projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Communication Framework: Sending functions
[linux-2.6-block.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
f72ffdd6 3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
1da177e4
LT		 4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
a6fb0934 35#include <linux/kthread.h>
bff61975 36#include <linux/blkdev.h>
1da177e4 37#include <linux/sysctl.h>
bff61975 38#include <linux/seq_file.h>
ff01bb48 39#include <linux/fs.h>
d7603b7e 40#include <linux/poll.h>
16f17b39 41#include <linux/ctype.h>
e7d2860b 42#include <linux/string.h>
fb4d8c76
N		 43#include <linux/hdreg.h>
44#include <linux/proc_fs.h>
45#include <linux/random.h>
056075c7 46#include <linux/module.h>
fb4d8c76 47#include <linux/reboot.h>
32a7627c 48#include <linux/file.h>
aa98aa31 49#include <linux/compat.h>
25570727 50#include <linux/delay.h>
bff61975
N		 51#include <linux/raid/md_p.h>
52#include <linux/raid/md_u.h>
5a0e3ad6 53#include <linux/slab.h>
43b2e5d8 54#include "md.h"
ef740c37 55#include "bitmap.h"
edb39c9d 56#include "md-cluster.h"
1da177e4 57
1da177e4 58#ifndef MODULE
d710e138 59static void autostart_arrays(int part);
1da177e4
LT		 60#endif
61
01f96c0a
N		 62/* pers_list is a list of registered personalities protected
63 * by pers_lock.
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
66 */
2604b703 67static LIST_HEAD(pers_list);
1da177e4
LT		 68static DEFINE_SPINLOCK(pers_lock);
69
edb39c9d
GR		 70struct md_cluster_operations *md_cluster_ops;
71struct module *md_cluster_mod;
72EXPORT_SYMBOL(md_cluster_mod);
73
90b08710 74static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 75static struct workqueue_struct *md_wq;
76static struct workqueue_struct *md_misc_wq;
90b08710 77
746d3207
N		 78static int remove_and_add_spares(struct mddev *mddev,
79 struct md_rdev *this);
5aa61f42 80static void mddev_detach(struct mddev *mddev);
746d3207 81
1e50915f
RB		 82/*
83 * Default number of read corrections we'll attempt on an rdev
84 * before ejecting it from the array. We divide the read error
85 * count by 2 for every hour elapsed between read errors.
86 */
87#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 88/*
89 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
90 * is 1000 KB/sec, so the extra system load does not show up that much.
91 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 92 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 93 * subsystem is idle. There is also an 'absolute maximum' reconstruction
94 * speed limit - in case reconstruction slows down your system despite
95 * idle IO detection.
96 *
97 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 98 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 99 */
100
101static int sysctl_speed_limit_min = 1000;
102static int sysctl_speed_limit_max = 200000;
fd01b88c 103static inline int speed_min(struct mddev *mddev)
88202a0c
N		 104{
105 return mddev->sync_speed_min ?
106 mddev->sync_speed_min : sysctl_speed_limit_min;
107}
108
fd01b88c 109static inline int speed_max(struct mddev *mddev)
88202a0c
N		 110{
111 return mddev->sync_speed_max ?
112 mddev->sync_speed_max : sysctl_speed_limit_max;
113}
1da177e4
LT		 114
115static struct ctl_table_header *raid_table_header;
116
82592c38 117static struct ctl_table raid_table[] = {
1da177e4 118 {
1da177e4
LT		 119 .procname = "speed_limit_min",
120 .data = &sysctl_speed_limit_min,
121 .maxlen = sizeof(int),
80ca3a44 122 .mode = S_IRUGO|S_IWUSR,
6d456111 123 .proc_handler = proc_dointvec,
1da177e4
LT		 124 },
125 {
1da177e4
LT		 126 .procname = "speed_limit_max",
127 .data = &sysctl_speed_limit_max,
128 .maxlen = sizeof(int),
80ca3a44 129 .mode = S_IRUGO|S_IWUSR,
6d456111 130 .proc_handler = proc_dointvec,
1da177e4 131 },
894d2491 132 { }
1da177e4
LT		 133};
134
82592c38 135static struct ctl_table raid_dir_table[] = {
1da177e4 136 {
1da177e4
LT		 137 .procname = "raid",
138 .maxlen = 0,
80ca3a44 139 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 140 .child = raid_table,
141 },
894d2491 142 { }
1da177e4
LT		 143};
144
82592c38 145static struct ctl_table raid_root_table[] = {
1da177e4 146 {
1da177e4
LT		 147 .procname = "dev",
148 .maxlen = 0,
149 .mode = 0555,
150 .child = raid_dir_table,
151 },
894d2491 152 { }
1da177e4
LT		 153};
154
83d5cde4 155static const struct block_device_operations md_fops;
1da177e4 156
f91de92e
N		 157static int start_readonly;
158
a167f663
N		 159/* bio_clone_mddev
160 * like bio_clone, but with a local bio set
161 */
162
a167f663 163struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 164 struct mddev *mddev)
a167f663
N		 165{
166 struct bio *b;
a167f663
N		 167
168 if (!mddev || !mddev->bio_set)
169 return bio_alloc(gfp_mask, nr_iovecs);
170
395c72a7 171 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
a167f663
N		 172 if (!b)
173 return NULL;
a167f663
N		 174 return b;
175}
176EXPORT_SYMBOL_GPL(bio_alloc_mddev);
177
178struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
fd01b88c 179 struct mddev *mddev)
a167f663 180{
a167f663
N		 181 if (!mddev || !mddev->bio_set)
182 return bio_clone(bio, gfp_mask);
183
bf800ef1 184 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
a167f663
N		 185}
186EXPORT_SYMBOL_GPL(bio_clone_mddev);
187
d7603b7e
N		 188/*
189 * We have a system wide 'event count' that is incremented
190 * on any 'interesting' event, and readers of /proc/mdstat
191 * can use 'poll' or 'select' to find out when the event
192 * count increases.
193 *
194 * Events are:
195 * start array, stop array, error, add device, remove device,
196 * start build, activate spare
197 */
2989ddbd 198static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 199static atomic_t md_event_count;
fd01b88c 200void md_new_event(struct mddev *mddev)
d7603b7e
N		 201{
202 atomic_inc(&md_event_count);
203 wake_up(&md_event_waiters);
204}
29269553 205EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 206
c331eb04
N		 207/* Alternate version that can be called from interrupts
208 * when calling sysfs_notify isn't needed.
209 */
fd01b88c 210static void md_new_event_inintr(struct mddev *mddev)
c331eb04
N		 211{
212 atomic_inc(&md_event_count);
213 wake_up(&md_event_waiters);
214}
215
1da177e4
LT		 216/*
217 * Enables to iterate over all existing md arrays
218 * all_mddevs_lock protects this list.
219 */
220static LIST_HEAD(all_mddevs);
221static DEFINE_SPINLOCK(all_mddevs_lock);
222
1da177e4
LT		 223/*
224 * iterates through all used mddevs in the system.
225 * We take care to grab the all_mddevs_lock whenever navigating
226 * the list, and to always hold a refcount when unlocked.
227 * Any code which breaks out of this loop while own
228 * a reference to the current mddev and must mddev_put it.
229 */
fd01b88c 230#define for_each_mddev(_mddev,_tmp) \
1da177e4 231 \
f72ffdd6 232 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N		 233 _tmp = all_mddevs.next; \
234 _mddev = NULL;}); \
235 ({ if (_tmp != &all_mddevs) \
236 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 237 spin_unlock(&all_mddevs_lock); \
fd01b88c
N		 238 if (_mddev) mddev_put(_mddev); \
239 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
240 _tmp != &all_mddevs;}); \
1da177e4 241 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 242 _tmp = _tmp->next;}) \
1da177e4
LT		 243 )
244
409c57f3
N		 245/* Rather than calling directly into the personality make_request function,
246 * IO requests come here first so that we can check if the device is
247 * being suspended pending a reconfiguration.
248 * We hold a refcount over the call to ->make_request. By the time that
249 * call has finished, the bio has been linked into some internal structure
250 * and so is visible to ->quiesce(), so we don't need the refcount any more.
251 */
5a7bbad2 252static void md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 253{
49077326 254 const int rw = bio_data_dir(bio);
fd01b88c 255 struct mddev *mddev = q->queuedata;
e91ece55 256 unsigned int sectors;
49077326 257
0ca69886
N		 258 if (mddev == NULL || mddev->pers == NULL
259 || !mddev->ready) {
409c57f3 260 bio_io_error(bio);
5a7bbad2 261 return;
409c57f3 262 }
bbfa57c0
SR		 263 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
264 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
265 return;
266 }
0ca69886 267 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 268 rcu_read_lock();
e9c7469b 269 if (mddev->suspended) {
409c57f3
N		 270 DEFINE_WAIT(__wait);
271 for (;;) {
272 prepare_to_wait(&mddev->sb_wait, &__wait,
273 TASK_UNINTERRUPTIBLE);
e9c7469b 274 if (!mddev->suspended)
409c57f3
N		 275 break;
276 rcu_read_unlock();
277 schedule();
278 rcu_read_lock();
279 }
280 finish_wait(&mddev->sb_wait, &__wait);
281 }
282 atomic_inc(&mddev->active_io);
283 rcu_read_unlock();
49077326 284
e91ece55
CM		 285 /*
286 * save the sectors now since our bio can
287 * go away inside make_request
288 */
289 sectors = bio_sectors(bio);
5a7bbad2 290 mddev->pers->make_request(mddev, bio);
49077326 291
18c0b223 292 generic_start_io_acct(rw, sectors, &mddev->gendisk->part0);
49077326 293
409c57f3
N		 294 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
295 wake_up(&mddev->sb_wait);
409c57f3
N		 296}
297
9e35b99c
N		 298/* mddev_suspend makes sure no new requests are submitted
299 * to the device, and that any requests that have been submitted
300 * are completely handled.
afa0f557
N		 301 * Once mddev_detach() is called and completes, the module will be
302 * completely unused.
9e35b99c 303 */
fd01b88c 304void mddev_suspend(struct mddev *mddev)
409c57f3
N		 305{
306 BUG_ON(mddev->suspended);
307 mddev->suspended = 1;
308 synchronize_rcu();
309 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
310 mddev->pers->quiesce(mddev, 1);
0d9f4f13
JB		 311
312 del_timer_sync(&mddev->safemode_timer);
409c57f3 313}
390ee602 314EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 315
fd01b88c 316void mddev_resume(struct mddev *mddev)
409c57f3
N		 317{
318 mddev->suspended = 0;
319 wake_up(&mddev->sb_wait);
320 mddev->pers->quiesce(mddev, 0);
0fd018af 321
47525e59 322 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
0fd018af
JB		 323 md_wakeup_thread(mddev->thread);
324 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 325}
390ee602 326EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 327
fd01b88c 328int mddev_congested(struct mddev *mddev, int bits)
3fa841d7 329{
5c675f83
N		 330 struct md_personality *pers = mddev->pers;
331 int ret = 0;
332
333 rcu_read_lock();
334 if (mddev->suspended)
335 ret = 1;
336 else if (pers && pers->congested)
337 ret = pers->congested(mddev, bits);
338 rcu_read_unlock();
339 return ret;
340}
341EXPORT_SYMBOL_GPL(mddev_congested);
342static int md_congested(void *data, int bits)
343{
344 struct mddev *mddev = data;
345 return mddev_congested(mddev, bits);
3fa841d7 346}
3fa841d7 347
64590f45
N		 348static int md_mergeable_bvec(struct request_queue *q,
349 struct bvec_merge_data *bvm,
350 struct bio_vec *biovec)
351{
352 struct mddev *mddev = q->queuedata;
353 int ret;
354 rcu_read_lock();
355 if (mddev->suspended) {
356 /* Must always allow one vec */
357 if (bvm->bi_size == 0)
358 ret = biovec->bv_len;
359 else
360 ret = 0;
361 } else {
362 struct md_personality *pers = mddev->pers;
363 if (pers && pers->mergeable_bvec)
364 ret = pers->mergeable_bvec(mddev, bvm, biovec);
365 else
366 ret = biovec->bv_len;
367 }
368 rcu_read_unlock();
369 return ret;
370}
a2826aa9 371/*
e9c7469b 372 * Generic flush handling for md
a2826aa9
N		 373 */
374
e9c7469b 375static void md_end_flush(struct bio *bio, int err)
a2826aa9 376{
3cb03002 377 struct md_rdev *rdev = bio->bi_private;
fd01b88c 378 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 379
380 rdev_dec_pending(rdev, mddev);
381
382 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 383 /* The pre-request flush has finished */
e804ac78 384 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 385 }
386 bio_put(bio);
387}
388
a7a07e69
N		 389static void md_submit_flush_data(struct work_struct *ws);
390
a035fc3e 391static void submit_flushes(struct work_struct *ws)
a2826aa9 392{
fd01b88c 393 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 394 struct md_rdev *rdev;
a2826aa9 395
a7a07e69
N		 396 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
397 atomic_set(&mddev->flush_pending, 1);
a2826aa9 398 rcu_read_lock();
dafb20fa 399 rdev_for_each_rcu(rdev, mddev)
a2826aa9
N		 400 if (rdev->raid_disk >= 0 &&
401 !test_bit(Faulty, &rdev->flags)) {
402 /* Take two references, one is dropped
403 * when request finishes, one after
404 * we reclaim rcu_read_lock
405 */
406 struct bio *bi;
407 atomic_inc(&rdev->nr_pending);
408 atomic_inc(&rdev->nr_pending);
409 rcu_read_unlock();
b5e1b8ce 410 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
e9c7469b 411 bi->bi_end_io = md_end_flush;
a2826aa9
N		 412 bi->bi_private = rdev;
413 bi->bi_bdev = rdev->bdev;
414 atomic_inc(&mddev->flush_pending);
e9c7469b 415 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 416 rcu_read_lock();
417 rdev_dec_pending(rdev, mddev);
418 }
419 rcu_read_unlock();
a7a07e69
N		 420 if (atomic_dec_and_test(&mddev->flush_pending))
421 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 422}
423
e9c7469b 424static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 425{
fd01b88c 426 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 427 struct bio *bio = mddev->flush_bio;
a2826aa9 428
4f024f37 429 if (bio->bi_iter.bi_size == 0)
a2826aa9
N		 430 /* an empty barrier - all done */
431 bio_endio(bio, 0);
432 else {
e9c7469b 433 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 434 mddev->pers->make_request(mddev, bio);
a2826aa9 435 }
2b74e12e
N		 436
437 mddev->flush_bio = NULL;
438 wake_up(&mddev->sb_wait);
a2826aa9
N		 439}
440
fd01b88c 441void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9 442{
85572d7c 443 spin_lock_irq(&mddev->lock);
a2826aa9 444 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 445 !mddev->flush_bio,
85572d7c 446 mddev->lock);
e9c7469b 447 mddev->flush_bio = bio;
85572d7c 448 spin_unlock_irq(&mddev->lock);
a2826aa9 449
a035fc3e
N		 450 INIT_WORK(&mddev->flush_work, submit_flushes);
451 queue_work(md_wq, &mddev->flush_work);
a2826aa9 452}
e9c7469b 453EXPORT_SYMBOL(md_flush_request);
409c57f3 454
74018dc3 455void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
97658cdd 456{
9cbb1750
N		 457 struct mddev *mddev = cb->data;
458 md_wakeup_thread(mddev->thread);
459 kfree(cb);
97658cdd 460}
9cbb1750 461EXPORT_SYMBOL(md_unplug);
2ac87401 462
fd01b88c 463static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 464{
465 atomic_inc(&mddev->active);
466 return mddev;
467}
468
5fd3a17e 469static void mddev_delayed_delete(struct work_struct *ws);
d3374825 470
fd01b88c 471static void mddev_put(struct mddev *mddev)
1da177e4 472{
a167f663
N		 473 struct bio_set *bs = NULL;
474
1da177e4
LT		 475 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
476 return;
d3374825 477 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 478 mddev->ctime == 0 && !mddev->hold_active) {
479 /* Array is not configured at all, and not held active,
480 * so destroy it */
af8a2434 481 list_del_init(&mddev->all_mddevs);
a167f663
N		 482 bs = mddev->bio_set;
483 mddev->bio_set = NULL;
d3374825 484 if (mddev->gendisk) {
e804ac78
TH		 485 /* We did a probe so need to clean up. Call
486 * queue_work inside the spinlock so that
487 * flush_workqueue() after mddev_find will
488 * succeed in waiting for the work to be done.
d3374825
N		 489 */
490 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 491 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 492 } else
493 kfree(mddev);
494 }
495 spin_unlock(&all_mddevs_lock);
a167f663
N		 496 if (bs)
497 bioset_free(bs);
1da177e4
LT		 498}
499
fd01b88c 500void mddev_init(struct mddev *mddev)
fafd7fb0
N		 501{
502 mutex_init(&mddev->open_mutex);
503 mutex_init(&mddev->reconfig_mutex);
504 mutex_init(&mddev->bitmap_info.mutex);
505 INIT_LIST_HEAD(&mddev->disks);
506 INIT_LIST_HEAD(&mddev->all_mddevs);
507 init_timer(&mddev->safemode_timer);
508 atomic_set(&mddev->active, 1);
509 atomic_set(&mddev->openers, 0);
510 atomic_set(&mddev->active_io, 0);
85572d7c 511 spin_lock_init(&mddev->lock);
fafd7fb0
N		 512 atomic_set(&mddev->flush_pending, 0);
513 init_waitqueue_head(&mddev->sb_wait);
514 init_waitqueue_head(&mddev->recovery_wait);
515 mddev->reshape_position = MaxSector;
2c810cdd 516 mddev->reshape_backwards = 0;
c4a39551 517 mddev->last_sync_action = "none";
fafd7fb0
N		 518 mddev->resync_min = 0;
519 mddev->resync_max = MaxSector;
520 mddev->level = LEVEL_NONE;
521}
390ee602 522EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 523
f72ffdd6 524static struct mddev *mddev_find(dev_t unit)
1da177e4 525{
fd01b88c 526 struct mddev *mddev, *new = NULL;
1da177e4 527
8f5f02c4
N		 528 if (unit && MAJOR(unit) != MD_MAJOR)
529 unit &= ~((1<<MdpMinorShift)-1);
530
1da177e4
LT		 531 retry:
532 spin_lock(&all_mddevs_lock);
efeb53c0
N		 533
534 if (unit) {
535 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
536 if (mddev->unit == unit) {
537 mddev_get(mddev);
538 spin_unlock(&all_mddevs_lock);
539 kfree(new);
540 return mddev;
541 }
542
543 if (new) {
544 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 545 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 546 new->hold_active = UNTIL_IOCTL;
547 return new;
1da177e4 548 }
efeb53c0
N		 549 } else if (new) {
550 /* find an unused unit number */
551 static int next_minor = 512;
552 int start = next_minor;
553 int is_free = 0;
554 int dev = 0;
555 while (!is_free) {
556 dev = MKDEV(MD_MAJOR, next_minor);
557 next_minor++;
558 if (next_minor > MINORMASK)
559 next_minor = 0;
560 if (next_minor == start) {
561 /* Oh dear, all in use. */
562 spin_unlock(&all_mddevs_lock);
563 kfree(new);
564 return NULL;
565 }
f72ffdd6 566
efeb53c0
N		 567 is_free = 1;
568 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
569 if (mddev->unit == dev) {
570 is_free = 0;
571 break;
572 }
573 }
574 new->unit = dev;
575 new->md_minor = MINOR(dev);
576 new->hold_active = UNTIL_STOP;
1da177e4
LT		 577 list_add(&new->all_mddevs, &all_mddevs);
578 spin_unlock(&all_mddevs_lock);
579 return new;
580 }
581 spin_unlock(&all_mddevs_lock);
582
9ffae0cf 583 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 584 if (!new)
585 return NULL;
586
1da177e4
LT		 587 new->unit = unit;
588 if (MAJOR(unit) == MD_MAJOR)
589 new->md_minor = MINOR(unit);
590 else
591 new->md_minor = MINOR(unit) >> MdpMinorShift;
592
fafd7fb0 593 mddev_init(new);
1da177e4 594
1da177e4
LT		 595 goto retry;
596}
597
b6eb127d
N		 598static struct attribute_group md_redundancy_group;
599
5c47daf6 600void mddev_unlock(struct mddev *mddev)
1da177e4 601{
a64c876f 602 if (mddev->to_remove) {
b6eb127d
N		 603 /* These cannot be removed under reconfig_mutex as
604 * an access to the files will try to take reconfig_mutex
605 * while holding the file unremovable, which leads to
606 * a deadlock.
bb4f1e9d
N		 607 * So hold set sysfs_active while the remove in happeing,
608 * and anything else which might set ->to_remove or my
609 * otherwise change the sysfs namespace will fail with
610 * -EBUSY if sysfs_active is still set.
611 * We set sysfs_active under reconfig_mutex and elsewhere
612 * test it under the same mutex to ensure its correct value
613 * is seen.
b6eb127d 614 */
a64c876f
N		 615 struct attribute_group *to_remove = mddev->to_remove;
616 mddev->to_remove = NULL;
bb4f1e9d 617 mddev->sysfs_active = 1;
b6eb127d
N		 618 mutex_unlock(&mddev->reconfig_mutex);
619
00bcb4ac
N		 620 if (mddev->kobj.sd) {
621 if (to_remove != &md_redundancy_group)
622 sysfs_remove_group(&mddev->kobj, to_remove);
623 if (mddev->pers == NULL ||
624 mddev->pers->sync_request == NULL) {
625 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
626 if (mddev->sysfs_action)
627 sysfs_put(mddev->sysfs_action);
628 mddev->sysfs_action = NULL;
629 }
a64c876f 630 }
bb4f1e9d 631 mddev->sysfs_active = 0;
b6eb127d
N		 632 } else
633 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 634
751e67ca
CD		 635 /* As we've dropped the mutex we need a spinlock to
636 * make sure the thread doesn't disappear
01f96c0a
N		 637 */
638 spin_lock(&pers_lock);
005eca5e 639 md_wakeup_thread(mddev->thread);
01f96c0a 640 spin_unlock(&pers_lock);
1da177e4 641}
5c47daf6 642EXPORT_SYMBOL_GPL(mddev_unlock);
1da177e4 643
1ca69c4b
N		 644static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
645{
646 struct md_rdev *rdev;
647
648 rdev_for_each_rcu(rdev, mddev)
649 if (rdev->desc_nr == nr)
650 return rdev;
651
652 return NULL;
653}
654
655static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
1da177e4 656{
3cb03002 657 struct md_rdev *rdev;
1da177e4 658
dafb20fa 659 rdev_for_each(rdev, mddev)
1da177e4
LT		 660 if (rdev->bdev->bd_dev == dev)
661 return rdev;
159ec1fc 662
1da177e4
LT		 663 return NULL;
664}
665
1ca69c4b
N		 666static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
667{
668 struct md_rdev *rdev;
669
670 rdev_for_each_rcu(rdev, mddev)
671 if (rdev->bdev->bd_dev == dev)
672 return rdev;
673
674 return NULL;
675}
676
84fc4b56 677static struct md_personality *find_pers(int level, char *clevel)
2604b703 678{
84fc4b56 679 struct md_personality *pers;
d9d166c2
N		 680 list_for_each_entry(pers, &pers_list, list) {
681 if (level != LEVEL_NONE && pers->level == level)
2604b703 682 return pers;
d9d166c2
N		 683 if (strcmp(pers->name, clevel)==0)
684 return pers;
685 }
2604b703
N		 686 return NULL;
687}
688
b73df2d3 689/* return the offset of the super block in 512byte sectors */
3cb03002 690static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 691{
57b2caa3 692 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 693 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 694}
695
f72ffdd6 696static int alloc_disk_sb(struct md_rdev *rdev)
1da177e4 697{
1da177e4
LT		 698 rdev->sb_page = alloc_page(GFP_KERNEL);
699 if (!rdev->sb_page) {
700 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 701 return -ENOMEM;
1da177e4
LT		 702 }
703
704 return 0;
705}
706
545c8795 707void md_rdev_clear(struct md_rdev *rdev)
1da177e4
LT		 708{
709 if (rdev->sb_page) {
2d1f3b5d 710 put_page(rdev->sb_page);
1da177e4
LT		 711 rdev->sb_loaded = 0;
712 rdev->sb_page = NULL;
0f420358 713 rdev->sb_start = 0;
dd8ac336 714 rdev->sectors = 0;
1da177e4 715 }
2699b672
N		 716 if (rdev->bb_page) {
717 put_page(rdev->bb_page);
718 rdev->bb_page = NULL;
719 }
4fa2f327
N		 720 kfree(rdev->badblocks.page);
721 rdev->badblocks.page = NULL;
1da177e4 722}
545c8795 723EXPORT_SYMBOL_GPL(md_rdev_clear);
1da177e4 724
6712ecf8 725static void super_written(struct bio *bio, int error)
7bfa19f2 726{
3cb03002 727 struct md_rdev *rdev = bio->bi_private;
fd01b88c 728 struct mddev *mddev = rdev->mddev;
7bfa19f2 729
3a0f5bbb
N		 730 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
731 printk("md: super_written gets error=%d, uptodate=%d\n",
732 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
733 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 734 md_error(mddev, rdev);
3a0f5bbb 735 }
7bfa19f2 736
a9701a30
N		 737 if (atomic_dec_and_test(&mddev->pending_writes))
738 wake_up(&mddev->sb_wait);
f8b58edf 739 bio_put(bio);
7bfa19f2
N		 740}
741
fd01b88c 742void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 743 sector_t sector, int size, struct page *page)
744{
745 /* write first size bytes of page to sector of rdev
746 * Increment mddev->pending_writes before returning
747 * and decrement it on completion, waking up sb_wait
748 * if zero is reached.
749 * If an error occurred, call md_error
750 */
a167f663 751 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 752
a6ff7e08 753 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
4f024f37 754 bio->bi_iter.bi_sector = sector;
7bfa19f2
N		 755 bio_add_page(bio, page, size, 0);
756 bio->bi_private = rdev;
757 bio->bi_end_io = super_written;
a9701a30 758
7bfa19f2 759 atomic_inc(&mddev->pending_writes);
a5bf4df0 760 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 761}
762
fd01b88c 763void md_super_wait(struct mddev *mddev)
a9701a30 764{
e9c7469b 765 /* wait for all superblock writes that were scheduled to complete */
1967cd56 766 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
7bfa19f2
N		 767}
768
3cb03002 769int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 770 struct page *page, int rw, bool metadata_op)
1da177e4 771{
a167f663 772 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 773 int ret;
774
a6ff7e08
JB		 775 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
776 rdev->meta_bdev : rdev->bdev;
ccebd4c4 777 if (metadata_op)
4f024f37 778 bio->bi_iter.bi_sector = sector + rdev->sb_start;
1fdd6fc9
N		 779 else if (rdev->mddev->reshape_position != MaxSector &&
780 (rdev->mddev->reshape_backwards ==
781 (sector >= rdev->mddev->reshape_position)))
4f024f37 782 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
ccebd4c4 783 else
4f024f37 784 bio->bi_iter.bi_sector = sector + rdev->data_offset;
1da177e4 785 bio_add_page(bio, page, size, 0);
c170bbb4 786 submit_bio_wait(rw, bio);
1da177e4
LT		 787
788 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
789 bio_put(bio);
790 return ret;
791}
a8745db2 792EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 793
f72ffdd6 794static int read_disk_sb(struct md_rdev *rdev, int size)
1da177e4
LT		 795{
796 char b[BDEVNAME_SIZE];
403df478 797
1da177e4
LT		 798 if (rdev->sb_loaded)
799 return 0;
800
ccebd4c4 801 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 802 goto fail;
803 rdev->sb_loaded = 1;
804 return 0;
805
806fail:
807 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
808 bdevname(rdev->bdev,b));
809 return -EINVAL;
810}
811
812static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
813{
f72ffdd6 814 return sb1->set_uuid0 == sb2->set_uuid0 &&
05710466
AN		 815 sb1->set_uuid1 == sb2->set_uuid1 &&
816 sb1->set_uuid2 == sb2->set_uuid2 &&
817 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 818}
819
1da177e4
LT		 820static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
821{
822 int ret;
823 mdp_super_t *tmp1, *tmp2;
824
825 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
826 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
827
828 if (!tmp1 || !tmp2) {
829 ret = 0;
35020f1a 830 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 831 goto abort;
832 }
833
834 *tmp1 = *sb1;
835 *tmp2 = *sb2;
836
837 /*
838 * nr_disks is not constant
839 */
840 tmp1->nr_disks = 0;
841 tmp2->nr_disks = 0;
842
ce0c8e05 843 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 844abort:
990a8baf
JJ		 845 kfree(tmp1);
846 kfree(tmp2);
1da177e4
LT		 847 return ret;
848}
849
4d167f09
N		 850static u32 md_csum_fold(u32 csum)
851{
852 csum = (csum & 0xffff) + (csum >> 16);
853 return (csum & 0xffff) + (csum >> 16);
854}
855
f72ffdd6 856static unsigned int calc_sb_csum(mdp_super_t *sb)
1da177e4 857{
4d167f09
N		 858 u64 newcsum = 0;
859 u32 *sb32 = (u32*)sb;
860 int i;
1da177e4
LT		 861 unsigned int disk_csum, csum;
862
863 disk_csum = sb->sb_csum;
864 sb->sb_csum = 0;
4d167f09
N		 865
866 for (i = 0; i < MD_SB_BYTES/4 ; i++)
867 newcsum += sb32[i];
868 csum = (newcsum & 0xffffffff) + (newcsum>>32);
869
4d167f09
N		 870#ifdef CONFIG_ALPHA
871 /* This used to use csum_partial, which was wrong for several
872 * reasons including that different results are returned on
873 * different architectures. It isn't critical that we get exactly
874 * the same return value as before (we always csum_fold before
875 * testing, and that removes any differences). However as we
876 * know that csum_partial always returned a 16bit value on
877 * alphas, do a fold to maximise conformity to previous behaviour.
878 */
879 sb->sb_csum = md_csum_fold(disk_csum);
880#else
1da177e4 881 sb->sb_csum = disk_csum;
4d167f09 882#endif
1da177e4
LT		 883 return csum;
884}
885
1da177e4
LT		 886/*
887 * Handle superblock details.
888 * We want to be able to handle multiple superblock formats
889 * so we have a common interface to them all, and an array of
890 * different handlers.
891 * We rely on user-space to write the initial superblock, and support
892 * reading and updating of superblocks.
893 * Interface methods are:
3cb03002 894 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 895 * loads and validates a superblock on dev.
896 * if refdev != NULL, compare superblocks on both devices
897 * Return:
898 * 0 - dev has a superblock that is compatible with refdev
899 * 1 - dev has a superblock that is compatible and newer than refdev
900 * so dev should be used as the refdev in future
901 * -EINVAL superblock incompatible or invalid
902 * -othererror e.g. -EIO
903 *
fd01b88c 904 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 905 * Verify that dev is acceptable into mddev.
906 * The first time, mddev->raid_disks will be 0, and data from
907 * dev should be merged in. Subsequent calls check that dev
908 * is new enough. Return 0 or -EINVAL
909 *
fd01b88c 910 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 911 * Update the superblock for rdev with data in mddev
912 * This does not write to disc.
913 *
914 */
915
916struct super_type {
0cd17fec
CW		 917 char *name;
918 struct module *owner;
c6563a8c
N		 919 int (*load_super)(struct md_rdev *rdev,
920 struct md_rdev *refdev,
0cd17fec 921 int minor_version);
c6563a8c
N		 922 int (*validate_super)(struct mddev *mddev,
923 struct md_rdev *rdev);
924 void (*sync_super)(struct mddev *mddev,
925 struct md_rdev *rdev);
3cb03002 926 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 927 sector_t num_sectors);
c6563a8c
N		 928 int (*allow_new_offset)(struct md_rdev *rdev,
929 unsigned long long new_offset);
1da177e4
LT		 930};
931
0894cc30
AN		 932/*
933 * Check that the given mddev has no bitmap.
934 *
935 * This function is called from the run method of all personalities that do not
936 * support bitmaps. It prints an error message and returns non-zero if mddev
937 * has a bitmap. Otherwise, it returns 0.
938 *
939 */
fd01b88c 940int md_check_no_bitmap(struct mddev *mddev)
0894cc30 941{
c3d9714e 942 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 943 return 0;
944 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
945 mdname(mddev), mddev->pers->name);
946 return 1;
947}
948EXPORT_SYMBOL(md_check_no_bitmap);
949
1da177e4 950/*
f72ffdd6 951 * load_super for 0.90.0
1da177e4 952 */
3cb03002 953static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 954{
955 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
956 mdp_super_t *sb;
957 int ret;
1da177e4
LT		 958
959 /*
0f420358 960 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 961 * it's at the end of the disk.
962 *
963 * It also happens to be a multiple of 4Kb.
964 */
57b2caa3 965 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 966
0002b271 967 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 968 if (ret) return ret;
969
970 ret = -EINVAL;
971
972 bdevname(rdev->bdev, b);
65a06f06 973 sb = page_address(rdev->sb_page);
1da177e4
LT		 974
975 if (sb->md_magic != MD_SB_MAGIC) {
976 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
977 b);
978 goto abort;
979 }
980
981 if (sb->major_version != 0 ||
f6705578
N		 982 sb->minor_version < 90 ||
983 sb->minor_version > 91) {
1da177e4
LT		 984 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
985 sb->major_version, sb->minor_version,
986 b);
987 goto abort;
988 }
989
990 if (sb->raid_disks <= 0)
991 goto abort;
992
4d167f09 993 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 994 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
995 b);
996 goto abort;
997 }
998
999 rdev->preferred_minor = sb->md_minor;
1000 rdev->data_offset = 0;
c6563a8c 1001 rdev->new_data_offset = 0;
0002b271 1002 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1003 rdev->badblocks.shift = -1;
1da177e4
LT		 1004
1005 if (sb->level == LEVEL_MULTIPATH)
1006 rdev->desc_nr = -1;
1007 else
1008 rdev->desc_nr = sb->this_disk.number;
1009
9a7b2b0f 1010 if (!refdev) {
1da177e4 1011 ret = 1;
9a7b2b0f 1012 } else {
1da177e4 1013 __u64 ev1, ev2;
65a06f06 1014 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1015 if (!uuid_equal(refsb, sb)) {
1016 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1017 b, bdevname(refdev->bdev,b2));
1018 goto abort;
1019 }
1020 if (!sb_equal(refsb, sb)) {
1021 printk(KERN_WARNING "md: %s has same UUID"
1022 " but different superblock to %s\n",
1023 b, bdevname(refdev->bdev, b2));
1024 goto abort;
1025 }
1026 ev1 = md_event(sb);
1027 ev2 = md_event(refsb);
1028 if (ev1 > ev2)
1029 ret = 1;
f72ffdd6 1030 else
1da177e4
LT		 1031 ret = 0;
1032 }
8190e754 1033 rdev->sectors = rdev->sb_start;
667a5313
N		 1034 /* Limit to 4TB as metadata cannot record more than that.
1035 * (not needed for Linear and RAID0 as metadata doesn't
1036 * record this size)
1037 */
1038 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
27a7b260 1039 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1040
27a7b260 1041 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1042 /* "this cannot possibly happen" ... */
1043 ret = -EINVAL;
1044
1da177e4
LT		 1045 abort:
1046 return ret;
1047}
1048
1049/*
1050 * validate_super for 0.90.0
1051 */
fd01b88c 1052static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1053{
1054 mdp_disk_t *desc;
65a06f06 1055 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1056 __u64 ev1 = md_event(sb);
1da177e4 1057
41158c7e 1058 rdev->raid_disk = -1;
c5d79adb
N		 1059 clear_bit(Faulty, &rdev->flags);
1060 clear_bit(In_sync, &rdev->flags);
8313b8e5 1061 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1062 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1063
1da177e4
LT		 1064 if (mddev->raid_disks == 0) {
1065 mddev->major_version = 0;
1066 mddev->minor_version = sb->minor_version;
1067 mddev->patch_version = sb->patch_version;
e691063a 1068 mddev->external = 0;
9d8f0363 1069 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1070 mddev->ctime = sb->ctime;
1071 mddev->utime = sb->utime;
1072 mddev->level = sb->level;
d9d166c2 1073 mddev->clevel[0] = 0;
1da177e4
LT		 1074 mddev->layout = sb->layout;
1075 mddev->raid_disks = sb->raid_disks;
27a7b260 1076 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1077 mddev->events = ev1;
c3d9714e 1078 mddev->bitmap_info.offset = 0;
6409bb05
N		 1079 mddev->bitmap_info.space = 0;
1080 /* bitmap can use 60 K after the 4K superblocks */
c3d9714e 1081 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 1082 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
2c810cdd 1083 mddev->reshape_backwards = 0;
1da177e4 1084
f6705578
N		 1085 if (mddev->minor_version >= 91) {
1086 mddev->reshape_position = sb->reshape_position;
1087 mddev->delta_disks = sb->delta_disks;
1088 mddev->new_level = sb->new_level;
1089 mddev->new_layout = sb->new_layout;
664e7c41 1090 mddev->new_chunk_sectors = sb->new_chunk >> 9;
2c810cdd
N		 1091 if (mddev->delta_disks < 0)
1092 mddev->reshape_backwards = 1;
f6705578
N		 1093 } else {
1094 mddev->reshape_position = MaxSector;
1095 mddev->delta_disks = 0;
1096 mddev->new_level = mddev->level;
1097 mddev->new_layout = mddev->layout;
664e7c41 1098 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1099 }
1100
1da177e4
LT		 1101 if (sb->state & (1<<MD_SB_CLEAN))
1102 mddev->recovery_cp = MaxSector;
1103 else {
f72ffdd6 1104 if (sb->events_hi == sb->cp_events_hi &&
1da177e4
LT		 1105 sb->events_lo == sb->cp_events_lo) {
1106 mddev->recovery_cp = sb->recovery_cp;
1107 } else
1108 mddev->recovery_cp = 0;
1109 }
1110
1111 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1112 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1113 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1114 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1115
1116 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1117
1118 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
6409bb05 1119 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1120 mddev->bitmap_info.offset =
1121 mddev->bitmap_info.default_offset;
6409bb05 1122 mddev->bitmap_info.space =
c9ad020f 1123 mddev->bitmap_info.default_space;
6409bb05 1124 }
a654b9d8 1125
41158c7e 1126 } else if (mddev->pers == NULL) {
be6800a7
N		 1127 /* Insist on good event counter while assembling, except
1128 * for spares (which don't need an event count) */
1da177e4 1129 ++ev1;
be6800a7
N		 1130 if (sb->disks[rdev->desc_nr].state & (
1131 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
f72ffdd6 1132 if (ev1 < mddev->events)
be6800a7 1133 return -EINVAL;
41158c7e
N		 1134 } else if (mddev->bitmap) {
1135 /* if adding to array with a bitmap, then we can accept an
1136 * older device ... but not too old.
