projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: Skip cluster setup in case of error while reading bitmap
[linux-2.6-block.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
f72ffdd6 3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
1da177e4
LT		 4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
a6fb0934 35#include <linux/kthread.h>
bff61975 36#include <linux/blkdev.h>
1da177e4 37#include <linux/sysctl.h>
bff61975 38#include <linux/seq_file.h>
ff01bb48 39#include <linux/fs.h>
d7603b7e 40#include <linux/poll.h>
16f17b39 41#include <linux/ctype.h>
e7d2860b 42#include <linux/string.h>
fb4d8c76
N		 43#include <linux/hdreg.h>
44#include <linux/proc_fs.h>
45#include <linux/random.h>
056075c7 46#include <linux/module.h>
fb4d8c76 47#include <linux/reboot.h>
32a7627c 48#include <linux/file.h>
aa98aa31 49#include <linux/compat.h>
25570727 50#include <linux/delay.h>
bff61975
N		 51#include <linux/raid/md_p.h>
52#include <linux/raid/md_u.h>
5a0e3ad6 53#include <linux/slab.h>
43b2e5d8 54#include "md.h"
ef740c37 55#include "bitmap.h"
edb39c9d 56#include "md-cluster.h"
1da177e4 57
1da177e4 58#ifndef MODULE
d710e138 59static void autostart_arrays(int part);
1da177e4
LT		 60#endif
61
01f96c0a
N		 62/* pers_list is a list of registered personalities protected
63 * by pers_lock.
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
66 */
2604b703 67static LIST_HEAD(pers_list);
1da177e4
LT		 68static DEFINE_SPINLOCK(pers_lock);
69
edb39c9d 70struct md_cluster_operations *md_cluster_ops;
589a1c49 71EXPORT_SYMBOL(md_cluster_ops);
edb39c9d
GR		 72struct module *md_cluster_mod;
73EXPORT_SYMBOL(md_cluster_mod);
74
90b08710 75static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 76static struct workqueue_struct *md_wq;
77static struct workqueue_struct *md_misc_wq;
90b08710 78
746d3207
N		 79static int remove_and_add_spares(struct mddev *mddev,
80 struct md_rdev *this);
5aa61f42 81static void mddev_detach(struct mddev *mddev);
746d3207 82
1e50915f
RB		 83/*
84 * Default number of read corrections we'll attempt on an rdev
85 * before ejecting it from the array. We divide the read error
86 * count by 2 for every hour elapsed between read errors.
87 */
88#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 89/*
90 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
91 * is 1000 KB/sec, so the extra system load does not show up that much.
92 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 93 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 94 * subsystem is idle. There is also an 'absolute maximum' reconstruction
95 * speed limit - in case reconstruction slows down your system despite
96 * idle IO detection.
97 *
98 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 99 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 100 */
101
102static int sysctl_speed_limit_min = 1000;
103static int sysctl_speed_limit_max = 200000;
fd01b88c 104static inline int speed_min(struct mddev *mddev)
88202a0c
N		 105{
106 return mddev->sync_speed_min ?
107 mddev->sync_speed_min : sysctl_speed_limit_min;
108}
109
fd01b88c 110static inline int speed_max(struct mddev *mddev)
88202a0c
N		 111{
112 return mddev->sync_speed_max ?
113 mddev->sync_speed_max : sysctl_speed_limit_max;
114}
1da177e4
LT		 115
116static struct ctl_table_header *raid_table_header;
117
82592c38 118static struct ctl_table raid_table[] = {
1da177e4 119 {
1da177e4
LT		 120 .procname = "speed_limit_min",
121 .data = &sysctl_speed_limit_min,
122 .maxlen = sizeof(int),
80ca3a44 123 .mode = S_IRUGO|S_IWUSR,
6d456111 124 .proc_handler = proc_dointvec,
1da177e4
LT		 125 },
126 {
1da177e4
LT		 127 .procname = "speed_limit_max",
128 .data = &sysctl_speed_limit_max,
129 .maxlen = sizeof(int),
80ca3a44 130 .mode = S_IRUGO|S_IWUSR,
6d456111 131 .proc_handler = proc_dointvec,
1da177e4 132 },
894d2491 133 { }
1da177e4
LT		 134};
135
82592c38 136static struct ctl_table raid_dir_table[] = {
1da177e4 137 {
1da177e4
LT		 138 .procname = "raid",
139 .maxlen = 0,
80ca3a44 140 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 141 .child = raid_table,
142 },
894d2491 143 { }
1da177e4
LT		 144};
145
82592c38 146static struct ctl_table raid_root_table[] = {
1da177e4 147 {
1da177e4
LT		 148 .procname = "dev",
149 .maxlen = 0,
150 .mode = 0555,
151 .child = raid_dir_table,
152 },
894d2491 153 { }
1da177e4
LT		 154};
155
83d5cde4 156static const struct block_device_operations md_fops;
1da177e4 157
f91de92e
N		 158static int start_readonly;
159
a167f663
N		 160/* bio_clone_mddev
161 * like bio_clone, but with a local bio set
162 */
163
a167f663 164struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 165 struct mddev *mddev)
a167f663
N		 166{
167 struct bio *b;
a167f663
N		 168
169 if (!mddev || !mddev->bio_set)
170 return bio_alloc(gfp_mask, nr_iovecs);
171
395c72a7 172 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
a167f663
N		 173 if (!b)
174 return NULL;
a167f663
N		 175 return b;
176}
177EXPORT_SYMBOL_GPL(bio_alloc_mddev);
178
179struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
fd01b88c 180 struct mddev *mddev)
a167f663 181{
a167f663
N		 182 if (!mddev || !mddev->bio_set)
183 return bio_clone(bio, gfp_mask);
184
bf800ef1 185 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
a167f663
N		 186}
187EXPORT_SYMBOL_GPL(bio_clone_mddev);
188
d7603b7e
N		 189/*
190 * We have a system wide 'event count' that is incremented
191 * on any 'interesting' event, and readers of /proc/mdstat
192 * can use 'poll' or 'select' to find out when the event
193 * count increases.
194 *
195 * Events are:
196 * start array, stop array, error, add device, remove device,
197 * start build, activate spare
198 */
2989ddbd 199static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 200static atomic_t md_event_count;
fd01b88c 201void md_new_event(struct mddev *mddev)
d7603b7e
N		 202{
203 atomic_inc(&md_event_count);
204 wake_up(&md_event_waiters);
205}
29269553 206EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 207
c331eb04
N		 208/* Alternate version that can be called from interrupts
209 * when calling sysfs_notify isn't needed.
210 */
fd01b88c 211static void md_new_event_inintr(struct mddev *mddev)
c331eb04
N		 212{
213 atomic_inc(&md_event_count);
214 wake_up(&md_event_waiters);
215}
216
1da177e4
LT		 217/*
218 * Enables to iterate over all existing md arrays
219 * all_mddevs_lock protects this list.
220 */
221static LIST_HEAD(all_mddevs);
222static DEFINE_SPINLOCK(all_mddevs_lock);
223
1da177e4
LT		 224/*
225 * iterates through all used mddevs in the system.
226 * We take care to grab the all_mddevs_lock whenever navigating
227 * the list, and to always hold a refcount when unlocked.
228 * Any code which breaks out of this loop while own
229 * a reference to the current mddev and must mddev_put it.
230 */
fd01b88c 231#define for_each_mddev(_mddev,_tmp) \
1da177e4 232 \
f72ffdd6 233 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N		 234 _tmp = all_mddevs.next; \
235 _mddev = NULL;}); \
236 ({ if (_tmp != &all_mddevs) \
237 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 238 spin_unlock(&all_mddevs_lock); \
fd01b88c
N		 239 if (_mddev) mddev_put(_mddev); \
240 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
241 _tmp != &all_mddevs;}); \
1da177e4 242 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 243 _tmp = _tmp->next;}) \
1da177e4
LT		 244 )
245
409c57f3
N		 246/* Rather than calling directly into the personality make_request function,
247 * IO requests come here first so that we can check if the device is
248 * being suspended pending a reconfiguration.
249 * We hold a refcount over the call to ->make_request. By the time that
250 * call has finished, the bio has been linked into some internal structure
251 * and so is visible to ->quiesce(), so we don't need the refcount any more.
252 */
5a7bbad2 253static void md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 254{
49077326 255 const int rw = bio_data_dir(bio);
fd01b88c 256 struct mddev *mddev = q->queuedata;
e91ece55 257 unsigned int sectors;
74672d06 258 int cpu;
49077326 259
0ca69886
N		 260 if (mddev == NULL || mddev->pers == NULL
261 || !mddev->ready) {
409c57f3 262 bio_io_error(bio);
5a7bbad2 263 return;
409c57f3 264 }
bbfa57c0
SR		 265 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
266 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
267 return;
268 }
0ca69886 269 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 270 rcu_read_lock();
e9c7469b 271 if (mddev->suspended) {
409c57f3
N		 272 DEFINE_WAIT(__wait);
273 for (;;) {
274 prepare_to_wait(&mddev->sb_wait, &__wait,
275 TASK_UNINTERRUPTIBLE);
e9c7469b 276 if (!mddev->suspended)
409c57f3
N		 277 break;
278 rcu_read_unlock();
279 schedule();
280 rcu_read_lock();
281 }
282 finish_wait(&mddev->sb_wait, &__wait);
283 }
284 atomic_inc(&mddev->active_io);
285 rcu_read_unlock();
49077326 286
e91ece55
CM		 287 /*
288 * save the sectors now since our bio can
289 * go away inside make_request
290 */
291 sectors = bio_sectors(bio);
5a7bbad2 292 mddev->pers->make_request(mddev, bio);
49077326 293
74672d06
GZ		 294 cpu = part_stat_lock();
295 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
296 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
297 part_stat_unlock();
49077326 298
409c57f3
N		 299 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
300 wake_up(&mddev->sb_wait);
409c57f3
N		 301}
302
9e35b99c
N		 303/* mddev_suspend makes sure no new requests are submitted
304 * to the device, and that any requests that have been submitted
305 * are completely handled.
afa0f557
N		 306 * Once mddev_detach() is called and completes, the module will be
307 * completely unused.
