { | |
035328c2 | 5828 | if (mddev->bitmap_info.file) |
c3d9714e | 5829 | fput(mddev->bitmap_info.file); |
c3d9714e | 5830 | mddev->bitmap_info.file = NULL; |
36fa3063 N |
5831 | } |
5832 | ||
32a7627c N |
5833 | return err; |
5834 | } | |
5835 | ||
1da177e4 LT |
5836 | /* |
5837 | * set_array_info is used two different ways | |
5838 | * The original usage is when creating a new array. | |
5839 | * In this usage, raid_disks is > 0 and it together with | |
5840 | * level, size, not_persistent,layout,chunksize determine the | |
5841 | * shape of the array. | |
5842 | * This will always create an array with a type-0.90.0 superblock. | |
5843 | * The newer usage is when assembling an array. | |
5844 | * In this case raid_disks will be 0, and the major_version field is | |
5845 | * use to determine which style super-blocks are to be found on the devices. | |
5846 | * The minor and patch _version numbers are also kept incase the | |
5847 | * super_block handler wishes to interpret them. | |
5848 | */ | |
f72ffdd6 | 5849 | static int set_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
5850 | { |
5851 | ||
5852 | if (info->raid_disks == 0) { | |
5853 | /* just setting version number for superblock loading */ | |
5854 | if (info->major_version < 0 || | |
50511da3 | 5855 | info->major_version >= ARRAY_SIZE(super_types) || |
1da177e4 LT |
5856 | super_types[info->major_version].name == NULL) { |
5857 | /* maybe try to auto-load a module? */ | |
f72ffdd6 | 5858 | printk(KERN_INFO |
1da177e4 LT |
5859 | "md: superblock version %d not known\n", |
5860 | info->major_version); | |
5861 | return -EINVAL; | |
5862 | } | |
5863 | mddev->major_version = info->major_version; | |
5864 | mddev->minor_version = info->minor_version; | |
5865 | mddev->patch_version = info->patch_version; | |
3f9d7b0d | 5866 | mddev->persistent = !info->not_persistent; |
cbd19983 N |
5867 | /* ensure mddev_put doesn't delete this now that there |
5868 | * is some minimal configuration. | |
5869 | */ | |
5870 | mddev->ctime = get_seconds(); | |
1da177e4 LT |
5871 | return 0; |
5872 | } | |
5873 | mddev->major_version = MD_MAJOR_VERSION; | |
5874 | mddev->minor_version = MD_MINOR_VERSION; | |
5875 | mddev->patch_version = MD_PATCHLEVEL_VERSION; | |
5876 | mddev->ctime = get_seconds(); | |
5877 | ||
5878 | mddev->level = info->level; | |
17115e03 | 5879 | mddev->clevel[0] = 0; |
58c0fed4 | 5880 | mddev->dev_sectors = 2 * (sector_t)info->size; |
1da177e4 LT |
5881 | mddev->raid_disks = info->raid_disks; |
5882 | /* don't set md_minor, it is determined by which /dev/md* was | |
5883 | * openned | |
5884 | */ | |
5885 | if (info->state & (1<<MD_SB_CLEAN)) | |
5886 | mddev->recovery_cp = MaxSector; | |
5887 | else | |
5888 | mddev->recovery_cp = 0; | |
5889 | mddev->persistent = ! info->not_persistent; | |
e691063a | 5890 | mddev->external = 0; |
1da177e4 LT |
5891 | |
5892 | mddev->layout = info->layout; | |
9d8f0363 | 5893 | mddev->chunk_sectors = info->chunk_size >> 9; |
1da177e4 LT |
5894 | |
5895 | mddev->max_disks = MD_SB_DISKS; | |
5896 | ||
e691063a N |
5897 | if (mddev->persistent) |
5898 | mddev->flags = 0; | |
850b2b42 | 5899 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 5900 | |
c3d9714e | 5901 | mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; |
6409bb05 | 5902 | mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9); |
c3d9714e | 5903 | mddev->bitmap_info.offset = 0; |
b2a2703c | 5904 | |
f6705578 N |
5905 | mddev->reshape_position = MaxSector; |
5906 | ||
1da177e4 LT |
5907 | /* |
5908 | * Generate a 128 bit UUID | |
5909 | */ | |
5910 | get_random_bytes(mddev->uuid, 16); | |
5911 | ||
f6705578 | 5912 | mddev->new_level = mddev->level; |
664e7c41 | 5913 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
f6705578 N |
5914 | mddev->new_layout = mddev->layout; |
5915 | mddev->delta_disks = 0; | |
2c810cdd | 5916 | mddev->reshape_backwards = 0; |
f6705578 | 5917 | |
1da177e4 LT |
5918 | return 0; |
5919 | } | |
5920 | ||
fd01b88c | 5921 | void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors) |
1f403624 | 5922 | { |
b522adcd DW |
5923 | WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__); |
5924 | ||
5925 | if (mddev->external_size) | |
5926 | return; | |
5927 | ||
1f403624 DW |
5928 | mddev->array_sectors = array_sectors; |
5929 | } | |
5930 | EXPORT_SYMBOL(md_set_array_sectors); | |
5931 | ||
fd01b88c | 5932 | static int update_size(struct mddev *mddev, sector_t num_sectors) |
a35b0d69 | 5933 | { |
3cb03002 | 5934 | struct md_rdev *rdev; |
a35b0d69 | 5935 | int rv; |
d71f9f88 | 5936 | int fit = (num_sectors == 0); |
a35b0d69 N |
5937 | |
5938 | if (mddev->pers->resize == NULL) | |
5939 | return -EINVAL; | |
d71f9f88 AN |
5940 | /* The "num_sectors" is the number of sectors of each device that |
5941 | * is used. This can only make sense for arrays with redundancy. | |
5942 | * linear and raid0 always use whatever space is available. We can only | |
5943 | * consider changing this number if no resync or reconstruction is | |
5944 | * happening, and if the new size is acceptable. It must fit before the | |
0f420358 | 5945 | * sb_start or, if that is <data_offset, it must fit before the size |
d71f9f88 AN |
5946 | * of each device. If num_sectors is zero, we find the largest size |
5947 | * that fits. | |
a35b0d69 N |
5948 | */ |
5949 | if (mddev->sync_thread) | |
5950 | return -EBUSY; | |
bd8839e0 N |
5951 | if (mddev->ro) |
5952 | return -EROFS; | |
a4a6125a | 5953 | |
dafb20fa | 5954 | rdev_for_each(rdev, mddev) { |
dd8ac336 | 5955 | sector_t avail = rdev->sectors; |
01ab5662 | 5956 | |
d71f9f88 AN |
5957 | if (fit && (num_sectors == 0 || num_sectors > avail)) |
5958 | num_sectors = avail; | |
5959 | if (avail < num_sectors) | |
a35b0d69 N |
5960 | return -ENOSPC; |
5961 | } | |
d71f9f88 | 5962 | rv = mddev->pers->resize(mddev, num_sectors); |
449aad3e N |
5963 | if (!rv) |
5964 | revalidate_disk(mddev->gendisk); | |
a35b0d69 N |
5965 | return rv; |
5966 | } | |
5967 | ||
fd01b88c | 5968 | static int update_raid_disks(struct mddev *mddev, int raid_disks) |
da943b99 N |
5969 | { |
5970 | int rv; | |
c6563a8c | 5971 | struct md_rdev *rdev; |
da943b99 | 5972 | /* change the number of raid disks */ |
63c70c4f | 5973 | if (mddev->pers->check_reshape == NULL) |
da943b99 | 5974 | return -EINVAL; |
bd8839e0 N |
5975 | if (mddev->ro) |
5976 | return -EROFS; | |
da943b99 | 5977 | if (raid_disks <= 0 || |
233fca36 | 5978 | (mddev->max_disks && raid_disks >= mddev->max_disks)) |
da943b99 | 5979 | return -EINVAL; |
63c70c4f | 5980 | if (mddev->sync_thread || mddev->reshape_position != MaxSector) |
da943b99 | 5981 | return -EBUSY; |
c6563a8c N |
5982 | |
5983 | rdev_for_each(rdev, mddev) { | |
5984 | if (mddev->raid_disks < raid_disks && | |
5985 | rdev->data_offset < rdev->new_data_offset) | |
5986 | return -EINVAL; | |
5987 | if (mddev->raid_disks > raid_disks && | |
5988 | rdev->data_offset > rdev->new_data_offset) | |
5989 | return -EINVAL; | |
5990 | } | |
5991 | ||
63c70c4f | 5992 | mddev->delta_disks = raid_disks - mddev->raid_disks; |
2c810cdd N |
5993 | if (mddev->delta_disks < 0) |
5994 | mddev->reshape_backwards = 1; | |
5995 | else if (mddev->delta_disks > 0) | |
5996 | mddev->reshape_backwards = 0; | |
63c70c4f N |
5997 | |
5998 | rv = mddev->pers->check_reshape(mddev); | |
2c810cdd | 5999 | if (rv < 0) { |
de171cb9 | 6000 | mddev->delta_disks = 0; |
2c810cdd N |
6001 | mddev->reshape_backwards = 0; |
6002 | } | |
da943b99 N |
6003 | return rv; |
6004 | } | |
6005 | ||
1da177e4 LT |
6006 | /* |
6007 | * update_array_info is used to change the configuration of an | |
6008 | * on-line array. | |
6009 | * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size | |
6010 | * fields in the info are checked against the array. | |
6011 | * Any differences that cannot be handled will cause an error. | |
6012 | * Normally, only one change can be managed at a time. | |
6013 | */ | |
fd01b88c | 6014 | static int update_array_info(struct mddev *mddev, mdu_array_info_t *info) |
1da177e4 LT |
6015 | { |
6016 | int rv = 0; | |
6017 | int cnt = 0; | |
36fa3063 N |
6018 | int state = 0; |
6019 | ||
6020 | /* calculate expected state,ignoring low bits */ | |
c3d9714e | 6021 | if (mddev->bitmap && mddev->bitmap_info.offset) |
36fa3063 | 6022 | state |= (1 << MD_SB_BITMAP_PRESENT); |
1da177e4 LT |
6023 | |
6024 | if (mddev->major_version != info->major_version || | |
6025 | mddev->minor_version != info->minor_version || | |
6026 | /* mddev->patch_version != info->patch_version || */ | |
6027 | mddev->ctime != info->ctime || | |
6028 | mddev->level != info->level || | |
6029 | /* mddev->layout != info->layout || */ | |
6030 | !mddev->persistent != info->not_persistent|| | |
9d8f0363 | 6031 | mddev->chunk_sectors != info->chunk_size >> 9 || |
36fa3063 N |
6032 | /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */ |
6033 | ((state^info->state) & 0xfffffe00) | |
6034 | ) | |
1da177e4 LT |
6035 | return -EINVAL; |
6036 | /* Check there is only one change */ | |
58c0fed4 AN |
6037 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
6038 | cnt++; | |
6039 | if (mddev->raid_disks != info->raid_disks) | |
6040 | cnt++; | |
6041 | if (mddev->layout != info->layout) | |
6042 | cnt++; | |
6043 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) | |
6044 | cnt++; | |
6045 | if (cnt == 0) | |
6046 | return 0; | |
6047 | if (cnt > 1) | |
6048 | return -EINVAL; | |
1da177e4 LT |
6049 | |
6050 | if (mddev->layout != info->layout) { | |
6051 | /* Change layout | |
6052 | * we don't need to do anything at the md level, the | |
6053 | * personality will take care of it all. | |
6054 | */ | |
50ac168a | 6055 | if (mddev->pers->check_reshape == NULL) |
1da177e4 | 6056 | return -EINVAL; |
597a711b N |
6057 | else { |
6058 | mddev->new_layout = info->layout; | |
50ac168a | 6059 | rv = mddev->pers->check_reshape(mddev); |
597a711b N |
6060 | if (rv) |
6061 | mddev->new_layout = mddev->layout; | |
6062 | return rv; | |
6063 | } | |
1da177e4 | 6064 | } |
58c0fed4 | 6065 | if (info->size >= 0 && mddev->dev_sectors / 2 != info->size) |
d71f9f88 | 6066 | rv = update_size(mddev, (sector_t)info->size * 2); |
a35b0d69 | 6067 | |
da943b99 N |
6068 | if (mddev->raid_disks != info->raid_disks) |
6069 | rv = update_raid_disks(mddev, info->raid_disks); | |
6070 | ||
36fa3063 | 6071 | if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) { |
d66b1b39 | 6072 | if (mddev->pers->quiesce == NULL || mddev->thread == NULL) |
36fa3063 N |
6073 | return -EINVAL; |
6074 | if (mddev->recovery || mddev->sync_thread) | |
6075 | return -EBUSY; | |
6076 | if (info->state & (1<<MD_SB_BITMAP_PRESENT)) { | |
6077 | /* add the bitmap */ | |
6078 | if (mddev->bitmap) | |
6079 | return -EEXIST; | |
c3d9714e | 6080 | if (mddev->bitmap_info.default_offset == 0) |
36fa3063 | 6081 | return -EINVAL; |
c3d9714e N |
6082 | mddev->bitmap_info.offset = |
6083 | mddev->bitmap_info.default_offset; | |
6409bb05 N |
6084 | mddev->bitmap_info.space = |
6085 | mddev->bitmap_info.default_space; | |
36fa3063 N |
6086 | mddev->pers->quiesce(mddev, 1); |
6087 | rv = bitmap_create(mddev); | |
69e51b44 N |
6088 | if (!rv) |
6089 | rv = bitmap_load(mddev); | |
36fa3063 N |
6090 | if (rv) |
6091 | bitmap_destroy(mddev); | |
6092 | mddev->pers->quiesce(mddev, 0); | |
6093 | } else { | |
6094 | /* remove the bitmap */ | |
6095 | if (!mddev->bitmap) | |
6096 | return -ENOENT; | |
1ec885cd | 6097 | if (mddev->bitmap->storage.file) |
36fa3063 N |
6098 | return -EINVAL; |
6099 | mddev->pers->quiesce(mddev, 1); | |
6100 | bitmap_destroy(mddev); | |
6101 | mddev->pers->quiesce(mddev, 0); | |
c3d9714e | 6102 | mddev->bitmap_info.offset = 0; |
36fa3063 N |
6103 | } |
6104 | } | |
850b2b42 | 6105 | md_update_sb(mddev, 1); |
1da177e4 LT |
6106 | return rv; |
6107 | } | |
6108 | ||
fd01b88c | 6109 | static int set_disk_faulty(struct mddev *mddev, dev_t dev) |
1da177e4 | 6110 | { |
3cb03002 | 6111 | struct md_rdev *rdev; |
1ca69c4b | 6112 | int err = 0; |
1da177e4 LT |
6113 | |
6114 | if (mddev->pers == NULL) | |
6115 | return -ENODEV; | |
6116 | ||
1ca69c4b N |
6117 | rcu_read_lock(); |
6118 | rdev = find_rdev_rcu(mddev, dev); | |
1da177e4 | 6119 | if (!rdev) |
1ca69c4b N |
6120 | err = -ENODEV; |
6121 | else { | |
6122 | md_error(mddev, rdev); | |
6123 | if (!test_bit(Faulty, &rdev->flags)) | |
6124 | err = -EBUSY; | |
6125 | } | |
6126 | rcu_read_unlock(); | |
6127 | return err; | |
1da177e4 LT |
6128 | } |
6129 | ||
2f9618ce AN |
6130 | /* |
6131 | * We have a problem here : there is no easy way to give a CHS | |
6132 | * virtual geometry. We currently pretend that we have a 2 heads | |
6133 | * 4 sectors (with a BIG number of cylinders...). This drives | |
6134 | * dosfs just mad... ;-) | |
6135 | */ | |
a885c8c4 CH |
6136 | static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
6137 | { | |
fd01b88c | 6138 | struct mddev *mddev = bdev->bd_disk->private_data; |
a885c8c4 CH |
6139 | |
6140 | geo->heads = 2; | |
6141 | geo->sectors = 4; | |
49ce6cea | 6142 | geo->cylinders = mddev->array_sectors / 8; |
a885c8c4 CH |
6143 | return 0; |
6144 | } | |
6145 | ||
cb335f88 NS |
6146 | static inline bool md_ioctl_valid(unsigned int cmd) |
6147 | { | |
6148 | switch (cmd) { | |
6149 | case ADD_NEW_DISK: | |
