projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: Export and rename find_rdev_nr_rcu
[deliverable/linux.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 */
5a7bbad2 253static void md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 254{
49077326 255 const int rw = bio_data_dir(bio);
fd01b88c 256 struct mddev *mddev = q->queuedata;
e91ece55 257 unsigned int sectors;
49077326 258
0ca69886
N		 259 if (mddev == NULL || mddev->pers == NULL
260 || !mddev->ready) {
409c57f3 261 bio_io_error(bio);
5a7bbad2 262 return;
409c57f3 263 }
bbfa57c0
SR		 264 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
265 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
266 return;
267 }
0ca69886 268 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 269 rcu_read_lock();
e9c7469b 270 if (mddev->suspended) {
409c57f3
N		 271 DEFINE_WAIT(__wait);
272 for (;;) {
273 prepare_to_wait(&mddev->sb_wait, &__wait,
274 TASK_UNINTERRUPTIBLE);
e9c7469b 275 if (!mddev->suspended)
409c57f3
N		 276 break;
277 rcu_read_unlock();
278 schedule();
279 rcu_read_lock();
280 }
281 finish_wait(&mddev->sb_wait, &__wait);
282 }
283 atomic_inc(&mddev->active_io);
284 rcu_read_unlock();
49077326 285
e91ece55
CM		 286 /*
287 * save the sectors now since our bio can
288 * go away inside make_request
289 */
290 sectors = bio_sectors(bio);
5a7bbad2 291 mddev->pers->make_request(mddev, bio);
49077326 292
18c0b223 293 generic_start_io_acct(rw, sectors, &mddev->gendisk->part0);
49077326 294
409c57f3
N		 295 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
296 wake_up(&mddev->sb_wait);
409c57f3
N		 297}
298
9e35b99c
N		 299/* mddev_suspend makes sure no new requests are submitted
300 * to the device, and that any requests that have been submitted
301 * are completely handled.
afa0f557
N		 302 * Once mddev_detach() is called and completes, the module will be
303 * completely unused.
9e35b99c 304 */
fd01b88c 305void mddev_suspend(struct mddev *mddev)
409c57f3
N		 306{
307 BUG_ON(mddev->suspended);
308 mddev->suspended = 1;
309 synchronize_rcu();
310 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
311 mddev->pers->quiesce(mddev, 1);
0d9f4f13
JB		 312
313 del_timer_sync(&mddev->safemode_timer);
409c57f3 314}
390ee602 315EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 316
fd01b88c 317void mddev_resume(struct mddev *mddev)
409c57f3
N		 318{
319 mddev->suspended = 0;
320 wake_up(&mddev->sb_wait);
321 mddev->pers->quiesce(mddev, 0);
0fd018af 322
47525e59 323 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
0fd018af
JB		 324 md_wakeup_thread(mddev->thread);
325 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 326}
390ee602 327EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 328
fd01b88c 329int mddev_congested(struct mddev *mddev, int bits)
3fa841d7 330{
5c675f83
N		 331 struct md_personality *pers = mddev->pers;
332 int ret = 0;
333
334 rcu_read_lock();
335 if (mddev->suspended)
336 ret = 1;
337 else if (pers && pers->congested)
338 ret = pers->congested(mddev, bits);
339 rcu_read_unlock();
340 return ret;
341}
342EXPORT_SYMBOL_GPL(mddev_congested);
343static int md_congested(void *data, int bits)
344{
345 struct mddev *mddev = data;
346 return mddev_congested(mddev, bits);
3fa841d7 347}
3fa841d7 348
64590f45
N		 349static int md_mergeable_bvec(struct request_queue *q,
350 struct bvec_merge_data *bvm,
351 struct bio_vec *biovec)
352{
353 struct mddev *mddev = q->queuedata;
354 int ret;
355 rcu_read_lock();
356 if (mddev->suspended) {
357 /* Must always allow one vec */
358 if (bvm->bi_size == 0)
359 ret = biovec->bv_len;
360 else
361 ret = 0;
362 } else {
363 struct md_personality *pers = mddev->pers;
364 if (pers && pers->mergeable_bvec)
365 ret = pers->mergeable_bvec(mddev, bvm, biovec);
366 else
367 ret = biovec->bv_len;
368 }
369 rcu_read_unlock();
370 return ret;
371}
a2826aa9 372/*
e9c7469b 373 * Generic flush handling for md
a2826aa9
N		 374 */
375
e9c7469b 376static void md_end_flush(struct bio *bio, int err)
a2826aa9 377{
3cb03002 378 struct md_rdev *rdev = bio->bi_private;
fd01b88c 379 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 380
381 rdev_dec_pending(rdev, mddev);
382
383 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 384 /* The pre-request flush has finished */
e804ac78 385 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 386 }
387 bio_put(bio);
388}
389
a7a07e69
N		 390static void md_submit_flush_data(struct work_struct *ws);
391
a035fc3e 392static void submit_flushes(struct work_struct *ws)
a2826aa9 393{
fd01b88c 394 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 395 struct md_rdev *rdev;
a2826aa9 396
a7a07e69
N		 397 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
398 atomic_set(&mddev->flush_pending, 1);
a2826aa9 399 rcu_read_lock();
dafb20fa 400 rdev_for_each_rcu(rdev, mddev)
a2826aa9
N		 401 if (rdev->raid_disk >= 0 &&
402 !test_bit(Faulty, &rdev->flags)) {
403 /* Take two references, one is dropped
404 * when request finishes, one after
405 * we reclaim rcu_read_lock
406 */
407 struct bio *bi;
408 atomic_inc(&rdev->nr_pending);
409 atomic_inc(&rdev->nr_pending);
410 rcu_read_unlock();
b5e1b8ce 411 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
e9c7469b 412 bi->bi_end_io = md_end_flush;
a2826aa9
N		 413 bi->bi_private = rdev;
414 bi->bi_bdev = rdev->bdev;
415 atomic_inc(&mddev->flush_pending);
e9c7469b 416 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 417 rcu_read_lock();
418 rdev_dec_pending(rdev, mddev);
419 }
420 rcu_read_unlock();
a7a07e69
N		 421 if (atomic_dec_and_test(&mddev->flush_pending))
422 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 423}
424
e9c7469b 425static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 426{
fd01b88c 427 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 428 struct bio *bio = mddev->flush_bio;
a2826aa9 429
4f024f37 430 if (bio->bi_iter.bi_size == 0)
a2826aa9
N		 431 /* an empty barrier - all done */
432 bio_endio(bio, 0);
433 else {
e9c7469b 434 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 435 mddev->pers->make_request(mddev, bio);
a2826aa9 436 }
2b74e12e
N		 437
438 mddev->flush_bio = NULL;
439 wake_up(&mddev->sb_wait);
a2826aa9
N		 440}
441
fd01b88c 442void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9 443{
85572d7c 444 spin_lock_irq(&mddev->lock);
a2826aa9 445 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 446 !mddev->flush_bio,
85572d7c 447 mddev->lock);
e9c7469b 448 mddev->flush_bio = bio;
85572d7c 449 spin_unlock_irq(&mddev->lock);
a2826aa9 450
a035fc3e
N		 451 INIT_WORK(&mddev->flush_work, submit_flushes);
452 queue_work(md_wq, &mddev->flush_work);
a2826aa9 453}
e9c7469b 454EXPORT_SYMBOL(md_flush_request);
409c57f3 455
74018dc3 456void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
97658cdd 457{
9cbb1750
N		 458 struct mddev *mddev = cb->data;
459 md_wakeup_thread(mddev->thread);
460 kfree(cb);
97658cdd 461}
9cbb1750 462EXPORT_SYMBOL(md_unplug);
2ac87401 463
fd01b88c 464static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 465{
466 atomic_inc(&mddev->active);
467 return mddev;
468}
469
5fd3a17e 470static void mddev_delayed_delete(struct work_struct *ws);
d3374825 471
fd01b88c 472static void mddev_put(struct mddev *mddev)
1da177e4 473{
a167f663
N		 474 struct bio_set *bs = NULL;
475
1da177e4
LT		 476 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
477 return;
d3374825 478 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 479 mddev->ctime == 0 && !mddev->hold_active) {
480 /* Array is not configured at all, and not held active,
481 * so destroy it */
af8a2434 482 list_del_init(&mddev->all_mddevs);
a167f663
N		 483 bs = mddev->bio_set;
484 mddev->bio_set = NULL;
d3374825 485 if (mddev->gendisk) {
e804ac78
TH		 486 /* We did a probe so need to clean up. Call
487 * queue_work inside the spinlock so that
488 * flush_workqueue() after mddev_find will
489 * succeed in waiting for the work to be done.
d3374825
N		 490 */
491 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 492 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 493 } else
494 kfree(mddev);
495 }
496 spin_unlock(&all_mddevs_lock);
a167f663
N		 497 if (bs)
498 bioset_free(bs);
1da177e4
LT		 499}
500
fd01b88c 501void mddev_init(struct mddev *mddev)
fafd7fb0
N		 502{
503 mutex_init(&mddev->open_mutex);
504 mutex_init(&mddev->reconfig_mutex);
505 mutex_init(&mddev->bitmap_info.mutex);
506 INIT_LIST_HEAD(&mddev->disks);
507 INIT_LIST_HEAD(&mddev->all_mddevs);
508 init_timer(&mddev->safemode_timer);
509 atomic_set(&mddev->active, 1);
510 atomic_set(&mddev->openers, 0);
511 atomic_set(&mddev->active_io, 0);
85572d7c 512 spin_lock_init(&mddev->lock);
fafd7fb0
N		 513 atomic_set(&mddev->flush_pending, 0);
514 init_waitqueue_head(&mddev->sb_wait);
515 init_waitqueue_head(&mddev->recovery_wait);
516 mddev->reshape_position = MaxSector;
2c810cdd 517 mddev->reshape_backwards = 0;
c4a39551 518 mddev->last_sync_action = "none";
fafd7fb0
N		 519 mddev->resync_min = 0;
520 mddev->resync_max = MaxSector;
521 mddev->level = LEVEL_NONE;
522}
390ee602 523EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 524
f72ffdd6 525static struct mddev *mddev_find(dev_t unit)
1da177e4 526{
fd01b88c 527 struct mddev *mddev, *new = NULL;
1da177e4 528
8f5f02c4
N		 529 if (unit && MAJOR(unit) != MD_MAJOR)
530 unit &= ~((1<<MdpMinorShift)-1);
531
1da177e4
LT		 532 retry:
533 spin_lock(&all_mddevs_lock);
efeb53c0
N		 534
535 if (unit) {
536 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
537 if (mddev->unit == unit) {
538 mddev_get(mddev);
539 spin_unlock(&all_mddevs_lock);
540 kfree(new);
541 return mddev;
542 }
543
544 if (new) {
545 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 546 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 547 new->hold_active = UNTIL_IOCTL;
548 return new;
1da177e4 549 }
efeb53c0
N		 550 } else if (new) {
551 /* find an unused unit number */
552 static int next_minor = 512;
553 int start = next_minor;
554 int is_free = 0;
555 int dev = 0;
556 while (!is_free) {
557 dev = MKDEV(MD_MAJOR, next_minor);
558 next_minor++;
559 if (next_minor > MINORMASK)
560 next_minor = 0;
561 if (next_minor == start) {
562 /* Oh dear, all in use. */
563 spin_unlock(&all_mddevs_lock);
564 kfree(new);
565 return NULL;
566 }
f72ffdd6 567
efeb53c0
N		 568 is_free = 1;
569 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
570 if (mddev->unit == dev) {
571 is_free = 0;
572 break;
573 }
574 }
575 new->unit = dev;
576 new->md_minor = MINOR(dev);
577 new->hold_active = UNTIL_STOP;
1da177e4
LT		 578 list_add(&new->all_mddevs, &all_mddevs);
579 spin_unlock(&all_mddevs_lock);
580 return new;
581 }
582 spin_unlock(&all_mddevs_lock);
583
9ffae0cf 584 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 585 if (!new)
586 return NULL;
587
1da177e4
LT		 588 new->unit = unit;
589 if (MAJOR(unit) == MD_MAJOR)
590 new->md_minor = MINOR(unit);
591 else
592 new->md_minor = MINOR(unit) >> MdpMinorShift;
593
fafd7fb0 594 mddev_init(new);
1da177e4 595
1da177e4
LT		 596 goto retry;
597}
598
b6eb127d
N		 599static struct attribute_group md_redundancy_group;
600
5c47daf6 601void mddev_unlock(struct mddev *mddev)
1da177e4 602{
a64c876f 603 if (mddev->to_remove) {
b6eb127d
N		 604 /* These cannot be removed under reconfig_mutex as
605 * an access to the files will try to take reconfig_mutex
606 * while holding the file unremovable, which leads to
607 * a deadlock.
bb4f1e9d
N		 608 * So hold set sysfs_active while the remove in happeing,
609 * and anything else which might set ->to_remove or my
610 * otherwise change the sysfs namespace will fail with
611 * -EBUSY if sysfs_active is still set.
612 * We set sysfs_active under reconfig_mutex and elsewhere
613 * test it under the same mutex to ensure its correct value
614 * is seen.
b6eb127d 615 */
a64c876f
N		 616 struct attribute_group *to_remove = mddev->to_remove;
617 mddev->to_remove = NULL;
bb4f1e9d 618 mddev->sysfs_active = 1;
b6eb127d
N		 619 mutex_unlock(&mddev->reconfig_mutex);
620
00bcb4ac
N		 621 if (mddev->kobj.sd) {
622 if (to_remove != &md_redundancy_group)
623 sysfs_remove_group(&mddev->kobj, to_remove);
624 if (mddev->pers == NULL ||
625 mddev->pers->sync_request == NULL) {
626 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
627 if (mddev->sysfs_action)
628 sysfs_put(mddev->sysfs_action);
629 mddev->sysfs_action = NULL;
630 }
a64c876f 631 }
bb4f1e9d 632 mddev->sysfs_active = 0;
b6eb127d
N		 633 } else
634 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 635
751e67ca
CD		 636 /* As we've dropped the mutex we need a spinlock to
637 * make sure the thread doesn't disappear
01f96c0a
N		 638 */
639 spin_lock(&pers_lock);
005eca5e 640 md_wakeup_thread(mddev->thread);
01f96c0a 641 spin_unlock(&pers_lock);
1da177e4 642}
5c47daf6 643EXPORT_SYMBOL_GPL(mddev_unlock);
1da177e4 644
57d051dc 645struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
1ca69c4b
N		 646{
647 struct md_rdev *rdev;
648
649 rdev_for_each_rcu(rdev, mddev)
650 if (rdev->desc_nr == nr)
651 return rdev;
652
653 return NULL;
654}
57d051dc 655EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
1ca69c4b
N		 656
657static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
1da177e4 658{
3cb03002 659 struct md_rdev *rdev;
1da177e4 660
dafb20fa 661 rdev_for_each(rdev, mddev)
1da177e4
LT		 662 if (rdev->bdev->bd_dev == dev)
663 return rdev;
159ec1fc 664
1da177e4
LT		 665 return NULL;
666}
667
1ca69c4b
N		 668static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
669{
670 struct md_rdev *rdev;
671
672 rdev_for_each_rcu(rdev, mddev)
673 if (rdev->bdev->bd_dev == dev)
674 return rdev;
675
676 return NULL;
677}
678
84fc4b56 679static struct md_personality *find_pers(int level, char *clevel)
2604b703 680{
84fc4b56 681 struct md_personality *pers;
d9d166c2
N		 682 list_for_each_entry(pers, &pers_list, list) {
683 if (level != LEVEL_NONE && pers->level == level)
2604b703 684 return pers;
d9d166c2
N		 685 if (strcmp(pers->name, clevel)==0)
686 return pers;
687 }
2604b703
N		 688 return NULL;
689}
690
b73df2d3 691/* return the offset of the super block in 512byte sectors */
3cb03002 692static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 693{
57b2caa3 694 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 695 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 696}
697
f72ffdd6 698static int alloc_disk_sb(struct md_rdev *rdev)
1da177e4 699{
1da177e4
LT		 700 rdev->sb_page = alloc_page(GFP_KERNEL);
701 if (!rdev->sb_page) {
702 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 703 return -ENOMEM;
1da177e4
LT		 704 }
705
706 return 0;
707}
708
545c8795 709void md_rdev_clear(struct md_rdev *rdev)
1da177e4
LT		 710{
711 if (rdev->sb_page) {
2d1f3b5d 712 put_page(rdev->sb_page);
1da177e4
LT		 713 rdev->sb_loaded = 0;
714 rdev->sb_page = NULL;
0f420358 715 rdev->sb_start = 0;
dd8ac336 716 rdev->sectors = 0;
1da177e4 717 }
2699b672
N		 718 if (rdev->bb_page) {
719 put_page(rdev->bb_page);
720 rdev->bb_page = NULL;
721 }
4fa2f327
N		 722 kfree(rdev->badblocks.page);
723 rdev->badblocks.page = NULL;
1da177e4 724}
545c8795 725EXPORT_SYMBOL_GPL(md_rdev_clear);
1da177e4 726
6712ecf8 727static void super_written(struct bio *bio, int error)
7bfa19f2 728{
3cb03002 729 struct md_rdev *rdev = bio->bi_private;
fd01b88c 730 struct mddev *mddev = rdev->mddev;
7bfa19f2 731
3a0f5bbb
N		 732 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
733 printk("md: super_written gets error=%d, uptodate=%d\n",
734 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
735 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 736 md_error(mddev, rdev);
3a0f5bbb 737 }
7bfa19f2 738
a9701a30
N		 739 if (atomic_dec_and_test(&mddev->pending_writes))
740 wake_up(&mddev->sb_wait);
f8b58edf 741 bio_put(bio);
7bfa19f2
N		 742}
743
fd01b88c 744void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 745 sector_t sector, int size, struct page *page)
746{
747 /* write first size bytes of page to sector of rdev
748 * Increment mddev->pending_writes before returning
749 * and decrement it on completion, waking up sb_wait
750 * if zero is reached.
751 * If an error occurred, call md_error
752 */
a167f663 753 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 754
a6ff7e08 755 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
4f024f37 756 bio->bi_iter.bi_sector = sector;
7bfa19f2
N		 757 bio_add_page(bio, page, size, 0);
758 bio->bi_private = rdev;
759 bio->bi_end_io = super_written;
a9701a30 760
7bfa19f2 761 atomic_inc(&mddev->pending_writes);
a5bf4df0 762 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 763}
764
fd01b88c 765void md_super_wait(struct mddev *mddev)
a9701a30 766{
e9c7469b 767 /* wait for all superblock writes that were scheduled to complete */
1967cd56 768 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
7bfa19f2
N		 769}
770
3cb03002 771int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 772 struct page *page, int rw, bool metadata_op)
1da177e4 773{
a167f663 774 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 775 int ret;
776
a6ff7e08
JB		 777 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
778 rdev->meta_bdev : rdev->bdev;
ccebd4c4 779 if (metadata_op)
4f024f37 780 bio->bi_iter.bi_sector = sector + rdev->sb_start;
1fdd6fc9
N		 781 else if (rdev->mddev->reshape_position != MaxSector &&
782 (rdev->mddev->reshape_backwards ==
783 (sector >= rdev->mddev->reshape_position)))
4f024f37 784 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
ccebd4c4 785 else
4f024f37 786 bio->bi_iter.bi_sector = sector + rdev->data_offset;
1da177e4 787 bio_add_page(bio, page, size, 0);
c170bbb4 788 submit_bio_wait(rw, bio);
1da177e4
LT		 789
790 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
791 bio_put(bio);
792 return ret;
793}
a8745db2 794EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 795
f72ffdd6 796static int read_disk_sb(struct md_rdev *rdev, int size)
1da177e4
LT		 797{
798 char b[BDEVNAME_SIZE];
403df478 799
1da177e4
LT		 800 if (rdev->sb_loaded)
801 return 0;
802
ccebd4c4 803 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 804 goto fail;
805 rdev->sb_loaded = 1;
806 return 0;
807
808fail:
809 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
810 bdevname(rdev->bdev,b));
811 return -EINVAL;
812}
813
814static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
815{
f72ffdd6 816 return sb1->set_uuid0 == sb2->set_uuid0 &&
05710466
AN		 817 sb1->set_uuid1 == sb2->set_uuid1 &&
818 sb1->set_uuid2 == sb2->set_uuid2 &&
819 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 820}
821
1da177e4
LT		 822static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
823{
824 int ret;
825 mdp_super_t *tmp1, *tmp2;
826
827 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
828 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
829
830 if (!tmp1 || !tmp2) {
831 ret = 0;
35020f1a 832 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 833 goto abort;
834 }
835
836 *tmp1 = *sb1;
837 *tmp2 = *sb2;
838
839 /*
840 * nr_disks is not constant
841 */
842 tmp1->nr_disks = 0;
843 tmp2->nr_disks = 0;
844
ce0c8e05 845 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 846abort:
990a8baf
JJ		 847 kfree(tmp1);
848 kfree(tmp2);
1da177e4
LT		 849 return ret;
850}
851
4d167f09
N		 852static u32 md_csum_fold(u32 csum)
853{
854 csum = (csum & 0xffff) + (csum >> 16);
855 return (csum & 0xffff) + (csum >> 16);
856}
857
f72ffdd6 858static unsigned int calc_sb_csum(mdp_super_t *sb)
1da177e4 859{
4d167f09
N		 860 u64 newcsum = 0;
861 u32 *sb32 = (u32*)sb;
862 int i;
1da177e4
LT		 863 unsigned int disk_csum, csum;
864
865 disk_csum = sb->sb_csum;
866 sb->sb_csum = 0;
4d167f09
N		 867
868 for (i = 0; i < MD_SB_BYTES/4 ; i++)
869 newcsum += sb32[i];
870 csum = (newcsum & 0xffffffff) + (newcsum>>32);
871
4d167f09
N		 872#ifdef CONFIG_ALPHA
873 /* This used to use csum_partial, which was wrong for several
874 * reasons including that different results are returned on
875 * different architectures. It isn't critical that we get exactly
876 * the same return value as before (we always csum_fold before
877 * testing, and that removes any differences). However as we
878 * know that csum_partial always returned a 16bit value on
879 * alphas, do a fold to maximise conformity to previous behaviour.
880 */
881 sb->sb_csum = md_csum_fold(disk_csum);
882#else
1da177e4 883 sb->sb_csum = disk_csum;
4d167f09 884#endif
1da177e4
LT		 885 return csum;
886}
887
1da177e4
LT		 888/*
889 * Handle superblock details.
890 * We want to be able to handle multiple superblock formats
891 * so we have a common interface to them all, and an array of
892 * different handlers.
893 * We rely on user-space to write the initial superblock, and support
894 * reading and updating of superblocks.
895 * Interface methods are:
3cb03002 896 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 897 * loads and validates a superblock on dev.
898 * if refdev != NULL, compare superblocks on both devices
899 * Return:
900 * 0 - dev has a superblock that is compatible with refdev
901 * 1 - dev has a superblock that is compatible and newer than refdev
902 * so dev should be used as the refdev in future
903 * -EINVAL superblock incompatible or invalid
904 * -othererror e.g. -EIO
905 *
fd01b88c 906 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 907 * Verify that dev is acceptable into mddev.
908 * The first time, mddev->raid_disks will be 0, and data from
909 * dev should be merged in. Subsequent calls check that dev
910 * is new enough. Return 0 or -EINVAL
911 *
fd01b88c 912 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 913 * Update the superblock for rdev with data in mddev
914 * This does not write to disc.
915 *
916 */
917
918struct super_type {
0cd17fec
CW		 919 char *name;
920 struct module *owner;
c6563a8c
N		 921 int (*load_super)(struct md_rdev *rdev,
922 struct md_rdev *refdev,
0cd17fec 923 int minor_version);
c6563a8c
N		 924 int (*validate_super)(struct mddev *mddev,
925 struct md_rdev *rdev);
926 void (*sync_super)(struct mddev *mddev,
927 struct md_rdev *rdev);
3cb03002 928 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 929 sector_t num_sectors);
c6563a8c
N		 930 int (*allow_new_offset)(struct md_rdev *rdev,
931 unsigned long long new_offset);
1da177e4
LT		 932};
933
0894cc30
AN		 934/*
935 * Check that the given mddev has no bitmap.
936 *
937 * This function is called from the run method of all personalities that do not
938 * support bitmaps. It prints an error message and returns non-zero if mddev
939 * has a bitmap. Otherwise, it returns 0.
940 *
941 */
fd01b88c 942int md_check_no_bitmap(struct mddev *mddev)
0894cc30 943{
c3d9714e 944 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 945 return 0;
946 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
947 mdname(mddev), mddev->pers->name);
948 return 1;
949}
950EXPORT_SYMBOL(md_check_no_bitmap);
951
1da177e4 952/*
f72ffdd6 953 * load_super for 0.90.0
1da177e4 954 */
3cb03002 955static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 956{
957 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
958 mdp_super_t *sb;
959 int ret;
1da177e4
LT		 960
961 /*
0f420358 962 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 963 * it's at the end of the disk.
964 *
965 * It also happens to be a multiple of 4Kb.
966 */
57b2caa3 967 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 968
0002b271 969 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 970 if (ret) return ret;
971
972 ret = -EINVAL;
973
974 bdevname(rdev->bdev, b);
65a06f06 975 sb = page_address(rdev->sb_page);
1da177e4
LT		 976
977 if (sb->md_magic != MD_SB_MAGIC) {
978 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
979 b);
980 goto abort;
981 }
982
983 if (sb->major_version != 0 ||
f6705578
N		 984 sb->minor_version < 90 ||
985 sb->minor_version > 91) {
1da177e4
LT		 986 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
987 sb->major_version, sb->minor_version,
988 b);
989 goto abort;
990 }
991
992 if (sb->raid_disks <= 0)
993 goto abort;
994
4d167f09 995 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 996 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
997 b);
998 goto abort;
999 }
1000
1001 rdev->preferred_minor = sb->md_minor;
1002 rdev->data_offset = 0;
c6563a8c 1003 rdev->new_data_offset = 0;
0002b271 1004 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1005 rdev->badblocks.shift = -1;
1da177e4
LT		 1006
1007 if (sb->level == LEVEL_MULTIPATH)
1008 rdev->desc_nr = -1;
1009 else
1010 rdev->desc_nr = sb->this_disk.number;
1011
9a7b2b0f 1012 if (!refdev) {
1da177e4 1013 ret = 1;
9a7b2b0f 1014 } else {
1da177e4 1015 __u64 ev1, ev2;
65a06f06 1016 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1017 if (!uuid_equal(refsb, sb)) {
1018 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1019 b, bdevname(refdev->bdev,b2));
1020 goto abort;
1021 }
1022 if (!sb_equal(refsb, sb)) {
1023 printk(KERN_WARNING "md: %s has same UUID"
1024 " but different superblock to %s\n",
1025 b, bdevname(refdev->bdev, b2));
1026 goto abort;
1027 }
1028 ev1 = md_event(sb);
1029 ev2 = md_event(refsb);
1030 if (ev1 > ev2)
1031 ret = 1;
f72ffdd6 1032 else
1da177e4
LT		 1033 ret = 0;
1034 }
8190e754 1035 rdev->sectors = rdev->sb_start;
667a5313
N		 1036 /* Limit to 4TB as metadata cannot record more than that.
1037 * (not needed for Linear and RAID0 as metadata doesn't
1038 * record this size)
1039 */
1040 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
27a7b260 1041 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1042
27a7b260 1043 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1044 /* "this cannot possibly happen" ... */
1045 ret = -EINVAL;
1046
1da177e4
LT		 1047 abort:
1048 return ret;
1049}
1050
1051/*
1052 * validate_super for 0.90.0
1053 */
fd01b88c 1054static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1055{
1056 mdp_disk_t *desc;
65a06f06 1057 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1058 __u64 ev1 = md_event(sb);
1da177e4 1059
41158c7e 1060 rdev->raid_disk = -1;
c5d79adb
N		 1061 clear_bit(Faulty, &rdev->flags);
1062 clear_bit(In_sync, &rdev->flags);
8313b8e5 1063 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1064 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1065
1da177e4
LT		 1066 if (mddev->raid_disks == 0) {
1067 mddev->major_version = 0;
1068 mddev->minor_version = sb->minor_version;
1069 mddev->patch_version = sb->patch_version;
e691063a 1070 mddev->external = 0;
9d8f0363 1071 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1072 mddev->ctime = sb->ctime;
1073 mddev->utime = sb->utime;
1074 mddev->level = sb->level;
d9d166c2 1075 mddev->clevel[0] = 0;
1da177e4
LT		 1076 mddev->layout = sb->layout;
1077 mddev->raid_disks = sb->raid_disks;
27a7b260 1078 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1079 mddev->events = ev1;
c3d9714e 1080 mddev->bitmap_info.offset = 0;
6409bb05
N		 1081 mddev->bitmap_info.space = 0;
1082 /* bitmap can use 60 K after the 4K superblocks */
c3d9714e 1083 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 1084 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
2c810cdd 1085 mddev->reshape_backwards = 0;
1da177e4 1086
f6705578
N		 1087 if (mddev->minor_version >= 91) {
1088 mddev->reshape_position = sb->reshape_position;
1089 mddev->delta_disks = sb->delta_disks;
1090 mddev->new_level = sb->new_level;
1091 mddev->new_layout = sb->new_layout;
664e7c41 1092 mddev->new_chunk_sectors = sb->new_chunk >> 9;
2c810cdd
N		 1093 if (mddev->delta_disks < 0)
1094 mddev->reshape_backwards = 1;
f6705578
N		 1095 } else {
1096 mddev->reshape_position = MaxSector;
1097 mddev->delta_disks = 0;
1098 mddev->new_level = mddev->level;
1099 mddev->new_layout = mddev->layout;
664e7c41 1100 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1101 }
1102
1da177e4
LT		 1103 if (sb->state & (1<<MD_SB_CLEAN))
1104 mddev->recovery_cp = MaxSector;
1105 else {
f72ffdd6 1106 if (sb->events_hi == sb->cp_events_hi &&
1da177e4
LT		 1107 sb->events_lo == sb->cp_events_lo) {
1108 mddev->recovery_cp = sb->recovery_cp;
1109 } else
1110 mddev->recovery_cp = 0;
1111 }
1112
1113 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1114 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1115 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1116 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1117
1118 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1119
1120 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
6409bb05 1121 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1122 mddev->bitmap_info.offset =
1123 mddev->bitmap_info.default_offset;
6409bb05 1124 mddev->bitmap_info.space =
c9ad020f 1125 mddev->bitmap_info.default_space;
6409bb05 1126 }
a654b9d8 1127
41158c7e 1128 } else if (mddev->pers == NULL) {
be6800a7
N		 1129 /* Insist on good event counter while assembling, except
1130 * for spares (which don't need an event count) */
1da177e4 1131 ++ev1;
be6800a7
N		 1132 if (sb->disks[rdev->desc_nr].state & (
1133 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
f72ffdd6 1134 if (ev1 < mddev->events)
be6800a7 1135 return -EINVAL;
41158c7e
N		 1136 } else if (mddev->bitmap) {
1137 /* if adding to array with a bitmap, then we can accept an
1138 * older device ... but not too old.
1139 */
41158c7e
N		 1140 if (ev1 < mddev->bitmap->events_cleared)
1141 return 0;
8313b8e5
N		 1142 if (ev1 < mddev->events)
1143 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1144 } else {
1145 if (ev1 < mddev->events)
1146 /* just a hot-add of a new device, leave raid_disk at -1 */
1147 return 0;
1148 }
41158c7e 1149
1da177e4 1150 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1151 desc = sb->disks + rdev->desc_nr;
1152
1153 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1154 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1155 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1156 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1157 set_bit(In_sync, &rdev->flags);
1da177e4 1158 rdev->raid_disk = desc->raid_disk;
f466722c 1159 rdev->saved_raid_disk = desc->raid_disk;
0261cd9f
N		 1160 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1161 /* active but not in sync implies recovery up to
1162 * reshape position. We don't know exactly where
1163 * that is, so set to zero for now */
1164 if (mddev->minor_version >= 91) {
1165 rdev->recovery_offset = 0;
1166 rdev->raid_disk = desc->raid_disk;
1167 }
1da177e4 1168 }
8ddf9efe
N		 1169 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1170 set_bit(WriteMostly, &rdev->flags);
41158c7e 1171 } else /* MULTIPATH are always insync */
b2d444d7 1172 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1173 return 0;
1174}
1175
1176/*
1177 * sync_super for 0.90.0
1178 */
fd01b88c 1179static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1180{
1181 mdp_super_t *sb;
3cb03002 1182 struct md_rdev *rdev2;
1da177e4 1183 int next_spare = mddev->raid_disks;
19133a42 1184
1da177e4
LT		 1185 /* make rdev->sb match mddev data..
1186 *
1187 * 1/ zero out disks
1188 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1189 * 3/ any empty disks < next_spare become removed
1190 *
1191 * disks[0] gets initialised to REMOVED because
1192 * we cannot be sure from other fields if it has
1193 * been initialised or not.
1194 */
1195 int i;
1196 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1197
61181565
N		 1198 rdev->sb_size = MD_SB_BYTES;
1199
65a06f06 1200 sb = page_address(rdev->sb_page);
1da177e4
LT		 1201
1202 memset(sb, 0, sizeof(*sb));
1203
1204 sb->md_magic = MD_SB_MAGIC;
1205 sb->major_version = mddev->major_version;
1da177e4
LT		 1206 sb->patch_version = mddev->patch_version;
1207 sb->gvalid_words = 0; /* ignored */
1208 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1209 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1210 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1211 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1212
1213 sb->ctime = mddev->ctime;
1214 sb->level = mddev->level;
58c0fed4 1215 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1216 sb->raid_disks = mddev->raid_disks;
1217 sb->md_minor = mddev->md_minor;
e691063a 1218 sb->not_persistent = 0;
1da177e4
LT		 1219 sb->utime = mddev->utime;
1220 sb->state = 0;
1221 sb->events_hi = (mddev->events>>32);
1222 sb->events_lo = (u32)mddev->events;
1223
f6705578
N		 1224 if (mddev->reshape_position == MaxSector)
1225 sb->minor_version = 90;
1226 else {
1227 sb->minor_version = 91;
1228 sb->reshape_position = mddev->reshape_position;
1229 sb->new_level = mddev->new_level;
1230 sb->delta_disks = mddev->delta_disks;
1231 sb->new_layout = mddev->new_layout;
664e7c41 1232 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1233 }
1234 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1235 if (mddev->in_sync)
1236 {
1237 sb->recovery_cp = mddev->recovery_cp;
1238 sb->cp_events_hi = (mddev->events>>32);
1239 sb->cp_events_lo = (u32)mddev->events;
1240 if (mddev->recovery_cp == MaxSector)
1241 sb->state = (1<< MD_SB_CLEAN);
1242 } else
1243 sb->recovery_cp = 0;
1244
1245 sb->layout = mddev->layout;
9d8f0363 1246 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1247
c3d9714e 1248 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1249 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1250
1da177e4 1251 sb->disks[0].state = (1<<MD_DISK_REMOVED);
dafb20fa 1252 rdev_for_each(rdev2, mddev) {
1da177e4 1253 mdp_disk_t *d;
86e6ffdd 1254 int desc_nr;
0261cd9f
N		 1255 int is_active = test_bit(In_sync, &rdev2->flags);
1256
1257 if (rdev2->raid_disk >= 0 &&
1258 sb->minor_version >= 91)
1259 /* we have nowhere to store the recovery_offset,
1260 * but if it is not below the reshape_position,
1261 * we can piggy-back on that.
1262 */
1263 is_active = 1;
1264 if (rdev2->raid_disk < 0 ||
1265 test_bit(Faulty, &rdev2->flags))
1266 is_active = 0;
1267 if (is_active)
86e6ffdd 1268 desc_nr = rdev2->raid_disk;
1da177e4 1269 else
86e6ffdd 1270 desc_nr = next_spare++;
19133a42 1271 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1272 d = &sb->disks[rdev2->desc_nr];
1273 nr_disks++;
1274 d->number = rdev2->desc_nr;
1275 d->major = MAJOR(rdev2->bdev->bd_dev);
1276 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1277 if (is_active)
1da177e4
LT		 1278 d->raid_disk = rdev2->raid_disk;
1279 else
1280 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1281 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1282 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1283 else if (is_active) {
1da177e4 1284 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1285 if (test_bit(In_sync, &rdev2->flags))
1286 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1287 active++;
1288 working++;
1289 } else {
1290 d->state = 0;
1291 spare++;
1292 working++;
1293 }
8ddf9efe
N		 1294 if (test_bit(WriteMostly, &rdev2->flags))
1295 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1296 }
1da177e4
LT		 1297 /* now set the "removed" and "faulty" bits on any missing devices */
1298 for (i=0 ; i < mddev->raid_disks ; i++) {
1299 mdp_disk_t *d = &sb->disks[i];
1300 if (d->state == 0 && d->number == 0) {
1301 d->number = i;
1302 d->raid_disk = i;
1303 d->state = (1<<MD_DISK_REMOVED);
1304 d->state |= (1<<MD_DISK_FAULTY);
1305 failed++;
1306 }
1307 }
1308 sb->nr_disks = nr_disks;
1309 sb->active_disks = active;
1310 sb->working_disks = working;
1311 sb->failed_disks = failed;
1312 sb->spare_disks = spare;
1313
1314 sb->this_disk = sb->disks[rdev->desc_nr];
1315 sb->sb_csum = calc_sb_csum(sb);
1316}
1317
0cd17fec
CW		 1318/*
1319 * rdev_size_change for 0.90.0
1320 */
1321static unsigned long long
3cb03002 1322super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1323{
58c0fed4 1324 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1325 return 0; /* component must fit device */
c3d9714e 1326 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1327 return 0; /* can't move bitmap */
57b2caa3 1328 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1329 if (!num_sectors || num_sectors > rdev->sb_start)
1330 num_sectors = rdev->sb_start;
27a7b260
N		 1331 /* Limit to 4TB as metadata cannot record more than that.
1332 * 4TB == 2^32 KB, or 2*2^32 sectors.