9e35b99c 308 */
fd01b88c 309void mddev_suspend(struct mddev *mddev)
409c57f3
N		 310{
311 BUG_ON(mddev->suspended);
312 mddev->suspended = 1;
313 synchronize_rcu();
314 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
315 mddev->pers->quiesce(mddev, 1);
0d9f4f13
JB		 316
317 del_timer_sync(&mddev->safemode_timer);
409c57f3 318}
390ee602 319EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 320
fd01b88c 321void mddev_resume(struct mddev *mddev)
409c57f3
N		 322{
323 mddev->suspended = 0;
324 wake_up(&mddev->sb_wait);
325 mddev->pers->quiesce(mddev, 0);
0fd018af 326
47525e59 327 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
0fd018af
JB		 328 md_wakeup_thread(mddev->thread);
329 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 330}
390ee602 331EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 332
fd01b88c 333int mddev_congested(struct mddev *mddev, int bits)
3fa841d7 334{
5c675f83
N		 335 struct md_personality *pers = mddev->pers;
336 int ret = 0;
337
338 rcu_read_lock();
339 if (mddev->suspended)
340 ret = 1;
341 else if (pers && pers->congested)
342 ret = pers->congested(mddev, bits);
343 rcu_read_unlock();
344 return ret;
345}
346EXPORT_SYMBOL_GPL(mddev_congested);
347static int md_congested(void *data, int bits)
348{
349 struct mddev *mddev = data;
350 return mddev_congested(mddev, bits);
3fa841d7 351}
3fa841d7 352
64590f45
N		 353static int md_mergeable_bvec(struct request_queue *q,
354 struct bvec_merge_data *bvm,
355 struct bio_vec *biovec)
356{
357 struct mddev *mddev = q->queuedata;
358 int ret;
359 rcu_read_lock();
360 if (mddev->suspended) {
361 /* Must always allow one vec */
362 if (bvm->bi_size == 0)
363 ret = biovec->bv_len;
364 else
365 ret = 0;
366 } else {
367 struct md_personality *pers = mddev->pers;
368 if (pers && pers->mergeable_bvec)
369 ret = pers->mergeable_bvec(mddev, bvm, biovec);
370 else
371 ret = biovec->bv_len;
372 }
373 rcu_read_unlock();
374 return ret;
375}
a2826aa9 376/*
e9c7469b 377 * Generic flush handling for md
a2826aa9
N		 378 */
379
e9c7469b 380static void md_end_flush(struct bio *bio, int err)
a2826aa9 381{
3cb03002 382 struct md_rdev *rdev = bio->bi_private;
fd01b88c 383 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 384
385 rdev_dec_pending(rdev, mddev);
386
387 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 388 /* The pre-request flush has finished */
e804ac78 389 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 390 }
391 bio_put(bio);
392}
393
a7a07e69
N		 394static void md_submit_flush_data(struct work_struct *ws);
395
a035fc3e 396static void submit_flushes(struct work_struct *ws)
a2826aa9 397{
fd01b88c 398 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 399 struct md_rdev *rdev;
a2826aa9 400
a7a07e69
N		 401 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
402 atomic_set(&mddev->flush_pending, 1);
a2826aa9 403 rcu_read_lock();
dafb20fa 404 rdev_for_each_rcu(rdev, mddev)
a2826aa9
N		 405 if (rdev->raid_disk >= 0 &&
406 !test_bit(Faulty, &rdev->flags)) {
407 /* Take two references, one is dropped
408 * when request finishes, one after
409 * we reclaim rcu_read_lock
410 */
411 struct bio *bi;
412 atomic_inc(&rdev->nr_pending);
413 atomic_inc(&rdev->nr_pending);
414 rcu_read_unlock();
b5e1b8ce 415 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
e9c7469b 416 bi->bi_end_io = md_end_flush;
a2826aa9
N		 417 bi->bi_private = rdev;
418 bi->bi_bdev = rdev->bdev;
419 atomic_inc(&mddev->flush_pending);
e9c7469b 420 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 421 rcu_read_lock();
422 rdev_dec_pending(rdev, mddev);
423 }
424 rcu_read_unlock();
a7a07e69
N		 425 if (atomic_dec_and_test(&mddev->flush_pending))
426 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 427}
428
e9c7469b 429static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 430{
fd01b88c 431 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 432 struct bio *bio = mddev->flush_bio;
a2826aa9 433
4f024f37 434 if (bio->bi_iter.bi_size == 0)
a2826aa9
N		 435 /* an empty barrier - all done */
436 bio_endio(bio, 0);
437 else {
e9c7469b 438 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 439 mddev->pers->make_request(mddev, bio);
a2826aa9 440 }
2b74e12e
N		 441
442 mddev->flush_bio = NULL;
443 wake_up(&mddev->sb_wait);
a2826aa9
N		 444}
445
fd01b88c 446void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9 447{
85572d7c 448 spin_lock_irq(&mddev->lock);
a2826aa9 449 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 450 !mddev->flush_bio,
85572d7c 451 mddev->lock);
e9c7469b 452 mddev->flush_bio = bio;
85572d7c 453 spin_unlock_irq(&mddev->lock);
a2826aa9 454
a035fc3e
N		 455 INIT_WORK(&mddev->flush_work, submit_flushes);
456 queue_work(md_wq, &mddev->flush_work);
a2826aa9 457}
e9c7469b 458EXPORT_SYMBOL(md_flush_request);
409c57f3 459
74018dc3 460void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
97658cdd 461{
9cbb1750
N		 462 struct mddev *mddev = cb->data;
463 md_wakeup_thread(mddev->thread);
464 kfree(cb);
97658cdd 465}
9cbb1750 466EXPORT_SYMBOL(md_unplug);
2ac87401 467
fd01b88c 468static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 469{
470 atomic_inc(&mddev->active);
471 return mddev;
472}
473
5fd3a17e 474static void mddev_delayed_delete(struct work_struct *ws);
d3374825 475
fd01b88c 476static void mddev_put(struct mddev *mddev)
1da177e4 477{
a167f663
N		 478 struct bio_set *bs = NULL;
479
1da177e4
LT		 480 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
481 return;
d3374825 482 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 483 mddev->ctime == 0 && !mddev->hold_active) {
484 /* Array is not configured at all, and not held active,
485 * so destroy it */
af8a2434 486 list_del_init(&mddev->all_mddevs);
a167f663
N		 487 bs = mddev->bio_set;
488 mddev->bio_set = NULL;
d3374825 489 if (mddev->gendisk) {
e804ac78
TH		 490 /* We did a probe so need to clean up. Call
491 * queue_work inside the spinlock so that
492 * flush_workqueue() after mddev_find will
493 * succeed in waiting for the work to be done.
d3374825
N		 494 */
495 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 496 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 497 } else
498 kfree(mddev);
499 }
500 spin_unlock(&all_mddevs_lock);
a167f663
N		 501 if (bs)
502 bioset_free(bs);
1da177e4
LT		 503}
504
fd01b88c 505void mddev_init(struct mddev *mddev)
fafd7fb0
N		 506{
507 mutex_init(&mddev->open_mutex);
508 mutex_init(&mddev->reconfig_mutex);
509 mutex_init(&mddev->bitmap_info.mutex);
510 INIT_LIST_HEAD(&mddev->disks);
511 INIT_LIST_HEAD(&mddev->all_mddevs);
512 init_timer(&mddev->safemode_timer);
513 atomic_set(&mddev->active, 1);
514 atomic_set(&mddev->openers, 0);
515 atomic_set(&mddev->active_io, 0);
85572d7c 516 spin_lock_init(&mddev->lock);
fafd7fb0
N		 517 atomic_set(&mddev->flush_pending, 0);
518 init_waitqueue_head(&mddev->sb_wait);
519 init_waitqueue_head(&mddev->recovery_wait);
520 mddev->reshape_position = MaxSector;
2c810cdd 521 mddev->reshape_backwards = 0;
c4a39551 522 mddev->last_sync_action = "none";
fafd7fb0
N		 523 mddev->resync_min = 0;
524 mddev->resync_max = MaxSector;
525 mddev->level = LEVEL_NONE;
526}
390ee602 527EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 528
f72ffdd6 529static struct mddev *mddev_find(dev_t unit)
1da177e4 530{
fd01b88c 531 struct mddev *mddev, *new = NULL;
1da177e4 532
8f5f02c4
N		 533 if (unit && MAJOR(unit) != MD_MAJOR)
534 unit &= ~((1<<MdpMinorShift)-1);
535
1da177e4
LT		 536 retry:
537 spin_lock(&all_mddevs_lock);
efeb53c0
N		 538
539 if (unit) {
540 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
541 if (mddev->unit == unit) {
542 mddev_get(mddev);
543 spin_unlock(&all_mddevs_lock);
544 kfree(new);
545 return mddev;
546 }
547
548 if (new) {
549 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 550 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 551 new->hold_active = UNTIL_IOCTL;
552 return new;
1da177e4 553 }
efeb53c0
N		 554 } else if (new) {
555 /* find an unused unit number */
556 static int next_minor = 512;
557 int start = next_minor;
558 int is_free = 0;
559 int dev = 0;
560 while (!is_free) {
561 dev = MKDEV(MD_MAJOR, next_minor);
562 next_minor++;
563 if (next_minor > MINORMASK)
564 next_minor = 0;
565 if (next_minor == start) {
566 /* Oh dear, all in use. */
567 spin_unlock(&all_mddevs_lock);
568 kfree(new);
569 return NULL;
570 }
f72ffdd6 571
efeb53c0
N		 572 is_free = 1;
573 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
574 if (mddev->unit == dev) {
575 is_free = 0;
576 break;
577 }
578 }
579 new->unit = dev;
580 new->md_minor = MINOR(dev);
581 new->hold_active = UNTIL_STOP;
1da177e4
LT		 582 list_add(&new->all_mddevs, &all_mddevs);
583 spin_unlock(&all_mddevs_lock);
584 return new;
585 }
586 spin_unlock(&all_mddevs_lock);
587
9ffae0cf 588 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 589 if (!new)
590 return NULL;
591
1da177e4
LT		 592 new->unit = unit;
593 if (MAJOR(unit) == MD_MAJOR)
594 new->md_minor = MINOR(unit);
595 else
596 new->md_minor = MINOR(unit) >> MdpMinorShift;
597
fafd7fb0 598 mddev_init(new);
1da177e4 599
1da177e4
LT		 600 goto retry;
601}
602
b6eb127d
N		 603static struct attribute_group md_redundancy_group;
604
5c47daf6 605void mddev_unlock(struct mddev *mddev)
1da177e4 606{
a64c876f 607 if (mddev->to_remove) {
b6eb127d
N		 608 /* These cannot be removed under reconfig_mutex as
609 * an access to the files will try to take reconfig_mutex
610 * while holding the file unremovable, which leads to
611 * a deadlock.
bb4f1e9d
N		 612 * So hold set sysfs_active while the remove in happeing,
613 * and anything else which might set ->to_remove or my
614 * otherwise change the sysfs namespace will fail with
615 * -EBUSY if sysfs_active is still set.
616 * We set sysfs_active under reconfig_mutex and elsewhere
617 * test it under the same mutex to ensure its correct value
618 * is seen.
b6eb127d 619 */
a64c876f
N		 620 struct attribute_group *to_remove = mddev->to_remove;
621 mddev->to_remove = NULL;
bb4f1e9d 622 mddev->sysfs_active = 1;
b6eb127d
N		 623 mutex_unlock(&mddev->reconfig_mutex);
624
00bcb4ac
N		 625 if (mddev->kobj.sd) {
626 if (to_remove != &md_redundancy_group)
627 sysfs_remove_group(&mddev->kobj, to_remove);
628 if (mddev->pers == NULL ||
629 mddev->pers->sync_request == NULL) {
630 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
631 if (mddev->sysfs_action)
632 sysfs_put(mddev->sysfs_action);
633 mddev->sysfs_action = NULL;
634 }
a64c876f 635 }
bb4f1e9d 636 mddev->sysfs_active = 0;
b6eb127d
N		 637 } else
638 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 639
751e67ca
CD		 640 /* As we've dropped the mutex we need a spinlock to
641 * make sure the thread doesn't disappear
01f96c0a
N		 642 */
643 spin_lock(&pers_lock);
005eca5e 644 md_wakeup_thread(mddev->thread);
01f96c0a 645 spin_unlock(&pers_lock);
1da177e4 646}
5c47daf6 647EXPORT_SYMBOL_GPL(mddev_unlock);
1da177e4 648
57d051dc 649struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
1ca69c4b
N		 650{
651 struct md_rdev *rdev;
652
653 rdev_for_each_rcu(rdev, mddev)
654 if (rdev->desc_nr == nr)
655 return rdev;
656
657 return NULL;
658}
57d051dc 659EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
1ca69c4b
N		 660
661static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
1da177e4 662{
3cb03002 663 struct md_rdev *rdev;
1da177e4 664
dafb20fa 665 rdev_for_each(rdev, mddev)
1da177e4
LT		 666 if (rdev->bdev->bd_dev == dev)
667 return rdev;
159ec1fc 668
1da177e4
LT		 669 return NULL;
670}
671
1ca69c4b
N		 672static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
673{
674 struct md_rdev *rdev;
675
676 rdev_for_each_rcu(rdev, mddev)
677 if (rdev->bdev->bd_dev == dev)
678 return rdev;
679
680 return NULL;
681}
682
84fc4b56 683static struct md_personality *find_pers(int level, char *clevel)
2604b703 684{
84fc4b56 685 struct md_personality *pers;
d9d166c2
N		 686 list_for_each_entry(pers, &pers_list, list) {
687 if (level != LEVEL_NONE && pers->level == level)
2604b703 688 return pers;
d9d166c2
N		 689 if (strcmp(pers->name, clevel)==0)
690 return pers;
691 }
2604b703
N		 692 return NULL;
693}
694
b73df2d3 695/* return the offset of the super block in 512byte sectors */
3cb03002 696static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 697{
57b2caa3 698 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 699 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 700}
701
f72ffdd6 702static int alloc_disk_sb(struct md_rdev *rdev)
1da177e4 703{
1da177e4
LT		 704 rdev->sb_page = alloc_page(GFP_KERNEL);
705 if (!rdev->sb_page) {
706 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 707 return -ENOMEM;
1da177e4
LT		 708 }
709
710 return 0;
711}
712
545c8795 713void md_rdev_clear(struct md_rdev *rdev)
1da177e4
LT		 714{
715 if (rdev->sb_page) {
2d1f3b5d 716 put_page(rdev->sb_page);
1da177e4
LT		 717 rdev->sb_loaded = 0;
718 rdev->sb_page = NULL;
0f420358 719 rdev->sb_start = 0;
dd8ac336 720 rdev->sectors = 0;
1da177e4 721 }
2699b672
N		 722 if (rdev->bb_page) {
723 put_page(rdev->bb_page);
724 rdev->bb_page = NULL;
725 }
4fa2f327
N		 726 kfree(rdev->badblocks.page);
727 rdev->badblocks.page = NULL;
1da177e4 728}
545c8795 729EXPORT_SYMBOL_GPL(md_rdev_clear);
1da177e4 730
6712ecf8 731static void super_written(struct bio *bio, int error)
7bfa19f2 732{
3cb03002 733 struct md_rdev *rdev = bio->bi_private;
fd01b88c 734 struct mddev *mddev = rdev->mddev;
7bfa19f2 735
3a0f5bbb
N		 736 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
737 printk("md: super_written gets error=%d, uptodate=%d\n",
738 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
739 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 740 md_error(mddev, rdev);
3a0f5bbb 741 }
7bfa19f2 742
a9701a30
N		 743 if (atomic_dec_and_test(&mddev->pending_writes))
744 wake_up(&mddev->sb_wait);
f8b58edf 745 bio_put(bio);
7bfa19f2
N		 746}
747
fd01b88c 748void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 749 sector_t sector, int size, struct page *page)
750{
751 /* write first size bytes of page to sector of rdev
752 * Increment mddev->pending_writes before returning
753 * and decrement it on completion, waking up sb_wait
754 * if zero is reached.
