projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: fix bug due to nested suspend
[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
NB		 2776 */
2777
2778 if (rdev->raid_disk != -1)
2779 return -EBUSY;
2780
c6751b2b
N		 2781 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2782 return -EBUSY;
2783
6c2fce2e
NB		 2784 if (rdev->mddev->pers->hot_add_disk == NULL)
2785 return -EINVAL;
2786
ba1b41b6
N		 2787 if (slot >= rdev->mddev->raid_disks &&
2788 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2789 return -ENOSPC;
2790
6c2fce2e
NB		 2791 rdev->raid_disk = slot;
2792 if (test_bit(In_sync, &rdev->flags))
2793 rdev->saved_raid_disk = slot;
2794 else
2795 rdev->saved_raid_disk = -1;
d30519fc 2796 clear_bit(In_sync, &rdev->flags);
8313b8e5 2797 clear_bit(Bitmap_sync, &rdev->flags);
2910ff17
GR		 2798 remove_and_add_spares(rdev->mddev, rdev);
2799 if (rdev->raid_disk == -1)
2800 return -EBUSY;
6c2fce2e 2801 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2802 } else {
ba1b41b6
N		 2803 if (slot >= rdev->mddev->raid_disks &&
2804 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2805 return -ENOSPC;
2806 rdev->raid_disk = slot;
2807 /* assume it is working */
c5d79adb
N		 2808 clear_bit(Faulty, &rdev->flags);
2809 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2810 set_bit(In_sync, &rdev->flags);
00bcb4ac 2811 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2812 }
014236d2
N		 2813 return len;
2814}
2815
014236d2 2816static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2817__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2818
93c8cad0 2819static ssize_t
3cb03002 2820offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2821{
6961ece4 2822 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2823}
2824
2825static ssize_t
3cb03002 2826offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2827{
c6563a8c 2828 unsigned long long offset;
b29bebd6 2829 if (kstrtoull(buf, 10, &offset) < 0)
93c8cad0 2830 return -EINVAL;
8ed0a521 2831 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2832 return -EBUSY;
dd8ac336 2833 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2834 /* Must set offset before size, so overlap checks
2835 * can be sane */
2836 return -EBUSY;
93c8cad0 2837 rdev->data_offset = offset;
25f7fd47 2838 rdev->new_data_offset = offset;
93c8cad0
N		 2839 return len;
2840}
2841
2842static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2843__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2844
c6563a8c
N		 2845static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2846{
2847 return sprintf(page, "%llu\n",
2848 (unsigned long long)rdev->new_data_offset);
2849}
2850
2851static ssize_t new_offset_store(struct md_rdev *rdev,
2852 const char *buf, size_t len)
2853{
2854 unsigned long long new_offset;
2855 struct mddev *mddev = rdev->mddev;
2856
b29bebd6 2857 if (kstrtoull(buf, 10, &new_offset) < 0)
c6563a8c
N		 2858 return -EINVAL;
2859
f851b60d
N		 2860 if (mddev->sync_thread ||
2861 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
c6563a8c
N		 2862 return -EBUSY;
2863 if (new_offset == rdev->data_offset)
2864 /* reset is always permitted */
2865 ;
2866 else if (new_offset > rdev->data_offset) {
2867 /* must not push array size beyond rdev_sectors */
2868 if (new_offset - rdev->data_offset
2869 + mddev->dev_sectors > rdev->sectors)
2870 return -E2BIG;
2871 }
2872 /* Metadata worries about other space details. */
2873
2874 /* decreasing the offset is inconsistent with a backwards
2875 * reshape.
2876 */
2877 if (new_offset < rdev->data_offset &&
2878 mddev->reshape_backwards)
2879 return -EINVAL;
2880 /* Increasing offset is inconsistent with forwards
2881 * reshape. reshape_direction should be set to
2882 * 'backwards' first.
2883 */
2884 if (new_offset > rdev->data_offset &&
2885 !mddev->reshape_backwards)
2886 return -EINVAL;
2887
2888 if (mddev->pers && mddev->persistent &&
2889 !super_types[mddev->major_version]
2890 .allow_new_offset(rdev, new_offset))
2891 return -E2BIG;
2892 rdev->new_data_offset = new_offset;
2893 if (new_offset > rdev->data_offset)
2894 mddev->reshape_backwards = 1;
2895 else if (new_offset < rdev->data_offset)
2896 mddev->reshape_backwards = 0;
2897
2898 return len;
2899}
2900static struct rdev_sysfs_entry rdev_new_offset =
2901__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2902
83303b61 2903static ssize_t
3cb03002 2904rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2905{
dd8ac336 2906 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2907}
2908
c5d79adb
N		 2909static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2910{
2911 /* check if two start/length pairs overlap */
2912 if (s1+l1 <= s2)
2913 return 0;
2914 if (s2+l2 <= s1)
2915 return 0;
2916 return 1;
2917}
2918
b522adcd
DW		 2919static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2920{
2921 unsigned long long blocks;
2922 sector_t new;
2923
b29bebd6 2924 if (kstrtoull(buf, 10, &blocks) < 0)
b522adcd
DW		 2925 return -EINVAL;
2926
2927 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2928 return -EINVAL; /* sector conversion overflow */
2929
2930 new = blocks * 2;
2931 if (new != blocks * 2)
2932 return -EINVAL; /* unsigned long long to sector_t overflow */
2933
2934 *sectors = new;
2935 return 0;
2936}
2937
83303b61 2938static ssize_t
3cb03002 2939rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2940{
fd01b88c 2941 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2942 sector_t oldsectors = rdev->sectors;
b522adcd 2943 sector_t sectors;
27c529bb 2944
f2076e7d
SL		 2945 if (test_bit(Journal, &rdev->flags))
2946 return -EBUSY;
b522adcd 2947 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2948 return -EINVAL;
c6563a8c
N		 2949 if (rdev->data_offset != rdev->new_data_offset)
2950 return -EINVAL; /* too confusing */
0cd17fec 2951 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2952 if (my_mddev->persistent) {
dd8ac336
AN		 2953 sectors = super_types[my_mddev->major_version].
2954 rdev_size_change(rdev, sectors);
2955 if (!sectors)
0cd17fec 2956 return -EBUSY;
dd8ac336 2957 } else if (!sectors)
77304d2a 2958 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2959 rdev->data_offset;
a6468539
N		 2960 if (!my_mddev->pers->resize)
2961 /* Cannot change size for RAID0 or Linear etc */
2962 return -EINVAL;
0cd17fec 2963 }
dd8ac336 2964 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2965 return -EINVAL; /* component must fit device */
0cd17fec 2966
dd8ac336
AN		 2967 rdev->sectors = sectors;
2968 if (sectors > oldsectors && my_mddev->external) {
8b1afc3d
N		 2969 /* Need to check that all other rdevs with the same
2970 * ->bdev do not overlap. 'rcu' is sufficient to walk
2971 * the rdev lists safely.
2972 * This check does not provide a hard guarantee, it
2973 * just helps avoid dangerous mistakes.
c5d79adb 2974 */
fd01b88c 2975 struct mddev *mddev;
c5d79adb 2976 int overlap = 0;
159ec1fc 2977 struct list_head *tmp;
c5d79adb 2978
8b1afc3d 2979 rcu_read_lock();
29ac4aa3 2980 for_each_mddev(mddev, tmp) {
3cb03002 2981 struct md_rdev *rdev2;
c5d79adb 2982
dafb20fa 2983 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 2984 if (rdev->bdev == rdev2->bdev &&
2985 rdev != rdev2 &&
2986 overlaps(rdev->data_offset, rdev->sectors,
2987 rdev2->data_offset,
2988 rdev2->sectors)) {
c5d79adb
N		 2989 overlap = 1;
2990 break;
2991 }
c5d79adb
N		 2992 if (overlap) {
2993 mddev_put(mddev);
2994 break;
2995 }
2996 }
8b1afc3d 2997 rcu_read_unlock();
c5d79adb
N		 2998 if (overlap) {
2999 /* Someone else could have slipped in a size
3000 * change here, but doing so is just silly.
dd8ac336 3001 * We put oldsectors back because we *know* it is
c5d79adb
N		 3002 * safe, and trust userspace not to race with
3003 * itself
3004 */
dd8ac336 3005 rdev->sectors = oldsectors;
c5d79adb
N		 3006 return -EBUSY;
3007 }
3008 }
83303b61
N		 3009 return len;
3010}
3011
3012static struct rdev_sysfs_entry rdev_size =
80ca3a44 3013__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 3014
3cb03002 3015static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 3016{
3017 unsigned long long recovery_start = rdev->recovery_offset;
3018
3019 if (test_bit(In_sync, &rdev->flags) ||
3020 recovery_start == MaxSector)
3021 return sprintf(page, "none\n");
3022
3023 return sprintf(page, "%llu\n", recovery_start);
3024}
3025
3cb03002 3026static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 3027{
3028 unsigned long long recovery_start;
3029
3030 if (cmd_match(buf, "none"))
3031 recovery_start = MaxSector;
b29bebd6 3032 else if (kstrtoull(buf, 10, &recovery_start))
06e3c817
DW		 3033 return -EINVAL;
3034
3035 if (rdev->mddev->pers &&
3036 rdev->raid_disk >= 0)
3037 return -EBUSY;
3038
3039 rdev->recovery_offset = recovery_start;
3040 if (recovery_start == MaxSector)
3041 set_bit(In_sync, &rdev->flags);
3042 else
3043 clear_bit(In_sync, &rdev->flags);
3044 return len;
3045}
3046
3047static struct rdev_sysfs_entry rdev_recovery_start =
3048__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3049
16c791a5
N		 3050static ssize_t
3051badblocks_show(struct badblocks *bb, char *page, int unack);
3052static ssize_t
3053badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3054
3cb03002 3055static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3056{
3057 return badblocks_show(&rdev->badblocks, page, 0);
3058}
3cb03002 3059static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 3060{
de393cde
N		 3061 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3062 /* Maybe that ack was all we needed */
3063 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3064 wake_up(&rdev->blocked_wait);
3065 return rv;
16c791a5
N		 3066}
3067static struct rdev_sysfs_entry rdev_bad_blocks =
3068__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3069
3cb03002 3070static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3071{
3072 return badblocks_show(&rdev->badblocks, page, 1);
3073}
3cb03002 3074static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 3075{
3076 return badblocks_store(&rdev->badblocks, page, len, 1);
3077}
3078static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3079__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3080
86e6ffdd
N		 3081static struct attribute *rdev_default_attrs[] = {
3082 &rdev_state.attr,
4dbcdc75 3083 &rdev_errors.attr,
014236d2 3084 &rdev_slot.attr,
93c8cad0 3085 &rdev_offset.attr,
c6563a8c 3086 &rdev_new_offset.attr,
83303b61 3087 &rdev_size.attr,
06e3c817 3088 &rdev_recovery_start.attr,
16c791a5
N		 3089 &rdev_bad_blocks.attr,
3090 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 3091 NULL,
3092};
3093static ssize_t
3094rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3095{
3096 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3097 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
86e6ffdd
N		 3098
3099 if (!entry->show)
3100 return -EIO;
758bfc8a
N		 3101 if (!rdev->mddev)
3102 return -EBUSY;
3103 return entry->show(rdev, page);
86e6ffdd
N		 3104}
3105
3106static ssize_t
3107rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3108 const char *page, size_t length)
3109{
3110 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3111 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3112 ssize_t rv;
fd01b88c 3113 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3114
3115 if (!entry->store)
3116 return -EIO;
67463acb
N		 3117 if (!capable(CAP_SYS_ADMIN))
3118 return -EACCES;
27c529bb 3119 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3120 if (!rv) {
27c529bb
N		 3121 if (rdev->mddev == NULL)
3122 rv = -EBUSY;
3123 else
3124 rv = entry->store(rdev, page, length);
6a51830e 3125 mddev_unlock(mddev);
ca388059
N		 3126 }
3127 return rv;
86e6ffdd
N		 3128}
3129
3130static void rdev_free(struct kobject *ko)
3131{
3cb03002 3132 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3133 kfree(rdev);
3134}
52cf25d0 3135static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3136 .show = rdev_attr_show,
3137 .store = rdev_attr_store,
3138};
3139static struct kobj_type rdev_ktype = {
3140 .release = rdev_free,
3141 .sysfs_ops = &rdev_sysfs_ops,
3142 .default_attrs = rdev_default_attrs,
3143};
3144
3cb03002 3145int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3146{
3147 rdev->desc_nr = -1;
3148 rdev->saved_raid_disk = -1;
3149 rdev->raid_disk = -1;
3150 rdev->flags = 0;
3151 rdev->data_offset = 0;
c6563a8c 3152 rdev->new_data_offset = 0;
e8bb9a83
N		 3153 rdev->sb_events = 0;
3154 rdev->last_read_error.tv_sec = 0;
3155 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3156 rdev->sb_loaded = 0;
3157 rdev->bb_page = NULL;
e8bb9a83
N		 3158 atomic_set(&rdev->nr_pending, 0);
3159 atomic_set(&rdev->read_errors, 0);
3160 atomic_set(&rdev->corrected_errors, 0);
3161
3162 INIT_LIST_HEAD(&rdev->same_set);
3163 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3164
3165 /* Add space to store bad block list.
3166 * This reserves the space even on arrays where it cannot
3167 * be used - I wonder if that matters
3168 */
3169 rdev->badblocks.count = 0;
486adf72 3170 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
2230dfe4
N		 3171 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3172 seqlock_init(&rdev->badblocks.lock);
3173 if (rdev->badblocks.page == NULL)
3174 return -ENOMEM;
3175
3176 return 0;
e8bb9a83
N		 3177}
3178EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3179/*
3180 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3181 *
3182 * mark the device faulty if:
3183 *
3184 * - the device is nonexistent (zero size)
3185 * - the device has no valid superblock
3186 *
3187 * a faulty rdev _never_ has rdev->sb set.
3188 */
3cb03002 3189static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3190{
3191 char b[BDEVNAME_SIZE];
3192 int err;
3cb03002 3193 struct md_rdev *rdev;
1da177e4
LT		 3194 sector_t size;
3195
9ffae0cf 3196 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3197 if (!rdev) {
3198 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3199 return ERR_PTR(-ENOMEM);
3200 }
1da177e4 3201
2230dfe4
N		 3202 err = md_rdev_init(rdev);
3203 if (err)
3204 goto abort_free;
3205 err = alloc_disk_sb(rdev);
3206 if (err)
1da177e4
LT		 3207 goto abort_free;
3208
c5d79adb 3209 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3210 if (err)
3211 goto abort_free;
3212
f9cb074b 3213 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3214
77304d2a 3215 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4 3216 if (!size) {
f72ffdd6 3217 printk(KERN_WARNING
1da177e4
LT		 3218 "md: %s has zero or unknown size, marking faulty!\n",
3219 bdevname(rdev->bdev,b));
3220 err = -EINVAL;
3221 goto abort_free;
3222 }
3223
3224 if (super_format >= 0) {
3225 err = super_types[super_format].
3226 load_super(rdev, NULL, super_minor);
3227 if (err == -EINVAL) {
df968c4e
N		 3228 printk(KERN_WARNING
3229 "md: %s does not have a valid v%d.%d "
3230 "superblock, not importing!\n",
3231 bdevname(rdev->bdev,b),
3232 super_format, super_minor);
1da177e4
LT		 3233 goto abort_free;
3234 }
3235 if (err < 0) {
f72ffdd6 3236 printk(KERN_WARNING
1da177e4
LT		 3237 "md: could not read %s's sb, not importing!\n",
3238 bdevname(rdev->bdev,b));
3239 goto abort_free;
3240 }
3241 }
6bfe0b49 3242
1da177e4
LT		 3243 return rdev;
3244
3245abort_free:
2699b672
N		 3246 if (rdev->bdev)
3247 unlock_rdev(rdev);
545c8795 3248 md_rdev_clear(rdev);
1da177e4
LT		 3249 kfree(rdev);
3250 return ERR_PTR(err);
3251}
3252
3253/*
3254 * Check a full RAID array for plausibility
3255 */
3256
f72ffdd6 3257static void analyze_sbs(struct mddev *mddev)
1da177e4
LT		 3258{
3259 int i;
3cb03002 3260 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3261 char b[BDEVNAME_SIZE];
3262
3263 freshest = NULL;
dafb20fa 3264 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3265 switch (super_types[mddev->major_version].
3266 load_super(rdev, freshest, mddev->minor_version)) {
3267 case 1:
3268 freshest = rdev;
3269 break;
3270 case 0:
3271 break;
3272 default:
3273 printk( KERN_ERR \
3274 "md: fatal superblock inconsistency in %s"
f72ffdd6 3275 " -- removing from array\n",
1da177e4 3276 bdevname(rdev->bdev,b));
fb56dfef 3277 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3278 }
3279
1da177e4
LT		 3280 super_types[mddev->major_version].
3281 validate_super(mddev, freshest);
3282
3283 i = 0;
dafb20fa 3284 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3285 if (mddev->max_disks &&
3286 (rdev->desc_nr >= mddev->max_disks ||
3287 i > mddev->max_disks)) {
de01dfad
N		 3288 printk(KERN_WARNING
3289 "md: %s: %s: only %d devices permitted\n",
3290 mdname(mddev), bdevname(rdev->bdev, b),
3291 mddev->max_disks);
fb56dfef 3292 md_kick_rdev_from_array(rdev);
de01dfad
N		 3293 continue;
3294 }
1aee41f6 3295 if (rdev != freshest) {
1da177e4
LT		 3296 if (super_types[mddev->major_version].
3297 validate_super(mddev, rdev)) {
3298 printk(KERN_WARNING "md: kicking non-fresh %s"
3299 " from array!\n",
3300 bdevname(rdev->bdev,b));
fb56dfef 3301 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3302 continue;
3303 }
1aee41f6 3304 }
1da177e4
LT		 3305 if (mddev->level == LEVEL_MULTIPATH) {
3306 rdev->desc_nr = i++;
3307 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3308 set_bit(In_sync, &rdev->flags);
f2076e7d
SL		 3309 } else if (rdev->raid_disk >=
3310 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3311 !test_bit(Journal, &rdev->flags)) {
a778b73f
N		 3312 rdev->raid_disk = -1;
3313 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3314 }
3315 }
1da177e4
LT		 3316}
3317
72e02075
N		 3318/* Read a fixed-point number.
3319 * Numbers in sysfs attributes should be in "standard" units where
3320 * possible, so time should be in seconds.
f72ffdd6 3321 * However we internally use a a much smaller unit such as
72e02075
N		 3322 * milliseconds or jiffies.
3323 * This function takes a decimal number with a possible fractional
3324 * component, and produces an integer which is the result of
3325 * multiplying that number by 10^'scale'.
3326 * all without any floating-point arithmetic.
3327 */
3328int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3329{
3330 unsigned long result = 0;
3331 long decimals = -1;
3332 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3333 if (*cp == '.')
3334 decimals = 0;
3335 else if (decimals < scale) {
3336 unsigned int value;
3337 value = *cp - '0';
3338 result = result * 10 + value;
3339 if (decimals >= 0)
3340 decimals++;
3341 }
3342 cp++;
3343 }
3344 if (*cp == '\n')
3345 cp++;
3346 if (*cp)
3347 return -EINVAL;
3348 if (decimals < 0)
3349 decimals = 0;
3350 while (decimals < scale) {
3351 result *= 10;
3352 decimals ++;
3353 }
3354 *res = result;
3355 return 0;
3356}
3357
16f17b39 3358static ssize_t
fd01b88c 3359safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3360{
3361 int msec = (mddev->safemode_delay*1000)/HZ;
3362 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3363}
3364static ssize_t
fd01b88c 3365safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3366{
16f17b39 3367 unsigned long msec;
97ce0a7f 3368
28c1b9fd
GR		 3369 if (mddev_is_clustered(mddev)) {
3370 pr_info("md: Safemode is disabled for clustered mode\n");
3371 return -EINVAL;
3372 }
3373
72e02075 3374 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3375 return -EINVAL;
16f17b39
N		 3376 if (msec == 0)
3377 mddev->safemode_delay = 0;
3378 else {
19052c0e 3379 unsigned long old_delay = mddev->safemode_delay;
1b30e66f
N		 3380 unsigned long new_delay = (msec*HZ)/1000;
3381
3382 if (new_delay == 0)
3383 new_delay = 1;
3384 mddev->safemode_delay = new_delay;
3385 if (new_delay < old_delay || old_delay == 0)
3386 mod_timer(&mddev->safemode_timer, jiffies+1);
16f17b39
N		 3387 }
3388 return len;
3389}
3390static struct md_sysfs_entry md_safe_delay =
80ca3a44 3391__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3392
eae1701f 3393static ssize_t
fd01b88c 3394level_show(struct mddev *mddev, char *page)
eae1701f 3395{
36d091f4
N		 3396 struct md_personality *p;
3397 int ret;
3398 spin_lock(&mddev->lock);
3399 p = mddev->pers;
d9d166c2 3400 if (p)
36d091f4 3401 ret = sprintf(page, "%s\n", p->name);
d9d166c2 3402 else if (mddev->clevel[0])
36d091f4 3403 ret = sprintf(page, "%s\n", mddev->clevel);
d9d166c2 3404 else if (mddev->level != LEVEL_NONE)
36d091f4 3405 ret = sprintf(page, "%d\n", mddev->level);
d9d166c2 3406 else
36d091f4
N		 3407 ret = 0;
3408 spin_unlock(&mddev->lock);
3409 return ret;
eae1701f
N		 3410}
3411
d9d166c2 3412static ssize_t
fd01b88c 3413level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3414{
f2859af6 3415 char clevel[16];
6791875e
N		 3416 ssize_t rv;
3417 size_t slen = len;
db721d32 3418 struct md_personality *pers, *oldpers;
f2859af6 3419 long level;
db721d32 3420 void *priv, *oldpriv;
3cb03002 3421 struct md_rdev *rdev;
245f46c2 3422
6791875e
N		 3423 if (slen == 0 || slen >= sizeof(clevel))
3424 return -EINVAL;
3425
3426 rv = mddev_lock(mddev);
3427 if (rv)
3428 return rv;
3429
245f46c2 3430 if (mddev->pers == NULL) {
6791875e
N		 3431 strncpy(mddev->clevel, buf, slen);
3432 if (mddev->clevel[slen-1] == '\n')
3433 slen--;
3434 mddev->clevel[slen] = 0;
245f46c2 3435 mddev->level = LEVEL_NONE;
6791875e
N		 3436 rv = len;
3437 goto out_unlock;
245f46c2 3438 }
6791875e 3439 rv = -EROFS;
bd8839e0 3440 if (mddev->ro)
6791875e 3441 goto out_unlock;
245f46c2
N		 3442
3443 /* request to change the personality. Need to ensure:
3444 * - array is not engaged in resync/recovery/reshape
3445 * - old personality can be suspended
3446 * - new personality will access other array.
3447 */
3448
6791875e 3449 rv = -EBUSY;
bb4f1e9d 3450 if (mddev->sync_thread ||
f851b60d 3451 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
bb4f1e9d
N		 3452 mddev->reshape_position != MaxSector ||
3453 mddev->sysfs_active)
6791875e 3454 goto out_unlock;
245f46c2 3455
6791875e 3456 rv = -EINVAL;
245f46c2
N		 3457 if (!mddev->pers->quiesce) {
3458 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3459 mdname(mddev), mddev->pers->name);
6791875e 3460 goto out_unlock;
245f46c2
N		 3461 }
3462
3463 /* Now find the new personality */
6791875e
N		 3464 strncpy(clevel, buf, slen);
3465 if (clevel[slen-1] == '\n')
3466 slen--;
3467 clevel[slen] = 0;
b29bebd6 3468 if (kstrtol(clevel, 10, &level))
f2859af6 3469 level = LEVEL_NONE;
245f46c2 3470
f2859af6
DW		 3471 if (request_module("md-%s", clevel) != 0)
3472 request_module("md-level-%s", clevel);
245f46c2 3473 spin_lock(&pers_lock);
f2859af6 3474 pers = find_pers(level, clevel);
245f46c2
N		 3475 if (!pers || !try_module_get(pers->owner)) {
3476 spin_unlock(&pers_lock);
f2859af6 3477 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
6791875e
N		 3478 rv = -EINVAL;
3479 goto out_unlock;
245f46c2
N		 3480 }
3481 spin_unlock(&pers_lock);
3482
3483 if (pers == mddev->pers) {
3484 /* Nothing to do! */
3485 module_put(pers->owner);
6791875e
N		 3486 rv = len;
3487 goto out_unlock;
245f46c2
N		 3488 }
3489 if (!pers->takeover) {
3490 module_put(pers->owner);
3491 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3492 mdname(mddev), clevel);
6791875e
N		 3493 rv = -EINVAL;
3494 goto out_unlock;
245f46c2
N		 3495 }
3496
dafb20fa 3497 rdev_for_each(rdev, mddev)
e93f68a1
N		 3498 rdev->new_raid_disk = rdev->raid_disk;
3499
245f46c2
N		 3500 /* ->takeover must set new_* and/or delta_disks
3501 * if it succeeds, and may set them when it fails.
3502 */
3503 priv = pers->takeover(mddev);
3504 if (IS_ERR(priv)) {
3505 mddev->new_level = mddev->level;
3506 mddev->new_layout = mddev->layout;
664e7c41 3507 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3508 mddev->raid_disks -= mddev->delta_disks;
3509 mddev->delta_disks = 0;
2c810cdd 3510 mddev->reshape_backwards = 0;
245f46c2
N		 3511 module_put(pers->owner);
3512 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3513 mdname(mddev), clevel);
6791875e
N		 3514 rv = PTR_ERR(priv);
3515 goto out_unlock;
245f46c2
N		 3516 }
3517
3518 /* Looks like we have a winner */
3519 mddev_suspend(mddev);
5aa61f42 3520 mddev_detach(mddev);
36d091f4
N		 3521
3522 spin_lock(&mddev->lock);
db721d32
N		 3523 oldpers = mddev->pers;
3524 oldpriv = mddev->private;
3525 mddev->pers = pers;
3526 mddev->private = priv;
3527 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3528 mddev->level = mddev->new_level;
3529 mddev->layout = mddev->new_layout;
3530 mddev->chunk_sectors = mddev->new_chunk_sectors;
3531 mddev->delta_disks = 0;
3532 mddev->reshape_backwards = 0;
3533 mddev->degraded = 0;
36d091f4 3534 spin_unlock(&mddev->lock);
db721d32
N		 3535
3536 if (oldpers->sync_request == NULL &&
3537 mddev->external) {
3538 /* We are converting from a no-redundancy array
3539 * to a redundancy array and metadata is managed
3540 * externally so we need to be sure that writes
3541 * won't block due to a need to transition
3542 * clean->dirty
3543 * until external management is started.
