projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md-cluster: update the documentation
[linux-2.6-block.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
f72ffdd6 3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
1da177e4
LT		 4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
a6fb0934 35#include <linux/kthread.h>
bff61975 36#include <linux/blkdev.h>
1da177e4 37#include <linux/sysctl.h>
bff61975 38#include <linux/seq_file.h>
ff01bb48 39#include <linux/fs.h>
d7603b7e 40#include <linux/poll.h>
16f17b39 41#include <linux/ctype.h>
e7d2860b 42#include <linux/string.h>
fb4d8c76
N		 43#include <linux/hdreg.h>
44#include <linux/proc_fs.h>
45#include <linux/random.h>
056075c7 46#include <linux/module.h>
fb4d8c76 47#include <linux/reboot.h>
32a7627c 48#include <linux/file.h>
aa98aa31 49#include <linux/compat.h>
25570727 50#include <linux/delay.h>
bff61975
N		 51#include <linux/raid/md_p.h>
52#include <linux/raid/md_u.h>
5a0e3ad6 53#include <linux/slab.h>
43b2e5d8 54#include "md.h"
ef740c37 55#include "bitmap.h"
edb39c9d 56#include "md-cluster.h"
1da177e4 57
1da177e4 58#ifndef MODULE
d710e138 59static void autostart_arrays(int part);
1da177e4
LT		 60#endif
61
01f96c0a
N		 62/* pers_list is a list of registered personalities protected
63 * by pers_lock.
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
66 */
2604b703 67static LIST_HEAD(pers_list);
1da177e4
LT		 68static DEFINE_SPINLOCK(pers_lock);
69
edb39c9d 70struct md_cluster_operations *md_cluster_ops;
589a1c49 71EXPORT_SYMBOL(md_cluster_ops);
edb39c9d
GR		 72struct module *md_cluster_mod;
73EXPORT_SYMBOL(md_cluster_mod);
74
90b08710 75static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 76static struct workqueue_struct *md_wq;
77static struct workqueue_struct *md_misc_wq;
90b08710 78
746d3207
N		 79static int remove_and_add_spares(struct mddev *mddev,
80 struct md_rdev *this);
5aa61f42 81static void mddev_detach(struct mddev *mddev);
746d3207 82
1e50915f
RB		 83/*
84 * Default number of read corrections we'll attempt on an rdev
85 * before ejecting it from the array. We divide the read error
86 * count by 2 for every hour elapsed between read errors.
87 */
88#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 89/*
90 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
91 * is 1000 KB/sec, so the extra system load does not show up that much.
92 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 93 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 94 * subsystem is idle. There is also an 'absolute maximum' reconstruction
95 * speed limit - in case reconstruction slows down your system despite
96 * idle IO detection.
97 *
98 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 99 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 100 */
101
102static int sysctl_speed_limit_min = 1000;
103static int sysctl_speed_limit_max = 200000;
fd01b88c 104static inline int speed_min(struct mddev *mddev)
88202a0c
N		 105{
106 return mddev->sync_speed_min ?
107 mddev->sync_speed_min : sysctl_speed_limit_min;
108}
109
fd01b88c 110static inline int speed_max(struct mddev *mddev)
88202a0c
N		 111{
112 return mddev->sync_speed_max ?
113 mddev->sync_speed_max : sysctl_speed_limit_max;
114}
1da177e4
LT		 115
116static struct ctl_table_header *raid_table_header;
117
82592c38 118static struct ctl_table raid_table[] = {
1da177e4 119 {
1da177e4
LT		 120 .procname = "speed_limit_min",
121 .data = &sysctl_speed_limit_min,
122 .maxlen = sizeof(int),
80ca3a44 123 .mode = S_IRUGO|S_IWUSR,
6d456111 124 .proc_handler = proc_dointvec,
1da177e4
LT		 125 },
126 {
1da177e4
LT		 127 .procname = "speed_limit_max",
128 .data = &sysctl_speed_limit_max,
129 .maxlen = sizeof(int),
80ca3a44 130 .mode = S_IRUGO|S_IWUSR,
6d456111 131 .proc_handler = proc_dointvec,
1da177e4 132 },
894d2491 133 { }
1da177e4
LT		 134};
135
82592c38 136static struct ctl_table raid_dir_table[] = {
1da177e4 137 {
1da177e4
LT		 138 .procname = "raid",
139 .maxlen = 0,
80ca3a44 140 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 141 .child = raid_table,
142 },
894d2491 143 { }
1da177e4
LT		 144};
145
82592c38 146static struct ctl_table raid_root_table[] = {
1da177e4 147 {
1da177e4
LT		 148 .procname = "dev",
149 .maxlen = 0,
150 .mode = 0555,
151 .child = raid_dir_table,
152 },
894d2491 153 { }
1da177e4
LT		 154};
155
83d5cde4 156static const struct block_device_operations md_fops;
1da177e4 157
f91de92e
N		 158static int start_readonly;
159
a167f663
N		 160/* bio_clone_mddev
161 * like bio_clone, but with a local bio set
162 */
163
a167f663 164struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 165 struct mddev *mddev)
a167f663
N		 166{
167 struct bio *b;
a167f663
N		 168
169 if (!mddev || !mddev->bio_set)
170 return bio_alloc(gfp_mask, nr_iovecs);
171
395c72a7 172 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
a167f663
N		 173 if (!b)
174 return NULL;
a167f663
N		 175 return b;
176}
177EXPORT_SYMBOL_GPL(bio_alloc_mddev);
178
179struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
fd01b88c 180 struct mddev *mddev)
a167f663 181{
a167f663
N		 182 if (!mddev || !mddev->bio_set)
183 return bio_clone(bio, gfp_mask);
184
bf800ef1 185 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
a167f663
N		 186}
187EXPORT_SYMBOL_GPL(bio_clone_mddev);
188
d7603b7e
N		 189/*
190 * We have a system wide 'event count' that is incremented
191 * on any 'interesting' event, and readers of /proc/mdstat
192 * can use 'poll' or 'select' to find out when the event
193 * count increases.
194 *
195 * Events are:
196 * start array, stop array, error, add device, remove device,
197 * start build, activate spare
198 */
2989ddbd 199static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 200static atomic_t md_event_count;
fd01b88c 201void md_new_event(struct mddev *mddev)
d7603b7e
N		 202{
203 atomic_inc(&md_event_count);
204 wake_up(&md_event_waiters);
205}
29269553 206EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 207
c331eb04
N		 208/* Alternate version that can be called from interrupts
209 * when calling sysfs_notify isn't needed.
210 */
fd01b88c 211static void md_new_event_inintr(struct mddev *mddev)
c331eb04
N		 212{
213 atomic_inc(&md_event_count);
214 wake_up(&md_event_waiters);
215}
216
1da177e4
LT		 217/*
218 * Enables to iterate over all existing md arrays
219 * all_mddevs_lock protects this list.
220 */
221static LIST_HEAD(all_mddevs);
222static DEFINE_SPINLOCK(all_mddevs_lock);
223
1da177e4
LT		 224/*
225 * iterates through all used mddevs in the system.
226 * We take care to grab the all_mddevs_lock whenever navigating
227 * the list, and to always hold a refcount when unlocked.
228 * Any code which breaks out of this loop while own
229 * a reference to the current mddev and must mddev_put it.
230 */
fd01b88c 231#define for_each_mddev(_mddev,_tmp) \
1da177e4 232 \
f72ffdd6 233 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N		 234 _tmp = all_mddevs.next; \
235 _mddev = NULL;}); \
236 ({ if (_tmp != &all_mddevs) \
237 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 238 spin_unlock(&all_mddevs_lock); \
fd01b88c
N		 239 if (_mddev) mddev_put(_mddev); \
240 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
241 _tmp != &all_mddevs;}); \
1da177e4 242 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 243 _tmp = _tmp->next;}) \
1da177e4
LT		 244 )
245
409c57f3
N		 246/* Rather than calling directly into the personality make_request function,
247 * IO requests come here first so that we can check if the device is
248 * being suspended pending a reconfiguration.
249 * We hold a refcount over the call to ->make_request. By the time that
250 * call has finished, the bio has been linked into some internal structure
251 * and so is visible to ->quiesce(), so we don't need the refcount any more.
252 */
dece1635 253static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 254{
49077326 255 const int rw = bio_data_dir(bio);
fd01b88c 256 struct mddev *mddev = q->queuedata;
e91ece55 257 unsigned int sectors;
74672d06 258 int cpu;
49077326 259
54efd50b
KO		 260 blk_queue_split(q, &bio, q->bio_split);
261
0ca69886
N		 262 if (mddev == NULL || mddev->pers == NULL
263 || !mddev->ready) {
409c57f3 264 bio_io_error(bio);
dece1635 265 return BLK_QC_T_NONE;
409c57f3 266 }
bbfa57c0 267 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
4246a0b6
CH		 268 if (bio_sectors(bio) != 0)
269 bio->bi_error = -EROFS;
270 bio_endio(bio);
dece1635 271 return BLK_QC_T_NONE;
bbfa57c0 272 }
0ca69886 273 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 274 rcu_read_lock();
e9c7469b 275 if (mddev->suspended) {
409c57f3
N		 276 DEFINE_WAIT(__wait);
277 for (;;) {
278 prepare_to_wait(&mddev->sb_wait, &__wait,
279 TASK_UNINTERRUPTIBLE);
e9c7469b 280 if (!mddev->suspended)
409c57f3
N		 281 break;
282 rcu_read_unlock();
283 schedule();
284 rcu_read_lock();
285 }
286 finish_wait(&mddev->sb_wait, &__wait);
287 }
288 atomic_inc(&mddev->active_io);
289 rcu_read_unlock();
49077326 290
e91ece55
CM		 291 /*
292 * save the sectors now since our bio can
293 * go away inside make_request
294 */
295 sectors = bio_sectors(bio);
5a7bbad2 296 mddev->pers->make_request(mddev, bio);
49077326 297
74672d06
GZ		 298 cpu = part_stat_lock();
299 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
300 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
301 part_stat_unlock();
49077326 302
409c57f3
N		 303 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
304 wake_up(&mddev->sb_wait);
dece1635
JA		 305
306 return BLK_QC_T_NONE;
409c57f3
N		 307}
308
9e35b99c
N		 309/* mddev_suspend makes sure no new requests are submitted
310 * to the device, and that any requests that have been submitted
311 * are completely handled.
afa0f557
N		 312 * Once mddev_detach() is called and completes, the module will be
313 * completely unused.
9e35b99c 314 */
fd01b88c 315void mddev_suspend(struct mddev *mddev)
409c57f3 316{
0dc10e50
MP		 317 if (mddev->suspended++)
318 return;
409c57f3
N		 319 synchronize_rcu();
320 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
321 mddev->pers->quiesce(mddev, 1);
0d9f4f13
JB		 322
323 del_timer_sync(&mddev->safemode_timer);
409c57f3 324}
390ee602 325EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 326
fd01b88c 327void mddev_resume(struct mddev *mddev)
409c57f3 328{
0dc10e50
MP		 329 if (--mddev->suspended)
330 return;
409c57f3
N		 331 wake_up(&mddev->sb_wait);
332 mddev->pers->quiesce(mddev, 0);
0fd018af 333
47525e59 334 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
0fd018af
JB		 335 md_wakeup_thread(mddev->thread);
336 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 337}
390ee602 338EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 339
fd01b88c 340int mddev_congested(struct mddev *mddev, int bits)
3fa841d7 341{
5c675f83
N		 342 struct md_personality *pers = mddev->pers;
343 int ret = 0;
344
345 rcu_read_lock();
346 if (mddev->suspended)
347 ret = 1;
348 else if (pers && pers->congested)
349 ret = pers->congested(mddev, bits);
350 rcu_read_unlock();
351 return ret;
352}
353EXPORT_SYMBOL_GPL(mddev_congested);
354static int md_congested(void *data, int bits)
355{
356 struct mddev *mddev = data;
357 return mddev_congested(mddev, bits);
3fa841d7 358}
3fa841d7 359
a2826aa9 360/*
e9c7469b 361 * Generic flush handling for md
a2826aa9
N		 362 */
363
4246a0b6 364static void md_end_flush(struct bio *bio)
a2826aa9 365{
3cb03002 366 struct md_rdev *rdev = bio->bi_private;
fd01b88c 367 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 368
369 rdev_dec_pending(rdev, mddev);
370
371 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 372 /* The pre-request flush has finished */
e804ac78 373 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 374 }
375 bio_put(bio);
376}
377
a7a07e69
N		 378static void md_submit_flush_data(struct work_struct *ws);
379
a035fc3e 380static void submit_flushes(struct work_struct *ws)
a2826aa9 381{
fd01b88c 382 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 383 struct md_rdev *rdev;
a2826aa9 384
a7a07e69
N		 385 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
386 atomic_set(&mddev->flush_pending, 1);
a2826aa9 387 rcu_read_lock();
dafb20fa 388 rdev_for_each_rcu(rdev, mddev)
a2826aa9
N		 389 if (rdev->raid_disk >= 0 &&
390 !test_bit(Faulty, &rdev->flags)) {
391 /* Take two references, one is dropped
392 * when request finishes, one after
393 * we reclaim rcu_read_lock
394 */
395 struct bio *bi;
396 atomic_inc(&rdev->nr_pending);
397 atomic_inc(&rdev->nr_pending);
398 rcu_read_unlock();
b5e1b8ce 399 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
e9c7469b 400 bi->bi_end_io = md_end_flush;
a2826aa9
N		 401 bi->bi_private = rdev;
402 bi->bi_bdev = rdev->bdev;
403 atomic_inc(&mddev->flush_pending);
e9c7469b 404 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 405 rcu_read_lock();
406 rdev_dec_pending(rdev, mddev);
407 }
408 rcu_read_unlock();
a7a07e69
N		 409 if (atomic_dec_and_test(&mddev->flush_pending))
410 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 411}
412
e9c7469b 413static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 414{
fd01b88c 415 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 416 struct bio *bio = mddev->flush_bio;
a2826aa9 417
4f024f37 418 if (bio->bi_iter.bi_size == 0)
a2826aa9 419 /* an empty barrier - all done */
4246a0b6 420 bio_endio(bio);
a2826aa9 421 else {
e9c7469b 422 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 423 mddev->pers->make_request(mddev, bio);
a2826aa9 424 }
2b74e12e
N		 425
426 mddev->flush_bio = NULL;
427 wake_up(&mddev->sb_wait);
a2826aa9
N		 428}
429
fd01b88c 430void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9 431{
85572d7c 432 spin_lock_irq(&mddev->lock);
a2826aa9 433 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 434 !mddev->flush_bio,
85572d7c 435 mddev->lock);
e9c7469b 436 mddev->flush_bio = bio;
85572d7c 437 spin_unlock_irq(&mddev->lock);
a2826aa9 438
a035fc3e
N		 439 INIT_WORK(&mddev->flush_work, submit_flushes);
440 queue_work(md_wq, &mddev->flush_work);
a2826aa9 441}
e9c7469b 442EXPORT_SYMBOL(md_flush_request);
409c57f3 443
74018dc3 444void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
97658cdd 445{
9cbb1750
N		 446 struct mddev *mddev = cb->data;
447 md_wakeup_thread(mddev->thread);
448 kfree(cb);
97658cdd 449}
9cbb1750 450EXPORT_SYMBOL(md_unplug);
2ac87401 451
fd01b88c 452static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 453{
454 atomic_inc(&mddev->active);
455 return mddev;
456}
457
5fd3a17e 458static void mddev_delayed_delete(struct work_struct *ws);
d3374825 459
fd01b88c 460static void mddev_put(struct mddev *mddev)
1da177e4 461{
a167f663
N		 462 struct bio_set *bs = NULL;
463
1da177e4
LT		 464 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
465 return;
d3374825 466 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 467 mddev->ctime == 0 && !mddev->hold_active) {
468 /* Array is not configured at all, and not held active,
469 * so destroy it */
af8a2434 470 list_del_init(&mddev->all_mddevs);
a167f663
N		 471 bs = mddev->bio_set;
472 mddev->bio_set = NULL;
d3374825 473 if (mddev->gendisk) {
e804ac78
TH		 474 /* We did a probe so need to clean up. Call
475 * queue_work inside the spinlock so that
476 * flush_workqueue() after mddev_find will
477 * succeed in waiting for the work to be done.
d3374825
N		 478 */
479 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 480 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 481 } else
482 kfree(mddev);
483 }
484 spin_unlock(&all_mddevs_lock);
a167f663
N		 485 if (bs)
486 bioset_free(bs);
1da177e4
LT		 487}
488
25b2edfa
SL		 489static void md_safemode_timeout(unsigned long data);
490
fd01b88c 491void mddev_init(struct mddev *mddev)
fafd7fb0
N		 492{
493 mutex_init(&mddev->open_mutex);
494 mutex_init(&mddev->reconfig_mutex);
495 mutex_init(&mddev->bitmap_info.mutex);
496 INIT_LIST_HEAD(&mddev->disks);
497 INIT_LIST_HEAD(&mddev->all_mddevs);
25b2edfa
SL		 498 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
499 (unsigned long) mddev);
fafd7fb0
N		 500 atomic_set(&mddev->active, 1);
501 atomic_set(&mddev->openers, 0);
502 atomic_set(&mddev->active_io, 0);
85572d7c 503 spin_lock_init(&mddev->lock);
fafd7fb0
N		 504 atomic_set(&mddev->flush_pending, 0);
505 init_waitqueue_head(&mddev->sb_wait);
506 init_waitqueue_head(&mddev->recovery_wait);
507 mddev->reshape_position = MaxSector;
2c810cdd 508 mddev->reshape_backwards = 0;
c4a39551 509 mddev->last_sync_action = "none";
fafd7fb0
N		 510 mddev->resync_min = 0;
511 mddev->resync_max = MaxSector;
512 mddev->level = LEVEL_NONE;
513}
390ee602 514EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 515
f72ffdd6 516static struct mddev *mddev_find(dev_t unit)
1da177e4 517{
fd01b88c 518 struct mddev *mddev, *new = NULL;
1da177e4 519
8f5f02c4
N		 520 if (unit && MAJOR(unit) != MD_MAJOR)
521 unit &= ~((1<<MdpMinorShift)-1);
522
1da177e4
LT		 523 retry:
524 spin_lock(&all_mddevs_lock);
efeb53c0
N		 525
526 if (unit) {
527 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
528 if (mddev->unit == unit) {
529 mddev_get(mddev);
530 spin_unlock(&all_mddevs_lock);
531 kfree(new);
532 return mddev;
533 }
534
535 if (new) {
536 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 537 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 538 new->hold_active = UNTIL_IOCTL;
539 return new;
1da177e4 540 }
efeb53c0
N		 541 } else if (new) {
542 /* find an unused unit number */
543 static int next_minor = 512;
544 int start = next_minor;
545 int is_free = 0;
546 int dev = 0;
547 while (!is_free) {
548 dev = MKDEV(MD_MAJOR, next_minor);
549 next_minor++;
550 if (next_minor > MINORMASK)
551 next_minor = 0;
552 if (next_minor == start) {
553 /* Oh dear, all in use. */
554 spin_unlock(&all_mddevs_lock);
555 kfree(new);
556 return NULL;
557 }
f72ffdd6 558
efeb53c0
N		 559 is_free = 1;
560 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
561 if (mddev->unit == dev) {
562 is_free = 0;
563 break;
564 }
565 }
566 new->unit = dev;
567 new->md_minor = MINOR(dev);
568 new->hold_active = UNTIL_STOP;
1da177e4
LT		 569 list_add(&new->all_mddevs, &all_mddevs);
570 spin_unlock(&all_mddevs_lock);
571 return new;
572 }
573 spin_unlock(&all_mddevs_lock);
574
9ffae0cf 575 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 576 if (!new)
577 return NULL;
578
1da177e4
LT		 579 new->unit = unit;
580 if (MAJOR(unit) == MD_MAJOR)
581 new->md_minor = MINOR(unit);
582 else
583 new->md_minor = MINOR(unit) >> MdpMinorShift;
584
fafd7fb0 585 mddev_init(new);
1da177e4 586
1da177e4
LT		 587 goto retry;
588}
589
b6eb127d
N		 590static struct attribute_group md_redundancy_group;
591
5c47daf6 592void mddev_unlock(struct mddev *mddev)
1da177e4 593{
a64c876f 594 if (mddev->to_remove) {
b6eb127d
N		 595 /* These cannot be removed under reconfig_mutex as
596 * an access to the files will try to take reconfig_mutex
597 * while holding the file unremovable, which leads to
598 * a deadlock.
bb4f1e9d
N		 599 * So hold set sysfs_active while the remove in happeing,
600 * and anything else which might set ->to_remove or my
601 * otherwise change the sysfs namespace will fail with
602 * -EBUSY if sysfs_active is still set.
603 * We set sysfs_active under reconfig_mutex and elsewhere
604 * test it under the same mutex to ensure its correct value
605 * is seen.
b6eb127d 606 */
a64c876f
N		 607 struct attribute_group *to_remove = mddev->to_remove;
608 mddev->to_remove = NULL;
bb4f1e9d 609 mddev->sysfs_active = 1;
b6eb127d
N		 610 mutex_unlock(&mddev->reconfig_mutex);
611
00bcb4ac
N		 612 if (mddev->kobj.sd) {
613 if (to_remove != &md_redundancy_group)
614 sysfs_remove_group(&mddev->kobj, to_remove);
615 if (mddev->pers == NULL ||
616 mddev->pers->sync_request == NULL) {
617 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
618 if (mddev->sysfs_action)
619 sysfs_put(mddev->sysfs_action);
620 mddev->sysfs_action = NULL;
621 }
a64c876f 622 }
bb4f1e9d 623 mddev->sysfs_active = 0;
b6eb127d
N		 624 } else
625 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 626
751e67ca
CD		 627 /* As we've dropped the mutex we need a spinlock to
628 * make sure the thread doesn't disappear
01f96c0a
N		 629 */
630 spin_lock(&pers_lock);
005eca5e 631 md_wakeup_thread(mddev->thread);
01f96c0a 632 spin_unlock(&pers_lock);
1da177e4 633}
5c47daf6 634EXPORT_SYMBOL_GPL(mddev_unlock);
1da177e4 635
57d051dc 636struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
1ca69c4b
N		 637{
638 struct md_rdev *rdev;
639
640 rdev_for_each_rcu(rdev, mddev)
641 if (rdev->desc_nr == nr)
642 return rdev;
643
644 return NULL;
645}
57d051dc 646EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
1ca69c4b
N		 647
648static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
1da177e4 649{
3cb03002 650 struct md_rdev *rdev;
1da177e4 651
dafb20fa 652 rdev_for_each(rdev, mddev)
1da177e4
LT		 653 if (rdev->bdev->bd_dev == dev)
654 return rdev;
159ec1fc 655
1da177e4
LT		 656 return NULL;
657}
658
1ca69c4b
N		 659static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
660{
661 struct md_rdev *rdev;
662
663 rdev_for_each_rcu(rdev, mddev)
664 if (rdev->bdev->bd_dev == dev)
665 return rdev;
666
667 return NULL;
668}
669
84fc4b56 670static struct md_personality *find_pers(int level, char *clevel)
2604b703 671{
84fc4b56 672 struct md_personality *pers;
d9d166c2
N		 673 list_for_each_entry(pers, &pers_list, list) {
674 if (level != LEVEL_NONE && pers->level == level)
2604b703 675 return pers;
d9d166c2
N		 676 if (strcmp(pers->name, clevel)==0)
677 return pers;
678 }
2604b703
N		 679 return NULL;
680}
681
b73df2d3 682/* return the offset of the super block in 512byte sectors */
3cb03002 683static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 684{
57b2caa3 685 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 686 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 687}
688
f72ffdd6 689static int alloc_disk_sb(struct md_rdev *rdev)
1da177e4 690{
1da177e4
LT		 691 rdev->sb_page = alloc_page(GFP_KERNEL);
692 if (!rdev->sb_page) {
693 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 694 return -ENOMEM;
1da177e4
LT		 695 }
696
697 return 0;
698}
699
545c8795 700void md_rdev_clear(struct md_rdev *rdev)
1da177e4
LT		 701{
702 if (rdev->sb_page) {
2d1f3b5d 703 put_page(rdev->sb_page);
1da177e4
LT		 704 rdev->sb_loaded = 0;
705 rdev->sb_page = NULL;
0f420358 706 rdev->sb_start = 0;
dd8ac336 707 rdev->sectors = 0;
1da177e4 708 }
2699b672
N		 709 if (rdev->bb_page) {
710 put_page(rdev->bb_page);
711 rdev->bb_page = NULL;
712 }
4fa2f327
N		 713 kfree(rdev->badblocks.page);
714 rdev->badblocks.page = NULL;
1da177e4 715}
545c8795 716EXPORT_SYMBOL_GPL(md_rdev_clear);
1da177e4 717
4246a0b6 718static void super_written(struct bio *bio)
7bfa19f2 719{
3cb03002 720 struct md_rdev *rdev = bio->bi_private;
fd01b88c 721 struct mddev *mddev = rdev->mddev;
7bfa19f2 722
4246a0b6
CH		 723 if (bio->bi_error) {
724 printk("md: super_written gets error=%d\n", bio->bi_error);
a9701a30 725 md_error(mddev, rdev);
3a0f5bbb 726 }
7bfa19f2 727
a9701a30
N		 728 if (atomic_dec_and_test(&mddev->pending_writes))
729 wake_up(&mddev->sb_wait);
f8b58edf 730 bio_put(bio);
7bfa19f2
N		 731}
732
fd01b88c 733void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 734 sector_t sector, int size, struct page *page)
735{
736 /* write first size bytes of page to sector of rdev
737 * Increment mddev->pending_writes before returning
738 * and decrement it on completion, waking up sb_wait
739 * if zero is reached.
740 * If an error occurred, call md_error
741 */
a167f663 742 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 743
a6ff7e08 744 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
4f024f37 745 bio->bi_iter.bi_sector = sector;
7bfa19f2
N		 746 bio_add_page(bio, page, size, 0);
747 bio->bi_private = rdev;
748 bio->bi_end_io = super_written;
a9701a30 749
7bfa19f2 750 atomic_inc(&mddev->pending_writes);
a5bf4df0 751 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 752}
753
fd01b88c 754void md_super_wait(struct mddev *mddev)
a9701a30 755{
e9c7469b 756 /* wait for all superblock writes that were scheduled to complete */
1967cd56 757 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
7bfa19f2
N		 758}
759
3cb03002 760int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 761 struct page *page, int rw, bool metadata_op)
1da177e4 762{
a167f663 763 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 764 int ret;
765
a6ff7e08
JB		 766 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
767 rdev->meta_bdev : rdev->bdev;
ccebd4c4 768 if (metadata_op)
4f024f37 769 bio->bi_iter.bi_sector = sector + rdev->sb_start;
1fdd6fc9
N		 770 else if (rdev->mddev->reshape_position != MaxSector &&
771 (rdev->mddev->reshape_backwards ==
772 (sector >= rdev->mddev->reshape_position)))
4f024f37 773 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
ccebd4c4 774 else
4f024f37 775 bio->bi_iter.bi_sector = sector + rdev->data_offset;
1da177e4 776 bio_add_page(bio, page, size, 0);
c170bbb4 777 submit_bio_wait(rw, bio);
1da177e4 778
4246a0b6 779 ret = !bio->bi_error;
1da177e4
LT		 780 bio_put(bio);
781 return ret;
782}
a8745db2 783EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 784
f72ffdd6 785static int read_disk_sb(struct md_rdev *rdev, int size)
1da177e4
LT		 786{
787 char b[BDEVNAME_SIZE];
403df478 788
1da177e4
LT		 789 if (rdev->sb_loaded)
790 return 0;
791
ccebd4c4 792 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 793 goto fail;
794 rdev->sb_loaded = 1;
795 return 0;
796
797fail:
798 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
799 bdevname(rdev->bdev,b));
800 return -EINVAL;
801}
802
803static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
804{
f72ffdd6 805 return sb1->set_uuid0 == sb2->set_uuid0 &&
05710466
AN		 806 sb1->set_uuid1 == sb2->set_uuid1 &&
807 sb1->set_uuid2 == sb2->set_uuid2 &&
808 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 809}
810
1da177e4
LT		 811static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
812{
813 int ret;
814 mdp_super_t *tmp1, *tmp2;
815
816 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
817 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
818
819 if (!tmp1 || !tmp2) {
820 ret = 0;
35020f1a 821 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 822 goto abort;
823 }
824
825 *tmp1 = *sb1;
826 *tmp2 = *sb2;
827
828 /*
829 * nr_disks is not constant
830 */
831 tmp1->nr_disks = 0;
832 tmp2->nr_disks = 0;
833
ce0c8e05 834 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 835abort:
990a8baf
JJ		 836 kfree(tmp1);
837 kfree(tmp2);
1da177e4
LT		 838 return ret;
839}
840
4d167f09
N		 841static u32 md_csum_fold(u32 csum)
842{
843 csum = (csum & 0xffff) + (csum >> 16);
844 return (csum & 0xffff) + (csum >> 16);
845}
846
f72ffdd6 847static unsigned int calc_sb_csum(mdp_super_t *sb)
1da177e4 848{
4d167f09
N		 849 u64 newcsum = 0;
850 u32 *sb32 = (u32*)sb;
851 int i;
1da177e4
LT		 852 unsigned int disk_csum, csum;
853
854 disk_csum = sb->sb_csum;
855 sb->sb_csum = 0;
4d167f09
N		 856
857 for (i = 0; i < MD_SB_BYTES/4 ; i++)
858 newcsum += sb32[i];
859 csum = (newcsum & 0xffffffff) + (newcsum>>32);
860
4d167f09
N		 861#ifdef CONFIG_ALPHA
862 /* This used to use csum_partial, which was wrong for several
863 * reasons including that different results are returned on
864 * different architectures. It isn't critical that we get exactly
865 * the same return value as before (we always csum_fold before
866 * testing, and that removes any differences). However as we
867 * know that csum_partial always returned a 16bit value on
868 * alphas, do a fold to maximise conformity to previous behaviour.
869 */
870 sb->sb_csum = md_csum_fold(disk_csum);
871#else
1da177e4 872 sb->sb_csum = disk_csum;
4d167f09 873#endif
1da177e4
LT		 874 return csum;
875}
876
1da177e4
LT		 877/*
878 * Handle superblock details.
879 * We want to be able to handle multiple superblock formats
880 * so we have a common interface to them all, and an array of
881 * different handlers.
882 * We rely on user-space to write the initial superblock, and support
883 * reading and updating of superblocks.
884 * Interface methods are:
3cb03002 885 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 886 * loads and validates a superblock on dev.
887 * if refdev != NULL, compare superblocks on both devices
888 * Return:
889 * 0 - dev has a superblock that is compatible with refdev
890 * 1 - dev has a superblock that is compatible and newer than refdev
891 * so dev should be used as the refdev in future
892 * -EINVAL superblock incompatible or invalid
893 * -othererror e.g. -EIO
894 *
fd01b88c 895 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 896 * Verify that dev is acceptable into mddev.
897 * The first time, mddev->raid_disks will be 0, and data from
898 * dev should be merged in. Subsequent calls check that dev
899 * is new enough. Return 0 or -EINVAL
900 *
fd01b88c 901 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 902 * Update the superblock for rdev with data in mddev
903 * This does not write to disc.
904 *
905 */
906
907struct super_type {
0cd17fec
CW		 908 char *name;
909 struct module *owner;
c6563a8c
N		 910 int (*load_super)(struct md_rdev *rdev,
911 struct md_rdev *refdev,
0cd17fec 912 int minor_version);
c6563a8c
N		 913 int (*validate_super)(struct mddev *mddev,
914 struct md_rdev *rdev);
915 void (*sync_super)(struct mddev *mddev,
916 struct md_rdev *rdev);
3cb03002 917 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 918 sector_t num_sectors);
c6563a8c
N		 919 int (*allow_new_offset)(struct md_rdev *rdev,
920 unsigned long long new_offset);
1da177e4
LT		 921};
922
0894cc30
AN		 923/*
924 * Check that the given mddev has no bitmap.
925 *
926 * This function is called from the run method of all personalities that do not
927 * support bitmaps. It prints an error message and returns non-zero if mddev
928 * has a bitmap. Otherwise, it returns 0.
929 *
930 */
fd01b88c 931int md_check_no_bitmap(struct mddev *mddev)
0894cc30 932{
c3d9714e 933 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 934 return 0;
935 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
936 mdname(mddev), mddev->pers->name);
937 return 1;
938}
939EXPORT_SYMBOL(md_check_no_bitmap);
940
1da177e4 941/*
f72ffdd6 942 * load_super for 0.90.0
1da177e4 943 */
3cb03002 944static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 945{
946 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
947 mdp_super_t *sb;
948 int ret;
1da177e4
LT		 949
950 /*
0f420358 951 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 952 * it's at the end of the disk.
953 *
954 * It also happens to be a multiple of 4Kb.
955 */
57b2caa3 956 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 957
0002b271 958 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 959 if (ret) return ret;
960
961 ret = -EINVAL;
962
963 bdevname(rdev->bdev, b);
65a06f06 964 sb = page_address(rdev->sb_page);
1da177e4
LT		 965
966 if (sb->md_magic != MD_SB_MAGIC) {
967 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
968 b);
969 goto abort;
970 }
971
972 if (sb->major_version != 0 ||
f6705578
N		 973 sb->minor_version < 90 ||
974 sb->minor_version > 91) {
1da177e4
LT		 975 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
976 sb->major_version, sb->minor_version,
977 b);
978 goto abort;
979 }
980
981 if (sb->raid_disks <= 0)
982 goto abort;
983
4d167f09 984 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 985 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
986 b);
987 goto abort;
988 }
989
990 rdev->preferred_minor = sb->md_minor;
991 rdev->data_offset = 0;
c6563a8c 992 rdev->new_data_offset = 0;
0002b271 993 rdev->sb_size = MD_SB_BYTES;
9f2f3830 994 rdev->badblocks.shift = -1;
1da177e4
LT		 995
996 if (sb->level == LEVEL_MULTIPATH)
997 rdev->desc_nr = -1;
998 else
999 rdev->desc_nr = sb->this_disk.number;
1000
9a7b2b0f 1001 if (!refdev) {
1da177e4 1002 ret = 1;
9a7b2b0f 1003 } else {
1da177e4 1004 __u64 ev1, ev2;
65a06f06 1005 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1006 if (!uuid_equal(refsb, sb)) {
1007 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1008 b, bdevname(refdev->bdev,b2));
1009 goto abort;
1010 }
1011 if (!sb_equal(refsb, sb)) {
1012 printk(KERN_WARNING "md: %s has same UUID"
1013 " but different superblock to %s\n",
1014 b, bdevname(refdev->bdev, b2));
1015 goto abort;
1016 }
1017 ev1 = md_event(sb);
1018 ev2 = md_event(refsb);
1019 if (ev1 > ev2)
1020 ret = 1;
f72ffdd6 1021 else
1da177e4
LT		 1022 ret = 0;
1023 }
8190e754 1024 rdev->sectors = rdev->sb_start;
667a5313
N		 1025 /* Limit to 4TB as metadata cannot record more than that.
1026 * (not needed for Linear and RAID0 as metadata doesn't
1027 * record this size)
1028 */
1029 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
27a7b260 1030 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1031
27a7b260 1032 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1033 /* "this cannot possibly happen" ... */
1034 ret = -EINVAL;
1035
1da177e4
LT		 1036 abort:
1037 return ret;
1038}
1039
1040/*
1041 * validate_super for 0.90.0
1042 */
fd01b88c 1043static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1044{
1045 mdp_disk_t *desc;
65a06f06 1046 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1047 __u64 ev1 = md_event(sb);
1da177e4 1048
41158c7e 1049 rdev->raid_disk = -1;
c5d79adb
N		 1050 clear_bit(Faulty, &rdev->flags);
1051 clear_bit(In_sync, &rdev->flags);
8313b8e5 1052 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1053 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1054
1da177e4
LT		 1055 if (mddev->raid_disks == 0) {
1056 mddev->major_version = 0;
1057 mddev->minor_version = sb->minor_version;
1058 mddev->patch_version = sb->patch_version;
e691063a 1059 mddev->external = 0;
9d8f0363 1060 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1061 mddev->ctime = sb->ctime;
1062 mddev->utime = sb->utime;
1063 mddev->level = sb->level;
d9d166c2 1064 mddev->clevel[0] = 0;
1da177e4
LT		 1065 mddev->layout = sb->layout;
1066 mddev->raid_disks = sb->raid_disks;
27a7b260 1067 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1068 mddev->events = ev1;
c3d9714e 1069 mddev->bitmap_info.offset = 0;
6409bb05
N		 1070 mddev->bitmap_info.space = 0;
1071 /* bitmap can use 60 K after the 4K superblocks */
c3d9714e 1072 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 1073 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
2c810cdd 1074 mddev->reshape_backwards = 0;
1da177e4 1075
f6705578
N		 1076 if (mddev->minor_version >= 91) {
1077 mddev->reshape_position = sb->reshape_position;
1078 mddev->delta_disks = sb->delta_disks;
1079 mddev->new_level = sb->new_level;
1080 mddev->new_layout = sb->new_layout;
664e7c41 1081 mddev->new_chunk_sectors = sb->new_chunk >> 9;
2c810cdd
N		 1082 if (mddev->delta_disks < 0)
1083 mddev->reshape_backwards = 1;
f6705578
N		 1084 } else {
1085 mddev->reshape_position = MaxSector;
1086 mddev->delta_disks = 0;
1087 mddev->new_level = mddev->level;
1088 mddev->new_layout = mddev->layout;
664e7c41 1089 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1090 }
1091
1da177e4
LT		 1092 if (sb->state & (1<<MD_SB_CLEAN))
1093 mddev->recovery_cp = MaxSector;
1094 else {
f72ffdd6 1095 if (sb->events_hi == sb->cp_events_hi &&
1da177e4
LT		 1096 sb->events_lo == sb->cp_events_lo) {
1097 mddev->recovery_cp = sb->recovery_cp;
1098 } else
1099 mddev->recovery_cp = 0;
1100 }
1101
1102 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1103 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1104 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1105 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1106
1107 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1108
1109 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
6409bb05 1110 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1111 mddev->bitmap_info.offset =
1112 mddev->bitmap_info.default_offset;
6409bb05 1113 mddev->bitmap_info.space =
c9ad020f 1114 mddev->bitmap_info.default_space;
6409bb05 1115 }
a654b9d8 1116
41158c7e 1117 } else if (mddev->pers == NULL) {
be6800a7
N		 1118 /* Insist on good event counter while assembling, except
1119 * for spares (which don't need an event count) */
1da177e4 1120 ++ev1;
be6800a7
N		 1121 if (sb->disks[rdev->desc_nr].state & (
1122 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
f72ffdd6 1123 if (ev1 < mddev->events)
be6800a7 1124 return -EINVAL;
41158c7e
N		 1125 } else if (mddev->bitmap) {
1126 /* if adding to array with a bitmap, then we can accept an
1127 * older device ... but not too old.
1128 */
41158c7e
N		 1129 if (ev1 < mddev->bitmap->events_cleared)
1130 return 0;
8313b8e5
N		 1131 if (ev1 < mddev->events)
1132 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1133 } else {
1134 if (ev1 < mddev->events)
1135 /* just a hot-add of a new device, leave raid_disk at -1 */
1136 return 0;
1137 }
41158c7e 1138
1da177e4 1139 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1140 desc = sb->disks + rdev->desc_nr;
1141
1142 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1143 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1144 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1145 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1146 set_bit(In_sync, &rdev->flags);
1da177e4 1147 rdev->raid_disk = desc->raid_disk;
f466722c 1148 rdev->saved_raid_disk = desc->raid_disk;
0261cd9f
N		 1149 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1150 /* active but not in sync implies recovery up to
1151 * reshape position. We don't know exactly where
1152 * that is, so set to zero for now */
1153 if (mddev->minor_version >= 91) {
1154 rdev->recovery_offset = 0;
1155 rdev->raid_disk = desc->raid_disk;
1156 }
1da177e4 1157 }
8ddf9efe
N		 1158 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1159 set_bit(WriteMostly, &rdev->flags);
41158c7e 1160 } else /* MULTIPATH are always insync */
b2d444d7 1161 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1162 return 0;
1163}
1164
1165/*
1166 * sync_super for 0.90.0
1167 */
fd01b88c 1168static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1169{
1170 mdp_super_t *sb;
3cb03002 1171 struct md_rdev *rdev2;
1da177e4 1172 int next_spare = mddev->raid_disks;
19133a42 1173
1da177e4
LT		 1174 /* make rdev->sb match mddev data..
1175 *
1176 * 1/ zero out disks
1177 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1178 * 3/ any empty disks < next_spare become removed
1179 *
1180 * disks[0] gets initialised to REMOVED because
1181 * we cannot be sure from other fields if it has
1182 * been initialised or not.