755 * If an error occurred, call md_error
756 */
a167f663 757 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 758
a6ff7e08 759 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
4f024f37 760 bio->bi_iter.bi_sector = sector;
7bfa19f2
N		 761 bio_add_page(bio, page, size, 0);
762 bio->bi_private = rdev;
763 bio->bi_end_io = super_written;
a9701a30 764
7bfa19f2 765 atomic_inc(&mddev->pending_writes);
a5bf4df0 766 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 767}
768
fd01b88c 769void md_super_wait(struct mddev *mddev)
a9701a30 770{
e9c7469b 771 /* wait for all superblock writes that were scheduled to complete */
1967cd56 772 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
7bfa19f2
N		 773}
774
3cb03002 775int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 776 struct page *page, int rw, bool metadata_op)
1da177e4 777{
a167f663 778 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 779 int ret;
780
a6ff7e08
JB		 781 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
782 rdev->meta_bdev : rdev->bdev;
ccebd4c4 783 if (metadata_op)
4f024f37 784 bio->bi_iter.bi_sector = sector + rdev->sb_start;
1fdd6fc9
N		 785 else if (rdev->mddev->reshape_position != MaxSector &&
786 (rdev->mddev->reshape_backwards ==
787 (sector >= rdev->mddev->reshape_position)))
4f024f37 788 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
ccebd4c4 789 else
4f024f37 790 bio->bi_iter.bi_sector = sector + rdev->data_offset;
1da177e4 791 bio_add_page(bio, page, size, 0);
c170bbb4 792 submit_bio_wait(rw, bio);
1da177e4
LT		 793
794 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
795 bio_put(bio);
796 return ret;
797}
a8745db2 798EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 799
f72ffdd6 800static int read_disk_sb(struct md_rdev *rdev, int size)
1da177e4
LT		 801{
802 char b[BDEVNAME_SIZE];
403df478 803
1da177e4
LT		 804 if (rdev->sb_loaded)
805 return 0;
806
ccebd4c4 807 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 808 goto fail;
809 rdev->sb_loaded = 1;
810 return 0;
811
812fail:
813 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
814 bdevname(rdev->bdev,b));
815 return -EINVAL;
816}
817
818static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
819{
f72ffdd6 820 return sb1->set_uuid0 == sb2->set_uuid0 &&
05710466
AN		 821 sb1->set_uuid1 == sb2->set_uuid1 &&
822 sb1->set_uuid2 == sb2->set_uuid2 &&
823 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 824}
825
1da177e4
LT		 826static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
827{
828 int ret;
829 mdp_super_t *tmp1, *tmp2;
830
831 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
832 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
833
834 if (!tmp1 || !tmp2) {
835 ret = 0;
35020f1a 836 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 837 goto abort;
838 }
839
840 *tmp1 = *sb1;
841 *tmp2 = *sb2;
842
843 /*
844 * nr_disks is not constant
845 */
846 tmp1->nr_disks = 0;
847 tmp2->nr_disks = 0;
848
ce0c8e05 849 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 850abort:
990a8baf
JJ		 851 kfree(tmp1);
852 kfree(tmp2);
1da177e4
LT		 853 return ret;
854}
855
4d167f09
N		 856static u32 md_csum_fold(u32 csum)
857{
858 csum = (csum & 0xffff) + (csum >> 16);
859 return (csum & 0xffff) + (csum >> 16);
860}
861
f72ffdd6 862static unsigned int calc_sb_csum(mdp_super_t *sb)
1da177e4 863{
4d167f09
N		 864 u64 newcsum = 0;
865 u32 *sb32 = (u32*)sb;
866 int i;
1da177e4
LT		 867 unsigned int disk_csum, csum;
868
869 disk_csum = sb->sb_csum;
870 sb->sb_csum = 0;
4d167f09
N		 871
872 for (i = 0; i < MD_SB_BYTES/4 ; i++)
873 newcsum += sb32[i];
874 csum = (newcsum & 0xffffffff) + (newcsum>>32);
875
4d167f09
N		 876#ifdef CONFIG_ALPHA
877 /* This used to use csum_partial, which was wrong for several
878 * reasons including that different results are returned on
879 * different architectures. It isn't critical that we get exactly
880 * the same return value as before (we always csum_fold before
881 * testing, and that removes any differences). However as we
882 * know that csum_partial always returned a 16bit value on
883 * alphas, do a fold to maximise conformity to previous behaviour.
884 */
885 sb->sb_csum = md_csum_fold(disk_csum);
886#else
1da177e4 887 sb->sb_csum = disk_csum;
4d167f09 888#endif
1da177e4
LT		 889 return csum;
890}
891
1da177e4
LT		 892/*
893 * Handle superblock details.
894 * We want to be able to handle multiple superblock formats
895 * so we have a common interface to them all, and an array of
896 * different handlers.
897 * We rely on user-space to write the initial superblock, and support
898 * reading and updating of superblocks.
899 * Interface methods are:
3cb03002 900 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 901 * loads and validates a superblock on dev.
902 * if refdev != NULL, compare superblocks on both devices
903 * Return:
904 * 0 - dev has a superblock that is compatible with refdev
905 * 1 - dev has a superblock that is compatible and newer than refdev
906 * so dev should be used as the refdev in future
907 * -EINVAL superblock incompatible or invalid
908 * -othererror e.g. -EIO
909 *
fd01b88c 910 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 911 * Verify that dev is acceptable into mddev.
912 * The first time, mddev->raid_disks will be 0, and data from
913 * dev should be merged in. Subsequent calls check that dev
914 * is new enough. Return 0 or -EINVAL
915 *
fd01b88c 916 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 917 * Update the superblock for rdev with data in mddev
918 * This does not write to disc.
919 *
920 */
921
922struct super_type {
0cd17fec
CW		 923 char *name;
924 struct module *owner;
c6563a8c
N		 925 int (*load_super)(struct md_rdev *rdev,
926 struct md_rdev *refdev,
0cd17fec 927 int minor_version);
c6563a8c
N		 928 int (*validate_super)(struct mddev *mddev,
929 struct md_rdev *rdev);
930 void (*sync_super)(struct mddev *mddev,
931 struct md_rdev *rdev);
3cb03002 932 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 933 sector_t num_sectors);
c6563a8c
N		 934 int (*allow_new_offset)(struct md_rdev *rdev,
935 unsigned long long new_offset);
1da177e4
LT		 936};
937
0894cc30
AN		 938/*
939 * Check that the given mddev has no bitmap.
940 *
941 * This function is called from the run method of all personalities that do not
942 * support bitmaps. It prints an error message and returns non-zero if mddev
943 * has a bitmap. Otherwise, it returns 0.
944 *
945 */
fd01b88c 946int md_check_no_bitmap(struct mddev *mddev)
0894cc30 947{
c3d9714e 948 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 949 return 0;
950 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
951 mdname(mddev), mddev->pers->name);
952 return 1;
953}
954EXPORT_SYMBOL(md_check_no_bitmap);
955
1da177e4 956/*
f72ffdd6 957 * load_super for 0.90.0
1da177e4 958 */
3cb03002 959static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 960{
961 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
962 mdp_super_t *sb;
963 int ret;
1da177e4
LT		 964
965 /*
0f420358 966 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 967 * it's at the end of the disk.
968 *
969 * It also happens to be a multiple of 4Kb.
970 */
57b2caa3 971 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 972
0002b271 973 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 974 if (ret) return ret;
975
976 ret = -EINVAL;
977
978 bdevname(rdev->bdev, b);
65a06f06 979 sb = page_address(rdev->sb_page);
1da177e4
LT		 980
981 if (sb->md_magic != MD_SB_MAGIC) {
982 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
983 b);
984 goto abort;
985 }
986
987 if (sb->major_version != 0 ||
f6705578
N		 988 sb->minor_version < 90 ||
989 sb->minor_version > 91) {
1da177e4
LT		 990 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
991 sb->major_version, sb->minor_version,
992 b);
993 goto abort;
994 }
995
996 if (sb->raid_disks <= 0)
997 goto abort;
998
4d167f09 999 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1000 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1001 b);
1002 goto abort;
1003 }
1004
1005 rdev->preferred_minor = sb->md_minor;
1006 rdev->data_offset = 0;
c6563a8c 1007 rdev->new_data_offset = 0;
0002b271 1008 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1009 rdev->badblocks.shift = -1;
1da177e4
LT		 1010
1011 if (sb->level == LEVEL_MULTIPATH)
1012 rdev->desc_nr = -1;
1013 else
1014 rdev->desc_nr = sb->this_disk.number;
1015
9a7b2b0f 1016 if (!refdev) {
1da177e4 1017 ret = 1;
9a7b2b0f 1018 } else {
1da177e4 1019 __u64 ev1, ev2;
65a06f06 1020 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1021 if (!uuid_equal(refsb, sb)) {
1022 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1023 b, bdevname(refdev->bdev,b2));
1024 goto abort;
1025 }
1026 if (!sb_equal(refsb, sb)) {
1027 printk(KERN_WARNING "md: %s has same UUID"
1028 " but different superblock to %s\n",
1029 b, bdevname(refdev->bdev, b2));
1030 goto abort;
1031 }
1032 ev1 = md_event(sb);
1033 ev2 = md_event(refsb);
1034 if (ev1 > ev2)
1035 ret = 1;
f72ffdd6 1036 else
1da177e4
LT		 1037 ret = 0;
1038 }
8190e754 1039 rdev->sectors = rdev->sb_start;
667a5313
N		 1040 /* Limit to 4TB as metadata cannot record more than that.