3544 */
3545 mddev->in_sync = 0;
3546 mddev->safemode_delay = 0;
3547 mddev->safemode = 0;
3548 }
f72ffdd6 3549
db721d32
N		 3550 oldpers->free(mddev, oldpriv);
3551
3552 if (oldpers->sync_request == NULL &&
a64c876f
N		 3553 pers->sync_request != NULL) {
3554 /* need to add the md_redundancy_group */
3555 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3556 printk(KERN_WARNING
3557 "md: cannot register extra attributes for %s\n",
3558 mdname(mddev));
388975cc 3559 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
f72ffdd6 3560 }
db721d32 3561 if (oldpers->sync_request != NULL &&
a64c876f
N		 3562 pers->sync_request == NULL) {
3563 /* need to remove the md_redundancy_group */
3564 if (mddev->to_remove == NULL)
3565 mddev->to_remove = &md_redundancy_group;
3566 }
3567
dafb20fa 3568 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3569 if (rdev->raid_disk < 0)
3570 continue;
bf2cb0da 3571 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3572 rdev->new_raid_disk = -1;
3573 if (rdev->new_raid_disk == rdev->raid_disk)
3574 continue;
36fad858 3575 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3576 }
dafb20fa 3577 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3578 if (rdev->raid_disk < 0)
3579 continue;
3580 if (rdev->new_raid_disk == rdev->raid_disk)
3581 continue;
3582 rdev->raid_disk = rdev->new_raid_disk;
3583 if (rdev->raid_disk < 0)
3a981b03 3584 clear_bit(In_sync, &rdev->flags);
e93f68a1 3585 else {
36fad858
NK		 3586 if (sysfs_link_rdev(mddev, rdev))
3587 printk(KERN_WARNING "md: cannot register rd%d"
3588 " for %s after level change\n",
3589 rdev->raid_disk, mdname(mddev));
3a981b03 3590 }
e93f68a1
N		 3591 }
3592
db721d32 3593 if (pers->sync_request == NULL) {
9af204cf
TM		 3594 /* this is now an array without redundancy, so
3595 * it must always be in_sync
3596 */
3597 mddev->in_sync = 1;
3598 del_timer_sync(&mddev->safemode_timer);
3599 }
02e5f5c0 3600 blk_set_stacking_limits(&mddev->queue->limits);
245f46c2 3601 pers->run(mddev);
245f46c2 3602 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3603 mddev_resume(mddev);
830778a1
N		 3604 if (!mddev->thread)
3605 md_update_sb(mddev, 1);
5cac7861 3606 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3607 md_new_event(mddev);
6791875e
N		 3608 rv = len;
3609out_unlock:
3610 mddev_unlock(mddev);
d9d166c2
N		 3611 return rv;
3612}
3613
3614static struct md_sysfs_entry md_level =
80ca3a44 3615__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3616
d4dbd025 3617static ssize_t
fd01b88c 3618layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3619{
3620 /* just a number, not meaningful for all levels */
08a02ecd
N		 3621 if (mddev->reshape_position != MaxSector &&
3622 mddev->layout != mddev->new_layout)
3623 return sprintf(page, "%d (%d)\n",
3624 mddev->new_layout, mddev->layout);
d4dbd025
N		 3625 return sprintf(page, "%d\n", mddev->layout);
3626}
3627
3628static ssize_t
fd01b88c 3629layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025 3630{
4c9309c0 3631 unsigned int n;
6791875e 3632 int err;
d4dbd025 3633
4c9309c0
AD		 3634 err = kstrtouint(buf, 10, &n);
3635 if (err < 0)
3636 return err;
6791875e
N		 3637 err = mddev_lock(mddev);
3638 if (err)
3639 return err;
d4dbd025 3640
b3546035 3641 if (mddev->pers) {
50ac168a 3642 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3643 err = -EBUSY;
3644 else if (mddev->ro)
3645 err = -EROFS;
3646 else {
3647 mddev->new_layout = n;
3648 err = mddev->pers->check_reshape(mddev);
3649 if (err)
3650 mddev->new_layout = mddev->layout;
597a711b 3651 }
b3546035 3652 } else {
08a02ecd 3653 mddev->new_layout = n;
b3546035
N		 3654 if (mddev->reshape_position == MaxSector)
3655 mddev->layout = n;
3656 }
6791875e
N		 3657 mddev_unlock(mddev);
3658 return err ?: len;
d4dbd025
N		 3659}
3660static struct md_sysfs_entry md_layout =
80ca3a44 3661__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025 3662
eae1701f 3663static ssize_t
fd01b88c 3664raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3665{
bb636547
N		 3666 if (mddev->raid_disks == 0)
3667 return 0;
08a02ecd
N		 3668 if (mddev->reshape_position != MaxSector &&
3669 mddev->delta_disks != 0)
3670 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3671 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3672 return sprintf(page, "%d\n", mddev->raid_disks);
3673}
3674
fd01b88c 3675static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3676
3677static ssize_t
fd01b88c 3678raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3679{
4c9309c0 3680 unsigned int n;
6791875e 3681 int err;
da943b99 3682
4c9309c0
AD		 3683 err = kstrtouint(buf, 10, &n);
3684 if (err < 0)
3685 return err;
da943b99 3686
6791875e
N		 3687 err = mddev_lock(mddev);
3688 if (err)
3689 return err;
da943b99 3690 if (mddev->pers)
6791875e 3691 err = update_raid_disks(mddev, n);
08a02ecd 3692 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3693 struct md_rdev *rdev;
08a02ecd 3694 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c 3695
6791875e 3696 err = -EINVAL;
c6563a8c
N		 3697 rdev_for_each(rdev, mddev) {
3698 if (olddisks < n &&
3699 rdev->data_offset < rdev->new_data_offset)
6791875e 3700 goto out_unlock;
c6563a8c
N		 3701 if (olddisks > n &&
3702 rdev->data_offset > rdev->new_data_offset)
6791875e 3703 goto out_unlock;
c6563a8c 3704 }
6791875e 3705 err = 0;
08a02ecd
N		 3706 mddev->delta_disks = n - olddisks;
3707 mddev->raid_disks = n;
2c810cdd 3708 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3709 } else
da943b99 3710 mddev->raid_disks = n;
6791875e
N		 3711out_unlock:
3712 mddev_unlock(mddev);
3713 return err ? err : len;
da943b99
N		 3714}
3715static struct md_sysfs_entry md_raid_disks =
80ca3a44 3716__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3717
3b34380a 3718static ssize_t
fd01b88c 3719chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3720{
08a02ecd 3721 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3722 mddev->chunk_sectors != mddev->new_chunk_sectors)
3723 return sprintf(page, "%d (%d)\n",
3724 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3725 mddev->chunk_sectors << 9);
3726 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3727}
3728
3729static ssize_t
fd01b88c 3730chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3731{
4c9309c0 3732 unsigned long n;
6791875e 3733 int err;
3b34380a 3734
4c9309c0
AD		 3735 err = kstrtoul(buf, 10, &n);
3736 if (err < 0)
3737 return err;
3b34380a 3738
6791875e
N		 3739 err = mddev_lock(mddev);
3740 if (err)
3741 return err;
b3546035 3742 if (mddev->pers) {
50ac168a 3743 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3744 err = -EBUSY;
3745 else if (mddev->ro)
3746 err = -EROFS;
3747 else {
3748 mddev->new_chunk_sectors = n >> 9;
3749 err = mddev->pers->check_reshape(mddev);
3750 if (err)
3751 mddev->new_chunk_sectors = mddev->chunk_sectors;
597a711b 3752 }
b3546035 3753 } else {
664e7c41 3754 mddev->new_chunk_sectors = n >> 9;
b3546035 3755 if (mddev->reshape_position == MaxSector)
9d8f0363 3756 mddev->chunk_sectors = n >> 9;
b3546035 3757 }
6791875e
N		 3758 mddev_unlock(mddev);
3759 return err ?: len;
3b34380a
N		 3760}
3761static struct md_sysfs_entry md_chunk_size =
80ca3a44 3762__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3763
a94213b1 3764static ssize_t
fd01b88c 3765resync_start_show(struct mddev *mddev, char *page)
a94213b1 3766{
d1a7c503
N		 3767 if (mddev->recovery_cp == MaxSector)
3768 return sprintf(page, "none\n");
a94213b1
N		 3769 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3770}
3771
3772static ssize_t
fd01b88c 3773resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3774{
4c9309c0 3775 unsigned long long n;
6791875e 3776 int err;
4c9309c0
AD		 3777
3778 if (cmd_match(buf, "none"))
3779 n = MaxSector;
3780 else {
3781 err = kstrtoull(buf, 10, &n);
3782 if (err < 0)
3783 return err;
3784 if (n != (sector_t)n)
3785 return -EINVAL;
3786 }
a94213b1 3787
6791875e
N		 3788 err = mddev_lock(mddev);
3789 if (err)
3790 return err;
b098636c 3791 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6791875e 3792 err = -EBUSY;
a94213b1 3793
6791875e
N		 3794 if (!err) {
3795 mddev->recovery_cp = n;
3796 if (mddev->pers)
3797 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3798 }
3799 mddev_unlock(mddev);
3800 return err ?: len;
a94213b1
N		 3801}
3802static struct md_sysfs_entry md_resync_start =
750f199e
N		 3803__ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3804 resync_start_show, resync_start_store);
a94213b1 3805
9e653b63
N		 3806/*
3807 * The array state can be:
3808 *
3809 * clear
3810 * No devices, no size, no level
3811 * Equivalent to STOP_ARRAY ioctl
3812 * inactive
3813 * May have some settings, but array is not active
3814 * all IO results in error
3815 * When written, doesn't tear down array, but just stops it
3816 * suspended (not supported yet)
3817 * All IO requests will block. The array can be reconfigured.
910d8cb3 3818 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3819 * readonly
3820 * no resync can happen. no superblocks get written.
3821 * write requests fail
3822 * read-auto
3823 * like readonly, but behaves like 'clean' on a write request.
3824 *
3825 * clean - no pending writes, but otherwise active.
3826 * When written to inactive array, starts without resync
3827 * If a write request arrives then
3828 * if metadata is known, mark 'dirty' and switch to 'active'.
3829 * if not known, block and switch to write-pending
3830 * If written to an active array that has pending writes, then fails.
3831 * active
3832 * fully active: IO and resync can be happening.
3833 * When written to inactive array, starts with resync
3834 *
3835 * write-pending
3836 * clean, but writes are blocked waiting for 'active' to be written.
3837 *
3838 * active-idle
3839 * like active, but no writes have been seen for a while (100msec).
3840 *
3841 */
3842enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3843 write_pending, active_idle, bad_word};
05381954 3844static char *array_states[] = {
9e653b63
N		 3845 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3846 "write-pending", "active-idle", NULL };
3847
3848static int match_word(const char *word, char **list)
3849{
3850 int n;
3851 for (n=0; list[n]; n++)
3852 if (cmd_match(word, list[n]))
3853 break;
3854 return n;
3855}
3856
3857static ssize_t
fd01b88c 3858array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3859{
3860 enum array_state st = inactive;
3861
3862 if (mddev->pers)
3863 switch(mddev->ro) {
3864 case 1:
3865 st = readonly;
3866 break;
3867 case 2:
3868 st = read_auto;
3869 break;
3870 case 0:
3871 if (mddev->in_sync)
3872 st = clean;
070dc6dd 3873 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3874 st = write_pending;
9e653b63
N		 3875 else if (mddev->safemode)
3876 st = active_idle;
3877 else
3878 st = active;
3879 }
3880 else {
3881 if (list_empty(&mddev->disks) &&
3882 mddev->raid_disks == 0 &&
58c0fed4 3883 mddev->dev_sectors == 0)
9e653b63
N		 3884 st = clear;
3885 else
3886 st = inactive;
3887 }
3888 return sprintf(page, "%s\n", array_states[st]);
3889}
3890
f72ffdd6
N		 3891static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3892static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3893static int do_md_run(struct mddev *mddev);
fd01b88c 3894static int restart_array(struct mddev *mddev);
9e653b63
N		 3895
3896static ssize_t
fd01b88c 3897array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63 3898{
6791875e 3899 int err;
9e653b63 3900 enum array_state st = match_word(buf, array_states);
6791875e
N		 3901
3902 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3903 /* don't take reconfig_mutex when toggling between
3904 * clean and active
3905 */
3906 spin_lock(&mddev->lock);
3907 if (st == active) {
3908 restart_array(mddev);
3909 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3910 wake_up(&mddev->sb_wait);
3911 err = 0;
3912 } else /* st == clean */ {
3913 restart_array(mddev);
3914 if (atomic_read(&mddev->writes_pending) == 0) {
3915 if (mddev->in_sync == 0) {
3916 mddev->in_sync = 1;
3917 if (mddev->safemode == 1)
3918 mddev->safemode = 0;
3919 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3920 }
3921 err = 0;
3922 } else
3923 err = -EBUSY;
3924 }
3925 spin_unlock(&mddev->lock);
c008f1d3 3926 return err ?: len;
6791875e
N		 3927 }
3928 err = mddev_lock(mddev);
3929 if (err)
3930 return err;
3931 err = -EINVAL;
9e653b63
N		 3932 switch(st) {
3933 case bad_word:
3934 break;
3935 case clear:
3936 /* stopping an active array */
a05b7ea0 3937 err = do_md_stop(mddev, 0, NULL);
9e653b63
N		 3938 break;
3939 case inactive:
3940 /* stopping an active array */
90cf195d 3941 if (mddev->pers)
a05b7ea0 3942 err = do_md_stop(mddev, 2, NULL);
90cf195d 3943 else
e691063a 3944 err = 0; /* already inactive */
9e653b63
N		 3945 break;
3946 case suspended:
3947 break; /* not supported yet */
3948 case readonly:
3949 if (mddev->pers)
a05b7ea0 3950 err = md_set_readonly(mddev, NULL);
9e653b63
N		 3951 else {
3952 mddev->ro = 1;
648b629e 3953 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3954 err = do_md_run(mddev);
3955 }
3956 break;
3957 case read_auto:
9e653b63 3958 if (mddev->pers) {
80268ee9 3959 if (mddev->ro == 0)
a05b7ea0 3960 err = md_set_readonly(mddev, NULL);
80268ee9 3961 else if (mddev->ro == 1)
648b629e
N		 3962 err = restart_array(mddev);
3963 if (err == 0) {
3964 mddev->ro = 2;
3965 set_disk_ro(mddev->gendisk, 0);
3966 }
9e653b63
N		 3967 } else {
3968 mddev->ro = 2;
3969 err = do_md_run(mddev);
3970 }
3971 break;
3972 case clean:
3973 if (mddev->pers) {
339421de
SL		 3974 err = restart_array(mddev);
3975 if (err)
3976 break;
85572d7c 3977 spin_lock(&mddev->lock);
9e653b63 3978 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3979 if (mddev->in_sync == 0) {
3980 mddev->in_sync = 1;
31a59e34
N		 3981 if (mddev->safemode == 1)
3982 mddev->safemode = 0;
070dc6dd 3983 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3984 }
3985 err = 0;
3986 } else
3987 err = -EBUSY;
85572d7c 3988 spin_unlock(&mddev->lock);
5bf29597
N		 3989 } else
3990 err = -EINVAL;
9e653b63
N		 3991 break;
3992 case active:
3993 if (mddev->pers) {
339421de
SL		 3994 err = restart_array(mddev);
3995 if (err)
3996 break;
070dc6dd 3997 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3998 wake_up(&mddev->sb_wait);
3999 err = 0;
4000 } else {
4001 mddev->ro = 0;
648b629e 4002 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 4003 err = do_md_run(mddev);
4004 }
4005 break;
4006 case write_pending:
4007 case active_idle:
4008 /* these cannot be set */
4009 break;
4010 }
6791875e
N		 4011
4012 if (!err) {
1d23f178
N		 4013 if (mddev->hold_active == UNTIL_IOCTL)
4014 mddev->hold_active = 0;
00bcb4ac 4015 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4016 }
6791875e
N		 4017 mddev_unlock(mddev);
4018 return err ?: len;
9e653b63 4019}
80ca3a44 4020static struct md_sysfs_entry md_array_state =
750f199e 4021__ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 4022
1e50915f 4023static ssize_t
fd01b88c 4024max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 4025 return sprintf(page, "%d\n",
4026 atomic_read(&mddev->max_corr_read_errors));
4027}
4028
4029static ssize_t
fd01b88c 4030max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f 4031{
4c9309c0
AD		 4032 unsigned int n;
4033 int rv;
1e50915f 4034
4c9309c0
AD		 4035 rv = kstrtouint(buf, 10, &n);
4036 if (rv < 0)
4037 return rv;
4038 atomic_set(&mddev->max_corr_read_errors, n);
4039 return len;
1e50915f
RB		 4040}
4041
4042static struct md_sysfs_entry max_corr_read_errors =
4043__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4044 max_corrected_read_errors_store);
4045
6d7ff738 4046static ssize_t
fd01b88c 4047null_show(struct mddev *mddev, char *page)
6d7ff738
N		 4048{
4049 return -EINVAL;
4050}
4051
4052static ssize_t
fd01b88c 4053new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 4054{
4055 /* buf must be %d:%d\n? giving major and minor numbers */
4056 /* The new device is added to the array.
4057 * If the array has a persistent superblock, we read the
4058 * superblock to initialise info and check validity.
4059 * Otherwise, only checking done is that in bind_rdev_to_array,
4060 * which mainly checks size.
4061 */
4062 char *e;
4063 int major = simple_strtoul(buf, &e, 10);
4064 int minor;
4065 dev_t dev;
3cb03002 4066 struct md_rdev *rdev;
6d7ff738
N		 4067 int err;
4068
4069 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4070 return -EINVAL;
4071 minor = simple_strtoul(e+1, &e, 10);
4072 if (*e && *e != '\n')
4073 return -EINVAL;
4074 dev = MKDEV(major, minor);
4075 if (major != MAJOR(dev) ||
4076 minor != MINOR(dev))
4077 return -EOVERFLOW;
4078
6791875e
N		 4079 flush_workqueue(md_misc_wq);
4080
4081 err = mddev_lock(mddev);
4082 if (err)
4083 return err;
6d7ff738
N		 4084 if (mddev->persistent) {
4085 rdev = md_import_device(dev, mddev->major_version,
4086 mddev->minor_version);
4087 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 4088 struct md_rdev *rdev0
4089 = list_entry(mddev->disks.next,
4090 struct md_rdev, same_set);
6d7ff738
N		 4091 err = super_types[mddev->major_version]
4092 .load_super(rdev, rdev0, mddev->minor_version);
4093 if (err < 0)
4094 goto out;
4095 }
c5d79adb
N		 4096 } else if (mddev->external)
4097 rdev = md_import_device(dev, -2, -1);
4098 else
6d7ff738
N		 4099 rdev = md_import_device(dev, -1, -1);
4100
9a8c0fa8
N		 4101 if (IS_ERR(rdev)) {
4102 mddev_unlock(mddev);
6d7ff738 4103 return PTR_ERR(rdev);
9a8c0fa8 4104 }
6d7ff738
N		 4105 err = bind_rdev_to_array(rdev, mddev);
4106 out:
4107 if (err)
4108 export_rdev(rdev);
6791875e 4109 mddev_unlock(mddev);
6d7ff738
N		 4110 return err ? err : len;
4111}
4112
4113static struct md_sysfs_entry md_new_device =
80ca3a44 4114__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 4115
9b1d1dac 4116static ssize_t
fd01b88c 4117bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 4118{
4119 char *end;
4120 unsigned long chunk, end_chunk;
6791875e 4121 int err;
9b1d1dac 4122
6791875e
N		 4123 err = mddev_lock(mddev);
4124 if (err)
4125 return err;
9b1d1dac
PC		 4126 if (!mddev->bitmap)
4127 goto out;
4128 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4129 while (*buf) {
4130 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4131 if (buf == end) break;
4132 if (*end == '-') { /* range */
4133 buf = end + 1;
4134 end_chunk = simple_strtoul(buf, &end, 0);
4135 if (buf == end) break;
4136 }
4137 if (*end && !isspace(*end)) break;
4138 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 4139 buf = skip_spaces(end);
9b1d1dac
PC		 4140 }
4141 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4142out:
6791875e 4143 mddev_unlock(mddev);
9b1d1dac
PC		 4144 return len;
4145}
4146
4147static struct md_sysfs_entry md_bitmap =
4148__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4149
a35b0d69 4150static ssize_t
fd01b88c 4151size_show(struct mddev *mddev, char *page)
a35b0d69 4152{
58c0fed4
AN		 4153 return sprintf(page, "%llu\n",
4154 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 4155}
4156
fd01b88c 4157static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 4158
4159static ssize_t
fd01b88c 4160size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 4161{
4162 /* If array is inactive, we can reduce the component size, but
4163 * not increase it (except from 0).
4164 * If array is active, we can try an on-line resize
4165 */
b522adcd
DW		 4166 sector_t sectors;
4167 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4168
58c0fed4
AN		 4169 if (err < 0)
4170 return err;
6791875e
N		 4171 err = mddev_lock(mddev);
4172 if (err)
4173 return err;
a35b0d69 4174 if (mddev->pers) {
58c0fed4 4175 err = update_size(mddev, sectors);
850b2b42 4176 md_update_sb(mddev, 1);
a35b0d69 4177 } else {
58c0fed4
AN		 4178 if (mddev->dev_sectors == 0 ||
4179 mddev->dev_sectors > sectors)
4180 mddev->dev_sectors = sectors;
a35b0d69
N		 4181 else
4182 err = -ENOSPC;
4183 }
6791875e 4184 mddev_unlock(mddev);
a35b0d69
N		 4185 return err ? err : len;
4186}
4187
4188static struct md_sysfs_entry md_size =
80ca3a44 4189__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4190
83f0d77a 4191/* Metadata version.
e691063a
N		 4192 * This is one of
4193 * 'none' for arrays with no metadata (good luck...)
4194 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4195 * or N.M for internally known formats
4196 */
4197static ssize_t
fd01b88c 4198metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4199{
4200 if (mddev->persistent)
4201 return sprintf(page, "%d.%d\n",
4202 mddev->major_version, mddev->minor_version);
e691063a
N		 4203 else if (mddev->external)
4204 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4205 else
4206 return sprintf(page, "none\n");
4207}
4208
4209static ssize_t
fd01b88c 4210metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4211{
4212 int major, minor;
4213 char *e;
6791875e 4214 int err;
ea43ddd8
N		 4215 /* Changing the details of 'external' metadata is
4216 * always permitted. Otherwise there must be
4217 * no devices attached to the array.
4218 */
6791875e
N		 4219
4220 err = mddev_lock(mddev);
4221 if (err)
4222 return err;
4223 err = -EBUSY;
ea43ddd8
N		 4224 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4225 ;
4226 else if (!list_empty(&mddev->disks))
6791875e 4227 goto out_unlock;
8bb93aac 4228
6791875e 4229 err = 0;
8bb93aac
N		 4230 if (cmd_match(buf, "none")) {
4231 mddev->persistent = 0;
e691063a
N		 4232 mddev->external = 0;
4233 mddev->major_version = 0;
4234 mddev->minor_version = 90;
6791875e 4235 goto out_unlock;
e691063a
N		 4236 }
4237 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4238 size_t namelen = len-9;
e691063a
N		 4239 if (namelen >= sizeof(mddev->metadata_type))
4240 namelen = sizeof(mddev->metadata_type)-1;
4241 strncpy(mddev->metadata_type, buf+9, namelen);
4242 mddev->metadata_type[namelen] = 0;
4243 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4244 mddev->metadata_type[--namelen] = 0;
4245 mddev->persistent = 0;
4246 mddev->external = 1;
8bb93aac
N		 4247 mddev->major_version = 0;
4248 mddev->minor_version = 90;
6791875e 4249 goto out_unlock;
8bb93aac
N		 4250 }
4251 major = simple_strtoul(buf, &e, 10);
6791875e 4252 err = -EINVAL;
8bb93aac 4253 if (e==buf || *e != '.')