6150 | case BLKROSET: | |
6151 | case GET_ARRAY_INFO: | |
6152 | case GET_BITMAP_FILE: | |
6153 | case GET_DISK_INFO: | |
6154 | case HOT_ADD_DISK: | |
6155 | case HOT_REMOVE_DISK: | |
cb335f88 NS |
6156 | case RAID_AUTORUN: |
6157 | case RAID_VERSION: | |
6158 | case RESTART_ARRAY_RW: | |
6159 | case RUN_ARRAY: | |
6160 | case SET_ARRAY_INFO: | |
6161 | case SET_BITMAP_FILE: | |
6162 | case SET_DISK_FAULTY: | |
6163 | case STOP_ARRAY: | |
6164 | case STOP_ARRAY_RO: | |
6165 | return true; | |
6166 | default: | |
6167 | return false; | |
6168 | } | |
6169 | } | |
6170 | ||
a39907fa | 6171 | static int md_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
6172 | unsigned int cmd, unsigned long arg) |
6173 | { | |
6174 | int err = 0; | |
6175 | void __user *argp = (void __user *)arg; | |
fd01b88c | 6176 | struct mddev *mddev = NULL; |
e2218350 | 6177 | int ro; |
1da177e4 | 6178 | |
cb335f88 NS |
6179 | if (!md_ioctl_valid(cmd)) |
6180 | return -ENOTTY; | |
6181 | ||
506c9e44 N |
6182 | switch (cmd) { |
6183 | case RAID_VERSION: | |
6184 | case GET_ARRAY_INFO: | |
6185 | case GET_DISK_INFO: | |
6186 | break; | |
6187 | default: | |
6188 | if (!capable(CAP_SYS_ADMIN)) | |
6189 | return -EACCES; | |
6190 | } | |
1da177e4 LT |
6191 | |
6192 | /* | |
6193 | * Commands dealing with the RAID driver but not any | |
6194 | * particular array: | |
6195 | */ | |
c02c0aeb N |
6196 | switch (cmd) { |
6197 | case RAID_VERSION: | |
6198 | err = get_version(argp); | |
3adc28d8 | 6199 | goto out; |
1da177e4 | 6200 | |
1da177e4 | 6201 | #ifndef MODULE |
c02c0aeb N |
6202 | case RAID_AUTORUN: |
6203 | err = 0; | |
6204 | autostart_arrays(arg); | |
3adc28d8 | 6205 | goto out; |
1da177e4 | 6206 | #endif |
c02c0aeb | 6207 | default:; |
1da177e4 LT |
6208 | } |
6209 | ||
6210 | /* | |
6211 | * Commands creating/starting a new array: | |
6212 | */ | |
6213 | ||
a39907fa | 6214 | mddev = bdev->bd_disk->private_data; |
1da177e4 LT |
6215 | |
6216 | if (!mddev) { | |
6217 | BUG(); | |
3adc28d8 | 6218 | goto out; |
1da177e4 LT |
6219 | } |
6220 | ||
1ca69c4b N |
6221 | /* Some actions do not requires the mutex */ |
6222 | switch (cmd) { | |
6223 | case GET_ARRAY_INFO: | |
6224 | if (!mddev->raid_disks && !mddev->external) | |
6225 | err = -ENODEV; | |
6226 | else | |
6227 | err = get_array_info(mddev, argp); | |
3adc28d8 | 6228 | goto out; |
1ca69c4b N |
6229 | |
6230 | case GET_DISK_INFO: | |
6231 | if (!mddev->raid_disks && !mddev->external) | |
6232 | err = -ENODEV; | |
6233 | else | |
6234 | err = get_disk_info(mddev, argp); | |
3adc28d8 | 6235 | goto out; |
1ca69c4b N |
6236 | |
6237 | case SET_DISK_FAULTY: | |
6238 | err = set_disk_faulty(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6239 | goto out; |
1ca69c4b N |
6240 | } |
6241 | ||
a7a3f08d N |
6242 | if (cmd == ADD_NEW_DISK) |
6243 | /* need to ensure md_delayed_delete() has completed */ | |
6244 | flush_workqueue(md_misc_wq); | |
6245 | ||
90f5f7ad HR |
6246 | if (cmd == HOT_REMOVE_DISK) |
6247 | /* need to ensure recovery thread has run */ | |
6248 | wait_event_interruptible_timeout(mddev->sb_wait, | |
6249 | !test_bit(MD_RECOVERY_NEEDED, | |
6250 | &mddev->flags), | |
6251 | msecs_to_jiffies(5000)); | |
260fa034 N |
6252 | if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) { |
6253 | /* Need to flush page cache, and ensure no-one else opens | |
6254 | * and writes | |
6255 | */ | |
6256 | mutex_lock(&mddev->open_mutex); | |
9ba3b7f5 | 6257 | if (mddev->pers && atomic_read(&mddev->openers) > 1) { |
260fa034 N |
6258 | mutex_unlock(&mddev->open_mutex); |
6259 | err = -EBUSY; | |
3adc28d8 | 6260 | goto out; |
260fa034 N |
6261 | } |
6262 | set_bit(MD_STILL_CLOSED, &mddev->flags); | |
6263 | mutex_unlock(&mddev->open_mutex); | |
6264 | sync_blockdev(bdev); | |
6265 | } | |
1da177e4 LT |
6266 | err = mddev_lock(mddev); |
6267 | if (err) { | |
f72ffdd6 | 6268 | printk(KERN_INFO |
1da177e4 LT |
6269 | "md: ioctl lock interrupted, reason %d, cmd %d\n", |
6270 | err, cmd); | |
3adc28d8 | 6271 | goto out; |
1da177e4 LT |
6272 | } |
6273 | ||
c02c0aeb N |
6274 | if (cmd == SET_ARRAY_INFO) { |
6275 | mdu_array_info_t info; | |
6276 | if (!arg) | |
6277 | memset(&info, 0, sizeof(info)); | |
6278 | else if (copy_from_user(&info, argp, sizeof(info))) { | |
6279 | err = -EFAULT; | |
3adc28d8 | 6280 | goto unlock; |
c02c0aeb N |
6281 | } |
6282 | if (mddev->pers) { | |
6283 | err = update_array_info(mddev, &info); | |
6284 | if (err) { | |
6285 | printk(KERN_WARNING "md: couldn't update" | |
6286 | " array info. %d\n", err); | |
3adc28d8 | 6287 | goto unlock; |
1da177e4 | 6288 | } |
3adc28d8 | 6289 | goto unlock; |
c02c0aeb N |
6290 | } |
6291 | if (!list_empty(&mddev->disks)) { | |
6292 | printk(KERN_WARNING | |
6293 | "md: array %s already has disks!\n", | |
6294 | mdname(mddev)); | |
6295 | err = -EBUSY; | |
3adc28d8 | 6296 | goto unlock; |
c02c0aeb N |
6297 | } |
6298 | if (mddev->raid_disks) { | |
6299 | printk(KERN_WARNING | |
6300 | "md: array %s already initialised!\n", | |
6301 | mdname(mddev)); | |
6302 | err = -EBUSY; | |
3adc28d8 | 6303 | goto unlock; |
c02c0aeb N |
6304 | } |
6305 | err = set_array_info(mddev, &info); | |
6306 | if (err) { | |
6307 | printk(KERN_WARNING "md: couldn't set" | |
6308 | " array info. %d\n", err); | |
3adc28d8 | 6309 | goto unlock; |
c02c0aeb | 6310 | } |
3adc28d8 | 6311 | goto unlock; |
1da177e4 LT |
6312 | } |
6313 | ||
6314 | /* | |
6315 | * Commands querying/configuring an existing array: | |
6316 | */ | |
32a7627c | 6317 | /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, |
3f9d7b0d | 6318 | * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */ |
a17184a9 N |
6319 | if ((!mddev->raid_disks && !mddev->external) |
6320 | && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY | |
6321 | && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE | |
6322 | && cmd != GET_BITMAP_FILE) { | |
1da177e4 | 6323 | err = -ENODEV; |
3adc28d8 | 6324 | goto unlock; |
1da177e4 LT |
6325 | } |
6326 | ||
6327 | /* | |
6328 | * Commands even a read-only array can execute: | |
6329 | */ | |
c02c0aeb N |
6330 | switch (cmd) { |
6331 | case GET_BITMAP_FILE: | |
6332 | err = get_bitmap_file(mddev, argp); | |
3adc28d8 | 6333 | goto unlock; |
32a7627c | 6334 | |
c02c0aeb N |
6335 | case RESTART_ARRAY_RW: |
6336 | err = restart_array(mddev); | |
3adc28d8 | 6337 | goto unlock; |
1da177e4 | 6338 | |
c02c0aeb N |
6339 | case STOP_ARRAY: |
6340 | err = do_md_stop(mddev, 0, bdev); | |
3adc28d8 | 6341 | goto unlock; |
1da177e4 | 6342 | |
c02c0aeb N |
6343 | case STOP_ARRAY_RO: |
6344 | err = md_set_readonly(mddev, bdev); | |
3adc28d8 | 6345 | goto unlock; |
1da177e4 | 6346 | |
3ea8929d N |
6347 | case HOT_REMOVE_DISK: |
6348 | err = hot_remove_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6349 | goto unlock; |
3ea8929d | 6350 | |
7ceb17e8 N |
6351 | case ADD_NEW_DISK: |
6352 | /* We can support ADD_NEW_DISK on read-only arrays | |
6353 | * on if we are re-adding a preexisting device. | |
6354 | * So require mddev->pers and MD_DISK_SYNC. | |
6355 | */ | |
6356 | if (mddev->pers) { | |
6357 | mdu_disk_info_t info; | |
6358 | if (copy_from_user(&info, argp, sizeof(info))) | |
6359 | err = -EFAULT; | |
6360 | else if (!(info.state & (1<<MD_DISK_SYNC))) | |
6361 | /* Need to clear read-only for this */ | |
6362 | break; | |
6363 | else | |
6364 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 6365 | goto unlock; |
7ceb17e8 N |
6366 | } |
6367 | break; | |
6368 | ||
c02c0aeb N |
6369 | case BLKROSET: |
6370 | if (get_user(ro, (int __user *)(arg))) { | |
6371 | err = -EFAULT; | |
3adc28d8 | 6372 | goto unlock; |
c02c0aeb N |
6373 | } |
6374 | err = -EINVAL; | |
e2218350 | 6375 | |
c02c0aeb N |
6376 | /* if the bdev is going readonly the value of mddev->ro |
6377 | * does not matter, no writes are coming | |
6378 | */ | |
6379 | if (ro) | |
3adc28d8 | 6380 | goto unlock; |
e2218350 | 6381 | |
c02c0aeb N |
6382 | /* are we are already prepared for writes? */ |
6383 | if (mddev->ro != 1) | |
3adc28d8 | 6384 | goto unlock; |
e2218350 | 6385 | |
c02c0aeb N |
6386 | /* transitioning to readauto need only happen for |
6387 | * arrays that call md_write_start | |
6388 | */ | |
6389 | if (mddev->pers) { | |
6390 | err = restart_array(mddev); | |
6391 | if (err == 0) { | |
6392 | mddev->ro = 2; | |
6393 | set_disk_ro(mddev->gendisk, 0); | |
e2218350 | 6394 | } |
c02c0aeb | 6395 | } |
3adc28d8 | 6396 | goto unlock; |
1da177e4 LT |
6397 | } |
6398 | ||
6399 | /* | |
6400 | * The remaining ioctls are changing the state of the | |
f91de92e | 6401 | * superblock, so we do not allow them on read-only arrays. |
1da177e4 | 6402 | */ |
326eb17d | 6403 | if (mddev->ro && mddev->pers) { |
f91de92e N |
6404 | if (mddev->ro == 2) { |
6405 | mddev->ro = 0; | |
00bcb4ac | 6406 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
0fd62b86 | 6407 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
f3378b48 N |
6408 | /* mddev_unlock will wake thread */ |
6409 | /* If a device failed while we were read-only, we | |
6410 | * need to make sure the metadata is updated now. | |
6411 | */ | |
6412 | if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) { | |
6413 | mddev_unlock(mddev); | |
6414 | wait_event(mddev->sb_wait, | |
6415 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) && | |
6416 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
29f097c4 | 6417 | mddev_lock_nointr(mddev); |
f3378b48 | 6418 | } |
f91de92e N |
6419 | } else { |
6420 | err = -EROFS; | |
3adc28d8 | 6421 | goto unlock; |
f91de92e | 6422 | } |
1da177e4 LT |
6423 | } |
6424 | ||
c02c0aeb N |
6425 | switch (cmd) { |
6426 | case ADD_NEW_DISK: | |
1da177e4 | 6427 | { |
c02c0aeb N |
6428 | mdu_disk_info_t info; |
6429 | if (copy_from_user(&info, argp, sizeof(info))) | |
6430 | err = -EFAULT; | |
6431 | else | |
6432 | err = add_new_disk(mddev, &info); | |
3adc28d8 | 6433 | goto unlock; |
c02c0aeb | 6434 | } |
1da177e4 | 6435 | |
c02c0aeb N |
6436 | case HOT_ADD_DISK: |
6437 | err = hot_add_disk(mddev, new_decode_dev(arg)); | |
3adc28d8 | 6438 | goto unlock; |
1da177e4 | 6439 | |
c02c0aeb N |
6440 | case RUN_ARRAY: |
6441 | err = do_md_run(mddev); | |
3adc28d8 | 6442 | goto unlock; |
1da177e4 | 6443 | |
c02c0aeb N |
6444 | case SET_BITMAP_FILE: |
6445 | err = set_bitmap_file(mddev, (int)arg); | |
3adc28d8 | 6446 | goto unlock; |
32a7627c | 6447 | |
c02c0aeb N |
6448 | default: |
6449 | err = -EINVAL; | |
3adc28d8 | 6450 | goto unlock; |
1da177e4 LT |
6451 | } |
6452 | ||
3adc28d8 | 6453 | unlock: |
d3374825 N |
6454 | if (mddev->hold_active == UNTIL_IOCTL && |
6455 | err != -EINVAL) | |
6456 | mddev->hold_active = 0; | |
1da177e4 | 6457 | mddev_unlock(mddev); |
3adc28d8 | 6458 | out: |
1da177e4 LT |
6459 | return err; |
6460 | } | |
aa98aa31 AB |
6461 | #ifdef CONFIG_COMPAT |
6462 | static int md_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
6463 | unsigned int cmd, unsigned long arg) | |
6464 | { | |
6465 | switch (cmd) { | |
6466 | case HOT_REMOVE_DISK: | |
6467 | case HOT_ADD_DISK: | |
6468 | case SET_DISK_FAULTY: | |
6469 | case SET_BITMAP_FILE: | |
6470 | /* These take in integer arg, do not convert */ | |
6471 | break; | |
6472 | default: | |
6473 | arg = (unsigned long)compat_ptr(arg); | |
6474 | break; | |
6475 | } | |
6476 | ||
6477 | return md_ioctl(bdev, mode, cmd, arg); | |
6478 | } | |
6479 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 6480 | |
a39907fa | 6481 | static int md_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
6482 | { |
6483 | /* | |
6484 | * Succeed if we can lock the mddev, which confirms that | |
6485 | * it isn't being stopped right now. | |
6486 | */ | |
fd01b88c | 6487 | struct mddev *mddev = mddev_find(bdev->bd_dev); |
1da177e4 LT |
6488 | int err; |
6489 | ||
0c098220 YL |
6490 | if (!mddev) |
6491 | return -ENODEV; | |
6492 | ||
d3374825 N |
6493 | if (mddev->gendisk != bdev->bd_disk) { |
6494 | /* we are racing with mddev_put which is discarding this | |
6495 | * bd_disk. | |
6496 | */ | |
6497 | mddev_put(mddev); | |
6498 | /* Wait until bdev->bd_disk is definitely gone */ | |
e804ac78 | 6499 | flush_workqueue(md_misc_wq); |
d3374825 N |
6500 | /* Then retry the open from the top */ |
6501 | return -ERESTARTSYS; | |
6502 | } | |
6503 | BUG_ON(mddev != bdev->bd_disk->private_data); | |
6504 | ||
c8c00a69 | 6505 | if ((err = mutex_lock_interruptible(&mddev->open_mutex))) |
1da177e4 LT |
6506 | goto out; |
6507 | ||
6508 | err = 0; | |
f2ea68cf | 6509 | atomic_inc(&mddev->openers); |
260fa034 | 6510 | clear_bit(MD_STILL_CLOSED, &mddev->flags); |
c8c00a69 | 6511 | mutex_unlock(&mddev->open_mutex); |
1da177e4 | 6512 | |
f0b4f7e2 | 6513 | check_disk_change(bdev); |
1da177e4 LT |
6514 | out: |
6515 | return err; | |
6516 | } | |
6517 | ||
db2a144b | 6518 | static void md_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 6519 | { |
f72ffdd6 | 6520 | struct mddev *mddev = disk->private_data; |
1da177e4 | 6521 | |
52e5f9d1 | 6522 | BUG_ON(!mddev); |
f2ea68cf | 6523 | atomic_dec(&mddev->openers); |
1da177e4 | 6524 | mddev_put(mddev); |
1da177e4 | 6525 | } |
f0b4f7e2 N |
6526 | |
6527 | static int md_media_changed(struct gendisk *disk) | |
6528 | { | |
fd01b88c | 6529 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6530 | |
6531 | return mddev->changed; | |