1041 * (not needed for Linear and RAID0 as metadata doesn't
1042 * record this size)
1043 */
1044 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
27a7b260 1045 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1046
27a7b260 1047 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1048 /* "this cannot possibly happen" ... */
1049 ret = -EINVAL;
1050
1da177e4
LT		 1051 abort:
1052 return ret;
1053}
1054
1055/*
1056 * validate_super for 0.90.0
1057 */
fd01b88c 1058static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1059{
1060 mdp_disk_t *desc;
65a06f06 1061 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1062 __u64 ev1 = md_event(sb);
1da177e4 1063
41158c7e 1064 rdev->raid_disk = -1;
c5d79adb
N		 1065 clear_bit(Faulty, &rdev->flags);
1066 clear_bit(In_sync, &rdev->flags);
8313b8e5 1067 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1068 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1069
1da177e4
LT		 1070 if (mddev->raid_disks == 0) {
1071 mddev->major_version = 0;
1072 mddev->minor_version = sb->minor_version;
1073 mddev->patch_version = sb->patch_version;
e691063a 1074 mddev->external = 0;
9d8f0363 1075 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1076 mddev->ctime = sb->ctime;
1077 mddev->utime = sb->utime;
1078 mddev->level = sb->level;
d9d166c2 1079 mddev->clevel[0] = 0;
1da177e4
LT		 1080 mddev->layout = sb->layout;
1081 mddev->raid_disks = sb->raid_disks;
27a7b260 1082 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1083 mddev->events = ev1;
c3d9714e 1084 mddev->bitmap_info.offset = 0;
6409bb05
N		 1085 mddev->bitmap_info.space = 0;
1086 /* bitmap can use 60 K after the 4K superblocks */
c3d9714e 1087 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 1088 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
2c810cdd 1089 mddev->reshape_backwards = 0;
1da177e4 1090
f6705578
N		 1091 if (mddev->minor_version >= 91) {
1092 mddev->reshape_position = sb->reshape_position;
1093 mddev->delta_disks = sb->delta_disks;
1094 mddev->new_level = sb->new_level;
1095 mddev->new_layout = sb->new_layout;
664e7c41 1096 mddev->new_chunk_sectors = sb->new_chunk >> 9;
2c810cdd
N		 1097 if (mddev->delta_disks < 0)
1098 mddev->reshape_backwards = 1;
f6705578
N		 1099 } else {
1100 mddev->reshape_position = MaxSector;
1101 mddev->delta_disks = 0;
1102 mddev->new_level = mddev->level;
1103 mddev->new_layout = mddev->layout;
664e7c41 1104 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1105 }
1106
1da177e4
LT		 1107 if (sb->state & (1<<MD_SB_CLEAN))
1108 mddev->recovery_cp = MaxSector;
1109 else {
f72ffdd6 1110 if (sb->events_hi == sb->cp_events_hi &&
1da177e4
LT		 1111 sb->events_lo == sb->cp_events_lo) {
1112 mddev->recovery_cp = sb->recovery_cp;
1113 } else
1114 mddev->recovery_cp = 0;
1115 }
1116
1117 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1118 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1119 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1120 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1121
1122 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1123
1124 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
6409bb05 1125 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1126 mddev->bitmap_info.offset =
1127 mddev->bitmap_info.default_offset;
6409bb05 1128 mddev->bitmap_info.space =
c9ad020f 1129 mddev->bitmap_info.default_space;
6409bb05 1130 }
a654b9d8 1131
41158c7e 1132 } else if (mddev->pers == NULL) {
be6800a7
N		 1133 /* Insist on good event counter while assembling, except
1134 * for spares (which don't need an event count) */
1da177e4 1135 ++ev1;
be6800a7
N		 1136 if (sb->disks[rdev->desc_nr].state & (
1137 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
f72ffdd6 1138 if (ev1 < mddev->events)
be6800a7 1139 return -EINVAL;
41158c7e
N		 1140 } else if (mddev->bitmap) {
1141 /* if adding to array with a bitmap, then we can accept an
1142 * older device ... but not too old.
1143 */
41158c7e
N		 1144 if (ev1 < mddev->bitmap->events_cleared)
1145 return 0;
8313b8e5
N		 1146 if (ev1 < mddev->events)
1147 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1148 } else {
1149 if (ev1 < mddev->events)
1150 /* just a hot-add of a new device, leave raid_disk at -1 */
1151 return 0;
1152 }
41158c7e 1153
1da177e4 1154 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1155 desc = sb->disks + rdev->desc_nr;
1156
1157 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1158 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1159 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1160 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1161 set_bit(In_sync, &rdev->flags);
1da177e4 1162 rdev->raid_disk = desc->raid_disk;
f466722c 1163 rdev->saved_raid_disk = desc->raid_disk;
0261cd9f
N		 1164 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1165 /* active but not in sync implies recovery up to
1166 * reshape position. We don't know exactly where
1167 * that is, so set to zero for now */
1168 if (mddev->minor_version >= 91) {
1169 rdev->recovery_offset = 0;
1170 rdev->raid_disk = desc->raid_disk;
1171 }
1da177e4 1172 }
8ddf9efe
N		 1173 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1174 set_bit(WriteMostly, &rdev->flags);
41158c7e 1175 } else /* MULTIPATH are always insync */
b2d444d7 1176 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1177 return 0;
1178}
1179
1180/*
1181 * sync_super for 0.90.0
1182 */
fd01b88c 1183static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1184{
1185 mdp_super_t *sb;
3cb03002 1186 struct md_rdev *rdev2;
1da177e4 1187 int next_spare = mddev->raid_disks;
19133a42 1188
1da177e4
LT		 1189 /* make rdev->sb match mddev data..
1190 *
1191 * 1/ zero out disks
1192 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1193 * 3/ any empty disks < next_spare become removed
1194 *
1195 * disks[0] gets initialised to REMOVED because
1196 * we cannot be sure from other fields if it has
1197 * been initialised or not.
1198 */
1199 int i;
1200 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1201
61181565
N		 1202 rdev->sb_size = MD_SB_BYTES;
1203
65a06f06 1204 sb = page_address(rdev->sb_page);
1da177e4
LT		 1205
1206 memset(sb, 0, sizeof(*sb));
1207
1208 sb->md_magic = MD_SB_MAGIC;
1209 sb->major_version = mddev->major_version;
1da177e4
LT		 1210 sb->patch_version = mddev->patch_version;
1211 sb->gvalid_words = 0; /* ignored */
1212 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1213 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1214 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1215 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1216
1217 sb->ctime = mddev->ctime;
1218 sb->level = mddev->level;
58c0fed4 1219 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1220 sb->raid_disks = mddev->raid_disks;
1221 sb->md_minor = mddev->md_minor;
e691063a 1222 sb->not_persistent = 0;
1da177e4
LT		 1223 sb->utime = mddev->utime;
1224 sb->state = 0;
1225 sb->events_hi = (mddev->events>>32);
1226 sb->events_lo = (u32)mddev->events;
1227
f6705578
N		 1228 if (mddev->reshape_position == MaxSector)
1229 sb->minor_version = 90;
1230 else {
1231 sb->minor_version = 91;
1232 sb->reshape_position = mddev->reshape_position;
1233 sb->new_level = mddev->new_level;
1234 sb->delta_disks = mddev->delta_disks;
1235 sb->new_layout = mddev->new_layout;
664e7c41 1236 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1237 }
1238 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1239 if (mddev->in_sync)
1240 {
1241 sb->recovery_cp = mddev->recovery_cp;
1242 sb->cp_events_hi = (mddev->events>>32);
1243 sb->cp_events_lo = (u32)mddev->events;
1244 if (mddev->recovery_cp == MaxSector)
1245 sb->state = (1<< MD_SB_CLEAN);
1246 } else
1247 sb->recovery_cp = 0;
1248
1249 sb->layout = mddev->layout;
9d8f0363 1250 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1251
c3d9714e 1252 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1253 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1254
1da177e4 1255 sb->disks[0].state = (1<<MD_DISK_REMOVED);
dafb20fa 1256 rdev_for_each(rdev2, mddev) {
1da177e4 1257 mdp_disk_t *d;
86e6ffdd 1258 int desc_nr;
0261cd9f
N		 1259 int is_active = test_bit(In_sync, &rdev2->flags);
1260
1261 if (rdev2->raid_disk >= 0 &&
1262 sb->minor_version >= 91)
1263 /* we have nowhere to store the recovery_offset,
1264 * but if it is not below the reshape_position,
1265 * we can piggy-back on that.
1266 */
1267 is_active = 1;
1268 if (rdev2->raid_disk < 0 ||
1269 test_bit(Faulty, &rdev2->flags))
1270 is_active = 0;
1271 if (is_active)
86e6ffdd 1272 desc_nr = rdev2->raid_disk;
1da177e4 1273 else
86e6ffdd 1274 desc_nr = next_spare++;
19133a42 1275 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1276 d = &sb->disks[rdev2->desc_nr];
1277 nr_disks++;
1278 d->number = rdev2->desc_nr;
1279 d->major = MAJOR(rdev2->bdev->bd_dev);
1280 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1281 if (is_active)
1da177e4
LT		 1282 d->raid_disk = rdev2->raid_disk;
1283 else
1284 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1285 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1286 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1287 else if (is_active) {
1da177e4 1288 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1289 if (test_bit(In_sync, &rdev2->flags))
1290 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1291 active++;
1292 working++;
1293 } else {
1294 d->state = 0;
1295 spare++;
1296 working++;
1297 }
8ddf9efe
N		 1298 if (test_bit(WriteMostly, &rdev2->flags))
1299 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1300 }
1da177e4
LT		 1301 /* now set the "removed" and "faulty" bits on any missing devices */
1302 for (i=0 ; i < mddev->raid_disks ; i++) {
1303 mdp_disk_t *d = &sb->disks[i];
1304 if (d->state == 0 && d->number == 0) {
1305 d->number = i;
1306 d->raid_disk = i;
1307 d->state = (1<<MD_DISK_REMOVED);
1308 d->state |= (1<<MD_DISK_FAULTY);
1309 failed++;
1310 }
1311 }
1312 sb->nr_disks = nr_disks;
1313 sb->active_disks = active;
1314 sb->working_disks = working;
1315 sb->failed_disks = failed;
1316 sb->spare_disks = spare;
1317
1318 sb->this_disk = sb->disks[rdev->desc_nr];
1319 sb->sb_csum = calc_sb_csum(sb);
1320}
1321
0cd17fec
CW		 1322/*
1323 * rdev_size_change for 0.90.0
1324 */
1325static unsigned long long
3cb03002 1326super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1327{
58c0fed4 1328 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1329 return 0; /* component must fit device */
c3d9714e 1330 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1331 return 0; /* can't move bitmap */
57b2caa3 1332 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1333 if (!num_sectors || num_sectors > rdev->sb_start)
1334 num_sectors = rdev->sb_start;
27a7b260
N		 1335 /* Limit to 4TB as metadata cannot record more than that.
1336 * 4TB == 2^32 KB, or 2*2^32 sectors.
1337 */
667a5313 1338 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
27a7b260 1339 num_sectors = (2ULL << 32) - 2;
0f420358 1340 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1341 rdev->sb_page);
1342 md_super_wait(rdev->mddev);
c26a44ed 1343 return num_sectors;
0cd17fec
CW		 1344}
1345
c6563a8c
N		 1346static int
1347super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1348{
1349 /* non-zero offset changes not possible with v0.90 */
1350 return new_offset == 0;
1351}
0cd17fec 1352
1da177e4
LT		 1353/*
1354 * version 1 superblock
1355 */
1356
f72ffdd6 1357static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1da177e4 1358{
1c05b4bc
N		 1359 __le32 disk_csum;
1360 u32 csum;
1da177e4
LT		 1361 unsigned long long newcsum;
1362 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1363 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1364
1365 disk_csum = sb->sb_csum;
1366 sb->sb_csum = 0;
1367 newcsum = 0;
1f3c9907 1368 for (; size >= 4; size -= 4)
1da177e4
LT		 1369 newcsum += le32_to_cpu(*isuper++);
1370
1371 if (size == 2)
1c05b4bc 1372 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1373
1374 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1375 sb->sb_csum = disk_csum;
1376 return cpu_to_le32(csum);
1377}
1378
2699b672
N		 1379static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1380 int acknowledged);
3cb03002 1381static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1382{
1383 struct mdp_superblock_1 *sb;
1384 int ret;
0f420358 1385 sector_t sb_start;
c6563a8c 1386 sector_t sectors;
1da177e4 1387 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1388 int bmask;
1da177e4
LT		 1389
1390 /*
0f420358 1391 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1392 * It is always aligned to a 4K boundary and
1393 * depeding on minor_version, it can be:
1394 * 0: At least 8K, but less than 12K, from end of device
1395 * 1: At start of device
1396 * 2: 4K from start of device.
1397 */
1398 switch(minor_version) {
1399 case 0:
77304d2a 1400 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1401 sb_start -= 8*2;
1402 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1403 break;
1404 case 1:
0f420358 1405 sb_start = 0;
1da177e4
LT		 1406 break;
1407 case 2:
0f420358 1408 sb_start = 8;
1da177e4
LT		 1409 break;
1410 default:
1411 return -EINVAL;
1412 }
0f420358 1413 rdev->sb_start = sb_start;
1da177e4 1414
0002b271
N		 1415 /* superblock is rarely larger than 1K, but it can be larger,
1416 * and it is safe to read 4k, so we do that
1417 */
1418 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1419 if (ret) return ret;
1420
65a06f06 1421 sb = page_address(rdev->sb_page);
1da177e4
LT		 1422
1423 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1424 sb->major_version != cpu_to_le32(1) ||
1425 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1426 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1427 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1428 return -EINVAL;
1429
1430 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1431 printk("md: invalid superblock checksum on %s\n",
1432 bdevname(rdev->bdev,b));
1433 return -EINVAL;
1434 }
1435 if (le64_to_cpu(sb->data_size) < 10) {
1436 printk("md: data_size too small on %s\n",
1437 bdevname(rdev->bdev,b));
1438 return -EINVAL;
1439 }
c6563a8c
N		 1440 if (sb->pad0 ||
1441 sb->pad3[0] ||
1442 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1443 /* Some padding is non-zero, might be a new feature */
1444 return -EINVAL;
e11e93fa 1445
1da177e4
LT		 1446 rdev->preferred_minor = 0xffff;
1447 rdev->data_offset = le64_to_cpu(sb->data_offset);
c6563a8c
N		 1448 rdev->new_data_offset = rdev->data_offset;
1449 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1450 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1451 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
4dbcdc75 1452 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1453
0002b271 1454 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1455 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1456 if (rdev->sb_size & bmask)
a1801f85
N		 1457 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1458
1459 if (minor_version
0f420358 1460 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1461 return -EINVAL;
c6563a8c
N		 1462 if (minor_version
1463 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1464 return -EINVAL;
0002b271 1465
31b65a0d
N		 1466 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1467 rdev->desc_nr = -1;
1468 else
1469 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1470
2699b672
N		 1471 if (!rdev->bb_page) {
1472 rdev->bb_page = alloc_page(GFP_KERNEL);
1473 if (!rdev->bb_page)
1474 return -ENOMEM;
1475 }
1476 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1477 rdev->badblocks.count == 0) {
1478 /* need to load the bad block list.