6791875e 4254 goto out_unlock;
8bb93aac
N		 4255 buf = e+1;
4256 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4257 if (e==buf || (*e && *e != '\n') )
6791875e
N		 4258 goto out_unlock;
4259 err = -ENOENT;
50511da3 4260 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
6791875e 4261 goto out_unlock;
8bb93aac
N		 4262 mddev->major_version = major;
4263 mddev->minor_version = minor;
4264 mddev->persistent = 1;
e691063a 4265 mddev->external = 0;
6791875e
N		 4266 err = 0;
4267out_unlock:
4268 mddev_unlock(mddev);
4269 return err ?: len;
8bb93aac
N		 4270}
4271
4272static struct md_sysfs_entry md_metadata =
750f199e 4273__ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4274
24dd469d 4275static ssize_t
fd01b88c 4276action_show(struct mddev *mddev, char *page)
24dd469d 4277{
7eec314d 4278 char *type = "idle";
b7b17c9b
N		 4279 unsigned long recovery = mddev->recovery;
4280 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
b6a9ce68 4281 type = "frozen";
b7b17c9b
N		 4282 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4283 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4284 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
ccfcc3c1 4285 type = "reshape";
b7b17c9b
N		 4286 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4287 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
24dd469d 4288 type = "resync";
b7b17c9b 4289 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
24dd469d
N		 4290 type = "check";
4291 else
4292 type = "repair";
b7b17c9b 4293 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
24dd469d 4294 type = "recover";
985ca973
N		 4295 else if (mddev->reshape_position != MaxSector)
4296 type = "reshape";
24dd469d
N		 4297 }
4298 return sprintf(page, "%s\n", type);
4299}
4300
4301static ssize_t
fd01b88c 4302action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4303{
7eec314d
N		 4304 if (!mddev->pers || !mddev->pers->sync_request)
4305 return -EINVAL;
4306
b6a9ce68
N		 4307
4308 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
56ccc112
N		 4309 if (cmd_match(page, "frozen"))
4310 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4311 else
4312 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
8e8e2518
N		 4313 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4314 mddev_lock(mddev) == 0) {
4315 flush_workqueue(md_misc_wq);
4316 if (mddev->sync_thread) {
4317 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6791875e 4318 md_reap_sync_thread(mddev);
6791875e 4319 }
8e8e2518 4320 mddev_unlock(mddev);
7eec314d 4321 }
03c902e1
N		 4322 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4323 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4324 return -EBUSY;
72a23c21 4325 else if (cmd_match(page, "resync"))
56ccc112 4326 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
72a23c21 4327 else if (cmd_match(page, "recover")) {
56ccc112 4328 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
72a23c21 4329 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
72a23c21 4330 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4331 int err;
4332 if (mddev->pers->start_reshape == NULL)
4333 return -EINVAL;
6791875e
N		 4334 err = mddev_lock(mddev);
4335 if (!err) {
56ccc112 4336 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6791875e
N		 4337 err = mddev->pers->start_reshape(mddev);
4338 mddev_unlock(mddev);
4339 }
16484bf5
N		 4340 if (err)
4341 return err;
a99ac971 4342 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4343 } else {
bce74dac 4344 if (cmd_match(page, "check"))
7eec314d 4345 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4346 else if (!cmd_match(page, "repair"))
7eec314d 4347 return -EINVAL;
56ccc112 4348 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
7eec314d
N		 4349 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4350 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4351 }
48c26ddc
N		 4352 if (mddev->ro == 2) {
4353 /* A write to sync_action is enough to justify
4354 * canceling read-auto mode
4355 */
4356 mddev->ro = 0;
4357 md_wakeup_thread(mddev->sync_thread);
4358 }
03c902e1 4359 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4360 md_wakeup_thread(mddev->thread);
00bcb4ac 4361 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4362 return len;
4363}
4364
c4a39551 4365static struct md_sysfs_entry md_scan_mode =
750f199e 4366__ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
c4a39551
JB		 4367
4368static ssize_t
4369last_sync_action_show(struct mddev *mddev, char *page)
4370{
4371 return sprintf(page, "%s\n", mddev->last_sync_action);
4372}
4373
4374static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4375
9d88883e 4376static ssize_t
fd01b88c 4377mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4378{
4379 return sprintf(page, "%llu\n",
7f7583d4
JM		 4380 (unsigned long long)
4381 atomic64_read(&mddev->resync_mismatches));
9d88883e
N		 4382}
4383
80ca3a44 4384static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4385
88202a0c 4386static ssize_t
fd01b88c 4387sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4388{
4389 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4390 mddev->sync_speed_min ? "local": "system");
4391}
4392
4393static ssize_t
fd01b88c 4394sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c 4395{
4c9309c0
AD		 4396 unsigned int min;
4397 int rv;
4398
88202a0c 4399 if (strncmp(buf, "system", 6)==0) {
4c9309c0
AD		 4400 min = 0;
4401 } else {
4402 rv = kstrtouint(buf, 10, &min);
4403 if (rv < 0)
4404 return rv;
4405 if (min == 0)
4406 return -EINVAL;
88202a0c 4407 }
88202a0c
N		 4408 mddev->sync_speed_min = min;
4409 return len;
4410}
4411
4412static struct md_sysfs_entry md_sync_min =
4413__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4414
4415static ssize_t
fd01b88c 4416sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4417{
4418 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4419 mddev->sync_speed_max ? "local": "system");
4420}
4421
4422static ssize_t
fd01b88c 4423sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c 4424{
4c9309c0
AD		 4425 unsigned int max;
4426 int rv;
4427
88202a0c 4428 if (strncmp(buf, "system", 6)==0) {
4c9309c0
AD		 4429 max = 0;
4430 } else {
4431 rv = kstrtouint(buf, 10, &max);
4432 if (rv < 0)
4433 return rv;
4434 if (max == 0)
4435 return -EINVAL;
88202a0c 4436 }
88202a0c
N		 4437 mddev->sync_speed_max = max;
4438 return len;
4439}
4440
4441static struct md_sysfs_entry md_sync_max =
4442__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4443
d7f3d291 4444static ssize_t
fd01b88c 4445degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4446{
4447 return sprintf(page, "%d\n", mddev->degraded);
4448}
4449static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4450
90b08710 4451static ssize_t
fd01b88c 4452sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4453{
4454 return sprintf(page, "%d\n", mddev->parallel_resync);
4455}
4456
4457static ssize_t
fd01b88c 4458sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4459{
4460 long n;
4461
b29bebd6 4462 if (kstrtol(buf, 10, &n))
90b08710
BS		 4463 return -EINVAL;
4464
4465 if (n != 0 && n != 1)
4466 return -EINVAL;
4467
4468 mddev->parallel_resync = n;
4469
4470 if (mddev->sync_thread)
4471 wake_up(&resync_wait);
4472
4473 return len;
4474}
4475
4476/* force parallel resync, even with shared block devices */
4477static struct md_sysfs_entry md_sync_force_parallel =
4478__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4479 sync_force_parallel_show, sync_force_parallel_store);
4480
88202a0c 4481static ssize_t
fd01b88c 4482sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4483{
4484 unsigned long resync, dt, db;
d1a7c503
N		 4485 if (mddev->curr_resync == 0)
4486 return sprintf(page, "none\n");
9687a60c
AN		 4487 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4488 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4489 if (!dt) dt++;
9687a60c
AN		 4490 db = resync - mddev->resync_mark_cnt;
4491 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4492}
4493
80ca3a44 4494static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4495
4496static ssize_t
fd01b88c 4497sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4498{
13ae864b 4499 unsigned long long max_sectors, resync;
88202a0c 4500
acb180b0
N		 4501 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4502 return sprintf(page, "none\n");
4503
72f36d59
N		 4504 if (mddev->curr_resync == 1 ||
4505 mddev->curr_resync == 2)
4506 return sprintf(page, "delayed\n");
4507
c804cdec
N		 4508 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4509 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 4510 max_sectors = mddev->resync_max_sectors;
88202a0c 4511 else
58c0fed4 4512 max_sectors = mddev->dev_sectors;
88202a0c 4513
acb180b0 4514 resync = mddev->curr_resync_completed;
13ae864b 4515 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4516}
4517
750f199e
N		 4518static struct md_sysfs_entry md_sync_completed =
4519 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
88202a0c 4520
5e96ee65 4521static ssize_t
fd01b88c 4522min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4523{
4524 return sprintf(page, "%llu\n",
4525 (unsigned long long)mddev->resync_min);
4526}
4527static ssize_t
fd01b88c 4528min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4529{
4530 unsigned long long min;
23da422b 4531 int err;
23da422b 4532
b29bebd6 4533 if (kstrtoull(buf, 10, &min))
5e96ee65 4534 return -EINVAL;
23da422b
N		 4535
4536 spin_lock(&mddev->lock);
4537 err = -EINVAL;
5e96ee65 4538 if (min > mddev->resync_max)
23da422b
N		 4539 goto out_unlock;
4540
4541 err = -EBUSY;
5e96ee65 4542 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4543 goto out_unlock;
5e96ee65 4544
50c37b13
N		 4545 /* Round down to multiple of 4K for safety */
4546 mddev->resync_min = round_down(min, 8);
23da422b 4547 err = 0;
5e96ee65 4548
23da422b
N		 4549out_unlock:
4550 spin_unlock(&mddev->lock);
4551 return err ?: len;
5e96ee65
NB		 4552}
4553
4554static struct md_sysfs_entry md_min_sync =
4555__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4556
c6207277 4557static ssize_t
fd01b88c 4558max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4559{
4560 if (mddev->resync_max == MaxSector)
4561 return sprintf(page, "max\n");
4562 else
4563 return sprintf(page, "%llu\n",
4564 (unsigned long long)mddev->resync_max);
4565}
4566static ssize_t
fd01b88c 4567max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277 4568{
23da422b
N		 4569 int err;
4570 spin_lock(&mddev->lock);
c6207277
N		 4571 if (strncmp(buf, "max", 3) == 0)
4572 mddev->resync_max = MaxSector;
4573 else {
5e96ee65 4574 unsigned long long max;
23da422b
N		 4575 int chunk;
4576
4577 err = -EINVAL;
b29bebd6 4578 if (kstrtoull(buf, 10, &max))
23da422b 4579 goto out_unlock;
5e96ee65 4580 if (max < mddev->resync_min)
23da422b
N		 4581 goto out_unlock;
4582
4583 err = -EBUSY;
c6207277 4584 if (max < mddev->resync_max &&
4d484a4a 4585 mddev->ro == 0 &&
c6207277 4586 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4587 goto out_unlock;
c6207277
N		 4588
4589 /* Must be a multiple of chunk_size */
23da422b
N		 4590 chunk = mddev->chunk_sectors;
4591 if (chunk) {
2ac06c33 4592 sector_t temp = max;
23da422b
N		 4593
4594 err = -EINVAL;
4595 if (sector_div(temp, chunk))
4596 goto out_unlock;
c6207277
N		 4597 }
4598 mddev->resync_max = max;
4599 }
4600 wake_up(&mddev->recovery_wait);
23da422b
N		 4601 err = 0;
4602out_unlock:
4603 spin_unlock(&mddev->lock);
4604 return err ?: len;
c6207277
N		 4605}
4606
4607static struct md_sysfs_entry md_max_sync =
4608__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4609
e464eafd 4610static ssize_t
fd01b88c 4611suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4612{
4613 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4614}
4615
4616static ssize_t
fd01b88c 4617suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd 4618{
4c9309c0 4619 unsigned long long old, new;
6791875e 4620 int err;
e464eafd 4621
4c9309c0
AD		 4622 err = kstrtoull(buf, 10, &new);
4623 if (err < 0)
4624 return err;
4625 if (new != (sector_t)new)
e464eafd 4626 return -EINVAL;
23ddff37 4627
6791875e
N		 4628 err = mddev_lock(mddev);
4629 if (err)
4630 return err;
4631 err = -EINVAL;
4632 if (mddev->pers == NULL ||
4633 mddev->pers->quiesce == NULL)
4634 goto unlock;
4635 old = mddev->suspend_lo;
23ddff37
N		 4636 mddev->suspend_lo = new;
4637 if (new >= old)
4638 /* Shrinking suspended region */
e464eafd 4639 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4640 else {
4641 /* Expanding suspended region - need to wait */
4642 mddev->pers->quiesce(mddev, 1);
4643 mddev->pers->quiesce(mddev, 0);
4644 }
6791875e
N		 4645 err = 0;
4646unlock:
4647 mddev_unlock(mddev);
4648 return err ?: len;
e464eafd
N		 4649}
4650static struct md_sysfs_entry md_suspend_lo =
4651__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4652
e464eafd 4653static ssize_t
fd01b88c 4654suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4655{
4656 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4657}
4658
4659static ssize_t
fd01b88c 4660suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd 4661{
4c9309c0 4662 unsigned long long old, new;
6791875e 4663 int err;
e464eafd 4664
4c9309c0
AD		 4665 err = kstrtoull(buf, 10, &new);
4666 if (err < 0)
4667 return err;
4668 if (new != (sector_t)new)
e464eafd 4669 return -EINVAL;
23ddff37 4670
6791875e
N		 4671 err = mddev_lock(mddev);
4672 if (err)
4673 return err;
4674 err = -EINVAL;
4675 if (mddev->pers == NULL ||
4676 mddev->pers->quiesce == NULL)
4677 goto unlock;
4678 old = mddev->suspend_hi;
23ddff37
N		 4679 mddev->suspend_hi = new;
4680 if (new <= old)
4681 /* Shrinking suspended region */
4682 mddev->pers->quiesce(mddev, 2);
4683 else {
4684 /* Expanding suspended region - need to wait */
e464eafd
N		 4685 mddev->pers->quiesce(mddev, 1);
4686 mddev->pers->quiesce(mddev, 0);
23ddff37 4687 }
6791875e
N		 4688 err = 0;
4689unlock:
4690 mddev_unlock(mddev);
4691 return err ?: len;
e464eafd
N		 4692}
4693static struct md_sysfs_entry md_suspend_hi =
4694__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4695
08a02ecd 4696static ssize_t
fd01b88c 4697reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4698{
4699 if (mddev->reshape_position != MaxSector)
4700 return sprintf(page, "%llu\n",
4701 (unsigned long long)mddev->reshape_position);
4702 strcpy(page, "none\n");
4703 return 5;
4704}
4705
4706static ssize_t
fd01b88c 4707reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd 4708{
c6563a8c 4709 struct md_rdev *rdev;
4c9309c0 4710 unsigned long long new;
6791875e 4711 int err;
6791875e 4712
4c9309c0
AD		 4713 err = kstrtoull(buf, 10, &new);
4714 if (err < 0)
4715 return err;
4716 if (new != (sector_t)new)
08a02ecd 4717 return -EINVAL;
6791875e
N		 4718 err = mddev_lock(mddev);
4719 if (err)
4720 return err;
4721 err = -EBUSY;
4722 if (mddev->pers)
4723 goto unlock;
08a02ecd
N		 4724 mddev->reshape_position = new;
4725 mddev->delta_disks = 0;
2c810cdd 4726 mddev->reshape_backwards = 0;
08a02ecd
N		 4727 mddev->new_level = mddev->level;
4728 mddev->new_layout = mddev->layout;
664e7c41 4729 mddev->new_chunk_sectors = mddev->chunk_sectors;
c6563a8c
N		 4730 rdev_for_each(rdev, mddev)
4731 rdev->new_data_offset = rdev->data_offset;
6791875e
N		 4732 err = 0;
4733unlock:
4734 mddev_unlock(mddev);
4735 return err ?: len;
08a02ecd
N		 4736}
4737
4738static struct md_sysfs_entry md_reshape_position =
4739__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4740 reshape_position_store);
4741
2c810cdd
N		 4742static ssize_t
4743reshape_direction_show(struct mddev *mddev, char *page)
4744{
4745 return sprintf(page, "%s\n",
4746 mddev->reshape_backwards ? "backwards" : "forwards");
4747}
4748
4749static ssize_t
4750reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4751{
4752 int backwards = 0;
6791875e
N		 4753 int err;
4754
2c810cdd
N		 4755 if (cmd_match(buf, "forwards"))
4756 backwards = 0;
4757 else if (cmd_match(buf, "backwards"))
4758 backwards = 1;
4759 else
4760 return -EINVAL;
4761 if (mddev->reshape_backwards == backwards)
4762 return len;
4763
6791875e
N		 4764 err = mddev_lock(mddev);
4765 if (err)
4766 return err;
2c810cdd
N		 4767 /* check if we are allowed to change */
4768 if (mddev->delta_disks)
6791875e
N		 4769 err = -EBUSY;
4770 else if (mddev->persistent &&
2c810cdd 4771 mddev->major_version == 0)
6791875e
N		 4772 err = -EINVAL;
4773 else
4774 mddev->reshape_backwards = backwards;
4775 mddev_unlock(mddev);
4776 return err ?: len;
2c810cdd
N		 4777}
4778
4779static struct md_sysfs_entry md_reshape_direction =
4780__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4781 reshape_direction_store);
4782
b522adcd 4783static ssize_t
fd01b88c 4784array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4785{
4786 if (mddev->external_size)
4787 return sprintf(page, "%llu\n",
4788 (unsigned long long)mddev->array_sectors/2);
4789 else
4790 return sprintf(page, "default\n");
4791}
4792
4793static ssize_t
fd01b88c 4794array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4795{
4796 sector_t sectors;
6791875e
N		 4797 int err;
4798
4799 err = mddev_lock(mddev);
4800 if (err)
4801 return err;
b522adcd
DW		 4802
4803 if (strncmp(buf, "default", 7) == 0) {
4804 if (mddev->pers)
4805 sectors = mddev->pers->size(mddev, 0, 0);
4806 else
4807 sectors = mddev->array_sectors;
4808
4809 mddev->external_size = 0;
4810 } else {
4811 if (strict_blocks_to_sectors(buf, &sectors) < 0)
6791875e
N		 4812 err = -EINVAL;
4813 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4814 err = -E2BIG;
4815 else
4816 mddev->external_size = 1;
b522adcd
DW		 4817 }
4818
6791875e
N		 4819 if (!err) {
4820 mddev->array_sectors = sectors;
4821 if (mddev->pers) {
4822 set_capacity(mddev->gendisk, mddev->array_sectors);
4823 revalidate_disk(mddev->gendisk);
4824 }
cbe6ef1d 4825 }
6791875e
N		 4826 mddev_unlock(mddev);
4827 return err ?: len;
b522adcd
DW		 4828}
4829
4830static struct md_sysfs_entry md_array_size =
4831__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4832 array_size_store);
e464eafd 4833
eae1701f
N		 4834static struct attribute *md_default_attrs[] = {
4835 &md_level.attr,
d4dbd025 4836 &md_layout.attr,
eae1701f 4837 &md_raid_disks.attr,
3b34380a 4838 &md_chunk_size.attr,
a35b0d69 4839 &md_size.attr,
a94213b1 4840 &md_resync_start.attr,
8bb93aac 4841 &md_metadata.attr,
6d7ff738 4842 &md_new_device.attr,
16f17b39 4843 &md_safe_delay.attr,
9e653b63 4844 &md_array_state.attr,
08a02ecd 4845 &md_reshape_position.attr,
2c810cdd 4846 &md_reshape_direction.attr,
b522adcd 4847 &md_array_size.attr,
1e50915f 4848 &max_corr_read_errors.attr,
411036fa
N		 4849 NULL,
4850};
4851
4852static struct attribute *md_redundancy_attrs[] = {
24dd469d 4853 &md_scan_mode.attr,
c4a39551 4854 &md_last_scan_mode.attr,
9d88883e 4855 &md_mismatches.attr,
88202a0c
N		 4856 &md_sync_min.attr,
4857 &md_sync_max.attr,
4858 &md_sync_speed.attr,
90b08710 4859 &md_sync_force_parallel.attr,
88202a0c 4860 &md_sync_completed.attr,
5e96ee65 4861 &md_min_sync.attr,
c6207277 4862 &md_max_sync.attr,
e464eafd
N		 4863 &md_suspend_lo.attr,
4864 &md_suspend_hi.attr,
9b1d1dac 4865 &md_bitmap.attr,
d7f3d291 4866 &md_degraded.attr,
eae1701f
N		 4867 NULL,
4868};
411036fa
N		 4869static struct attribute_group md_redundancy_group = {
4870 .name = NULL,
4871 .attrs = md_redundancy_attrs,
4872};
4873
eae1701f
N		 4874static ssize_t
4875md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4876{
4877 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4878 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4879 ssize_t rv;
eae1701f
N		 4880
4881 if (!entry->show)
4882 return -EIO;
af8a2434
N		 4883 spin_lock(&all_mddevs_lock);
4884 if (list_empty(&mddev->all_mddevs)) {
4885 spin_unlock(&all_mddevs_lock);
4886 return -EBUSY;
4887 }
4888 mddev_get(mddev);
4889 spin_unlock(&all_mddevs_lock);
4890
b7b17c9b 4891 rv = entry->show(mddev, page);
af8a2434 4892 mddev_put(mddev);
96de1e66 4893 return rv;
eae1701f
N		 4894}
4895
4896static ssize_t
4897md_attr_store(struct kobject *kobj, struct attribute *attr,
4898 const char *page, size_t length)
4899{
4900 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4901 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4902 ssize_t rv;
eae1701f
N		 4903
4904 if (!entry->store)
4905 return -EIO;
67463acb
N		 4906 if (!capable(CAP_SYS_ADMIN))
4907 return -EACCES;
af8a2434
N		 4908 spin_lock(&all_mddevs_lock);
4909 if (list_empty(&mddev->all_mddevs)) {
4910 spin_unlock(&all_mddevs_lock);
4911 return -EBUSY;
4912 }
4913 mddev_get(mddev);
4914 spin_unlock(&all_mddevs_lock);
6791875e 4915 rv = entry->store(mddev, page, length);
af8a2434 4916 mddev_put(mddev);
96de1e66 4917 return rv;
eae1701f
N		 4918}
4919
4920static void md_free(struct kobject *ko)
4921{
fd01b88c 4922 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4923
4924 if (mddev->sysfs_state)
4925 sysfs_put(mddev->sysfs_state);
4926
6cd18e71
N		 4927 if (mddev->queue)
4928 blk_cleanup_queue(mddev->queue);
a21d1504
N		 4929 if (mddev->gendisk) {
4930 del_gendisk(mddev->gendisk);
4931 put_disk(mddev->gendisk);
4932 }
a21d1504 4933
eae1701f
N		 4934 kfree(mddev);
4935}
4936
52cf25d0 4937static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4938 .show = md_attr_show,
4939 .store = md_attr_store,
4940};
4941static struct kobj_type md_ktype = {
4942 .release = md_free,
4943 .sysfs_ops = &md_sysfs_ops,
4944 .default_attrs = md_default_attrs,
4945};
4946
1da177e4
LT		 4947int mdp_major = 0;
4948
5fd3a17e
DW		 4949static void mddev_delayed_delete(struct work_struct *ws)
4950{
fd01b88c 4951 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4952
43a70507 4953 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4954 kobject_del(&mddev->kobj);
4955 kobject_put(&mddev->kobj);
4956}
4957
efeb53c0 4958static int md_alloc(dev_t dev, char *name)
1da177e4 4959{
48c9c27b 4960 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4961 struct mddev *mddev = mddev_find(dev);
1da177e4 4962 struct gendisk *disk;
efeb53c0
N		 4963 int partitioned;
4964 int shift;
4965 int unit;
3830c62f 4966 int error;
1da177e4
LT		 4967
4968 if (!mddev)
efeb53c0
N		 4969 return -ENODEV;
4970
4971 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4972 shift = partitioned ? MdpMinorShift : 0;
4973 unit = MINOR(mddev->unit) >> shift;
1da177e4 4974
e804ac78
TH		 4975 /* wait for any previous instance of this device to be
4976 * completely removed (mddev_delayed_delete).
d3374825 4977 */
e804ac78 4978 flush_workqueue(md_misc_wq);
d3374825 4979
48c9c27b 4980 mutex_lock(&disks_mutex);
0909dc44
N		 4981 error = -EEXIST;
4982 if (mddev->gendisk)
4983 goto abort;
efeb53c0
N		 4984
4985 if (name) {
4986 /* Need to ensure that 'name' is not a duplicate.
4987 */
fd01b88c 4988 struct mddev *mddev2;
efeb53c0
N		 4989 spin_lock(&all_mddevs_lock);
4990
4991 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4992 if (mddev2->gendisk &&
4993 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4994 spin_unlock(&all_mddevs_lock);
0909dc44 4995 goto abort;
efeb53c0
N		 4996 }
4997 spin_unlock(&all_mddevs_lock);
1da177e4 4998 }
8b765398 4999
0909dc44 5000 error = -ENOMEM;
8b765398 5001 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 5002 if (!mddev->queue)
5003 goto abort;
409c57f3
N		 5004 mddev->queue->queuedata = mddev;
5005
409c57f3 5006 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 5007 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 5008
1da177e4
LT		 5009 disk = alloc_disk(1 << shift);
5010 if (!disk) {
8b765398
N		 5011 blk_cleanup_queue(mddev->queue);
5012 mddev->queue = NULL;
0909dc44 5013 goto abort;
1da177e4 5014 }
efeb53c0 5015 disk->major = MAJOR(mddev->unit);
1da177e4 5016 disk->first_minor = unit << shift;
efeb53c0
N		 5017 if (name)
5018 strcpy(disk->disk_name, name);
5019 else if (partitioned)
1da177e4 5020 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 5021 else
1da177e4 5022 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 5023 disk->fops = &md_fops;
5024 disk->private_data = mddev;
5025 disk->queue = mddev->queue;
b0140891 5026 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 5027 /* Allow extended partitions. This makes the
d3374825 5028 * 'mdp' device redundant, but we can't really
92850bbd
N		 5029 * remove it now.
5030 */
5031 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 5032 mddev->gendisk = disk;
b0140891
N		 5033 /* As soon as we call add_disk(), another thread could get
5034 * through to md_open, so make sure it doesn't get too far
5035 */
5036 mutex_lock(&mddev->open_mutex);
5037 add_disk(disk);
5038
ed9e1982
TH		 5039 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5040 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 5041 if (error) {
5042 /* This isn't possible, but as kobject_init_and_add is marked
5043 * __must_check, we must do something with the result
5044 */
5e55e2f5
N		 5045 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5046 disk->disk_name);
0909dc44
N		 5047 error = 0;
5048 }
00bcb4ac
N		 5049 if (mddev->kobj.sd &&
5050 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 5051 printk(KERN_DEBUG "pointless warning\n");
b0140891 5052 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 5053 abort:
5054 mutex_unlock(&disks_mutex);
00bcb4ac 5055 if (!error && mddev->kobj.sd) {
3830c62f 5056 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 5057 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 5058 }
d3374825 5059 mddev_put(mddev);
0909dc44 5060 return error;
efeb53c0
N		 5061}
5062
5063static struct kobject *md_probe(dev_t dev, int *part, void *data)
5064{
5065 md_alloc(dev, NULL);
1da177e4
LT		 5066 return NULL;
5067}
5068
efeb53c0
N		 5069static int add_named_array(const char *val, struct kernel_param *kp)
5070{
5071 /* val must be "md_*" where * is not all digits.
5072 * We allocate an array with a large free minor number, and
5073 * set the name to val. val must not already be an active name.
5074 */
5075 int len = strlen(val);
5076 char buf[DISK_NAME_LEN];
5077
5078 while (len && val[len-1] == '\n')
5079 len--;
5080 if (len >= DISK_NAME_LEN)
5081 return -E2BIG;
5082 strlcpy(buf, val, len+1);
5083 if (strncmp(buf, "md_", 3) != 0)
5084 return -EINVAL;
5085 return md_alloc(0, buf);
5086}
5087
1da177e4
LT		 5088static void md_safemode_timeout(unsigned long data)
5089{
fd01b88c 5090 struct mddev *mddev = (struct mddev *) data;
1da177e4 5091
0fd62b86
NB		 5092 if (!atomic_read(&mddev->writes_pending)) {
5093 mddev->safemode = 1;
5094 if (mddev->external)
00bcb4ac 5095 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 5096 }
1da177e4
LT		 5097 md_wakeup_thread(mddev->thread);
5098}
5099
6ff8d8ec 5100static int start_dirty_degraded;
1da177e4 5101
fd01b88c 5102int md_run(struct mddev *mddev)
1da177e4 5103{
2604b703 5104 int err;
3cb03002 5105 struct md_rdev *rdev;
84fc4b56 5106 struct md_personality *pers;
1da177e4 5107
a757e64c
N		 5108 if (list_empty(&mddev->disks))
5109 /* cannot run an array with no devices.. */
1da177e4 5110 return -EINVAL;
1da177e4
LT		 5111
5112 if (mddev->pers)
5113 return -EBUSY;
bb4f1e9d
N		 5114 /* Cannot run until previous stop completes properly */
5115 if (mddev->sysfs_active)
5116 return -EBUSY;
b6eb127d 5117
1da177e4
LT		 5118 /*
5119 * Analyze all RAID superblock(s)
5120 */
1ec4a939
N		 5121 if (!mddev->raid_disks) {
5122 if (!mddev->persistent)
5123 return -EINVAL;
a757e64c 5124 analyze_sbs(mddev);
1ec4a939 5125 }
1da177e4 5126
d9d166c2
N		 5127 if (mddev->level != LEVEL_NONE)
5128 request_module("md-level-%d", mddev->level);
5129 else if (mddev->clevel[0])
5130 request_module("md-%s", mddev->clevel);
1da177e4
LT		 5131
5132 /*
5133 * Drop all container device buffers, from now on
5134 * the only valid external interface is through the md
5135 * device.
1da177e4 5136 */
dafb20fa 5137 rdev_for_each(rdev, mddev) {
b2d444d7 5138 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5139 continue;
5140 sync_blockdev(rdev->bdev);
f98393a6 5141 invalidate_bdev(rdev->bdev);
f0d76d70
N		 5142
5143 /* perform some consistency tests on the device.
5144 * We don't want the data to overlap the metadata,
58c0fed4 5145 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 5146 */
a6ff7e08
JB		 5147 if (rdev->meta_bdev) {
5148 /* Nothing to check */;
5149 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 5150 if (mddev->dev_sectors &&
5151 rdev->data_offset + mddev->dev_sectors
0f420358 5152 > rdev->sb_start) {
f0d76d70
N		 5153 printk("md: %s: data overlaps metadata\n",
5154 mdname(mddev));
5155 return -EINVAL;
5156 }
5157 } else {
0f420358 5158 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 5159 > rdev->data_offset) {
5160 printk("md: %s: metadata overlaps data\n",
5161 mdname(mddev));
5162 return -EINVAL;
5163 }
5164 }
00bcb4ac 5165 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 5166 }
5167
a167f663 5168 if (mddev->bio_set == NULL)
395c72a7 5169 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
a167f663 5170
1da177e4 5171 spin_lock(&pers_lock);
d9d166c2 5172 pers = find_pers(mddev->level, mddev->clevel);
2604b703 5173 if (!pers || !try_module_get(pers->owner)) {
1da177e4 5174 spin_unlock(&pers_lock);
d9d166c2
N		 5175 if (mddev->level != LEVEL_NONE)
5176 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5177 mddev->level);
5178 else
5179 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5180 mddev->clevel);
1da177e4
LT		 5181 return -EINVAL;
5182 }
1da177e4 5183 spin_unlock(&pers_lock);
34817e8c
N		 5184 if (mddev->level != pers->level) {
5185 mddev->level = pers->level;
5186 mddev->new_level = pers->level;
5187 }
d9d166c2 5188 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 5189
f6705578 5190 if (mddev->reshape_position != MaxSector &&
63c70c4f 5191 pers->start_reshape == NULL) {
f6705578 5192 /* This personality cannot handle reshaping... */
f6705578
N		 5193 module_put(pers->owner);
5194 return -EINVAL;
5195 }
5196
7dd5e7c3
N		 5197 if (pers->sync_request) {
5198 /* Warn if this is a potentially silly
5199 * configuration.
5200 */
5201 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5202 struct md_rdev *rdev2;
7dd5e7c3 5203 int warned = 0;
159ec1fc 5204
dafb20fa
N		 5205 rdev_for_each(rdev, mddev)
5206 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N		 5207 if (rdev < rdev2 &&
5208 rdev->bdev->bd_contains ==
5209 rdev2->bdev->bd_contains) {
5210 printk(KERN_WARNING
5211 "%s: WARNING: %s appears to be"
5212 " on the same physical disk as"
5213 " %s.\n",
5214 mdname(mddev),
5215 bdevname(rdev->bdev,b),
5216 bdevname(rdev2->bdev,b2));
5217 warned = 1;
5218 }
5219 }
159ec1fc 5220
7dd5e7c3
N		 5221 if (warned)
5222 printk(KERN_WARNING
5223 "True protection against single-disk"
5224 " failure might be compromised.\n");
5225 }
5226
657390d2 5227 mddev->recovery = 0;
58c0fed4
AN		 5228 /* may be over-ridden by personality */
5229 mddev->resync_max_sectors = mddev->dev_sectors;
5230
6ff8d8ec 5231 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5232
0f9552b5 5233 if (start_readonly && mddev->ro == 0)
f91de92e
N		 5234 mddev->ro = 2; /* read-only, but switch on first write */
5235
36d091f4 5236 err = pers->run(mddev);
13e53df3
AN		 5237 if (err)
5238 printk(KERN_ERR "md: pers->run() failed ...\n");
36d091f4 5239 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
b522adcd
DW		 5240 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5241 " but 'external_size' not in effect?\n", __func__);
5242 printk(KERN_ERR
5243 "md: invalid array_size %llu > default size %llu\n",
5244 (unsigned long long)mddev->array_sectors / 2,
36d091f4 5245 (unsigned long long)pers->size(mddev, 0, 0) / 2);
b522adcd 5246 err = -EINVAL;
b522adcd 5247 }
36d091f4 5248 if (err == 0 && pers->sync_request &&
ef99bf48 5249 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
f9209a32
GR		 5250 struct bitmap *bitmap;
5251
5252 bitmap = bitmap_create(mddev, -1);
5253 if (IS_ERR(bitmap)) {
5254 err = PTR_ERR(bitmap);
b15c2e57
N		 5255 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5256 mdname(mddev), err);
f9209a32
GR		 5257 } else
5258 mddev->bitmap = bitmap;
5259
b15c2e57 5260 }
1da177e4 5261 if (err) {
5aa61f42 5262 mddev_detach(mddev);
0c35bd47
N		 5263 if (mddev->private)
5264 pers->free(mddev, mddev->private);
bd691922 5265 mddev->private = NULL;
36d091f4 5266 module_put(pers->owner);
32a7627c
N		 5267 bitmap_destroy(mddev);
5268 return err;
1da177e4 5269 }
5c675f83
N		 5270 if (mddev->queue) {
5271 mddev->queue->backing_dev_info.congested_data = mddev;
5272 mddev->queue->backing_dev_info.congested_fn = md_congested;
5273 }
36d091f4 5274 if (pers->sync_request) {
00bcb4ac
N		 5275 if (mddev->kobj.sd &&
5276 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 5277 printk(KERN_WARNING
5278 "md: cannot register extra attributes for %s\n",
5279 mdname(mddev));
00bcb4ac 5280 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5281 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 5282 mddev->ro = 0;
5283
f72ffdd6 5284 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 5285 atomic_set(&mddev->max_corr_read_errors,
5286 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4 5287 mddev->safemode = 0;
28c1b9fd
GR		 5288 if (mddev_is_clustered(mddev))
5289 mddev->safemode_delay = 0;
5290 else
5291 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5292 mddev->in_sync = 1;
0ca69886 5293 smp_wmb();
36d091f4
N		 5294 spin_lock(&mddev->lock);
5295 mddev->pers = pers;
0ca69886 5296 mddev->ready = 1;
36d091f4 5297 spin_unlock(&mddev->lock);
dafb20fa 5298 rdev_for_each(rdev, mddev)
36fad858
NK		 5299 if (rdev->raid_disk >= 0)
5300 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 5301 /* failure here is OK */;
f72ffdd6 5302
a4a3d26d
N		 5303 if (mddev->degraded && !mddev->ro)
5304 /* This ensures that recovering status is reported immediately
5305 * via sysfs - until a lack of spares is confirmed.