1183 */
1184 int i;
1185 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1186
61181565
N		 1187 rdev->sb_size = MD_SB_BYTES;
1188
65a06f06 1189 sb = page_address(rdev->sb_page);
1da177e4
LT		 1190
1191 memset(sb, 0, sizeof(*sb));
1192
1193 sb->md_magic = MD_SB_MAGIC;
1194 sb->major_version = mddev->major_version;
1da177e4
LT		 1195 sb->patch_version = mddev->patch_version;
1196 sb->gvalid_words = 0; /* ignored */
1197 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1198 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1199 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1200 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1201
1202 sb->ctime = mddev->ctime;
1203 sb->level = mddev->level;
58c0fed4 1204 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1205 sb->raid_disks = mddev->raid_disks;
1206 sb->md_minor = mddev->md_minor;
e691063a 1207 sb->not_persistent = 0;
1da177e4
LT		 1208 sb->utime = mddev->utime;
1209 sb->state = 0;
1210 sb->events_hi = (mddev->events>>32);
1211 sb->events_lo = (u32)mddev->events;
1212
f6705578
N		 1213 if (mddev->reshape_position == MaxSector)
1214 sb->minor_version = 90;
1215 else {
1216 sb->minor_version = 91;
1217 sb->reshape_position = mddev->reshape_position;
1218 sb->new_level = mddev->new_level;
1219 sb->delta_disks = mddev->delta_disks;
1220 sb->new_layout = mddev->new_layout;
664e7c41 1221 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1222 }
1223 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1224 if (mddev->in_sync)
1225 {
1226 sb->recovery_cp = mddev->recovery_cp;
1227 sb->cp_events_hi = (mddev->events>>32);
1228 sb->cp_events_lo = (u32)mddev->events;
1229 if (mddev->recovery_cp == MaxSector)
1230 sb->state = (1<< MD_SB_CLEAN);
1231 } else
1232 sb->recovery_cp = 0;
1233
1234 sb->layout = mddev->layout;
9d8f0363 1235 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1236
c3d9714e 1237 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1238 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1239
1da177e4 1240 sb->disks[0].state = (1<<MD_DISK_REMOVED);
dafb20fa 1241 rdev_for_each(rdev2, mddev) {
1da177e4 1242 mdp_disk_t *d;
86e6ffdd 1243 int desc_nr;
0261cd9f
N		 1244 int is_active = test_bit(In_sync, &rdev2->flags);
1245
1246 if (rdev2->raid_disk >= 0 &&
1247 sb->minor_version >= 91)
1248 /* we have nowhere to store the recovery_offset,
1249 * but if it is not below the reshape_position,
1250 * we can piggy-back on that.
1251 */
1252 is_active = 1;
1253 if (rdev2->raid_disk < 0 ||
1254 test_bit(Faulty, &rdev2->flags))
1255 is_active = 0;
1256 if (is_active)
86e6ffdd 1257 desc_nr = rdev2->raid_disk;
1da177e4 1258 else
86e6ffdd 1259 desc_nr = next_spare++;
19133a42 1260 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1261 d = &sb->disks[rdev2->desc_nr];
1262 nr_disks++;
1263 d->number = rdev2->desc_nr;
1264 d->major = MAJOR(rdev2->bdev->bd_dev);
1265 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1266 if (is_active)
1da177e4
LT		 1267 d->raid_disk = rdev2->raid_disk;
1268 else
1269 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1270 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1271 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1272 else if (is_active) {
1da177e4 1273 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1274 if (test_bit(In_sync, &rdev2->flags))
1275 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1276 active++;
1277 working++;
1278 } else {
1279 d->state = 0;
1280 spare++;
1281 working++;
1282 }
8ddf9efe
N		 1283 if (test_bit(WriteMostly, &rdev2->flags))
1284 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1285 }
1da177e4
LT		 1286 /* now set the "removed" and "faulty" bits on any missing devices */
1287 for (i=0 ; i < mddev->raid_disks ; i++) {
1288 mdp_disk_t *d = &sb->disks[i];
1289 if (d->state == 0 && d->number == 0) {
1290 d->number = i;
1291 d->raid_disk = i;
1292 d->state = (1<<MD_DISK_REMOVED);
1293 d->state |= (1<<MD_DISK_FAULTY);
1294 failed++;
1295 }
1296 }
1297 sb->nr_disks = nr_disks;
1298 sb->active_disks = active;
1299 sb->working_disks = working;
1300 sb->failed_disks = failed;
1301 sb->spare_disks = spare;
1302
1303 sb->this_disk = sb->disks[rdev->desc_nr];
1304 sb->sb_csum = calc_sb_csum(sb);
1305}
1306
0cd17fec
CW		 1307/*
1308 * rdev_size_change for 0.90.0
1309 */
1310static unsigned long long
3cb03002 1311super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1312{
58c0fed4 1313 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1314 return 0; /* component must fit device */
c3d9714e 1315 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1316 return 0; /* can't move bitmap */
57b2caa3 1317 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1318 if (!num_sectors || num_sectors > rdev->sb_start)
1319 num_sectors = rdev->sb_start;
27a7b260
N		 1320 /* Limit to 4TB as metadata cannot record more than that.
1321 * 4TB == 2^32 KB, or 2*2^32 sectors.
1322 */
667a5313 1323 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
27a7b260 1324 num_sectors = (2ULL << 32) - 2;
0f420358 1325 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1326 rdev->sb_page);
1327 md_super_wait(rdev->mddev);
c26a44ed 1328 return num_sectors;
0cd17fec
CW		 1329}
1330
c6563a8c
N		 1331static int
1332super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1333{
1334 /* non-zero offset changes not possible with v0.90 */
1335 return new_offset == 0;
1336}
0cd17fec 1337
1da177e4
LT		 1338/*
1339 * version 1 superblock
1340 */
1341
f72ffdd6 1342static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1da177e4 1343{
1c05b4bc
N		 1344 __le32 disk_csum;
1345 u32 csum;
1da177e4
LT		 1346 unsigned long long newcsum;
1347 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1348 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1349
1350 disk_csum = sb->sb_csum;
1351 sb->sb_csum = 0;
1352 newcsum = 0;
1f3c9907 1353 for (; size >= 4; size -= 4)
1da177e4
LT		 1354 newcsum += le32_to_cpu(*isuper++);
1355
1356 if (size == 2)
1c05b4bc 1357 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1358
1359 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1360 sb->sb_csum = disk_csum;
1361 return cpu_to_le32(csum);
1362}
1363
2699b672
N		 1364static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1365 int acknowledged);
3cb03002 1366static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1367{
1368 struct mdp_superblock_1 *sb;
1369 int ret;
0f420358 1370 sector_t sb_start;
c6563a8c 1371 sector_t sectors;
1da177e4 1372 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1373 int bmask;
1da177e4
LT		 1374
1375 /*
0f420358 1376 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1377 * It is always aligned to a 4K boundary and
1378 * depeding on minor_version, it can be:
1379 * 0: At least 8K, but less than 12K, from end of device
1380 * 1: At start of device
1381 * 2: 4K from start of device.
1382 */
1383 switch(minor_version) {
1384 case 0:
77304d2a 1385 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1386 sb_start -= 8*2;
1387 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1388 break;
1389 case 1:
0f420358 1390 sb_start = 0;
1da177e4
LT		 1391 break;
1392 case 2:
0f420358 1393 sb_start = 8;
1da177e4
LT		 1394 break;
1395 default:
1396 return -EINVAL;
1397 }
0f420358 1398 rdev->sb_start = sb_start;
1da177e4 1399
0002b271
N		 1400 /* superblock is rarely larger than 1K, but it can be larger,
1401 * and it is safe to read 4k, so we do that
1402 */
1403 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1404 if (ret) return ret;
1405
65a06f06 1406 sb = page_address(rdev->sb_page);
1da177e4
LT		 1407
1408 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1409 sb->major_version != cpu_to_le32(1) ||
1410 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1411 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1412 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1413 return -EINVAL;
1414
1415 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1416 printk("md: invalid superblock checksum on %s\n",
1417 bdevname(rdev->bdev,b));
1418 return -EINVAL;
1419 }
1420 if (le64_to_cpu(sb->data_size) < 10) {
1421 printk("md: data_size too small on %s\n",
1422 bdevname(rdev->bdev,b));
1423 return -EINVAL;
1424 }
c6563a8c
N		 1425 if (sb->pad0 ||
1426 sb->pad3[0] ||
1427 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1428 /* Some padding is non-zero, might be a new feature */
1429 return -EINVAL;
e11e93fa 1430
1da177e4
LT		 1431 rdev->preferred_minor = 0xffff;
1432 rdev->data_offset = le64_to_cpu(sb->data_offset);
c6563a8c
N		 1433 rdev->new_data_offset = rdev->data_offset;
1434 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1435 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1436 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
4dbcdc75 1437 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1438
0002b271 1439 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1440 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1441 if (rdev->sb_size & bmask)
a1801f85
N		 1442 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1443
1444 if (minor_version
0f420358 1445 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1446 return -EINVAL;
c6563a8c
N		 1447 if (minor_version
1448 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1449 return -EINVAL;
0002b271 1450
31b65a0d
N		 1451 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1452 rdev->desc_nr = -1;
1453 else
1454 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1455
2699b672
N		 1456 if (!rdev->bb_page) {
1457 rdev->bb_page = alloc_page(GFP_KERNEL);
1458 if (!rdev->bb_page)
1459 return -ENOMEM;
1460 }
1461 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1462 rdev->badblocks.count == 0) {
1463 /* need to load the bad block list.
1464 * Currently we limit it to one page.
1465 */
1466 s32 offset;
1467 sector_t bb_sector;
1468 u64 *bbp;
1469 int i;
1470 int sectors = le16_to_cpu(sb->bblog_size);
1471 if (sectors > (PAGE_SIZE / 512))
1472 return -EINVAL;
1473 offset = le32_to_cpu(sb->bblog_offset);
1474 if (offset == 0)
1475 return -EINVAL;
1476 bb_sector = (long long)offset;
1477 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1478 rdev->bb_page, READ, true))
1479 return -EIO;
1480 bbp = (u64 *)page_address(rdev->bb_page);
1481 rdev->badblocks.shift = sb->bblog_shift;
1482 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1483 u64 bb = le64_to_cpu(*bbp);
1484 int count = bb & (0x3ff);
1485 u64 sector = bb >> 10;
1486 sector <<= sb->bblog_shift;
1487 count <<= sb->bblog_shift;
1488 if (bb + 1 == 0)
1489 break;
1490 if (md_set_badblocks(&rdev->badblocks,
1491 sector, count, 1) == 0)
1492 return -EINVAL;
1493 }
486adf72
N		 1494 } else if (sb->bblog_offset != 0)
1495 rdev->badblocks.shift = 0;
2699b672 1496
9a7b2b0f 1497 if (!refdev) {
8ed75463 1498 ret = 1;
9a7b2b0f 1499 } else {
1da177e4 1500 __u64 ev1, ev2;
65a06f06 1501 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1502
1503 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1504 sb->level != refsb->level ||
1505 sb->layout != refsb->layout ||
1506 sb->chunksize != refsb->chunksize) {
1507 printk(KERN_WARNING "md: %s has strangely different"
1508 " superblock to %s\n",
1509 bdevname(rdev->bdev,b),
1510 bdevname(refdev->bdev,b2));
1511 return -EINVAL;
1512 }
1513 ev1 = le64_to_cpu(sb->events);
1514 ev2 = le64_to_cpu(refsb->events);
1515
1516 if (ev1 > ev2)
8ed75463
N		 1517 ret = 1;
1518 else
1519 ret = 0;
1da177e4 1520 }
c6563a8c
N		 1521 if (minor_version) {
1522 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1523 sectors -= rdev->data_offset;
1524 } else
1525 sectors = rdev->sb_start;
1526 if (sectors < le64_to_cpu(sb->data_size))
1da177e4 1527 return -EINVAL;
dd8ac336 1528 rdev->sectors = le64_to_cpu(sb->data_size);
8ed75463 1529 return ret;
1da177e4
LT		 1530}
1531
fd01b88c 1532static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1533{
65a06f06 1534 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1535 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1536
41158c7e 1537 rdev->raid_disk = -1;
c5d79adb
N		 1538 clear_bit(Faulty, &rdev->flags);
1539 clear_bit(In_sync, &rdev->flags);
8313b8e5 1540 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1541 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1542
1da177e4
LT		 1543 if (mddev->raid_disks == 0) {
1544 mddev->major_version = 1;
1545 mddev->patch_version = 0;
e691063a 1546 mddev->external = 0;
9d8f0363 1547 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1548 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1549 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1550 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1551 mddev->clevel[0] = 0;
1da177e4
LT		 1552 mddev->layout = le32_to_cpu(sb->layout);
1553 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1554 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1555 mddev->events = ev1;
c3d9714e 1556 mddev->bitmap_info.offset = 0;
6409bb05
N		 1557 mddev->bitmap_info.space = 0;
1558 /* Default location for bitmap is 1K after superblock
1559 * using 3K - total of 4K
1560 */
c3d9714e 1561 mddev->bitmap_info.default_offset = 1024 >> 9;
6409bb05 1562 mddev->bitmap_info.default_space = (4096-1024) >> 9;
2c810cdd
N		 1563 mddev->reshape_backwards = 0;
1564
1da177e4
LT		 1565 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1566 memcpy(mddev->uuid, sb->set_uuid, 16);
1567
1568 mddev->max_disks = (4096-256)/2;
a654b9d8 1569
71c0805c 1570 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
6409bb05 1571 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1572 mddev->bitmap_info.offset =
1573 (__s32)le32_to_cpu(sb->bitmap_offset);
6409bb05
N		 1574 /* Metadata doesn't record how much space is available.
1575 * For 1.0, we assume we can use up to the superblock
1576 * if before, else to 4K beyond superblock.
1577 * For others, assume no change is possible.
1578 */
1579 if (mddev->minor_version > 0)
1580 mddev->bitmap_info.space = 0;
1581 else if (mddev->bitmap_info.offset > 0)
1582 mddev->bitmap_info.space =
1583 8 - mddev->bitmap_info.offset;
1584 else
1585 mddev->bitmap_info.space =
1586 -mddev->bitmap_info.offset;
1587 }
e11e93fa 1588
f6705578
N		 1589 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1590 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1591 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1592 mddev->new_level = le32_to_cpu(sb->new_level);
1593 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1594 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
2c810cdd
N		 1595 if (mddev->delta_disks < 0 ||
1596 (mddev->delta_disks == 0 &&
1597 (le32_to_cpu(sb->feature_map)
1598 & MD_FEATURE_RESHAPE_BACKWARDS)))
1599 mddev->reshape_backwards = 1;
f6705578
N		 1600 } else {
1601 mddev->reshape_position = MaxSector;
1602 mddev->delta_disks = 0;
1603 mddev->new_level = mddev->level;
1604 mddev->new_layout = mddev->layout;
664e7c41 1605 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1606 }
1607
41158c7e 1608 } else if (mddev->pers == NULL) {
be6800a7
N		 1609 /* Insist of good event counter while assembling, except for
1610 * spares (which don't need an event count) */
1da177e4 1611 ++ev1;
be6800a7
N		 1612 if (rdev->desc_nr >= 0 &&
1613 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
a3dfbdaa
SL		 1614 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1615 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
be6800a7
N		 1616 if (ev1 < mddev->events)
1617 return -EINVAL;
41158c7e
N		 1618 } else if (mddev->bitmap) {
1619 /* If adding to array with a bitmap, then we can accept an
1620 * older device, but not too old.
1621 */
41158c7e
N		 1622 if (ev1 < mddev->bitmap->events_cleared)
1623 return 0;
8313b8e5
N		 1624 if (ev1 < mddev->events)
1625 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1626 } else {
1627 if (ev1 < mddev->events)
1628 /* just a hot-add of a new device, leave raid_disk at -1 */
1629 return 0;
1630 }
1da177e4
LT		 1631 if (mddev->level != LEVEL_MULTIPATH) {
1632 int role;
3673f305
N		 1633 if (rdev->desc_nr < 0 ||
1634 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
c4d4c91b 1635 role = MD_DISK_ROLE_SPARE;
3673f305
N		 1636 rdev->desc_nr = -1;
1637 } else
1638 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4 1639 switch(role) {
c4d4c91b 1640 case MD_DISK_ROLE_SPARE: /* spare */
1da177e4 1641 break;
c4d4c91b 1642 case MD_DISK_ROLE_FAULTY: /* faulty */
b2d444d7 1643 set_bit(Faulty, &rdev->flags);
1da177e4 1644 break;
bac624f3
SL		 1645 case MD_DISK_ROLE_JOURNAL: /* journal device */
1646 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1647 /* journal device without journal feature */
1648 printk(KERN_WARNING
1649 "md: journal device provided without journal feature, ignoring the device\n");
1650 return -EINVAL;
1651 }
1652 set_bit(Journal, &rdev->flags);
3069aa8d 1653 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
bd18f646
SL		 1654 if (mddev->recovery_cp == MaxSector)
1655 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
9b15603d 1656 rdev->raid_disk = 0;
bac624f3 1657 break;
1da177e4 1658 default:
f466722c 1659 rdev->saved_raid_disk = role;
5fd6c1dc 1660 if ((le32_to_cpu(sb->feature_map) &
f466722c 1661 MD_FEATURE_RECOVERY_OFFSET)) {
5fd6c1dc 1662 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
f466722c
N		 1663 if (!(le32_to_cpu(sb->feature_map) &
1664 MD_FEATURE_RECOVERY_BITMAP))
1665 rdev->saved_raid_disk = -1;
1666 } else
5fd6c1dc 1667 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1668 rdev->raid_disk = role;
1669 break;
1670 }
8ddf9efe
N		 1671 if (sb->devflags & WriteMostly1)
1672 set_bit(WriteMostly, &rdev->flags);
2d78f8c4
N		 1673 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1674 set_bit(Replacement, &rdev->flags);
a97b7896
SL		 1675 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1676 set_bit(MD_HAS_JOURNAL, &mddev->flags);
41158c7e 1677 } else /* MULTIPATH are always insync */
b2d444d7 1678 set_bit(In_sync, &rdev->flags);
41158c7e 1679
1da177e4
LT		 1680 return 0;
1681}
1682
fd01b88c 1683static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1684{
1685 struct mdp_superblock_1 *sb;
3cb03002 1686 struct md_rdev *rdev2;
1da177e4
LT		 1687 int max_dev, i;
1688 /* make rdev->sb match mddev and rdev data. */
1689
65a06f06 1690 sb = page_address(rdev->sb_page);
1da177e4
LT		 1691
1692 sb->feature_map = 0;
1693 sb->pad0 = 0;
5fd6c1dc 1694 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1695 memset(sb->pad3, 0, sizeof(sb->pad3));
1696
1697 sb->utime = cpu_to_le64((__u64)mddev->utime);
1698 sb->events = cpu_to_le64(mddev->events);
1699 if (mddev->in_sync)
1700 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
bd18f646
SL		 1701 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1702 sb->resync_offset = cpu_to_le64(MaxSector);
1da177e4
LT		 1703 else
1704 sb->resync_offset = cpu_to_le64(0);
1705
1c05b4bc 1706 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1707
f0ca340c 1708 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1709 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1710 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1711 sb->level = cpu_to_le32(mddev->level);
1712 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1713
aeb9b211
N		 1714 if (test_bit(WriteMostly, &rdev->flags))
1715 sb->devflags |= WriteMostly1;
1716 else
1717 sb->devflags &= ~WriteMostly1;
c6563a8c
N		 1718 sb->data_offset = cpu_to_le64(rdev->data_offset);
1719 sb->data_size = cpu_to_le64(rdev->sectors);
aeb9b211 1720
c3d9714e
N		 1721 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1722 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1723 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1724 }
5fd6c1dc 1725
f2076e7d 1726 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
97e4f42d 1727 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1728 sb->feature_map |=
1729 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1730 sb->recovery_offset =
1731 cpu_to_le64(rdev->recovery_offset);
f466722c
N		 1732 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1733 sb->feature_map |=
1734 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
5fd6c1dc 1735 }
3069aa8d
SL		 1736 /* Note: recovery_offset and journal_tail share space */
1737 if (test_bit(Journal, &rdev->flags))
1738 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2d78f8c4
N		 1739 if (test_bit(Replacement, &rdev->flags))
1740 sb->feature_map |=
1741 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1742
f6705578
N		 1743 if (mddev->reshape_position != MaxSector) {
1744 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1745 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1746 sb->new_layout = cpu_to_le32(mddev->new_layout);
1747 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1748 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1749 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2c810cdd
N		 1750 if (mddev->delta_disks == 0 &&
1751 mddev->reshape_backwards)
1752 sb->feature_map
1753 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
c6563a8c
N		 1754 if (rdev->new_data_offset != rdev->data_offset) {
1755 sb->feature_map
1756 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1757 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1758 - rdev->data_offset));
1759 }
f6705578 1760 }
a654b9d8 1761
3c462c88
GR		 1762 if (mddev_is_clustered(mddev))
1763 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1764
2699b672
N		 1765 if (rdev->badblocks.count == 0)
1766 /* Nothing to do for bad blocks*/ ;
1767 else if (sb->bblog_offset == 0)
1768 /* Cannot record bad blocks on this device */
1769 md_error(mddev, rdev);
1770 else {
1771 struct badblocks *bb = &rdev->badblocks;
1772 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1773 u64 *p = bb->page;
1774 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1775 if (bb->changed) {
1776 unsigned seq;
1777
1778retry:
1779 seq = read_seqbegin(&bb->lock);
1780
1781 memset(bbp, 0xff, PAGE_SIZE);
1782
1783 for (i = 0 ; i < bb->count ; i++) {
35f9ac2d 1784 u64 internal_bb = p[i];
2699b672
N		 1785 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1786 | BB_LEN(internal_bb));
35f9ac2d 1787 bbp[i] = cpu_to_le64(store_bb);
2699b672 1788 }
d0962936 1789 bb->changed = 0;
2699b672
N		 1790 if (read_seqretry(&bb->lock, seq))
1791 goto retry;
1792
1793 bb->sector = (rdev->sb_start +
1794 (int)le32_to_cpu(sb->bblog_offset));
1795 bb->size = le16_to_cpu(sb->bblog_size);
2699b672
N		 1796 }
1797 }
1798
1da177e4 1799 max_dev = 0;
dafb20fa 1800 rdev_for_each(rdev2, mddev)
1da177e4
LT		 1801 if (rdev2->desc_nr+1 > max_dev)
1802 max_dev = rdev2->desc_nr+1;
a778b73f 1803
70471daf
N		 1804 if (max_dev > le32_to_cpu(sb->max_dev)) {
1805 int bmask;
a778b73f 1806 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1807 rdev->sb_size = max_dev * 2 + 256;
1808 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1809 if (rdev->sb_size & bmask)
1810 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1811 } else
1812 max_dev = le32_to_cpu(sb->max_dev);
1813
1da177e4 1814 for (i=0; i<max_dev;i++)
c4d4c91b 1815 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
f72ffdd6 1816
a97b7896
SL		 1817 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1818 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
f72ffdd6 1819
dafb20fa 1820 rdev_for_each(rdev2, mddev) {
1da177e4 1821 i = rdev2->desc_nr;
b2d444d7 1822 if (test_bit(Faulty, &rdev2->flags))
c4d4c91b 1823 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
b2d444d7 1824 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1825 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
a97b7896 1826 else if (test_bit(Journal, &rdev2->flags))
bac624f3 1827 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
93be75ff 1828 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1829 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4 1830 else
c4d4c91b 1831 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1da177e4
LT		 1832 }
1833
1da177e4
LT		 1834 sb->sb_csum = calc_sb_1_csum(sb);
1835}
1836
0cd17fec 1837static unsigned long long
3cb03002 1838super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 1839{
1840 struct mdp_superblock_1 *sb;
15f4a5fd 1841 sector_t max_sectors;
58c0fed4 1842 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1843 return 0; /* component must fit device */
c6563a8c
N		 1844 if (rdev->data_offset != rdev->new_data_offset)
1845 return 0; /* too confusing */
0f420358 1846 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1847 /* minor versions 1 and 2; superblock before data */
77304d2a 1848 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1849 max_sectors -= rdev->data_offset;
1850 if (!num_sectors || num_sectors > max_sectors)
1851 num_sectors = max_sectors;
c3d9714e 1852 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1853 /* minor version 0 with bitmap we can't move */
1854 return 0;
1855 } else {
1856 /* minor version 0; superblock after data */
0f420358 1857 sector_t sb_start;
77304d2a 1858 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1859 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1860 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1861 if (!num_sectors || num_sectors > max_sectors)
1862 num_sectors = max_sectors;
0f420358 1863 rdev->sb_start = sb_start;
0cd17fec 1864 }
65a06f06 1865 sb = page_address(rdev->sb_page);
15f4a5fd 1866 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1867 sb->super_offset = rdev->sb_start;
0cd17fec 1868 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1869 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1870 rdev->sb_page);
1871 md_super_wait(rdev->mddev);
c26a44ed 1872 return num_sectors;
c6563a8c
N		 1873
1874}
1875
1876static int
1877super_1_allow_new_offset(struct md_rdev *rdev,
1878 unsigned long long new_offset)
1879{
1880 /* All necessary checks on new >= old have been done */
1881 struct bitmap *bitmap;
1882 if (new_offset >= rdev->data_offset)
1883 return 1;
1884
1885 /* with 1.0 metadata, there is no metadata to tread on
1886 * so we can always move back */
1887 if (rdev->mddev->minor_version == 0)
1888 return 1;
1889
1890 /* otherwise we must be sure not to step on
1891 * any metadata, so stay:
1892 * 36K beyond start of superblock
1893 * beyond end of badblocks
1894 * beyond write-intent bitmap
1895 */
1896 if (rdev->sb_start + (32+4)*2 > new_offset)
1897 return 0;
1898 bitmap = rdev->mddev->bitmap;
1899 if (bitmap && !rdev->mddev->bitmap_info.file &&
1900 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1ec885cd 1901 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
c6563a8c
N		 1902 return 0;
1903 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1904 return 0;
1905
1906 return 1;
0cd17fec 1907}
1da177e4 1908
75c96f85 1909static struct super_type super_types[] = {
1da177e4
LT		 1910 [0] = {
1911 .name = "0.90.0",
1912 .owner = THIS_MODULE,
0cd17fec
CW		 1913 .load_super = super_90_load,
1914 .validate_super = super_90_validate,
1915 .sync_super = super_90_sync,
1916 .rdev_size_change = super_90_rdev_size_change,
c6563a8c 1917 .allow_new_offset = super_90_allow_new_offset,
1da177e4
LT		 1918 },
1919 [1] = {
1920 .name = "md-1",
1921 .owner = THIS_MODULE,
0cd17fec
CW		 1922 .load_super = super_1_load,
1923 .validate_super = super_1_validate,
1924 .sync_super = super_1_sync,
1925 .rdev_size_change = super_1_rdev_size_change,
c6563a8c 1926 .allow_new_offset = super_1_allow_new_offset,
1da177e4
LT		 1927 },
1928};
1da177e4 1929
fd01b88c 1930static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 1931{
1932 if (mddev->sync_super) {
1933 mddev->sync_super(mddev, rdev);
1934 return;
1935 }
1936
1937 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1938
1939 super_types[mddev->major_version].sync_super(mddev, rdev);
1940}
1941
fd01b88c 1942static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1943{
3cb03002 1944 struct md_rdev *rdev, *rdev2;
1da177e4 1945
4b80991c 1946 rcu_read_lock();
0b020e85
SL		 1947 rdev_for_each_rcu(rdev, mddev1) {
1948 if (test_bit(Faulty, &rdev->flags) ||
1949 test_bit(Journal, &rdev->flags) ||
1950 rdev->raid_disk == -1)
1951 continue;
1952 rdev_for_each_rcu(rdev2, mddev2) {
1953 if (test_bit(Faulty, &rdev2->flags) ||
1954 test_bit(Journal, &rdev2->flags) ||
1955 rdev2->raid_disk == -1)
1956 continue;
7dd5e7c3 1957 if (rdev->bdev->bd_contains ==
4b80991c
N		 1958 rdev2->bdev->bd_contains) {
1959 rcu_read_unlock();
7dd5e7c3 1960 return 1;
4b80991c 1961 }
0b020e85
SL		 1962 }
1963 }
4b80991c 1964 rcu_read_unlock();
1da177e4
LT		 1965 return 0;
1966}
1967
1968static LIST_HEAD(pending_raid_disks);
1969
ac5e7113
AN		 1970/*
1971 * Try to register data integrity profile for an mddev
1972 *
1973 * This is called when an array is started and after a disk has been kicked
1974 * from the array. It only succeeds if all working and active component devices
1975 * are integrity capable with matching profiles.
1976 */
fd01b88c 1977int md_integrity_register(struct mddev *mddev)
ac5e7113 1978{
3cb03002 1979 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 1980
1981 if (list_empty(&mddev->disks))
1982 return 0; /* nothing to do */
629acb6a
JB		 1983 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1984 return 0; /* shouldn't register, or already is */
dafb20fa 1985 rdev_for_each(rdev, mddev) {
ac5e7113
AN		 1986 /* skip spares and non-functional disks */
1987 if (test_bit(Faulty, &rdev->flags))
1988 continue;
1989 if (rdev->raid_disk < 0)
1990 continue;
ac5e7113
AN		 1991 if (!reference) {
1992 /* Use the first rdev as the reference */
1993 reference = rdev;
1994 continue;
1995 }
1996 /* does this rdev's profile match the reference profile? */
1997 if (blk_integrity_compare(reference->bdev->bd_disk,
1998 rdev->bdev->bd_disk) < 0)
1999 return -EINVAL;
2000 }
89078d57
MP		 2001 if (!reference || !bdev_get_integrity(reference->bdev))
2002 return 0;
ac5e7113
AN		 2003 /*
2004 * All component devices are integrity capable and have matching
2005 * profiles, register the common profile for the md device.
2006 */
25520d55
MP		 2007 blk_integrity_register(mddev->gendisk,
2008 bdev_get_integrity(reference->bdev));
2009
a91a2785
MP		 2010 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2011 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2012 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2013 mdname(mddev));
2014 return -EINVAL;
2015 }
ac5e7113
AN		 2016 return 0;
2017}
2018EXPORT_SYMBOL(md_integrity_register);
2019
2020/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 2021void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 2022{
2863b9eb
JB		 2023 struct blk_integrity *bi_rdev;
2024 struct blk_integrity *bi_mddev;
2025
2026 if (!mddev->gendisk)
2027 return;
2028
2029 bi_rdev = bdev_get_integrity(rdev->bdev);
2030 bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 2031
ac5e7113 2032 if (!bi_mddev) /* nothing to do */
3f9d99c1 2033 return;
ac5e7113 2034 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 2035 return;
ac5e7113
AN		 2036 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2037 rdev->bdev->bd_disk) >= 0)
2038 return;
c7bfced9 2039 WARN_ON_ONCE(!mddev->suspended);
ac5e7113
AN		 2040 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2041 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2042}
ac5e7113 2043EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2044
f72ffdd6 2045static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
1da177e4 2046{
7dd5e7c3 2047 char b[BDEVNAME_SIZE];
f637b9f9 2048 struct kobject *ko;
5e55e2f5 2049 int err;
1da177e4 2050
11e2ede0
DW		 2051 /* prevent duplicates */
2052 if (find_rdev(mddev, rdev->bdev->bd_dev))
2053 return -EEXIST;
2054
dd8ac336
AN		 2055 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2056 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2057 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2058 if (mddev->pers) {
2059 /* Cannot change size, so fail
2060 * If mddev->level <= 0, then we don't care
2061 * about aligning sizes (e.g. linear)
2062 */
2063 if (mddev->level > 0)
2064 return -ENOSPC;
2065 } else
dd8ac336 2066 mddev->dev_sectors = rdev->sectors;
2bf071bf 2067 }
1da177e4
LT		 2068
2069 /* Verify rdev->desc_nr is unique.
2070 * If it is -1, assign a free number, else
2071 * check number is not in use
2072 */
4878e9eb 2073 rcu_read_lock();
1da177e4
LT		 2074 if (rdev->desc_nr < 0) {
2075 int choice = 0;
4878e9eb
N		 2076 if (mddev->pers)
2077 choice = mddev->raid_disks;
57d051dc 2078 while (md_find_rdev_nr_rcu(mddev, choice))
1da177e4
LT		 2079 choice++;
2080 rdev->desc_nr = choice;
2081 } else {
57d051dc 2082 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
4878e9eb 2083 rcu_read_unlock();
1da177e4 2084 return -EBUSY;
4878e9eb 2085 }
1da177e4 2086 }
4878e9eb 2087 rcu_read_unlock();
de01dfad
N		 2088 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2089 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2090 mdname(mddev), mddev->max_disks);
2091 return -EBUSY;
2092 }
19133a42 2093 bdevname(rdev->bdev,b);
90a9befb 2094 strreplace(b, '/', '!');
649316b2 2095
1da177e4 2096 rdev->mddev = mddev;
19133a42 2097 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2098
b2d6db58 2099 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2100 goto fail;
86e6ffdd 2101
0762b8bd 2102 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2103 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2104 /* failure here is OK */;
2105 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2106
4b80991c 2107 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2108 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2109
2110 /* May as well allow recovery to be retried once */
5389042f 2111 mddev->recovery_disabled++;
3f9d99c1 2112
1da177e4 2113 return 0;
5e55e2f5
N		 2114
2115 fail:
2116 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2117 b, mdname(mddev));
2118 return err;
1da177e4
LT		 2119}
2120
177a99b2 2121static void md_delayed_delete(struct work_struct *ws)
5792a285 2122{
3cb03002 2123 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2124 kobject_del(&rdev->kobj);
177a99b2 2125 kobject_put(&rdev->kobj);
5792a285
N		 2126}
2127
f72ffdd6 2128static void unbind_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2129{
2130 char b[BDEVNAME_SIZE];
403df478 2131
49731baa 2132 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2133 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2134 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2135 rdev->mddev = NULL;
86e6ffdd 2136 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2137 sysfs_put(rdev->sysfs_state);
2138 rdev->sysfs_state = NULL;
2230dfe4 2139 rdev->badblocks.count = 0;
5792a285 2140 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2141 * writing to 'remove' to "dev/state". We also need
2142 * to delay it due to rcu usage.
5792a285 2143 */
4b80991c 2144 synchronize_rcu();
177a99b2
N		 2145 INIT_WORK(&rdev->del_work, md_delayed_delete);
2146 kobject_get(&rdev->kobj);
e804ac78 2147 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2148}
2149
2150/*
2151 * prevent the device from being mounted, repartitioned or
2152 * otherwise reused by a RAID array (or any other kernel
2153 * subsystem), by bd_claiming the device.
2154 */
3cb03002 2155static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2156{
2157 int err = 0;
2158 struct block_device *bdev;
2159 char b[BDEVNAME_SIZE];
2160
d4d77629 2161 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2162 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2163 if (IS_ERR(bdev)) {
2164 printk(KERN_ERR "md: could not open %s.\n",
2165 __bdevname(dev, b));
2166 return PTR_ERR(bdev);
2167 }
1da177e4
LT		 2168 rdev->bdev = bdev;
2169 return err;
2170}
2171
3cb03002 2172static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2173{
2174 struct block_device *bdev = rdev->bdev;
2175 rdev->bdev = NULL;
e525fd89 2176 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2177}
2178
2179void md_autodetect_dev(dev_t dev);
2180
f72ffdd6 2181static void export_rdev(struct md_rdev *rdev)
1da177e4
LT		 2182{
2183 char b[BDEVNAME_SIZE];
403df478 2184
1da177e4
LT		 2185 printk(KERN_INFO "md: export_rdev(%s)\n",
2186 bdevname(rdev->bdev,b));
545c8795 2187 md_rdev_clear(rdev);
1da177e4 2188#ifndef MODULE
d0fae18f
N		 2189 if (test_bit(AutoDetected, &rdev->flags))
2190 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2191#endif
2192 unlock_rdev(rdev);
86e6ffdd 2193 kobject_put(&rdev->kobj);
1da177e4
LT		 2194}
2195
fb56dfef 2196void md_kick_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2197{
2198 unbind_rdev_from_array(rdev);
2199 export_rdev(rdev);
2200}
fb56dfef 2201EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
1da177e4 2202
fd01b88c 2203static void export_array(struct mddev *mddev)
1da177e4 2204{
0638bb0e 2205 struct md_rdev *rdev;
1da177e4 2206
0638bb0e
N		 2207 while (!list_empty(&mddev->disks)) {
2208 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2209 same_set);
fb56dfef 2210 md_kick_rdev_from_array(rdev);
1da177e4 2211 }
1da177e4
LT		 2212 mddev->raid_disks = 0;
2213 mddev->major_version = 0;
2214}
2215
f72ffdd6 2216static void sync_sbs(struct mddev *mddev, int nospares)
1da177e4 2217{
42543769
N		 2218 /* Update each superblock (in-memory image), but
2219 * if we are allowed to, skip spares which already
2220 * have the right event counter, or have one earlier
2221 * (which would mean they aren't being marked as dirty
2222 * with the rest of the array)
2223 */
3cb03002 2224 struct md_rdev *rdev;
dafb20fa 2225 rdev_for_each(rdev, mddev) {
42543769
N		 2226 if (rdev->sb_events == mddev->events ||
2227 (nospares &&
2228 rdev->raid_disk < 0 &&
42543769
N		 2229 rdev->sb_events+1 == mddev->events)) {
2230 /* Don't update this superblock */
2231 rdev->sb_loaded = 2;
2232 } else {
076f968b 2233 sync_super(mddev, rdev);
42543769
N		 2234 rdev->sb_loaded = 1;
2235 }
1da177e4
LT		 2236 }
2237}
2238
2aa82191
GR		 2239static bool does_sb_need_changing(struct mddev *mddev)
2240{
2241 struct md_rdev *rdev;
2242 struct mdp_superblock_1 *sb;
2243 int role;
2244
2245 /* Find a good rdev */
2246 rdev_for_each(rdev, mddev)
2247 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2248 break;
2249
2250 /* No good device found. */
2251 if (!rdev)
2252 return false;
2253
2254 sb = page_address(rdev->sb_page);
2255 /* Check if a device has become faulty or a spare become active */
2256 rdev_for_each(rdev, mddev) {
2257 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2258 /* Device activated? */
2259 if (role == 0xffff && rdev->raid_disk >=0 &&
2260 !test_bit(Faulty, &rdev->flags))
2261 return true;
2262 /* Device turned faulty? */
2263 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2264 return true;
2265 }
2266
2267 /* Check if any mddev parameters have changed */
2268 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2269 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2aa82191
GR		 2270 (mddev->layout != le64_to_cpu(sb->layout)) ||
2271 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2272 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2273 return true;
2274
2275 return false;
2276}
2277
1aee41f6 2278void md_update_sb(struct mddev *mddev, int force_change)
1da177e4 2279{
3cb03002 2280 struct md_rdev *rdev;
06d91a5f 2281 int sync_req;
42543769 2282 int nospares = 0;
2699b672 2283 int any_badblocks_changed = 0;
23b63f9f 2284 int ret = -1;
1da177e4 2285
d87f064f
N		 2286 if (mddev->ro) {
2287 if (force_change)
2288 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2289 return;
2290 }
2aa82191
GR		 2291
2292 if (mddev_is_clustered(mddev)) {
2293 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2294 force_change = 1;
23b63f9f 2295 ret = md_cluster_ops->metadata_update_start(mddev);
2aa82191
GR		 2296 /* Has someone else has updated the sb */
2297 if (!does_sb_need_changing(mddev)) {
23b63f9f
GJ		 2298 if (ret == 0)
2299 md_cluster_ops->metadata_update_cancel(mddev);
2aa82191
GR		 2300 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2301 return;
2302 }
2303 }
1da177e4 2304repeat:
3a3a5ddb 2305 /* First make sure individual recovery_offsets are correct */
dafb20fa 2306 rdev_for_each(rdev, mddev) {
3a3a5ddb
N		 2307 if (rdev->raid_disk >= 0 &&
2308 mddev->delta_disks >= 0 &&
f2076e7d 2309 !test_bit(Journal, &rdev->flags) &&
3a3a5ddb
N		 2310 !test_bit(In_sync, &rdev->flags) &&
2311 mddev->curr_resync_completed > rdev->recovery_offset)
2312 rdev->recovery_offset = mddev->curr_resync_completed;
2313
f72ffdd6 2314 }
bd52b746 2315 if (!mddev->persistent) {
070dc6dd 2316 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2317 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2318 if (!mddev->external) {
d97a41dc 2319 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
dafb20fa 2320 rdev_for_each(rdev, mddev) {
de393cde 2321 if (rdev->badblocks.changed) {
d0962936 2322 rdev->badblocks.changed = 0;
de393cde
N		 2323 md_ack_all_badblocks(&rdev->badblocks);
2324 md_error(mddev, rdev);
2325 }
2326 clear_bit(Blocked, &rdev->flags);
2327 clear_bit(BlockedBadBlocks, &rdev->flags);
2328 wake_up(&rdev->blocked_wait);
2329 }
2330 }
3a3a5ddb
N		 2331 wake_up(&mddev->sb_wait);
2332 return;
2333 }
2334
85572d7c 2335 spin_lock(&mddev->lock);
84692195 2336
3a3a5ddb
N		 2337 mddev->utime = get_seconds();
2338
850b2b42
N		 2339 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2340 force_change = 1;
2341 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2342 /* just a clean<-> dirty transition, possibly leave spares alone,
2343 * though if events isn't the right even/odd, we will have to do
2344 * spares after all
2345 */
2346 nospares = 1;
2347 if (force_change)
2348 nospares = 0;
2349 if (mddev->degraded)
84692195
N		 2350 /* If the array is degraded, then skipping spares is both
2351 * dangerous and fairly pointless.