1333 */
667a5313 1334 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
27a7b260 1335 num_sectors = (2ULL << 32) - 2;
0f420358 1336 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1337 rdev->sb_page);
1338 md_super_wait(rdev->mddev);
c26a44ed 1339 return num_sectors;
0cd17fec
CW		 1340}
1341
c6563a8c
N		 1342static int
1343super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1344{
1345 /* non-zero offset changes not possible with v0.90 */
1346 return new_offset == 0;
1347}
0cd17fec 1348
1da177e4
LT		 1349/*
1350 * version 1 superblock
1351 */
1352
f72ffdd6 1353static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1da177e4 1354{
1c05b4bc
N		 1355 __le32 disk_csum;
1356 u32 csum;
1da177e4
LT		 1357 unsigned long long newcsum;
1358 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1359 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1360
1361 disk_csum = sb->sb_csum;
1362 sb->sb_csum = 0;
1363 newcsum = 0;
1f3c9907 1364 for (; size >= 4; size -= 4)
1da177e4
LT		 1365 newcsum += le32_to_cpu(*isuper++);
1366
1367 if (size == 2)
1c05b4bc 1368 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1369
1370 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1371 sb->sb_csum = disk_csum;
1372 return cpu_to_le32(csum);
1373}
1374
2699b672
N		 1375static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1376 int acknowledged);
3cb03002 1377static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1378{
1379 struct mdp_superblock_1 *sb;
1380 int ret;
0f420358 1381 sector_t sb_start;
c6563a8c 1382 sector_t sectors;
1da177e4 1383 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1384 int bmask;
1da177e4
LT		 1385
1386 /*
0f420358 1387 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1388 * It is always aligned to a 4K boundary and
1389 * depeding on minor_version, it can be:
1390 * 0: At least 8K, but less than 12K, from end of device
1391 * 1: At start of device
1392 * 2: 4K from start of device.
1393 */
1394 switch(minor_version) {
1395 case 0:
77304d2a 1396 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1397 sb_start -= 8*2;
1398 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1399 break;
1400 case 1:
0f420358 1401 sb_start = 0;
1da177e4
LT		 1402 break;
1403 case 2:
0f420358 1404 sb_start = 8;
1da177e4
LT		 1405 break;
1406 default:
1407 return -EINVAL;
1408 }
0f420358 1409 rdev->sb_start = sb_start;
1da177e4 1410
0002b271
N		 1411 /* superblock is rarely larger than 1K, but it can be larger,
1412 * and it is safe to read 4k, so we do that
1413 */
1414 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1415 if (ret) return ret;
1416
65a06f06 1417 sb = page_address(rdev->sb_page);
1da177e4
LT		 1418
1419 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1420 sb->major_version != cpu_to_le32(1) ||
1421 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1422 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1423 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1424 return -EINVAL;
1425
1426 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1427 printk("md: invalid superblock checksum on %s\n",
1428 bdevname(rdev->bdev,b));
1429 return -EINVAL;
1430 }
1431 if (le64_to_cpu(sb->data_size) < 10) {
1432 printk("md: data_size too small on %s\n",
1433 bdevname(rdev->bdev,b));
1434 return -EINVAL;
1435 }
c6563a8c
N		 1436 if (sb->pad0 ||
1437 sb->pad3[0] ||
1438 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1439 /* Some padding is non-zero, might be a new feature */
1440 return -EINVAL;
e11e93fa 1441
1da177e4
LT		 1442 rdev->preferred_minor = 0xffff;
1443 rdev->data_offset = le64_to_cpu(sb->data_offset);
c6563a8c
N		 1444 rdev->new_data_offset = rdev->data_offset;
1445 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1446 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1447 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
4dbcdc75 1448 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1449
0002b271 1450 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1451 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1452 if (rdev->sb_size & bmask)
a1801f85
N		 1453 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1454
1455 if (minor_version
0f420358 1456 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1457 return -EINVAL;
c6563a8c
N		 1458 if (minor_version
1459 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1460 return -EINVAL;
0002b271 1461
31b65a0d
N		 1462 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1463 rdev->desc_nr = -1;
1464 else
1465 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1466
2699b672
N		 1467 if (!rdev->bb_page) {
1468 rdev->bb_page = alloc_page(GFP_KERNEL);
1469 if (!rdev->bb_page)
1470 return -ENOMEM;
1471 }
1472 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1473 rdev->badblocks.count == 0) {
1474 /* need to load the bad block list.
1475 * Currently we limit it to one page.
1476 */
1477 s32 offset;
1478 sector_t bb_sector;
1479 u64 *bbp;
1480 int i;
1481 int sectors = le16_to_cpu(sb->bblog_size);
1482 if (sectors > (PAGE_SIZE / 512))
1483 return -EINVAL;
1484 offset = le32_to_cpu(sb->bblog_offset);
1485 if (offset == 0)
1486 return -EINVAL;
1487 bb_sector = (long long)offset;
1488 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1489 rdev->bb_page, READ, true))
1490 return -EIO;
1491 bbp = (u64 *)page_address(rdev->bb_page);
1492 rdev->badblocks.shift = sb->bblog_shift;
1493 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1494 u64 bb = le64_to_cpu(*bbp);
1495 int count = bb & (0x3ff);
1496 u64 sector = bb >> 10;
1497 sector <<= sb->bblog_shift;
1498 count <<= sb->bblog_shift;
1499 if (bb + 1 == 0)
1500 break;
1501 if (md_set_badblocks(&rdev->badblocks,
1502 sector, count, 1) == 0)
1503 return -EINVAL;
1504 }
486adf72
N		 1505 } else if (sb->bblog_offset != 0)
1506 rdev->badblocks.shift = 0;
2699b672 1507
9a7b2b0f 1508 if (!refdev) {
8ed75463 1509 ret = 1;
9a7b2b0f 1510 } else {
1da177e4 1511 __u64 ev1, ev2;
65a06f06 1512 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1513
1514 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1515 sb->level != refsb->level ||
1516 sb->layout != refsb->layout ||
1517 sb->chunksize != refsb->chunksize) {
1518 printk(KERN_WARNING "md: %s has strangely different"
1519 " superblock to %s\n",
1520 bdevname(rdev->bdev,b),
1521 bdevname(refdev->bdev,b2));
1522 return -EINVAL;
1523 }
1524 ev1 = le64_to_cpu(sb->events);
1525 ev2 = le64_to_cpu(refsb->events);
1526
1527 if (ev1 > ev2)
8ed75463
N		 1528 ret = 1;
1529 else
1530 ret = 0;
1da177e4 1531 }
c6563a8c
N		 1532 if (minor_version) {
1533 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1534 sectors -= rdev->data_offset;
1535 } else
1536 sectors = rdev->sb_start;
1537 if (sectors < le64_to_cpu(sb->data_size))
1da177e4 1538 return -EINVAL;
dd8ac336 1539 rdev->sectors = le64_to_cpu(sb->data_size);
8ed75463 1540 return ret;
1da177e4
LT		 1541}
1542
fd01b88c 1543static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1544{
65a06f06 1545 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1546 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1547
41158c7e 1548 rdev->raid_disk = -1;
c5d79adb
N		 1549 clear_bit(Faulty, &rdev->flags);
1550 clear_bit(In_sync, &rdev->flags);
8313b8e5 1551 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1552 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1553
1da177e4
LT		 1554 if (mddev->raid_disks == 0) {
1555 mddev->major_version = 1;
1556 mddev->patch_version = 0;
e691063a 1557 mddev->external = 0;
9d8f0363 1558 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1559 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1560 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1561 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1562 mddev->clevel[0] = 0;
1da177e4
LT		 1563 mddev->layout = le32_to_cpu(sb->layout);
1564 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1565 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1566 mddev->events = ev1;
c3d9714e 1567 mddev->bitmap_info.offset = 0;
6409bb05
N		 1568 mddev->bitmap_info.space = 0;
1569 /* Default location for bitmap is 1K after superblock
1570 * using 3K - total of 4K
1571 */
c3d9714e 1572 mddev->bitmap_info.default_offset = 1024 >> 9;
6409bb05 1573 mddev->bitmap_info.default_space = (4096-1024) >> 9;
2c810cdd
N		 1574 mddev->reshape_backwards = 0;
1575
1da177e4
LT		 1576 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1577 memcpy(mddev->uuid, sb->set_uuid, 16);
1578
1579 mddev->max_disks = (4096-256)/2;
a654b9d8 1580
71c0805c 1581 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
6409bb05 1582 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1583 mddev->bitmap_info.offset =
1584 (__s32)le32_to_cpu(sb->bitmap_offset);
6409bb05
N		 1585 /* Metadata doesn't record how much space is available.
1586 * For 1.0, we assume we can use up to the superblock
1587 * if before, else to 4K beyond superblock.
1588 * For others, assume no change is possible.
1589 */
1590 if (mddev->minor_version > 0)
1591 mddev->bitmap_info.space = 0;
1592 else if (mddev->bitmap_info.offset > 0)
1593 mddev->bitmap_info.space =
1594 8 - mddev->bitmap_info.offset;
1595 else
1596 mddev->bitmap_info.space =
1597 -mddev->bitmap_info.offset;
1598 }
e11e93fa 1599
f6705578
N		 1600 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1601 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1602 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1603 mddev->new_level = le32_to_cpu(sb->new_level);
1604 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1605 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
2c810cdd
N		 1606 if (mddev->delta_disks < 0 ||
1607 (mddev->delta_disks == 0 &&
1608 (le32_to_cpu(sb->feature_map)
1609 & MD_FEATURE_RESHAPE_BACKWARDS)))
1610 mddev->reshape_backwards = 1;
f6705578
N		 1611 } else {
1612 mddev->reshape_position = MaxSector;
1613 mddev->delta_disks = 0;
1614 mddev->new_level = mddev->level;
1615 mddev->new_layout = mddev->layout;
664e7c41 1616 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1617 }
1618
41158c7e 1619 } else if (mddev->pers == NULL) {
be6800a7
N		 1620 /* Insist of good event counter while assembling, except for
1621 * spares (which don't need an event count) */
1da177e4 1622 ++ev1;
be6800a7
N		 1623 if (rdev->desc_nr >= 0 &&
1624 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1625 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1626 if (ev1 < mddev->events)
1627 return -EINVAL;
41158c7e
N		 1628 } else if (mddev->bitmap) {
1629 /* If adding to array with a bitmap, then we can accept an
1630 * older device, but not too old.
1631 */
41158c7e
N		 1632 if (ev1 < mddev->bitmap->events_cleared)
1633 return 0;
8313b8e5
N		 1634 if (ev1 < mddev->events)
1635 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1636 } else {
1637 if (ev1 < mddev->events)
1638 /* just a hot-add of a new device, leave raid_disk at -1 */
1639 return 0;
1640 }
1da177e4
LT		 1641 if (mddev->level != LEVEL_MULTIPATH) {
1642 int role;
3673f305
N		 1643 if (rdev->desc_nr < 0 ||
1644 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1645 role = 0xffff;
1646 rdev->desc_nr = -1;
1647 } else
1648 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1649 switch(role) {
1650 case 0xffff: /* spare */
1da177e4
LT		 1651 break;
1652 case 0xfffe: /* faulty */
b2d444d7 1653 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1654 break;
1655 default:
f466722c 1656 rdev->saved_raid_disk = role;
5fd6c1dc 1657 if ((le32_to_cpu(sb->feature_map) &
f466722c 1658 MD_FEATURE_RECOVERY_OFFSET)) {
5fd6c1dc 1659 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
f466722c
N		 1660 if (!(le32_to_cpu(sb->feature_map) &
1661 MD_FEATURE_RECOVERY_BITMAP))
1662 rdev->saved_raid_disk = -1;
1663 } else
5fd6c1dc 1664 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1665 rdev->raid_disk = role;
1666 break;
1667 }
8ddf9efe
N		 1668 if (sb->devflags & WriteMostly1)
1669 set_bit(WriteMostly, &rdev->flags);
2d78f8c4
N		 1670 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1671 set_bit(Replacement, &rdev->flags);
41158c7e 1672 } else /* MULTIPATH are always insync */
b2d444d7 1673 set_bit(In_sync, &rdev->flags);
41158c7e 1674
1da177e4
LT		 1675 return 0;
1676}
1677
fd01b88c 1678static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1679{
1680 struct mdp_superblock_1 *sb;
3cb03002 1681 struct md_rdev *rdev2;
1da177e4
LT		 1682 int max_dev, i;
1683 /* make rdev->sb match mddev and rdev data. */
1684
65a06f06 1685 sb = page_address(rdev->sb_page);
1da177e4
LT		 1686
1687 sb->feature_map = 0;
1688 sb->pad0 = 0;
5fd6c1dc 1689 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1690 memset(sb->pad3, 0, sizeof(sb->pad3));
1691
1692 sb->utime = cpu_to_le64((__u64)mddev->utime);
1693 sb->events = cpu_to_le64(mddev->events);
1694 if (mddev->in_sync)
1695 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1696 else
1697 sb->resync_offset = cpu_to_le64(0);
1698
1c05b4bc 1699 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1700
f0ca340c 1701 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1702 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1703 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1704 sb->level = cpu_to_le32(mddev->level);
1705 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1706
aeb9b211
N		 1707 if (test_bit(WriteMostly, &rdev->flags))
1708 sb->devflags |= WriteMostly1;
1709 else
1710 sb->devflags &= ~WriteMostly1;
c6563a8c
N		 1711 sb->data_offset = cpu_to_le64(rdev->data_offset);
1712 sb->data_size = cpu_to_le64(rdev->sectors);
aeb9b211 1713
c3d9714e
N		 1714 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1715 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1716 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1717 }
5fd6c1dc
N		 1718
1719 if (rdev->raid_disk >= 0 &&
97e4f42d 1720 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1721 sb->feature_map |=
1722 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1723 sb->recovery_offset =
1724 cpu_to_le64(rdev->recovery_offset);
f466722c
N		 1725 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1726 sb->feature_map |=
1727 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
5fd6c1dc 1728 }
2d78f8c4
N		 1729 if (test_bit(Replacement, &rdev->flags))
1730 sb->feature_map |=
1731 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1732
f6705578
N		 1733 if (mddev->reshape_position != MaxSector) {
1734 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1735 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1736 sb->new_layout = cpu_to_le32(mddev->new_layout);
1737 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1738 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1739 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2c810cdd
N		 1740 if (mddev->delta_disks == 0 &&
1741 mddev->reshape_backwards)
1742 sb->feature_map
1743 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
c6563a8c
N		 1744 if (rdev->new_data_offset != rdev->data_offset) {
1745 sb->feature_map
1746 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1747 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1748 - rdev->data_offset));
1749 }
f6705578 1750 }
a654b9d8 1751
2699b672
N		 1752 if (rdev->badblocks.count == 0)
1753 /* Nothing to do for bad blocks*/ ;
1754 else if (sb->bblog_offset == 0)
1755 /* Cannot record bad blocks on this device */
1756 md_error(mddev, rdev);
1757 else {
1758 struct badblocks *bb = &rdev->badblocks;
1759 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1760 u64 *p = bb->page;
1761 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1762 if (bb->changed) {
1763 unsigned seq;
1764
1765retry:
1766 seq = read_seqbegin(&bb->lock);
1767
1768 memset(bbp, 0xff, PAGE_SIZE);
1769
1770 for (i = 0 ; i < bb->count ; i++) {
35f9ac2d 1771 u64 internal_bb = p[i];
2699b672
N		 1772 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1773 | BB_LEN(internal_bb));
35f9ac2d 1774 bbp[i] = cpu_to_le64(store_bb);
2699b672 1775 }
d0962936 1776 bb->changed = 0;
2699b672
N		 1777 if (read_seqretry(&bb->lock, seq))
1778 goto retry;
1779
1780 bb->sector = (rdev->sb_start +
1781 (int)le32_to_cpu(sb->bblog_offset));
1782 bb->size = le16_to_cpu(sb->bblog_size);
2699b672
N		 1783 }
1784 }
1785
1da177e4 1786 max_dev = 0;
dafb20fa 1787 rdev_for_each(rdev2, mddev)
1da177e4
LT		 1788 if (rdev2->desc_nr+1 > max_dev)
1789 max_dev = rdev2->desc_nr+1;
a778b73f 1790
70471daf
N		 1791 if (max_dev > le32_to_cpu(sb->max_dev)) {
1792 int bmask;
a778b73f 1793 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1794 rdev->sb_size = max_dev * 2 + 256;
1795 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1796 if (rdev->sb_size & bmask)
1797 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1798 } else
1799 max_dev = le32_to_cpu(sb->max_dev);
1800
1da177e4
LT		 1801 for (i=0; i<max_dev;i++)
1802 sb->dev_roles[i] = cpu_to_le16(0xfffe);
f72ffdd6 1803
dafb20fa 1804 rdev_for_each(rdev2, mddev) {
1da177e4 1805 i = rdev2->desc_nr;
b2d444d7 1806 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1807 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1808 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1809 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1810 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1811 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1812 else
1813 sb->dev_roles[i] = cpu_to_le16(0xffff);
1814 }
1815
1da177e4
LT		 1816 sb->sb_csum = calc_sb_1_csum(sb);
1817}
1818
0cd17fec 1819static unsigned long long
3cb03002 1820super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 1821{
1822 struct mdp_superblock_1 *sb;
15f4a5fd 1823 sector_t max_sectors;
58c0fed4 1824 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1825 return 0; /* component must fit device */
c6563a8c
N		 1826 if (rdev->data_offset != rdev->new_data_offset)
1827 return 0; /* too confusing */
0f420358 1828 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1829 /* minor versions 1 and 2; superblock before data */
77304d2a 1830 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1831 max_sectors -= rdev->data_offset;
1832 if (!num_sectors || num_sectors > max_sectors)
1833 num_sectors = max_sectors;
c3d9714e 1834 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1835 /* minor version 0 with bitmap we can't move */
1836 return 0;
1837 } else {
1838 /* minor version 0; superblock after data */
0f420358 1839 sector_t sb_start;
77304d2a 1840 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1841 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1842 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1843 if (!num_sectors || num_sectors > max_sectors)
1844 num_sectors = max_sectors;
0f420358 1845 rdev->sb_start = sb_start;
0cd17fec 1846 }
65a06f06 1847 sb = page_address(rdev->sb_page);
15f4a5fd 1848 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1849 sb->super_offset = rdev->sb_start;
0cd17fec 1850 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1851 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1852 rdev->sb_page);
1853 md_super_wait(rdev->mddev);
c26a44ed 1854 return num_sectors;
c6563a8c
N		 1855
1856}
1857
1858static int
1859super_1_allow_new_offset(struct md_rdev *rdev,
1860 unsigned long long new_offset)
1861{
1862 /* All necessary checks on new >= old have been done */
1863 struct bitmap *bitmap;
1864 if (new_offset >= rdev->data_offset)
1865 return 1;
1866
1867 /* with 1.0 metadata, there is no metadata to tread on
1868 * so we can always move back */
1869 if (rdev->mddev->minor_version == 0)
1870 return 1;
1871
1872 /* otherwise we must be sure not to step on
1873 * any metadata, so stay:
1874 * 36K beyond start of superblock
1875 * beyond end of badblocks
1876 * beyond write-intent bitmap
1877 */
1878 if (rdev->sb_start + (32+4)*2 > new_offset)
1879 return 0;
1880 bitmap = rdev->mddev->bitmap;
1881 if (bitmap && !rdev->mddev->bitmap_info.file &&
1882 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1ec885cd 1883 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
c6563a8c
N		 1884 return 0;
1885 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1886 return 0;
1887
1888 return 1;
0cd17fec 1889}
1da177e4 1890
75c96f85 1891static struct super_type super_types[] = {
1da177e4
LT		 1892 [0] = {
1893 .name = "0.90.0",
1894 .owner = THIS_MODULE,
0cd17fec
CW		 1895 .load_super = super_90_load,
1896 .validate_super = super_90_validate,
1897 .sync_super = super_90_sync,
1898 .rdev_size_change = super_90_rdev_size_change,
c6563a8c 1899 .allow_new_offset = super_90_allow_new_offset,
1da177e4
LT		 1900 },
1901 [1] = {
1902 .name = "md-1",
1903 .owner = THIS_MODULE,
0cd17fec
CW		 1904 .load_super = super_1_load,
1905 .validate_super = super_1_validate,
1906 .sync_super = super_1_sync,
1907 .rdev_size_change = super_1_rdev_size_change,
c6563a8c 1908 .allow_new_offset = super_1_allow_new_offset,
1da177e4
LT		 1909 },
1910};
1da177e4 1911
fd01b88c 1912static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 1913{
1914 if (mddev->sync_super) {
1915 mddev->sync_super(mddev, rdev);
1916 return;
1917 }
1918
1919 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1920
1921 super_types[mddev->major_version].sync_super(mddev, rdev);
1922}
1923
fd01b88c 1924static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1925{
3cb03002 1926 struct md_rdev *rdev, *rdev2;
1da177e4 1927
4b80991c
N		 1928 rcu_read_lock();
1929 rdev_for_each_rcu(rdev, mddev1)
1930 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1931 if (rdev->bdev->bd_contains ==
4b80991c
N		 1932 rdev2->bdev->bd_contains) {
1933 rcu_read_unlock();
7dd5e7c3 1934 return 1;
4b80991c
N		 1935 }
1936 rcu_read_unlock();
1da177e4
LT		 1937 return 0;
1938}
1939
1940static LIST_HEAD(pending_raid_disks);
1941
ac5e7113
AN		 1942/*
1943 * Try to register data integrity profile for an mddev
1944 *
1945 * This is called when an array is started and after a disk has been kicked
1946 * from the array. It only succeeds if all working and active component devices
1947 * are integrity capable with matching profiles.
1948 */
fd01b88c 1949int md_integrity_register(struct mddev *mddev)
ac5e7113 1950{
3cb03002 1951 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 1952
1953 if (list_empty(&mddev->disks))
1954 return 0; /* nothing to do */
629acb6a
JB		 1955 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1956 return 0; /* shouldn't register, or already is */
dafb20fa 1957 rdev_for_each(rdev, mddev) {
ac5e7113
AN		 1958 /* skip spares and non-functional disks */
1959 if (test_bit(Faulty, &rdev->flags))
1960 continue;
1961 if (rdev->raid_disk < 0)
1962 continue;
ac5e7113
AN		 1963 if (!reference) {
1964 /* Use the first rdev as the reference */
1965 reference = rdev;
1966 continue;
1967 }
1968 /* does this rdev's profile match the reference profile? */
1969 if (blk_integrity_compare(reference->bdev->bd_disk,
1970 rdev->bdev->bd_disk) < 0)
1971 return -EINVAL;
1972 }
89078d57
MP		 1973 if (!reference || !bdev_get_integrity(reference->bdev))
1974 return 0;
ac5e7113
AN		 1975 /*
1976 * All component devices are integrity capable and have matching
1977 * profiles, register the common profile for the md device.
1978 */
1979 if (blk_integrity_register(mddev->gendisk,
1980 bdev_get_integrity(reference->bdev)) != 0) {
1981 printk(KERN_ERR "md: failed to register integrity for %s\n",
1982 mdname(mddev));
1983 return -EINVAL;
1984 }
a91a2785
MP		 1985 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1986 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1987 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1988 mdname(mddev));
1989 return -EINVAL;
1990 }
ac5e7113
AN		 1991 return 0;
1992}
1993EXPORT_SYMBOL(md_integrity_register);
1994
1995/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 1996void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 1997{
2863b9eb
JB		 1998 struct blk_integrity *bi_rdev;
1999 struct blk_integrity *bi_mddev;
2000
2001 if (!mddev->gendisk)
2002 return;
2003
2004 bi_rdev = bdev_get_integrity(rdev->bdev);
2005 bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 2006
ac5e7113 2007 if (!bi_mddev) /* nothing to do */
3f9d99c1 2008 return;
ac5e7113 2009 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 2010 return;
ac5e7113
AN		 2011 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2012 rdev->bdev->bd_disk) >= 0)
2013 return;
2014 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2015 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2016}
ac5e7113 2017EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2018
f72ffdd6 2019static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
1da177e4 2020{
7dd5e7c3 2021 char b[BDEVNAME_SIZE];
f637b9f9 2022 struct kobject *ko;
1edf80d3 2023 char *s;
5e55e2f5 2024 int err;
1da177e4 2025
11e2ede0
DW		 2026 /* prevent duplicates */
2027 if (find_rdev(mddev, rdev->bdev->bd_dev))
2028 return -EEXIST;
2029
dd8ac336
AN		 2030 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2031 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2032 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2033 if (mddev->pers) {
2034 /* Cannot change size, so fail
2035 * If mddev->level <= 0, then we don't care
2036 * about aligning sizes (e.g. linear)
2037 */
2038 if (mddev->level > 0)
2039 return -ENOSPC;
2040 } else
dd8ac336 2041 mddev->dev_sectors = rdev->sectors;
2bf071bf 2042 }
1da177e4
LT		 2043
2044 /* Verify rdev->desc_nr is unique.
2045 * If it is -1, assign a free number, else
2046 * check number is not in use
2047 */
4878e9eb 2048 rcu_read_lock();
1da177e4
LT		 2049 if (rdev->desc_nr < 0) {
2050 int choice = 0;
4878e9eb
N		 2051 if (mddev->pers)
2052 choice = mddev->raid_disks;
57d051dc 2053 while (md_find_rdev_nr_rcu(mddev, choice))
1da177e4
LT		 2054 choice++;
2055 rdev->desc_nr = choice;
2056 } else {
57d051dc 2057 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
4878e9eb 2058 rcu_read_unlock();
1da177e4 2059 return -EBUSY;
4878e9eb 2060 }
1da177e4 2061 }
4878e9eb 2062 rcu_read_unlock();
de01dfad
N		 2063 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2064 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2065 mdname(mddev), mddev->max_disks);
2066 return -EBUSY;
2067 }
19133a42 2068 bdevname(rdev->bdev,b);
649316b2 2069 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2070 *s = '!';
649316b2 2071
1da177e4 2072 rdev->mddev = mddev;
19133a42 2073 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2074
b2d6db58 2075 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2076 goto fail;
86e6ffdd 2077
0762b8bd 2078 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2079 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2080 /* failure here is OK */;
2081 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2082
4b80991c 2083 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2084 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2085
2086 /* May as well allow recovery to be retried once */
5389042f 2087 mddev->recovery_disabled++;
3f9d99c1 2088
1da177e4 2089 return 0;
5e55e2f5
N		 2090
2091 fail:
2092 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2093 b, mdname(mddev));
2094 return err;
1da177e4
LT		 2095}
2096
177a99b2 2097static void md_delayed_delete(struct work_struct *ws)
5792a285 2098{
3cb03002 2099 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2100 kobject_del(&rdev->kobj);
177a99b2 2101 kobject_put(&rdev->kobj);
5792a285
N		 2102}
2103
f72ffdd6 2104static void unbind_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2105{
2106 char b[BDEVNAME_SIZE];
403df478 2107
49731baa 2108 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2109 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2110 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2111 rdev->mddev = NULL;
86e6ffdd 2112 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2113 sysfs_put(rdev->sysfs_state);
2114 rdev->sysfs_state = NULL;
2230dfe4 2115 rdev->badblocks.count = 0;
5792a285 2116 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2117 * writing to 'remove' to "dev/state". We also need
2118 * to delay it due to rcu usage.
5792a285 2119 */
4b80991c 2120 synchronize_rcu();
177a99b2
N		 2121 INIT_WORK(&rdev->del_work, md_delayed_delete);
2122 kobject_get(&rdev->kobj);
e804ac78 2123 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2124}
2125
2126/*
2127 * prevent the device from being mounted, repartitioned or
2128 * otherwise reused by a RAID array (or any other kernel
2129 * subsystem), by bd_claiming the device.
2130 */
3cb03002 2131static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2132{
2133 int err = 0;
2134 struct block_device *bdev;
2135 char b[BDEVNAME_SIZE];
2136
d4d77629 2137 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2138 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2139 if (IS_ERR(bdev)) {
2140 printk(KERN_ERR "md: could not open %s.\n",
2141 __bdevname(dev, b));
2142 return PTR_ERR(bdev);
2143 }
1da177e4
LT		 2144 rdev->bdev = bdev;
2145 return err;
2146}
2147
3cb03002 2148static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2149{
2150 struct block_device *bdev = rdev->bdev;
2151 rdev->bdev = NULL;
e525fd89 2152 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2153}
2154
2155void md_autodetect_dev(dev_t dev);
2156
f72ffdd6 2157static void export_rdev(struct md_rdev *rdev)
1da177e4
LT		 2158{
2159 char b[BDEVNAME_SIZE];
403df478 2160
1da177e4
LT		 2161 printk(KERN_INFO "md: export_rdev(%s)\n",
2162 bdevname(rdev->bdev,b));
545c8795 2163 md_rdev_clear(rdev);
1da177e4 2164#ifndef MODULE
d0fae18f
N		 2165 if (test_bit(AutoDetected, &rdev->flags))
2166 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2167#endif
2168 unlock_rdev(rdev);
86e6ffdd 2169 kobject_put(&rdev->kobj);
1da177e4
LT		 2170}
2171
fb56dfef 2172void md_kick_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2173{
2174 unbind_rdev_from_array(rdev);
2175 export_rdev(rdev);
2176}
fb56dfef 2177EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
1da177e4 2178
fd01b88c 2179static void export_array(struct mddev *mddev)
1da177e4 2180{
0638bb0e 2181 struct md_rdev *rdev;
1da177e4 2182
0638bb0e
N		 2183 while (!list_empty(&mddev->disks)) {
2184 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2185 same_set);
fb56dfef 2186 md_kick_rdev_from_array(rdev);
1da177e4 2187 }
1da177e4
LT		 2188 mddev->raid_disks = 0;
2189 mddev->major_version = 0;
2190}
2191
f72ffdd6 2192static void sync_sbs(struct mddev *mddev, int nospares)
1da177e4 2193{
42543769
N		 2194 /* Update each superblock (in-memory image), but
2195 * if we are allowed to, skip spares which already
2196 * have the right event counter, or have one earlier
2197 * (which would mean they aren't being marked as dirty
2198 * with the rest of the array)
2199 */
3cb03002 2200 struct md_rdev *rdev;
dafb20fa 2201 rdev_for_each(rdev, mddev) {
42543769
N		 2202 if (rdev->sb_events == mddev->events ||
2203 (nospares &&
2204 rdev->raid_disk < 0 &&
42543769
N		 2205 rdev->sb_events+1 == mddev->events)) {
2206 /* Don't update this superblock */
2207 rdev->sb_loaded = 2;
2208 } else {
076f968b 2209 sync_super(mddev, rdev);
42543769
N		 2210 rdev->sb_loaded = 1;
2211 }
1da177e4
LT		 2212 }
2213}
2214
1aee41f6 2215void md_update_sb(struct mddev *mddev, int force_change)
1da177e4 2216{
3cb03002 2217 struct md_rdev *rdev;
06d91a5f 2218 int sync_req;
42543769 2219 int nospares = 0;
2699b672 2220 int any_badblocks_changed = 0;
1da177e4 2221
d87f064f
N		 2222 if (mddev->ro) {
2223 if (force_change)
2224 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2225 return;
2226 }
1da177e4 2227repeat:
3a3a5ddb 2228 /* First make sure individual recovery_offsets are correct */
dafb20fa 2229 rdev_for_each(rdev, mddev) {
3a3a5ddb
N		 2230 if (rdev->raid_disk >= 0 &&
2231 mddev->delta_disks >= 0 &&
2232 !test_bit(In_sync, &rdev->flags) &&
2233 mddev->curr_resync_completed > rdev->recovery_offset)
2234 rdev->recovery_offset = mddev->curr_resync_completed;
2235
f72ffdd6 2236 }
bd52b746 2237 if (!mddev->persistent) {
070dc6dd 2238 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2239 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2240 if (!mddev->external) {
d97a41dc 2241 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
dafb20fa 2242 rdev_for_each(rdev, mddev) {
de393cde 2243 if (rdev->badblocks.changed) {
d0962936 2244 rdev->badblocks.changed = 0;
de393cde
N		 2245 md_ack_all_badblocks(&rdev->badblocks);
2246 md_error(mddev, rdev);
2247 }
2248 clear_bit(Blocked, &rdev->flags);
2249 clear_bit(BlockedBadBlocks, &rdev->flags);
2250 wake_up(&rdev->blocked_wait);
2251 }
2252 }
3a3a5ddb
N		 2253 wake_up(&mddev->sb_wait);
2254 return;
2255 }
2256
85572d7c 2257 spin_lock(&mddev->lock);
84692195 2258
3a3a5ddb
N		 2259 mddev->utime = get_seconds();
2260
850b2b42
N		 2261 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2262 force_change = 1;
2263 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2264 /* just a clean<-> dirty transition, possibly leave spares alone,
2265 * though if events isn't the right even/odd, we will have to do
2266 * spares after all
2267 */
2268 nospares = 1;
2269 if (force_change)
2270 nospares = 0;
2271 if (mddev->degraded)
84692195
N		 2272 /* If the array is degraded, then skipping spares is both
2273 * dangerous and fairly pointless.
2274 * Dangerous because a device that was removed from the array
2275 * might have a event_count that still looks up-to-date,
2276 * so it can be re-added without a resync.
2277 * Pointless because if there are any spares to skip,
2278 * then a recovery will happen and soon that array won't
2279 * be degraded any more and the spare can go back to sleep then.
2280 */
850b2b42 2281 nospares = 0;
84692195 2282
06d91a5f 2283 sync_req = mddev->in_sync;
42543769
N		 2284
2285 /* If this is just a dirty<->clean transition, and the array is clean
2286 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2287 if (nospares
42543769 2288 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2289 && mddev->can_decrease_events
2290 && mddev->events != 1) {
42543769 2291 mddev->events--;
a8707c08
N		 2292 mddev->can_decrease_events = 0;
2293 } else {
42543769
N		 2294 /* otherwise we have to go forward and ... */
2295 mddev->events ++;
a8707c08 2296 mddev->can_decrease_events = nospares;
42543769 2297 }
1da177e4 2298
403df478
N		 2299 /*
2300 * This 64-bit counter should never wrap.
2301 * Either we are in around ~1 trillion A.C., assuming
2302 * 1 reboot per second, or we have a bug...
2303 */
2304 WARN_ON(mddev->events == 0);
2699b672 2305
dafb20fa 2306 rdev_for_each(rdev, mddev) {
2699b672
N		 2307 if (rdev->badblocks.changed)
2308 any_badblocks_changed++;
de393cde
N		 2309 if (test_bit(Faulty, &rdev->flags))
2310 set_bit(FaultRecorded, &rdev->flags);
2311 }
2699b672 2312
e691063a 2313 sync_sbs(mddev, nospares);
85572d7c 2314 spin_unlock(&mddev->lock);
1da177e4 2315
36a4e1fe
N		 2316 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2317 mdname(mddev), mddev->in_sync);
1da177e4 2318
4ad13663 2319 bitmap_update_sb(mddev->bitmap);
dafb20fa 2320 rdev_for_each(rdev, mddev) {
1da177e4 2321 char b[BDEVNAME_SIZE];
36a4e1fe 2322
42543769
N		 2323 if (rdev->sb_loaded != 1)
2324 continue; /* no noise on spare devices */
1da177e4 2325
f466722c 2326 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2327 md_super_write(mddev,rdev,
0f420358 2328 rdev->sb_start, rdev->sb_size,
7bfa19f2 2329 rdev->sb_page);
36a4e1fe
N		 2330 pr_debug("md: (write) %s's sb offset: %llu\n",
2331 bdevname(rdev->bdev, b),
2332 (unsigned long long)rdev->sb_start);
42543769 2333 rdev->sb_events = mddev->events;
2699b672
N		 2334 if (rdev->badblocks.size) {
2335 md_super_write(mddev, rdev,
2336 rdev->badblocks.sector,
2337 rdev->badblocks.size << 9,
2338 rdev->bb_page);
2339 rdev->badblocks.size = 0;
2340 }
7bfa19f2 2341
f466722c 2342 } else
36a4e1fe
N		 2343 pr_debug("md: %s (skipping faulty)\n",
2344 bdevname(rdev->bdev, b));
d70ed2e4 2345
7bfa19f2 2346 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2347 /* only need to write one superblock... */
2348 break;
2349 }
a9701a30 2350 md_super_wait(mddev);
850b2b42 2351 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2352
85572d7c 2353 spin_lock(&mddev->lock);
850b2b42
N		 2354 if (mddev->in_sync != sync_req ||
2355 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2356 /* have to write it out again */
85572d7c 2357 spin_unlock(&mddev->lock);
06d91a5f
N		 2358 goto repeat;
2359 }
850b2b42 2360 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
85572d7c 2361 spin_unlock(&mddev->lock);
3d310eb7 2362 wake_up(&mddev->sb_wait);
acb180b0
N		 2363 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2364 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2365
dafb20fa 2366 rdev_for_each(rdev, mddev) {
de393cde
N		 2367 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2368 clear_bit(Blocked, &rdev->flags);
2369
2370 if (any_badblocks_changed)
2699b672 2371 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2372 clear_bit(BlockedBadBlocks, &rdev->flags);
2373 wake_up(&rdev->blocked_wait);
2374 }
1da177e4 2375}
1aee41f6 2376EXPORT_SYMBOL(md_update_sb);
1da177e4 2377
7f6ce769 2378/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2379 * We want to accept with case. For this we use cmd_match.