5306 */
5307 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
1da177e4 5308 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f72ffdd6 5309
7a0a5355 5310 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 5311 md_update_sb(mddev, 0);
1da177e4 5312
d7603b7e 5313 md_new_event(mddev);
00bcb4ac
N		 5314 sysfs_notify_dirent_safe(mddev->sysfs_state);
5315 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 5316 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 5317 return 0;
5318}
390ee602 5319EXPORT_SYMBOL_GPL(md_run);
1da177e4 5320
fd01b88c 5321static int do_md_run(struct mddev *mddev)
fe60b014
N		 5322{
5323 int err;
5324
5325 err = md_run(mddev);
5326 if (err)
5327 goto out;
69e51b44
N		 5328 err = bitmap_load(mddev);
5329 if (err) {
5330 bitmap_destroy(mddev);
5331 goto out;
5332 }
0fd018af 5333
28c1b9fd
GR		 5334 if (mddev_is_clustered(mddev))
5335 md_allow_write(mddev);
5336
0fd018af
JB		 5337 md_wakeup_thread(mddev->thread);
5338 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5339
fe60b014
N		 5340 set_capacity(mddev->gendisk, mddev->array_sectors);
5341 revalidate_disk(mddev->gendisk);
f0b4f7e2 5342 mddev->changed = 1;
fe60b014
N		 5343 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5344out:
5345 return err;
5346}
5347
fd01b88c 5348static int restart_array(struct mddev *mddev)
1da177e4
LT		 5349{
5350 struct gendisk *disk = mddev->gendisk;
1da177e4 5351
80fab1d7 5352 /* Complain if it has no devices */
1da177e4 5353 if (list_empty(&mddev->disks))
80fab1d7
AN		 5354 return -ENXIO;
5355 if (!mddev->pers)
5356 return -EINVAL;
5357 if (!mddev->ro)
5358 return -EBUSY;
339421de
SL		 5359 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5360 struct md_rdev *rdev;
5361 bool has_journal = false;
5362
5363 rcu_read_lock();
5364 rdev_for_each_rcu(rdev, mddev) {
5365 if (test_bit(Journal, &rdev->flags) &&
5366 !test_bit(Faulty, &rdev->flags)) {
5367 has_journal = true;
5368 break;
5369 }
5370 }
5371 rcu_read_unlock();
5372
5373 /* Don't restart rw with journal missing/faulty */
5374 if (!has_journal)
5375 return -EINVAL;
5376 }
5377
80fab1d7
AN		 5378 mddev->safemode = 0;
5379 mddev->ro = 0;
5380 set_disk_ro(disk, 0);
5381 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5382 mdname(mddev));
5383 /* Kick recovery or resync if necessary */
5384 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5385 md_wakeup_thread(mddev->thread);
5386 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5387 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5388 return 0;
1da177e4
LT		 5389}
5390
fd01b88c 5391static void md_clean(struct mddev *mddev)
6177b472
N		 5392{
5393 mddev->array_sectors = 0;
5394 mddev->external_size = 0;
5395 mddev->dev_sectors = 0;
5396 mddev->raid_disks = 0;
5397 mddev->recovery_cp = 0;
5398 mddev->resync_min = 0;
5399 mddev->resync_max = MaxSector;
5400 mddev->reshape_position = MaxSector;
5401 mddev->external = 0;
5402 mddev->persistent = 0;
5403 mddev->level = LEVEL_NONE;
5404 mddev->clevel[0] = 0;
5405 mddev->flags = 0;
5406 mddev->ro = 0;
5407 mddev->metadata_type[0] = 0;
5408 mddev->chunk_sectors = 0;
5409 mddev->ctime = mddev->utime = 0;
5410 mddev->layout = 0;
5411 mddev->max_disks = 0;
5412 mddev->events = 0;
a8707c08 5413 mddev->can_decrease_events = 0;
6177b472 5414 mddev->delta_disks = 0;
2c810cdd 5415 mddev->reshape_backwards = 0;
6177b472
N		 5416 mddev->new_level = LEVEL_NONE;
5417 mddev->new_layout = 0;
5418 mddev->new_chunk_sectors = 0;
5419 mddev->curr_resync = 0;
7f7583d4 5420 atomic64_set(&mddev->resync_mismatches, 0);
6177b472
N		 5421 mddev->suspend_lo = mddev->suspend_hi = 0;
5422 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5423 mddev->recovery = 0;
5424 mddev->in_sync = 0;
f0b4f7e2 5425 mddev->changed = 0;
6177b472 5426 mddev->degraded = 0;
6177b472 5427 mddev->safemode = 0;
bd691922 5428 mddev->private = NULL;
6177b472
N		 5429 mddev->bitmap_info.offset = 0;
5430 mddev->bitmap_info.default_offset = 0;
6409bb05 5431 mddev->bitmap_info.default_space = 0;
6177b472
N		 5432 mddev->bitmap_info.chunksize = 0;
5433 mddev->bitmap_info.daemon_sleep = 0;
5434 mddev->bitmap_info.max_write_behind = 0;
5435}
5436
fd01b88c 5437static void __md_stop_writes(struct mddev *mddev)
a047e125 5438{
6b6204ee 5439 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
f851b60d 5440 flush_workqueue(md_misc_wq);
a047e125 5441 if (mddev->sync_thread) {
a047e125 5442 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 5443 md_reap_sync_thread(mddev);
a047e125
N		 5444 }
5445
5446 del_timer_sync(&mddev->safemode_timer);
5447
5448 bitmap_flush(mddev);
5449 md_super_wait(mddev);
5450
b6d428c6 5451 if (mddev->ro == 0 &&
28c1b9fd
GR		 5452 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5453 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
a047e125 5454 /* mark array as shutdown cleanly */
28c1b9fd
GR		 5455 if (!mddev_is_clustered(mddev))
5456 mddev->in_sync = 1;
a047e125
N		 5457 md_update_sb(mddev, 1);
5458 }
5459}
defad61a 5460
fd01b88c 5461void md_stop_writes(struct mddev *mddev)
defad61a 5462{
29f097c4 5463 mddev_lock_nointr(mddev);
defad61a
N		 5464 __md_stop_writes(mddev);
5465 mddev_unlock(mddev);
5466}
390ee602 5467EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5468
5aa61f42
N		 5469static void mddev_detach(struct mddev *mddev)
5470{
5471 struct bitmap *bitmap = mddev->bitmap;
5472 /* wait for behind writes to complete */
5473 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5474 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5475 mdname(mddev));
5476 /* need to kick something here to make sure I/O goes? */
5477 wait_event(bitmap->behind_wait,
5478 atomic_read(&bitmap->behind_writes) == 0);
5479 }
36d091f4 5480 if (mddev->pers && mddev->pers->quiesce) {
5aa61f42
N		 5481 mddev->pers->quiesce(mddev, 1);
5482 mddev->pers->quiesce(mddev, 0);
5483 }
5484 md_unregister_thread(&mddev->thread);
5485 if (mddev->queue)
5486 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5487}
5488
5eff3c43 5489static void __md_stop(struct mddev *mddev)
6177b472 5490{
36d091f4 5491 struct md_personality *pers = mddev->pers;
5aa61f42 5492 mddev_detach(mddev);
ee5d004f
N		 5493 /* Ensure ->event_work is done */
5494 flush_workqueue(md_misc_wq);
36d091f4
N		 5495 spin_lock(&mddev->lock);
5496 mddev->ready = 0;
6177b472 5497 mddev->pers = NULL;
36d091f4
N		 5498 spin_unlock(&mddev->lock);
5499 pers->free(mddev, mddev->private);
bd691922 5500 mddev->private = NULL;
36d091f4
N		 5501 if (pers->sync_request && mddev->to_remove == NULL)
5502 mddev->to_remove = &md_redundancy_group;
5503 module_put(pers->owner);
cca9cf90 5504 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5505}
5eff3c43
N		 5506
5507void md_stop(struct mddev *mddev)
5508{
5509 /* stop the array and free an attached data structures.
5510 * This is called from dm-raid
5511 */
5512 __md_stop(mddev);
5513 bitmap_destroy(mddev);
5514 if (mddev->bio_set)
5515 bioset_free(mddev->bio_set);
5516}
5517
390ee602 5518EXPORT_SYMBOL_GPL(md_stop);
6177b472 5519
a05b7ea0 5520static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N		 5521{
5522 int err = 0;
30b8feb7
N		 5523 int did_freeze = 0;
5524
5525 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5526 did_freeze = 1;
5527 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5528 md_wakeup_thread(mddev->thread);
5529 }
f851b60d 5530 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5531 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5532 if (mddev->sync_thread)
30b8feb7
N		 5533 /* Thread might be blocked waiting for metadata update
5534 * which will now never happen */
5535 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5536
88724bfa
N		 5537 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5538 return -EBUSY;
30b8feb7 5539 mddev_unlock(mddev);
f851b60d
N		 5540 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5541 &mddev->recovery));
88724bfa
N		 5542 wait_event(mddev->sb_wait,
5543 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
30b8feb7
N		 5544 mddev_lock_nointr(mddev);
5545
a4bd82d0 5546 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5547 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7 5548 mddev->sync_thread ||
f851b60d 5549 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5550 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
a4bd82d0 5551 printk("md: %s still in use.\n",mdname(mddev));
30b8feb7
N		 5552 if (did_freeze) {
5553 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5554 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5555 md_wakeup_thread(mddev->thread);
5556 }
a4bd82d0
N		 5557 err = -EBUSY;
5558 goto out;
5559 }
5560 if (mddev->pers) {
defad61a 5561 __md_stop_writes(mddev);
a4bd82d0
N		 5562
5563 err = -ENXIO;
5564 if (mddev->ro==1)
5565 goto out;
5566 mddev->ro = 1;
5567 set_disk_ro(mddev->gendisk, 1);
5568 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d
N		 5569 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5570 md_wakeup_thread(mddev->thread);
00bcb4ac 5571 sysfs_notify_dirent_safe(mddev->sysfs_state);
30b8feb7 5572 err = 0;
a4bd82d0
N		 5573 }
5574out:
5575 mutex_unlock(&mddev->open_mutex);
5576 return err;
5577}
5578
9e653b63
N		 5579/* mode:
5580 * 0 - completely stop and dis-assemble array
9e653b63
N		 5581 * 2 - stop but do not disassemble array
5582 */
f72ffdd6 5583static int do_md_stop(struct mddev *mddev, int mode,
a05b7ea0 5584 struct block_device *bdev)
1da177e4 5585{
1da177e4 5586 struct gendisk *disk = mddev->gendisk;
3cb03002 5587 struct md_rdev *rdev;
30b8feb7
N		 5588 int did_freeze = 0;
5589
5590 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5591 did_freeze = 1;
5592 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5593 md_wakeup_thread(mddev->thread);
5594 }
f851b60d 5595 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5596 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5597 if (mddev->sync_thread)
30b8feb7
N		 5598 /* Thread might be blocked waiting for metadata update
5599 * which will now never happen */
5600 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5601
30b8feb7 5602 mddev_unlock(mddev);
f851b60d
N		 5603 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5604 !test_bit(MD_RECOVERY_RUNNING,
5605 &mddev->recovery)));
30b8feb7 5606 mddev_lock_nointr(mddev);
1da177e4 5607
c8c00a69 5608 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5609 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7
N		 5610 mddev->sysfs_active ||
5611 mddev->sync_thread ||
f851b60d 5612 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5613 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
df5b20cf 5614 printk("md: %s still in use.\n",mdname(mddev));
6e17b027 5615 mutex_unlock(&mddev->open_mutex);
30b8feb7
N		 5616 if (did_freeze) {
5617 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5618 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5619 md_wakeup_thread(mddev->thread);
5620 }
260fa034
N		 5621 return -EBUSY;
5622 }
6e17b027 5623 if (mddev->pers) {
a4bd82d0
N		 5624 if (mddev->ro)
5625 set_disk_ro(disk, 0);
409c57f3 5626
defad61a 5627 __md_stop_writes(mddev);
5eff3c43 5628 __md_stop(mddev);
a4bd82d0 5629 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5630
a4bd82d0 5631 /* tell userspace to handle 'inactive' */
00bcb4ac 5632 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5633
dafb20fa 5634 rdev_for_each(rdev, mddev)
36fad858
NK		 5635 if (rdev->raid_disk >= 0)
5636 sysfs_unlink_rdev(mddev, rdev);
c4647292 5637
a4bd82d0 5638 set_capacity(disk, 0);
6e17b027 5639 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5640 mddev->changed = 1;
a4bd82d0 5641 revalidate_disk(disk);
0d4ca600 5642
a4bd82d0
N		 5643 if (mddev->ro)
5644 mddev->ro = 0;
6e17b027
N		 5645 } else
5646 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5647 /*
5648 * Free resources if final stop
5649 */
9e653b63 5650 if (mode == 0) {
1da177e4
LT		 5651 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5652
978f946b 5653 bitmap_destroy(mddev);
c3d9714e 5654 if (mddev->bitmap_info.file) {
4af1a041
N		 5655 struct file *f = mddev->bitmap_info.file;
5656 spin_lock(&mddev->lock);
c3d9714e 5657 mddev->bitmap_info.file = NULL;
4af1a041
N		 5658 spin_unlock(&mddev->lock);
5659 fput(f);
978f946b 5660 }
c3d9714e 5661 mddev->bitmap_info.offset = 0;
978f946b 5662
1da177e4
LT		 5663 export_array(mddev);
5664
6177b472 5665 md_clean(mddev);
934d9c23 5666 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5667 if (mddev->hold_active == UNTIL_STOP)
5668 mddev->hold_active = 0;
a4bd82d0 5669 }
d7603b7e 5670 md_new_event(mddev);
00bcb4ac 5671 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5672 return 0;
1da177e4
LT		 5673}
5674
fdee8ae4 5675#ifndef MODULE
fd01b88c 5676static void autorun_array(struct mddev *mddev)
1da177e4 5677{
3cb03002 5678 struct md_rdev *rdev;
1da177e4
LT		 5679 int err;
5680
a757e64c 5681 if (list_empty(&mddev->disks))
1da177e4 5682 return;
1da177e4
LT		 5683
5684 printk(KERN_INFO "md: running: ");
5685
dafb20fa 5686 rdev_for_each(rdev, mddev) {
1da177e4
LT		 5687 char b[BDEVNAME_SIZE];
5688 printk("<%s>", bdevname(rdev->bdev,b));
5689 }
5690 printk("\n");
5691
d710e138 5692 err = do_md_run(mddev);
1da177e4
LT		 5693 if (err) {
5694 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
a05b7ea0 5695 do_md_stop(mddev, 0, NULL);
1da177e4
LT		 5696 }
5697}
5698
5699/*
5700 * lets try to run arrays based on all disks that have arrived
5701 * until now. (those are in pending_raid_disks)
5702 *
5703 * the method: pick the first pending disk, collect all disks with
5704 * the same UUID, remove all from the pending list and put them into
5705 * the 'same_array' list. Then order this list based on superblock
5706 * update time (freshest comes first), kick out 'old' disks and
5707 * compare superblocks. If everything's fine then run it.
5708 *
5709 * If "unit" is allocated, then bump its reference count
5710 */
5711static void autorun_devices(int part)
5712{
3cb03002 5713 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5714 struct mddev *mddev;
1da177e4
LT		 5715 char b[BDEVNAME_SIZE];
5716
5717 printk(KERN_INFO "md: autorun ...\n");
5718 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5719 int unit;
1da177e4 5720 dev_t dev;
ad01c9e3 5721 LIST_HEAD(candidates);
1da177e4 5722 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5723 struct md_rdev, same_set);
1da177e4
LT		 5724
5725 printk(KERN_INFO "md: considering %s ...\n",
5726 bdevname(rdev0->bdev,b));
5727 INIT_LIST_HEAD(&candidates);
159ec1fc 5728 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5729 if (super_90_load(rdev, rdev0, 0) >= 0) {
5730 printk(KERN_INFO "md: adding %s ...\n",
5731 bdevname(rdev->bdev,b));
5732 list_move(&rdev->same_set, &candidates);
5733 }
5734 /*
5735 * now we have a set of devices, with all of them having
5736 * mostly sane superblocks. It's time to allocate the
5737 * mddev.
5738 */
e8703fe1
N		 5739 if (part) {
5740 dev = MKDEV(mdp_major,
5741 rdev0->preferred_minor << MdpMinorShift);
5742 unit = MINOR(dev) >> MdpMinorShift;
5743 } else {
5744 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5745 unit = MINOR(dev);
5746 }
5747 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5748 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5749 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5750 break;
5751 }
1da177e4
LT		 5752
5753 md_probe(dev, NULL, NULL);
5754 mddev = mddev_find(dev);
9bbbca3a
NB		 5755 if (!mddev || !mddev->gendisk) {
5756 if (mddev)
5757 mddev_put(mddev);
5758 printk(KERN_ERR
1da177e4
LT		 5759 "md: cannot allocate memory for md drive.\n");
5760 break;
5761 }
f72ffdd6 5762 if (mddev_lock(mddev))
1da177e4
LT		 5763 printk(KERN_WARNING "md: %s locked, cannot run\n",
5764 mdname(mddev));
5765 else if (mddev->raid_disks || mddev->major_version
5766 || !list_empty(&mddev->disks)) {
f72ffdd6 5767 printk(KERN_WARNING
1da177e4
LT		 5768 "md: %s already running, cannot run %s\n",
5769 mdname(mddev), bdevname(rdev0->bdev,b));
5770 mddev_unlock(mddev);
5771 } else {
5772 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5773 mddev->persistent = 1;
159ec1fc 5774 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5775 list_del_init(&rdev->same_set);
5776 if (bind_rdev_to_array(rdev, mddev))
5777 export_rdev(rdev);
5778 }
5779 autorun_array(mddev);
5780 mddev_unlock(mddev);
5781 }
5782 /* on success, candidates will be empty, on error
5783 * it won't...
5784 */
159ec1fc 5785 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5786 list_del_init(&rdev->same_set);
1da177e4 5787 export_rdev(rdev);
4b80991c 5788 }
1da177e4
LT		 5789 mddev_put(mddev);
5790 }
5791 printk(KERN_INFO "md: ... autorun DONE.\n");
5792}
fdee8ae4 5793#endif /* !MODULE */
1da177e4 5794
f72ffdd6 5795static int get_version(void __user *arg)
1da177e4
LT		 5796{
5797 mdu_version_t ver;
5798
5799 ver.major = MD_MAJOR_VERSION;
5800 ver.minor = MD_MINOR_VERSION;
5801 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5802
5803 if (copy_to_user(arg, &ver, sizeof(ver)))
5804 return -EFAULT;
5805
5806 return 0;
5807}
5808
f72ffdd6 5809static int get_array_info(struct mddev *mddev, void __user *arg)
1da177e4
LT		 5810{
5811 mdu_array_info_t info;
a9f326eb 5812 int nr,working,insync,failed,spare;
3cb03002 5813 struct md_rdev *rdev;
1da177e4 5814
1ca69c4b
N		 5815 nr = working = insync = failed = spare = 0;
5816 rcu_read_lock();
5817 rdev_for_each_rcu(rdev, mddev) {
1da177e4 5818 nr++;
b2d444d7 5819 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5820 failed++;
5821 else {
5822 working++;
b2d444d7 5823 if (test_bit(In_sync, &rdev->flags))
f72ffdd6 5824 insync++;
1da177e4
LT		 5825 else
5826 spare++;
5827 }
5828 }
1ca69c4b 5829 rcu_read_unlock();
1da177e4
LT		 5830
5831 info.major_version = mddev->major_version;
5832 info.minor_version = mddev->minor_version;
5833 info.patch_version = MD_PATCHLEVEL_VERSION;
5834 info.ctime = mddev->ctime;
5835 info.level = mddev->level;
58c0fed4
AN		 5836 info.size = mddev->dev_sectors / 2;
5837 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5838 info.size = -1;
1da177e4
LT		 5839 info.nr_disks = nr;
5840 info.raid_disks = mddev->raid_disks;
5841 info.md_minor = mddev->md_minor;
5842 info.not_persistent= !mddev->persistent;
5843
5844 info.utime = mddev->utime;
5845 info.state = 0;
5846 if (mddev->in_sync)
5847 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5848 if (mddev->bitmap && mddev->bitmap_info.offset)
9bd35920 5849 info.state |= (1<<MD_SB_BITMAP_PRESENT);
ca8895d9
GR		 5850 if (mddev_is_clustered(mddev))
5851 info.state |= (1<<MD_SB_CLUSTERED);
a9f326eb 5852 info.active_disks = insync;
1da177e4
LT		 5853 info.working_disks = working;
5854 info.failed_disks = failed;
5855 info.spare_disks = spare;
5856
5857 info.layout = mddev->layout;
9d8f0363 5858 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5859
5860 if (copy_to_user(arg, &info, sizeof(info)))
5861 return -EFAULT;
5862
5863 return 0;
5864}
5865
f72ffdd6 5866static int get_bitmap_file(struct mddev *mddev, void __user * arg)
32a7627c
N		 5867{
5868 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
f4ad3d38 5869 char *ptr;
4af1a041 5870 int err;
32a7627c 5871
b6878d9e 5872 file = kzalloc(sizeof(*file), GFP_NOIO);
32a7627c 5873 if (!file)
4af1a041 5874 return -ENOMEM;
32a7627c 5875
4af1a041
N		 5876 err = 0;
5877 spin_lock(&mddev->lock);
25eafe1a
BR		 5878 /* bitmap enabled */
5879 if (mddev->bitmap_info.file) {
5880 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5881 sizeof(file->pathname));
5882 if (IS_ERR(ptr))
5883 err = PTR_ERR(ptr);
5884 else
5885 memmove(file->pathname, ptr,
5886 sizeof(file->pathname)-(ptr-file->pathname));
5887 }
4af1a041 5888 spin_unlock(&mddev->lock);
32a7627c 5889
4af1a041
N		 5890 if (err == 0 &&
5891 copy_to_user(arg, file, sizeof(*file)))
32a7627c 5892 err = -EFAULT;
4af1a041 5893
32a7627c
N		 5894 kfree(file);
5895 return err;
5896}
5897
f72ffdd6 5898static int get_disk_info(struct mddev *mddev, void __user * arg)
1da177e4
LT		 5899{
5900 mdu_disk_info_t info;
3cb03002 5901 struct md_rdev *rdev;
1da177e4
LT		 5902
5903 if (copy_from_user(&info, arg, sizeof(info)))
5904 return -EFAULT;
5905
1ca69c4b 5906 rcu_read_lock();
57d051dc 5907 rdev = md_find_rdev_nr_rcu(mddev, info.number);
1da177e4
LT		 5908 if (rdev) {
5909 info.major = MAJOR(rdev->bdev->bd_dev);
5910 info.minor = MINOR(rdev->bdev->bd_dev);
5911 info.raid_disk = rdev->raid_disk;
5912 info.state = 0;
b2d444d7 5913 if (test_bit(Faulty, &rdev->flags))
1da177e4 5914 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5915 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5916 info.state |= (1<<MD_DISK_ACTIVE);
5917 info.state |= (1<<MD_DISK_SYNC);
5918 }
9efdca16 5919 if (test_bit(Journal, &rdev->flags))
bac624f3 5920 info.state |= (1<<MD_DISK_JOURNAL);
8ddf9efe
N		 5921 if (test_bit(WriteMostly, &rdev->flags))
5922 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5923 } else {
5924 info.major = info.minor = 0;
5925 info.raid_disk = -1;
5926 info.state = (1<<MD_DISK_REMOVED);
5927 }
1ca69c4b 5928 rcu_read_unlock();
1da177e4
LT		 5929
5930 if (copy_to_user(arg, &info, sizeof(info)))
5931 return -EFAULT;
5932
5933 return 0;
5934}
5935
f72ffdd6 5936static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
1da177e4
LT		 5937{
5938 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5939 struct md_rdev *rdev;
1da177e4
LT		 5940 dev_t dev = MKDEV(info->major,info->minor);
5941
1aee41f6
GR		 5942 if (mddev_is_clustered(mddev) &&
5943 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
fa8259da 5944 pr_err("%s: Cannot add to clustered mddev.\n",
1aee41f6
GR		 5945 mdname(mddev));
5946 return -EINVAL;
5947 }
5948
1da177e4
LT		 5949 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5950 return -EOVERFLOW;
5951
5952 if (!mddev->raid_disks) {
5953 int err;
5954 /* expecting a device which has a superblock */
5955 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5956 if (IS_ERR(rdev)) {
f72ffdd6 5957 printk(KERN_WARNING
1da177e4
LT		 5958 "md: md_import_device returned %ld\n",
5959 PTR_ERR(rdev));
5960 return PTR_ERR(rdev);
5961 }
5962 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5963 struct md_rdev *rdev0
5964 = list_entry(mddev->disks.next,
5965 struct md_rdev, same_set);
a9f326eb 5966 err = super_types[mddev->major_version]
1da177e4
LT		 5967 .load_super(rdev, rdev0, mddev->minor_version);
5968 if (err < 0) {
f72ffdd6 5969 printk(KERN_WARNING
1da177e4 5970 "md: %s has different UUID to %s\n",
f72ffdd6 5971 bdevname(rdev->bdev,b),
1da177e4
LT		 5972 bdevname(rdev0->bdev,b2));
5973 export_rdev(rdev);
5974 return -EINVAL;
5975 }
5976 }
5977 err = bind_rdev_to_array(rdev, mddev);
5978 if (err)
5979 export_rdev(rdev);
5980 return err;
5981 }
5982
5983 /*
5984 * add_new_disk can be used once the array is assembled
5985 * to add "hot spares". They must already have a superblock
5986 * written
5987 */
5988 if (mddev->pers) {
5989 int err;
5990 if (!mddev->pers->hot_add_disk) {
f72ffdd6 5991 printk(KERN_WARNING
1da177e4
LT		 5992 "%s: personality does not support diskops!\n",
5993 mdname(mddev));
5994 return -EINVAL;
5995 }
7b1e35f6
N		 5996 if (mddev->persistent)
5997 rdev = md_import_device(dev, mddev->major_version,
5998 mddev->minor_version);
5999 else
6000 rdev = md_import_device(dev, -1, -1);
1da177e4 6001 if (IS_ERR(rdev)) {
f72ffdd6 6002 printk(KERN_WARNING
1da177e4
LT		 6003 "md: md_import_device returned %ld\n",
6004 PTR_ERR(rdev));
6005 return PTR_ERR(rdev);
6006 }
1a855a06 6007 /* set saved_raid_disk if appropriate */
41158c7e
N		 6008 if (!mddev->persistent) {
6009 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 6010 info->raid_disk < mddev->raid_disks) {
41158c7e 6011 rdev->raid_disk = info->raid_disk;
bf572541 6012 set_bit(In_sync, &rdev->flags);
8313b8e5 6013 clear_bit(Bitmap_sync, &rdev->flags);
bf572541 6014 } else
41158c7e 6015 rdev->raid_disk = -1;
f466722c 6016 rdev->saved_raid_disk = rdev->raid_disk;
41158c7e
N		 6017 } else
6018 super_types[mddev->major_version].
6019 validate_super(mddev, rdev);
bedd86b7 6020 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 6021 rdev->raid_disk != info->raid_disk) {
bedd86b7
N		 6022 /* This was a hot-add request, but events doesn't
6023 * match, so reject it.