2352 * Dangerous because a device that was removed from the array
2353 * might have a event_count that still looks up-to-date,
2354 * so it can be re-added without a resync.
2355 * Pointless because if there are any spares to skip,
2356 * then a recovery will happen and soon that array won't
2357 * be degraded any more and the spare can go back to sleep then.
2358 */
850b2b42 2359 nospares = 0;
84692195 2360
06d91a5f 2361 sync_req = mddev->in_sync;
42543769
N		 2362
2363 /* If this is just a dirty<->clean transition, and the array is clean
2364 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2365 if (nospares
42543769 2366 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2367 && mddev->can_decrease_events
2368 && mddev->events != 1) {
42543769 2369 mddev->events--;
a8707c08
N		 2370 mddev->can_decrease_events = 0;
2371 } else {
42543769
N		 2372 /* otherwise we have to go forward and ... */
2373 mddev->events ++;
a8707c08 2374 mddev->can_decrease_events = nospares;
42543769 2375 }
1da177e4 2376
403df478
N		 2377 /*
2378 * This 64-bit counter should never wrap.
2379 * Either we are in around ~1 trillion A.C., assuming
2380 * 1 reboot per second, or we have a bug...
2381 */
2382 WARN_ON(mddev->events == 0);
2699b672 2383
dafb20fa 2384 rdev_for_each(rdev, mddev) {
2699b672
N		 2385 if (rdev->badblocks.changed)
2386 any_badblocks_changed++;
de393cde
N		 2387 if (test_bit(Faulty, &rdev->flags))
2388 set_bit(FaultRecorded, &rdev->flags);
2389 }
2699b672 2390
e691063a 2391 sync_sbs(mddev, nospares);
85572d7c 2392 spin_unlock(&mddev->lock);
1da177e4 2393
36a4e1fe
N		 2394 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2395 mdname(mddev), mddev->in_sync);
1da177e4 2396
4ad13663 2397 bitmap_update_sb(mddev->bitmap);
dafb20fa 2398 rdev_for_each(rdev, mddev) {
1da177e4 2399 char b[BDEVNAME_SIZE];
36a4e1fe 2400
42543769
N		 2401 if (rdev->sb_loaded != 1)
2402 continue; /* no noise on spare devices */
1da177e4 2403
f466722c 2404 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2405 md_super_write(mddev,rdev,
0f420358 2406 rdev->sb_start, rdev->sb_size,
7bfa19f2 2407 rdev->sb_page);
36a4e1fe
N		 2408 pr_debug("md: (write) %s's sb offset: %llu\n",
2409 bdevname(rdev->bdev, b),
2410 (unsigned long long)rdev->sb_start);
42543769 2411 rdev->sb_events = mddev->events;
2699b672
N		 2412 if (rdev->badblocks.size) {
2413 md_super_write(mddev, rdev,
2414 rdev->badblocks.sector,
2415 rdev->badblocks.size << 9,
2416 rdev->bb_page);
2417 rdev->badblocks.size = 0;
2418 }
7bfa19f2 2419
f466722c 2420 } else
36a4e1fe
N		 2421 pr_debug("md: %s (skipping faulty)\n",
2422 bdevname(rdev->bdev, b));
d70ed2e4 2423
7bfa19f2 2424 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2425 /* only need to write one superblock... */
2426 break;
2427 }
a9701a30 2428 md_super_wait(mddev);
850b2b42 2429 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2430
85572d7c 2431 spin_lock(&mddev->lock);
850b2b42
N		 2432 if (mddev->in_sync != sync_req ||
2433 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2434 /* have to write it out again */
85572d7c 2435 spin_unlock(&mddev->lock);
06d91a5f
N		 2436 goto repeat;
2437 }
850b2b42 2438 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
85572d7c 2439 spin_unlock(&mddev->lock);
3d310eb7 2440 wake_up(&mddev->sb_wait);
acb180b0
N		 2441 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2442 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2443
dafb20fa 2444 rdev_for_each(rdev, mddev) {
de393cde
N		 2445 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2446 clear_bit(Blocked, &rdev->flags);
2447
2448 if (any_badblocks_changed)
2699b672 2449 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2450 clear_bit(BlockedBadBlocks, &rdev->flags);
2451 wake_up(&rdev->blocked_wait);
2452 }
2aa82191 2453
23b63f9f 2454 if (mddev_is_clustered(mddev) && ret == 0)
2aa82191 2455 md_cluster_ops->metadata_update_finish(mddev);
1da177e4 2456}
1aee41f6 2457EXPORT_SYMBOL(md_update_sb);
1da177e4 2458
a6da4ef8
GR		 2459static int add_bound_rdev(struct md_rdev *rdev)
2460{
2461 struct mddev *mddev = rdev->mddev;
2462 int err = 0;
2463
2464 if (!mddev->pers->hot_remove_disk) {
2465 /* If there is hot_add_disk but no hot_remove_disk
2466 * then added disks for geometry changes,
2467 * and should be added immediately.
2468 */
2469 super_types[mddev->major_version].
2470 validate_super(mddev, rdev);
2471 err = mddev->pers->hot_add_disk(mddev, rdev);
2472 if (err) {
2473 unbind_rdev_from_array(rdev);
2474 export_rdev(rdev);
2475 return err;
2476 }
2477 }
2478 sysfs_notify_dirent_safe(rdev->sysfs_state);
2479
2480 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2481 if (mddev->degraded)
2482 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2483 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2484 md_new_event(mddev);
2485 md_wakeup_thread(mddev->thread);
2486 return 0;
2487}
1da177e4 2488
7f6ce769 2489/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2490 * We want to accept with case. For this we use cmd_match.
2491 */
2492static int cmd_match(const char *cmd, const char *str)
2493{
2494 /* See if cmd, written into a sysfs file, matches
2495 * str. They must either be the same, or cmd can
2496 * have a trailing newline
2497 */
2498 while (*cmd && *str && *cmd == *str) {
2499 cmd++;
2500 str++;
2501 }
2502 if (*cmd == '\n')
2503 cmd++;
2504 if (*str || *cmd)
2505 return 0;
2506 return 1;
2507}
2508
86e6ffdd
N		 2509struct rdev_sysfs_entry {
2510 struct attribute attr;
3cb03002
N		 2511 ssize_t (*show)(struct md_rdev *, char *);
2512 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2513};
2514
2515static ssize_t
3cb03002 2516state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2517{
2518 char *sep = "";
20a49ff6 2519 size_t len = 0;
758bfc8a 2520 unsigned long flags = ACCESS_ONCE(rdev->flags);
86e6ffdd 2521
758bfc8a 2522 if (test_bit(Faulty, &flags) ||
de393cde 2523 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2524 len+= sprintf(page+len, "%sfaulty",sep);
2525 sep = ",";
2526 }
758bfc8a 2527 if (test_bit(In_sync, &flags)) {
86e6ffdd
N		 2528 len += sprintf(page+len, "%sin_sync",sep);
2529 sep = ",";
2530 }
ac6096e9
SL		 2531 if (test_bit(Journal, &flags)) {
2532 len += sprintf(page+len, "%sjournal",sep);
2533 sep = ",";
2534 }
758bfc8a 2535 if (test_bit(WriteMostly, &flags)) {
f655675b
N		 2536 len += sprintf(page+len, "%swrite_mostly",sep);
2537 sep = ",";
2538 }
758bfc8a 2539 if (test_bit(Blocked, &flags) ||
52c64152 2540 (rdev->badblocks.unacked_exist
758bfc8a 2541 && !test_bit(Faulty, &flags))) {
6bfe0b49
DW		 2542 len += sprintf(page+len, "%sblocked", sep);
2543 sep = ",";
2544 }
758bfc8a 2545 if (!test_bit(Faulty, &flags) &&
f2076e7d 2546 !test_bit(Journal, &flags) &&
758bfc8a 2547 !test_bit(In_sync, &flags)) {
86e6ffdd
N		 2548 len += sprintf(page+len, "%sspare", sep);
2549 sep = ",";
2550 }
758bfc8a 2551 if (test_bit(WriteErrorSeen, &flags)) {
d7a9d443
N		 2552 len += sprintf(page+len, "%swrite_error", sep);
2553 sep = ",";
2554 }
758bfc8a 2555 if (test_bit(WantReplacement, &flags)) {
2d78f8c4
N		 2556 len += sprintf(page+len, "%swant_replacement", sep);
2557 sep = ",";
2558 }
758bfc8a 2559 if (test_bit(Replacement, &flags)) {
2d78f8c4
N		 2560 len += sprintf(page+len, "%sreplacement", sep);
2561 sep = ",";
2562 }
2563
86e6ffdd
N		 2564 return len+sprintf(page+len, "\n");
2565}
2566
45dc2de1 2567static ssize_t
3cb03002 2568state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2569{
2570 /* can write
de393cde 2571 * faulty - simulates an error
45dc2de1 2572 * remove - disconnects the device
f655675b
N		 2573 * writemostly - sets write_mostly
2574 * -writemostly - clears write_mostly
de393cde
N		 2575 * blocked - sets the Blocked flags
2576 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2577 * insync - sets Insync providing device isn't active
f466722c
N		 2578 * -insync - clear Insync for a device with a slot assigned,
2579 * so that it gets rebuilt based on bitmap
d7a9d443
N		 2580 * write_error - sets WriteErrorSeen
2581 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2582 */
2583 int err = -EINVAL;
2584 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2585 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2586 if (test_bit(Faulty, &rdev->flags))
2587 err = 0;
2588 else
2589 err = -EBUSY;
45dc2de1
N		 2590 } else if (cmd_match(buf, "remove")) {
2591 if (rdev->raid_disk >= 0)
2592 err = -EBUSY;
2593 else {
fd01b88c 2594 struct mddev *mddev = rdev->mddev;
45dc2de1 2595 err = 0;
a9720903
GJ		 2596 if (mddev_is_clustered(mddev))
2597 err = md_cluster_ops->remove_disk(mddev, rdev);
2598
2599 if (err == 0) {
2600 md_kick_rdev_from_array(rdev);
2601 if (mddev->pers)
2602 md_update_sb(mddev, 1);
2603 md_new_event(mddev);
2604 }
45dc2de1 2605 }
f655675b
N		 2606 } else if (cmd_match(buf, "writemostly")) {
2607 set_bit(WriteMostly, &rdev->flags);
2608 err = 0;
2609 } else if (cmd_match(buf, "-writemostly")) {
2610 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2611 err = 0;
2612 } else if (cmd_match(buf, "blocked")) {
2613 set_bit(Blocked, &rdev->flags);
2614 err = 0;
2615 } else if (cmd_match(buf, "-blocked")) {
de393cde 2616 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2617 rdev->badblocks.unacked_exist) {
de393cde
N		 2618 /* metadata handler doesn't understand badblocks,
2619 * so we need to fail the device
2620 */
2621 md_error(rdev->mddev, rdev);
2622 }
6bfe0b49 2623 clear_bit(Blocked, &rdev->flags);
de393cde 2624 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2625 wake_up(&rdev->blocked_wait);
2626 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2627 md_wakeup_thread(rdev->mddev->thread);
2628
6d56e278
N		 2629 err = 0;
2630 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2631 set_bit(In_sync, &rdev->flags);
f655675b 2632 err = 0;
f2076e7d
SL		 2633 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2634 !test_bit(Journal, &rdev->flags)) {
e1960f8c
N		 2635 if (rdev->mddev->pers == NULL) {
2636 clear_bit(In_sync, &rdev->flags);
2637 rdev->saved_raid_disk = rdev->raid_disk;
2638 rdev->raid_disk = -1;
2639 err = 0;
2640 }
d7a9d443
N		 2641 } else if (cmd_match(buf, "write_error")) {
2642 set_bit(WriteErrorSeen, &rdev->flags);
2643 err = 0;
2644 } else if (cmd_match(buf, "-write_error")) {
2645 clear_bit(WriteErrorSeen, &rdev->flags);
2646 err = 0;
2d78f8c4
N		 2647 } else if (cmd_match(buf, "want_replacement")) {
2648 /* Any non-spare device that is not a replacement can
2649 * become want_replacement at any time, but we then need to
2650 * check if recovery is needed.
2651 */
2652 if (rdev->raid_disk >= 0 &&
f2076e7d 2653 !test_bit(Journal, &rdev->flags) &&
2d78f8c4
N		 2654 !test_bit(Replacement, &rdev->flags))
2655 set_bit(WantReplacement, &rdev->flags);
2656 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2657 md_wakeup_thread(rdev->mddev->thread);
2658 err = 0;
2659 } else if (cmd_match(buf, "-want_replacement")) {
2660 /* Clearing 'want_replacement' is always allowed.
2661 * Once replacements starts it is too late though.
2662 */
2663 err = 0;
2664 clear_bit(WantReplacement, &rdev->flags);
2665 } else if (cmd_match(buf, "replacement")) {
2666 /* Can only set a device as a replacement when array has not
2667 * yet been started. Once running, replacement is automatic
2668 * from spares, or by assigning 'slot'.
2669 */
2670 if (rdev->mddev->pers)
2671 err = -EBUSY;
2672 else {
2673 set_bit(Replacement, &rdev->flags);
2674 err = 0;
2675 }
2676 } else if (cmd_match(buf, "-replacement")) {
2677 /* Similarly, can only clear Replacement before start */
2678 if (rdev->mddev->pers)
2679 err = -EBUSY;
2680 else {
2681 clear_bit(Replacement, &rdev->flags);
2682 err = 0;
2683 }
a6da4ef8
GR		 2684 } else if (cmd_match(buf, "re-add")) {
2685 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
97f6cd39
GR		 2686 /* clear_bit is performed _after_ all the devices
2687 * have their local Faulty bit cleared. If any writes
2688 * happen in the meantime in the local node, they
2689 * will land in the local bitmap, which will be synced
2690 * by this node eventually
2691 */
2692 if (!mddev_is_clustered(rdev->mddev) ||
2693 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2694 clear_bit(Faulty, &rdev->flags);
2695 err = add_bound_rdev(rdev);
2696 }
a6da4ef8
GR		 2697 } else
2698 err = -EBUSY;
45dc2de1 2699 }
00bcb4ac
N		 2700 if (!err)
2701 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2702 return err ? err : len;
2703}
80ca3a44 2704static struct rdev_sysfs_entry rdev_state =
750f199e 2705__ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2706
4dbcdc75 2707static ssize_t
3cb03002 2708errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2709{
2710 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2711}
2712
2713static ssize_t
3cb03002 2714errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75 2715{
4c9309c0
AD		 2716 unsigned int n;
2717 int rv;
2718
2719 rv = kstrtouint(buf, 10, &n);
2720 if (rv < 0)
2721 return rv;
2722 atomic_set(&rdev->corrected_errors, n);
2723 return len;
4dbcdc75
N		 2724}
2725static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2726__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2727
014236d2 2728static ssize_t
3cb03002 2729slot_show(struct md_rdev *rdev, char *page)
014236d2 2730{
f2076e7d
SL		 2731 if (test_bit(Journal, &rdev->flags))
2732 return sprintf(page, "journal\n");
2733 else if (rdev->raid_disk < 0)
014236d2
N		 2734 return sprintf(page, "none\n");
2735 else
2736 return sprintf(page, "%d\n", rdev->raid_disk);
2737}
2738
2739static ssize_t
3cb03002 2740slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2 2741{
4c9309c0 2742 int slot;
c303da6d 2743 int err;
4c9309c0 2744
f2076e7d
SL		 2745 if (test_bit(Journal, &rdev->flags))
2746 return -EBUSY;
014236d2
N		 2747 if (strncmp(buf, "none", 4)==0)
2748 slot = -1;
4c9309c0
AD		 2749 else {
2750 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2751 if (err < 0)
2752 return err;
2753 }
6c2fce2e 2754 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2755 /* Setting 'slot' on an active array requires also
2756 * updating the 'rd%d' link, and communicating
2757 * with the personality with ->hot_*_disk.
2758 * For now we only support removing
2759 * failed/spare devices. This normally happens automatically,
2760 * but not when the metadata is externally managed.
2761 */
c303da6d
N		 2762 if (rdev->raid_disk == -1)
2763 return -EEXIST;
2764 /* personality does all needed checks */
01393f3d 2765 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d 2766 return -EINVAL;
746d3207
N		 2767 clear_bit(Blocked, &rdev->flags);
2768 remove_and_add_spares(rdev->mddev, rdev);
2769 if (rdev->raid_disk >= 0)
2770 return -EBUSY;
c303da6d
N		 2771 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2772 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2773 } else if (rdev->mddev->pers) {
6c2fce2e 2774 /* Activating a spare .. or possibly reactivating
6d56e278 2775 * if we ever get bitmaps working here.
6c2fce2e 2776 */
cb01c549 2777 int err;
6c2fce2e
NB		 2778
2779 if (rdev->raid_disk != -1)
2780 return -EBUSY;
2781
c6751b2b
N		 2782 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2783 return -EBUSY;
2784
6c2fce2e
NB		 2785 if (rdev->mddev->pers->hot_add_disk == NULL)
2786 return -EINVAL;
2787
ba1b41b6
N		 2788 if (slot >= rdev->mddev->raid_disks &&
2789 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2790 return -ENOSPC;
2791
6c2fce2e
NB		 2792 rdev->raid_disk = slot;
2793 if (test_bit(In_sync, &rdev->flags))
2794 rdev->saved_raid_disk = slot;
2795 else
2796 rdev->saved_raid_disk = -1;
d30519fc 2797 clear_bit(In_sync, &rdev->flags);
8313b8e5 2798 clear_bit(Bitmap_sync, &rdev->flags);
cb01c549
GR		 2799 err = rdev->mddev->pers->
2800 hot_add_disk(rdev->mddev, rdev);
2801 if (err) {
2802 rdev->raid_disk = -1;
2803 return err;
2804 } else
2805 sysfs_notify_dirent_safe(rdev->sysfs_state);
2806 if (sysfs_link_rdev(rdev->mddev, rdev))
2807 /* failure here is OK */;
6c2fce2e 2808 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2809 } else {
ba1b41b6
N		 2810 if (slot >= rdev->mddev->raid_disks &&
2811 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2812 return -ENOSPC;
2813 rdev->raid_disk = slot;
2814 /* assume it is working */
c5d79adb
N		 2815 clear_bit(Faulty, &rdev->flags);
2816 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2817 set_bit(In_sync, &rdev->flags);
00bcb4ac 2818 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2819 }
014236d2
N		 2820 return len;
2821}
2822
014236d2 2823static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2824__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2825
93c8cad0 2826static ssize_t
3cb03002 2827offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2828{
6961ece4 2829 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2830}
2831
2832static ssize_t
3cb03002 2833offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2834{
c6563a8c 2835 unsigned long long offset;
b29bebd6 2836 if (kstrtoull(buf, 10, &offset) < 0)
93c8cad0 2837 return -EINVAL;
8ed0a521 2838 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2839 return -EBUSY;
dd8ac336 2840 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2841 /* Must set offset before size, so overlap checks
2842 * can be sane */
2843 return -EBUSY;
93c8cad0 2844 rdev->data_offset = offset;
25f7fd47 2845 rdev->new_data_offset = offset;
93c8cad0
N		 2846 return len;
2847}
2848
2849static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2850__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2851
c6563a8c
N		 2852static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2853{
2854 return sprintf(page, "%llu\n",
2855 (unsigned long long)rdev->new_data_offset);
2856}
2857
2858static ssize_t new_offset_store(struct md_rdev *rdev,
2859 const char *buf, size_t len)
2860{
2861 unsigned long long new_offset;
2862 struct mddev *mddev = rdev->mddev;
2863
b29bebd6 2864 if (kstrtoull(buf, 10, &new_offset) < 0)
c6563a8c
N		 2865 return -EINVAL;
2866
f851b60d
N		 2867 if (mddev->sync_thread ||
2868 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
c6563a8c
N		 2869 return -EBUSY;
2870 if (new_offset == rdev->data_offset)
2871 /* reset is always permitted */
2872 ;
2873 else if (new_offset > rdev->data_offset) {
2874 /* must not push array size beyond rdev_sectors */
2875 if (new_offset - rdev->data_offset
2876 + mddev->dev_sectors > rdev->sectors)
2877 return -E2BIG;
2878 }
2879 /* Metadata worries about other space details. */
2880
2881 /* decreasing the offset is inconsistent with a backwards
2882 * reshape.
2883 */
2884 if (new_offset < rdev->data_offset &&
2885 mddev->reshape_backwards)
2886 return -EINVAL;
2887 /* Increasing offset is inconsistent with forwards
2888 * reshape. reshape_direction should be set to
2889 * 'backwards' first.
2890 */
2891 if (new_offset > rdev->data_offset &&
2892 !mddev->reshape_backwards)
2893 return -EINVAL;
2894
2895 if (mddev->pers && mddev->persistent &&
2896 !super_types[mddev->major_version]
2897 .allow_new_offset(rdev, new_offset))
2898 return -E2BIG;
2899 rdev->new_data_offset = new_offset;
2900 if (new_offset > rdev->data_offset)
2901 mddev->reshape_backwards = 1;
2902 else if (new_offset < rdev->data_offset)
2903 mddev->reshape_backwards = 0;
2904
2905 return len;
2906}
2907static struct rdev_sysfs_entry rdev_new_offset =
2908__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2909
83303b61 2910static ssize_t
3cb03002 2911rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2912{
dd8ac336 2913 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2914}
2915
c5d79adb
N		 2916static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2917{
2918 /* check if two start/length pairs overlap */
2919 if (s1+l1 <= s2)
2920 return 0;
2921 if (s2+l2 <= s1)
2922 return 0;
2923 return 1;
2924}
2925
b522adcd
DW		 2926static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2927{
2928 unsigned long long blocks;
2929 sector_t new;
2930
b29bebd6 2931 if (kstrtoull(buf, 10, &blocks) < 0)
b522adcd
DW		 2932 return -EINVAL;
2933
2934 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2935 return -EINVAL; /* sector conversion overflow */
2936
2937 new = blocks * 2;
2938 if (new != blocks * 2)
2939 return -EINVAL; /* unsigned long long to sector_t overflow */
2940
2941 *sectors = new;
2942 return 0;
2943}
2944
83303b61 2945static ssize_t
3cb03002 2946rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2947{
fd01b88c 2948 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2949 sector_t oldsectors = rdev->sectors;
b522adcd 2950 sector_t sectors;
27c529bb 2951
f2076e7d
SL		 2952 if (test_bit(Journal, &rdev->flags))
2953 return -EBUSY;
b522adcd 2954 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2955 return -EINVAL;
c6563a8c
N		 2956 if (rdev->data_offset != rdev->new_data_offset)
2957 return -EINVAL; /* too confusing */
0cd17fec 2958 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2959 if (my_mddev->persistent) {
dd8ac336
AN		 2960 sectors = super_types[my_mddev->major_version].
2961 rdev_size_change(rdev, sectors);
2962 if (!sectors)
0cd17fec 2963 return -EBUSY;
dd8ac336 2964 } else if (!sectors)
77304d2a 2965 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2966 rdev->data_offset;
a6468539
N		 2967 if (!my_mddev->pers->resize)
2968 /* Cannot change size for RAID0 or Linear etc */
2969 return -EINVAL;
0cd17fec 2970 }
dd8ac336 2971 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2972 return -EINVAL; /* component must fit device */
0cd17fec 2973
dd8ac336
AN		 2974 rdev->sectors = sectors;
2975 if (sectors > oldsectors && my_mddev->external) {
8b1afc3d
N		 2976 /* Need to check that all other rdevs with the same
2977 * ->bdev do not overlap. 'rcu' is sufficient to walk
2978 * the rdev lists safely.
2979 * This check does not provide a hard guarantee, it
2980 * just helps avoid dangerous mistakes.
c5d79adb 2981 */
fd01b88c 2982 struct mddev *mddev;
c5d79adb 2983 int overlap = 0;
159ec1fc 2984 struct list_head *tmp;
c5d79adb 2985
8b1afc3d 2986 rcu_read_lock();
29ac4aa3 2987 for_each_mddev(mddev, tmp) {
3cb03002 2988 struct md_rdev *rdev2;
c5d79adb 2989
dafb20fa 2990 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 2991 if (rdev->bdev == rdev2->bdev &&
2992 rdev != rdev2 &&
2993 overlaps(rdev->data_offset, rdev->sectors,
2994 rdev2->data_offset,
2995 rdev2->sectors)) {
c5d79adb
N		 2996 overlap = 1;
2997 break;
2998 }
c5d79adb
N		 2999 if (overlap) {
3000 mddev_put(mddev);
3001 break;
3002 }
3003 }
8b1afc3d 3004 rcu_read_unlock();
c5d79adb
N		 3005 if (overlap) {
3006 /* Someone else could have slipped in a size
3007 * change here, but doing so is just silly.
dd8ac336 3008 * We put oldsectors back because we *know* it is
c5d79adb
N		 3009 * safe, and trust userspace not to race with
3010 * itself
3011 */
dd8ac336 3012 rdev->sectors = oldsectors;
c5d79adb
N		 3013 return -EBUSY;
3014 }
3015 }
83303b61
N		 3016 return len;
3017}
3018
3019static struct rdev_sysfs_entry rdev_size =
80ca3a44 3020__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 3021
3cb03002 3022static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 3023{
3024 unsigned long long recovery_start = rdev->recovery_offset;
3025
3026 if (test_bit(In_sync, &rdev->flags) ||
3027 recovery_start == MaxSector)
3028 return sprintf(page, "none\n");
3029
3030 return sprintf(page, "%llu\n", recovery_start);
3031}
3032
3cb03002 3033static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 3034{
3035 unsigned long long recovery_start;
3036
3037 if (cmd_match(buf, "none"))
3038 recovery_start = MaxSector;
b29bebd6 3039 else if (kstrtoull(buf, 10, &recovery_start))
06e3c817
DW		 3040 return -EINVAL;
3041
3042 if (rdev->mddev->pers &&
3043 rdev->raid_disk >= 0)
3044 return -EBUSY;
3045
3046 rdev->recovery_offset = recovery_start;
3047 if (recovery_start == MaxSector)
3048 set_bit(In_sync, &rdev->flags);
3049 else
3050 clear_bit(In_sync, &rdev->flags);
3051 return len;
3052}
3053
3054static struct rdev_sysfs_entry rdev_recovery_start =
3055__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3056
16c791a5
N		 3057static ssize_t
3058badblocks_show(struct badblocks *bb, char *page, int unack);
3059static ssize_t
3060badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3061
3cb03002 3062static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3063{
3064 return badblocks_show(&rdev->badblocks, page, 0);
3065}
3cb03002 3066static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 3067{
de393cde
N		 3068 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3069 /* Maybe that ack was all we needed */
3070 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3071 wake_up(&rdev->blocked_wait);
3072 return rv;
16c791a5
N		 3073}
3074static struct rdev_sysfs_entry rdev_bad_blocks =
3075__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3076
3cb03002 3077static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3078{
3079 return badblocks_show(&rdev->badblocks, page, 1);
3080}
3cb03002 3081static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 3082{
3083 return badblocks_store(&rdev->badblocks, page, len, 1);
3084}
3085static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3086__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3087
86e6ffdd
N		 3088static struct attribute *rdev_default_attrs[] = {
3089 &rdev_state.attr,
4dbcdc75 3090 &rdev_errors.attr,
014236d2 3091 &rdev_slot.attr,
93c8cad0 3092 &rdev_offset.attr,
c6563a8c 3093 &rdev_new_offset.attr,
83303b61 3094 &rdev_size.attr,
06e3c817 3095 &rdev_recovery_start.attr,
16c791a5
N		 3096 &rdev_bad_blocks.attr,
3097 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 3098 NULL,
3099};
3100static ssize_t
3101rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3102{
3103 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3104 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
86e6ffdd
N		 3105
3106 if (!entry->show)
3107 return -EIO;
758bfc8a
N		 3108 if (!rdev->mddev)
3109 return -EBUSY;
3110 return entry->show(rdev, page);
86e6ffdd
N		 3111}
3112
3113static ssize_t
3114rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3115 const char *page, size_t length)
3116{
3117 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3118 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3119 ssize_t rv;
fd01b88c 3120 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3121
3122 if (!entry->store)
3123 return -EIO;
67463acb
N		 3124 if (!capable(CAP_SYS_ADMIN))
3125 return -EACCES;
27c529bb 3126 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3127 if (!rv) {
27c529bb
N		 3128 if (rdev->mddev == NULL)
3129 rv = -EBUSY;
3130 else
3131 rv = entry->store(rdev, page, length);
6a51830e 3132 mddev_unlock(mddev);
ca388059
N		 3133 }
3134 return rv;
86e6ffdd
N		 3135}
3136
3137static void rdev_free(struct kobject *ko)
3138{
3cb03002 3139 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3140 kfree(rdev);
3141}
52cf25d0 3142static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3143 .show = rdev_attr_show,
3144 .store = rdev_attr_store,
3145};
3146static struct kobj_type rdev_ktype = {
3147 .release = rdev_free,
3148 .sysfs_ops = &rdev_sysfs_ops,
3149 .default_attrs = rdev_default_attrs,
3150};
3151
3cb03002 3152int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3153{
3154 rdev->desc_nr = -1;
3155 rdev->saved_raid_disk = -1;
3156 rdev->raid_disk = -1;
3157 rdev->flags = 0;
3158 rdev->data_offset = 0;
c6563a8c 3159 rdev->new_data_offset = 0;
e8bb9a83
N		 3160 rdev->sb_events = 0;
3161 rdev->last_read_error.tv_sec = 0;
3162 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3163 rdev->sb_loaded = 0;
3164 rdev->bb_page = NULL;
e8bb9a83
N		 3165 atomic_set(&rdev->nr_pending, 0);
3166 atomic_set(&rdev->read_errors, 0);
3167 atomic_set(&rdev->corrected_errors, 0);
3168
3169 INIT_LIST_HEAD(&rdev->same_set);
3170 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3171
3172 /* Add space to store bad block list.
3173 * This reserves the space even on arrays where it cannot
3174 * be used - I wonder if that matters
3175 */
3176 rdev->badblocks.count = 0;
486adf72 3177 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
2230dfe4
N		 3178 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3179 seqlock_init(&rdev->badblocks.lock);
3180 if (rdev->badblocks.page == NULL)
3181 return -ENOMEM;
3182
3183 return 0;
e8bb9a83
N		 3184}
3185EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3186/*
3187 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3188 *
3189 * mark the device faulty if:
3190 *
3191 * - the device is nonexistent (zero size)
3192 * - the device has no valid superblock
3193 *
3194 * a faulty rdev _never_ has rdev->sb set.
3195 */
3cb03002 3196static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3197{
3198 char b[BDEVNAME_SIZE];
3199 int err;
3cb03002 3200 struct md_rdev *rdev;
1da177e4
LT		 3201 sector_t size;
3202
9ffae0cf 3203 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3204 if (!rdev) {
3205 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3206 return ERR_PTR(-ENOMEM);
3207 }
1da177e4 3208
2230dfe4
N		 3209 err = md_rdev_init(rdev);
3210 if (err)
3211 goto abort_free;
3212 err = alloc_disk_sb(rdev);
3213 if (err)
1da177e4
LT		 3214 goto abort_free;
3215
c5d79adb 3216 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3217 if (err)
3218 goto abort_free;
3219
f9cb074b 3220 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3221
77304d2a 3222 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4 3223 if (!size) {
f72ffdd6 3224 printk(KERN_WARNING
1da177e4
LT		 3225 "md: %s has zero or unknown size, marking faulty!\n",
3226 bdevname(rdev->bdev,b));
3227 err = -EINVAL;
3228 goto abort_free;
3229 }
3230
3231 if (super_format >= 0) {
3232 err = super_types[super_format].
3233 load_super(rdev, NULL, super_minor);
3234 if (err == -EINVAL) {
df968c4e
N		 3235 printk(KERN_WARNING
3236 "md: %s does not have a valid v%d.%d "
3237 "superblock, not importing!\n",
3238 bdevname(rdev->bdev,b),
3239 super_format, super_minor);
1da177e4
LT		 3240 goto abort_free;
3241 }
3242 if (err < 0) {
f72ffdd6 3243 printk(KERN_WARNING
1da177e4
LT		 3244 "md: could not read %s's sb, not importing!\n",
3245 bdevname(rdev->bdev,b));
3246 goto abort_free;
3247 }
3248 }
6bfe0b49 3249
1da177e4
LT		 3250 return rdev;
3251
3252abort_free:
2699b672
N		 3253 if (rdev->bdev)
3254 unlock_rdev(rdev);
545c8795 3255 md_rdev_clear(rdev);
1da177e4
LT		 3256 kfree(rdev);
3257 return ERR_PTR(err);
3258}
3259
3260/*
3261 * Check a full RAID array for plausibility
3262 */
3263
f72ffdd6 3264static void analyze_sbs(struct mddev *mddev)
1da177e4
LT		 3265{
3266 int i;
3cb03002 3267 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3268 char b[BDEVNAME_SIZE];
3269
3270 freshest = NULL;
dafb20fa 3271 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3272 switch (super_types[mddev->major_version].
3273 load_super(rdev, freshest, mddev->minor_version)) {
3274 case 1:
3275 freshest = rdev;
3276 break;
3277 case 0:
3278 break;
3279 default:
3280 printk( KERN_ERR \
3281 "md: fatal superblock inconsistency in %s"
f72ffdd6 3282 " -- removing from array\n",
1da177e4 3283 bdevname(rdev->bdev,b));
fb56dfef 3284 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3285 }
3286
1da177e4
LT		 3287 super_types[mddev->major_version].
3288 validate_super(mddev, freshest);
3289
3290 i = 0;
dafb20fa 3291 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3292 if (mddev->max_disks &&
3293 (rdev->desc_nr >= mddev->max_disks ||
3294 i > mddev->max_disks)) {
de01dfad
N		 3295 printk(KERN_WARNING
3296 "md: %s: %s: only %d devices permitted\n",
3297 mdname(mddev), bdevname(rdev->bdev, b),
3298 mddev->max_disks);
fb56dfef 3299 md_kick_rdev_from_array(rdev);
de01dfad
N		 3300 continue;
3301 }
1aee41f6 3302 if (rdev != freshest) {
1da177e4
LT		 3303 if (super_types[mddev->major_version].
3304 validate_super(mddev, rdev)) {
3305 printk(KERN_WARNING "md: kicking non-fresh %s"
3306 " from array!\n",
3307 bdevname(rdev->bdev,b));
fb56dfef 3308 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3309 continue;
3310 }
1aee41f6 3311 }
1da177e4
LT		 3312 if (mddev->level == LEVEL_MULTIPATH) {
3313 rdev->desc_nr = i++;
3314 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3315 set_bit(In_sync, &rdev->flags);
f2076e7d
SL		 3316 } else if (rdev->raid_disk >=
3317 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3318 !test_bit(Journal, &rdev->flags)) {
a778b73f
N		 3319 rdev->raid_disk = -1;
3320 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3321 }
3322 }
1da177e4
LT		 3323}
3324
72e02075
N		 3325/* Read a fixed-point number.
3326 * Numbers in sysfs attributes should be in "standard" units where
3327 * possible, so time should be in seconds.
f72ffdd6 3328 * However we internally use a a much smaller unit such as
72e02075
N		 3329 * milliseconds or jiffies.
3330 * This function takes a decimal number with a possible fractional
3331 * component, and produces an integer which is the result of
3332 * multiplying that number by 10^'scale'.
3333 * all without any floating-point arithmetic.
3334 */
3335int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3336{
3337 unsigned long result = 0;
3338 long decimals = -1;
3339 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3340 if (*cp == '.')
3341 decimals = 0;
3342 else if (decimals < scale) {
3343 unsigned int value;
3344 value = *cp - '0';
3345 result = result * 10 + value;
3346 if (decimals >= 0)
3347 decimals++;
3348 }
3349 cp++;
3350 }
3351 if (*cp == '\n')
3352 cp++;
3353 if (*cp)
3354 return -EINVAL;
3355 if (decimals < 0)
3356 decimals = 0;
3357 while (decimals < scale) {
3358 result *= 10;
3359 decimals ++;
3360 }
3361 *res = result;
3362 return 0;
3363}
3364
16f17b39 3365static ssize_t
fd01b88c 3366safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3367{
3368 int msec = (mddev->safemode_delay*1000)/HZ;
3369 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3370}
3371static ssize_t
fd01b88c 3372safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3373{
16f17b39 3374 unsigned long msec;
97ce0a7f 3375
28c1b9fd
GR		 3376 if (mddev_is_clustered(mddev)) {
3377 pr_info("md: Safemode is disabled for clustered mode\n");
3378 return -EINVAL;
3379 }
3380
72e02075 3381 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3382 return -EINVAL;
16f17b39
N		 3383 if (msec == 0)
3384 mddev->safemode_delay = 0;
3385 else {
19052c0e 3386 unsigned long old_delay = mddev->safemode_delay;
1b30e66f
N		 3387 unsigned long new_delay = (msec*HZ)/1000;
3388
3389 if (new_delay == 0)
3390 new_delay = 1;
3391 mddev->safemode_delay = new_delay;
3392 if (new_delay < old_delay || old_delay == 0)
3393 mod_timer(&mddev->safemode_timer, jiffies+1);
16f17b39
N		 3394 }
3395 return len;
3396}
3397static struct md_sysfs_entry md_safe_delay =
80ca3a44 3398__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3399
eae1701f 3400static ssize_t
fd01b88c 3401level_show(struct mddev *mddev, char *page)
eae1701f 3402{
36d091f4
N		 3403 struct md_personality *p;
3404 int ret;
3405 spin_lock(&mddev->lock);
3406 p = mddev->pers;
d9d166c2 3407 if (p)
36d091f4 3408 ret = sprintf(page, "%s\n", p->name);
d9d166c2 3409 else if (mddev->clevel[0])
36d091f4 3410 ret = sprintf(page, "%s\n", mddev->clevel);
d9d166c2 3411 else if (mddev->level != LEVEL_NONE)
36d091f4 3412 ret = sprintf(page, "%d\n", mddev->level);
d9d166c2 3413 else
36d091f4
N		 3414 ret = 0;
3415 spin_unlock(&mddev->lock);
3416 return ret;
eae1701f
N		 3417}
3418
d9d166c2 3419static ssize_t
fd01b88c 3420level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3421{
f2859af6 3422 char clevel[16];
6791875e
N		 3423 ssize_t rv;
3424 size_t slen = len;
db721d32 3425 struct md_personality *pers, *oldpers;
f2859af6 3426 long level;
db721d32 3427 void *priv, *oldpriv;
3cb03002 3428 struct md_rdev *rdev;
245f46c2 3429
6791875e
N		 3430 if (slen == 0 || slen >= sizeof(clevel))
3431 return -EINVAL;
3432
3433 rv = mddev_lock(mddev);
3434 if (rv)
3435 return rv;
3436
245f46c2 3437 if (mddev->pers == NULL) {
6791875e
N		 3438 strncpy(mddev->clevel, buf, slen);
3439 if (mddev->clevel[slen-1] == '\n')
3440 slen--;
3441 mddev->clevel[slen] = 0;
245f46c2 3442 mddev->level = LEVEL_NONE;
6791875e
N		 3443 rv = len;
3444 goto out_unlock;
245f46c2 3445 }
6791875e 3446 rv = -EROFS;
bd8839e0 3447 if (mddev->ro)
6791875e 3448 goto out_unlock;
245f46c2
N		 3449
3450 /* request to change the personality. Need to ensure:
3451 * - array is not engaged in resync/recovery/reshape
3452 * - old personality can be suspended
3453 * - new personality will access other array.
3454 */
3455
6791875e 3456 rv = -EBUSY;
bb4f1e9d 3457 if (mddev->sync_thread ||
f851b60d 3458 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
bb4f1e9d
N		 3459 mddev->reshape_position != MaxSector ||
3460 mddev->sysfs_active)
6791875e 3461 goto out_unlock;
245f46c2 3462
6791875e 3463 rv = -EINVAL;
245f46c2
N		 3464 if (!mddev->pers->quiesce) {
3465 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3466 mdname(mddev), mddev->pers->name);
6791875e 3467 goto out_unlock;
245f46c2
N		 3468 }
3469
3470 /* Now find the new personality */
6791875e
N		 3471 strncpy(clevel, buf, slen);
3472 if (clevel[slen-1] == '\n')
3473 slen--;
3474 clevel[slen] = 0;
b29bebd6 3475 if (kstrtol(clevel, 10, &level))
f2859af6 3476 level = LEVEL_NONE;
245f46c2 3477
f2859af6
DW		 3478 if (request_module("md-%s", clevel) != 0)
3479 request_module("md-level-%s", clevel);
245f46c2 3480 spin_lock(&pers_lock);
f2859af6 3481 pers = find_pers(level, clevel);
245f46c2
N		 3482 if (!pers || !try_module_get(pers->owner)) {
3483 spin_unlock(&pers_lock);
f2859af6 3484 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
6791875e
N		 3485 rv = -EINVAL;
3486 goto out_unlock;
245f46c2
N		 3487 }
3488 spin_unlock(&pers_lock);
3489
3490 if (pers == mddev->pers) {
3491 /* Nothing to do! */
3492 module_put(pers->owner);
6791875e
N		 3493 rv = len;
3494 goto out_unlock;
245f46c2
N		 3495 }
3496 if (!pers->takeover) {
3497 module_put(pers->owner);
3498 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3499 mdname(mddev), clevel);
6791875e
N		 3500 rv = -EINVAL;
3501 goto out_unlock;
245f46c2
N		 3502 }
3503
dafb20fa 3504 rdev_for_each(rdev, mddev)
e93f68a1
N		 3505 rdev->new_raid_disk = rdev->raid_disk;
3506
245f46c2
N		 3507 /* ->takeover must set new_* and/or delta_disks
3508 * if it succeeds, and may set them when it fails.