2380 */
2381static int cmd_match(const char *cmd, const char *str)
2382{
2383 /* See if cmd, written into a sysfs file, matches
2384 * str. They must either be the same, or cmd can
2385 * have a trailing newline
2386 */
2387 while (*cmd && *str && *cmd == *str) {
2388 cmd++;
2389 str++;
2390 }
2391 if (*cmd == '\n')
2392 cmd++;
2393 if (*str || *cmd)
2394 return 0;
2395 return 1;
2396}
2397
86e6ffdd
N		 2398struct rdev_sysfs_entry {
2399 struct attribute attr;
3cb03002
N		 2400 ssize_t (*show)(struct md_rdev *, char *);
2401 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2402};
2403
2404static ssize_t
3cb03002 2405state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2406{
2407 char *sep = "";
20a49ff6 2408 size_t len = 0;
758bfc8a 2409 unsigned long flags = ACCESS_ONCE(rdev->flags);
86e6ffdd 2410
758bfc8a 2411 if (test_bit(Faulty, &flags) ||
de393cde 2412 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2413 len+= sprintf(page+len, "%sfaulty",sep);
2414 sep = ",";
2415 }
758bfc8a 2416 if (test_bit(In_sync, &flags)) {
86e6ffdd
N		 2417 len += sprintf(page+len, "%sin_sync",sep);
2418 sep = ",";
2419 }
758bfc8a 2420 if (test_bit(WriteMostly, &flags)) {
f655675b
N		 2421 len += sprintf(page+len, "%swrite_mostly",sep);
2422 sep = ",";
2423 }
758bfc8a 2424 if (test_bit(Blocked, &flags) ||
52c64152 2425 (rdev->badblocks.unacked_exist
758bfc8a 2426 && !test_bit(Faulty, &flags))) {
6bfe0b49
DW		 2427 len += sprintf(page+len, "%sblocked", sep);
2428 sep = ",";
2429 }
758bfc8a
N		 2430 if (!test_bit(Faulty, &flags) &&
2431 !test_bit(In_sync, &flags)) {
86e6ffdd
N		 2432 len += sprintf(page+len, "%sspare", sep);
2433 sep = ",";
2434 }
758bfc8a 2435 if (test_bit(WriteErrorSeen, &flags)) {
d7a9d443
N		 2436 len += sprintf(page+len, "%swrite_error", sep);
2437 sep = ",";
2438 }
758bfc8a 2439 if (test_bit(WantReplacement, &flags)) {
2d78f8c4
N		 2440 len += sprintf(page+len, "%swant_replacement", sep);
2441 sep = ",";
2442 }
758bfc8a 2443 if (test_bit(Replacement, &flags)) {
2d78f8c4
N		 2444 len += sprintf(page+len, "%sreplacement", sep);
2445 sep = ",";
2446 }
2447
86e6ffdd
N		 2448 return len+sprintf(page+len, "\n");
2449}
2450
45dc2de1 2451static ssize_t
3cb03002 2452state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2453{
2454 /* can write
de393cde 2455 * faulty - simulates an error
45dc2de1 2456 * remove - disconnects the device
f655675b
N		 2457 * writemostly - sets write_mostly
2458 * -writemostly - clears write_mostly
de393cde
N		 2459 * blocked - sets the Blocked flags
2460 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2461 * insync - sets Insync providing device isn't active
f466722c
N		 2462 * -insync - clear Insync for a device with a slot assigned,
2463 * so that it gets rebuilt based on bitmap
d7a9d443
N		 2464 * write_error - sets WriteErrorSeen
2465 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2466 */
2467 int err = -EINVAL;
2468 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2469 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2470 if (test_bit(Faulty, &rdev->flags))
2471 err = 0;
2472 else
2473 err = -EBUSY;
45dc2de1
N		 2474 } else if (cmd_match(buf, "remove")) {
2475 if (rdev->raid_disk >= 0)
2476 err = -EBUSY;
2477 else {
fd01b88c 2478 struct mddev *mddev = rdev->mddev;
293467aa
GR		 2479 if (mddev_is_clustered(mddev))
2480 md_cluster_ops->metadata_update_start(mddev);
fb56dfef 2481 md_kick_rdev_from_array(rdev);
3f9d7b0d
N		 2482 if (mddev->pers)
2483 md_update_sb(mddev, 1);
45dc2de1 2484 md_new_event(mddev);
293467aa
GR		 2485 if (mddev_is_clustered(mddev))
2486 md_cluster_ops->metadata_update_finish(mddev);
45dc2de1
N		 2487 err = 0;
2488 }
f655675b
N		 2489 } else if (cmd_match(buf, "writemostly")) {
2490 set_bit(WriteMostly, &rdev->flags);
2491 err = 0;
2492 } else if (cmd_match(buf, "-writemostly")) {
2493 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2494 err = 0;
2495 } else if (cmd_match(buf, "blocked")) {
2496 set_bit(Blocked, &rdev->flags);
2497 err = 0;
2498 } else if (cmd_match(buf, "-blocked")) {
de393cde 2499 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2500 rdev->badblocks.unacked_exist) {
de393cde
N		 2501 /* metadata handler doesn't understand badblocks,
2502 * so we need to fail the device
2503 */
2504 md_error(rdev->mddev, rdev);
2505 }
6bfe0b49 2506 clear_bit(Blocked, &rdev->flags);
de393cde 2507 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2508 wake_up(&rdev->blocked_wait);
2509 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2510 md_wakeup_thread(rdev->mddev->thread);
2511
6d56e278
N		 2512 err = 0;
2513 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2514 set_bit(In_sync, &rdev->flags);
f655675b 2515 err = 0;
f466722c 2516 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
e1960f8c
N		 2517 if (rdev->mddev->pers == NULL) {
2518 clear_bit(In_sync, &rdev->flags);
2519 rdev->saved_raid_disk = rdev->raid_disk;
2520 rdev->raid_disk = -1;
2521 err = 0;
2522 }
d7a9d443
N		 2523 } else if (cmd_match(buf, "write_error")) {
2524 set_bit(WriteErrorSeen, &rdev->flags);
2525 err = 0;
2526 } else if (cmd_match(buf, "-write_error")) {
2527 clear_bit(WriteErrorSeen, &rdev->flags);
2528 err = 0;
2d78f8c4
N		 2529 } else if (cmd_match(buf, "want_replacement")) {
2530 /* Any non-spare device that is not a replacement can
2531 * become want_replacement at any time, but we then need to
2532 * check if recovery is needed.
2533 */
2534 if (rdev->raid_disk >= 0 &&
2535 !test_bit(Replacement, &rdev->flags))
2536 set_bit(WantReplacement, &rdev->flags);
2537 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2538 md_wakeup_thread(rdev->mddev->thread);
2539 err = 0;
2540 } else if (cmd_match(buf, "-want_replacement")) {
2541 /* Clearing 'want_replacement' is always allowed.
2542 * Once replacements starts it is too late though.
2543 */
2544 err = 0;
2545 clear_bit(WantReplacement, &rdev->flags);
2546 } else if (cmd_match(buf, "replacement")) {
2547 /* Can only set a device as a replacement when array has not
2548 * yet been started. Once running, replacement is automatic
2549 * from spares, or by assigning 'slot'.
2550 */
2551 if (rdev->mddev->pers)
2552 err = -EBUSY;
2553 else {
2554 set_bit(Replacement, &rdev->flags);
2555 err = 0;
2556 }
2557 } else if (cmd_match(buf, "-replacement")) {
2558 /* Similarly, can only clear Replacement before start */
2559 if (rdev->mddev->pers)
2560 err = -EBUSY;
2561 else {
2562 clear_bit(Replacement, &rdev->flags);
2563 err = 0;
2564 }
45dc2de1 2565 }
00bcb4ac
N		 2566 if (!err)
2567 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2568 return err ? err : len;
2569}
80ca3a44
N		 2570static struct rdev_sysfs_entry rdev_state =
2571__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2572
4dbcdc75 2573static ssize_t
3cb03002 2574errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2575{
2576 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2577}
2578
2579static ssize_t
3cb03002 2580errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N		 2581{
2582 char *e;
2583 unsigned long n = simple_strtoul(buf, &e, 10);
2584 if (*buf && (*e == 0 || *e == '\n')) {
2585 atomic_set(&rdev->corrected_errors, n);
2586 return len;
2587 }
2588 return -EINVAL;
2589}
2590static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2591__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2592
014236d2 2593static ssize_t
3cb03002 2594slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2595{
2596 if (rdev->raid_disk < 0)
2597 return sprintf(page, "none\n");
2598 else
2599 return sprintf(page, "%d\n", rdev->raid_disk);
2600}
2601
2602static ssize_t
3cb03002 2603slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N		 2604{
2605 char *e;
c303da6d 2606 int err;
014236d2
N		 2607 int slot = simple_strtoul(buf, &e, 10);
2608 if (strncmp(buf, "none", 4)==0)
2609 slot = -1;
2610 else if (e==buf || (*e && *e!= '\n'))
2611 return -EINVAL;
6c2fce2e 2612 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2613 /* Setting 'slot' on an active array requires also
2614 * updating the 'rd%d' link, and communicating
2615 * with the personality with ->hot_*_disk.
2616 * For now we only support removing
2617 * failed/spare devices. This normally happens automatically,
2618 * but not when the metadata is externally managed.
2619 */
c303da6d
N		 2620 if (rdev->raid_disk == -1)
2621 return -EEXIST;
2622 /* personality does all needed checks */
01393f3d 2623 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d 2624 return -EINVAL;
746d3207
N		 2625 clear_bit(Blocked, &rdev->flags);
2626 remove_and_add_spares(rdev->mddev, rdev);
2627 if (rdev->raid_disk >= 0)
2628 return -EBUSY;
c303da6d
N		 2629 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2630 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2631 } else if (rdev->mddev->pers) {
6c2fce2e 2632 /* Activating a spare .. or possibly reactivating
6d56e278 2633 * if we ever get bitmaps working here.
6c2fce2e
NB		 2634 */
2635
2636 if (rdev->raid_disk != -1)
2637 return -EBUSY;
2638
c6751b2b
N		 2639 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2640 return -EBUSY;
2641
6c2fce2e
NB		 2642 if (rdev->mddev->pers->hot_add_disk == NULL)
2643 return -EINVAL;
2644
ba1b41b6
N		 2645 if (slot >= rdev->mddev->raid_disks &&
2646 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2647 return -ENOSPC;
2648
6c2fce2e
NB		 2649 rdev->raid_disk = slot;
2650 if (test_bit(In_sync, &rdev->flags))
2651 rdev->saved_raid_disk = slot;
2652 else
2653 rdev->saved_raid_disk = -1;
d30519fc 2654 clear_bit(In_sync, &rdev->flags);
8313b8e5 2655 clear_bit(Bitmap_sync, &rdev->flags);
6c2fce2e
NB		 2656 err = rdev->mddev->pers->
2657 hot_add_disk(rdev->mddev, rdev);
199050ea 2658 if (err) {
6c2fce2e 2659 rdev->raid_disk = -1;
6c2fce2e 2660 return err;
52664732 2661 } else
00bcb4ac 2662 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2663 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2664 /* failure here is OK */;
6c2fce2e 2665 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2666 } else {
ba1b41b6
N		 2667 if (slot >= rdev->mddev->raid_disks &&
2668 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2669 return -ENOSPC;
2670 rdev->raid_disk = slot;
2671 /* assume it is working */
c5d79adb
N		 2672 clear_bit(Faulty, &rdev->flags);
2673 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2674 set_bit(In_sync, &rdev->flags);
00bcb4ac 2675 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2676 }
014236d2
N		 2677 return len;
2678}
2679
014236d2 2680static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2681__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2682
93c8cad0 2683static ssize_t
3cb03002 2684offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2685{
6961ece4 2686 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2687}
2688
2689static ssize_t
3cb03002 2690offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2691{
c6563a8c 2692 unsigned long long offset;
b29bebd6 2693 if (kstrtoull(buf, 10, &offset) < 0)
93c8cad0 2694 return -EINVAL;
8ed0a521 2695 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2696 return -EBUSY;
dd8ac336 2697 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2698 /* Must set offset before size, so overlap checks
2699 * can be sane */
2700 return -EBUSY;
93c8cad0 2701 rdev->data_offset = offset;
25f7fd47 2702 rdev->new_data_offset = offset;
93c8cad0
N		 2703 return len;
2704}
2705
2706static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2707__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2708
c6563a8c
N		 2709static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2710{
2711 return sprintf(page, "%llu\n",
2712 (unsigned long long)rdev->new_data_offset);
2713}
2714
2715static ssize_t new_offset_store(struct md_rdev *rdev,
2716 const char *buf, size_t len)
2717{
2718 unsigned long long new_offset;
2719 struct mddev *mddev = rdev->mddev;
2720
b29bebd6 2721 if (kstrtoull(buf, 10, &new_offset) < 0)
c6563a8c
N		 2722 return -EINVAL;
2723
f851b60d
N		 2724 if (mddev->sync_thread ||
2725 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
c6563a8c
N		 2726 return -EBUSY;
2727 if (new_offset == rdev->data_offset)
2728 /* reset is always permitted */
2729 ;
2730 else if (new_offset > rdev->data_offset) {
2731 /* must not push array size beyond rdev_sectors */
2732 if (new_offset - rdev->data_offset
2733 + mddev->dev_sectors > rdev->sectors)
2734 return -E2BIG;
2735 }
2736 /* Metadata worries about other space details. */
2737
2738 /* decreasing the offset is inconsistent with a backwards
2739 * reshape.
2740 */
2741 if (new_offset < rdev->data_offset &&
2742 mddev->reshape_backwards)
2743 return -EINVAL;
2744 /* Increasing offset is inconsistent with forwards
2745 * reshape. reshape_direction should be set to
2746 * 'backwards' first.
2747 */
2748 if (new_offset > rdev->data_offset &&
2749 !mddev->reshape_backwards)
2750 return -EINVAL;
2751
2752 if (mddev->pers && mddev->persistent &&
2753 !super_types[mddev->major_version]
2754 .allow_new_offset(rdev, new_offset))
2755 return -E2BIG;
2756 rdev->new_data_offset = new_offset;
2757 if (new_offset > rdev->data_offset)
2758 mddev->reshape_backwards = 1;
2759 else if (new_offset < rdev->data_offset)
2760 mddev->reshape_backwards = 0;
2761
2762 return len;
2763}
2764static struct rdev_sysfs_entry rdev_new_offset =
2765__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2766
83303b61 2767static ssize_t
3cb03002 2768rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2769{
dd8ac336 2770 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2771}
2772
c5d79adb
N		 2773static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2774{
2775 /* check if two start/length pairs overlap */
2776 if (s1+l1 <= s2)
2777 return 0;
2778 if (s2+l2 <= s1)
2779 return 0;
2780 return 1;
2781}
2782
b522adcd
DW		 2783static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2784{
2785 unsigned long long blocks;
2786 sector_t new;
2787
b29bebd6 2788 if (kstrtoull(buf, 10, &blocks) < 0)
b522adcd
DW		 2789 return -EINVAL;
2790
2791 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2792 return -EINVAL; /* sector conversion overflow */
2793
2794 new = blocks * 2;
2795 if (new != blocks * 2)
2796 return -EINVAL; /* unsigned long long to sector_t overflow */
2797
2798 *sectors = new;
2799 return 0;
2800}
2801
83303b61 2802static ssize_t
3cb03002 2803rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2804{
fd01b88c 2805 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2806 sector_t oldsectors = rdev->sectors;
b522adcd 2807 sector_t sectors;
27c529bb 2808
b522adcd 2809 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2810 return -EINVAL;
c6563a8c
N		 2811 if (rdev->data_offset != rdev->new_data_offset)
2812 return -EINVAL; /* too confusing */
0cd17fec 2813 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2814 if (my_mddev->persistent) {
dd8ac336
AN		 2815 sectors = super_types[my_mddev->major_version].
2816 rdev_size_change(rdev, sectors);
2817 if (!sectors)
0cd17fec 2818 return -EBUSY;
dd8ac336 2819 } else if (!sectors)
77304d2a 2820 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2821 rdev->data_offset;
a6468539
N		 2822 if (!my_mddev->pers->resize)
2823 /* Cannot change size for RAID0 or Linear etc */
2824 return -EINVAL;
0cd17fec 2825 }
dd8ac336 2826 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2827 return -EINVAL; /* component must fit device */
0cd17fec 2828
dd8ac336
AN		 2829 rdev->sectors = sectors;
2830 if (sectors > oldsectors && my_mddev->external) {
8b1afc3d
N		 2831 /* Need to check that all other rdevs with the same
2832 * ->bdev do not overlap. 'rcu' is sufficient to walk
2833 * the rdev lists safely.
2834 * This check does not provide a hard guarantee, it
2835 * just helps avoid dangerous mistakes.
c5d79adb 2836 */
fd01b88c 2837 struct mddev *mddev;
c5d79adb 2838 int overlap = 0;
159ec1fc 2839 struct list_head *tmp;
c5d79adb 2840
8b1afc3d 2841 rcu_read_lock();
29ac4aa3 2842 for_each_mddev(mddev, tmp) {
3cb03002 2843 struct md_rdev *rdev2;
c5d79adb 2844
dafb20fa 2845 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 2846 if (rdev->bdev == rdev2->bdev &&
2847 rdev != rdev2 &&
2848 overlaps(rdev->data_offset, rdev->sectors,
2849 rdev2->data_offset,
2850 rdev2->sectors)) {
c5d79adb
N		 2851 overlap = 1;
2852 break;
2853 }
c5d79adb
N		 2854 if (overlap) {
2855 mddev_put(mddev);
2856 break;
2857 }
2858 }
8b1afc3d 2859 rcu_read_unlock();
c5d79adb
N		 2860 if (overlap) {
2861 /* Someone else could have slipped in a size
2862 * change here, but doing so is just silly.
dd8ac336 2863 * We put oldsectors back because we *know* it is
c5d79adb
N		 2864 * safe, and trust userspace not to race with
2865 * itself
2866 */
dd8ac336 2867 rdev->sectors = oldsectors;
c5d79adb
N		 2868 return -EBUSY;
2869 }
2870 }
83303b61
N		 2871 return len;
2872}
2873
2874static struct rdev_sysfs_entry rdev_size =
80ca3a44 2875__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2876
3cb03002 2877static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 2878{
2879 unsigned long long recovery_start = rdev->recovery_offset;
2880
2881 if (test_bit(In_sync, &rdev->flags) ||
2882 recovery_start == MaxSector)
2883 return sprintf(page, "none\n");
2884
2885 return sprintf(page, "%llu\n", recovery_start);
2886}
2887
3cb03002 2888static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 2889{
2890 unsigned long long recovery_start;
2891
2892 if (cmd_match(buf, "none"))
2893 recovery_start = MaxSector;
b29bebd6 2894 else if (kstrtoull(buf, 10, &recovery_start))
06e3c817
DW		 2895 return -EINVAL;
2896
2897 if (rdev->mddev->pers &&
2898 rdev->raid_disk >= 0)
2899 return -EBUSY;
2900
2901 rdev->recovery_offset = recovery_start;
2902 if (recovery_start == MaxSector)
2903 set_bit(In_sync, &rdev->flags);
2904 else
2905 clear_bit(In_sync, &rdev->flags);
2906 return len;
2907}
2908
2909static struct rdev_sysfs_entry rdev_recovery_start =
2910__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2911
16c791a5
N		 2912static ssize_t
2913badblocks_show(struct badblocks *bb, char *page, int unack);
2914static ssize_t
2915badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2916
3cb03002 2917static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2918{
2919 return badblocks_show(&rdev->badblocks, page, 0);
2920}
3cb03002 2921static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 2922{
de393cde
N		 2923 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2924 /* Maybe that ack was all we needed */
2925 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2926 wake_up(&rdev->blocked_wait);
2927 return rv;
16c791a5
N		 2928}
2929static struct rdev_sysfs_entry rdev_bad_blocks =
2930__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2931
3cb03002 2932static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2933{
2934 return badblocks_show(&rdev->badblocks, page, 1);
2935}
3cb03002 2936static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 2937{
2938 return badblocks_store(&rdev->badblocks, page, len, 1);
2939}
2940static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2941__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2942
86e6ffdd
N		 2943static struct attribute *rdev_default_attrs[] = {
2944 &rdev_state.attr,
4dbcdc75 2945 &rdev_errors.attr,
014236d2 2946 &rdev_slot.attr,
93c8cad0 2947 &rdev_offset.attr,
c6563a8c 2948 &rdev_new_offset.attr,
83303b61 2949 &rdev_size.attr,
06e3c817 2950 &rdev_recovery_start.attr,
16c791a5
N		 2951 &rdev_bad_blocks.attr,
2952 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 2953 NULL,
2954};
2955static ssize_t
2956rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2957{
2958 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 2959 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
86e6ffdd
N		 2960
2961 if (!entry->show)
2962 return -EIO;
758bfc8a
N		 2963 if (!rdev->mddev)
2964 return -EBUSY;
2965 return entry->show(rdev, page);
86e6ffdd
N		 2966}
2967
2968static ssize_t
2969rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2970 const char *page, size_t length)
2971{
2972 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 2973 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 2974 ssize_t rv;
fd01b88c 2975 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 2976
2977 if (!entry->store)
2978 return -EIO;
67463acb
N		 2979 if (!capable(CAP_SYS_ADMIN))
2980 return -EACCES;
27c529bb 2981 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 2982 if (!rv) {
27c529bb
N		 2983 if (rdev->mddev == NULL)
2984 rv = -EBUSY;
2985 else
2986 rv = entry->store(rdev, page, length);
6a51830e 2987 mddev_unlock(mddev);
ca388059
N		 2988 }
2989 return rv;
86e6ffdd
N		 2990}
2991
2992static void rdev_free(struct kobject *ko)
2993{
3cb03002 2994 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 2995 kfree(rdev);
2996}
52cf25d0 2997static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 2998 .show = rdev_attr_show,
2999 .store = rdev_attr_store,
3000};
3001static struct kobj_type rdev_ktype = {
3002 .release = rdev_free,
3003 .sysfs_ops = &rdev_sysfs_ops,
3004 .default_attrs = rdev_default_attrs,
3005};
3006
3cb03002 3007int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3008{
3009 rdev->desc_nr = -1;
3010 rdev->saved_raid_disk = -1;
3011 rdev->raid_disk = -1;
3012 rdev->flags = 0;
3013 rdev->data_offset = 0;
c6563a8c 3014 rdev->new_data_offset = 0;
e8bb9a83
N		 3015 rdev->sb_events = 0;
3016 rdev->last_read_error.tv_sec = 0;
3017 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3018 rdev->sb_loaded = 0;
3019 rdev->bb_page = NULL;
e8bb9a83
N		 3020 atomic_set(&rdev->nr_pending, 0);
3021 atomic_set(&rdev->read_errors, 0);
3022 atomic_set(&rdev->corrected_errors, 0);
3023
3024 INIT_LIST_HEAD(&rdev->same_set);
3025 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3026
3027 /* Add space to store bad block list.
3028 * This reserves the space even on arrays where it cannot
3029 * be used - I wonder if that matters
3030 */
3031 rdev->badblocks.count = 0;
486adf72 3032 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
2230dfe4
N		 3033 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3034 seqlock_init(&rdev->badblocks.lock);
3035 if (rdev->badblocks.page == NULL)
3036 return -ENOMEM;
3037
3038 return 0;
e8bb9a83
N		 3039}
3040EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3041/*
3042 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3043 *
3044 * mark the device faulty if:
3045 *
3046 * - the device is nonexistent (zero size)
3047 * - the device has no valid superblock
3048 *
3049 * a faulty rdev _never_ has rdev->sb set.
3050 */
3cb03002 3051static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3052{
3053 char b[BDEVNAME_SIZE];
3054 int err;
3cb03002 3055 struct md_rdev *rdev;
1da177e4
LT		 3056 sector_t size;
3057
9ffae0cf 3058 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3059 if (!rdev) {
3060 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3061 return ERR_PTR(-ENOMEM);
3062 }
1da177e4 3063
2230dfe4
N		 3064 err = md_rdev_init(rdev);
3065 if (err)
3066 goto abort_free;
3067 err = alloc_disk_sb(rdev);
3068 if (err)
1da177e4
LT		 3069 goto abort_free;
3070
c5d79adb 3071 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3072 if (err)
3073 goto abort_free;
3074
f9cb074b 3075 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3076
77304d2a 3077 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4 3078 if (!size) {
f72ffdd6 3079 printk(KERN_WARNING
1da177e4
LT		 3080 "md: %s has zero or unknown size, marking faulty!\n",
3081 bdevname(rdev->bdev,b));
3082 err = -EINVAL;
3083 goto abort_free;
3084 }
3085
3086 if (super_format >= 0) {
3087 err = super_types[super_format].
3088 load_super(rdev, NULL, super_minor);
3089 if (err == -EINVAL) {
df968c4e
N		 3090 printk(KERN_WARNING
3091 "md: %s does not have a valid v%d.%d "
3092 "superblock, not importing!\n",
3093 bdevname(rdev->bdev,b),
3094 super_format, super_minor);
1da177e4
LT		 3095 goto abort_free;
3096 }
3097 if (err < 0) {
f72ffdd6 3098 printk(KERN_WARNING
1da177e4
LT		 3099 "md: could not read %s's sb, not importing!\n",
3100 bdevname(rdev->bdev,b));
3101 goto abort_free;
3102 }
3103 }
6bfe0b49 3104
1da177e4
LT		 3105 return rdev;
3106
3107abort_free:
2699b672
N		 3108 if (rdev->bdev)
3109 unlock_rdev(rdev);
545c8795 3110 md_rdev_clear(rdev);
1da177e4
LT		 3111 kfree(rdev);
3112 return ERR_PTR(err);
3113}
3114
3115/*
3116 * Check a full RAID array for plausibility
3117 */
3118
f72ffdd6 3119static void analyze_sbs(struct mddev *mddev)
1da177e4
LT		 3120{
3121 int i;
3cb03002 3122 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3123 char b[BDEVNAME_SIZE];
3124
3125 freshest = NULL;
dafb20fa 3126 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3127 switch (super_types[mddev->major_version].
3128 load_super(rdev, freshest, mddev->minor_version)) {
3129 case 1:
3130 freshest = rdev;
3131 break;
3132 case 0:
3133 break;
3134 default:
3135 printk( KERN_ERR \
3136 "md: fatal superblock inconsistency in %s"
f72ffdd6 3137 " -- removing from array\n",
1da177e4 3138 bdevname(rdev->bdev,b));
fb56dfef 3139 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3140 }
3141
1da177e4
LT		 3142 super_types[mddev->major_version].
3143 validate_super(mddev, freshest);
3144
3145 i = 0;
dafb20fa 3146 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3147 if (mddev->max_disks &&
3148 (rdev->desc_nr >= mddev->max_disks ||
3149 i > mddev->max_disks)) {
de01dfad
N		 3150 printk(KERN_WARNING
3151 "md: %s: %s: only %d devices permitted\n",
3152 mdname(mddev), bdevname(rdev->bdev, b),
3153 mddev->max_disks);
fb56dfef 3154 md_kick_rdev_from_array(rdev);
de01dfad
N		 3155 continue;
3156 }
1aee41f6 3157 if (rdev != freshest) {
1da177e4
LT		 3158 if (super_types[mddev->major_version].
3159 validate_super(mddev, rdev)) {
3160 printk(KERN_WARNING "md: kicking non-fresh %s"
3161 " from array!\n",
3162 bdevname(rdev->bdev,b));
fb56dfef 3163 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3164 continue;
3165 }
1aee41f6
GR		 3166 /* No device should have a Candidate flag
3167 * when reading devices
3168 */
3169 if (test_bit(Candidate, &rdev->flags)) {
3170 pr_info("md: kicking Cluster Candidate %s from array!\n",
3171 bdevname(rdev->bdev, b));
fb56dfef 3172 md_kick_rdev_from_array(rdev);
1aee41f6
GR		 3173 }
3174 }
1da177e4
LT		 3175 if (mddev->level == LEVEL_MULTIPATH) {
3176 rdev->desc_nr = i++;
3177 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3178 set_bit(In_sync, &rdev->flags);
5e5e3e78 3179 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3180 rdev->raid_disk = -1;
3181 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3182 }
3183 }
1da177e4
LT		 3184}
3185
72e02075
N		 3186/* Read a fixed-point number.
3187 * Numbers in sysfs attributes should be in "standard" units where
3188 * possible, so time should be in seconds.
f72ffdd6 3189 * However we internally use a a much smaller unit such as
72e02075
N		 3190 * milliseconds or jiffies.
3191 * This function takes a decimal number with a possible fractional
3192 * component, and produces an integer which is the result of
3193 * multiplying that number by 10^'scale'.
3194 * all without any floating-point arithmetic.
3195 */
3196int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3197{
3198 unsigned long result = 0;
3199 long decimals = -1;
3200 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3201 if (*cp == '.')
3202 decimals = 0;
3203 else if (decimals < scale) {
3204 unsigned int value;
3205 value = *cp - '0';
3206 result = result * 10 + value;
3207 if (decimals >= 0)
3208 decimals++;
3209 }
3210 cp++;
3211 }
3212 if (*cp == '\n')
3213 cp++;
3214 if (*cp)
3215 return -EINVAL;
3216 if (decimals < 0)
3217 decimals = 0;
3218 while (decimals < scale) {
3219 result *= 10;
3220 decimals ++;
3221 }
3222 *res = result;
3223 return 0;
3224}
3225
19052c0e
N		 3226static void md_safemode_timeout(unsigned long data);
3227
16f17b39 3228static ssize_t
fd01b88c 3229safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3230{
3231 int msec = (mddev->safemode_delay*1000)/HZ;
3232 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3233}
3234static ssize_t
fd01b88c 3235safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3236{
16f17b39 3237 unsigned long msec;
97ce0a7f 3238
72e02075 3239 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3240 return -EINVAL;
16f17b39
N		 3241 if (msec == 0)
3242 mddev->safemode_delay = 0;
3243 else {
19052c0e 3244 unsigned long old_delay = mddev->safemode_delay;
1b30e66f
N		 3245 unsigned long new_delay = (msec*HZ)/1000;
3246
3247 if (new_delay == 0)
3248 new_delay = 1;
3249 mddev->safemode_delay = new_delay;
3250 if (new_delay < old_delay || old_delay == 0)
3251 mod_timer(&mddev->safemode_timer, jiffies+1);
16f17b39
N		 3252 }
3253 return len;
3254}
3255static struct md_sysfs_entry md_safe_delay =
80ca3a44 3256__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3257
eae1701f 3258static ssize_t
fd01b88c 3259level_show(struct mddev *mddev, char *page)
eae1701f 3260{
36d091f4
N		 3261 struct md_personality *p;
3262 int ret;
3263 spin_lock(&mddev->lock);
3264 p = mddev->pers;
d9d166c2 3265 if (p)
36d091f4 3266 ret = sprintf(page, "%s\n", p->name);
d9d166c2 3267 else if (mddev->clevel[0])
36d091f4 3268 ret = sprintf(page, "%s\n", mddev->clevel);
d9d166c2 3269 else if (mddev->level != LEVEL_NONE)
36d091f4 3270 ret = sprintf(page, "%d\n", mddev->level);
d9d166c2 3271 else
36d091f4
N		 3272 ret = 0;
3273 spin_unlock(&mddev->lock);
3274 return ret;
eae1701f
N		 3275}
3276
d9d166c2 3277static ssize_t
fd01b88c 3278level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3279{
f2859af6 3280 char clevel[16];
6791875e
N		 3281 ssize_t rv;
3282 size_t slen = len;
db721d32 3283 struct md_personality *pers, *oldpers;
f2859af6 3284 long level;
db721d32 3285 void *priv, *oldpriv;
3cb03002 3286 struct md_rdev *rdev;
245f46c2 3287
6791875e
N		 3288 if (slen == 0 || slen >= sizeof(clevel))
3289 return -EINVAL;
3290
3291 rv = mddev_lock(mddev);
3292 if (rv)
3293 return rv;
3294
245f46c2 3295 if (mddev->pers == NULL) {
6791875e
N		 3296 strncpy(mddev->clevel, buf, slen);
3297 if (mddev->clevel[slen-1] == '\n')
3298 slen--;
3299 mddev->clevel[slen] = 0;
245f46c2 3300 mddev->level = LEVEL_NONE;
6791875e
N		 3301 rv = len;
3302 goto out_unlock;
245f46c2 3303 }
6791875e 3304 rv = -EROFS;
bd8839e0 3305 if (mddev->ro)
6791875e 3306 goto out_unlock;
245f46c2
N		 3307
3308 /* request to change the personality. Need to ensure:
3309 * - array is not engaged in resync/recovery/reshape
3310 * - old personality can be suspended
3311 * - new personality will access other array.
3312 */
3313
6791875e 3314 rv = -EBUSY;
bb4f1e9d 3315 if (mddev->sync_thread ||
f851b60d 3316 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
bb4f1e9d
N		 3317 mddev->reshape_position != MaxSector ||
3318 mddev->sysfs_active)
6791875e 3319 goto out_unlock;
245f46c2 3320
6791875e 3321 rv = -EINVAL;
245f46c2
N		 3322 if (!mddev->pers->quiesce) {
3323 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3324 mdname(mddev), mddev->pers->name);
6791875e 3325 goto out_unlock;
245f46c2
N		 3326 }
3327
3328 /* Now find the new personality */
6791875e
N		 3329 strncpy(clevel, buf, slen);
3330 if (clevel[slen-1] == '\n')
3331 slen--;
3332 clevel[slen] = 0;
b29bebd6 3333 if (kstrtol(clevel, 10, &level))
f2859af6 3334 level = LEVEL_NONE;
245f46c2 3335
f2859af6
DW		 3336 if (request_module("md-%s", clevel) != 0)
3337 request_module("md-level-%s", clevel);
245f46c2 3338 spin_lock(&pers_lock);
f2859af6 3339 pers = find_pers(level, clevel);
245f46c2
N		 3340 if (!pers || !try_module_get(pers->owner)) {
3341 spin_unlock(&pers_lock);
f2859af6 3342 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
6791875e
N		 3343 rv = -EINVAL;
3344 goto out_unlock;
245f46c2
N		 3345 }
3346 spin_unlock(&pers_lock);
3347
3348 if (pers == mddev->pers) {
3349 /* Nothing to do! */
3350 module_put(pers->owner);
6791875e
N		 3351 rv = len;
3352 goto out_unlock;
245f46c2
N		 3353 }
3354 if (!pers->takeover) {
3355 module_put(pers->owner);
3356 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3357 mdname(mddev), clevel);
6791875e
N		 3358 rv = -EINVAL;
3359 goto out_unlock;
245f46c2
N		 3360 }
3361
dafb20fa 3362 rdev_for_each(rdev, mddev)
e93f68a1
N		 3363 rdev->new_raid_disk = rdev->raid_disk;
3364
245f46c2
N		 3365 /* ->takeover must set new_* and/or delta_disks
3366 * if it succeeds, and may set them when it fails.
3367 */
3368 priv = pers->takeover(mddev);
3369 if (IS_ERR(priv)) {
3370 mddev->new_level = mddev->level;
3371 mddev->new_layout = mddev->layout;
664e7c41 3372 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3373 mddev->raid_disks -= mddev->delta_disks;
3374 mddev->delta_disks = 0;
2c810cdd 3375 mddev->reshape_backwards = 0;
245f46c2
N		 3376 module_put(pers->owner);
3377 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3378 mdname(mddev), clevel);
6791875e
N		 3379 rv = PTR_ERR(priv);
3380 goto out_unlock;
245f46c2
N		 3381 }
3382
3383 /* Looks like we have a winner */
3384 mddev_suspend(mddev);
5aa61f42 3385 mddev_detach(mddev);
36d091f4
N		 3386
3387 spin_lock(&mddev->lock);
db721d32
N		 3388 oldpers = mddev->pers;
3389 oldpriv = mddev->private;
3390 mddev->pers = pers;
3391 mddev->private = priv;
3392 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3393 mddev->level = mddev->new_level;
3394 mddev->layout = mddev->new_layout;
3395 mddev->chunk_sectors = mddev->new_chunk_sectors;
3396 mddev->delta_disks = 0;
3397 mddev->reshape_backwards = 0;
3398 mddev->degraded = 0;
36d091f4 3399 spin_unlock(&mddev->lock);
db721d32
N		 3400
3401 if (oldpers->sync_request == NULL &&
3402 mddev->external) {
3403 /* We are converting from a no-redundancy array
3404 * to a redundancy array and metadata is managed
3405 * externally so we need to be sure that writes
3406 * won't block due to a need to transition
3407 * clean->dirty
3408 * until external management is started.