6024 */
6025 export_rdev(rdev);
6026 return -EINVAL;
6027 }
6028
b2d444d7 6029 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 6030 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6031 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 6032 else
6033 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 6034
bac624f3
SL		 6035 if (info->state & (1<<MD_DISK_JOURNAL))
6036 set_bit(Journal, &rdev->flags);
1aee41f6
GR		 6037 /*
6038 * check whether the device shows up in other nodes
6039 */
6040 if (mddev_is_clustered(mddev)) {
dbb64f86 6041 if (info->state & (1 << MD_DISK_CANDIDATE))
1aee41f6 6042 set_bit(Candidate, &rdev->flags);
dbb64f86 6043 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
1aee41f6 6044 /* --add initiated by this node */
dbb64f86 6045 err = md_cluster_ops->add_new_disk(mddev, rdev);
1aee41f6 6046 if (err) {
1aee41f6
GR		 6047 export_rdev(rdev);
6048 return err;
6049 }
6050 }
6051 }
6052
1da177e4
LT		 6053 rdev->raid_disk = -1;
6054 err = bind_rdev_to_array(rdev, mddev);
dbb64f86 6055
1da177e4
LT		 6056 if (err)
6057 export_rdev(rdev);
dbb64f86
GR		 6058
6059 if (mddev_is_clustered(mddev)) {
6060 if (info->state & (1 << MD_DISK_CANDIDATE))
6061 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6062 else {
6063 if (err)
6064 md_cluster_ops->add_new_disk_cancel(mddev);
6065 else
6066 err = add_bound_rdev(rdev);
6067 }
6068
6069 } else if (!err)
a6da4ef8 6070 err = add_bound_rdev(rdev);
dbb64f86 6071
1da177e4
LT		 6072 return err;
6073 }
6074
6075 /* otherwise, add_new_disk is only allowed
6076 * for major_version==0 superblocks
6077 */
6078 if (mddev->major_version != 0) {
6079 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6080 mdname(mddev));
6081 return -EINVAL;
6082 }
6083
6084 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6085 int err;
d710e138 6086 rdev = md_import_device(dev, -1, 0);
1da177e4 6087 if (IS_ERR(rdev)) {
f72ffdd6 6088 printk(KERN_WARNING
1da177e4
LT		 6089 "md: error, md_import_device() returned %ld\n",
6090 PTR_ERR(rdev));
6091 return PTR_ERR(rdev);
6092 }
6093 rdev->desc_nr = info->number;
6094 if (info->raid_disk < mddev->raid_disks)
6095 rdev->raid_disk = info->raid_disk;
6096 else
6097 rdev->raid_disk = -1;
6098
1da177e4 6099 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 6100 if (info->state & (1<<MD_DISK_SYNC))
6101 set_bit(In_sync, &rdev->flags);
1da177e4 6102
8ddf9efe
N		 6103 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6104 set_bit(WriteMostly, &rdev->flags);
6105
1da177e4
LT		 6106 if (!mddev->persistent) {
6107 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 6108 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6109 } else
57b2caa3 6110 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 6111 rdev->sectors = rdev->sb_start;
1da177e4 6112
2bf071bf
N		 6113 err = bind_rdev_to_array(rdev, mddev);
6114 if (err) {
6115 export_rdev(rdev);
6116 return err;
6117 }
1da177e4
LT		 6118 }
6119
6120 return 0;
6121}
6122
f72ffdd6 6123static int hot_remove_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 6124{
6125 char b[BDEVNAME_SIZE];
3cb03002 6126 struct md_rdev *rdev;
23b63f9f 6127 int ret = -1;
1da177e4 6128
1da177e4
LT		 6129 rdev = find_rdev(mddev, dev);
6130 if (!rdev)
6131 return -ENXIO;
6132
293467aa 6133 if (mddev_is_clustered(mddev))
23b63f9f 6134 ret = md_cluster_ops->metadata_update_start(mddev);
293467aa 6135
2910ff17
GR		 6136 if (rdev->raid_disk < 0)
6137 goto kick_rdev;
293467aa 6138
3ea8929d
N		 6139 clear_bit(Blocked, &rdev->flags);
6140 remove_and_add_spares(mddev, rdev);
6141
1da177e4
LT		 6142 if (rdev->raid_disk >= 0)
6143 goto busy;
6144
2910ff17 6145kick_rdev:
23b63f9f 6146 if (mddev_is_clustered(mddev) && ret == 0)
88bcfef7
GR		 6147 md_cluster_ops->remove_disk(mddev, rdev);
6148
fb56dfef 6149 md_kick_rdev_from_array(rdev);
850b2b42 6150 md_update_sb(mddev, 1);
d7603b7e 6151 md_new_event(mddev);
1da177e4
LT		 6152
6153 return 0;
6154busy:
23b63f9f 6155 if (mddev_is_clustered(mddev) && ret == 0)
293467aa 6156 md_cluster_ops->metadata_update_cancel(mddev);
2910ff17 6157
fdefa4d8 6158 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 6159 bdevname(rdev->bdev,b), mdname(mddev));
6160 return -EBUSY;
6161}
6162
f72ffdd6 6163static int hot_add_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 6164{
6165 char b[BDEVNAME_SIZE];
6166 int err;
3cb03002 6167 struct md_rdev *rdev;
1da177e4
LT		 6168
6169 if (!mddev->pers)
6170 return -ENODEV;
6171
6172 if (mddev->major_version != 0) {
6173 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6174 " version-0 superblocks.\n",
6175 mdname(mddev));
6176 return -EINVAL;
6177 }
6178 if (!mddev->pers->hot_add_disk) {
f72ffdd6 6179 printk(KERN_WARNING
1da177e4
LT		 6180 "%s: personality does not support diskops!\n",
6181 mdname(mddev));
6182 return -EINVAL;
6183 }
6184
d710e138 6185 rdev = md_import_device(dev, -1, 0);
1da177e4 6186 if (IS_ERR(rdev)) {
f72ffdd6 6187 printk(KERN_WARNING
1da177e4
LT		 6188 "md: error, md_import_device() returned %ld\n",
6189 PTR_ERR(rdev));
6190 return -EINVAL;
6191 }
6192
6193 if (mddev->persistent)
57b2caa3 6194 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 6195 else
77304d2a 6196 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 6197
8190e754 6198 rdev->sectors = rdev->sb_start;
1da177e4 6199
b2d444d7 6200 if (test_bit(Faulty, &rdev->flags)) {
f72ffdd6 6201 printk(KERN_WARNING
1da177e4
LT		 6202 "md: can not hot-add faulty %s disk to %s!\n",
6203 bdevname(rdev->bdev,b), mdname(mddev));
6204 err = -EINVAL;
6205 goto abort_export;
6206 }
293467aa 6207
b2d444d7 6208 clear_bit(In_sync, &rdev->flags);
1da177e4 6209 rdev->desc_nr = -1;
5842730d 6210 rdev->saved_raid_disk = -1;
2bf071bf
N		 6211 err = bind_rdev_to_array(rdev, mddev);
6212 if (err)
2aa82191 6213 goto abort_export;
1da177e4
LT		 6214
6215 /*
6216 * The rest should better be atomic, we can have disk failures
6217 * noticed in interrupt contexts ...
6218 */
6219
1da177e4
LT		 6220 rdev->raid_disk = -1;
6221
850b2b42 6222 md_update_sb(mddev, 1);
1da177e4
LT		 6223 /*
6224 * Kick recovery, maybe this spare has to be added to the
6225 * array immediately.
6226 */
6227 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6228 md_wakeup_thread(mddev->thread);
d7603b7e 6229 md_new_event(mddev);
1da177e4
LT		 6230 return 0;
6231
1da177e4
LT		 6232abort_export:
6233 export_rdev(rdev);
6234 return err;
6235}
6236
fd01b88c 6237static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c 6238{
035328c2 6239 int err = 0;
32a7627c 6240
36fa3063 6241 if (mddev->pers) {
d66b1b39 6242 if (!mddev->pers->quiesce || !mddev->thread)
36fa3063
N		 6243 return -EBUSY;
6244 if (mddev->recovery || mddev->sync_thread)
6245 return -EBUSY;
6246 /* we should be able to change the bitmap.. */
6247 }
32a7627c 6248
36fa3063 6249 if (fd >= 0) {
035328c2 6250 struct inode *inode;
1e594bb2
N		 6251 struct file *f;
6252
6253 if (mddev->bitmap || mddev->bitmap_info.file)
36fa3063 6254 return -EEXIST; /* cannot add when bitmap is present */
1e594bb2 6255 f = fget(fd);
32a7627c 6256
1e594bb2 6257 if (f == NULL) {
36fa3063
N		 6258 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6259 mdname(mddev));
6260 return -EBADF;
6261 }
6262
1e594bb2 6263 inode = f->f_mapping->host;
035328c2
N		 6264 if (!S_ISREG(inode->i_mode)) {
6265 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6266 mdname(mddev));
6267 err = -EBADF;
1e594bb2 6268 } else if (!(f->f_mode & FMODE_WRITE)) {
035328c2
N		 6269 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6270 mdname(mddev));
6271 err = -EBADF;
6272 } else if (atomic_read(&inode->i_writecount) != 1) {
36fa3063
N		 6273 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6274 mdname(mddev));
035328c2
N		 6275 err = -EBUSY;
6276 }
6277 if (err) {
1e594bb2 6278 fput(f);
36fa3063
N		 6279 return err;
6280 }
1e594bb2 6281 mddev->bitmap_info.file = f;
c3d9714e 6282 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 6283 } else if (mddev->bitmap == NULL)
6284 return -ENOENT; /* cannot remove what isn't there */
6285 err = 0;
6286 if (mddev->pers) {
6287 mddev->pers->quiesce(mddev, 1);
69e51b44 6288 if (fd >= 0) {
f9209a32
GR		 6289 struct bitmap *bitmap;
6290
6291 bitmap = bitmap_create(mddev, -1);
6292 if (!IS_ERR(bitmap)) {
6293 mddev->bitmap = bitmap;
69e51b44 6294 err = bitmap_load(mddev);
ba599aca
N		 6295 } else
6296 err = PTR_ERR(bitmap);
69e51b44 6297 }
d7375ab3 6298 if (fd < 0 || err) {
36fa3063 6299 bitmap_destroy(mddev);
d7375ab3
N		 6300 fd = -1; /* make sure to put the file */
6301 }
36fa3063 6302 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 6303 }
6304 if (fd < 0) {
4af1a041
N		 6305 struct file *f = mddev->bitmap_info.file;
6306 if (f) {
6307 spin_lock(&mddev->lock);
6308 mddev->bitmap_info.file = NULL;
6309 spin_unlock(&mddev->lock);
6310 fput(f);
6311 }
36fa3063
N		 6312 }
6313
32a7627c
N		 6314 return err;
6315}
6316
1da177e4
LT		 6317/*
6318 * set_array_info is used two different ways
6319 * The original usage is when creating a new array.
6320 * In this usage, raid_disks is > 0 and it together with
6321 * level, size, not_persistent,layout,chunksize determine the
6322 * shape of the array.
6323 * This will always create an array with a type-0.90.0 superblock.
6324 * The newer usage is when assembling an array.
6325 * In this case raid_disks will be 0, and the major_version field is
6326 * use to determine which style super-blocks are to be found on the devices.
6327 * The minor and patch _version numbers are also kept incase the
6328 * super_block handler wishes to interpret them.
6329 */
f72ffdd6 6330static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6331{
6332
6333 if (info->raid_disks == 0) {
6334 /* just setting version number for superblock loading */
6335 if (info->major_version < 0 ||
50511da3 6336 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 6337 super_types[info->major_version].name == NULL) {
6338 /* maybe try to auto-load a module? */
f72ffdd6 6339 printk(KERN_INFO
1da177e4
LT		 6340 "md: superblock version %d not known\n",
6341 info->major_version);
6342 return -EINVAL;
6343 }
6344 mddev->major_version = info->major_version;
6345 mddev->minor_version = info->minor_version;
6346 mddev->patch_version = info->patch_version;
3f9d7b0d 6347 mddev->persistent = !info->not_persistent;
cbd19983
N		 6348 /* ensure mddev_put doesn't delete this now that there
6349 * is some minimal configuration.
6350 */
6351 mddev->ctime = get_seconds();
1da177e4
LT		 6352 return 0;
6353 }
6354 mddev->major_version = MD_MAJOR_VERSION;
6355 mddev->minor_version = MD_MINOR_VERSION;
6356 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6357 mddev->ctime = get_seconds();
6358
6359 mddev->level = info->level;
17115e03 6360 mddev->clevel[0] = 0;
58c0fed4 6361 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 6362 mddev->raid_disks = info->raid_disks;
6363 /* don't set md_minor, it is determined by which /dev/md* was
6364 * openned
6365 */
6366 if (info->state & (1<<MD_SB_CLEAN))
6367 mddev->recovery_cp = MaxSector;
6368 else
6369 mddev->recovery_cp = 0;
6370 mddev->persistent = ! info->not_persistent;
e691063a 6371 mddev->external = 0;
1da177e4
LT		 6372
6373 mddev->layout = info->layout;
9d8f0363 6374 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 6375
6376 mddev->max_disks = MD_SB_DISKS;
6377
e691063a
N		 6378 if (mddev->persistent)
6379 mddev->flags = 0;
850b2b42 6380 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6381
c3d9714e 6382 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6383 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6384 mddev->bitmap_info.offset = 0;
b2a2703c 6385
f6705578
N		 6386 mddev->reshape_position = MaxSector;
6387
1da177e4
LT		 6388 /*
6389 * Generate a 128 bit UUID
6390 */
6391 get_random_bytes(mddev->uuid, 16);
6392
f6705578 6393 mddev->new_level = mddev->level;
664e7c41 6394 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 6395 mddev->new_layout = mddev->layout;
6396 mddev->delta_disks = 0;
2c810cdd 6397 mddev->reshape_backwards = 0;
f6705578 6398
1da177e4
LT		 6399 return 0;
6400}
6401
fd01b88c 6402void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6403{
b522adcd
DW		 6404 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6405
6406 if (mddev->external_size)
6407 return;
6408
1f403624
DW		 6409 mddev->array_sectors = array_sectors;
6410}
6411EXPORT_SYMBOL(md_set_array_sectors);
6412
fd01b88c 6413static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6414{
3cb03002 6415 struct md_rdev *rdev;
a35b0d69 6416 int rv;
d71f9f88 6417 int fit = (num_sectors == 0);
a35b0d69
N		 6418
6419 if (mddev->pers->resize == NULL)
6420 return -EINVAL;
d71f9f88
AN		 6421 /* The "num_sectors" is the number of sectors of each device that
6422 * is used. This can only make sense for arrays with redundancy.
6423 * linear and raid0 always use whatever space is available. We can only
6424 * consider changing this number if no resync or reconstruction is
6425 * happening, and if the new size is acceptable. It must fit before the
0f420358 6426 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 6427 * of each device. If num_sectors is zero, we find the largest size
6428 * that fits.
a35b0d69 6429 */
f851b60d
N		 6430 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6431 mddev->sync_thread)
a35b0d69 6432 return -EBUSY;
bd8839e0
N		 6433 if (mddev->ro)
6434 return -EROFS;
a4a6125a 6435
dafb20fa 6436 rdev_for_each(rdev, mddev) {
dd8ac336 6437 sector_t avail = rdev->sectors;
01ab5662 6438
d71f9f88
AN		 6439 if (fit && (num_sectors == 0 || num_sectors > avail))
6440 num_sectors = avail;
6441 if (avail < num_sectors)
a35b0d69
N		 6442 return -ENOSPC;
6443 }
d71f9f88 6444 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 6445 if (!rv)
6446 revalidate_disk(mddev->gendisk);
a35b0d69
N		 6447 return rv;
6448}
6449
fd01b88c 6450static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 6451{
6452 int rv;
c6563a8c 6453 struct md_rdev *rdev;
da943b99 6454 /* change the number of raid disks */
63c70c4f 6455 if (mddev->pers->check_reshape == NULL)
da943b99 6456 return -EINVAL;
bd8839e0
N		 6457 if (mddev->ro)
6458 return -EROFS;
da943b99 6459 if (raid_disks <= 0 ||
233fca36 6460 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 6461 return -EINVAL;
f851b60d
N		 6462 if (mddev->sync_thread ||
6463 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6464 mddev->reshape_position != MaxSector)
da943b99 6465 return -EBUSY;
c6563a8c
N		 6466
6467 rdev_for_each(rdev, mddev) {
6468 if (mddev->raid_disks < raid_disks &&
6469 rdev->data_offset < rdev->new_data_offset)
6470 return -EINVAL;
6471 if (mddev->raid_disks > raid_disks &&
6472 rdev->data_offset > rdev->new_data_offset)
6473 return -EINVAL;
6474 }
6475
63c70c4f 6476 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N		 6477 if (mddev->delta_disks < 0)
6478 mddev->reshape_backwards = 1;
6479 else if (mddev->delta_disks > 0)
6480 mddev->reshape_backwards = 0;
63c70c4f
N		 6481
6482 rv = mddev->pers->check_reshape(mddev);
2c810cdd 6483 if (rv < 0) {
de171cb9 6484 mddev->delta_disks = 0;
2c810cdd
N		 6485 mddev->reshape_backwards = 0;
6486 }
da943b99
N		 6487 return rv;
6488}
6489
1da177e4
LT		 6490/*
6491 * update_array_info is used to change the configuration of an
6492 * on-line array.
6493 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6494 * fields in the info are checked against the array.
6495 * Any differences that cannot be handled will cause an error.
6496 * Normally, only one change can be managed at a time.
6497 */
fd01b88c 6498static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6499{
6500 int rv = 0;
6501 int cnt = 0;
36fa3063
N		 6502 int state = 0;
6503
6504 /* calculate expected state,ignoring low bits */
c3d9714e 6505 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 6506 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 6507
6508 if (mddev->major_version != info->major_version ||
6509 mddev->minor_version != info->minor_version ||
6510/* mddev->patch_version != info->patch_version || */
6511 mddev->ctime != info->ctime ||
6512 mddev->level != info->level ||
6513/* mddev->layout != info->layout || */
4e023612 6514 mddev->persistent != !info->not_persistent ||
9d8f0363 6515 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 6516 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6517 ((state^info->state) & 0xfffffe00)
6518 )
1da177e4
LT		 6519 return -EINVAL;
6520 /* Check there is only one change */
58c0fed4
AN		 6521 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6522 cnt++;
6523 if (mddev->raid_disks != info->raid_disks)
6524 cnt++;
6525 if (mddev->layout != info->layout)
6526 cnt++;
6527 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6528 cnt++;
6529 if (cnt == 0)
6530 return 0;
6531 if (cnt > 1)
6532 return -EINVAL;
1da177e4
LT		 6533
6534 if (mddev->layout != info->layout) {
6535 /* Change layout
6536 * we don't need to do anything at the md level, the
6537 * personality will take care of it all.
6538 */
50ac168a 6539 if (mddev->pers->check_reshape == NULL)
1da177e4 6540 return -EINVAL;
597a711b
N		 6541 else {
6542 mddev->new_layout = info->layout;
50ac168a 6543 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 6544 if (rv)
6545 mddev->new_layout = mddev->layout;
6546 return rv;
6547 }
1da177e4 6548 }
58c0fed4 6549 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6550 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6551
da943b99
N		 6552 if (mddev->raid_disks != info->raid_disks)
6553 rv = update_raid_disks(mddev, info->raid_disks);
6554
36fa3063 6555 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
293467aa
GR		 6556 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6557 rv = -EINVAL;
6558 goto err;
6559 }
6560 if (mddev->recovery || mddev->sync_thread) {
6561 rv = -EBUSY;
6562 goto err;
6563 }
36fa3063 6564 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
f9209a32 6565 struct bitmap *bitmap;
36fa3063 6566 /* add the bitmap */
293467aa
GR		 6567 if (mddev->bitmap) {
6568 rv = -EEXIST;
6569 goto err;
6570 }
6571 if (mddev->bitmap_info.default_offset == 0) {
6572 rv = -EINVAL;
6573 goto err;
6574 }
c3d9714e
N		 6575 mddev->bitmap_info.offset =
6576 mddev->bitmap_info.default_offset;
6409bb05
N		 6577 mddev->bitmap_info.space =
6578 mddev->bitmap_info.default_space;
36fa3063 6579 mddev->pers->quiesce(mddev, 1);
f9209a32
GR		 6580 bitmap = bitmap_create(mddev, -1);
6581 if (!IS_ERR(bitmap)) {
6582 mddev->bitmap = bitmap;
69e51b44 6583 rv = bitmap_load(mddev);
ba599aca
N		 6584 } else
6585 rv = PTR_ERR(bitmap);
36fa3063
N		 6586 if (rv)
6587 bitmap_destroy(mddev);
6588 mddev->pers->quiesce(mddev, 0);
6589 } else {
6590 /* remove the bitmap */
293467aa
GR		 6591 if (!mddev->bitmap) {
6592 rv = -ENOENT;
6593 goto err;
6594 }
6595 if (mddev->bitmap->storage.file) {
6596 rv = -EINVAL;
6597 goto err;
6598 }
36fa3063
N		 6599 mddev->pers->quiesce(mddev, 1);
6600 bitmap_destroy(mddev);
6601 mddev->pers->quiesce(mddev, 0);
c3d9714e 6602 mddev->bitmap_info.offset = 0;
36fa3063
N		 6603 }
6604 }
850b2b42 6605 md_update_sb(mddev, 1);
293467aa
GR		 6606 return rv;
6607err:
1da177e4
LT		 6608 return rv;
6609}
6610
fd01b88c 6611static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6612{
3cb03002 6613 struct md_rdev *rdev;
1ca69c4b 6614 int err = 0;
1da177e4
LT		 6615
6616 if (mddev->pers == NULL)
6617 return -ENODEV;
6618
1ca69c4b
N		 6619 rcu_read_lock();
6620 rdev = find_rdev_rcu(mddev, dev);
1da177e4 6621 if (!rdev)
1ca69c4b
N		 6622 err = -ENODEV;
6623 else {
6624 md_error(mddev, rdev);
6625 if (!test_bit(Faulty, &rdev->flags))
6626 err = -EBUSY;
6627 }
6628 rcu_read_unlock();
6629 return err;
1da177e4
LT		 6630}
6631
2f9618ce
AN		 6632/*
6633 * We have a problem here : there is no easy way to give a CHS
6634 * virtual geometry. We currently pretend that we have a 2 heads
6635 * 4 sectors (with a BIG number of cylinders...). This drives
6636 * dosfs just mad... ;-)
6637 */
a885c8c4
CH		 6638static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6639{
fd01b88c 6640 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6641
6642 geo->heads = 2;
6643 geo->sectors = 4;
49ce6cea 6644 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6645 return 0;
6646}
6647
cb335f88
NS		 6648static inline bool md_ioctl_valid(unsigned int cmd)
6649{
6650 switch (cmd) {
6651 case ADD_NEW_DISK:
6652 case BLKROSET:
6653 case GET_ARRAY_INFO:
6654 case GET_BITMAP_FILE:
6655 case GET_DISK_INFO:
6656 case HOT_ADD_DISK:
6657 case HOT_REMOVE_DISK:
cb335f88
NS		 6658 case RAID_AUTORUN:
6659 case RAID_VERSION:
6660 case RESTART_ARRAY_RW:
6661 case RUN_ARRAY:
6662 case SET_ARRAY_INFO:
6663 case SET_BITMAP_FILE:
6664 case SET_DISK_FAULTY:
6665 case STOP_ARRAY:
6666 case STOP_ARRAY_RO:
1aee41f6 6667 case CLUSTERED_DISK_NACK:
cb335f88
NS		 6668 return true;
6669 default:
6670 return false;
6671 }
6672}
6673
a39907fa 6674static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6675 unsigned int cmd, unsigned long arg)
6676{
6677 int err = 0;
6678 void __user *argp = (void __user *)arg;
fd01b88c 6679 struct mddev *mddev = NULL;
e2218350 6680 int ro;
1da177e4 6681
cb335f88
NS		 6682 if (!md_ioctl_valid(cmd))
6683 return -ENOTTY;
6684
506c9e44
N		 6685 switch (cmd) {
6686 case RAID_VERSION:
6687 case GET_ARRAY_INFO:
6688 case GET_DISK_INFO:
6689 break;
6690 default:
6691 if (!capable(CAP_SYS_ADMIN))
6692 return -EACCES;
6693 }
1da177e4
LT		 6694
6695 /*
6696 * Commands dealing with the RAID driver but not any
6697 * particular array:
6698 */
c02c0aeb
N		 6699 switch (cmd) {
6700 case RAID_VERSION:
6701 err = get_version(argp);
3adc28d8 6702 goto out;
1da177e4 6703
1da177e4 6704#ifndef MODULE
c02c0aeb
N		 6705 case RAID_AUTORUN:
6706 err = 0;
6707 autostart_arrays(arg);
3adc28d8 6708 goto out;
1da177e4 6709#endif
c02c0aeb 6710 default:;
1da177e4
LT		 6711 }
6712
6713 /*
6714 * Commands creating/starting a new array:
6715 */
6716
a39907fa 6717 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6718
6719 if (!mddev) {
6720 BUG();
3adc28d8 6721 goto out;
1da177e4
LT		 6722 }
6723
1ca69c4b
N		 6724 /* Some actions do not requires the mutex */
6725 switch (cmd) {
6726 case GET_ARRAY_INFO:
6727 if (!mddev->raid_disks && !mddev->external)
6728 err = -ENODEV;
6729 else
6730 err = get_array_info(mddev, argp);
3adc28d8 6731 goto out;
1ca69c4b
N		 6732
6733 case GET_DISK_INFO:
6734 if (!mddev->raid_disks && !mddev->external)
6735 err = -ENODEV;
6736 else
6737 err = get_disk_info(mddev, argp);
3adc28d8 6738 goto out;
1ca69c4b
N		 6739
6740 case SET_DISK_FAULTY:
6741 err = set_disk_faulty(mddev, new_decode_dev(arg));
3adc28d8 6742 goto out;
4af1a041
N		 6743
6744 case GET_BITMAP_FILE:
6745 err = get_bitmap_file(mddev, argp);
6746 goto out;
6747
1ca69c4b
N		 6748 }
6749
a7a3f08d
N		 6750 if (cmd == ADD_NEW_DISK)
6751 /* need to ensure md_delayed_delete() has completed */
6752 flush_workqueue(md_misc_wq);
6753
90f5f7ad
HR		 6754 if (cmd == HOT_REMOVE_DISK)
6755 /* need to ensure recovery thread has run */
6756 wait_event_interruptible_timeout(mddev->sb_wait,
6757 !test_bit(MD_RECOVERY_NEEDED,
6758 &mddev->flags),
6759 msecs_to_jiffies(5000));
260fa034
N		 6760 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6761 /* Need to flush page cache, and ensure no-one else opens
6762 * and writes
6763 */
6764 mutex_lock(&mddev->open_mutex);
9ba3b7f5 6765 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
260fa034
N		 6766 mutex_unlock(&mddev->open_mutex);
6767 err = -EBUSY;
3adc28d8 6768 goto out;
260fa034
N		 6769 }
6770 set_bit(MD_STILL_CLOSED, &mddev->flags);
6771 mutex_unlock(&mddev->open_mutex);
6772 sync_blockdev(bdev);
6773 }
1da177e4
LT		 6774 err = mddev_lock(mddev);
6775 if (err) {
f72ffdd6 6776 printk(KERN_INFO
1da177e4
LT		 6777 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6778 err, cmd);
3adc28d8 6779 goto out;
1da177e4
LT		 6780 }
6781
c02c0aeb
N		 6782 if (cmd == SET_ARRAY_INFO) {
6783 mdu_array_info_t info;
6784 if (!arg)
6785 memset(&info, 0, sizeof(info));
6786 else if (copy_from_user(&info, argp, sizeof(info))) {
6787 err = -EFAULT;
3adc28d8 6788 goto unlock;
c02c0aeb
N		 6789 }
6790 if (mddev->pers) {
6791 err = update_array_info(mddev, &info);
6792 if (err) {
6793 printk(KERN_WARNING "md: couldn't update"
6794 " array info. %d\n", err);
3adc28d8 6795 goto unlock;
1da177e4 6796 }
3adc28d8 6797 goto unlock;
c02c0aeb
N		 6798 }
6799 if (!list_empty(&mddev->disks)) {
6800 printk(KERN_WARNING
6801 "md: array %s already has disks!\n",
6802 mdname(mddev));
6803 err = -EBUSY;
3adc28d8 6804 goto unlock;
c02c0aeb
N		 6805 }
6806 if (mddev->raid_disks) {
6807 printk(KERN_WARNING
6808 "md: array %s already initialised!\n",
6809 mdname(mddev));
6810 err = -EBUSY;
3adc28d8 6811 goto unlock;
c02c0aeb
N		 6812 }
6813 err = set_array_info(mddev, &info);
6814 if (err) {
6815 printk(KERN_WARNING "md: couldn't set"
6816 " array info. %d\n", err);
3adc28d8 6817 goto unlock;
c02c0aeb 6818 }
3adc28d8 6819 goto unlock;
1da177e4
LT		 6820 }
6821
6822 /*
6823 * Commands querying/configuring an existing array:
6824 */
32a7627c 6825 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6826 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6827 if ((!mddev->raid_disks && !mddev->external)
6828 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6829 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6830 && cmd != GET_BITMAP_FILE) {
1da177e4 6831 err = -ENODEV;
3adc28d8 6832 goto unlock;
1da177e4
LT		 6833 }
6834
6835 /*
6836 * Commands even a read-only array can execute:
6837 */
c02c0aeb 6838 switch (cmd) {
c02c0aeb
N		 6839 case RESTART_ARRAY_RW:
6840 err = restart_array(mddev);
3adc28d8 6841 goto unlock;
1da177e4 6842
c02c0aeb
N		 6843 case STOP_ARRAY:
6844 err = do_md_stop(mddev, 0, bdev);
3adc28d8 6845 goto unlock;
1da177e4 6846
c02c0aeb
N		 6847 case STOP_ARRAY_RO:
6848 err = md_set_readonly(mddev, bdev);
3adc28d8 6849 goto unlock;
1da177e4 6850
3ea8929d
N		 6851 case HOT_REMOVE_DISK:
6852 err = hot_remove_disk(mddev, new_decode_dev(arg));
3adc28d8 6853 goto unlock;
3ea8929d 6854
7ceb17e8
N		 6855 case ADD_NEW_DISK:
6856 /* We can support ADD_NEW_DISK on read-only arrays
6857 * on if we are re-adding a preexisting device.