1479 * Currently we limit it to one page.
1480 */
1481 s32 offset;
1482 sector_t bb_sector;
1483 u64 *bbp;
1484 int i;
1485 int sectors = le16_to_cpu(sb->bblog_size);
1486 if (sectors > (PAGE_SIZE / 512))
1487 return -EINVAL;
1488 offset = le32_to_cpu(sb->bblog_offset);
1489 if (offset == 0)
1490 return -EINVAL;
1491 bb_sector = (long long)offset;
1492 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1493 rdev->bb_page, READ, true))
1494 return -EIO;
1495 bbp = (u64 *)page_address(rdev->bb_page);
1496 rdev->badblocks.shift = sb->bblog_shift;
1497 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1498 u64 bb = le64_to_cpu(*bbp);
1499 int count = bb & (0x3ff);
1500 u64 sector = bb >> 10;
1501 sector <<= sb->bblog_shift;
1502 count <<= sb->bblog_shift;
1503 if (bb + 1 == 0)
1504 break;
1505 if (md_set_badblocks(&rdev->badblocks,
1506 sector, count, 1) == 0)
1507 return -EINVAL;
1508 }
486adf72
N		 1509 } else if (sb->bblog_offset != 0)
1510 rdev->badblocks.shift = 0;
2699b672 1511
9a7b2b0f 1512 if (!refdev) {
8ed75463 1513 ret = 1;
9a7b2b0f 1514 } else {
1da177e4 1515 __u64 ev1, ev2;
65a06f06 1516 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1517
1518 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1519 sb->level != refsb->level ||
1520 sb->layout != refsb->layout ||
1521 sb->chunksize != refsb->chunksize) {
1522 printk(KERN_WARNING "md: %s has strangely different"
1523 " superblock to %s\n",
1524 bdevname(rdev->bdev,b),
1525 bdevname(refdev->bdev,b2));
1526 return -EINVAL;
1527 }
1528 ev1 = le64_to_cpu(sb->events);
1529 ev2 = le64_to_cpu(refsb->events);
1530
1531 if (ev1 > ev2)
8ed75463
N		 1532 ret = 1;
1533 else
1534 ret = 0;
1da177e4 1535 }
c6563a8c
N		 1536 if (minor_version) {
1537 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1538 sectors -= rdev->data_offset;
1539 } else
1540 sectors = rdev->sb_start;
1541 if (sectors < le64_to_cpu(sb->data_size))
1da177e4 1542 return -EINVAL;
dd8ac336 1543 rdev->sectors = le64_to_cpu(sb->data_size);
8ed75463 1544 return ret;
1da177e4
LT		 1545}
1546
fd01b88c 1547static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1548{
65a06f06 1549 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1550 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1551
41158c7e 1552 rdev->raid_disk = -1;
c5d79adb
N		 1553 clear_bit(Faulty, &rdev->flags);
1554 clear_bit(In_sync, &rdev->flags);
8313b8e5 1555 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1556 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1557
1da177e4
LT		 1558 if (mddev->raid_disks == 0) {
1559 mddev->major_version = 1;
1560 mddev->patch_version = 0;
e691063a 1561 mddev->external = 0;
9d8f0363 1562 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1563 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1564 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1565 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1566 mddev->clevel[0] = 0;
1da177e4
LT		 1567 mddev->layout = le32_to_cpu(sb->layout);
1568 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1569 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1570 mddev->events = ev1;
c3d9714e 1571 mddev->bitmap_info.offset = 0;
6409bb05
N		 1572 mddev->bitmap_info.space = 0;
1573 /* Default location for bitmap is 1K after superblock
1574 * using 3K - total of 4K
1575 */
c3d9714e 1576 mddev->bitmap_info.default_offset = 1024 >> 9;
6409bb05 1577 mddev->bitmap_info.default_space = (4096-1024) >> 9;
2c810cdd
N		 1578 mddev->reshape_backwards = 0;
1579
1da177e4
LT		 1580 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1581 memcpy(mddev->uuid, sb->set_uuid, 16);
1582
1583 mddev->max_disks = (4096-256)/2;
a654b9d8 1584
71c0805c 1585 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
6409bb05 1586 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1587 mddev->bitmap_info.offset =
1588 (__s32)le32_to_cpu(sb->bitmap_offset);
6409bb05
N		 1589 /* Metadata doesn't record how much space is available.
1590 * For 1.0, we assume we can use up to the superblock
1591 * if before, else to 4K beyond superblock.
1592 * For others, assume no change is possible.
1593 */
1594 if (mddev->minor_version > 0)
1595 mddev->bitmap_info.space = 0;
1596 else if (mddev->bitmap_info.offset > 0)
1597 mddev->bitmap_info.space =
1598 8 - mddev->bitmap_info.offset;
1599 else
1600 mddev->bitmap_info.space =
1601 -mddev->bitmap_info.offset;
1602 }
e11e93fa 1603
f6705578
N		 1604 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1605 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1606 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1607 mddev->new_level = le32_to_cpu(sb->new_level);
1608 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1609 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
2c810cdd
N		 1610 if (mddev->delta_disks < 0 ||
1611 (mddev->delta_disks == 0 &&
1612 (le32_to_cpu(sb->feature_map)
1613 & MD_FEATURE_RESHAPE_BACKWARDS)))
1614 mddev->reshape_backwards = 1;
f6705578
N		 1615 } else {
1616 mddev->reshape_position = MaxSector;
1617 mddev->delta_disks = 0;
1618 mddev->new_level = mddev->level;
1619 mddev->new_layout = mddev->layout;
664e7c41 1620 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1621 }
1622
41158c7e 1623 } else if (mddev->pers == NULL) {
be6800a7
N		 1624 /* Insist of good event counter while assembling, except for
1625 * spares (which don't need an event count) */
1da177e4 1626 ++ev1;
be6800a7
N		 1627 if (rdev->desc_nr >= 0 &&
1628 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1629 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1630 if (ev1 < mddev->events)
1631 return -EINVAL;
41158c7e
N		 1632 } else if (mddev->bitmap) {
1633 /* If adding to array with a bitmap, then we can accept an
1634 * older device, but not too old.
1635 */
41158c7e
N		 1636 if (ev1 < mddev->bitmap->events_cleared)
1637 return 0;
8313b8e5
N		 1638 if (ev1 < mddev->events)
1639 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1640 } else {
1641 if (ev1 < mddev->events)
1642 /* just a hot-add of a new device, leave raid_disk at -1 */
1643 return 0;
1644 }
1da177e4
LT		 1645 if (mddev->level != LEVEL_MULTIPATH) {
1646 int role;
3673f305
N		 1647 if (rdev->desc_nr < 0 ||
1648 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1649 role = 0xffff;
1650 rdev->desc_nr = -1;
1651 } else
1652 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1653 switch(role) {
1654 case 0xffff: /* spare */
1da177e4
LT		 1655 break;
1656 case 0xfffe: /* faulty */
b2d444d7 1657 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1658 break;
1659 default:
f466722c 1660 rdev->saved_raid_disk = role;
5fd6c1dc 1661 if ((le32_to_cpu(sb->feature_map) &
f466722c 1662 MD_FEATURE_RECOVERY_OFFSET)) {
5fd6c1dc 1663 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
f466722c
N		 1664 if (!(le32_to_cpu(sb->feature_map) &
1665 MD_FEATURE_RECOVERY_BITMAP))
1666 rdev->saved_raid_disk = -1;
1667 } else
5fd6c1dc 1668 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1669 rdev->raid_disk = role;
1670 break;
1671 }
8ddf9efe
N		 1672 if (sb->devflags & WriteMostly1)
1673 set_bit(WriteMostly, &rdev->flags);
2d78f8c4
N		 1674 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1675 set_bit(Replacement, &rdev->flags);
41158c7e 1676 } else /* MULTIPATH are always insync */
b2d444d7 1677 set_bit(In_sync, &rdev->flags);
41158c7e 1678
1da177e4
LT		 1679 return 0;
1680}
1681
fd01b88c 1682static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1683{
1684 struct mdp_superblock_1 *sb;
3cb03002 1685 struct md_rdev *rdev2;
1da177e4
LT		 1686 int max_dev, i;
1687 /* make rdev->sb match mddev and rdev data. */
1688
65a06f06 1689 sb = page_address(rdev->sb_page);
1da177e4
LT		 1690
1691 sb->feature_map = 0;
1692 sb->pad0 = 0;
5fd6c1dc 1693 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1694 memset(sb->pad3, 0, sizeof(sb->pad3));
1695
1696 sb->utime = cpu_to_le64((__u64)mddev->utime);
1697 sb->events = cpu_to_le64(mddev->events);
1698 if (mddev->in_sync)
1699 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1700 else
1701 sb->resync_offset = cpu_to_le64(0);
1702
1c05b4bc 1703 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1704
f0ca340c 1705 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1706 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1707 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1708 sb->level = cpu_to_le32(mddev->level);
1709 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1710
aeb9b211
N		 1711 if (test_bit(WriteMostly, &rdev->flags))
1712 sb->devflags |= WriteMostly1;
1713 else
1714 sb->devflags &= ~WriteMostly1;
c6563a8c
N		 1715 sb->data_offset = cpu_to_le64(rdev->data_offset);
1716 sb->data_size = cpu_to_le64(rdev->sectors);
aeb9b211 1717
c3d9714e
N		 1718 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1719 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1720 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1721 }
5fd6c1dc
N		 1722
1723 if (rdev->raid_disk >= 0 &&
97e4f42d 1724 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1725 sb->feature_map |=
1726 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1727 sb->recovery_offset =
1728 cpu_to_le64(rdev->recovery_offset);
f466722c
N		 1729 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1730 sb->feature_map |=
1731 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
5fd6c1dc 1732 }
2d78f8c4
N		 1733 if (test_bit(Replacement, &rdev->flags))
1734 sb->feature_map |=
1735 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1736
f6705578
N		 1737 if (mddev->reshape_position != MaxSector) {
1738 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1739 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1740 sb->new_layout = cpu_to_le32(mddev->new_layout);
1741 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1742 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1743 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2c810cdd
N		 1744 if (mddev->delta_disks == 0 &&
1745 mddev->reshape_backwards)
1746 sb->feature_map
1747 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
c6563a8c
N		 1748 if (rdev->new_data_offset != rdev->data_offset) {
1749 sb->feature_map
1750 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1751 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1752 - rdev->data_offset));
1753 }
f6705578 1754 }
a654b9d8 1755
2699b672
N		 1756 if (rdev->badblocks.count == 0)
1757 /* Nothing to do for bad blocks*/ ;
1758 else if (sb->bblog_offset == 0)
1759 /* Cannot record bad blocks on this device */
1760 md_error(mddev, rdev);
1761 else {
1762 struct badblocks *bb = &rdev->badblocks;
1763 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1764 u64 *p = bb->page;
1765 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1766 if (bb->changed) {
1767 unsigned seq;
1768
1769retry:
1770 seq = read_seqbegin(&bb->lock);
1771
1772 memset(bbp, 0xff, PAGE_SIZE);
1773
1774 for (i = 0 ; i < bb->count ; i++) {
35f9ac2d 1775 u64 internal_bb = p[i];
2699b672
N		 1776 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1777 | BB_LEN(internal_bb));
35f9ac2d 1778 bbp[i] = cpu_to_le64(store_bb);
2699b672 1779 }
d0962936 1780 bb->changed = 0;
2699b672
N		 1781 if (read_seqretry(&bb->lock, seq))
1782 goto retry;
1783
1784 bb->sector = (rdev->sb_start +
1785 (int)le32_to_cpu(sb->bblog_offset));
1786 bb->size = le16_to_cpu(sb->bblog_size);
2699b672
N		 1787 }
1788 }
1789
1da177e4 1790 max_dev = 0;
dafb20fa 1791 rdev_for_each(rdev2, mddev)
1da177e4
LT		 1792 if (rdev2->desc_nr+1 > max_dev)
1793 max_dev = rdev2->desc_nr+1;
a778b73f 1794
70471daf
N		 1795 if (max_dev > le32_to_cpu(sb->max_dev)) {
1796 int bmask;
a778b73f 1797 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1798 rdev->sb_size = max_dev * 2 + 256;