3409 */
3410 mddev->in_sync = 0;
3411 mddev->safemode_delay = 0;
3412 mddev->safemode = 0;
3413 }
f72ffdd6 3414
db721d32
N		 3415 oldpers->free(mddev, oldpriv);
3416
3417 if (oldpers->sync_request == NULL &&
a64c876f
N		 3418 pers->sync_request != NULL) {
3419 /* need to add the md_redundancy_group */
3420 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3421 printk(KERN_WARNING
3422 "md: cannot register extra attributes for %s\n",
3423 mdname(mddev));
388975cc 3424 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
f72ffdd6 3425 }
db721d32 3426 if (oldpers->sync_request != NULL &&
a64c876f
N		 3427 pers->sync_request == NULL) {
3428 /* need to remove the md_redundancy_group */
3429 if (mddev->to_remove == NULL)
3430 mddev->to_remove = &md_redundancy_group;
3431 }
3432
dafb20fa 3433 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3434 if (rdev->raid_disk < 0)
3435 continue;
bf2cb0da 3436 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3437 rdev->new_raid_disk = -1;
3438 if (rdev->new_raid_disk == rdev->raid_disk)
3439 continue;
36fad858 3440 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3441 }
dafb20fa 3442 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3443 if (rdev->raid_disk < 0)
3444 continue;
3445 if (rdev->new_raid_disk == rdev->raid_disk)
3446 continue;
3447 rdev->raid_disk = rdev->new_raid_disk;
3448 if (rdev->raid_disk < 0)
3a981b03 3449 clear_bit(In_sync, &rdev->flags);
e93f68a1 3450 else {
36fad858
NK		 3451 if (sysfs_link_rdev(mddev, rdev))
3452 printk(KERN_WARNING "md: cannot register rd%d"
3453 " for %s after level change\n",
3454 rdev->raid_disk, mdname(mddev));
3a981b03 3455 }
e93f68a1
N		 3456 }
3457
db721d32 3458 if (pers->sync_request == NULL) {
9af204cf
TM		 3459 /* this is now an array without redundancy, so
3460 * it must always be in_sync
3461 */
3462 mddev->in_sync = 1;
3463 del_timer_sync(&mddev->safemode_timer);
3464 }
02e5f5c0 3465 blk_set_stacking_limits(&mddev->queue->limits);
245f46c2 3466 pers->run(mddev);
245f46c2 3467 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3468 mddev_resume(mddev);
830778a1
N		 3469 if (!mddev->thread)
3470 md_update_sb(mddev, 1);
5cac7861 3471 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3472 md_new_event(mddev);
6791875e
N		 3473 rv = len;
3474out_unlock:
3475 mddev_unlock(mddev);
d9d166c2
N		 3476 return rv;
3477}
3478
3479static struct md_sysfs_entry md_level =
80ca3a44 3480__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3481
d4dbd025 3482static ssize_t
fd01b88c 3483layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3484{
3485 /* just a number, not meaningful for all levels */
08a02ecd
N		 3486 if (mddev->reshape_position != MaxSector &&
3487 mddev->layout != mddev->new_layout)
3488 return sprintf(page, "%d (%d)\n",
3489 mddev->new_layout, mddev->layout);
d4dbd025
N		 3490 return sprintf(page, "%d\n", mddev->layout);
3491}
3492
3493static ssize_t
fd01b88c 3494layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N		 3495{
3496 char *e;
3497 unsigned long n = simple_strtoul(buf, &e, 10);
6791875e 3498 int err;
d4dbd025
N		 3499
3500 if (!*buf || (*e && *e != '\n'))
3501 return -EINVAL;
6791875e
N		 3502 err = mddev_lock(mddev);
3503 if (err)
3504 return err;
d4dbd025 3505
b3546035 3506 if (mddev->pers) {
50ac168a 3507 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3508 err = -EBUSY;
3509 else if (mddev->ro)
3510 err = -EROFS;
3511 else {
3512 mddev->new_layout = n;
3513 err = mddev->pers->check_reshape(mddev);
3514 if (err)
3515 mddev->new_layout = mddev->layout;
597a711b 3516 }
b3546035 3517 } else {
08a02ecd 3518 mddev->new_layout = n;
b3546035
N		 3519 if (mddev->reshape_position == MaxSector)
3520 mddev->layout = n;
3521 }
6791875e
N		 3522 mddev_unlock(mddev);
3523 return err ?: len;
d4dbd025
N		 3524}
3525static struct md_sysfs_entry md_layout =
80ca3a44 3526__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025 3527
eae1701f 3528static ssize_t
fd01b88c 3529raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3530{
bb636547
N		 3531 if (mddev->raid_disks == 0)
3532 return 0;
08a02ecd
N		 3533 if (mddev->reshape_position != MaxSector &&
3534 mddev->delta_disks != 0)
3535 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3536 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3537 return sprintf(page, "%d\n", mddev->raid_disks);
3538}
3539
fd01b88c 3540static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3541
3542static ssize_t
fd01b88c 3543raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3544{
da943b99 3545 char *e;
6791875e 3546 int err;
da943b99
N		 3547 unsigned long n = simple_strtoul(buf, &e, 10);
3548
3549 if (!*buf || (*e && *e != '\n'))
3550 return -EINVAL;
3551
6791875e
N		 3552 err = mddev_lock(mddev);
3553 if (err)
3554 return err;
da943b99 3555 if (mddev->pers)
6791875e 3556 err = update_raid_disks(mddev, n);
08a02ecd 3557 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3558 struct md_rdev *rdev;
08a02ecd 3559 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c 3560
6791875e 3561 err = -EINVAL;
c6563a8c
N		 3562 rdev_for_each(rdev, mddev) {
3563 if (olddisks < n &&
3564 rdev->data_offset < rdev->new_data_offset)
6791875e 3565 goto out_unlock;
c6563a8c
N		 3566 if (olddisks > n &&
3567 rdev->data_offset > rdev->new_data_offset)
6791875e 3568 goto out_unlock;
c6563a8c 3569 }
6791875e 3570 err = 0;
08a02ecd
N		 3571 mddev->delta_disks = n - olddisks;
3572 mddev->raid_disks = n;
2c810cdd 3573 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3574 } else
da943b99 3575 mddev->raid_disks = n;
6791875e
N		 3576out_unlock:
3577 mddev_unlock(mddev);
3578 return err ? err : len;
da943b99
N		 3579}
3580static struct md_sysfs_entry md_raid_disks =
80ca3a44 3581__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3582
3b34380a 3583static ssize_t
fd01b88c 3584chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3585{
08a02ecd 3586 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3587 mddev->chunk_sectors != mddev->new_chunk_sectors)
3588 return sprintf(page, "%d (%d)\n",
3589 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3590 mddev->chunk_sectors << 9);
3591 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3592}
3593
3594static ssize_t
fd01b88c 3595chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3596{
6791875e 3597 int err;
3b34380a
N		 3598 char *e;
3599 unsigned long n = simple_strtoul(buf, &e, 10);
3600
3b34380a
N		 3601 if (!*buf || (*e && *e != '\n'))
3602 return -EINVAL;
3603
6791875e
N		 3604 err = mddev_lock(mddev);
3605 if (err)
3606 return err;
b3546035 3607 if (mddev->pers) {
50ac168a 3608 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3609 err = -EBUSY;
3610 else if (mddev->ro)
3611 err = -EROFS;
3612 else {
3613 mddev->new_chunk_sectors = n >> 9;
3614 err = mddev->pers->check_reshape(mddev);
3615 if (err)
3616 mddev->new_chunk_sectors = mddev->chunk_sectors;
597a711b 3617 }
b3546035 3618 } else {
664e7c41 3619 mddev->new_chunk_sectors = n >> 9;
b3546035 3620 if (mddev->reshape_position == MaxSector)
9d8f0363 3621 mddev->chunk_sectors = n >> 9;
b3546035 3622 }
6791875e
N		 3623 mddev_unlock(mddev);
3624 return err ?: len;
3b34380a
N		 3625}
3626static struct md_sysfs_entry md_chunk_size =
80ca3a44 3627__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3628
a94213b1 3629static ssize_t
fd01b88c 3630resync_start_show(struct mddev *mddev, char *page)
a94213b1 3631{
d1a7c503
N		 3632 if (mddev->recovery_cp == MaxSector)
3633 return sprintf(page, "none\n");
a94213b1
N		 3634 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3635}
3636
3637static ssize_t
fd01b88c 3638resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3639{
6791875e 3640 int err;
a94213b1
N		 3641 char *e;
3642 unsigned long long n = simple_strtoull(buf, &e, 10);
3643
6791875e
N		 3644 err = mddev_lock(mddev);
3645 if (err)
3646 return err;
b098636c 3647 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6791875e
N		 3648 err = -EBUSY;
3649 else if (cmd_match(buf, "none"))
06e3c817
DW		 3650 n = MaxSector;
3651 else if (!*buf || (*e && *e != '\n'))
6791875e 3652 err = -EINVAL;
a94213b1 3653
6791875e
N		 3654 if (!err) {
3655 mddev->recovery_cp = n;
3656 if (mddev->pers)
3657 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3658 }
3659 mddev_unlock(mddev);
3660 return err ?: len;
a94213b1
N		 3661}
3662static struct md_sysfs_entry md_resync_start =
80ca3a44 3663__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3664
9e653b63
N		 3665/*
3666 * The array state can be:
3667 *
3668 * clear
3669 * No devices, no size, no level
3670 * Equivalent to STOP_ARRAY ioctl
3671 * inactive
3672 * May have some settings, but array is not active
3673 * all IO results in error
3674 * When written, doesn't tear down array, but just stops it
3675 * suspended (not supported yet)
3676 * All IO requests will block. The array can be reconfigured.
910d8cb3 3677 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3678 * readonly
3679 * no resync can happen. no superblocks get written.
3680 * write requests fail
3681 * read-auto
3682 * like readonly, but behaves like 'clean' on a write request.
3683 *
3684 * clean - no pending writes, but otherwise active.
3685 * When written to inactive array, starts without resync
3686 * If a write request arrives then
3687 * if metadata is known, mark 'dirty' and switch to 'active'.
3688 * if not known, block and switch to write-pending
3689 * If written to an active array that has pending writes, then fails.
3690 * active
3691 * fully active: IO and resync can be happening.
3692 * When written to inactive array, starts with resync
3693 *
3694 * write-pending
3695 * clean, but writes are blocked waiting for 'active' to be written.
3696 *
3697 * active-idle
3698 * like active, but no writes have been seen for a while (100msec).
3699 *
3700 */
3701enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3702 write_pending, active_idle, bad_word};
05381954 3703static char *array_states[] = {
9e653b63
N		 3704 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3705 "write-pending", "active-idle", NULL };
3706
3707static int match_word(const char *word, char **list)
3708{
3709 int n;
3710 for (n=0; list[n]; n++)
3711 if (cmd_match(word, list[n]))
3712 break;
3713 return n;
3714}
3715
3716static ssize_t
fd01b88c 3717array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3718{
3719 enum array_state st = inactive;
3720
3721 if (mddev->pers)
3722 switch(mddev->ro) {
3723 case 1:
3724 st = readonly;
3725 break;
3726 case 2:
3727 st = read_auto;
3728 break;
3729 case 0:
3730 if (mddev->in_sync)
3731 st = clean;
070dc6dd 3732 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3733 st = write_pending;
9e653b63
N		 3734 else if (mddev->safemode)
3735 st = active_idle;
3736 else
3737 st = active;
3738 }
3739 else {
3740 if (list_empty(&mddev->disks) &&
3741 mddev->raid_disks == 0 &&
58c0fed4 3742 mddev->dev_sectors == 0)
9e653b63
N		 3743 st = clear;
3744 else
3745 st = inactive;
3746 }
3747 return sprintf(page, "%s\n", array_states[st]);
3748}
3749
f72ffdd6
N		 3750static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3751static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3752static int do_md_run(struct mddev *mddev);
fd01b88c 3753static int restart_array(struct mddev *mddev);
9e653b63
N		 3754
3755static ssize_t
fd01b88c 3756array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63 3757{
6791875e 3758 int err;
9e653b63 3759 enum array_state st = match_word(buf, array_states);
6791875e
N		 3760
3761 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3762 /* don't take reconfig_mutex when toggling between
3763 * clean and active
3764 */
3765 spin_lock(&mddev->lock);
3766 if (st == active) {
3767 restart_array(mddev);
3768 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3769 wake_up(&mddev->sb_wait);
3770 err = 0;
3771 } else /* st == clean */ {
3772 restart_array(mddev);
3773 if (atomic_read(&mddev->writes_pending) == 0) {
3774 if (mddev->in_sync == 0) {
3775 mddev->in_sync = 1;
3776 if (mddev->safemode == 1)
3777 mddev->safemode = 0;
3778 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3779 }
3780 err = 0;
3781 } else
3782 err = -EBUSY;
3783 }
3784 spin_unlock(&mddev->lock);
3785 return err;
3786 }
3787 err = mddev_lock(mddev);
3788 if (err)
3789 return err;
3790 err = -EINVAL;
9e653b63
N		 3791 switch(st) {
3792 case bad_word:
3793 break;
3794 case clear:
3795 /* stopping an active array */
a05b7ea0 3796 err = do_md_stop(mddev, 0, NULL);
9e653b63
N		 3797 break;
3798 case inactive:
3799 /* stopping an active array */
90cf195d 3800 if (mddev->pers)
a05b7ea0 3801 err = do_md_stop(mddev, 2, NULL);
90cf195d 3802 else
e691063a 3803 err = 0; /* already inactive */
9e653b63
N		 3804 break;
3805 case suspended:
3806 break; /* not supported yet */
3807 case readonly:
3808 if (mddev->pers)
a05b7ea0 3809 err = md_set_readonly(mddev, NULL);
9e653b63
N		 3810 else {
3811 mddev->ro = 1;
648b629e 3812 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3813 err = do_md_run(mddev);
3814 }
3815 break;
3816 case read_auto:
9e653b63 3817 if (mddev->pers) {
80268ee9 3818 if (mddev->ro == 0)
a05b7ea0 3819 err = md_set_readonly(mddev, NULL);
80268ee9 3820 else if (mddev->ro == 1)
648b629e
N		 3821 err = restart_array(mddev);
3822 if (err == 0) {
3823 mddev->ro = 2;
3824 set_disk_ro(mddev->gendisk, 0);
3825 }
9e653b63
N		 3826 } else {
3827 mddev->ro = 2;
3828 err = do_md_run(mddev);
3829 }
3830 break;
3831 case clean:
3832 if (mddev->pers) {
3833 restart_array(mddev);
85572d7c 3834 spin_lock(&mddev->lock);
9e653b63 3835 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3836 if (mddev->in_sync == 0) {
3837 mddev->in_sync = 1;
31a59e34
N		 3838 if (mddev->safemode == 1)
3839 mddev->safemode = 0;
070dc6dd 3840 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3841 }
3842 err = 0;
3843 } else
3844 err = -EBUSY;
85572d7c 3845 spin_unlock(&mddev->lock);
5bf29597
N		 3846 } else
3847 err = -EINVAL;
9e653b63
N		 3848 break;
3849 case active:
3850 if (mddev->pers) {
3851 restart_array(mddev);
070dc6dd 3852 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3853 wake_up(&mddev->sb_wait);
3854 err = 0;
3855 } else {
3856 mddev->ro = 0;
648b629e 3857 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3858 err = do_md_run(mddev);
3859 }
3860 break;
3861 case write_pending:
3862 case active_idle:
3863 /* these cannot be set */
3864 break;
3865 }
6791875e
N		 3866
3867 if (!err) {
1d23f178
N		 3868 if (mddev->hold_active == UNTIL_IOCTL)
3869 mddev->hold_active = 0;
00bcb4ac 3870 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 3871 }
6791875e
N		 3872 mddev_unlock(mddev);
3873 return err ?: len;
9e653b63 3874}
80ca3a44
N		 3875static struct md_sysfs_entry md_array_state =
3876__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3877
1e50915f 3878static ssize_t
fd01b88c 3879max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3880 return sprintf(page, "%d\n",
3881 atomic_read(&mddev->max_corr_read_errors));
3882}
3883
3884static ssize_t
fd01b88c 3885max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB		 3886{
3887 char *e;
3888 unsigned long n = simple_strtoul(buf, &e, 10);
3889
3890 if (*buf && (*e == 0 || *e == '\n')) {
3891 atomic_set(&mddev->max_corr_read_errors, n);
3892 return len;
3893 }
3894 return -EINVAL;
3895}
3896
3897static struct md_sysfs_entry max_corr_read_errors =
3898__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3899 max_corrected_read_errors_store);
3900
6d7ff738 3901static ssize_t
fd01b88c 3902null_show(struct mddev *mddev, char *page)
6d7ff738
N		 3903{
3904 return -EINVAL;
3905}
3906
3907static ssize_t
fd01b88c 3908new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 3909{
3910 /* buf must be %d:%d\n? giving major and minor numbers */
3911 /* The new device is added to the array.
3912 * If the array has a persistent superblock, we read the
3913 * superblock to initialise info and check validity.
3914 * Otherwise, only checking done is that in bind_rdev_to_array,
3915 * which mainly checks size.
3916 */
3917 char *e;
3918 int major = simple_strtoul(buf, &e, 10);
3919 int minor;
3920 dev_t dev;
3cb03002 3921 struct md_rdev *rdev;
6d7ff738
N		 3922 int err;
3923
3924 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3925 return -EINVAL;
3926 minor = simple_strtoul(e+1, &e, 10);
3927 if (*e && *e != '\n')
3928 return -EINVAL;
3929 dev = MKDEV(major, minor);
3930 if (major != MAJOR(dev) ||
3931 minor != MINOR(dev))
3932 return -EOVERFLOW;
3933
6791875e
N		 3934 flush_workqueue(md_misc_wq);
3935
3936 err = mddev_lock(mddev);
3937 if (err)
3938 return err;
6d7ff738
N		 3939 if (mddev->persistent) {
3940 rdev = md_import_device(dev, mddev->major_version,
3941 mddev->minor_version);
3942 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 3943 struct md_rdev *rdev0
3944 = list_entry(mddev->disks.next,
3945 struct md_rdev, same_set);
6d7ff738
N		 3946 err = super_types[mddev->major_version]
3947 .load_super(rdev, rdev0, mddev->minor_version);
3948 if (err < 0)
3949 goto out;
3950 }
c5d79adb
N		 3951 } else if (mddev->external)
3952 rdev = md_import_device(dev, -2, -1);
3953 else
6d7ff738
N		 3954 rdev = md_import_device(dev, -1, -1);
3955
3956 if (IS_ERR(rdev))
3957 return PTR_ERR(rdev);
3958 err = bind_rdev_to_array(rdev, mddev);
3959 out:
3960 if (err)
3961 export_rdev(rdev);
6791875e 3962 mddev_unlock(mddev);
6d7ff738
N		 3963 return err ? err : len;
3964}
3965
3966static struct md_sysfs_entry md_new_device =
80ca3a44 3967__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3968
9b1d1dac 3969static ssize_t
fd01b88c 3970bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 3971{
3972 char *end;
3973 unsigned long chunk, end_chunk;
6791875e 3974 int err;
9b1d1dac 3975
6791875e
N		 3976 err = mddev_lock(mddev);
3977 if (err)
3978 return err;
9b1d1dac
PC		 3979 if (!mddev->bitmap)
3980 goto out;
3981 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3982 while (*buf) {
3983 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3984 if (buf == end) break;
3985 if (*end == '-') { /* range */
3986 buf = end + 1;
3987 end_chunk = simple_strtoul(buf, &end, 0);
3988 if (buf == end) break;
3989 }
3990 if (*end && !isspace(*end)) break;
3991 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 3992 buf = skip_spaces(end);
9b1d1dac
PC		 3993 }
3994 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3995out:
6791875e 3996 mddev_unlock(mddev);
9b1d1dac
PC		 3997 return len;
3998}
3999
4000static struct md_sysfs_entry md_bitmap =
4001__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4002
a35b0d69 4003static ssize_t
fd01b88c 4004size_show(struct mddev *mddev, char *page)
a35b0d69 4005{
58c0fed4
AN		 4006 return sprintf(page, "%llu\n",
4007 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 4008}
4009
fd01b88c 4010static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 4011
4012static ssize_t
fd01b88c 4013size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 4014{
4015 /* If array is inactive, we can reduce the component size, but
4016 * not increase it (except from 0).
4017 * If array is active, we can try an on-line resize
4018 */
b522adcd
DW		 4019 sector_t sectors;
4020 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4021
58c0fed4
AN		 4022 if (err < 0)
4023 return err;
6791875e
N		 4024 err = mddev_lock(mddev);
4025 if (err)
4026 return err;
a35b0d69 4027 if (mddev->pers) {
293467aa
GR		 4028 if (mddev_is_clustered(mddev))
4029 md_cluster_ops->metadata_update_start(mddev);
58c0fed4 4030 err = update_size(mddev, sectors);
850b2b42 4031 md_update_sb(mddev, 1);
293467aa
GR		 4032 if (mddev_is_clustered(mddev))
4033 md_cluster_ops->metadata_update_finish(mddev);
a35b0d69 4034 } else {
58c0fed4
AN		 4035 if (mddev->dev_sectors == 0 ||
4036 mddev->dev_sectors > sectors)
4037 mddev->dev_sectors = sectors;
a35b0d69
N		 4038 else
4039 err = -ENOSPC;
4040 }
6791875e 4041 mddev_unlock(mddev);
a35b0d69
N		 4042 return err ? err : len;
4043}
4044
4045static struct md_sysfs_entry md_size =
80ca3a44 4046__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4047
83f0d77a 4048/* Metadata version.
e691063a
N		 4049 * This is one of
4050 * 'none' for arrays with no metadata (good luck...)
4051 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4052 * or N.M for internally known formats
4053 */
4054static ssize_t
fd01b88c 4055metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4056{
4057 if (mddev->persistent)
4058 return sprintf(page, "%d.%d\n",
4059 mddev->major_version, mddev->minor_version);
e691063a
N		 4060 else if (mddev->external)
4061 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4062 else
4063 return sprintf(page, "none\n");
4064}
4065
4066static ssize_t
fd01b88c 4067metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4068{
4069 int major, minor;
4070 char *e;
6791875e 4071 int err;
ea43ddd8
N		 4072 /* Changing the details of 'external' metadata is
4073 * always permitted. Otherwise there must be
4074 * no devices attached to the array.
4075 */
6791875e
N		 4076
4077 err = mddev_lock(mddev);
4078 if (err)
4079 return err;
4080 err = -EBUSY;
ea43ddd8
N		 4081 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4082 ;
4083 else if (!list_empty(&mddev->disks))
6791875e 4084 goto out_unlock;
8bb93aac 4085
6791875e 4086 err = 0;
8bb93aac
N		 4087 if (cmd_match(buf, "none")) {
4088 mddev->persistent = 0;
e691063a
N		 4089 mddev->external = 0;
4090 mddev->major_version = 0;
4091 mddev->minor_version = 90;
6791875e 4092 goto out_unlock;
e691063a
N		 4093 }
4094 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4095 size_t namelen = len-9;
e691063a
N		 4096 if (namelen >= sizeof(mddev->metadata_type))
4097 namelen = sizeof(mddev->metadata_type)-1;
4098 strncpy(mddev->metadata_type, buf+9, namelen);
4099 mddev->metadata_type[namelen] = 0;
4100 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4101 mddev->metadata_type[--namelen] = 0;
4102 mddev->persistent = 0;
4103 mddev->external = 1;
8bb93aac
N		 4104 mddev->major_version = 0;
4105 mddev->minor_version = 90;
6791875e 4106 goto out_unlock;
8bb93aac
N		 4107 }
4108 major = simple_strtoul(buf, &e, 10);
6791875e 4109 err = -EINVAL;
8bb93aac 4110 if (e==buf || *e != '.')
6791875e 4111 goto out_unlock;
8bb93aac
N		 4112 buf = e+1;
4113 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4114 if (e==buf || (*e && *e != '\n') )
6791875e
N		 4115 goto out_unlock;
4116 err = -ENOENT;
50511da3 4117 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
6791875e 4118 goto out_unlock;
8bb93aac
N		 4119 mddev->major_version = major;
4120 mddev->minor_version = minor;
4121 mddev->persistent = 1;
e691063a 4122 mddev->external = 0;
6791875e
N		 4123 err = 0;
4124out_unlock:
4125 mddev_unlock(mddev);
4126 return err ?: len;
8bb93aac
N		 4127}
4128
4129static struct md_sysfs_entry md_metadata =
80ca3a44 4130__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4131
24dd469d 4132static ssize_t
fd01b88c 4133action_show(struct mddev *mddev, char *page)
24dd469d 4134{
7eec314d 4135 char *type = "idle";
b7b17c9b
N		 4136 unsigned long recovery = mddev->recovery;
4137 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
b6a9ce68 4138 type = "frozen";
b7b17c9b
N		 4139 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4140 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4141 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
ccfcc3c1 4142 type = "reshape";
b7b17c9b
N		 4143 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4144 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
24dd469d 4145 type = "resync";
b7b17c9b 4146 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
24dd469d
N		 4147 type = "check";
4148 else
4149 type = "repair";
b7b17c9b 4150 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
24dd469d
N		 4151 type = "recover";
4152 }
4153 return sprintf(page, "%s\n", type);
4154}
4155
4156static ssize_t
fd01b88c 4157action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4158{
7eec314d
N		 4159 if (!mddev->pers || !mddev->pers->sync_request)
4160 return -EINVAL;
4161
b6a9ce68
N		 4162 if (cmd_match(page, "frozen"))
4163 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4164 else
4165 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4166
4167 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
f851b60d 4168 flush_workqueue(md_misc_wq);
7eec314d
N		 4169 if (mddev->sync_thread) {
4170 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6791875e
N		 4171 if (mddev_lock(mddev) == 0) {
4172 md_reap_sync_thread(mddev);
4173 mddev_unlock(mddev);
4174 }
7eec314d 4175 }
03c902e1
N		 4176 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4177 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4178 return -EBUSY;
72a23c21
NB		 4179 else if (cmd_match(page, "resync"))
4180 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4181 else if (cmd_match(page, "recover")) {
4182 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4183 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4184 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4185 int err;
4186 if (mddev->pers->start_reshape == NULL)
4187 return -EINVAL;
6791875e
N		 4188 err = mddev_lock(mddev);
4189 if (!err) {
4190 err = mddev->pers->start_reshape(mddev);
4191 mddev_unlock(mddev);
4192 }
16484bf5
N		 4193 if (err)
4194 return err;
a99ac971 4195 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4196 } else {
bce74dac 4197 if (cmd_match(page, "check"))
7eec314d 4198 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4199 else if (!cmd_match(page, "repair"))
7eec314d
N		 4200 return -EINVAL;
4201 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4202 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4203 }
48c26ddc
N		 4204 if (mddev->ro == 2) {
4205 /* A write to sync_action is enough to justify
4206 * canceling read-auto mode
4207 */
4208 mddev->ro = 0;
4209 md_wakeup_thread(mddev->sync_thread);
4210 }
03c902e1 4211 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4212 md_wakeup_thread(mddev->thread);
00bcb4ac 4213 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4214 return len;
4215}
4216
c4a39551
JB		 4217static struct md_sysfs_entry md_scan_mode =
4218__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4219
4220static ssize_t
4221last_sync_action_show(struct mddev *mddev, char *page)
4222{
4223 return sprintf(page, "%s\n", mddev->last_sync_action);
4224}
4225
4226static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4227
9d88883e 4228static ssize_t
fd01b88c 4229mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4230{
4231 return sprintf(page, "%llu\n",
7f7583d4
JM		 4232 (unsigned long long)
4233 atomic64_read(&mddev->resync_mismatches));
9d88883e
N		 4234}
4235
80ca3a44 4236static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4237
88202a0c 4238static ssize_t
fd01b88c 4239sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4240{
4241 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4242 mddev->sync_speed_min ? "local": "system");
4243}
4244
4245static ssize_t
fd01b88c 4246sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4247{
4248 int min;
4249 char *e;
4250 if (strncmp(buf, "system", 6)==0) {
4251 mddev->sync_speed_min = 0;
4252 return len;
4253 }
4254 min = simple_strtoul(buf, &e, 10);
4255 if (buf == e || (*e && *e != '\n') || min <= 0)
4256 return -EINVAL;
4257 mddev->sync_speed_min = min;
4258 return len;
4259}
4260
4261static struct md_sysfs_entry md_sync_min =
4262__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4263
4264static ssize_t
fd01b88c 4265sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4266{
4267 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4268 mddev->sync_speed_max ? "local": "system");
4269}
4270
4271static ssize_t
fd01b88c 4272sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4273{
4274 int max;
4275 char *e;
4276 if (strncmp(buf, "system", 6)==0) {
4277 mddev->sync_speed_max = 0;
4278 return len;
4279 }
4280 max = simple_strtoul(buf, &e, 10);
4281 if (buf == e || (*e && *e != '\n') || max <= 0)
4282 return -EINVAL;
4283 mddev->sync_speed_max = max;
4284 return len;
4285}
4286
4287static struct md_sysfs_entry md_sync_max =
4288__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4289
d7f3d291 4290static ssize_t
fd01b88c 4291degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4292{
4293 return sprintf(page, "%d\n", mddev->degraded);
4294}
4295static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4296
90b08710 4297static ssize_t
fd01b88c 4298sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4299{
4300 return sprintf(page, "%d\n", mddev->parallel_resync);
4301}
4302
4303static ssize_t
fd01b88c 4304sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4305{
4306 long n;
4307
b29bebd6 4308 if (kstrtol(buf, 10, &n))
90b08710
BS		 4309 return -EINVAL;
4310
4311 if (n != 0 && n != 1)
4312 return -EINVAL;
4313
4314 mddev->parallel_resync = n;
4315
4316 if (mddev->sync_thread)
4317 wake_up(&resync_wait);
4318
4319 return len;
4320}
4321
4322/* force parallel resync, even with shared block devices */
4323static struct md_sysfs_entry md_sync_force_parallel =
4324__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4325 sync_force_parallel_show, sync_force_parallel_store);
4326
88202a0c 4327static ssize_t
fd01b88c 4328sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4329{
4330 unsigned long resync, dt, db;
d1a7c503
N		 4331 if (mddev->curr_resync == 0)
4332 return sprintf(page, "none\n");
9687a60c
AN		 4333 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4334 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4335 if (!dt) dt++;
9687a60c
AN		 4336 db = resync - mddev->resync_mark_cnt;
4337 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4338}
4339
80ca3a44 4340static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4341
4342static ssize_t
fd01b88c 4343sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4344{
13ae864b 4345 unsigned long long max_sectors, resync;
88202a0c 4346
acb180b0
N		 4347 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4348 return sprintf(page, "none\n");
4349
72f36d59
N		 4350 if (mddev->curr_resync == 1 ||
4351 mddev->curr_resync == 2)
4352 return sprintf(page, "delayed\n");
4353
c804cdec
N		 4354 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4355 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 4356 max_sectors = mddev->resync_max_sectors;
88202a0c 4357 else
58c0fed4 4358 max_sectors = mddev->dev_sectors;
88202a0c 4359
acb180b0 4360 resync = mddev->curr_resync_completed;
13ae864b 4361 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4362}
4363
80ca3a44 4364static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 4365
5e96ee65 4366static ssize_t
fd01b88c 4367min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4368{
4369 return sprintf(page, "%llu\n",
4370 (unsigned long long)mddev->resync_min);
4371}
4372static ssize_t
fd01b88c 4373min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4374{
4375 unsigned long long min;
23da422b
N		 4376 int err;
4377 int chunk;
4378
b29bebd6 4379 if (kstrtoull(buf, 10, &min))
5e96ee65 4380 return -EINVAL;
23da422b
N		 4381
4382 spin_lock(&mddev->lock);
4383 err = -EINVAL;
5e96ee65 4384 if (min > mddev->resync_max)
23da422b
N		 4385 goto out_unlock;
4386
4387 err = -EBUSY;
5e96ee65 4388 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4389 goto out_unlock;
5e96ee65
NB		 4390
4391 /* Must be a multiple of chunk_size */
23da422b
N		 4392 chunk = mddev->chunk_sectors;
4393 if (chunk) {
2ac06c33 4394 sector_t temp = min;
23da422b
N		 4395
4396 err = -EINVAL;
4397 if (sector_div(temp, chunk))
4398 goto out_unlock;
5e96ee65
NB		 4399 }
4400 mddev->resync_min = min;
23da422b 4401 err = 0;
5e96ee65 4402
23da422b
N		 4403out_unlock:
4404 spin_unlock(&mddev->lock);
4405 return err ?: len;
5e96ee65
NB		 4406}
4407
4408static struct md_sysfs_entry md_min_sync =
4409__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4410
c6207277 4411static ssize_t
fd01b88c 4412max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4413{
4414 if (mddev->resync_max == MaxSector)
4415 return sprintf(page, "max\n");
4416 else
4417 return sprintf(page, "%llu\n",
4418 (unsigned long long)mddev->resync_max);
4419}
4420static ssize_t
fd01b88c 4421max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277 4422{
23da422b
N		 4423 int err;
4424 spin_lock(&mddev->lock);
c6207277
N		 4425 if (strncmp(buf, "max", 3) == 0)
4426 mddev->resync_max = MaxSector;
4427 else {
5e96ee65 4428 unsigned long long max;
23da422b
N		 4429 int chunk;
4430
4431 err = -EINVAL;
b29bebd6 4432 if (kstrtoull(buf, 10, &max))
23da422b 4433 goto out_unlock;
5e96ee65 4434 if (max < mddev->resync_min)
23da422b
N		 4435 goto out_unlock;
4436
4437 err = -EBUSY;
c6207277 4438 if (max < mddev->resync_max &&
4d484a4a 4439 mddev->ro == 0 &&
c6207277 4440 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4441 goto out_unlock;
c6207277
N		 4442
4443 /* Must be a multiple of chunk_size */
23da422b
N		 4444 chunk = mddev->chunk_sectors;
4445 if (chunk) {
2ac06c33 4446 sector_t temp = max;
23da422b
N		 4447
4448 err = -EINVAL;
4449 if (sector_div(temp, chunk))
4450 goto out_unlock;
c6207277
N		 4451 }
4452 mddev->resync_max = max;
4453 }
4454 wake_up(&mddev->recovery_wait);
23da422b
N		 4455 err = 0;
4456out_unlock:
4457 spin_unlock(&mddev->lock);
4458 return err ?: len;
c6207277
N		 4459}
4460
4461static struct md_sysfs_entry md_max_sync =
4462__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4463
e464eafd 4464static ssize_t
fd01b88c 4465suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4466{
4467 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4468}
4469
4470static ssize_t
fd01b88c 4471suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4472{
4473 char *e;
4474 unsigned long long new = simple_strtoull(buf, &e, 10);
6791875e
N		 4475 unsigned long long old;
4476 int err;
e464eafd 4477
e464eafd
N		 4478 if (buf == e || (*e && *e != '\n'))
4479 return -EINVAL;
23ddff37 4480
6791875e
N		 4481 err = mddev_lock(mddev);
4482 if (err)
4483 return err;
4484 err = -EINVAL;
4485 if (mddev->pers == NULL ||
4486 mddev->pers->quiesce == NULL)
4487 goto unlock;
4488 old = mddev->suspend_lo;
23ddff37
N		 4489 mddev->suspend_lo = new;
4490 if (new >= old)
4491 /* Shrinking suspended region */
e464eafd 4492 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4493 else {
4494 /* Expanding suspended region - need to wait */
4495 mddev->pers->quiesce(mddev, 1);
4496 mddev->pers->quiesce(mddev, 0);
4497 }
6791875e
N		 4498 err = 0;
4499unlock:
4500 mddev_unlock(mddev);
4501 return err ?: len;
e464eafd
N		 4502}
4503static struct md_sysfs_entry md_suspend_lo =
4504__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4505
e464eafd 4506static ssize_t
fd01b88c 4507suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4508{
4509 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4510}
4511
4512static ssize_t
fd01b88c 4513suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4514{
4515 char *e;
4516 unsigned long long new = simple_strtoull(buf, &e, 10);
6791875e
N		 4517 unsigned long long old;
4518 int err;
e464eafd 4519
e464eafd
N		 4520 if (buf == e || (*e && *e != '\n'))
4521 return -EINVAL;
23ddff37 4522
6791875e
N		 4523 err = mddev_lock(mddev);
4524 if (err)
4525 return err;
4526 err = -EINVAL;
4527 if (mddev->pers == NULL ||
4528 mddev->pers->quiesce == NULL)
4529 goto unlock;
4530 old = mddev->suspend_hi;
23ddff37
N		 4531 mddev->suspend_hi = new;
4532 if (new <= old)
4533 /* Shrinking suspended region */
4534 mddev->pers->quiesce(mddev, 2);
4535 else {
4536 /* Expanding suspended region - need to wait */
e464eafd
N		 4537 mddev->pers->quiesce(mddev, 1);
4538 mddev->pers->quiesce(mddev, 0);
23ddff37 4539 }
6791875e
N		 4540 err = 0;
4541unlock:
4542 mddev_unlock(mddev);
4543 return err ?: len;
e464eafd
N		 4544}
4545static struct md_sysfs_entry md_suspend_hi =
4546__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4547
08a02ecd 4548static ssize_t
fd01b88c 4549reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4550{
4551 if (mddev->reshape_position != MaxSector)
4552 return sprintf(page, "%llu\n",
4553 (unsigned long long)mddev->reshape_position);
4554 strcpy(page, "none\n");
4555 return 5;
4556}
4557
4558static ssize_t
fd01b88c 4559reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd 4560{
c6563a8c 4561 struct md_rdev *rdev;
08a02ecd 4562 char *e;
6791875e 4563 int err;
08a02ecd 4564 unsigned long long new = simple_strtoull(buf, &e, 10);
6791875e 4565
08a02ecd
N		 4566 if (buf == e || (*e && *e != '\n'))
4567 return -EINVAL;
6791875e
N		 4568 err = mddev_lock(mddev);
4569 if (err)
4570 return err;
4571 err = -EBUSY;
4572 if (mddev->pers)
4573 goto unlock;
08a02ecd
N		 4574 mddev->reshape_position = new;
4575 mddev->delta_disks = 0;
2c810cdd 4576 mddev->reshape_backwards = 0;
08a02ecd
N		 4577 mddev->new_level = mddev->level;
4578 mddev->new_layout = mddev->layout;
664e7c41 4579 mddev->new_chunk_sectors = mddev->chunk_sectors;
c6563a8c
N		 4580 rdev_for_each(rdev, mddev)
4581 rdev->new_data_offset = rdev->data_offset;
6791875e
N		 4582 err = 0;
4583unlock:
4584 mddev_unlock(mddev);
4585 return err ?: len;
08a02ecd
N		 4586}
4587
4588static struct md_sysfs_entry md_reshape_position =
4589__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4590 reshape_position_store);
4591
2c810cdd
N		 4592static ssize_t
4593reshape_direction_show(struct mddev *mddev, char *page)
4594{
4595 return sprintf(page, "%s\n",
4596 mddev->reshape_backwards ? "backwards" : "forwards");
4597}
4598
4599static ssize_t
4600reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4601{
4602 int backwards = 0;
6791875e
N		 4603 int err;
4604
2c810cdd
N		 4605 if (cmd_match(buf, "forwards"))
4606 backwards = 0;
4607 else if (cmd_match(buf, "backwards"))
4608 backwards = 1;
4609 else
4610 return -EINVAL;
4611 if (mddev->reshape_backwards == backwards)
4612 return len;
4613
6791875e
N		 4614 err = mddev_lock(mddev);
4615 if (err)
4616 return err;
2c810cdd
N		 4617 /* check if we are allowed to change */
4618 if (mddev->delta_disks)
6791875e
N		 4619 err = -EBUSY;
4620 else if (mddev->persistent &&
2c810cdd 4621 mddev->major_version == 0)
6791875e
N		 4622 err = -EINVAL;
4623 else
4624 mddev->reshape_backwards = backwards;
4625 mddev_unlock(mddev);
4626 return err ?: len;
2c810cdd
N		 4627}
4628
4629static struct md_sysfs_entry md_reshape_direction =
4630__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4631 reshape_direction_store);
4632
b522adcd 4633static ssize_t
fd01b88c 4634array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4635{
4636 if (mddev->external_size)
4637 return sprintf(page, "%llu\n",
4638 (unsigned long long)mddev->array_sectors/2);
4639 else
4640 return sprintf(page, "default\n");
4641}
4642
4643static ssize_t
fd01b88c 4644array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4645{
4646 sector_t sectors;
6791875e
N		 4647 int err;
4648
4649 err = mddev_lock(mddev);
4650 if (err)
4651 return err;
b522adcd
DW		 4652
4653 if (strncmp(buf, "default", 7) == 0) {
4654 if (mddev->pers)
4655 sectors = mddev->pers->size(mddev, 0, 0);
4656 else
4657 sectors = mddev->array_sectors;
4658
4659 mddev->external_size = 0;
4660 } else {
4661 if (strict_blocks_to_sectors(buf, &sectors) < 0)
6791875e
N		 4662 err = -EINVAL;
4663 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4664 err = -E2BIG;
4665 else
4666 mddev->external_size = 1;
b522adcd
DW		 4667 }
4668
6791875e
N		 4669 if (!err) {
4670 mddev->array_sectors = sectors;
4671 if (mddev->pers) {
4672 set_capacity(mddev->gendisk, mddev->array_sectors);
4673 revalidate_disk(mddev->gendisk);
4674 }
cbe6ef1d 4675 }
6791875e
N		 4676 mddev_unlock(mddev);
4677 return err ?: len;
b522adcd
DW		 4678}
4679
4680static struct md_sysfs_entry md_array_size =
4681__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4682 array_size_store);
e464eafd 4683
eae1701f
N		 4684static struct attribute *md_default_attrs[] = {
4685 &md_level.attr,
d4dbd025 4686 &md_layout.attr,
eae1701f 4687 &md_raid_disks.attr,
3b34380a 4688 &md_chunk_size.attr,
a35b0d69 4689 &md_size.attr,
a94213b1 4690 &md_resync_start.attr,
8bb93aac 4691 &md_metadata.attr,
6d7ff738 4692 &md_new_device.attr,
16f17b39 4693 &md_safe_delay.attr,
9e653b63 4694 &md_array_state.attr,
08a02ecd 4695 &md_reshape_position.attr,
2c810cdd 4696 &md_reshape_direction.attr,
b522adcd 4697 &md_array_size.attr,
1e50915f 4698 &max_corr_read_errors.attr,
411036fa
N		 4699 NULL,
4700};
4701
4702static struct attribute *md_redundancy_attrs[] = {
24dd469d 4703 &md_scan_mode.attr,
c4a39551 4704 &md_last_scan_mode.attr,
9d88883e 4705 &md_mismatches.attr,
88202a0c
N		 4706 &md_sync_min.attr,
4707 &md_sync_max.attr,
4708 &md_sync_speed.attr,
90b08710 4709 &md_sync_force_parallel.attr,
88202a0c 4710 &md_sync_completed.attr,
5e96ee65 4711 &md_min_sync.attr,
c6207277 4712 &md_max_sync.attr,
e464eafd
N		 4713 &md_suspend_lo.attr,
4714 &md_suspend_hi.attr,
9b1d1dac 4715 &md_bitmap.attr,
d7f3d291 4716 &md_degraded.attr,
eae1701f
N		 4717 NULL,
4718};
411036fa
N		 4719static struct attribute_group md_redundancy_group = {
4720 .name = NULL,
4721 .attrs = md_redundancy_attrs,
4722};
4723
eae1701f
N		 4724static ssize_t
4725md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4726{
4727 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4728 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4729 ssize_t rv;
eae1701f
N		 4730
4731 if (!entry->show)
4732 return -EIO;
af8a2434
N		 4733 spin_lock(&all_mddevs_lock);
4734 if (list_empty(&mddev->all_mddevs)) {
4735 spin_unlock(&all_mddevs_lock);
4736 return -EBUSY;
4737 }
4738 mddev_get(mddev);
4739 spin_unlock(&all_mddevs_lock);
4740
b7b17c9b 4741 rv = entry->show(mddev, page);
af8a2434 4742 mddev_put(mddev);
96de1e66 4743 return rv;
eae1701f
N		 4744}
4745
4746static ssize_t
4747md_attr_store(struct kobject *kobj, struct attribute *attr,
4748 const char *page, size_t length)
4749{
4750 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4751 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4752 ssize_t rv;
eae1701f
N		 4753
4754 if (!entry->store)
4755 return -EIO;
67463acb
N		 4756 if (!capable(CAP_SYS_ADMIN))
4757 return -EACCES;
af8a2434
N		 4758 spin_lock(&all_mddevs_lock);
4759 if (list_empty(&mddev->all_mddevs)) {
4760 spin_unlock(&all_mddevs_lock);
4761 return -EBUSY;
4762 }
4763 mddev_get(mddev);
4764 spin_unlock(&all_mddevs_lock);
6791875e 4765 rv = entry->store(mddev, page, length);
af8a2434 4766 mddev_put(mddev);
96de1e66 4767 return rv;
eae1701f
N		 4768}
4769
4770static void md_free(struct kobject *ko)
4771{
fd01b88c 4772 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4773
4774 if (mddev->sysfs_state)
4775 sysfs_put(mddev->sysfs_state);
4776
4777 if (mddev->gendisk) {
4778 del_gendisk(mddev->gendisk);
4779 put_disk(mddev->gendisk);
4780 }
4781 if (mddev->queue)
4782 blk_cleanup_queue(mddev->queue);
4783
eae1701f
N		 4784 kfree(mddev);
4785}
4786
52cf25d0 4787static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4788 .show = md_attr_show,
4789 .store = md_attr_store,
4790};
4791static struct kobj_type md_ktype = {
4792 .release = md_free,
4793 .sysfs_ops = &md_sysfs_ops,
4794 .default_attrs = md_default_attrs,
4795};
4796
1da177e4
LT		 4797int mdp_major = 0;
4798
5fd3a17e
DW		 4799static void mddev_delayed_delete(struct work_struct *ws)
4800{
fd01b88c 4801 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4802
43a70507 4803 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4804 kobject_del(&mddev->kobj);
4805 kobject_put(&mddev->kobj);
4806}
4807
efeb53c0 4808static int md_alloc(dev_t dev, char *name)
1da177e4 4809{
48c9c27b 4810 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4811 struct mddev *mddev = mddev_find(dev);
1da177e4 4812 struct gendisk *disk;
efeb53c0
N		 4813 int partitioned;
4814 int shift;
4815 int unit;
3830c62f 4816 int error;
1da177e4
LT		 4817
4818 if (!mddev)
efeb53c0
N		 4819 return -ENODEV;
4820
4821 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4822 shift = partitioned ? MdpMinorShift : 0;
4823 unit = MINOR(mddev->unit) >> shift;
1da177e4 4824
e804ac78
TH		 4825 /* wait for any previous instance of this device to be
4826 * completely removed (mddev_delayed_delete).
d3374825 4827 */
e804ac78 4828 flush_workqueue(md_misc_wq);
d3374825 4829
48c9c27b 4830 mutex_lock(&disks_mutex);
0909dc44
N		 4831 error = -EEXIST;
4832 if (mddev->gendisk)
4833 goto abort;
efeb53c0
N		 4834
4835 if (name) {
4836 /* Need to ensure that 'name' is not a duplicate.