6858 * So require mddev->pers and MD_DISK_SYNC.
6859 */
6860 if (mddev->pers) {
6861 mdu_disk_info_t info;
6862 if (copy_from_user(&info, argp, sizeof(info)))
6863 err = -EFAULT;
6864 else if (!(info.state & (1<<MD_DISK_SYNC)))
6865 /* Need to clear read-only for this */
6866 break;
6867 else
6868 err = add_new_disk(mddev, &info);
3adc28d8 6869 goto unlock;
7ceb17e8
N		 6870 }
6871 break;
6872
c02c0aeb
N		 6873 case BLKROSET:
6874 if (get_user(ro, (int __user *)(arg))) {
6875 err = -EFAULT;
3adc28d8 6876 goto unlock;
c02c0aeb
N		 6877 }
6878 err = -EINVAL;
e2218350 6879
c02c0aeb
N		 6880 /* if the bdev is going readonly the value of mddev->ro
6881 * does not matter, no writes are coming
6882 */
6883 if (ro)
3adc28d8 6884 goto unlock;
e2218350 6885
c02c0aeb
N		 6886 /* are we are already prepared for writes? */
6887 if (mddev->ro != 1)
3adc28d8 6888 goto unlock;
e2218350 6889
c02c0aeb
N		 6890 /* transitioning to readauto need only happen for
6891 * arrays that call md_write_start
6892 */
6893 if (mddev->pers) {
6894 err = restart_array(mddev);
6895 if (err == 0) {
6896 mddev->ro = 2;
6897 set_disk_ro(mddev->gendisk, 0);
e2218350 6898 }
c02c0aeb 6899 }
3adc28d8 6900 goto unlock;
1da177e4
LT		 6901 }
6902
6903 /*
6904 * The remaining ioctls are changing the state of the
f91de92e 6905 * superblock, so we do not allow them on read-only arrays.
1da177e4 6906 */
326eb17d 6907 if (mddev->ro && mddev->pers) {
f91de92e
N		 6908 if (mddev->ro == 2) {
6909 mddev->ro = 0;
00bcb4ac 6910 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 6911 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N		 6912 /* mddev_unlock will wake thread */
6913 /* If a device failed while we were read-only, we
6914 * need to make sure the metadata is updated now.
6915 */
6916 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6917 mddev_unlock(mddev);
6918 wait_event(mddev->sb_wait,
6919 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6920 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
29f097c4 6921 mddev_lock_nointr(mddev);
f3378b48 6922 }
f91de92e
N		 6923 } else {
6924 err = -EROFS;
3adc28d8 6925 goto unlock;
f91de92e 6926 }
1da177e4
LT		 6927 }
6928
c02c0aeb
N		 6929 switch (cmd) {
6930 case ADD_NEW_DISK:
1da177e4 6931 {
c02c0aeb
N		 6932 mdu_disk_info_t info;
6933 if (copy_from_user(&info, argp, sizeof(info)))
6934 err = -EFAULT;
6935 else
6936 err = add_new_disk(mddev, &info);
3adc28d8 6937 goto unlock;
c02c0aeb 6938 }
1da177e4 6939
1aee41f6
GR		 6940 case CLUSTERED_DISK_NACK:
6941 if (mddev_is_clustered(mddev))
6942 md_cluster_ops->new_disk_ack(mddev, false);
6943 else
6944 err = -EINVAL;
6945 goto unlock;
6946
c02c0aeb
N		 6947 case HOT_ADD_DISK:
6948 err = hot_add_disk(mddev, new_decode_dev(arg));
3adc28d8 6949 goto unlock;
1da177e4 6950
c02c0aeb
N		 6951 case RUN_ARRAY:
6952 err = do_md_run(mddev);
3adc28d8 6953 goto unlock;
1da177e4 6954
c02c0aeb
N		 6955 case SET_BITMAP_FILE:
6956 err = set_bitmap_file(mddev, (int)arg);
3adc28d8 6957 goto unlock;
32a7627c 6958
c02c0aeb
N		 6959 default:
6960 err = -EINVAL;
3adc28d8 6961 goto unlock;
1da177e4
LT		 6962 }
6963
3adc28d8 6964unlock:
d3374825
N		 6965 if (mddev->hold_active == UNTIL_IOCTL &&
6966 err != -EINVAL)
6967 mddev->hold_active = 0;
1da177e4 6968 mddev_unlock(mddev);
3adc28d8 6969out:
1da177e4
LT		 6970 return err;
6971}
aa98aa31
AB		 6972#ifdef CONFIG_COMPAT
6973static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6974 unsigned int cmd, unsigned long arg)
6975{
6976 switch (cmd) {
6977 case HOT_REMOVE_DISK:
6978 case HOT_ADD_DISK:
6979 case SET_DISK_FAULTY:
6980 case SET_BITMAP_FILE:
6981 /* These take in integer arg, do not convert */
6982 break;
6983 default:
6984 arg = (unsigned long)compat_ptr(arg);
6985 break;
6986 }
6987
6988 return md_ioctl(bdev, mode, cmd, arg);
6989}
6990#endif /* CONFIG_COMPAT */
1da177e4 6991
a39907fa 6992static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6993{
6994 /*
6995 * Succeed if we can lock the mddev, which confirms that
6996 * it isn't being stopped right now.
6997 */
fd01b88c 6998 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6999 int err;
7000
0c098220
YL		 7001 if (!mddev)
7002 return -ENODEV;
7003
d3374825
N		 7004 if (mddev->gendisk != bdev->bd_disk) {
7005 /* we are racing with mddev_put which is discarding this
7006 * bd_disk.
7007 */
7008 mddev_put(mddev);
7009 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 7010 flush_workqueue(md_misc_wq);
d3374825
N		 7011 /* Then retry the open from the top */
7012 return -ERESTARTSYS;
7013 }
7014 BUG_ON(mddev != bdev->bd_disk->private_data);
7015
c8c00a69 7016 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 7017 goto out;
7018
7019 err = 0;
f2ea68cf 7020 atomic_inc(&mddev->openers);
260fa034 7021 clear_bit(MD_STILL_CLOSED, &mddev->flags);
c8c00a69 7022 mutex_unlock(&mddev->open_mutex);
1da177e4 7023
f0b4f7e2 7024 check_disk_change(bdev);
1da177e4
LT		 7025 out:
7026 return err;
7027}
7028
db2a144b 7029static void md_release(struct gendisk *disk, fmode_t mode)
1da177e4 7030{
f72ffdd6 7031 struct mddev *mddev = disk->private_data;
1da177e4 7032
52e5f9d1 7033 BUG_ON(!mddev);
f2ea68cf 7034 atomic_dec(&mddev->openers);
1da177e4 7035 mddev_put(mddev);
1da177e4 7036}
f0b4f7e2
N		 7037
7038static int md_media_changed(struct gendisk *disk)
7039{
fd01b88c 7040 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 7041
7042 return mddev->changed;
7043}
7044
7045static int md_revalidate(struct gendisk *disk)
7046{
fd01b88c 7047 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 7048
7049 mddev->changed = 0;
7050 return 0;
7051}
83d5cde4 7052static const struct block_device_operations md_fops =
1da177e4
LT		 7053{
7054 .owner = THIS_MODULE,
a39907fa
AV		 7055 .open = md_open,
7056 .release = md_release,
b492b852 7057 .ioctl = md_ioctl,
aa98aa31
AB		 7058#ifdef CONFIG_COMPAT
7059 .compat_ioctl = md_compat_ioctl,
7060#endif
a885c8c4 7061 .getgeo = md_getgeo,
f0b4f7e2
N		 7062 .media_changed = md_media_changed,
7063 .revalidate_disk= md_revalidate,
1da177e4
LT		 7064};
7065
f72ffdd6 7066static int md_thread(void *arg)
1da177e4 7067{
2b8bf345 7068 struct md_thread *thread = arg;
1da177e4 7069
1da177e4
LT		 7070 /*
7071 * md_thread is a 'system-thread', it's priority should be very
7072 * high. We avoid resource deadlocks individually in each
7073 * raid personality. (RAID5 does preallocation) We also use RR and
7074 * the very same RT priority as kswapd, thus we will never get
7075 * into a priority inversion deadlock.
7076 *
7077 * we definitely have to have equal or higher priority than
7078 * bdflush, otherwise bdflush will deadlock if there are too
7079 * many dirty RAID5 blocks.
7080 */
1da177e4 7081
6985c43f 7082 allow_signal(SIGKILL);
a6fb0934 7083 while (!kthread_should_stop()) {
1da177e4 7084
93588e22
N		 7085 /* We need to wait INTERRUPTIBLE so that
7086 * we don't add to the load-average.
7087 * That means we need to be sure no signals are
7088 * pending
7089 */
7090 if (signal_pending(current))
7091 flush_signals(current);
7092
7093 wait_event_interruptible_timeout
7094 (thread->wqueue,
7095 test_bit(THREAD_WAKEUP, &thread->flags)
7096 || kthread_should_stop(),
7097 thread->timeout);
1da177e4 7098
6c987910
N		 7099 clear_bit(THREAD_WAKEUP, &thread->flags);
7100 if (!kthread_should_stop())
4ed8731d 7101 thread->run(thread);
1da177e4 7102 }
a6fb0934 7103
1da177e4
LT		 7104 return 0;
7105}
7106
2b8bf345 7107void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 7108{
7109 if (thread) {
36a4e1fe 7110 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 7111 set_bit(THREAD_WAKEUP, &thread->flags);
7112 wake_up(&thread->wqueue);
7113 }
7114}
6c144d31 7115EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 7116
4ed8731d
SL		 7117struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7118 struct mddev *mddev, const char *name)
1da177e4 7119{
2b8bf345 7120 struct md_thread *thread;
1da177e4 7121
2b8bf345 7122 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 7123 if (!thread)
7124 return NULL;
7125
1da177e4
LT		 7126 init_waitqueue_head(&thread->wqueue);
7127
1da177e4
LT		 7128 thread->run = run;
7129 thread->mddev = mddev;
32a7627c 7130 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 7131 thread->tsk = kthread_run(md_thread, thread,
7132 "%s_%s",
7133 mdname(thread->mddev),
0232605d 7134 name);
a6fb0934 7135 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 7136 kfree(thread);
7137 return NULL;
7138 }
1da177e4
LT		 7139 return thread;
7140}
6c144d31 7141EXPORT_SYMBOL(md_register_thread);
1da177e4 7142
2b8bf345 7143void md_unregister_thread(struct md_thread **threadp)
1da177e4 7144{
2b8bf345 7145 struct md_thread *thread = *threadp;
e0cf8f04
N		 7146 if (!thread)
7147 return;
36a4e1fe 7148 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 7149 /* Locking ensures that mddev_unlock does not wake_up a
7150 * non-existent thread
7151 */
7152 spin_lock(&pers_lock);
7153 *threadp = NULL;
7154 spin_unlock(&pers_lock);
a6fb0934
N		 7155
7156 kthread_stop(thread->tsk);
1da177e4
LT		 7157 kfree(thread);
7158}
6c144d31 7159EXPORT_SYMBOL(md_unregister_thread);
1da177e4 7160
fd01b88c 7161void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7162{
b2d444d7 7163 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 7164 return;
6bfe0b49 7165
de393cde 7166 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 7167 return;
7168 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 7169 if (mddev->degraded)
7170 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 7171 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 7172 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7173 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7174 md_wakeup_thread(mddev->thread);
768a418d 7175 if (mddev->event_work.func)
e804ac78 7176 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 7177 md_new_event_inintr(mddev);
1da177e4 7178}
6c144d31 7179EXPORT_SYMBOL(md_error);
1da177e4
LT		 7180
7181/* seq_file implementation /proc/mdstat */
7182
7183static void status_unused(struct seq_file *seq)
7184{
7185 int i = 0;
3cb03002 7186 struct md_rdev *rdev;
1da177e4
LT		 7187
7188 seq_printf(seq, "unused devices: ");
7189
159ec1fc 7190 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 7191 char b[BDEVNAME_SIZE];
7192 i++;
7193 seq_printf(seq, "%s ",
7194 bdevname(rdev->bdev,b));
7195 }
7196 if (!i)
7197 seq_printf(seq, "<none>");
7198
7199 seq_printf(seq, "\n");
7200}
7201
f7851be7 7202static int status_resync(struct seq_file *seq, struct mddev *mddev)
1da177e4 7203{
dd71cf6b
N		 7204 sector_t max_sectors, resync, res;
7205 unsigned long dt, db;
7206 sector_t rt;
4588b42e
N		 7207 int scale;
7208 unsigned int per_milli;
1da177e4 7209
c804cdec
N		 7210 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7211 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 7212 max_sectors = mddev->resync_max_sectors;
1da177e4 7213 else
dd71cf6b 7214 max_sectors = mddev->dev_sectors;
1da177e4 7215
f7851be7
N		 7216 resync = mddev->curr_resync;
7217 if (resync <= 3) {
7218 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7219 /* Still cleaning up */
7220 resync = max_sectors;
7221 } else
7222 resync -= atomic_read(&mddev->recovery_active);
7223
7224 if (resync == 0) {
7225 if (mddev->recovery_cp < MaxSector) {
7226 seq_printf(seq, "\tresync=PENDING");
7227 return 1;
7228 }
7229 return 0;
7230 }
7231 if (resync < 3) {
7232 seq_printf(seq, "\tresync=DELAYED");
7233 return 1;
7234 }
7235
403df478 7236 WARN_ON(max_sectors == 0);
4588b42e 7237 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 7238 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 7239 * u32, as those are the requirements for sector_div.
7240 * Thus 'scale' must be at least 10
7241 */
7242 scale = 10;
7243 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 7244 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 7245 scale++;
7246 }
7247 res = (resync>>scale)*1000;
dd71cf6b 7248 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 7249
7250 per_milli = res;
1da177e4 7251 {
4588b42e 7252 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 7253 seq_printf(seq, "[");
7254 for (i = 0; i < x; i++)
7255 seq_printf(seq, "=");
7256 seq_printf(seq, ">");
7257 for (i = 0; i < y; i++)
7258 seq_printf(seq, ".");
7259 seq_printf(seq, "] ");
7260 }
4588b42e 7261 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 7262 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7263 "reshape" :
61df9d91
N		 7264 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7265 "check" :
7266 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7267 "resync" : "recovery"))),
7268 per_milli/10, per_milli % 10,
dd71cf6b
N		 7269 (unsigned long long) resync/2,
7270 (unsigned long long) max_sectors/2);
1da177e4
LT		 7271
7272 /*
1da177e4
LT		 7273 * dt: time from mark until now
7274 * db: blocks written from mark until now
7275 * rt: remaining time
dd71cf6b
N		 7276 *
7277 * rt is a sector_t, so could be 32bit or 64bit.
7278 * So we divide before multiply in case it is 32bit and close
7279 * to the limit.
25985edc 7280 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 7281 * near the end of resync when the number of remaining sectors
7282 * is close to 'db'.
7283 * We then divide rt by 32 after multiplying by db to compensate.
7284 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 7285 */
7286 dt = ((jiffies - mddev->resync_mark) / HZ);
7287 if (!dt) dt++;
ff4e8d9a
N		 7288 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7289 - mddev->resync_mark_cnt;
1da177e4 7290
dd71cf6b
N		 7291 rt = max_sectors - resync; /* number of remaining sectors */
7292 sector_div(rt, db/32+1);
7293 rt *= dt;
7294 rt >>= 5;
7295
7296 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7297 ((unsigned long)rt % 60)/6);
1da177e4 7298
ff4e8d9a 7299 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
f7851be7 7300 return 1;
1da177e4
LT		 7301}
7302
7303static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7304{
7305 struct list_head *tmp;
7306 loff_t l = *pos;
fd01b88c 7307 struct mddev *mddev;
1da177e4
LT		 7308
7309 if (l >= 0x10000)
7310 return NULL;
7311 if (!l--)
7312 /* header */
7313 return (void*)1;
7314
7315 spin_lock(&all_mddevs_lock);
7316 list_for_each(tmp,&all_mddevs)
7317 if (!l--) {
fd01b88c 7318 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 7319 mddev_get(mddev);
7320 spin_unlock(&all_mddevs_lock);
7321 return mddev;
7322 }
7323 spin_unlock(&all_mddevs_lock);
7324 if (!l--)
7325 return (void*)2;/* tail */
7326 return NULL;
7327}
7328
7329static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7330{
7331 struct list_head *tmp;
fd01b88c 7332 struct mddev *next_mddev, *mddev = v;
f72ffdd6 7333
1da177e4
LT		 7334 ++*pos;
7335 if (v == (void*)2)
7336 return NULL;
7337
7338 spin_lock(&all_mddevs_lock);
7339 if (v == (void*)1)
7340 tmp = all_mddevs.next;
7341 else
7342 tmp = mddev->all_mddevs.next;
7343 if (tmp != &all_mddevs)
fd01b88c 7344 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 7345 else {
7346 next_mddev = (void*)2;
7347 *pos = 0x10000;
f72ffdd6 7348 }
1da177e4
LT		 7349 spin_unlock(&all_mddevs_lock);
7350
7351 if (v != (void*)1)
7352 mddev_put(mddev);
7353 return next_mddev;
7354
7355}
7356
7357static void md_seq_stop(struct seq_file *seq, void *v)
7358{
fd01b88c 7359 struct mddev *mddev = v;
1da177e4
LT		 7360
7361 if (mddev && v != (void*)1 && v != (void*)2)
7362 mddev_put(mddev);
7363}
7364
7365static int md_seq_show(struct seq_file *seq, void *v)
7366{
fd01b88c 7367 struct mddev *mddev = v;
dd8ac336 7368 sector_t sectors;
3cb03002 7369 struct md_rdev *rdev;
1da177e4
LT		 7370
7371 if (v == (void*)1) {
84fc4b56 7372 struct md_personality *pers;
1da177e4
LT		 7373 seq_printf(seq, "Personalities : ");
7374 spin_lock(&pers_lock);
2604b703
N		 7375 list_for_each_entry(pers, &pers_list, list)
7376 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 7377
7378 spin_unlock(&pers_lock);
7379 seq_printf(seq, "\n");
f1514638 7380 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7381 return 0;
7382 }
7383 if (v == (void*)2) {
7384 status_unused(seq);
7385 return 0;
7386 }
7387
36d091f4 7388 spin_lock(&mddev->lock);
1da177e4
LT		 7389 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7390 seq_printf(seq, "%s : %sactive", mdname(mddev),
7391 mddev->pers ? "" : "in");
7392 if (mddev->pers) {
f91de92e 7393 if (mddev->ro==1)
1da177e4 7394 seq_printf(seq, " (read-only)");
f91de92e 7395 if (mddev->ro==2)
52720ae7 7396 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 7397 seq_printf(seq, " %s", mddev->pers->name);
7398 }
7399
dd8ac336 7400 sectors = 0;
f97fcad3
N		 7401 rcu_read_lock();
7402 rdev_for_each_rcu(rdev, mddev) {
1da177e4
LT		 7403 char b[BDEVNAME_SIZE];
7404 seq_printf(seq, " %s[%d]",
7405 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 7406 if (test_bit(WriteMostly, &rdev->flags))
7407 seq_printf(seq, "(W)");
9efdca16
SL		 7408 if (test_bit(Journal, &rdev->flags))
7409 seq_printf(seq, "(J)");
b2d444d7 7410 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7411 seq_printf(seq, "(F)");
7412 continue;
2d78f8c4
N		 7413 }
7414 if (rdev->raid_disk < 0)
b325a32e 7415 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 7416 if (test_bit(Replacement, &rdev->flags))
7417 seq_printf(seq, "(R)");
dd8ac336 7418 sectors += rdev->sectors;
1da177e4 7419 }
f97fcad3 7420 rcu_read_unlock();
1da177e4
LT		 7421
7422 if (!list_empty(&mddev->disks)) {
7423 if (mddev->pers)
7424 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 7425 (unsigned long long)
7426 mddev->array_sectors / 2);
1da177e4
LT		 7427 else
7428 seq_printf(seq, "\n %llu blocks",
dd8ac336 7429 (unsigned long long)sectors / 2);
1da177e4 7430 }
1cd6bf19
N		 7431 if (mddev->persistent) {
7432 if (mddev->major_version != 0 ||
7433 mddev->minor_version != 90) {
7434 seq_printf(seq," super %d.%d",
7435 mddev->major_version,
7436 mddev->minor_version);
7437 }
e691063a
N		 7438 } else if (mddev->external)
7439 seq_printf(seq, " super external:%s",
7440 mddev->metadata_type);
7441 else
1cd6bf19 7442 seq_printf(seq, " super non-persistent");
1da177e4
LT		 7443
7444 if (mddev->pers) {
d710e138 7445 mddev->pers->status(seq, mddev);
f72ffdd6 7446 seq_printf(seq, "\n ");
8e1b39d6 7447 if (mddev->pers->sync_request) {
f7851be7 7448 if (status_resync(seq, mddev))
8e1b39d6 7449 seq_printf(seq, "\n ");
8e1b39d6 7450 }
32a7627c
N		 7451 } else
7452 seq_printf(seq, "\n ");
7453
57148964 7454 bitmap_status(seq, mddev->bitmap);
1da177e4
LT		 7455
7456 seq_printf(seq, "\n");
7457 }
36d091f4 7458 spin_unlock(&mddev->lock);
f72ffdd6 7459
1da177e4
LT		 7460 return 0;
7461}
7462
110518bc 7463static const struct seq_operations md_seq_ops = {
1da177e4
LT		 7464 .start = md_seq_start,
7465 .next = md_seq_next,
7466 .stop = md_seq_stop,
7467 .show = md_seq_show,
7468};
7469
7470static int md_seq_open(struct inode *inode, struct file *file)
7471{
f1514638 7472 struct seq_file *seq;
1da177e4
LT		 7473 int error;
7474
7475 error = seq_open(file, &md_seq_ops);
d7603b7e 7476 if (error)
f1514638
KS		 7477 return error;
7478
7479 seq = file->private_data;
7480 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7481 return error;
7482}
7483
e2f23b60 7484static int md_unloading;
d7603b7e
N		 7485static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7486{
f1514638 7487 struct seq_file *seq = filp->private_data;
d7603b7e
N		 7488 int mask;
7489
e2f23b60 7490 if (md_unloading)
7d7e64f2 7491 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
d7603b7e
N		 7492 poll_wait(filp, &md_event_waiters, wait);
7493
7494 /* always allow read */
7495 mask = POLLIN | POLLRDNORM;
7496
f1514638 7497 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 7498 mask |= POLLERR | POLLPRI;
7499 return mask;
7500}
7501
fa027c2a 7502static const struct file_operations md_seq_fops = {
e24650c2 7503 .owner = THIS_MODULE,
1da177e4
LT		 7504 .open = md_seq_open,
7505 .read = seq_read,
7506 .llseek = seq_lseek,
c3f94b40 7507 .release = seq_release_private,
d7603b7e 7508 .poll = mdstat_poll,
1da177e4
LT		 7509};
7510
84fc4b56 7511int register_md_personality(struct md_personality *p)
1da177e4 7512{
50bd3774
CY		 7513 printk(KERN_INFO "md: %s personality registered for level %d\n",
7514 p->name, p->level);
1da177e4 7515 spin_lock(&pers_lock);
2604b703 7516 list_add_tail(&p->list, &pers_list);
1da177e4
LT		 7517 spin_unlock(&pers_lock);
7518 return 0;
7519}
6c144d31 7520EXPORT_SYMBOL(register_md_personality);
1da177e4 7521
84fc4b56 7522int unregister_md_personality(struct md_personality *p)
1da177e4 7523{
2604b703 7524 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7525 spin_lock(&pers_lock);
2604b703 7526 list_del_init(&p->list);
1da177e4
LT		 7527 spin_unlock(&pers_lock);
7528 return 0;
7529}
6c144d31 7530EXPORT_SYMBOL(unregister_md_personality);
1da177e4 7531
6022e75b
N		 7532int register_md_cluster_operations(struct md_cluster_operations *ops,
7533 struct module *module)
edb39c9d 7534{
6022e75b 7535 int ret = 0;
edb39c9d 7536 spin_lock(&pers_lock);
6022e75b
N		 7537 if (md_cluster_ops != NULL)
7538 ret = -EALREADY;
7539 else {
7540 md_cluster_ops = ops;
7541 md_cluster_mod = module;
7542 }
edb39c9d 7543 spin_unlock(&pers_lock);
6022e75b 7544 return ret;
edb39c9d
GR		 7545}
7546EXPORT_SYMBOL(register_md_cluster_operations);
7547
7548int unregister_md_cluster_operations(void)
7549{
7550 spin_lock(&pers_lock);
7551 md_cluster_ops = NULL;
7552 spin_unlock(&pers_lock);
7553 return 0;
7554}
7555EXPORT_SYMBOL(unregister_md_cluster_operations);
7556
7557int md_setup_cluster(struct mddev *mddev, int nodes)
7558{
7559 int err;
7560
7561 err = request_module("md-cluster");
7562 if (err) {
7563 pr_err("md-cluster module not found.\n");
b0c26a79 7564 return -ENOENT;
edb39c9d
GR		 7565 }
7566
7567 spin_lock(&pers_lock);
7568 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7569 spin_unlock(&pers_lock);
7570 return -ENOENT;
7571 }
7572 spin_unlock(&pers_lock);
7573
cf921cc1 7574 return md_cluster_ops->join(mddev, nodes);
edb39c9d
GR		 7575}
7576
7577void md_cluster_stop(struct mddev *mddev)
7578{
c4ce867f
GR		 7579 if (!md_cluster_ops)
7580 return;
edb39c9d
GR		 7581 md_cluster_ops->leave(mddev);
7582 module_put(md_cluster_mod);
7583}
7584
fd01b88c 7585static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7586{
f72ffdd6 7587 struct md_rdev *rdev;
1da177e4 7588 int idle;
eea1bf38 7589 int curr_events;
1da177e4
LT		 7590
7591 idle = 1;
4b80991c
N		 7592 rcu_read_lock();
7593 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7594 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 7595 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7596 (int)part_stat_read(&disk->part0, sectors[1]) -
7597 atomic_read(&disk->sync_io);
713f6ab1
N		 7598 /* sync IO will cause sync_io to increase before the disk_stats
7599 * as sync_io is counted when a request starts, and
7600 * disk_stats is counted when it completes.
7601 * So resync activity will cause curr_events to be smaller than
7602 * when there was no such activity.
7603 * non-sync IO will cause disk_stat to increase without
7604 * increasing sync_io so curr_events will (eventually)
7605 * be larger than it was before. Once it becomes
7606 * substantially larger, the test below will cause
7607 * the array to appear non-idle, and resync will slow
7608 * down.
7609 * If there is a lot of outstanding resync activity when
7610 * we set last_event to curr_events, then all that activity
7611 * completing might cause the array to appear non-idle
7612 * and resync will be slowed down even though there might
7613 * not have been non-resync activity. This will only
7614 * happen once though. 'last_events' will soon reflect
7615 * the state where there is little or no outstanding
7616 * resync requests, and further resync activity will
7617 * always make curr_events less than last_events.
c0e48521 7618 *
1da177e4 7619 */
eea1bf38 7620 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 7621 rdev->last_events = curr_events;
7622 idle = 0;
7623 }
7624 }
4b80991c 7625 rcu_read_unlock();
1da177e4
LT		 7626 return idle;
7627}
7628
fd01b88c 7629void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 7630{
7631 /* another "blocks" (512byte) blocks have been synced */
7632 atomic_sub(blocks, &mddev->recovery_active);
7633 wake_up(&mddev->recovery_wait);
7634 if (!ok) {
dfc70645 7635 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 7636 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT		 7637 md_wakeup_thread(mddev->thread);
7638 // stop recovery, signal do_sync ....
7639 }
7640}
6c144d31 7641EXPORT_SYMBOL(md_done_sync);
1da177e4 7642
06d91a5f
N		 7643/* md_write_start(mddev, bi)
7644 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 7645 * in superblock) before writing, schedule a superblock update
7646 * and wait for it to complete.