6532 | } | |
6533 | ||
6534 | static int md_revalidate(struct gendisk *disk) | |
6535 | { | |
fd01b88c | 6536 | struct mddev *mddev = disk->private_data; |
f0b4f7e2 N |
6537 | |
6538 | mddev->changed = 0; | |
6539 | return 0; | |
6540 | } | |
83d5cde4 | 6541 | static const struct block_device_operations md_fops = |
1da177e4 LT |
6542 | { |
6543 | .owner = THIS_MODULE, | |
a39907fa AV |
6544 | .open = md_open, |
6545 | .release = md_release, | |
b492b852 | 6546 | .ioctl = md_ioctl, |
aa98aa31 AB |
6547 | #ifdef CONFIG_COMPAT |
6548 | .compat_ioctl = md_compat_ioctl, | |
6549 | #endif | |
a885c8c4 | 6550 | .getgeo = md_getgeo, |
f0b4f7e2 N |
6551 | .media_changed = md_media_changed, |
6552 | .revalidate_disk= md_revalidate, | |
1da177e4 LT |
6553 | }; |
6554 | ||
f72ffdd6 | 6555 | static int md_thread(void *arg) |
1da177e4 | 6556 | { |
2b8bf345 | 6557 | struct md_thread *thread = arg; |
1da177e4 | 6558 | |
1da177e4 LT |
6559 | /* |
6560 | * md_thread is a 'system-thread', it's priority should be very | |
6561 | * high. We avoid resource deadlocks individually in each | |
6562 | * raid personality. (RAID5 does preallocation) We also use RR and | |
6563 | * the very same RT priority as kswapd, thus we will never get | |
6564 | * into a priority inversion deadlock. | |
6565 | * | |
6566 | * we definitely have to have equal or higher priority than | |
6567 | * bdflush, otherwise bdflush will deadlock if there are too | |
6568 | * many dirty RAID5 blocks. | |
6569 | */ | |
1da177e4 | 6570 | |
6985c43f | 6571 | allow_signal(SIGKILL); |
a6fb0934 | 6572 | while (!kthread_should_stop()) { |
1da177e4 | 6573 | |
93588e22 N |
6574 | /* We need to wait INTERRUPTIBLE so that |
6575 | * we don't add to the load-average. | |
6576 | * That means we need to be sure no signals are | |
6577 | * pending | |
6578 | */ | |
6579 | if (signal_pending(current)) | |
6580 | flush_signals(current); | |
6581 | ||
6582 | wait_event_interruptible_timeout | |
6583 | (thread->wqueue, | |
6584 | test_bit(THREAD_WAKEUP, &thread->flags) | |
6585 | || kthread_should_stop(), | |
6586 | thread->timeout); | |
1da177e4 | 6587 | |
6c987910 N |
6588 | clear_bit(THREAD_WAKEUP, &thread->flags); |
6589 | if (!kthread_should_stop()) | |
4ed8731d | 6590 | thread->run(thread); |
1da177e4 | 6591 | } |
a6fb0934 | 6592 | |
1da177e4 LT |
6593 | return 0; |
6594 | } | |
6595 | ||
2b8bf345 | 6596 | void md_wakeup_thread(struct md_thread *thread) |
1da177e4 LT |
6597 | { |
6598 | if (thread) { | |
36a4e1fe | 6599 | pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm); |
1da177e4 LT |
6600 | set_bit(THREAD_WAKEUP, &thread->flags); |
6601 | wake_up(&thread->wqueue); | |
6602 | } | |
6603 | } | |
6c144d31 | 6604 | EXPORT_SYMBOL(md_wakeup_thread); |
1da177e4 | 6605 | |
4ed8731d SL |
6606 | struct md_thread *md_register_thread(void (*run) (struct md_thread *), |
6607 | struct mddev *mddev, const char *name) | |
1da177e4 | 6608 | { |
2b8bf345 | 6609 | struct md_thread *thread; |
1da177e4 | 6610 | |
2b8bf345 | 6611 | thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL); |
1da177e4 LT |
6612 | if (!thread) |
6613 | return NULL; | |
6614 | ||
1da177e4 LT |
6615 | init_waitqueue_head(&thread->wqueue); |
6616 | ||
1da177e4 LT |
6617 | thread->run = run; |
6618 | thread->mddev = mddev; | |
32a7627c | 6619 | thread->timeout = MAX_SCHEDULE_TIMEOUT; |
0da3c619 N |
6620 | thread->tsk = kthread_run(md_thread, thread, |
6621 | "%s_%s", | |
6622 | mdname(thread->mddev), | |
0232605d | 6623 | name); |
a6fb0934 | 6624 | if (IS_ERR(thread->tsk)) { |
1da177e4 LT |
6625 | kfree(thread); |
6626 | return NULL; | |
6627 | } | |
1da177e4 LT |
6628 | return thread; |
6629 | } | |
6c144d31 | 6630 | EXPORT_SYMBOL(md_register_thread); |
1da177e4 | 6631 | |
2b8bf345 | 6632 | void md_unregister_thread(struct md_thread **threadp) |
1da177e4 | 6633 | { |
2b8bf345 | 6634 | struct md_thread *thread = *threadp; |
e0cf8f04 N |
6635 | if (!thread) |
6636 | return; | |
36a4e1fe | 6637 | pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk)); |
01f96c0a N |
6638 | /* Locking ensures that mddev_unlock does not wake_up a |
6639 | * non-existent thread | |
6640 | */ | |
6641 | spin_lock(&pers_lock); | |
6642 | *threadp = NULL; | |
6643 | spin_unlock(&pers_lock); | |
a6fb0934 N |
6644 | |
6645 | kthread_stop(thread->tsk); | |
1da177e4 LT |
6646 | kfree(thread); |
6647 | } | |
6c144d31 | 6648 | EXPORT_SYMBOL(md_unregister_thread); |
1da177e4 | 6649 | |
fd01b88c | 6650 | void md_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 6651 | { |
b2d444d7 | 6652 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
1da177e4 | 6653 | return; |
6bfe0b49 | 6654 | |
de393cde | 6655 | if (!mddev->pers || !mddev->pers->error_handler) |
1da177e4 LT |
6656 | return; |
6657 | mddev->pers->error_handler(mddev,rdev); | |
72a23c21 NB |
6658 | if (mddev->degraded) |
6659 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); | |
00bcb4ac | 6660 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
6661 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
6662 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
6663 | md_wakeup_thread(mddev->thread); | |
768a418d | 6664 | if (mddev->event_work.func) |
e804ac78 | 6665 | queue_work(md_misc_wq, &mddev->event_work); |
c331eb04 | 6666 | md_new_event_inintr(mddev); |
1da177e4 | 6667 | } |
6c144d31 | 6668 | EXPORT_SYMBOL(md_error); |
1da177e4 LT |
6669 | |
6670 | /* seq_file implementation /proc/mdstat */ | |
6671 | ||
6672 | static void status_unused(struct seq_file *seq) | |
6673 | { | |
6674 | int i = 0; | |
3cb03002 | 6675 | struct md_rdev *rdev; |
1da177e4 LT |
6676 | |
6677 | seq_printf(seq, "unused devices: "); | |
6678 | ||
159ec1fc | 6679 | list_for_each_entry(rdev, &pending_raid_disks, same_set) { |
1da177e4 LT |
6680 | char b[BDEVNAME_SIZE]; |
6681 | i++; | |
6682 | seq_printf(seq, "%s ", | |
6683 | bdevname(rdev->bdev,b)); | |
6684 | } | |
6685 | if (!i) | |
6686 | seq_printf(seq, "<none>"); | |
6687 | ||
6688 | seq_printf(seq, "\n"); | |
6689 | } | |
6690 | ||
f72ffdd6 | 6691 | static void status_resync(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 6692 | { |
dd71cf6b N |
6693 | sector_t max_sectors, resync, res; |
6694 | unsigned long dt, db; | |
6695 | sector_t rt; | |
4588b42e N |
6696 | int scale; |
6697 | unsigned int per_milli; | |
1da177e4 | 6698 | |
72f36d59 N |
6699 | if (mddev->curr_resync <= 3) |
6700 | resync = 0; | |
6701 | else | |
6702 | resync = mddev->curr_resync | |
6703 | - atomic_read(&mddev->recovery_active); | |
1da177e4 | 6704 | |
c804cdec N |
6705 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
6706 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
dd71cf6b | 6707 | max_sectors = mddev->resync_max_sectors; |
1da177e4 | 6708 | else |
dd71cf6b | 6709 | max_sectors = mddev->dev_sectors; |
1da177e4 | 6710 | |
403df478 | 6711 | WARN_ON(max_sectors == 0); |
4588b42e | 6712 | /* Pick 'scale' such that (resync>>scale)*1000 will fit |
dd71cf6b | 6713 | * in a sector_t, and (max_sectors>>scale) will fit in a |
4588b42e N |
6714 | * u32, as those are the requirements for sector_div. |
6715 | * Thus 'scale' must be at least 10 | |
6716 | */ | |
6717 | scale = 10; | |
6718 | if (sizeof(sector_t) > sizeof(unsigned long)) { | |
dd71cf6b | 6719 | while ( max_sectors/2 > (1ULL<<(scale+32))) |
4588b42e N |
6720 | scale++; |
6721 | } | |
6722 | res = (resync>>scale)*1000; | |
dd71cf6b | 6723 | sector_div(res, (u32)((max_sectors>>scale)+1)); |
4588b42e N |
6724 | |
6725 | per_milli = res; | |
1da177e4 | 6726 | { |
4588b42e | 6727 | int i, x = per_milli/50, y = 20-x; |
1da177e4 LT |
6728 | seq_printf(seq, "["); |
6729 | for (i = 0; i < x; i++) | |
6730 | seq_printf(seq, "="); | |
6731 | seq_printf(seq, ">"); | |
6732 | for (i = 0; i < y; i++) | |
6733 | seq_printf(seq, "."); | |
6734 | seq_printf(seq, "] "); | |
6735 | } | |
4588b42e | 6736 | seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)", |
ccfcc3c1 N |
6737 | (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)? |
6738 | "reshape" : | |
61df9d91 N |
6739 | (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)? |
6740 | "check" : | |
6741 | (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? | |
6742 | "resync" : "recovery"))), | |
6743 | per_milli/10, per_milli % 10, | |
dd71cf6b N |
6744 | (unsigned long long) resync/2, |
6745 | (unsigned long long) max_sectors/2); | |
1da177e4 LT |
6746 | |
6747 | /* | |
1da177e4 LT |
6748 | * dt: time from mark until now |
6749 | * db: blocks written from mark until now | |
6750 | * rt: remaining time | |
dd71cf6b N |
6751 | * |
6752 | * rt is a sector_t, so could be 32bit or 64bit. | |
6753 | * So we divide before multiply in case it is 32bit and close | |
6754 | * to the limit. | |
25985edc | 6755 | * We scale the divisor (db) by 32 to avoid losing precision |
dd71cf6b N |
6756 | * near the end of resync when the number of remaining sectors |
6757 | * is close to 'db'. | |
6758 | * We then divide rt by 32 after multiplying by db to compensate. | |
6759 | * The '+1' avoids division by zero if db is very small. | |
1da177e4 LT |
6760 | */ |
6761 | dt = ((jiffies - mddev->resync_mark) / HZ); | |
6762 | if (!dt) dt++; | |
ff4e8d9a N |
6763 | db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active)) |
6764 | - mddev->resync_mark_cnt; | |
1da177e4 | 6765 | |
dd71cf6b N |
6766 | rt = max_sectors - resync; /* number of remaining sectors */ |
6767 | sector_div(rt, db/32+1); | |
6768 | rt *= dt; | |
6769 | rt >>= 5; | |
6770 | ||
6771 | seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60, | |
6772 | ((unsigned long)rt % 60)/6); | |
1da177e4 | 6773 | |
ff4e8d9a | 6774 | seq_printf(seq, " speed=%ldK/sec", db/2/dt); |
1da177e4 LT |
6775 | } |
6776 | ||
6777 | static void *md_seq_start(struct seq_file *seq, loff_t *pos) | |
6778 | { | |
6779 | struct list_head *tmp; | |
6780 | loff_t l = *pos; | |
fd01b88c | 6781 | struct mddev *mddev; |
1da177e4 LT |
6782 | |
6783 | if (l >= 0x10000) | |
6784 | return NULL; | |
6785 | if (!l--) | |
6786 | /* header */ | |
6787 | return (void*)1; | |
6788 | ||
6789 | spin_lock(&all_mddevs_lock); | |
6790 | list_for_each(tmp,&all_mddevs) | |
6791 | if (!l--) { | |
fd01b88c | 6792 | mddev = list_entry(tmp, struct mddev, all_mddevs); |
1da177e4 LT |
6793 | mddev_get(mddev); |
6794 | spin_unlock(&all_mddevs_lock); | |
6795 | return mddev; | |
6796 | } | |
6797 | spin_unlock(&all_mddevs_lock); | |
6798 | if (!l--) | |
6799 | return (void*)2;/* tail */ | |
6800 | return NULL; | |
6801 | } | |
6802 | ||
6803 | static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
6804 | { | |
6805 | struct list_head *tmp; | |
fd01b88c | 6806 | struct mddev *next_mddev, *mddev = v; |
f72ffdd6 | 6807 | |
1da177e4 LT |
6808 | ++*pos; |
6809 | if (v == (void*)2) | |
6810 | return NULL; | |
6811 | ||
6812 | spin_lock(&all_mddevs_lock); | |
6813 | if (v == (void*)1) | |
6814 | tmp = all_mddevs.next; | |
6815 | else | |
6816 | tmp = mddev->all_mddevs.next; | |
6817 | if (tmp != &all_mddevs) | |
fd01b88c | 6818 | next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs)); |
1da177e4 LT |
6819 | else { |
6820 | next_mddev = (void*)2; | |
6821 | *pos = 0x10000; | |
f72ffdd6 | 6822 | } |
1da177e4 LT |
6823 | spin_unlock(&all_mddevs_lock); |
6824 | ||
6825 | if (v != (void*)1) | |
6826 | mddev_put(mddev); | |
6827 | return next_mddev; | |
6828 | ||
6829 | } | |
6830 | ||
6831 | static void md_seq_stop(struct seq_file *seq, void *v) | |
6832 | { | |
fd01b88c | 6833 | struct mddev *mddev = v; |
1da177e4 LT |
6834 | |
6835 | if (mddev && v != (void*)1 && v != (void*)2) | |
6836 | mddev_put(mddev); | |
6837 | } | |
6838 | ||
6839 | static int md_seq_show(struct seq_file *seq, void *v) | |
6840 | { | |
fd01b88c | 6841 | struct mddev *mddev = v; |
dd8ac336 | 6842 | sector_t sectors; |
3cb03002 | 6843 | struct md_rdev *rdev; |
1da177e4 LT |
6844 | |
6845 | if (v == (void*)1) { | |
84fc4b56 | 6846 | struct md_personality *pers; |
1da177e4 LT |
6847 | seq_printf(seq, "Personalities : "); |
6848 | spin_lock(&pers_lock); | |
2604b703 N |
6849 | list_for_each_entry(pers, &pers_list, list) |
6850 | seq_printf(seq, "[%s] ", pers->name); | |
1da177e4 LT |
6851 | |
6852 | spin_unlock(&pers_lock); | |
6853 | seq_printf(seq, "\n"); | |
f1514638 | 6854 | seq->poll_event = atomic_read(&md_event_count); |
1da177e4 LT |
6855 | return 0; |
6856 | } | |
6857 | if (v == (void*)2) { | |
6858 | status_unused(seq); | |
6859 | return 0; | |
6860 | } | |
6861 | ||
5dc5cf7d | 6862 | if (mddev_lock(mddev) < 0) |
1da177e4 | 6863 | return -EINTR; |
5dc5cf7d | 6864 | |
1da177e4 LT |
6865 | if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { |
6866 | seq_printf(seq, "%s : %sactive", mdname(mddev), | |
6867 | mddev->pers ? "" : "in"); | |
6868 | if (mddev->pers) { | |
f91de92e | 6869 | if (mddev->ro==1) |
1da177e4 | 6870 | seq_printf(seq, " (read-only)"); |
f91de92e | 6871 | if (mddev->ro==2) |
52720ae7 | 6872 | seq_printf(seq, " (auto-read-only)"); |
1da177e4 LT |
6873 | seq_printf(seq, " %s", mddev->pers->name); |
6874 | } | |
6875 | ||
dd8ac336 | 6876 | sectors = 0; |
dafb20fa | 6877 | rdev_for_each(rdev, mddev) { |
1da177e4 LT |
6878 | char b[BDEVNAME_SIZE]; |
6879 | seq_printf(seq, " %s[%d]", | |
6880 | bdevname(rdev->bdev,b), rdev->desc_nr); | |