4837 */
fd01b88c 4838 struct mddev *mddev2;
efeb53c0
N		 4839 spin_lock(&all_mddevs_lock);
4840
4841 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4842 if (mddev2->gendisk &&
4843 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4844 spin_unlock(&all_mddevs_lock);
0909dc44 4845 goto abort;
efeb53c0
N		 4846 }
4847 spin_unlock(&all_mddevs_lock);
1da177e4 4848 }
8b765398 4849
0909dc44 4850 error = -ENOMEM;
8b765398 4851 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4852 if (!mddev->queue)
4853 goto abort;
409c57f3
N		 4854 mddev->queue->queuedata = mddev;
4855
409c57f3 4856 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 4857 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 4858
1da177e4
LT		 4859 disk = alloc_disk(1 << shift);
4860 if (!disk) {
8b765398
N		 4861 blk_cleanup_queue(mddev->queue);
4862 mddev->queue = NULL;
0909dc44 4863 goto abort;
1da177e4 4864 }
efeb53c0 4865 disk->major = MAJOR(mddev->unit);
1da177e4 4866 disk->first_minor = unit << shift;
efeb53c0
N		 4867 if (name)
4868 strcpy(disk->disk_name, name);
4869 else if (partitioned)
1da177e4 4870 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4871 else
1da177e4 4872 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4873 disk->fops = &md_fops;
4874 disk->private_data = mddev;
4875 disk->queue = mddev->queue;
b0140891 4876 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4877 /* Allow extended partitions. This makes the
d3374825 4878 * 'mdp' device redundant, but we can't really
92850bbd
N		 4879 * remove it now.
4880 */
4881 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4882 mddev->gendisk = disk;
b0140891
N		 4883 /* As soon as we call add_disk(), another thread could get
4884 * through to md_open, so make sure it doesn't get too far
4885 */
4886 mutex_lock(&mddev->open_mutex);
4887 add_disk(disk);
4888
ed9e1982
TH		 4889 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4890 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4891 if (error) {
4892 /* This isn't possible, but as kobject_init_and_add is marked
4893 * __must_check, we must do something with the result
4894 */
5e55e2f5
N		 4895 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4896 disk->disk_name);
0909dc44
N		 4897 error = 0;
4898 }
00bcb4ac
N		 4899 if (mddev->kobj.sd &&
4900 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4901 printk(KERN_DEBUG "pointless warning\n");
b0140891 4902 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4903 abort:
4904 mutex_unlock(&disks_mutex);
00bcb4ac 4905 if (!error && mddev->kobj.sd) {
3830c62f 4906 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4907 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4908 }
d3374825 4909 mddev_put(mddev);
0909dc44 4910 return error;
efeb53c0
N		 4911}
4912
4913static struct kobject *md_probe(dev_t dev, int *part, void *data)
4914{
4915 md_alloc(dev, NULL);
1da177e4
LT		 4916 return NULL;
4917}
4918
efeb53c0
N		 4919static int add_named_array(const char *val, struct kernel_param *kp)
4920{
4921 /* val must be "md_*" where * is not all digits.
4922 * We allocate an array with a large free minor number, and
4923 * set the name to val. val must not already be an active name.
4924 */
4925 int len = strlen(val);
4926 char buf[DISK_NAME_LEN];
4927
4928 while (len && val[len-1] == '\n')
4929 len--;
4930 if (len >= DISK_NAME_LEN)
4931 return -E2BIG;
4932 strlcpy(buf, val, len+1);
4933 if (strncmp(buf, "md_", 3) != 0)
4934 return -EINVAL;
4935 return md_alloc(0, buf);
4936}
4937
1da177e4
LT		 4938static void md_safemode_timeout(unsigned long data)
4939{
fd01b88c 4940 struct mddev *mddev = (struct mddev *) data;
1da177e4 4941
0fd62b86
NB		 4942 if (!atomic_read(&mddev->writes_pending)) {
4943 mddev->safemode = 1;
4944 if (mddev->external)
00bcb4ac 4945 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4946 }
1da177e4
LT		 4947 md_wakeup_thread(mddev->thread);
4948}
4949
6ff8d8ec 4950static int start_dirty_degraded;
1da177e4 4951
fd01b88c 4952int md_run(struct mddev *mddev)
1da177e4 4953{
2604b703 4954 int err;
3cb03002 4955 struct md_rdev *rdev;
84fc4b56 4956 struct md_personality *pers;
1da177e4 4957
a757e64c
N		 4958 if (list_empty(&mddev->disks))
4959 /* cannot run an array with no devices.. */
1da177e4 4960 return -EINVAL;
1da177e4
LT		 4961
4962 if (mddev->pers)
4963 return -EBUSY;
bb4f1e9d
N		 4964 /* Cannot run until previous stop completes properly */
4965 if (mddev->sysfs_active)
4966 return -EBUSY;
b6eb127d 4967
1da177e4
LT		 4968 /*
4969 * Analyze all RAID superblock(s)
4970 */
1ec4a939
N		 4971 if (!mddev->raid_disks) {
4972 if (!mddev->persistent)
4973 return -EINVAL;
a757e64c 4974 analyze_sbs(mddev);
1ec4a939 4975 }
1da177e4 4976
d9d166c2
N		 4977 if (mddev->level != LEVEL_NONE)
4978 request_module("md-level-%d", mddev->level);
4979 else if (mddev->clevel[0])
4980 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4981
4982 /*
4983 * Drop all container device buffers, from now on
4984 * the only valid external interface is through the md
4985 * device.
1da177e4 4986 */
dafb20fa 4987 rdev_for_each(rdev, mddev) {
b2d444d7 4988 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4989 continue;
4990 sync_blockdev(rdev->bdev);
f98393a6 4991 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4992
4993 /* perform some consistency tests on the device.
4994 * We don't want the data to overlap the metadata,
58c0fed4 4995 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4996 */
a6ff7e08
JB		 4997 if (rdev->meta_bdev) {
4998 /* Nothing to check */;
4999 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 5000 if (mddev->dev_sectors &&
5001 rdev->data_offset + mddev->dev_sectors
0f420358 5002 > rdev->sb_start) {
f0d76d70
N		 5003 printk("md: %s: data overlaps metadata\n",
5004 mdname(mddev));
5005 return -EINVAL;
5006 }
5007 } else {
0f420358 5008 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 5009 > rdev->data_offset) {
5010 printk("md: %s: metadata overlaps data\n",
5011 mdname(mddev));
5012 return -EINVAL;
5013 }
5014 }
00bcb4ac 5015 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 5016 }
5017
a167f663 5018 if (mddev->bio_set == NULL)
395c72a7 5019 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
a167f663 5020
1da177e4 5021 spin_lock(&pers_lock);
d9d166c2 5022 pers = find_pers(mddev->level, mddev->clevel);
2604b703 5023 if (!pers || !try_module_get(pers->owner)) {
1da177e4 5024 spin_unlock(&pers_lock);
d9d166c2
N		 5025 if (mddev->level != LEVEL_NONE)
5026 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5027 mddev->level);
5028 else
5029 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5030 mddev->clevel);
1da177e4
LT		 5031 return -EINVAL;
5032 }
1da177e4 5033 spin_unlock(&pers_lock);
34817e8c
N		 5034 if (mddev->level != pers->level) {
5035 mddev->level = pers->level;
5036 mddev->new_level = pers->level;
5037 }
d9d166c2 5038 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 5039
f6705578 5040 if (mddev->reshape_position != MaxSector &&
63c70c4f 5041 pers->start_reshape == NULL) {
f6705578 5042 /* This personality cannot handle reshaping... */
f6705578
N		 5043 module_put(pers->owner);
5044 return -EINVAL;
5045 }
5046
7dd5e7c3
N		 5047 if (pers->sync_request) {
5048 /* Warn if this is a potentially silly
5049 * configuration.
5050 */
5051 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5052 struct md_rdev *rdev2;
7dd5e7c3 5053 int warned = 0;
159ec1fc 5054
dafb20fa
N		 5055 rdev_for_each(rdev, mddev)
5056 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N		 5057 if (rdev < rdev2 &&
5058 rdev->bdev->bd_contains ==
5059 rdev2->bdev->bd_contains) {
5060 printk(KERN_WARNING
5061 "%s: WARNING: %s appears to be"
5062 " on the same physical disk as"
5063 " %s.\n",
5064 mdname(mddev),
5065 bdevname(rdev->bdev,b),
5066 bdevname(rdev2->bdev,b2));
5067 warned = 1;
5068 }
5069 }
159ec1fc 5070
7dd5e7c3
N		 5071 if (warned)
5072 printk(KERN_WARNING
5073 "True protection against single-disk"
5074 " failure might be compromised.\n");
5075 }
5076
657390d2 5077 mddev->recovery = 0;
58c0fed4
AN		 5078 /* may be over-ridden by personality */
5079 mddev->resync_max_sectors = mddev->dev_sectors;
5080
6ff8d8ec 5081 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5082
0f9552b5 5083 if (start_readonly && mddev->ro == 0)
f91de92e
N		 5084 mddev->ro = 2; /* read-only, but switch on first write */
5085
36d091f4 5086 err = pers->run(mddev);
13e53df3
AN		 5087 if (err)
5088 printk(KERN_ERR "md: pers->run() failed ...\n");
36d091f4 5089 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
b522adcd
DW		 5090 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5091 " but 'external_size' not in effect?\n", __func__);
5092 printk(KERN_ERR
5093 "md: invalid array_size %llu > default size %llu\n",
5094 (unsigned long long)mddev->array_sectors / 2,
36d091f4 5095 (unsigned long long)pers->size(mddev, 0, 0) / 2);
b522adcd 5096 err = -EINVAL;
b522adcd 5097 }
36d091f4 5098 if (err == 0 && pers->sync_request &&
ef99bf48 5099 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
f9209a32
GR		 5100 struct bitmap *bitmap;
5101
5102 bitmap = bitmap_create(mddev, -1);
5103 if (IS_ERR(bitmap)) {
5104 err = PTR_ERR(bitmap);
b15c2e57
N		 5105 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5106 mdname(mddev), err);
f9209a32
GR		 5107 } else
5108 mddev->bitmap = bitmap;
5109
b15c2e57 5110 }
1da177e4 5111 if (err) {
5aa61f42 5112 mddev_detach(mddev);
36d091f4
N		 5113 pers->free(mddev, mddev->private);
5114 module_put(pers->owner);
32a7627c
N		 5115 bitmap_destroy(mddev);
5116 return err;
1da177e4 5117 }
5c675f83
N		 5118 if (mddev->queue) {
5119 mddev->queue->backing_dev_info.congested_data = mddev;
5120 mddev->queue->backing_dev_info.congested_fn = md_congested;
64590f45 5121 blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec);
5c675f83 5122 }
36d091f4 5123 if (pers->sync_request) {
00bcb4ac
N		 5124 if (mddev->kobj.sd &&
5125 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 5126 printk(KERN_WARNING
5127 "md: cannot register extra attributes for %s\n",
5128 mdname(mddev));
00bcb4ac 5129 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5130 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 5131 mddev->ro = 0;
5132
f72ffdd6 5133 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 5134 atomic_set(&mddev->max_corr_read_errors,
5135 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 5136 mddev->safemode = 0;
5137 mddev->safemode_timer.function = md_safemode_timeout;
5138 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 5139 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5140 mddev->in_sync = 1;
0ca69886 5141 smp_wmb();
36d091f4
N		 5142 spin_lock(&mddev->lock);
5143 mddev->pers = pers;
0ca69886 5144 mddev->ready = 1;
36d091f4 5145 spin_unlock(&mddev->lock);
dafb20fa 5146 rdev_for_each(rdev, mddev)
36fad858
NK		 5147 if (rdev->raid_disk >= 0)
5148 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 5149 /* failure here is OK */;
f72ffdd6 5150
1da177e4 5151 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f72ffdd6 5152
7a0a5355 5153 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 5154 md_update_sb(mddev, 0);
1da177e4 5155
d7603b7e 5156 md_new_event(mddev);
00bcb4ac
N		 5157 sysfs_notify_dirent_safe(mddev->sysfs_state);
5158 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 5159 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 5160 return 0;
5161}
390ee602 5162EXPORT_SYMBOL_GPL(md_run);
1da177e4 5163
fd01b88c 5164static int do_md_run(struct mddev *mddev)
fe60b014
N		 5165{
5166 int err;
5167
5168 err = md_run(mddev);
5169 if (err)
5170 goto out;
69e51b44
N		 5171 err = bitmap_load(mddev);
5172 if (err) {
5173 bitmap_destroy(mddev);
5174 goto out;
5175 }
0fd018af
JB		 5176
5177 md_wakeup_thread(mddev->thread);
5178 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5179
fe60b014
N		 5180 set_capacity(mddev->gendisk, mddev->array_sectors);
5181 revalidate_disk(mddev->gendisk);
f0b4f7e2 5182 mddev->changed = 1;
fe60b014
N		 5183 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5184out:
5185 return err;
5186}
5187
fd01b88c 5188static int restart_array(struct mddev *mddev)
1da177e4
LT		 5189{
5190 struct gendisk *disk = mddev->gendisk;
1da177e4 5191
80fab1d7 5192 /* Complain if it has no devices */
1da177e4 5193 if (list_empty(&mddev->disks))
80fab1d7
AN		 5194 return -ENXIO;
5195 if (!mddev->pers)
5196 return -EINVAL;
5197 if (!mddev->ro)
5198 return -EBUSY;
5199 mddev->safemode = 0;
5200 mddev->ro = 0;
5201 set_disk_ro(disk, 0);
5202 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5203 mdname(mddev));
5204 /* Kick recovery or resync if necessary */
5205 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5206 md_wakeup_thread(mddev->thread);
5207 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5208 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5209 return 0;
1da177e4
LT		 5210}
5211
fd01b88c 5212static void md_clean(struct mddev *mddev)
6177b472
N		 5213{
5214 mddev->array_sectors = 0;
5215 mddev->external_size = 0;
5216 mddev->dev_sectors = 0;
5217 mddev->raid_disks = 0;
5218 mddev->recovery_cp = 0;
5219 mddev->resync_min = 0;
5220 mddev->resync_max = MaxSector;
5221 mddev->reshape_position = MaxSector;
5222 mddev->external = 0;
5223 mddev->persistent = 0;
5224 mddev->level = LEVEL_NONE;
5225 mddev->clevel[0] = 0;
5226 mddev->flags = 0;
5227 mddev->ro = 0;
5228 mddev->metadata_type[0] = 0;
5229 mddev->chunk_sectors = 0;
5230 mddev->ctime = mddev->utime = 0;
5231 mddev->layout = 0;
5232 mddev->max_disks = 0;
5233 mddev->events = 0;
a8707c08 5234 mddev->can_decrease_events = 0;
6177b472 5235 mddev->delta_disks = 0;
2c810cdd 5236 mddev->reshape_backwards = 0;
6177b472
N		 5237 mddev->new_level = LEVEL_NONE;
5238 mddev->new_layout = 0;
5239 mddev->new_chunk_sectors = 0;
5240 mddev->curr_resync = 0;
7f7583d4 5241 atomic64_set(&mddev->resync_mismatches, 0);
6177b472
N		 5242 mddev->suspend_lo = mddev->suspend_hi = 0;
5243 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5244 mddev->recovery = 0;
5245 mddev->in_sync = 0;
f0b4f7e2 5246 mddev->changed = 0;
6177b472 5247 mddev->degraded = 0;
6177b472 5248 mddev->safemode = 0;
050b6615 5249 mddev->merge_check_needed = 0;
6177b472
N		 5250 mddev->bitmap_info.offset = 0;
5251 mddev->bitmap_info.default_offset = 0;
6409bb05 5252 mddev->bitmap_info.default_space = 0;
6177b472
N		 5253 mddev->bitmap_info.chunksize = 0;
5254 mddev->bitmap_info.daemon_sleep = 0;
5255 mddev->bitmap_info.max_write_behind = 0;
5256}
5257
fd01b88c 5258static void __md_stop_writes(struct mddev *mddev)
a047e125 5259{
293467aa
GR		 5260 if (mddev_is_clustered(mddev))
5261 md_cluster_ops->metadata_update_start(mddev);
6b6204ee 5262 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
f851b60d 5263 flush_workqueue(md_misc_wq);
a047e125 5264 if (mddev->sync_thread) {
a047e125 5265 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 5266 md_reap_sync_thread(mddev);
a047e125
N		 5267 }
5268
5269 del_timer_sync(&mddev->safemode_timer);
5270
5271 bitmap_flush(mddev);
5272 md_super_wait(mddev);
5273
b6d428c6 5274 if (mddev->ro == 0 &&
7a0a5355 5275 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
a047e125
N		 5276 /* mark array as shutdown cleanly */
5277 mddev->in_sync = 1;
5278 md_update_sb(mddev, 1);
5279 }
293467aa
GR		 5280 if (mddev_is_clustered(mddev))
5281 md_cluster_ops->metadata_update_finish(mddev);
a047e125 5282}
defad61a 5283
fd01b88c 5284void md_stop_writes(struct mddev *mddev)
defad61a 5285{
29f097c4 5286 mddev_lock_nointr(mddev);
defad61a
N		 5287 __md_stop_writes(mddev);
5288 mddev_unlock(mddev);
5289}
390ee602 5290EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5291
5aa61f42
N		 5292static void mddev_detach(struct mddev *mddev)
5293{
5294 struct bitmap *bitmap = mddev->bitmap;
5295 /* wait for behind writes to complete */
5296 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5297 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5298 mdname(mddev));
5299 /* need to kick something here to make sure I/O goes? */
5300 wait_event(bitmap->behind_wait,
5301 atomic_read(&bitmap->behind_writes) == 0);
5302 }
36d091f4 5303 if (mddev->pers && mddev->pers->quiesce) {
5aa61f42
N		 5304 mddev->pers->quiesce(mddev, 1);
5305 mddev->pers->quiesce(mddev, 0);
5306 }
5307 md_unregister_thread(&mddev->thread);
5308 if (mddev->queue)
5309 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5310}
5311
5eff3c43 5312static void __md_stop(struct mddev *mddev)
6177b472 5313{
36d091f4 5314 struct md_personality *pers = mddev->pers;
5aa61f42 5315 mddev_detach(mddev);
36d091f4
N		 5316 spin_lock(&mddev->lock);
5317 mddev->ready = 0;
6177b472 5318 mddev->pers = NULL;
36d091f4
N		 5319 spin_unlock(&mddev->lock);
5320 pers->free(mddev, mddev->private);
5321 if (pers->sync_request && mddev->to_remove == NULL)
5322 mddev->to_remove = &md_redundancy_group;
5323 module_put(pers->owner);
cca9cf90 5324 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5325}
5eff3c43
N		 5326
5327void md_stop(struct mddev *mddev)
5328{
5329 /* stop the array and free an attached data structures.
5330 * This is called from dm-raid
5331 */
5332 __md_stop(mddev);
5333 bitmap_destroy(mddev);
5334 if (mddev->bio_set)
5335 bioset_free(mddev->bio_set);
5336}
5337
390ee602 5338EXPORT_SYMBOL_GPL(md_stop);
6177b472 5339
a05b7ea0 5340static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N		 5341{
5342 int err = 0;
30b8feb7
N		 5343 int did_freeze = 0;
5344
5345 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5346 did_freeze = 1;
5347 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5348 md_wakeup_thread(mddev->thread);
5349 }
f851b60d 5350 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5351 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5352 if (mddev->sync_thread)
30b8feb7
N		 5353 /* Thread might be blocked waiting for metadata update
5354 * which will now never happen */
5355 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5356
30b8feb7 5357 mddev_unlock(mddev);
f851b60d
N		 5358 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5359 &mddev->recovery));
30b8feb7
N		 5360 mddev_lock_nointr(mddev);
5361
a4bd82d0 5362 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5363 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7 5364 mddev->sync_thread ||
f851b60d 5365 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5366 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
a4bd82d0 5367 printk("md: %s still in use.\n",mdname(mddev));
30b8feb7
N		 5368 if (did_freeze) {
5369 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5370 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5371 md_wakeup_thread(mddev->thread);
5372 }
a4bd82d0
N		 5373 err = -EBUSY;
5374 goto out;
5375 }
5376 if (mddev->pers) {
defad61a 5377 __md_stop_writes(mddev);
a4bd82d0
N		 5378
5379 err = -ENXIO;
5380 if (mddev->ro==1)
5381 goto out;
5382 mddev->ro = 1;
5383 set_disk_ro(mddev->gendisk, 1);
5384 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d
N		 5385 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5386 md_wakeup_thread(mddev->thread);
00bcb4ac 5387 sysfs_notify_dirent_safe(mddev->sysfs_state);
30b8feb7 5388 err = 0;
a4bd82d0
N		 5389 }
5390out:
5391 mutex_unlock(&mddev->open_mutex);
5392 return err;
5393}
5394
9e653b63
N		 5395/* mode:
5396 * 0 - completely stop and dis-assemble array
9e653b63
N		 5397 * 2 - stop but do not disassemble array
5398 */
f72ffdd6 5399static int do_md_stop(struct mddev *mddev, int mode,
a05b7ea0 5400 struct block_device *bdev)
1da177e4 5401{
1da177e4 5402 struct gendisk *disk = mddev->gendisk;
3cb03002 5403 struct md_rdev *rdev;
30b8feb7
N		 5404 int did_freeze = 0;
5405
5406 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5407 did_freeze = 1;
5408 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5409 md_wakeup_thread(mddev->thread);
5410 }
f851b60d 5411 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5412 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5413 if (mddev->sync_thread)
30b8feb7
N		 5414 /* Thread might be blocked waiting for metadata update
5415 * which will now never happen */
5416 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5417
30b8feb7 5418 mddev_unlock(mddev);
f851b60d
N		 5419 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5420 !test_bit(MD_RECOVERY_RUNNING,
5421 &mddev->recovery)));
30b8feb7 5422 mddev_lock_nointr(mddev);
1da177e4 5423
c8c00a69 5424 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5425 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7
N		 5426 mddev->sysfs_active ||
5427 mddev->sync_thread ||
f851b60d 5428 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5429 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
df5b20cf 5430 printk("md: %s still in use.\n",mdname(mddev));
6e17b027 5431 mutex_unlock(&mddev->open_mutex);
30b8feb7
N		 5432 if (did_freeze) {
5433 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5434 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5435 md_wakeup_thread(mddev->thread);
5436 }
260fa034
N		 5437 return -EBUSY;
5438 }
6e17b027 5439 if (mddev->pers) {
a4bd82d0
N		 5440 if (mddev->ro)
5441 set_disk_ro(disk, 0);
409c57f3 5442
defad61a 5443 __md_stop_writes(mddev);
5eff3c43 5444 __md_stop(mddev);
a4bd82d0 5445 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5446 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5447
a4bd82d0 5448 /* tell userspace to handle 'inactive' */
00bcb4ac 5449 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5450
dafb20fa 5451 rdev_for_each(rdev, mddev)
36fad858
NK		 5452 if (rdev->raid_disk >= 0)
5453 sysfs_unlink_rdev(mddev, rdev);
c4647292 5454
a4bd82d0 5455 set_capacity(disk, 0);
6e17b027 5456 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5457 mddev->changed = 1;
a4bd82d0 5458 revalidate_disk(disk);
0d4ca600 5459
a4bd82d0
N		 5460 if (mddev->ro)
5461 mddev->ro = 0;
6e17b027
N		 5462 } else
5463 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5464 /*
5465 * Free resources if final stop
5466 */
9e653b63 5467 if (mode == 0) {
1da177e4
LT		 5468 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5469
978f946b 5470 bitmap_destroy(mddev);
c3d9714e 5471 if (mddev->bitmap_info.file) {
4af1a041
N		 5472 struct file *f = mddev->bitmap_info.file;
5473 spin_lock(&mddev->lock);
c3d9714e 5474 mddev->bitmap_info.file = NULL;
4af1a041
N		 5475 spin_unlock(&mddev->lock);
5476 fput(f);
978f946b 5477 }
c3d9714e 5478 mddev->bitmap_info.offset = 0;
978f946b 5479
1da177e4
LT		 5480 export_array(mddev);
5481
6177b472 5482 md_clean(mddev);
934d9c23 5483 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5484 if (mddev->hold_active == UNTIL_STOP)
5485 mddev->hold_active = 0;
a4bd82d0 5486 }
3f9d99c1 5487 blk_integrity_unregister(disk);
d7603b7e 5488 md_new_event(mddev);
00bcb4ac 5489 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5490 return 0;
1da177e4
LT		 5491}
5492
fdee8ae4 5493#ifndef MODULE
fd01b88c 5494static void autorun_array(struct mddev *mddev)
1da177e4 5495{
3cb03002 5496 struct md_rdev *rdev;
1da177e4
LT		 5497 int err;
5498
a757e64c 5499 if (list_empty(&mddev->disks))
1da177e4 5500 return;
1da177e4
LT		 5501
5502 printk(KERN_INFO "md: running: ");
5503
dafb20fa 5504 rdev_for_each(rdev, mddev) {
1da177e4
LT		 5505 char b[BDEVNAME_SIZE];
5506 printk("<%s>", bdevname(rdev->bdev,b));
5507 }
5508 printk("\n");
5509
d710e138 5510 err = do_md_run(mddev);
1da177e4
LT		 5511 if (err) {
5512 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
a05b7ea0 5513 do_md_stop(mddev, 0, NULL);
1da177e4
LT		 5514 }
5515}
5516
5517/*
5518 * lets try to run arrays based on all disks that have arrived
5519 * until now. (those are in pending_raid_disks)
5520 *
5521 * the method: pick the first pending disk, collect all disks with
5522 * the same UUID, remove all from the pending list and put them into
5523 * the 'same_array' list. Then order this list based on superblock
5524 * update time (freshest comes first), kick out 'old' disks and
5525 * compare superblocks. If everything's fine then run it.
5526 *
5527 * If "unit" is allocated, then bump its reference count
5528 */
5529static void autorun_devices(int part)
5530{
3cb03002 5531 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5532 struct mddev *mddev;
1da177e4
LT		 5533 char b[BDEVNAME_SIZE];
5534
5535 printk(KERN_INFO "md: autorun ...\n");
5536 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5537 int unit;
1da177e4 5538 dev_t dev;
ad01c9e3 5539 LIST_HEAD(candidates);
1da177e4 5540 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5541 struct md_rdev, same_set);
1da177e4
LT		 5542
5543 printk(KERN_INFO "md: considering %s ...\n",
5544 bdevname(rdev0->bdev,b));
5545 INIT_LIST_HEAD(&candidates);
159ec1fc 5546 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5547 if (super_90_load(rdev, rdev0, 0) >= 0) {
5548 printk(KERN_INFO "md: adding %s ...\n",
5549 bdevname(rdev->bdev,b));
5550 list_move(&rdev->same_set, &candidates);
5551 }
5552 /*
5553 * now we have a set of devices, with all of them having
5554 * mostly sane superblocks. It's time to allocate the
5555 * mddev.
5556 */
e8703fe1
N		 5557 if (part) {
5558 dev = MKDEV(mdp_major,
5559 rdev0->preferred_minor << MdpMinorShift);
5560 unit = MINOR(dev) >> MdpMinorShift;
5561 } else {
5562 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5563 unit = MINOR(dev);
5564 }
5565 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5566 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5567 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5568 break;
5569 }
1da177e4
LT		 5570
5571 md_probe(dev, NULL, NULL);
5572 mddev = mddev_find(dev);
9bbbca3a
NB		 5573 if (!mddev || !mddev->gendisk) {
5574 if (mddev)
5575 mddev_put(mddev);
5576 printk(KERN_ERR
1da177e4
LT		 5577 "md: cannot allocate memory for md drive.\n");
5578 break;
5579 }
f72ffdd6 5580 if (mddev_lock(mddev))
1da177e4
LT		 5581 printk(KERN_WARNING "md: %s locked, cannot run\n",
5582 mdname(mddev));
5583 else if (mddev->raid_disks || mddev->major_version
5584 || !list_empty(&mddev->disks)) {
f72ffdd6 5585 printk(KERN_WARNING
1da177e4
LT		 5586 "md: %s already running, cannot run %s\n",
5587 mdname(mddev), bdevname(rdev0->bdev,b));
5588 mddev_unlock(mddev);
5589 } else {
5590 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5591 mddev->persistent = 1;
159ec1fc 5592 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5593 list_del_init(&rdev->same_set);
5594 if (bind_rdev_to_array(rdev, mddev))
5595 export_rdev(rdev);
5596 }
5597 autorun_array(mddev);
5598 mddev_unlock(mddev);
5599 }
5600 /* on success, candidates will be empty, on error
5601 * it won't...
5602 */
159ec1fc 5603 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5604 list_del_init(&rdev->same_set);
1da177e4 5605 export_rdev(rdev);
4b80991c 5606 }
1da177e4
LT		 5607 mddev_put(mddev);
5608 }
5609 printk(KERN_INFO "md: ... autorun DONE.\n");
5610}
fdee8ae4 5611#endif /* !MODULE */
1da177e4 5612
f72ffdd6 5613static int get_version(void __user *arg)
1da177e4
LT		 5614{
5615 mdu_version_t ver;
5616
5617 ver.major = MD_MAJOR_VERSION;
5618 ver.minor = MD_MINOR_VERSION;
5619 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5620
5621 if (copy_to_user(arg, &ver, sizeof(ver)))
5622 return -EFAULT;
5623
5624 return 0;
5625}
5626
f72ffdd6 5627static int get_array_info(struct mddev *mddev, void __user *arg)
1da177e4
LT		 5628{
5629 mdu_array_info_t info;
a9f326eb 5630 int nr,working,insync,failed,spare;
3cb03002 5631 struct md_rdev *rdev;
1da177e4 5632
1ca69c4b
N		 5633 nr = working = insync = failed = spare = 0;
5634 rcu_read_lock();
5635 rdev_for_each_rcu(rdev, mddev) {
1da177e4 5636 nr++;
b2d444d7 5637 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5638 failed++;
5639 else {
5640 working++;
b2d444d7 5641 if (test_bit(In_sync, &rdev->flags))
f72ffdd6 5642 insync++;
1da177e4
LT		 5643 else
5644 spare++;
5645 }
5646 }
1ca69c4b 5647 rcu_read_unlock();
1da177e4
LT		 5648
5649 info.major_version = mddev->major_version;
5650 info.minor_version = mddev->minor_version;
5651 info.patch_version = MD_PATCHLEVEL_VERSION;
5652 info.ctime = mddev->ctime;
5653 info.level = mddev->level;
58c0fed4
AN		 5654 info.size = mddev->dev_sectors / 2;
5655 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5656 info.size = -1;
1da177e4
LT		 5657 info.nr_disks = nr;
5658 info.raid_disks = mddev->raid_disks;
5659 info.md_minor = mddev->md_minor;
5660 info.not_persistent= !mddev->persistent;
5661
5662 info.utime = mddev->utime;
5663 info.state = 0;
5664 if (mddev->in_sync)
5665 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5666 if (mddev->bitmap && mddev->bitmap_info.offset)
9bd35920 5667 info.state |= (1<<MD_SB_BITMAP_PRESENT);
ca8895d9
GR		 5668 if (mddev_is_clustered(mddev))
5669 info.state |= (1<<MD_SB_CLUSTERED);
a9f326eb 5670 info.active_disks = insync;
1da177e4
LT		 5671 info.working_disks = working;
5672 info.failed_disks = failed;
5673 info.spare_disks = spare;
5674
5675 info.layout = mddev->layout;
9d8f0363 5676 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5677
5678 if (copy_to_user(arg, &info, sizeof(info)))
5679 return -EFAULT;
5680
5681 return 0;
5682}
5683
f72ffdd6 5684static int get_bitmap_file(struct mddev *mddev, void __user * arg)
32a7627c
N		 5685{
5686 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
f4ad3d38 5687 char *ptr;
4af1a041 5688 int err;
32a7627c 5689
60559da4 5690 file = kmalloc(sizeof(*file), GFP_NOIO);
32a7627c 5691 if (!file)
4af1a041 5692 return -ENOMEM;
32a7627c 5693
4af1a041
N		 5694 err = 0;
5695 spin_lock(&mddev->lock);
32a7627c 5696 /* bitmap disabled, zero the first byte and copy out */
4af1a041 5697 if (!mddev->bitmap_info.file)
32a7627c 5698 file->pathname[0] = '\0';
4af1a041
N		 5699 else if ((ptr = d_path(&mddev->bitmap_info.file->f_path,
5700 file->pathname, sizeof(file->pathname))),
5701 IS_ERR(ptr))
5702 err = PTR_ERR(ptr);
5703 else
5704 memmove(file->pathname, ptr,
5705 sizeof(file->pathname)-(ptr-file->pathname));
5706 spin_unlock(&mddev->lock);
32a7627c 5707
4af1a041
N		 5708 if (err == 0 &&
5709 copy_to_user(arg, file, sizeof(*file)))
32a7627c 5710 err = -EFAULT;
4af1a041 5711
32a7627c
N		 5712 kfree(file);
5713 return err;
5714}
5715
f72ffdd6 5716static int get_disk_info(struct mddev *mddev, void __user * arg)
1da177e4
LT		 5717{
5718 mdu_disk_info_t info;
3cb03002 5719 struct md_rdev *rdev;
1da177e4
LT		 5720
5721 if (copy_from_user(&info, arg, sizeof(info)))
5722 return -EFAULT;
5723
1ca69c4b 5724 rcu_read_lock();
57d051dc 5725 rdev = md_find_rdev_nr_rcu(mddev, info.number);
1da177e4
LT		 5726 if (rdev) {
5727 info.major = MAJOR(rdev->bdev->bd_dev);
5728 info.minor = MINOR(rdev->bdev->bd_dev);
5729 info.raid_disk = rdev->raid_disk;
5730 info.state = 0;
b2d444d7 5731 if (test_bit(Faulty, &rdev->flags))
1da177e4 5732 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5733 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5734 info.state |= (1<<MD_DISK_ACTIVE);
5735 info.state |= (1<<MD_DISK_SYNC);
5736 }
8ddf9efe
N		 5737 if (test_bit(WriteMostly, &rdev->flags))
5738 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5739 } else {
5740 info.major = info.minor = 0;
5741 info.raid_disk = -1;
5742 info.state = (1<<MD_DISK_REMOVED);
5743 }
1ca69c4b 5744 rcu_read_unlock();
1da177e4
LT		 5745
5746 if (copy_to_user(arg, &info, sizeof(info)))
5747 return -EFAULT;
5748
5749 return 0;
5750}
5751
f72ffdd6 5752static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
1da177e4
LT		 5753{
5754 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5755 struct md_rdev *rdev;
1da177e4
LT		 5756 dev_t dev = MKDEV(info->major,info->minor);
5757
1aee41f6
GR		 5758 if (mddev_is_clustered(mddev) &&
5759 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
fa8259da 5760 pr_err("%s: Cannot add to clustered mddev.\n",
1aee41f6
GR		 5761 mdname(mddev));
5762 return -EINVAL;
5763 }
5764
1da177e4
LT		 5765 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5766 return -EOVERFLOW;
5767
5768 if (!mddev->raid_disks) {
5769 int err;
5770 /* expecting a device which has a superblock */
5771 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5772 if (IS_ERR(rdev)) {
f72ffdd6 5773 printk(KERN_WARNING
1da177e4
LT		 5774 "md: md_import_device returned %ld\n",
5775 PTR_ERR(rdev));
5776 return PTR_ERR(rdev);
5777 }
5778 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5779 struct md_rdev *rdev0
5780 = list_entry(mddev->disks.next,
5781 struct md_rdev, same_set);
a9f326eb 5782 err = super_types[mddev->major_version]
1da177e4
LT		 5783 .load_super(rdev, rdev0, mddev->minor_version);
5784 if (err < 0) {
f72ffdd6 5785 printk(KERN_WARNING
1da177e4 5786 "md: %s has different UUID to %s\n",
f72ffdd6 5787 bdevname(rdev->bdev,b),
1da177e4
LT		 5788 bdevname(rdev0->bdev,b2));
5789 export_rdev(rdev);
5790 return -EINVAL;
5791 }
5792 }
5793 err = bind_rdev_to_array(rdev, mddev);
5794 if (err)
5795 export_rdev(rdev);
5796 return err;
5797 }
5798
5799 /*
5800 * add_new_disk can be used once the array is assembled
5801 * to add "hot spares". They must already have a superblock
5802 * written
5803 */
5804 if (mddev->pers) {
5805 int err;
5806 if (!mddev->pers->hot_add_disk) {
f72ffdd6 5807 printk(KERN_WARNING
1da177e4
LT		 5808 "%s: personality does not support diskops!\n",
5809 mdname(mddev));
5810 return -EINVAL;
5811 }
7b1e35f6
N		 5812 if (mddev->persistent)
5813 rdev = md_import_device(dev, mddev->major_version,
5814 mddev->minor_version);
5815 else
5816 rdev = md_import_device(dev, -1, -1);
1da177e4 5817 if (IS_ERR(rdev)) {
f72ffdd6 5818 printk(KERN_WARNING
1da177e4
LT		 5819 "md: md_import_device returned %ld\n",
5820 PTR_ERR(rdev));
5821 return PTR_ERR(rdev);
5822 }
1a855a06 5823 /* set saved_raid_disk if appropriate */
41158c7e
N		 5824 if (!mddev->persistent) {
5825 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5826 info->raid_disk < mddev->raid_disks) {
41158c7e 5827 rdev->raid_disk = info->raid_disk;
bf572541 5828 set_bit(In_sync, &rdev->flags);
8313b8e5 5829 clear_bit(Bitmap_sync, &rdev->flags);
bf572541 5830 } else
41158c7e 5831 rdev->raid_disk = -1;
f466722c 5832 rdev->saved_raid_disk = rdev->raid_disk;
41158c7e
N		 5833 } else
5834 super_types[mddev->major_version].