3509 */
3510 priv = pers->takeover(mddev);
3511 if (IS_ERR(priv)) {
3512 mddev->new_level = mddev->level;
3513 mddev->new_layout = mddev->layout;
664e7c41 3514 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3515 mddev->raid_disks -= mddev->delta_disks;
3516 mddev->delta_disks = 0;
2c810cdd 3517 mddev->reshape_backwards = 0;
245f46c2
N		 3518 module_put(pers->owner);
3519 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3520 mdname(mddev), clevel);
6791875e
N		 3521 rv = PTR_ERR(priv);
3522 goto out_unlock;
245f46c2
N		 3523 }
3524
3525 /* Looks like we have a winner */
3526 mddev_suspend(mddev);
5aa61f42 3527 mddev_detach(mddev);
36d091f4
N		 3528
3529 spin_lock(&mddev->lock);
db721d32
N		 3530 oldpers = mddev->pers;
3531 oldpriv = mddev->private;
3532 mddev->pers = pers;
3533 mddev->private = priv;
3534 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3535 mddev->level = mddev->new_level;
3536 mddev->layout = mddev->new_layout;
3537 mddev->chunk_sectors = mddev->new_chunk_sectors;
3538 mddev->delta_disks = 0;
3539 mddev->reshape_backwards = 0;
3540 mddev->degraded = 0;
36d091f4 3541 spin_unlock(&mddev->lock);
db721d32
N		 3542
3543 if (oldpers->sync_request == NULL &&
3544 mddev->external) {
3545 /* We are converting from a no-redundancy array
3546 * to a redundancy array and metadata is managed
3547 * externally so we need to be sure that writes
3548 * won't block due to a need to transition
3549 * clean->dirty
3550 * until external management is started.
3551 */
3552 mddev->in_sync = 0;
3553 mddev->safemode_delay = 0;
3554 mddev->safemode = 0;
3555 }
f72ffdd6 3556
db721d32
N		 3557 oldpers->free(mddev, oldpriv);
3558
3559 if (oldpers->sync_request == NULL &&
a64c876f
N		 3560 pers->sync_request != NULL) {
3561 /* need to add the md_redundancy_group */
3562 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3563 printk(KERN_WARNING
3564 "md: cannot register extra attributes for %s\n",
3565 mdname(mddev));
388975cc 3566 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
f72ffdd6 3567 }
db721d32 3568 if (oldpers->sync_request != NULL &&
a64c876f
N		 3569 pers->sync_request == NULL) {
3570 /* need to remove the md_redundancy_group */
3571 if (mddev->to_remove == NULL)
3572 mddev->to_remove = &md_redundancy_group;
3573 }
3574
dafb20fa 3575 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3576 if (rdev->raid_disk < 0)
3577 continue;
bf2cb0da 3578 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3579 rdev->new_raid_disk = -1;
3580 if (rdev->new_raid_disk == rdev->raid_disk)
3581 continue;
36fad858 3582 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3583 }
dafb20fa 3584 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3585 if (rdev->raid_disk < 0)
3586 continue;
3587 if (rdev->new_raid_disk == rdev->raid_disk)
3588 continue;
3589 rdev->raid_disk = rdev->new_raid_disk;
3590 if (rdev->raid_disk < 0)
3a981b03 3591 clear_bit(In_sync, &rdev->flags);
e93f68a1 3592 else {
36fad858
NK		 3593 if (sysfs_link_rdev(mddev, rdev))
3594 printk(KERN_WARNING "md: cannot register rd%d"
3595 " for %s after level change\n",
3596 rdev->raid_disk, mdname(mddev));
3a981b03 3597 }
e93f68a1
N		 3598 }
3599
db721d32 3600 if (pers->sync_request == NULL) {
9af204cf
TM		 3601 /* this is now an array without redundancy, so
3602 * it must always be in_sync
3603 */
3604 mddev->in_sync = 1;
3605 del_timer_sync(&mddev->safemode_timer);
3606 }
02e5f5c0 3607 blk_set_stacking_limits(&mddev->queue->limits);
245f46c2 3608 pers->run(mddev);
245f46c2 3609 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3610 mddev_resume(mddev);
830778a1
N		 3611 if (!mddev->thread)
3612 md_update_sb(mddev, 1);
5cac7861 3613 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3614 md_new_event(mddev);
6791875e
N		 3615 rv = len;
3616out_unlock:
3617 mddev_unlock(mddev);
d9d166c2
N		 3618 return rv;
3619}
3620
3621static struct md_sysfs_entry md_level =
80ca3a44 3622__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3623
d4dbd025 3624static ssize_t
fd01b88c 3625layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3626{
3627 /* just a number, not meaningful for all levels */
08a02ecd
N		 3628 if (mddev->reshape_position != MaxSector &&
3629 mddev->layout != mddev->new_layout)
3630 return sprintf(page, "%d (%d)\n",
3631 mddev->new_layout, mddev->layout);
d4dbd025
N		 3632 return sprintf(page, "%d\n", mddev->layout);
3633}
3634
3635static ssize_t
fd01b88c 3636layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025 3637{
4c9309c0 3638 unsigned int n;
6791875e 3639 int err;
d4dbd025 3640
4c9309c0
AD		 3641 err = kstrtouint(buf, 10, &n);
3642 if (err < 0)
3643 return err;
6791875e
N		 3644 err = mddev_lock(mddev);
3645 if (err)
3646 return err;
d4dbd025 3647
b3546035 3648 if (mddev->pers) {
50ac168a 3649 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3650 err = -EBUSY;
3651 else if (mddev->ro)
3652 err = -EROFS;
3653 else {
3654 mddev->new_layout = n;
3655 err = mddev->pers->check_reshape(mddev);
3656 if (err)
3657 mddev->new_layout = mddev->layout;
597a711b 3658 }
b3546035 3659 } else {
08a02ecd 3660 mddev->new_layout = n;
b3546035
N		 3661 if (mddev->reshape_position == MaxSector)
3662 mddev->layout = n;
3663 }
6791875e
N		 3664 mddev_unlock(mddev);
3665 return err ?: len;
d4dbd025
N		 3666}
3667static struct md_sysfs_entry md_layout =
80ca3a44 3668__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025 3669
eae1701f 3670static ssize_t
fd01b88c 3671raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3672{
bb636547
N		 3673 if (mddev->raid_disks == 0)
3674 return 0;
08a02ecd
N		 3675 if (mddev->reshape_position != MaxSector &&
3676 mddev->delta_disks != 0)
3677 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3678 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3679 return sprintf(page, "%d\n", mddev->raid_disks);
3680}
3681
fd01b88c 3682static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3683
3684static ssize_t
fd01b88c 3685raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3686{
4c9309c0 3687 unsigned int n;
6791875e 3688 int err;
da943b99 3689
4c9309c0
AD		 3690 err = kstrtouint(buf, 10, &n);
3691 if (err < 0)
3692 return err;
da943b99 3693
6791875e
N		 3694 err = mddev_lock(mddev);
3695 if (err)
3696 return err;
da943b99 3697 if (mddev->pers)
6791875e 3698 err = update_raid_disks(mddev, n);
08a02ecd 3699 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3700 struct md_rdev *rdev;
08a02ecd 3701 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c 3702
6791875e 3703 err = -EINVAL;
c6563a8c
N		 3704 rdev_for_each(rdev, mddev) {
3705 if (olddisks < n &&
3706 rdev->data_offset < rdev->new_data_offset)
6791875e 3707 goto out_unlock;
c6563a8c
N		 3708 if (olddisks > n &&
3709 rdev->data_offset > rdev->new_data_offset)
6791875e 3710 goto out_unlock;
c6563a8c 3711 }
6791875e 3712 err = 0;
08a02ecd
N		 3713 mddev->delta_disks = n - olddisks;
3714 mddev->raid_disks = n;
2c810cdd 3715 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3716 } else
da943b99 3717 mddev->raid_disks = n;
6791875e
N		 3718out_unlock:
3719 mddev_unlock(mddev);
3720 return err ? err : len;
da943b99
N		 3721}
3722static struct md_sysfs_entry md_raid_disks =
80ca3a44 3723__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3724
3b34380a 3725static ssize_t
fd01b88c 3726chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3727{
08a02ecd 3728 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3729 mddev->chunk_sectors != mddev->new_chunk_sectors)
3730 return sprintf(page, "%d (%d)\n",
3731 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3732 mddev->chunk_sectors << 9);
3733 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3734}
3735
3736static ssize_t
fd01b88c 3737chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3738{
4c9309c0 3739 unsigned long n;
6791875e 3740 int err;
3b34380a 3741
4c9309c0
AD		 3742 err = kstrtoul(buf, 10, &n);
3743 if (err < 0)
3744 return err;
3b34380a 3745
6791875e
N		 3746 err = mddev_lock(mddev);
3747 if (err)
3748 return err;
b3546035 3749 if (mddev->pers) {
50ac168a 3750 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3751 err = -EBUSY;
3752 else if (mddev->ro)
3753 err = -EROFS;
3754 else {
3755 mddev->new_chunk_sectors = n >> 9;
3756 err = mddev->pers->check_reshape(mddev);
3757 if (err)
3758 mddev->new_chunk_sectors = mddev->chunk_sectors;
597a711b 3759 }
b3546035 3760 } else {
664e7c41 3761 mddev->new_chunk_sectors = n >> 9;
b3546035 3762 if (mddev->reshape_position == MaxSector)
9d8f0363 3763 mddev->chunk_sectors = n >> 9;
b3546035 3764 }
6791875e
N		 3765 mddev_unlock(mddev);
3766 return err ?: len;
3b34380a
N		 3767}
3768static struct md_sysfs_entry md_chunk_size =
80ca3a44 3769__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3770
a94213b1 3771static ssize_t
fd01b88c 3772resync_start_show(struct mddev *mddev, char *page)
a94213b1 3773{
d1a7c503
N		 3774 if (mddev->recovery_cp == MaxSector)
3775 return sprintf(page, "none\n");
a94213b1
N		 3776 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3777}
3778
3779static ssize_t
fd01b88c 3780resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3781{
4c9309c0 3782 unsigned long long n;
6791875e 3783 int err;
4c9309c0
AD		 3784
3785 if (cmd_match(buf, "none"))
3786 n = MaxSector;
3787 else {
3788 err = kstrtoull(buf, 10, &n);
3789 if (err < 0)
3790 return err;
3791 if (n != (sector_t)n)
3792 return -EINVAL;
3793 }
a94213b1 3794
6791875e
N		 3795 err = mddev_lock(mddev);
3796 if (err)
3797 return err;
b098636c 3798 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6791875e 3799 err = -EBUSY;
a94213b1 3800
6791875e
N		 3801 if (!err) {
3802 mddev->recovery_cp = n;
3803 if (mddev->pers)
3804 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3805 }
3806 mddev_unlock(mddev);
3807 return err ?: len;
a94213b1
N		 3808}
3809static struct md_sysfs_entry md_resync_start =
750f199e
N		 3810__ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3811 resync_start_show, resync_start_store);
a94213b1 3812
9e653b63
N		 3813/*
3814 * The array state can be:
3815 *
3816 * clear
3817 * No devices, no size, no level
3818 * Equivalent to STOP_ARRAY ioctl
3819 * inactive
3820 * May have some settings, but array is not active
3821 * all IO results in error
3822 * When written, doesn't tear down array, but just stops it
3823 * suspended (not supported yet)
3824 * All IO requests will block. The array can be reconfigured.
910d8cb3 3825 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3826 * readonly
3827 * no resync can happen. no superblocks get written.
3828 * write requests fail
3829 * read-auto
3830 * like readonly, but behaves like 'clean' on a write request.
3831 *
3832 * clean - no pending writes, but otherwise active.
3833 * When written to inactive array, starts without resync
3834 * If a write request arrives then
3835 * if metadata is known, mark 'dirty' and switch to 'active'.
3836 * if not known, block and switch to write-pending
3837 * If written to an active array that has pending writes, then fails.
3838 * active
3839 * fully active: IO and resync can be happening.
3840 * When written to inactive array, starts with resync
3841 *
3842 * write-pending
3843 * clean, but writes are blocked waiting for 'active' to be written.
3844 *
3845 * active-idle
3846 * like active, but no writes have been seen for a while (100msec).
3847 *
3848 */
3849enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3850 write_pending, active_idle, bad_word};
05381954 3851static char *array_states[] = {
9e653b63
N		 3852 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3853 "write-pending", "active-idle", NULL };
3854
3855static int match_word(const char *word, char **list)
3856{
3857 int n;
3858 for (n=0; list[n]; n++)
3859 if (cmd_match(word, list[n]))
3860 break;
3861 return n;
3862}
3863
3864static ssize_t
fd01b88c 3865array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3866{
3867 enum array_state st = inactive;
3868
3869 if (mddev->pers)
3870 switch(mddev->ro) {
3871 case 1:
3872 st = readonly;
3873 break;
3874 case 2:
3875 st = read_auto;
3876 break;
3877 case 0:
3878 if (mddev->in_sync)
3879 st = clean;
070dc6dd 3880 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3881 st = write_pending;
9e653b63
N		 3882 else if (mddev->safemode)
3883 st = active_idle;
3884 else
3885 st = active;
3886 }
3887 else {
3888 if (list_empty(&mddev->disks) &&
3889 mddev->raid_disks == 0 &&
58c0fed4 3890 mddev->dev_sectors == 0)
9e653b63
N		 3891 st = clear;
3892 else
3893 st = inactive;
3894 }
3895 return sprintf(page, "%s\n", array_states[st]);
3896}
3897
f72ffdd6
N		 3898static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3899static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3900static int do_md_run(struct mddev *mddev);
fd01b88c 3901static int restart_array(struct mddev *mddev);
9e653b63
N		 3902
3903static ssize_t
fd01b88c 3904array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63 3905{
6791875e 3906 int err;
9e653b63 3907 enum array_state st = match_word(buf, array_states);
6791875e
N		 3908
3909 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3910 /* don't take reconfig_mutex when toggling between
3911 * clean and active
3912 */
3913 spin_lock(&mddev->lock);
3914 if (st == active) {
3915 restart_array(mddev);
3916 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3917 wake_up(&mddev->sb_wait);
3918 err = 0;
3919 } else /* st == clean */ {
3920 restart_array(mddev);
3921 if (atomic_read(&mddev->writes_pending) == 0) {
3922 if (mddev->in_sync == 0) {
3923 mddev->in_sync = 1;
3924 if (mddev->safemode == 1)
3925 mddev->safemode = 0;
3926 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3927 }
3928 err = 0;
3929 } else
3930 err = -EBUSY;
3931 }
3932 spin_unlock(&mddev->lock);
c008f1d3 3933 return err ?: len;
6791875e
N		 3934 }
3935 err = mddev_lock(mddev);
3936 if (err)
3937 return err;
3938 err = -EINVAL;
9e653b63
N		 3939 switch(st) {
3940 case bad_word:
3941 break;
3942 case clear:
3943 /* stopping an active array */
a05b7ea0 3944 err = do_md_stop(mddev, 0, NULL);
9e653b63
N		 3945 break;
3946 case inactive:
3947 /* stopping an active array */
90cf195d 3948 if (mddev->pers)
a05b7ea0 3949 err = do_md_stop(mddev, 2, NULL);
90cf195d 3950 else
e691063a 3951 err = 0; /* already inactive */
9e653b63
N		 3952 break;
3953 case suspended:
3954 break; /* not supported yet */
3955 case readonly:
3956 if (mddev->pers)
a05b7ea0 3957 err = md_set_readonly(mddev, NULL);
9e653b63
N		 3958 else {
3959 mddev->ro = 1;
648b629e 3960 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3961 err = do_md_run(mddev);
3962 }
3963 break;
3964 case read_auto:
9e653b63 3965 if (mddev->pers) {
80268ee9 3966 if (mddev->ro == 0)
a05b7ea0 3967 err = md_set_readonly(mddev, NULL);
80268ee9 3968 else if (mddev->ro == 1)
648b629e
N		 3969 err = restart_array(mddev);
3970 if (err == 0) {
3971 mddev->ro = 2;
3972 set_disk_ro(mddev->gendisk, 0);
3973 }
9e653b63
N		 3974 } else {
3975 mddev->ro = 2;
3976 err = do_md_run(mddev);
3977 }
3978 break;
3979 case clean:
3980 if (mddev->pers) {
339421de
SL		 3981 err = restart_array(mddev);
3982 if (err)
3983 break;
85572d7c 3984 spin_lock(&mddev->lock);
9e653b63 3985 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3986 if (mddev->in_sync == 0) {
3987 mddev->in_sync = 1;
31a59e34
N		 3988 if (mddev->safemode == 1)
3989 mddev->safemode = 0;
070dc6dd 3990 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3991 }
3992 err = 0;
3993 } else
3994 err = -EBUSY;
85572d7c 3995 spin_unlock(&mddev->lock);
5bf29597
N		 3996 } else
3997 err = -EINVAL;
9e653b63
N		 3998 break;
3999 case active:
4000 if (mddev->pers) {
339421de
SL		 4001 err = restart_array(mddev);
4002 if (err)
4003 break;
070dc6dd 4004 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 4005 wake_up(&mddev->sb_wait);
4006 err = 0;
4007 } else {
4008 mddev->ro = 0;
648b629e 4009 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 4010 err = do_md_run(mddev);
4011 }
4012 break;
4013 case write_pending:
4014 case active_idle:
4015 /* these cannot be set */
4016 break;
4017 }
6791875e
N		 4018
4019 if (!err) {
1d23f178
N		 4020 if (mddev->hold_active == UNTIL_IOCTL)
4021 mddev->hold_active = 0;
00bcb4ac 4022 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4023 }
6791875e
N		 4024 mddev_unlock(mddev);
4025 return err ?: len;
9e653b63 4026}
80ca3a44 4027static struct md_sysfs_entry md_array_state =
750f199e 4028__ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 4029
1e50915f 4030static ssize_t
fd01b88c 4031max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 4032 return sprintf(page, "%d\n",
4033 atomic_read(&mddev->max_corr_read_errors));
4034}
4035
4036static ssize_t
fd01b88c 4037max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f 4038{
4c9309c0
AD		 4039 unsigned int n;
4040 int rv;
1e50915f 4041
4c9309c0
AD		 4042 rv = kstrtouint(buf, 10, &n);
4043 if (rv < 0)
4044 return rv;
4045 atomic_set(&mddev->max_corr_read_errors, n);
4046 return len;
1e50915f
RB		 4047}
4048
4049static struct md_sysfs_entry max_corr_read_errors =
4050__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4051 max_corrected_read_errors_store);
4052
6d7ff738 4053static ssize_t
fd01b88c 4054null_show(struct mddev *mddev, char *page)
6d7ff738
N		 4055{
4056 return -EINVAL;
4057}
4058
4059static ssize_t
fd01b88c 4060new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 4061{
4062 /* buf must be %d:%d\n? giving major and minor numbers */
4063 /* The new device is added to the array.
4064 * If the array has a persistent superblock, we read the
4065 * superblock to initialise info and check validity.
4066 * Otherwise, only checking done is that in bind_rdev_to_array,
4067 * which mainly checks size.
4068 */
4069 char *e;
4070 int major = simple_strtoul(buf, &e, 10);
4071 int minor;
4072 dev_t dev;
3cb03002 4073 struct md_rdev *rdev;
6d7ff738
N		 4074 int err;
4075
4076 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4077 return -EINVAL;
4078 minor = simple_strtoul(e+1, &e, 10);
4079 if (*e && *e != '\n')
4080 return -EINVAL;
4081 dev = MKDEV(major, minor);
4082 if (major != MAJOR(dev) ||
4083 minor != MINOR(dev))
4084 return -EOVERFLOW;
4085
6791875e
N		 4086 flush_workqueue(md_misc_wq);
4087
4088 err = mddev_lock(mddev);
4089 if (err)
4090 return err;
6d7ff738
N		 4091 if (mddev->persistent) {
4092 rdev = md_import_device(dev, mddev->major_version,
4093 mddev->minor_version);
4094 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 4095 struct md_rdev *rdev0
4096 = list_entry(mddev->disks.next,
4097 struct md_rdev, same_set);
6d7ff738
N		 4098 err = super_types[mddev->major_version]
4099 .load_super(rdev, rdev0, mddev->minor_version);
4100 if (err < 0)
4101 goto out;
4102 }
c5d79adb
N		 4103 } else if (mddev->external)
4104 rdev = md_import_device(dev, -2, -1);
4105 else
6d7ff738
N		 4106 rdev = md_import_device(dev, -1, -1);
4107
9a8c0fa8
N		 4108 if (IS_ERR(rdev)) {
4109 mddev_unlock(mddev);
6d7ff738 4110 return PTR_ERR(rdev);
9a8c0fa8 4111 }
6d7ff738
N		 4112 err = bind_rdev_to_array(rdev, mddev);
4113 out:
4114 if (err)
4115 export_rdev(rdev);
6791875e 4116 mddev_unlock(mddev);
6d7ff738
N		 4117 return err ? err : len;
4118}
4119
4120static struct md_sysfs_entry md_new_device =
80ca3a44 4121__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 4122
9b1d1dac 4123static ssize_t
fd01b88c 4124bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 4125{
4126 char *end;
4127 unsigned long chunk, end_chunk;
6791875e 4128 int err;
9b1d1dac 4129
6791875e
N		 4130 err = mddev_lock(mddev);
4131 if (err)
4132 return err;
9b1d1dac
PC		 4133 if (!mddev->bitmap)
4134 goto out;
4135 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4136 while (*buf) {
4137 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4138 if (buf == end) break;
4139 if (*end == '-') { /* range */
4140 buf = end + 1;
4141 end_chunk = simple_strtoul(buf, &end, 0);
4142 if (buf == end) break;
4143 }
4144 if (*end && !isspace(*end)) break;
4145 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 4146 buf = skip_spaces(end);
9b1d1dac
PC		 4147 }
4148 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4149out:
6791875e 4150 mddev_unlock(mddev);
9b1d1dac
PC		 4151 return len;
4152}
4153
4154static struct md_sysfs_entry md_bitmap =
4155__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4156
a35b0d69 4157static ssize_t
fd01b88c 4158size_show(struct mddev *mddev, char *page)
a35b0d69 4159{
58c0fed4
AN		 4160 return sprintf(page, "%llu\n",
4161 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 4162}
4163
fd01b88c 4164static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 4165
4166static ssize_t
fd01b88c 4167size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 4168{
4169 /* If array is inactive, we can reduce the component size, but
4170 * not increase it (except from 0).
4171 * If array is active, we can try an on-line resize
4172 */
b522adcd
DW		 4173 sector_t sectors;
4174 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4175
58c0fed4
AN		 4176 if (err < 0)
4177 return err;
6791875e
N		 4178 err = mddev_lock(mddev);
4179 if (err)
4180 return err;
a35b0d69 4181 if (mddev->pers) {
58c0fed4 4182 err = update_size(mddev, sectors);
850b2b42 4183 md_update_sb(mddev, 1);
a35b0d69 4184 } else {
58c0fed4
AN		 4185 if (mddev->dev_sectors == 0 ||
4186 mddev->dev_sectors > sectors)
4187 mddev->dev_sectors = sectors;
a35b0d69
N		 4188 else
4189 err = -ENOSPC;
4190 }
6791875e 4191 mddev_unlock(mddev);
a35b0d69
N		 4192 return err ? err : len;
4193}
4194
4195static struct md_sysfs_entry md_size =
80ca3a44 4196__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4197
83f0d77a 4198/* Metadata version.
e691063a
N		 4199 * This is one of
4200 * 'none' for arrays with no metadata (good luck...)
4201 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4202 * or N.M for internally known formats
4203 */
4204static ssize_t
fd01b88c 4205metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4206{
4207 if (mddev->persistent)
4208 return sprintf(page, "%d.%d\n",
4209 mddev->major_version, mddev->minor_version);
e691063a
N		 4210 else if (mddev->external)
4211 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4212 else
4213 return sprintf(page, "none\n");
4214}
4215
4216static ssize_t
fd01b88c 4217metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4218{
4219 int major, minor;
4220 char *e;
6791875e 4221 int err;
ea43ddd8
N		 4222 /* Changing the details of 'external' metadata is
4223 * always permitted. Otherwise there must be
4224 * no devices attached to the array.
4225 */
6791875e
N		 4226
4227 err = mddev_lock(mddev);
4228 if (err)
4229 return err;
4230 err = -EBUSY;
ea43ddd8
N		 4231 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4232 ;
4233 else if (!list_empty(&mddev->disks))
6791875e 4234 goto out_unlock;
8bb93aac 4235
6791875e 4236 err = 0;
8bb93aac
N		 4237 if (cmd_match(buf, "none")) {
4238 mddev->persistent = 0;
e691063a
N		 4239 mddev->external = 0;
4240 mddev->major_version = 0;
4241 mddev->minor_version = 90;
6791875e 4242 goto out_unlock;
e691063a
N		 4243 }
4244 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4245 size_t namelen = len-9;
e691063a
N		 4246 if (namelen >= sizeof(mddev->metadata_type))
4247 namelen = sizeof(mddev->metadata_type)-1;
4248 strncpy(mddev->metadata_type, buf+9, namelen);
4249 mddev->metadata_type[namelen] = 0;
4250 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4251 mddev->metadata_type[--namelen] = 0;
4252 mddev->persistent = 0;
4253 mddev->external = 1;
8bb93aac
N		 4254 mddev->major_version = 0;
4255 mddev->minor_version = 90;
6791875e 4256 goto out_unlock;
8bb93aac
N		 4257 }
4258 major = simple_strtoul(buf, &e, 10);
6791875e 4259 err = -EINVAL;
8bb93aac 4260 if (e==buf || *e != '.')
6791875e 4261 goto out_unlock;
8bb93aac
N		 4262 buf = e+1;
4263 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4264 if (e==buf || (*e && *e != '\n') )
6791875e
N		 4265 goto out_unlock;
4266 err = -ENOENT;
50511da3 4267 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
6791875e 4268 goto out_unlock;
8bb93aac
N		 4269 mddev->major_version = major;
4270 mddev->minor_version = minor;
4271 mddev->persistent = 1;
e691063a 4272 mddev->external = 0;
6791875e
N		 4273 err = 0;
4274out_unlock:
4275 mddev_unlock(mddev);
4276 return err ?: len;
8bb93aac
N		 4277}
4278
4279static struct md_sysfs_entry md_metadata =
750f199e 4280__ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4281
24dd469d 4282static ssize_t
fd01b88c 4283action_show(struct mddev *mddev, char *page)
24dd469d 4284{
7eec314d 4285 char *type = "idle";
b7b17c9b
N		 4286 unsigned long recovery = mddev->recovery;
4287 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
b6a9ce68 4288 type = "frozen";
b7b17c9b
N		 4289 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4290 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4291 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
ccfcc3c1 4292 type = "reshape";
b7b17c9b
N		 4293 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4294 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
24dd469d 4295 type = "resync";
b7b17c9b 4296 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
24dd469d
N		 4297 type = "check";
4298 else
4299 type = "repair";
b7b17c9b 4300 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
24dd469d 4301 type = "recover";
985ca973
N		 4302 else if (mddev->reshape_position != MaxSector)
4303 type = "reshape";
24dd469d
N		 4304 }
4305 return sprintf(page, "%s\n", type);
4306}
4307
4308static ssize_t
fd01b88c 4309action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4310{
7eec314d
N		 4311 if (!mddev->pers || !mddev->pers->sync_request)
4312 return -EINVAL;
4313
b6a9ce68
N		 4314
4315 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
56ccc112
N		 4316 if (cmd_match(page, "frozen"))
4317 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4318 else
4319 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
8e8e2518
N		 4320 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4321 mddev_lock(mddev) == 0) {
4322 flush_workqueue(md_misc_wq);
4323 if (mddev->sync_thread) {
4324 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6791875e 4325 md_reap_sync_thread(mddev);
6791875e 4326 }
8e8e2518 4327 mddev_unlock(mddev);
7eec314d 4328 }
312045ee 4329 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
24dd469d 4330 return -EBUSY;
72a23c21 4331 else if (cmd_match(page, "resync"))
56ccc112 4332 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
72a23c21 4333 else if (cmd_match(page, "recover")) {
56ccc112 4334 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
72a23c21 4335 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
72a23c21 4336 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4337 int err;
4338 if (mddev->pers->start_reshape == NULL)
4339 return -EINVAL;
6791875e
N		 4340 err = mddev_lock(mddev);
4341 if (!err) {
312045ee
N		 4342 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4343 err = -EBUSY;
4344 else {
4345 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4346 err = mddev->pers->start_reshape(mddev);
4347 }
6791875e
N		 4348 mddev_unlock(mddev);
4349 }
16484bf5
N		 4350 if (err)
4351 return err;
a99ac971 4352 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4353 } else {
bce74dac 4354 if (cmd_match(page, "check"))
7eec314d 4355 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4356 else if (!cmd_match(page, "repair"))
7eec314d 4357 return -EINVAL;
56ccc112 4358 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
7eec314d
N		 4359 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4360 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4361 }
48c26ddc
N		 4362 if (mddev->ro == 2) {
4363 /* A write to sync_action is enough to justify
4364 * canceling read-auto mode
4365 */
4366 mddev->ro = 0;
4367 md_wakeup_thread(mddev->sync_thread);
4368 }
03c902e1 4369 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4370 md_wakeup_thread(mddev->thread);
00bcb4ac 4371 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4372 return len;
4373}
4374
c4a39551 4375static struct md_sysfs_entry md_scan_mode =
750f199e 4376__ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
c4a39551
JB		 4377
4378static ssize_t
4379last_sync_action_show(struct mddev *mddev, char *page)
4380{
4381 return sprintf(page, "%s\n", mddev->last_sync_action);
4382}
4383
4384static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4385
9d88883e 4386static ssize_t
fd01b88c 4387mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4388{
4389 return sprintf(page, "%llu\n",
7f7583d4
JM		 4390 (unsigned long long)
4391 atomic64_read(&mddev->resync_mismatches));
9d88883e
N		 4392}
4393
80ca3a44 4394static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4395
88202a0c 4396static ssize_t
fd01b88c 4397sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4398{
4399 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4400 mddev->sync_speed_min ? "local": "system");
4401}
4402
4403static ssize_t
fd01b88c 4404sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c 4405{
4c9309c0
AD		 4406 unsigned int min;
4407 int rv;
4408
88202a0c 4409 if (strncmp(buf, "system", 6)==0) {
4c9309c0
AD		 4410 min = 0;
4411 } else {
4412 rv = kstrtouint(buf, 10, &min);
4413 if (rv < 0)
4414 return rv;
4415 if (min == 0)
4416 return -EINVAL;
88202a0c 4417 }
88202a0c
N		 4418 mddev->sync_speed_min = min;
4419 return len;
4420}
4421
4422static struct md_sysfs_entry md_sync_min =
4423__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4424
4425static ssize_t
fd01b88c 4426sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4427{
4428 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4429 mddev->sync_speed_max ? "local": "system");
4430}
4431
4432static ssize_t
fd01b88c 4433sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c 4434{
4c9309c0
AD		 4435 unsigned int max;
4436 int rv;
4437
88202a0c 4438 if (strncmp(buf, "system", 6)==0) {
4c9309c0
AD		 4439 max = 0;
4440 } else {
4441 rv = kstrtouint(buf, 10, &max);
4442 if (rv < 0)
4443 return rv;
4444 if (max == 0)
4445 return -EINVAL;
88202a0c 4446 }
88202a0c
N		 4447 mddev->sync_speed_max = max;
4448 return len;
4449}
4450
4451static struct md_sysfs_entry md_sync_max =
4452__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4453
d7f3d291 4454static ssize_t
fd01b88c 4455degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4456{
4457 return sprintf(page, "%d\n", mddev->degraded);
4458}
4459static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4460
90b08710 4461static ssize_t
fd01b88c 4462sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4463{
4464 return sprintf(page, "%d\n", mddev->parallel_resync);
4465}
4466
4467static ssize_t
fd01b88c 4468sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4469{
4470 long n;
4471
b29bebd6 4472 if (kstrtol(buf, 10, &n))
90b08710
BS		 4473 return -EINVAL;
4474
4475 if (n != 0 && n != 1)
4476 return -EINVAL;
4477
4478 mddev->parallel_resync = n;
4479
4480 if (mddev->sync_thread)
4481 wake_up(&resync_wait);
4482
4483 return len;
4484}
4485
4486/* force parallel resync, even with shared block devices */
4487static struct md_sysfs_entry md_sync_force_parallel =
4488__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4489 sync_force_parallel_show, sync_force_parallel_store);
4490
88202a0c 4491static ssize_t
fd01b88c 4492sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4493{
4494 unsigned long resync, dt, db;
d1a7c503
N		 4495 if (mddev->curr_resync == 0)
4496 return sprintf(page, "none\n");
9687a60c
AN		 4497 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4498 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4499 if (!dt) dt++;
9687a60c
AN		 4500 db = resync - mddev->resync_mark_cnt;
4501 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4502}
4503
80ca3a44 4504static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4505
4506static ssize_t
fd01b88c 4507sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4508{
13ae864b 4509 unsigned long long max_sectors, resync;
88202a0c 4510
acb180b0
N		 4511 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4512 return sprintf(page, "none\n");
4513
72f36d59
N		 4514 if (mddev->curr_resync == 1 ||
4515 mddev->curr_resync == 2)
4516 return sprintf(page, "delayed\n");
4517
c804cdec
N		 4518 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4519 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 4520 max_sectors = mddev->resync_max_sectors;
88202a0c 4521 else
58c0fed4 4522 max_sectors = mddev->dev_sectors;
88202a0c 4523
acb180b0 4524 resync = mddev->curr_resync_completed;
13ae864b 4525 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4526}
4527
750f199e
N		 4528static struct md_sysfs_entry md_sync_completed =
4529 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
88202a0c 4530
5e96ee65 4531static ssize_t
fd01b88c 4532min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4533{
4534 return sprintf(page, "%llu\n",
4535 (unsigned long long)mddev->resync_min);
4536}
4537static ssize_t
fd01b88c 4538min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4539{
4540 unsigned long long min;
23da422b 4541 int err;
23da422b 4542
b29bebd6 4543 if (kstrtoull(buf, 10, &min))
5e96ee65 4544 return -EINVAL;
23da422b
N		 4545
4546 spin_lock(&mddev->lock);
4547 err = -EINVAL;
5e96ee65 4548 if (min > mddev->resync_max)
23da422b
N		 4549 goto out_unlock;
4550
4551 err = -EBUSY;
5e96ee65 4552 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4553 goto out_unlock;
5e96ee65 4554
50c37b13
N		 4555 /* Round down to multiple of 4K for safety */
4556 mddev->resync_min = round_down(min, 8);
23da422b 4557 err = 0;
5e96ee65 4558
23da422b
N		 4559out_unlock:
4560 spin_unlock(&mddev->lock);
4561 return err ?: len;
5e96ee65
NB		 4562}
4563
4564static struct md_sysfs_entry md_min_sync =
4565__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4566
c6207277 4567static ssize_t
fd01b88c 4568max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4569{
4570 if (mddev->resync_max == MaxSector)
4571 return sprintf(page, "max\n");
4572 else
4573 return sprintf(page, "%llu\n",
4574 (unsigned long long)mddev->resync_max);
4575}
4576static ssize_t
fd01b88c 4577max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277 4578{
23da422b
N		 4579 int err;
4580 spin_lock(&mddev->lock);
c6207277
N		 4581 if (strncmp(buf, "max", 3) == 0)
4582 mddev->resync_max = MaxSector;
4583 else {
5e96ee65 4584 unsigned long long max;
23da422b
N		 4585 int chunk;
4586
4587 err = -EINVAL;
b29bebd6 4588 if (kstrtoull(buf, 10, &max))
23da422b 4589 goto out_unlock;
5e96ee65 4590 if (max < mddev->resync_min)
23da422b
N		 4591 goto out_unlock;
4592
4593 err = -EBUSY;
c6207277 4594 if (max < mddev->resync_max &&
4d484a4a 4595 mddev->ro == 0 &&
c6207277 4596 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4597 goto out_unlock;
c6207277
N		 4598
4599 /* Must be a multiple of chunk_size */
23da422b
N		 4600 chunk = mddev->chunk_sectors;
4601 if (chunk) {
2ac06c33 4602 sector_t temp = max;
23da422b
N		 4603
4604 err = -EINVAL;
4605 if (sector_div(temp, chunk))
4606 goto out_unlock;
c6207277
N		 4607 }
4608 mddev->resync_max = max;
4609 }
4610 wake_up(&mddev->recovery_wait);
23da422b
N		 4611 err = 0;
4612out_unlock:
4613 spin_unlock(&mddev->lock);
4614 return err ?: len;
c6207277
N		 4615}
4616
4617static struct md_sysfs_entry md_max_sync =
4618__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4619
e464eafd 4620static ssize_t
fd01b88c 4621suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4622{
4623 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4624}
4625
4626static ssize_t
fd01b88c 4627suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd 4628{
4c9309c0 4629 unsigned long long old, new;
6791875e 4630 int err;
e464eafd 4631
4c9309c0
AD		 4632 err = kstrtoull(buf, 10, &new);
4633 if (err < 0)
4634 return err;
4635 if (new != (sector_t)new)
e464eafd 4636 return -EINVAL;
23ddff37 4637
6791875e
N		 4638 err = mddev_lock(mddev);
4639 if (err)
4640 return err;
4641 err = -EINVAL;
4642 if (mddev->pers == NULL ||
4643 mddev->pers->quiesce == NULL)
4644 goto unlock;
4645 old = mddev->suspend_lo;
23ddff37
N		 4646 mddev->suspend_lo = new;
4647 if (new >= old)
4648 /* Shrinking suspended region */
e464eafd 4649 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4650 else {
4651 /* Expanding suspended region - need to wait */
4652 mddev->pers->quiesce(mddev, 1);
4653 mddev->pers->quiesce(mddev, 0);
4654 }
6791875e
N		 4655 err = 0;
4656unlock:
4657 mddev_unlock(mddev);
4658 return err ?: len;
e464eafd
N		 4659}
4660static struct md_sysfs_entry md_suspend_lo =
4661__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4662
e464eafd 4663static ssize_t
fd01b88c 4664suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4665{
4666 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4667}
4668
4669static ssize_t
fd01b88c 4670suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd 4671{
4c9309c0 4672 unsigned long long old, new;
6791875e 4673 int err;
e464eafd 4674
4c9309c0
AD		 4675 err = kstrtoull(buf, 10, &new);
4676 if (err < 0)
4677 return err;
4678 if (new != (sector_t)new)
e464eafd 4679 return -EINVAL;
23ddff37 4680
6791875e
N		 4681 err = mddev_lock(mddev);
4682 if (err)
4683 return err;
4684 err = -EINVAL;
4685 if (mddev->pers == NULL ||
4686 mddev->pers->quiesce == NULL)
4687 goto unlock;
4688 old = mddev->suspend_hi;
23ddff37
N		 4689 mddev->suspend_hi = new;
4690 if (new <= old)
4691 /* Shrinking suspended region */
4692 mddev->pers->quiesce(mddev, 2);
4693 else {
4694 /* Expanding suspended region - need to wait */
e464eafd
N		 4695 mddev->pers->quiesce(mddev, 1);
4696 mddev->pers->quiesce(mddev, 0);
23ddff37 4697 }
6791875e
N		 4698 err = 0;
4699unlock:
4700 mddev_unlock(mddev);
4701 return err ?: len;
e464eafd
N		 4702}
4703static struct md_sysfs_entry md_suspend_hi =
4704__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4705
08a02ecd 4706static ssize_t
fd01b88c 4707reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4708{
4709 if (mddev->reshape_position != MaxSector)
4710 return sprintf(page, "%llu\n",
4711 (unsigned long long)mddev->reshape_position);
4712 strcpy(page, "none\n");
4713 return 5;
4714}
4715
4716static ssize_t
fd01b88c 4717reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd 4718{
c6563a8c 4719 struct md_rdev *rdev;
4c9309c0 4720 unsigned long long new;
6791875e 4721 int err;
6791875e 4722
4c9309c0
AD		 4723 err = kstrtoull(buf, 10, &new);
4724 if (err < 0)
4725 return err;
4726 if (new != (sector_t)new)
08a02ecd 4727 return -EINVAL;
6791875e
N		 4728 err = mddev_lock(mddev);
4729 if (err)
4730 return err;
4731 err = -EBUSY;
4732 if (mddev->pers)
4733 goto unlock;
08a02ecd
N		 4734 mddev->reshape_position = new;
4735 mddev->delta_disks = 0;
2c810cdd 4736 mddev->reshape_backwards = 0;
08a02ecd
N		 4737 mddev->new_level = mddev->level;
4738 mddev->new_layout = mddev->layout;
664e7c41 4739 mddev->new_chunk_sectors = mddev->chunk_sectors;
c6563a8c
N		 4740 rdev_for_each(rdev, mddev)
4741 rdev->new_data_offset = rdev->data_offset;
6791875e
N		 4742 err = 0;
4743unlock:
4744 mddev_unlock(mddev);
4745 return err ?: len;
08a02ecd
N		 4746}
4747
4748static struct md_sysfs_entry md_reshape_position =
4749__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4750 reshape_position_store);
4751
2c810cdd
N		 4752static ssize_t
4753reshape_direction_show(struct mddev *mddev, char *page)
4754{
4755 return sprintf(page, "%s\n",
4756 mddev->reshape_backwards ? "backwards" : "forwards");
4757}
4758
4759static ssize_t
4760reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4761{
4762 int backwards = 0;
6791875e
N		 4763 int err;
4764
2c810cdd
N		 4765 if (cmd_match(buf, "forwards"))
4766 backwards = 0;
4767 else if (cmd_match(buf, "backwards"))
4768 backwards = 1;
4769 else
4770 return -EINVAL;
4771 if (mddev->reshape_backwards == backwards)
4772 return len;
4773
6791875e
N		 4774 err = mddev_lock(mddev);
4775 if (err)
4776 return err;
2c810cdd
N		 4777 /* check if we are allowed to change */
4778 if (mddev->delta_disks)
6791875e
N		 4779 err = -EBUSY;
4780 else if (mddev->persistent &&
2c810cdd 4781 mddev->major_version == 0)
6791875e
N		 4782 err = -EINVAL;
4783 else
4784 mddev->reshape_backwards = backwards;
4785 mddev_unlock(mddev);
4786 return err ?: len;
2c810cdd
N		 4787}
4788
4789static struct md_sysfs_entry md_reshape_direction =
4790__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4791 reshape_direction_store);
4792
b522adcd 4793static ssize_t
fd01b88c 4794array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4795{
4796 if (mddev->external_size)
4797 return sprintf(page, "%llu\n",
4798 (unsigned long long)mddev->array_sectors/2);
4799 else
4800 return sprintf(page, "default\n");
4801}
4802
4803static ssize_t
fd01b88c 4804array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4805{
4806 sector_t sectors;
6791875e
N		 4807 int err;
4808
4809 err = mddev_lock(mddev);
4810 if (err)
4811 return err;
b522adcd
DW		 4812
4813 if (strncmp(buf, "default", 7) == 0) {
4814 if (mddev->pers)
4815 sectors = mddev->pers->size(mddev, 0, 0);
4816 else
4817 sectors = mddev->array_sectors;
4818
4819 mddev->external_size = 0;
4820 } else {
4821 if (strict_blocks_to_sectors(buf, &sectors) < 0)
6791875e
N		 4822 err = -EINVAL;
4823 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4824 err = -E2BIG;
4825 else
4826 mddev->external_size = 1;
b522adcd
DW		 4827 }
4828
6791875e
N		 4829 if (!err) {
4830 mddev->array_sectors = sectors;
4831 if (mddev->pers) {
4832 set_capacity(mddev->gendisk, mddev->array_sectors);
4833 revalidate_disk(mddev->gendisk);
4834 }
cbe6ef1d 4835 }
6791875e
N		 4836 mddev_unlock(mddev);
4837 return err ?: len;
b522adcd
DW		 4838}
4839
4840static struct md_sysfs_entry md_array_size =
4841__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4842 array_size_store);
e464eafd 4843
eae1701f
N		 4844static struct attribute *md_default_attrs[] = {
4845 &md_level.attr,
d4dbd025 4846 &md_layout.attr,
eae1701f 4847 &md_raid_disks.attr,
3b34380a 4848 &md_chunk_size.attr,
a35b0d69 4849 &md_size.attr,
a94213b1 4850 &md_resync_start.attr,
8bb93aac 4851 &md_metadata.attr,
6d7ff738 4852 &md_new_device.attr,
16f17b39 4853 &md_safe_delay.attr,
9e653b63 4854 &md_array_state.attr,
08a02ecd 4855 &md_reshape_position.attr,
2c810cdd 4856 &md_reshape_direction.attr,
b522adcd 4857 &md_array_size.attr,
1e50915f 4858 &max_corr_read_errors.attr,
411036fa
N		 4859 NULL,
4860};
4861
4862static struct attribute *md_redundancy_attrs[] = {
24dd469d 4863 &md_scan_mode.attr,
c4a39551 4864 &md_last_scan_mode.attr,
9d88883e 4865 &md_mismatches.attr,
88202a0c
N		 4866 &md_sync_min.attr,
4867 &md_sync_max.attr,
4868 &md_sync_speed.attr,
90b08710 4869 &md_sync_force_parallel.attr,
88202a0c 4870 &md_sync_completed.attr,
5e96ee65 4871 &md_min_sync.attr,
c6207277 4872 &md_max_sync.attr,
e464eafd
N		 4873 &md_suspend_lo.attr,
4874 &md_suspend_hi.attr,
9b1d1dac 4875 &md_bitmap.attr,
d7f3d291 4876 &md_degraded.attr,
eae1701f
N		 4877 NULL,
4878};
411036fa
N		 4879static struct attribute_group md_redundancy_group = {
4880 .name = NULL,
4881 .attrs = md_redundancy_attrs,
4882};
4883
eae1701f
N		 4884static ssize_t
4885md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4886{
4887 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4888 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4889 ssize_t rv;
eae1701f
N		 4890
4891 if (!entry->show)
4892 return -EIO;
af8a2434
N		 4893 spin_lock(&all_mddevs_lock);
4894 if (list_empty(&mddev->all_mddevs)) {
4895 spin_unlock(&all_mddevs_lock);
4896 return -EBUSY;
4897 }
4898 mddev_get(mddev);
4899 spin_unlock(&all_mddevs_lock);
4900
b7b17c9b 4901 rv = entry->show(mddev, page);
af8a2434 4902 mddev_put(mddev);
96de1e66 4903 return rv;
eae1701f
N		 4904}
4905
4906static ssize_t
4907md_attr_store(struct kobject *kobj, struct attribute *attr,
4908 const char *page, size_t length)
4909{
4910 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4911 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4912 ssize_t rv;
eae1701f
N		 4913
4914 if (!entry->store)
4915 return -EIO;
67463acb
N		 4916 if (!capable(CAP_SYS_ADMIN))
4917 return -EACCES;
af8a2434
N		 4918 spin_lock(&all_mddevs_lock);
4919 if (list_empty(&mddev->all_mddevs)) {
4920 spin_unlock(&all_mddevs_lock);
4921 return -EBUSY;
4922 }
4923 mddev_get(mddev);
4924 spin_unlock(&all_mddevs_lock);
6791875e 4925 rv = entry->store(mddev, page, length);
af8a2434 4926 mddev_put(mddev);
96de1e66 4927 return rv;
eae1701f
N		 4928}
4929
4930static void md_free(struct kobject *ko)
4931{
fd01b88c 4932 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4933
4934 if (mddev->sysfs_state)
4935 sysfs_put(mddev->sysfs_state);
4936
6cd18e71
N		 4937 if (mddev->queue)
4938 blk_cleanup_queue(mddev->queue);
a21d1504
N		 4939 if (mddev->gendisk) {
4940 del_gendisk(mddev->gendisk);
4941 put_disk(mddev->gendisk);
4942 }
a21d1504 4943
eae1701f
N		 4944 kfree(mddev);
4945}
4946
52cf25d0 4947static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4948 .show = md_attr_show,
4949 .store = md_attr_store,
4950};
4951static struct kobj_type md_ktype = {
4952 .release = md_free,
4953 .sysfs_ops = &md_sysfs_ops,
4954 .default_attrs = md_default_attrs,
4955};
4956
1da177e4
LT		 4957int mdp_major = 0;
4958
5fd3a17e
DW		 4959static void mddev_delayed_delete(struct work_struct *ws)
4960{
fd01b88c 4961 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4962
43a70507 4963 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4964 kobject_del(&mddev->kobj);
4965 kobject_put(&mddev->kobj);
4966}
4967
efeb53c0 4968static int md_alloc(dev_t dev, char *name)
1da177e4 4969{
48c9c27b 4970 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4971 struct mddev *mddev = mddev_find(dev);
1da177e4 4972 struct gendisk *disk;
efeb53c0
N		 4973 int partitioned;
4974 int shift;
4975 int unit;
3830c62f 4976 int error;
1da177e4
LT		 4977
4978 if (!mddev)
efeb53c0
N		 4979 return -ENODEV;
4980
4981 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4982 shift = partitioned ? MdpMinorShift : 0;
4983 unit = MINOR(mddev->unit) >> shift;
1da177e4 4984
e804ac78
TH		 4985 /* wait for any previous instance of this device to be
4986 * completely removed (mddev_delayed_delete).