8ddf9efe N |
6881 | if (test_bit(WriteMostly, &rdev->flags)) |
6882 | seq_printf(seq, "(W)"); | |
b2d444d7 | 6883 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
6884 | seq_printf(seq, "(F)"); |
6885 | continue; | |
2d78f8c4 N |
6886 | } |
6887 | if (rdev->raid_disk < 0) | |
b325a32e | 6888 | seq_printf(seq, "(S)"); /* spare */ |
2d78f8c4 N |
6889 | if (test_bit(Replacement, &rdev->flags)) |
6890 | seq_printf(seq, "(R)"); | |
dd8ac336 | 6891 | sectors += rdev->sectors; |
1da177e4 LT |
6892 | } |
6893 | ||
6894 | if (!list_empty(&mddev->disks)) { | |
6895 | if (mddev->pers) | |
6896 | seq_printf(seq, "\n %llu blocks", | |
f233ea5c AN |
6897 | (unsigned long long) |
6898 | mddev->array_sectors / 2); | |
1da177e4 LT |
6899 | else |
6900 | seq_printf(seq, "\n %llu blocks", | |
dd8ac336 | 6901 | (unsigned long long)sectors / 2); |
1da177e4 | 6902 | } |
1cd6bf19 N |
6903 | if (mddev->persistent) { |
6904 | if (mddev->major_version != 0 || | |
6905 | mddev->minor_version != 90) { | |
6906 | seq_printf(seq," super %d.%d", | |
6907 | mddev->major_version, | |
6908 | mddev->minor_version); | |
6909 | } | |
e691063a N |
6910 | } else if (mddev->external) |
6911 | seq_printf(seq, " super external:%s", | |
6912 | mddev->metadata_type); | |
6913 | else | |
1cd6bf19 | 6914 | seq_printf(seq, " super non-persistent"); |
1da177e4 LT |
6915 | |
6916 | if (mddev->pers) { | |
d710e138 | 6917 | mddev->pers->status(seq, mddev); |
f72ffdd6 | 6918 | seq_printf(seq, "\n "); |
8e1b39d6 N |
6919 | if (mddev->pers->sync_request) { |
6920 | if (mddev->curr_resync > 2) { | |
d710e138 | 6921 | status_resync(seq, mddev); |
8e1b39d6 | 6922 | seq_printf(seq, "\n "); |
72f36d59 | 6923 | } else if (mddev->curr_resync >= 1) |
8e1b39d6 N |
6924 | seq_printf(seq, "\tresync=DELAYED\n "); |
6925 | else if (mddev->recovery_cp < MaxSector) | |
6926 | seq_printf(seq, "\tresync=PENDING\n "); | |
6927 | } | |
32a7627c N |
6928 | } else |
6929 | seq_printf(seq, "\n "); | |
6930 | ||
57148964 | 6931 | bitmap_status(seq, mddev->bitmap); |
1da177e4 LT |
6932 | |
6933 | seq_printf(seq, "\n"); | |
6934 | } | |
6935 | mddev_unlock(mddev); | |
f72ffdd6 | 6936 | |
1da177e4 LT |
6937 | return 0; |
6938 | } | |
6939 | ||
110518bc | 6940 | static const struct seq_operations md_seq_ops = { |
1da177e4 LT |
6941 | .start = md_seq_start, |
6942 | .next = md_seq_next, | |
6943 | .stop = md_seq_stop, | |
6944 | .show = md_seq_show, | |
6945 | }; | |
6946 | ||
6947 | static int md_seq_open(struct inode *inode, struct file *file) | |
6948 | { | |
f1514638 | 6949 | struct seq_file *seq; |
1da177e4 LT |
6950 | int error; |
6951 | ||
6952 | error = seq_open(file, &md_seq_ops); | |
d7603b7e | 6953 | if (error) |
f1514638 KS |
6954 | return error; |
6955 | ||
6956 | seq = file->private_data; | |
6957 | seq->poll_event = atomic_read(&md_event_count); | |
1da177e4 LT |
6958 | return error; |
6959 | } | |
6960 | ||
e2f23b60 | 6961 | static int md_unloading; |
d7603b7e N |
6962 | static unsigned int mdstat_poll(struct file *filp, poll_table *wait) |
6963 | { | |
f1514638 | 6964 | struct seq_file *seq = filp->private_data; |
d7603b7e N |
6965 | int mask; |
6966 | ||
e2f23b60 N |
6967 | if (md_unloading) |
6968 | return POLLIN|POLLRDNORM|POLLERR|POLLPRI;; | |
d7603b7e N |
6969 | poll_wait(filp, &md_event_waiters, wait); |
6970 | ||
6971 | /* always allow read */ | |
6972 | mask = POLLIN | POLLRDNORM; | |
6973 | ||
f1514638 | 6974 | if (seq->poll_event != atomic_read(&md_event_count)) |
d7603b7e N |
6975 | mask |= POLLERR | POLLPRI; |
6976 | return mask; | |
6977 | } | |
6978 | ||
fa027c2a | 6979 | static const struct file_operations md_seq_fops = { |
e24650c2 | 6980 | .owner = THIS_MODULE, |
1da177e4 LT |
6981 | .open = md_seq_open, |
6982 | .read = seq_read, | |
6983 | .llseek = seq_lseek, | |
c3f94b40 | 6984 | .release = seq_release_private, |
d7603b7e | 6985 | .poll = mdstat_poll, |
1da177e4 LT |
6986 | }; |
6987 | ||
84fc4b56 | 6988 | int register_md_personality(struct md_personality *p) |
1da177e4 | 6989 | { |
50bd3774 CY |
6990 | printk(KERN_INFO "md: %s personality registered for level %d\n", |
6991 | p->name, p->level); | |
1da177e4 | 6992 | spin_lock(&pers_lock); |
2604b703 | 6993 | list_add_tail(&p->list, &pers_list); |
1da177e4 LT |
6994 | spin_unlock(&pers_lock); |
6995 | return 0; | |
6996 | } | |
6c144d31 | 6997 | EXPORT_SYMBOL(register_md_personality); |
1da177e4 | 6998 | |
84fc4b56 | 6999 | int unregister_md_personality(struct md_personality *p) |
1da177e4 | 7000 | { |
2604b703 | 7001 | printk(KERN_INFO "md: %s personality unregistered\n", p->name); |
1da177e4 | 7002 | spin_lock(&pers_lock); |
2604b703 | 7003 | list_del_init(&p->list); |
1da177e4 LT |
7004 | spin_unlock(&pers_lock); |
7005 | return 0; | |
7006 | } | |
6c144d31 | 7007 | EXPORT_SYMBOL(unregister_md_personality); |
1da177e4 | 7008 | |
fd01b88c | 7009 | static int is_mddev_idle(struct mddev *mddev, int init) |
1da177e4 | 7010 | { |
f72ffdd6 | 7011 | struct md_rdev *rdev; |
1da177e4 | 7012 | int idle; |
eea1bf38 | 7013 | int curr_events; |
1da177e4 LT |
7014 | |
7015 | idle = 1; | |
4b80991c N |
7016 | rcu_read_lock(); |
7017 | rdev_for_each_rcu(rdev, mddev) { | |
1da177e4 | 7018 | struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; |
eea1bf38 N |
7019 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
7020 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
7021 | atomic_read(&disk->sync_io); | |
713f6ab1 N |
7022 | /* sync IO will cause sync_io to increase before the disk_stats |
7023 | * as sync_io is counted when a request starts, and | |
7024 | * disk_stats is counted when it completes. | |
7025 | * So resync activity will cause curr_events to be smaller than | |
7026 | * when there was no such activity. | |
7027 | * non-sync IO will cause disk_stat to increase without | |
7028 | * increasing sync_io so curr_events will (eventually) | |
7029 | * be larger than it was before. Once it becomes | |
7030 | * substantially larger, the test below will cause | |
7031 | * the array to appear non-idle, and resync will slow | |
7032 | * down. | |
7033 | * If there is a lot of outstanding resync activity when | |
7034 | * we set last_event to curr_events, then all that activity | |
7035 | * completing might cause the array to appear non-idle | |
7036 | * and resync will be slowed down even though there might | |
7037 | * not have been non-resync activity. This will only | |
7038 | * happen once though. 'last_events' will soon reflect | |
7039 | * the state where there is little or no outstanding | |
7040 | * resync requests, and further resync activity will | |
7041 | * always make curr_events less than last_events. | |
c0e48521 | 7042 | * |
1da177e4 | 7043 | */ |
eea1bf38 | 7044 | if (init || curr_events - rdev->last_events > 64) { |
1da177e4 LT |
7045 | rdev->last_events = curr_events; |
7046 | idle = 0; | |
7047 | } | |
7048 | } | |
4b80991c | 7049 | rcu_read_unlock(); |
1da177e4 LT |
7050 | return idle; |
7051 | } | |
7052 | ||
fd01b88c | 7053 | void md_done_sync(struct mddev *mddev, int blocks, int ok) |
1da177e4 LT |
7054 | { |
7055 | /* another "blocks" (512byte) blocks have been synced */ | |
7056 | atomic_sub(blocks, &mddev->recovery_active); | |
7057 | wake_up(&mddev->recovery_wait); | |
7058 | if (!ok) { | |
dfc70645 | 7059 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
0a19caab | 7060 | set_bit(MD_RECOVERY_ERROR, &mddev->recovery); |
1da177e4 LT |
7061 | md_wakeup_thread(mddev->thread); |
7062 | // stop recovery, signal do_sync .... | |
7063 | } | |
7064 | } | |
6c144d31 | 7065 | EXPORT_SYMBOL(md_done_sync); |
1da177e4 | 7066 | |
06d91a5f N |
7067 | /* md_write_start(mddev, bi) |
7068 | * If we need to update some array metadata (e.g. 'active' flag | |
3d310eb7 N |
7069 | * in superblock) before writing, schedule a superblock update |
7070 | * and wait for it to complete. | |
06d91a5f | 7071 | */ |
fd01b88c | 7072 | void md_write_start(struct mddev *mddev, struct bio *bi) |
1da177e4 | 7073 | { |
0fd62b86 | 7074 | int did_change = 0; |
06d91a5f | 7075 | if (bio_data_dir(bi) != WRITE) |
3d310eb7 | 7076 | return; |
06d91a5f | 7077 | |
f91de92e N |
7078 | BUG_ON(mddev->ro == 1); |
7079 | if (mddev->ro == 2) { | |
7080 | /* need to switch to read/write */ | |
7081 | mddev->ro = 0; | |
7082 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7083 | md_wakeup_thread(mddev->thread); | |
25156198 | 7084 | md_wakeup_thread(mddev->sync_thread); |
0fd62b86 | 7085 | did_change = 1; |
f91de92e | 7086 | } |
06d91a5f | 7087 | atomic_inc(&mddev->writes_pending); |
31a59e34 N |
7088 | if (mddev->safemode == 1) |
7089 | mddev->safemode = 0; | |
06d91a5f | 7090 | if (mddev->in_sync) { |
a9701a30 | 7091 | spin_lock_irq(&mddev->write_lock); |
3d310eb7 N |
7092 | if (mddev->in_sync) { |
7093 | mddev->in_sync = 0; | |
850b2b42 | 7094 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
070dc6dd | 7095 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
3d310eb7 | 7096 | md_wakeup_thread(mddev->thread); |
0fd62b86 | 7097 | did_change = 1; |
3d310eb7 | 7098 | } |
a9701a30 | 7099 | spin_unlock_irq(&mddev->write_lock); |
06d91a5f | 7100 | } |
0fd62b86 | 7101 | if (did_change) |
00bcb4ac | 7102 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
09a44cc1 | 7103 | wait_event(mddev->sb_wait, |
09a44cc1 | 7104 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); |
1da177e4 | 7105 | } |
6c144d31 | 7106 | EXPORT_SYMBOL(md_write_start); |
1da177e4 | 7107 | |
fd01b88c | 7108 | void md_write_end(struct mddev *mddev) |
1da177e4 LT |
7109 | { |
7110 | if (atomic_dec_and_test(&mddev->writes_pending)) { | |
7111 | if (mddev->safemode == 2) | |
7112 | md_wakeup_thread(mddev->thread); | |
16f17b39 | 7113 | else if (mddev->safemode_delay) |
1da177e4 LT |
7114 | mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); |
7115 | } | |
7116 | } | |
6c144d31 | 7117 | EXPORT_SYMBOL(md_write_end); |
1da177e4 | 7118 | |
2a2275d6 N |
7119 | /* md_allow_write(mddev) |
7120 | * Calling this ensures that the array is marked 'active' so that writes | |
7121 | * may proceed without blocking. It is important to call this before | |
7122 | * attempting a GFP_KERNEL allocation while holding the mddev lock. | |
7123 | * Must be called with mddev_lock held. | |
b5470dc5 DW |
7124 | * |
7125 | * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock | |
7126 | * is dropped, so return -EAGAIN after notifying userspace. | |
2a2275d6 | 7127 | */ |
fd01b88c | 7128 | int md_allow_write(struct mddev *mddev) |
2a2275d6 N |
7129 | { |
7130 | if (!mddev->pers) | |
b5470dc5 | 7131 | return 0; |
2a2275d6 | 7132 | if (mddev->ro) |
b5470dc5 | 7133 | return 0; |
1a0fd497 | 7134 | if (!mddev->pers->sync_request) |
b5470dc5 | 7135 | return 0; |
2a2275d6 N |
7136 | |
7137 | spin_lock_irq(&mddev->write_lock); | |
7138 | if (mddev->in_sync) { | |
7139 | mddev->in_sync = 0; | |
7140 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); | |
070dc6dd | 7141 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
2a2275d6 N |
7142 | if (mddev->safemode_delay && |
7143 | mddev->safemode == 0) | |
7144 | mddev->safemode = 1; | |
7145 | spin_unlock_irq(&mddev->write_lock); | |
7146 | md_update_sb(mddev, 0); | |
00bcb4ac | 7147 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
2a2275d6 N |
7148 | } else |
7149 | spin_unlock_irq(&mddev->write_lock); | |
b5470dc5 | 7150 | |
070dc6dd | 7151 | if (test_bit(MD_CHANGE_PENDING, &mddev->flags)) |
b5470dc5 DW |
7152 | return -EAGAIN; |
7153 | else | |
7154 | return 0; | |
2a2275d6 N |
7155 | } |
7156 | EXPORT_SYMBOL_GPL(md_allow_write); | |
7157 | ||
1da177e4 LT |
7158 | #define SYNC_MARKS 10 |
7159 | #define SYNC_MARK_STEP (3*HZ) | |
54f89341 | 7160 | #define UPDATE_FREQUENCY (5*60*HZ) |
4ed8731d | 7161 | void md_do_sync(struct md_thread *thread) |
1da177e4 | 7162 | { |
4ed8731d | 7163 | struct mddev *mddev = thread->mddev; |
fd01b88c | 7164 | struct mddev *mddev2; |
1da177e4 LT |
7165 | unsigned int currspeed = 0, |
7166 | window; | |
ac7e50a3 | 7167 | sector_t max_sectors,j, io_sectors, recovery_done; |
1da177e4 | 7168 | unsigned long mark[SYNC_MARKS]; |
54f89341 | 7169 | unsigned long update_time; |
1da177e4 LT |
7170 | sector_t mark_cnt[SYNC_MARKS]; |
7171 | int last_mark,m; | |
7172 | struct list_head *tmp; | |
7173 | sector_t last_check; | |
57afd89f | 7174 | int skipped = 0; |
3cb03002 | 7175 | struct md_rdev *rdev; |
c4a39551 | 7176 | char *desc, *action = NULL; |
7c2c57c9 | 7177 | struct blk_plug plug; |
1da177e4 LT |
7178 | |
7179 | /* just incase thread restarts... */ | |
7180 | if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) | |
7181 | return; | |
3991b31e N |
7182 | if (mddev->ro) {/* never try to sync a read-only array */ |
7183 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5fd6c1dc | 7184 | return; |
3991b31e | 7185 | } |
1da177e4 | 7186 | |
61df9d91 | 7187 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
c4a39551 | 7188 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { |
61df9d91 | 7189 | desc = "data-check"; |
c4a39551 JB |
7190 | action = "check"; |
7191 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
61df9d91 | 7192 | desc = "requested-resync"; |
c4a39551 JB |
7193 | action = "repair"; |
7194 | } else | |
61df9d91 N |
7195 | desc = "resync"; |