1137 */
41158c7e
N		 1138 if (ev1 < mddev->bitmap->events_cleared)
1139 return 0;
8313b8e5
N		 1140 if (ev1 < mddev->events)
1141 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1142 } else {
1143 if (ev1 < mddev->events)
1144 /* just a hot-add of a new device, leave raid_disk at -1 */
1145 return 0;
1146 }
41158c7e 1147
1da177e4 1148 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1149 desc = sb->disks + rdev->desc_nr;
1150
1151 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1152 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1153 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1154 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1155 set_bit(In_sync, &rdev->flags);
1da177e4 1156 rdev->raid_disk = desc->raid_disk;
f466722c 1157 rdev->saved_raid_disk = desc->raid_disk;
0261cd9f
N		 1158 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1159 /* active but not in sync implies recovery up to
1160 * reshape position. We don't know exactly where
1161 * that is, so set to zero for now */
1162 if (mddev->minor_version >= 91) {
1163 rdev->recovery_offset = 0;
1164 rdev->raid_disk = desc->raid_disk;
1165 }
1da177e4 1166 }
8ddf9efe
N		 1167 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1168 set_bit(WriteMostly, &rdev->flags);
41158c7e 1169 } else /* MULTIPATH are always insync */
b2d444d7 1170 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1171 return 0;
1172}
1173
1174/*
1175 * sync_super for 0.90.0
1176 */
fd01b88c 1177static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1178{
1179 mdp_super_t *sb;
3cb03002 1180 struct md_rdev *rdev2;
1da177e4 1181 int next_spare = mddev->raid_disks;
19133a42 1182
1da177e4
LT		 1183 /* make rdev->sb match mddev data..
1184 *
1185 * 1/ zero out disks
1186 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1187 * 3/ any empty disks < next_spare become removed
1188 *
1189 * disks[0] gets initialised to REMOVED because
1190 * we cannot be sure from other fields if it has
1191 * been initialised or not.
1192 */
1193 int i;
1194 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1195
61181565
N		 1196 rdev->sb_size = MD_SB_BYTES;
1197
65a06f06 1198 sb = page_address(rdev->sb_page);
1da177e4
LT		 1199
1200 memset(sb, 0, sizeof(*sb));
1201
1202 sb->md_magic = MD_SB_MAGIC;
1203 sb->major_version = mddev->major_version;
1da177e4
LT		 1204 sb->patch_version = mddev->patch_version;
1205 sb->gvalid_words = 0; /* ignored */
1206 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1207 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1208 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1209 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1210
1211 sb->ctime = mddev->ctime;
1212 sb->level = mddev->level;
58c0fed4 1213 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1214 sb->raid_disks = mddev->raid_disks;
1215 sb->md_minor = mddev->md_minor;
e691063a 1216 sb->not_persistent = 0;
1da177e4
LT		 1217 sb->utime = mddev->utime;
1218 sb->state = 0;
1219 sb->events_hi = (mddev->events>>32);
1220 sb->events_lo = (u32)mddev->events;
1221
f6705578
N		 1222 if (mddev->reshape_position == MaxSector)
1223 sb->minor_version = 90;
1224 else {
1225 sb->minor_version = 91;
1226 sb->reshape_position = mddev->reshape_position;
1227 sb->new_level = mddev->new_level;
1228 sb->delta_disks = mddev->delta_disks;
1229 sb->new_layout = mddev->new_layout;
664e7c41 1230 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1231 }
1232 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1233 if (mddev->in_sync)
1234 {
1235 sb->recovery_cp = mddev->recovery_cp;
1236 sb->cp_events_hi = (mddev->events>>32);
1237 sb->cp_events_lo = (u32)mddev->events;
1238 if (mddev->recovery_cp == MaxSector)
1239 sb->state = (1<< MD_SB_CLEAN);
1240 } else
1241 sb->recovery_cp = 0;
1242
1243 sb->layout = mddev->layout;
9d8f0363 1244 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1245
c3d9714e 1246 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1247 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1248
1da177e4 1249 sb->disks[0].state = (1<<MD_DISK_REMOVED);
dafb20fa 1250 rdev_for_each(rdev2, mddev) {
1da177e4 1251 mdp_disk_t *d;
86e6ffdd 1252 int desc_nr;
0261cd9f
N		 1253 int is_active = test_bit(In_sync, &rdev2->flags);
1254
1255 if (rdev2->raid_disk >= 0 &&
1256 sb->minor_version >= 91)
1257 /* we have nowhere to store the recovery_offset,
1258 * but if it is not below the reshape_position,
1259 * we can piggy-back on that.
1260 */
1261 is_active = 1;
1262 if (rdev2->raid_disk < 0 ||
1263 test_bit(Faulty, &rdev2->flags))
1264 is_active = 0;
1265 if (is_active)
86e6ffdd 1266 desc_nr = rdev2->raid_disk;
1da177e4 1267 else
86e6ffdd 1268 desc_nr = next_spare++;
19133a42 1269 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1270 d = &sb->disks[rdev2->desc_nr];
1271 nr_disks++;
1272 d->number = rdev2->desc_nr;
1273 d->major = MAJOR(rdev2->bdev->bd_dev);
1274 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1275 if (is_active)
1da177e4
LT		 1276 d->raid_disk = rdev2->raid_disk;
1277 else
1278 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1279 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1280 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1281 else if (is_active) {
1da177e4 1282 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1283 if (test_bit(In_sync, &rdev2->flags))
1284 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1285 active++;
1286 working++;
1287 } else {
1288 d->state = 0;
1289 spare++;
1290 working++;
1291 }
8ddf9efe
N		 1292 if (test_bit(WriteMostly, &rdev2->flags))
1293 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1294 }
1da177e4
LT		 1295 /* now set the "removed" and "faulty" bits on any missing devices */
1296 for (i=0 ; i < mddev->raid_disks ; i++) {
1297 mdp_disk_t *d = &sb->disks[i];
1298 if (d->state == 0 && d->number == 0) {
1299 d->number = i;
1300 d->raid_disk = i;
1301 d->state = (1<<MD_DISK_REMOVED);
1302 d->state |= (1<<MD_DISK_FAULTY);
1303 failed++;
1304 }
1305 }
1306 sb->nr_disks = nr_disks;
1307 sb->active_disks = active;
1308 sb->working_disks = working;
1309 sb->failed_disks = failed;
1310 sb->spare_disks = spare;
1311
1312 sb->this_disk = sb->disks[rdev->desc_nr];
1313 sb->sb_csum = calc_sb_csum(sb);
1314}
1315
0cd17fec
CW		 1316/*
1317 * rdev_size_change for 0.90.0
1318 */
1319static unsigned long long
3cb03002 1320super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1321{
58c0fed4 1322 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1323 return 0; /* component must fit device */
c3d9714e 1324 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1325 return 0; /* can't move bitmap */
57b2caa3 1326 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1327 if (!num_sectors || num_sectors > rdev->sb_start)
1328 num_sectors = rdev->sb_start;
27a7b260
N		 1329 /* Limit to 4TB as metadata cannot record more than that.
1330 * 4TB == 2^32 KB, or 2*2^32 sectors.
1331 */
667a5313 1332 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
27a7b260 1333 num_sectors = (2ULL << 32) - 2;
0f420358 1334 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1335 rdev->sb_page);
1336 md_super_wait(rdev->mddev);
c26a44ed 1337 return num_sectors;
0cd17fec
CW		 1338}
1339
c6563a8c
N		 1340static int
1341super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1342{
1343 /* non-zero offset changes not possible with v0.90 */
1344 return new_offset == 0;
1345}
0cd17fec 1346
1da177e4
LT		 1347/*
1348 * version 1 superblock
1349 */
1350
f72ffdd6 1351static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1da177e4 1352{
1c05b4bc
N		 1353 __le32 disk_csum;
1354 u32 csum;
1da177e4
LT		 1355 unsigned long long newcsum;
1356 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1357 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1358
1359 disk_csum = sb->sb_csum;
1360 sb->sb_csum = 0;
1361 newcsum = 0;
1f3c9907 1362 for (; size >= 4; size -= 4)
1da177e4
LT		 1363 newcsum += le32_to_cpu(*isuper++);
1364
1365 if (size == 2)
1c05b4bc 1366 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1367
1368 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1369 sb->sb_csum = disk_csum;
1370 return cpu_to_le32(csum);
1371}
1372
2699b672
N		 1373static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1374 int acknowledged);
3cb03002 1375static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1376{
1377 struct mdp_superblock_1 *sb;
1378 int ret;
0f420358 1379 sector_t sb_start;
c6563a8c 1380 sector_t sectors;
1da177e4 1381 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1382 int bmask;
1da177e4
LT		 1383
1384 /*
0f420358 1385 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1386 * It is always aligned to a 4K boundary and
1387 * depeding on minor_version, it can be:
1388 * 0: At least 8K, but less than 12K, from end of device
1389 * 1: At start of device
1390 * 2: 4K from start of device.
1391 */
1392 switch(minor_version) {
1393 case 0:
77304d2a 1394 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1395 sb_start -= 8*2;
1396 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1397 break;
1398 case 1:
0f420358 1399 sb_start = 0;
1da177e4
LT		 1400 break;
1401 case 2:
0f420358 1402 sb_start = 8;
1da177e4
LT		 1403 break;
1404 default:
1405 return -EINVAL;
1406 }
0f420358 1407 rdev->sb_start = sb_start;
1da177e4 1408
0002b271
N		 1409 /* superblock is rarely larger than 1K, but it can be larger,
1410 * and it is safe to read 4k, so we do that
1411 */
1412 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1413 if (ret) return ret;
1414
65a06f06 1415 sb = page_address(rdev->sb_page);
1da177e4
LT		 1416
1417 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1418 sb->major_version != cpu_to_le32(1) ||
1419 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1420 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1421 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1422 return -EINVAL;
1423
1424 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1425 printk("md: invalid superblock checksum on %s\n",
1426 bdevname(rdev->bdev,b));
1427 return -EINVAL;
1428 }
1429 if (le64_to_cpu(sb->data_size) < 10) {
1430 printk("md: data_size too small on %s\n",
1431 bdevname(rdev->bdev,b));
1432 return -EINVAL;
1433 }
c6563a8c
N		 1434 if (sb->pad0 ||
1435 sb->pad3[0] ||
1436 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1437 /* Some padding is non-zero, might be a new feature */
1438 return -EINVAL;
e11e93fa 1439
1da177e4
LT		 1440 rdev->preferred_minor = 0xffff;
1441 rdev->data_offset = le64_to_cpu(sb->data_offset);
c6563a8c
N		 1442 rdev->new_data_offset = rdev->data_offset;
1443 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1444 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1445 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
4dbcdc75 1446 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1447
0002b271 1448 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1449 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1450 if (rdev->sb_size & bmask)
a1801f85
N		 1451 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1452
1453 if (minor_version
0f420358 1454 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1455 return -EINVAL;
c6563a8c
N		 1456 if (minor_version
1457 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1458 return -EINVAL;
0002b271 1459
31b65a0d
N		 1460 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1461 rdev->desc_nr = -1;
1462 else
1463 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1464
2699b672
N		 1465 if (!rdev->bb_page) {
1466 rdev->bb_page = alloc_page(GFP_KERNEL);
1467 if (!rdev->bb_page)
1468 return -ENOMEM;
1469 }
1470 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1471 rdev->badblocks.count == 0) {
1472 /* need to load the bad block list.
1473 * Currently we limit it to one page.
1474 */
1475 s32 offset;
1476 sector_t bb_sector;
1477 u64 *bbp;
1478 int i;
1479 int sectors = le16_to_cpu(sb->bblog_size);
1480 if (sectors > (PAGE_SIZE / 512))
1481 return -EINVAL;
1482 offset = le32_to_cpu(sb->bblog_offset);
1483 if (offset == 0)
1484 return -EINVAL;
1485 bb_sector = (long long)offset;
1486 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1487 rdev->bb_page, READ, true))
1488 return -EIO;
1489 bbp = (u64 *)page_address(rdev->bb_page);
1490 rdev->badblocks.shift = sb->bblog_shift;
1491 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1492 u64 bb = le64_to_cpu(*bbp);
1493 int count = bb & (0x3ff);
1494 u64 sector = bb >> 10;
1495 sector <<= sb->bblog_shift;
1496 count <<= sb->bblog_shift;
1497 if (bb + 1 == 0)
1498 break;
1499 if (md_set_badblocks(&rdev->badblocks,
1500 sector, count, 1) == 0)
1501 return -EINVAL;
1502 }
486adf72
N		 1503 } else if (sb->bblog_offset != 0)
1504 rdev->badblocks.shift = 0;
2699b672 1505
9a7b2b0f 1506 if (!refdev) {
8ed75463 1507 ret = 1;
9a7b2b0f 1508 } else {
1da177e4 1509 __u64 ev1, ev2;
65a06f06 1510 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1511
1512 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1513 sb->level != refsb->level ||
1514 sb->layout != refsb->layout ||
1515 sb->chunksize != refsb->chunksize) {
1516 printk(KERN_WARNING "md: %s has strangely different"
1517 " superblock to %s\n",
1518 bdevname(rdev->bdev,b),
1519 bdevname(refdev->bdev,b2));
1520 return -EINVAL;
1521 }
1522 ev1 = le64_to_cpu(sb->events);
1523 ev2 = le64_to_cpu(refsb->events);
1524
1525 if (ev1 > ev2)
8ed75463
N		 1526 ret = 1;
1527 else
1528 ret = 0;
1da177e4 1529 }
c6563a8c
N		 1530 if (minor_version) {
1531 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1532 sectors -= rdev->data_offset;
1533 } else
1534 sectors = rdev->sb_start;
1535 if (sectors < le64_to_cpu(sb->data_size))
1da177e4 1536 return -EINVAL;
dd8ac336 1537 rdev->sectors = le64_to_cpu(sb->data_size);
8ed75463 1538 return ret;
1da177e4
LT		 1539}
1540
fd01b88c 1541static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1542{
65a06f06 1543 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1544 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1545
41158c7e 1546 rdev->raid_disk = -1;
c5d79adb
N		 1547 clear_bit(Faulty, &rdev->flags);
1548 clear_bit(In_sync, &rdev->flags);
8313b8e5 1549 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1550 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1551
1da177e4
LT		 1552 if (mddev->raid_disks == 0) {
1553 mddev->major_version = 1;
1554 mddev->patch_version = 0;
e691063a 1555 mddev->external = 0;
9d8f0363 1556 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1557 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1558 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1559 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1560 mddev->clevel[0] = 0;
1da177e4
LT		 1561 mddev->layout = le32_to_cpu(sb->layout);
1562 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1563 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1564 mddev->events = ev1;
c3d9714e 1565 mddev->bitmap_info.offset = 0;
6409bb05
N		 1566 mddev->bitmap_info.space = 0;
1567 /* Default location for bitmap is 1K after superblock
1568 * using 3K - total of 4K
1569 */
c3d9714e 1570 mddev->bitmap_info.default_offset = 1024 >> 9;
6409bb05 1571 mddev->bitmap_info.default_space = (4096-1024) >> 9;
2c810cdd
N		 1572 mddev->reshape_backwards = 0;
1573
1da177e4
LT		 1574 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1575 memcpy(mddev->uuid, sb->set_uuid, 16);
1576
1577 mddev->max_disks = (4096-256)/2;
a654b9d8 1578
71c0805c 1579 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
6409bb05 1580 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1581 mddev->bitmap_info.offset =
1582 (__s32)le32_to_cpu(sb->bitmap_offset);
6409bb05
N		 1583 /* Metadata doesn't record how much space is available.
1584 * For 1.0, we assume we can use up to the superblock
1585 * if before, else to 4K beyond superblock.
1586 * For others, assume no change is possible.
1587 */
1588 if (mddev->minor_version > 0)
1589 mddev->bitmap_info.space = 0;
1590 else if (mddev->bitmap_info.offset > 0)
1591 mddev->bitmap_info.space =
1592 8 - mddev->bitmap_info.offset;
1593 else
1594 mddev->bitmap_info.space =
1595 -mddev->bitmap_info.offset;
1596 }
e11e93fa 1597
f6705578
N		 1598 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1599 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1600 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1601 mddev->new_level = le32_to_cpu(sb->new_level);
1602 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1603 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
2c810cdd
N		 1604 if (mddev->delta_disks < 0 ||
1605 (mddev->delta_disks == 0 &&
1606 (le32_to_cpu(sb->feature_map)
1607 & MD_FEATURE_RESHAPE_BACKWARDS)))
1608 mddev->reshape_backwards = 1;
f6705578
N		 1609 } else {
1610 mddev->reshape_position = MaxSector;
1611 mddev->delta_disks = 0;
1612 mddev->new_level = mddev->level;
1613 mddev->new_layout = mddev->layout;
664e7c41 1614 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1615 }
1616
41158c7e 1617 } else if (mddev->pers == NULL) {
be6800a7
N		 1618 /* Insist of good event counter while assembling, except for
1619 * spares (which don't need an event count) */
1da177e4 1620 ++ev1;
be6800a7
N		 1621 if (rdev->desc_nr >= 0 &&
1622 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1623 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1624 if (ev1 < mddev->events)
1625 return -EINVAL;
41158c7e
N		 1626 } else if (mddev->bitmap) {
1627 /* If adding to array with a bitmap, then we can accept an
1628 * older device, but not too old.
1629 */
41158c7e
N		 1630 if (ev1 < mddev->bitmap->events_cleared)
1631 return 0;
8313b8e5
N		 1632 if (ev1 < mddev->events)
1633 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1634 } else {
1635 if (ev1 < mddev->events)
1636 /* just a hot-add of a new device, leave raid_disk at -1 */
1637 return 0;
1638 }
1da177e4
LT		 1639 if (mddev->level != LEVEL_MULTIPATH) {
1640 int role;
3673f305
N		 1641 if (rdev->desc_nr < 0 ||
1642 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1643 role = 0xffff;
1644 rdev->desc_nr = -1;
1645 } else
1646 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1647 switch(role) {
1648 case 0xffff: /* spare */
1da177e4
LT		 1649 break;
1650 case 0xfffe: /* faulty */
b2d444d7 1651 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1652 break;
1653 default:
f466722c 1654 rdev->saved_raid_disk = role;
5fd6c1dc 1655 if ((le32_to_cpu(sb->feature_map) &
f466722c 1656 MD_FEATURE_RECOVERY_OFFSET)) {
5fd6c1dc 1657 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
f466722c
N		 1658 if (!(le32_to_cpu(sb->feature_map) &
1659 MD_FEATURE_RECOVERY_BITMAP))
1660 rdev->saved_raid_disk = -1;
1661 } else
5fd6c1dc 1662 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1663 rdev->raid_disk = role;
1664 break;
1665 }
8ddf9efe
N		 1666 if (sb->devflags & WriteMostly1)
1667 set_bit(WriteMostly, &rdev->flags);
2d78f8c4
N		 1668 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1669 set_bit(Replacement, &rdev->flags);
41158c7e 1670 } else /* MULTIPATH are always insync */
b2d444d7 1671 set_bit(In_sync, &rdev->flags);
41158c7e 1672
1da177e4
LT		 1673 return 0;
1674}
1675
fd01b88c 1676static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1677{
1678 struct mdp_superblock_1 *sb;
3cb03002 1679 struct md_rdev *rdev2;
1da177e4
LT		 1680 int max_dev, i;
1681 /* make rdev->sb match mddev and rdev data. */
1682
65a06f06 1683 sb = page_address(rdev->sb_page);
1da177e4
LT		 1684
1685 sb->feature_map = 0;
1686 sb->pad0 = 0;
5fd6c1dc 1687 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1688 memset(sb->pad3, 0, sizeof(sb->pad3));
1689
1690 sb->utime = cpu_to_le64((__u64)mddev->utime);
1691 sb->events = cpu_to_le64(mddev->events);
1692 if (mddev->in_sync)
1693 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1694 else
1695 sb->resync_offset = cpu_to_le64(0);
1696
1c05b4bc 1697 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1698
f0ca340c 1699 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1700 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1701 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1702 sb->level = cpu_to_le32(mddev->level);
1703 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1704
aeb9b211
N		 1705 if (test_bit(WriteMostly, &rdev->flags))
1706 sb->devflags |= WriteMostly1;
1707 else
1708 sb->devflags &= ~WriteMostly1;
c6563a8c
N		 1709 sb->data_offset = cpu_to_le64(rdev->data_offset);
1710 sb->data_size = cpu_to_le64(rdev->sectors);
aeb9b211 1711
c3d9714e
N		 1712 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1713 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1714 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1715 }
5fd6c1dc
N		 1716
1717 if (rdev->raid_disk >= 0 &&
97e4f42d 1718 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1719 sb->feature_map |=
1720 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1721 sb->recovery_offset =
1722 cpu_to_le64(rdev->recovery_offset);
f466722c
N		 1723 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1724 sb->feature_map |=
1725 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
5fd6c1dc 1726 }
2d78f8c4
N		 1727 if (test_bit(Replacement, &rdev->flags))
1728 sb->feature_map |=
1729 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1730
f6705578
N		 1731 if (mddev->reshape_position != MaxSector) {
1732 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1733 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1734 sb->new_layout = cpu_to_le32(mddev->new_layout);
1735 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1736 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1737 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2c810cdd
N		 1738 if (mddev->delta_disks == 0 &&
1739 mddev->reshape_backwards)
1740 sb->feature_map
1741 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
c6563a8c
N		 1742 if (rdev->new_data_offset != rdev->data_offset) {
1743 sb->feature_map
1744 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1745 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1746 - rdev->data_offset));
1747 }
f6705578 1748 }
a654b9d8 1749
2699b672
N		 1750 if (rdev->badblocks.count == 0)
1751 /* Nothing to do for bad blocks*/ ;
1752 else if (sb->bblog_offset == 0)
1753 /* Cannot record bad blocks on this device */
1754 md_error(mddev, rdev);
1755 else {
1756 struct badblocks *bb = &rdev->badblocks;
1757 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1758 u64 *p = bb->page;
1759 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1760 if (bb->changed) {
1761 unsigned seq;
1762
1763retry:
1764 seq = read_seqbegin(&bb->lock);
1765
1766 memset(bbp, 0xff, PAGE_SIZE);
1767
1768 for (i = 0 ; i < bb->count ; i++) {
35f9ac2d 1769 u64 internal_bb = p[i];
2699b672
N		 1770 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1771 | BB_LEN(internal_bb));
35f9ac2d 1772 bbp[i] = cpu_to_le64(store_bb);
2699b672 1773 }
d0962936 1774 bb->changed = 0;
2699b672
N		 1775 if (read_seqretry(&bb->lock, seq))
1776 goto retry;
1777
1778 bb->sector = (rdev->sb_start +
1779 (int)le32_to_cpu(sb->bblog_offset));
1780 bb->size = le16_to_cpu(sb->bblog_size);
2699b672
N		 1781 }
1782 }
1783
1da177e4 1784 max_dev = 0;
dafb20fa 1785 rdev_for_each(rdev2, mddev)
1da177e4
LT		 1786 if (rdev2->desc_nr+1 > max_dev)
1787 max_dev = rdev2->desc_nr+1;
a778b73f 1788
70471daf
N		 1789 if (max_dev > le32_to_cpu(sb->max_dev)) {
1790 int bmask;
a778b73f 1791 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1792 rdev->sb_size = max_dev * 2 + 256;
1793 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1794 if (rdev->sb_size & bmask)
1795 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1796 } else
1797 max_dev = le32_to_cpu(sb->max_dev);
1798
1da177e4
LT		 1799 for (i=0; i<max_dev;i++)
1800 sb->dev_roles[i] = cpu_to_le16(0xfffe);
f72ffdd6 1801
dafb20fa 1802 rdev_for_each(rdev2, mddev) {
1da177e4 1803 i = rdev2->desc_nr;
b2d444d7 1804 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1805 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1806 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1807 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1808 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1809 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1810 else
1811 sb->dev_roles[i] = cpu_to_le16(0xffff);
1812 }
1813
1da177e4
LT		 1814 sb->sb_csum = calc_sb_1_csum(sb);
1815}
1816
0cd17fec 1817static unsigned long long
3cb03002 1818super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 1819{
1820 struct mdp_superblock_1 *sb;
15f4a5fd 1821 sector_t max_sectors;
58c0fed4 1822 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1823 return 0; /* component must fit device */
c6563a8c
N		 1824 if (rdev->data_offset != rdev->new_data_offset)
1825 return 0; /* too confusing */
0f420358 1826 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1827 /* minor versions 1 and 2; superblock before data */
77304d2a 1828 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1829 max_sectors -= rdev->data_offset;
1830 if (!num_sectors || num_sectors > max_sectors)
1831 num_sectors = max_sectors;
c3d9714e 1832 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1833 /* minor version 0 with bitmap we can't move */
1834 return 0;
1835 } else {
1836 /* minor version 0; superblock after data */
0f420358 1837 sector_t sb_start;
77304d2a 1838 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1839 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1840 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1841 if (!num_sectors || num_sectors > max_sectors)
1842 num_sectors = max_sectors;
0f420358 1843 rdev->sb_start = sb_start;
0cd17fec 1844 }
65a06f06 1845 sb = page_address(rdev->sb_page);
15f4a5fd 1846 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1847 sb->super_offset = rdev->sb_start;
0cd17fec 1848 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1849 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1850 rdev->sb_page);
1851 md_super_wait(rdev->mddev);
c26a44ed 1852 return num_sectors;
c6563a8c
N		 1853
1854}
1855
1856static int
1857super_1_allow_new_offset(struct md_rdev *rdev,
1858 unsigned long long new_offset)
1859{
1860 /* All necessary checks on new >= old have been done */
1861 struct bitmap *bitmap;
1862 if (new_offset >= rdev->data_offset)
1863 return 1;
1864
1865 /* with 1.0 metadata, there is no metadata to tread on
1866 * so we can always move back */
1867 if (rdev->mddev->minor_version == 0)
1868 return 1;
1869
1870 /* otherwise we must be sure not to step on
1871 * any metadata, so stay:
1872 * 36K beyond start of superblock
1873 * beyond end of badblocks
1874 * beyond write-intent bitmap
1875 */
1876 if (rdev->sb_start + (32+4)*2 > new_offset)
1877 return 0;
1878 bitmap = rdev->mddev->bitmap;
1879 if (bitmap && !rdev->mddev->bitmap_info.file &&
1880 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1ec885cd 1881 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
c6563a8c
N		 1882 return 0;
1883 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1884 return 0;
1885
1886 return 1;
0cd17fec 1887}
1da177e4 1888
75c96f85 1889static struct super_type super_types[] = {
1da177e4
LT		 1890 [0] = {
1891 .name = "0.90.0",
1892 .owner = THIS_MODULE,
0cd17fec
CW		 1893 .load_super = super_90_load,
1894 .validate_super = super_90_validate,
1895 .sync_super = super_90_sync,
1896 .rdev_size_change = super_90_rdev_size_change,
c6563a8c 1897 .allow_new_offset = super_90_allow_new_offset,
1da177e4
LT		 1898 },
1899 [1] = {
1900 .name = "md-1",
1901 .owner = THIS_MODULE,
0cd17fec
CW		 1902 .load_super = super_1_load,
1903 .validate_super = super_1_validate,
1904 .sync_super = super_1_sync,
1905 .rdev_size_change = super_1_rdev_size_change,
c6563a8c 1906 .allow_new_offset = super_1_allow_new_offset,
1da177e4
LT		 1907 },
1908};
1da177e4 1909
fd01b88c 1910static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 1911{
1912 if (mddev->sync_super) {
1913 mddev->sync_super(mddev, rdev);
1914 return;
1915 }
1916
1917 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1918
1919 super_types[mddev->major_version].sync_super(mddev, rdev);
1920}
1921
fd01b88c 1922static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1923{
3cb03002 1924 struct md_rdev *rdev, *rdev2;
1da177e4 1925
4b80991c
N		 1926 rcu_read_lock();
1927 rdev_for_each_rcu(rdev, mddev1)
1928 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1929 if (rdev->bdev->bd_contains ==
4b80991c
N		 1930 rdev2->bdev->bd_contains) {
1931 rcu_read_unlock();
7dd5e7c3 1932 return 1;
4b80991c
N		 1933 }
1934 rcu_read_unlock();
1da177e4
LT		 1935 return 0;
1936}
1937
1938static LIST_HEAD(pending_raid_disks);
1939
ac5e7113
AN		 1940/*
1941 * Try to register data integrity profile for an mddev
1942 *
1943 * This is called when an array is started and after a disk has been kicked
1944 * from the array. It only succeeds if all working and active component devices
1945 * are integrity capable with matching profiles.
1946 */
fd01b88c 1947int md_integrity_register(struct mddev *mddev)
ac5e7113 1948{
3cb03002 1949 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 1950
1951 if (list_empty(&mddev->disks))
1952 return 0; /* nothing to do */
629acb6a
JB		 1953 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1954 return 0; /* shouldn't register, or already is */
dafb20fa 1955 rdev_for_each(rdev, mddev) {
ac5e7113
AN		 1956 /* skip spares and non-functional disks */
1957 if (test_bit(Faulty, &rdev->flags))
1958 continue;
1959 if (rdev->raid_disk < 0)
1960 continue;
ac5e7113
AN		 1961 if (!reference) {
1962 /* Use the first rdev as the reference */
1963 reference = rdev;
1964 continue;
1965 }
1966 /* does this rdev's profile match the reference profile? */
1967 if (blk_integrity_compare(reference->bdev->bd_disk,
1968 rdev->bdev->bd_disk) < 0)
1969 return -EINVAL;
1970 }
89078d57
MP		 1971 if (!reference || !bdev_get_integrity(reference->bdev))
1972 return 0;
ac5e7113
AN		 1973 /*
1974 * All component devices are integrity capable and have matching
1975 * profiles, register the common profile for the md device.
1976 */
1977 if (blk_integrity_register(mddev->gendisk,
1978 bdev_get_integrity(reference->bdev)) != 0) {
1979 printk(KERN_ERR "md: failed to register integrity for %s\n",
1980 mdname(mddev));
1981 return -EINVAL;
1982 }
a91a2785
MP		 1983 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1984 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1985 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1986 mdname(mddev));
1987 return -EINVAL;
1988 }
ac5e7113
AN		 1989 return 0;
1990}
1991EXPORT_SYMBOL(md_integrity_register);
1992
1993/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 1994void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 1995{
2863b9eb
JB		 1996 struct blk_integrity *bi_rdev;
1997 struct blk_integrity *bi_mddev;
1998
1999 if (!mddev->gendisk)
2000 return;
2001
2002 bi_rdev = bdev_get_integrity(rdev->bdev);
2003 bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 2004
ac5e7113 2005 if (!bi_mddev) /* nothing to do */
3f9d99c1 2006 return;
ac5e7113 2007 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 2008 return;
ac5e7113
AN		 2009 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2010 rdev->bdev->bd_disk) >= 0)
2011 return;
2012 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2013 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2014}
ac5e7113 2015EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2016
f72ffdd6 2017static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
1da177e4 2018{
7dd5e7c3 2019 char b[BDEVNAME_SIZE];
f637b9f9 2020 struct kobject *ko;
1edf80d3 2021 char *s;
5e55e2f5 2022 int err;
1da177e4 2023
11e2ede0
DW		 2024 /* prevent duplicates */
2025 if (find_rdev(mddev, rdev->bdev->bd_dev))
2026 return -EEXIST;
2027
dd8ac336
AN		 2028 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2029 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2030 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2031 if (mddev->pers) {
2032 /* Cannot change size, so fail
2033 * If mddev->level <= 0, then we don't care
2034 * about aligning sizes (e.g. linear)
2035 */
2036 if (mddev->level > 0)
2037 return -ENOSPC;
2038 } else
dd8ac336 2039 mddev->dev_sectors = rdev->sectors;
2bf071bf 2040 }
1da177e4
LT		 2041
2042 /* Verify rdev->desc_nr is unique.
2043 * If it is -1, assign a free number, else
2044 * check number is not in use
2045 */
4878e9eb 2046 rcu_read_lock();
1da177e4
LT		 2047 if (rdev->desc_nr < 0) {
2048 int choice = 0;
4878e9eb
N		 2049 if (mddev->pers)
2050 choice = mddev->raid_disks;
2051 while (find_rdev_nr_rcu(mddev, choice))
1da177e4
LT		 2052 choice++;
2053 rdev->desc_nr = choice;
2054 } else {
4878e9eb
N		 2055 if (find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2056 rcu_read_unlock();
1da177e4 2057 return -EBUSY;
4878e9eb 2058 }
1da177e4 2059 }
4878e9eb 2060 rcu_read_unlock();
de01dfad
N		 2061 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2062 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2063 mdname(mddev), mddev->max_disks);
2064 return -EBUSY;
2065 }
19133a42 2066 bdevname(rdev->bdev,b);
649316b2 2067 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2068 *s = '!';
649316b2 2069
1da177e4 2070 rdev->mddev = mddev;
19133a42 2071 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2072
b2d6db58 2073 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2074 goto fail;
86e6ffdd 2075
0762b8bd 2076 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2077 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2078 /* failure here is OK */;
2079 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2080
4b80991c 2081 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2082 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2083
2084 /* May as well allow recovery to be retried once */
5389042f 2085 mddev->recovery_disabled++;
3f9d99c1 2086
1da177e4 2087 return 0;
5e55e2f5
N		 2088
2089 fail:
2090 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2091 b, mdname(mddev));
2092 return err;
1da177e4
LT		 2093}
2094
177a99b2 2095static void md_delayed_delete(struct work_struct *ws)
5792a285 2096{
3cb03002 2097 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2098 kobject_del(&rdev->kobj);
177a99b2 2099 kobject_put(&rdev->kobj);
5792a285
N		 2100}
2101
f72ffdd6 2102static void unbind_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2103{
2104 char b[BDEVNAME_SIZE];
403df478 2105
49731baa 2106 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2107 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2108 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2109 rdev->mddev = NULL;
86e6ffdd 2110 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2111 sysfs_put(rdev->sysfs_state);
2112 rdev->sysfs_state = NULL;
2230dfe4 2113 rdev->badblocks.count = 0;
5792a285 2114 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2115 * writing to 'remove' to "dev/state". We also need
2116 * to delay it due to rcu usage.
5792a285 2117 */
4b80991c 2118 synchronize_rcu();
177a99b2
N		 2119 INIT_WORK(&rdev->del_work, md_delayed_delete);
2120 kobject_get(&rdev->kobj);
e804ac78 2121 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2122}
2123
2124/*
2125 * prevent the device from being mounted, repartitioned or
2126 * otherwise reused by a RAID array (or any other kernel
2127 * subsystem), by bd_claiming the device.
2128 */
3cb03002 2129static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2130{
2131 int err = 0;
2132 struct block_device *bdev;
2133 char b[BDEVNAME_SIZE];
2134
d4d77629 2135 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2136 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2137 if (IS_ERR(bdev)) {
2138 printk(KERN_ERR "md: could not open %s.\n",
2139 __bdevname(dev, b));
2140 return PTR_ERR(bdev);
2141 }
1da177e4
LT		 2142 rdev->bdev = bdev;
2143 return err;
2144}
2145
3cb03002 2146static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2147{
2148 struct block_device *bdev = rdev->bdev;
2149 rdev->bdev = NULL;
e525fd89 2150 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2151}
2152
2153void md_autodetect_dev(dev_t dev);
2154
f72ffdd6 2155static void export_rdev(struct md_rdev *rdev)
1da177e4
LT		 2156{
2157 char b[BDEVNAME_SIZE];
403df478 2158
1da177e4
LT		 2159 printk(KERN_INFO "md: export_rdev(%s)\n",
2160 bdevname(rdev->bdev,b));
545c8795 2161 md_rdev_clear(rdev);
1da177e4 2162#ifndef MODULE
d0fae18f
N		 2163 if (test_bit(AutoDetected, &rdev->flags))
2164 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2165#endif
2166 unlock_rdev(rdev);
86e6ffdd 2167 kobject_put(&rdev->kobj);
1da177e4
LT		 2168}
2169
f72ffdd6 2170static void kick_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2171{
2172 unbind_rdev_from_array(rdev);
2173 export_rdev(rdev);
2174}
2175
fd01b88c 2176static void export_array(struct mddev *mddev)
1da177e4 2177{
0638bb0e 2178 struct md_rdev *rdev;
1da177e4 2179
0638bb0e
N		 2180 while (!list_empty(&mddev->disks)) {
2181 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2182 same_set);
1da177e4
LT		 2183 kick_rdev_from_array(rdev);
2184 }
1da177e4
LT		 2185 mddev->raid_disks = 0;
2186 mddev->major_version = 0;
2187}
2188
f72ffdd6 2189static void sync_sbs(struct mddev *mddev, int nospares)
1da177e4 2190{
42543769
N		 2191 /* Update each superblock (in-memory image), but
2192 * if we are allowed to, skip spares which already
2193 * have the right event counter, or have one earlier
2194 * (which would mean they aren't being marked as dirty
2195 * with the rest of the array)
2196 */
3cb03002 2197 struct md_rdev *rdev;
dafb20fa 2198 rdev_for_each(rdev, mddev) {
42543769
N		 2199 if (rdev->sb_events == mddev->events ||
2200 (nospares &&
2201 rdev->raid_disk < 0 &&
42543769
N		 2202 rdev->sb_events+1 == mddev->events)) {
2203 /* Don't update this superblock */
2204 rdev->sb_loaded = 2;
2205 } else {
076f968b 2206 sync_super(mddev, rdev);
42543769
N		 2207 rdev->sb_loaded = 1;
2208 }
1da177e4
LT		 2209 }
2210}
2211
f72ffdd6 2212static void md_update_sb(struct mddev *mddev, int force_change)
1da177e4 2213{
3cb03002 2214 struct md_rdev *rdev;
06d91a5f 2215 int sync_req;
42543769 2216 int nospares = 0;
2699b672 2217 int any_badblocks_changed = 0;
1da177e4 2218
d87f064f
N		 2219 if (mddev->ro) {
2220 if (force_change)
2221 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2222 return;
2223 }
1da177e4 2224repeat:
3a3a5ddb 2225 /* First make sure individual recovery_offsets are correct */
dafb20fa 2226 rdev_for_each(rdev, mddev) {
3a3a5ddb
N		 2227 if (rdev->raid_disk >= 0 &&
2228 mddev->delta_disks >= 0 &&
2229 !test_bit(In_sync, &rdev->flags) &&
2230 mddev->curr_resync_completed > rdev->recovery_offset)
2231 rdev->recovery_offset = mddev->curr_resync_completed;
2232
f72ffdd6 2233 }
bd52b746 2234 if (!mddev->persistent) {
070dc6dd 2235 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2236 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2237 if (!mddev->external) {
d97a41dc 2238 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
dafb20fa 2239 rdev_for_each(rdev, mddev) {
de393cde 2240 if (rdev->badblocks.changed) {
d0962936 2241 rdev->badblocks.changed = 0;
de393cde
N		 2242 md_ack_all_badblocks(&rdev->badblocks);
2243 md_error(mddev, rdev);
2244 }
2245 clear_bit(Blocked, &rdev->flags);
2246 clear_bit(BlockedBadBlocks, &rdev->flags);
2247 wake_up(&rdev->blocked_wait);
2248 }
2249 }
3a3a5ddb
N		 2250 wake_up(&mddev->sb_wait);
2251 return;
2252 }
2253
85572d7c 2254 spin_lock(&mddev->lock);
84692195 2255
3a3a5ddb
N		 2256 mddev->utime = get_seconds();
2257
850b2b42
N		 2258 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2259 force_change = 1;
2260 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2261 /* just a clean<-> dirty transition, possibly leave spares alone,
2262 * though if events isn't the right even/odd, we will have to do
2263 * spares after all
2264 */
2265 nospares = 1;
2266 if (force_change)
2267 nospares = 0;
2268 if (mddev->degraded)
84692195
N		 2269 /* If the array is degraded, then skipping spares is both
2270 * dangerous and fairly pointless.