06d91a5f 7647 */
fd01b88c 7648void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7649{
0fd62b86 7650 int did_change = 0;
06d91a5f 7651 if (bio_data_dir(bi) != WRITE)
3d310eb7 7652 return;
06d91a5f 7653
f91de92e
N		 7654 BUG_ON(mddev->ro == 1);
7655 if (mddev->ro == 2) {
7656 /* need to switch to read/write */
7657 mddev->ro = 0;
7658 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7659 md_wakeup_thread(mddev->thread);
25156198 7660 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7661 did_change = 1;
f91de92e 7662 }
06d91a5f 7663 atomic_inc(&mddev->writes_pending);
31a59e34
N		 7664 if (mddev->safemode == 1)
7665 mddev->safemode = 0;
06d91a5f 7666 if (mddev->in_sync) {
85572d7c 7667 spin_lock(&mddev->lock);
3d310eb7
N		 7668 if (mddev->in_sync) {
7669 mddev->in_sync = 0;
850b2b42 7670 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7671 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7672 md_wakeup_thread(mddev->thread);
0fd62b86 7673 did_change = 1;
3d310eb7 7674 }
85572d7c 7675 spin_unlock(&mddev->lock);
06d91a5f 7676 }
0fd62b86 7677 if (did_change)
00bcb4ac 7678 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7679 wait_event(mddev->sb_wait,
09a44cc1 7680 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4 7681}
6c144d31 7682EXPORT_SYMBOL(md_write_start);
1da177e4 7683
fd01b88c 7684void md_write_end(struct mddev *mddev)
1da177e4
LT		 7685{
7686 if (atomic_dec_and_test(&mddev->writes_pending)) {
7687 if (mddev->safemode == 2)
7688 md_wakeup_thread(mddev->thread);
16f17b39 7689 else if (mddev->safemode_delay)
1da177e4
LT		 7690 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7691 }
7692}
6c144d31 7693EXPORT_SYMBOL(md_write_end);
1da177e4 7694
2a2275d6
N		 7695/* md_allow_write(mddev)
7696 * Calling this ensures that the array is marked 'active' so that writes
7697 * may proceed without blocking. It is important to call this before
7698 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7699 * Must be called with mddev_lock held.
b5470dc5
DW		 7700 *
7701 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7702 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7703 */
fd01b88c 7704int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7705{
7706 if (!mddev->pers)
b5470dc5 7707 return 0;
2a2275d6 7708 if (mddev->ro)
b5470dc5 7709 return 0;
1a0fd497 7710 if (!mddev->pers->sync_request)
b5470dc5 7711 return 0;
2a2275d6 7712
85572d7c 7713 spin_lock(&mddev->lock);
2a2275d6
N		 7714 if (mddev->in_sync) {
7715 mddev->in_sync = 0;
7716 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7717 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7718 if (mddev->safemode_delay &&
7719 mddev->safemode == 0)
7720 mddev->safemode = 1;
85572d7c 7721 spin_unlock(&mddev->lock);
2a2275d6 7722 md_update_sb(mddev, 0);
00bcb4ac 7723 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6 7724 } else
85572d7c 7725 spin_unlock(&mddev->lock);
b5470dc5 7726
070dc6dd 7727 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7728 return -EAGAIN;
7729 else
7730 return 0;
2a2275d6
N		 7731}
7732EXPORT_SYMBOL_GPL(md_allow_write);
7733
1da177e4
LT		 7734#define SYNC_MARKS 10
7735#define SYNC_MARK_STEP (3*HZ)
54f89341 7736#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 7737void md_do_sync(struct md_thread *thread)
1da177e4 7738{
4ed8731d 7739 struct mddev *mddev = thread->mddev;
fd01b88c 7740 struct mddev *mddev2;
1da177e4
LT		 7741 unsigned int currspeed = 0,
7742 window;
ac7e50a3 7743 sector_t max_sectors,j, io_sectors, recovery_done;
1da177e4 7744 unsigned long mark[SYNC_MARKS];
54f89341 7745 unsigned long update_time;
1da177e4
LT		 7746 sector_t mark_cnt[SYNC_MARKS];
7747 int last_mark,m;
7748 struct list_head *tmp;
7749 sector_t last_check;
57afd89f 7750 int skipped = 0;
3cb03002 7751 struct md_rdev *rdev;
c4a39551 7752 char *desc, *action = NULL;
7c2c57c9 7753 struct blk_plug plug;
c186b128 7754 bool cluster_resync_finished = false;
1da177e4
LT		 7755
7756 /* just incase thread restarts... */
7757 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7758 return;
3991b31e
N		 7759 if (mddev->ro) {/* never try to sync a read-only array */
7760 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5fd6c1dc 7761 return;
3991b31e 7762 }
1da177e4 7763
61df9d91 7764 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
c4a39551 7765 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
61df9d91 7766 desc = "data-check";
c4a39551
JB		 7767 action = "check";
7768 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
61df9d91 7769 desc = "requested-resync";
c4a39551
JB		 7770 action = "repair";
7771 } else
61df9d91
N		 7772 desc = "resync";
7773 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7774 desc = "reshape";
7775 else
7776 desc = "recovery";
7777
c4a39551
JB		 7778 mddev->last_sync_action = action ?: desc;
7779
1da177e4
LT		 7780 /* we overload curr_resync somewhat here.
7781 * 0 == not engaged in resync at all
7782 * 2 == checking that there is no conflict with another sync
7783 * 1 == like 2, but have yielded to allow conflicting resync to
7784 * commense
7785 * other == active in resync - this many blocks
7786 *
7787 * Before starting a resync we must have set curr_resync to
7788 * 2, and then checked that every "conflicting" array has curr_resync
7789 * less than ours. When we find one that is the same or higher
7790 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7791 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7792 * This will mean we have to start checking from the beginning again.
7793 *
7794 */
7795
7796 do {
7797 mddev->curr_resync = 2;
7798
7799 try_again:
404e4b43 7800 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7801 goto skip;
29ac4aa3 7802 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7803 if (mddev2 == mddev)
7804 continue;
90b08710
BS		 7805 if (!mddev->parallel_resync
7806 && mddev2->curr_resync
7807 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7808 DEFINE_WAIT(wq);
7809 if (mddev < mddev2 && mddev->curr_resync == 2) {
7810 /* arbitrarily yield */
7811 mddev->curr_resync = 1;
7812 wake_up(&resync_wait);
7813 }
7814 if (mddev > mddev2 && mddev->curr_resync == 1)
7815 /* no need to wait here, we can wait the next
7816 * time 'round when curr_resync == 2
7817 */
7818 continue;
9744197c
N		 7819 /* We need to wait 'interruptible' so as not to
7820 * contribute to the load average, and not to
7821 * be caught by 'softlockup'
7822 */
7823 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
c91abf5a 7824 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8712e553 7825 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7826 printk(KERN_INFO "md: delaying %s of %s"
7827 " until %s has finished (they"
1da177e4 7828 " share one or more physical units)\n",
61df9d91 7829 desc, mdname(mddev), mdname(mddev2));
1da177e4 7830 mddev_put(mddev2);
9744197c
N		 7831 if (signal_pending(current))
7832 flush_signals(current);
1da177e4
LT		 7833 schedule();
7834 finish_wait(&resync_wait, &wq);
7835 goto try_again;
7836 }
7837 finish_wait(&resync_wait, &wq);
7838 }
7839 }
7840 } while (mddev->curr_resync < 2);
7841
5fd6c1dc 7842 j = 0;
9d88883e 7843 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7844 /* resync follows the size requested by the personality,
57afd89f 7845 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7846 */
7847 max_sectors = mddev->resync_max_sectors;
7f7583d4 7848 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 7849 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7850 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7851 j = mddev->resync_min;
7852 else if (!mddev->bitmap)
5fd6c1dc 7853 j = mddev->recovery_cp;
5e96ee65 7854
ccfcc3c1 7855 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7856 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7857 else {
1da177e4 7858 /* recovery follows the physical size of devices */
58c0fed4 7859 max_sectors = mddev->dev_sectors;
5fd6c1dc 7860 j = MaxSector;
4e59ca7d 7861 rcu_read_lock();
dafb20fa 7862 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7863 if (rdev->raid_disk >= 0 &&
f2076e7d 7864 !test_bit(Journal, &rdev->flags) &&
5fd6c1dc
N		 7865 !test_bit(Faulty, &rdev->flags) &&
7866 !test_bit(In_sync, &rdev->flags) &&
7867 rdev->recovery_offset < j)
7868 j = rdev->recovery_offset;
4e59ca7d 7869 rcu_read_unlock();
133d4527
N		 7870
7871 /* If there is a bitmap, we need to make sure all
7872 * writes that started before we added a spare
7873 * complete before we start doing a recovery.
7874 * Otherwise the write might complete and (via
7875 * bitmap_endwrite) set a bit in the bitmap after the
7876 * recovery has checked that bit and skipped that
7877 * region.
7878 */
7879 if (mddev->bitmap) {
7880 mddev->pers->quiesce(mddev, 1);
7881 mddev->pers->quiesce(mddev, 0);
7882 }
5fd6c1dc 7883 }
1da177e4 7884
61df9d91
N		 7885 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7886 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7887 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7888 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7889 "(but not more than %d KB/sec) for %s.\n",
7890 speed_max(mddev), desc);
1da177e4 7891
eea1bf38 7892 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7893
57afd89f 7894 io_sectors = 0;
1da177e4
LT		 7895 for (m = 0; m < SYNC_MARKS; m++) {
7896 mark[m] = jiffies;
57afd89f 7897 mark_cnt[m] = io_sectors;
1da177e4
LT		 7898 }
7899 last_mark = 0;
7900 mddev->resync_mark = mark[last_mark];
7901 mddev->resync_mark_cnt = mark_cnt[last_mark];
7902
7903 /*
7904 * Tune reconstruction:
7905 */
7906 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7907 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7908 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7909
7910 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7911 last_check = 0;
7912
7913 if (j>2) {
c91abf5a 7914 printk(KERN_INFO
61df9d91
N		 7915 "md: resuming %s of %s from checkpoint.\n",
7916 desc, mdname(mddev));
1da177e4 7917 mddev->curr_resync = j;
72f36d59
N		 7918 } else
7919 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 7920 mddev->curr_resync_completed = j;
72f36d59
N		 7921 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7922 md_new_event(mddev);
54f89341 7923 update_time = jiffies;
1da177e4 7924
7c2c57c9 7925 blk_start_plug(&plug);
1da177e4 7926 while (j < max_sectors) {
57afd89f 7927 sector_t sectors;
1da177e4 7928
57afd89f 7929 skipped = 0;
97e4f42d 7930
7a91ee1f
N		 7931 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7932 ((mddev->curr_resync > mddev->curr_resync_completed &&
7933 (mddev->curr_resync - mddev->curr_resync_completed)
7934 > (max_sectors >> 4)) ||
54f89341 7935 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f 7936 (j - mddev->curr_resync_completed)*2
c5e19d90
N		 7937 >= mddev->resync_max - mddev->curr_resync_completed ||
7938 mddev->curr_resync_completed > mddev->resync_max
7a91ee1f 7939 )) {
97e4f42d 7940 /* time to update curr_resync_completed */
97e4f42d
N		 7941 wait_event(mddev->recovery_wait,
7942 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7943 mddev->curr_resync_completed = j;
35d78c66 7944 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7945 j > mddev->recovery_cp)
7946 mddev->recovery_cp = j;
54f89341 7947 update_time = jiffies;
070dc6dd 7948 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7949 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7950 }
acb180b0 7951
c91abf5a
N		 7952 while (j >= mddev->resync_max &&
7953 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
e62e58a5
N		 7954 /* As this condition is controlled by user-space,
7955 * we can block indefinitely, so use '_interruptible'
7956 * to avoid triggering warnings.
7957 */
7958 flush_signals(current); /* just in case */
7959 wait_event_interruptible(mddev->recovery_wait,
7960 mddev->resync_max > j
c91abf5a
N		 7961 || test_bit(MD_RECOVERY_INTR,
7962 &mddev->recovery));
e62e58a5 7963 }
acb180b0 7964
c91abf5a
N		 7965 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7966 break;
acb180b0 7967
09314799 7968 sectors = mddev->pers->sync_request(mddev, j, &skipped);
57afd89f 7969 if (sectors == 0) {
dfc70645 7970 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
c91abf5a 7971 break;
1da177e4 7972 }
57afd89f
N		 7973
7974 if (!skipped) { /* actual IO requested */
7975 io_sectors += sectors;
7976 atomic_add(sectors, &mddev->recovery_active);
7977 }
7978
e875ecea
N		 7979 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7980 break;
7981
1da177e4 7982 j += sectors;
5ed1df2e
N		 7983 if (j > max_sectors)
7984 /* when skipping, extra large numbers can be returned. */
7985 j = max_sectors;
72f36d59
N		 7986 if (j > 2)
7987 mddev->curr_resync = j;
ff4e8d9a 7988 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7989 if (last_check == 0)
e875ecea 7990 /* this is the earliest that rebuild will be
d7603b7e
N		 7991 * visible in /proc/mdstat
7992 */
7993 md_new_event(mddev);
57afd89f
N		 7994
7995 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7996 continue;
7997
57afd89f 7998 last_check = io_sectors;
1da177e4
LT		 7999 repeat:
8000 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8001 /* step marks */
8002 int next = (last_mark+1) % SYNC_MARKS;
8003
8004 mddev->resync_mark = mark[next];
8005 mddev->resync_mark_cnt = mark_cnt[next];
8006 mark[next] = jiffies;
57afd89f 8007 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 8008 last_mark = next;
8009 }
8010
c91abf5a
N		 8011 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8012 break;
1da177e4
LT		 8013
8014 /*
8015 * this loop exits only if either when we are slower than
8016 * the 'hard' speed limit, or the system was IO-idle for
8017 * a jiffy.
8018 * the system might be non-idle CPU-wise, but we only care
8019 * about not overloading the IO subsystem. (things like an
8020 * e2fsck being done on the RAID array should execute fast)
8021 */
1da177e4
LT		 8022 cond_resched();
8023
ac7e50a3
XN		 8024 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8025 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
57afd89f 8026 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 8027
88202a0c 8028 if (currspeed > speed_min(mddev)) {
ac8fa419 8029 if (currspeed > speed_max(mddev)) {
c0e48521 8030 msleep(500);
1da177e4
LT		 8031 goto repeat;
8032 }
ac8fa419
N		 8033 if (!is_mddev_idle(mddev, 0)) {
8034 /*
8035 * Give other IO more of a chance.
8036 * The faster the devices, the less we wait.
8037 */
8038 wait_event(mddev->recovery_wait,
8039 !atomic_read(&mddev->recovery_active));
8040 }
1da177e4
LT		 8041 }
8042 }
c91abf5a
N		 8043 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8044 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8045 ? "interrupted" : "done");
1da177e4
LT		 8046 /*
8047 * this also signals 'finished resyncing' to md_stop
8048 */
7c2c57c9 8049 blk_finish_plug(&plug);
1da177e4
LT		 8050 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8051
5ed1df2e
N		 8052 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8053 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8054 mddev->curr_resync > 2) {
8055 mddev->curr_resync_completed = mddev->curr_resync;
8056 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8057 }
c186b128
GR		 8058 /* tell personality and other nodes that we are finished */
8059 if (mddev_is_clustered(mddev)) {
8060 md_cluster_ops->resync_finish(mddev);
8061 cluster_resync_finished = true;
8062 }
09314799 8063 mddev->pers->sync_request(mddev, max_sectors, &skipped);
1da177e4 8064
dfc70645 8065 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 8066 mddev->curr_resync > 2) {
8067 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8068 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8069 if (mddev->curr_resync >= mddev->recovery_cp) {
8070 printk(KERN_INFO
61df9d91
N		 8071 "md: checkpointing %s of %s.\n",
8072 desc, mdname(mddev));
0a19caab 8073 if (test_bit(MD_RECOVERY_ERROR,
8074 &mddev->recovery))
8075 mddev->recovery_cp =
8076 mddev->curr_resync_completed;
8077 else
8078 mddev->recovery_cp =
8079 mddev->curr_resync;
5fd6c1dc
N		 8080 }
8081 } else
8082 mddev->recovery_cp = MaxSector;
8083 } else {
8084 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8085 mddev->curr_resync = MaxSector;
4e59ca7d 8086 rcu_read_lock();
dafb20fa 8087 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 8088 if (rdev->raid_disk >= 0 &&
70fffd0b 8089 mddev->delta_disks >= 0 &&
f2076e7d 8090 !test_bit(Journal, &rdev->flags) &&
5fd6c1dc
N		 8091 !test_bit(Faulty, &rdev->flags) &&
8092 !test_bit(In_sync, &rdev->flags) &&
8093 rdev->recovery_offset < mddev->curr_resync)
8094 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 8095 rcu_read_unlock();
5fd6c1dc 8096 }
1da177e4 8097 }
db91ff55 8098 skip:
17571284 8099 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 8100
c186b128
GR		 8101 if (mddev_is_clustered(mddev) &&
8102 test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8103 !cluster_resync_finished)
8104 md_cluster_ops->resync_finish(mddev);
8105
23da422b 8106 spin_lock(&mddev->lock);
c07b70ad
N		 8107 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8108 /* We completed so min/max setting can be forgotten if used. */
8109 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8110 mddev->resync_min = 0;
8111 mddev->resync_max = MaxSector;
8112 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8113 mddev->resync_min = mddev->curr_resync_completed;
f7851be7 8114 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 8115 mddev->curr_resync = 0;
23da422b
N		 8116 spin_unlock(&mddev->lock);
8117
1da177e4 8118 wake_up(&resync_wait);
1da177e4 8119 md_wakeup_thread(mddev->thread);
c6207277 8120 return;
1da177e4 8121}
29269553 8122EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 8123
746d3207
N		 8124static int remove_and_add_spares(struct mddev *mddev,
8125 struct md_rdev *this)
b4c4c7b8 8126{
3cb03002 8127 struct md_rdev *rdev;
b4c4c7b8 8128 int spares = 0;
f2a371c5 8129 int removed = 0;
b4c4c7b8 8130
dafb20fa 8131 rdev_for_each(rdev, mddev)
746d3207
N		 8132 if ((this == NULL || rdev == this) &&
8133 rdev->raid_disk >= 0 &&
6bfe0b49 8134 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8 8135 (test_bit(Faulty, &rdev->flags) ||
f2076e7d
SL		 8136 (!test_bit(In_sync, &rdev->flags) &&
8137 !test_bit(Journal, &rdev->flags))) &&
b4c4c7b8
N		 8138 atomic_read(&rdev->nr_pending)==0) {
8139 if (mddev->pers->hot_remove_disk(
b8321b68 8140 mddev, rdev) == 0) {
36fad858 8141 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 8142 rdev->raid_disk = -1;
f2a371c5 8143 removed++;
b4c4c7b8
N		 8144 }
8145 }
90584fc9
JB		 8146 if (removed && mddev->kobj.sd)
8147 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 8148
2910ff17 8149 if (this && removed)
746d3207
N		 8150 goto no_add;
8151
dafb20fa 8152 rdev_for_each(rdev, mddev) {
2910ff17
GR		 8153 if (this && this != rdev)
8154 continue;
dbb64f86
GR		 8155 if (test_bit(Candidate, &rdev->flags))
8156 continue;
7bfec5f3
N		 8157 if (rdev->raid_disk >= 0 &&
8158 !test_bit(In_sync, &rdev->flags) &&
f2076e7d 8159 !test_bit(Journal, &rdev->flags) &&
7bfec5f3
N		 8160 !test_bit(Faulty, &rdev->flags))
8161 spares++;
7ceb17e8
N		 8162 if (rdev->raid_disk >= 0)
8163 continue;
8164 if (test_bit(Faulty, &rdev->flags))
8165 continue;
3069aa8d
SL		 8166 if (test_bit(Journal, &rdev->flags))
8167 continue;
7ceb17e8 8168 if (mddev->ro &&
8313b8e5
N		 8169 ! (rdev->saved_raid_disk >= 0 &&
8170 !test_bit(Bitmap_sync, &rdev->flags)))
7ceb17e8
N		 8171 continue;
8172
d01552a7 8173 rdev->recovery_offset = 0;
7ceb17e8
N		 8174 if (mddev->pers->
8175 hot_add_disk(mddev, rdev) == 0) {
8176 if (sysfs_link_rdev(mddev, rdev))
8177 /* failure here is OK */;
8178 spares++;
8179 md_new_event(mddev);
8180 set_bit(MD_CHANGE_DEVS, &mddev->flags);
dfc70645 8181 }
b4c4c7b8 8182 }
746d3207 8183no_add:
6dafab6b
N		 8184 if (removed)
8185 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 8186 return spares;
8187}
7ebc0be7 8188
ac05f256
N		 8189static void md_start_sync(struct work_struct *ws)
8190{
8191 struct mddev *mddev = container_of(ws, struct mddev, del_work);
c186b128
GR		 8192 int ret = 0;
8193
8194 if (mddev_is_clustered(mddev)) {
8195 ret = md_cluster_ops->resync_start(mddev);
8196 if (ret) {
8197 mddev->sync_thread = NULL;
8198 goto out;
8199 }
8200 }
ac05f256
N		 8201
8202 mddev->sync_thread = md_register_thread(md_do_sync,
8203 mddev,
8204 "resync");
c186b128 8205out:
ac05f256 8206 if (!mddev->sync_thread) {
c186b128
GR		 8207 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8208 printk(KERN_ERR "%s: could not start resync"
8209 " thread...\n",
8210 mdname(mddev));
ac05f256
N		 8211 /* leave the spares where they are, it shouldn't hurt */
8212 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8213 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8214 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8215 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8216 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8217 wake_up(&resync_wait);
ac05f256
N		 8218 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8219 &mddev->recovery))
8220 if (mddev->sysfs_action)
8221 sysfs_notify_dirent_safe(mddev->sysfs_action);
8222 } else
8223 md_wakeup_thread(mddev->sync_thread);
8224 sysfs_notify_dirent_safe(mddev->sysfs_action);
8225 md_new_event(mddev);
8226}
8227
1da177e4
LT		 8228/*
8229 * This routine is regularly called by all per-raid-array threads to
8230 * deal with generic issues like resync and super-block update.
8231 * Raid personalities that don't have a thread (linear/raid0) do not
8232 * need this as they never do any recovery or update the superblock.
8233 *
8234 * It does not do any resync itself, but rather "forks" off other threads
8235 * to do that as needed.
8236 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8237 * "->recovery" and create a thread at ->sync_thread.
dfc70645 8238 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 8239 * and wakeups up this thread which will reap the thread and finish up.
8240 * This thread also removes any faulty devices (with nr_pending == 0).
8241 *
8242 * The overall approach is:
8243 * 1/ if the superblock needs updating, update it.
8244 * 2/ If a recovery thread is running, don't do anything else.
8245 * 3/ If recovery has finished, clean up, possibly marking spares active.
8246 * 4/ If there are any faulty devices, remove them.
8247 * 5/ If array is degraded, try to add spares devices
8248 * 6/ If array has spares or is not in-sync, start a resync thread.
8249 */
fd01b88c 8250void md_check_recovery(struct mddev *mddev)
1da177e4 8251{
68866e42
JB		 8252 if (mddev->suspended)
8253 return;
8254
5f40402d 8255 if (mddev->bitmap)
aa5cbd10 8256 bitmap_daemon_work(mddev);
1da177e4 8257
fca4d848 8258 if (signal_pending(current)) {
31a59e34 8259 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 8260 printk(KERN_INFO "md: %s in immediate safe mode\n",
8261 mdname(mddev));
8262 mddev->safemode = 2;
8263 }
8264 flush_signals(current);
8265 }
8266
c89a8eee
N		 8267 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8268 return;
1da177e4 8269 if ( ! (
142d44c3 8270 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 8271 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 8272 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 8273 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 8274 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8275 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 8276 ))
8277 return;
fca4d848 8278
df5b89b3 8279 if (mddev_trylock(mddev)) {
b4c4c7b8 8280 int spares = 0;
fca4d848 8281
c89a8eee 8282 if (mddev->ro) {
ab16bfc7
NB		 8283 struct md_rdev *rdev;
8284 if (!mddev->external && mddev->in_sync)
8285 /* 'Blocked' flag not needed as failed devices
8286 * will be recorded if array switched to read/write.
8287 * Leaving it set will prevent the device
8288 * from being removed.
8289 */
8290 rdev_for_each(rdev, mddev)
8291 clear_bit(Blocked, &rdev->flags);
7ceb17e8
N		 8292 /* On a read-only array we can:
8293 * - remove failed devices
8294 * - add already-in_sync devices if the array itself
8295 * is in-sync.
8296 * As we only add devices that are already in-sync,
8297 * we can activate the spares immediately.
c89a8eee 8298 */
7ceb17e8 8299 remove_and_add_spares(mddev, NULL);
8313b8e5
N		 8300 /* There is no thread, but we need to call
8301 * ->spare_active and clear saved_raid_disk
8302 */
2ac295a5 8303 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8313b8e5 8304 md_reap_sync_thread(mddev);
a4a3d26d 8305 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8313b8e5 8306 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d4929add 8307 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
c89a8eee
N		 8308 goto unlock;
8309 }
8310
31a59e34 8311 if (!mddev->external) {
0fd62b86 8312 int did_change = 0;
85572d7c 8313 spin_lock(&mddev->lock);
31a59e34
N		 8314 if (mddev->safemode &&
8315 !atomic_read(&mddev->writes_pending) &&
8316 !mddev->in_sync &&
8317 mddev->recovery_cp == MaxSector) {
8318 mddev->in_sync = 1;
0fd62b86 8319 did_change = 1;
070dc6dd 8320 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 8321 }
8322 if (mddev->safemode == 1)
8323 mddev->safemode = 0;
85572d7c 8324 spin_unlock(&mddev->lock);
0fd62b86 8325 if (did_change)
00bcb4ac 8326 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 8327 }
fca4d848 8328
2aa82191 8329 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 8330 md_update_sb(mddev, 0);
06d91a5f 8331
1da177e4
LT		 8332 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8333 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8334 /* resync/recovery still happening */
8335 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8336 goto unlock;
8337 }
8338 if (mddev->sync_thread) {
a91d5ac0 8339 md_reap_sync_thread(mddev);
1da177e4
LT		 8340 goto unlock;
8341 }
72a23c21
NB		 8342 /* Set RUNNING before clearing NEEDED to avoid
8343 * any transients in the value of "sync_action".
8344 */
72f36d59 8345 mddev->curr_resync_completed = 0;
23da422b 8346 spin_lock(&mddev->lock);
72a23c21 8347 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
23da422b 8348 spin_unlock(&mddev->lock);
24dd469d
N		 8349 /* Clear some bits that don't mean anything, but
8350 * might be left set
8351 */
24dd469d
N		 8352 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8353 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 8354
ed209584
N		 8355 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8356 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
ac05f256 8357 goto not_running;
1da177e4
LT		 8358 /* no recovery is running.
8359 * remove any failed drives, then
8360 * add spares if possible.
72f36d59 8361 * Spares are also removed and re-added, to allow
1da177e4
LT		 8362 * the personality to fail the re-add.
8363 */
1da177e4 8364
b4c4c7b8 8365 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 8366 if (mddev->pers->check_reshape == NULL ||
8367 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8 8368 /* Cannot proceed */
ac05f256 8369 goto not_running;
b4c4c7b8 8370 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 8371 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 8372 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N		 8373 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8374 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 8375 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 8376 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8377 } else if (mddev->recovery_cp < MaxSector) {
8378 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 8379 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8380 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8381 /* nothing to be done ... */
ac05f256 8382 goto not_running;
24dd469d 8383
1da177e4 8384 if (mddev->pers->sync_request) {
ef99bf48 8385 if (spares) {
a654b9d8
N		 8386 /* We are adding a device or devices to an array
8387 * which has the bitmap stored on all devices.