d3374825 4987 */
e804ac78 4988 flush_workqueue(md_misc_wq);
d3374825 4989
48c9c27b 4990 mutex_lock(&disks_mutex);
0909dc44
N		 4991 error = -EEXIST;
4992 if (mddev->gendisk)
4993 goto abort;
efeb53c0
N		 4994
4995 if (name) {
4996 /* Need to ensure that 'name' is not a duplicate.
4997 */
fd01b88c 4998 struct mddev *mddev2;
efeb53c0
N		 4999 spin_lock(&all_mddevs_lock);
5000
5001 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5002 if (mddev2->gendisk &&
5003 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5004 spin_unlock(&all_mddevs_lock);
0909dc44 5005 goto abort;
efeb53c0
N		 5006 }
5007 spin_unlock(&all_mddevs_lock);
1da177e4 5008 }
8b765398 5009
0909dc44 5010 error = -ENOMEM;
8b765398 5011 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 5012 if (!mddev->queue)
5013 goto abort;
409c57f3
N		 5014 mddev->queue->queuedata = mddev;
5015
409c57f3 5016 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 5017 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 5018
1da177e4
LT		 5019 disk = alloc_disk(1 << shift);
5020 if (!disk) {
8b765398
N		 5021 blk_cleanup_queue(mddev->queue);
5022 mddev->queue = NULL;
0909dc44 5023 goto abort;
1da177e4 5024 }
efeb53c0 5025 disk->major = MAJOR(mddev->unit);
1da177e4 5026 disk->first_minor = unit << shift;
efeb53c0
N		 5027 if (name)
5028 strcpy(disk->disk_name, name);
5029 else if (partitioned)
1da177e4 5030 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 5031 else
1da177e4 5032 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 5033 disk->fops = &md_fops;
5034 disk->private_data = mddev;
5035 disk->queue = mddev->queue;
b0140891 5036 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 5037 /* Allow extended partitions. This makes the
d3374825 5038 * 'mdp' device redundant, but we can't really
92850bbd
N		 5039 * remove it now.
5040 */
5041 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 5042 mddev->gendisk = disk;
b0140891
N		 5043 /* As soon as we call add_disk(), another thread could get
5044 * through to md_open, so make sure it doesn't get too far
5045 */
5046 mutex_lock(&mddev->open_mutex);
5047 add_disk(disk);
5048
ed9e1982
TH		 5049 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5050 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 5051 if (error) {
5052 /* This isn't possible, but as kobject_init_and_add is marked
5053 * __must_check, we must do something with the result
5054 */
5e55e2f5
N		 5055 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5056 disk->disk_name);
0909dc44
N		 5057 error = 0;
5058 }
00bcb4ac
N		 5059 if (mddev->kobj.sd &&
5060 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 5061 printk(KERN_DEBUG "pointless warning\n");
b0140891 5062 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 5063 abort:
5064 mutex_unlock(&disks_mutex);
00bcb4ac 5065 if (!error && mddev->kobj.sd) {
3830c62f 5066 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 5067 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 5068 }
d3374825 5069 mddev_put(mddev);
0909dc44 5070 return error;
efeb53c0
N		 5071}
5072
5073static struct kobject *md_probe(dev_t dev, int *part, void *data)
5074{
5075 md_alloc(dev, NULL);
1da177e4
LT		 5076 return NULL;
5077}
5078
efeb53c0
N		 5079static int add_named_array(const char *val, struct kernel_param *kp)
5080{
5081 /* val must be "md_*" where * is not all digits.
5082 * We allocate an array with a large free minor number, and
5083 * set the name to val. val must not already be an active name.
5084 */
5085 int len = strlen(val);
5086 char buf[DISK_NAME_LEN];
5087
5088 while (len && val[len-1] == '\n')
5089 len--;
5090 if (len >= DISK_NAME_LEN)
5091 return -E2BIG;
5092 strlcpy(buf, val, len+1);
5093 if (strncmp(buf, "md_", 3) != 0)
5094 return -EINVAL;
5095 return md_alloc(0, buf);
5096}
5097
1da177e4
LT		 5098static void md_safemode_timeout(unsigned long data)
5099{
fd01b88c 5100 struct mddev *mddev = (struct mddev *) data;
1da177e4 5101
0fd62b86
NB		 5102 if (!atomic_read(&mddev->writes_pending)) {
5103 mddev->safemode = 1;
5104 if (mddev->external)
00bcb4ac 5105 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 5106 }
1da177e4
LT		 5107 md_wakeup_thread(mddev->thread);
5108}
5109
6ff8d8ec 5110static int start_dirty_degraded;
1da177e4 5111
fd01b88c 5112int md_run(struct mddev *mddev)
1da177e4 5113{
2604b703 5114 int err;
3cb03002 5115 struct md_rdev *rdev;
84fc4b56 5116 struct md_personality *pers;
1da177e4 5117
a757e64c
N		 5118 if (list_empty(&mddev->disks))
5119 /* cannot run an array with no devices.. */
1da177e4 5120 return -EINVAL;
1da177e4
LT		 5121
5122 if (mddev->pers)
5123 return -EBUSY;
bb4f1e9d
N		 5124 /* Cannot run until previous stop completes properly */
5125 if (mddev->sysfs_active)
5126 return -EBUSY;
b6eb127d 5127
1da177e4
LT		 5128 /*
5129 * Analyze all RAID superblock(s)
5130 */
1ec4a939
N		 5131 if (!mddev->raid_disks) {
5132 if (!mddev->persistent)
5133 return -EINVAL;
a757e64c 5134 analyze_sbs(mddev);
1ec4a939 5135 }
1da177e4 5136
d9d166c2
N		 5137 if (mddev->level != LEVEL_NONE)
5138 request_module("md-level-%d", mddev->level);
5139 else if (mddev->clevel[0])
5140 request_module("md-%s", mddev->clevel);
1da177e4
LT		 5141
5142 /*
5143 * Drop all container device buffers, from now on
5144 * the only valid external interface is through the md
5145 * device.
1da177e4 5146 */
dafb20fa 5147 rdev_for_each(rdev, mddev) {
b2d444d7 5148 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5149 continue;
5150 sync_blockdev(rdev->bdev);
f98393a6 5151 invalidate_bdev(rdev->bdev);
f0d76d70
N		 5152
5153 /* perform some consistency tests on the device.
5154 * We don't want the data to overlap the metadata,
58c0fed4 5155 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 5156 */
a6ff7e08
JB		 5157 if (rdev->meta_bdev) {
5158 /* Nothing to check */;
5159 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 5160 if (mddev->dev_sectors &&
5161 rdev->data_offset + mddev->dev_sectors
0f420358 5162 > rdev->sb_start) {
f0d76d70
N		 5163 printk("md: %s: data overlaps metadata\n",
5164 mdname(mddev));
5165 return -EINVAL;
5166 }
5167 } else {
0f420358 5168 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 5169 > rdev->data_offset) {
5170 printk("md: %s: metadata overlaps data\n",
5171 mdname(mddev));
5172 return -EINVAL;
5173 }
5174 }
00bcb4ac 5175 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 5176 }
5177
a167f663 5178 if (mddev->bio_set == NULL)
395c72a7 5179 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
a167f663 5180
1da177e4 5181 spin_lock(&pers_lock);
d9d166c2 5182 pers = find_pers(mddev->level, mddev->clevel);
2604b703 5183 if (!pers || !try_module_get(pers->owner)) {
1da177e4 5184 spin_unlock(&pers_lock);
d9d166c2
N		 5185 if (mddev->level != LEVEL_NONE)
5186 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5187 mddev->level);
5188 else
5189 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5190 mddev->clevel);
1da177e4
LT		 5191 return -EINVAL;
5192 }
1da177e4 5193 spin_unlock(&pers_lock);
34817e8c
N		 5194 if (mddev->level != pers->level) {
5195 mddev->level = pers->level;
5196 mddev->new_level = pers->level;
5197 }
d9d166c2 5198 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 5199
f6705578 5200 if (mddev->reshape_position != MaxSector &&
63c70c4f 5201 pers->start_reshape == NULL) {
f6705578 5202 /* This personality cannot handle reshaping... */
f6705578
N		 5203 module_put(pers->owner);
5204 return -EINVAL;
5205 }
5206
7dd5e7c3
N		 5207 if (pers->sync_request) {
5208 /* Warn if this is a potentially silly
5209 * configuration.
5210 */
5211 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5212 struct md_rdev *rdev2;
7dd5e7c3 5213 int warned = 0;
159ec1fc 5214
dafb20fa
N		 5215 rdev_for_each(rdev, mddev)
5216 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N		 5217 if (rdev < rdev2 &&
5218 rdev->bdev->bd_contains ==
5219 rdev2->bdev->bd_contains) {
5220 printk(KERN_WARNING
5221 "%s: WARNING: %s appears to be"
5222 " on the same physical disk as"
5223 " %s.\n",
5224 mdname(mddev),
5225 bdevname(rdev->bdev,b),
5226 bdevname(rdev2->bdev,b2));
5227 warned = 1;
5228 }
5229 }
159ec1fc 5230
7dd5e7c3
N		 5231 if (warned)
5232 printk(KERN_WARNING
5233 "True protection against single-disk"
5234 " failure might be compromised.\n");
5235 }
5236
657390d2 5237 mddev->recovery = 0;
58c0fed4
AN		 5238 /* may be over-ridden by personality */
5239 mddev->resync_max_sectors = mddev->dev_sectors;
5240
6ff8d8ec 5241 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5242
0f9552b5 5243 if (start_readonly && mddev->ro == 0)
f91de92e
N		 5244 mddev->ro = 2; /* read-only, but switch on first write */
5245
36d091f4 5246 err = pers->run(mddev);
13e53df3
AN		 5247 if (err)
5248 printk(KERN_ERR "md: pers->run() failed ...\n");
36d091f4 5249 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
b522adcd
DW		 5250 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5251 " but 'external_size' not in effect?\n", __func__);
5252 printk(KERN_ERR
5253 "md: invalid array_size %llu > default size %llu\n",
5254 (unsigned long long)mddev->array_sectors / 2,
36d091f4 5255 (unsigned long long)pers->size(mddev, 0, 0) / 2);
b522adcd 5256 err = -EINVAL;
b522adcd 5257 }
36d091f4 5258 if (err == 0 && pers->sync_request &&
ef99bf48 5259 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
f9209a32
GR		 5260 struct bitmap *bitmap;
5261
5262 bitmap = bitmap_create(mddev, -1);
5263 if (IS_ERR(bitmap)) {
5264 err = PTR_ERR(bitmap);
b15c2e57
N		 5265 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5266 mdname(mddev), err);
f9209a32
GR		 5267 } else
5268 mddev->bitmap = bitmap;
5269
b15c2e57 5270 }
1da177e4 5271 if (err) {
5aa61f42 5272 mddev_detach(mddev);
0c35bd47
N		 5273 if (mddev->private)
5274 pers->free(mddev, mddev->private);
bd691922 5275 mddev->private = NULL;
36d091f4 5276 module_put(pers->owner);
32a7627c
N		 5277 bitmap_destroy(mddev);
5278 return err;
1da177e4 5279 }
5c675f83
N		 5280 if (mddev->queue) {
5281 mddev->queue->backing_dev_info.congested_data = mddev;
5282 mddev->queue->backing_dev_info.congested_fn = md_congested;
5283 }
36d091f4 5284 if (pers->sync_request) {
00bcb4ac
N		 5285 if (mddev->kobj.sd &&
5286 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 5287 printk(KERN_WARNING
5288 "md: cannot register extra attributes for %s\n",
5289 mdname(mddev));
00bcb4ac 5290 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5291 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 5292 mddev->ro = 0;
5293
f72ffdd6 5294 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 5295 atomic_set(&mddev->max_corr_read_errors,
5296 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4 5297 mddev->safemode = 0;
28c1b9fd
GR		 5298 if (mddev_is_clustered(mddev))
5299 mddev->safemode_delay = 0;
5300 else
5301 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5302 mddev->in_sync = 1;
0ca69886 5303 smp_wmb();
36d091f4
N		 5304 spin_lock(&mddev->lock);
5305 mddev->pers = pers;
0ca69886 5306 mddev->ready = 1;
36d091f4 5307 spin_unlock(&mddev->lock);
dafb20fa 5308 rdev_for_each(rdev, mddev)
36fad858
NK		 5309 if (rdev->raid_disk >= 0)
5310 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 5311 /* failure here is OK */;
f72ffdd6 5312
a4a3d26d
N		 5313 if (mddev->degraded && !mddev->ro)
5314 /* This ensures that recovering status is reported immediately
5315 * via sysfs - until a lack of spares is confirmed.
5316 */
5317 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
1da177e4 5318 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f72ffdd6 5319
7a0a5355 5320 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 5321 md_update_sb(mddev, 0);
1da177e4 5322
d7603b7e 5323 md_new_event(mddev);
00bcb4ac
N		 5324 sysfs_notify_dirent_safe(mddev->sysfs_state);
5325 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 5326 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 5327 return 0;
5328}
390ee602 5329EXPORT_SYMBOL_GPL(md_run);
1da177e4 5330
fd01b88c 5331static int do_md_run(struct mddev *mddev)
fe60b014
N		 5332{
5333 int err;
5334
5335 err = md_run(mddev);
5336 if (err)
5337 goto out;
69e51b44
N		 5338 err = bitmap_load(mddev);
5339 if (err) {
5340 bitmap_destroy(mddev);
5341 goto out;
5342 }
0fd018af 5343
28c1b9fd
GR		 5344 if (mddev_is_clustered(mddev))
5345 md_allow_write(mddev);
5346
0fd018af
JB		 5347 md_wakeup_thread(mddev->thread);
5348 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5349
fe60b014
N		 5350 set_capacity(mddev->gendisk, mddev->array_sectors);
5351 revalidate_disk(mddev->gendisk);
f0b4f7e2 5352 mddev->changed = 1;
fe60b014
N		 5353 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5354out:
5355 return err;
5356}
5357
fd01b88c 5358static int restart_array(struct mddev *mddev)
1da177e4
LT		 5359{
5360 struct gendisk *disk = mddev->gendisk;
1da177e4 5361
80fab1d7 5362 /* Complain if it has no devices */
1da177e4 5363 if (list_empty(&mddev->disks))
80fab1d7
AN		 5364 return -ENXIO;
5365 if (!mddev->pers)
5366 return -EINVAL;
5367 if (!mddev->ro)
5368 return -EBUSY;
339421de
SL		 5369 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5370 struct md_rdev *rdev;
5371 bool has_journal = false;
5372
5373 rcu_read_lock();
5374 rdev_for_each_rcu(rdev, mddev) {
5375 if (test_bit(Journal, &rdev->flags) &&
5376 !test_bit(Faulty, &rdev->flags)) {
5377 has_journal = true;
5378 break;
5379 }
5380 }
5381 rcu_read_unlock();
5382
5383 /* Don't restart rw with journal missing/faulty */
5384 if (!has_journal)
5385 return -EINVAL;
5386 }
5387
80fab1d7
AN		 5388 mddev->safemode = 0;
5389 mddev->ro = 0;
5390 set_disk_ro(disk, 0);
5391 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5392 mdname(mddev));
5393 /* Kick recovery or resync if necessary */
5394 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5395 md_wakeup_thread(mddev->thread);
5396 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5397 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5398 return 0;
1da177e4
LT		 5399}
5400
fd01b88c 5401static void md_clean(struct mddev *mddev)
6177b472
N		 5402{
5403 mddev->array_sectors = 0;
5404 mddev->external_size = 0;
5405 mddev->dev_sectors = 0;
5406 mddev->raid_disks = 0;
5407 mddev->recovery_cp = 0;
5408 mddev->resync_min = 0;
5409 mddev->resync_max = MaxSector;
5410 mddev->reshape_position = MaxSector;
5411 mddev->external = 0;
5412 mddev->persistent = 0;
5413 mddev->level = LEVEL_NONE;
5414 mddev->clevel[0] = 0;
5415 mddev->flags = 0;
5416 mddev->ro = 0;
5417 mddev->metadata_type[0] = 0;
5418 mddev->chunk_sectors = 0;
5419 mddev->ctime = mddev->utime = 0;
5420 mddev->layout = 0;
5421 mddev->max_disks = 0;
5422 mddev->events = 0;
a8707c08 5423 mddev->can_decrease_events = 0;
6177b472 5424 mddev->delta_disks = 0;
2c810cdd 5425 mddev->reshape_backwards = 0;
6177b472
N		 5426 mddev->new_level = LEVEL_NONE;
5427 mddev->new_layout = 0;
5428 mddev->new_chunk_sectors = 0;
5429 mddev->curr_resync = 0;
7f7583d4 5430 atomic64_set(&mddev->resync_mismatches, 0);
6177b472
N		 5431 mddev->suspend_lo = mddev->suspend_hi = 0;
5432 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5433 mddev->recovery = 0;
5434 mddev->in_sync = 0;
f0b4f7e2 5435 mddev->changed = 0;
6177b472 5436 mddev->degraded = 0;
6177b472 5437 mddev->safemode = 0;
bd691922 5438 mddev->private = NULL;
6177b472
N		 5439 mddev->bitmap_info.offset = 0;
5440 mddev->bitmap_info.default_offset = 0;
6409bb05 5441 mddev->bitmap_info.default_space = 0;
6177b472
N		 5442 mddev->bitmap_info.chunksize = 0;
5443 mddev->bitmap_info.daemon_sleep = 0;
5444 mddev->bitmap_info.max_write_behind = 0;
5445}
5446
fd01b88c 5447static void __md_stop_writes(struct mddev *mddev)
a047e125 5448{
6b6204ee 5449 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
f851b60d 5450 flush_workqueue(md_misc_wq);
a047e125 5451 if (mddev->sync_thread) {
a047e125 5452 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 5453 md_reap_sync_thread(mddev);
a047e125
N		 5454 }
5455
5456 del_timer_sync(&mddev->safemode_timer);
5457
5458 bitmap_flush(mddev);
5459 md_super_wait(mddev);
5460
b6d428c6 5461 if (mddev->ro == 0 &&
28c1b9fd
GR		 5462 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5463 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
a047e125 5464 /* mark array as shutdown cleanly */
28c1b9fd
GR		 5465 if (!mddev_is_clustered(mddev))
5466 mddev->in_sync = 1;
a047e125
N		 5467 md_update_sb(mddev, 1);
5468 }
5469}
defad61a 5470
fd01b88c 5471void md_stop_writes(struct mddev *mddev)
defad61a 5472{
29f097c4 5473 mddev_lock_nointr(mddev);
defad61a
N		 5474 __md_stop_writes(mddev);
5475 mddev_unlock(mddev);
5476}
390ee602 5477EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5478
5aa61f42
N		 5479static void mddev_detach(struct mddev *mddev)
5480{
5481 struct bitmap *bitmap = mddev->bitmap;
5482 /* wait for behind writes to complete */
5483 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5484 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5485 mdname(mddev));
5486 /* need to kick something here to make sure I/O goes? */
5487 wait_event(bitmap->behind_wait,
5488 atomic_read(&bitmap->behind_writes) == 0);
5489 }
36d091f4 5490 if (mddev->pers && mddev->pers->quiesce) {
5aa61f42
N		 5491 mddev->pers->quiesce(mddev, 1);
5492 mddev->pers->quiesce(mddev, 0);
5493 }
5494 md_unregister_thread(&mddev->thread);
5495 if (mddev->queue)
5496 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5497}
5498
5eff3c43 5499static void __md_stop(struct mddev *mddev)
6177b472 5500{
36d091f4 5501 struct md_personality *pers = mddev->pers;
5aa61f42 5502 mddev_detach(mddev);
ee5d004f
N		 5503 /* Ensure ->event_work is done */
5504 flush_workqueue(md_misc_wq);
36d091f4
N		 5505 spin_lock(&mddev->lock);
5506 mddev->ready = 0;
6177b472 5507 mddev->pers = NULL;
36d091f4
N		 5508 spin_unlock(&mddev->lock);
5509 pers->free(mddev, mddev->private);
bd691922 5510 mddev->private = NULL;
36d091f4
N		 5511 if (pers->sync_request && mddev->to_remove == NULL)
5512 mddev->to_remove = &md_redundancy_group;
5513 module_put(pers->owner);
cca9cf90 5514 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5515}
5eff3c43
N		 5516
5517void md_stop(struct mddev *mddev)
5518{
5519 /* stop the array and free an attached data structures.
5520 * This is called from dm-raid
5521 */
5522 __md_stop(mddev);
5523 bitmap_destroy(mddev);
5524 if (mddev->bio_set)
5525 bioset_free(mddev->bio_set);
5526}
5527
390ee602 5528EXPORT_SYMBOL_GPL(md_stop);
6177b472 5529
a05b7ea0 5530static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N		 5531{
5532 int err = 0;
30b8feb7
N		 5533 int did_freeze = 0;
5534
5535 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5536 did_freeze = 1;
5537 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5538 md_wakeup_thread(mddev->thread);
5539 }
f851b60d 5540 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5541 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5542 if (mddev->sync_thread)
30b8feb7
N		 5543 /* Thread might be blocked waiting for metadata update
5544 * which will now never happen */
5545 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5546
88724bfa
N		 5547 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5548 return -EBUSY;
30b8feb7 5549 mddev_unlock(mddev);
f851b60d
N		 5550 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5551 &mddev->recovery));
88724bfa
N		 5552 wait_event(mddev->sb_wait,
5553 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
30b8feb7
N		 5554 mddev_lock_nointr(mddev);
5555
a4bd82d0 5556 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5557 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7 5558 mddev->sync_thread ||
f851b60d 5559 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5560 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
a4bd82d0 5561 printk("md: %s still in use.\n",mdname(mddev));
30b8feb7
N		 5562 if (did_freeze) {
5563 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5564 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5565 md_wakeup_thread(mddev->thread);
5566 }
a4bd82d0
N		 5567 err = -EBUSY;
5568 goto out;
5569 }
5570 if (mddev->pers) {
defad61a 5571 __md_stop_writes(mddev);
a4bd82d0
N		 5572
5573 err = -ENXIO;
5574 if (mddev->ro==1)
5575 goto out;
5576 mddev->ro = 1;
5577 set_disk_ro(mddev->gendisk, 1);
5578 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d
N		 5579 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5580 md_wakeup_thread(mddev->thread);
00bcb4ac 5581 sysfs_notify_dirent_safe(mddev->sysfs_state);
30b8feb7 5582 err = 0;
a4bd82d0
N		 5583 }
5584out:
5585 mutex_unlock(&mddev->open_mutex);
5586 return err;
5587}
5588
9e653b63
N		 5589/* mode:
5590 * 0 - completely stop and dis-assemble array
9e653b63
N		 5591 * 2 - stop but do not disassemble array
5592 */
f72ffdd6 5593static int do_md_stop(struct mddev *mddev, int mode,
a05b7ea0 5594 struct block_device *bdev)
1da177e4 5595{
1da177e4 5596 struct gendisk *disk = mddev->gendisk;
3cb03002 5597 struct md_rdev *rdev;
30b8feb7
N		 5598 int did_freeze = 0;
5599
5600 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5601 did_freeze = 1;
5602 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5603 md_wakeup_thread(mddev->thread);
5604 }
f851b60d 5605 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5606 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5607 if (mddev->sync_thread)
30b8feb7
N		 5608 /* Thread might be blocked waiting for metadata update
5609 * which will now never happen */
5610 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5611
30b8feb7 5612 mddev_unlock(mddev);
f851b60d
N		 5613 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5614 !test_bit(MD_RECOVERY_RUNNING,
5615 &mddev->recovery)));
30b8feb7 5616 mddev_lock_nointr(mddev);
1da177e4 5617
c8c00a69 5618 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5619 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7
N		 5620 mddev->sysfs_active ||
5621 mddev->sync_thread ||
f851b60d 5622 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5623 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
df5b20cf 5624 printk("md: %s still in use.\n",mdname(mddev));
6e17b027 5625 mutex_unlock(&mddev->open_mutex);
30b8feb7
N		 5626 if (did_freeze) {
5627 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5628 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5629 md_wakeup_thread(mddev->thread);
5630 }
260fa034
N		 5631 return -EBUSY;
5632 }
6e17b027 5633 if (mddev->pers) {
a4bd82d0
N		 5634 if (mddev->ro)
5635 set_disk_ro(disk, 0);
409c57f3 5636
defad61a 5637 __md_stop_writes(mddev);
5eff3c43 5638 __md_stop(mddev);
a4bd82d0 5639 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5640
a4bd82d0 5641 /* tell userspace to handle 'inactive' */
00bcb4ac 5642 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5643
dafb20fa 5644 rdev_for_each(rdev, mddev)
36fad858
NK		 5645 if (rdev->raid_disk >= 0)
5646 sysfs_unlink_rdev(mddev, rdev);
c4647292 5647
a4bd82d0 5648 set_capacity(disk, 0);
6e17b027 5649 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5650 mddev->changed = 1;
a4bd82d0 5651 revalidate_disk(disk);
0d4ca600 5652
a4bd82d0
N		 5653 if (mddev->ro)
5654 mddev->ro = 0;
6e17b027
N		 5655 } else
5656 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5657 /*
5658 * Free resources if final stop
5659 */
9e653b63 5660 if (mode == 0) {
1da177e4
LT		 5661 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5662
978f946b 5663 bitmap_destroy(mddev);
c3d9714e 5664 if (mddev->bitmap_info.file) {
4af1a041
N		 5665 struct file *f = mddev->bitmap_info.file;
5666 spin_lock(&mddev->lock);
c3d9714e 5667 mddev->bitmap_info.file = NULL;
4af1a041
N		 5668 spin_unlock(&mddev->lock);
5669 fput(f);
978f946b 5670 }
c3d9714e 5671 mddev->bitmap_info.offset = 0;
978f946b 5672
1da177e4
LT		 5673 export_array(mddev);
5674
6177b472 5675 md_clean(mddev);
934d9c23 5676 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5677 if (mddev->hold_active == UNTIL_STOP)
5678 mddev->hold_active = 0;
a4bd82d0 5679 }
d7603b7e 5680 md_new_event(mddev);
00bcb4ac 5681 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5682 return 0;
1da177e4
LT		 5683}
5684
fdee8ae4 5685#ifndef MODULE
fd01b88c 5686static void autorun_array(struct mddev *mddev)
1da177e4 5687{
3cb03002 5688 struct md_rdev *rdev;
1da177e4
LT		 5689 int err;
5690
a757e64c 5691 if (list_empty(&mddev->disks))
1da177e4 5692 return;
1da177e4
LT		 5693
5694 printk(KERN_INFO "md: running: ");
5695
dafb20fa 5696 rdev_for_each(rdev, mddev) {
1da177e4
LT		 5697 char b[BDEVNAME_SIZE];
5698 printk("<%s>", bdevname(rdev->bdev,b));
5699 }
5700 printk("\n");
5701
d710e138 5702 err = do_md_run(mddev);
1da177e4
LT		 5703 if (err) {
5704 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
a05b7ea0 5705 do_md_stop(mddev, 0, NULL);
1da177e4
LT		 5706 }
5707}
5708
5709/*
5710 * lets try to run arrays based on all disks that have arrived
5711 * until now. (those are in pending_raid_disks)
5712 *
5713 * the method: pick the first pending disk, collect all disks with
5714 * the same UUID, remove all from the pending list and put them into
5715 * the 'same_array' list. Then order this list based on superblock
5716 * update time (freshest comes first), kick out 'old' disks and
5717 * compare superblocks. If everything's fine then run it.
5718 *
5719 * If "unit" is allocated, then bump its reference count
5720 */
5721static void autorun_devices(int part)
5722{
3cb03002 5723 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5724 struct mddev *mddev;
1da177e4
LT		 5725 char b[BDEVNAME_SIZE];
5726
5727 printk(KERN_INFO "md: autorun ...\n");
5728 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5729 int unit;
1da177e4 5730 dev_t dev;
ad01c9e3 5731 LIST_HEAD(candidates);
1da177e4 5732 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5733 struct md_rdev, same_set);
1da177e4
LT		 5734
5735 printk(KERN_INFO "md: considering %s ...\n",
5736 bdevname(rdev0->bdev,b));
5737 INIT_LIST_HEAD(&candidates);
159ec1fc 5738 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5739 if (super_90_load(rdev, rdev0, 0) >= 0) {
5740 printk(KERN_INFO "md: adding %s ...\n",
5741 bdevname(rdev->bdev,b));
5742 list_move(&rdev->same_set, &candidates);
5743 }
5744 /*
5745 * now we have a set of devices, with all of them having
5746 * mostly sane superblocks. It's time to allocate the
5747 * mddev.
5748 */
e8703fe1
N		 5749 if (part) {
5750 dev = MKDEV(mdp_major,
5751 rdev0->preferred_minor << MdpMinorShift);
5752 unit = MINOR(dev) >> MdpMinorShift;
5753 } else {
5754 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5755 unit = MINOR(dev);
5756 }
5757 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5758 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5759 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5760 break;
5761 }
1da177e4
LT		 5762
5763 md_probe(dev, NULL, NULL);
5764 mddev = mddev_find(dev);
9bbbca3a
NB		 5765 if (!mddev || !mddev->gendisk) {
5766 if (mddev)
5767 mddev_put(mddev);
5768 printk(KERN_ERR
1da177e4
LT		 5769 "md: cannot allocate memory for md drive.\n");
5770 break;
5771 }
f72ffdd6 5772 if (mddev_lock(mddev))
1da177e4
LT		 5773 printk(KERN_WARNING "md: %s locked, cannot run\n",
5774 mdname(mddev));
5775 else if (mddev->raid_disks || mddev->major_version
5776 || !list_empty(&mddev->disks)) {
f72ffdd6 5777 printk(KERN_WARNING
1da177e4
LT		 5778 "md: %s already running, cannot run %s\n",
5779 mdname(mddev), bdevname(rdev0->bdev,b));
5780 mddev_unlock(mddev);
5781 } else {
5782 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5783 mddev->persistent = 1;
159ec1fc 5784 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5785 list_del_init(&rdev->same_set);
5786 if (bind_rdev_to_array(rdev, mddev))
5787 export_rdev(rdev);
5788 }
5789 autorun_array(mddev);
5790 mddev_unlock(mddev);
5791 }
5792 /* on success, candidates will be empty, on error
5793 * it won't...