2271 * Dangerous because a device that was removed from the array
2272 * might have a event_count that still looks up-to-date,
2273 * so it can be re-added without a resync.
2274 * Pointless because if there are any spares to skip,
2275 * then a recovery will happen and soon that array won't
2276 * be degraded any more and the spare can go back to sleep then.
2277 */
850b2b42 2278 nospares = 0;
84692195 2279
06d91a5f 2280 sync_req = mddev->in_sync;
42543769
N		 2281
2282 /* If this is just a dirty<->clean transition, and the array is clean
2283 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2284 if (nospares
42543769 2285 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2286 && mddev->can_decrease_events
2287 && mddev->events != 1) {
42543769 2288 mddev->events--;
a8707c08
N		 2289 mddev->can_decrease_events = 0;
2290 } else {
42543769
N		 2291 /* otherwise we have to go forward and ... */
2292 mddev->events ++;
a8707c08 2293 mddev->can_decrease_events = nospares;
42543769 2294 }
1da177e4 2295
403df478
N		 2296 /*
2297 * This 64-bit counter should never wrap.
2298 * Either we are in around ~1 trillion A.C., assuming
2299 * 1 reboot per second, or we have a bug...
2300 */
2301 WARN_ON(mddev->events == 0);
2699b672 2302
dafb20fa 2303 rdev_for_each(rdev, mddev) {
2699b672
N		 2304 if (rdev->badblocks.changed)
2305 any_badblocks_changed++;
de393cde
N		 2306 if (test_bit(Faulty, &rdev->flags))
2307 set_bit(FaultRecorded, &rdev->flags);
2308 }
2699b672 2309
e691063a 2310 sync_sbs(mddev, nospares);
85572d7c 2311 spin_unlock(&mddev->lock);
1da177e4 2312
36a4e1fe
N		 2313 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2314 mdname(mddev), mddev->in_sync);
1da177e4 2315
4ad13663 2316 bitmap_update_sb(mddev->bitmap);
dafb20fa 2317 rdev_for_each(rdev, mddev) {
1da177e4 2318 char b[BDEVNAME_SIZE];
36a4e1fe 2319
42543769
N		 2320 if (rdev->sb_loaded != 1)
2321 continue; /* no noise on spare devices */
1da177e4 2322
f466722c 2323 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2324 md_super_write(mddev,rdev,
0f420358 2325 rdev->sb_start, rdev->sb_size,
7bfa19f2 2326 rdev->sb_page);
36a4e1fe
N		 2327 pr_debug("md: (write) %s's sb offset: %llu\n",
2328 bdevname(rdev->bdev, b),
2329 (unsigned long long)rdev->sb_start);
42543769 2330 rdev->sb_events = mddev->events;
2699b672
N		 2331 if (rdev->badblocks.size) {
2332 md_super_write(mddev, rdev,
2333 rdev->badblocks.sector,
2334 rdev->badblocks.size << 9,
2335 rdev->bb_page);
2336 rdev->badblocks.size = 0;
2337 }
7bfa19f2 2338
f466722c 2339 } else
36a4e1fe
N		 2340 pr_debug("md: %s (skipping faulty)\n",
2341 bdevname(rdev->bdev, b));
d70ed2e4 2342
7bfa19f2 2343 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2344 /* only need to write one superblock... */
2345 break;
2346 }
a9701a30 2347 md_super_wait(mddev);
850b2b42 2348 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2349
85572d7c 2350 spin_lock(&mddev->lock);
850b2b42
N		 2351 if (mddev->in_sync != sync_req ||
2352 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2353 /* have to write it out again */
85572d7c 2354 spin_unlock(&mddev->lock);
06d91a5f
N		 2355 goto repeat;
2356 }
850b2b42 2357 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
85572d7c 2358 spin_unlock(&mddev->lock);
3d310eb7 2359 wake_up(&mddev->sb_wait);
acb180b0
N		 2360 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2361 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2362
dafb20fa 2363 rdev_for_each(rdev, mddev) {
de393cde
N		 2364 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2365 clear_bit(Blocked, &rdev->flags);
2366
2367 if (any_badblocks_changed)
2699b672 2368 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2369 clear_bit(BlockedBadBlocks, &rdev->flags);
2370 wake_up(&rdev->blocked_wait);
2371 }
1da177e4
LT		 2372}
2373
7f6ce769 2374/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2375 * We want to accept with case. For this we use cmd_match.
2376 */
2377static int cmd_match(const char *cmd, const char *str)
2378{
2379 /* See if cmd, written into a sysfs file, matches
2380 * str. They must either be the same, or cmd can
2381 * have a trailing newline
2382 */
2383 while (*cmd && *str && *cmd == *str) {
2384 cmd++;
2385 str++;
2386 }
2387 if (*cmd == '\n')
2388 cmd++;
2389 if (*str || *cmd)
2390 return 0;
2391 return 1;
2392}
2393
86e6ffdd
N		 2394struct rdev_sysfs_entry {
2395 struct attribute attr;
3cb03002
N		 2396 ssize_t (*show)(struct md_rdev *, char *);
2397 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2398};
2399
2400static ssize_t
3cb03002 2401state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2402{
2403 char *sep = "";
20a49ff6 2404 size_t len = 0;
758bfc8a 2405 unsigned long flags = ACCESS_ONCE(rdev->flags);
86e6ffdd 2406
758bfc8a 2407 if (test_bit(Faulty, &flags) ||
de393cde 2408 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2409 len+= sprintf(page+len, "%sfaulty",sep);
2410 sep = ",";
2411 }
758bfc8a 2412 if (test_bit(In_sync, &flags)) {
86e6ffdd
N		 2413 len += sprintf(page+len, "%sin_sync",sep);
2414 sep = ",";
2415 }
758bfc8a 2416 if (test_bit(WriteMostly, &flags)) {
f655675b
N		 2417 len += sprintf(page+len, "%swrite_mostly",sep);
2418 sep = ",";
2419 }
758bfc8a 2420 if (test_bit(Blocked, &flags) ||
52c64152 2421 (rdev->badblocks.unacked_exist
758bfc8a 2422 && !test_bit(Faulty, &flags))) {
6bfe0b49
DW		 2423 len += sprintf(page+len, "%sblocked", sep);
2424 sep = ",";
2425 }
758bfc8a
N		 2426 if (!test_bit(Faulty, &flags) &&
2427 !test_bit(In_sync, &flags)) {
86e6ffdd
N		 2428 len += sprintf(page+len, "%sspare", sep);
2429 sep = ",";
2430 }
758bfc8a 2431 if (test_bit(WriteErrorSeen, &flags)) {
d7a9d443
N		 2432 len += sprintf(page+len, "%swrite_error", sep);
2433 sep = ",";
2434 }
758bfc8a 2435 if (test_bit(WantReplacement, &flags)) {
2d78f8c4
N		 2436 len += sprintf(page+len, "%swant_replacement", sep);
2437 sep = ",";
2438 }
758bfc8a 2439 if (test_bit(Replacement, &flags)) {
2d78f8c4
N		 2440 len += sprintf(page+len, "%sreplacement", sep);
2441 sep = ",";
2442 }
2443
86e6ffdd
N		 2444 return len+sprintf(page+len, "\n");
2445}
2446
45dc2de1 2447static ssize_t
3cb03002 2448state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2449{
2450 /* can write
de393cde 2451 * faulty - simulates an error
45dc2de1 2452 * remove - disconnects the device
f655675b
N		 2453 * writemostly - sets write_mostly
2454 * -writemostly - clears write_mostly
de393cde
N		 2455 * blocked - sets the Blocked flags
2456 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2457 * insync - sets Insync providing device isn't active
f466722c
N		 2458 * -insync - clear Insync for a device with a slot assigned,
2459 * so that it gets rebuilt based on bitmap
d7a9d443
N		 2460 * write_error - sets WriteErrorSeen
2461 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2462 */
2463 int err = -EINVAL;
2464 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2465 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2466 if (test_bit(Faulty, &rdev->flags))
2467 err = 0;
2468 else
2469 err = -EBUSY;
45dc2de1
N		 2470 } else if (cmd_match(buf, "remove")) {
2471 if (rdev->raid_disk >= 0)
2472 err = -EBUSY;
2473 else {
fd01b88c 2474 struct mddev *mddev = rdev->mddev;
45dc2de1 2475 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2476 if (mddev->pers)
2477 md_update_sb(mddev, 1);
45dc2de1
N		 2478 md_new_event(mddev);
2479 err = 0;
2480 }
f655675b
N		 2481 } else if (cmd_match(buf, "writemostly")) {
2482 set_bit(WriteMostly, &rdev->flags);
2483 err = 0;
2484 } else if (cmd_match(buf, "-writemostly")) {
2485 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2486 err = 0;
2487 } else if (cmd_match(buf, "blocked")) {
2488 set_bit(Blocked, &rdev->flags);
2489 err = 0;
2490 } else if (cmd_match(buf, "-blocked")) {
de393cde 2491 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2492 rdev->badblocks.unacked_exist) {
de393cde
N		 2493 /* metadata handler doesn't understand badblocks,
2494 * so we need to fail the device
2495 */
2496 md_error(rdev->mddev, rdev);
2497 }
6bfe0b49 2498 clear_bit(Blocked, &rdev->flags);
de393cde 2499 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2500 wake_up(&rdev->blocked_wait);
2501 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2502 md_wakeup_thread(rdev->mddev->thread);
2503
6d56e278
N		 2504 err = 0;
2505 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2506 set_bit(In_sync, &rdev->flags);
f655675b 2507 err = 0;
f466722c 2508 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
e1960f8c
N		 2509 if (rdev->mddev->pers == NULL) {
2510 clear_bit(In_sync, &rdev->flags);
2511 rdev->saved_raid_disk = rdev->raid_disk;
2512 rdev->raid_disk = -1;
2513 err = 0;
2514 }
d7a9d443
N		 2515 } else if (cmd_match(buf, "write_error")) {
2516 set_bit(WriteErrorSeen, &rdev->flags);
2517 err = 0;
2518 } else if (cmd_match(buf, "-write_error")) {
2519 clear_bit(WriteErrorSeen, &rdev->flags);
2520 err = 0;
2d78f8c4
N		 2521 } else if (cmd_match(buf, "want_replacement")) {
2522 /* Any non-spare device that is not a replacement can
2523 * become want_replacement at any time, but we then need to
2524 * check if recovery is needed.
2525 */
2526 if (rdev->raid_disk >= 0 &&
2527 !test_bit(Replacement, &rdev->flags))
2528 set_bit(WantReplacement, &rdev->flags);
2529 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2530 md_wakeup_thread(rdev->mddev->thread);
2531 err = 0;
2532 } else if (cmd_match(buf, "-want_replacement")) {
2533 /* Clearing 'want_replacement' is always allowed.
2534 * Once replacements starts it is too late though.
2535 */
2536 err = 0;
2537 clear_bit(WantReplacement, &rdev->flags);
2538 } else if (cmd_match(buf, "replacement")) {
2539 /* Can only set a device as a replacement when array has not
2540 * yet been started. Once running, replacement is automatic
2541 * from spares, or by assigning 'slot'.
2542 */
2543 if (rdev->mddev->pers)
2544 err = -EBUSY;
2545 else {
2546 set_bit(Replacement, &rdev->flags);
2547 err = 0;
2548 }
2549 } else if (cmd_match(buf, "-replacement")) {
2550 /* Similarly, can only clear Replacement before start */
2551 if (rdev->mddev->pers)
2552 err = -EBUSY;
2553 else {
2554 clear_bit(Replacement, &rdev->flags);
2555 err = 0;
2556 }
45dc2de1 2557 }
00bcb4ac
N		 2558 if (!err)
2559 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2560 return err ? err : len;
2561}
80ca3a44
N		 2562static struct rdev_sysfs_entry rdev_state =
2563__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2564
4dbcdc75 2565static ssize_t
3cb03002 2566errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2567{
2568 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2569}
2570
2571static ssize_t
3cb03002 2572errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N		 2573{
2574 char *e;
2575 unsigned long n = simple_strtoul(buf, &e, 10);
2576 if (*buf && (*e == 0 || *e == '\n')) {
2577 atomic_set(&rdev->corrected_errors, n);
2578 return len;
2579 }
2580 return -EINVAL;
2581}
2582static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2583__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2584
014236d2 2585static ssize_t
3cb03002 2586slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2587{
2588 if (rdev->raid_disk < 0)
2589 return sprintf(page, "none\n");
2590 else
2591 return sprintf(page, "%d\n", rdev->raid_disk);
2592}
2593
2594static ssize_t
3cb03002 2595slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N		 2596{
2597 char *e;
c303da6d 2598 int err;
014236d2
N		 2599 int slot = simple_strtoul(buf, &e, 10);
2600 if (strncmp(buf, "none", 4)==0)
2601 slot = -1;
2602 else if (e==buf || (*e && *e!= '\n'))
2603 return -EINVAL;
6c2fce2e 2604 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2605 /* Setting 'slot' on an active array requires also
2606 * updating the 'rd%d' link, and communicating
2607 * with the personality with ->hot_*_disk.
2608 * For now we only support removing
2609 * failed/spare devices. This normally happens automatically,
2610 * but not when the metadata is externally managed.
2611 */
c303da6d
N		 2612 if (rdev->raid_disk == -1)
2613 return -EEXIST;
2614 /* personality does all needed checks */
01393f3d 2615 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d 2616 return -EINVAL;
746d3207
N		 2617 clear_bit(Blocked, &rdev->flags);
2618 remove_and_add_spares(rdev->mddev, rdev);
2619 if (rdev->raid_disk >= 0)
2620 return -EBUSY;
c303da6d
N		 2621 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2622 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2623 } else if (rdev->mddev->pers) {
6c2fce2e 2624 /* Activating a spare .. or possibly reactivating
6d56e278 2625 * if we ever get bitmaps working here.
6c2fce2e
NB		 2626 */
2627
2628 if (rdev->raid_disk != -1)
2629 return -EBUSY;
2630
c6751b2b
N		 2631 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2632 return -EBUSY;
2633
6c2fce2e
NB		 2634 if (rdev->mddev->pers->hot_add_disk == NULL)
2635 return -EINVAL;
2636
ba1b41b6
N		 2637 if (slot >= rdev->mddev->raid_disks &&
2638 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2639 return -ENOSPC;
2640
6c2fce2e
NB		 2641 rdev->raid_disk = slot;
2642 if (test_bit(In_sync, &rdev->flags))
2643 rdev->saved_raid_disk = slot;
2644 else
2645 rdev->saved_raid_disk = -1;
d30519fc 2646 clear_bit(In_sync, &rdev->flags);
8313b8e5 2647 clear_bit(Bitmap_sync, &rdev->flags);
6c2fce2e
NB		 2648 err = rdev->mddev->pers->
2649 hot_add_disk(rdev->mddev, rdev);
199050ea 2650 if (err) {
6c2fce2e 2651 rdev->raid_disk = -1;
6c2fce2e 2652 return err;
52664732 2653 } else
00bcb4ac 2654 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2655 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2656 /* failure here is OK */;
6c2fce2e 2657 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2658 } else {
ba1b41b6
N		 2659 if (slot >= rdev->mddev->raid_disks &&
2660 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2661 return -ENOSPC;
2662 rdev->raid_disk = slot;
2663 /* assume it is working */
c5d79adb
N		 2664 clear_bit(Faulty, &rdev->flags);
2665 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2666 set_bit(In_sync, &rdev->flags);
00bcb4ac 2667 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2668 }
014236d2
N		 2669 return len;
2670}
2671
014236d2 2672static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2673__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2674
93c8cad0 2675static ssize_t
3cb03002 2676offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2677{
6961ece4 2678 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2679}
2680
2681static ssize_t
3cb03002 2682offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2683{
c6563a8c 2684 unsigned long long offset;
b29bebd6 2685 if (kstrtoull(buf, 10, &offset) < 0)
93c8cad0 2686 return -EINVAL;
8ed0a521 2687 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2688 return -EBUSY;
dd8ac336 2689 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2690 /* Must set offset before size, so overlap checks
2691 * can be sane */
2692 return -EBUSY;
93c8cad0 2693 rdev->data_offset = offset;
25f7fd47 2694 rdev->new_data_offset = offset;
93c8cad0
N		 2695 return len;
2696}
2697
2698static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2699__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2700
c6563a8c
N		 2701static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2702{
2703 return sprintf(page, "%llu\n",
2704 (unsigned long long)rdev->new_data_offset);
2705}
2706
2707static ssize_t new_offset_store(struct md_rdev *rdev,
2708 const char *buf, size_t len)
2709{
2710 unsigned long long new_offset;
2711 struct mddev *mddev = rdev->mddev;
2712
b29bebd6 2713 if (kstrtoull(buf, 10, &new_offset) < 0)
c6563a8c
N		 2714 return -EINVAL;
2715
f851b60d
N		 2716 if (mddev->sync_thread ||
2717 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
c6563a8c
N		 2718 return -EBUSY;
2719 if (new_offset == rdev->data_offset)
2720 /* reset is always permitted */
2721 ;
2722 else if (new_offset > rdev->data_offset) {
2723 /* must not push array size beyond rdev_sectors */
2724 if (new_offset - rdev->data_offset
2725 + mddev->dev_sectors > rdev->sectors)
2726 return -E2BIG;
2727 }
2728 /* Metadata worries about other space details. */
2729
2730 /* decreasing the offset is inconsistent with a backwards
2731 * reshape.
2732 */
2733 if (new_offset < rdev->data_offset &&
2734 mddev->reshape_backwards)
2735 return -EINVAL;
2736 /* Increasing offset is inconsistent with forwards
2737 * reshape. reshape_direction should be set to
2738 * 'backwards' first.
2739 */
2740 if (new_offset > rdev->data_offset &&
2741 !mddev->reshape_backwards)
2742 return -EINVAL;
2743
2744 if (mddev->pers && mddev->persistent &&
2745 !super_types[mddev->major_version]
2746 .allow_new_offset(rdev, new_offset))
2747 return -E2BIG;
2748 rdev->new_data_offset = new_offset;
2749 if (new_offset > rdev->data_offset)
2750 mddev->reshape_backwards = 1;
2751 else if (new_offset < rdev->data_offset)
2752 mddev->reshape_backwards = 0;
2753
2754 return len;
2755}
2756static struct rdev_sysfs_entry rdev_new_offset =
2757__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2758
83303b61 2759static ssize_t
3cb03002 2760rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2761{
dd8ac336 2762 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2763}
2764
c5d79adb
N		 2765static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2766{
2767 /* check if two start/length pairs overlap */
2768 if (s1+l1 <= s2)
2769 return 0;
2770 if (s2+l2 <= s1)
2771 return 0;
2772 return 1;
2773}
2774
b522adcd
DW		 2775static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2776{
2777 unsigned long long blocks;
2778 sector_t new;
2779
b29bebd6 2780 if (kstrtoull(buf, 10, &blocks) < 0)
b522adcd
DW		 2781 return -EINVAL;
2782
2783 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2784 return -EINVAL; /* sector conversion overflow */
2785
2786 new = blocks * 2;
2787 if (new != blocks * 2)
2788 return -EINVAL; /* unsigned long long to sector_t overflow */
2789
2790 *sectors = new;
2791 return 0;
2792}
2793
83303b61 2794static ssize_t
3cb03002 2795rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2796{
fd01b88c 2797 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2798 sector_t oldsectors = rdev->sectors;
b522adcd 2799 sector_t sectors;
27c529bb 2800
b522adcd 2801 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2802 return -EINVAL;
c6563a8c
N		 2803 if (rdev->data_offset != rdev->new_data_offset)
2804 return -EINVAL; /* too confusing */
0cd17fec 2805 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2806 if (my_mddev->persistent) {
dd8ac336
AN		 2807 sectors = super_types[my_mddev->major_version].
2808 rdev_size_change(rdev, sectors);
2809 if (!sectors)
0cd17fec 2810 return -EBUSY;
dd8ac336 2811 } else if (!sectors)
77304d2a 2812 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2813 rdev->data_offset;
a6468539
N		 2814 if (!my_mddev->pers->resize)
2815 /* Cannot change size for RAID0 or Linear etc */
2816 return -EINVAL;
0cd17fec 2817 }
dd8ac336 2818 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2819 return -EINVAL; /* component must fit device */
0cd17fec 2820
dd8ac336
AN		 2821 rdev->sectors = sectors;
2822 if (sectors > oldsectors && my_mddev->external) {
8b1afc3d
N		 2823 /* Need to check that all other rdevs with the same
2824 * ->bdev do not overlap. 'rcu' is sufficient to walk
2825 * the rdev lists safely.
2826 * This check does not provide a hard guarantee, it
2827 * just helps avoid dangerous mistakes.
c5d79adb 2828 */
fd01b88c 2829 struct mddev *mddev;
c5d79adb 2830 int overlap = 0;
159ec1fc 2831 struct list_head *tmp;
c5d79adb 2832
8b1afc3d 2833 rcu_read_lock();
29ac4aa3 2834 for_each_mddev(mddev, tmp) {
3cb03002 2835 struct md_rdev *rdev2;
c5d79adb 2836
dafb20fa 2837 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 2838 if (rdev->bdev == rdev2->bdev &&
2839 rdev != rdev2 &&
2840 overlaps(rdev->data_offset, rdev->sectors,
2841 rdev2->data_offset,
2842 rdev2->sectors)) {
c5d79adb
N		 2843 overlap = 1;
2844 break;
2845 }
c5d79adb
N		 2846 if (overlap) {
2847 mddev_put(mddev);
2848 break;
2849 }
2850 }
8b1afc3d 2851 rcu_read_unlock();
c5d79adb
N		 2852 if (overlap) {
2853 /* Someone else could have slipped in a size
2854 * change here, but doing so is just silly.
dd8ac336 2855 * We put oldsectors back because we *know* it is
c5d79adb
N		 2856 * safe, and trust userspace not to race with
2857 * itself
2858 */
dd8ac336 2859 rdev->sectors = oldsectors;
c5d79adb
N		 2860 return -EBUSY;
2861 }
2862 }
83303b61
N		 2863 return len;
2864}
2865
2866static struct rdev_sysfs_entry rdev_size =
80ca3a44 2867__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2868
3cb03002 2869static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 2870{
2871 unsigned long long recovery_start = rdev->recovery_offset;
2872
2873 if (test_bit(In_sync, &rdev->flags) ||
2874 recovery_start == MaxSector)
2875 return sprintf(page, "none\n");
2876
2877 return sprintf(page, "%llu\n", recovery_start);
2878}
2879
3cb03002 2880static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 2881{
2882 unsigned long long recovery_start;
2883
2884 if (cmd_match(buf, "none"))
2885 recovery_start = MaxSector;
b29bebd6 2886 else if (kstrtoull(buf, 10, &recovery_start))
06e3c817
DW		 2887 return -EINVAL;
2888
2889 if (rdev->mddev->pers &&
2890 rdev->raid_disk >= 0)
2891 return -EBUSY;
2892
2893 rdev->recovery_offset = recovery_start;
2894 if (recovery_start == MaxSector)
2895 set_bit(In_sync, &rdev->flags);
2896 else
2897 clear_bit(In_sync, &rdev->flags);
2898 return len;
2899}
2900
2901static struct rdev_sysfs_entry rdev_recovery_start =
2902__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2903
16c791a5
N		 2904static ssize_t
2905badblocks_show(struct badblocks *bb, char *page, int unack);
2906static ssize_t
2907badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2908
3cb03002 2909static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2910{
2911 return badblocks_show(&rdev->badblocks, page, 0);
2912}
3cb03002 2913static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 2914{
de393cde
N		 2915 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2916 /* Maybe that ack was all we needed */
2917 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2918 wake_up(&rdev->blocked_wait);
2919 return rv;
16c791a5
N		 2920}
2921static struct rdev_sysfs_entry rdev_bad_blocks =
2922__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2923
3cb03002 2924static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2925{
2926 return badblocks_show(&rdev->badblocks, page, 1);
2927}
3cb03002 2928static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 2929{
2930 return badblocks_store(&rdev->badblocks, page, len, 1);
2931}
2932static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2933__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2934
86e6ffdd
N		 2935static struct attribute *rdev_default_attrs[] = {
2936 &rdev_state.attr,
4dbcdc75 2937 &rdev_errors.attr,
014236d2 2938 &rdev_slot.attr,
93c8cad0 2939 &rdev_offset.attr,
c6563a8c 2940 &rdev_new_offset.attr,
83303b61 2941 &rdev_size.attr,
06e3c817 2942 &rdev_recovery_start.attr,
16c791a5
N		 2943 &rdev_bad_blocks.attr,
2944 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 2945 NULL,
2946};
2947static ssize_t
2948rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2949{
2950 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 2951 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
86e6ffdd
N		 2952
2953 if (!entry->show)
2954 return -EIO;
758bfc8a
N		 2955 if (!rdev->mddev)
2956 return -EBUSY;
2957 return entry->show(rdev, page);
86e6ffdd
N		 2958}
2959
2960static ssize_t
2961rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2962 const char *page, size_t length)
2963{
2964 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 2965 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 2966 ssize_t rv;
fd01b88c 2967 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 2968
2969 if (!entry->store)
2970 return -EIO;
67463acb
N		 2971 if (!capable(CAP_SYS_ADMIN))
2972 return -EACCES;
27c529bb 2973 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 2974 if (!rv) {
27c529bb
N		 2975 if (rdev->mddev == NULL)
2976 rv = -EBUSY;
2977 else
2978 rv = entry->store(rdev, page, length);
6a51830e 2979 mddev_unlock(mddev);
ca388059
N		 2980 }
2981 return rv;
86e6ffdd
N		 2982}
2983
2984static void rdev_free(struct kobject *ko)
2985{
3cb03002 2986 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 2987 kfree(rdev);
2988}
52cf25d0 2989static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 2990 .show = rdev_attr_show,
2991 .store = rdev_attr_store,
2992};
2993static struct kobj_type rdev_ktype = {
2994 .release = rdev_free,
2995 .sysfs_ops = &rdev_sysfs_ops,
2996 .default_attrs = rdev_default_attrs,
2997};
2998
3cb03002 2999int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3000{
3001 rdev->desc_nr = -1;
3002 rdev->saved_raid_disk = -1;
3003 rdev->raid_disk = -1;
3004 rdev->flags = 0;
3005 rdev->data_offset = 0;
c6563a8c 3006 rdev->new_data_offset = 0;
e8bb9a83
N		 3007 rdev->sb_events = 0;
3008 rdev->last_read_error.tv_sec = 0;
3009 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3010 rdev->sb_loaded = 0;
3011 rdev->bb_page = NULL;
e8bb9a83
N		 3012 atomic_set(&rdev->nr_pending, 0);
3013 atomic_set(&rdev->read_errors, 0);
3014 atomic_set(&rdev->corrected_errors, 0);
3015
3016 INIT_LIST_HEAD(&rdev->same_set);
3017 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3018
3019 /* Add space to store bad block list.
3020 * This reserves the space even on arrays where it cannot
3021 * be used - I wonder if that matters
3022 */
3023 rdev->badblocks.count = 0;
486adf72 3024 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
2230dfe4
N		 3025 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3026 seqlock_init(&rdev->badblocks.lock);
3027 if (rdev->badblocks.page == NULL)
3028 return -ENOMEM;
3029
3030 return 0;
e8bb9a83
N		 3031}
3032EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3033/*
3034 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3035 *
3036 * mark the device faulty if:
3037 *
3038 * - the device is nonexistent (zero size)
3039 * - the device has no valid superblock
3040 *
3041 * a faulty rdev _never_ has rdev->sb set.
3042 */
3cb03002 3043static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3044{
3045 char b[BDEVNAME_SIZE];
3046 int err;
3cb03002 3047 struct md_rdev *rdev;
1da177e4
LT		 3048 sector_t size;
3049
9ffae0cf 3050 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3051 if (!rdev) {
3052 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3053 return ERR_PTR(-ENOMEM);
3054 }
1da177e4 3055
2230dfe4
N		 3056 err = md_rdev_init(rdev);
3057 if (err)
3058 goto abort_free;
3059 err = alloc_disk_sb(rdev);
3060 if (err)
1da177e4
LT		 3061 goto abort_free;
3062
c5d79adb 3063 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3064 if (err)
3065 goto abort_free;
3066
f9cb074b 3067 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3068
77304d2a 3069 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4 3070 if (!size) {
f72ffdd6 3071 printk(KERN_WARNING
1da177e4
LT		 3072 "md: %s has zero or unknown size, marking faulty!\n",
3073 bdevname(rdev->bdev,b));
3074 err = -EINVAL;
3075 goto abort_free;
3076 }
3077
3078 if (super_format >= 0) {
3079 err = super_types[super_format].
3080 load_super(rdev, NULL, super_minor);
3081 if (err == -EINVAL) {
df968c4e
N		 3082 printk(KERN_WARNING
3083 "md: %s does not have a valid v%d.%d "
3084 "superblock, not importing!\n",
3085 bdevname(rdev->bdev,b),
3086 super_format, super_minor);
1da177e4
LT		 3087 goto abort_free;
3088 }
3089 if (err < 0) {
f72ffdd6 3090 printk(KERN_WARNING
1da177e4
LT		 3091 "md: could not read %s's sb, not importing!\n",
3092 bdevname(rdev->bdev,b));
3093 goto abort_free;
3094 }
3095 }
6bfe0b49 3096
1da177e4
LT		 3097 return rdev;
3098
3099abort_free:
2699b672
N		 3100 if (rdev->bdev)
3101 unlock_rdev(rdev);
545c8795 3102 md_rdev_clear(rdev);
1da177e4
LT		 3103 kfree(rdev);
3104 return ERR_PTR(err);
3105}
3106
3107/*
3108 * Check a full RAID array for plausibility
3109 */
3110
f72ffdd6 3111static void analyze_sbs(struct mddev *mddev)
1da177e4
LT		 3112{
3113 int i;
3cb03002 3114 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3115 char b[BDEVNAME_SIZE];
3116
3117 freshest = NULL;
dafb20fa 3118 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3119 switch (super_types[mddev->major_version].
3120 load_super(rdev, freshest, mddev->minor_version)) {
3121 case 1:
3122 freshest = rdev;
3123 break;
3124 case 0:
3125 break;
3126 default:
3127 printk( KERN_ERR \
3128 "md: fatal superblock inconsistency in %s"
f72ffdd6 3129 " -- removing from array\n",
1da177e4
LT		 3130 bdevname(rdev->bdev,b));
3131 kick_rdev_from_array(rdev);
3132 }
3133
1da177e4
LT		 3134 super_types[mddev->major_version].
3135 validate_super(mddev, freshest);
3136
3137 i = 0;
dafb20fa 3138 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3139 if (mddev->max_disks &&
3140 (rdev->desc_nr >= mddev->max_disks ||
3141 i > mddev->max_disks)) {
de01dfad
N		 3142 printk(KERN_WARNING
3143 "md: %s: %s: only %d devices permitted\n",
3144 mdname(mddev), bdevname(rdev->bdev, b),
3145 mddev->max_disks);
3146 kick_rdev_from_array(rdev);
3147 continue;
3148 }
1da177e4
LT		 3149 if (rdev != freshest)
3150 if (super_types[mddev->major_version].
3151 validate_super(mddev, rdev)) {
3152 printk(KERN_WARNING "md: kicking non-fresh %s"
3153 " from array!\n",
3154 bdevname(rdev->bdev,b));
3155 kick_rdev_from_array(rdev);
3156 continue;
3157 }
3158 if (mddev->level == LEVEL_MULTIPATH) {
3159 rdev->desc_nr = i++;
3160 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3161 set_bit(In_sync, &rdev->flags);
5e5e3e78 3162 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3163 rdev->raid_disk = -1;
3164 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3165 }
3166 }
1da177e4
LT		 3167}
3168
72e02075
N		 3169/* Read a fixed-point number.
3170 * Numbers in sysfs attributes should be in "standard" units where
3171 * possible, so time should be in seconds.
f72ffdd6 3172 * However we internally use a a much smaller unit such as
72e02075
N		 3173 * milliseconds or jiffies.
3174 * This function takes a decimal number with a possible fractional
3175 * component, and produces an integer which is the result of
3176 * multiplying that number by 10^'scale'.
3177 * all without any floating-point arithmetic.
3178 */
3179int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3180{
3181 unsigned long result = 0;
3182 long decimals = -1;
3183 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3184 if (*cp == '.')
3185 decimals = 0;
3186 else if (decimals < scale) {
3187 unsigned int value;
3188 value = *cp - '0';
3189 result = result * 10 + value;
3190 if (decimals >= 0)
3191 decimals++;
3192 }
3193 cp++;
3194 }
3195 if (*cp == '\n')
3196 cp++;
3197 if (*cp)
3198 return -EINVAL;
3199 if (decimals < 0)
3200 decimals = 0;
3201 while (decimals < scale) {
3202 result *= 10;
3203 decimals ++;
3204 }
3205 *res = result;
3206 return 0;
3207}
3208
19052c0e
N		 3209static void md_safemode_timeout(unsigned long data);
3210
16f17b39 3211static ssize_t
fd01b88c 3212safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3213{
3214 int msec = (mddev->safemode_delay*1000)/HZ;
3215 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3216}
3217static ssize_t
fd01b88c 3218safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3219{
16f17b39 3220 unsigned long msec;
97ce0a7f 3221
72e02075 3222 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3223 return -EINVAL;
16f17b39
N		 3224 if (msec == 0)
3225 mddev->safemode_delay = 0;
3226 else {
19052c0e 3227 unsigned long old_delay = mddev->safemode_delay;
1b30e66f
N		 3228 unsigned long new_delay = (msec*HZ)/1000;
3229
3230 if (new_delay == 0)
3231 new_delay = 1;
3232 mddev->safemode_delay = new_delay;
3233 if (new_delay < old_delay || old_delay == 0)
3234 mod_timer(&mddev->safemode_timer, jiffies+1);
16f17b39
N		 3235 }
3236 return len;
3237}
3238static struct md_sysfs_entry md_safe_delay =
80ca3a44 3239__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3240
eae1701f 3241static ssize_t
fd01b88c 3242level_show(struct mddev *mddev, char *page)
eae1701f 3243{
36d091f4
N		 3244 struct md_personality *p;
3245 int ret;
3246 spin_lock(&mddev->lock);
3247 p = mddev->pers;
d9d166c2 3248 if (p)
36d091f4 3249 ret = sprintf(page, "%s\n", p->name);
d9d166c2 3250 else if (mddev->clevel[0])
36d091f4 3251 ret = sprintf(page, "%s\n", mddev->clevel);
d9d166c2 3252 else if (mddev->level != LEVEL_NONE)
36d091f4 3253 ret = sprintf(page, "%d\n", mddev->level);
d9d166c2 3254 else
36d091f4
N		 3255 ret = 0;
3256 spin_unlock(&mddev->lock);
3257 return ret;
eae1701f
N		 3258}
3259
d9d166c2 3260static ssize_t
fd01b88c 3261level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3262{
f2859af6 3263 char clevel[16];
6791875e
N		 3264 ssize_t rv;
3265 size_t slen = len;
db721d32 3266 struct md_personality *pers, *oldpers;
f2859af6 3267 long level;
db721d32 3268 void *priv, *oldpriv;
3cb03002 3269 struct md_rdev *rdev;
245f46c2 3270
6791875e
N		 3271 if (slen == 0 || slen >= sizeof(clevel))
3272 return -EINVAL;
3273
3274 rv = mddev_lock(mddev);
3275 if (rv)
3276 return rv;
3277
245f46c2 3278 if (mddev->pers == NULL) {
6791875e
N		 3279 strncpy(mddev->clevel, buf, slen);
3280 if (mddev->clevel[slen-1] == '\n')
3281 slen--;
3282 mddev->clevel[slen] = 0;
245f46c2 3283 mddev->level = LEVEL_NONE;
6791875e
N		 3284 rv = len;
3285 goto out_unlock;
245f46c2 3286 }
6791875e 3287 rv = -EROFS;
bd8839e0 3288 if (mddev->ro)
6791875e 3289 goto out_unlock;
245f46c2
N		 3290
3291 /* request to change the personality. Need to ensure:
3292 * - array is not engaged in resync/recovery/reshape
3293 * - old personality can be suspended
3294 * - new personality will access other array.
3295 */
3296
6791875e 3297 rv = -EBUSY;
bb4f1e9d 3298 if (mddev->sync_thread ||
f851b60d 3299 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
bb4f1e9d
N		 3300 mddev->reshape_position != MaxSector ||
3301 mddev->sysfs_active)
6791875e 3302 goto out_unlock;
245f46c2 3303
6791875e 3304 rv = -EINVAL;
245f46c2
N		 3305 if (!mddev->pers->quiesce) {
3306 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3307 mdname(mddev), mddev->pers->name);
6791875e 3308 goto out_unlock;
245f46c2
N		 3309 }
3310
3311 /* Now find the new personality */
6791875e
N		 3312 strncpy(clevel, buf, slen);
3313 if (clevel[slen-1] == '\n')
3314 slen--;
3315 clevel[slen] = 0;
b29bebd6 3316 if (kstrtol(clevel, 10, &level))
f2859af6 3317 level = LEVEL_NONE;
245f46c2 3318
f2859af6
DW		 3319 if (request_module("md-%s", clevel) != 0)
3320 request_module("md-level-%s", clevel);
245f46c2 3321 spin_lock(&pers_lock);
f2859af6 3322 pers = find_pers(level, clevel);
245f46c2
N		 3323 if (!pers || !try_module_get(pers->owner)) {
3324 spin_unlock(&pers_lock);
f2859af6 3325 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
6791875e
N		 3326 rv = -EINVAL;
3327 goto out_unlock;
245f46c2
N		 3328 }
3329 spin_unlock(&pers_lock);
3330
3331 if (pers == mddev->pers) {
3332 /* Nothing to do! */
3333 module_put(pers->owner);
6791875e
N		 3334 rv = len;
3335 goto out_unlock;
245f46c2
N		 3336 }
3337 if (!pers->takeover) {
3338 module_put(pers->owner);
3339 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3340 mdname(mddev), clevel);
6791875e
N		 3341 rv = -EINVAL;
3342 goto out_unlock;
245f46c2
N		 3343 }
3344
dafb20fa 3345 rdev_for_each(rdev, mddev)
e93f68a1
N		 3346 rdev->new_raid_disk = rdev->raid_disk;
3347
245f46c2
N		 3348 /* ->takeover must set new_* and/or delta_disks
3349 * if it succeeds, and may set them when it fails.