5835 validate_super(mddev, rdev);
bedd86b7 5836 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 5837 rdev->raid_disk != info->raid_disk) {
bedd86b7
N		 5838 /* This was a hot-add request, but events doesn't
5839 * match, so reject it.
5840 */
5841 export_rdev(rdev);
5842 return -EINVAL;
5843 }
5844
b2d444d7 5845 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5846 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5847 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5848 else
5849 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5850
1aee41f6
GR		 5851 /*
5852 * check whether the device shows up in other nodes
5853 */
5854 if (mddev_is_clustered(mddev)) {
5855 if (info->state & (1 << MD_DISK_CANDIDATE)) {
5856 /* Through --cluster-confirm */
5857 set_bit(Candidate, &rdev->flags);
fa8259da
GR		 5858 err = md_cluster_ops->new_disk_ack(mddev, true);
5859 if (err) {
5860 export_rdev(rdev);
5861 return err;
5862 }
1aee41f6
GR		 5863 } else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
5864 /* --add initiated by this node */
5865 err = md_cluster_ops->add_new_disk_start(mddev, rdev);
5866 if (err) {
5867 md_cluster_ops->add_new_disk_finish(mddev);
5868 export_rdev(rdev);
5869 return err;
5870 }
5871 }
5872 }
5873
1da177e4
LT		 5874 rdev->raid_disk = -1;
5875 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5876 if (!err && !mddev->pers->hot_remove_disk) {
5877 /* If there is hot_add_disk but no hot_remove_disk
5878 * then added disks for geometry changes,
5879 * and should be added immediately.
5880 */
5881 super_types[mddev->major_version].
5882 validate_super(mddev, rdev);
5883 err = mddev->pers->hot_add_disk(mddev, rdev);
5884 if (err)
5885 unbind_rdev_from_array(rdev);
5886 }
1da177e4
LT		 5887 if (err)
5888 export_rdev(rdev);
52664732 5889 else
00bcb4ac 5890 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5891
7ceb17e8 5892 set_bit(MD_CHANGE_DEVS, &mddev->flags);
72a23c21
NB		 5893 if (mddev->degraded)
5894 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5895 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5896 if (!err)
5897 md_new_event(mddev);
005eca5e 5898 md_wakeup_thread(mddev->thread);
1aee41f6
GR		 5899 if (mddev_is_clustered(mddev) &&
5900 (info->state & (1 << MD_DISK_CLUSTER_ADD)))
5901 md_cluster_ops->add_new_disk_finish(mddev);
1da177e4
LT		 5902 return err;
5903 }
5904
5905 /* otherwise, add_new_disk is only allowed
5906 * for major_version==0 superblocks
5907 */
5908 if (mddev->major_version != 0) {
5909 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5910 mdname(mddev));
5911 return -EINVAL;
5912 }
5913
5914 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5915 int err;
d710e138 5916 rdev = md_import_device(dev, -1, 0);
1da177e4 5917 if (IS_ERR(rdev)) {
f72ffdd6 5918 printk(KERN_WARNING
1da177e4
LT		 5919 "md: error, md_import_device() returned %ld\n",
5920 PTR_ERR(rdev));
5921 return PTR_ERR(rdev);
5922 }
5923 rdev->desc_nr = info->number;
5924 if (info->raid_disk < mddev->raid_disks)
5925 rdev->raid_disk = info->raid_disk;
5926 else
5927 rdev->raid_disk = -1;
5928
1da177e4 5929 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5930 if (info->state & (1<<MD_DISK_SYNC))
5931 set_bit(In_sync, &rdev->flags);
1da177e4 5932
8ddf9efe
N		 5933 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5934 set_bit(WriteMostly, &rdev->flags);
5935
1da177e4
LT		 5936 if (!mddev->persistent) {
5937 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5938 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5939 } else
57b2caa3 5940 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5941 rdev->sectors = rdev->sb_start;
1da177e4 5942
2bf071bf
N		 5943 err = bind_rdev_to_array(rdev, mddev);
5944 if (err) {
5945 export_rdev(rdev);
5946 return err;
5947 }
1da177e4
LT		 5948 }
5949
5950 return 0;
5951}
5952
f72ffdd6 5953static int hot_remove_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 5954{
5955 char b[BDEVNAME_SIZE];
3cb03002 5956 struct md_rdev *rdev;
1da177e4 5957
1da177e4
LT		 5958 rdev = find_rdev(mddev, dev);
5959 if (!rdev)
5960 return -ENXIO;
5961
293467aa
GR		 5962 if (mddev_is_clustered(mddev))
5963 md_cluster_ops->metadata_update_start(mddev);
5964
3ea8929d
N		 5965 clear_bit(Blocked, &rdev->flags);
5966 remove_and_add_spares(mddev, rdev);
5967
1da177e4
LT		 5968 if (rdev->raid_disk >= 0)
5969 goto busy;
5970
fb56dfef 5971 md_kick_rdev_from_array(rdev);
850b2b42 5972 md_update_sb(mddev, 1);
d7603b7e 5973 md_new_event(mddev);
1da177e4 5974
293467aa
GR		 5975 if (mddev_is_clustered(mddev))
5976 md_cluster_ops->metadata_update_finish(mddev);
5977
1da177e4
LT		 5978 return 0;
5979busy:
293467aa
GR		 5980 if (mddev_is_clustered(mddev))
5981 md_cluster_ops->metadata_update_cancel(mddev);
fdefa4d8 5982 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5983 bdevname(rdev->bdev,b), mdname(mddev));
5984 return -EBUSY;
5985}
5986
f72ffdd6 5987static int hot_add_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 5988{
5989 char b[BDEVNAME_SIZE];
5990 int err;
3cb03002 5991 struct md_rdev *rdev;
1da177e4
LT		 5992
5993 if (!mddev->pers)
5994 return -ENODEV;
5995
5996 if (mddev->major_version != 0) {
5997 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5998 " version-0 superblocks.\n",
5999 mdname(mddev));
6000 return -EINVAL;
6001 }
6002 if (!mddev->pers->hot_add_disk) {
f72ffdd6 6003 printk(KERN_WARNING
1da177e4
LT		 6004 "%s: personality does not support diskops!\n",
6005 mdname(mddev));
6006 return -EINVAL;
6007 }
6008
d710e138 6009 rdev = md_import_device(dev, -1, 0);
1da177e4 6010 if (IS_ERR(rdev)) {
f72ffdd6 6011 printk(KERN_WARNING
1da177e4
LT		 6012 "md: error, md_import_device() returned %ld\n",
6013 PTR_ERR(rdev));
6014 return -EINVAL;
6015 }
6016
6017 if (mddev->persistent)
57b2caa3 6018 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 6019 else
77304d2a 6020 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 6021
8190e754 6022 rdev->sectors = rdev->sb_start;
1da177e4 6023
b2d444d7 6024 if (test_bit(Faulty, &rdev->flags)) {
f72ffdd6 6025 printk(KERN_WARNING
1da177e4
LT		 6026 "md: can not hot-add faulty %s disk to %s!\n",
6027 bdevname(rdev->bdev,b), mdname(mddev));
6028 err = -EINVAL;
6029 goto abort_export;
6030 }
293467aa
GR		 6031
6032 if (mddev_is_clustered(mddev))
6033 md_cluster_ops->metadata_update_start(mddev);
b2d444d7 6034 clear_bit(In_sync, &rdev->flags);
1da177e4 6035 rdev->desc_nr = -1;
5842730d 6036 rdev->saved_raid_disk = -1;
2bf071bf
N		 6037 err = bind_rdev_to_array(rdev, mddev);
6038 if (err)
293467aa 6039 goto abort_clustered;
1da177e4
LT		 6040
6041 /*
6042 * The rest should better be atomic, we can have disk failures
6043 * noticed in interrupt contexts ...
6044 */
6045
1da177e4
LT		 6046 rdev->raid_disk = -1;
6047
850b2b42 6048 md_update_sb(mddev, 1);
1da177e4 6049
293467aa
GR		 6050 if (mddev_is_clustered(mddev))
6051 md_cluster_ops->metadata_update_finish(mddev);
1da177e4
LT		 6052 /*
6053 * Kick recovery, maybe this spare has to be added to the
6054 * array immediately.
6055 */
6056 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6057 md_wakeup_thread(mddev->thread);
d7603b7e 6058 md_new_event(mddev);
1da177e4
LT		 6059 return 0;
6060
293467aa
GR		 6061abort_clustered:
6062 if (mddev_is_clustered(mddev))
6063 md_cluster_ops->metadata_update_cancel(mddev);
1da177e4
LT		 6064abort_export:
6065 export_rdev(rdev);
6066 return err;
6067}
6068
fd01b88c 6069static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c 6070{
035328c2 6071 int err = 0;
32a7627c 6072
36fa3063 6073 if (mddev->pers) {
d66b1b39 6074 if (!mddev->pers->quiesce || !mddev->thread)
36fa3063
N		 6075 return -EBUSY;
6076 if (mddev->recovery || mddev->sync_thread)
6077 return -EBUSY;
6078 /* we should be able to change the bitmap.. */
6079 }
32a7627c 6080
36fa3063 6081 if (fd >= 0) {
035328c2 6082 struct inode *inode;
1e594bb2
N		 6083 struct file *f;
6084
6085 if (mddev->bitmap || mddev->bitmap_info.file)
36fa3063 6086 return -EEXIST; /* cannot add when bitmap is present */
1e594bb2 6087 f = fget(fd);
32a7627c 6088
1e594bb2 6089 if (f == NULL) {
36fa3063
N		 6090 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6091 mdname(mddev));
6092 return -EBADF;
6093 }
6094
1e594bb2 6095 inode = f->f_mapping->host;
035328c2
N		 6096 if (!S_ISREG(inode->i_mode)) {
6097 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6098 mdname(mddev));
6099 err = -EBADF;
1e594bb2 6100 } else if (!(f->f_mode & FMODE_WRITE)) {
035328c2
N		 6101 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6102 mdname(mddev));
6103 err = -EBADF;
6104 } else if (atomic_read(&inode->i_writecount) != 1) {
36fa3063
N		 6105 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6106 mdname(mddev));
035328c2
N		 6107 err = -EBUSY;
6108 }
6109 if (err) {
1e594bb2 6110 fput(f);
36fa3063
N		 6111 return err;
6112 }
1e594bb2 6113 mddev->bitmap_info.file = f;
c3d9714e 6114 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 6115 } else if (mddev->bitmap == NULL)
6116 return -ENOENT; /* cannot remove what isn't there */
6117 err = 0;
6118 if (mddev->pers) {
6119 mddev->pers->quiesce(mddev, 1);
69e51b44 6120 if (fd >= 0) {
f9209a32
GR		 6121 struct bitmap *bitmap;
6122
6123 bitmap = bitmap_create(mddev, -1);
6124 if (!IS_ERR(bitmap)) {
6125 mddev->bitmap = bitmap;
69e51b44 6126 err = bitmap_load(mddev);
ba599aca
N		 6127 } else
6128 err = PTR_ERR(bitmap);
69e51b44 6129 }
d7375ab3 6130 if (fd < 0 || err) {
36fa3063 6131 bitmap_destroy(mddev);
d7375ab3
N		 6132 fd = -1; /* make sure to put the file */
6133 }
36fa3063 6134 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 6135 }
6136 if (fd < 0) {
4af1a041
N		 6137 struct file *f = mddev->bitmap_info.file;
6138 if (f) {
6139 spin_lock(&mddev->lock);
6140 mddev->bitmap_info.file = NULL;
6141 spin_unlock(&mddev->lock);
6142 fput(f);
6143 }
36fa3063
N		 6144 }
6145
32a7627c
N		 6146 return err;
6147}
6148
1da177e4
LT		 6149/*
6150 * set_array_info is used two different ways
6151 * The original usage is when creating a new array.
6152 * In this usage, raid_disks is > 0 and it together with
6153 * level, size, not_persistent,layout,chunksize determine the
6154 * shape of the array.
6155 * This will always create an array with a type-0.90.0 superblock.
6156 * The newer usage is when assembling an array.
6157 * In this case raid_disks will be 0, and the major_version field is
6158 * use to determine which style super-blocks are to be found on the devices.
6159 * The minor and patch _version numbers are also kept incase the
6160 * super_block handler wishes to interpret them.
6161 */
f72ffdd6 6162static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6163{
6164
6165 if (info->raid_disks == 0) {
6166 /* just setting version number for superblock loading */
6167 if (info->major_version < 0 ||
50511da3 6168 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 6169 super_types[info->major_version].name == NULL) {
6170 /* maybe try to auto-load a module? */
f72ffdd6 6171 printk(KERN_INFO
1da177e4
LT		 6172 "md: superblock version %d not known\n",
6173 info->major_version);
6174 return -EINVAL;
6175 }
6176 mddev->major_version = info->major_version;
6177 mddev->minor_version = info->minor_version;
6178 mddev->patch_version = info->patch_version;
3f9d7b0d 6179 mddev->persistent = !info->not_persistent;
cbd19983
N		 6180 /* ensure mddev_put doesn't delete this now that there
6181 * is some minimal configuration.
6182 */
6183 mddev->ctime = get_seconds();
1da177e4
LT		 6184 return 0;
6185 }
6186 mddev->major_version = MD_MAJOR_VERSION;
6187 mddev->minor_version = MD_MINOR_VERSION;
6188 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6189 mddev->ctime = get_seconds();
6190
6191 mddev->level = info->level;
17115e03 6192 mddev->clevel[0] = 0;
58c0fed4 6193 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 6194 mddev->raid_disks = info->raid_disks;
6195 /* don't set md_minor, it is determined by which /dev/md* was
6196 * openned
6197 */
6198 if (info->state & (1<<MD_SB_CLEAN))
6199 mddev->recovery_cp = MaxSector;
6200 else
6201 mddev->recovery_cp = 0;
6202 mddev->persistent = ! info->not_persistent;
e691063a 6203 mddev->external = 0;
1da177e4
LT		 6204
6205 mddev->layout = info->layout;
9d8f0363 6206 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 6207
6208 mddev->max_disks = MD_SB_DISKS;
6209
e691063a
N		 6210 if (mddev->persistent)
6211 mddev->flags = 0;
850b2b42 6212 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6213
c3d9714e 6214 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6215 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6216 mddev->bitmap_info.offset = 0;
b2a2703c 6217
f6705578
N		 6218 mddev->reshape_position = MaxSector;
6219
1da177e4
LT		 6220 /*
6221 * Generate a 128 bit UUID
6222 */
6223 get_random_bytes(mddev->uuid, 16);
6224
f6705578 6225 mddev->new_level = mddev->level;
664e7c41 6226 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 6227 mddev->new_layout = mddev->layout;
6228 mddev->delta_disks = 0;
2c810cdd 6229 mddev->reshape_backwards = 0;
f6705578 6230
1da177e4
LT		 6231 return 0;
6232}
6233
fd01b88c 6234void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6235{
b522adcd
DW		 6236 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6237
6238 if (mddev->external_size)
6239 return;
6240
1f403624
DW		 6241 mddev->array_sectors = array_sectors;
6242}
6243EXPORT_SYMBOL(md_set_array_sectors);
6244
fd01b88c 6245static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6246{
3cb03002 6247 struct md_rdev *rdev;
a35b0d69 6248 int rv;
d71f9f88 6249 int fit = (num_sectors == 0);
a35b0d69
N		 6250
6251 if (mddev->pers->resize == NULL)
6252 return -EINVAL;
d71f9f88
AN		 6253 /* The "num_sectors" is the number of sectors of each device that
6254 * is used. This can only make sense for arrays with redundancy.
6255 * linear and raid0 always use whatever space is available. We can only
6256 * consider changing this number if no resync or reconstruction is
6257 * happening, and if the new size is acceptable. It must fit before the
0f420358 6258 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 6259 * of each device. If num_sectors is zero, we find the largest size
6260 * that fits.
a35b0d69 6261 */
f851b60d
N		 6262 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6263 mddev->sync_thread)
a35b0d69 6264 return -EBUSY;
bd8839e0
N		 6265 if (mddev->ro)
6266 return -EROFS;
a4a6125a 6267
dafb20fa 6268 rdev_for_each(rdev, mddev) {
dd8ac336 6269 sector_t avail = rdev->sectors;
01ab5662 6270
d71f9f88
AN		 6271 if (fit && (num_sectors == 0 || num_sectors > avail))
6272 num_sectors = avail;
6273 if (avail < num_sectors)
a35b0d69
N		 6274 return -ENOSPC;
6275 }
d71f9f88 6276 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 6277 if (!rv)
6278 revalidate_disk(mddev->gendisk);
a35b0d69
N		 6279 return rv;
6280}
6281
fd01b88c 6282static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 6283{
6284 int rv;
c6563a8c 6285 struct md_rdev *rdev;
da943b99 6286 /* change the number of raid disks */
63c70c4f 6287 if (mddev->pers->check_reshape == NULL)
da943b99 6288 return -EINVAL;
bd8839e0
N		 6289 if (mddev->ro)
6290 return -EROFS;
da943b99 6291 if (raid_disks <= 0 ||
233fca36 6292 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 6293 return -EINVAL;
f851b60d
N		 6294 if (mddev->sync_thread ||
6295 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6296 mddev->reshape_position != MaxSector)
da943b99 6297 return -EBUSY;
c6563a8c
N		 6298
6299 rdev_for_each(rdev, mddev) {
6300 if (mddev->raid_disks < raid_disks &&
6301 rdev->data_offset < rdev->new_data_offset)
6302 return -EINVAL;
6303 if (mddev->raid_disks > raid_disks &&
6304 rdev->data_offset > rdev->new_data_offset)
6305 return -EINVAL;
6306 }
6307
63c70c4f 6308 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N		 6309 if (mddev->delta_disks < 0)
6310 mddev->reshape_backwards = 1;
6311 else if (mddev->delta_disks > 0)
6312 mddev->reshape_backwards = 0;
63c70c4f
N		 6313
6314 rv = mddev->pers->check_reshape(mddev);
2c810cdd 6315 if (rv < 0) {
de171cb9 6316 mddev->delta_disks = 0;
2c810cdd
N		 6317 mddev->reshape_backwards = 0;
6318 }
da943b99
N		 6319 return rv;
6320}
6321
1da177e4
LT		 6322/*
6323 * update_array_info is used to change the configuration of an
6324 * on-line array.
6325 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6326 * fields in the info are checked against the array.
6327 * Any differences that cannot be handled will cause an error.
6328 * Normally, only one change can be managed at a time.
6329 */
fd01b88c 6330static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6331{
6332 int rv = 0;
6333 int cnt = 0;
36fa3063
N		 6334 int state = 0;
6335
6336 /* calculate expected state,ignoring low bits */
c3d9714e 6337 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 6338 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 6339
6340 if (mddev->major_version != info->major_version ||
6341 mddev->minor_version != info->minor_version ||
6342/* mddev->patch_version != info->patch_version || */
6343 mddev->ctime != info->ctime ||
6344 mddev->level != info->level ||
6345/* mddev->layout != info->layout || */
6346 !mddev->persistent != info->not_persistent||
9d8f0363 6347 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 6348 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6349 ((state^info->state) & 0xfffffe00)
6350 )
1da177e4
LT		 6351 return -EINVAL;
6352 /* Check there is only one change */
58c0fed4
AN		 6353 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6354 cnt++;
6355 if (mddev->raid_disks != info->raid_disks)
6356 cnt++;
6357 if (mddev->layout != info->layout)
6358 cnt++;
6359 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6360 cnt++;
6361 if (cnt == 0)
6362 return 0;
6363 if (cnt > 1)
6364 return -EINVAL;
1da177e4
LT		 6365
6366 if (mddev->layout != info->layout) {
6367 /* Change layout
6368 * we don't need to do anything at the md level, the
6369 * personality will take care of it all.
6370 */
50ac168a 6371 if (mddev->pers->check_reshape == NULL)
1da177e4 6372 return -EINVAL;
597a711b
N		 6373 else {
6374 mddev->new_layout = info->layout;
50ac168a 6375 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 6376 if (rv)
6377 mddev->new_layout = mddev->layout;
6378 return rv;
6379 }
1da177e4 6380 }
293467aa
GR		 6381 if (mddev_is_clustered(mddev))
6382 md_cluster_ops->metadata_update_start(mddev);
58c0fed4 6383 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6384 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6385
da943b99
N		 6386 if (mddev->raid_disks != info->raid_disks)
6387 rv = update_raid_disks(mddev, info->raid_disks);
6388
36fa3063 6389 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
293467aa
GR		 6390 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6391 rv = -EINVAL;
6392 goto err;
6393 }
6394 if (mddev->recovery || mddev->sync_thread) {
6395 rv = -EBUSY;
6396 goto err;
6397 }
36fa3063 6398 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
f9209a32 6399 struct bitmap *bitmap;
36fa3063 6400 /* add the bitmap */
293467aa
GR		 6401 if (mddev->bitmap) {
6402 rv = -EEXIST;
6403 goto err;
6404 }
6405 if (mddev->bitmap_info.default_offset == 0) {
6406 rv = -EINVAL;
6407 goto err;
6408 }
c3d9714e
N		 6409 mddev->bitmap_info.offset =
6410 mddev->bitmap_info.default_offset;
6409bb05
N		 6411 mddev->bitmap_info.space =
6412 mddev->bitmap_info.default_space;
36fa3063 6413 mddev->pers->quiesce(mddev, 1);
f9209a32
GR		 6414 bitmap = bitmap_create(mddev, -1);
6415 if (!IS_ERR(bitmap)) {
6416 mddev->bitmap = bitmap;
69e51b44 6417 rv = bitmap_load(mddev);
ba599aca
N		 6418 } else
6419 rv = PTR_ERR(bitmap);
36fa3063
N		 6420 if (rv)
6421 bitmap_destroy(mddev);
6422 mddev->pers->quiesce(mddev, 0);
6423 } else {
6424 /* remove the bitmap */
293467aa
GR		 6425 if (!mddev->bitmap) {
6426 rv = -ENOENT;
6427 goto err;
6428 }
6429 if (mddev->bitmap->storage.file) {
6430 rv = -EINVAL;
6431 goto err;
6432 }
36fa3063
N		 6433 mddev->pers->quiesce(mddev, 1);
6434 bitmap_destroy(mddev);
6435 mddev->pers->quiesce(mddev, 0);
c3d9714e 6436 mddev->bitmap_info.offset = 0;
36fa3063
N		 6437 }
6438 }
850b2b42 6439 md_update_sb(mddev, 1);
293467aa
GR		 6440 if (mddev_is_clustered(mddev))
6441 md_cluster_ops->metadata_update_finish(mddev);
6442 return rv;
6443err:
6444 if (mddev_is_clustered(mddev))
6445 md_cluster_ops->metadata_update_cancel(mddev);
1da177e4
LT		 6446 return rv;
6447}
6448
fd01b88c 6449static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6450{
3cb03002 6451 struct md_rdev *rdev;
1ca69c4b 6452 int err = 0;
1da177e4
LT		 6453
6454 if (mddev->pers == NULL)
6455 return -ENODEV;
6456
1ca69c4b
N		 6457 rcu_read_lock();
6458 rdev = find_rdev_rcu(mddev, dev);
1da177e4 6459 if (!rdev)
1ca69c4b
N		 6460 err = -ENODEV;
6461 else {
6462 md_error(mddev, rdev);
6463 if (!test_bit(Faulty, &rdev->flags))
6464 err = -EBUSY;
6465 }
6466 rcu_read_unlock();
6467 return err;
1da177e4
LT		 6468}
6469
2f9618ce
AN		 6470/*
6471 * We have a problem here : there is no easy way to give a CHS
6472 * virtual geometry. We currently pretend that we have a 2 heads
6473 * 4 sectors (with a BIG number of cylinders...). This drives
6474 * dosfs just mad... ;-)
6475 */
a885c8c4
CH		 6476static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6477{
fd01b88c 6478 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6479
6480 geo->heads = 2;
6481 geo->sectors = 4;
49ce6cea 6482 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6483 return 0;
6484}
6485
cb335f88
NS		 6486static inline bool md_ioctl_valid(unsigned int cmd)
6487{
6488 switch (cmd) {
6489 case ADD_NEW_DISK:
6490 case BLKROSET:
6491 case GET_ARRAY_INFO:
6492 case GET_BITMAP_FILE:
6493 case GET_DISK_INFO:
6494 case HOT_ADD_DISK:
6495 case HOT_REMOVE_DISK:
cb335f88
NS		 6496 case RAID_AUTORUN:
6497 case RAID_VERSION:
6498 case RESTART_ARRAY_RW:
6499 case RUN_ARRAY:
6500 case SET_ARRAY_INFO:
6501 case SET_BITMAP_FILE:
6502 case SET_DISK_FAULTY:
6503 case STOP_ARRAY:
6504 case STOP_ARRAY_RO:
1aee41f6 6505 case CLUSTERED_DISK_NACK:
cb335f88
NS		 6506 return true;
6507 default:
6508 return false;
6509 }
6510}
6511
a39907fa 6512static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6513 unsigned int cmd, unsigned long arg)
6514{
6515 int err = 0;
6516 void __user *argp = (void __user *)arg;
fd01b88c 6517 struct mddev *mddev = NULL;
e2218350 6518 int ro;
1da177e4 6519
cb335f88
NS		 6520 if (!md_ioctl_valid(cmd))
6521 return -ENOTTY;
6522
506c9e44
N		 6523 switch (cmd) {
6524 case RAID_VERSION:
6525 case GET_ARRAY_INFO:
6526 case GET_DISK_INFO:
6527 break;
6528 default:
6529 if (!capable(CAP_SYS_ADMIN))
6530 return -EACCES;
6531 }
1da177e4
LT		 6532
6533 /*
6534 * Commands dealing with the RAID driver but not any
6535 * particular array:
6536 */
c02c0aeb
N		 6537 switch (cmd) {
6538 case RAID_VERSION:
6539 err = get_version(argp);
3adc28d8 6540 goto out;
1da177e4 6541
1da177e4 6542#ifndef MODULE
c02c0aeb
N		 6543 case RAID_AUTORUN:
6544 err = 0;
6545 autostart_arrays(arg);
3adc28d8 6546 goto out;
1da177e4 6547#endif
c02c0aeb 6548 default:;
1da177e4
LT		 6549 }
6550
6551 /*
6552 * Commands creating/starting a new array:
6553 */
6554
a39907fa 6555 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6556
6557 if (!mddev) {
6558 BUG();
3adc28d8 6559 goto out;
1da177e4
LT		 6560 }
6561
1ca69c4b
N		 6562 /* Some actions do not requires the mutex */
6563 switch (cmd) {
6564 case GET_ARRAY_INFO:
6565 if (!mddev->raid_disks && !mddev->external)
6566 err = -ENODEV;
6567 else
6568 err = get_array_info(mddev, argp);
3adc28d8 6569 goto out;
1ca69c4b
N		 6570
6571 case GET_DISK_INFO:
6572 if (!mddev->raid_disks && !mddev->external)
6573 err = -ENODEV;
6574 else
6575 err = get_disk_info(mddev, argp);
3adc28d8 6576 goto out;
1ca69c4b
N		 6577
6578 case SET_DISK_FAULTY:
6579 err = set_disk_faulty(mddev, new_decode_dev(arg));
3adc28d8 6580 goto out;
4af1a041
N		 6581
6582 case GET_BITMAP_FILE:
6583 err = get_bitmap_file(mddev, argp);
6584 goto out;
6585
1ca69c4b
N		 6586 }
6587
a7a3f08d
N		 6588 if (cmd == ADD_NEW_DISK)
6589 /* need to ensure md_delayed_delete() has completed */
6590 flush_workqueue(md_misc_wq);
6591
90f5f7ad
HR		 6592 if (cmd == HOT_REMOVE_DISK)
6593 /* need to ensure recovery thread has run */
6594 wait_event_interruptible_timeout(mddev->sb_wait,
6595 !test_bit(MD_RECOVERY_NEEDED,
6596 &mddev->flags),
6597 msecs_to_jiffies(5000));
260fa034
N		 6598 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6599 /* Need to flush page cache, and ensure no-one else opens
6600 * and writes
6601 */
6602 mutex_lock(&mddev->open_mutex);
9ba3b7f5 6603 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
260fa034
N		 6604 mutex_unlock(&mddev->open_mutex);
6605 err = -EBUSY;
3adc28d8 6606 goto out;
260fa034
N		 6607 }
6608 set_bit(MD_STILL_CLOSED, &mddev->flags);
6609 mutex_unlock(&mddev->open_mutex);
6610 sync_blockdev(bdev);
6611 }
1da177e4
LT		 6612 err = mddev_lock(mddev);
6613 if (err) {
f72ffdd6 6614 printk(KERN_INFO
1da177e4
LT		 6615 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6616 err, cmd);
3adc28d8 6617 goto out;
1da177e4
LT		 6618 }
6619
c02c0aeb
N		 6620 if (cmd == SET_ARRAY_INFO) {
6621 mdu_array_info_t info;
6622 if (!arg)
6623 memset(&info, 0, sizeof(info));
6624 else if (copy_from_user(&info, argp, sizeof(info))) {
6625 err = -EFAULT;
3adc28d8 6626 goto unlock;
c02c0aeb
N		 6627 }
6628 if (mddev->pers) {
6629 err = update_array_info(mddev, &info);
6630 if (err) {
6631 printk(KERN_WARNING "md: couldn't update"
6632 " array info. %d\n", err);
3adc28d8 6633 goto unlock;
1da177e4 6634 }
3adc28d8 6635 goto unlock;
c02c0aeb
N		 6636 }
6637 if (!list_empty(&mddev->disks)) {
6638 printk(KERN_WARNING
6639 "md: array %s already has disks!\n",
6640 mdname(mddev));
6641 err = -EBUSY;
3adc28d8 6642 goto unlock;
c02c0aeb
N		 6643 }
6644 if (mddev->raid_disks) {
6645 printk(KERN_WARNING
6646 "md: array %s already initialised!\n",
6647 mdname(mddev));
6648 err = -EBUSY;
3adc28d8 6649 goto unlock;
c02c0aeb
N		 6650 }
6651 err = set_array_info(mddev, &info);
6652 if (err) {
6653 printk(KERN_WARNING "md: couldn't set"
6654 " array info. %d\n", err);
3adc28d8 6655 goto unlock;
c02c0aeb 6656 }
3adc28d8 6657 goto unlock;
1da177e4
LT		 6658 }
6659
6660 /*
6661 * Commands querying/configuring an existing array:
6662 */
32a7627c 6663 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6664 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6665 if ((!mddev->raid_disks && !mddev->external)
6666 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6667 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6668 && cmd != GET_BITMAP_FILE) {
1da177e4 6669 err = -ENODEV;
3adc28d8 6670 goto unlock;
1da177e4
LT		 6671 }
6672
6673 /*
6674 * Commands even a read-only array can execute:
6675 */
c02c0aeb 6676 switch (cmd) {
c02c0aeb
N		 6677 case RESTART_ARRAY_RW:
6678 err = restart_array(mddev);
3adc28d8 6679 goto unlock;
1da177e4 6680
c02c0aeb
N		 6681 case STOP_ARRAY:
6682 err = do_md_stop(mddev, 0, bdev);
3adc28d8 6683 goto unlock;
1da177e4 6684
c02c0aeb
N		 6685 case STOP_ARRAY_RO:
6686 err = md_set_readonly(mddev, bdev);
3adc28d8 6687 goto unlock;
1da177e4 6688
3ea8929d
N		 6689 case HOT_REMOVE_DISK:
6690 err = hot_remove_disk(mddev, new_decode_dev(arg));
3adc28d8 6691 goto unlock;
3ea8929d 6692
7ceb17e8
N		 6693 case ADD_NEW_DISK:
6694 /* We can support ADD_NEW_DISK on read-only arrays
6695 * on if we are re-adding a preexisting device.
6696 * So require mddev->pers and MD_DISK_SYNC.
6697 */
6698 if (mddev->pers) {
6699 mdu_disk_info_t info;
6700 if (copy_from_user(&info, argp, sizeof(info)))
6701 err = -EFAULT;
6702 else if (!(info.state & (1<<MD_DISK_SYNC)))
6703 /* Need to clear read-only for this */
6704 break;
6705 else
6706 err = add_new_disk(mddev, &info);
3adc28d8 6707 goto unlock;
7ceb17e8
N		 6708 }
6709 break;
6710
c02c0aeb
N		 6711 case BLKROSET:
6712 if (get_user(ro, (int __user *)(arg))) {
6713 err = -EFAULT;
3adc28d8 6714 goto unlock;
c02c0aeb
N		 6715 }
6716 err = -EINVAL;
e2218350 6717
c02c0aeb
N		 6718 /* if the bdev is going readonly the value of mddev->ro
6719 * does not matter, no writes are coming
6720 */
6721 if (ro)
3adc28d8 6722 goto unlock;
e2218350 6723
c02c0aeb
N		 6724 /* are we are already prepared for writes? */
6725 if (mddev->ro != 1)
3adc28d8 6726 goto unlock;
e2218350 6727
c02c0aeb
N		 6728 /* transitioning to readauto need only happen for
6729 * arrays that call md_write_start
6730 */
6731 if (mddev->pers) {
6732 err = restart_array(mddev);
6733 if (err == 0) {
6734 mddev->ro = 2;
6735 set_disk_ro(mddev->gendisk, 0);
e2218350 6736 }
c02c0aeb 6737 }
3adc28d8 6738 goto unlock;
1da177e4
LT		 6739 }
6740
6741 /*
6742 * The remaining ioctls are changing the state of the
f91de92e 6743 * superblock, so we do not allow them on read-only arrays.
1da177e4 6744 */
326eb17d 6745 if (mddev->ro && mddev->pers) {
f91de92e
N		 6746 if (mddev->ro == 2) {
6747 mddev->ro = 0;
00bcb4ac 6748 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 6749 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N		 6750 /* mddev_unlock will wake thread */
6751 /* If a device failed while we were read-only, we
6752 * need to make sure the metadata is updated now.