7196 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
7197 | desc = "reshape"; | |
7198 | else | |
7199 | desc = "recovery"; | |
7200 | ||
c4a39551 JB |
7201 | mddev->last_sync_action = action ?: desc; |
7202 | ||
1da177e4 LT |
7203 | /* we overload curr_resync somewhat here. |
7204 | * 0 == not engaged in resync at all | |
7205 | * 2 == checking that there is no conflict with another sync | |
7206 | * 1 == like 2, but have yielded to allow conflicting resync to | |
7207 | * commense | |
7208 | * other == active in resync - this many blocks | |
7209 | * | |
7210 | * Before starting a resync we must have set curr_resync to | |
7211 | * 2, and then checked that every "conflicting" array has curr_resync | |
7212 | * less than ours. When we find one that is the same or higher | |
7213 | * we wait on resync_wait. To avoid deadlock, we reduce curr_resync | |
7214 | * to 1 if we choose to yield (based arbitrarily on address of mddev structure). | |
7215 | * This will mean we have to start checking from the beginning again. | |
7216 | * | |
7217 | */ | |
7218 | ||
7219 | do { | |
7220 | mddev->curr_resync = 2; | |
7221 | ||
7222 | try_again: | |
404e4b43 | 7223 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
1da177e4 | 7224 | goto skip; |
29ac4aa3 | 7225 | for_each_mddev(mddev2, tmp) { |
1da177e4 LT |
7226 | if (mddev2 == mddev) |
7227 | continue; | |
90b08710 BS |
7228 | if (!mddev->parallel_resync |
7229 | && mddev2->curr_resync | |
7230 | && match_mddev_units(mddev, mddev2)) { | |
1da177e4 LT |
7231 | DEFINE_WAIT(wq); |
7232 | if (mddev < mddev2 && mddev->curr_resync == 2) { | |
7233 | /* arbitrarily yield */ | |
7234 | mddev->curr_resync = 1; | |
7235 | wake_up(&resync_wait); | |
7236 | } | |
7237 | if (mddev > mddev2 && mddev->curr_resync == 1) | |
7238 | /* no need to wait here, we can wait the next | |
7239 | * time 'round when curr_resync == 2 | |
7240 | */ | |
7241 | continue; | |
9744197c N |
7242 | /* We need to wait 'interruptible' so as not to |
7243 | * contribute to the load average, and not to | |
7244 | * be caught by 'softlockup' | |
7245 | */ | |
7246 | prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); | |
c91abf5a | 7247 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
8712e553 | 7248 | mddev2->curr_resync >= mddev->curr_resync) { |
61df9d91 N |
7249 | printk(KERN_INFO "md: delaying %s of %s" |
7250 | " until %s has finished (they" | |
1da177e4 | 7251 | " share one or more physical units)\n", |
61df9d91 | 7252 | desc, mdname(mddev), mdname(mddev2)); |
1da177e4 | 7253 | mddev_put(mddev2); |
9744197c N |
7254 | if (signal_pending(current)) |
7255 | flush_signals(current); | |
1da177e4 LT |
7256 | schedule(); |
7257 | finish_wait(&resync_wait, &wq); | |
7258 | goto try_again; | |
7259 | } | |
7260 | finish_wait(&resync_wait, &wq); | |
7261 | } | |
7262 | } | |
7263 | } while (mddev->curr_resync < 2); | |
7264 | ||
5fd6c1dc | 7265 | j = 0; |
9d88883e | 7266 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1da177e4 | 7267 | /* resync follows the size requested by the personality, |
57afd89f | 7268 | * which defaults to physical size, but can be virtual size |
1da177e4 LT |
7269 | */ |
7270 | max_sectors = mddev->resync_max_sectors; | |
7f7583d4 | 7271 | atomic64_set(&mddev->resync_mismatches, 0); |
5fd6c1dc | 7272 | /* we don't use the checkpoint if there's a bitmap */ |
5e96ee65 NB |
7273 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) |
7274 | j = mddev->resync_min; | |
7275 | else if (!mddev->bitmap) | |
5fd6c1dc | 7276 | j = mddev->recovery_cp; |
5e96ee65 | 7277 | |
ccfcc3c1 | 7278 | } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
c804cdec | 7279 | max_sectors = mddev->resync_max_sectors; |
5fd6c1dc | 7280 | else { |
1da177e4 | 7281 | /* recovery follows the physical size of devices */ |
58c0fed4 | 7282 | max_sectors = mddev->dev_sectors; |
5fd6c1dc | 7283 | j = MaxSector; |
4e59ca7d | 7284 | rcu_read_lock(); |
dafb20fa | 7285 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc N |
7286 | if (rdev->raid_disk >= 0 && |
7287 | !test_bit(Faulty, &rdev->flags) && | |
7288 | !test_bit(In_sync, &rdev->flags) && | |
7289 | rdev->recovery_offset < j) | |
7290 | j = rdev->recovery_offset; | |
4e59ca7d | 7291 | rcu_read_unlock(); |
133d4527 N |
7292 | |
7293 | /* If there is a bitmap, we need to make sure all | |
7294 | * writes that started before we added a spare | |
7295 | * complete before we start doing a recovery. | |
7296 | * Otherwise the write might complete and (via | |
7297 | * bitmap_endwrite) set a bit in the bitmap after the | |
7298 | * recovery has checked that bit and skipped that | |
7299 | * region. | |
7300 | */ | |
7301 | if (mddev->bitmap) { | |
7302 | mddev->pers->quiesce(mddev, 1); | |
7303 | mddev->pers->quiesce(mddev, 0); | |
7304 | } | |
5fd6c1dc | 7305 | } |
1da177e4 | 7306 | |
61df9d91 N |
7307 | printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev)); |
7308 | printk(KERN_INFO "md: minimum _guaranteed_ speed:" | |
7309 | " %d KB/sec/disk.\n", speed_min(mddev)); | |
338cec32 | 7310 | printk(KERN_INFO "md: using maximum available idle IO bandwidth " |
61df9d91 N |
7311 | "(but not more than %d KB/sec) for %s.\n", |
7312 | speed_max(mddev), desc); | |
1da177e4 | 7313 | |
eea1bf38 | 7314 | is_mddev_idle(mddev, 1); /* this initializes IO event counters */ |
5fd6c1dc | 7315 | |
57afd89f | 7316 | io_sectors = 0; |
1da177e4 LT |
7317 | for (m = 0; m < SYNC_MARKS; m++) { |
7318 | mark[m] = jiffies; | |
57afd89f | 7319 | mark_cnt[m] = io_sectors; |
1da177e4 LT |
7320 | } |
7321 | last_mark = 0; | |
7322 | mddev->resync_mark = mark[last_mark]; | |
7323 | mddev->resync_mark_cnt = mark_cnt[last_mark]; | |
7324 | ||
7325 | /* | |
7326 | * Tune reconstruction: | |
7327 | */ | |
7328 | window = 32*(PAGE_SIZE/512); | |
ac42450c JB |
7329 | printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n", |
7330 | window/2, (unsigned long long)max_sectors/2); | |
1da177e4 LT |
7331 | |
7332 | atomic_set(&mddev->recovery_active, 0); | |
1da177e4 LT |
7333 | last_check = 0; |
7334 | ||
7335 | if (j>2) { | |
c91abf5a | 7336 | printk(KERN_INFO |
61df9d91 N |
7337 | "md: resuming %s of %s from checkpoint.\n", |
7338 | desc, mdname(mddev)); | |
1da177e4 | 7339 | mddev->curr_resync = j; |
72f36d59 N |
7340 | } else |
7341 | mddev->curr_resync = 3; /* no longer delayed */ | |
75d3da43 | 7342 | mddev->curr_resync_completed = j; |
72f36d59 N |
7343 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
7344 | md_new_event(mddev); | |
54f89341 | 7345 | update_time = jiffies; |
1da177e4 | 7346 | |
7c2c57c9 | 7347 | blk_start_plug(&plug); |
1da177e4 | 7348 | while (j < max_sectors) { |
57afd89f | 7349 | sector_t sectors; |
1da177e4 | 7350 | |
57afd89f | 7351 | skipped = 0; |
97e4f42d | 7352 | |
7a91ee1f N |
7353 | if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7354 | ((mddev->curr_resync > mddev->curr_resync_completed && | |
7355 | (mddev->curr_resync - mddev->curr_resync_completed) | |
7356 | > (max_sectors >> 4)) || | |
54f89341 | 7357 | time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) || |
7a91ee1f N |
7358 | (j - mddev->curr_resync_completed)*2 |
7359 | >= mddev->resync_max - mddev->curr_resync_completed | |
7360 | )) { | |
97e4f42d | 7361 | /* time to update curr_resync_completed */ |
97e4f42d N |
7362 | wait_event(mddev->recovery_wait, |
7363 | atomic_read(&mddev->recovery_active) == 0); | |
75d3da43 | 7364 | mddev->curr_resync_completed = j; |
35d78c66 | 7365 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && |
7366 | j > mddev->recovery_cp) | |
7367 | mddev->recovery_cp = j; | |
54f89341 | 7368 | update_time = jiffies; |
070dc6dd | 7369 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
acb180b0 | 7370 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
97e4f42d | 7371 | } |
acb180b0 | 7372 | |
c91abf5a N |
7373 | while (j >= mddev->resync_max && |
7374 | !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
e62e58a5 N |
7375 | /* As this condition is controlled by user-space, |
7376 | * we can block indefinitely, so use '_interruptible' | |
7377 | * to avoid triggering warnings. | |
7378 | */ | |
7379 | flush_signals(current); /* just in case */ | |
7380 | wait_event_interruptible(mddev->recovery_wait, | |
7381 | mddev->resync_max > j | |
c91abf5a N |
7382 | || test_bit(MD_RECOVERY_INTR, |
7383 | &mddev->recovery)); | |
e62e58a5 | 7384 | } |
acb180b0 | 7385 | |
c91abf5a N |
7386 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7387 | break; | |
acb180b0 | 7388 | |
57afd89f | 7389 | sectors = mddev->pers->sync_request(mddev, j, &skipped, |
c6207277 | 7390 | currspeed < speed_min(mddev)); |
57afd89f | 7391 | if (sectors == 0) { |
dfc70645 | 7392 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
c91abf5a | 7393 | break; |
1da177e4 | 7394 | } |
57afd89f N |
7395 | |
7396 | if (!skipped) { /* actual IO requested */ | |
7397 | io_sectors += sectors; | |
7398 | atomic_add(sectors, &mddev->recovery_active); | |
7399 | } | |
7400 | ||
e875ecea N |
7401 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7402 | break; | |
7403 | ||
1da177e4 | 7404 | j += sectors; |
72f36d59 N |
7405 | if (j > 2) |
7406 | mddev->curr_resync = j; | |
ff4e8d9a | 7407 | mddev->curr_mark_cnt = io_sectors; |
d7603b7e | 7408 | if (last_check == 0) |
e875ecea | 7409 | /* this is the earliest that rebuild will be |
d7603b7e N |
7410 | * visible in /proc/mdstat |
7411 | */ | |
7412 | md_new_event(mddev); | |
57afd89f N |
7413 | |
7414 | if (last_check + window > io_sectors || j == max_sectors) | |
1da177e4 LT |
7415 | continue; |
7416 | ||
57afd89f | 7417 | last_check = io_sectors; |
1da177e4 LT |
7418 | repeat: |
7419 | if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { | |
7420 | /* step marks */ | |
7421 | int next = (last_mark+1) % SYNC_MARKS; | |
7422 | ||
7423 | mddev->resync_mark = mark[next]; | |
7424 | mddev->resync_mark_cnt = mark_cnt[next]; | |
7425 | mark[next] = jiffies; | |
57afd89f | 7426 | mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); |
1da177e4 LT |
7427 | last_mark = next; |
7428 | } | |
7429 | ||
c91abf5a N |
7430 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) |
7431 | break; | |
1da177e4 LT |
7432 | |
7433 | /* | |
7434 | * this loop exits only if either when we are slower than | |
7435 | * the 'hard' speed limit, or the system was IO-idle for | |
7436 | * a jiffy. | |
7437 | * the system might be non-idle CPU-wise, but we only care | |
7438 | * about not overloading the IO subsystem. (things like an | |
7439 | * e2fsck being done on the RAID array should execute fast) | |
7440 | */ | |
1da177e4 LT |
7441 | cond_resched(); |
7442 | ||
ac7e50a3 XN |
7443 | recovery_done = io_sectors - atomic_read(&mddev->recovery_active); |
7444 | currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2 | |
57afd89f | 7445 | /((jiffies-mddev->resync_mark)/HZ +1) +1; |
1da177e4 | 7446 | |
88202a0c N |
7447 | if (currspeed > speed_min(mddev)) { |
7448 | if ((currspeed > speed_max(mddev)) || | |
eea1bf38 | 7449 | !is_mddev_idle(mddev, 0)) { |
c0e48521 | 7450 | msleep(500); |
1da177e4 LT |
7451 | goto repeat; |
7452 | } | |
7453 | } | |
7454 | } | |
c91abf5a N |
7455 | printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc, |
7456 | test_bit(MD_RECOVERY_INTR, &mddev->recovery) | |
7457 | ? "interrupted" : "done"); | |
1da177e4 LT |
7458 | /* |
7459 | * this also signals 'finished resyncing' to md_stop | |
7460 | */ | |
7c2c57c9 | 7461 | blk_finish_plug(&plug); |
1da177e4 LT |
7462 | wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); |
7463 | ||
7464 | /* tell personality that we are finished */ | |
57afd89f | 7465 | mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); |
1da177e4 | 7466 | |
dfc70645 | 7467 | if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) && |
5fd6c1dc N |
7468 | mddev->curr_resync > 2) { |
7469 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
7470 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7471 | if (mddev->curr_resync >= mddev->recovery_cp) { | |
7472 | printk(KERN_INFO | |
61df9d91 N |
7473 | "md: checkpointing %s of %s.\n", |
7474 | desc, mdname(mddev)); | |
0a19caab | 7475 | if (test_bit(MD_RECOVERY_ERROR, |
7476 | &mddev->recovery)) | |
7477 | mddev->recovery_cp = | |
7478 | mddev->curr_resync_completed; | |
7479 | else | |
7480 | mddev->recovery_cp = | |
7481 | mddev->curr_resync; | |
5fd6c1dc N |
7482 | } |
7483 | } else | |
7484 | mddev->recovery_cp = MaxSector; | |
7485 | } else { | |
7486 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
7487 | mddev->curr_resync = MaxSector; | |
4e59ca7d | 7488 | rcu_read_lock(); |
dafb20fa | 7489 | rdev_for_each_rcu(rdev, mddev) |
5fd6c1dc | 7490 | if (rdev->raid_disk >= 0 && |
70fffd0b | 7491 | mddev->delta_disks >= 0 && |
5fd6c1dc N |
7492 | !test_bit(Faulty, &rdev->flags) && |
7493 | !test_bit(In_sync, &rdev->flags) && | |
7494 | rdev->recovery_offset < mddev->curr_resync) | |
7495 | rdev->recovery_offset = mddev->curr_resync; | |
4e59ca7d | 7496 | rcu_read_unlock(); |
5fd6c1dc | 7497 | } |
1da177e4 | 7498 | } |
db91ff55 | 7499 | skip: |
17571284 | 7500 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
1da177e4 | 7501 | |
c07b70ad N |
7502 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { |
7503 | /* We completed so min/max setting can be forgotten if used. */ | |
7504 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7505 | mddev->resync_min = 0; | |
7506 | mddev->resync_max = MaxSector; | |
7507 | } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
7508 | mddev->resync_min = mddev->curr_resync_completed; | |
1da177e4 LT |
7509 | mddev->curr_resync = 0; |
7510 | wake_up(&resync_wait); | |
7511 | set_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
7512 | md_wakeup_thread(mddev->thread); | |
c6207277 | 7513 | return; |
1da177e4 | 7514 | } |
29269553 | 7515 | EXPORT_SYMBOL_GPL(md_do_sync); |
1da177e4 | 7516 | |
746d3207 N |
7517 | static int remove_and_add_spares(struct mddev *mddev, |
7518 | struct md_rdev *this) | |
b4c4c7b8 | 7519 | { |
3cb03002 | 7520 | struct md_rdev *rdev; |
b4c4c7b8 | 7521 | int spares = 0; |
f2a371c5 | 7522 | int removed = 0; |
b4c4c7b8 | 7523 | |
dafb20fa | 7524 | rdev_for_each(rdev, mddev) |
746d3207 N |
7525 | if ((this == NULL || rdev == this) && |
7526 | rdev->raid_disk >= 0 && | |
6bfe0b49 | 7527 | !test_bit(Blocked, &rdev->flags) && |
b4c4c7b8 N |
7528 | (test_bit(Faulty, &rdev->flags) || |
7529 | ! test_bit(In_sync, &rdev->flags)) && | |
7530 | atomic_read(&rdev->nr_pending)==0) { | |
7531 | if (mddev->pers->hot_remove_disk( | |
b8321b68 | 7532 | mddev, rdev) == 0) { |
36fad858 | 7533 | sysfs_unlink_rdev(mddev, rdev); |
b4c4c7b8 | 7534 | rdev->raid_disk = -1; |
f2a371c5 | 7535 | removed++; |
b4c4c7b8 N |
7536 | } |
7537 | } | |
90584fc9 JB |
7538 | if (removed && mddev->kobj.sd) |
7539 | sysfs_notify(&mddev->kobj, NULL, "degraded"); | |
b4c4c7b8 | 7540 | |
746d3207 N |
7541 | if (this) |
7542 | goto no_add; | |
7543 | ||
dafb20fa | 7544 | rdev_for_each(rdev, mddev) { |
7bfec5f3 N |
7545 | if (rdev->raid_disk >= 0 && |
7546 | !test_bit(In_sync, &rdev->flags) && | |
7547 | !test_bit(Faulty, &rdev->flags)) | |
7548 | spares++; | |
7ceb17e8 N |
7549 | if (rdev->raid_disk >= 0) |
7550 | continue; | |
7551 | if (test_bit(Faulty, &rdev->flags)) | |
7552 | continue; | |
7553 | if (mddev->ro && | |
8313b8e5 N |
7554 | ! (rdev->saved_raid_disk >= 0 && |
7555 | !test_bit(Bitmap_sync, &rdev->flags))) | |
7ceb17e8 N |
7556 | continue; |
7557 | ||
7eb41885 N |
7558 | if (rdev->saved_raid_disk < 0) |
7559 | rdev->recovery_offset = 0; | |
7ceb17e8 N |
7560 | if (mddev->pers-> |
7561 | hot_add_disk(mddev, rdev) == 0) { | |
7562 | if (sysfs_link_rdev(mddev, rdev)) | |
7563 | /* failure here is OK */; | |
7564 | spares++; | |
7565 | md_new_event(mddev); | |
7566 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
dfc70645 | 7567 | } |
b4c4c7b8 | 7568 | } |
746d3207 | 7569 | no_add: |
6dafab6b N |
7570 | if (removed) |
7571 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
b4c4c7b8 N |
7572 | return spares; |
7573 | } | |
7ebc0be7 | 7574 | |
ac05f256 N |
7575 | static void md_start_sync(struct work_struct *ws) |
7576 | { | |
7577 | struct mddev *mddev = container_of(ws, struct mddev, del_work); | |
7578 | ||
7579 | mddev->sync_thread = md_register_thread(md_do_sync, | |
7580 | mddev, | |
7581 | "resync"); | |
7582 | if (!mddev->sync_thread) { | |
7583 | printk(KERN_ERR "%s: could not start resync" | |
7584 | " thread...\n", | |
7585 | mdname(mddev)); | |
7586 | /* leave the spares where they are, it shouldn't hurt */ | |
7587 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7588 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7589 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7590 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7591 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7592 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7593 | &mddev->recovery)) | |
7594 | if (mddev->sysfs_action) | |
7595 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
7596 | } else | |
7597 | md_wakeup_thread(mddev->sync_thread); | |
7598 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
7599 | md_new_event(mddev); | |
7600 | } | |
7601 | ||
1da177e4 LT |
7602 | /* |
7603 | * This routine is regularly called by all per-raid-array threads to | |
7604 | * deal with generic issues like resync and super-block update. | |
7605 | * Raid personalities that don't have a thread (linear/raid0) do not | |
7606 | * need this as they never do any recovery or update the superblock. | |
7607 | * | |
7608 | * It does not do any resync itself, but rather "forks" off other threads | |
7609 | * to do that as needed. | |
7610 | * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in | |
7611 | * "->recovery" and create a thread at ->sync_thread. | |
dfc70645 | 7612 | * When the thread finishes it sets MD_RECOVERY_DONE |
1da177e4 LT |
7613 | * and wakeups up this thread which will reap the thread and finish up. |
7614 | * This thread also removes any faulty devices (with nr_pending == 0). | |
7615 | * | |
7616 | * The overall approach is: | |
7617 | * 1/ if the superblock needs updating, update it. | |
7618 | * 2/ If a recovery thread is running, don't do anything else. | |
7619 | * 3/ If recovery has finished, clean up, possibly marking spares active. | |
7620 | * 4/ If there are any faulty devices, remove them. | |
7621 | * 5/ If array is degraded, try to add spares devices | |
7622 | * 6/ If array has spares or is not in-sync, start a resync thread. | |
7623 | */ | |
fd01b88c | 7624 | void md_check_recovery(struct mddev *mddev) |
1da177e4 | 7625 | { |
68866e42 JB |
7626 | if (mddev->suspended) |
7627 | return; | |
7628 | ||
5f40402d | 7629 | if (mddev->bitmap) |
aa5cbd10 | 7630 | bitmap_daemon_work(mddev); |
1da177e4 | 7631 | |
fca4d848 | 7632 | if (signal_pending(current)) { |
31a59e34 | 7633 | if (mddev->pers->sync_request && !mddev->external) { |
fca4d848 N |
7634 | printk(KERN_INFO "md: %s in immediate safe mode\n", |
7635 | mdname(mddev)); | |
7636 | mddev->safemode = 2; | |
7637 | } | |
7638 | flush_signals(current); | |
7639 | } | |
7640 | ||
c89a8eee N |
7641 | if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) |
7642 | return; | |
1da177e4 | 7643 | if ( ! ( |
142d44c3 | 7644 | (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) || |
1da177e4 | 7645 | test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
fca4d848 | 7646 | test_bit(MD_RECOVERY_DONE, &mddev->recovery) || |
31a59e34 | 7647 | (mddev->external == 0 && mddev->safemode == 1) || |
fca4d848 N |
7648 | (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) |
7649 | && !mddev->in_sync && mddev->recovery_cp == MaxSector) | |
1da177e4 LT |
7650 | )) |
7651 | return; | |
fca4d848 | 7652 | |
df5b89b3 | 7653 | if (mddev_trylock(mddev)) { |
b4c4c7b8 | 7654 | int spares = 0; |
fca4d848 | 7655 | |
c89a8eee | 7656 | if (mddev->ro) { |
7ceb17e8 N |
7657 | /* On a read-only array we can: |
7658 | * - remove failed devices | |
7659 | * - add already-in_sync devices if the array itself | |
7660 | * is in-sync. | |
7661 | * As we only add devices that are already in-sync, | |
7662 | * we can activate the spares immediately. | |
c89a8eee | 7663 | */ |
7ceb17e8 | 7664 | remove_and_add_spares(mddev, NULL); |
8313b8e5 N |
7665 | /* There is no thread, but we need to call |
7666 | * ->spare_active and clear saved_raid_disk | |
7667 | */ | |
2ac295a5 | 7668 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
8313b8e5 N |
7669 | md_reap_sync_thread(mddev); |
7670 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
c89a8eee N |
7671 | goto unlock; |
7672 | } | |
7673 | ||
31a59e34 | 7674 | if (!mddev->external) { |
0fd62b86 | 7675 | int did_change = 0; |
31a59e34 N |
7676 | spin_lock_irq(&mddev->write_lock); |
7677 | if (mddev->safemode && | |
7678 | !atomic_read(&mddev->writes_pending) && | |
7679 | !mddev->in_sync && | |
7680 | mddev->recovery_cp == MaxSector) { | |
7681 | mddev->in_sync = 1; | |
0fd62b86 | 7682 | did_change = 1; |
070dc6dd | 7683 | set_bit(MD_CHANGE_CLEAN, &mddev->flags); |
31a59e34 N |
7684 | } |
7685 | if (mddev->safemode == 1) | |
7686 | mddev->safemode = 0; | |
7687 | spin_unlock_irq(&mddev->write_lock); | |
0fd62b86 | 7688 | if (did_change) |
00bcb4ac | 7689 | sysfs_notify_dirent_safe(mddev->sysfs_state); |
fca4d848 | 7690 | } |
fca4d848 | 7691 | |
7a0a5355 | 7692 | if (mddev->flags & MD_UPDATE_SB_FLAGS) |
850b2b42 | 7693 | md_update_sb(mddev, 0); |
06d91a5f | 7694 | |
1da177e4 LT |
7695 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && |
7696 | !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { | |
7697 | /* resync/recovery still happening */ | |
7698 | clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7699 | goto unlock; | |
7700 | } | |
7701 | if (mddev->sync_thread) { | |
a91d5ac0 | 7702 | md_reap_sync_thread(mddev); |
1da177e4 LT |
7703 | goto unlock; |
7704 | } | |
72a23c21 NB |
7705 | /* Set RUNNING before clearing NEEDED to avoid |
7706 | * any transients in the value of "sync_action". | |
7707 | */ | |
72f36d59 | 7708 | mddev->curr_resync_completed = 0; |
72a23c21 | 7709 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); |
24dd469d N |
7710 | /* Clear some bits that don't mean anything, but |
7711 | * might be left set | |
7712 | */ | |
24dd469d N |
7713 | clear_bit(MD_RECOVERY_INTR, &mddev->recovery); |
7714 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); | |
1da177e4 | 7715 | |
ed209584 N |
7716 | if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || |
7717 | test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) | |
ac05f256 | 7718 | goto not_running; |
1da177e4 LT |
7719 | /* no recovery is running. |
7720 | * remove any failed drives, then | |
7721 | * add spares if possible. | |
72f36d59 | 7722 | * Spares are also removed and re-added, to allow |
1da177e4 LT |
7723 | * the personality to fail the re-add. |
7724 | */ | |
1da177e4 | 7725 | |
b4c4c7b8 | 7726 | if (mddev->reshape_position != MaxSector) { |
50ac168a N |
7727 | if (mddev->pers->check_reshape == NULL || |
7728 | mddev->pers->check_reshape(mddev) != 0) | |
b4c4c7b8 | 7729 | /* Cannot proceed */ |
ac05f256 | 7730 | goto not_running; |
b4c4c7b8 | 7731 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
72a23c21 | 7732 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
746d3207 | 7733 | } else if ((spares = remove_and_add_spares(mddev, NULL))) { |
24dd469d N |
7734 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7735 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
56ac36d7 | 7736 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); |
72a23c21 | 7737 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7738 | } else if (mddev->recovery_cp < MaxSector) { |
7739 | set_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
72a23c21 | 7740 | clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
24dd469d N |
7741 | } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
7742 | /* nothing to be done ... */ | |
ac05f256 | 7743 | goto not_running; |
24dd469d | 7744 | |
1da177e4 | 7745 | if (mddev->pers->sync_request) { |
ef99bf48 | 7746 | if (spares) { |
a654b9d8 N |
7747 | /* We are adding a device or devices to an array |
7748 | * which has the bitmap stored on all devices. | |
7749 | * So make sure all bitmap pages get written | |
7750 | */ | |
7751 | bitmap_write_all(mddev->bitmap); | |
7752 | } | |
ac05f256 N |
7753 | INIT_WORK(&mddev->del_work, md_start_sync); |
7754 | queue_work(md_misc_wq, &mddev->del_work); | |
7755 | goto unlock; | |
1da177e4 | 7756 | } |
ac05f256 | 7757 | not_running: |
72a23c21 NB |
7758 | if (!mddev->sync_thread) { |
7759 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7760 | if (test_and_clear_bit(MD_RECOVERY_RECOVER, | |
7761 | &mddev->recovery)) | |
0c3573f1 | 7762 | if (mddev->sysfs_action) |
00bcb4ac | 7763 | sysfs_notify_dirent_safe(mddev->sysfs_action); |
72a23c21 | 7764 | } |
ac05f256 N |
7765 | unlock: |
7766 | wake_up(&mddev->sb_wait); | |
1da177e4 LT |
7767 | mddev_unlock(mddev); |
7768 | } | |
7769 | } | |
6c144d31 | 7770 | EXPORT_SYMBOL(md_check_recovery); |
1da177e4 | 7771 | |
a91d5ac0 JB |
7772 | void md_reap_sync_thread(struct mddev *mddev) |
7773 | { | |
7774 | struct md_rdev *rdev; | |
7775 | ||
7776 | /* resync has finished, collect result */ | |
7777 | md_unregister_thread(&mddev->sync_thread); | |
30b8feb7 | 7778 | wake_up(&resync_wait); |
a91d5ac0 JB |
7779 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) && |
7780 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
7781 | /* success...*/ | |
7782 | /* activate any spares */ | |
7783 | if (mddev->pers->spare_active(mddev)) { | |
7784 | sysfs_notify(&mddev->kobj, NULL, | |
7785 | "degraded"); | |
7786 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
7787 | } | |
7788 | } | |
7789 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
7790 | mddev->pers->finish_reshape) | |
7791 | mddev->pers->finish_reshape(mddev); | |
7792 | ||
7793 | /* If array is no-longer degraded, then any saved_raid_disk | |
f466722c | 7794 | * information must be scrapped. |
a91d5ac0 | 7795 | */ |
f466722c N |
7796 | if (!mddev->degraded) |
7797 | rdev_for_each(rdev, mddev) | |
a91d5ac0 JB |
7798 | rdev->saved_raid_disk = -1; |
7799 | ||
7800 | md_update_sb(mddev, 1); | |
7801 | clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7802 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7803 | clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7804 | clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery); | |