5794 */
159ec1fc 5795 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5796 list_del_init(&rdev->same_set);
1da177e4 5797 export_rdev(rdev);
4b80991c 5798 }
1da177e4
LT		 5799 mddev_put(mddev);
5800 }
5801 printk(KERN_INFO "md: ... autorun DONE.\n");
5802}
fdee8ae4 5803#endif /* !MODULE */
1da177e4 5804
f72ffdd6 5805static int get_version(void __user *arg)
1da177e4
LT		 5806{
5807 mdu_version_t ver;
5808
5809 ver.major = MD_MAJOR_VERSION;
5810 ver.minor = MD_MINOR_VERSION;
5811 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5812
5813 if (copy_to_user(arg, &ver, sizeof(ver)))
5814 return -EFAULT;
5815
5816 return 0;
5817}
5818
f72ffdd6 5819static int get_array_info(struct mddev *mddev, void __user *arg)
1da177e4
LT		 5820{
5821 mdu_array_info_t info;
a9f326eb 5822 int nr,working,insync,failed,spare;
3cb03002 5823 struct md_rdev *rdev;
1da177e4 5824
1ca69c4b
N		 5825 nr = working = insync = failed = spare = 0;
5826 rcu_read_lock();
5827 rdev_for_each_rcu(rdev, mddev) {
1da177e4 5828 nr++;
b2d444d7 5829 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5830 failed++;
5831 else {
5832 working++;
b2d444d7 5833 if (test_bit(In_sync, &rdev->flags))
f72ffdd6 5834 insync++;
1da177e4
LT		 5835 else
5836 spare++;
5837 }
5838 }
1ca69c4b 5839 rcu_read_unlock();
1da177e4
LT		 5840
5841 info.major_version = mddev->major_version;
5842 info.minor_version = mddev->minor_version;
5843 info.patch_version = MD_PATCHLEVEL_VERSION;
5844 info.ctime = mddev->ctime;
5845 info.level = mddev->level;
58c0fed4
AN		 5846 info.size = mddev->dev_sectors / 2;
5847 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5848 info.size = -1;
1da177e4
LT		 5849 info.nr_disks = nr;
5850 info.raid_disks = mddev->raid_disks;
5851 info.md_minor = mddev->md_minor;
5852 info.not_persistent= !mddev->persistent;
5853
5854 info.utime = mddev->utime;
5855 info.state = 0;
5856 if (mddev->in_sync)
5857 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5858 if (mddev->bitmap && mddev->bitmap_info.offset)
9bd35920 5859 info.state |= (1<<MD_SB_BITMAP_PRESENT);
ca8895d9
GR		 5860 if (mddev_is_clustered(mddev))
5861 info.state |= (1<<MD_SB_CLUSTERED);
a9f326eb 5862 info.active_disks = insync;
1da177e4
LT		 5863 info.working_disks = working;
5864 info.failed_disks = failed;
5865 info.spare_disks = spare;
5866
5867 info.layout = mddev->layout;
9d8f0363 5868 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5869
5870 if (copy_to_user(arg, &info, sizeof(info)))
5871 return -EFAULT;
5872
5873 return 0;
5874}
5875
f72ffdd6 5876static int get_bitmap_file(struct mddev *mddev, void __user * arg)
32a7627c
N		 5877{
5878 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
f4ad3d38 5879 char *ptr;
4af1a041 5880 int err;
32a7627c 5881
b6878d9e 5882 file = kzalloc(sizeof(*file), GFP_NOIO);
32a7627c 5883 if (!file)
4af1a041 5884 return -ENOMEM;
32a7627c 5885
4af1a041
N		 5886 err = 0;
5887 spin_lock(&mddev->lock);
25eafe1a
BR		 5888 /* bitmap enabled */
5889 if (mddev->bitmap_info.file) {
5890 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5891 sizeof(file->pathname));
5892 if (IS_ERR(ptr))
5893 err = PTR_ERR(ptr);
5894 else
5895 memmove(file->pathname, ptr,
5896 sizeof(file->pathname)-(ptr-file->pathname));
5897 }
4af1a041 5898 spin_unlock(&mddev->lock);
32a7627c 5899
4af1a041
N		 5900 if (err == 0 &&
5901 copy_to_user(arg, file, sizeof(*file)))
32a7627c 5902 err = -EFAULT;
4af1a041 5903
32a7627c
N		 5904 kfree(file);
5905 return err;
5906}
5907
f72ffdd6 5908static int get_disk_info(struct mddev *mddev, void __user * arg)
1da177e4
LT		 5909{
5910 mdu_disk_info_t info;
3cb03002 5911 struct md_rdev *rdev;
1da177e4
LT		 5912
5913 if (copy_from_user(&info, arg, sizeof(info)))
5914 return -EFAULT;
5915
1ca69c4b 5916 rcu_read_lock();
57d051dc 5917 rdev = md_find_rdev_nr_rcu(mddev, info.number);
1da177e4
LT		 5918 if (rdev) {
5919 info.major = MAJOR(rdev->bdev->bd_dev);
5920 info.minor = MINOR(rdev->bdev->bd_dev);
5921 info.raid_disk = rdev->raid_disk;
5922 info.state = 0;
b2d444d7 5923 if (test_bit(Faulty, &rdev->flags))
1da177e4 5924 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5925 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5926 info.state |= (1<<MD_DISK_ACTIVE);
5927 info.state |= (1<<MD_DISK_SYNC);
5928 }
9efdca16 5929 if (test_bit(Journal, &rdev->flags))
bac624f3 5930 info.state |= (1<<MD_DISK_JOURNAL);
8ddf9efe
N		 5931 if (test_bit(WriteMostly, &rdev->flags))
5932 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5933 } else {
5934 info.major = info.minor = 0;
5935 info.raid_disk = -1;
5936 info.state = (1<<MD_DISK_REMOVED);
5937 }
1ca69c4b 5938 rcu_read_unlock();
1da177e4
LT		 5939
5940 if (copy_to_user(arg, &info, sizeof(info)))
5941 return -EFAULT;
5942
5943 return 0;
5944}
5945
f72ffdd6 5946static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
1da177e4
LT		 5947{
5948 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5949 struct md_rdev *rdev;
1da177e4
LT		 5950 dev_t dev = MKDEV(info->major,info->minor);
5951
1aee41f6
GR		 5952 if (mddev_is_clustered(mddev) &&
5953 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
fa8259da 5954 pr_err("%s: Cannot add to clustered mddev.\n",
1aee41f6
GR		 5955 mdname(mddev));
5956 return -EINVAL;
5957 }
5958
1da177e4
LT		 5959 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5960 return -EOVERFLOW;
5961
5962 if (!mddev->raid_disks) {
5963 int err;
5964 /* expecting a device which has a superblock */
5965 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5966 if (IS_ERR(rdev)) {
f72ffdd6 5967 printk(KERN_WARNING
1da177e4
LT		 5968 "md: md_import_device returned %ld\n",
5969 PTR_ERR(rdev));
5970 return PTR_ERR(rdev);
5971 }
5972 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5973 struct md_rdev *rdev0
5974 = list_entry(mddev->disks.next,
5975 struct md_rdev, same_set);
a9f326eb 5976 err = super_types[mddev->major_version]
1da177e4
LT		 5977 .load_super(rdev, rdev0, mddev->minor_version);
5978 if (err < 0) {
f72ffdd6 5979 printk(KERN_WARNING
1da177e4 5980 "md: %s has different UUID to %s\n",
f72ffdd6 5981 bdevname(rdev->bdev,b),
1da177e4
LT		 5982 bdevname(rdev0->bdev,b2));
5983 export_rdev(rdev);
5984 return -EINVAL;
5985 }
5986 }
5987 err = bind_rdev_to_array(rdev, mddev);
5988 if (err)
5989 export_rdev(rdev);
5990 return err;
5991 }
5992
5993 /*
5994 * add_new_disk can be used once the array is assembled
5995 * to add "hot spares". They must already have a superblock
5996 * written
5997 */
5998 if (mddev->pers) {
5999 int err;
6000 if (!mddev->pers->hot_add_disk) {
f72ffdd6 6001 printk(KERN_WARNING
1da177e4
LT		 6002 "%s: personality does not support diskops!\n",
6003 mdname(mddev));
6004 return -EINVAL;
6005 }
7b1e35f6
N		 6006 if (mddev->persistent)
6007 rdev = md_import_device(dev, mddev->major_version,
6008 mddev->minor_version);
6009 else
6010 rdev = md_import_device(dev, -1, -1);
1da177e4 6011 if (IS_ERR(rdev)) {
f72ffdd6 6012 printk(KERN_WARNING
1da177e4
LT		 6013 "md: md_import_device returned %ld\n",
6014 PTR_ERR(rdev));
6015 return PTR_ERR(rdev);
6016 }
1a855a06 6017 /* set saved_raid_disk if appropriate */
41158c7e
N		 6018 if (!mddev->persistent) {
6019 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 6020 info->raid_disk < mddev->raid_disks) {
41158c7e 6021 rdev->raid_disk = info->raid_disk;
bf572541 6022 set_bit(In_sync, &rdev->flags);
8313b8e5 6023 clear_bit(Bitmap_sync, &rdev->flags);
bf572541 6024 } else
41158c7e 6025 rdev->raid_disk = -1;
f466722c 6026 rdev->saved_raid_disk = rdev->raid_disk;
41158c7e
N		 6027 } else
6028 super_types[mddev->major_version].
6029 validate_super(mddev, rdev);
bedd86b7 6030 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 6031 rdev->raid_disk != info->raid_disk) {
bedd86b7
N		 6032 /* This was a hot-add request, but events doesn't
6033 * match, so reject it.
6034 */
6035 export_rdev(rdev);
6036 return -EINVAL;
6037 }
6038
b2d444d7 6039 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 6040 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6041 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 6042 else
6043 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 6044
bac624f3
SL		 6045 if (info->state & (1<<MD_DISK_JOURNAL))
6046 set_bit(Journal, &rdev->flags);
1aee41f6
GR		 6047 /*
6048 * check whether the device shows up in other nodes
6049 */
6050 if (mddev_is_clustered(mddev)) {
dbb64f86 6051 if (info->state & (1 << MD_DISK_CANDIDATE))
1aee41f6 6052 set_bit(Candidate, &rdev->flags);
dbb64f86 6053 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
1aee41f6 6054 /* --add initiated by this node */
dbb64f86 6055 err = md_cluster_ops->add_new_disk(mddev, rdev);
1aee41f6 6056 if (err) {
1aee41f6
GR		 6057 export_rdev(rdev);
6058 return err;
6059 }
6060 }
6061 }
6062
1da177e4
LT		 6063 rdev->raid_disk = -1;
6064 err = bind_rdev_to_array(rdev, mddev);
dbb64f86 6065
1da177e4
LT		 6066 if (err)
6067 export_rdev(rdev);
dbb64f86
GR		 6068
6069 if (mddev_is_clustered(mddev)) {
6070 if (info->state & (1 << MD_DISK_CANDIDATE))
6071 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6072 else {
6073 if (err)
6074 md_cluster_ops->add_new_disk_cancel(mddev);
6075 else
6076 err = add_bound_rdev(rdev);
6077 }
6078
6079 } else if (!err)
a6da4ef8 6080 err = add_bound_rdev(rdev);
dbb64f86 6081
1da177e4
LT		 6082 return err;
6083 }
6084
6085 /* otherwise, add_new_disk is only allowed
6086 * for major_version==0 superblocks
6087 */
6088 if (mddev->major_version != 0) {
6089 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6090 mdname(mddev));
6091 return -EINVAL;
6092 }
6093
6094 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6095 int err;
d710e138 6096 rdev = md_import_device(dev, -1, 0);
1da177e4 6097 if (IS_ERR(rdev)) {
f72ffdd6 6098 printk(KERN_WARNING
1da177e4
LT		 6099 "md: error, md_import_device() returned %ld\n",
6100 PTR_ERR(rdev));
6101 return PTR_ERR(rdev);
6102 }
6103 rdev->desc_nr = info->number;
6104 if (info->raid_disk < mddev->raid_disks)
6105 rdev->raid_disk = info->raid_disk;
6106 else
6107 rdev->raid_disk = -1;
6108
1da177e4 6109 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 6110 if (info->state & (1<<MD_DISK_SYNC))
6111 set_bit(In_sync, &rdev->flags);
1da177e4 6112
8ddf9efe
N		 6113 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6114 set_bit(WriteMostly, &rdev->flags);
6115
1da177e4
LT		 6116 if (!mddev->persistent) {
6117 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 6118 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6119 } else
57b2caa3 6120 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 6121 rdev->sectors = rdev->sb_start;
1da177e4 6122
2bf071bf
N		 6123 err = bind_rdev_to_array(rdev, mddev);
6124 if (err) {
6125 export_rdev(rdev);
6126 return err;
6127 }
1da177e4
LT		 6128 }
6129
6130 return 0;
6131}
6132
f72ffdd6 6133static int hot_remove_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 6134{
6135 char b[BDEVNAME_SIZE];
3cb03002 6136 struct md_rdev *rdev;
1da177e4 6137
1da177e4
LT		 6138 rdev = find_rdev(mddev, dev);
6139 if (!rdev)
6140 return -ENXIO;
6141
2910ff17
GR		 6142 if (rdev->raid_disk < 0)
6143 goto kick_rdev;
293467aa 6144
3ea8929d
N		 6145 clear_bit(Blocked, &rdev->flags);
6146 remove_and_add_spares(mddev, rdev);
6147
1da177e4
LT		 6148 if (rdev->raid_disk >= 0)
6149 goto busy;
6150
2910ff17 6151kick_rdev:
54a88392 6152 if (mddev_is_clustered(mddev))
88bcfef7
GR		 6153 md_cluster_ops->remove_disk(mddev, rdev);
6154
fb56dfef 6155 md_kick_rdev_from_array(rdev);
850b2b42 6156 md_update_sb(mddev, 1);
d7603b7e 6157 md_new_event(mddev);
1da177e4
LT		 6158
6159 return 0;
6160busy:
fdefa4d8 6161 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 6162 bdevname(rdev->bdev,b), mdname(mddev));
6163 return -EBUSY;
6164}
6165
f72ffdd6 6166static int hot_add_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 6167{
6168 char b[BDEVNAME_SIZE];
6169 int err;
3cb03002 6170 struct md_rdev *rdev;
1da177e4
LT		 6171
6172 if (!mddev->pers)
6173 return -ENODEV;
6174
6175 if (mddev->major_version != 0) {
6176 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6177 " version-0 superblocks.\n",
6178 mdname(mddev));
6179 return -EINVAL;
6180 }
6181 if (!mddev->pers->hot_add_disk) {
f72ffdd6 6182 printk(KERN_WARNING
1da177e4
LT		 6183 "%s: personality does not support diskops!\n",
6184 mdname(mddev));
6185 return -EINVAL;
6186 }
6187
d710e138 6188 rdev = md_import_device(dev, -1, 0);
1da177e4 6189 if (IS_ERR(rdev)) {
f72ffdd6 6190 printk(KERN_WARNING
1da177e4
LT		 6191 "md: error, md_import_device() returned %ld\n",
6192 PTR_ERR(rdev));
6193 return -EINVAL;
6194 }
6195
6196 if (mddev->persistent)
57b2caa3 6197 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 6198 else
77304d2a 6199 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 6200
8190e754 6201 rdev->sectors = rdev->sb_start;
1da177e4 6202
b2d444d7 6203 if (test_bit(Faulty, &rdev->flags)) {
f72ffdd6 6204 printk(KERN_WARNING
1da177e4
LT		 6205 "md: can not hot-add faulty %s disk to %s!\n",
6206 bdevname(rdev->bdev,b), mdname(mddev));
6207 err = -EINVAL;
6208 goto abort_export;
6209 }
293467aa 6210
b2d444d7 6211 clear_bit(In_sync, &rdev->flags);
1da177e4 6212 rdev->desc_nr = -1;
5842730d 6213 rdev->saved_raid_disk = -1;
2bf071bf
N		 6214 err = bind_rdev_to_array(rdev, mddev);
6215 if (err)
2aa82191 6216 goto abort_export;
1da177e4
LT		 6217
6218 /*
6219 * The rest should better be atomic, we can have disk failures
6220 * noticed in interrupt contexts ...
6221 */
6222
1da177e4
LT		 6223 rdev->raid_disk = -1;
6224
850b2b42 6225 md_update_sb(mddev, 1);
1da177e4
LT		 6226 /*
6227 * Kick recovery, maybe this spare has to be added to the
6228 * array immediately.
6229 */
6230 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6231 md_wakeup_thread(mddev->thread);
d7603b7e 6232 md_new_event(mddev);
1da177e4
LT		 6233 return 0;
6234
1da177e4
LT		 6235abort_export:
6236 export_rdev(rdev);
6237 return err;
6238}
6239
fd01b88c 6240static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c 6241{
035328c2 6242 int err = 0;
32a7627c 6243
36fa3063 6244 if (mddev->pers) {
d66b1b39 6245 if (!mddev->pers->quiesce || !mddev->thread)
36fa3063
N		 6246 return -EBUSY;
6247 if (mddev->recovery || mddev->sync_thread)
6248 return -EBUSY;
6249 /* we should be able to change the bitmap.. */
6250 }
32a7627c 6251
36fa3063 6252 if (fd >= 0) {
035328c2 6253 struct inode *inode;
1e594bb2
N		 6254 struct file *f;
6255
6256 if (mddev->bitmap || mddev->bitmap_info.file)
36fa3063 6257 return -EEXIST; /* cannot add when bitmap is present */
1e594bb2 6258 f = fget(fd);
32a7627c 6259
1e594bb2 6260 if (f == NULL) {
36fa3063
N		 6261 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6262 mdname(mddev));
6263 return -EBADF;
6264 }
6265
1e594bb2 6266 inode = f->f_mapping->host;
035328c2
N		 6267 if (!S_ISREG(inode->i_mode)) {
6268 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6269 mdname(mddev));
6270 err = -EBADF;
1e594bb2 6271 } else if (!(f->f_mode & FMODE_WRITE)) {
035328c2
N		 6272 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6273 mdname(mddev));
6274 err = -EBADF;
6275 } else if (atomic_read(&inode->i_writecount) != 1) {
36fa3063
N		 6276 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6277 mdname(mddev));
035328c2
N		 6278 err = -EBUSY;
6279 }
6280 if (err) {
1e594bb2 6281 fput(f);
36fa3063
N		 6282 return err;
6283 }
1e594bb2 6284 mddev->bitmap_info.file = f;
c3d9714e 6285 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 6286 } else if (mddev->bitmap == NULL)
6287 return -ENOENT; /* cannot remove what isn't there */
6288 err = 0;
6289 if (mddev->pers) {
6290 mddev->pers->quiesce(mddev, 1);
69e51b44 6291 if (fd >= 0) {
f9209a32
GR		 6292 struct bitmap *bitmap;
6293
6294 bitmap = bitmap_create(mddev, -1);
6295 if (!IS_ERR(bitmap)) {
6296 mddev->bitmap = bitmap;
69e51b44 6297 err = bitmap_load(mddev);
ba599aca
N		 6298 } else
6299 err = PTR_ERR(bitmap);
69e51b44 6300 }
d7375ab3 6301 if (fd < 0 || err) {
36fa3063 6302 bitmap_destroy(mddev);
d7375ab3
N		 6303 fd = -1; /* make sure to put the file */
6304 }
36fa3063 6305 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 6306 }
6307 if (fd < 0) {
4af1a041
N		 6308 struct file *f = mddev->bitmap_info.file;
6309 if (f) {
6310 spin_lock(&mddev->lock);
6311 mddev->bitmap_info.file = NULL;
6312 spin_unlock(&mddev->lock);
6313 fput(f);
6314 }
36fa3063
N		 6315 }
6316
32a7627c
N		 6317 return err;
6318}
6319
1da177e4
LT		 6320/*
6321 * set_array_info is used two different ways
6322 * The original usage is when creating a new array.
6323 * In this usage, raid_disks is > 0 and it together with
6324 * level, size, not_persistent,layout,chunksize determine the
6325 * shape of the array.
6326 * This will always create an array with a type-0.90.0 superblock.
6327 * The newer usage is when assembling an array.
6328 * In this case raid_disks will be 0, and the major_version field is
6329 * use to determine which style super-blocks are to be found on the devices.
6330 * The minor and patch _version numbers are also kept incase the
6331 * super_block handler wishes to interpret them.
6332 */
f72ffdd6 6333static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6334{
6335
6336 if (info->raid_disks == 0) {
6337 /* just setting version number for superblock loading */
6338 if (info->major_version < 0 ||
50511da3 6339 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 6340 super_types[info->major_version].name == NULL) {
6341 /* maybe try to auto-load a module? */
f72ffdd6 6342 printk(KERN_INFO
1da177e4
LT		 6343 "md: superblock version %d not known\n",
6344 info->major_version);
6345 return -EINVAL;
6346 }
6347 mddev->major_version = info->major_version;
6348 mddev->minor_version = info->minor_version;
6349 mddev->patch_version = info->patch_version;
3f9d7b0d 6350 mddev->persistent = !info->not_persistent;
cbd19983
N		 6351 /* ensure mddev_put doesn't delete this now that there
6352 * is some minimal configuration.
6353 */
6354 mddev->ctime = get_seconds();
1da177e4
LT		 6355 return 0;
6356 }
6357 mddev->major_version = MD_MAJOR_VERSION;
6358 mddev->minor_version = MD_MINOR_VERSION;
6359 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6360 mddev->ctime = get_seconds();
6361
6362 mddev->level = info->level;
17115e03 6363 mddev->clevel[0] = 0;
58c0fed4 6364 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 6365 mddev->raid_disks = info->raid_disks;
6366 /* don't set md_minor, it is determined by which /dev/md* was
6367 * openned
6368 */
6369 if (info->state & (1<<MD_SB_CLEAN))
6370 mddev->recovery_cp = MaxSector;
6371 else
6372 mddev->recovery_cp = 0;
6373 mddev->persistent = ! info->not_persistent;
e691063a 6374 mddev->external = 0;
1da177e4
LT		 6375
6376 mddev->layout = info->layout;
9d8f0363 6377 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 6378
6379 mddev->max_disks = MD_SB_DISKS;
6380
e691063a
N		 6381 if (mddev->persistent)
6382 mddev->flags = 0;
850b2b42 6383 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6384
c3d9714e 6385 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6386 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6387 mddev->bitmap_info.offset = 0;
b2a2703c 6388
f6705578
N		 6389 mddev->reshape_position = MaxSector;
6390
1da177e4
LT		 6391 /*
6392 * Generate a 128 bit UUID
6393 */
6394 get_random_bytes(mddev->uuid, 16);
6395
f6705578 6396 mddev->new_level = mddev->level;
664e7c41 6397 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 6398 mddev->new_layout = mddev->layout;
6399 mddev->delta_disks = 0;
2c810cdd 6400 mddev->reshape_backwards = 0;
f6705578 6401
1da177e4
LT		 6402 return 0;
6403}
6404
fd01b88c 6405void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6406{
b522adcd
DW		 6407 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6408
6409 if (mddev->external_size)
6410 return;
6411
1f403624
DW		 6412 mddev->array_sectors = array_sectors;
6413}
6414EXPORT_SYMBOL(md_set_array_sectors);
6415
fd01b88c 6416static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6417{
3cb03002 6418 struct md_rdev *rdev;
a35b0d69 6419 int rv;
d71f9f88 6420 int fit = (num_sectors == 0);
a35b0d69
N		 6421
6422 if (mddev->pers->resize == NULL)
6423 return -EINVAL;
d71f9f88
AN		 6424 /* The "num_sectors" is the number of sectors of each device that
6425 * is used. This can only make sense for arrays with redundancy.
6426 * linear and raid0 always use whatever space is available. We can only
6427 * consider changing this number if no resync or reconstruction is
6428 * happening, and if the new size is acceptable. It must fit before the
0f420358 6429 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 6430 * of each device. If num_sectors is zero, we find the largest size
6431 * that fits.
a35b0d69 6432 */
f851b60d
N		 6433 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6434 mddev->sync_thread)
a35b0d69 6435 return -EBUSY;
bd8839e0
N		 6436 if (mddev->ro)
6437 return -EROFS;
a4a6125a 6438
dafb20fa 6439 rdev_for_each(rdev, mddev) {
dd8ac336 6440 sector_t avail = rdev->sectors;
01ab5662 6441
d71f9f88
AN		 6442 if (fit && (num_sectors == 0 || num_sectors > avail))
6443 num_sectors = avail;
6444 if (avail < num_sectors)
a35b0d69
N		 6445 return -ENOSPC;
6446 }
d71f9f88 6447 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 6448 if (!rv)
6449 revalidate_disk(mddev->gendisk);
a35b0d69
N		 6450 return rv;
6451}
6452
fd01b88c 6453static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 6454{
6455 int rv;
c6563a8c 6456 struct md_rdev *rdev;
da943b99 6457 /* change the number of raid disks */
63c70c4f 6458 if (mddev->pers->check_reshape == NULL)
da943b99 6459 return -EINVAL;
bd8839e0
N		 6460 if (mddev->ro)
6461 return -EROFS;
da943b99 6462 if (raid_disks <= 0 ||
233fca36 6463 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 6464 return -EINVAL;
f851b60d
N		 6465 if (mddev->sync_thread ||
6466 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6467 mddev->reshape_position != MaxSector)
da943b99 6468 return -EBUSY;
c6563a8c
N		 6469
6470 rdev_for_each(rdev, mddev) {
6471 if (mddev->raid_disks < raid_disks &&
6472 rdev->data_offset < rdev->new_data_offset)
6473 return -EINVAL;
6474 if (mddev->raid_disks > raid_disks &&
6475 rdev->data_offset > rdev->new_data_offset)
6476 return -EINVAL;
6477 }
6478
63c70c4f 6479 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N		 6480 if (mddev->delta_disks < 0)
6481 mddev->reshape_backwards = 1;
6482 else if (mddev->delta_disks > 0)
6483 mddev->reshape_backwards = 0;
63c70c4f
N		 6484
6485 rv = mddev->pers->check_reshape(mddev);
2c810cdd 6486 if (rv < 0) {
de171cb9 6487 mddev->delta_disks = 0;
2c810cdd
N		 6488 mddev->reshape_backwards = 0;
6489 }
da943b99
N		 6490 return rv;
6491}
6492
1da177e4
LT		 6493/*
6494 * update_array_info is used to change the configuration of an
6495 * on-line array.
6496 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6497 * fields in the info are checked against the array.
6498 * Any differences that cannot be handled will cause an error.
6499 * Normally, only one change can be managed at a time.
6500 */
fd01b88c 6501static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6502{
6503 int rv = 0;
6504 int cnt = 0;
36fa3063
N		 6505 int state = 0;
6506
6507 /* calculate expected state,ignoring low bits */
c3d9714e 6508 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 6509 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 6510
6511 if (mddev->major_version != info->major_version ||
6512 mddev->minor_version != info->minor_version ||
6513/* mddev->patch_version != info->patch_version || */
6514 mddev->ctime != info->ctime ||
6515 mddev->level != info->level ||
6516/* mddev->layout != info->layout || */
4e023612 6517 mddev->persistent != !info->not_persistent ||
9d8f0363 6518 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 6519 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6520 ((state^info->state) & 0xfffffe00)
6521 )
1da177e4
LT		 6522 return -EINVAL;
6523 /* Check there is only one change */
58c0fed4
AN		 6524 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6525 cnt++;
6526 if (mddev->raid_disks != info->raid_disks)
6527 cnt++;
6528 if (mddev->layout != info->layout)
6529 cnt++;
6530 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6531 cnt++;
6532 if (cnt == 0)
6533 return 0;
6534 if (cnt > 1)
6535 return -EINVAL;
1da177e4
LT		 6536
6537 if (mddev->layout != info->layout) {
6538 /* Change layout
6539 * we don't need to do anything at the md level, the
6540 * personality will take care of it all.
6541 */
50ac168a 6542 if (mddev->pers->check_reshape == NULL)
1da177e4 6543 return -EINVAL;
597a711b
N		 6544 else {
6545 mddev->new_layout = info->layout;
50ac168a 6546 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 6547 if (rv)
6548 mddev->new_layout = mddev->layout;
6549 return rv;
6550 }
1da177e4 6551 }
58c0fed4 6552 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6553 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6554
da943b99
N		 6555 if (mddev->raid_disks != info->raid_disks)
6556 rv = update_raid_disks(mddev, info->raid_disks);
6557
36fa3063 6558 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
293467aa
GR		 6559 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6560 rv = -EINVAL;
6561 goto err;
6562 }
6563 if (mddev->recovery || mddev->sync_thread) {
6564 rv = -EBUSY;
6565 goto err;
6566 }
36fa3063 6567 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
f9209a32 6568 struct bitmap *bitmap;
36fa3063 6569 /* add the bitmap */
293467aa
GR		 6570 if (mddev->bitmap) {
6571 rv = -EEXIST;
6572 goto err;
6573 }
6574 if (mddev->bitmap_info.default_offset == 0) {
6575 rv = -EINVAL;
6576 goto err;
6577 }
c3d9714e
N		 6578 mddev->bitmap_info.offset =
6579 mddev->bitmap_info.default_offset;
6409bb05
N		 6580 mddev->bitmap_info.space =
6581 mddev->bitmap_info.default_space;
36fa3063 6582 mddev->pers->quiesce(mddev, 1);
f9209a32
GR		 6583 bitmap = bitmap_create(mddev, -1);
6584 if (!IS_ERR(bitmap)) {
6585 mddev->bitmap = bitmap;
69e51b44 6586 rv = bitmap_load(mddev);
ba599aca
N		 6587 } else
6588 rv = PTR_ERR(bitmap);
36fa3063
N		 6589 if (rv)
6590 bitmap_destroy(mddev);
6591 mddev->pers->quiesce(mddev, 0);
6592 } else {
6593 /* remove the bitmap */
293467aa
GR		 6594 if (!mddev->bitmap) {
6595 rv = -ENOENT;
6596 goto err;
6597 }
6598 if (mddev->bitmap->storage.file) {
6599 rv = -EINVAL;
6600 goto err;
6601 }
f6a2dc64
GJ		 6602 if (mddev->bitmap_info.nodes) {
6603 /* hold PW on all the bitmap lock */
6604 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6605 printk("md: can't change bitmap to none since the"
6606 " array is in use by more than one node\n");
6607 rv = -EPERM;
6608 md_cluster_ops->unlock_all_bitmaps(mddev);
6609 goto err;
6610 }
6611
6612 mddev->bitmap_info.nodes = 0;
6613 md_cluster_ops->leave(mddev);
6614 }
36fa3063
N		 6615 mddev->pers->quiesce(mddev, 1);
6616 bitmap_destroy(mddev);
6617 mddev->pers->quiesce(mddev, 0);
c3d9714e 6618 mddev->bitmap_info.offset = 0;
36fa3063
N		 6619 }
6620 }
850b2b42 6621 md_update_sb(mddev, 1);
293467aa
GR		 6622 return rv;
6623err:
1da177e4
LT		 6624 return rv;
6625}
6626
fd01b88c 6627static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6628{
3cb03002 6629 struct md_rdev *rdev;
1ca69c4b 6630 int err = 0;
1da177e4
LT		 6631
6632 if (mddev->pers == NULL)
6633 return -ENODEV;
6634
1ca69c4b
N		 6635 rcu_read_lock();
6636 rdev = find_rdev_rcu(mddev, dev);
1da177e4 6637 if (!rdev)
1ca69c4b
N		 6638 err = -ENODEV;
6639 else {
6640 md_error(mddev, rdev);
6641 if (!test_bit(Faulty, &rdev->flags))
6642 err = -EBUSY;
6643 }
6644 rcu_read_unlock();
6645 return err;
1da177e4
LT		 6646}
6647
2f9618ce
AN		 6648/*
6649 * We have a problem here : there is no easy way to give a CHS
6650 * virtual geometry. We currently pretend that we have a 2 heads
6651 * 4 sectors (with a BIG number of cylinders...). This drives
6652 * dosfs just mad... ;-)
6653 */
a885c8c4
CH		 6654static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6655{
fd01b88c 6656 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6657
6658 geo->heads = 2;
6659 geo->sectors = 4;
49ce6cea 6660 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6661 return 0;
6662}
6663
cb335f88
NS		 6664static inline bool md_ioctl_valid(unsigned int cmd)
6665{
6666 switch (cmd) {
6667 case ADD_NEW_DISK:
6668 case BLKROSET:
6669 case GET_ARRAY_INFO:
6670 case GET_BITMAP_FILE:
6671 case GET_DISK_INFO:
6672 case HOT_ADD_DISK:
6673 case HOT_REMOVE_DISK:
cb335f88
NS		 6674 case RAID_AUTORUN:
6675 case RAID_VERSION:
6676 case RESTART_ARRAY_RW:
6677 case RUN_ARRAY:
6678 case SET_ARRAY_INFO:
6679 case SET_BITMAP_FILE:
6680 case SET_DISK_FAULTY:
6681 case STOP_ARRAY:
6682 case STOP_ARRAY_RO:
1aee41f6 6683 case CLUSTERED_DISK_NACK:
cb335f88
NS		 6684 return true;
6685 default:
6686 return false;
6687 }
6688}
6689
a39907fa 6690static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6691 unsigned int cmd, unsigned long arg)
6692{
6693 int err = 0;
6694 void __user *argp = (void __user *)arg;
fd01b88c 6695 struct mddev *mddev = NULL;
e2218350 6696 int ro;
1da177e4 6697
cb335f88
NS		 6698 if (!md_ioctl_valid(cmd))
6699 return -ENOTTY;
6700
506c9e44
N		 6701 switch (cmd) {
6702 case RAID_VERSION:
6703 case GET_ARRAY_INFO:
6704 case GET_DISK_INFO:
6705 break;
6706 default:
6707 if (!capable(CAP_SYS_ADMIN))
6708 return -EACCES;
6709 }
1da177e4
LT		 6710
6711 /*
6712 * Commands dealing with the RAID driver but not any
6713 * particular array:
6714 */
c02c0aeb
N		 6715 switch (cmd) {
6716 case RAID_VERSION:
6717 err = get_version(argp);
3adc28d8 6718 goto out;
1da177e4 6719
1da177e4 6720#ifndef MODULE
c02c0aeb
N		 6721 case RAID_AUTORUN:
6722 err = 0;
6723 autostart_arrays(arg);
3adc28d8 6724 goto out;
1da177e4 6725#endif
c02c0aeb 6726 default:;
1da177e4
LT		 6727 }
6728
6729 /*
6730 * Commands creating/starting a new array:
6731 */
6732
a39907fa 6733 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6734
6735 if (!mddev) {
6736 BUG();
3adc28d8 6737 goto out;
1da177e4
LT		 6738 }
6739
1ca69c4b
N		 6740 /* Some actions do not requires the mutex */
6741 switch (cmd) {
6742 case GET_ARRAY_INFO:
6743 if (!mddev->raid_disks && !mddev->external)
6744 err = -ENODEV;
6745 else
6746 err = get_array_info(mddev, argp);
3adc28d8 6747 goto out;
1ca69c4b
N		 6748
6749 case GET_DISK_INFO:
6750 if (!mddev->raid_disks && !mddev->external)
6751 err = -ENODEV;
6752 else
6753 err = get_disk_info(mddev, argp);
3adc28d8 6754 goto out;
1ca69c4b
N		 6755
6756 case SET_DISK_FAULTY:
6757 err = set_disk_faulty(mddev, new_decode_dev(arg));
3adc28d8 6758 goto out;
4af1a041
N		 6759
6760 case GET_BITMAP_FILE:
6761 err = get_bitmap_file(mddev, argp);
6762 goto out;
6763
1ca69c4b
N		 6764 }
6765
a7a3f08d
N		 6766 if (cmd == ADD_NEW_DISK)
6767 /* need to ensure md_delayed_delete() has completed */
6768 flush_workqueue(md_misc_wq);
6769
90f5f7ad
HR		 6770 if (cmd == HOT_REMOVE_DISK)
6771 /* need to ensure recovery thread has run */
6772 wait_event_interruptible_timeout(mddev->sb_wait,
6773 !test_bit(MD_RECOVERY_NEEDED,
6774 &mddev->flags),
6775 msecs_to_jiffies(5000));
260fa034
N		 6776 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6777 /* Need to flush page cache, and ensure no-one else opens
6778 * and writes
6779 */
6780 mutex_lock(&mddev->open_mutex);
9ba3b7f5 6781 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
260fa034
N		 6782 mutex_unlock(&mddev->open_mutex);
6783 err = -EBUSY;
3adc28d8 6784 goto out;
260fa034
N		 6785 }
6786 set_bit(MD_STILL_CLOSED, &mddev->flags);
6787 mutex_unlock(&mddev->open_mutex);
6788 sync_blockdev(bdev);
6789 }
1da177e4
LT		 6790 err = mddev_lock(mddev);
6791 if (err) {
f72ffdd6 6792 printk(KERN_INFO
1da177e4
LT		 6793 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6794 err, cmd);
3adc28d8 6795 goto out;
1da177e4
LT		 6796 }
6797
c02c0aeb
N		 6798 if (cmd == SET_ARRAY_INFO) {
6799 mdu_array_info_t info;
6800 if (!arg)
6801 memset(&info, 0, sizeof(info));
6802 else if (copy_from_user(&info, argp, sizeof(info))) {
6803 err = -EFAULT;
3adc28d8 6804 goto unlock;
c02c0aeb
N		 6805 }
6806 if (mddev->pers) {
6807 err = update_array_info(mddev, &info);
6808 if (err) {
6809 printk(KERN_WARNING "md: couldn't update"
6810 " array info. %d\n", err);
3adc28d8 6811 goto unlock;
1da177e4 6812 }
3adc28d8 6813 goto unlock;
c02c0aeb
N		 6814 }
6815 if (!list_empty(&mddev->disks)) {
6816 printk(KERN_WARNING
6817 "md: array %s already has disks!\n",
6818 mdname(mddev));
6819 err = -EBUSY;
3adc28d8 6820 goto unlock;
c02c0aeb
N		 6821 }
6822 if (mddev->raid_disks) {
6823 printk(KERN_WARNING
6824 "md: array %s already initialised!\n",
6825 mdname(mddev));
6826 err = -EBUSY;
3adc28d8 6827 goto unlock;
c02c0aeb
N		 6828 }
6829 err = set_array_info(mddev, &info);
6830 if (err) {
6831 printk(KERN_WARNING "md: couldn't set"
6832 " array info. %d\n", err);
3adc28d8 6833 goto unlock;
c02c0aeb 6834 }
3adc28d8 6835 goto unlock;
1da177e4
LT		 6836 }
6837
6838 /*
6839 * Commands querying/configuring an existing array:
6840 */
32a7627c 6841 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6842 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6843 if ((!mddev->raid_disks && !mddev->external)
6844 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6845 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6846 && cmd != GET_BITMAP_FILE) {
1da177e4 6847 err = -ENODEV;
3adc28d8 6848 goto unlock;
1da177e4
LT		 6849 }
6850
6851 /*
6852 * Commands even a read-only array can execute:
6853 */
c02c0aeb 6854 switch (cmd) {
c02c0aeb
N		 6855 case RESTART_ARRAY_RW:
6856 err = restart_array(mddev);
3adc28d8 6857 goto unlock;
1da177e4 6858
c02c0aeb
N		 6859 case STOP_ARRAY:
6860 err = do_md_stop(mddev, 0, bdev);
3adc28d8 6861 goto unlock;
1da177e4 6862
c02c0aeb
N		 6863 case STOP_ARRAY_RO:
6864 err = md_set_readonly(mddev, bdev);
3adc28d8 6865 goto unlock;
1da177e4 6866
3ea8929d
N		 6867 case HOT_REMOVE_DISK:
6868 err = hot_remove_disk(mddev, new_decode_dev(arg));
3adc28d8 6869 goto unlock;
3ea8929d 6870
7ceb17e8
N		 6871 case ADD_NEW_DISK:
6872 /* We can support ADD_NEW_DISK on read-only arrays
6873 * on if we are re-adding a preexisting device.
6874 * So require mddev->pers and MD_DISK_SYNC.
6875 */
6876 if (mddev->pers) {
6877 mdu_disk_info_t info;
6878 if (copy_from_user(&info, argp, sizeof(info)))
6879 err = -EFAULT;
6880 else if (!(info.state & (1<<MD_DISK_SYNC)))
6881 /* Need to clear read-only for this */
6882 break;
6883 else
6884 err = add_new_disk(mddev, &info);
3adc28d8 6885 goto unlock;
7ceb17e8
N		 6886 }
6887 break;
6888
c02c0aeb
N		 6889 case BLKROSET:
6890 if (get_user(ro, (int __user *)(arg))) {
6891 err = -EFAULT;
3adc28d8 6892 goto unlock;
c02c0aeb
N		 6893 }
6894 err = -EINVAL;
e2218350 6895
c02c0aeb
N		 6896 /* if the bdev is going readonly the value of mddev->ro
6897 * does not matter, no writes are coming
6898 */
6899 if (ro)
3adc28d8 6900 goto unlock;
e2218350 6901
c02c0aeb
N		 6902 /* are we are already prepared for writes? */
6903 if (mddev->ro != 1)
3adc28d8 6904 goto unlock;
e2218350 6905
c02c0aeb
N		 6906 /* transitioning to readauto need only happen for
6907 * arrays that call md_write_start
6908 */
6909 if (mddev->pers) {
6910 err = restart_array(mddev);
6911 if (err == 0) {
6912 mddev->ro = 2;
6913 set_disk_ro(mddev->gendisk, 0);
e2218350 6914 }
c02c0aeb 6915 }
3adc28d8 6916 goto unlock;
1da177e4
LT		 6917 }
6918
6919 /*
6920 * The remaining ioctls are changing the state of the
f91de92e 6921 * superblock, so we do not allow them on read-only arrays.
1da177e4 6922 */
326eb17d 6923 if (mddev->ro && mddev->pers) {
f91de92e
N		 6924 if (mddev->ro == 2) {
6925 mddev->ro = 0;
00bcb4ac 6926 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 6927 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N		 6928 /* mddev_unlock will wake thread */
6929 /* If a device failed while we were read-only, we
6930 * need to make sure the metadata is updated now.