6753 */
6754 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6755 mddev_unlock(mddev);
6756 wait_event(mddev->sb_wait,
6757 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6758 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
29f097c4 6759 mddev_lock_nointr(mddev);
f3378b48 6760 }
f91de92e
N		 6761 } else {
6762 err = -EROFS;
3adc28d8 6763 goto unlock;
f91de92e 6764 }
1da177e4
LT		 6765 }
6766
c02c0aeb
N		 6767 switch (cmd) {
6768 case ADD_NEW_DISK:
1da177e4 6769 {
c02c0aeb
N		 6770 mdu_disk_info_t info;
6771 if (copy_from_user(&info, argp, sizeof(info)))
6772 err = -EFAULT;
6773 else
6774 err = add_new_disk(mddev, &info);
3adc28d8 6775 goto unlock;
c02c0aeb 6776 }
1da177e4 6777
1aee41f6
GR		 6778 case CLUSTERED_DISK_NACK:
6779 if (mddev_is_clustered(mddev))
6780 md_cluster_ops->new_disk_ack(mddev, false);
6781 else
6782 err = -EINVAL;
6783 goto unlock;
6784
c02c0aeb
N		 6785 case HOT_ADD_DISK:
6786 err = hot_add_disk(mddev, new_decode_dev(arg));
3adc28d8 6787 goto unlock;
1da177e4 6788
c02c0aeb
N		 6789 case RUN_ARRAY:
6790 err = do_md_run(mddev);
3adc28d8 6791 goto unlock;
1da177e4 6792
c02c0aeb
N		 6793 case SET_BITMAP_FILE:
6794 err = set_bitmap_file(mddev, (int)arg);
3adc28d8 6795 goto unlock;
32a7627c 6796
c02c0aeb
N		 6797 default:
6798 err = -EINVAL;
3adc28d8 6799 goto unlock;
1da177e4
LT		 6800 }
6801
3adc28d8 6802unlock:
d3374825
N		 6803 if (mddev->hold_active == UNTIL_IOCTL &&
6804 err != -EINVAL)
6805 mddev->hold_active = 0;
1da177e4 6806 mddev_unlock(mddev);
3adc28d8 6807out:
1da177e4
LT		 6808 return err;
6809}
aa98aa31
AB		 6810#ifdef CONFIG_COMPAT
6811static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6812 unsigned int cmd, unsigned long arg)
6813{
6814 switch (cmd) {
6815 case HOT_REMOVE_DISK:
6816 case HOT_ADD_DISK:
6817 case SET_DISK_FAULTY:
6818 case SET_BITMAP_FILE:
6819 /* These take in integer arg, do not convert */
6820 break;
6821 default:
6822 arg = (unsigned long)compat_ptr(arg);
6823 break;
6824 }
6825
6826 return md_ioctl(bdev, mode, cmd, arg);
6827}
6828#endif /* CONFIG_COMPAT */
1da177e4 6829
a39907fa 6830static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6831{
6832 /*
6833 * Succeed if we can lock the mddev, which confirms that
6834 * it isn't being stopped right now.
6835 */
fd01b88c 6836 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6837 int err;
6838
0c098220
YL		 6839 if (!mddev)
6840 return -ENODEV;
6841
d3374825
N		 6842 if (mddev->gendisk != bdev->bd_disk) {
6843 /* we are racing with mddev_put which is discarding this
6844 * bd_disk.
6845 */
6846 mddev_put(mddev);
6847 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6848 flush_workqueue(md_misc_wq);
d3374825
N		 6849 /* Then retry the open from the top */
6850 return -ERESTARTSYS;
6851 }
6852 BUG_ON(mddev != bdev->bd_disk->private_data);
6853
c8c00a69 6854 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6855 goto out;
6856
6857 err = 0;
f2ea68cf 6858 atomic_inc(&mddev->openers);
260fa034 6859 clear_bit(MD_STILL_CLOSED, &mddev->flags);
c8c00a69 6860 mutex_unlock(&mddev->open_mutex);
1da177e4 6861
f0b4f7e2 6862 check_disk_change(bdev);
1da177e4
LT		 6863 out:
6864 return err;
6865}
6866
db2a144b 6867static void md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6868{
f72ffdd6 6869 struct mddev *mddev = disk->private_data;
1da177e4 6870
52e5f9d1 6871 BUG_ON(!mddev);
f2ea68cf 6872 atomic_dec(&mddev->openers);
1da177e4 6873 mddev_put(mddev);
1da177e4 6874}
f0b4f7e2
N		 6875
6876static int md_media_changed(struct gendisk *disk)
6877{
fd01b88c 6878 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6879
6880 return mddev->changed;
6881}
6882
6883static int md_revalidate(struct gendisk *disk)
6884{
fd01b88c 6885 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6886
6887 mddev->changed = 0;
6888 return 0;
6889}
83d5cde4 6890static const struct block_device_operations md_fops =
1da177e4
LT		 6891{
6892 .owner = THIS_MODULE,
a39907fa
AV		 6893 .open = md_open,
6894 .release = md_release,
b492b852 6895 .ioctl = md_ioctl,
aa98aa31
AB		 6896#ifdef CONFIG_COMPAT
6897 .compat_ioctl = md_compat_ioctl,
6898#endif
a885c8c4 6899 .getgeo = md_getgeo,
f0b4f7e2
N		 6900 .media_changed = md_media_changed,
6901 .revalidate_disk= md_revalidate,
1da177e4
LT		 6902};
6903
f72ffdd6 6904static int md_thread(void *arg)
1da177e4 6905{
2b8bf345 6906 struct md_thread *thread = arg;
1da177e4 6907
1da177e4
LT		 6908 /*
6909 * md_thread is a 'system-thread', it's priority should be very
6910 * high. We avoid resource deadlocks individually in each
6911 * raid personality. (RAID5 does preallocation) We also use RR and
6912 * the very same RT priority as kswapd, thus we will never get
6913 * into a priority inversion deadlock.
6914 *
6915 * we definitely have to have equal or higher priority than
6916 * bdflush, otherwise bdflush will deadlock if there are too
6917 * many dirty RAID5 blocks.
6918 */
1da177e4 6919
6985c43f 6920 allow_signal(SIGKILL);
a6fb0934 6921 while (!kthread_should_stop()) {
1da177e4 6922
93588e22
N		 6923 /* We need to wait INTERRUPTIBLE so that
6924 * we don't add to the load-average.
6925 * That means we need to be sure no signals are
6926 * pending
6927 */
6928 if (signal_pending(current))
6929 flush_signals(current);
6930
6931 wait_event_interruptible_timeout
6932 (thread->wqueue,
6933 test_bit(THREAD_WAKEUP, &thread->flags)
6934 || kthread_should_stop(),
6935 thread->timeout);
1da177e4 6936
6c987910
N		 6937 clear_bit(THREAD_WAKEUP, &thread->flags);
6938 if (!kthread_should_stop())
4ed8731d 6939 thread->run(thread);
1da177e4 6940 }
a6fb0934 6941
1da177e4
LT		 6942 return 0;
6943}
6944
2b8bf345 6945void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6946{
6947 if (thread) {
36a4e1fe 6948 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6949 set_bit(THREAD_WAKEUP, &thread->flags);
6950 wake_up(&thread->wqueue);
6951 }
6952}
6c144d31 6953EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 6954
4ed8731d
SL		 6955struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6956 struct mddev *mddev, const char *name)
1da177e4 6957{
2b8bf345 6958 struct md_thread *thread;
1da177e4 6959
2b8bf345 6960 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6961 if (!thread)
6962 return NULL;
6963
1da177e4
LT		 6964 init_waitqueue_head(&thread->wqueue);
6965
1da177e4
LT		 6966 thread->run = run;
6967 thread->mddev = mddev;
32a7627c 6968 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6969 thread->tsk = kthread_run(md_thread, thread,
6970 "%s_%s",
6971 mdname(thread->mddev),
0232605d 6972 name);
a6fb0934 6973 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6974 kfree(thread);
6975 return NULL;
6976 }
1da177e4
LT		 6977 return thread;
6978}
6c144d31 6979EXPORT_SYMBOL(md_register_thread);
1da177e4 6980
2b8bf345 6981void md_unregister_thread(struct md_thread **threadp)
1da177e4 6982{
2b8bf345 6983 struct md_thread *thread = *threadp;
e0cf8f04
N		 6984 if (!thread)
6985 return;
36a4e1fe 6986 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 6987 /* Locking ensures that mddev_unlock does not wake_up a
6988 * non-existent thread
6989 */
6990 spin_lock(&pers_lock);
6991 *threadp = NULL;
6992 spin_unlock(&pers_lock);
a6fb0934
N		 6993
6994 kthread_stop(thread->tsk);
1da177e4
LT		 6995 kfree(thread);
6996}
6c144d31 6997EXPORT_SYMBOL(md_unregister_thread);
1da177e4 6998
fd01b88c 6999void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7000{
b2d444d7 7001 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 7002 return;
6bfe0b49 7003
de393cde 7004 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 7005 return;
7006 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 7007 if (mddev->degraded)
7008 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 7009 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 7010 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7011 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7012 md_wakeup_thread(mddev->thread);
768a418d 7013 if (mddev->event_work.func)
e804ac78 7014 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 7015 md_new_event_inintr(mddev);
1da177e4 7016}
6c144d31 7017EXPORT_SYMBOL(md_error);
1da177e4
LT		 7018
7019/* seq_file implementation /proc/mdstat */
7020
7021static void status_unused(struct seq_file *seq)
7022{
7023 int i = 0;
3cb03002 7024 struct md_rdev *rdev;
1da177e4
LT		 7025
7026 seq_printf(seq, "unused devices: ");
7027
159ec1fc 7028 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 7029 char b[BDEVNAME_SIZE];
7030 i++;
7031 seq_printf(seq, "%s ",
7032 bdevname(rdev->bdev,b));
7033 }
7034 if (!i)
7035 seq_printf(seq, "<none>");
7036
7037 seq_printf(seq, "\n");
7038}
7039
f72ffdd6 7040static void status_resync(struct seq_file *seq, struct mddev *mddev)
1da177e4 7041{
dd71cf6b
N		 7042 sector_t max_sectors, resync, res;
7043 unsigned long dt, db;
7044 sector_t rt;
4588b42e
N		 7045 int scale;
7046 unsigned int per_milli;
1da177e4 7047
72f36d59
N		 7048 if (mddev->curr_resync <= 3)
7049 resync = 0;
7050 else
7051 resync = mddev->curr_resync
7052 - atomic_read(&mddev->recovery_active);
1da177e4 7053
c804cdec
N		 7054 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7055 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 7056 max_sectors = mddev->resync_max_sectors;
1da177e4 7057 else
dd71cf6b 7058 max_sectors = mddev->dev_sectors;
1da177e4 7059
403df478 7060 WARN_ON(max_sectors == 0);
4588b42e 7061 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 7062 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 7063 * u32, as those are the requirements for sector_div.
7064 * Thus 'scale' must be at least 10
7065 */
7066 scale = 10;
7067 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 7068 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 7069 scale++;
7070 }
7071 res = (resync>>scale)*1000;
dd71cf6b 7072 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 7073
7074 per_milli = res;
1da177e4 7075 {
4588b42e 7076 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 7077 seq_printf(seq, "[");
7078 for (i = 0; i < x; i++)
7079 seq_printf(seq, "=");
7080 seq_printf(seq, ">");
7081 for (i = 0; i < y; i++)
7082 seq_printf(seq, ".");
7083 seq_printf(seq, "] ");
7084 }
4588b42e 7085 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 7086 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7087 "reshape" :
61df9d91
N		 7088 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7089 "check" :
7090 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7091 "resync" : "recovery"))),
7092 per_milli/10, per_milli % 10,
dd71cf6b
N		 7093 (unsigned long long) resync/2,
7094 (unsigned long long) max_sectors/2);
1da177e4
LT		 7095
7096 /*
1da177e4
LT		 7097 * dt: time from mark until now
7098 * db: blocks written from mark until now
7099 * rt: remaining time
dd71cf6b
N		 7100 *
7101 * rt is a sector_t, so could be 32bit or 64bit.
7102 * So we divide before multiply in case it is 32bit and close
7103 * to the limit.
25985edc 7104 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 7105 * near the end of resync when the number of remaining sectors
7106 * is close to 'db'.
7107 * We then divide rt by 32 after multiplying by db to compensate.
7108 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 7109 */
7110 dt = ((jiffies - mddev->resync_mark) / HZ);
7111 if (!dt) dt++;
ff4e8d9a
N		 7112 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7113 - mddev->resync_mark_cnt;
1da177e4 7114
dd71cf6b
N		 7115 rt = max_sectors - resync; /* number of remaining sectors */
7116 sector_div(rt, db/32+1);
7117 rt *= dt;
7118 rt >>= 5;
7119
7120 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7121 ((unsigned long)rt % 60)/6);
1da177e4 7122
ff4e8d9a 7123 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 7124}
7125
7126static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7127{
7128 struct list_head *tmp;
7129 loff_t l = *pos;
fd01b88c 7130 struct mddev *mddev;
1da177e4
LT		 7131
7132 if (l >= 0x10000)
7133 return NULL;
7134 if (!l--)
7135 /* header */
7136 return (void*)1;
7137
7138 spin_lock(&all_mddevs_lock);
7139 list_for_each(tmp,&all_mddevs)
7140 if (!l--) {
fd01b88c 7141 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 7142 mddev_get(mddev);
7143 spin_unlock(&all_mddevs_lock);
7144 return mddev;
7145 }
7146 spin_unlock(&all_mddevs_lock);
7147 if (!l--)
7148 return (void*)2;/* tail */
7149 return NULL;
7150}
7151
7152static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7153{
7154 struct list_head *tmp;
fd01b88c 7155 struct mddev *next_mddev, *mddev = v;
f72ffdd6 7156
1da177e4
LT		 7157 ++*pos;
7158 if (v == (void*)2)
7159 return NULL;
7160
7161 spin_lock(&all_mddevs_lock);
7162 if (v == (void*)1)
7163 tmp = all_mddevs.next;
7164 else
7165 tmp = mddev->all_mddevs.next;
7166 if (tmp != &all_mddevs)
fd01b88c 7167 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 7168 else {
7169 next_mddev = (void*)2;
7170 *pos = 0x10000;
f72ffdd6 7171 }
1da177e4
LT		 7172 spin_unlock(&all_mddevs_lock);
7173
7174 if (v != (void*)1)
7175 mddev_put(mddev);
7176 return next_mddev;
7177
7178}
7179
7180static void md_seq_stop(struct seq_file *seq, void *v)
7181{
fd01b88c 7182 struct mddev *mddev = v;
1da177e4
LT		 7183
7184 if (mddev && v != (void*)1 && v != (void*)2)
7185 mddev_put(mddev);
7186}
7187
7188static int md_seq_show(struct seq_file *seq, void *v)
7189{
fd01b88c 7190 struct mddev *mddev = v;
dd8ac336 7191 sector_t sectors;
3cb03002 7192 struct md_rdev *rdev;
1da177e4
LT		 7193
7194 if (v == (void*)1) {
84fc4b56 7195 struct md_personality *pers;
1da177e4
LT		 7196 seq_printf(seq, "Personalities : ");
7197 spin_lock(&pers_lock);
2604b703
N		 7198 list_for_each_entry(pers, &pers_list, list)
7199 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 7200
7201 spin_unlock(&pers_lock);
7202 seq_printf(seq, "\n");
f1514638 7203 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7204 return 0;
7205 }
7206 if (v == (void*)2) {
7207 status_unused(seq);
7208 return 0;
7209 }
7210
36d091f4 7211 spin_lock(&mddev->lock);
1da177e4
LT		 7212 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7213 seq_printf(seq, "%s : %sactive", mdname(mddev),
7214 mddev->pers ? "" : "in");
7215 if (mddev->pers) {
f91de92e 7216 if (mddev->ro==1)
1da177e4 7217 seq_printf(seq, " (read-only)");
f91de92e 7218 if (mddev->ro==2)
52720ae7 7219 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 7220 seq_printf(seq, " %s", mddev->pers->name);
7221 }
7222
dd8ac336 7223 sectors = 0;
f97fcad3
N		 7224 rcu_read_lock();
7225 rdev_for_each_rcu(rdev, mddev) {
1da177e4
LT		 7226 char b[BDEVNAME_SIZE];
7227 seq_printf(seq, " %s[%d]",
7228 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 7229 if (test_bit(WriteMostly, &rdev->flags))
7230 seq_printf(seq, "(W)");
b2d444d7 7231 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7232 seq_printf(seq, "(F)");
7233 continue;
2d78f8c4
N		 7234 }
7235 if (rdev->raid_disk < 0)
b325a32e 7236 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 7237 if (test_bit(Replacement, &rdev->flags))
7238 seq_printf(seq, "(R)");
dd8ac336 7239 sectors += rdev->sectors;
1da177e4 7240 }
f97fcad3 7241 rcu_read_unlock();
1da177e4
LT		 7242
7243 if (!list_empty(&mddev->disks)) {
7244 if (mddev->pers)
7245 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 7246 (unsigned long long)
7247 mddev->array_sectors / 2);
1da177e4
LT		 7248 else
7249 seq_printf(seq, "\n %llu blocks",
dd8ac336 7250 (unsigned long long)sectors / 2);
1da177e4 7251 }
1cd6bf19
N		 7252 if (mddev->persistent) {
7253 if (mddev->major_version != 0 ||
7254 mddev->minor_version != 90) {
7255 seq_printf(seq," super %d.%d",
7256 mddev->major_version,
7257 mddev->minor_version);
7258 }
e691063a
N		 7259 } else if (mddev->external)
7260 seq_printf(seq, " super external:%s",
7261 mddev->metadata_type);
7262 else
1cd6bf19 7263 seq_printf(seq, " super non-persistent");
1da177e4
LT		 7264
7265 if (mddev->pers) {
d710e138 7266 mddev->pers->status(seq, mddev);
f72ffdd6 7267 seq_printf(seq, "\n ");
8e1b39d6
N		 7268 if (mddev->pers->sync_request) {
7269 if (mddev->curr_resync > 2) {
d710e138 7270 status_resync(seq, mddev);
8e1b39d6 7271 seq_printf(seq, "\n ");
72f36d59 7272 } else if (mddev->curr_resync >= 1)
8e1b39d6
N		 7273 seq_printf(seq, "\tresync=DELAYED\n ");
7274 else if (mddev->recovery_cp < MaxSector)
7275 seq_printf(seq, "\tresync=PENDING\n ");
7276 }
32a7627c
N		 7277 } else
7278 seq_printf(seq, "\n ");
7279
57148964 7280 bitmap_status(seq, mddev->bitmap);
1da177e4
LT		 7281
7282 seq_printf(seq, "\n");
7283 }
36d091f4 7284 spin_unlock(&mddev->lock);
f72ffdd6 7285
1da177e4
LT		 7286 return 0;
7287}
7288
110518bc 7289static const struct seq_operations md_seq_ops = {
1da177e4
LT		 7290 .start = md_seq_start,
7291 .next = md_seq_next,
7292 .stop = md_seq_stop,
7293 .show = md_seq_show,
7294};
7295
7296static int md_seq_open(struct inode *inode, struct file *file)
7297{
f1514638 7298 struct seq_file *seq;
1da177e4
LT		 7299 int error;
7300
7301 error = seq_open(file, &md_seq_ops);
d7603b7e 7302 if (error)
f1514638
KS		 7303 return error;
7304
7305 seq = file->private_data;
7306 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7307 return error;
7308}
7309
e2f23b60 7310static int md_unloading;
d7603b7e
N		 7311static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7312{
f1514638 7313 struct seq_file *seq = filp->private_data;
d7603b7e
N		 7314 int mask;
7315
e2f23b60 7316 if (md_unloading)
7d7e64f2 7317 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
d7603b7e
N		 7318 poll_wait(filp, &md_event_waiters, wait);
7319
7320 /* always allow read */
7321 mask = POLLIN | POLLRDNORM;
7322
f1514638 7323 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 7324 mask |= POLLERR | POLLPRI;
7325 return mask;
7326}
7327
fa027c2a 7328static const struct file_operations md_seq_fops = {
e24650c2 7329 .owner = THIS_MODULE,
1da177e4
LT		 7330 .open = md_seq_open,
7331 .read = seq_read,
7332 .llseek = seq_lseek,
c3f94b40 7333 .release = seq_release_private,
d7603b7e 7334 .poll = mdstat_poll,
1da177e4
LT		 7335};
7336
84fc4b56 7337int register_md_personality(struct md_personality *p)
1da177e4 7338{
50bd3774
CY		 7339 printk(KERN_INFO "md: %s personality registered for level %d\n",
7340 p->name, p->level);
1da177e4 7341 spin_lock(&pers_lock);
2604b703 7342 list_add_tail(&p->list, &pers_list);
1da177e4
LT		 7343 spin_unlock(&pers_lock);
7344 return 0;
7345}
6c144d31 7346EXPORT_SYMBOL(register_md_personality);
1da177e4 7347
84fc4b56 7348int unregister_md_personality(struct md_personality *p)
1da177e4 7349{
2604b703 7350 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7351 spin_lock(&pers_lock);
2604b703 7352 list_del_init(&p->list);
1da177e4
LT		 7353 spin_unlock(&pers_lock);
7354 return 0;
7355}
6c144d31 7356EXPORT_SYMBOL(unregister_md_personality);
1da177e4 7357
edb39c9d
GR		 7358int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module)
7359{
7360 if (md_cluster_ops != NULL)
7361 return -EALREADY;
7362 spin_lock(&pers_lock);
7363 md_cluster_ops = ops;
7364 md_cluster_mod = module;
7365 spin_unlock(&pers_lock);
7366 return 0;
7367}
7368EXPORT_SYMBOL(register_md_cluster_operations);
7369
7370int unregister_md_cluster_operations(void)
7371{
7372 spin_lock(&pers_lock);
7373 md_cluster_ops = NULL;
7374 spin_unlock(&pers_lock);
7375 return 0;
7376}
7377EXPORT_SYMBOL(unregister_md_cluster_operations);
7378
7379int md_setup_cluster(struct mddev *mddev, int nodes)
7380{
7381 int err;
7382
7383 err = request_module("md-cluster");
7384 if (err) {
7385 pr_err("md-cluster module not found.\n");
7386 return err;
7387 }
7388
7389 spin_lock(&pers_lock);
7390 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7391 spin_unlock(&pers_lock);
7392 return -ENOENT;
7393 }
7394 spin_unlock(&pers_lock);
7395
cf921cc1 7396 return md_cluster_ops->join(mddev, nodes);
edb39c9d
GR		 7397}
7398
7399void md_cluster_stop(struct mddev *mddev)
7400{
c4ce867f
GR		 7401 if (!md_cluster_ops)
7402 return;
edb39c9d
GR		 7403 md_cluster_ops->leave(mddev);
7404 module_put(md_cluster_mod);
7405}
7406
fd01b88c 7407static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7408{
f72ffdd6 7409 struct md_rdev *rdev;
1da177e4 7410 int idle;
eea1bf38 7411 int curr_events;
1da177e4
LT		 7412
7413 idle = 1;
4b80991c
N		 7414 rcu_read_lock();
7415 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7416 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 7417 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7418 (int)part_stat_read(&disk->part0, sectors[1]) -
7419 atomic_read(&disk->sync_io);
713f6ab1
N		 7420 /* sync IO will cause sync_io to increase before the disk_stats
7421 * as sync_io is counted when a request starts, and
7422 * disk_stats is counted when it completes.
7423 * So resync activity will cause curr_events to be smaller than
7424 * when there was no such activity.
7425 * non-sync IO will cause disk_stat to increase without
7426 * increasing sync_io so curr_events will (eventually)
7427 * be larger than it was before. Once it becomes
7428 * substantially larger, the test below will cause
7429 * the array to appear non-idle, and resync will slow
7430 * down.
7431 * If there is a lot of outstanding resync activity when
7432 * we set last_event to curr_events, then all that activity
7433 * completing might cause the array to appear non-idle
7434 * and resync will be slowed down even though there might
7435 * not have been non-resync activity. This will only
7436 * happen once though. 'last_events' will soon reflect
7437 * the state where there is little or no outstanding
7438 * resync requests, and further resync activity will
7439 * always make curr_events less than last_events.
c0e48521 7440 *
1da177e4 7441 */
eea1bf38 7442 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 7443 rdev->last_events = curr_events;
7444 idle = 0;
7445 }
7446 }
4b80991c 7447 rcu_read_unlock();
1da177e4
LT		 7448 return idle;
7449}
7450
fd01b88c 7451void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 7452{
7453 /* another "blocks" (512byte) blocks have been synced */
7454 atomic_sub(blocks, &mddev->recovery_active);
7455 wake_up(&mddev->recovery_wait);
7456 if (!ok) {
dfc70645 7457 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 7458 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT		 7459 md_wakeup_thread(mddev->thread);
7460 // stop recovery, signal do_sync ....
7461 }
7462}
6c144d31 7463EXPORT_SYMBOL(md_done_sync);
1da177e4 7464
06d91a5f
N		 7465/* md_write_start(mddev, bi)
7466 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 7467 * in superblock) before writing, schedule a superblock update
7468 * and wait for it to complete.
06d91a5f 7469 */
fd01b88c 7470void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7471{
0fd62b86 7472 int did_change = 0;
06d91a5f 7473 if (bio_data_dir(bi) != WRITE)
3d310eb7 7474 return;
06d91a5f 7475
f91de92e
N		 7476 BUG_ON(mddev->ro == 1);
7477 if (mddev->ro == 2) {
7478 /* need to switch to read/write */
7479 mddev->ro = 0;
7480 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7481 md_wakeup_thread(mddev->thread);
25156198 7482 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7483 did_change = 1;
f91de92e 7484 }
06d91a5f 7485 atomic_inc(&mddev->writes_pending);
31a59e34
N		 7486 if (mddev->safemode == 1)
7487 mddev->safemode = 0;
06d91a5f 7488 if (mddev->in_sync) {
85572d7c 7489 spin_lock(&mddev->lock);
3d310eb7
N		 7490 if (mddev->in_sync) {
7491 mddev->in_sync = 0;
850b2b42 7492 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7493 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7494 md_wakeup_thread(mddev->thread);
0fd62b86 7495 did_change = 1;
3d310eb7 7496 }
85572d7c 7497 spin_unlock(&mddev->lock);
06d91a5f 7498 }
0fd62b86 7499 if (did_change)
00bcb4ac 7500 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7501 wait_event(mddev->sb_wait,
09a44cc1 7502 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4 7503}
6c144d31 7504EXPORT_SYMBOL(md_write_start);
1da177e4 7505
fd01b88c 7506void md_write_end(struct mddev *mddev)
1da177e4
LT		 7507{
7508 if (atomic_dec_and_test(&mddev->writes_pending)) {
7509 if (mddev->safemode == 2)
7510 md_wakeup_thread(mddev->thread);
16f17b39 7511 else if (mddev->safemode_delay)
1da177e4
LT		 7512 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7513 }
7514}
6c144d31 7515EXPORT_SYMBOL(md_write_end);
1da177e4 7516
2a2275d6
N		 7517/* md_allow_write(mddev)
7518 * Calling this ensures that the array is marked 'active' so that writes
7519 * may proceed without blocking. It is important to call this before
7520 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7521 * Must be called with mddev_lock held.
b5470dc5
DW		 7522 *
7523 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7524 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7525 */
fd01b88c 7526int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7527{
7528 if (!mddev->pers)
b5470dc5 7529 return 0;
2a2275d6 7530 if (mddev->ro)
b5470dc5 7531 return 0;
1a0fd497 7532 if (!mddev->pers->sync_request)
b5470dc5 7533 return 0;
2a2275d6 7534
85572d7c 7535 spin_lock(&mddev->lock);
2a2275d6
N		 7536 if (mddev->in_sync) {
7537 mddev->in_sync = 0;
7538 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7539 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7540 if (mddev->safemode_delay &&
7541 mddev->safemode == 0)
7542 mddev->safemode = 1;
85572d7c 7543 spin_unlock(&mddev->lock);
293467aa
GR		 7544 if (mddev_is_clustered(mddev))
7545 md_cluster_ops->metadata_update_start(mddev);
2a2275d6 7546 md_update_sb(mddev, 0);
293467aa
GR		 7547 if (mddev_is_clustered(mddev))
7548 md_cluster_ops->metadata_update_finish(mddev);
00bcb4ac 7549 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6 7550 } else
85572d7c 7551 spin_unlock(&mddev->lock);
b5470dc5 7552
070dc6dd 7553 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7554 return -EAGAIN;
7555 else
7556 return 0;
2a2275d6
N		 7557}
7558EXPORT_SYMBOL_GPL(md_allow_write);
7559
1da177e4
LT		 7560#define SYNC_MARKS 10
7561#define SYNC_MARK_STEP (3*HZ)
54f89341 7562#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 7563void md_do_sync(struct md_thread *thread)
1da177e4 7564{
4ed8731d 7565 struct mddev *mddev = thread->mddev;
fd01b88c 7566 struct mddev *mddev2;
1da177e4
LT		 7567 unsigned int currspeed = 0,
7568 window;
ac7e50a3 7569 sector_t max_sectors,j, io_sectors, recovery_done;
1da177e4 7570 unsigned long mark[SYNC_MARKS];
54f89341 7571 unsigned long update_time;
1da177e4
LT		 7572 sector_t mark_cnt[SYNC_MARKS];
7573 int last_mark,m;
7574 struct list_head *tmp;
7575 sector_t last_check;
57afd89f 7576 int skipped = 0;
3cb03002 7577 struct md_rdev *rdev;
c4a39551 7578 char *desc, *action = NULL;
7c2c57c9 7579 struct blk_plug plug;
1da177e4
LT		 7580
7581 /* just incase thread restarts... */
7582 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7583 return;
3991b31e
N		 7584 if (mddev->ro) {/* never try to sync a read-only array */
7585 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5fd6c1dc 7586 return;
3991b31e 7587 }
1da177e4 7588
61df9d91 7589 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
c4a39551 7590 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
61df9d91 7591 desc = "data-check";
c4a39551
JB		 7592 action = "check";
7593 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
61df9d91 7594 desc = "requested-resync";
c4a39551
JB		 7595 action = "repair";
7596 } else
61df9d91
N		 7597 desc = "resync";
7598 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7599 desc = "reshape";
7600 else
7601 desc = "recovery";
7602
c4a39551
JB		 7603 mddev->last_sync_action = action ?: desc;
7604
1da177e4
LT		 7605 /* we overload curr_resync somewhat here.
7606 * 0 == not engaged in resync at all
7607 * 2 == checking that there is no conflict with another sync
7608 * 1 == like 2, but have yielded to allow conflicting resync to
7609 * commense
7610 * other == active in resync - this many blocks
7611 *
7612 * Before starting a resync we must have set curr_resync to
7613 * 2, and then checked that every "conflicting" array has curr_resync
7614 * less than ours. When we find one that is the same or higher
7615 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7616 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7617 * This will mean we have to start checking from the beginning again.
7618 *
7619 */
7620
7621 do {
7622 mddev->curr_resync = 2;
7623
7624 try_again:
404e4b43 7625 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7626 goto skip;
29ac4aa3 7627 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7628 if (mddev2 == mddev)
7629 continue;
90b08710
BS		 7630 if (!mddev->parallel_resync
7631 && mddev2->curr_resync
7632 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7633 DEFINE_WAIT(wq);
7634 if (mddev < mddev2 && mddev->curr_resync == 2) {
7635 /* arbitrarily yield */
7636 mddev->curr_resync = 1;
7637 wake_up(&resync_wait);
7638 }
7639 if (mddev > mddev2 && mddev->curr_resync == 1)
7640 /* no need to wait here, we can wait the next
7641 * time 'round when curr_resync == 2
7642 */
7643 continue;
9744197c
N		 7644 /* We need to wait 'interruptible' so as not to
7645 * contribute to the load average, and not to
7646 * be caught by 'softlockup'
7647 */
7648 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
c91abf5a 7649 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8712e553 7650 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7651 printk(KERN_INFO "md: delaying %s of %s"
7652 " until %s has finished (they"
1da177e4 7653 " share one or more physical units)\n",
61df9d91 7654 desc, mdname(mddev), mdname(mddev2));
1da177e4 7655 mddev_put(mddev2);
9744197c
N		 7656 if (signal_pending(current))
7657 flush_signals(current);
1da177e4
LT		 7658 schedule();
7659 finish_wait(&resync_wait, &wq);
7660 goto try_again;
7661 }
7662 finish_wait(&resync_wait, &wq);
7663 }
7664 }
7665 } while (mddev->curr_resync < 2);
7666
5fd6c1dc 7667 j = 0;
9d88883e 7668 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7669 /* resync follows the size requested by the personality,
57afd89f 7670 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7671 */
7672 max_sectors = mddev->resync_max_sectors;
7f7583d4 7673 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 7674 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7675 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7676 j = mddev->resync_min;
7677 else if (!mddev->bitmap)
5fd6c1dc 7678 j = mddev->recovery_cp;
5e96ee65 7679
ccfcc3c1 7680 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7681 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7682 else {
1da177e4 7683 /* recovery follows the physical size of devices */
58c0fed4 7684 max_sectors = mddev->dev_sectors;
5fd6c1dc 7685 j = MaxSector;
4e59ca7d 7686 rcu_read_lock();
dafb20fa 7687 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc
N		 7688 if (rdev->raid_disk >= 0 &&
7689 !test_bit(Faulty, &rdev->flags) &&
7690 !test_bit(In_sync, &rdev->flags) &&
7691 rdev->recovery_offset < j)
7692 j = rdev->recovery_offset;
4e59ca7d 7693 rcu_read_unlock();
133d4527
N		 7694
7695 /* If there is a bitmap, we need to make sure all
7696 * writes that started before we added a spare
7697 * complete before we start doing a recovery.
7698 * Otherwise the write might complete and (via
7699 * bitmap_endwrite) set a bit in the bitmap after the
7700 * recovery has checked that bit and skipped that
7701 * region.
7702 */
7703 if (mddev->bitmap) {
7704 mddev->pers->quiesce(mddev, 1);
7705 mddev->pers->quiesce(mddev, 0);
7706 }
5fd6c1dc 7707 }
1da177e4 7708
61df9d91
N		 7709 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7710 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7711 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7712 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7713 "(but not more than %d KB/sec) for %s.\n",
7714 speed_max(mddev), desc);
1da177e4 7715
eea1bf38 7716 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7717
57afd89f 7718 io_sectors = 0;
1da177e4
LT		 7719 for (m = 0; m < SYNC_MARKS; m++) {
7720 mark[m] = jiffies;
57afd89f 7721 mark_cnt[m] = io_sectors;
1da177e4
LT		 7722 }
7723 last_mark = 0;
7724 mddev->resync_mark = mark[last_mark];
7725 mddev->resync_mark_cnt = mark_cnt[last_mark];
7726
7727 /*
7728 * Tune reconstruction:
7729 */
7730 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7731 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7732 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7733
7734 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7735 last_check = 0;
7736
7737 if (j>2) {
c91abf5a 7738 printk(KERN_INFO
61df9d91
N		 7739 "md: resuming %s of %s from checkpoint.\n",
7740 desc, mdname(mddev));
1da177e4 7741 mddev->curr_resync = j;
72f36d59
N		 7742 } else
7743 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 7744 mddev->curr_resync_completed = j;
72f36d59
N		 7745 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7746 md_new_event(mddev);
54f89341 7747 update_time = jiffies;
1da177e4 7748
96ae923a 7749 if (mddev_is_clustered(mddev))
965400eb 7750 md_cluster_ops->resync_start(mddev, j, max_sectors);
96ae923a 7751
7c2c57c9 7752 blk_start_plug(&plug);
1da177e4 7753 while (j < max_sectors) {
57afd89f 7754 sector_t sectors;
1da177e4 7755
57afd89f 7756 skipped = 0;
97e4f42d 7757
7a91ee1f
N		 7758 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7759 ((mddev->curr_resync > mddev->curr_resync_completed &&
7760 (mddev->curr_resync - mddev->curr_resync_completed)
7761 > (max_sectors >> 4)) ||
54f89341 7762 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f
N		 7763 (j - mddev->curr_resync_completed)*2
7764 >= mddev->resync_max - mddev->curr_resync_completed
7765 )) {
97e4f42d 7766 /* time to update curr_resync_completed */
97e4f42d
N		 7767 wait_event(mddev->recovery_wait,
7768 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7769 mddev->curr_resync_completed = j;
35d78c66 7770 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7771 j > mddev->recovery_cp)
7772 mddev->recovery_cp = j;
54f89341 7773 update_time = jiffies;
070dc6dd 7774 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7775 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7776 }
acb180b0 7777
c91abf5a
N		 7778 while (j >= mddev->resync_max &&
7779 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
e62e58a5
N		 7780 /* As this condition is controlled by user-space,
7781 * we can block indefinitely, so use '_interruptible'
7782 * to avoid triggering warnings.
7783 */
7784 flush_signals(current); /* just in case */
7785 wait_event_interruptible(mddev->recovery_wait,
7786 mddev->resync_max > j
c91abf5a
N		 7787 || test_bit(MD_RECOVERY_INTR,
7788 &mddev->recovery));
e62e58a5 7789 }
acb180b0 7790
c91abf5a
N		 7791 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7792 break;
acb180b0 7793
57afd89f 7794 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7795 currspeed < speed_min(mddev));
57afd89f 7796 if (sectors == 0) {
dfc70645 7797 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
c91abf5a 7798 break;
1da177e4 7799 }
57afd89f
N		 7800
7801 if (!skipped) { /* actual IO requested */
7802 io_sectors += sectors;
7803 atomic_add(sectors, &mddev->recovery_active);
7804 }
7805
e875ecea
N		 7806 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7807 break;
7808
1da177e4 7809 j += sectors;
72f36d59
N		 7810 if (j > 2)
7811 mddev->curr_resync = j;
96ae923a
GR		 7812 if (mddev_is_clustered(mddev))
7813 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
ff4e8d9a 7814 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7815 if (last_check == 0)
e875ecea 7816 /* this is the earliest that rebuild will be
d7603b7e
N		 7817 * visible in /proc/mdstat
7818 */
7819 md_new_event(mddev);
57afd89f
N		 7820
7821 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7822 continue;
7823
57afd89f 7824 last_check = io_sectors;
1da177e4
LT		 7825 repeat:
7826 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7827 /* step marks */
7828 int next = (last_mark+1) % SYNC_MARKS;
7829
7830 mddev->resync_mark = mark[next];
7831 mddev->resync_mark_cnt = mark_cnt[next];
7832 mark[next] = jiffies;
57afd89f 7833 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7834 last_mark = next;
7835 }
7836
c91abf5a
N		 7837 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7838 break;
1da177e4
LT		 7839
7840 /*
7841 * this loop exits only if either when we are slower than
7842 * the 'hard' speed limit, or the system was IO-idle for
7843 * a jiffy.