1799 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1800 if (rdev->sb_size & bmask)
1801 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1802 } else
1803 max_dev = le32_to_cpu(sb->max_dev);
1804
1da177e4
LT		 1805 for (i=0; i<max_dev;i++)
1806 sb->dev_roles[i] = cpu_to_le16(0xfffe);
f72ffdd6 1807
dafb20fa 1808 rdev_for_each(rdev2, mddev) {
1da177e4 1809 i = rdev2->desc_nr;
b2d444d7 1810 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1811 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1812 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1813 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1814 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1815 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1816 else
1817 sb->dev_roles[i] = cpu_to_le16(0xffff);
1818 }
1819
1da177e4
LT		 1820 sb->sb_csum = calc_sb_1_csum(sb);
1821}
1822
0cd17fec 1823static unsigned long long
3cb03002 1824super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 1825{
1826 struct mdp_superblock_1 *sb;
15f4a5fd 1827 sector_t max_sectors;
58c0fed4 1828 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1829 return 0; /* component must fit device */
c6563a8c
N		 1830 if (rdev->data_offset != rdev->new_data_offset)
1831 return 0; /* too confusing */
0f420358 1832 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1833 /* minor versions 1 and 2; superblock before data */
77304d2a 1834 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1835 max_sectors -= rdev->data_offset;
1836 if (!num_sectors || num_sectors > max_sectors)
1837 num_sectors = max_sectors;
c3d9714e 1838 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1839 /* minor version 0 with bitmap we can't move */
1840 return 0;
1841 } else {
1842 /* minor version 0; superblock after data */
0f420358 1843 sector_t sb_start;
77304d2a 1844 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1845 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1846 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1847 if (!num_sectors || num_sectors > max_sectors)
1848 num_sectors = max_sectors;
0f420358 1849 rdev->sb_start = sb_start;
0cd17fec 1850 }
65a06f06 1851 sb = page_address(rdev->sb_page);
15f4a5fd 1852 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1853 sb->super_offset = rdev->sb_start;
0cd17fec 1854 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1855 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1856 rdev->sb_page);
1857 md_super_wait(rdev->mddev);
c26a44ed 1858 return num_sectors;
c6563a8c
N		 1859
1860}
1861
1862static int
1863super_1_allow_new_offset(struct md_rdev *rdev,
1864 unsigned long long new_offset)
1865{
1866 /* All necessary checks on new >= old have been done */
1867 struct bitmap *bitmap;
1868 if (new_offset >= rdev->data_offset)
1869 return 1;
1870
1871 /* with 1.0 metadata, there is no metadata to tread on
1872 * so we can always move back */
1873 if (rdev->mddev->minor_version == 0)
1874 return 1;
1875
1876 /* otherwise we must be sure not to step on
1877 * any metadata, so stay:
1878 * 36K beyond start of superblock
1879 * beyond end of badblocks
1880 * beyond write-intent bitmap
1881 */
1882 if (rdev->sb_start + (32+4)*2 > new_offset)
1883 return 0;
1884 bitmap = rdev->mddev->bitmap;
1885 if (bitmap && !rdev->mddev->bitmap_info.file &&
1886 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1ec885cd 1887 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
c6563a8c
N		 1888 return 0;
1889 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1890 return 0;
1891
1892 return 1;
0cd17fec 1893}
1da177e4 1894
75c96f85 1895static struct super_type super_types[] = {
1da177e4
LT		 1896 [0] = {
1897 .name = "0.90.0",
1898 .owner = THIS_MODULE,
0cd17fec
CW		 1899 .load_super = super_90_load,
1900 .validate_super = super_90_validate,
1901 .sync_super = super_90_sync,
1902 .rdev_size_change = super_90_rdev_size_change,
c6563a8c 1903 .allow_new_offset = super_90_allow_new_offset,
1da177e4
LT		 1904 },
1905 [1] = {
1906 .name = "md-1",
1907 .owner = THIS_MODULE,
0cd17fec
CW		 1908 .load_super = super_1_load,
1909 .validate_super = super_1_validate,
1910 .sync_super = super_1_sync,
1911 .rdev_size_change = super_1_rdev_size_change,
c6563a8c 1912 .allow_new_offset = super_1_allow_new_offset,
1da177e4
LT		 1913 },
1914};
1da177e4 1915
fd01b88c 1916static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 1917{
1918 if (mddev->sync_super) {
1919 mddev->sync_super(mddev, rdev);
1920 return;
1921 }
1922
1923 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1924
1925 super_types[mddev->major_version].sync_super(mddev, rdev);
1926}
1927
fd01b88c 1928static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1929{
3cb03002 1930 struct md_rdev *rdev, *rdev2;
1da177e4 1931
4b80991c
N		 1932 rcu_read_lock();
1933 rdev_for_each_rcu(rdev, mddev1)
1934 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1935 if (rdev->bdev->bd_contains ==
4b80991c
N		 1936 rdev2->bdev->bd_contains) {
1937 rcu_read_unlock();
7dd5e7c3 1938 return 1;
4b80991c
N		 1939 }
1940 rcu_read_unlock();
1da177e4
LT		 1941 return 0;
1942}
1943
1944static LIST_HEAD(pending_raid_disks);
1945
ac5e7113
AN		 1946/*
1947 * Try to register data integrity profile for an mddev
1948 *
1949 * This is called when an array is started and after a disk has been kicked
1950 * from the array. It only succeeds if all working and active component devices
1951 * are integrity capable with matching profiles.
1952 */
fd01b88c 1953int md_integrity_register(struct mddev *mddev)
ac5e7113 1954{
3cb03002 1955 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 1956
1957 if (list_empty(&mddev->disks))
1958 return 0; /* nothing to do */
629acb6a
JB		 1959 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1960 return 0; /* shouldn't register, or already is */
dafb20fa 1961 rdev_for_each(rdev, mddev) {
ac5e7113
AN		 1962 /* skip spares and non-functional disks */
1963 if (test_bit(Faulty, &rdev->flags))
1964 continue;
1965 if (rdev->raid_disk < 0)
1966 continue;
ac5e7113
AN		 1967 if (!reference) {
1968 /* Use the first rdev as the reference */
1969 reference = rdev;
1970 continue;
1971 }
1972 /* does this rdev's profile match the reference profile? */
1973 if (blk_integrity_compare(reference->bdev->bd_disk,
1974 rdev->bdev->bd_disk) < 0)
1975 return -EINVAL;
1976 }
89078d57
MP		 1977 if (!reference || !bdev_get_integrity(reference->bdev))
1978 return 0;
ac5e7113
AN		 1979 /*
1980 * All component devices are integrity capable and have matching
1981 * profiles, register the common profile for the md device.
1982 */
1983 if (blk_integrity_register(mddev->gendisk,
1984 bdev_get_integrity(reference->bdev)) != 0) {
1985 printk(KERN_ERR "md: failed to register integrity for %s\n",
1986 mdname(mddev));
1987 return -EINVAL;
1988 }
a91a2785
MP		 1989 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1990 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1991 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1992 mdname(mddev));
1993 return -EINVAL;
1994 }
ac5e7113
AN		 1995 return 0;
1996}
1997EXPORT_SYMBOL(md_integrity_register);
1998
1999/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 2000void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 2001{
2863b9eb
JB		 2002 struct blk_integrity *bi_rdev;
2003 struct blk_integrity *bi_mddev;
2004
2005 if (!mddev->gendisk)
2006 return;
2007
2008 bi_rdev = bdev_get_integrity(rdev->bdev);
2009 bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 2010
ac5e7113 2011 if (!bi_mddev) /* nothing to do */
3f9d99c1 2012 return;
ac5e7113 2013 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 2014 return;
ac5e7113
AN		 2015 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2016 rdev->bdev->bd_disk) >= 0)
2017 return;
2018 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2019 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2020}
ac5e7113 2021EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2022
f72ffdd6 2023static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
1da177e4 2024{
7dd5e7c3 2025 char b[BDEVNAME_SIZE];
f637b9f9 2026 struct kobject *ko;
5e55e2f5 2027 int err;
1da177e4 2028
11e2ede0
DW		 2029 /* prevent duplicates */
2030 if (find_rdev(mddev, rdev->bdev->bd_dev))
2031 return -EEXIST;
2032
dd8ac336
AN		 2033 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2034 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2035 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2036 if (mddev->pers) {
2037 /* Cannot change size, so fail
2038 * If mddev->level <= 0, then we don't care
2039 * about aligning sizes (e.g. linear)
2040 */
2041 if (mddev->level > 0)
2042 return -ENOSPC;
2043 } else
dd8ac336 2044 mddev->dev_sectors = rdev->sectors;
2bf071bf 2045 }
1da177e4
LT		 2046
2047 /* Verify rdev->desc_nr is unique.
2048 * If it is -1, assign a free number, else
2049 * check number is not in use
2050 */
4878e9eb 2051 rcu_read_lock();
1da177e4
LT		 2052 if (rdev->desc_nr < 0) {
2053 int choice = 0;
4878e9eb
N		 2054 if (mddev->pers)
2055 choice = mddev->raid_disks;
57d051dc 2056 while (md_find_rdev_nr_rcu(mddev, choice))
1da177e4
LT		 2057 choice++;
2058 rdev->desc_nr = choice;
2059 } else {
57d051dc 2060 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
4878e9eb 2061 rcu_read_unlock();
1da177e4 2062 return -EBUSY;
4878e9eb 2063 }
1da177e4 2064 }
4878e9eb 2065 rcu_read_unlock();
de01dfad
N		 2066 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2067 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2068 mdname(mddev), mddev->max_disks);
2069 return -EBUSY;
2070 }
19133a42 2071 bdevname(rdev->bdev,b);
90a9befb 2072 strreplace(b, '/', '!');
649316b2 2073
1da177e4 2074 rdev->mddev = mddev;
19133a42 2075 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2076
b2d6db58 2077 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2078 goto fail;
86e6ffdd 2079
0762b8bd 2080 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2081 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2082 /* failure here is OK */;
2083 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2084
4b80991c 2085 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2086 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2087
2088 /* May as well allow recovery to be retried once */
5389042f 2089 mddev->recovery_disabled++;
3f9d99c1 2090
1da177e4 2091 return 0;
5e55e2f5
N		 2092
2093 fail:
2094 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2095 b, mdname(mddev));
2096 return err;
1da177e4
LT		 2097}
2098
177a99b2 2099static void md_delayed_delete(struct work_struct *ws)
5792a285 2100{
3cb03002 2101 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2102 kobject_del(&rdev->kobj);
177a99b2 2103 kobject_put(&rdev->kobj);
5792a285
N		 2104}
2105
f72ffdd6 2106static void unbind_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2107{
2108 char b[BDEVNAME_SIZE];
403df478 2109
49731baa 2110 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2111 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2112 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2113 rdev->mddev = NULL;
86e6ffdd 2114 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2115 sysfs_put(rdev->sysfs_state);
2116 rdev->sysfs_state = NULL;
2230dfe4 2117 rdev->badblocks.count = 0;
5792a285 2118 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2119 * writing to 'remove' to "dev/state". We also need
2120 * to delay it due to rcu usage.
5792a285 2121 */
4b80991c 2122 synchronize_rcu();
177a99b2
N		 2123 INIT_WORK(&rdev->del_work, md_delayed_delete);
2124 kobject_get(&rdev->kobj);
e804ac78 2125 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2126}
2127
2128/*
2129 * prevent the device from being mounted, repartitioned or
2130 * otherwise reused by a RAID array (or any other kernel
2131 * subsystem), by bd_claiming the device.