8388 * So make sure all bitmap pages get written
8389 */
8390 bitmap_write_all(mddev->bitmap);
8391 }
ac05f256
N		 8392 INIT_WORK(&mddev->del_work, md_start_sync);
8393 queue_work(md_misc_wq, &mddev->del_work);
8394 goto unlock;
1da177e4 8395 }
ac05f256 8396 not_running:
72a23c21
NB		 8397 if (!mddev->sync_thread) {
8398 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8399 wake_up(&resync_wait);
72a23c21
NB		 8400 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8401 &mddev->recovery))
0c3573f1 8402 if (mddev->sysfs_action)
00bcb4ac 8403 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 8404 }
ac05f256
N		 8405 unlock:
8406 wake_up(&mddev->sb_wait);
1da177e4
LT		 8407 mddev_unlock(mddev);
8408 }
8409}
6c144d31 8410EXPORT_SYMBOL(md_check_recovery);
1da177e4 8411
a91d5ac0
JB		 8412void md_reap_sync_thread(struct mddev *mddev)
8413{
8414 struct md_rdev *rdev;
8415
8416 /* resync has finished, collect result */
8417 md_unregister_thread(&mddev->sync_thread);
8418 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8419 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8420 /* success...*/
8421 /* activate any spares */
8422 if (mddev->pers->spare_active(mddev)) {
8423 sysfs_notify(&mddev->kobj, NULL,
8424 "degraded");
8425 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8426 }
8427 }
8428 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8429 mddev->pers->finish_reshape)
8430 mddev->pers->finish_reshape(mddev);
8431
8432 /* If array is no-longer degraded, then any saved_raid_disk
f466722c 8433 * information must be scrapped.
a91d5ac0 8434 */
f466722c
N		 8435 if (!mddev->degraded)
8436 rdev_for_each(rdev, mddev)
a91d5ac0
JB		 8437 rdev->saved_raid_disk = -1;
8438
8439 md_update_sb(mddev, 1);
8440 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
ea358cd0 8441 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
a91d5ac0
JB		 8442 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8443 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8444 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8445 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
f851b60d 8446 wake_up(&resync_wait);
a91d5ac0
JB		 8447 /* flag recovery needed just to double check */
8448 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8449 sysfs_notify_dirent_safe(mddev->sysfs_action);
8450 md_new_event(mddev);
8451 if (mddev->event_work.func)
8452 queue_work(md_misc_wq, &mddev->event_work);
8453}
6c144d31 8454EXPORT_SYMBOL(md_reap_sync_thread);
a91d5ac0 8455
fd01b88c 8456void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 8457{
00bcb4ac 8458 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 8459 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 8460 !test_bit(Blocked, &rdev->flags) &&
8461 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 8462 msecs_to_jiffies(5000));
8463 rdev_dec_pending(rdev, mddev);
8464}
8465EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8466
c6563a8c
N		 8467void md_finish_reshape(struct mddev *mddev)
8468{
8469 /* called be personality module when reshape completes. */
8470 struct md_rdev *rdev;
8471
8472 rdev_for_each(rdev, mddev) {
8473 if (rdev->data_offset > rdev->new_data_offset)
8474 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8475 else
8476 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8477 rdev->data_offset = rdev->new_data_offset;
8478 }
8479}
8480EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N		 8481
8482/* Bad block management.
8483 * We can record which blocks on each device are 'bad' and so just
8484 * fail those blocks, or that stripe, rather than the whole device.
8485 * Entries in the bad-block table are 64bits wide. This comprises:
8486 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8487 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8488 * A 'shift' can be set so that larger blocks are tracked and
8489 * consequently larger devices can be covered.
8490 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8491 *
8492 * Locking of the bad-block table uses a seqlock so md_is_badblock
8493 * might need to retry if it is very unlucky.
8494 * We will sometimes want to check for bad blocks in a bi_end_io function,
8495 * so we use the write_seqlock_irq variant.
8496 *
8497 * When looking for a bad block we specify a range and want to
8498 * know if any block in the range is bad. So we binary-search
8499 * to the last range that starts at-or-before the given endpoint,
8500 * (or "before the sector after the target range")
8501 * then see if it ends after the given start.
8502 * We return
8503 * 0 if there are no known bad blocks in the range
8504 * 1 if there are known bad block which are all acknowledged
8505 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8506 * plus the start/length of the first bad section we overlap.
8507 */
8508int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8509 sector_t *first_bad, int *bad_sectors)
8510{
8511 int hi;
ab05613a 8512 int lo;
2230dfe4 8513 u64 *p = bb->page;
ab05613a 8514 int rv;
2230dfe4
N		 8515 sector_t target = s + sectors;
8516 unsigned seq;
8517
8518 if (bb->shift > 0) {
8519 /* round the start down, and the end up */
8520 s >>= bb->shift;
8521 target += (1<<bb->shift) - 1;
8522 target >>= bb->shift;
8523 sectors = target - s;
8524 }
8525 /* 'target' is now the first block after the bad range */
8526
8527retry:
8528 seq = read_seqbegin(&bb->lock);
ab05613a 8529 lo = 0;
8530 rv = 0;
2230dfe4
N		 8531 hi = bb->count;
8532
8533 /* Binary search between lo and hi for 'target'
8534 * i.e. for the last range that starts before 'target'
8535 */
8536 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8537 * are known not to be the last range before target.
8538 * VARIANT: hi-lo is the number of possible
8539 * ranges, and decreases until it reaches 1
8540 */
8541 while (hi - lo > 1) {
8542 int mid = (lo + hi) / 2;
8543 sector_t a = BB_OFFSET(p[mid]);
8544 if (a < target)
8545 /* This could still be the one, earlier ranges
8546 * could not. */
8547 lo = mid;
8548 else
8549 /* This and later ranges are definitely out. */
8550 hi = mid;
8551 }
8552 /* 'lo' might be the last that started before target, but 'hi' isn't */
8553 if (hi > lo) {
8554 /* need to check all range that end after 's' to see if
8555 * any are unacknowledged.
8556 */
8557 while (lo >= 0 &&
8558 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8559 if (BB_OFFSET(p[lo]) < target) {
8560 /* starts before the end, and finishes after
8561 * the start, so they must overlap
8562 */
8563 if (rv != -1 && BB_ACK(p[lo]))
8564 rv = 1;
8565 else
8566 rv = -1;
8567 *first_bad = BB_OFFSET(p[lo]);
8568 *bad_sectors = BB_LEN(p[lo]);
8569 }
8570 lo--;
8571 }
8572 }
8573
8574 if (read_seqretry(&bb->lock, seq))
8575 goto retry;
8576
8577 return rv;
8578}
8579EXPORT_SYMBOL_GPL(md_is_badblock);
8580
8581/*
8582 * Add a range of bad blocks to the table.
8583 * This might extend the table, or might contract it
8584 * if two adjacent ranges can be merged.
8585 * We binary-search to find the 'insertion' point, then
8586 * decide how best to handle it.
8587 */
8588static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8589 int acknowledged)
8590{
8591 u64 *p;
8592 int lo, hi;
8593 int rv = 1;
905b0297 8594 unsigned long flags;
2230dfe4
N		 8595
8596 if (bb->shift < 0)
8597 /* badblocks are disabled */
8598 return 0;
8599
8600 if (bb->shift) {
8601 /* round the start down, and the end up */
8602 sector_t next = s + sectors;
8603 s >>= bb->shift;
8604 next += (1<<bb->shift) - 1;
8605 next >>= bb->shift;
8606 sectors = next - s;
8607 }
8608
905b0297 8609 write_seqlock_irqsave(&bb->lock, flags);
2230dfe4
N		 8610
8611 p = bb->page;
8612 lo = 0;
8613 hi = bb->count;
8614 /* Find the last range that starts at-or-before 's' */
8615 while (hi - lo > 1) {
8616 int mid = (lo + hi) / 2;
8617 sector_t a = BB_OFFSET(p[mid]);
8618 if (a <= s)
8619 lo = mid;
8620 else
8621 hi = mid;
8622 }
8623 if (hi > lo && BB_OFFSET(p[lo]) > s)
8624 hi = lo;
8625
8626 if (hi > lo) {
8627 /* we found a range that might merge with the start
8628 * of our new range
8629 */
8630 sector_t a = BB_OFFSET(p[lo]);
8631 sector_t e = a + BB_LEN(p[lo]);
8632 int ack = BB_ACK(p[lo]);
8633 if (e >= s) {
8634 /* Yes, we can merge with a previous range */
8635 if (s == a && s + sectors >= e)
8636 /* new range covers old */
8637 ack = acknowledged;
8638 else
8639 ack = ack && acknowledged;
8640
8641 if (e < s + sectors)
8642 e = s + sectors;
8643 if (e - a <= BB_MAX_LEN) {
8644 p[lo] = BB_MAKE(a, e-a, ack);
8645 s = e;
8646 } else {
8647 /* does not all fit in one range,
8648 * make p[lo] maximal
8649 */
8650 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8651 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8652 s = a + BB_MAX_LEN;
8653 }
8654 sectors = e - s;
8655 }
8656 }
8657 if (sectors && hi < bb->count) {
8658 /* 'hi' points to the first range that starts after 's'.
8659 * Maybe we can merge with the start of that range */
8660 sector_t a = BB_OFFSET(p[hi]);
8661 sector_t e = a + BB_LEN(p[hi]);
8662 int ack = BB_ACK(p[hi]);
8663 if (a <= s + sectors) {
8664 /* merging is possible */
8665 if (e <= s + sectors) {
8666 /* full overlap */
8667 e = s + sectors;
8668 ack = acknowledged;
8669 } else
8670 ack = ack && acknowledged;
8671
8672 a = s;
8673 if (e - a <= BB_MAX_LEN) {
8674 p[hi] = BB_MAKE(a, e-a, ack);
8675 s = e;
8676 } else {
8677 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8678 s = a + BB_MAX_LEN;
8679 }
8680 sectors = e - s;
8681 lo = hi;
8682 hi++;
8683 }
8684 }
8685 if (sectors == 0 && hi < bb->count) {
8686 /* we might be able to combine lo and hi */
8687 /* Note: 's' is at the end of 'lo' */
8688 sector_t a = BB_OFFSET(p[hi]);
8689 int lolen = BB_LEN(p[lo]);
8690 int hilen = BB_LEN(p[hi]);
8691 int newlen = lolen + hilen - (s - a);
8692 if (s >= a && newlen < BB_MAX_LEN) {
8693 /* yes, we can combine them */
8694 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8695 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8696 memmove(p + hi, p + hi + 1,
8697 (bb->count - hi - 1) * 8);
8698 bb->count--;
8699 }
8700 }
8701 while (sectors) {
8702 /* didn't merge (it all).
8703 * Need to add a range just before 'hi' */
8704 if (bb->count >= MD_MAX_BADBLOCKS) {
8705 /* No room for more */
8706 rv = 0;
8707 break;
8708 } else {
8709 int this_sectors = sectors;
8710 memmove(p + hi + 1, p + hi,
8711 (bb->count - hi) * 8);
8712 bb->count++;
8713
8714 if (this_sectors > BB_MAX_LEN)
8715 this_sectors = BB_MAX_LEN;
8716 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8717 sectors -= this_sectors;
8718 s += this_sectors;
8719 }
8720 }
8721
8722 bb->changed = 1;
de393cde
N		 8723 if (!acknowledged)
8724 bb->unacked_exist = 1;
905b0297 8725 write_sequnlock_irqrestore(&bb->lock, flags);
2230dfe4
N		 8726
8727 return rv;
8728}
8729
3cb03002 8730int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8731 int is_new)
2230dfe4 8732{
c6563a8c
N		 8733 int rv;
8734 if (is_new)
8735 s += rdev->new_data_offset;
8736 else
8737 s += rdev->data_offset;
8738 rv = md_set_badblocks(&rdev->badblocks,
8739 s, sectors, 0);
2230dfe4
N		 8740 if (rv) {
8741 /* Make sure they get written out promptly */
8bd2f0a0 8742 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4 8743 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
55ce74d4 8744 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
2230dfe4
N		 8745 md_wakeup_thread(rdev->mddev->thread);
8746 }
8747 return rv;
8748}
8749EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8750
8751/*
8752 * Remove a range of bad blocks from the table.
8753 * This may involve extending the table if we spilt a region,
8754 * but it must not fail. So if the table becomes full, we just
8755 * drop the remove request.
8756 */
8757static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8758{
8759 u64 *p;
8760 int lo, hi;
8761 sector_t target = s + sectors;
8762 int rv = 0;
8763
8764 if (bb->shift > 0) {
8765 /* When clearing we round the start up and the end down.
8766 * This should not matter as the shift should align with
8767 * the block size and no rounding should ever be needed.
8768 * However it is better the think a block is bad when it
8769 * isn't than to think a block is not bad when it is.
8770 */
8771 s += (1<<bb->shift) - 1;
8772 s >>= bb->shift;
8773 target >>= bb->shift;
8774 sectors = target - s;
8775 }
8776
8777 write_seqlock_irq(&bb->lock);
8778
8779 p = bb->page;
8780 lo = 0;
8781 hi = bb->count;
8782 /* Find the last range that starts before 'target' */
8783 while (hi - lo > 1) {
8784 int mid = (lo + hi) / 2;
8785 sector_t a = BB_OFFSET(p[mid]);
8786 if (a < target)
8787 lo = mid;
8788 else
8789 hi = mid;
8790 }
8791 if (hi > lo) {
8792 /* p[lo] is the last range that could overlap the
8793 * current range. Earlier ranges could also overlap,
8794 * but only this one can overlap the end of the range.
8795 */
8796 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8797 /* Partial overlap, leave the tail of this range */
8798 int ack = BB_ACK(p[lo]);
8799 sector_t a = BB_OFFSET(p[lo]);
8800 sector_t end = a + BB_LEN(p[lo]);
8801
8802 if (a < s) {
8803 /* we need to split this range */
8804 if (bb->count >= MD_MAX_BADBLOCKS) {
8b32bf5e 8805 rv = -ENOSPC;
2230dfe4
N		 8806 goto out;
8807 }
8808 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8809 bb->count++;
8810 p[lo] = BB_MAKE(a, s-a, ack);
8811 lo++;
8812 }
8813 p[lo] = BB_MAKE(target, end - target, ack);
8814 /* there is no longer an overlap */
8815 hi = lo;
8816 lo--;
8817 }
8818 while (lo >= 0 &&
8819 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8820 /* This range does overlap */
8821 if (BB_OFFSET(p[lo]) < s) {
8822 /* Keep the early parts of this range. */
8823 int ack = BB_ACK(p[lo]);
8824 sector_t start = BB_OFFSET(p[lo]);
8825 p[lo] = BB_MAKE(start, s - start, ack);
8826 /* now low doesn't overlap, so.. */
8827 break;
8828 }
8829 lo--;
8830 }
8831 /* 'lo' is strictly before, 'hi' is strictly after,
8832 * anything between needs to be discarded
8833 */
8834 if (hi - lo > 1) {
8835 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8836 bb->count -= (hi - lo - 1);
8837 }
8838 }
8839
8840 bb->changed = 1;
8841out:
8842 write_sequnlock_irq(&bb->lock);
8843 return rv;
8844}
8845
c6563a8c
N		 8846int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8847 int is_new)
2230dfe4 8848{
c6563a8c
N		 8849 if (is_new)
8850 s += rdev->new_data_offset;
8851 else
8852 s += rdev->data_offset;
2230dfe4 8853 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8854 s, sectors);
2230dfe4
N		 8855}
8856EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8857
8858/*
8859 * Acknowledge all bad blocks in a list.
8860 * This only succeeds if ->changed is clear. It is used by
8861 * in-kernel metadata updates
8862 */
8863void md_ack_all_badblocks(struct badblocks *bb)
8864{
8865 if (bb->page == NULL || bb->changed)
8866 /* no point even trying */
8867 return;
8868 write_seqlock_irq(&bb->lock);
8869
ecb178bb 8870 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N		 8871 u64 *p = bb->page;
8872 int i;
8873 for (i = 0; i < bb->count ; i++) {
8874 if (!BB_ACK(p[i])) {
8875 sector_t start = BB_OFFSET(p[i]);
8876 int len = BB_LEN(p[i]);
8877 p[i] = BB_MAKE(start, len, 1);
8878 }
8879 }
de393cde 8880 bb->unacked_exist = 0;
2230dfe4
N		 8881 }
8882 write_sequnlock_irq(&bb->lock);
8883}
8884EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8885
16c791a5
N		 8886/* sysfs access to bad-blocks list.
8887 * We present two files.
8888 * 'bad-blocks' lists sector numbers and lengths of ranges that
8889 * are recorded as bad. The list is truncated to fit within
8890 * the one-page limit of sysfs.
8891 * Writing "sector length" to this file adds an acknowledged
8892 * bad block list.
8893 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8894 * been acknowledged. Writing to this file adds bad blocks
8895 * without acknowledging them. This is largely for testing.
8896 */
8897
8898static ssize_t
8899badblocks_show(struct badblocks *bb, char *page, int unack)
8900{
8901 size_t len;
8902 int i;
8903 u64 *p = bb->page;
8904 unsigned seq;
8905
8906 if (bb->shift < 0)
8907 return 0;
8908
8909retry:
8910 seq = read_seqbegin(&bb->lock);
8911
8912 len = 0;
8913 i = 0;
8914
8915 while (len < PAGE_SIZE && i < bb->count) {
8916 sector_t s = BB_OFFSET(p[i]);
8917 unsigned int length = BB_LEN(p[i]);
8918 int ack = BB_ACK(p[i]);
8919 i++;
8920
8921 if (unack && ack)
8922 continue;
8923
8924 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8925 (unsigned long long)s << bb->shift,
8926 length << bb->shift);
8927 }
de393cde
N		 8928 if (unack && len == 0)
8929 bb->unacked_exist = 0;
16c791a5
N		 8930
8931 if (read_seqretry(&bb->lock, seq))
8932 goto retry;
8933
8934 return len;
8935}
8936
8937#define DO_DEBUG 1
8938
8939static ssize_t
8940badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8941{
8942 unsigned long long sector;
8943 int length;
8944 char newline;
8945#ifdef DO_DEBUG
8946 /* Allow clearing via sysfs *only* for testing/debugging.
8947 * Normally only a successful write may clear a badblock
8948 */
8949 int clear = 0;
8950 if (page[0] == '-') {
8951 clear = 1;
8952 page++;
8953 }
8954#endif /* DO_DEBUG */
8955
8956 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8957 case 3:
8958 if (newline != '\n')
8959 return -EINVAL;
8960 case 2:
8961 if (length <= 0)
8962 return -EINVAL;
8963 break;
8964 default:
8965 return -EINVAL;
8966 }
8967
8968#ifdef DO_DEBUG
8969 if (clear) {
8970 md_clear_badblocks(bb, sector, length);
8971 return len;
8972 }
8973#endif /* DO_DEBUG */
8974 if (md_set_badblocks(bb, sector, length, !unack))
8975 return len;
8976 else
8977 return -ENOSPC;
8978}
8979
75c96f85
AB		 8980static int md_notify_reboot(struct notifier_block *this,
8981 unsigned long code, void *x)
1da177e4
LT		 8982{
8983 struct list_head *tmp;
fd01b88c 8984 struct mddev *mddev;
2dba6a91 8985 int need_delay = 0;
1da177e4 8986
c744a65c
N		 8987 for_each_mddev(mddev, tmp) {
8988 if (mddev_trylock(mddev)) {
30b8aa91
N		 8989 if (mddev->pers)
8990 __md_stop_writes(mddev);
0f62fb22
N		 8991 if (mddev->persistent)
8992 mddev->safemode = 2;
c744a65c 8993 mddev_unlock(mddev);
2dba6a91 8994 }
c744a65c 8995 need_delay = 1;
1da177e4 8996 }
c744a65c
N		 8997 /*
8998 * certain more exotic SCSI devices are known to be
8999 * volatile wrt too early system reboots. While the
9000 * right place to handle this issue is the given
9001 * driver, we do want to have a safe RAID driver ...
9002 */
9003 if (need_delay)
9004 mdelay(1000*1);
9005
1da177e4
LT		 9006 return NOTIFY_DONE;
9007}
9008
75c96f85 9009static struct notifier_block md_notifier = {
1da177e4
LT		 9010 .notifier_call = md_notify_reboot,
9011 .next = NULL,
9012 .priority = INT_MAX, /* before any real devices */
9013};
9014
9015static void md_geninit(void)
9016{
36a4e1fe 9017 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 9018
c7705f34 9019 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 9020}
9021
75c96f85 9022static int __init md_init(void)
1da177e4 9023{
e804ac78
TH		 9024 int ret = -ENOMEM;
9025
ada609ee 9026 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 9027 if (!md_wq)
9028 goto err_wq;
9029
9030 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9031 if (!md_misc_wq)
9032 goto err_misc_wq;
9033
9034 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9035 goto err_md;
9036
9037 if ((ret = register_blkdev(0, "mdp")) < 0)
9038 goto err_mdp;
9039 mdp_major = ret;
9040
af5628f0 9041 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
e8703fe1
N		 9042 md_probe, NULL, NULL);
9043 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 9044 md_probe, NULL, NULL);
9045
1da177e4 9046 register_reboot_notifier(&md_notifier);
0b4d4147 9047 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 9048
9049 md_geninit();
d710e138 9050 return 0;
1da177e4 9051
e804ac78
TH		 9052err_mdp:
9053 unregister_blkdev(MD_MAJOR, "md");
9054err_md:
9055 destroy_workqueue(md_misc_wq);
9056err_misc_wq:
9057 destroy_workqueue(md_wq);
9058err_wq:
9059 return ret;
9060}
1da177e4 9061
70bcecdb 9062static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
1d7e3e96 9063{
70bcecdb
GR		 9064 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9065 struct md_rdev *rdev2;
9066 int role, ret;
9067 char b[BDEVNAME_SIZE];
1d7e3e96 9068
70bcecdb
GR		 9069 /* Check for change of roles in the active devices */
9070 rdev_for_each(rdev2, mddev) {
9071 if (test_bit(Faulty, &rdev2->flags))
9072 continue;
9073
9074 /* Check if the roles changed */
9075 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
dbb64f86
GR		 9076
9077 if (test_bit(Candidate, &rdev2->flags)) {
9078 if (role == 0xfffe) {
9079 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9080 md_kick_rdev_from_array(rdev2);
9081 continue;
9082 }
9083 else
9084 clear_bit(Candidate, &rdev2->flags);
9085 }
9086
70bcecdb
GR		 9087 if (role != rdev2->raid_disk) {
9088 /* got activated */
9089 if (rdev2->raid_disk == -1 && role != 0xffff) {
9090 rdev2->saved_raid_disk = role;
9091 ret = remove_and_add_spares(mddev, rdev2);
9092 pr_info("Activated spare: %s\n",
9093 bdevname(rdev2->bdev,b));
9094 continue;
9095 }
9096 /* device faulty
9097 * We just want to do the minimum to mark the disk
9098 * as faulty. The recovery is performed by the
9099 * one who initiated the error.
9100 */
9101 if ((role == 0xfffe) || (role == 0xfffd)) {
9102 md_error(mddev, rdev2);
9103 clear_bit(Blocked, &rdev2->flags);
9104 }
9105 }
1d7e3e96 9106 }
70bcecdb 9107
28c1b9fd
GR		 9108 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9109 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
70bcecdb
GR		 9110
9111 /* Finally set the event to be up to date */
9112 mddev->events = le64_to_cpu(sb->events);
9113}
9114
9115static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9116{
9117 int err;
9118 struct page *swapout = rdev->sb_page;
9119 struct mdp_superblock_1 *sb;
9120
9121 /* Store the sb page of the rdev in the swapout temporary
9122 * variable in case we err in the future
9123 */
9124 rdev->sb_page = NULL;
9125 alloc_disk_sb(rdev);
9126 ClearPageUptodate(rdev->sb_page);
9127 rdev->sb_loaded = 0;
9128 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
9129
9130 if (err < 0) {
9131 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9132 __func__, __LINE__, rdev->desc_nr, err);
9133 put_page(rdev->sb_page);
9134 rdev->sb_page = swapout;
9135 rdev->sb_loaded = 1;
9136 return err;
1d7e3e96
GR		 9137 }
9138
70bcecdb
GR		 9139 sb = page_address(rdev->sb_page);
9140 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9141 * is not set
9142 */
9143
9144 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9145 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9146
9147 /* The other node finished recovery, call spare_active to set
9148 * device In_sync and mddev->degraded
9149 */
9150 if (rdev->recovery_offset == MaxSector &&
9151 !test_bit(In_sync, &rdev->flags) &&
9152 mddev->pers->spare_active(mddev))
9153 sysfs_notify(&mddev->kobj, NULL, "degraded");
9154
9155 put_page(swapout);
9156 return 0;
9157}
9158
9159void md_reload_sb(struct mddev *mddev, int nr)
9160{
9161 struct md_rdev *rdev;
9162 int err;
9163
9164 /* Find the rdev */
9165 rdev_for_each_rcu(rdev, mddev) {
9166 if (rdev->desc_nr == nr)
9167 break;
9168 }
9169
9170 if (!rdev || rdev->desc_nr != nr) {
9171 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9172 return;
9173 }
9174
9175 err = read_rdev(mddev, rdev);
9176 if (err < 0)
9177 return;
9178
9179 check_sb_changes(mddev, rdev);
9180
9181 /* Read all rdev's to update recovery_offset */
9182 rdev_for_each_rcu(rdev, mddev)
9183 read_rdev(mddev, rdev);
1d7e3e96
GR		 9184}
9185EXPORT_SYMBOL(md_reload_sb);
9186
1da177e4
LT		 9187#ifndef MODULE
9188
9189/*
9190 * Searches all registered partitions for autorun RAID arrays
9191 * at boot time.
9192 */
4d936ec1
ME		 9193
9194static LIST_HEAD(all_detected_devices);
9195struct detected_devices_node {
9196 struct list_head list;
9197 dev_t dev;
9198};
1da177e4
LT		 9199
9200void md_autodetect_dev(dev_t dev)
9201{
4d936ec1
ME		 9202 struct detected_devices_node *node_detected_dev;
9203
9204 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9205 if (node_detected_dev) {
9206 node_detected_dev->dev = dev;
9207 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9208 } else {
9209 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
9210 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
9211 }
1da177e4
LT		 9212}
9213
1da177e4
LT		 9214static void autostart_arrays(int part)
9215{
3cb03002 9216 struct md_rdev *rdev;
4d936ec1
ME		 9217 struct detected_devices_node *node_detected_dev;
9218 dev_t dev;
9219 int i_scanned, i_passed;
1da177e4 9220
4d936ec1
ME		 9221 i_scanned = 0;
9222 i_passed = 0;
1da177e4 9223
4d936ec1 9224 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 9225
4d936ec1
ME		 9226 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9227 i_scanned++;
9228 node_detected_dev = list_entry(all_detected_devices.next,
9229 struct detected_devices_node, list);
9230 list_del(&node_detected_dev->list);
9231 dev = node_detected_dev->dev;
9232 kfree(node_detected_dev);
df968c4e 9233 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 9234 if (IS_ERR(rdev))
9235 continue;
9236
403df478 9237 if (test_bit(Faulty, &rdev->flags))
1da177e4 9238 continue;
403df478 9239
d0fae18f 9240 set_bit(AutoDetected, &rdev->flags);
1da177e4 9241 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 9242 i_passed++;
1da177e4 9243 }
4d936ec1
ME		 9244
9245 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
9246 i_scanned, i_passed);
1da177e4
LT		 9247
9248 autorun_devices(part);
9249}
9250
fdee8ae4 9251#endif /* !MODULE */
1da177e4
LT		 9252
9253static __exit void md_exit(void)
9254{
fd01b88c 9255 struct mddev *mddev;
1da177e4 9256 struct list_head *tmp;
e2f23b60 9257 int delay = 1;
8ab5e4c1 9258
af5628f0 9259 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
e8703fe1 9260 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 9261
3dbd8c2e 9262 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 9263 unregister_blkdev(mdp_major, "mdp");
9264 unregister_reboot_notifier(&md_notifier);
9265 unregister_sysctl_table(raid_table_header);
e2f23b60
N		 9266
9267 /* We cannot unload the modules while some process is
9268 * waiting for us in select() or poll() - wake them up
9269 */
9270 md_unloading = 1;
9271 while (waitqueue_active(&md_event_waiters)) {
9272 /* not safe to leave yet */
9273 wake_up(&md_event_waiters);
9274 msleep(delay);
9275 delay += delay;
9276 }
1da177e4 9277 remove_proc_entry("mdstat", NULL);
e2f23b60 9278
29ac4aa3 9279 for_each_mddev(mddev, tmp) {
1da177e4 9280 export_array(mddev);
d3374825 9281 mddev->hold_active = 0;
1da177e4 9282 }
e804ac78
TH		 9283 destroy_workqueue(md_misc_wq);
9284 destroy_workqueue(md_wq);
1da177e4
LT		 9285}
9286
685784aa 9287subsys_initcall(md_init);
1da177e4
LT		 9288module_exit(md_exit)
9289
f91de92e
N		 9290static int get_ro(char *buffer, struct kernel_param *kp)
9291{
9292 return sprintf(buffer, "%d", start_readonly);
9293}
9294static int set_ro(const char *val, struct kernel_param *kp)
9295{
4c9309c0 9296 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
f91de92e
N		 9297}
9298
80ca3a44
N		 9299module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9300module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
efeb53c0 9301module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 9302
1da177e4 9303MODULE_LICENSE("GPL");
0efb9e61 9304MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 9305MODULE_ALIAS("md");
72008652 9306MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)