3350 */
3351 priv = pers->takeover(mddev);
3352 if (IS_ERR(priv)) {
3353 mddev->new_level = mddev->level;
3354 mddev->new_layout = mddev->layout;
664e7c41 3355 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3356 mddev->raid_disks -= mddev->delta_disks;
3357 mddev->delta_disks = 0;
2c810cdd 3358 mddev->reshape_backwards = 0;
245f46c2
N		 3359 module_put(pers->owner);
3360 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3361 mdname(mddev), clevel);
6791875e
N		 3362 rv = PTR_ERR(priv);
3363 goto out_unlock;
245f46c2
N		 3364 }
3365
3366 /* Looks like we have a winner */
3367 mddev_suspend(mddev);
5aa61f42 3368 mddev_detach(mddev);
36d091f4
N		 3369
3370 spin_lock(&mddev->lock);
db721d32
N		 3371 oldpers = mddev->pers;
3372 oldpriv = mddev->private;
3373 mddev->pers = pers;
3374 mddev->private = priv;
3375 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3376 mddev->level = mddev->new_level;
3377 mddev->layout = mddev->new_layout;
3378 mddev->chunk_sectors = mddev->new_chunk_sectors;
3379 mddev->delta_disks = 0;
3380 mddev->reshape_backwards = 0;
3381 mddev->degraded = 0;
36d091f4 3382 spin_unlock(&mddev->lock);
db721d32
N		 3383
3384 if (oldpers->sync_request == NULL &&
3385 mddev->external) {
3386 /* We are converting from a no-redundancy array
3387 * to a redundancy array and metadata is managed
3388 * externally so we need to be sure that writes
3389 * won't block due to a need to transition
3390 * clean->dirty
3391 * until external management is started.
3392 */
3393 mddev->in_sync = 0;
3394 mddev->safemode_delay = 0;
3395 mddev->safemode = 0;
3396 }
f72ffdd6 3397
db721d32
N		 3398 oldpers->free(mddev, oldpriv);
3399
3400 if (oldpers->sync_request == NULL &&
a64c876f
N		 3401 pers->sync_request != NULL) {
3402 /* need to add the md_redundancy_group */
3403 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3404 printk(KERN_WARNING
3405 "md: cannot register extra attributes for %s\n",
3406 mdname(mddev));
388975cc 3407 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
f72ffdd6 3408 }
db721d32 3409 if (oldpers->sync_request != NULL &&
a64c876f
N		 3410 pers->sync_request == NULL) {
3411 /* need to remove the md_redundancy_group */
3412 if (mddev->to_remove == NULL)
3413 mddev->to_remove = &md_redundancy_group;
3414 }
3415
dafb20fa 3416 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3417 if (rdev->raid_disk < 0)
3418 continue;
bf2cb0da 3419 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3420 rdev->new_raid_disk = -1;
3421 if (rdev->new_raid_disk == rdev->raid_disk)
3422 continue;
36fad858 3423 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3424 }
dafb20fa 3425 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3426 if (rdev->raid_disk < 0)
3427 continue;
3428 if (rdev->new_raid_disk == rdev->raid_disk)
3429 continue;
3430 rdev->raid_disk = rdev->new_raid_disk;
3431 if (rdev->raid_disk < 0)
3a981b03 3432 clear_bit(In_sync, &rdev->flags);
e93f68a1 3433 else {
36fad858
NK		 3434 if (sysfs_link_rdev(mddev, rdev))
3435 printk(KERN_WARNING "md: cannot register rd%d"
3436 " for %s after level change\n",
3437 rdev->raid_disk, mdname(mddev));
3a981b03 3438 }
e93f68a1
N		 3439 }
3440
db721d32 3441 if (pers->sync_request == NULL) {
9af204cf
TM		 3442 /* this is now an array without redundancy, so
3443 * it must always be in_sync
3444 */
3445 mddev->in_sync = 1;
3446 del_timer_sync(&mddev->safemode_timer);
3447 }
02e5f5c0 3448 blk_set_stacking_limits(&mddev->queue->limits);
245f46c2 3449 pers->run(mddev);
245f46c2 3450 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3451 mddev_resume(mddev);
830778a1
N		 3452 if (!mddev->thread)
3453 md_update_sb(mddev, 1);
5cac7861 3454 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3455 md_new_event(mddev);
6791875e
N		 3456 rv = len;
3457out_unlock:
3458 mddev_unlock(mddev);
d9d166c2
N		 3459 return rv;
3460}
3461
3462static struct md_sysfs_entry md_level =
80ca3a44 3463__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3464
d4dbd025 3465static ssize_t
fd01b88c 3466layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3467{
3468 /* just a number, not meaningful for all levels */
08a02ecd
N		 3469 if (mddev->reshape_position != MaxSector &&
3470 mddev->layout != mddev->new_layout)
3471 return sprintf(page, "%d (%d)\n",
3472 mddev->new_layout, mddev->layout);
d4dbd025
N		 3473 return sprintf(page, "%d\n", mddev->layout);
3474}
3475
3476static ssize_t
fd01b88c 3477layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N		 3478{
3479 char *e;
3480 unsigned long n = simple_strtoul(buf, &e, 10);
6791875e 3481 int err;
d4dbd025
N		 3482
3483 if (!*buf || (*e && *e != '\n'))
3484 return -EINVAL;
6791875e
N		 3485 err = mddev_lock(mddev);
3486 if (err)
3487 return err;
d4dbd025 3488
b3546035 3489 if (mddev->pers) {
50ac168a 3490 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3491 err = -EBUSY;
3492 else if (mddev->ro)
3493 err = -EROFS;
3494 else {
3495 mddev->new_layout = n;
3496 err = mddev->pers->check_reshape(mddev);
3497 if (err)
3498 mddev->new_layout = mddev->layout;
597a711b 3499 }
b3546035 3500 } else {
08a02ecd 3501 mddev->new_layout = n;
b3546035
N		 3502 if (mddev->reshape_position == MaxSector)
3503 mddev->layout = n;
3504 }
6791875e
N		 3505 mddev_unlock(mddev);
3506 return err ?: len;
d4dbd025
N		 3507}
3508static struct md_sysfs_entry md_layout =
80ca3a44 3509__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025 3510
eae1701f 3511static ssize_t
fd01b88c 3512raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3513{
bb636547
N		 3514 if (mddev->raid_disks == 0)
3515 return 0;
08a02ecd
N		 3516 if (mddev->reshape_position != MaxSector &&
3517 mddev->delta_disks != 0)
3518 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3519 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3520 return sprintf(page, "%d\n", mddev->raid_disks);
3521}
3522
fd01b88c 3523static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3524
3525static ssize_t
fd01b88c 3526raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3527{
da943b99 3528 char *e;
6791875e 3529 int err;
da943b99
N		 3530 unsigned long n = simple_strtoul(buf, &e, 10);
3531
3532 if (!*buf || (*e && *e != '\n'))
3533 return -EINVAL;
3534
6791875e
N		 3535 err = mddev_lock(mddev);
3536 if (err)
3537 return err;
da943b99 3538 if (mddev->pers)
6791875e 3539 err = update_raid_disks(mddev, n);
08a02ecd 3540 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3541 struct md_rdev *rdev;
08a02ecd 3542 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c 3543
6791875e 3544 err = -EINVAL;
c6563a8c
N		 3545 rdev_for_each(rdev, mddev) {
3546 if (olddisks < n &&
3547 rdev->data_offset < rdev->new_data_offset)
6791875e 3548 goto out_unlock;
c6563a8c
N		 3549 if (olddisks > n &&
3550 rdev->data_offset > rdev->new_data_offset)
6791875e 3551 goto out_unlock;
c6563a8c 3552 }
6791875e 3553 err = 0;
08a02ecd
N		 3554 mddev->delta_disks = n - olddisks;
3555 mddev->raid_disks = n;
2c810cdd 3556 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3557 } else
da943b99 3558 mddev->raid_disks = n;
6791875e
N		 3559out_unlock:
3560 mddev_unlock(mddev);
3561 return err ? err : len;
da943b99
N		 3562}
3563static struct md_sysfs_entry md_raid_disks =
80ca3a44 3564__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3565
3b34380a 3566static ssize_t
fd01b88c 3567chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3568{
08a02ecd 3569 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3570 mddev->chunk_sectors != mddev->new_chunk_sectors)
3571 return sprintf(page, "%d (%d)\n",
3572 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3573 mddev->chunk_sectors << 9);
3574 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3575}
3576
3577static ssize_t
fd01b88c 3578chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3579{
6791875e 3580 int err;
3b34380a
N		 3581 char *e;
3582 unsigned long n = simple_strtoul(buf, &e, 10);
3583
3b34380a
N		 3584 if (!*buf || (*e && *e != '\n'))
3585 return -EINVAL;
3586
6791875e
N		 3587 err = mddev_lock(mddev);
3588 if (err)
3589 return err;
b3546035 3590 if (mddev->pers) {
50ac168a 3591 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3592 err = -EBUSY;
3593 else if (mddev->ro)
3594 err = -EROFS;
3595 else {
3596 mddev->new_chunk_sectors = n >> 9;
3597 err = mddev->pers->check_reshape(mddev);
3598 if (err)
3599 mddev->new_chunk_sectors = mddev->chunk_sectors;
597a711b 3600 }
b3546035 3601 } else {
664e7c41 3602 mddev->new_chunk_sectors = n >> 9;
b3546035 3603 if (mddev->reshape_position == MaxSector)
9d8f0363 3604 mddev->chunk_sectors = n >> 9;
b3546035 3605 }
6791875e
N		 3606 mddev_unlock(mddev);
3607 return err ?: len;
3b34380a
N		 3608}
3609static struct md_sysfs_entry md_chunk_size =
80ca3a44 3610__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3611
a94213b1 3612static ssize_t
fd01b88c 3613resync_start_show(struct mddev *mddev, char *page)
a94213b1 3614{
d1a7c503
N		 3615 if (mddev->recovery_cp == MaxSector)
3616 return sprintf(page, "none\n");
a94213b1
N		 3617 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3618}
3619
3620static ssize_t
fd01b88c 3621resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3622{
6791875e 3623 int err;
a94213b1
N		 3624 char *e;
3625 unsigned long long n = simple_strtoull(buf, &e, 10);
3626
6791875e
N		 3627 err = mddev_lock(mddev);
3628 if (err)
3629 return err;
b098636c 3630 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6791875e
N		 3631 err = -EBUSY;
3632 else if (cmd_match(buf, "none"))
06e3c817
DW		 3633 n = MaxSector;
3634 else if (!*buf || (*e && *e != '\n'))
6791875e 3635 err = -EINVAL;
a94213b1 3636
6791875e
N		 3637 if (!err) {
3638 mddev->recovery_cp = n;
3639 if (mddev->pers)
3640 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3641 }
3642 mddev_unlock(mddev);
3643 return err ?: len;
a94213b1
N		 3644}
3645static struct md_sysfs_entry md_resync_start =
80ca3a44 3646__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3647
9e653b63
N		 3648/*
3649 * The array state can be:
3650 *
3651 * clear
3652 * No devices, no size, no level
3653 * Equivalent to STOP_ARRAY ioctl
3654 * inactive
3655 * May have some settings, but array is not active
3656 * all IO results in error
3657 * When written, doesn't tear down array, but just stops it
3658 * suspended (not supported yet)
3659 * All IO requests will block. The array can be reconfigured.
910d8cb3 3660 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3661 * readonly
3662 * no resync can happen. no superblocks get written.
3663 * write requests fail
3664 * read-auto
3665 * like readonly, but behaves like 'clean' on a write request.
3666 *
3667 * clean - no pending writes, but otherwise active.
3668 * When written to inactive array, starts without resync
3669 * If a write request arrives then
3670 * if metadata is known, mark 'dirty' and switch to 'active'.
3671 * if not known, block and switch to write-pending
3672 * If written to an active array that has pending writes, then fails.
3673 * active
3674 * fully active: IO and resync can be happening.
3675 * When written to inactive array, starts with resync
3676 *
3677 * write-pending
3678 * clean, but writes are blocked waiting for 'active' to be written.
3679 *
3680 * active-idle
3681 * like active, but no writes have been seen for a while (100msec).
3682 *
3683 */
3684enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3685 write_pending, active_idle, bad_word};
05381954 3686static char *array_states[] = {
9e653b63
N		 3687 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3688 "write-pending", "active-idle", NULL };
3689
3690static int match_word(const char *word, char **list)
3691{
3692 int n;
3693 for (n=0; list[n]; n++)
3694 if (cmd_match(word, list[n]))
3695 break;
3696 return n;
3697}
3698
3699static ssize_t
fd01b88c 3700array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3701{
3702 enum array_state st = inactive;
3703
3704 if (mddev->pers)
3705 switch(mddev->ro) {
3706 case 1:
3707 st = readonly;
3708 break;
3709 case 2:
3710 st = read_auto;
3711 break;
3712 case 0:
3713 if (mddev->in_sync)
3714 st = clean;
070dc6dd 3715 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3716 st = write_pending;
9e653b63
N		 3717 else if (mddev->safemode)
3718 st = active_idle;
3719 else
3720 st = active;
3721 }
3722 else {
3723 if (list_empty(&mddev->disks) &&
3724 mddev->raid_disks == 0 &&
58c0fed4 3725 mddev->dev_sectors == 0)
9e653b63
N		 3726 st = clear;
3727 else
3728 st = inactive;
3729 }
3730 return sprintf(page, "%s\n", array_states[st]);
3731}
3732
f72ffdd6
N		 3733static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3734static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3735static int do_md_run(struct mddev *mddev);
fd01b88c 3736static int restart_array(struct mddev *mddev);
9e653b63
N		 3737
3738static ssize_t
fd01b88c 3739array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63 3740{
6791875e 3741 int err;
9e653b63 3742 enum array_state st = match_word(buf, array_states);
6791875e
N		 3743
3744 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3745 /* don't take reconfig_mutex when toggling between
3746 * clean and active
3747 */
3748 spin_lock(&mddev->lock);
3749 if (st == active) {
3750 restart_array(mddev);
3751 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3752 wake_up(&mddev->sb_wait);
3753 err = 0;
3754 } else /* st == clean */ {
3755 restart_array(mddev);
3756 if (atomic_read(&mddev->writes_pending) == 0) {
3757 if (mddev->in_sync == 0) {
3758 mddev->in_sync = 1;
3759 if (mddev->safemode == 1)
3760 mddev->safemode = 0;
3761 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3762 }
3763 err = 0;
3764 } else
3765 err = -EBUSY;
3766 }
3767 spin_unlock(&mddev->lock);
3768 return err;
3769 }
3770 err = mddev_lock(mddev);
3771 if (err)
3772 return err;
3773 err = -EINVAL;
9e653b63
N		 3774 switch(st) {
3775 case bad_word:
3776 break;
3777 case clear:
3778 /* stopping an active array */
a05b7ea0 3779 err = do_md_stop(mddev, 0, NULL);
9e653b63
N		 3780 break;
3781 case inactive:
3782 /* stopping an active array */
90cf195d 3783 if (mddev->pers)
a05b7ea0 3784 err = do_md_stop(mddev, 2, NULL);
90cf195d 3785 else
e691063a 3786 err = 0; /* already inactive */
9e653b63
N		 3787 break;
3788 case suspended:
3789 break; /* not supported yet */
3790 case readonly:
3791 if (mddev->pers)
a05b7ea0 3792 err = md_set_readonly(mddev, NULL);
9e653b63
N		 3793 else {
3794 mddev->ro = 1;
648b629e 3795 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3796 err = do_md_run(mddev);
3797 }
3798 break;
3799 case read_auto:
9e653b63 3800 if (mddev->pers) {
80268ee9 3801 if (mddev->ro == 0)
a05b7ea0 3802 err = md_set_readonly(mddev, NULL);
80268ee9 3803 else if (mddev->ro == 1)
648b629e
N		 3804 err = restart_array(mddev);
3805 if (err == 0) {
3806 mddev->ro = 2;
3807 set_disk_ro(mddev->gendisk, 0);
3808 }
9e653b63
N		 3809 } else {
3810 mddev->ro = 2;
3811 err = do_md_run(mddev);
3812 }
3813 break;
3814 case clean:
3815 if (mddev->pers) {
3816 restart_array(mddev);
85572d7c 3817 spin_lock(&mddev->lock);
9e653b63 3818 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3819 if (mddev->in_sync == 0) {
3820 mddev->in_sync = 1;
31a59e34
N		 3821 if (mddev->safemode == 1)
3822 mddev->safemode = 0;
070dc6dd 3823 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3824 }
3825 err = 0;
3826 } else
3827 err = -EBUSY;
85572d7c 3828 spin_unlock(&mddev->lock);
5bf29597
N		 3829 } else
3830 err = -EINVAL;
9e653b63
N		 3831 break;
3832 case active:
3833 if (mddev->pers) {
3834 restart_array(mddev);
070dc6dd 3835 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3836 wake_up(&mddev->sb_wait);
3837 err = 0;
3838 } else {
3839 mddev->ro = 0;
648b629e 3840 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3841 err = do_md_run(mddev);
3842 }
3843 break;
3844 case write_pending:
3845 case active_idle:
3846 /* these cannot be set */
3847 break;
3848 }
6791875e
N		 3849
3850 if (!err) {
1d23f178
N		 3851 if (mddev->hold_active == UNTIL_IOCTL)
3852 mddev->hold_active = 0;
00bcb4ac 3853 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 3854 }
6791875e
N		 3855 mddev_unlock(mddev);
3856 return err ?: len;
9e653b63 3857}
80ca3a44
N		 3858static struct md_sysfs_entry md_array_state =
3859__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3860
1e50915f 3861static ssize_t
fd01b88c 3862max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3863 return sprintf(page, "%d\n",
3864 atomic_read(&mddev->max_corr_read_errors));
3865}
3866
3867static ssize_t
fd01b88c 3868max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB		 3869{
3870 char *e;
3871 unsigned long n = simple_strtoul(buf, &e, 10);
3872
3873 if (*buf && (*e == 0 || *e == '\n')) {
3874 atomic_set(&mddev->max_corr_read_errors, n);
3875 return len;
3876 }
3877 return -EINVAL;
3878}
3879
3880static struct md_sysfs_entry max_corr_read_errors =
3881__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3882 max_corrected_read_errors_store);
3883
6d7ff738 3884static ssize_t
fd01b88c 3885null_show(struct mddev *mddev, char *page)
6d7ff738
N		 3886{
3887 return -EINVAL;
3888}
3889
3890static ssize_t
fd01b88c 3891new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 3892{
3893 /* buf must be %d:%d\n? giving major and minor numbers */
3894 /* The new device is added to the array.
3895 * If the array has a persistent superblock, we read the
3896 * superblock to initialise info and check validity.
3897 * Otherwise, only checking done is that in bind_rdev_to_array,
3898 * which mainly checks size.
3899 */
3900 char *e;
3901 int major = simple_strtoul(buf, &e, 10);
3902 int minor;
3903 dev_t dev;
3cb03002 3904 struct md_rdev *rdev;
6d7ff738
N		 3905 int err;
3906
3907 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3908 return -EINVAL;
3909 minor = simple_strtoul(e+1, &e, 10);
3910 if (*e && *e != '\n')
3911 return -EINVAL;
3912 dev = MKDEV(major, minor);
3913 if (major != MAJOR(dev) ||
3914 minor != MINOR(dev))
3915 return -EOVERFLOW;
3916
6791875e
N		 3917 flush_workqueue(md_misc_wq);
3918
3919 err = mddev_lock(mddev);
3920 if (err)
3921 return err;
6d7ff738
N		 3922 if (mddev->persistent) {
3923 rdev = md_import_device(dev, mddev->major_version,
3924 mddev->minor_version);
3925 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 3926 struct md_rdev *rdev0
3927 = list_entry(mddev->disks.next,
3928 struct md_rdev, same_set);
6d7ff738
N		 3929 err = super_types[mddev->major_version]
3930 .load_super(rdev, rdev0, mddev->minor_version);
3931 if (err < 0)
3932 goto out;
3933 }
c5d79adb
N		 3934 } else if (mddev->external)
3935 rdev = md_import_device(dev, -2, -1);
3936 else
6d7ff738
N		 3937 rdev = md_import_device(dev, -1, -1);
3938
3939 if (IS_ERR(rdev))
3940 return PTR_ERR(rdev);
3941 err = bind_rdev_to_array(rdev, mddev);
3942 out:
3943 if (err)
3944 export_rdev(rdev);
6791875e 3945 mddev_unlock(mddev);
6d7ff738
N		 3946 return err ? err : len;
3947}
3948
3949static struct md_sysfs_entry md_new_device =
80ca3a44 3950__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3951
9b1d1dac 3952static ssize_t
fd01b88c 3953bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 3954{
3955 char *end;
3956 unsigned long chunk, end_chunk;
6791875e 3957 int err;
9b1d1dac 3958
6791875e
N		 3959 err = mddev_lock(mddev);
3960 if (err)
3961 return err;
9b1d1dac
PC		 3962 if (!mddev->bitmap)
3963 goto out;
3964 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3965 while (*buf) {
3966 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3967 if (buf == end) break;
3968 if (*end == '-') { /* range */
3969 buf = end + 1;
3970 end_chunk = simple_strtoul(buf, &end, 0);
3971 if (buf == end) break;
3972 }
3973 if (*end && !isspace(*end)) break;
3974 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 3975 buf = skip_spaces(end);
9b1d1dac
PC		 3976 }
3977 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3978out:
6791875e 3979 mddev_unlock(mddev);
9b1d1dac
PC		 3980 return len;
3981}
3982
3983static struct md_sysfs_entry md_bitmap =
3984__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3985
a35b0d69 3986static ssize_t
fd01b88c 3987size_show(struct mddev *mddev, char *page)
a35b0d69 3988{
58c0fed4
AN		 3989 return sprintf(page, "%llu\n",
3990 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 3991}
3992
fd01b88c 3993static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 3994
3995static ssize_t
fd01b88c 3996size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 3997{
3998 /* If array is inactive, we can reduce the component size, but
3999 * not increase it (except from 0).
4000 * If array is active, we can try an on-line resize
4001 */
b522adcd
DW		 4002 sector_t sectors;
4003 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4004
58c0fed4
AN		 4005 if (err < 0)
4006 return err;
6791875e
N		 4007 err = mddev_lock(mddev);
4008 if (err)
4009 return err;
a35b0d69 4010 if (mddev->pers) {
58c0fed4 4011 err = update_size(mddev, sectors);
850b2b42 4012 md_update_sb(mddev, 1);
a35b0d69 4013 } else {
58c0fed4
AN		 4014 if (mddev->dev_sectors == 0 ||
4015 mddev->dev_sectors > sectors)
4016 mddev->dev_sectors = sectors;
a35b0d69
N		 4017 else
4018 err = -ENOSPC;
4019 }
6791875e 4020 mddev_unlock(mddev);
a35b0d69
N		 4021 return err ? err : len;
4022}
4023
4024static struct md_sysfs_entry md_size =
80ca3a44 4025__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4026
83f0d77a 4027/* Metadata version.
e691063a
N		 4028 * This is one of
4029 * 'none' for arrays with no metadata (good luck...)
4030 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4031 * or N.M for internally known formats
4032 */
4033static ssize_t
fd01b88c 4034metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4035{
4036 if (mddev->persistent)
4037 return sprintf(page, "%d.%d\n",
4038 mddev->major_version, mddev->minor_version);
e691063a
N		 4039 else if (mddev->external)
4040 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4041 else
4042 return sprintf(page, "none\n");
4043}
4044
4045static ssize_t
fd01b88c 4046metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4047{
4048 int major, minor;
4049 char *e;
6791875e 4050 int err;
ea43ddd8
N		 4051 /* Changing the details of 'external' metadata is
4052 * always permitted. Otherwise there must be
4053 * no devices attached to the array.
4054 */
6791875e
N		 4055
4056 err = mddev_lock(mddev);
4057 if (err)
4058 return err;
4059 err = -EBUSY;
ea43ddd8
N		 4060 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4061 ;
4062 else if (!list_empty(&mddev->disks))
6791875e 4063 goto out_unlock;
8bb93aac 4064
6791875e 4065 err = 0;
8bb93aac
N		 4066 if (cmd_match(buf, "none")) {
4067 mddev->persistent = 0;
e691063a
N		 4068 mddev->external = 0;
4069 mddev->major_version = 0;
4070 mddev->minor_version = 90;
6791875e 4071 goto out_unlock;
e691063a
N		 4072 }
4073 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4074 size_t namelen = len-9;
e691063a
N		 4075 if (namelen >= sizeof(mddev->metadata_type))
4076 namelen = sizeof(mddev->metadata_type)-1;
4077 strncpy(mddev->metadata_type, buf+9, namelen);
4078 mddev->metadata_type[namelen] = 0;
4079 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4080 mddev->metadata_type[--namelen] = 0;
4081 mddev->persistent = 0;
4082 mddev->external = 1;
8bb93aac
N		 4083 mddev->major_version = 0;
4084 mddev->minor_version = 90;
6791875e 4085 goto out_unlock;
8bb93aac
N		 4086 }
4087 major = simple_strtoul(buf, &e, 10);
6791875e 4088 err = -EINVAL;
8bb93aac 4089 if (e==buf || *e != '.')
6791875e 4090 goto out_unlock;
8bb93aac
N		 4091 buf = e+1;
4092 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4093 if (e==buf || (*e && *e != '\n') )
6791875e
N		 4094 goto out_unlock;
4095 err = -ENOENT;
50511da3 4096 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
6791875e 4097 goto out_unlock;
8bb93aac
N		 4098 mddev->major_version = major;
4099 mddev->minor_version = minor;
4100 mddev->persistent = 1;
e691063a 4101 mddev->external = 0;
6791875e
N		 4102 err = 0;
4103out_unlock:
4104 mddev_unlock(mddev);
4105 return err ?: len;
8bb93aac
N		 4106}
4107
4108static struct md_sysfs_entry md_metadata =
80ca3a44 4109__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4110
24dd469d 4111static ssize_t
fd01b88c 4112action_show(struct mddev *mddev, char *page)
24dd469d 4113{
7eec314d 4114 char *type = "idle";
b7b17c9b
N		 4115 unsigned long recovery = mddev->recovery;
4116 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
b6a9ce68 4117 type = "frozen";
b7b17c9b
N		 4118 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4119 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4120 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
ccfcc3c1 4121 type = "reshape";
b7b17c9b
N		 4122 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4123 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
24dd469d 4124 type = "resync";
b7b17c9b 4125 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
24dd469d
N		 4126 type = "check";
4127 else
4128 type = "repair";
b7b17c9b 4129 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
24dd469d
N		 4130 type = "recover";
4131 }
4132 return sprintf(page, "%s\n", type);
4133}
4134
4135static ssize_t
fd01b88c 4136action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4137{
7eec314d
N		 4138 if (!mddev->pers || !mddev->pers->sync_request)
4139 return -EINVAL;
4140
b6a9ce68
N		 4141 if (cmd_match(page, "frozen"))
4142 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4143 else
4144 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4145
4146 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
f851b60d 4147 flush_workqueue(md_misc_wq);
7eec314d
N		 4148 if (mddev->sync_thread) {
4149 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6791875e
N		 4150 if (mddev_lock(mddev) == 0) {
4151 md_reap_sync_thread(mddev);
4152 mddev_unlock(mddev);
4153 }
7eec314d 4154 }
03c902e1
N		 4155 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4156 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4157 return -EBUSY;
72a23c21
NB		 4158 else if (cmd_match(page, "resync"))
4159 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4160 else if (cmd_match(page, "recover")) {
4161 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4162 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4163 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4164 int err;
4165 if (mddev->pers->start_reshape == NULL)
4166 return -EINVAL;
6791875e
N		 4167 err = mddev_lock(mddev);
4168 if (!err) {
4169 err = mddev->pers->start_reshape(mddev);
4170 mddev_unlock(mddev);
4171 }
16484bf5
N		 4172 if (err)
4173 return err;
a99ac971 4174 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4175 } else {
bce74dac 4176 if (cmd_match(page, "check"))
7eec314d 4177 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4178 else if (!cmd_match(page, "repair"))
7eec314d
N		 4179 return -EINVAL;
4180 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4181 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4182 }
48c26ddc
N		 4183 if (mddev->ro == 2) {
4184 /* A write to sync_action is enough to justify
4185 * canceling read-auto mode
4186 */
4187 mddev->ro = 0;
4188 md_wakeup_thread(mddev->sync_thread);
4189 }
03c902e1 4190 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4191 md_wakeup_thread(mddev->thread);
00bcb4ac 4192 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4193 return len;
4194}
4195
c4a39551
JB		 4196static struct md_sysfs_entry md_scan_mode =
4197__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4198
4199static ssize_t
4200last_sync_action_show(struct mddev *mddev, char *page)
4201{
4202 return sprintf(page, "%s\n", mddev->last_sync_action);
4203}
4204
4205static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4206
9d88883e 4207static ssize_t
fd01b88c 4208mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4209{
4210 return sprintf(page, "%llu\n",
7f7583d4
JM		 4211 (unsigned long long)
4212 atomic64_read(&mddev->resync_mismatches));
9d88883e
N		 4213}
4214
80ca3a44 4215static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4216
88202a0c 4217static ssize_t
fd01b88c 4218sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4219{
4220 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4221 mddev->sync_speed_min ? "local": "system");
4222}
4223
4224static ssize_t
fd01b88c 4225sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4226{
4227 int min;
4228 char *e;
4229 if (strncmp(buf, "system", 6)==0) {
4230 mddev->sync_speed_min = 0;
4231 return len;
4232 }
4233 min = simple_strtoul(buf, &e, 10);
4234 if (buf == e || (*e && *e != '\n') || min <= 0)
4235 return -EINVAL;
4236 mddev->sync_speed_min = min;
4237 return len;
4238}
4239
4240static struct md_sysfs_entry md_sync_min =
4241__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4242
4243static ssize_t
fd01b88c 4244sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4245{
4246 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4247 mddev->sync_speed_max ? "local": "system");
4248}
4249
4250static ssize_t
fd01b88c 4251sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4252{
4253 int max;
4254 char *e;
4255 if (strncmp(buf, "system", 6)==0) {
4256 mddev->sync_speed_max = 0;
4257 return len;
4258 }
4259 max = simple_strtoul(buf, &e, 10);
4260 if (buf == e || (*e && *e != '\n') || max <= 0)
4261 return -EINVAL;
4262 mddev->sync_speed_max = max;
4263 return len;
4264}
4265
4266static struct md_sysfs_entry md_sync_max =
4267__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4268
d7f3d291 4269static ssize_t
fd01b88c 4270degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4271{
4272 return sprintf(page, "%d\n", mddev->degraded);
4273}
4274static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4275
90b08710 4276static ssize_t
fd01b88c 4277sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4278{
4279 return sprintf(page, "%d\n", mddev->parallel_resync);
4280}
4281
4282static ssize_t
fd01b88c 4283sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4284{
4285 long n;
4286
b29bebd6 4287 if (kstrtol(buf, 10, &n))
90b08710
BS		 4288 return -EINVAL;
4289
4290 if (n != 0 && n != 1)
4291 return -EINVAL;
4292
4293 mddev->parallel_resync = n;
4294
4295 if (mddev->sync_thread)
4296 wake_up(&resync_wait);
4297
4298 return len;
4299}
4300
4301/* force parallel resync, even with shared block devices */
4302static struct md_sysfs_entry md_sync_force_parallel =
4303__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4304 sync_force_parallel_show, sync_force_parallel_store);
4305
88202a0c 4306static ssize_t
fd01b88c 4307sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4308{
4309 unsigned long resync, dt, db;
d1a7c503
N		 4310 if (mddev->curr_resync == 0)
4311 return sprintf(page, "none\n");
9687a60c
AN		 4312 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4313 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4314 if (!dt) dt++;
9687a60c
AN		 4315 db = resync - mddev->resync_mark_cnt;
4316 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4317}
4318
80ca3a44 4319static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4320
4321static ssize_t
fd01b88c 4322sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4323{
13ae864b 4324 unsigned long long max_sectors, resync;
88202a0c 4325
acb180b0
N		 4326 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4327 return sprintf(page, "none\n");
4328
72f36d59
N		 4329 if (mddev->curr_resync == 1 ||
4330 mddev->curr_resync == 2)
4331 return sprintf(page, "delayed\n");
4332
c804cdec
N		 4333 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4334 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 4335 max_sectors = mddev->resync_max_sectors;
88202a0c 4336 else
58c0fed4 4337 max_sectors = mddev->dev_sectors;
88202a0c 4338
acb180b0 4339 resync = mddev->curr_resync_completed;
13ae864b 4340 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4341}
4342
80ca3a44 4343static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 4344
5e96ee65 4345static ssize_t
fd01b88c 4346min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4347{
4348 return sprintf(page, "%llu\n",
4349 (unsigned long long)mddev->resync_min);
4350}
4351static ssize_t
fd01b88c 4352min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4353{
4354 unsigned long long min;
23da422b
N		 4355 int err;
4356 int chunk;
4357
b29bebd6 4358 if (kstrtoull(buf, 10, &min))
5e96ee65 4359 return -EINVAL;
23da422b
N		 4360
4361 spin_lock(&mddev->lock);
4362 err = -EINVAL;
5e96ee65 4363 if (min > mddev->resync_max)
23da422b
N		 4364 goto out_unlock;
4365
4366 err = -EBUSY;
5e96ee65 4367 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4368 goto out_unlock;
5e96ee65
NB		 4369
4370 /* Must be a multiple of chunk_size */
23da422b
N		 4371 chunk = mddev->chunk_sectors;
4372 if (chunk) {
2ac06c33 4373 sector_t temp = min;
23da422b
N		 4374
4375 err = -EINVAL;
4376 if (sector_div(temp, chunk))
4377 goto out_unlock;
5e96ee65
NB		 4378 }
4379 mddev->resync_min = min;
23da422b 4380 err = 0;
5e96ee65 4381
23da422b
N		 4382out_unlock:
4383 spin_unlock(&mddev->lock);
4384 return err ?: len;
5e96ee65
NB		 4385}
4386
4387static struct md_sysfs_entry md_min_sync =
4388__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4389
c6207277 4390static ssize_t
fd01b88c 4391max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4392{
4393 if (mddev->resync_max == MaxSector)
4394 return sprintf(page, "max\n");
4395 else
4396 return sprintf(page, "%llu\n",
4397 (unsigned long long)mddev->resync_max);
4398}
4399static ssize_t
fd01b88c 4400max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277 4401{
23da422b
N		 4402 int err;
4403 spin_lock(&mddev->lock);
c6207277
N		 4404 if (strncmp(buf, "max", 3) == 0)
4405 mddev->resync_max = MaxSector;
4406 else {
5e96ee65 4407 unsigned long long max;
23da422b
N		 4408 int chunk;
4409
4410 err = -EINVAL;
b29bebd6 4411 if (kstrtoull(buf, 10, &max))
23da422b 4412 goto out_unlock;
5e96ee65 4413 if (max < mddev->resync_min)
23da422b
N		 4414 goto out_unlock;
4415
4416 err = -EBUSY;
c6207277 4417 if (max < mddev->resync_max &&
4d484a4a 4418 mddev->ro == 0 &&
c6207277 4419 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4420 goto out_unlock;
c6207277
N		 4421
4422 /* Must be a multiple of chunk_size */
23da422b
N		 4423 chunk = mddev->chunk_sectors;
4424 if (chunk) {
2ac06c33 4425 sector_t temp = max;
23da422b
N		 4426
4427 err = -EINVAL;
4428 if (sector_div(temp, chunk))
4429 goto out_unlock;
c6207277
N		 4430 }
4431 mddev->resync_max = max;
4432 }
4433 wake_up(&mddev->recovery_wait);
23da422b
N		 4434 err = 0;
4435out_unlock:
4436 spin_unlock(&mddev->lock);
4437 return err ?: len;
c6207277
N		 4438}
4439
4440static struct md_sysfs_entry md_max_sync =
4441__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4442
e464eafd 4443static ssize_t
fd01b88c 4444suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4445{
4446 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4447}
4448
4449static ssize_t
fd01b88c 4450suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4451{
4452 char *e;
4453 unsigned long long new = simple_strtoull(buf, &e, 10);
6791875e
N		 4454 unsigned long long old;
4455 int err;
e464eafd 4456
e464eafd
N		 4457 if (buf == e || (*e && *e != '\n'))
4458 return -EINVAL;
23ddff37 4459
6791875e
N		 4460 err = mddev_lock(mddev);
4461 if (err)
4462 return err;
4463 err = -EINVAL;
4464 if (mddev->pers == NULL ||
4465 mddev->pers->quiesce == NULL)
4466 goto unlock;
4467 old = mddev->suspend_lo;
23ddff37
N		 4468 mddev->suspend_lo = new;
4469 if (new >= old)
4470 /* Shrinking suspended region */
e464eafd 4471 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4472 else {
4473 /* Expanding suspended region - need to wait */
4474 mddev->pers->quiesce(mddev, 1);
4475 mddev->pers->quiesce(mddev, 0);
4476 }
6791875e
N		 4477 err = 0;
4478unlock:
4479 mddev_unlock(mddev);
4480 return err ?: len;
e464eafd
N		 4481}
4482static struct md_sysfs_entry md_suspend_lo =
4483__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4484
e464eafd 4485static ssize_t
fd01b88c 4486suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4487{
4488 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4489}
4490
4491static ssize_t
fd01b88c 4492suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4493{
4494 char *e;
4495 unsigned long long new = simple_strtoull(buf, &e, 10);
6791875e
N		 4496 unsigned long long old;
4497 int err;
e464eafd 4498
e464eafd
N		 4499 if (buf == e || (*e && *e != '\n'))
4500 return -EINVAL;
23ddff37 4501
6791875e
N		 4502 err = mddev_lock(mddev);
4503 if (err)
4504 return err;
4505 err = -EINVAL;
4506 if (mddev->pers == NULL ||
4507 mddev->pers->quiesce == NULL)
4508 goto unlock;
4509 old = mddev->suspend_hi;
23ddff37
N		 4510 mddev->suspend_hi = new;
4511 if (new <= old)
4512 /* Shrinking suspended region */
4513 mddev->pers->quiesce(mddev, 2);
4514 else {
4515 /* Expanding suspended region - need to wait */
e464eafd
N		 4516 mddev->pers->quiesce(mddev, 1);
4517 mddev->pers->quiesce(mddev, 0);
23ddff37 4518 }
6791875e
N		 4519 err = 0;
4520unlock:
4521 mddev_unlock(mddev);
4522 return err ?: len;
e464eafd
N		 4523}
4524static struct md_sysfs_entry md_suspend_hi =
4525__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4526
08a02ecd 4527static ssize_t
fd01b88c 4528reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4529{
4530 if (mddev->reshape_position != MaxSector)
4531 return sprintf(page, "%llu\n",
4532 (unsigned long long)mddev->reshape_position);
4533 strcpy(page, "none\n");
4534 return 5;
4535}
4536
4537static ssize_t
fd01b88c 4538reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd 4539{
c6563a8c 4540 struct md_rdev *rdev;
08a02ecd 4541 char *e;
6791875e 4542 int err;
08a02ecd 4543 unsigned long long new = simple_strtoull(buf, &e, 10);
6791875e 4544
08a02ecd
N		 4545 if (buf == e || (*e && *e != '\n'))
4546 return -EINVAL;
6791875e
N		 4547 err = mddev_lock(mddev);
4548 if (err)
4549 return err;
4550 err = -EBUSY;
4551 if (mddev->pers)
4552 goto unlock;
08a02ecd
N		 4553 mddev->reshape_position = new;
4554 mddev->delta_disks = 0;
2c810cdd 4555 mddev->reshape_backwards = 0;
08a02ecd
N		 4556 mddev->new_level = mddev->level;
4557 mddev->new_layout = mddev->layout;
664e7c41 4558 mddev->new_chunk_sectors = mddev->chunk_sectors;
c6563a8c
N		 4559 rdev_for_each(rdev, mddev)
4560 rdev->new_data_offset = rdev->data_offset;
6791875e
N		 4561 err = 0;
4562unlock:
4563 mddev_unlock(mddev);
4564 return err ?: len;
08a02ecd
N		 4565}
4566
4567static struct md_sysfs_entry md_reshape_position =
4568__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4569 reshape_position_store);
4570
2c810cdd
N		 4571static ssize_t
4572reshape_direction_show(struct mddev *mddev, char *page)
4573{
4574 return sprintf(page, "%s\n",
4575 mddev->reshape_backwards ? "backwards" : "forwards");
4576}
4577
4578static ssize_t
4579reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4580{
4581 int backwards = 0;
6791875e
N		 4582 int err;
4583
2c810cdd
N		 4584 if (cmd_match(buf, "forwards"))
4585 backwards = 0;
4586 else if (cmd_match(buf, "backwards"))
4587 backwards = 1;
4588 else
4589 return -EINVAL;
4590 if (mddev->reshape_backwards == backwards)
4591 return len;
4592
6791875e
N		 4593 err = mddev_lock(mddev);
4594 if (err)
4595 return err;
2c810cdd
N		 4596 /* check if we are allowed to change */
4597 if (mddev->delta_disks)
6791875e
N		 4598 err = -EBUSY;
4599 else if (mddev->persistent &&
2c810cdd 4600 mddev->major_version == 0)
6791875e
N		 4601 err = -EINVAL;
4602 else
4603 mddev->reshape_backwards = backwards;
4604 mddev_unlock(mddev);
4605 return err ?: len;
2c810cdd
N		 4606}
4607
4608static struct md_sysfs_entry md_reshape_direction =
4609__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4610 reshape_direction_store);
4611
b522adcd 4612static ssize_t
fd01b88c 4613array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4614{
4615 if (mddev->external_size)
4616 return sprintf(page, "%llu\n",
4617 (unsigned long long)mddev->array_sectors/2);
4618 else
4619 return sprintf(page, "default\n");
4620}
4621
4622static ssize_t
fd01b88c 4623array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4624{
4625 sector_t sectors;
6791875e
N		 4626 int err;
4627
4628 err = mddev_lock(mddev);
4629 if (err)
4630 return err;
b522adcd
DW		 4631
4632 if (strncmp(buf, "default", 7) == 0) {
4633 if (mddev->pers)
4634 sectors = mddev->pers->size(mddev, 0, 0);
4635 else
4636 sectors = mddev->array_sectors;
4637
4638 mddev->external_size = 0;
4639 } else {
4640 if (strict_blocks_to_sectors(buf, &sectors) < 0)
6791875e
N		 4641 err = -EINVAL;
4642 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4643 err = -E2BIG;
4644 else
4645 mddev->external_size = 1;
b522adcd
DW		 4646 }
4647
6791875e
N		 4648 if (!err) {
4649 mddev->array_sectors = sectors;
4650 if (mddev->pers) {
4651 set_capacity(mddev->gendisk, mddev->array_sectors);
4652 revalidate_disk(mddev->gendisk);
4653 }
cbe6ef1d 4654 }
6791875e
N		 4655 mddev_unlock(mddev);
4656 return err ?: len;
b522adcd
DW		 4657}
4658
4659static struct md_sysfs_entry md_array_size =
4660__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4661 array_size_store);
e464eafd 4662
eae1701f
N		 4663static struct attribute *md_default_attrs[] = {
4664 &md_level.attr,
d4dbd025 4665 &md_layout.attr,
eae1701f 4666 &md_raid_disks.attr,
3b34380a 4667 &md_chunk_size.attr,
a35b0d69 4668 &md_size.attr,
a94213b1 4669 &md_resync_start.attr,
8bb93aac 4670 &md_metadata.attr,
6d7ff738 4671 &md_new_device.attr,
16f17b39 4672 &md_safe_delay.attr,
9e653b63 4673 &md_array_state.attr,
08a02ecd 4674 &md_reshape_position.attr,
2c810cdd 4675 &md_reshape_direction.attr,
b522adcd 4676 &md_array_size.attr,
1e50915f 4677 &max_corr_read_errors.attr,
411036fa
N		 4678 NULL,
4679};
4680
4681static struct attribute *md_redundancy_attrs[] = {
24dd469d 4682 &md_scan_mode.attr,
c4a39551 4683 &md_last_scan_mode.attr,
9d88883e 4684 &md_mismatches.attr,
88202a0c
N		 4685 &md_sync_min.attr,
4686 &md_sync_max.attr,
4687 &md_sync_speed.attr,
90b08710 4688 &md_sync_force_parallel.attr,
88202a0c 4689 &md_sync_completed.attr,
5e96ee65 4690 &md_min_sync.attr,
c6207277 4691 &md_max_sync.attr,
e464eafd
N		 4692 &md_suspend_lo.attr,
4693 &md_suspend_hi.attr,
9b1d1dac 4694 &md_bitmap.attr,
d7f3d291 4695 &md_degraded.attr,
eae1701f
N		 4696 NULL,
4697};
411036fa
N		 4698static struct attribute_group md_redundancy_group = {
4699 .name = NULL,
4700 .attrs = md_redundancy_attrs,
4701};
4702
eae1701f
N		 4703static ssize_t
4704md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4705{
4706 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4707 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4708 ssize_t rv;
eae1701f
N		 4709
4710 if (!entry->show)
4711 return -EIO;
af8a2434
N		 4712 spin_lock(&all_mddevs_lock);
4713 if (list_empty(&mddev->all_mddevs)) {
4714 spin_unlock(&all_mddevs_lock);
4715 return -EBUSY;
4716 }
4717 mddev_get(mddev);
4718 spin_unlock(&all_mddevs_lock);
4719
b7b17c9b 4720 rv = entry->show(mddev, page);
af8a2434 4721 mddev_put(mddev);
96de1e66 4722 return rv;
eae1701f
N		 4723}
4724
4725static ssize_t
4726md_attr_store(struct kobject *kobj, struct attribute *attr,
4727 const char *page, size_t length)
4728{
4729 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4730 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4731 ssize_t rv;
eae1701f
N		 4732
4733 if (!entry->store)
4734 return -EIO;
67463acb
N		 4735 if (!capable(CAP_SYS_ADMIN))
4736 return -EACCES;
af8a2434
N		 4737 spin_lock(&all_mddevs_lock);
4738 if (list_empty(&mddev->all_mddevs)) {
4739 spin_unlock(&all_mddevs_lock);
4740 return -EBUSY;
4741 }
4742 mddev_get(mddev);
4743 spin_unlock(&all_mddevs_lock);
6791875e 4744 rv = entry->store(mddev, page, length);
af8a2434 4745 mddev_put(mddev);
96de1e66 4746 return rv;
eae1701f
N		 4747}
4748
4749static void md_free(struct kobject *ko)
4750{
fd01b88c 4751 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4752
4753 if (mddev->sysfs_state)
4754 sysfs_put(mddev->sysfs_state);
4755
4756 if (mddev->gendisk) {
4757 del_gendisk(mddev->gendisk);
4758 put_disk(mddev->gendisk);
4759 }
4760 if (mddev->queue)
4761 blk_cleanup_queue(mddev->queue);
4762
eae1701f
N		 4763 kfree(mddev);
4764}
4765
52cf25d0 4766static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4767 .show = md_attr_show,
4768 .store = md_attr_store,
4769};
4770static struct kobj_type md_ktype = {
4771 .release = md_free,
4772 .sysfs_ops = &md_sysfs_ops,
4773 .default_attrs = md_default_attrs,
4774};
4775
1da177e4
LT		 4776int mdp_major = 0;
4777
5fd3a17e
DW		 4778static void mddev_delayed_delete(struct work_struct *ws)
4779{
fd01b88c 4780 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4781
43a70507 4782 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4783 kobject_del(&mddev->kobj);
4784 kobject_put(&mddev->kobj);
4785}
4786
efeb53c0 4787static int md_alloc(dev_t dev, char *name)
1da177e4 4788{
48c9c27b 4789 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4790 struct mddev *mddev = mddev_find(dev);
1da177e4 4791 struct gendisk *disk;
efeb53c0
N		 4792 int partitioned;
4793 int shift;
4794 int unit;
3830c62f 4795 int error;
1da177e4
LT		 4796
4797 if (!mddev)
efeb53c0
N		 4798 return -ENODEV;
4799
4800 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4801 shift = partitioned ? MdpMinorShift : 0;
4802 unit = MINOR(mddev->unit) >> shift;
1da177e4 4803
e804ac78
TH		 4804 /* wait for any previous instance of this device to be
4805 * completely removed (mddev_delayed_delete).