7844 * the system might be non-idle CPU-wise, but we only care
7845 * about not overloading the IO subsystem. (things like an
7846 * e2fsck being done on the RAID array should execute fast)
7847 */
1da177e4
LT		 7848 cond_resched();
7849
ac7e50a3
XN		 7850 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7851 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
57afd89f 7852 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7853
88202a0c
N		 7854 if (currspeed > speed_min(mddev)) {
7855 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7856 !is_mddev_idle(mddev, 0)) {
c0e48521 7857 msleep(500);
1da177e4
LT		 7858 goto repeat;
7859 }
7860 }
7861 }
c91abf5a
N		 7862 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7863 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7864 ? "interrupted" : "done");
1da177e4
LT		 7865 /*
7866 * this also signals 'finished resyncing' to md_stop
7867 */
7c2c57c9 7868 blk_finish_plug(&plug);
1da177e4
LT		 7869 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7870
7871 /* tell personality that we are finished */
57afd89f 7872 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7873
96ae923a 7874 if (mddev_is_clustered(mddev))
965400eb 7875 md_cluster_ops->resync_finish(mddev);
96ae923a 7876
dfc70645 7877 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7878 mddev->curr_resync > 2) {
7879 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7880 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7881 if (mddev->curr_resync >= mddev->recovery_cp) {
7882 printk(KERN_INFO
61df9d91
N		 7883 "md: checkpointing %s of %s.\n",
7884 desc, mdname(mddev));
0a19caab 7885 if (test_bit(MD_RECOVERY_ERROR,
7886 &mddev->recovery))
7887 mddev->recovery_cp =
7888 mddev->curr_resync_completed;
7889 else
7890 mddev->recovery_cp =
7891 mddev->curr_resync;
5fd6c1dc
N		 7892 }
7893 } else
7894 mddev->recovery_cp = MaxSector;
7895 } else {
7896 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7897 mddev->curr_resync = MaxSector;
4e59ca7d 7898 rcu_read_lock();
dafb20fa 7899 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7900 if (rdev->raid_disk >= 0 &&
70fffd0b 7901 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7902 !test_bit(Faulty, &rdev->flags) &&
7903 !test_bit(In_sync, &rdev->flags) &&
7904 rdev->recovery_offset < mddev->curr_resync)
7905 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7906 rcu_read_unlock();
5fd6c1dc 7907 }
1da177e4 7908 }
db91ff55 7909 skip:
17571284 7910 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7911
23da422b 7912 spin_lock(&mddev->lock);
c07b70ad
N		 7913 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7914 /* We completed so min/max setting can be forgotten if used. */
7915 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7916 mddev->resync_min = 0;
7917 mddev->resync_max = MaxSector;
7918 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7919 mddev->resync_min = mddev->curr_resync_completed;
1da177e4 7920 mddev->curr_resync = 0;
23da422b
N		 7921 spin_unlock(&mddev->lock);
7922
1da177e4
LT		 7923 wake_up(&resync_wait);
7924 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7925 md_wakeup_thread(mddev->thread);
c6207277 7926 return;
1da177e4 7927}
29269553 7928EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7929
746d3207
N		 7930static int remove_and_add_spares(struct mddev *mddev,
7931 struct md_rdev *this)
b4c4c7b8 7932{
3cb03002 7933 struct md_rdev *rdev;
b4c4c7b8 7934 int spares = 0;
f2a371c5 7935 int removed = 0;
b4c4c7b8 7936
dafb20fa 7937 rdev_for_each(rdev, mddev)
746d3207
N		 7938 if ((this == NULL || rdev == this) &&
7939 rdev->raid_disk >= 0 &&
6bfe0b49 7940 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7941 (test_bit(Faulty, &rdev->flags) ||
7942 ! test_bit(In_sync, &rdev->flags)) &&
7943 atomic_read(&rdev->nr_pending)==0) {
7944 if (mddev->pers->hot_remove_disk(
b8321b68 7945 mddev, rdev) == 0) {
36fad858 7946 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 7947 rdev->raid_disk = -1;
f2a371c5 7948 removed++;
b4c4c7b8
N		 7949 }
7950 }
90584fc9
JB		 7951 if (removed && mddev->kobj.sd)
7952 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 7953
746d3207
N		 7954 if (this)
7955 goto no_add;
7956
dafb20fa 7957 rdev_for_each(rdev, mddev) {
7bfec5f3
N		 7958 if (rdev->raid_disk >= 0 &&
7959 !test_bit(In_sync, &rdev->flags) &&
7960 !test_bit(Faulty, &rdev->flags))
7961 spares++;
7ceb17e8
N		 7962 if (rdev->raid_disk >= 0)
7963 continue;
7964 if (test_bit(Faulty, &rdev->flags))
7965 continue;
7966 if (mddev->ro &&
8313b8e5
N		 7967 ! (rdev->saved_raid_disk >= 0 &&
7968 !test_bit(Bitmap_sync, &rdev->flags)))
7ceb17e8
N		 7969 continue;
7970
7eb41885
N		 7971 if (rdev->saved_raid_disk < 0)
7972 rdev->recovery_offset = 0;
7ceb17e8
N		 7973 if (mddev->pers->
7974 hot_add_disk(mddev, rdev) == 0) {
7975 if (sysfs_link_rdev(mddev, rdev))
7976 /* failure here is OK */;
7977 spares++;
7978 md_new_event(mddev);
7979 set_bit(MD_CHANGE_DEVS, &mddev->flags);
dfc70645 7980 }
b4c4c7b8 7981 }
746d3207 7982no_add:
6dafab6b
N		 7983 if (removed)
7984 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 7985 return spares;
7986}
7ebc0be7 7987
ac05f256
N		 7988static void md_start_sync(struct work_struct *ws)
7989{
7990 struct mddev *mddev = container_of(ws, struct mddev, del_work);
7991
7992 mddev->sync_thread = md_register_thread(md_do_sync,
7993 mddev,
7994 "resync");
7995 if (!mddev->sync_thread) {
7996 printk(KERN_ERR "%s: could not start resync"
7997 " thread...\n",
7998 mdname(mddev));
7999 /* leave the spares where they are, it shouldn't hurt */
8000 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8001 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8002 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8003 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8004 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8005 wake_up(&resync_wait);
ac05f256
N		 8006 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8007 &mddev->recovery))
8008 if (mddev->sysfs_action)
8009 sysfs_notify_dirent_safe(mddev->sysfs_action);
8010 } else
8011 md_wakeup_thread(mddev->sync_thread);
8012 sysfs_notify_dirent_safe(mddev->sysfs_action);
8013 md_new_event(mddev);
8014}
8015
1da177e4
LT		 8016/*
8017 * This routine is regularly called by all per-raid-array threads to
8018 * deal with generic issues like resync and super-block update.
8019 * Raid personalities that don't have a thread (linear/raid0) do not
8020 * need this as they never do any recovery or update the superblock.
8021 *
8022 * It does not do any resync itself, but rather "forks" off other threads
8023 * to do that as needed.
8024 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8025 * "->recovery" and create a thread at ->sync_thread.
dfc70645 8026 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 8027 * and wakeups up this thread which will reap the thread and finish up.
8028 * This thread also removes any faulty devices (with nr_pending == 0).
8029 *
8030 * The overall approach is:
8031 * 1/ if the superblock needs updating, update it.
8032 * 2/ If a recovery thread is running, don't do anything else.
8033 * 3/ If recovery has finished, clean up, possibly marking spares active.
8034 * 4/ If there are any faulty devices, remove them.
8035 * 5/ If array is degraded, try to add spares devices
8036 * 6/ If array has spares or is not in-sync, start a resync thread.
8037 */
fd01b88c 8038void md_check_recovery(struct mddev *mddev)
1da177e4 8039{
68866e42
JB		 8040 if (mddev->suspended)
8041 return;
8042
5f40402d 8043 if (mddev->bitmap)
aa5cbd10 8044 bitmap_daemon_work(mddev);
1da177e4 8045
fca4d848 8046 if (signal_pending(current)) {
31a59e34 8047 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 8048 printk(KERN_INFO "md: %s in immediate safe mode\n",
8049 mdname(mddev));
8050 mddev->safemode = 2;
8051 }
8052 flush_signals(current);
8053 }
8054
c89a8eee
N		 8055 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8056 return;
1da177e4 8057 if ( ! (
142d44c3 8058 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 8059 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 8060 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 8061 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 8062 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8063 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 8064 ))
8065 return;
fca4d848 8066
df5b89b3 8067 if (mddev_trylock(mddev)) {
b4c4c7b8 8068 int spares = 0;
fca4d848 8069
c89a8eee 8070 if (mddev->ro) {
7ceb17e8
N		 8071 /* On a read-only array we can:
8072 * - remove failed devices
8073 * - add already-in_sync devices if the array itself
8074 * is in-sync.
8075 * As we only add devices that are already in-sync,
8076 * we can activate the spares immediately.
c89a8eee 8077 */
7ceb17e8 8078 remove_and_add_spares(mddev, NULL);
8313b8e5
N		 8079 /* There is no thread, but we need to call
8080 * ->spare_active and clear saved_raid_disk
8081 */
2ac295a5 8082 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8313b8e5
N		 8083 md_reap_sync_thread(mddev);
8084 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
c89a8eee
N		 8085 goto unlock;
8086 }
8087
31a59e34 8088 if (!mddev->external) {
0fd62b86 8089 int did_change = 0;
85572d7c 8090 spin_lock(&mddev->lock);
31a59e34
N		 8091 if (mddev->safemode &&
8092 !atomic_read(&mddev->writes_pending) &&
8093 !mddev->in_sync &&
8094 mddev->recovery_cp == MaxSector) {
8095 mddev->in_sync = 1;
0fd62b86 8096 did_change = 1;
070dc6dd 8097 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 8098 }
8099 if (mddev->safemode == 1)
8100 mddev->safemode = 0;
85572d7c 8101 spin_unlock(&mddev->lock);
0fd62b86 8102 if (did_change)
00bcb4ac 8103 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 8104 }
fca4d848 8105
293467aa
GR		 8106 if (mddev->flags & MD_UPDATE_SB_FLAGS) {
8107 if (mddev_is_clustered(mddev))
8108 md_cluster_ops->metadata_update_start(mddev);
850b2b42 8109 md_update_sb(mddev, 0);
293467aa
GR		 8110 if (mddev_is_clustered(mddev))
8111 md_cluster_ops->metadata_update_finish(mddev);
8112 }
06d91a5f 8113
1da177e4
LT		 8114 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8115 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8116 /* resync/recovery still happening */
8117 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8118 goto unlock;
8119 }
8120 if (mddev->sync_thread) {
a91d5ac0 8121 md_reap_sync_thread(mddev);
1da177e4
LT		 8122 goto unlock;
8123 }
72a23c21
NB		 8124 /* Set RUNNING before clearing NEEDED to avoid
8125 * any transients in the value of "sync_action".
8126 */
72f36d59 8127 mddev->curr_resync_completed = 0;
23da422b 8128 spin_lock(&mddev->lock);
72a23c21 8129 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
23da422b 8130 spin_unlock(&mddev->lock);
24dd469d
N		 8131 /* Clear some bits that don't mean anything, but
8132 * might be left set
8133 */
24dd469d
N		 8134 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8135 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 8136
ed209584
N		 8137 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8138 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
ac05f256 8139 goto not_running;
1da177e4
LT		 8140 /* no recovery is running.
8141 * remove any failed drives, then
8142 * add spares if possible.
72f36d59 8143 * Spares are also removed and re-added, to allow
1da177e4
LT		 8144 * the personality to fail the re-add.
8145 */
1da177e4 8146
b4c4c7b8 8147 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 8148 if (mddev->pers->check_reshape == NULL ||
8149 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8 8150 /* Cannot proceed */
ac05f256 8151 goto not_running;
b4c4c7b8 8152 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 8153 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 8154 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N		 8155 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8156 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 8157 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 8158 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8159 } else if (mddev->recovery_cp < MaxSector) {
8160 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 8161 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8162 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8163 /* nothing to be done ... */
ac05f256 8164 goto not_running;
24dd469d 8165
1da177e4 8166 if (mddev->pers->sync_request) {
ef99bf48 8167 if (spares) {
a654b9d8
N		 8168 /* We are adding a device or devices to an array
8169 * which has the bitmap stored on all devices.
8170 * So make sure all bitmap pages get written
8171 */
8172 bitmap_write_all(mddev->bitmap);
8173 }
ac05f256
N		 8174 INIT_WORK(&mddev->del_work, md_start_sync);
8175 queue_work(md_misc_wq, &mddev->del_work);
8176 goto unlock;
1da177e4 8177 }
ac05f256 8178 not_running:
72a23c21
NB		 8179 if (!mddev->sync_thread) {
8180 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8181 wake_up(&resync_wait);
72a23c21
NB		 8182 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8183 &mddev->recovery))
0c3573f1 8184 if (mddev->sysfs_action)
00bcb4ac 8185 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 8186 }
ac05f256
N		 8187 unlock:
8188 wake_up(&mddev->sb_wait);
1da177e4
LT		 8189 mddev_unlock(mddev);
8190 }
8191}
6c144d31 8192EXPORT_SYMBOL(md_check_recovery);
1da177e4 8193
a91d5ac0
JB		 8194void md_reap_sync_thread(struct mddev *mddev)
8195{
8196 struct md_rdev *rdev;
8197
8198 /* resync has finished, collect result */
8199 md_unregister_thread(&mddev->sync_thread);
8200 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8201 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8202 /* success...*/
8203 /* activate any spares */
8204 if (mddev->pers->spare_active(mddev)) {
8205 sysfs_notify(&mddev->kobj, NULL,
8206 "degraded");
8207 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8208 }
8209 }
293467aa
GR		 8210 if (mddev_is_clustered(mddev))
8211 md_cluster_ops->metadata_update_start(mddev);
a91d5ac0
JB		 8212 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8213 mddev->pers->finish_reshape)
8214 mddev->pers->finish_reshape(mddev);
8215
8216 /* If array is no-longer degraded, then any saved_raid_disk
f466722c 8217 * information must be scrapped.
a91d5ac0 8218 */
f466722c
N		 8219 if (!mddev->degraded)
8220 rdev_for_each(rdev, mddev)
a91d5ac0
JB		 8221 rdev->saved_raid_disk = -1;
8222
8223 md_update_sb(mddev, 1);
293467aa
GR		 8224 if (mddev_is_clustered(mddev))
8225 md_cluster_ops->metadata_update_finish(mddev);
a91d5ac0
JB		 8226 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8227 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8228 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8229 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8230 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
f851b60d 8231 wake_up(&resync_wait);
a91d5ac0
JB		 8232 /* flag recovery needed just to double check */
8233 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8234 sysfs_notify_dirent_safe(mddev->sysfs_action);
8235 md_new_event(mddev);
8236 if (mddev->event_work.func)
8237 queue_work(md_misc_wq, &mddev->event_work);
8238}
6c144d31 8239EXPORT_SYMBOL(md_reap_sync_thread);
a91d5ac0 8240
fd01b88c 8241void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 8242{
00bcb4ac 8243 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 8244 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 8245 !test_bit(Blocked, &rdev->flags) &&
8246 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 8247 msecs_to_jiffies(5000));
8248 rdev_dec_pending(rdev, mddev);
8249}
8250EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8251
c6563a8c
N		 8252void md_finish_reshape(struct mddev *mddev)
8253{
8254 /* called be personality module when reshape completes. */
8255 struct md_rdev *rdev;
8256
8257 rdev_for_each(rdev, mddev) {
8258 if (rdev->data_offset > rdev->new_data_offset)
8259 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8260 else
8261 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8262 rdev->data_offset = rdev->new_data_offset;
8263 }
8264}
8265EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N		 8266
8267/* Bad block management.
8268 * We can record which blocks on each device are 'bad' and so just
8269 * fail those blocks, or that stripe, rather than the whole device.
8270 * Entries in the bad-block table are 64bits wide. This comprises:
8271 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8272 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8273 * A 'shift' can be set so that larger blocks are tracked and
8274 * consequently larger devices can be covered.
8275 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8276 *
8277 * Locking of the bad-block table uses a seqlock so md_is_badblock
8278 * might need to retry if it is very unlucky.
8279 * We will sometimes want to check for bad blocks in a bi_end_io function,
8280 * so we use the write_seqlock_irq variant.
8281 *
8282 * When looking for a bad block we specify a range and want to
8283 * know if any block in the range is bad. So we binary-search
8284 * to the last range that starts at-or-before the given endpoint,
8285 * (or "before the sector after the target range")
8286 * then see if it ends after the given start.
8287 * We return
8288 * 0 if there are no known bad blocks in the range
8289 * 1 if there are known bad block which are all acknowledged
8290 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8291 * plus the start/length of the first bad section we overlap.
8292 */
8293int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8294 sector_t *first_bad, int *bad_sectors)
8295{
8296 int hi;
ab05613a 8297 int lo;
2230dfe4 8298 u64 *p = bb->page;
ab05613a 8299 int rv;
2230dfe4
N		 8300 sector_t target = s + sectors;
8301 unsigned seq;
8302
8303 if (bb->shift > 0) {
8304 /* round the start down, and the end up */
8305 s >>= bb->shift;
8306 target += (1<<bb->shift) - 1;
8307 target >>= bb->shift;
8308 sectors = target - s;
8309 }
8310 /* 'target' is now the first block after the bad range */
8311
8312retry:
8313 seq = read_seqbegin(&bb->lock);
ab05613a 8314 lo = 0;
8315 rv = 0;
2230dfe4
N		 8316 hi = bb->count;
8317
8318 /* Binary search between lo and hi for 'target'
8319 * i.e. for the last range that starts before 'target'
8320 */
8321 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8322 * are known not to be the last range before target.
8323 * VARIANT: hi-lo is the number of possible
8324 * ranges, and decreases until it reaches 1
8325 */
8326 while (hi - lo > 1) {
8327 int mid = (lo + hi) / 2;
8328 sector_t a = BB_OFFSET(p[mid]);
8329 if (a < target)
8330 /* This could still be the one, earlier ranges
8331 * could not. */
8332 lo = mid;
8333 else
8334 /* This and later ranges are definitely out. */
8335 hi = mid;
8336 }
8337 /* 'lo' might be the last that started before target, but 'hi' isn't */
8338 if (hi > lo) {
8339 /* need to check all range that end after 's' to see if
8340 * any are unacknowledged.
8341 */
8342 while (lo >= 0 &&
8343 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8344 if (BB_OFFSET(p[lo]) < target) {
8345 /* starts before the end, and finishes after
8346 * the start, so they must overlap
8347 */
8348 if (rv != -1 && BB_ACK(p[lo]))
8349 rv = 1;
8350 else
8351 rv = -1;
8352 *first_bad = BB_OFFSET(p[lo]);
8353 *bad_sectors = BB_LEN(p[lo]);
8354 }
8355 lo--;
8356 }
8357 }
8358
8359 if (read_seqretry(&bb->lock, seq))
8360 goto retry;
8361
8362 return rv;
8363}
8364EXPORT_SYMBOL_GPL(md_is_badblock);
8365
8366/*
8367 * Add a range of bad blocks to the table.
8368 * This might extend the table, or might contract it
8369 * if two adjacent ranges can be merged.
8370 * We binary-search to find the 'insertion' point, then
8371 * decide how best to handle it.
8372 */
8373static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8374 int acknowledged)
8375{
8376 u64 *p;
8377 int lo, hi;
8378 int rv = 1;
905b0297 8379 unsigned long flags;
2230dfe4
N		 8380
8381 if (bb->shift < 0)
8382 /* badblocks are disabled */
8383 return 0;
8384
8385 if (bb->shift) {
8386 /* round the start down, and the end up */
8387 sector_t next = s + sectors;
8388 s >>= bb->shift;
8389 next += (1<<bb->shift) - 1;
8390 next >>= bb->shift;
8391 sectors = next - s;
8392 }
8393
905b0297 8394 write_seqlock_irqsave(&bb->lock, flags);
2230dfe4
N		 8395
8396 p = bb->page;
8397 lo = 0;
8398 hi = bb->count;
8399 /* Find the last range that starts at-or-before 's' */
8400 while (hi - lo > 1) {
8401 int mid = (lo + hi) / 2;
8402 sector_t a = BB_OFFSET(p[mid]);
8403 if (a <= s)
8404 lo = mid;
8405 else
8406 hi = mid;
8407 }
8408 if (hi > lo && BB_OFFSET(p[lo]) > s)
8409 hi = lo;
8410
8411 if (hi > lo) {
8412 /* we found a range that might merge with the start
8413 * of our new range
8414 */
8415 sector_t a = BB_OFFSET(p[lo]);
8416 sector_t e = a + BB_LEN(p[lo]);
8417 int ack = BB_ACK(p[lo]);
8418 if (e >= s) {
8419 /* Yes, we can merge with a previous range */
8420 if (s == a && s + sectors >= e)
8421 /* new range covers old */
8422 ack = acknowledged;
8423 else
8424 ack = ack && acknowledged;
8425
8426 if (e < s + sectors)
8427 e = s + sectors;
8428 if (e - a <= BB_MAX_LEN) {
8429 p[lo] = BB_MAKE(a, e-a, ack);
8430 s = e;
8431 } else {
8432 /* does not all fit in one range,
8433 * make p[lo] maximal
8434 */
8435 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8436 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8437 s = a + BB_MAX_LEN;
8438 }
8439 sectors = e - s;
8440 }
8441 }
8442 if (sectors && hi < bb->count) {
8443 /* 'hi' points to the first range that starts after 's'.
8444 * Maybe we can merge with the start of that range */
8445 sector_t a = BB_OFFSET(p[hi]);
8446 sector_t e = a + BB_LEN(p[hi]);
8447 int ack = BB_ACK(p[hi]);
8448 if (a <= s + sectors) {
8449 /* merging is possible */
8450 if (e <= s + sectors) {
8451 /* full overlap */
8452 e = s + sectors;
8453 ack = acknowledged;
8454 } else
8455 ack = ack && acknowledged;
8456
8457 a = s;
8458 if (e - a <= BB_MAX_LEN) {
8459 p[hi] = BB_MAKE(a, e-a, ack);
8460 s = e;
8461 } else {
8462 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8463 s = a + BB_MAX_LEN;
8464 }
8465 sectors = e - s;
8466 lo = hi;
8467 hi++;
8468 }
8469 }
8470 if (sectors == 0 && hi < bb->count) {
8471 /* we might be able to combine lo and hi */
8472 /* Note: 's' is at the end of 'lo' */
8473 sector_t a = BB_OFFSET(p[hi]);
8474 int lolen = BB_LEN(p[lo]);
8475 int hilen = BB_LEN(p[hi]);
8476 int newlen = lolen + hilen - (s - a);
8477 if (s >= a && newlen < BB_MAX_LEN) {
8478 /* yes, we can combine them */
8479 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8480 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8481 memmove(p + hi, p + hi + 1,
8482 (bb->count - hi - 1) * 8);
8483 bb->count--;
8484 }
8485 }
8486 while (sectors) {
8487 /* didn't merge (it all).
8488 * Need to add a range just before 'hi' */
8489 if (bb->count >= MD_MAX_BADBLOCKS) {
8490 /* No room for more */
8491 rv = 0;
8492 break;
8493 } else {
8494 int this_sectors = sectors;
8495 memmove(p + hi + 1, p + hi,
8496 (bb->count - hi) * 8);
8497 bb->count++;
8498
8499 if (this_sectors > BB_MAX_LEN)
8500 this_sectors = BB_MAX_LEN;
8501 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8502 sectors -= this_sectors;
8503 s += this_sectors;
8504 }
8505 }
8506
8507 bb->changed = 1;
de393cde
N		 8508 if (!acknowledged)
8509 bb->unacked_exist = 1;
905b0297 8510 write_sequnlock_irqrestore(&bb->lock, flags);
2230dfe4
N		 8511
8512 return rv;
8513}
8514
3cb03002 8515int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8516 int is_new)
2230dfe4 8517{
c6563a8c
N		 8518 int rv;
8519 if (is_new)
8520 s += rdev->new_data_offset;
8521 else
8522 s += rdev->data_offset;
8523 rv = md_set_badblocks(&rdev->badblocks,
8524 s, sectors, 0);
2230dfe4
N		 8525 if (rv) {
8526 /* Make sure they get written out promptly */
8bd2f0a0 8527 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 8528 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8529 md_wakeup_thread(rdev->mddev->thread);
8530 }
8531 return rv;
8532}
8533EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8534
8535/*
8536 * Remove a range of bad blocks from the table.
8537 * This may involve extending the table if we spilt a region,
8538 * but it must not fail. So if the table becomes full, we just
8539 * drop the remove request.
8540 */
8541static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8542{
8543 u64 *p;
8544 int lo, hi;
8545 sector_t target = s + sectors;
8546 int rv = 0;
8547
8548 if (bb->shift > 0) {
8549 /* When clearing we round the start up and the end down.
8550 * This should not matter as the shift should align with
8551 * the block size and no rounding should ever be needed.
8552 * However it is better the think a block is bad when it
8553 * isn't than to think a block is not bad when it is.
8554 */
8555 s += (1<<bb->shift) - 1;
8556 s >>= bb->shift;
8557 target >>= bb->shift;
8558 sectors = target - s;
8559 }
8560
8561 write_seqlock_irq(&bb->lock);
8562
8563 p = bb->page;
8564 lo = 0;
8565 hi = bb->count;
8566 /* Find the last range that starts before 'target' */
8567 while (hi - lo > 1) {
8568 int mid = (lo + hi) / 2;
8569 sector_t a = BB_OFFSET(p[mid]);
8570 if (a < target)
8571 lo = mid;
8572 else
8573 hi = mid;
8574 }
8575 if (hi > lo) {
8576 /* p[lo] is the last range that could overlap the
8577 * current range. Earlier ranges could also overlap,
8578 * but only this one can overlap the end of the range.
8579 */
8580 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8581 /* Partial overlap, leave the tail of this range */
8582 int ack = BB_ACK(p[lo]);
8583 sector_t a = BB_OFFSET(p[lo]);
8584 sector_t end = a + BB_LEN(p[lo]);
8585
8586 if (a < s) {
8587 /* we need to split this range */
8588 if (bb->count >= MD_MAX_BADBLOCKS) {
8b32bf5e 8589 rv = -ENOSPC;
2230dfe4
N		 8590 goto out;
8591 }
8592 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8593 bb->count++;
8594 p[lo] = BB_MAKE(a, s-a, ack);
8595 lo++;
8596 }
8597 p[lo] = BB_MAKE(target, end - target, ack);
8598 /* there is no longer an overlap */
8599 hi = lo;
8600 lo--;
8601 }
8602 while (lo >= 0 &&
8603 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8604 /* This range does overlap */
8605 if (BB_OFFSET(p[lo]) < s) {
8606 /* Keep the early parts of this range. */
8607 int ack = BB_ACK(p[lo]);
8608 sector_t start = BB_OFFSET(p[lo]);
8609 p[lo] = BB_MAKE(start, s - start, ack);
8610 /* now low doesn't overlap, so.. */
8611 break;
8612 }
8613 lo--;
8614 }
8615 /* 'lo' is strictly before, 'hi' is strictly after,
8616 * anything between needs to be discarded
8617 */
8618 if (hi - lo > 1) {
8619 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8620 bb->count -= (hi - lo - 1);
8621 }
8622 }
8623
8624 bb->changed = 1;
8625out:
8626 write_sequnlock_irq(&bb->lock);
8627 return rv;
8628}
8629
c6563a8c
N		 8630int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8631 int is_new)
2230dfe4 8632{
c6563a8c
N		 8633 if (is_new)
8634 s += rdev->new_data_offset;
8635 else
8636 s += rdev->data_offset;
2230dfe4 8637 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8638 s, sectors);
2230dfe4
N		 8639}
8640EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8641
8642/*
8643 * Acknowledge all bad blocks in a list.
8644 * This only succeeds if ->changed is clear. It is used by
8645 * in-kernel metadata updates
8646 */
8647void md_ack_all_badblocks(struct badblocks *bb)
8648{
8649 if (bb->page == NULL || bb->changed)
8650 /* no point even trying */
8651 return;
8652 write_seqlock_irq(&bb->lock);
8653
ecb178bb 8654 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N		 8655 u64 *p = bb->page;
8656 int i;
8657 for (i = 0; i < bb->count ; i++) {
8658 if (!BB_ACK(p[i])) {
8659 sector_t start = BB_OFFSET(p[i]);
8660 int len = BB_LEN(p[i]);
8661 p[i] = BB_MAKE(start, len, 1);
8662 }
8663 }
de393cde 8664 bb->unacked_exist = 0;
2230dfe4
N		 8665 }
8666 write_sequnlock_irq(&bb->lock);
8667}
8668EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8669
16c791a5
N		 8670/* sysfs access to bad-blocks list.
8671 * We present two files.
8672 * 'bad-blocks' lists sector numbers and lengths of ranges that
8673 * are recorded as bad. The list is truncated to fit within
8674 * the one-page limit of sysfs.
8675 * Writing "sector length" to this file adds an acknowledged
8676 * bad block list.
8677 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8678 * been acknowledged. Writing to this file adds bad blocks
8679 * without acknowledging them. This is largely for testing.
8680 */
8681
8682static ssize_t
8683badblocks_show(struct badblocks *bb, char *page, int unack)
8684{
8685 size_t len;
8686 int i;
8687 u64 *p = bb->page;
8688 unsigned seq;
8689
8690 if (bb->shift < 0)
8691 return 0;
8692
8693retry:
8694 seq = read_seqbegin(&bb->lock);
8695
8696 len = 0;
8697 i = 0;
8698
8699 while (len < PAGE_SIZE && i < bb->count) {
8700 sector_t s = BB_OFFSET(p[i]);
8701 unsigned int length = BB_LEN(p[i]);
8702 int ack = BB_ACK(p[i]);
8703 i++;
8704
8705 if (unack && ack)
8706 continue;
8707
8708 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8709 (unsigned long long)s << bb->shift,
8710 length << bb->shift);
8711 }
de393cde
N		 8712 if (unack && len == 0)
8713 bb->unacked_exist = 0;
16c791a5
N		 8714
8715 if (read_seqretry(&bb->lock, seq))
8716 goto retry;
8717
8718 return len;
8719}
8720
8721#define DO_DEBUG 1
8722
8723static ssize_t
8724badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8725{
8726 unsigned long long sector;
8727 int length;
8728 char newline;
8729#ifdef DO_DEBUG
8730 /* Allow clearing via sysfs *only* for testing/debugging.
8731 * Normally only a successful write may clear a badblock
8732 */
8733 int clear = 0;
8734 if (page[0] == '-') {
8735 clear = 1;
8736 page++;
8737 }
8738#endif /* DO_DEBUG */
8739
8740 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8741 case 3:
8742 if (newline != '\n')
8743 return -EINVAL;
8744 case 2:
8745 if (length <= 0)
8746 return -EINVAL;
8747 break;
8748 default:
8749 return -EINVAL;
8750 }
8751
8752#ifdef DO_DEBUG
8753 if (clear) {
8754 md_clear_badblocks(bb, sector, length);
8755 return len;
8756 }
8757#endif /* DO_DEBUG */
8758 if (md_set_badblocks(bb, sector, length, !unack))
8759 return len;
8760 else
8761 return -ENOSPC;
8762}
8763
75c96f85
AB		 8764static int md_notify_reboot(struct notifier_block *this,
8765 unsigned long code, void *x)
1da177e4
LT		 8766{
8767 struct list_head *tmp;
fd01b88c 8768 struct mddev *mddev;
2dba6a91 8769 int need_delay = 0;
1da177e4 8770
c744a65c
N		 8771 for_each_mddev(mddev, tmp) {
8772 if (mddev_trylock(mddev)) {
30b8aa91
N		 8773 if (mddev->pers)
8774 __md_stop_writes(mddev);
0f62fb22
N		 8775 if (mddev->persistent)
8776 mddev->safemode = 2;
c744a65c 8777 mddev_unlock(mddev);
2dba6a91 8778 }
c744a65c 8779 need_delay = 1;
1da177e4 8780 }
c744a65c
N		 8781 /*
8782 * certain more exotic SCSI devices are known to be
8783 * volatile wrt too early system reboots. While the
8784 * right place to handle this issue is the given
8785 * driver, we do want to have a safe RAID driver ...
8786 */
8787 if (need_delay)
8788 mdelay(1000*1);
8789
1da177e4
LT		 8790 return NOTIFY_DONE;
8791}
8792
75c96f85 8793static struct notifier_block md_notifier = {
1da177e4
LT		 8794 .notifier_call = md_notify_reboot,
8795 .next = NULL,
8796 .priority = INT_MAX, /* before any real devices */
8797};
8798
8799static void md_geninit(void)
8800{
36a4e1fe 8801 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8802
c7705f34 8803 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8804}
8805
75c96f85 8806static int __init md_init(void)
1da177e4 8807{
e804ac78
TH		 8808 int ret = -ENOMEM;
8809
ada609ee 8810 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8811 if (!md_wq)
8812 goto err_wq;
8813
8814 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8815 if (!md_misc_wq)
8816 goto err_misc_wq;
8817
8818 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8819 goto err_md;
8820
8821 if ((ret = register_blkdev(0, "mdp")) < 0)
8822 goto err_mdp;
8823 mdp_major = ret;
8824
af5628f0 8825 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
e8703fe1
N		 8826 md_probe, NULL, NULL);
8827 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8828 md_probe, NULL, NULL);
8829
1da177e4 8830 register_reboot_notifier(&md_notifier);
0b4d4147 8831 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8832
8833 md_geninit();
d710e138 8834 return 0;
1da177e4 8835
e804ac78
TH		 8836err_mdp:
8837 unregister_blkdev(MD_MAJOR, "md");
8838err_md:
8839 destroy_workqueue(md_misc_wq);
8840err_misc_wq:
8841 destroy_workqueue(md_wq);
8842err_wq:
8843 return ret;
8844}
1da177e4 8845
1d7e3e96
GR		 8846void md_reload_sb(struct mddev *mddev)
8847{
8848 struct md_rdev *rdev, *tmp;
8849
8850 rdev_for_each_safe(rdev, tmp, mddev) {
8851 rdev->sb_loaded = 0;
8852 ClearPageUptodate(rdev->sb_page);
8853 }
8854 mddev->raid_disks = 0;
8855 analyze_sbs(mddev);
8856 rdev_for_each_safe(rdev, tmp, mddev) {
8857 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8858 /* since we don't write to faulty devices, we figure out if the
8859 * disk is faulty by comparing events
8860 */
8861 if (mddev->events > sb->events)
8862 set_bit(Faulty, &rdev->flags);
8863 }
8864
8865}
8866EXPORT_SYMBOL(md_reload_sb);
8867
1da177e4
LT		 8868#ifndef MODULE
8869
8870/*
8871 * Searches all registered partitions for autorun RAID arrays
8872 * at boot time.
8873 */
4d936ec1
ME		 8874
8875static LIST_HEAD(all_detected_devices);
8876struct detected_devices_node {
8877 struct list_head list;
8878 dev_t dev;
8879};
1da177e4
LT		 8880
8881void md_autodetect_dev(dev_t dev)
8882{
4d936ec1
ME		 8883 struct detected_devices_node *node_detected_dev;
8884
8885 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8886 if (node_detected_dev) {
8887 node_detected_dev->dev = dev;
8888 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8889 } else {
8890 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8891 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8892 }
1da177e4
LT		 8893}
8894
1da177e4
LT		 8895static void autostart_arrays(int part)
8896{
3cb03002 8897 struct md_rdev *rdev;
4d936ec1
ME		 8898 struct detected_devices_node *node_detected_dev;
8899 dev_t dev;
8900 int i_scanned, i_passed;
1da177e4 8901
4d936ec1
ME		 8902 i_scanned = 0;
8903 i_passed = 0;
1da177e4 8904
4d936ec1 8905 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8906
4d936ec1
ME		 8907 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8908 i_scanned++;
8909 node_detected_dev = list_entry(all_detected_devices.next,
8910 struct detected_devices_node, list);
8911 list_del(&node_detected_dev->list);
8912 dev = node_detected_dev->dev;
8913 kfree(node_detected_dev);
df968c4e 8914 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8915 if (IS_ERR(rdev))
8916 continue;
8917
403df478 8918 if (test_bit(Faulty, &rdev->flags))
1da177e4 8919 continue;
403df478 8920
d0fae18f 8921 set_bit(AutoDetected, &rdev->flags);
1da177e4 8922 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8923 i_passed++;
1da177e4 8924 }
4d936ec1
ME		 8925
8926 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8927 i_scanned, i_passed);
1da177e4
LT		 8928
8929 autorun_devices(part);
8930}
8931
fdee8ae4 8932#endif /* !MODULE */
1da177e4
LT		 8933
8934static __exit void md_exit(void)
8935{
fd01b88c 8936 struct mddev *mddev;
1da177e4 8937 struct list_head *tmp;
e2f23b60 8938 int delay = 1;
8ab5e4c1 8939
af5628f0 8940 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
e8703fe1 8941 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8942
3dbd8c2e 8943 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8944 unregister_blkdev(mdp_major, "mdp");
8945 unregister_reboot_notifier(&md_notifier);
8946 unregister_sysctl_table(raid_table_header);
e2f23b60
N		 8947
8948 /* We cannot unload the modules while some process is
8949 * waiting for us in select() or poll() - wake them up
8950 */
8951 md_unloading = 1;
8952 while (waitqueue_active(&md_event_waiters)) {
8953 /* not safe to leave yet */
8954 wake_up(&md_event_waiters);
8955 msleep(delay);
8956 delay += delay;
8957 }
1da177e4 8958 remove_proc_entry("mdstat", NULL);
e2f23b60 8959
29ac4aa3 8960 for_each_mddev(mddev, tmp) {
1da177e4 8961 export_array(mddev);
d3374825 8962 mddev->hold_active = 0;
1da177e4 8963 }
e804ac78
TH		 8964 destroy_workqueue(md_misc_wq);
8965 destroy_workqueue(md_wq);
1da177e4
LT		 8966}
8967
685784aa 8968subsys_initcall(md_init);
1da177e4
LT		 8969module_exit(md_exit)
8970
f91de92e
N		 8971static int get_ro(char *buffer, struct kernel_param *kp)
8972{
8973 return sprintf(buffer, "%d", start_readonly);
8974}
8975static int set_ro(const char *val, struct kernel_param *kp)
8976{
8977 char *e;
8978 int num = simple_strtoul(val, &e, 10);
8979 if (*val && (*e == '\0' || *e == '\n')) {
8980 start_readonly = num;
4dbcdc75 8981 return 0;
f91de92e
N		 8982 }
8983 return -EINVAL;
8984}
8985
80ca3a44
N		 8986module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8987module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
efeb53c0 8988module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8989
1da177e4 8990MODULE_LICENSE("GPL");
0efb9e61 8991MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8992MODULE_ALIAS("md");
72008652 8993MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux kernel - Deliverables
This page took 2.313603 seconds and 5 git commands to generate.