2132 */
3cb03002 2133static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2134{
2135 int err = 0;
2136 struct block_device *bdev;
2137 char b[BDEVNAME_SIZE];
2138
d4d77629 2139 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2140 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2141 if (IS_ERR(bdev)) {
2142 printk(KERN_ERR "md: could not open %s.\n",
2143 __bdevname(dev, b));
2144 return PTR_ERR(bdev);
2145 }
1da177e4
LT		 2146 rdev->bdev = bdev;
2147 return err;
2148}
2149
3cb03002 2150static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2151{
2152 struct block_device *bdev = rdev->bdev;
2153 rdev->bdev = NULL;
e525fd89 2154 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2155}
2156
2157void md_autodetect_dev(dev_t dev);
2158
f72ffdd6 2159static void export_rdev(struct md_rdev *rdev)
1da177e4
LT		 2160{
2161 char b[BDEVNAME_SIZE];
403df478 2162
1da177e4
LT		 2163 printk(KERN_INFO "md: export_rdev(%s)\n",
2164 bdevname(rdev->bdev,b));
545c8795 2165 md_rdev_clear(rdev);
1da177e4 2166#ifndef MODULE
d0fae18f
N		 2167 if (test_bit(AutoDetected, &rdev->flags))
2168 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2169#endif
2170 unlock_rdev(rdev);
86e6ffdd 2171 kobject_put(&rdev->kobj);
1da177e4
LT		 2172}
2173
fb56dfef 2174void md_kick_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2175{
2176 unbind_rdev_from_array(rdev);
2177 export_rdev(rdev);
2178}
fb56dfef 2179EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
1da177e4 2180
fd01b88c 2181static void export_array(struct mddev *mddev)
1da177e4 2182{
0638bb0e 2183 struct md_rdev *rdev;
1da177e4 2184
0638bb0e
N		 2185 while (!list_empty(&mddev->disks)) {
2186 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2187 same_set);
fb56dfef 2188 md_kick_rdev_from_array(rdev);
1da177e4 2189 }
1da177e4
LT		 2190 mddev->raid_disks = 0;
2191 mddev->major_version = 0;
2192}
2193
f72ffdd6 2194static void sync_sbs(struct mddev *mddev, int nospares)
1da177e4 2195{
42543769
N		 2196 /* Update each superblock (in-memory image), but
2197 * if we are allowed to, skip spares which already
2198 * have the right event counter, or have one earlier
2199 * (which would mean they aren't being marked as dirty
2200 * with the rest of the array)
2201 */
3cb03002 2202 struct md_rdev *rdev;
dafb20fa 2203 rdev_for_each(rdev, mddev) {
42543769
N		 2204 if (rdev->sb_events == mddev->events ||
2205 (nospares &&
2206 rdev->raid_disk < 0 &&
42543769
N		 2207 rdev->sb_events+1 == mddev->events)) {
2208 /* Don't update this superblock */
2209 rdev->sb_loaded = 2;
2210 } else {
076f968b 2211 sync_super(mddev, rdev);
42543769
N		 2212 rdev->sb_loaded = 1;
2213 }
1da177e4
LT		 2214 }
2215}
2216
1aee41f6 2217void md_update_sb(struct mddev *mddev, int force_change)
1da177e4 2218{
3cb03002 2219 struct md_rdev *rdev;
06d91a5f 2220 int sync_req;
42543769 2221 int nospares = 0;
2699b672 2222 int any_badblocks_changed = 0;
1da177e4 2223
d87f064f
N		 2224 if (mddev->ro) {
2225 if (force_change)
2226 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2227 return;
2228 }
1da177e4 2229repeat:
3a3a5ddb 2230 /* First make sure individual recovery_offsets are correct */
dafb20fa 2231 rdev_for_each(rdev, mddev) {
3a3a5ddb
N		 2232 if (rdev->raid_disk >= 0 &&
2233 mddev->delta_disks >= 0 &&
2234 !test_bit(In_sync, &rdev->flags) &&
2235 mddev->curr_resync_completed > rdev->recovery_offset)
2236 rdev->recovery_offset = mddev->curr_resync_completed;
2237
f72ffdd6 2238 }
bd52b746 2239 if (!mddev->persistent) {
070dc6dd 2240 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2241 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2242 if (!mddev->external) {
d97a41dc 2243 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
dafb20fa 2244 rdev_for_each(rdev, mddev) {
de393cde 2245 if (rdev->badblocks.changed) {
d0962936 2246 rdev->badblocks.changed = 0;
de393cde
N		 2247 md_ack_all_badblocks(&rdev->badblocks);
2248 md_error(mddev, rdev);
2249 }
2250 clear_bit(Blocked, &rdev->flags);
2251 clear_bit(BlockedBadBlocks, &rdev->flags);
2252 wake_up(&rdev->blocked_wait);
2253 }
2254 }
3a3a5ddb
N		 2255 wake_up(&mddev->sb_wait);
2256 return;
2257 }
2258
85572d7c 2259 spin_lock(&mddev->lock);
84692195 2260
3a3a5ddb
N		 2261 mddev->utime = get_seconds();
2262
850b2b42
N		 2263 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2264 force_change = 1;
2265 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2266 /* just a clean<-> dirty transition, possibly leave spares alone,
2267 * though if events isn't the right even/odd, we will have to do
2268 * spares after all
2269 */
2270 nospares = 1;
2271 if (force_change)
2272 nospares = 0;
2273 if (mddev->degraded)
84692195
N		 2274 /* If the array is degraded, then skipping spares is both
2275 * dangerous and fairly pointless.
2276 * Dangerous because a device that was removed from the array
2277 * might have a event_count that still looks up-to-date,
2278 * so it can be re-added without a resync.
2279 * Pointless because if there are any spares to skip,
2280 * then a recovery will happen and soon that array won't
2281 * be degraded any more and the spare can go back to sleep then.
2282 */
850b2b42 2283 nospares = 0;
84692195 2284
06d91a5f 2285 sync_req = mddev->in_sync;
42543769
N		 2286
2287 /* If this is just a dirty<->clean transition, and the array is clean
2288 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2289 if (nospares
42543769 2290 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2291 && mddev->can_decrease_events
2292 && mddev->events != 1) {
42543769 2293 mddev->events--;
a8707c08
N		 2294 mddev->can_decrease_events = 0;
2295 } else {
42543769
N		 2296 /* otherwise we have to go forward and ... */
2297 mddev->events ++;
a8707c08 2298 mddev->can_decrease_events = nospares;
42543769 2299 }
1da177e4 2300
403df478
N		 2301 /*
2302 * This 64-bit counter should never wrap.
2303 * Either we are in around ~1 trillion A.C., assuming
2304 * 1 reboot per second, or we have a bug...
2305 */
2306 WARN_ON(mddev->events == 0);
2699b672 2307
dafb20fa 2308 rdev_for_each(rdev, mddev) {
2699b672
N		 2309 if (rdev->badblocks.changed)
2310 any_badblocks_changed++;
de393cde
N		 2311 if (test_bit(Faulty, &rdev->flags))
2312 set_bit(FaultRecorded, &rdev->flags);
2313 }
2699b672 2314
e691063a 2315 sync_sbs(mddev, nospares);
85572d7c 2316 spin_unlock(&mddev->lock);
1da177e4 2317
36a4e1fe
N		 2318 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2319 mdname(mddev), mddev->in_sync);
1da177e4 2320
4ad13663 2321 bitmap_update_sb(mddev->bitmap);
dafb20fa 2322 rdev_for_each(rdev, mddev) {
1da177e4 2323 char b[BDEVNAME_SIZE];
36a4e1fe 2324
42543769
N		 2325 if (rdev->sb_loaded != 1)
2326 continue; /* no noise on spare devices */
1da177e4 2327
f466722c 2328 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2329 md_super_write(mddev,rdev,
0f420358 2330 rdev->sb_start, rdev->sb_size,
7bfa19f2 2331 rdev->sb_page);
36a4e1fe
N		 2332 pr_debug("md: (write) %s's sb offset: %llu\n",
2333 bdevname(rdev->bdev, b),
2334 (unsigned long long)rdev->sb_start);
42543769 2335 rdev->sb_events = mddev->events;
2699b672
N		 2336 if (rdev->badblocks.size) {
2337 md_super_write(mddev, rdev,
2338 rdev->badblocks.sector,
2339 rdev->badblocks.size << 9,
2340 rdev->bb_page);
2341 rdev->badblocks.size = 0;
2342 }
7bfa19f2 2343
f466722c 2344 } else
36a4e1fe
N		 2345 pr_debug("md: %s (skipping faulty)\n",
2346 bdevname(rdev->bdev, b));
d70ed2e4 2347
7bfa19f2 2348 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2349 /* only need to write one superblock... */
2350 break;
2351 }
a9701a30 2352 md_super_wait(mddev);
850b2b42 2353 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2354
85572d7c 2355 spin_lock(&mddev->lock);
850b2b42
N		 2356 if (mddev->in_sync != sync_req ||
2357 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2358 /* have to write it out again */
85572d7c 2359 spin_unlock(&mddev->lock);
06d91a5f
N		 2360 goto repeat;
2361 }
850b2b42 2362 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
85572d7c 2363 spin_unlock(&mddev->lock);
3d310eb7 2364 wake_up(&mddev->sb_wait);
acb180b0
N		 2365 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2366 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2367
dafb20fa 2368 rdev_for_each(rdev, mddev) {
de393cde
N		 2369 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2370 clear_bit(Blocked, &rdev->flags);
2371
2372 if (any_badblocks_changed)
2699b672 2373 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2374 clear_bit(BlockedBadBlocks, &rdev->flags);
2375 wake_up(&rdev->blocked_wait);
2376 }
1da177e4 2377}
1aee41f6 2378EXPORT_SYMBOL(md_update_sb);
1da177e4 2379
a6da4ef8
GR		 2380static int add_bound_rdev(struct md_rdev *rdev)
2381{
2382 struct mddev *mddev = rdev->mddev;
2383 int err = 0;
2384
2385 if (!mddev->pers->hot_remove_disk) {
2386 /* If there is hot_add_disk but no hot_remove_disk
2387 * then added disks for geometry changes,
2388 * and should be added immediately.
2389 */
2390 super_types[mddev->major_version].
2391 validate_super(mddev, rdev);
2392 err = mddev->pers->hot_add_disk(mddev, rdev);
2393 if (err) {
2394 unbind_rdev_from_array(rdev);
2395 export_rdev(rdev);
2396 return err;
2397 }
2398 }
2399 sysfs_notify_dirent_safe(rdev->sysfs_state);
2400
2401 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2402 if (mddev->degraded)
2403 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2404 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2405 md_new_event(mddev);
2406 md_wakeup_thread(mddev->thread);
2407 return 0;
2408}
1da177e4 2409
7f6ce769 2410/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2411 * We want to accept with case. For this we use cmd_match.