d3374825 4806 */
e804ac78 4807 flush_workqueue(md_misc_wq);
d3374825 4808
48c9c27b 4809 mutex_lock(&disks_mutex);
0909dc44
N		 4810 error = -EEXIST;
4811 if (mddev->gendisk)
4812 goto abort;
efeb53c0
N		 4813
4814 if (name) {
4815 /* Need to ensure that 'name' is not a duplicate.
4816 */
fd01b88c 4817 struct mddev *mddev2;
efeb53c0
N		 4818 spin_lock(&all_mddevs_lock);
4819
4820 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4821 if (mddev2->gendisk &&
4822 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4823 spin_unlock(&all_mddevs_lock);
0909dc44 4824 goto abort;
efeb53c0
N		 4825 }
4826 spin_unlock(&all_mddevs_lock);
1da177e4 4827 }
8b765398 4828
0909dc44 4829 error = -ENOMEM;
8b765398 4830 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4831 if (!mddev->queue)
4832 goto abort;
409c57f3
N		 4833 mddev->queue->queuedata = mddev;
4834
409c57f3 4835 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 4836 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 4837
1da177e4
LT		 4838 disk = alloc_disk(1 << shift);
4839 if (!disk) {
8b765398
N		 4840 blk_cleanup_queue(mddev->queue);
4841 mddev->queue = NULL;
0909dc44 4842 goto abort;
1da177e4 4843 }
efeb53c0 4844 disk->major = MAJOR(mddev->unit);
1da177e4 4845 disk->first_minor = unit << shift;
efeb53c0
N		 4846 if (name)
4847 strcpy(disk->disk_name, name);
4848 else if (partitioned)
1da177e4 4849 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4850 else
1da177e4 4851 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4852 disk->fops = &md_fops;
4853 disk->private_data = mddev;
4854 disk->queue = mddev->queue;
b0140891 4855 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4856 /* Allow extended partitions. This makes the
d3374825 4857 * 'mdp' device redundant, but we can't really
92850bbd
N		 4858 * remove it now.
4859 */
4860 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4861 mddev->gendisk = disk;
b0140891
N		 4862 /* As soon as we call add_disk(), another thread could get
4863 * through to md_open, so make sure it doesn't get too far
4864 */
4865 mutex_lock(&mddev->open_mutex);
4866 add_disk(disk);
4867
ed9e1982
TH		 4868 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4869 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4870 if (error) {
4871 /* This isn't possible, but as kobject_init_and_add is marked
4872 * __must_check, we must do something with the result
4873 */
5e55e2f5
N		 4874 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4875 disk->disk_name);
0909dc44
N		 4876 error = 0;
4877 }
00bcb4ac
N		 4878 if (mddev->kobj.sd &&
4879 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4880 printk(KERN_DEBUG "pointless warning\n");
b0140891 4881 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4882 abort:
4883 mutex_unlock(&disks_mutex);
00bcb4ac 4884 if (!error && mddev->kobj.sd) {
3830c62f 4885 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4886 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4887 }
d3374825 4888 mddev_put(mddev);
0909dc44 4889 return error;
efeb53c0
N		 4890}
4891
4892static struct kobject *md_probe(dev_t dev, int *part, void *data)
4893{
4894 md_alloc(dev, NULL);
1da177e4
LT		 4895 return NULL;
4896}
4897
efeb53c0
N		 4898static int add_named_array(const char *val, struct kernel_param *kp)
4899{
4900 /* val must be "md_*" where * is not all digits.
4901 * We allocate an array with a large free minor number, and
4902 * set the name to val. val must not already be an active name.
4903 */
4904 int len = strlen(val);
4905 char buf[DISK_NAME_LEN];
4906
4907 while (len && val[len-1] == '\n')
4908 len--;
4909 if (len >= DISK_NAME_LEN)
4910 return -E2BIG;
4911 strlcpy(buf, val, len+1);
4912 if (strncmp(buf, "md_", 3) != 0)
4913 return -EINVAL;
4914 return md_alloc(0, buf);
4915}
4916
1da177e4
LT		 4917static void md_safemode_timeout(unsigned long data)
4918{
fd01b88c 4919 struct mddev *mddev = (struct mddev *) data;
1da177e4 4920
0fd62b86
NB		 4921 if (!atomic_read(&mddev->writes_pending)) {
4922 mddev->safemode = 1;
4923 if (mddev->external)
00bcb4ac 4924 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4925 }
1da177e4
LT		 4926 md_wakeup_thread(mddev->thread);
4927}
4928
6ff8d8ec 4929static int start_dirty_degraded;
1da177e4 4930
fd01b88c 4931int md_run(struct mddev *mddev)
1da177e4 4932{
2604b703 4933 int err;
3cb03002 4934 struct md_rdev *rdev;
84fc4b56 4935 struct md_personality *pers;
1da177e4 4936
a757e64c
N		 4937 if (list_empty(&mddev->disks))
4938 /* cannot run an array with no devices.. */
1da177e4 4939 return -EINVAL;
1da177e4
LT		 4940
4941 if (mddev->pers)
4942 return -EBUSY;
bb4f1e9d
N		 4943 /* Cannot run until previous stop completes properly */
4944 if (mddev->sysfs_active)
4945 return -EBUSY;
b6eb127d 4946
1da177e4
LT		 4947 /*
4948 * Analyze all RAID superblock(s)
4949 */
1ec4a939
N		 4950 if (!mddev->raid_disks) {
4951 if (!mddev->persistent)
4952 return -EINVAL;
a757e64c 4953 analyze_sbs(mddev);
1ec4a939 4954 }
1da177e4 4955
d9d166c2
N		 4956 if (mddev->level != LEVEL_NONE)
4957 request_module("md-level-%d", mddev->level);
4958 else if (mddev->clevel[0])
4959 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4960
4961 /*
4962 * Drop all container device buffers, from now on
4963 * the only valid external interface is through the md
4964 * device.
1da177e4 4965 */
dafb20fa 4966 rdev_for_each(rdev, mddev) {
b2d444d7 4967 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4968 continue;
4969 sync_blockdev(rdev->bdev);
f98393a6 4970 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4971
4972 /* perform some consistency tests on the device.
4973 * We don't want the data to overlap the metadata,
58c0fed4 4974 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4975 */
a6ff7e08
JB		 4976 if (rdev->meta_bdev) {
4977 /* Nothing to check */;
4978 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4979 if (mddev->dev_sectors &&
4980 rdev->data_offset + mddev->dev_sectors
0f420358 4981 > rdev->sb_start) {
f0d76d70
N		 4982 printk("md: %s: data overlaps metadata\n",
4983 mdname(mddev));
4984 return -EINVAL;
4985 }
4986 } else {
0f420358 4987 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4988 > rdev->data_offset) {
4989 printk("md: %s: metadata overlaps data\n",
4990 mdname(mddev));
4991 return -EINVAL;
4992 }
4993 }
00bcb4ac 4994 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 4995 }
4996
a167f663 4997 if (mddev->bio_set == NULL)
395c72a7 4998 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
a167f663 4999
1da177e4 5000 spin_lock(&pers_lock);
d9d166c2 5001 pers = find_pers(mddev->level, mddev->clevel);
2604b703 5002 if (!pers || !try_module_get(pers->owner)) {
1da177e4 5003 spin_unlock(&pers_lock);
d9d166c2
N		 5004 if (mddev->level != LEVEL_NONE)
5005 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5006 mddev->level);
5007 else
5008 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5009 mddev->clevel);
1da177e4
LT		 5010 return -EINVAL;
5011 }
1da177e4 5012 spin_unlock(&pers_lock);
34817e8c
N		 5013 if (mddev->level != pers->level) {
5014 mddev->level = pers->level;
5015 mddev->new_level = pers->level;
5016 }
d9d166c2 5017 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 5018
f6705578 5019 if (mddev->reshape_position != MaxSector &&
63c70c4f 5020 pers->start_reshape == NULL) {
f6705578 5021 /* This personality cannot handle reshaping... */
f6705578
N		 5022 module_put(pers->owner);
5023 return -EINVAL;
5024 }
5025
7dd5e7c3
N		 5026 if (pers->sync_request) {
5027 /* Warn if this is a potentially silly
5028 * configuration.
5029 */
5030 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5031 struct md_rdev *rdev2;
7dd5e7c3 5032 int warned = 0;
159ec1fc 5033
dafb20fa
N		 5034 rdev_for_each(rdev, mddev)
5035 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N		 5036 if (rdev < rdev2 &&
5037 rdev->bdev->bd_contains ==
5038 rdev2->bdev->bd_contains) {
5039 printk(KERN_WARNING
5040 "%s: WARNING: %s appears to be"
5041 " on the same physical disk as"
5042 " %s.\n",
5043 mdname(mddev),
5044 bdevname(rdev->bdev,b),
5045 bdevname(rdev2->bdev,b2));
5046 warned = 1;
5047 }
5048 }
159ec1fc 5049
7dd5e7c3
N		 5050 if (warned)
5051 printk(KERN_WARNING
5052 "True protection against single-disk"
5053 " failure might be compromised.\n");
5054 }
5055
657390d2 5056 mddev->recovery = 0;
58c0fed4
AN		 5057 /* may be over-ridden by personality */
5058 mddev->resync_max_sectors = mddev->dev_sectors;
5059
6ff8d8ec 5060 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5061
0f9552b5 5062 if (start_readonly && mddev->ro == 0)
f91de92e
N		 5063 mddev->ro = 2; /* read-only, but switch on first write */
5064
36d091f4 5065 err = pers->run(mddev);
13e53df3
AN		 5066 if (err)
5067 printk(KERN_ERR "md: pers->run() failed ...\n");
36d091f4 5068 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
b522adcd
DW		 5069 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5070 " but 'external_size' not in effect?\n", __func__);
5071 printk(KERN_ERR
5072 "md: invalid array_size %llu > default size %llu\n",
5073 (unsigned long long)mddev->array_sectors / 2,
36d091f4 5074 (unsigned long long)pers->size(mddev, 0, 0) / 2);
b522adcd 5075 err = -EINVAL;
b522adcd 5076 }
36d091f4 5077 if (err == 0 && pers->sync_request &&
ef99bf48 5078 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
f9209a32
GR		 5079 struct bitmap *bitmap;
5080
5081 bitmap = bitmap_create(mddev, -1);
5082 if (IS_ERR(bitmap)) {
5083 err = PTR_ERR(bitmap);
b15c2e57
N		 5084 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5085 mdname(mddev), err);
f9209a32
GR		 5086 } else
5087 mddev->bitmap = bitmap;
5088
b15c2e57 5089 }
1da177e4 5090 if (err) {
5aa61f42 5091 mddev_detach(mddev);
36d091f4
N		 5092 pers->free(mddev, mddev->private);
5093 module_put(pers->owner);
32a7627c
N		 5094 bitmap_destroy(mddev);
5095 return err;
1da177e4 5096 }
5c675f83
N		 5097 if (mddev->queue) {
5098 mddev->queue->backing_dev_info.congested_data = mddev;
5099 mddev->queue->backing_dev_info.congested_fn = md_congested;
64590f45 5100 blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec);
5c675f83 5101 }
36d091f4 5102 if (pers->sync_request) {
00bcb4ac
N		 5103 if (mddev->kobj.sd &&
5104 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 5105 printk(KERN_WARNING
5106 "md: cannot register extra attributes for %s\n",
5107 mdname(mddev));
00bcb4ac 5108 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5109 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 5110 mddev->ro = 0;
5111
f72ffdd6 5112 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 5113 atomic_set(&mddev->max_corr_read_errors,
5114 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 5115 mddev->safemode = 0;
5116 mddev->safemode_timer.function = md_safemode_timeout;
5117 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 5118 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5119 mddev->in_sync = 1;
0ca69886 5120 smp_wmb();
36d091f4
N		 5121 spin_lock(&mddev->lock);
5122 mddev->pers = pers;
0ca69886 5123 mddev->ready = 1;
36d091f4 5124 spin_unlock(&mddev->lock);
dafb20fa 5125 rdev_for_each(rdev, mddev)
36fad858
NK		 5126 if (rdev->raid_disk >= 0)
5127 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 5128 /* failure here is OK */;
f72ffdd6 5129
1da177e4 5130 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f72ffdd6 5131
7a0a5355 5132 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 5133 md_update_sb(mddev, 0);
1da177e4 5134
d7603b7e 5135 md_new_event(mddev);
00bcb4ac
N		 5136 sysfs_notify_dirent_safe(mddev->sysfs_state);
5137 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 5138 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 5139 return 0;
5140}
390ee602 5141EXPORT_SYMBOL_GPL(md_run);
1da177e4 5142
fd01b88c 5143static int do_md_run(struct mddev *mddev)
fe60b014
N		 5144{
5145 int err;
5146
5147 err = md_run(mddev);
5148 if (err)
5149 goto out;
69e51b44
N		 5150 err = bitmap_load(mddev);
5151 if (err) {
5152 bitmap_destroy(mddev);
5153 goto out;
5154 }
0fd018af
JB		 5155
5156 md_wakeup_thread(mddev->thread);
5157 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5158
fe60b014
N		 5159 set_capacity(mddev->gendisk, mddev->array_sectors);
5160 revalidate_disk(mddev->gendisk);
f0b4f7e2 5161 mddev->changed = 1;
fe60b014
N		 5162 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5163out:
5164 return err;
5165}
5166
fd01b88c 5167static int restart_array(struct mddev *mddev)
1da177e4
LT		 5168{
5169 struct gendisk *disk = mddev->gendisk;
1da177e4 5170
80fab1d7 5171 /* Complain if it has no devices */
1da177e4 5172 if (list_empty(&mddev->disks))
80fab1d7
AN		 5173 return -ENXIO;
5174 if (!mddev->pers)
5175 return -EINVAL;
5176 if (!mddev->ro)
5177 return -EBUSY;
5178 mddev->safemode = 0;
5179 mddev->ro = 0;
5180 set_disk_ro(disk, 0);
5181 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5182 mdname(mddev));
5183 /* Kick recovery or resync if necessary */
5184 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5185 md_wakeup_thread(mddev->thread);
5186 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5187 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5188 return 0;
1da177e4
LT		 5189}
5190
fd01b88c 5191static void md_clean(struct mddev *mddev)
6177b472
N		 5192{
5193 mddev->array_sectors = 0;
5194 mddev->external_size = 0;
5195 mddev->dev_sectors = 0;
5196 mddev->raid_disks = 0;
5197 mddev->recovery_cp = 0;
5198 mddev->resync_min = 0;
5199 mddev->resync_max = MaxSector;
5200 mddev->reshape_position = MaxSector;
5201 mddev->external = 0;
5202 mddev->persistent = 0;
5203 mddev->level = LEVEL_NONE;
5204 mddev->clevel[0] = 0;
5205 mddev->flags = 0;
5206 mddev->ro = 0;
5207 mddev->metadata_type[0] = 0;
5208 mddev->chunk_sectors = 0;
5209 mddev->ctime = mddev->utime = 0;
5210 mddev->layout = 0;
5211 mddev->max_disks = 0;
5212 mddev->events = 0;
a8707c08 5213 mddev->can_decrease_events = 0;
6177b472 5214 mddev->delta_disks = 0;
2c810cdd 5215 mddev->reshape_backwards = 0;
6177b472
N		 5216 mddev->new_level = LEVEL_NONE;
5217 mddev->new_layout = 0;
5218 mddev->new_chunk_sectors = 0;
5219 mddev->curr_resync = 0;
7f7583d4 5220 atomic64_set(&mddev->resync_mismatches, 0);
6177b472
N		 5221 mddev->suspend_lo = mddev->suspend_hi = 0;
5222 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5223 mddev->recovery = 0;
5224 mddev->in_sync = 0;
f0b4f7e2 5225 mddev->changed = 0;
6177b472 5226 mddev->degraded = 0;
6177b472 5227 mddev->safemode = 0;
050b6615 5228 mddev->merge_check_needed = 0;
6177b472
N		 5229 mddev->bitmap_info.offset = 0;
5230 mddev->bitmap_info.default_offset = 0;
6409bb05 5231 mddev->bitmap_info.default_space = 0;
6177b472
N		 5232 mddev->bitmap_info.chunksize = 0;
5233 mddev->bitmap_info.daemon_sleep = 0;
5234 mddev->bitmap_info.max_write_behind = 0;
5235}
5236
fd01b88c 5237static void __md_stop_writes(struct mddev *mddev)
a047e125 5238{
6b6204ee 5239 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
f851b60d 5240 flush_workqueue(md_misc_wq);
a047e125 5241 if (mddev->sync_thread) {
a047e125 5242 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 5243 md_reap_sync_thread(mddev);
a047e125
N		 5244 }
5245
5246 del_timer_sync(&mddev->safemode_timer);
5247
5248 bitmap_flush(mddev);
5249 md_super_wait(mddev);
5250
b6d428c6 5251 if (mddev->ro == 0 &&
7a0a5355 5252 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
a047e125
N		 5253 /* mark array as shutdown cleanly */
5254 mddev->in_sync = 1;
5255 md_update_sb(mddev, 1);
5256 }
5257}
defad61a 5258
fd01b88c 5259void md_stop_writes(struct mddev *mddev)
defad61a 5260{
29f097c4 5261 mddev_lock_nointr(mddev);
defad61a
N		 5262 __md_stop_writes(mddev);
5263 mddev_unlock(mddev);
5264}
390ee602 5265EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5266
5aa61f42
N		 5267static void mddev_detach(struct mddev *mddev)
5268{
5269 struct bitmap *bitmap = mddev->bitmap;
5270 /* wait for behind writes to complete */
5271 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5272 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5273 mdname(mddev));
5274 /* need to kick something here to make sure I/O goes? */
5275 wait_event(bitmap->behind_wait,
5276 atomic_read(&bitmap->behind_writes) == 0);
5277 }
36d091f4 5278 if (mddev->pers && mddev->pers->quiesce) {
5aa61f42
N		 5279 mddev->pers->quiesce(mddev, 1);
5280 mddev->pers->quiesce(mddev, 0);
5281 }
5282 md_unregister_thread(&mddev->thread);
5283 if (mddev->queue)
5284 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5285}
5286
5eff3c43 5287static void __md_stop(struct mddev *mddev)
6177b472 5288{
36d091f4 5289 struct md_personality *pers = mddev->pers;
5aa61f42 5290 mddev_detach(mddev);
36d091f4
N		 5291 spin_lock(&mddev->lock);
5292 mddev->ready = 0;
6177b472 5293 mddev->pers = NULL;
36d091f4
N		 5294 spin_unlock(&mddev->lock);
5295 pers->free(mddev, mddev->private);
5296 if (pers->sync_request && mddev->to_remove == NULL)
5297 mddev->to_remove = &md_redundancy_group;
5298 module_put(pers->owner);
cca9cf90 5299 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5300}
5eff3c43
N		 5301
5302void md_stop(struct mddev *mddev)
5303{
5304 /* stop the array and free an attached data structures.
5305 * This is called from dm-raid
5306 */
5307 __md_stop(mddev);
5308 bitmap_destroy(mddev);
5309 if (mddev->bio_set)
5310 bioset_free(mddev->bio_set);
5311}
5312
390ee602 5313EXPORT_SYMBOL_GPL(md_stop);
6177b472 5314
a05b7ea0 5315static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N		 5316{
5317 int err = 0;
30b8feb7
N		 5318 int did_freeze = 0;
5319
5320 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5321 did_freeze = 1;
5322 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5323 md_wakeup_thread(mddev->thread);
5324 }
f851b60d 5325 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5326 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5327 if (mddev->sync_thread)
30b8feb7
N		 5328 /* Thread might be blocked waiting for metadata update
5329 * which will now never happen */
5330 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5331
30b8feb7 5332 mddev_unlock(mddev);
f851b60d
N		 5333 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5334 &mddev->recovery));
30b8feb7
N		 5335 mddev_lock_nointr(mddev);
5336
a4bd82d0 5337 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5338 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7 5339 mddev->sync_thread ||
f851b60d 5340 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5341 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
a4bd82d0 5342 printk("md: %s still in use.\n",mdname(mddev));
30b8feb7
N		 5343 if (did_freeze) {
5344 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5345 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5346 md_wakeup_thread(mddev->thread);
5347 }
a4bd82d0
N		 5348 err = -EBUSY;
5349 goto out;
5350 }
5351 if (mddev->pers) {
defad61a 5352 __md_stop_writes(mddev);
a4bd82d0
N		 5353
5354 err = -ENXIO;
5355 if (mddev->ro==1)
5356 goto out;
5357 mddev->ro = 1;
5358 set_disk_ro(mddev->gendisk, 1);
5359 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d
N		 5360 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5361 md_wakeup_thread(mddev->thread);
00bcb4ac 5362 sysfs_notify_dirent_safe(mddev->sysfs_state);
30b8feb7 5363 err = 0;
a4bd82d0
N		 5364 }
5365out:
5366 mutex_unlock(&mddev->open_mutex);
5367 return err;
5368}
5369
9e653b63
N		 5370/* mode:
5371 * 0 - completely stop and dis-assemble array
9e653b63
N		 5372 * 2 - stop but do not disassemble array
5373 */
f72ffdd6 5374static int do_md_stop(struct mddev *mddev, int mode,
a05b7ea0 5375 struct block_device *bdev)
1da177e4 5376{
1da177e4 5377 struct gendisk *disk = mddev->gendisk;
3cb03002 5378 struct md_rdev *rdev;
30b8feb7
N		 5379 int did_freeze = 0;
5380
5381 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5382 did_freeze = 1;
5383 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5384 md_wakeup_thread(mddev->thread);
5385 }
f851b60d 5386 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5387 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5388 if (mddev->sync_thread)
30b8feb7
N		 5389 /* Thread might be blocked waiting for metadata update
5390 * which will now never happen */
5391 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5392
30b8feb7 5393 mddev_unlock(mddev);
f851b60d
N		 5394 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5395 !test_bit(MD_RECOVERY_RUNNING,
5396 &mddev->recovery)));
30b8feb7 5397 mddev_lock_nointr(mddev);
1da177e4 5398
c8c00a69 5399 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5400 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7
N		 5401 mddev->sysfs_active ||
5402 mddev->sync_thread ||
f851b60d 5403 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5404 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
df5b20cf 5405 printk("md: %s still in use.\n",mdname(mddev));
6e17b027 5406 mutex_unlock(&mddev->open_mutex);
30b8feb7
N		 5407 if (did_freeze) {
5408 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5409 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5410 md_wakeup_thread(mddev->thread);
5411 }
260fa034
N		 5412 return -EBUSY;
5413 }
6e17b027 5414 if (mddev->pers) {
a4bd82d0
N		 5415 if (mddev->ro)
5416 set_disk_ro(disk, 0);
409c57f3 5417
defad61a 5418 __md_stop_writes(mddev);
5eff3c43 5419 __md_stop(mddev);
a4bd82d0 5420 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5421 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5422
a4bd82d0 5423 /* tell userspace to handle 'inactive' */
00bcb4ac 5424 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5425
dafb20fa 5426 rdev_for_each(rdev, mddev)
36fad858
NK		 5427 if (rdev->raid_disk >= 0)
5428 sysfs_unlink_rdev(mddev, rdev);
c4647292 5429
a4bd82d0 5430 set_capacity(disk, 0);
6e17b027 5431 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5432 mddev->changed = 1;
a4bd82d0 5433 revalidate_disk(disk);
0d4ca600 5434
a4bd82d0
N		 5435 if (mddev->ro)
5436 mddev->ro = 0;
6e17b027
N		 5437 } else
5438 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5439 /*
5440 * Free resources if final stop
5441 */
9e653b63 5442 if (mode == 0) {
1da177e4
LT		 5443 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5444
978f946b 5445 bitmap_destroy(mddev);
c3d9714e 5446 if (mddev->bitmap_info.file) {
4af1a041
N		 5447 struct file *f = mddev->bitmap_info.file;
5448 spin_lock(&mddev->lock);
c3d9714e 5449 mddev->bitmap_info.file = NULL;
4af1a041
N		 5450 spin_unlock(&mddev->lock);
5451 fput(f);
978f946b 5452 }
c3d9714e 5453 mddev->bitmap_info.offset = 0;
978f946b 5454
1da177e4
LT		 5455 export_array(mddev);
5456
6177b472 5457 md_clean(mddev);
934d9c23 5458 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5459 if (mddev->hold_active == UNTIL_STOP)
5460 mddev->hold_active = 0;
a4bd82d0 5461 }
3f9d99c1 5462 blk_integrity_unregister(disk);
d7603b7e 5463 md_new_event(mddev);
00bcb4ac 5464 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5465 return 0;
1da177e4
LT		 5466}
5467
fdee8ae4 5468#ifndef MODULE
fd01b88c 5469static void autorun_array(struct mddev *mddev)
1da177e4 5470{
3cb03002 5471 struct md_rdev *rdev;
1da177e4
LT		 5472 int err;
5473
a757e64c 5474 if (list_empty(&mddev->disks))
1da177e4 5475 return;
1da177e4
LT		 5476
5477 printk(KERN_INFO "md: running: ");
5478
dafb20fa 5479 rdev_for_each(rdev, mddev) {
1da177e4
LT		 5480 char b[BDEVNAME_SIZE];
5481 printk("<%s>", bdevname(rdev->bdev,b));
5482 }
5483 printk("\n");
5484
d710e138 5485 err = do_md_run(mddev);
1da177e4
LT		 5486 if (err) {
5487 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
a05b7ea0 5488 do_md_stop(mddev, 0, NULL);
1da177e4
LT		 5489 }
5490}
5491
5492/*
5493 * lets try to run arrays based on all disks that have arrived
5494 * until now. (those are in pending_raid_disks)
5495 *
5496 * the method: pick the first pending disk, collect all disks with
5497 * the same UUID, remove all from the pending list and put them into
5498 * the 'same_array' list. Then order this list based on superblock
5499 * update time (freshest comes first), kick out 'old' disks and
5500 * compare superblocks. If everything's fine then run it.
5501 *
5502 * If "unit" is allocated, then bump its reference count
5503 */
5504static void autorun_devices(int part)
5505{
3cb03002 5506 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5507 struct mddev *mddev;
1da177e4
LT		 5508 char b[BDEVNAME_SIZE];
5509
5510 printk(KERN_INFO "md: autorun ...\n");
5511 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5512 int unit;
1da177e4 5513 dev_t dev;
ad01c9e3 5514 LIST_HEAD(candidates);
1da177e4 5515 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5516 struct md_rdev, same_set);
1da177e4
LT		 5517
5518 printk(KERN_INFO "md: considering %s ...\n",
5519 bdevname(rdev0->bdev,b));
5520 INIT_LIST_HEAD(&candidates);
159ec1fc 5521 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5522 if (super_90_load(rdev, rdev0, 0) >= 0) {
5523 printk(KERN_INFO "md: adding %s ...\n",
5524 bdevname(rdev->bdev,b));
5525 list_move(&rdev->same_set, &candidates);
5526 }
5527 /*
5528 * now we have a set of devices, with all of them having
5529 * mostly sane superblocks. It's time to allocate the
5530 * mddev.
5531 */
e8703fe1
N		 5532 if (part) {
5533 dev = MKDEV(mdp_major,
5534 rdev0->preferred_minor << MdpMinorShift);
5535 unit = MINOR(dev) >> MdpMinorShift;
5536 } else {
5537 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5538 unit = MINOR(dev);
5539 }
5540 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5541 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5542 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5543 break;
5544 }
1da177e4
LT		 5545
5546 md_probe(dev, NULL, NULL);
5547 mddev = mddev_find(dev);
9bbbca3a
NB		 5548 if (!mddev || !mddev->gendisk) {
5549 if (mddev)
5550 mddev_put(mddev);
5551 printk(KERN_ERR
1da177e4
LT		 5552 "md: cannot allocate memory for md drive.\n");
5553 break;
5554 }
f72ffdd6 5555 if (mddev_lock(mddev))
1da177e4
LT		 5556 printk(KERN_WARNING "md: %s locked, cannot run\n",
5557 mdname(mddev));
5558 else if (mddev->raid_disks || mddev->major_version
5559 || !list_empty(&mddev->disks)) {
f72ffdd6 5560 printk(KERN_WARNING
1da177e4
LT		 5561 "md: %s already running, cannot run %s\n",
5562 mdname(mddev), bdevname(rdev0->bdev,b));
5563 mddev_unlock(mddev);
5564 } else {
5565 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5566 mddev->persistent = 1;
159ec1fc 5567 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5568 list_del_init(&rdev->same_set);
5569 if (bind_rdev_to_array(rdev, mddev))
5570 export_rdev(rdev);
5571 }
5572 autorun_array(mddev);
5573 mddev_unlock(mddev);
5574 }
5575 /* on success, candidates will be empty, on error
5576 * it won't...