7805 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7806 | /* flag recovery needed just to double check */ | |
7807 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
7808 | sysfs_notify_dirent_safe(mddev->sysfs_action); | |
7809 | md_new_event(mddev); | |
7810 | if (mddev->event_work.func) | |
7811 | queue_work(md_misc_wq, &mddev->event_work); | |
7812 | } | |
6c144d31 | 7813 | EXPORT_SYMBOL(md_reap_sync_thread); |
a91d5ac0 | 7814 | |
fd01b88c | 7815 | void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev) |
6bfe0b49 | 7816 | { |
00bcb4ac | 7817 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
6bfe0b49 | 7818 | wait_event_timeout(rdev->blocked_wait, |
de393cde N |
7819 | !test_bit(Blocked, &rdev->flags) && |
7820 | !test_bit(BlockedBadBlocks, &rdev->flags), | |
6bfe0b49 DW |
7821 | msecs_to_jiffies(5000)); |
7822 | rdev_dec_pending(rdev, mddev); | |
7823 | } | |
7824 | EXPORT_SYMBOL(md_wait_for_blocked_rdev); | |
7825 | ||
c6563a8c N |
7826 | void md_finish_reshape(struct mddev *mddev) |
7827 | { | |
7828 | /* called be personality module when reshape completes. */ | |
7829 | struct md_rdev *rdev; | |
7830 | ||
7831 | rdev_for_each(rdev, mddev) { | |
7832 | if (rdev->data_offset > rdev->new_data_offset) | |
7833 | rdev->sectors += rdev->data_offset - rdev->new_data_offset; | |
7834 | else | |
7835 | rdev->sectors -= rdev->new_data_offset - rdev->data_offset; | |
7836 | rdev->data_offset = rdev->new_data_offset; | |
7837 | } | |
7838 | } | |
7839 | EXPORT_SYMBOL(md_finish_reshape); | |
2230dfe4 N |
7840 | |
7841 | /* Bad block management. | |
7842 | * We can record which blocks on each device are 'bad' and so just | |
7843 | * fail those blocks, or that stripe, rather than the whole device. | |
7844 | * Entries in the bad-block table are 64bits wide. This comprises: | |
7845 | * Length of bad-range, in sectors: 0-511 for lengths 1-512 | |
7846 | * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes) | |
7847 | * A 'shift' can be set so that larger blocks are tracked and | |
7848 | * consequently larger devices can be covered. | |
7849 | * 'Acknowledged' flag - 1 bit. - the most significant bit. | |
7850 | * | |
7851 | * Locking of the bad-block table uses a seqlock so md_is_badblock | |
7852 | * might need to retry if it is very unlucky. | |
7853 | * We will sometimes want to check for bad blocks in a bi_end_io function, | |
7854 | * so we use the write_seqlock_irq variant. | |
7855 | * | |
7856 | * When looking for a bad block we specify a range and want to | |
7857 | * know if any block in the range is bad. So we binary-search | |
7858 | * to the last range that starts at-or-before the given endpoint, | |
7859 | * (or "before the sector after the target range") | |
7860 | * then see if it ends after the given start. | |
7861 | * We return | |
7862 | * 0 if there are no known bad blocks in the range | |
7863 | * 1 if there are known bad block which are all acknowledged | |
7864 | * -1 if there are bad blocks which have not yet been acknowledged in metadata. | |
7865 | * plus the start/length of the first bad section we overlap. | |
7866 | */ | |
7867 | int md_is_badblock(struct badblocks *bb, sector_t s, int sectors, | |
7868 | sector_t *first_bad, int *bad_sectors) | |
7869 | { | |
7870 | int hi; | |
ab05613a | 7871 | int lo; |
2230dfe4 | 7872 | u64 *p = bb->page; |
ab05613a | 7873 | int rv; |
2230dfe4 N |
7874 | sector_t target = s + sectors; |
7875 | unsigned seq; | |
7876 | ||
7877 | if (bb->shift > 0) { | |
7878 | /* round the start down, and the end up */ | |
7879 | s >>= bb->shift; | |
7880 | target += (1<<bb->shift) - 1; | |
7881 | target >>= bb->shift; | |
7882 | sectors = target - s; | |
7883 | } | |
7884 | /* 'target' is now the first block after the bad range */ | |
7885 | ||
7886 | retry: | |
7887 | seq = read_seqbegin(&bb->lock); | |
ab05613a | 7888 | lo = 0; |
7889 | rv = 0; | |
2230dfe4 N |
7890 | hi = bb->count; |
7891 | ||
7892 | /* Binary search between lo and hi for 'target' | |
7893 | * i.e. for the last range that starts before 'target' | |
7894 | */ | |
7895 | /* INVARIANT: ranges before 'lo' and at-or-after 'hi' | |
7896 | * are known not to be the last range before target. | |
7897 | * VARIANT: hi-lo is the number of possible | |
7898 | * ranges, and decreases until it reaches 1 | |
7899 | */ | |
7900 | while (hi - lo > 1) { | |
7901 | int mid = (lo + hi) / 2; | |
7902 | sector_t a = BB_OFFSET(p[mid]); | |
7903 | if (a < target) | |
7904 | /* This could still be the one, earlier ranges | |
7905 | * could not. */ | |
7906 | lo = mid; | |
7907 | else | |
7908 | /* This and later ranges are definitely out. */ | |
7909 | hi = mid; | |
7910 | } | |
7911 | /* 'lo' might be the last that started before target, but 'hi' isn't */ | |
7912 | if (hi > lo) { | |
7913 | /* need to check all range that end after 's' to see if | |
7914 | * any are unacknowledged. | |
7915 | */ | |
7916 | while (lo >= 0 && | |
7917 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
7918 | if (BB_OFFSET(p[lo]) < target) { | |
7919 | /* starts before the end, and finishes after | |
7920 | * the start, so they must overlap | |
7921 | */ | |
7922 | if (rv != -1 && BB_ACK(p[lo])) | |
7923 | rv = 1; | |
7924 | else | |
7925 | rv = -1; | |
7926 | *first_bad = BB_OFFSET(p[lo]); | |
7927 | *bad_sectors = BB_LEN(p[lo]); | |
7928 | } | |
7929 | lo--; | |
7930 | } | |
7931 | } | |
7932 | ||
7933 | if (read_seqretry(&bb->lock, seq)) | |
7934 | goto retry; | |
7935 | ||
7936 | return rv; | |
7937 | } | |
7938 | EXPORT_SYMBOL_GPL(md_is_badblock); | |
7939 | ||
7940 | /* | |
7941 | * Add a range of bad blocks to the table. | |
7942 | * This might extend the table, or might contract it | |
7943 | * if two adjacent ranges can be merged. | |
7944 | * We binary-search to find the 'insertion' point, then | |
7945 | * decide how best to handle it. | |
7946 | */ | |
7947 | static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors, | |
7948 | int acknowledged) | |
7949 | { | |
7950 | u64 *p; | |
7951 | int lo, hi; | |
7952 | int rv = 1; | |
905b0297 | 7953 | unsigned long flags; |
2230dfe4 N |
7954 | |
7955 | if (bb->shift < 0) | |
7956 | /* badblocks are disabled */ | |
7957 | return 0; | |
7958 | ||
7959 | if (bb->shift) { | |
7960 | /* round the start down, and the end up */ | |
7961 | sector_t next = s + sectors; | |
7962 | s >>= bb->shift; | |
7963 | next += (1<<bb->shift) - 1; | |
7964 | next >>= bb->shift; | |
7965 | sectors = next - s; | |
7966 | } | |
7967 | ||
905b0297 | 7968 | write_seqlock_irqsave(&bb->lock, flags); |
2230dfe4 N |
7969 | |
7970 | p = bb->page; | |
7971 | lo = 0; | |
7972 | hi = bb->count; | |
7973 | /* Find the last range that starts at-or-before 's' */ | |
7974 | while (hi - lo > 1) { | |
7975 | int mid = (lo + hi) / 2; | |
7976 | sector_t a = BB_OFFSET(p[mid]); | |
7977 | if (a <= s) | |
7978 | lo = mid; | |
7979 | else | |
7980 | hi = mid; | |
7981 | } | |
7982 | if (hi > lo && BB_OFFSET(p[lo]) > s) | |
7983 | hi = lo; | |
7984 | ||
7985 | if (hi > lo) { | |
7986 | /* we found a range that might merge with the start | |
7987 | * of our new range | |
7988 | */ | |
7989 | sector_t a = BB_OFFSET(p[lo]); | |
7990 | sector_t e = a + BB_LEN(p[lo]); | |
7991 | int ack = BB_ACK(p[lo]); | |
7992 | if (e >= s) { | |
7993 | /* Yes, we can merge with a previous range */ | |
7994 | if (s == a && s + sectors >= e) | |
7995 | /* new range covers old */ | |
7996 | ack = acknowledged; | |
7997 | else | |
7998 | ack = ack && acknowledged; | |
7999 | ||
8000 | if (e < s + sectors) | |
8001 | e = s + sectors; | |
8002 | if (e - a <= BB_MAX_LEN) { | |
8003 | p[lo] = BB_MAKE(a, e-a, ack); | |
8004 | s = e; | |
8005 | } else { | |
8006 | /* does not all fit in one range, | |
8007 | * make p[lo] maximal | |
8008 | */ | |
8009 | if (BB_LEN(p[lo]) != BB_MAX_LEN) | |
8010 | p[lo] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8011 | s = a + BB_MAX_LEN; | |
8012 | } | |
8013 | sectors = e - s; | |
8014 | } | |
8015 | } | |
8016 | if (sectors && hi < bb->count) { | |
8017 | /* 'hi' points to the first range that starts after 's'. | |
8018 | * Maybe we can merge with the start of that range */ | |
8019 | sector_t a = BB_OFFSET(p[hi]); | |
8020 | sector_t e = a + BB_LEN(p[hi]); | |
8021 | int ack = BB_ACK(p[hi]); | |
8022 | if (a <= s + sectors) { | |
8023 | /* merging is possible */ | |
8024 | if (e <= s + sectors) { | |
8025 | /* full overlap */ | |
8026 | e = s + sectors; | |
8027 | ack = acknowledged; | |
8028 | } else | |
8029 | ack = ack && acknowledged; | |
8030 | ||
8031 | a = s; | |
8032 | if (e - a <= BB_MAX_LEN) { | |
8033 | p[hi] = BB_MAKE(a, e-a, ack); | |
8034 | s = e; | |
8035 | } else { | |
8036 | p[hi] = BB_MAKE(a, BB_MAX_LEN, ack); | |
8037 | s = a + BB_MAX_LEN; | |
8038 | } | |
8039 | sectors = e - s; | |
8040 | lo = hi; | |
8041 | hi++; | |
8042 | } | |
8043 | } | |
8044 | if (sectors == 0 && hi < bb->count) { | |
8045 | /* we might be able to combine lo and hi */ | |
8046 | /* Note: 's' is at the end of 'lo' */ | |
8047 | sector_t a = BB_OFFSET(p[hi]); | |
8048 | int lolen = BB_LEN(p[lo]); | |
8049 | int hilen = BB_LEN(p[hi]); | |
8050 | int newlen = lolen + hilen - (s - a); | |
8051 | if (s >= a && newlen < BB_MAX_LEN) { | |
8052 | /* yes, we can combine them */ | |
8053 | int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]); | |
8054 | p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack); | |
8055 | memmove(p + hi, p + hi + 1, | |
8056 | (bb->count - hi - 1) * 8); | |
8057 | bb->count--; | |
8058 | } | |
8059 | } | |
8060 | while (sectors) { | |
8061 | /* didn't merge (it all). | |
8062 | * Need to add a range just before 'hi' */ | |
8063 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8064 | /* No room for more */ | |
8065 | rv = 0; | |
8066 | break; | |
8067 | } else { | |
8068 | int this_sectors = sectors; | |
8069 | memmove(p + hi + 1, p + hi, | |
8070 | (bb->count - hi) * 8); | |
8071 | bb->count++; | |
8072 | ||
8073 | if (this_sectors > BB_MAX_LEN) | |
8074 | this_sectors = BB_MAX_LEN; | |
8075 | p[hi] = BB_MAKE(s, this_sectors, acknowledged); | |
8076 | sectors -= this_sectors; | |
8077 | s += this_sectors; | |
8078 | } | |
8079 | } | |
8080 | ||
8081 | bb->changed = 1; | |
de393cde N |
8082 | if (!acknowledged) |
8083 | bb->unacked_exist = 1; | |
905b0297 | 8084 | write_sequnlock_irqrestore(&bb->lock, flags); |
2230dfe4 N |
8085 | |
8086 | return rv; | |
8087 | } | |
8088 | ||
3cb03002 | 8089 | int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
c6563a8c | 8090 | int is_new) |
2230dfe4 | 8091 | { |
c6563a8c N |
8092 | int rv; |
8093 | if (is_new) | |
8094 | s += rdev->new_data_offset; | |
8095 | else | |
8096 | s += rdev->data_offset; | |
8097 | rv = md_set_badblocks(&rdev->badblocks, | |
8098 | s, sectors, 0); | |
2230dfe4 N |
8099 | if (rv) { |
8100 | /* Make sure they get written out promptly */ | |
8bd2f0a0 | 8101 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
2230dfe4 N |
8102 | set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags); |
8103 | md_wakeup_thread(rdev->mddev->thread); | |
8104 | } | |
8105 | return rv; | |
8106 | } | |
8107 | EXPORT_SYMBOL_GPL(rdev_set_badblocks); | |
8108 | ||
8109 | /* | |
8110 | * Remove a range of bad blocks from the table. | |
8111 | * This may involve extending the table if we spilt a region, | |
8112 | * but it must not fail. So if the table becomes full, we just | |
8113 | * drop the remove request. | |
8114 | */ | |
8115 | static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors) | |
8116 | { | |
8117 | u64 *p; | |
8118 | int lo, hi; | |
8119 | sector_t target = s + sectors; | |
8120 | int rv = 0; | |
8121 | ||
8122 | if (bb->shift > 0) { | |
8123 | /* When clearing we round the start up and the end down. | |
8124 | * This should not matter as the shift should align with | |
8125 | * the block size and no rounding should ever be needed. | |
8126 | * However it is better the think a block is bad when it | |
8127 | * isn't than to think a block is not bad when it is. | |
8128 | */ | |
8129 | s += (1<<bb->shift) - 1; | |
8130 | s >>= bb->shift; | |
8131 | target >>= bb->shift; | |
8132 | sectors = target - s; | |
8133 | } | |
8134 | ||
8135 | write_seqlock_irq(&bb->lock); | |
8136 | ||
8137 | p = bb->page; | |
8138 | lo = 0; | |
8139 | hi = bb->count; | |
8140 | /* Find the last range that starts before 'target' */ | |
8141 | while (hi - lo > 1) { | |
8142 | int mid = (lo + hi) / 2; | |
8143 | sector_t a = BB_OFFSET(p[mid]); | |
8144 | if (a < target) | |
8145 | lo = mid; | |
8146 | else | |
8147 | hi = mid; | |
8148 | } | |
8149 | if (hi > lo) { | |
8150 | /* p[lo] is the last range that could overlap the | |
8151 | * current range. Earlier ranges could also overlap, | |
8152 | * but only this one can overlap the end of the range. | |
8153 | */ | |
8154 | if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) { | |
8155 | /* Partial overlap, leave the tail of this range */ | |
8156 | int ack = BB_ACK(p[lo]); | |
8157 | sector_t a = BB_OFFSET(p[lo]); | |
8158 | sector_t end = a + BB_LEN(p[lo]); | |
8159 | ||
8160 | if (a < s) { | |
8161 | /* we need to split this range */ | |
8162 | if (bb->count >= MD_MAX_BADBLOCKS) { | |
8b32bf5e | 8163 | rv = -ENOSPC; |
2230dfe4 N |