6931 */
6932 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6933 mddev_unlock(mddev);
6934 wait_event(mddev->sb_wait,
6935 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6936 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
29f097c4 6937 mddev_lock_nointr(mddev);
f3378b48 6938 }
f91de92e
N		 6939 } else {
6940 err = -EROFS;
3adc28d8 6941 goto unlock;
f91de92e 6942 }
1da177e4
LT		 6943 }
6944
c02c0aeb
N		 6945 switch (cmd) {
6946 case ADD_NEW_DISK:
1da177e4 6947 {
c02c0aeb
N		 6948 mdu_disk_info_t info;
6949 if (copy_from_user(&info, argp, sizeof(info)))
6950 err = -EFAULT;
6951 else
6952 err = add_new_disk(mddev, &info);
3adc28d8 6953 goto unlock;
c02c0aeb 6954 }
1da177e4 6955
1aee41f6
GR		 6956 case CLUSTERED_DISK_NACK:
6957 if (mddev_is_clustered(mddev))
6958 md_cluster_ops->new_disk_ack(mddev, false);
6959 else
6960 err = -EINVAL;
6961 goto unlock;
6962
c02c0aeb
N		 6963 case HOT_ADD_DISK:
6964 err = hot_add_disk(mddev, new_decode_dev(arg));
3adc28d8 6965 goto unlock;
1da177e4 6966
c02c0aeb
N		 6967 case RUN_ARRAY:
6968 err = do_md_run(mddev);
3adc28d8 6969 goto unlock;
1da177e4 6970
c02c0aeb
N		 6971 case SET_BITMAP_FILE:
6972 err = set_bitmap_file(mddev, (int)arg);
3adc28d8 6973 goto unlock;
32a7627c 6974
c02c0aeb
N		 6975 default:
6976 err = -EINVAL;
3adc28d8 6977 goto unlock;
1da177e4
LT		 6978 }
6979
3adc28d8 6980unlock:
d3374825
N		 6981 if (mddev->hold_active == UNTIL_IOCTL &&
6982 err != -EINVAL)
6983 mddev->hold_active = 0;
1da177e4 6984 mddev_unlock(mddev);
3adc28d8 6985out:
1da177e4
LT		 6986 return err;
6987}
aa98aa31
AB		 6988#ifdef CONFIG_COMPAT
6989static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6990 unsigned int cmd, unsigned long arg)
6991{
6992 switch (cmd) {
6993 case HOT_REMOVE_DISK:
6994 case HOT_ADD_DISK:
6995 case SET_DISK_FAULTY:
6996 case SET_BITMAP_FILE:
6997 /* These take in integer arg, do not convert */
6998 break;
6999 default:
7000 arg = (unsigned long)compat_ptr(arg);
7001 break;
7002 }
7003
7004 return md_ioctl(bdev, mode, cmd, arg);
7005}
7006#endif /* CONFIG_COMPAT */
1da177e4 7007
a39907fa 7008static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 7009{
7010 /*
7011 * Succeed if we can lock the mddev, which confirms that
7012 * it isn't being stopped right now.
7013 */
fd01b88c 7014 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 7015 int err;
7016
0c098220
YL		 7017 if (!mddev)
7018 return -ENODEV;
7019
d3374825
N		 7020 if (mddev->gendisk != bdev->bd_disk) {
7021 /* we are racing with mddev_put which is discarding this
7022 * bd_disk.
7023 */
7024 mddev_put(mddev);
7025 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 7026 flush_workqueue(md_misc_wq);
d3374825
N		 7027 /* Then retry the open from the top */
7028 return -ERESTARTSYS;
7029 }
7030 BUG_ON(mddev != bdev->bd_disk->private_data);
7031
c8c00a69 7032 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 7033 goto out;
7034
7035 err = 0;
f2ea68cf 7036 atomic_inc(&mddev->openers);
260fa034 7037 clear_bit(MD_STILL_CLOSED, &mddev->flags);
c8c00a69 7038 mutex_unlock(&mddev->open_mutex);
1da177e4 7039
f0b4f7e2 7040 check_disk_change(bdev);
1da177e4
LT		 7041 out:
7042 return err;
7043}
7044
db2a144b 7045static void md_release(struct gendisk *disk, fmode_t mode)
1da177e4 7046{
f72ffdd6 7047 struct mddev *mddev = disk->private_data;
1da177e4 7048
52e5f9d1 7049 BUG_ON(!mddev);
f2ea68cf 7050 atomic_dec(&mddev->openers);
1da177e4 7051 mddev_put(mddev);
1da177e4 7052}
f0b4f7e2
N		 7053
7054static int md_media_changed(struct gendisk *disk)
7055{
fd01b88c 7056 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 7057
7058 return mddev->changed;
7059}
7060
7061static int md_revalidate(struct gendisk *disk)
7062{
fd01b88c 7063 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 7064
7065 mddev->changed = 0;
7066 return 0;
7067}
83d5cde4 7068static const struct block_device_operations md_fops =
1da177e4
LT		 7069{
7070 .owner = THIS_MODULE,
a39907fa
AV		 7071 .open = md_open,
7072 .release = md_release,
b492b852 7073 .ioctl = md_ioctl,
aa98aa31
AB		 7074#ifdef CONFIG_COMPAT
7075 .compat_ioctl = md_compat_ioctl,
7076#endif
a885c8c4 7077 .getgeo = md_getgeo,
f0b4f7e2
N		 7078 .media_changed = md_media_changed,
7079 .revalidate_disk= md_revalidate,
1da177e4
LT		 7080};
7081
f72ffdd6 7082static int md_thread(void *arg)
1da177e4 7083{
2b8bf345 7084 struct md_thread *thread = arg;
1da177e4 7085
1da177e4
LT		 7086 /*
7087 * md_thread is a 'system-thread', it's priority should be very
7088 * high. We avoid resource deadlocks individually in each
7089 * raid personality. (RAID5 does preallocation) We also use RR and
7090 * the very same RT priority as kswapd, thus we will never get
7091 * into a priority inversion deadlock.
7092 *
7093 * we definitely have to have equal or higher priority than
7094 * bdflush, otherwise bdflush will deadlock if there are too
7095 * many dirty RAID5 blocks.
7096 */
1da177e4 7097
6985c43f 7098 allow_signal(SIGKILL);
a6fb0934 7099 while (!kthread_should_stop()) {
1da177e4 7100
93588e22
N		 7101 /* We need to wait INTERRUPTIBLE so that
7102 * we don't add to the load-average.
7103 * That means we need to be sure no signals are
7104 * pending
7105 */
7106 if (signal_pending(current))
7107 flush_signals(current);
7108
7109 wait_event_interruptible_timeout
7110 (thread->wqueue,
7111 test_bit(THREAD_WAKEUP, &thread->flags)
7112 || kthread_should_stop(),
7113 thread->timeout);
1da177e4 7114
6c987910
N		 7115 clear_bit(THREAD_WAKEUP, &thread->flags);
7116 if (!kthread_should_stop())
4ed8731d 7117 thread->run(thread);
1da177e4 7118 }
a6fb0934 7119
1da177e4
LT		 7120 return 0;
7121}
7122
2b8bf345 7123void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 7124{
7125 if (thread) {
36a4e1fe 7126 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 7127 set_bit(THREAD_WAKEUP, &thread->flags);
7128 wake_up(&thread->wqueue);
7129 }
7130}
6c144d31 7131EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 7132
4ed8731d
SL		 7133struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7134 struct mddev *mddev, const char *name)
1da177e4 7135{
2b8bf345 7136 struct md_thread *thread;
1da177e4 7137
2b8bf345 7138 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 7139 if (!thread)
7140 return NULL;
7141
1da177e4
LT		 7142 init_waitqueue_head(&thread->wqueue);
7143
1da177e4
LT		 7144 thread->run = run;
7145 thread->mddev = mddev;
32a7627c 7146 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 7147 thread->tsk = kthread_run(md_thread, thread,
7148 "%s_%s",
7149 mdname(thread->mddev),
0232605d 7150 name);
a6fb0934 7151 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 7152 kfree(thread);
7153 return NULL;
7154 }
1da177e4
LT		 7155 return thread;
7156}
6c144d31 7157EXPORT_SYMBOL(md_register_thread);
1da177e4 7158
2b8bf345 7159void md_unregister_thread(struct md_thread **threadp)
1da177e4 7160{
2b8bf345 7161 struct md_thread *thread = *threadp;
e0cf8f04
N		 7162 if (!thread)
7163 return;
36a4e1fe 7164 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 7165 /* Locking ensures that mddev_unlock does not wake_up a
7166 * non-existent thread
7167 */
7168 spin_lock(&pers_lock);
7169 *threadp = NULL;
7170 spin_unlock(&pers_lock);
a6fb0934
N		 7171
7172 kthread_stop(thread->tsk);
1da177e4
LT		 7173 kfree(thread);
7174}
6c144d31 7175EXPORT_SYMBOL(md_unregister_thread);
1da177e4 7176
fd01b88c 7177void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7178{
b2d444d7 7179 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 7180 return;
6bfe0b49 7181
de393cde 7182 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 7183 return;
7184 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 7185 if (mddev->degraded)
7186 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 7187 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 7188 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7189 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7190 md_wakeup_thread(mddev->thread);
768a418d 7191 if (mddev->event_work.func)
e804ac78 7192 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 7193 md_new_event_inintr(mddev);
1da177e4 7194}
6c144d31 7195EXPORT_SYMBOL(md_error);
1da177e4
LT		 7196
7197/* seq_file implementation /proc/mdstat */
7198
7199static void status_unused(struct seq_file *seq)
7200{
7201 int i = 0;
3cb03002 7202 struct md_rdev *rdev;
1da177e4
LT		 7203
7204 seq_printf(seq, "unused devices: ");
7205
159ec1fc 7206 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 7207 char b[BDEVNAME_SIZE];
7208 i++;
7209 seq_printf(seq, "%s ",
7210 bdevname(rdev->bdev,b));
7211 }
7212 if (!i)
7213 seq_printf(seq, "<none>");
7214
7215 seq_printf(seq, "\n");
7216}
7217
f7851be7 7218static int status_resync(struct seq_file *seq, struct mddev *mddev)
1da177e4 7219{
dd71cf6b
N		 7220 sector_t max_sectors, resync, res;
7221 unsigned long dt, db;
7222 sector_t rt;
4588b42e
N		 7223 int scale;
7224 unsigned int per_milli;
1da177e4 7225
c804cdec
N		 7226 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7227 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 7228 max_sectors = mddev->resync_max_sectors;
1da177e4 7229 else
dd71cf6b 7230 max_sectors = mddev->dev_sectors;
1da177e4 7231
f7851be7
N		 7232 resync = mddev->curr_resync;
7233 if (resync <= 3) {
7234 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7235 /* Still cleaning up */
7236 resync = max_sectors;
7237 } else
7238 resync -= atomic_read(&mddev->recovery_active);
7239
7240 if (resync == 0) {
7241 if (mddev->recovery_cp < MaxSector) {
7242 seq_printf(seq, "\tresync=PENDING");
7243 return 1;
7244 }
7245 return 0;
7246 }
7247 if (resync < 3) {
7248 seq_printf(seq, "\tresync=DELAYED");
7249 return 1;
7250 }
7251
403df478 7252 WARN_ON(max_sectors == 0);
4588b42e 7253 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 7254 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 7255 * u32, as those are the requirements for sector_div.
7256 * Thus 'scale' must be at least 10
7257 */
7258 scale = 10;
7259 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 7260 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 7261 scale++;
7262 }
7263 res = (resync>>scale)*1000;
dd71cf6b 7264 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 7265
7266 per_milli = res;
1da177e4 7267 {
4588b42e 7268 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 7269 seq_printf(seq, "[");
7270 for (i = 0; i < x; i++)
7271 seq_printf(seq, "=");
7272 seq_printf(seq, ">");
7273 for (i = 0; i < y; i++)
7274 seq_printf(seq, ".");
7275 seq_printf(seq, "] ");
7276 }
4588b42e 7277 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 7278 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7279 "reshape" :
61df9d91
N		 7280 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7281 "check" :
7282 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7283 "resync" : "recovery"))),
7284 per_milli/10, per_milli % 10,
dd71cf6b
N		 7285 (unsigned long long) resync/2,
7286 (unsigned long long) max_sectors/2);
1da177e4
LT		 7287
7288 /*
1da177e4
LT		 7289 * dt: time from mark until now
7290 * db: blocks written from mark until now
7291 * rt: remaining time
dd71cf6b
N		 7292 *
7293 * rt is a sector_t, so could be 32bit or 64bit.
7294 * So we divide before multiply in case it is 32bit and close
7295 * to the limit.
25985edc 7296 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 7297 * near the end of resync when the number of remaining sectors
7298 * is close to 'db'.
7299 * We then divide rt by 32 after multiplying by db to compensate.
7300 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 7301 */
7302 dt = ((jiffies - mddev->resync_mark) / HZ);
7303 if (!dt) dt++;
ff4e8d9a
N		 7304 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7305 - mddev->resync_mark_cnt;
1da177e4 7306
dd71cf6b
N		 7307 rt = max_sectors - resync; /* number of remaining sectors */
7308 sector_div(rt, db/32+1);
7309 rt *= dt;
7310 rt >>= 5;
7311
7312 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7313 ((unsigned long)rt % 60)/6);
1da177e4 7314
ff4e8d9a 7315 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
f7851be7 7316 return 1;
1da177e4
LT		 7317}
7318
7319static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7320{
7321 struct list_head *tmp;
7322 loff_t l = *pos;
fd01b88c 7323 struct mddev *mddev;
1da177e4
LT		 7324
7325 if (l >= 0x10000)
7326 return NULL;
7327 if (!l--)
7328 /* header */
7329 return (void*)1;
7330
7331 spin_lock(&all_mddevs_lock);
7332 list_for_each(tmp,&all_mddevs)
7333 if (!l--) {
fd01b88c 7334 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 7335 mddev_get(mddev);
7336 spin_unlock(&all_mddevs_lock);
7337 return mddev;
7338 }
7339 spin_unlock(&all_mddevs_lock);
7340 if (!l--)
7341 return (void*)2;/* tail */
7342 return NULL;
7343}
7344
7345static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7346{
7347 struct list_head *tmp;
fd01b88c 7348 struct mddev *next_mddev, *mddev = v;
f72ffdd6 7349
1da177e4
LT		 7350 ++*pos;
7351 if (v == (void*)2)
7352 return NULL;
7353
7354 spin_lock(&all_mddevs_lock);
7355 if (v == (void*)1)
7356 tmp = all_mddevs.next;
7357 else
7358 tmp = mddev->all_mddevs.next;
7359 if (tmp != &all_mddevs)
fd01b88c 7360 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 7361 else {
7362 next_mddev = (void*)2;
7363 *pos = 0x10000;
f72ffdd6 7364 }
1da177e4
LT		 7365 spin_unlock(&all_mddevs_lock);
7366
7367 if (v != (void*)1)
7368 mddev_put(mddev);
7369 return next_mddev;
7370
7371}
7372
7373static void md_seq_stop(struct seq_file *seq, void *v)
7374{
fd01b88c 7375 struct mddev *mddev = v;
1da177e4
LT		 7376
7377 if (mddev && v != (void*)1 && v != (void*)2)
7378 mddev_put(mddev);
7379}
7380
7381static int md_seq_show(struct seq_file *seq, void *v)
7382{
fd01b88c 7383 struct mddev *mddev = v;
dd8ac336 7384 sector_t sectors;
3cb03002 7385 struct md_rdev *rdev;
1da177e4
LT		 7386
7387 if (v == (void*)1) {
84fc4b56 7388 struct md_personality *pers;
1da177e4
LT		 7389 seq_printf(seq, "Personalities : ");
7390 spin_lock(&pers_lock);
2604b703
N		 7391 list_for_each_entry(pers, &pers_list, list)
7392 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 7393
7394 spin_unlock(&pers_lock);
7395 seq_printf(seq, "\n");
f1514638 7396 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7397 return 0;
7398 }
7399 if (v == (void*)2) {
7400 status_unused(seq);
7401 return 0;
7402 }
7403
36d091f4 7404 spin_lock(&mddev->lock);
1da177e4
LT		 7405 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7406 seq_printf(seq, "%s : %sactive", mdname(mddev),
7407 mddev->pers ? "" : "in");
7408 if (mddev->pers) {
f91de92e 7409 if (mddev->ro==1)
1da177e4 7410 seq_printf(seq, " (read-only)");
f91de92e 7411 if (mddev->ro==2)
52720ae7 7412 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 7413 seq_printf(seq, " %s", mddev->pers->name);
7414 }
7415
dd8ac336 7416 sectors = 0;
f97fcad3
N		 7417 rcu_read_lock();
7418 rdev_for_each_rcu(rdev, mddev) {
1da177e4
LT		 7419 char b[BDEVNAME_SIZE];
7420 seq_printf(seq, " %s[%d]",
7421 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 7422 if (test_bit(WriteMostly, &rdev->flags))
7423 seq_printf(seq, "(W)");
9efdca16
SL		 7424 if (test_bit(Journal, &rdev->flags))
7425 seq_printf(seq, "(J)");
b2d444d7 7426 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7427 seq_printf(seq, "(F)");
7428 continue;
2d78f8c4
N		 7429 }
7430 if (rdev->raid_disk < 0)
b325a32e 7431 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 7432 if (test_bit(Replacement, &rdev->flags))
7433 seq_printf(seq, "(R)");
dd8ac336 7434 sectors += rdev->sectors;
1da177e4 7435 }
f97fcad3 7436 rcu_read_unlock();
1da177e4
LT		 7437
7438 if (!list_empty(&mddev->disks)) {
7439 if (mddev->pers)
7440 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 7441 (unsigned long long)
7442 mddev->array_sectors / 2);
1da177e4
LT		 7443 else
7444 seq_printf(seq, "\n %llu blocks",
dd8ac336 7445 (unsigned long long)sectors / 2);
1da177e4 7446 }
1cd6bf19
N		 7447 if (mddev->persistent) {
7448 if (mddev->major_version != 0 ||
7449 mddev->minor_version != 90) {
7450 seq_printf(seq," super %d.%d",
7451 mddev->major_version,
7452 mddev->minor_version);
7453 }
e691063a
N		 7454 } else if (mddev->external)
7455 seq_printf(seq, " super external:%s",
7456 mddev->metadata_type);
7457 else
1cd6bf19 7458 seq_printf(seq, " super non-persistent");
1da177e4
LT		 7459
7460 if (mddev->pers) {
d710e138 7461 mddev->pers->status(seq, mddev);
f72ffdd6 7462 seq_printf(seq, "\n ");
8e1b39d6 7463 if (mddev->pers->sync_request) {
f7851be7 7464 if (status_resync(seq, mddev))
8e1b39d6 7465 seq_printf(seq, "\n ");
8e1b39d6 7466 }
32a7627c
N		 7467 } else
7468 seq_printf(seq, "\n ");
7469
57148964 7470 bitmap_status(seq, mddev->bitmap);
1da177e4
LT		 7471
7472 seq_printf(seq, "\n");
7473 }
36d091f4 7474 spin_unlock(&mddev->lock);
f72ffdd6 7475
1da177e4
LT		 7476 return 0;
7477}
7478
110518bc 7479static const struct seq_operations md_seq_ops = {
1da177e4
LT		 7480 .start = md_seq_start,
7481 .next = md_seq_next,
7482 .stop = md_seq_stop,
7483 .show = md_seq_show,
7484};
7485
7486static int md_seq_open(struct inode *inode, struct file *file)
7487{
f1514638 7488 struct seq_file *seq;
1da177e4
LT		 7489 int error;
7490
7491 error = seq_open(file, &md_seq_ops);
d7603b7e 7492 if (error)
f1514638
KS		 7493 return error;
7494
7495 seq = file->private_data;
7496 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7497 return error;
7498}
7499
e2f23b60 7500static int md_unloading;
d7603b7e
N		 7501static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7502{
f1514638 7503 struct seq_file *seq = filp->private_data;
d7603b7e
N		 7504 int mask;
7505
e2f23b60 7506 if (md_unloading)
7d7e64f2 7507 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
d7603b7e
N		 7508 poll_wait(filp, &md_event_waiters, wait);
7509
7510 /* always allow read */
7511 mask = POLLIN | POLLRDNORM;
7512
f1514638 7513 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 7514 mask |= POLLERR | POLLPRI;
7515 return mask;
7516}
7517
fa027c2a 7518static const struct file_operations md_seq_fops = {
e24650c2 7519 .owner = THIS_MODULE,
1da177e4
LT		 7520 .open = md_seq_open,
7521 .read = seq_read,
7522 .llseek = seq_lseek,
c3f94b40 7523 .release = seq_release_private,
d7603b7e 7524 .poll = mdstat_poll,
1da177e4
LT		 7525};
7526
84fc4b56 7527int register_md_personality(struct md_personality *p)
1da177e4 7528{
50bd3774
CY		 7529 printk(KERN_INFO "md: %s personality registered for level %d\n",
7530 p->name, p->level);
1da177e4 7531 spin_lock(&pers_lock);
2604b703 7532 list_add_tail(&p->list, &pers_list);
1da177e4
LT		 7533 spin_unlock(&pers_lock);
7534 return 0;
7535}
6c144d31 7536EXPORT_SYMBOL(register_md_personality);
1da177e4 7537
84fc4b56 7538int unregister_md_personality(struct md_personality *p)
1da177e4 7539{
2604b703 7540 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7541 spin_lock(&pers_lock);
2604b703 7542 list_del_init(&p->list);
1da177e4
LT		 7543 spin_unlock(&pers_lock);
7544 return 0;
7545}
6c144d31 7546EXPORT_SYMBOL(unregister_md_personality);
1da177e4 7547
6022e75b
N		 7548int register_md_cluster_operations(struct md_cluster_operations *ops,
7549 struct module *module)
edb39c9d 7550{
6022e75b 7551 int ret = 0;
edb39c9d 7552 spin_lock(&pers_lock);
6022e75b
N		 7553 if (md_cluster_ops != NULL)
7554 ret = -EALREADY;
7555 else {
7556 md_cluster_ops = ops;
7557 md_cluster_mod = module;
7558 }
edb39c9d 7559 spin_unlock(&pers_lock);
6022e75b 7560 return ret;
edb39c9d
GR		 7561}
7562EXPORT_SYMBOL(register_md_cluster_operations);
7563
7564int unregister_md_cluster_operations(void)
7565{
7566 spin_lock(&pers_lock);
7567 md_cluster_ops = NULL;
7568 spin_unlock(&pers_lock);
7569 return 0;
7570}
7571EXPORT_SYMBOL(unregister_md_cluster_operations);
7572
7573int md_setup_cluster(struct mddev *mddev, int nodes)
7574{
7575 int err;
7576
7577 err = request_module("md-cluster");
7578 if (err) {
7579 pr_err("md-cluster module not found.\n");
b0c26a79 7580 return -ENOENT;
edb39c9d
GR		 7581 }
7582
7583 spin_lock(&pers_lock);
7584 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7585 spin_unlock(&pers_lock);
7586 return -ENOENT;
7587 }
7588 spin_unlock(&pers_lock);
7589
cf921cc1 7590 return md_cluster_ops->join(mddev, nodes);
edb39c9d
GR		 7591}
7592
7593void md_cluster_stop(struct mddev *mddev)
7594{
c4ce867f
GR		 7595 if (!md_cluster_ops)
7596 return;
edb39c9d
GR		 7597 md_cluster_ops->leave(mddev);
7598 module_put(md_cluster_mod);
7599}
7600
fd01b88c 7601static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7602{
f72ffdd6 7603 struct md_rdev *rdev;
1da177e4 7604 int idle;
eea1bf38 7605 int curr_events;
1da177e4
LT		 7606
7607 idle = 1;
4b80991c
N		 7608 rcu_read_lock();
7609 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7610 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 7611 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7612 (int)part_stat_read(&disk->part0, sectors[1]) -
7613 atomic_read(&disk->sync_io);
713f6ab1
N		 7614 /* sync IO will cause sync_io to increase before the disk_stats
7615 * as sync_io is counted when a request starts, and
7616 * disk_stats is counted when it completes.
7617 * So resync activity will cause curr_events to be smaller than
7618 * when there was no such activity.
7619 * non-sync IO will cause disk_stat to increase without
7620 * increasing sync_io so curr_events will (eventually)
7621 * be larger than it was before. Once it becomes
7622 * substantially larger, the test below will cause
7623 * the array to appear non-idle, and resync will slow
7624 * down.
7625 * If there is a lot of outstanding resync activity when
7626 * we set last_event to curr_events, then all that activity
7627 * completing might cause the array to appear non-idle
7628 * and resync will be slowed down even though there might
7629 * not have been non-resync activity. This will only
7630 * happen once though. 'last_events' will soon reflect
7631 * the state where there is little or no outstanding
7632 * resync requests, and further resync activity will
7633 * always make curr_events less than last_events.
c0e48521 7634 *
1da177e4 7635 */
eea1bf38 7636 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 7637 rdev->last_events = curr_events;
7638 idle = 0;
7639 }
7640 }
4b80991c 7641 rcu_read_unlock();
1da177e4
LT		 7642 return idle;
7643}
7644
fd01b88c 7645void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 7646{
7647 /* another "blocks" (512byte) blocks have been synced */
7648 atomic_sub(blocks, &mddev->recovery_active);
7649 wake_up(&mddev->recovery_wait);
7650 if (!ok) {
dfc70645 7651 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 7652 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT		 7653 md_wakeup_thread(mddev->thread);
7654 // stop recovery, signal do_sync ....
7655 }
7656}
6c144d31 7657EXPORT_SYMBOL(md_done_sync);
1da177e4 7658
06d91a5f
N		 7659/* md_write_start(mddev, bi)
7660 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 7661 * in superblock) before writing, schedule a superblock update
7662 * and wait for it to complete.
06d91a5f 7663 */
fd01b88c 7664void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7665{
0fd62b86 7666 int did_change = 0;
06d91a5f 7667 if (bio_data_dir(bi) != WRITE)
3d310eb7 7668 return;
06d91a5f 7669
f91de92e
N		 7670 BUG_ON(mddev->ro == 1);
7671 if (mddev->ro == 2) {
7672 /* need to switch to read/write */
7673 mddev->ro = 0;
7674 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7675 md_wakeup_thread(mddev->thread);
25156198 7676 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7677 did_change = 1;
f91de92e 7678 }
06d91a5f 7679 atomic_inc(&mddev->writes_pending);
31a59e34
N		 7680 if (mddev->safemode == 1)
7681 mddev->safemode = 0;
06d91a5f 7682 if (mddev->in_sync) {
85572d7c 7683 spin_lock(&mddev->lock);
3d310eb7
N		 7684 if (mddev->in_sync) {
7685 mddev->in_sync = 0;
850b2b42 7686 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7687 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7688 md_wakeup_thread(mddev->thread);
0fd62b86 7689 did_change = 1;
3d310eb7 7690 }
85572d7c 7691 spin_unlock(&mddev->lock);
06d91a5f 7692 }
0fd62b86 7693 if (did_change)
00bcb4ac 7694 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7695 wait_event(mddev->sb_wait,
09a44cc1 7696 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4 7697}
6c144d31 7698EXPORT_SYMBOL(md_write_start);
1da177e4 7699
fd01b88c 7700void md_write_end(struct mddev *mddev)
1da177e4
LT		 7701{
7702 if (atomic_dec_and_test(&mddev->writes_pending)) {
7703 if (mddev->safemode == 2)
7704 md_wakeup_thread(mddev->thread);
16f17b39 7705 else if (mddev->safemode_delay)
1da177e4
LT		 7706 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7707 }
7708}
6c144d31 7709EXPORT_SYMBOL(md_write_end);
1da177e4 7710
2a2275d6
N		 7711/* md_allow_write(mddev)
7712 * Calling this ensures that the array is marked 'active' so that writes
7713 * may proceed without blocking. It is important to call this before
7714 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7715 * Must be called with mddev_lock held.
b5470dc5
DW		 7716 *
7717 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7718 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7719 */
fd01b88c 7720int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7721{
7722 if (!mddev->pers)
b5470dc5 7723 return 0;
2a2275d6 7724 if (mddev->ro)
b5470dc5 7725 return 0;
1a0fd497 7726 if (!mddev->pers->sync_request)
b5470dc5 7727 return 0;
2a2275d6 7728
85572d7c 7729 spin_lock(&mddev->lock);
2a2275d6
N		 7730 if (mddev->in_sync) {
7731 mddev->in_sync = 0;
7732 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7733 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7734 if (mddev->safemode_delay &&
7735 mddev->safemode == 0)
7736 mddev->safemode = 1;
85572d7c 7737 spin_unlock(&mddev->lock);
2a2275d6 7738 md_update_sb(mddev, 0);
00bcb4ac 7739 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6 7740 } else
85572d7c 7741 spin_unlock(&mddev->lock);
b5470dc5 7742
070dc6dd 7743 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7744 return -EAGAIN;
7745 else
7746 return 0;
2a2275d6
N		 7747}
7748EXPORT_SYMBOL_GPL(md_allow_write);
7749
1da177e4
LT		 7750#define SYNC_MARKS 10
7751#define SYNC_MARK_STEP (3*HZ)
54f89341 7752#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 7753void md_do_sync(struct md_thread *thread)
1da177e4 7754{
4ed8731d 7755 struct mddev *mddev = thread->mddev;
fd01b88c 7756 struct mddev *mddev2;
1da177e4
LT		 7757 unsigned int currspeed = 0,
7758 window;
ac7e50a3 7759 sector_t max_sectors,j, io_sectors, recovery_done;
1da177e4 7760 unsigned long mark[SYNC_MARKS];
54f89341 7761 unsigned long update_time;
1da177e4
LT		 7762 sector_t mark_cnt[SYNC_MARKS];
7763 int last_mark,m;
7764 struct list_head *tmp;
7765 sector_t last_check;
57afd89f 7766 int skipped = 0;
3cb03002 7767 struct md_rdev *rdev;
c4a39551 7768 char *desc, *action = NULL;
7c2c57c9 7769 struct blk_plug plug;
c186b128 7770 bool cluster_resync_finished = false;
1da177e4
LT		 7771
7772 /* just incase thread restarts... */
7773 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7774 return;
3991b31e
N		 7775 if (mddev->ro) {/* never try to sync a read-only array */
7776 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5fd6c1dc 7777 return;
3991b31e 7778 }
1da177e4 7779
61df9d91 7780 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
c4a39551 7781 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
61df9d91 7782 desc = "data-check";
c4a39551
JB		 7783 action = "check";
7784 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
61df9d91 7785 desc = "requested-resync";
c4a39551
JB		 7786 action = "repair";
7787 } else
61df9d91
N		 7788 desc = "resync";
7789 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7790 desc = "reshape";
7791 else
7792 desc = "recovery";
7793
c4a39551
JB		 7794 mddev->last_sync_action = action ?: desc;
7795
1da177e4
LT		 7796 /* we overload curr_resync somewhat here.
7797 * 0 == not engaged in resync at all
7798 * 2 == checking that there is no conflict with another sync
7799 * 1 == like 2, but have yielded to allow conflicting resync to
7800 * commense
7801 * other == active in resync - this many blocks
7802 *
7803 * Before starting a resync we must have set curr_resync to
7804 * 2, and then checked that every "conflicting" array has curr_resync
7805 * less than ours. When we find one that is the same or higher
7806 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7807 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7808 * This will mean we have to start checking from the beginning again.
7809 *
7810 */
7811
7812 do {
7813 mddev->curr_resync = 2;
7814
7815 try_again:
404e4b43 7816 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7817 goto skip;
29ac4aa3 7818 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7819 if (mddev2 == mddev)
7820 continue;
90b08710
BS		 7821 if (!mddev->parallel_resync
7822 && mddev2->curr_resync
7823 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7824 DEFINE_WAIT(wq);
7825 if (mddev < mddev2 && mddev->curr_resync == 2) {
7826 /* arbitrarily yield */
7827 mddev->curr_resync = 1;
7828 wake_up(&resync_wait);
7829 }
7830 if (mddev > mddev2 && mddev->curr_resync == 1)
7831 /* no need to wait here, we can wait the next
7832 * time 'round when curr_resync == 2
7833 */
7834 continue;
9744197c
N		 7835 /* We need to wait 'interruptible' so as not to
7836 * contribute to the load average, and not to
7837 * be caught by 'softlockup'
7838 */
7839 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
c91abf5a 7840 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8712e553 7841 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7842 printk(KERN_INFO "md: delaying %s of %s"
7843 " until %s has finished (they"
1da177e4 7844 " share one or more physical units)\n",
61df9d91 7845 desc, mdname(mddev), mdname(mddev2));
1da177e4 7846 mddev_put(mddev2);
9744197c
N		 7847 if (signal_pending(current))
7848 flush_signals(current);
1da177e4
LT		 7849 schedule();
7850 finish_wait(&resync_wait, &wq);
7851 goto try_again;
7852 }
7853 finish_wait(&resync_wait, &wq);
7854 }
7855 }
7856 } while (mddev->curr_resync < 2);
7857
5fd6c1dc 7858 j = 0;
9d88883e 7859 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7860 /* resync follows the size requested by the personality,
57afd89f 7861 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7862 */
7863 max_sectors = mddev->resync_max_sectors;
7f7583d4 7864 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 7865 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7866 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7867 j = mddev->resync_min;
7868 else if (!mddev->bitmap)
5fd6c1dc 7869 j = mddev->recovery_cp;
5e96ee65 7870
ccfcc3c1 7871 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7872 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7873 else {
1da177e4 7874 /* recovery follows the physical size of devices */
58c0fed4 7875 max_sectors = mddev->dev_sectors;
5fd6c1dc 7876 j = MaxSector;
4e59ca7d 7877 rcu_read_lock();
dafb20fa 7878 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7879 if (rdev->raid_disk >= 0 &&
f2076e7d 7880 !test_bit(Journal, &rdev->flags) &&
5fd6c1dc
N		 7881 !test_bit(Faulty, &rdev->flags) &&
7882 !test_bit(In_sync, &rdev->flags) &&
7883 rdev->recovery_offset < j)
7884 j = rdev->recovery_offset;
4e59ca7d 7885 rcu_read_unlock();
133d4527
N		 7886
7887 /* If there is a bitmap, we need to make sure all
7888 * writes that started before we added a spare
7889 * complete before we start doing a recovery.
7890 * Otherwise the write might complete and (via
7891 * bitmap_endwrite) set a bit in the bitmap after the
7892 * recovery has checked that bit and skipped that
7893 * region.
7894 */
7895 if (mddev->bitmap) {
7896 mddev->pers->quiesce(mddev, 1);
7897 mddev->pers->quiesce(mddev, 0);
7898 }
5fd6c1dc 7899 }
1da177e4 7900
61df9d91
N		 7901 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7902 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7903 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7904 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7905 "(but not more than %d KB/sec) for %s.\n",
7906 speed_max(mddev), desc);
1da177e4 7907
eea1bf38 7908 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7909
57afd89f 7910 io_sectors = 0;
1da177e4
LT		 7911 for (m = 0; m < SYNC_MARKS; m++) {
7912 mark[m] = jiffies;
57afd89f 7913 mark_cnt[m] = io_sectors;
1da177e4
LT		 7914 }
7915 last_mark = 0;
7916 mddev->resync_mark = mark[last_mark];
7917 mddev->resync_mark_cnt = mark_cnt[last_mark];
7918
7919 /*
7920 * Tune reconstruction:
7921 */
7922 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7923 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7924 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7925
7926 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7927 last_check = 0;
7928
7929 if (j>2) {
c91abf5a 7930 printk(KERN_INFO
61df9d91
N		 7931 "md: resuming %s of %s from checkpoint.\n",
7932 desc, mdname(mddev));
1da177e4 7933 mddev->curr_resync = j;
72f36d59
N		 7934 } else
7935 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 7936 mddev->curr_resync_completed = j;
72f36d59
N		 7937 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7938 md_new_event(mddev);
54f89341 7939 update_time = jiffies;
1da177e4 7940
7c2c57c9 7941 blk_start_plug(&plug);
1da177e4 7942 while (j < max_sectors) {
57afd89f 7943 sector_t sectors;
1da177e4 7944
57afd89f 7945 skipped = 0;
97e4f42d 7946
7a91ee1f
N		 7947 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7948 ((mddev->curr_resync > mddev->curr_resync_completed &&
7949 (mddev->curr_resync - mddev->curr_resync_completed)
7950 > (max_sectors >> 4)) ||
54f89341 7951 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f 7952 (j - mddev->curr_resync_completed)*2
c5e19d90
N		 7953 >= mddev->resync_max - mddev->curr_resync_completed ||
7954 mddev->curr_resync_completed > mddev->resync_max
7a91ee1f 7955 )) {
97e4f42d 7956 /* time to update curr_resync_completed */
97e4f42d
N		 7957 wait_event(mddev->recovery_wait,
7958 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7959 mddev->curr_resync_completed = j;
35d78c66 7960 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7961 j > mddev->recovery_cp)
7962 mddev->recovery_cp = j;
54f89341 7963 update_time = jiffies;
070dc6dd 7964 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7965 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7966 }
acb180b0 7967
c91abf5a
N		 7968 while (j >= mddev->resync_max &&
7969 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
e62e58a5
N		 7970 /* As this condition is controlled by user-space,
7971 * we can block indefinitely, so use '_interruptible'
7972 * to avoid triggering warnings.
7973 */
7974 flush_signals(current); /* just in case */
7975 wait_event_interruptible(mddev->recovery_wait,
7976 mddev->resync_max > j
c91abf5a
N		 7977 || test_bit(MD_RECOVERY_INTR,
7978 &mddev->recovery));
e62e58a5 7979 }
acb180b0 7980
c91abf5a
N		 7981 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7982 break;
acb180b0 7983
09314799 7984 sectors = mddev->pers->sync_request(mddev, j, &skipped);
57afd89f 7985 if (sectors == 0) {
dfc70645 7986 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
c91abf5a 7987 break;
1da177e4 7988 }
57afd89f
N		 7989
7990 if (!skipped) { /* actual IO requested */
7991 io_sectors += sectors;
7992 atomic_add(sectors, &mddev->recovery_active);
7993 }
7994
e875ecea
N		 7995 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7996 break;
7997
1da177e4 7998 j += sectors;
5ed1df2e
N		 7999 if (j > max_sectors)
8000 /* when skipping, extra large numbers can be returned. */
8001 j = max_sectors;
72f36d59
N		 8002 if (j > 2)
8003 mddev->curr_resync = j;
ff4e8d9a 8004 mddev->curr_mark_cnt = io_sectors;
d7603b7e 8005 if (last_check == 0)
e875ecea 8006 /* this is the earliest that rebuild will be
d7603b7e
N		 8007 * visible in /proc/mdstat
8008 */
8009 md_new_event(mddev);
57afd89f
N		 8010
8011 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 8012 continue;
8013
57afd89f 8014 last_check = io_sectors;
1da177e4
LT		 8015 repeat:
8016 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8017 /* step marks */
8018 int next = (last_mark+1) % SYNC_MARKS;
8019
8020 mddev->resync_mark = mark[next];
8021 mddev->resync_mark_cnt = mark_cnt[next];
8022 mark[next] = jiffies;
57afd89f 8023 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 8024 last_mark = next;
8025 }
8026
c91abf5a
N		 8027 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8028 break;
1da177e4
LT		 8029
8030 /*
8031 * this loop exits only if either when we are slower than
8032 * the 'hard' speed limit, or the system was IO-idle for
8033 * a jiffy.
8034 * the system might be non-idle CPU-wise, but we only care
8035 * about not overloading the IO subsystem. (things like an
8036 * e2fsck being done on the RAID array should execute fast)
8037 */
1da177e4
LT		 8038 cond_resched();
8039
ac7e50a3
XN		 8040 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8041 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
57afd89f 8042 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 8043
88202a0c 8044 if (currspeed > speed_min(mddev)) {
ac8fa419 8045 if (currspeed > speed_max(mddev)) {
c0e48521 8046 msleep(500);
1da177e4
LT		 8047 goto repeat;
8048 }
ac8fa419
N		 8049 if (!is_mddev_idle(mddev, 0)) {
8050 /*
8051 * Give other IO more of a chance.
8052 * The faster the devices, the less we wait.