5577 */
159ec1fc 5578 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5579 list_del_init(&rdev->same_set);
1da177e4 5580 export_rdev(rdev);
4b80991c 5581 }
1da177e4
LT		 5582 mddev_put(mddev);
5583 }
5584 printk(KERN_INFO "md: ... autorun DONE.\n");
5585}
fdee8ae4 5586#endif /* !MODULE */
1da177e4 5587
f72ffdd6 5588static int get_version(void __user *arg)
1da177e4
LT		 5589{
5590 mdu_version_t ver;
5591
5592 ver.major = MD_MAJOR_VERSION;
5593 ver.minor = MD_MINOR_VERSION;
5594 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5595
5596 if (copy_to_user(arg, &ver, sizeof(ver)))
5597 return -EFAULT;
5598
5599 return 0;
5600}
5601
f72ffdd6 5602static int get_array_info(struct mddev *mddev, void __user *arg)
1da177e4
LT		 5603{
5604 mdu_array_info_t info;
a9f326eb 5605 int nr,working,insync,failed,spare;
3cb03002 5606 struct md_rdev *rdev;
1da177e4 5607
1ca69c4b
N		 5608 nr = working = insync = failed = spare = 0;
5609 rcu_read_lock();
5610 rdev_for_each_rcu(rdev, mddev) {
1da177e4 5611 nr++;
b2d444d7 5612 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5613 failed++;
5614 else {
5615 working++;
b2d444d7 5616 if (test_bit(In_sync, &rdev->flags))
f72ffdd6 5617 insync++;
1da177e4
LT		 5618 else
5619 spare++;
5620 }
5621 }
1ca69c4b 5622 rcu_read_unlock();
1da177e4
LT		 5623
5624 info.major_version = mddev->major_version;
5625 info.minor_version = mddev->minor_version;
5626 info.patch_version = MD_PATCHLEVEL_VERSION;
5627 info.ctime = mddev->ctime;
5628 info.level = mddev->level;
58c0fed4
AN		 5629 info.size = mddev->dev_sectors / 2;
5630 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5631 info.size = -1;
1da177e4
LT		 5632 info.nr_disks = nr;
5633 info.raid_disks = mddev->raid_disks;
5634 info.md_minor = mddev->md_minor;
5635 info.not_persistent= !mddev->persistent;
5636
5637 info.utime = mddev->utime;
5638 info.state = 0;
5639 if (mddev->in_sync)
5640 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5641 if (mddev->bitmap && mddev->bitmap_info.offset)
9bd35920 5642 info.state |= (1<<MD_SB_BITMAP_PRESENT);
ca8895d9
GR		 5643 if (mddev_is_clustered(mddev))
5644 info.state |= (1<<MD_SB_CLUSTERED);
5645
a9f326eb 5646 info.active_disks = insync;
1da177e4
LT		 5647 info.working_disks = working;
5648 info.failed_disks = failed;
5649 info.spare_disks = spare;
5650
5651 info.layout = mddev->layout;
9d8f0363 5652 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5653
5654 if (copy_to_user(arg, &info, sizeof(info)))
5655 return -EFAULT;
5656
5657 return 0;
5658}
5659
f72ffdd6 5660static int get_bitmap_file(struct mddev *mddev, void __user * arg)
32a7627c
N		 5661{
5662 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
f4ad3d38 5663 char *ptr;
4af1a041 5664 int err;
32a7627c 5665
60559da4 5666 file = kmalloc(sizeof(*file), GFP_NOIO);
32a7627c 5667 if (!file)
4af1a041 5668 return -ENOMEM;
32a7627c 5669
4af1a041
N		 5670 err = 0;
5671 spin_lock(&mddev->lock);
32a7627c 5672 /* bitmap disabled, zero the first byte and copy out */
4af1a041 5673 if (!mddev->bitmap_info.file)
32a7627c 5674 file->pathname[0] = '\0';
4af1a041
N		 5675 else if ((ptr = d_path(&mddev->bitmap_info.file->f_path,
5676 file->pathname, sizeof(file->pathname))),
5677 IS_ERR(ptr))
5678 err = PTR_ERR(ptr);
5679 else
5680 memmove(file->pathname, ptr,
5681 sizeof(file->pathname)-(ptr-file->pathname));
5682 spin_unlock(&mddev->lock);
32a7627c 5683
4af1a041
N		 5684 if (err == 0 &&
5685 copy_to_user(arg, file, sizeof(*file)))
32a7627c 5686 err = -EFAULT;
4af1a041 5687
32a7627c
N		 5688 kfree(file);
5689 return err;
5690}
5691
f72ffdd6 5692static int get_disk_info(struct mddev *mddev, void __user * arg)
1da177e4
LT		 5693{
5694 mdu_disk_info_t info;
3cb03002 5695 struct md_rdev *rdev;
1da177e4
LT		 5696
5697 if (copy_from_user(&info, arg, sizeof(info)))
5698 return -EFAULT;
5699
1ca69c4b
N		 5700 rcu_read_lock();
5701 rdev = find_rdev_nr_rcu(mddev, info.number);
1da177e4
LT		 5702 if (rdev) {
5703 info.major = MAJOR(rdev->bdev->bd_dev);
5704 info.minor = MINOR(rdev->bdev->bd_dev);
5705 info.raid_disk = rdev->raid_disk;
5706 info.state = 0;
b2d444d7 5707 if (test_bit(Faulty, &rdev->flags))
1da177e4 5708 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5709 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5710 info.state |= (1<<MD_DISK_ACTIVE);
5711 info.state |= (1<<MD_DISK_SYNC);
5712 }
8ddf9efe
N		 5713 if (test_bit(WriteMostly, &rdev->flags))
5714 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5715 } else {
5716 info.major = info.minor = 0;
5717 info.raid_disk = -1;
5718 info.state = (1<<MD_DISK_REMOVED);
5719 }
1ca69c4b 5720 rcu_read_unlock();
1da177e4
LT		 5721
5722 if (copy_to_user(arg, &info, sizeof(info)))
5723 return -EFAULT;
5724
5725 return 0;
5726}
5727
f72ffdd6 5728static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
1da177e4
LT		 5729{
5730 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5731 struct md_rdev *rdev;
1da177e4
LT		 5732 dev_t dev = MKDEV(info->major,info->minor);
5733
5734 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5735 return -EOVERFLOW;
5736
5737 if (!mddev->raid_disks) {
5738 int err;
5739 /* expecting a device which has a superblock */
5740 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5741 if (IS_ERR(rdev)) {
f72ffdd6 5742 printk(KERN_WARNING
1da177e4
LT		 5743 "md: md_import_device returned %ld\n",
5744 PTR_ERR(rdev));
5745 return PTR_ERR(rdev);
5746 }
5747 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5748 struct md_rdev *rdev0
5749 = list_entry(mddev->disks.next,
5750 struct md_rdev, same_set);
a9f326eb 5751 err = super_types[mddev->major_version]
1da177e4
LT		 5752 .load_super(rdev, rdev0, mddev->minor_version);
5753 if (err < 0) {
f72ffdd6 5754 printk(KERN_WARNING
1da177e4 5755 "md: %s has different UUID to %s\n",
f72ffdd6 5756 bdevname(rdev->bdev,b),
1da177e4
LT		 5757 bdevname(rdev0->bdev,b2));
5758 export_rdev(rdev);
5759 return -EINVAL;
5760 }
5761 }
5762 err = bind_rdev_to_array(rdev, mddev);
5763 if (err)
5764 export_rdev(rdev);
5765 return err;
5766 }
5767
5768 /*
5769 * add_new_disk can be used once the array is assembled
5770 * to add "hot spares". They must already have a superblock
5771 * written
5772 */
5773 if (mddev->pers) {
5774 int err;
5775 if (!mddev->pers->hot_add_disk) {
f72ffdd6 5776 printk(KERN_WARNING
1da177e4
LT		 5777 "%s: personality does not support diskops!\n",
5778 mdname(mddev));
5779 return -EINVAL;
5780 }
7b1e35f6
N		 5781 if (mddev->persistent)
5782 rdev = md_import_device(dev, mddev->major_version,
5783 mddev->minor_version);
5784 else
5785 rdev = md_import_device(dev, -1, -1);
1da177e4 5786 if (IS_ERR(rdev)) {
f72ffdd6 5787 printk(KERN_WARNING
1da177e4
LT		 5788 "md: md_import_device returned %ld\n",
5789 PTR_ERR(rdev));
5790 return PTR_ERR(rdev);
5791 }
1a855a06 5792 /* set saved_raid_disk if appropriate */
41158c7e
N		 5793 if (!mddev->persistent) {
5794 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5795 info->raid_disk < mddev->raid_disks) {
41158c7e 5796 rdev->raid_disk = info->raid_disk;
bf572541 5797 set_bit(In_sync, &rdev->flags);
8313b8e5 5798 clear_bit(Bitmap_sync, &rdev->flags);
bf572541 5799 } else
41158c7e 5800 rdev->raid_disk = -1;
f466722c 5801 rdev->saved_raid_disk = rdev->raid_disk;
41158c7e
N		 5802 } else
5803 super_types[mddev->major_version].
5804 validate_super(mddev, rdev);
bedd86b7 5805 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 5806 rdev->raid_disk != info->raid_disk) {
bedd86b7
N		 5807 /* This was a hot-add request, but events doesn't
5808 * match, so reject it.
5809 */
5810 export_rdev(rdev);
5811 return -EINVAL;
5812 }
5813
b2d444d7 5814 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5815 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5816 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5817 else
5818 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5819
1da177e4
LT		 5820 rdev->raid_disk = -1;
5821 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5822 if (!err && !mddev->pers->hot_remove_disk) {
5823 /* If there is hot_add_disk but no hot_remove_disk
5824 * then added disks for geometry changes,
5825 * and should be added immediately.
5826 */
5827 super_types[mddev->major_version].
5828 validate_super(mddev, rdev);
5829 err = mddev->pers->hot_add_disk(mddev, rdev);
5830 if (err)
5831 unbind_rdev_from_array(rdev);
5832 }
1da177e4
LT		 5833 if (err)
5834 export_rdev(rdev);
52664732 5835 else
00bcb4ac 5836 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5837
7ceb17e8 5838 set_bit(MD_CHANGE_DEVS, &mddev->flags);
72a23c21
NB		 5839 if (mddev->degraded)
5840 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5841 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5842 if (!err)
5843 md_new_event(mddev);
005eca5e 5844 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5845 return err;
5846 }
5847
5848 /* otherwise, add_new_disk is only allowed
5849 * for major_version==0 superblocks
5850 */
5851 if (mddev->major_version != 0) {
5852 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5853 mdname(mddev));
5854 return -EINVAL;
5855 }
5856
5857 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5858 int err;
d710e138 5859 rdev = md_import_device(dev, -1, 0);
1da177e4 5860 if (IS_ERR(rdev)) {
f72ffdd6 5861 printk(KERN_WARNING
1da177e4
LT		 5862 "md: error, md_import_device() returned %ld\n",
5863 PTR_ERR(rdev));
5864 return PTR_ERR(rdev);
5865 }
5866 rdev->desc_nr = info->number;
5867 if (info->raid_disk < mddev->raid_disks)
5868 rdev->raid_disk = info->raid_disk;
5869 else
5870 rdev->raid_disk = -1;
5871
1da177e4 5872 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5873 if (info->state & (1<<MD_DISK_SYNC))
5874 set_bit(In_sync, &rdev->flags);
1da177e4 5875
8ddf9efe
N		 5876 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5877 set_bit(WriteMostly, &rdev->flags);
5878
1da177e4
LT		 5879 if (!mddev->persistent) {
5880 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5881 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5882 } else
57b2caa3 5883 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5884 rdev->sectors = rdev->sb_start;
1da177e4 5885
2bf071bf
N		 5886 err = bind_rdev_to_array(rdev, mddev);
5887 if (err) {
5888 export_rdev(rdev);
5889 return err;
5890 }
1da177e4
LT		 5891 }
5892
5893 return 0;
5894}
5895
f72ffdd6 5896static int hot_remove_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 5897{
5898 char b[BDEVNAME_SIZE];
3cb03002 5899 struct md_rdev *rdev;
1da177e4 5900
1da177e4
LT		 5901 rdev = find_rdev(mddev, dev);
5902 if (!rdev)
5903 return -ENXIO;
5904
3ea8929d
N		 5905 clear_bit(Blocked, &rdev->flags);
5906 remove_and_add_spares(mddev, rdev);
5907
1da177e4
LT		 5908 if (rdev->raid_disk >= 0)
5909 goto busy;
5910
5911 kick_rdev_from_array(rdev);
850b2b42 5912 md_update_sb(mddev, 1);
d7603b7e 5913 md_new_event(mddev);
1da177e4
LT		 5914
5915 return 0;
5916busy:
fdefa4d8 5917 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5918 bdevname(rdev->bdev,b), mdname(mddev));
5919 return -EBUSY;
5920}
5921
f72ffdd6 5922static int hot_add_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 5923{
5924 char b[BDEVNAME_SIZE];
5925 int err;
3cb03002 5926 struct md_rdev *rdev;
1da177e4
LT		 5927
5928 if (!mddev->pers)
5929 return -ENODEV;
5930
5931 if (mddev->major_version != 0) {
5932 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5933 " version-0 superblocks.\n",
5934 mdname(mddev));
5935 return -EINVAL;
5936 }
5937 if (!mddev->pers->hot_add_disk) {
f72ffdd6 5938 printk(KERN_WARNING
1da177e4
LT		 5939 "%s: personality does not support diskops!\n",
5940 mdname(mddev));
5941 return -EINVAL;
5942 }
5943
d710e138 5944 rdev = md_import_device(dev, -1, 0);
1da177e4 5945 if (IS_ERR(rdev)) {
f72ffdd6 5946 printk(KERN_WARNING
1da177e4
LT		 5947 "md: error, md_import_device() returned %ld\n",
5948 PTR_ERR(rdev));
5949 return -EINVAL;
5950 }
5951
5952 if (mddev->persistent)
57b2caa3 5953 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5954 else
77304d2a 5955 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5956
8190e754 5957 rdev->sectors = rdev->sb_start;
1da177e4 5958
b2d444d7 5959 if (test_bit(Faulty, &rdev->flags)) {
f72ffdd6 5960 printk(KERN_WARNING
1da177e4
LT		 5961 "md: can not hot-add faulty %s disk to %s!\n",
5962 bdevname(rdev->bdev,b), mdname(mddev));
5963 err = -EINVAL;
5964 goto abort_export;
5965 }
b2d444d7 5966 clear_bit(In_sync, &rdev->flags);
1da177e4 5967 rdev->desc_nr = -1;
5842730d 5968 rdev->saved_raid_disk = -1;
2bf071bf
N		 5969 err = bind_rdev_to_array(rdev, mddev);
5970 if (err)
5971 goto abort_export;
1da177e4
LT		 5972
5973 /*
5974 * The rest should better be atomic, we can have disk failures
5975 * noticed in interrupt contexts ...
5976 */
5977
1da177e4
LT		 5978 rdev->raid_disk = -1;
5979
850b2b42 5980 md_update_sb(mddev, 1);
1da177e4
LT		 5981
5982 /*
5983 * Kick recovery, maybe this spare has to be added to the
5984 * array immediately.
5985 */
5986 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5987 md_wakeup_thread(mddev->thread);
d7603b7e 5988 md_new_event(mddev);
1da177e4
LT		 5989 return 0;
5990
1da177e4
LT		 5991abort_export:
5992 export_rdev(rdev);
5993 return err;
5994}
5995
fd01b88c 5996static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c 5997{
035328c2 5998 int err = 0;
32a7627c 5999
36fa3063 6000 if (mddev->pers) {
d66b1b39 6001 if (!mddev->pers->quiesce || !mddev->thread)
36fa3063
N		 6002 return -EBUSY;
6003 if (mddev->recovery || mddev->sync_thread)
6004 return -EBUSY;
6005 /* we should be able to change the bitmap.. */
6006 }
32a7627c 6007
36fa3063 6008 if (fd >= 0) {
035328c2 6009 struct inode *inode;
1e594bb2
N		 6010 struct file *f;
6011
6012 if (mddev->bitmap || mddev->bitmap_info.file)
36fa3063 6013 return -EEXIST; /* cannot add when bitmap is present */
1e594bb2 6014 f = fget(fd);
32a7627c 6015
1e594bb2 6016 if (f == NULL) {
36fa3063
N		 6017 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6018 mdname(mddev));
6019 return -EBADF;
6020 }
6021
1e594bb2 6022 inode = f->f_mapping->host;
035328c2
N		 6023 if (!S_ISREG(inode->i_mode)) {
6024 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6025 mdname(mddev));
6026 err = -EBADF;
1e594bb2 6027 } else if (!(f->f_mode & FMODE_WRITE)) {
035328c2
N		 6028 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6029 mdname(mddev));
6030 err = -EBADF;
6031 } else if (atomic_read(&inode->i_writecount) != 1) {
36fa3063
N		 6032 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6033 mdname(mddev));
035328c2
N		 6034 err = -EBUSY;
6035 }
6036 if (err) {
1e594bb2 6037 fput(f);
36fa3063
N		 6038 return err;
6039 }
1e594bb2 6040 mddev->bitmap_info.file = f;
c3d9714e 6041 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 6042 } else if (mddev->bitmap == NULL)
6043 return -ENOENT; /* cannot remove what isn't there */
6044 err = 0;
6045 if (mddev->pers) {
6046 mddev->pers->quiesce(mddev, 1);
69e51b44 6047 if (fd >= 0) {
f9209a32
GR		 6048 struct bitmap *bitmap;
6049
6050 bitmap = bitmap_create(mddev, -1);
6051 if (!IS_ERR(bitmap)) {
6052 mddev->bitmap = bitmap;
69e51b44 6053 err = bitmap_load(mddev);
f9209a32 6054 }
69e51b44 6055 }
d7375ab3 6056 if (fd < 0 || err) {
36fa3063 6057 bitmap_destroy(mddev);
d7375ab3
N		 6058 fd = -1; /* make sure to put the file */
6059 }
36fa3063 6060 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 6061 }
6062 if (fd < 0) {
4af1a041
N		 6063 struct file *f = mddev->bitmap_info.file;
6064 if (f) {
6065 spin_lock(&mddev->lock);
6066 mddev->bitmap_info.file = NULL;
6067 spin_unlock(&mddev->lock);
6068 fput(f);
6069 }
36fa3063
N		 6070 }
6071
32a7627c
N		 6072 return err;
6073}
6074
1da177e4
LT		 6075/*
6076 * set_array_info is used two different ways
6077 * The original usage is when creating a new array.
6078 * In this usage, raid_disks is > 0 and it together with
6079 * level, size, not_persistent,layout,chunksize determine the
6080 * shape of the array.
6081 * This will always create an array with a type-0.90.0 superblock.
6082 * The newer usage is when assembling an array.
6083 * In this case raid_disks will be 0, and the major_version field is
6084 * use to determine which style super-blocks are to be found on the devices.
6085 * The minor and patch _version numbers are also kept incase the
6086 * super_block handler wishes to interpret them.
6087 */
f72ffdd6 6088static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6089{
6090
6091 if (info->raid_disks == 0) {
6092 /* just setting version number for superblock loading */
6093 if (info->major_version < 0 ||
50511da3 6094 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 6095 super_types[info->major_version].name == NULL) {
6096 /* maybe try to auto-load a module? */
f72ffdd6 6097 printk(KERN_INFO
1da177e4
LT		 6098 "md: superblock version %d not known\n",
6099 info->major_version);
6100 return -EINVAL;
6101 }
6102 mddev->major_version = info->major_version;
6103 mddev->minor_version = info->minor_version;
6104 mddev->patch_version = info->patch_version;
3f9d7b0d 6105 mddev->persistent = !info->not_persistent;
cbd19983
N		 6106 /* ensure mddev_put doesn't delete this now that there
6107 * is some minimal configuration.
6108 */
6109 mddev->ctime = get_seconds();
1da177e4
LT		 6110 return 0;
6111 }
6112 mddev->major_version = MD_MAJOR_VERSION;
6113 mddev->minor_version = MD_MINOR_VERSION;
6114 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6115 mddev->ctime = get_seconds();
6116
6117 mddev->level = info->level;
17115e03 6118 mddev->clevel[0] = 0;
58c0fed4 6119 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 6120 mddev->raid_disks = info->raid_disks;
6121 /* don't set md_minor, it is determined by which /dev/md* was
6122 * openned
6123 */
6124 if (info->state & (1<<MD_SB_CLEAN))
6125 mddev->recovery_cp = MaxSector;
6126 else
6127 mddev->recovery_cp = 0;
6128 mddev->persistent = ! info->not_persistent;
e691063a 6129 mddev->external = 0;
1da177e4
LT		 6130
6131 mddev->layout = info->layout;
9d8f0363 6132 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 6133
6134 mddev->max_disks = MD_SB_DISKS;
6135
e691063a
N		 6136 if (mddev->persistent)
6137 mddev->flags = 0;
850b2b42 6138 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6139
c3d9714e 6140 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6141 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6142 mddev->bitmap_info.offset = 0;
b2a2703c 6143
f6705578
N		 6144 mddev->reshape_position = MaxSector;
6145
1da177e4
LT		 6146 /*
6147 * Generate a 128 bit UUID
6148 */
6149 get_random_bytes(mddev->uuid, 16);
6150
f6705578 6151 mddev->new_level = mddev->level;
664e7c41 6152 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 6153 mddev->new_layout = mddev->layout;
6154 mddev->delta_disks = 0;
2c810cdd 6155 mddev->reshape_backwards = 0;
f6705578 6156
1da177e4
LT		 6157 return 0;
6158}
6159
fd01b88c 6160void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6161{
b522adcd
DW		 6162 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6163
6164 if (mddev->external_size)
6165 return;
6166
1f403624
DW		 6167 mddev->array_sectors = array_sectors;
6168}
6169EXPORT_SYMBOL(md_set_array_sectors);
6170
fd01b88c 6171static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6172{
3cb03002 6173 struct md_rdev *rdev;
a35b0d69 6174 int rv;
d71f9f88 6175 int fit = (num_sectors == 0);
a35b0d69
N		 6176
6177 if (mddev->pers->resize == NULL)
6178 return -EINVAL;
d71f9f88
AN		 6179 /* The "num_sectors" is the number of sectors of each device that
6180 * is used. This can only make sense for arrays with redundancy.
6181 * linear and raid0 always use whatever space is available. We can only
6182 * consider changing this number if no resync or reconstruction is
6183 * happening, and if the new size is acceptable. It must fit before the
0f420358 6184 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 6185 * of each device. If num_sectors is zero, we find the largest size
6186 * that fits.
a35b0d69 6187 */
f851b60d
N		 6188 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6189 mddev->sync_thread)
a35b0d69 6190 return -EBUSY;
bd8839e0
N		 6191 if (mddev->ro)
6192 return -EROFS;
a4a6125a 6193
dafb20fa 6194 rdev_for_each(rdev, mddev) {
dd8ac336 6195 sector_t avail = rdev->sectors;
01ab5662 6196
d71f9f88
AN		 6197 if (fit && (num_sectors == 0 || num_sectors > avail))
6198 num_sectors = avail;
6199 if (avail < num_sectors)
a35b0d69
N		 6200 return -ENOSPC;
6201 }
d71f9f88 6202 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 6203 if (!rv)
6204 revalidate_disk(mddev->gendisk);
a35b0d69
N		 6205 return rv;
6206}
6207
fd01b88c 6208static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 6209{
6210 int rv;
c6563a8c 6211 struct md_rdev *rdev;
da943b99 6212 /* change the number of raid disks */
63c70c4f 6213 if (mddev->pers->check_reshape == NULL)
da943b99 6214 return -EINVAL;
bd8839e0
N		 6215 if (mddev->ro)
6216 return -EROFS;
da943b99 6217 if (raid_disks <= 0 ||
233fca36 6218 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 6219 return -EINVAL;
f851b60d
N		 6220 if (mddev->sync_thread ||
6221 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6222 mddev->reshape_position != MaxSector)
da943b99 6223 return -EBUSY;
c6563a8c
N		 6224
6225 rdev_for_each(rdev, mddev) {
6226 if (mddev->raid_disks < raid_disks &&
6227 rdev->data_offset < rdev->new_data_offset)
6228 return -EINVAL;
6229 if (mddev->raid_disks > raid_disks &&
6230 rdev->data_offset > rdev->new_data_offset)
6231 return -EINVAL;
6232 }
6233
63c70c4f 6234 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N		 6235 if (mddev->delta_disks < 0)
6236 mddev->reshape_backwards = 1;
6237 else if (mddev->delta_disks > 0)
6238 mddev->reshape_backwards = 0;
63c70c4f
N		 6239
6240 rv = mddev->pers->check_reshape(mddev);
2c810cdd 6241 if (rv < 0) {
de171cb9 6242 mddev->delta_disks = 0;
2c810cdd
N		 6243 mddev->reshape_backwards = 0;
6244 }
da943b99
N		 6245 return rv;
6246}
6247
1da177e4
LT		 6248/*
6249 * update_array_info is used to change the configuration of an
6250 * on-line array.
6251 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6252 * fields in the info are checked against the array.
6253 * Any differences that cannot be handled will cause an error.
6254 * Normally, only one change can be managed at a time.
6255 */
fd01b88c 6256static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6257{
6258 int rv = 0;
6259 int cnt = 0;
36fa3063
N		 6260 int state = 0;
6261
6262 /* calculate expected state,ignoring low bits */
c3d9714e 6263 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 6264 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 6265
6266 if (mddev->major_version != info->major_version ||
6267 mddev->minor_version != info->minor_version ||
6268/* mddev->patch_version != info->patch_version || */
6269 mddev->ctime != info->ctime ||
6270 mddev->level != info->level ||
6271/* mddev->layout != info->layout || */
6272 !mddev->persistent != info->not_persistent||
9d8f0363 6273 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 6274 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6275 ((state^info->state) & 0xfffffe00)
6276 )
1da177e4
LT		 6277 return -EINVAL;
6278 /* Check there is only one change */
58c0fed4
AN		 6279 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6280 cnt++;
6281 if (mddev->raid_disks != info->raid_disks)
6282 cnt++;
6283 if (mddev->layout != info->layout)
6284 cnt++;
6285 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6286 cnt++;
6287 if (cnt == 0)
6288 return 0;
6289 if (cnt > 1)
6290 return -EINVAL;
1da177e4
LT		 6291
6292 if (mddev->layout != info->layout) {
6293 /* Change layout
6294 * we don't need to do anything at the md level, the
6295 * personality will take care of it all.
6296 */
50ac168a 6297 if (mddev->pers->check_reshape == NULL)
1da177e4 6298 return -EINVAL;
597a711b
N		 6299 else {
6300 mddev->new_layout = info->layout;
50ac168a 6301 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 6302 if (rv)
6303 mddev->new_layout = mddev->layout;
6304 return rv;
6305 }
1da177e4 6306 }
58c0fed4 6307 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6308 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6309
da943b99
N		 6310 if (mddev->raid_disks != info->raid_disks)
6311 rv = update_raid_disks(mddev, info->raid_disks);
6312
36fa3063 6313 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
d66b1b39 6314 if (mddev->pers->quiesce == NULL || mddev->thread == NULL)
36fa3063
N		 6315 return -EINVAL;
6316 if (mddev->recovery || mddev->sync_thread)
6317 return -EBUSY;
6318 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
f9209a32 6319 struct bitmap *bitmap;
36fa3063
N		 6320 /* add the bitmap */
6321 if (mddev->bitmap)
6322 return -EEXIST;
c3d9714e 6323 if (mddev->bitmap_info.default_offset == 0)
36fa3063 6324 return -EINVAL;
c3d9714e
N		 6325 mddev->bitmap_info.offset =
6326 mddev->bitmap_info.default_offset;
6409bb05
N		 6327 mddev->bitmap_info.space =
6328 mddev->bitmap_info.default_space;
36fa3063 6329 mddev->pers->quiesce(mddev, 1);
f9209a32
GR		 6330 bitmap = bitmap_create(mddev, -1);
6331 if (!IS_ERR(bitmap)) {
6332 mddev->bitmap = bitmap;
69e51b44 6333 rv = bitmap_load(mddev);
f9209a32 6334 }
36fa3063
N		 6335 if (rv)
6336 bitmap_destroy(mddev);
6337 mddev->pers->quiesce(mddev, 0);
6338 } else {
6339 /* remove the bitmap */
6340 if (!mddev->bitmap)
6341 return -ENOENT;
1ec885cd 6342 if (mddev->bitmap->storage.file)
36fa3063
N		 6343 return -EINVAL;
6344 mddev->pers->quiesce(mddev, 1);
6345 bitmap_destroy(mddev);
6346 mddev->pers->quiesce(mddev, 0);
c3d9714e 6347 mddev->bitmap_info.offset = 0;
36fa3063
N		 6348 }
6349 }
850b2b42 6350 md_update_sb(mddev, 1);
1da177e4
LT		 6351 return rv;
6352}
6353
fd01b88c 6354static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6355{
3cb03002 6356 struct md_rdev *rdev;
1ca69c4b 6357 int err = 0;
1da177e4
LT		 6358
6359 if (mddev->pers == NULL)
6360 return -ENODEV;
6361
1ca69c4b
N		 6362 rcu_read_lock();
6363 rdev = find_rdev_rcu(mddev, dev);
1da177e4 6364 if (!rdev)
1ca69c4b
N		 6365 err = -ENODEV;
6366 else {
6367 md_error(mddev, rdev);
6368 if (!test_bit(Faulty, &rdev->flags))
6369 err = -EBUSY;
6370 }
6371 rcu_read_unlock();
6372 return err;
1da177e4
LT		 6373}
6374
2f9618ce
AN		 6375/*
6376 * We have a problem here : there is no easy way to give a CHS
6377 * virtual geometry. We currently pretend that we have a 2 heads
6378 * 4 sectors (with a BIG number of cylinders...). This drives
6379 * dosfs just mad... ;-)
6380 */
a885c8c4
CH		 6381static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6382{
fd01b88c 6383 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6384
6385 geo->heads = 2;
6386 geo->sectors = 4;
49ce6cea 6387 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6388 return 0;
6389}
6390
cb335f88
NS		 6391static inline bool md_ioctl_valid(unsigned int cmd)
6392{
6393 switch (cmd) {
6394 case ADD_NEW_DISK:
6395 case BLKROSET:
6396 case GET_ARRAY_INFO:
6397 case GET_BITMAP_FILE:
6398 case GET_DISK_INFO:
6399 case HOT_ADD_DISK:
6400 case HOT_REMOVE_DISK:
cb335f88
NS		 6401 case RAID_AUTORUN:
6402 case RAID_VERSION:
6403 case RESTART_ARRAY_RW:
6404 case RUN_ARRAY:
6405 case SET_ARRAY_INFO:
6406 case SET_BITMAP_FILE:
6407 case SET_DISK_FAULTY:
6408 case STOP_ARRAY:
6409 case STOP_ARRAY_RO:
6410 return true;
6411 default:
6412 return false;
6413 }
6414}
6415
a39907fa 6416static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6417 unsigned int cmd, unsigned long arg)
6418{
6419 int err = 0;
6420 void __user *argp = (void __user *)arg;
fd01b88c 6421 struct mddev *mddev = NULL;
e2218350 6422 int ro;
1da177e4 6423
cb335f88
NS		 6424 if (!md_ioctl_valid(cmd))
6425 return -ENOTTY;
6426
506c9e44
N		 6427 switch (cmd) {
6428 case RAID_VERSION:
6429 case GET_ARRAY_INFO:
6430 case GET_DISK_INFO:
6431 break;
6432 default:
6433 if (!capable(CAP_SYS_ADMIN))
6434 return -EACCES;
6435 }
1da177e4
LT		 6436
6437 /*
6438 * Commands dealing with the RAID driver but not any
6439 * particular array:
6440 */
c02c0aeb
N		 6441 switch (cmd) {
6442 case RAID_VERSION:
6443 err = get_version(argp);
3adc28d8 6444 goto out;
1da177e4 6445
1da177e4 6446#ifndef MODULE
c02c0aeb
N		 6447 case RAID_AUTORUN:
6448 err = 0;
6449 autostart_arrays(arg);
3adc28d8 6450 goto out;
1da177e4 6451#endif
c02c0aeb 6452 default:;
1da177e4
LT		 6453 }
6454
6455 /*
6456 * Commands creating/starting a new array:
6457 */
6458
a39907fa 6459 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6460
6461 if (!mddev) {
6462 BUG();
3adc28d8 6463 goto out;
1da177e4
LT		 6464 }
6465
1ca69c4b
N		 6466 /* Some actions do not requires the mutex */
6467 switch (cmd) {
6468 case GET_ARRAY_INFO:
6469 if (!mddev->raid_disks && !mddev->external)
6470 err = -ENODEV;
6471 else
6472 err = get_array_info(mddev, argp);
3adc28d8 6473 goto out;
1ca69c4b
N		 6474
6475 case GET_DISK_INFO:
6476 if (!mddev->raid_disks && !mddev->external)
6477 err = -ENODEV;
6478 else
6479 err = get_disk_info(mddev, argp);
3adc28d8 6480 goto out;
1ca69c4b
N		 6481
6482 case SET_DISK_FAULTY:
6483 err = set_disk_faulty(mddev, new_decode_dev(arg));
3adc28d8 6484 goto out;
4af1a041
N		 6485
6486 case GET_BITMAP_FILE:
6487 err = get_bitmap_file(mddev, argp);
6488 goto out;
6489
1ca69c4b
N		 6490 }
6491
a7a3f08d
N		 6492 if (cmd == ADD_NEW_DISK)
6493 /* need to ensure md_delayed_delete() has completed */
6494 flush_workqueue(md_misc_wq);
6495
90f5f7ad
HR		 6496 if (cmd == HOT_REMOVE_DISK)
6497 /* need to ensure recovery thread has run */
6498 wait_event_interruptible_timeout(mddev->sb_wait,
6499 !test_bit(MD_RECOVERY_NEEDED,
6500 &mddev->flags),
6501 msecs_to_jiffies(5000));
260fa034
N		 6502 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6503 /* Need to flush page cache, and ensure no-one else opens
6504 * and writes
6505 */
6506 mutex_lock(&mddev->open_mutex);
9ba3b7f5 6507 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
260fa034
N		 6508 mutex_unlock(&mddev->open_mutex);
6509 err = -EBUSY;
3adc28d8 6510 goto out;
260fa034
N		 6511 }
6512 set_bit(MD_STILL_CLOSED, &mddev->flags);
6513 mutex_unlock(&mddev->open_mutex);
6514 sync_blockdev(bdev);
6515 }
1da177e4
LT		 6516 err = mddev_lock(mddev);
6517 if (err) {
f72ffdd6 6518 printk(KERN_INFO
1da177e4
LT		 6519 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6520 err, cmd);
3adc28d8 6521 goto out;
1da177e4
LT		 6522 }
6523
c02c0aeb
N		 6524 if (cmd == SET_ARRAY_INFO) {
6525 mdu_array_info_t info;
6526 if (!arg)
6527 memset(&info, 0, sizeof(info));
6528 else if (copy_from_user(&info, argp, sizeof(info))) {
6529 err = -EFAULT;
3adc28d8 6530 goto unlock;
c02c0aeb
N		 6531 }
6532 if (mddev->pers) {
6533 err = update_array_info(mddev, &info);
6534 if (err) {
6535 printk(KERN_WARNING "md: couldn't update"
6536 " array info. %d\n", err);
3adc28d8 6537 goto unlock;
1da177e4 6538 }
3adc28d8 6539 goto unlock;
c02c0aeb
N		 6540 }
6541 if (!list_empty(&mddev->disks)) {
6542 printk(KERN_WARNING
6543 "md: array %s already has disks!\n",
6544 mdname(mddev));
6545 err = -EBUSY;
3adc28d8 6546 goto unlock;
c02c0aeb
N		 6547 }
6548 if (mddev->raid_disks) {
6549 printk(KERN_WARNING
6550 "md: array %s already initialised!\n",
6551 mdname(mddev));
6552 err = -EBUSY;
3adc28d8 6553 goto unlock;
c02c0aeb
N		 6554 }
6555 err = set_array_info(mddev, &info);
6556 if (err) {
6557 printk(KERN_WARNING "md: couldn't set"
6558 " array info. %d\n", err);
3adc28d8 6559 goto unlock;
c02c0aeb 6560 }
3adc28d8 6561 goto unlock;
1da177e4
LT		 6562 }
6563
6564 /*
6565 * Commands querying/configuring an existing array:
6566 */
32a7627c 6567 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6568 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6569 if ((!mddev->raid_disks && !mddev->external)
6570 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6571 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6572 && cmd != GET_BITMAP_FILE) {
1da177e4 6573 err = -ENODEV;
3adc28d8 6574 goto unlock;
1da177e4
LT		 6575 }
6576
6577 /*
6578 * Commands even a read-only array can execute:
6579 */
c02c0aeb 6580 switch (cmd) {
c02c0aeb
N		 6581 case RESTART_ARRAY_RW:
6582 err = restart_array(mddev);
3adc28d8 6583 goto unlock;
1da177e4 6584
c02c0aeb
N		 6585 case STOP_ARRAY:
6586 err = do_md_stop(mddev, 0, bdev);
3adc28d8 6587 goto unlock;
1da177e4 6588
c02c0aeb
N		 6589 case STOP_ARRAY_RO:
6590 err = md_set_readonly(mddev, bdev);
3adc28d8 6591 goto unlock;
1da177e4 6592
3ea8929d
N		 6593 case HOT_REMOVE_DISK:
6594 err = hot_remove_disk(mddev, new_decode_dev(arg));
3adc28d8 6595 goto unlock;
3ea8929d 6596
7ceb17e8
N		 6597 case ADD_NEW_DISK:
6598 /* We can support ADD_NEW_DISK on read-only arrays
6599 * on if we are re-adding a preexisting device.
6600 * So require mddev->pers and MD_DISK_SYNC.