8164 | goto out; |
8165 | } | |
8166 | memmove(p+lo+1, p+lo, (bb->count - lo) * 8); | |
8167 | bb->count++; | |
8168 | p[lo] = BB_MAKE(a, s-a, ack); | |
8169 | lo++; | |
8170 | } | |
8171 | p[lo] = BB_MAKE(target, end - target, ack); | |
8172 | /* there is no longer an overlap */ | |
8173 | hi = lo; | |
8174 | lo--; | |
8175 | } | |
8176 | while (lo >= 0 && | |
8177 | BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) { | |
8178 | /* This range does overlap */ | |
8179 | if (BB_OFFSET(p[lo]) < s) { | |
8180 | /* Keep the early parts of this range. */ | |
8181 | int ack = BB_ACK(p[lo]); | |
8182 | sector_t start = BB_OFFSET(p[lo]); | |
8183 | p[lo] = BB_MAKE(start, s - start, ack); | |
8184 | /* now low doesn't overlap, so.. */ | |
8185 | break; | |
8186 | } | |
8187 | lo--; | |
8188 | } | |
8189 | /* 'lo' is strictly before, 'hi' is strictly after, | |
8190 | * anything between needs to be discarded | |
8191 | */ | |
8192 | if (hi - lo > 1) { | |
8193 | memmove(p+lo+1, p+hi, (bb->count - hi) * 8); | |
8194 | bb->count -= (hi - lo - 1); | |
8195 | } | |
8196 | } | |
8197 | ||
8198 | bb->changed = 1; | |
8199 | out: | |
8200 | write_sequnlock_irq(&bb->lock); | |
8201 | return rv; | |
8202 | } | |
8203 | ||
c6563a8c N |
8204 | int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
8205 | int is_new) | |
2230dfe4 | 8206 | { |
c6563a8c N |
8207 | if (is_new) |
8208 | s += rdev->new_data_offset; | |
8209 | else | |
8210 | s += rdev->data_offset; | |
2230dfe4 | 8211 | return md_clear_badblocks(&rdev->badblocks, |
c6563a8c | 8212 | s, sectors); |
2230dfe4 N |
8213 | } |
8214 | EXPORT_SYMBOL_GPL(rdev_clear_badblocks); | |
8215 | ||
8216 | /* | |
8217 | * Acknowledge all bad blocks in a list. | |
8218 | * This only succeeds if ->changed is clear. It is used by | |
8219 | * in-kernel metadata updates | |
8220 | */ | |
8221 | void md_ack_all_badblocks(struct badblocks *bb) | |
8222 | { | |
8223 | if (bb->page == NULL || bb->changed) | |
8224 | /* no point even trying */ | |
8225 | return; | |
8226 | write_seqlock_irq(&bb->lock); | |
8227 | ||
ecb178bb | 8228 | if (bb->changed == 0 && bb->unacked_exist) { |
2230dfe4 N |
8229 | u64 *p = bb->page; |
8230 | int i; | |
8231 | for (i = 0; i < bb->count ; i++) { | |
8232 | if (!BB_ACK(p[i])) { | |
8233 | sector_t start = BB_OFFSET(p[i]); | |
8234 | int len = BB_LEN(p[i]); | |
8235 | p[i] = BB_MAKE(start, len, 1); | |
8236 | } | |
8237 | } | |
de393cde | 8238 | bb->unacked_exist = 0; |
2230dfe4 N |
8239 | } |
8240 | write_sequnlock_irq(&bb->lock); | |
8241 | } | |
8242 | EXPORT_SYMBOL_GPL(md_ack_all_badblocks); | |
8243 | ||
16c791a5 N |
8244 | /* sysfs access to bad-blocks list. |
8245 | * We present two files. | |
8246 | * 'bad-blocks' lists sector numbers and lengths of ranges that | |
8247 | * are recorded as bad. The list is truncated to fit within | |
8248 | * the one-page limit of sysfs. | |
8249 | * Writing "sector length" to this file adds an acknowledged | |
8250 | * bad block list. | |
8251 | * 'unacknowledged-bad-blocks' lists bad blocks that have not yet | |
8252 | * been acknowledged. Writing to this file adds bad blocks | |
8253 | * without acknowledging them. This is largely for testing. | |
8254 | */ | |
8255 | ||
8256 | static ssize_t | |
8257 | badblocks_show(struct badblocks *bb, char *page, int unack) | |
8258 | { | |
8259 | size_t len; | |
8260 | int i; | |
8261 | u64 *p = bb->page; | |
8262 | unsigned seq; | |
8263 | ||
8264 | if (bb->shift < 0) | |
8265 | return 0; | |
8266 | ||
8267 | retry: | |
8268 | seq = read_seqbegin(&bb->lock); | |
8269 | ||
8270 | len = 0; | |
8271 | i = 0; | |
8272 | ||
8273 | while (len < PAGE_SIZE && i < bb->count) { | |
8274 | sector_t s = BB_OFFSET(p[i]); | |
8275 | unsigned int length = BB_LEN(p[i]); | |
8276 | int ack = BB_ACK(p[i]); | |
8277 | i++; | |
8278 | ||
8279 | if (unack && ack) | |
8280 | continue; | |
8281 | ||
8282 | len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n", | |
8283 | (unsigned long long)s << bb->shift, | |
8284 | length << bb->shift); | |
8285 | } | |
de393cde N |
8286 | if (unack && len == 0) |
8287 | bb->unacked_exist = 0; | |
16c791a5 N |
8288 | |
8289 | if (read_seqretry(&bb->lock, seq)) | |
8290 | goto retry; | |
8291 | ||
8292 | return len; | |
8293 | } | |
8294 | ||
8295 | #define DO_DEBUG 1 | |
8296 | ||
8297 | static ssize_t | |
8298 | badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack) | |
8299 | { | |
8300 | unsigned long long sector; | |
8301 | int length; | |
8302 | char newline; | |
8303 | #ifdef DO_DEBUG | |
8304 | /* Allow clearing via sysfs *only* for testing/debugging. | |
8305 | * Normally only a successful write may clear a badblock | |
8306 | */ | |
8307 | int clear = 0; | |
8308 | if (page[0] == '-') { | |
8309 | clear = 1; | |
8310 | page++; | |
8311 | } | |
8312 | #endif /* DO_DEBUG */ | |
8313 | ||
8314 | switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) { | |
8315 | case 3: | |
8316 | if (newline != '\n') | |
8317 | return -EINVAL; | |
8318 | case 2: | |
8319 | if (length <= 0) | |
8320 | return -EINVAL; | |
8321 | break; | |
8322 | default: | |
8323 | return -EINVAL; | |
8324 | } | |
8325 | ||
8326 | #ifdef DO_DEBUG | |
8327 | if (clear) { | |
8328 | md_clear_badblocks(bb, sector, length); | |
8329 | return len; | |
8330 | } | |
8331 | #endif /* DO_DEBUG */ | |
8332 | if (md_set_badblocks(bb, sector, length, !unack)) | |
8333 | return len; | |
8334 | else | |
8335 | return -ENOSPC; | |
8336 | } | |
8337 | ||
75c96f85 AB |
8338 | static int md_notify_reboot(struct notifier_block *this, |
8339 | unsigned long code, void *x) | |
1da177e4 LT |
8340 | { |
8341 | struct list_head *tmp; | |
fd01b88c | 8342 | struct mddev *mddev; |
2dba6a91 | 8343 | int need_delay = 0; |
1da177e4 | 8344 | |
c744a65c N |
8345 | for_each_mddev(mddev, tmp) { |
8346 | if (mddev_trylock(mddev)) { | |
30b8aa91 N |
8347 | if (mddev->pers) |
8348 | __md_stop_writes(mddev); | |
0f62fb22 N |
8349 | if (mddev->persistent) |
8350 | mddev->safemode = 2; | |
c744a65c | 8351 | mddev_unlock(mddev); |
2dba6a91 | 8352 | } |
c744a65c | 8353 | need_delay = 1; |
1da177e4 | 8354 | } |
c744a65c N |
8355 | /* |
8356 | * certain more exotic SCSI devices are known to be | |
8357 | * volatile wrt too early system reboots. While the | |
8358 | * right place to handle this issue is the given | |
8359 | * driver, we do want to have a safe RAID driver ... | |
8360 | */ | |
8361 | if (need_delay) | |
8362 | mdelay(1000*1); | |
8363 | ||
1da177e4 LT |
8364 | return NOTIFY_DONE; |
8365 | } | |
8366 | ||
75c96f85 | 8367 | static struct notifier_block md_notifier = { |
1da177e4 LT |
8368 | .notifier_call = md_notify_reboot, |
8369 | .next = NULL, | |
8370 | .priority = INT_MAX, /* before any real devices */ | |
8371 | }; | |
8372 | ||
8373 | static void md_geninit(void) | |
8374 | { | |
36a4e1fe | 8375 | pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); |
1da177e4 | 8376 | |
c7705f34 | 8377 | proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops); |
1da177e4 LT |
8378 | } |
8379 | ||
75c96f85 | 8380 | static int __init md_init(void) |
1da177e4 | 8381 | { |
e804ac78 TH |
8382 | int ret = -ENOMEM; |
8383 | ||
ada609ee | 8384 | md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0); |
e804ac78 TH |
8385 | if (!md_wq) |
8386 | goto err_wq; | |
8387 | ||
8388 | md_misc_wq = alloc_workqueue("md_misc", 0, 0); | |
8389 | if (!md_misc_wq) | |
8390 | goto err_misc_wq; | |
8391 | ||
8392 | if ((ret = register_blkdev(MD_MAJOR, "md")) < 0) | |
8393 | goto err_md; | |
8394 | ||
8395 | if ((ret = register_blkdev(0, "mdp")) < 0) | |
8396 | goto err_mdp; | |
8397 | mdp_major = ret; | |
8398 | ||
af5628f0 | 8399 | blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE, |
e8703fe1 N |
8400 | md_probe, NULL, NULL); |
8401 | blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE, | |
1da177e4 LT |
8402 | md_probe, NULL, NULL); |
8403 | ||
1da177e4 | 8404 | register_reboot_notifier(&md_notifier); |
0b4d4147 | 8405 | raid_table_header = register_sysctl_table(raid_root_table); |
1da177e4 LT |
8406 | |
8407 | md_geninit(); | |
d710e138 | 8408 | return 0; |
1da177e4 | 8409 | |
e804ac78 TH |
8410 | err_mdp: |
8411 | unregister_blkdev(MD_MAJOR, "md"); | |
8412 | err_md: | |
8413 | destroy_workqueue(md_misc_wq); | |
8414 | err_misc_wq: | |
8415 | destroy_workqueue(md_wq); | |
8416 | err_wq: | |
8417 | return ret; | |
8418 | } | |
1da177e4 LT |
8419 | |
8420 | #ifndef MODULE | |
8421 | ||
8422 | /* | |
8423 | * Searches all registered partitions for autorun RAID arrays | |
8424 | * at boot time. | |
8425 | */ | |
4d936ec1 ME |
8426 | |
8427 | static LIST_HEAD(all_detected_devices); | |
8428 | struct detected_devices_node { | |
8429 | struct list_head list; | |
8430 | dev_t dev; | |
8431 | }; | |
1da177e4 LT |
8432 | |
8433 | void md_autodetect_dev(dev_t dev) | |
8434 | { | |
4d936ec1 ME |
8435 | struct detected_devices_node *node_detected_dev; |
8436 | ||
8437 | node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL); | |
8438 | if (node_detected_dev) { | |
8439 | node_detected_dev->dev = dev; | |
8440 | list_add_tail(&node_detected_dev->list, &all_detected_devices); | |
8441 | } else { | |
8442 | printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed" | |
8443 | ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev)); | |
8444 | } | |
1da177e4 LT |
8445 | } |
8446 | ||
1da177e4 LT |
8447 | static void autostart_arrays(int part) |
8448 | { | |
3cb03002 | 8449 | struct md_rdev *rdev; |
4d936ec1 ME |
8450 | struct detected_devices_node *node_detected_dev; |
8451 | dev_t dev; | |
8452 | int i_scanned, i_passed; | |
1da177e4 | 8453 | |
4d936ec1 ME |
8454 | i_scanned = 0; |
8455 | i_passed = 0; | |
1da177e4 | 8456 | |
4d936ec1 | 8457 | printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); |
1da177e4 | 8458 | |
4d936ec1 ME |
8459 | while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) { |
8460 | i_scanned++; | |
8461 | node_detected_dev = list_entry(all_detected_devices.next, | |
8462 | struct detected_devices_node, list); | |
8463 | list_del(&node_detected_dev->list); | |
8464 | dev = node_detected_dev->dev; | |
8465 | kfree(node_detected_dev); | |
df968c4e | 8466 | rdev = md_import_device(dev,0, 90); |
1da177e4 LT |
8467 | if (IS_ERR(rdev)) |
8468 | continue; | |
8469 | ||
403df478 | 8470 | if (test_bit(Faulty, &rdev->flags)) |
1da177e4 | 8471 | continue; |
403df478 | 8472 | |
d0fae18f | 8473 | set_bit(AutoDetected, &rdev->flags); |
1da177e4 | 8474 | list_add(&rdev->same_set, &pending_raid_disks); |
4d936ec1 | 8475 | i_passed++; |
1da177e4 | 8476 | } |
4d936ec1 ME |
8477 | |
8478 | printk(KERN_INFO "md: Scanned %d and added %d devices.\n", | |
8479 | i_scanned, i_passed); | |
1da177e4 LT |
8480 | |
8481 | autorun_devices(part); | |
8482 | } | |
8483 | ||
fdee8ae4 | 8484 | #endif /* !MODULE */ |
1da177e4 LT |
8485 | |
8486 | static __exit void md_exit(void) | |
8487 | { | |
fd01b88c | 8488 | struct mddev *mddev; |
1da177e4 | 8489 | struct list_head *tmp; |
e2f23b60 | 8490 | int delay = 1; |
8ab5e4c1 | 8491 | |
af5628f0 | 8492 | blk_unregister_region(MKDEV(MD_MAJOR,0), 512); |
e8703fe1 | 8493 | blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS); |
1da177e4 | 8494 | |
3dbd8c2e | 8495 | unregister_blkdev(MD_MAJOR,"md"); |
1da177e4 LT |
8496 | unregister_blkdev(mdp_major, "mdp"); |
8497 | unregister_reboot_notifier(&md_notifier); | |
8498 | unregister_sysctl_table(raid_table_header); | |
e2f23b60 N |
8499 | |
8500 | /* We cannot unload the modules while some process is | |
8501 | * waiting for us in select() or poll() - wake them up | |
8502 | */ | |
8503 | md_unloading = 1; | |
8504 | while (waitqueue_active(&md_event_waiters)) { | |
8505 | /* not safe to leave yet */ | |
8506 | wake_up(&md_event_waiters); | |
8507 | msleep(delay); | |
8508 | delay += delay; | |
8509 | } | |
1da177e4 | 8510 | remove_proc_entry("mdstat", NULL); |
e2f23b60 | 8511 | |
29ac4aa3 | 8512 | for_each_mddev(mddev, tmp) { |
1da177e4 | 8513 | export_array(mddev); |
d3374825 | 8514 | mddev->hold_active = 0; |
1da177e4 | 8515 | } |
e804ac78 TH |
8516 | destroy_workqueue(md_misc_wq); |
8517 | destroy_workqueue(md_wq); | |
1da177e4 LT |
8518 | } |
8519 | ||
685784aa | 8520 | subsys_initcall(md_init); |
1da177e4 LT |
8521 | module_exit(md_exit) |
8522 | ||
f91de92e N |
8523 | static int get_ro(char *buffer, struct kernel_param *kp) |
8524 | { | |
8525 | return sprintf(buffer, "%d", start_readonly); | |
8526 | } | |
8527 | static int set_ro(const char *val, struct kernel_param *kp) | |
8528 | { | |
8529 | char *e; | |
8530 | int num = simple_strtoul(val, &e, 10); | |
8531 | if (*val && (*e == '\0' || *e == '\n')) { | |
8532 | start_readonly = num; | |
4dbcdc75 | 8533 | return 0; |
f91de92e N |
8534 | } |
8535 | return -EINVAL; | |
8536 | } | |
8537 | ||
80ca3a44 N |
8538 | module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR); |
8539 | module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR); | |
efeb53c0 | 8540 | module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR); |
f91de92e | 8541 | |
1da177e4 | 8542 | MODULE_LICENSE("GPL"); |
0efb9e61 | 8543 | MODULE_DESCRIPTION("MD RAID framework"); |
aa1595e9 | 8544 | MODULE_ALIAS("md"); |
72008652 | 8545 | MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR); |