8053 */
8054 wait_event(mddev->recovery_wait,
8055 !atomic_read(&mddev->recovery_active));
8056 }
1da177e4
LT		 8057 }
8058 }
c91abf5a
N		 8059 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8060 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8061 ? "interrupted" : "done");
1da177e4
LT		 8062 /*
8063 * this also signals 'finished resyncing' to md_stop
8064 */
7c2c57c9 8065 blk_finish_plug(&plug);
1da177e4
LT		 8066 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8067
5ed1df2e
N		 8068 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8069 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8070 mddev->curr_resync > 2) {
8071 mddev->curr_resync_completed = mddev->curr_resync;
8072 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8073 }
c186b128
GR		 8074 /* tell personality and other nodes that we are finished */
8075 if (mddev_is_clustered(mddev)) {
8076 md_cluster_ops->resync_finish(mddev);
8077 cluster_resync_finished = true;
8078 }
09314799 8079 mddev->pers->sync_request(mddev, max_sectors, &skipped);
1da177e4 8080
dfc70645 8081 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 8082 mddev->curr_resync > 2) {
8083 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8084 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8085 if (mddev->curr_resync >= mddev->recovery_cp) {
8086 printk(KERN_INFO
61df9d91
N		 8087 "md: checkpointing %s of %s.\n",
8088 desc, mdname(mddev));
0a19caab 8089 if (test_bit(MD_RECOVERY_ERROR,
8090 &mddev->recovery))
8091 mddev->recovery_cp =
8092 mddev->curr_resync_completed;
8093 else
8094 mddev->recovery_cp =
8095 mddev->curr_resync;
5fd6c1dc
N		 8096 }
8097 } else
8098 mddev->recovery_cp = MaxSector;
8099 } else {
8100 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8101 mddev->curr_resync = MaxSector;
4e59ca7d 8102 rcu_read_lock();
dafb20fa 8103 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 8104 if (rdev->raid_disk >= 0 &&
70fffd0b 8105 mddev->delta_disks >= 0 &&
f2076e7d 8106 !test_bit(Journal, &rdev->flags) &&
5fd6c1dc
N		 8107 !test_bit(Faulty, &rdev->flags) &&
8108 !test_bit(In_sync, &rdev->flags) &&
8109 rdev->recovery_offset < mddev->curr_resync)
8110 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 8111 rcu_read_unlock();
5fd6c1dc 8112 }
1da177e4 8113 }
db91ff55 8114 skip:
17571284 8115 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 8116
c186b128
GR		 8117 if (mddev_is_clustered(mddev) &&
8118 test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8119 !cluster_resync_finished)
8120 md_cluster_ops->resync_finish(mddev);
8121
23da422b 8122 spin_lock(&mddev->lock);
c07b70ad
N		 8123 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8124 /* We completed so min/max setting can be forgotten if used. */
8125 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8126 mddev->resync_min = 0;
8127 mddev->resync_max = MaxSector;
8128 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8129 mddev->resync_min = mddev->curr_resync_completed;
f7851be7 8130 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 8131 mddev->curr_resync = 0;
23da422b
N		 8132 spin_unlock(&mddev->lock);
8133
1da177e4 8134 wake_up(&resync_wait);
1da177e4 8135 md_wakeup_thread(mddev->thread);
c6207277 8136 return;
1da177e4 8137}
29269553 8138EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 8139
746d3207
N		 8140static int remove_and_add_spares(struct mddev *mddev,
8141 struct md_rdev *this)
b4c4c7b8 8142{
3cb03002 8143 struct md_rdev *rdev;
b4c4c7b8 8144 int spares = 0;
f2a371c5 8145 int removed = 0;
b4c4c7b8 8146
dafb20fa 8147 rdev_for_each(rdev, mddev)
746d3207
N		 8148 if ((this == NULL || rdev == this) &&
8149 rdev->raid_disk >= 0 &&
6bfe0b49 8150 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8 8151 (test_bit(Faulty, &rdev->flags) ||
f2076e7d
SL		 8152 (!test_bit(In_sync, &rdev->flags) &&
8153 !test_bit(Journal, &rdev->flags))) &&
b4c4c7b8
N		 8154 atomic_read(&rdev->nr_pending)==0) {
8155 if (mddev->pers->hot_remove_disk(
b8321b68 8156 mddev, rdev) == 0) {
36fad858 8157 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 8158 rdev->raid_disk = -1;
f2a371c5 8159 removed++;
b4c4c7b8
N		 8160 }
8161 }
90584fc9
JB		 8162 if (removed && mddev->kobj.sd)
8163 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 8164
2910ff17 8165 if (this && removed)
746d3207
N		 8166 goto no_add;
8167
dafb20fa 8168 rdev_for_each(rdev, mddev) {
2910ff17
GR		 8169 if (this && this != rdev)
8170 continue;
dbb64f86
GR		 8171 if (test_bit(Candidate, &rdev->flags))
8172 continue;
7bfec5f3
N		 8173 if (rdev->raid_disk >= 0 &&
8174 !test_bit(In_sync, &rdev->flags) &&
f2076e7d 8175 !test_bit(Journal, &rdev->flags) &&
7bfec5f3
N		 8176 !test_bit(Faulty, &rdev->flags))
8177 spares++;
7ceb17e8
N		 8178 if (rdev->raid_disk >= 0)
8179 continue;
8180 if (test_bit(Faulty, &rdev->flags))
8181 continue;
3069aa8d
SL		 8182 if (test_bit(Journal, &rdev->flags))
8183 continue;
7ceb17e8 8184 if (mddev->ro &&
8313b8e5
N		 8185 ! (rdev->saved_raid_disk >= 0 &&
8186 !test_bit(Bitmap_sync, &rdev->flags)))
7ceb17e8
N		 8187 continue;
8188
d01552a7 8189 rdev->recovery_offset = 0;
7ceb17e8
N		 8190 if (mddev->pers->
8191 hot_add_disk(mddev, rdev) == 0) {
8192 if (sysfs_link_rdev(mddev, rdev))
8193 /* failure here is OK */;
8194 spares++;
8195 md_new_event(mddev);
8196 set_bit(MD_CHANGE_DEVS, &mddev->flags);
dfc70645 8197 }
b4c4c7b8 8198 }
746d3207 8199no_add:
6dafab6b
N		 8200 if (removed)
8201 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 8202 return spares;
8203}
7ebc0be7 8204
ac05f256
N		 8205static void md_start_sync(struct work_struct *ws)
8206{
8207 struct mddev *mddev = container_of(ws, struct mddev, del_work);
c186b128
GR		 8208 int ret = 0;
8209
8210 if (mddev_is_clustered(mddev)) {
8211 ret = md_cluster_ops->resync_start(mddev);
8212 if (ret) {
8213 mddev->sync_thread = NULL;
8214 goto out;
8215 }
8216 }
ac05f256
N		 8217
8218 mddev->sync_thread = md_register_thread(md_do_sync,
8219 mddev,
8220 "resync");
c186b128 8221out:
ac05f256 8222 if (!mddev->sync_thread) {
c186b128
GR		 8223 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8224 printk(KERN_ERR "%s: could not start resync"
8225 " thread...\n",
8226 mdname(mddev));
ac05f256
N		 8227 /* leave the spares where they are, it shouldn't hurt */
8228 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8229 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8230 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8231 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8232 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8233 wake_up(&resync_wait);
ac05f256
N		 8234 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8235 &mddev->recovery))
8236 if (mddev->sysfs_action)
8237 sysfs_notify_dirent_safe(mddev->sysfs_action);
8238 } else
8239 md_wakeup_thread(mddev->sync_thread);
8240 sysfs_notify_dirent_safe(mddev->sysfs_action);
8241 md_new_event(mddev);
8242}
8243
1da177e4
LT		 8244/*
8245 * This routine is regularly called by all per-raid-array threads to
8246 * deal with generic issues like resync and super-block update.
8247 * Raid personalities that don't have a thread (linear/raid0) do not
8248 * need this as they never do any recovery or update the superblock.
8249 *
8250 * It does not do any resync itself, but rather "forks" off other threads
8251 * to do that as needed.
8252 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8253 * "->recovery" and create a thread at ->sync_thread.
dfc70645 8254 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 8255 * and wakeups up this thread which will reap the thread and finish up.
8256 * This thread also removes any faulty devices (with nr_pending == 0).
8257 *
8258 * The overall approach is:
8259 * 1/ if the superblock needs updating, update it.
8260 * 2/ If a recovery thread is running, don't do anything else.
8261 * 3/ If recovery has finished, clean up, possibly marking spares active.
8262 * 4/ If there are any faulty devices, remove them.
8263 * 5/ If array is degraded, try to add spares devices
8264 * 6/ If array has spares or is not in-sync, start a resync thread.
8265 */
fd01b88c 8266void md_check_recovery(struct mddev *mddev)
1da177e4 8267{
68866e42
JB		 8268 if (mddev->suspended)
8269 return;
8270
5f40402d 8271 if (mddev->bitmap)
aa5cbd10 8272 bitmap_daemon_work(mddev);
1da177e4 8273
fca4d848 8274 if (signal_pending(current)) {
31a59e34 8275 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 8276 printk(KERN_INFO "md: %s in immediate safe mode\n",
8277 mdname(mddev));
8278 mddev->safemode = 2;
8279 }
8280 flush_signals(current);
8281 }
8282
c89a8eee
N		 8283 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8284 return;
1da177e4 8285 if ( ! (
142d44c3 8286 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 8287 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 8288 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 8289 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 8290 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8291 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 8292 ))
8293 return;
fca4d848 8294
df5b89b3 8295 if (mddev_trylock(mddev)) {
b4c4c7b8 8296 int spares = 0;
fca4d848 8297
c89a8eee 8298 if (mddev->ro) {
ab16bfc7
NB		 8299 struct md_rdev *rdev;
8300 if (!mddev->external && mddev->in_sync)
8301 /* 'Blocked' flag not needed as failed devices
8302 * will be recorded if array switched to read/write.
8303 * Leaving it set will prevent the device
8304 * from being removed.
8305 */
8306 rdev_for_each(rdev, mddev)
8307 clear_bit(Blocked, &rdev->flags);
7ceb17e8
N		 8308 /* On a read-only array we can:
8309 * - remove failed devices
8310 * - add already-in_sync devices if the array itself
8311 * is in-sync.
8312 * As we only add devices that are already in-sync,
8313 * we can activate the spares immediately.
c89a8eee 8314 */
7ceb17e8 8315 remove_and_add_spares(mddev, NULL);
8313b8e5
N		 8316 /* There is no thread, but we need to call
8317 * ->spare_active and clear saved_raid_disk
8318 */
2ac295a5 8319 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8313b8e5 8320 md_reap_sync_thread(mddev);
a4a3d26d 8321 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8313b8e5 8322 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d4929add 8323 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
c89a8eee
N		 8324 goto unlock;
8325 }
8326
659b254f
GJ		 8327 if (mddev_is_clustered(mddev)) {
8328 struct md_rdev *rdev;
8329 /* kick the device if another node issued a
8330 * remove disk.
8331 */
8332 rdev_for_each(rdev, mddev) {
8333 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8334 rdev->raid_disk < 0)
8335 md_kick_rdev_from_array(rdev);
8336 }
8337 }
8338
31a59e34 8339 if (!mddev->external) {
0fd62b86 8340 int did_change = 0;
85572d7c 8341 spin_lock(&mddev->lock);
31a59e34
N		 8342 if (mddev->safemode &&
8343 !atomic_read(&mddev->writes_pending) &&
8344 !mddev->in_sync &&
8345 mddev->recovery_cp == MaxSector) {
8346 mddev->in_sync = 1;
0fd62b86 8347 did_change = 1;
070dc6dd 8348 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 8349 }
8350 if (mddev->safemode == 1)
8351 mddev->safemode = 0;
85572d7c 8352 spin_unlock(&mddev->lock);
0fd62b86 8353 if (did_change)
00bcb4ac 8354 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 8355 }
fca4d848 8356
2aa82191 8357 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 8358 md_update_sb(mddev, 0);
06d91a5f 8359
1da177e4
LT		 8360 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8361 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8362 /* resync/recovery still happening */
8363 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8364 goto unlock;
8365 }
8366 if (mddev->sync_thread) {
a91d5ac0 8367 md_reap_sync_thread(mddev);
1da177e4
LT		 8368 goto unlock;
8369 }
72a23c21
NB		 8370 /* Set RUNNING before clearing NEEDED to avoid
8371 * any transients in the value of "sync_action".
8372 */
72f36d59 8373 mddev->curr_resync_completed = 0;
23da422b 8374 spin_lock(&mddev->lock);
72a23c21 8375 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
23da422b 8376 spin_unlock(&mddev->lock);
24dd469d
N		 8377 /* Clear some bits that don't mean anything, but
8378 * might be left set
8379 */
24dd469d
N		 8380 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8381 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 8382
ed209584
N		 8383 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8384 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
ac05f256 8385 goto not_running;
1da177e4
LT		 8386 /* no recovery is running.
8387 * remove any failed drives, then
8388 * add spares if possible.
72f36d59 8389 * Spares are also removed and re-added, to allow
1da177e4
LT		 8390 * the personality to fail the re-add.
8391 */
1da177e4 8392
b4c4c7b8 8393 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 8394 if (mddev->pers->check_reshape == NULL ||
8395 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8 8396 /* Cannot proceed */
ac05f256 8397 goto not_running;
b4c4c7b8 8398 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 8399 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 8400 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N		 8401 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8402 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 8403 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 8404 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8405 } else if (mddev->recovery_cp < MaxSector) {
8406 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 8407 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8408 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8409 /* nothing to be done ... */
ac05f256 8410 goto not_running;
24dd469d 8411
1da177e4 8412 if (mddev->pers->sync_request) {
ef99bf48 8413 if (spares) {
a654b9d8
N		 8414 /* We are adding a device or devices to an array
8415 * which has the bitmap stored on all devices.
8416 * So make sure all bitmap pages get written
8417 */
8418 bitmap_write_all(mddev->bitmap);
8419 }
ac05f256
N		 8420 INIT_WORK(&mddev->del_work, md_start_sync);
8421 queue_work(md_misc_wq, &mddev->del_work);
8422 goto unlock;
1da177e4 8423 }
ac05f256 8424 not_running:
72a23c21
NB		 8425 if (!mddev->sync_thread) {
8426 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8427 wake_up(&resync_wait);
72a23c21
NB		 8428 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8429 &mddev->recovery))
0c3573f1 8430 if (mddev->sysfs_action)
00bcb4ac 8431 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 8432 }
ac05f256
N		 8433 unlock:
8434 wake_up(&mddev->sb_wait);
1da177e4
LT		 8435 mddev_unlock(mddev);
8436 }
8437}
6c144d31 8438EXPORT_SYMBOL(md_check_recovery);
1da177e4 8439
a91d5ac0
JB		 8440void md_reap_sync_thread(struct mddev *mddev)
8441{
8442 struct md_rdev *rdev;
8443
8444 /* resync has finished, collect result */
8445 md_unregister_thread(&mddev->sync_thread);
8446 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8447 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8448 /* success...*/
8449 /* activate any spares */
8450 if (mddev->pers->spare_active(mddev)) {
8451 sysfs_notify(&mddev->kobj, NULL,
8452 "degraded");
8453 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8454 }
8455 }
8456 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8457 mddev->pers->finish_reshape)
8458 mddev->pers->finish_reshape(mddev);
8459
8460 /* If array is no-longer degraded, then any saved_raid_disk
f466722c 8461 * information must be scrapped.
a91d5ac0 8462 */
f466722c
N		 8463 if (!mddev->degraded)
8464 rdev_for_each(rdev, mddev)
a91d5ac0
JB		 8465 rdev->saved_raid_disk = -1;
8466
8467 md_update_sb(mddev, 1);
8468 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
ea358cd0 8469 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
a91d5ac0
JB		 8470 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8471 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8472 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8473 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
f851b60d 8474 wake_up(&resync_wait);
a91d5ac0
JB		 8475 /* flag recovery needed just to double check */
8476 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8477 sysfs_notify_dirent_safe(mddev->sysfs_action);
8478 md_new_event(mddev);
8479 if (mddev->event_work.func)
8480 queue_work(md_misc_wq, &mddev->event_work);
8481}
6c144d31 8482EXPORT_SYMBOL(md_reap_sync_thread);
a91d5ac0 8483
fd01b88c 8484void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 8485{
00bcb4ac 8486 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 8487 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 8488 !test_bit(Blocked, &rdev->flags) &&
8489 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 8490 msecs_to_jiffies(5000));
8491 rdev_dec_pending(rdev, mddev);
8492}
8493EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8494
c6563a8c
N		 8495void md_finish_reshape(struct mddev *mddev)
8496{
8497 /* called be personality module when reshape completes. */
8498 struct md_rdev *rdev;
8499
8500 rdev_for_each(rdev, mddev) {
8501 if (rdev->data_offset > rdev->new_data_offset)
8502 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8503 else
8504 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8505 rdev->data_offset = rdev->new_data_offset;
8506 }
8507}
8508EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N		 8509
8510/* Bad block management.
8511 * We can record which blocks on each device are 'bad' and so just
8512 * fail those blocks, or that stripe, rather than the whole device.
8513 * Entries in the bad-block table are 64bits wide. This comprises:
8514 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8515 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8516 * A 'shift' can be set so that larger blocks are tracked and
8517 * consequently larger devices can be covered.
8518 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8519 *
8520 * Locking of the bad-block table uses a seqlock so md_is_badblock
8521 * might need to retry if it is very unlucky.
8522 * We will sometimes want to check for bad blocks in a bi_end_io function,
8523 * so we use the write_seqlock_irq variant.
8524 *
8525 * When looking for a bad block we specify a range and want to
8526 * know if any block in the range is bad. So we binary-search
8527 * to the last range that starts at-or-before the given endpoint,
8528 * (or "before the sector after the target range")
8529 * then see if it ends after the given start.
8530 * We return
8531 * 0 if there are no known bad blocks in the range
8532 * 1 if there are known bad block which are all acknowledged
8533 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8534 * plus the start/length of the first bad section we overlap.
8535 */
8536int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8537 sector_t *first_bad, int *bad_sectors)
8538{
8539 int hi;
ab05613a 8540 int lo;
2230dfe4 8541 u64 *p = bb->page;
ab05613a 8542 int rv;
2230dfe4
N		 8543 sector_t target = s + sectors;
8544 unsigned seq;
8545
8546 if (bb->shift > 0) {
8547 /* round the start down, and the end up */
8548 s >>= bb->shift;
8549 target += (1<<bb->shift) - 1;
8550 target >>= bb->shift;
8551 sectors = target - s;
8552 }
8553 /* 'target' is now the first block after the bad range */
8554
8555retry:
8556 seq = read_seqbegin(&bb->lock);
ab05613a 8557 lo = 0;
8558 rv = 0;
2230dfe4
N		 8559 hi = bb->count;
8560
8561 /* Binary search between lo and hi for 'target'
8562 * i.e. for the last range that starts before 'target'
8563 */
8564 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8565 * are known not to be the last range before target.
8566 * VARIANT: hi-lo is the number of possible
8567 * ranges, and decreases until it reaches 1
8568 */
8569 while (hi - lo > 1) {
8570 int mid = (lo + hi) / 2;
8571 sector_t a = BB_OFFSET(p[mid]);
8572 if (a < target)
8573 /* This could still be the one, earlier ranges
8574 * could not. */
8575 lo = mid;
8576 else
8577 /* This and later ranges are definitely out. */
8578 hi = mid;
8579 }
8580 /* 'lo' might be the last that started before target, but 'hi' isn't */
8581 if (hi > lo) {
8582 /* need to check all range that end after 's' to see if
8583 * any are unacknowledged.
8584 */
8585 while (lo >= 0 &&
8586 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8587 if (BB_OFFSET(p[lo]) < target) {
8588 /* starts before the end, and finishes after
8589 * the start, so they must overlap
8590 */
8591 if (rv != -1 && BB_ACK(p[lo]))
8592 rv = 1;
8593 else
8594 rv = -1;
8595 *first_bad = BB_OFFSET(p[lo]);
8596 *bad_sectors = BB_LEN(p[lo]);
8597 }
8598 lo--;
8599 }
8600 }
8601
8602 if (read_seqretry(&bb->lock, seq))
8603 goto retry;
8604
8605 return rv;
8606}
8607EXPORT_SYMBOL_GPL(md_is_badblock);
8608
8609/*
8610 * Add a range of bad blocks to the table.
8611 * This might extend the table, or might contract it
8612 * if two adjacent ranges can be merged.
8613 * We binary-search to find the 'insertion' point, then
8614 * decide how best to handle it.
8615 */
8616static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8617 int acknowledged)
8618{
8619 u64 *p;
8620 int lo, hi;
8621 int rv = 1;
905b0297 8622 unsigned long flags;
2230dfe4
N		 8623
8624 if (bb->shift < 0)
8625 /* badblocks are disabled */
8626 return 0;
8627
8628 if (bb->shift) {
8629 /* round the start down, and the end up */
8630 sector_t next = s + sectors;
8631 s >>= bb->shift;
8632 next += (1<<bb->shift) - 1;
8633 next >>= bb->shift;
8634 sectors = next - s;
8635 }
8636
905b0297 8637 write_seqlock_irqsave(&bb->lock, flags);
2230dfe4
N		 8638
8639 p = bb->page;
8640 lo = 0;
8641 hi = bb->count;
8642 /* Find the last range that starts at-or-before 's' */
8643 while (hi - lo > 1) {
8644 int mid = (lo + hi) / 2;
8645 sector_t a = BB_OFFSET(p[mid]);
8646 if (a <= s)
8647 lo = mid;
8648 else
8649 hi = mid;
8650 }
8651 if (hi > lo && BB_OFFSET(p[lo]) > s)
8652 hi = lo;
8653
8654 if (hi > lo) {
8655 /* we found a range that might merge with the start
8656 * of our new range
8657 */
8658 sector_t a = BB_OFFSET(p[lo]);
8659 sector_t e = a + BB_LEN(p[lo]);
8660 int ack = BB_ACK(p[lo]);
8661 if (e >= s) {
8662 /* Yes, we can merge with a previous range */
8663 if (s == a && s + sectors >= e)
8664 /* new range covers old */
8665 ack = acknowledged;
8666 else
8667 ack = ack && acknowledged;
8668
8669 if (e < s + sectors)
8670 e = s + sectors;
8671 if (e - a <= BB_MAX_LEN) {
8672 p[lo] = BB_MAKE(a, e-a, ack);
8673 s = e;
8674 } else {
8675 /* does not all fit in one range,
8676 * make p[lo] maximal
8677 */
8678 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8679 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8680 s = a + BB_MAX_LEN;
8681 }
8682 sectors = e - s;
8683 }
8684 }
8685 if (sectors && hi < bb->count) {
8686 /* 'hi' points to the first range that starts after 's'.
8687 * Maybe we can merge with the start of that range */
8688 sector_t a = BB_OFFSET(p[hi]);
8689 sector_t e = a + BB_LEN(p[hi]);
8690 int ack = BB_ACK(p[hi]);
8691 if (a <= s + sectors) {
8692 /* merging is possible */
8693 if (e <= s + sectors) {
8694 /* full overlap */
8695 e = s + sectors;
8696 ack = acknowledged;
8697 } else
8698 ack = ack && acknowledged;
8699
8700 a = s;
8701 if (e - a <= BB_MAX_LEN) {
8702 p[hi] = BB_MAKE(a, e-a, ack);
8703 s = e;
8704 } else {
8705 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8706 s = a + BB_MAX_LEN;
8707 }
8708 sectors = e - s;
8709 lo = hi;
8710 hi++;
8711 }
8712 }
8713 if (sectors == 0 && hi < bb->count) {
8714 /* we might be able to combine lo and hi */
8715 /* Note: 's' is at the end of 'lo' */
8716 sector_t a = BB_OFFSET(p[hi]);
8717 int lolen = BB_LEN(p[lo]);
8718 int hilen = BB_LEN(p[hi]);
8719 int newlen = lolen + hilen - (s - a);
8720 if (s >= a && newlen < BB_MAX_LEN) {
8721 /* yes, we can combine them */
8722 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8723 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8724 memmove(p + hi, p + hi + 1,
8725 (bb->count - hi - 1) * 8);
8726 bb->count--;
8727 }
8728 }
8729 while (sectors) {
8730 /* didn't merge (it all).
8731 * Need to add a range just before 'hi' */
8732 if (bb->count >= MD_MAX_BADBLOCKS) {
8733 /* No room for more */
8734 rv = 0;
8735 break;
8736 } else {
8737 int this_sectors = sectors;
8738 memmove(p + hi + 1, p + hi,
8739 (bb->count - hi) * 8);
8740 bb->count++;
8741
8742 if (this_sectors > BB_MAX_LEN)
8743 this_sectors = BB_MAX_LEN;
8744 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8745 sectors -= this_sectors;
8746 s += this_sectors;
8747 }
8748 }
8749
8750 bb->changed = 1;
de393cde
N		 8751 if (!acknowledged)
8752 bb->unacked_exist = 1;
905b0297 8753 write_sequnlock_irqrestore(&bb->lock, flags);
2230dfe4
N		 8754
8755 return rv;
8756}
8757
3cb03002 8758int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8759 int is_new)
2230dfe4 8760{
c6563a8c
N		 8761 int rv;
8762 if (is_new)
8763 s += rdev->new_data_offset;
8764 else
8765 s += rdev->data_offset;
8766 rv = md_set_badblocks(&rdev->badblocks,
8767 s, sectors, 0);
2230dfe4
N		 8768 if (rv) {
8769 /* Make sure they get written out promptly */
8bd2f0a0 8770 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4 8771 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
55ce74d4 8772 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
2230dfe4
N		 8773 md_wakeup_thread(rdev->mddev->thread);
8774 }
8775 return rv;
8776}
8777EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8778
8779/*
8780 * Remove a range of bad blocks from the table.
8781 * This may involve extending the table if we spilt a region,
8782 * but it must not fail. So if the table becomes full, we just
8783 * drop the remove request.
8784 */
8785static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8786{
8787 u64 *p;
8788 int lo, hi;
8789 sector_t target = s + sectors;
8790 int rv = 0;
8791
8792 if (bb->shift > 0) {
8793 /* When clearing we round the start up and the end down.
8794 * This should not matter as the shift should align with
8795 * the block size and no rounding should ever be needed.
8796 * However it is better the think a block is bad when it
8797 * isn't than to think a block is not bad when it is.
8798 */
8799 s += (1<<bb->shift) - 1;
8800 s >>= bb->shift;
8801 target >>= bb->shift;
8802 sectors = target - s;
8803 }
8804
8805 write_seqlock_irq(&bb->lock);
8806
8807 p = bb->page;
8808 lo = 0;
8809 hi = bb->count;
8810 /* Find the last range that starts before 'target' */
8811 while (hi - lo > 1) {
8812 int mid = (lo + hi) / 2;
8813 sector_t a = BB_OFFSET(p[mid]);
8814 if (a < target)
8815 lo = mid;
8816 else
8817 hi = mid;
8818 }
8819 if (hi > lo) {
8820 /* p[lo] is the last range that could overlap the
8821 * current range. Earlier ranges could also overlap,
8822 * but only this one can overlap the end of the range.
8823 */
8824 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8825 /* Partial overlap, leave the tail of this range */
8826 int ack = BB_ACK(p[lo]);
8827 sector_t a = BB_OFFSET(p[lo]);
8828 sector_t end = a + BB_LEN(p[lo]);
8829
8830 if (a < s) {
8831 /* we need to split this range */
8832 if (bb->count >= MD_MAX_BADBLOCKS) {
8b32bf5e 8833 rv = -ENOSPC;
2230dfe4
N		 8834 goto out;
8835 }
8836 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8837 bb->count++;
8838 p[lo] = BB_MAKE(a, s-a, ack);
8839 lo++;
8840 }
8841 p[lo] = BB_MAKE(target, end - target, ack);
8842 /* there is no longer an overlap */
8843 hi = lo;
8844 lo--;
8845 }
8846 while (lo >= 0 &&
8847 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8848 /* This range does overlap */
8849 if (BB_OFFSET(p[lo]) < s) {
8850 /* Keep the early parts of this range. */
8851 int ack = BB_ACK(p[lo]);
8852 sector_t start = BB_OFFSET(p[lo]);
8853 p[lo] = BB_MAKE(start, s - start, ack);
8854 /* now low doesn't overlap, so.. */
8855 break;
8856 }
8857 lo--;
8858 }
8859 /* 'lo' is strictly before, 'hi' is strictly after,
8860 * anything between needs to be discarded
8861 */
8862 if (hi - lo > 1) {
8863 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8864 bb->count -= (hi - lo - 1);
8865 }
8866 }
8867
8868 bb->changed = 1;
8869out:
8870 write_sequnlock_irq(&bb->lock);
8871 return rv;
8872}
8873
c6563a8c
N		 8874int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8875 int is_new)
2230dfe4 8876{
c6563a8c
N		 8877 if (is_new)
8878 s += rdev->new_data_offset;
8879 else
8880 s += rdev->data_offset;
2230dfe4 8881 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8882 s, sectors);
2230dfe4
N		 8883}
8884EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8885
8886/*
8887 * Acknowledge all bad blocks in a list.
8888 * This only succeeds if ->changed is clear. It is used by
8889 * in-kernel metadata updates
8890 */
8891void md_ack_all_badblocks(struct badblocks *bb)
8892{
8893 if (bb->page == NULL || bb->changed)
8894 /* no point even trying */
8895 return;
8896 write_seqlock_irq(&bb->lock);
8897
ecb178bb 8898 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N		 8899 u64 *p = bb->page;
8900 int i;
8901 for (i = 0; i < bb->count ; i++) {
8902 if (!BB_ACK(p[i])) {
8903 sector_t start = BB_OFFSET(p[i]);
8904 int len = BB_LEN(p[i]);
8905 p[i] = BB_MAKE(start, len, 1);
8906 }
8907 }
de393cde 8908 bb->unacked_exist = 0;
2230dfe4
N		 8909 }
8910 write_sequnlock_irq(&bb->lock);
8911}
8912EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8913
16c791a5
N		 8914/* sysfs access to bad-blocks list.
8915 * We present two files.
8916 * 'bad-blocks' lists sector numbers and lengths of ranges that
8917 * are recorded as bad. The list is truncated to fit within
8918 * the one-page limit of sysfs.
8919 * Writing "sector length" to this file adds an acknowledged
8920 * bad block list.
8921 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8922 * been acknowledged. Writing to this file adds bad blocks
8923 * without acknowledging them. This is largely for testing.
8924 */
8925
8926static ssize_t
8927badblocks_show(struct badblocks *bb, char *page, int unack)
8928{
8929 size_t len;
8930 int i;
8931 u64 *p = bb->page;
8932 unsigned seq;
8933
8934 if (bb->shift < 0)
8935 return 0;
8936
8937retry:
8938 seq = read_seqbegin(&bb->lock);
8939
8940 len = 0;
8941 i = 0;
8942
8943 while (len < PAGE_SIZE && i < bb->count) {
8944 sector_t s = BB_OFFSET(p[i]);
8945 unsigned int length = BB_LEN(p[i]);
8946 int ack = BB_ACK(p[i]);
8947 i++;
8948
8949 if (unack && ack)
8950 continue;
8951
8952 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8953 (unsigned long long)s << bb->shift,
8954 length << bb->shift);
8955 }
de393cde
N		 8956 if (unack && len == 0)
8957 bb->unacked_exist = 0;
16c791a5
N		 8958
8959 if (read_seqretry(&bb->lock, seq))
8960 goto retry;
8961
8962 return len;
8963}
8964
8965#define DO_DEBUG 1
8966
8967static ssize_t
8968badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8969{
8970 unsigned long long sector;
8971 int length;
8972 char newline;
8973#ifdef DO_DEBUG
8974 /* Allow clearing via sysfs *only* for testing/debugging.
8975 * Normally only a successful write may clear a badblock
8976 */
8977 int clear = 0;
8978 if (page[0] == '-') {
8979 clear = 1;
8980 page++;
8981 }
8982#endif /* DO_DEBUG */
8983
8984 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8985 case 3:
8986 if (newline != '\n')
8987 return -EINVAL;
8988 case 2:
8989 if (length <= 0)
8990 return -EINVAL;
8991 break;
8992 default:
8993 return -EINVAL;
8994 }
8995
8996#ifdef DO_DEBUG
8997 if (clear) {
8998 md_clear_badblocks(bb, sector, length);
8999 return len;
9000 }
9001#endif /* DO_DEBUG */
9002 if (md_set_badblocks(bb, sector, length, !unack))
9003 return len;
9004 else
9005 return -ENOSPC;
9006}
9007
75c96f85
AB		 9008static int md_notify_reboot(struct notifier_block *this,
9009 unsigned long code, void *x)
1da177e4
LT		 9010{
9011 struct list_head *tmp;
fd01b88c 9012 struct mddev *mddev;
2dba6a91 9013 int need_delay = 0;
1da177e4 9014
c744a65c
N		 9015 for_each_mddev(mddev, tmp) {
9016 if (mddev_trylock(mddev)) {
30b8aa91
N		 9017 if (mddev->pers)
9018 __md_stop_writes(mddev);
0f62fb22
N		 9019 if (mddev->persistent)
9020 mddev->safemode = 2;
c744a65c 9021 mddev_unlock(mddev);
2dba6a91 9022 }
c744a65c 9023 need_delay = 1;
1da177e4 9024 }
c744a65c
N		 9025 /*
9026 * certain more exotic SCSI devices are known to be
9027 * volatile wrt too early system reboots. While the
9028 * right place to handle this issue is the given
9029 * driver, we do want to have a safe RAID driver ...
9030 */
9031 if (need_delay)
9032 mdelay(1000*1);
9033
1da177e4
LT		 9034 return NOTIFY_DONE;
9035}
9036
75c96f85 9037static struct notifier_block md_notifier = {
1da177e4
LT		 9038 .notifier_call = md_notify_reboot,
9039 .next = NULL,
9040 .priority = INT_MAX, /* before any real devices */
9041};
9042
9043static void md_geninit(void)
9044{
36a4e1fe 9045 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 9046
c7705f34 9047 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 9048}
9049
75c96f85 9050static int __init md_init(void)
1da177e4 9051{
e804ac78
TH		 9052 int ret = -ENOMEM;
9053
ada609ee 9054 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 9055 if (!md_wq)
9056 goto err_wq;
9057
9058 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9059 if (!md_misc_wq)
9060 goto err_misc_wq;
9061
9062 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9063 goto err_md;
9064
9065 if ((ret = register_blkdev(0, "mdp")) < 0)
9066 goto err_mdp;
9067 mdp_major = ret;
9068
af5628f0 9069 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
e8703fe1
N		 9070 md_probe, NULL, NULL);
9071 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 9072 md_probe, NULL, NULL);
9073
1da177e4 9074 register_reboot_notifier(&md_notifier);
0b4d4147 9075 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 9076
9077 md_geninit();
d710e138 9078 return 0;
1da177e4 9079
e804ac78
TH		 9080err_mdp:
9081 unregister_blkdev(MD_MAJOR, "md");
9082err_md:
9083 destroy_workqueue(md_misc_wq);
9084err_misc_wq:
9085 destroy_workqueue(md_wq);
9086err_wq:
9087 return ret;
9088}
1da177e4 9089
70bcecdb 9090static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
1d7e3e96 9091{
70bcecdb
GR		 9092 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9093 struct md_rdev *rdev2;
9094 int role, ret;
9095 char b[BDEVNAME_SIZE];
1d7e3e96 9096
70bcecdb
GR		 9097 /* Check for change of roles in the active devices */
9098 rdev_for_each(rdev2, mddev) {
9099 if (test_bit(Faulty, &rdev2->flags))
9100 continue;
9101
9102 /* Check if the roles changed */
9103 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
dbb64f86
GR		 9104
9105 if (test_bit(Candidate, &rdev2->flags)) {
9106 if (role == 0xfffe) {
9107 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9108 md_kick_rdev_from_array(rdev2);
9109 continue;
9110 }
9111 else
9112 clear_bit(Candidate, &rdev2->flags);
9113 }
9114
70bcecdb
GR		 9115 if (role != rdev2->raid_disk) {
9116 /* got activated */
9117 if (rdev2->raid_disk == -1 && role != 0xffff) {
9118 rdev2->saved_raid_disk = role;
9119 ret = remove_and_add_spares(mddev, rdev2);
9120 pr_info("Activated spare: %s\n",
9121 bdevname(rdev2->bdev,b));
70bcecdb
GR		 9122 }
9123 /* device faulty
9124 * We just want to do the minimum to mark the disk
9125 * as faulty. The recovery is performed by the
9126 * one who initiated the error.
9127 */
9128 if ((role == 0xfffe) || (role == 0xfffd)) {
9129 md_error(mddev, rdev2);
9130 clear_bit(Blocked, &rdev2->flags);
9131 }
9132 }
1d7e3e96 9133 }
70bcecdb 9134
28c1b9fd
GR		 9135 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9136 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
70bcecdb
GR		 9137
9138 /* Finally set the event to be up to date */
9139 mddev->events = le64_to_cpu(sb->events);
9140}
9141
9142static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9143{
9144 int err;
9145 struct page *swapout = rdev->sb_page;
9146 struct mdp_superblock_1 *sb;
9147
9148 /* Store the sb page of the rdev in the swapout temporary
9149 * variable in case we err in the future
9150 */
9151 rdev->sb_page = NULL;
9152 alloc_disk_sb(rdev);
9153 ClearPageUptodate(rdev->sb_page);
9154 rdev->sb_loaded = 0;
9155 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
9156
9157 if (err < 0) {
9158 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9159 __func__, __LINE__, rdev->desc_nr, err);
9160 put_page(rdev->sb_page);
9161 rdev->sb_page = swapout;
9162 rdev->sb_loaded = 1;
9163 return err;
1d7e3e96
GR		 9164 }
9165
70bcecdb
GR		 9166 sb = page_address(rdev->sb_page);
9167 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9168 * is not set
9169 */
9170
9171 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9172 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9173
9174 /* The other node finished recovery, call spare_active to set
9175 * device In_sync and mddev->degraded
9176 */
9177 if (rdev->recovery_offset == MaxSector &&
9178 !test_bit(In_sync, &rdev->flags) &&
9179 mddev->pers->spare_active(mddev))
9180 sysfs_notify(&mddev->kobj, NULL, "degraded");
9181
9182 put_page(swapout);
9183 return 0;
9184}
9185
9186void md_reload_sb(struct mddev *mddev, int nr)
9187{
9188 struct md_rdev *rdev;
9189 int err;
9190
9191 /* Find the rdev */
9192 rdev_for_each_rcu(rdev, mddev) {
9193 if (rdev->desc_nr == nr)
9194 break;
9195 }
9196
9197 if (!rdev || rdev->desc_nr != nr) {
9198 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9199 return;
9200 }
9201
9202 err = read_rdev(mddev, rdev);
9203 if (err < 0)
9204 return;
9205
9206 check_sb_changes(mddev, rdev);
9207
9208 /* Read all rdev's to update recovery_offset */
9209 rdev_for_each_rcu(rdev, mddev)
9210 read_rdev(mddev, rdev);
1d7e3e96
GR		 9211}
9212EXPORT_SYMBOL(md_reload_sb);
9213
1da177e4
LT		 9214#ifndef MODULE
9215
9216/*
9217 * Searches all registered partitions for autorun RAID arrays
9218 * at boot time.
9219 */
4d936ec1
ME		 9220
9221static LIST_HEAD(all_detected_devices);
9222struct detected_devices_node {
9223 struct list_head list;
9224 dev_t dev;
9225};
1da177e4
LT		 9226
9227void md_autodetect_dev(dev_t dev)
9228{
4d936ec1
ME		 9229 struct detected_devices_node *node_detected_dev;
9230
9231 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9232 if (node_detected_dev) {
9233 node_detected_dev->dev = dev;
9234 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9235 } else {
9236 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
9237 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
9238 }
1da177e4
LT		 9239}
9240
1da177e4
LT		 9241static void autostart_arrays(int part)
9242{
3cb03002 9243 struct md_rdev *rdev;
4d936ec1
ME		 9244 struct detected_devices_node *node_detected_dev;
9245 dev_t dev;
9246 int i_scanned, i_passed;
1da177e4 9247
4d936ec1
ME		 9248 i_scanned = 0;
9249 i_passed = 0;
1da177e4 9250
4d936ec1 9251 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 9252
4d936ec1
ME		 9253 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9254 i_scanned++;
9255 node_detected_dev = list_entry(all_detected_devices.next,
9256 struct detected_devices_node, list);
9257 list_del(&node_detected_dev->list);
9258 dev = node_detected_dev->dev;
9259 kfree(node_detected_dev);
df968c4e 9260 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 9261 if (IS_ERR(rdev))
9262 continue;
9263
403df478 9264 if (test_bit(Faulty, &rdev->flags))
1da177e4 9265 continue;
403df478 9266
d0fae18f 9267 set_bit(AutoDetected, &rdev->flags);
1da177e4 9268 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 9269 i_passed++;
1da177e4 9270 }
4d936ec1
ME		 9271
9272 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
9273 i_scanned, i_passed);
1da177e4
LT		 9274
9275 autorun_devices(part);
9276}
9277
fdee8ae4 9278#endif /* !MODULE */
1da177e4
LT		 9279
9280static __exit void md_exit(void)
9281{
fd01b88c 9282 struct mddev *mddev;
1da177e4 9283 struct list_head *tmp;
e2f23b60 9284 int delay = 1;
8ab5e4c1 9285
af5628f0 9286 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
e8703fe1 9287 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 9288
3dbd8c2e 9289 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 9290 unregister_blkdev(mdp_major, "mdp");
9291 unregister_reboot_notifier(&md_notifier);
9292 unregister_sysctl_table(raid_table_header);
e2f23b60
N		 9293
9294 /* We cannot unload the modules while some process is
9295 * waiting for us in select() or poll() - wake them up
9296 */
9297 md_unloading = 1;
9298 while (waitqueue_active(&md_event_waiters)) {
9299 /* not safe to leave yet */
9300 wake_up(&md_event_waiters);
9301 msleep(delay);
9302 delay += delay;
9303 }
1da177e4 9304 remove_proc_entry("mdstat", NULL);
e2f23b60 9305
29ac4aa3 9306 for_each_mddev(mddev, tmp) {
1da177e4 9307 export_array(mddev);
d3374825 9308 mddev->hold_active = 0;
1da177e4 9309 }
e804ac78
TH		 9310 destroy_workqueue(md_misc_wq);
9311 destroy_workqueue(md_wq);
1da177e4
LT		 9312}
9313
685784aa 9314subsys_initcall(md_init);
1da177e4
LT		 9315module_exit(md_exit)
9316
f91de92e
N		 9317static int get_ro(char *buffer, struct kernel_param *kp)
9318{
9319 return sprintf(buffer, "%d", start_readonly);
9320}
9321static int set_ro(const char *val, struct kernel_param *kp)
9322{
4c9309c0 9323 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
f91de92e
N		 9324}
9325
80ca3a44
N		 9326module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9327module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
efeb53c0 9328module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 9329
1da177e4 9330MODULE_LICENSE("GPL");
0efb9e61 9331MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 9332MODULE_ALIAS("md");
72008652 9333MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)