6601 */
6602 if (mddev->pers) {
6603 mdu_disk_info_t info;
6604 if (copy_from_user(&info, argp, sizeof(info)))
6605 err = -EFAULT;
6606 else if (!(info.state & (1<<MD_DISK_SYNC)))
6607 /* Need to clear read-only for this */
6608 break;
6609 else
6610 err = add_new_disk(mddev, &info);
3adc28d8 6611 goto unlock;
7ceb17e8
N		 6612 }
6613 break;
6614
c02c0aeb
N		 6615 case BLKROSET:
6616 if (get_user(ro, (int __user *)(arg))) {
6617 err = -EFAULT;
3adc28d8 6618 goto unlock;
c02c0aeb
N		 6619 }
6620 err = -EINVAL;
e2218350 6621
c02c0aeb
N		 6622 /* if the bdev is going readonly the value of mddev->ro
6623 * does not matter, no writes are coming
6624 */
6625 if (ro)
3adc28d8 6626 goto unlock;
e2218350 6627
c02c0aeb
N		 6628 /* are we are already prepared for writes? */
6629 if (mddev->ro != 1)
3adc28d8 6630 goto unlock;
e2218350 6631
c02c0aeb
N		 6632 /* transitioning to readauto need only happen for
6633 * arrays that call md_write_start
6634 */
6635 if (mddev->pers) {
6636 err = restart_array(mddev);
6637 if (err == 0) {
6638 mddev->ro = 2;
6639 set_disk_ro(mddev->gendisk, 0);
e2218350 6640 }
c02c0aeb 6641 }
3adc28d8 6642 goto unlock;
1da177e4
LT		 6643 }
6644
6645 /*
6646 * The remaining ioctls are changing the state of the
f91de92e 6647 * superblock, so we do not allow them on read-only arrays.
1da177e4 6648 */
326eb17d 6649 if (mddev->ro && mddev->pers) {
f91de92e
N		 6650 if (mddev->ro == 2) {
6651 mddev->ro = 0;
00bcb4ac 6652 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 6653 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N		 6654 /* mddev_unlock will wake thread */
6655 /* If a device failed while we were read-only, we
6656 * need to make sure the metadata is updated now.
6657 */
6658 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6659 mddev_unlock(mddev);
6660 wait_event(mddev->sb_wait,
6661 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6662 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
29f097c4 6663 mddev_lock_nointr(mddev);
f3378b48 6664 }
f91de92e
N		 6665 } else {
6666 err = -EROFS;
3adc28d8 6667 goto unlock;
f91de92e 6668 }
1da177e4
LT		 6669 }
6670
c02c0aeb
N		 6671 switch (cmd) {
6672 case ADD_NEW_DISK:
1da177e4 6673 {
c02c0aeb
N		 6674 mdu_disk_info_t info;
6675 if (copy_from_user(&info, argp, sizeof(info)))
6676 err = -EFAULT;
6677 else
6678 err = add_new_disk(mddev, &info);
3adc28d8 6679 goto unlock;
c02c0aeb 6680 }
1da177e4 6681
c02c0aeb
N		 6682 case HOT_ADD_DISK:
6683 err = hot_add_disk(mddev, new_decode_dev(arg));
3adc28d8 6684 goto unlock;
1da177e4 6685
c02c0aeb
N		 6686 case RUN_ARRAY:
6687 err = do_md_run(mddev);
3adc28d8 6688 goto unlock;
1da177e4 6689
c02c0aeb
N		 6690 case SET_BITMAP_FILE:
6691 err = set_bitmap_file(mddev, (int)arg);
3adc28d8 6692 goto unlock;
32a7627c 6693
c02c0aeb
N		 6694 default:
6695 err = -EINVAL;
3adc28d8 6696 goto unlock;
1da177e4
LT		 6697 }
6698
3adc28d8 6699unlock:
d3374825
N		 6700 if (mddev->hold_active == UNTIL_IOCTL &&
6701 err != -EINVAL)
6702 mddev->hold_active = 0;
1da177e4 6703 mddev_unlock(mddev);
3adc28d8 6704out:
1da177e4
LT		 6705 return err;
6706}
aa98aa31
AB		 6707#ifdef CONFIG_COMPAT
6708static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6709 unsigned int cmd, unsigned long arg)
6710{
6711 switch (cmd) {
6712 case HOT_REMOVE_DISK:
6713 case HOT_ADD_DISK:
6714 case SET_DISK_FAULTY:
6715 case SET_BITMAP_FILE:
6716 /* These take in integer arg, do not convert */
6717 break;
6718 default:
6719 arg = (unsigned long)compat_ptr(arg);
6720 break;
6721 }
6722
6723 return md_ioctl(bdev, mode, cmd, arg);
6724}
6725#endif /* CONFIG_COMPAT */
1da177e4 6726
a39907fa 6727static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6728{
6729 /*
6730 * Succeed if we can lock the mddev, which confirms that
6731 * it isn't being stopped right now.
6732 */
fd01b88c 6733 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6734 int err;
6735
0c098220
YL		 6736 if (!mddev)
6737 return -ENODEV;
6738
d3374825
N		 6739 if (mddev->gendisk != bdev->bd_disk) {
6740 /* we are racing with mddev_put which is discarding this
6741 * bd_disk.
6742 */
6743 mddev_put(mddev);
6744 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6745 flush_workqueue(md_misc_wq);
d3374825
N		 6746 /* Then retry the open from the top */
6747 return -ERESTARTSYS;
6748 }
6749 BUG_ON(mddev != bdev->bd_disk->private_data);
6750
c8c00a69 6751 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6752 goto out;
6753
6754 err = 0;
f2ea68cf 6755 atomic_inc(&mddev->openers);
260fa034 6756 clear_bit(MD_STILL_CLOSED, &mddev->flags);
c8c00a69 6757 mutex_unlock(&mddev->open_mutex);
1da177e4 6758
f0b4f7e2 6759 check_disk_change(bdev);
1da177e4
LT		 6760 out:
6761 return err;
6762}
6763
db2a144b 6764static void md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6765{
f72ffdd6 6766 struct mddev *mddev = disk->private_data;
1da177e4 6767
52e5f9d1 6768 BUG_ON(!mddev);
f2ea68cf 6769 atomic_dec(&mddev->openers);
1da177e4 6770 mddev_put(mddev);
1da177e4 6771}
f0b4f7e2
N		 6772
6773static int md_media_changed(struct gendisk *disk)
6774{
fd01b88c 6775 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6776
6777 return mddev->changed;
6778}
6779
6780static int md_revalidate(struct gendisk *disk)
6781{
fd01b88c 6782 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6783
6784 mddev->changed = 0;
6785 return 0;
6786}
83d5cde4 6787static const struct block_device_operations md_fops =
1da177e4
LT		 6788{
6789 .owner = THIS_MODULE,
a39907fa
AV		 6790 .open = md_open,
6791 .release = md_release,
b492b852 6792 .ioctl = md_ioctl,
aa98aa31
AB		 6793#ifdef CONFIG_COMPAT
6794 .compat_ioctl = md_compat_ioctl,
6795#endif
a885c8c4 6796 .getgeo = md_getgeo,
f0b4f7e2
N		 6797 .media_changed = md_media_changed,
6798 .revalidate_disk= md_revalidate,
1da177e4
LT		 6799};
6800
f72ffdd6 6801static int md_thread(void *arg)
1da177e4 6802{
2b8bf345 6803 struct md_thread *thread = arg;
1da177e4 6804
1da177e4
LT		 6805 /*
6806 * md_thread is a 'system-thread', it's priority should be very
6807 * high. We avoid resource deadlocks individually in each
6808 * raid personality. (RAID5 does preallocation) We also use RR and
6809 * the very same RT priority as kswapd, thus we will never get
6810 * into a priority inversion deadlock.
6811 *
6812 * we definitely have to have equal or higher priority than
6813 * bdflush, otherwise bdflush will deadlock if there are too
6814 * many dirty RAID5 blocks.
6815 */
1da177e4 6816
6985c43f 6817 allow_signal(SIGKILL);
a6fb0934 6818 while (!kthread_should_stop()) {
1da177e4 6819
93588e22
N		 6820 /* We need to wait INTERRUPTIBLE so that
6821 * we don't add to the load-average.
6822 * That means we need to be sure no signals are
6823 * pending
6824 */
6825 if (signal_pending(current))
6826 flush_signals(current);
6827
6828 wait_event_interruptible_timeout
6829 (thread->wqueue,
6830 test_bit(THREAD_WAKEUP, &thread->flags)
6831 || kthread_should_stop(),
6832 thread->timeout);
1da177e4 6833
6c987910
N		 6834 clear_bit(THREAD_WAKEUP, &thread->flags);
6835 if (!kthread_should_stop())
4ed8731d 6836 thread->run(thread);
1da177e4 6837 }
a6fb0934 6838
1da177e4
LT		 6839 return 0;
6840}
6841
2b8bf345 6842void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6843{
6844 if (thread) {
36a4e1fe 6845 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6846 set_bit(THREAD_WAKEUP, &thread->flags);
6847 wake_up(&thread->wqueue);
6848 }
6849}
6c144d31 6850EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 6851
4ed8731d
SL		 6852struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6853 struct mddev *mddev, const char *name)
1da177e4 6854{
2b8bf345 6855 struct md_thread *thread;
1da177e4 6856
2b8bf345 6857 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6858 if (!thread)
6859 return NULL;
6860
1da177e4
LT		 6861 init_waitqueue_head(&thread->wqueue);
6862
1da177e4
LT		 6863 thread->run = run;
6864 thread->mddev = mddev;
32a7627c 6865 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6866 thread->tsk = kthread_run(md_thread, thread,
6867 "%s_%s",
6868 mdname(thread->mddev),
0232605d 6869 name);
a6fb0934 6870 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6871 kfree(thread);
6872 return NULL;
6873 }
1da177e4
LT		 6874 return thread;
6875}
6c144d31 6876EXPORT_SYMBOL(md_register_thread);
1da177e4 6877
2b8bf345 6878void md_unregister_thread(struct md_thread **threadp)
1da177e4 6879{
2b8bf345 6880 struct md_thread *thread = *threadp;
e0cf8f04
N		 6881 if (!thread)
6882 return;
36a4e1fe 6883 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 6884 /* Locking ensures that mddev_unlock does not wake_up a
6885 * non-existent thread
6886 */
6887 spin_lock(&pers_lock);
6888 *threadp = NULL;
6889 spin_unlock(&pers_lock);
a6fb0934
N		 6890
6891 kthread_stop(thread->tsk);
1da177e4
LT		 6892 kfree(thread);
6893}
6c144d31 6894EXPORT_SYMBOL(md_unregister_thread);
1da177e4 6895
fd01b88c 6896void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 6897{
b2d444d7 6898 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6899 return;
6bfe0b49 6900
de393cde 6901 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 6902 return;
6903 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6904 if (mddev->degraded)
6905 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6906 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6907 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6908 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6909 md_wakeup_thread(mddev->thread);
768a418d 6910 if (mddev->event_work.func)
e804ac78 6911 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6912 md_new_event_inintr(mddev);
1da177e4 6913}
6c144d31 6914EXPORT_SYMBOL(md_error);
1da177e4
LT		 6915
6916/* seq_file implementation /proc/mdstat */
6917
6918static void status_unused(struct seq_file *seq)
6919{
6920 int i = 0;
3cb03002 6921 struct md_rdev *rdev;
1da177e4
LT		 6922
6923 seq_printf(seq, "unused devices: ");
6924
159ec1fc 6925 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6926 char b[BDEVNAME_SIZE];
6927 i++;
6928 seq_printf(seq, "%s ",
6929 bdevname(rdev->bdev,b));
6930 }
6931 if (!i)
6932 seq_printf(seq, "<none>");
6933
6934 seq_printf(seq, "\n");
6935}
6936
f72ffdd6 6937static void status_resync(struct seq_file *seq, struct mddev *mddev)
1da177e4 6938{
dd71cf6b
N		 6939 sector_t max_sectors, resync, res;
6940 unsigned long dt, db;
6941 sector_t rt;
4588b42e
N		 6942 int scale;
6943 unsigned int per_milli;
1da177e4 6944
72f36d59
N		 6945 if (mddev->curr_resync <= 3)
6946 resync = 0;
6947 else
6948 resync = mddev->curr_resync
6949 - atomic_read(&mddev->recovery_active);
1da177e4 6950
c804cdec
N		 6951 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6952 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 6953 max_sectors = mddev->resync_max_sectors;
1da177e4 6954 else
dd71cf6b 6955 max_sectors = mddev->dev_sectors;
1da177e4 6956
403df478 6957 WARN_ON(max_sectors == 0);
4588b42e 6958 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6959 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6960 * u32, as those are the requirements for sector_div.
6961 * Thus 'scale' must be at least 10
6962 */
6963 scale = 10;
6964 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6965 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6966 scale++;
6967 }
6968 res = (resync>>scale)*1000;
dd71cf6b 6969 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6970
6971 per_milli = res;
1da177e4 6972 {
4588b42e 6973 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6974 seq_printf(seq, "[");
6975 for (i = 0; i < x; i++)
6976 seq_printf(seq, "=");
6977 seq_printf(seq, ">");
6978 for (i = 0; i < y; i++)
6979 seq_printf(seq, ".");
6980 seq_printf(seq, "] ");
6981 }
4588b42e 6982 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6983 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6984 "reshape" :
61df9d91
N		 6985 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6986 "check" :
6987 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6988 "resync" : "recovery"))),
6989 per_milli/10, per_milli % 10,
dd71cf6b
N		 6990 (unsigned long long) resync/2,
6991 (unsigned long long) max_sectors/2);
1da177e4
LT		 6992
6993 /*
1da177e4
LT		 6994 * dt: time from mark until now
6995 * db: blocks written from mark until now
6996 * rt: remaining time
dd71cf6b
N		 6997 *
6998 * rt is a sector_t, so could be 32bit or 64bit.
6999 * So we divide before multiply in case it is 32bit and close
7000 * to the limit.
25985edc 7001 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 7002 * near the end of resync when the number of remaining sectors
7003 * is close to 'db'.
7004 * We then divide rt by 32 after multiplying by db to compensate.
7005 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 7006 */
7007 dt = ((jiffies - mddev->resync_mark) / HZ);
7008 if (!dt) dt++;
ff4e8d9a
N		 7009 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7010 - mddev->resync_mark_cnt;
1da177e4 7011
dd71cf6b
N		 7012 rt = max_sectors - resync; /* number of remaining sectors */
7013 sector_div(rt, db/32+1);
7014 rt *= dt;
7015 rt >>= 5;
7016
7017 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7018 ((unsigned long)rt % 60)/6);
1da177e4 7019
ff4e8d9a 7020 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 7021}
7022
7023static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7024{
7025 struct list_head *tmp;
7026 loff_t l = *pos;
fd01b88c 7027 struct mddev *mddev;
1da177e4
LT		 7028
7029 if (l >= 0x10000)
7030 return NULL;
7031 if (!l--)
7032 /* header */
7033 return (void*)1;
7034
7035 spin_lock(&all_mddevs_lock);
7036 list_for_each(tmp,&all_mddevs)
7037 if (!l--) {
fd01b88c 7038 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 7039 mddev_get(mddev);
7040 spin_unlock(&all_mddevs_lock);
7041 return mddev;
7042 }
7043 spin_unlock(&all_mddevs_lock);
7044 if (!l--)
7045 return (void*)2;/* tail */
7046 return NULL;
7047}
7048
7049static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7050{
7051 struct list_head *tmp;
fd01b88c 7052 struct mddev *next_mddev, *mddev = v;
f72ffdd6 7053
1da177e4
LT		 7054 ++*pos;
7055 if (v == (void*)2)
7056 return NULL;
7057
7058 spin_lock(&all_mddevs_lock);
7059 if (v == (void*)1)
7060 tmp = all_mddevs.next;
7061 else
7062 tmp = mddev->all_mddevs.next;
7063 if (tmp != &all_mddevs)
fd01b88c 7064 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 7065 else {
7066 next_mddev = (void*)2;
7067 *pos = 0x10000;
f72ffdd6 7068 }
1da177e4
LT		 7069 spin_unlock(&all_mddevs_lock);
7070
7071 if (v != (void*)1)
7072 mddev_put(mddev);
7073 return next_mddev;
7074
7075}
7076
7077static void md_seq_stop(struct seq_file *seq, void *v)
7078{
fd01b88c 7079 struct mddev *mddev = v;
1da177e4
LT		 7080
7081 if (mddev && v != (void*)1 && v != (void*)2)
7082 mddev_put(mddev);
7083}
7084
7085static int md_seq_show(struct seq_file *seq, void *v)
7086{
fd01b88c 7087 struct mddev *mddev = v;
dd8ac336 7088 sector_t sectors;
3cb03002 7089 struct md_rdev *rdev;
1da177e4
LT		 7090
7091 if (v == (void*)1) {
84fc4b56 7092 struct md_personality *pers;
1da177e4
LT		 7093 seq_printf(seq, "Personalities : ");
7094 spin_lock(&pers_lock);
2604b703
N		 7095 list_for_each_entry(pers, &pers_list, list)
7096 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 7097
7098 spin_unlock(&pers_lock);
7099 seq_printf(seq, "\n");
f1514638 7100 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7101 return 0;
7102 }
7103 if (v == (void*)2) {
7104 status_unused(seq);
7105 return 0;
7106 }
7107
36d091f4 7108 spin_lock(&mddev->lock);
1da177e4
LT		 7109 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7110 seq_printf(seq, "%s : %sactive", mdname(mddev),
7111 mddev->pers ? "" : "in");
7112 if (mddev->pers) {
f91de92e 7113 if (mddev->ro==1)
1da177e4 7114 seq_printf(seq, " (read-only)");
f91de92e 7115 if (mddev->ro==2)
52720ae7 7116 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 7117 seq_printf(seq, " %s", mddev->pers->name);
7118 }
7119
dd8ac336 7120 sectors = 0;
f97fcad3
N		 7121 rcu_read_lock();
7122 rdev_for_each_rcu(rdev, mddev) {
1da177e4
LT		 7123 char b[BDEVNAME_SIZE];
7124 seq_printf(seq, " %s[%d]",
7125 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 7126 if (test_bit(WriteMostly, &rdev->flags))
7127 seq_printf(seq, "(W)");
b2d444d7 7128 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7129 seq_printf(seq, "(F)");
7130 continue;
2d78f8c4
N		 7131 }
7132 if (rdev->raid_disk < 0)
b325a32e 7133 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 7134 if (test_bit(Replacement, &rdev->flags))
7135 seq_printf(seq, "(R)");
dd8ac336 7136 sectors += rdev->sectors;
1da177e4 7137 }
f97fcad3 7138 rcu_read_unlock();
1da177e4
LT		 7139
7140 if (!list_empty(&mddev->disks)) {
7141 if (mddev->pers)
7142 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 7143 (unsigned long long)
7144 mddev->array_sectors / 2);
1da177e4
LT		 7145 else
7146 seq_printf(seq, "\n %llu blocks",
dd8ac336 7147 (unsigned long long)sectors / 2);
1da177e4 7148 }
1cd6bf19
N		 7149 if (mddev->persistent) {
7150 if (mddev->major_version != 0 ||
7151 mddev->minor_version != 90) {
7152 seq_printf(seq," super %d.%d",
7153 mddev->major_version,
7154 mddev->minor_version);
7155 }
e691063a
N		 7156 } else if (mddev->external)
7157 seq_printf(seq, " super external:%s",
7158 mddev->metadata_type);
7159 else
1cd6bf19 7160 seq_printf(seq, " super non-persistent");
1da177e4
LT		 7161
7162 if (mddev->pers) {
d710e138 7163 mddev->pers->status(seq, mddev);
f72ffdd6 7164 seq_printf(seq, "\n ");
8e1b39d6
N		 7165 if (mddev->pers->sync_request) {
7166 if (mddev->curr_resync > 2) {
d710e138 7167 status_resync(seq, mddev);
8e1b39d6 7168 seq_printf(seq, "\n ");
72f36d59 7169 } else if (mddev->curr_resync >= 1)
8e1b39d6
N		 7170 seq_printf(seq, "\tresync=DELAYED\n ");
7171 else if (mddev->recovery_cp < MaxSector)
7172 seq_printf(seq, "\tresync=PENDING\n ");
7173 }
32a7627c
N		 7174 } else
7175 seq_printf(seq, "\n ");
7176
57148964 7177 bitmap_status(seq, mddev->bitmap);
1da177e4
LT		 7178
7179 seq_printf(seq, "\n");
7180 }
36d091f4 7181 spin_unlock(&mddev->lock);
f72ffdd6 7182
1da177e4
LT		 7183 return 0;
7184}
7185
110518bc 7186static const struct seq_operations md_seq_ops = {
1da177e4
LT		 7187 .start = md_seq_start,
7188 .next = md_seq_next,
7189 .stop = md_seq_stop,
7190 .show = md_seq_show,
7191};
7192
7193static int md_seq_open(struct inode *inode, struct file *file)
7194{
f1514638 7195 struct seq_file *seq;
1da177e4
LT		 7196 int error;
7197
7198 error = seq_open(file, &md_seq_ops);
d7603b7e 7199 if (error)
f1514638
KS		 7200 return error;
7201
7202 seq = file->private_data;
7203 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7204 return error;
7205}
7206
e2f23b60 7207static int md_unloading;
d7603b7e
N		 7208static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7209{
f1514638 7210 struct seq_file *seq = filp->private_data;
d7603b7e
N		 7211 int mask;
7212
e2f23b60 7213 if (md_unloading)
7d7e64f2 7214 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
d7603b7e
N		 7215 poll_wait(filp, &md_event_waiters, wait);
7216
7217 /* always allow read */
7218 mask = POLLIN | POLLRDNORM;
7219
f1514638 7220 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 7221 mask |= POLLERR | POLLPRI;
7222 return mask;
7223}
7224
fa027c2a 7225static const struct file_operations md_seq_fops = {
e24650c2 7226 .owner = THIS_MODULE,
1da177e4
LT		 7227 .open = md_seq_open,
7228 .read = seq_read,
7229 .llseek = seq_lseek,
c3f94b40 7230 .release = seq_release_private,
d7603b7e 7231 .poll = mdstat_poll,
1da177e4
LT		 7232};
7233
84fc4b56 7234int register_md_personality(struct md_personality *p)
1da177e4 7235{
50bd3774
CY		 7236 printk(KERN_INFO "md: %s personality registered for level %d\n",
7237 p->name, p->level);
1da177e4 7238 spin_lock(&pers_lock);
2604b703 7239 list_add_tail(&p->list, &pers_list);
1da177e4
LT		 7240 spin_unlock(&pers_lock);
7241 return 0;
7242}
6c144d31 7243EXPORT_SYMBOL(register_md_personality);
1da177e4 7244
84fc4b56 7245int unregister_md_personality(struct md_personality *p)
1da177e4 7246{
2604b703 7247 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7248 spin_lock(&pers_lock);
2604b703 7249 list_del_init(&p->list);
1da177e4
LT		 7250 spin_unlock(&pers_lock);
7251 return 0;
7252}
6c144d31 7253EXPORT_SYMBOL(unregister_md_personality);
1da177e4 7254
edb39c9d
GR		 7255int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module)
7256{
7257 if (md_cluster_ops != NULL)
7258 return -EALREADY;
7259 spin_lock(&pers_lock);
7260 md_cluster_ops = ops;
7261 md_cluster_mod = module;
7262 spin_unlock(&pers_lock);
7263 return 0;
7264}
7265EXPORT_SYMBOL(register_md_cluster_operations);
7266
7267int unregister_md_cluster_operations(void)
7268{
7269 spin_lock(&pers_lock);
7270 md_cluster_ops = NULL;
7271 spin_unlock(&pers_lock);
7272 return 0;
7273}
7274EXPORT_SYMBOL(unregister_md_cluster_operations);
7275
7276int md_setup_cluster(struct mddev *mddev, int nodes)
7277{
7278 int err;
7279
7280 err = request_module("md-cluster");
7281 if (err) {
7282 pr_err("md-cluster module not found.\n");
7283 return err;
7284 }
7285
7286 spin_lock(&pers_lock);
7287 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7288 spin_unlock(&pers_lock);
7289 return -ENOENT;
7290 }
7291 spin_unlock(&pers_lock);
7292
cf921cc1 7293 return md_cluster_ops->join(mddev, nodes);
edb39c9d
GR		 7294}
7295
7296void md_cluster_stop(struct mddev *mddev)
7297{
c4ce867f
GR		 7298 if (!md_cluster_ops)
7299 return;
edb39c9d
GR		 7300 md_cluster_ops->leave(mddev);
7301 module_put(md_cluster_mod);
7302}
7303
fd01b88c 7304static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7305{
f72ffdd6 7306 struct md_rdev *rdev;
1da177e4 7307 int idle;
eea1bf38 7308 int curr_events;
1da177e4
LT		 7309
7310 idle = 1;
4b80991c
N		 7311 rcu_read_lock();
7312 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7313 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 7314 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7315 (int)part_stat_read(&disk->part0, sectors[1]) -
7316 atomic_read(&disk->sync_io);
713f6ab1
N		 7317 /* sync IO will cause sync_io to increase before the disk_stats
7318 * as sync_io is counted when a request starts, and
7319 * disk_stats is counted when it completes.
7320 * So resync activity will cause curr_events to be smaller than
7321 * when there was no such activity.
7322 * non-sync IO will cause disk_stat to increase without
7323 * increasing sync_io so curr_events will (eventually)
7324 * be larger than it was before. Once it becomes
7325 * substantially larger, the test below will cause
7326 * the array to appear non-idle, and resync will slow
7327 * down.
7328 * If there is a lot of outstanding resync activity when
7329 * we set last_event to curr_events, then all that activity
7330 * completing might cause the array to appear non-idle
7331 * and resync will be slowed down even though there might
7332 * not have been non-resync activity. This will only
7333 * happen once though. 'last_events' will soon reflect
7334 * the state where there is little or no outstanding
7335 * resync requests, and further resync activity will
7336 * always make curr_events less than last_events.
c0e48521 7337 *
1da177e4 7338 */
eea1bf38 7339 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 7340 rdev->last_events = curr_events;
7341 idle = 0;
7342 }
7343 }
4b80991c 7344 rcu_read_unlock();
1da177e4
LT		 7345 return idle;
7346}
7347
fd01b88c 7348void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 7349{
7350 /* another "blocks" (512byte) blocks have been synced */
7351 atomic_sub(blocks, &mddev->recovery_active);
7352 wake_up(&mddev->recovery_wait);
7353 if (!ok) {
dfc70645 7354 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 7355 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT		 7356 md_wakeup_thread(mddev->thread);
7357 // stop recovery, signal do_sync ....
7358 }
7359}
6c144d31 7360EXPORT_SYMBOL(md_done_sync);
1da177e4 7361
06d91a5f
N		 7362/* md_write_start(mddev, bi)
7363 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 7364 * in superblock) before writing, schedule a superblock update
7365 * and wait for it to complete.
06d91a5f 7366 */
fd01b88c 7367void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7368{
0fd62b86 7369 int did_change = 0;
06d91a5f 7370 if (bio_data_dir(bi) != WRITE)
3d310eb7 7371 return;
06d91a5f 7372
f91de92e
N		 7373 BUG_ON(mddev->ro == 1);
7374 if (mddev->ro == 2) {
7375 /* need to switch to read/write */
7376 mddev->ro = 0;
7377 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7378 md_wakeup_thread(mddev->thread);
25156198 7379 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7380 did_change = 1;
f91de92e 7381 }
06d91a5f 7382 atomic_inc(&mddev->writes_pending);
31a59e34
N		 7383 if (mddev->safemode == 1)
7384 mddev->safemode = 0;
06d91a5f 7385 if (mddev->in_sync) {
85572d7c 7386 spin_lock(&mddev->lock);
3d310eb7
N		 7387 if (mddev->in_sync) {
7388 mddev->in_sync = 0;
850b2b42 7389 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7390 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7391 md_wakeup_thread(mddev->thread);
0fd62b86 7392 did_change = 1;
3d310eb7 7393 }
85572d7c 7394 spin_unlock(&mddev->lock);
06d91a5f 7395 }
0fd62b86 7396 if (did_change)
00bcb4ac 7397 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7398 wait_event(mddev->sb_wait,
09a44cc1 7399 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4 7400}
6c144d31 7401EXPORT_SYMBOL(md_write_start);
1da177e4 7402
fd01b88c 7403void md_write_end(struct mddev *mddev)
1da177e4
LT		 7404{
7405 if (atomic_dec_and_test(&mddev->writes_pending)) {
7406 if (mddev->safemode == 2)
7407 md_wakeup_thread(mddev->thread);
16f17b39 7408 else if (mddev->safemode_delay)
1da177e4
LT		 7409 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7410 }
7411}
6c144d31 7412EXPORT_SYMBOL(md_write_end);
1da177e4 7413
2a2275d6
N		 7414/* md_allow_write(mddev)
7415 * Calling this ensures that the array is marked 'active' so that writes
7416 * may proceed without blocking. It is important to call this before
7417 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7418 * Must be called with mddev_lock held.
b5470dc5
DW		 7419 *
7420 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7421 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7422 */
fd01b88c 7423int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7424{
7425 if (!mddev->pers)
b5470dc5 7426 return 0;
2a2275d6 7427 if (mddev->ro)
b5470dc5 7428 return 0;
1a0fd497 7429 if (!mddev->pers->sync_request)
b5470dc5 7430 return 0;
2a2275d6 7431
85572d7c 7432 spin_lock(&mddev->lock);
2a2275d6
N		 7433 if (mddev->in_sync) {
7434 mddev->in_sync = 0;
7435 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7436 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7437 if (mddev->safemode_delay &&
7438 mddev->safemode == 0)
7439 mddev->safemode = 1;
85572d7c 7440 spin_unlock(&mddev->lock);
2a2275d6 7441 md_update_sb(mddev, 0);
00bcb4ac 7442 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6 7443 } else
85572d7c 7444 spin_unlock(&mddev->lock);
b5470dc5 7445
070dc6dd 7446 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7447 return -EAGAIN;
7448 else
7449 return 0;
2a2275d6
N		 7450}
7451EXPORT_SYMBOL_GPL(md_allow_write);
7452
1da177e4
LT		 7453#define SYNC_MARKS 10
7454#define SYNC_MARK_STEP (3*HZ)
54f89341 7455#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 7456void md_do_sync(struct md_thread *thread)
1da177e4 7457{
4ed8731d 7458 struct mddev *mddev = thread->mddev;
fd01b88c 7459 struct mddev *mddev2;
1da177e4
LT		 7460 unsigned int currspeed = 0,
7461 window;
ac7e50a3 7462 sector_t max_sectors,j, io_sectors, recovery_done;
1da177e4 7463 unsigned long mark[SYNC_MARKS];
54f89341 7464 unsigned long update_time;
1da177e4
LT		 7465 sector_t mark_cnt[SYNC_MARKS];
7466 int last_mark,m;
7467 struct list_head *tmp;
7468 sector_t last_check;
57afd89f 7469 int skipped = 0;
3cb03002 7470 struct md_rdev *rdev;
c4a39551 7471 char *desc, *action = NULL;
7c2c57c9 7472 struct blk_plug plug;
1da177e4
LT		 7473
7474 /* just incase thread restarts... */
7475 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7476 return;
3991b31e
N		 7477 if (mddev->ro) {/* never try to sync a read-only array */
7478 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5fd6c1dc 7479 return;
3991b31e 7480 }
1da177e4 7481
61df9d91 7482 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
c4a39551 7483 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
61df9d91 7484 desc = "data-check";
c4a39551
JB		 7485 action = "check";
7486 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
61df9d91 7487 desc = "requested-resync";
c4a39551
JB		 7488 action = "repair";
7489 } else
61df9d91
N		 7490 desc = "resync";
7491 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7492 desc = "reshape";
7493 else
7494 desc = "recovery";
7495
c4a39551
JB		 7496 mddev->last_sync_action = action ?: desc;
7497
1da177e4
LT		 7498 /* we overload curr_resync somewhat here.
7499 * 0 == not engaged in resync at all
7500 * 2 == checking that there is no conflict with another sync
7501 * 1 == like 2, but have yielded to allow conflicting resync to
7502 * commense
7503 * other == active in resync - this many blocks
7504 *
7505 * Before starting a resync we must have set curr_resync to
7506 * 2, and then checked that every "conflicting" array has curr_resync
7507 * less than ours. When we find one that is the same or higher
7508 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7509 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7510 * This will mean we have to start checking from the beginning again.
7511 *
7512 */
7513
7514 do {
7515 mddev->curr_resync = 2;
7516
7517 try_again:
404e4b43 7518 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7519 goto skip;
29ac4aa3 7520 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7521 if (mddev2 == mddev)
7522 continue;
90b08710
BS		 7523 if (!mddev->parallel_resync
7524 && mddev2->curr_resync
7525 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7526 DEFINE_WAIT(wq);
7527 if (mddev < mddev2 && mddev->curr_resync == 2) {
7528 /* arbitrarily yield */
7529 mddev->curr_resync = 1;
7530 wake_up(&resync_wait);
7531 }
7532 if (mddev > mddev2 && mddev->curr_resync == 1)
7533 /* no need to wait here, we can wait the next
7534 * time 'round when curr_resync == 2
7535 */
7536 continue;
9744197c
N		 7537 /* We need to wait 'interruptible' so as not to
7538 * contribute to the load average, and not to
7539 * be caught by 'softlockup'
7540 */
7541 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
c91abf5a 7542 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8712e553 7543 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7544 printk(KERN_INFO "md: delaying %s of %s"
7545 " until %s has finished (they"
1da177e4 7546 " share one or more physical units)\n",
61df9d91 7547 desc, mdname(mddev), mdname(mddev2));
1da177e4 7548 mddev_put(mddev2);
9744197c
N		 7549 if (signal_pending(current))
7550 flush_signals(current);
1da177e4
LT		 7551 schedule();
7552 finish_wait(&resync_wait, &wq);
7553 goto try_again;
7554 }
7555 finish_wait(&resync_wait, &wq);
7556 }
7557 }
7558 } while (mddev->curr_resync < 2);
7559
5fd6c1dc 7560 j = 0;
9d88883e 7561 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7562 /* resync follows the size requested by the personality,
57afd89f 7563 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7564 */
7565 max_sectors = mddev->resync_max_sectors;
7f7583d4 7566 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 7567 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7568 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7569 j = mddev->resync_min;
7570 else if (!mddev->bitmap)
5fd6c1dc 7571 j = mddev->recovery_cp;
5e96ee65 7572
ccfcc3c1 7573 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7574 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7575 else {
1da177e4 7576 /* recovery follows the physical size of devices */
58c0fed4 7577 max_sectors = mddev->dev_sectors;
5fd6c1dc 7578 j = MaxSector;
4e59ca7d 7579 rcu_read_lock();
dafb20fa 7580 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc
N		 7581 if (rdev->raid_disk >= 0 &&
7582 !test_bit(Faulty, &rdev->flags) &&
7583 !test_bit(In_sync, &rdev->flags) &&
7584 rdev->recovery_offset < j)
7585 j = rdev->recovery_offset;
4e59ca7d 7586 rcu_read_unlock();
133d4527
N		 7587
7588 /* If there is a bitmap, we need to make sure all
7589 * writes that started before we added a spare
7590 * complete before we start doing a recovery.
7591 * Otherwise the write might complete and (via
7592 * bitmap_endwrite) set a bit in the bitmap after the
7593 * recovery has checked that bit and skipped that
7594 * region.
7595 */
7596 if (mddev->bitmap) {
7597 mddev->pers->quiesce(mddev, 1);
7598 mddev->pers->quiesce(mddev, 0);
7599 }
5fd6c1dc 7600 }
1da177e4 7601
61df9d91
N		 7602 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7603 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7604 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7605 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7606 "(but not more than %d KB/sec) for %s.\n",
7607 speed_max(mddev), desc);
1da177e4 7608
eea1bf38 7609 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7610
57afd89f 7611 io_sectors = 0;
1da177e4
LT		 7612 for (m = 0; m < SYNC_MARKS; m++) {
7613 mark[m] = jiffies;
57afd89f 7614 mark_cnt[m] = io_sectors;
1da177e4
LT		 7615 }
7616 last_mark = 0;
7617 mddev->resync_mark = mark[last_mark];
7618 mddev->resync_mark_cnt = mark_cnt[last_mark];
7619
7620 /*
7621 * Tune reconstruction:
7622 */
7623 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7624 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7625 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7626
7627 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7628 last_check = 0;
7629
7630 if (j>2) {
c91abf5a 7631 printk(KERN_INFO
61df9d91
N		 7632 "md: resuming %s of %s from checkpoint.\n",
7633 desc, mdname(mddev));
1da177e4 7634 mddev->curr_resync = j;
72f36d59
N		 7635 } else
7636 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 7637 mddev->curr_resync_completed = j;
72f36d59
N		 7638 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7639 md_new_event(mddev);
54f89341 7640 update_time = jiffies;
1da177e4 7641
96ae923a
GR		 7642 if (mddev_is_clustered(mddev))
7643 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
7644
7c2c57c9 7645 blk_start_plug(&plug);
1da177e4 7646 while (j < max_sectors) {
57afd89f 7647 sector_t sectors;
1da177e4 7648
57afd89f 7649 skipped = 0;
97e4f42d 7650
7a91ee1f
N		 7651 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7652 ((mddev->curr_resync > mddev->curr_resync_completed &&
7653 (mddev->curr_resync - mddev->curr_resync_completed)
7654 > (max_sectors >> 4)) ||
54f89341 7655 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f
N		 7656 (j - mddev->curr_resync_completed)*2
7657 >= mddev->resync_max - mddev->curr_resync_completed
7658 )) {
97e4f42d 7659 /* time to update curr_resync_completed */
97e4f42d
N		 7660 wait_event(mddev->recovery_wait,
7661 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7662 mddev->curr_resync_completed = j;
35d78c66 7663 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7664 j > mddev->recovery_cp)
7665 mddev->recovery_cp = j;
54f89341 7666 update_time = jiffies;
070dc6dd 7667 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7668 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7669 }
acb180b0 7670
c91abf5a
N		 7671 while (j >= mddev->resync_max &&
7672 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
e62e58a5
N		 7673 /* As this condition is controlled by user-space,
7674 * we can block indefinitely, so use '_interruptible'
7675 * to avoid triggering warnings.
7676 */
7677 flush_signals(current); /* just in case */
7678 wait_event_interruptible(mddev->recovery_wait,
7679 mddev->resync_max > j
c91abf5a
N		 7680 || test_bit(MD_RECOVERY_INTR,
7681 &mddev->recovery));
e62e58a5 7682 }
acb180b0 7683
c91abf5a
N		 7684 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7685 break;
acb180b0 7686
57afd89f 7687 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7688 currspeed < speed_min(mddev));
57afd89f 7689 if (sectors == 0) {
dfc70645 7690 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
c91abf5a 7691 break;
1da177e4 7692 }
57afd89f
N		 7693
7694 if (!skipped) { /* actual IO requested */
7695 io_sectors += sectors;
7696 atomic_add(sectors, &mddev->recovery_active);
7697 }
7698
e875ecea
N		 7699 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7700 break;
7701
1da177e4 7702 j += sectors;
72f36d59
N		 7703 if (j > 2)
7704 mddev->curr_resync = j;
96ae923a
GR		 7705 if (mddev_is_clustered(mddev))
7706 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
ff4e8d9a 7707 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7708 if (last_check == 0)
e875ecea 7709 /* this is the earliest that rebuild will be
d7603b7e
N		 7710 * visible in /proc/mdstat
7711 */
7712 md_new_event(mddev);
57afd89f
N		 7713
7714 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7715 continue;
7716
57afd89f 7717 last_check = io_sectors;
1da177e4
LT		 7718 repeat:
7719 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7720 /* step marks */
7721 int next = (last_mark+1) % SYNC_MARKS;
7722
7723 mddev->resync_mark = mark[next];
7724 mddev->resync_mark_cnt = mark_cnt[next];
7725 mark[next] = jiffies;
57afd89f 7726 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7727 last_mark = next;
7728 }
7729
c91abf5a
N		 7730 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7731 break;
1da177e4
LT		 7732
7733 /*
7734 * this loop exits only if either when we are slower than
7735 * the 'hard' speed limit, or the system was IO-idle for
7736 * a jiffy.
7737 * the system might be non-idle CPU-wise, but we only care
7738 * about not overloading the IO subsystem. (things like an
7739 * e2fsck being done on the RAID array should execute fast)
7740 */
1da177e4
LT		 7741 cond_resched();
7742
ac7e50a3
XN		 7743 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7744 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
57afd89f 7745 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7746
88202a0c
N		 7747 if (currspeed > speed_min(mddev)) {
7748 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7749 !is_mddev_idle(mddev, 0)) {
c0e48521 7750 msleep(500);
1da177e4
LT		 7751 goto repeat;
7752 }
7753 }
7754 }
c91abf5a
N		 7755 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7756 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7757 ? "interrupted" : "done");
1da177e4
LT		 7758 /*
7759 * this also signals 'finished resyncing' to md_stop
7760 */
7c2c57c9 7761 blk_finish_plug(&plug);
1da177e4
LT		 7762 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7763
7764 /* tell personality that we are finished */
57afd89f 7765 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7766
96ae923a
GR		 7767 if (mddev_is_clustered(mddev))
7768 md_cluster_ops->resync_info_update(mddev, 0, 0);
7769
dfc70645 7770 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7771 mddev->curr_resync > 2) {
7772 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7773 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7774 if (mddev->curr_resync >= mddev->recovery_cp) {
7775 printk(KERN_INFO
61df9d91
N		 7776 "md: checkpointing %s of %s.\n",
7777 desc, mdname(mddev));
0a19caab 7778 if (test_bit(MD_RECOVERY_ERROR,
7779 &mddev->recovery))
7780 mddev->recovery_cp =
7781 mddev->curr_resync_completed;
7782 else
7783 mddev->recovery_cp =
7784 mddev->curr_resync;
5fd6c1dc
N		 7785 }
7786 } else
7787 mddev->recovery_cp = MaxSector;
7788 } else {
7789 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7790 mddev->curr_resync = MaxSector;
4e59ca7d 7791 rcu_read_lock();
dafb20fa 7792 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7793 if (rdev->raid_disk >= 0 &&
70fffd0b 7794 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7795 !test_bit(Faulty, &rdev->flags) &&
7796 !test_bit(In_sync, &rdev->flags) &&
7797 rdev->recovery_offset < mddev->curr_resync)
7798 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7799 rcu_read_unlock();
5fd6c1dc 7800 }
1da177e4 7801 }
db91ff55 7802 skip:
17571284 7803 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7804
23da422b 7805 spin_lock(&mddev->lock);
c07b70ad
N		 7806 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7807 /* We completed so min/max setting can be forgotten if used. */
7808 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7809 mddev->resync_min = 0;
7810 mddev->resync_max = MaxSector;
7811 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7812 mddev->resync_min = mddev->curr_resync_completed;
1da177e4 7813 mddev->curr_resync = 0;
23da422b
N		 7814 spin_unlock(&mddev->lock);
7815
1da177e4
LT		 7816 wake_up(&resync_wait);
7817 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7818 md_wakeup_thread(mddev->thread);
c6207277 7819 return;
1da177e4 7820}
29269553 7821EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7822
746d3207
N		 7823static int remove_and_add_spares(struct mddev *mddev,
7824 struct md_rdev *this)
b4c4c7b8 7825{
3cb03002 7826 struct md_rdev *rdev;
b4c4c7b8 7827 int spares = 0;
f2a371c5 7828 int removed = 0;
b4c4c7b8 7829
dafb20fa 7830 rdev_for_each(rdev, mddev)
746d3207
N		 7831 if ((this == NULL || rdev == this) &&
7832 rdev->raid_disk >= 0 &&
6bfe0b49 7833 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7834 (test_bit(Faulty, &rdev->flags) ||
7835 ! test_bit(In_sync, &rdev->flags)) &&
7836 atomic_read(&rdev->nr_pending)==0) {
7837 if (mddev->pers->hot_remove_disk(
b8321b68 7838 mddev, rdev) == 0) {
36fad858 7839 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 7840 rdev->raid_disk = -1;
f2a371c5 7841 removed++;
b4c4c7b8
N		 7842 }
7843 }
90584fc9
JB		 7844 if (removed && mddev->kobj.sd)
7845 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 7846
746d3207
N		 7847 if (this)
7848 goto no_add;
7849
dafb20fa 7850 rdev_for_each(rdev, mddev) {
7bfec5f3
N		 7851 if (rdev->raid_disk >= 0 &&
7852 !test_bit(In_sync, &rdev->flags) &&
7853 !test_bit(Faulty, &rdev->flags))
7854 spares++;
7ceb17e8
N		 7855 if (rdev->raid_disk >= 0)
7856 continue;
7857 if (test_bit(Faulty, &rdev->flags))
7858 continue;
7859 if (mddev->ro &&
8313b8e5
N		 7860 ! (rdev->saved_raid_disk >= 0 &&
7861 !test_bit(Bitmap_sync, &rdev->flags)))
7ceb17e8
N		 7862 continue;
7863
7eb41885
N		 7864 if (rdev->saved_raid_disk < 0)
7865 rdev->recovery_offset = 0;
7ceb17e8
N		 7866 if (mddev->pers->
7867 hot_add_disk(mddev, rdev) == 0) {
7868 if (sysfs_link_rdev(mddev, rdev))
7869 /* failure here is OK */;
7870 spares++;
7871 md_new_event(mddev);
7872 set_bit(MD_CHANGE_DEVS, &mddev->flags);
dfc70645 7873 }
b4c4c7b8 7874 }
746d3207 7875no_add:
6dafab6b
N		 7876 if (removed)
7877 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 7878 return spares;
7879}
7ebc0be7 7880
ac05f256
N		 7881static void md_start_sync(struct work_struct *ws)
7882{
7883 struct mddev *mddev = container_of(ws, struct mddev, del_work);
7884
7885 mddev->sync_thread = md_register_thread(md_do_sync,
7886 mddev,
7887 "resync");
7888 if (!mddev->sync_thread) {
7889 printk(KERN_ERR "%s: could not start resync"
7890 " thread...\n",
7891 mdname(mddev));
7892 /* leave the spares where they are, it shouldn't hurt */
7893 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7894 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7895 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7896 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7897 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 7898 wake_up(&resync_wait);
ac05f256
N		 7899 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7900 &mddev->recovery))
7901 if (mddev->sysfs_action)
7902 sysfs_notify_dirent_safe(mddev->sysfs_action);
7903 } else
7904 md_wakeup_thread(mddev->sync_thread);
7905 sysfs_notify_dirent_safe(mddev->sysfs_action);
7906 md_new_event(mddev);
7907}
7908
1da177e4
LT		 7909/*
7910 * This routine is regularly called by all per-raid-array threads to
7911 * deal with generic issues like resync and super-block update.
7912 * Raid personalities that don't have a thread (linear/raid0) do not
7913 * need this as they never do any recovery or update the superblock.
7914 *
7915 * It does not do any resync itself, but rather "forks" off other threads
7916 * to do that as needed.
7917 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7918 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7919 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7920 * and wakeups up this thread which will reap the thread and finish up.
7921 * This thread also removes any faulty devices (with nr_pending == 0).
7922 *
7923 * The overall approach is:
7924 * 1/ if the superblock needs updating, update it.
7925 * 2/ If a recovery thread is running, don't do anything else.
7926 * 3/ If recovery has finished, clean up, possibly marking spares active.
7927 * 4/ If there are any faulty devices, remove them.
7928 * 5/ If array is degraded, try to add spares devices
7929 * 6/ If array has spares or is not in-sync, start a resync thread.
7930 */
fd01b88c 7931void md_check_recovery(struct mddev *mddev)
1da177e4 7932{
68866e42
JB		 7933 if (mddev->suspended)
7934 return;
7935
5f40402d 7936 if (mddev->bitmap)
aa5cbd10 7937 bitmap_daemon_work(mddev);
1da177e4 7938
fca4d848 7939 if (signal_pending(current)) {
31a59e34 7940 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7941 printk(KERN_INFO "md: %s in immediate safe mode\n",
7942 mdname(mddev));
7943 mddev->safemode = 2;
7944 }
7945 flush_signals(current);
7946 }
7947
c89a8eee
N		 7948 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7949 return;
1da177e4 7950 if ( ! (
142d44c3 7951 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7952 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7953 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7954 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7955 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7956 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7957 ))
7958 return;
fca4d848 7959
df5b89b3 7960 if (mddev_trylock(mddev)) {
b4c4c7b8 7961 int spares = 0;
fca4d848 7962
c89a8eee 7963 if (mddev->ro) {
7ceb17e8
N		 7964 /* On a read-only array we can:
7965 * - remove failed devices
7966 * - add already-in_sync devices if the array itself
7967 * is in-sync.
7968 * As we only add devices that are already in-sync,
7969 * we can activate the spares immediately.
c89a8eee 7970 */
7ceb17e8 7971 remove_and_add_spares(mddev, NULL);
8313b8e5
N		 7972 /* There is no thread, but we need to call
7973 * ->spare_active and clear saved_raid_disk
7974 */
2ac295a5 7975 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8313b8e5
N		 7976 md_reap_sync_thread(mddev);
7977 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
c89a8eee
N		 7978 goto unlock;
7979 }
7980
31a59e34 7981 if (!mddev->external) {
0fd62b86 7982 int did_change = 0;
85572d7c 7983 spin_lock(&mddev->lock);
31a59e34
N		 7984 if (mddev->safemode &&
7985 !atomic_read(&mddev->writes_pending) &&
7986 !mddev->in_sync &&
7987 mddev->recovery_cp == MaxSector) {
7988 mddev->in_sync = 1;
0fd62b86 7989 did_change = 1;
070dc6dd 7990 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7991 }
7992 if (mddev->safemode == 1)
7993 mddev->safemode = 0;
85572d7c 7994 spin_unlock(&mddev->lock);
0fd62b86 7995 if (did_change)
00bcb4ac 7996 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7997 }
fca4d848 7998
7a0a5355 7999 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 8000 md_update_sb(mddev, 0);
06d91a5f 8001
1da177e4
LT		 8002 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8003 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8004 /* resync/recovery still happening */
8005 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8006 goto unlock;
8007 }
8008 if (mddev->sync_thread) {
a91d5ac0 8009 md_reap_sync_thread(mddev);
1da177e4
LT		 8010 goto unlock;
8011 }
72a23c21
NB		 8012 /* Set RUNNING before clearing NEEDED to avoid
8013 * any transients in the value of "sync_action".
8014 */
72f36d59 8015 mddev->curr_resync_completed = 0;
23da422b 8016 spin_lock(&mddev->lock);
72a23c21 8017 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
23da422b 8018 spin_unlock(&mddev->lock);
24dd469d
N		 8019 /* Clear some bits that don't mean anything, but
8020 * might be left set
8021 */
24dd469d
N		 8022 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8023 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 8024
ed209584
N		 8025 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8026 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
ac05f256 8027 goto not_running;
1da177e4
LT		 8028 /* no recovery is running.
8029 * remove any failed drives, then
8030 * add spares if possible.
72f36d59 8031 * Spares are also removed and re-added, to allow
1da177e4
LT		 8032 * the personality to fail the re-add.
8033 */
1da177e4 8034
b4c4c7b8 8035 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 8036 if (mddev->pers->check_reshape == NULL ||
8037 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8 8038 /* Cannot proceed */
ac05f256 8039 goto not_running;
b4c4c7b8 8040 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 8041 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 8042 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N		 8043 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8044 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 8045 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 8046 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8047 } else if (mddev->recovery_cp < MaxSector) {
8048 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 8049 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8050 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8051 /* nothing to be done ... */
ac05f256 8052 goto not_running;
24dd469d 8053
1da177e4 8054 if (mddev->pers->sync_request) {
ef99bf48 8055 if (spares) {
a654b9d8
N		 8056 /* We are adding a device or devices to an array
8057 * which has the bitmap stored on all devices.
8058 * So make sure all bitmap pages get written
8059 */
8060 bitmap_write_all(mddev->bitmap);
8061 }
ac05f256
N		 8062 INIT_WORK(&mddev->del_work, md_start_sync);
8063 queue_work(md_misc_wq, &mddev->del_work);
8064 goto unlock;
1da177e4 8065 }
ac05f256 8066 not_running:
72a23c21
NB		 8067 if (!mddev->sync_thread) {
8068 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8069 wake_up(&resync_wait);
72a23c21
NB		 8070 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8071 &mddev->recovery))
0c3573f1 8072 if (mddev->sysfs_action)
00bcb4ac 8073 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 8074 }
ac05f256
N		 8075 unlock:
8076 wake_up(&mddev->sb_wait);
1da177e4
LT		 8077 mddev_unlock(mddev);
8078 }
8079}
6c144d31 8080EXPORT_SYMBOL(md_check_recovery);
1da177e4 8081
a91d5ac0
JB		 8082void md_reap_sync_thread(struct mddev *mddev)
8083{
8084 struct md_rdev *rdev;
8085
8086 /* resync has finished, collect result */
8087 md_unregister_thread(&mddev->sync_thread);
8088 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8089 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8090 /* success...*/
8091 /* activate any spares */
8092 if (mddev->pers->spare_active(mddev)) {
8093 sysfs_notify(&mddev->kobj, NULL,
8094 "degraded");
8095 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8096 }
8097 }
8098 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8099 mddev->pers->finish_reshape)
8100 mddev->pers->finish_reshape(mddev);
8101
8102 /* If array is no-longer degraded, then any saved_raid_disk
f466722c 8103 * information must be scrapped.
a91d5ac0 8104 */
f466722c
N		 8105 if (!mddev->degraded)
8106 rdev_for_each(rdev, mddev)
a91d5ac0
JB		 8107 rdev->saved_raid_disk = -1;
8108
8109 md_update_sb(mddev, 1);
8110 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8111 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8112 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8113 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8114 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
f851b60d 8115 wake_up(&resync_wait);
a91d5ac0
JB		 8116 /* flag recovery needed just to double check */
8117 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8118 sysfs_notify_dirent_safe(mddev->sysfs_action);
8119 md_new_event(mddev);
8120 if (mddev->event_work.func)
8121 queue_work(md_misc_wq, &mddev->event_work);
8122}
6c144d31 8123EXPORT_SYMBOL(md_reap_sync_thread);
a91d5ac0 8124
fd01b88c 8125void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 8126{
00bcb4ac 8127 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 8128 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 8129 !test_bit(Blocked, &rdev->flags) &&
8130 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 8131 msecs_to_jiffies(5000));
8132 rdev_dec_pending(rdev, mddev);
8133}
8134EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8135
c6563a8c
N		 8136void md_finish_reshape(struct mddev *mddev)
8137{
8138 /* called be personality module when reshape completes. */
8139 struct md_rdev *rdev;
8140
8141 rdev_for_each(rdev, mddev) {
8142 if (rdev->data_offset > rdev->new_data_offset)
8143 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8144 else
8145 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8146 rdev->data_offset = rdev->new_data_offset;
8147 }
8148}
8149EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N		 8150
8151/* Bad block management.
8152 * We can record which blocks on each device are 'bad' and so just
8153 * fail those blocks, or that stripe, rather than the whole device.
8154 * Entries in the bad-block table are 64bits wide. This comprises:
8155 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8156 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8157 * A 'shift' can be set so that larger blocks are tracked and
8158 * consequently larger devices can be covered.
8159 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8160 *
8161 * Locking of the bad-block table uses a seqlock so md_is_badblock
8162 * might need to retry if it is very unlucky.
8163 * We will sometimes want to check for bad blocks in a bi_end_io function,
8164 * so we use the write_seqlock_irq variant.
8165 *
8166 * When looking for a bad block we specify a range and want to
8167 * know if any block in the range is bad. So we binary-search
8168 * to the last range that starts at-or-before the given endpoint,
8169 * (or "before the sector after the target range")
8170 * then see if it ends after the given start.
8171 * We return
8172 * 0 if there are no known bad blocks in the range
8173 * 1 if there are known bad block which are all acknowledged
8174 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8175 * plus the start/length of the first bad section we overlap.
8176 */
8177int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8178 sector_t *first_bad, int *bad_sectors)
8179{
8180 int hi;
ab05613a 8181 int lo;
2230dfe4 8182 u64 *p = bb->page;
ab05613a 8183 int rv;
2230dfe4
N		 8184 sector_t target = s + sectors;
8185 unsigned seq;
8186
8187 if (bb->shift > 0) {
8188 /* round the start down, and the end up */
8189 s >>= bb->shift;
8190 target += (1<<bb->shift) - 1;
8191 target >>= bb->shift;
8192 sectors = target - s;
8193 }
8194 /* 'target' is now the first block after the bad range */
8195
8196retry:
8197 seq = read_seqbegin(&bb->lock);
ab05613a 8198 lo = 0;
8199 rv = 0;
2230dfe4
N		 8200 hi = bb->count;
8201
8202 /* Binary search between lo and hi for 'target'
8203 * i.e. for the last range that starts before 'target'
8204 */
8205 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8206 * are known not to be the last range before target.
8207 * VARIANT: hi-lo is the number of possible
8208 * ranges, and decreases until it reaches 1
8209 */
8210 while (hi - lo > 1) {
8211 int mid = (lo + hi) / 2;
8212 sector_t a = BB_OFFSET(p[mid]);
8213 if (a < target)
8214 /* This could still be the one, earlier ranges
8215 * could not. */
8216 lo = mid;
8217 else
8218 /* This and later ranges are definitely out. */
8219 hi = mid;
8220 }
8221 /* 'lo' might be the last that started before target, but 'hi' isn't */
8222 if (hi > lo) {
8223 /* need to check all range that end after 's' to see if
8224 * any are unacknowledged.
8225 */
8226 while (lo >= 0 &&
8227 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8228 if (BB_OFFSET(p[lo]) < target) {
8229 /* starts before the end, and finishes after
8230 * the start, so they must overlap
8231 */
8232 if (rv != -1 && BB_ACK(p[lo]))
8233 rv = 1;
8234 else
8235 rv = -1;
8236 *first_bad = BB_OFFSET(p[lo]);
8237 *bad_sectors = BB_LEN(p[lo]);
8238 }
8239 lo--;
8240 }
8241 }
8242
8243 if (read_seqretry(&bb->lock, seq))
8244 goto retry;
8245
8246 return rv;
8247}
8248EXPORT_SYMBOL_GPL(md_is_badblock);
8249
8250/*
8251 * Add a range of bad blocks to the table.
8252 * This might extend the table, or might contract it
8253 * if two adjacent ranges can be merged.
8254 * We binary-search to find the 'insertion' point, then
8255 * decide how best to handle it.
8256 */
8257static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8258 int acknowledged)
8259{
8260 u64 *p;
8261 int lo, hi;
8262 int rv = 1;
905b0297 8263 unsigned long flags;
2230dfe4
N		 8264
8265 if (bb->shift < 0)
8266 /* badblocks are disabled */
8267 return 0;
8268
8269 if (bb->shift) {
8270 /* round the start down, and the end up */
8271 sector_t next = s + sectors;
8272 s >>= bb->shift;
8273 next += (1<<bb->shift) - 1;
8274 next >>= bb->shift;
8275 sectors = next - s;
8276 }
8277
905b0297 8278 write_seqlock_irqsave(&bb->lock, flags);
2230dfe4
N		 8279
8280 p = bb->page;
8281 lo = 0;
8282 hi = bb->count;
8283 /* Find the last range that starts at-or-before 's' */
8284 while (hi - lo > 1) {
8285 int mid = (lo + hi) / 2;
8286 sector_t a = BB_OFFSET(p[mid]);
8287 if (a <= s)
8288 lo = mid;
8289 else
8290 hi = mid;
8291 }
8292 if (hi > lo && BB_OFFSET(p[lo]) > s)
8293 hi = lo;
8294
8295 if (hi > lo) {
8296 /* we found a range that might merge with the start
8297 * of our new range
8298 */
8299 sector_t a = BB_OFFSET(p[lo]);
8300 sector_t e = a + BB_LEN(p[lo]);
8301 int ack = BB_ACK(p[lo]);
8302 if (e >= s) {
8303 /* Yes, we can merge with a previous range */
8304 if (s == a && s + sectors >= e)
8305 /* new range covers old */
8306 ack = acknowledged;
8307 else
8308 ack = ack && acknowledged;
8309
8310 if (e < s + sectors)
8311 e = s + sectors;
8312 if (e - a <= BB_MAX_LEN) {
8313 p[lo] = BB_MAKE(a, e-a, ack);
8314 s = e;
8315 } else {
8316 /* does not all fit in one range,
8317 * make p[lo] maximal
8318 */
8319 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8320 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8321 s = a + BB_MAX_LEN;
8322 }
8323 sectors = e - s;
8324 }
8325 }
8326 if (sectors && hi < bb->count) {
8327 /* 'hi' points to the first range that starts after 's'.
8328 * Maybe we can merge with the start of that range */
8329 sector_t a = BB_OFFSET(p[hi]);
8330 sector_t e = a + BB_LEN(p[hi]);
8331 int ack = BB_ACK(p[hi]);
8332 if (a <= s + sectors) {
8333 /* merging is possible */
8334 if (e <= s + sectors) {
8335 /* full overlap */
8336 e = s + sectors;
8337 ack = acknowledged;
8338 } else
8339 ack = ack && acknowledged;
8340
8341 a = s;
8342 if (e - a <= BB_MAX_LEN) {
8343 p[hi] = BB_MAKE(a, e-a, ack);
8344 s = e;
8345 } else {
8346 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8347 s = a + BB_MAX_LEN;
8348 }
8349 sectors = e - s;
8350 lo = hi;
8351 hi++;
8352 }
8353 }
8354 if (sectors == 0 && hi < bb->count) {
8355 /* we might be able to combine lo and hi */
8356 /* Note: 's' is at the end of 'lo' */
8357 sector_t a = BB_OFFSET(p[hi]);
8358 int lolen = BB_LEN(p[lo]);
8359 int hilen = BB_LEN(p[hi]);
8360 int newlen = lolen + hilen - (s - a);
8361 if (s >= a && newlen < BB_MAX_LEN) {
8362 /* yes, we can combine them */
8363 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8364 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8365 memmove(p + hi, p + hi + 1,
8366 (bb->count - hi - 1) * 8);
8367 bb->count--;
8368 }
8369 }
8370 while (sectors) {
8371 /* didn't merge (it all).
8372 * Need to add a range just before 'hi' */
8373 if (bb->count >= MD_MAX_BADBLOCKS) {
8374 /* No room for more */
8375 rv = 0;
8376 break;
8377 } else {
8378 int this_sectors = sectors;
8379 memmove(p + hi + 1, p + hi,
8380 (bb->count - hi) * 8);
8381 bb->count++;
8382
8383 if (this_sectors > BB_MAX_LEN)
8384 this_sectors = BB_MAX_LEN;
8385 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8386 sectors -= this_sectors;
8387 s += this_sectors;
8388 }
8389 }
8390
8391 bb->changed = 1;
de393cde
N		 8392 if (!acknowledged)
8393 bb->unacked_exist = 1;
905b0297 8394 write_sequnlock_irqrestore(&bb->lock, flags);
2230dfe4
N		 8395
8396 return rv;
8397}
8398
3cb03002 8399int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8400 int is_new)
2230dfe4 8401{
c6563a8c
N		 8402 int rv;
8403 if (is_new)
8404 s += rdev->new_data_offset;
8405 else
8406 s += rdev->data_offset;
8407 rv = md_set_badblocks(&rdev->badblocks,
8408 s, sectors, 0);
2230dfe4
N		 8409 if (rv) {
8410 /* Make sure they get written out promptly */
8bd2f0a0 8411 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 8412 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8413 md_wakeup_thread(rdev->mddev->thread);
8414 }
8415 return rv;
8416}
8417EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8418
8419/*
8420 * Remove a range of bad blocks from the table.
8421 * This may involve extending the table if we spilt a region,
8422 * but it must not fail. So if the table becomes full, we just
8423 * drop the remove request.
8424 */
8425static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8426{
8427 u64 *p;
8428 int lo, hi;
8429 sector_t target = s + sectors;
8430 int rv = 0;
8431
8432 if (bb->shift > 0) {
8433 /* When clearing we round the start up and the end down.
8434 * This should not matter as the shift should align with
8435 * the block size and no rounding should ever be needed.
8436 * However it is better the think a block is bad when it
8437 * isn't than to think a block is not bad when it is.
8438 */
8439 s += (1<<bb->shift) - 1;
8440 s >>= bb->shift;
8441 target >>= bb->shift;
8442 sectors = target - s;
8443 }
8444
8445 write_seqlock_irq(&bb->lock);
8446
8447 p = bb->page;
8448 lo = 0;
8449 hi = bb->count;
8450 /* Find the last range that starts before 'target' */
8451 while (hi - lo > 1) {
8452 int mid = (lo + hi) / 2;
8453 sector_t a = BB_OFFSET(p[mid]);
8454 if (a < target)
8455 lo = mid;
8456 else
8457 hi = mid;
8458 }
8459 if (hi > lo) {
8460 /* p[lo] is the last range that could overlap the
8461 * current range. Earlier ranges could also overlap,
8462 * but only this one can overlap the end of the range.
8463 */
8464 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8465 /* Partial overlap, leave the tail of this range */
8466 int ack = BB_ACK(p[lo]);
8467 sector_t a = BB_OFFSET(p[lo]);
8468 sector_t end = a + BB_LEN(p[lo]);
8469
8470 if (a < s) {
8471 /* we need to split this range */
8472 if (bb->count >= MD_MAX_BADBLOCKS) {
8b32bf5e 8473 rv = -ENOSPC;
2230dfe4
N		 8474 goto out;
8475 }
8476 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8477 bb->count++;
8478 p[lo] = BB_MAKE(a, s-a, ack);
8479 lo++;
8480 }
8481 p[lo] = BB_MAKE(target, end - target, ack);
8482 /* there is no longer an overlap */
8483 hi = lo;
8484 lo--;
8485 }
8486 while (lo >= 0 &&
8487 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8488 /* This range does overlap */
8489 if (BB_OFFSET(p[lo]) < s) {
8490 /* Keep the early parts of this range. */
8491 int ack = BB_ACK(p[lo]);
8492 sector_t start = BB_OFFSET(p[lo]);
8493 p[lo] = BB_MAKE(start, s - start, ack);
8494 /* now low doesn't overlap, so.. */
8495 break;
8496 }
8497 lo--;
8498 }
8499 /* 'lo' is strictly before, 'hi' is strictly after,
8500 * anything between needs to be discarded
8501 */
8502 if (hi - lo > 1) {
8503 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8504 bb->count -= (hi - lo - 1);
8505 }
8506 }
8507
8508 bb->changed = 1;
8509out:
8510 write_sequnlock_irq(&bb->lock);
8511 return rv;
8512}
8513
c6563a8c
N		 8514int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8515 int is_new)
2230dfe4 8516{
c6563a8c
N		 8517 if (is_new)
8518 s += rdev->new_data_offset;
8519 else
8520 s += rdev->data_offset;
2230dfe4 8521 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8522 s, sectors);
2230dfe4
N		 8523}
8524EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8525
8526/*
8527 * Acknowledge all bad blocks in a list.
8528 * This only succeeds if ->changed is clear. It is used by
8529 * in-kernel metadata updates
8530 */
8531void md_ack_all_badblocks(struct badblocks *bb)
8532{
8533 if (bb->page == NULL || bb->changed)
8534 /* no point even trying */
8535 return;
8536 write_seqlock_irq(&bb->lock);
8537
ecb178bb 8538 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N		 8539 u64 *p = bb->page;
8540 int i;
8541 for (i = 0; i < bb->count ; i++) {
8542 if (!BB_ACK(p[i])) {
8543 sector_t start = BB_OFFSET(p[i]);
8544 int len = BB_LEN(p[i]);
8545 p[i] = BB_MAKE(start, len, 1);
8546 }
8547 }
de393cde 8548 bb->unacked_exist = 0;
2230dfe4
N		 8549 }
8550 write_sequnlock_irq(&bb->lock);
8551}
8552EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8553
16c791a5
N		 8554/* sysfs access to bad-blocks list.
8555 * We present two files.
8556 * 'bad-blocks' lists sector numbers and lengths of ranges that
8557 * are recorded as bad. The list is truncated to fit within
8558 * the one-page limit of sysfs.
8559 * Writing "sector length" to this file adds an acknowledged
8560 * bad block list.
8561 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8562 * been acknowledged. Writing to this file adds bad blocks
8563 * without acknowledging them. This is largely for testing.
8564 */
8565
8566static ssize_t
8567badblocks_show(struct badblocks *bb, char *page, int unack)
8568{
8569 size_t len;
8570 int i;
8571 u64 *p = bb->page;
8572 unsigned seq;
8573
8574 if (bb->shift < 0)
8575 return 0;
8576
8577retry:
8578 seq = read_seqbegin(&bb->lock);
8579
8580 len = 0;
8581 i = 0;
8582
8583 while (len < PAGE_SIZE && i < bb->count) {
8584 sector_t s = BB_OFFSET(p[i]);
8585 unsigned int length = BB_LEN(p[i]);
8586 int ack = BB_ACK(p[i]);
8587 i++;
8588
8589 if (unack && ack)
8590 continue;
8591
8592 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8593 (unsigned long long)s << bb->shift,
8594 length << bb->shift);
8595 }
de393cde
N		 8596 if (unack && len == 0)
8597 bb->unacked_exist = 0;
16c791a5
N		 8598
8599 if (read_seqretry(&bb->lock, seq))
8600 goto retry;
8601
8602 return len;
8603}
8604
8605#define DO_DEBUG 1
8606
8607static ssize_t
8608badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8609{
8610 unsigned long long sector;
8611 int length;
8612 char newline;
8613#ifdef DO_DEBUG
8614 /* Allow clearing via sysfs *only* for testing/debugging.
8615 * Normally only a successful write may clear a badblock
8616 */
8617 int clear = 0;
8618 if (page[0] == '-') {
8619 clear = 1;
8620 page++;
8621 }
8622#endif /* DO_DEBUG */
8623
8624 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8625 case 3:
8626 if (newline != '\n')
8627 return -EINVAL;
8628 case 2:
8629 if (length <= 0)
8630 return -EINVAL;
8631 break;
8632 default:
8633 return -EINVAL;
8634 }
8635
8636#ifdef DO_DEBUG
8637 if (clear) {
8638 md_clear_badblocks(bb, sector, length);
8639 return len;
8640 }
8641#endif /* DO_DEBUG */
8642 if (md_set_badblocks(bb, sector, length, !unack))
8643 return len;
8644 else
8645 return -ENOSPC;
8646}
8647
75c96f85
AB		 8648static int md_notify_reboot(struct notifier_block *this,
8649 unsigned long code, void *x)
1da177e4
LT		 8650{
8651 struct list_head *tmp;
fd01b88c 8652 struct mddev *mddev;
2dba6a91 8653 int need_delay = 0;
1da177e4 8654
c744a65c
N		 8655 for_each_mddev(mddev, tmp) {
8656 if (mddev_trylock(mddev)) {
30b8aa91
N		 8657 if (mddev->pers)
8658 __md_stop_writes(mddev);
0f62fb22
N		 8659 if (mddev->persistent)
8660 mddev->safemode = 2;
c744a65c 8661 mddev_unlock(mddev);
2dba6a91 8662 }
c744a65c 8663 need_delay = 1;
1da177e4 8664 }
c744a65c
N		 8665 /*
8666 * certain more exotic SCSI devices are known to be
8667 * volatile wrt too early system reboots. While the
8668 * right place to handle this issue is the given
8669 * driver, we do want to have a safe RAID driver ...
8670 */
8671 if (need_delay)
8672 mdelay(1000*1);
8673
1da177e4
LT		 8674 return NOTIFY_DONE;
8675}
8676
75c96f85 8677static struct notifier_block md_notifier = {
1da177e4
LT		 8678 .notifier_call = md_notify_reboot,
8679 .next = NULL,
8680 .priority = INT_MAX, /* before any real devices */
8681};
8682
8683static void md_geninit(void)
8684{
36a4e1fe 8685 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8686
c7705f34 8687 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8688}
8689
75c96f85 8690static int __init md_init(void)
1da177e4 8691{
e804ac78
TH		 8692 int ret = -ENOMEM;
8693
ada609ee 8694 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8695 if (!md_wq)
8696 goto err_wq;
8697
8698 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8699 if (!md_misc_wq)
8700 goto err_misc_wq;
8701
8702 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8703 goto err_md;
8704
8705 if ((ret = register_blkdev(0, "mdp")) < 0)
8706 goto err_mdp;
8707 mdp_major = ret;
8708
af5628f0 8709 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
e8703fe1
N		 8710 md_probe, NULL, NULL);
8711 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8712 md_probe, NULL, NULL);
8713
1da177e4 8714 register_reboot_notifier(&md_notifier);
0b4d4147 8715 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8716
8717 md_geninit();
d710e138 8718 return 0;
1da177e4 8719
e804ac78
TH		 8720err_mdp:
8721 unregister_blkdev(MD_MAJOR, "md");
8722err_md:
8723 destroy_workqueue(md_misc_wq);
8724err_misc_wq:
8725 destroy_workqueue(md_wq);
8726err_wq:
8727 return ret;
8728}
1da177e4
LT		 8729
8730#ifndef MODULE
8731
8732/*
8733 * Searches all registered partitions for autorun RAID arrays
8734 * at boot time.
8735 */
4d936ec1
ME		 8736
8737static LIST_HEAD(all_detected_devices);
8738struct detected_devices_node {
8739 struct list_head list;
8740 dev_t dev;
8741};
1da177e4
LT		 8742
8743void md_autodetect_dev(dev_t dev)
8744{
4d936ec1
ME		 8745 struct detected_devices_node *node_detected_dev;
8746
8747 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8748 if (node_detected_dev) {
8749 node_detected_dev->dev = dev;
8750 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8751 } else {
8752 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8753 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8754 }
1da177e4
LT		 8755}
8756
1da177e4
LT		 8757static void autostart_arrays(int part)
8758{
3cb03002 8759 struct md_rdev *rdev;
4d936ec1
ME		 8760 struct detected_devices_node *node_detected_dev;
8761 dev_t dev;
8762 int i_scanned, i_passed;
1da177e4 8763
4d936ec1
ME		 8764 i_scanned = 0;
8765 i_passed = 0;
1da177e4 8766
4d936ec1 8767 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8768
4d936ec1
ME		 8769 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8770 i_scanned++;
8771 node_detected_dev = list_entry(all_detected_devices.next,
8772 struct detected_devices_node, list);
8773 list_del(&node_detected_dev->list);
8774 dev = node_detected_dev->dev;
8775 kfree(node_detected_dev);
df968c4e 8776 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8777 if (IS_ERR(rdev))
8778 continue;
8779
403df478 8780 if (test_bit(Faulty, &rdev->flags))
1da177e4 8781 continue;
403df478 8782
d0fae18f 8783 set_bit(AutoDetected, &rdev->flags);
1da177e4 8784 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8785 i_passed++;
1da177e4 8786 }
4d936ec1
ME		 8787
8788 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8789 i_scanned, i_passed);
1da177e4
LT		 8790
8791 autorun_devices(part);
8792}
8793
fdee8ae4 8794#endif /* !MODULE */
1da177e4
LT		 8795
8796static __exit void md_exit(void)
8797{
fd01b88c 8798 struct mddev *mddev;
1da177e4 8799 struct list_head *tmp;
e2f23b60 8800 int delay = 1;
8ab5e4c1 8801
af5628f0 8802 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
e8703fe1 8803 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8804
3dbd8c2e 8805 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8806 unregister_blkdev(mdp_major, "mdp");
8807 unregister_reboot_notifier(&md_notifier);
8808 unregister_sysctl_table(raid_table_header);
e2f23b60
N		 8809
8810 /* We cannot unload the modules while some process is
8811 * waiting for us in select() or poll() - wake them up
8812 */
8813 md_unloading = 1;
8814 while (waitqueue_active(&md_event_waiters)) {
8815 /* not safe to leave yet */
8816 wake_up(&md_event_waiters);
8817 msleep(delay);
8818 delay += delay;
8819 }
1da177e4 8820 remove_proc_entry("mdstat", NULL);
e2f23b60 8821
29ac4aa3 8822 for_each_mddev(mddev, tmp) {
1da177e4 8823 export_array(mddev);
d3374825 8824 mddev->hold_active = 0;
1da177e4 8825 }
e804ac78
TH		 8826 destroy_workqueue(md_misc_wq);
8827 destroy_workqueue(md_wq);
1da177e4
LT		 8828}
8829
685784aa 8830subsys_initcall(md_init);
1da177e4
LT		 8831module_exit(md_exit)
8832
f91de92e
N		 8833static int get_ro(char *buffer, struct kernel_param *kp)
8834{
8835 return sprintf(buffer, "%d", start_readonly);
8836}
8837static int set_ro(const char *val, struct kernel_param *kp)
8838{
8839 char *e;
8840 int num = simple_strtoul(val, &e, 10);
8841 if (*val && (*e == '\0' || *e == '\n')) {
8842 start_readonly = num;
4dbcdc75 8843 return 0;
f91de92e
N		 8844 }
8845 return -EINVAL;
8846}
8847
80ca3a44
N		 8848module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8849module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
efeb53c0 8850module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8851
1da177e4 8852MODULE_LICENSE("GPL");
0efb9e61 8853MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8854MODULE_ALIAS("md");
72008652 8855MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)