2412 */
2413static int cmd_match(const char *cmd, const char *str)
2414{
2415 /* See if cmd, written into a sysfs file, matches
2416 * str. They must either be the same, or cmd can
2417 * have a trailing newline
2418 */
2419 while (*cmd && *str && *cmd == *str) {
2420 cmd++;
2421 str++;
2422 }
2423 if (*cmd == '\n')
2424 cmd++;
2425 if (*str || *cmd)
2426 return 0;
2427 return 1;
2428}
2429
86e6ffdd
N		 2430struct rdev_sysfs_entry {
2431 struct attribute attr;
3cb03002
N		 2432 ssize_t (*show)(struct md_rdev *, char *);
2433 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2434};
2435
2436static ssize_t
3cb03002 2437state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2438{
2439 char *sep = "";
20a49ff6 2440 size_t len = 0;
758bfc8a 2441 unsigned long flags = ACCESS_ONCE(rdev->flags);
86e6ffdd 2442
758bfc8a 2443 if (test_bit(Faulty, &flags) ||
de393cde 2444 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2445 len+= sprintf(page+len, "%sfaulty",sep);
2446 sep = ",";
2447 }
758bfc8a 2448 if (test_bit(In_sync, &flags)) {
86e6ffdd
N		 2449 len += sprintf(page+len, "%sin_sync",sep);
2450 sep = ",";
2451 }
758bfc8a 2452 if (test_bit(WriteMostly, &flags)) {
f655675b
N		 2453 len += sprintf(page+len, "%swrite_mostly",sep);
2454 sep = ",";
2455 }
758bfc8a 2456 if (test_bit(Blocked, &flags) ||
52c64152 2457 (rdev->badblocks.unacked_exist
758bfc8a 2458 && !test_bit(Faulty, &flags))) {
6bfe0b49
DW		 2459 len += sprintf(page+len, "%sblocked", sep);
2460 sep = ",";
2461 }
758bfc8a
N		 2462 if (!test_bit(Faulty, &flags) &&
2463 !test_bit(In_sync, &flags)) {
86e6ffdd
N		 2464 len += sprintf(page+len, "%sspare", sep);
2465 sep = ",";
2466 }
758bfc8a 2467 if (test_bit(WriteErrorSeen, &flags)) {
d7a9d443
N		 2468 len += sprintf(page+len, "%swrite_error", sep);
2469 sep = ",";
2470 }
758bfc8a 2471 if (test_bit(WantReplacement, &flags)) {
2d78f8c4
N		 2472 len += sprintf(page+len, "%swant_replacement", sep);
2473 sep = ",";
2474 }
758bfc8a 2475 if (test_bit(Replacement, &flags)) {
2d78f8c4
N		 2476 len += sprintf(page+len, "%sreplacement", sep);
2477 sep = ",";
2478 }
2479
86e6ffdd
N		 2480 return len+sprintf(page+len, "\n");
2481}
2482
45dc2de1 2483static ssize_t
3cb03002 2484state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2485{
2486 /* can write
de393cde 2487 * faulty - simulates an error
45dc2de1 2488 * remove - disconnects the device
f655675b
N		 2489 * writemostly - sets write_mostly
2490 * -writemostly - clears write_mostly
de393cde
N		 2491 * blocked - sets the Blocked flags
2492 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2493 * insync - sets Insync providing device isn't active
f466722c
N		 2494 * -insync - clear Insync for a device with a slot assigned,
2495 * so that it gets rebuilt based on bitmap
d7a9d443
N		 2496 * write_error - sets WriteErrorSeen
2497 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2498 */
2499 int err = -EINVAL;
2500 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2501 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2502 if (test_bit(Faulty, &rdev->flags))
2503 err = 0;
2504 else
2505 err = -EBUSY;
45dc2de1
N		 2506 } else if (cmd_match(buf, "remove")) {
2507 if (rdev->raid_disk >= 0)
2508 err = -EBUSY;
2509 else {
fd01b88c 2510 struct mddev *mddev = rdev->mddev;
293467aa 2511 if (mddev_is_clustered(mddev))
88bcfef7 2512 md_cluster_ops->remove_disk(mddev, rdev);
fb56dfef 2513 md_kick_rdev_from_array(rdev);
88bcfef7
GR		 2514 if (mddev_is_clustered(mddev))
2515 md_cluster_ops->metadata_update_start(mddev);
3f9d7b0d
N		 2516 if (mddev->pers)
2517 md_update_sb(mddev, 1);
45dc2de1 2518 md_new_event(mddev);
293467aa
GR		 2519 if (mddev_is_clustered(mddev))
2520 md_cluster_ops->metadata_update_finish(mddev);
45dc2de1
N		 2521 err = 0;
2522 }
f655675b
N		 2523 } else if (cmd_match(buf, "writemostly")) {
2524 set_bit(WriteMostly, &rdev->flags);
2525 err = 0;
2526 } else if (cmd_match(buf, "-writemostly")) {
2527 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2528 err = 0;
2529 } else if (cmd_match(buf, "blocked")) {
2530 set_bit(Blocked, &rdev->flags);
2531 err = 0;
2532 } else if (cmd_match(buf, "-blocked")) {
de393cde 2533 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2534 rdev->badblocks.unacked_exist) {
de393cde
N		 2535 /* metadata handler doesn't understand badblocks,
2536 * so we need to fail the device
2537 */
2538 md_error(rdev->mddev, rdev);
2539 }
6bfe0b49 2540 clear_bit(Blocked, &rdev->flags);
de393cde 2541 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2542 wake_up(&rdev->blocked_wait);
2543 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2544 md_wakeup_thread(rdev->mddev->thread);
2545
6d56e278
N		 2546 err = 0;
2547 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2548 set_bit(In_sync, &rdev->flags);
f655675b 2549 err = 0;
f466722c 2550 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
e1960f8c
N		 2551 if (rdev->mddev->pers == NULL) {
2552 clear_bit(In_sync, &rdev->flags);
2553 rdev->saved_raid_disk = rdev->raid_disk;
2554 rdev->raid_disk = -1;
2555 err = 0;
2556 }
d7a9d443
N		 2557 } else if (cmd_match(buf, "write_error")) {
2558 set_bit(WriteErrorSeen, &rdev->flags);
2559 err = 0;
2560 } else if (cmd_match(buf, "-write_error")) {
2561 clear_bit(WriteErrorSeen, &rdev->flags);
2562 err = 0;
2d78f8c4
N		 2563 } else if (cmd_match(buf, "want_replacement")) {
2564 /* Any non-spare device that is not a replacement can
2565 * become want_replacement at any time, but we then need to
2566 * check if recovery is needed.
2567 */
2568 if (rdev->raid_disk >= 0 &&
2569 !test_bit(Replacement, &rdev->flags))
2570 set_bit(WantReplacement, &rdev->flags);
2571 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2572 md_wakeup_thread(rdev->mddev->thread);
2573 err = 0;
2574 } else if (cmd_match(buf, "-want_replacement")) {
2575 /* Clearing 'want_replacement' is always allowed.
2576 * Once replacements starts it is too late though.
2577 */
2578 err = 0;
2579 clear_bit(WantReplacement, &rdev->flags);
2580 } else if (cmd_match(buf, "replacement")) {
2581 /* Can only set a device as a replacement when array has not
2582 * yet been started. Once running, replacement is automatic
2583 * from spares, or by assigning 'slot'.
2584 */
2585 if (rdev->mddev->pers)
2586 err = -EBUSY;
2587 else {
2588 set_bit(Replacement, &rdev->flags);
2589 err = 0;
2590 }
2591 } else if (cmd_match(buf, "-replacement")) {
2592 /* Similarly, can only clear Replacement before start */
2593 if (rdev->mddev->pers)
2594 err = -EBUSY;
2595 else {
2596 clear_bit(Replacement, &rdev->flags);
2597 err = 0;
2598 }
a6da4ef8
GR		 2599 } else if (cmd_match(buf, "re-add")) {
2600 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
97f6cd39
GR		 2601 /* clear_bit is performed _after_ all the devices
2602 * have their local Faulty bit cleared. If any writes
2603 * happen in the meantime in the local node, they
2604 * will land in the local bitmap, which will be synced
2605 * by this node eventually
2606 */
2607 if (!mddev_is_clustered(rdev->mddev) ||
2608 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2609 clear_bit(Faulty, &rdev->flags);
2610 err = add_bound_rdev(rdev);
2611 }
a6da4ef8
GR		 2612 } else
2613 err = -EBUSY;
45dc2de1 2614 }
00bcb4ac
N		 2615 if (!err)
2616 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2617 return err ? err : len;
2618}
80ca3a44 2619static struct rdev_sysfs_entry rdev_state =
750f199e 2620__ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2621
4dbcdc75 2622static ssize_t
3cb03002 2623errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2624{
2625 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2626}
2627
2628static ssize_t
3cb03002 2629errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75 2630{
4c9309c0
AD		 2631 unsigned int n;
2632 int rv;
2633
2634 rv = kstrtouint(buf, 10, &n);
2635 if (rv < 0)
2636 return rv;
2637 atomic_set(&rdev->corrected_errors, n);
2638 return len;
4dbcdc75
N		 2639}
2640static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2641__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2642
014236d2 2643static ssize_t
3cb03002 2644slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2645{
2646 if (rdev->raid_disk < 0)
2647 return sprintf(page, "none\n");
2648 else
2649 return sprintf(page, "%d\n", rdev->raid_disk);
2650}
2651
2652static ssize_t
3cb03002 2653slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2 2654{
4c9309c0 2655 int slot;
c303da6d 2656 int err;
4c9309c0 2657
014236d2
N		 2658 if (strncmp(buf, "none", 4)==0)
2659 slot = -1;
4c9309c0
AD		 2660 else {
2661 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2662 if (err < 0)
2663 return err;
2664 }
6c2fce2e 2665 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2666 /* Setting 'slot' on an active array requires also
2667 * updating the 'rd%d' link, and communicating
2668 * with the personality with ->hot_*_disk.
2669 * For now we only support removing
2670 * failed/spare devices. This normally happens automatically,
2671 * but not when the metadata is externally managed.
2672 */
c303da6d
N		 2673 if (rdev->raid_disk == -1)
2674 return -EEXIST;
2675 /* personality does all needed checks */
01393f3d 2676 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d 2677 return -EINVAL;
746d3207
N		 2678 clear_bit(Blocked, &rdev->flags);
2679 remove_and_add_spares(rdev->mddev, rdev);
2680 if (rdev->raid_disk >= 0)
2681 return -EBUSY;
c303da6d
N		 2682 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2683 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2684 } else if (rdev->mddev->pers) {
6c2fce2e 2685 /* Activating a spare .. or possibly reactivating
6d56e278 2686 * if we ever get bitmaps working here.
6c2fce2e
NB		 2687 */
2688
2689 if (rdev->raid_disk != -1)
2690 return -EBUSY;
2691
c6751b2b
N		 2692 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2693 return -EBUSY;
2694
6c2fce2e
NB		 2695 if (rdev->mddev->pers->hot_add_disk == NULL)
2696 return -EINVAL;
2697
ba1b41b6
N		 2698 if (slot >= rdev->mddev->raid_disks &&
2699 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2700 return -ENOSPC;
2701
6c2fce2e
NB		 2702 rdev->raid_disk = slot;
2703 if (test_bit(In_sync, &rdev->flags))
2704 rdev->saved_raid_disk = slot;
2705 else
2706 rdev->saved_raid_disk = -1;
d30519fc 2707 clear_bit(In_sync, &rdev->flags);
8313b8e5 2708 clear_bit(Bitmap_sync, &rdev->flags);
6c2fce2e
NB		 2709 err = rdev->mddev->pers->
2710 hot_add_disk(rdev->mddev, rdev);
199050ea 2711 if (err) {
6c2fce2e 2712 rdev->raid_disk = -1;
6c2fce2e 2713 return err;
52664732 2714 } else
00bcb4ac 2715 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2716 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2717 /* failure here is OK */;
6c2fce2e 2718 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2719 } else {
ba1b41b6
N		 2720 if (slot >= rdev->mddev->raid_disks &&
2721 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2722 return -ENOSPC;
2723 rdev->raid_disk = slot;
2724 /* assume it is working */
c5d79adb
N		 2725 clear_bit(Faulty, &rdev->flags);
2726 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2727 set_bit(In_sync, &rdev->flags);
00bcb4ac 2728 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2729 }
014236d2
N		 2730 return len;
2731}
2732
014236d2 2733static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2734__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2735
93c8cad0 2736static ssize_t
3cb03002 2737offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2738{
6961ece4 2739 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2740}
2741
2742static ssize_t
3cb03002 2743offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2744{
c6563a8c 2745 unsigned long long offset;
b29bebd6 2746 if (kstrtoull(buf, 10, &offset) < 0)
93c8cad0 2747 return -EINVAL;
8ed0a521 2748 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2749 return -EBUSY;
dd8ac336 2750 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2751 /* Must set offset before size, so overlap checks
2752 * can be sane */
2753 return -EBUSY;
93c8cad0 2754 rdev->data_offset = offset;
25f7fd47 2755 rdev->new_data_offset = offset;
93c8cad0
N		 2756 return len;
2757}
2758
2759static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2760__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2761
c6563a8c
N		 2762static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2763{
2764 return sprintf(page, "%llu\n",
2765 (unsigned long long)rdev->new_data_offset);
2766}
2767
2768static ssize_t new_offset_store(struct md_rdev *rdev,
2769 const char *buf, size_t len)
2770{
2771 unsigned long long new_offset;
2772 struct mddev *mddev = rdev->mddev;
2773
b29bebd6 2774 if (kstrtoull(buf, 10, &new_offset) < 0)
c6563a8c
N		 2775 return -EINVAL;
2776
f851b60d
N		 2777 if (mddev->sync_thread ||
2778 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
c6563a8c
N		 2779 return -EBUSY;
2780 if (new_offset == rdev->data_offset)
2781 /* reset is always permitted */
2782 ;
2783 else if (new_offset > rdev->data_offset) {
2784 /* must not push array size beyond rdev_sectors */
2785 if (new_offset - rdev->data_offset
2786 + mddev->dev_sectors > rdev->sectors)
2787 return -E2BIG;
2788 }
2789 /* Metadata worries about other space details. */
2790
2791 /* decreasing the offset is inconsistent with a backwards
2792 * reshape.
2793 */
2794 if (new_offset < rdev->data_offset &&
2795 mddev->reshape_backwards)
2796 return -EINVAL;
2797 /* Increasing offset is inconsistent with forwards
2798 * reshape. reshape_direction should be set to
2799 * 'backwards' first.
2800 */
2801 if (new_offset > rdev->data_offset &&
2802 !mddev->reshape_backwards)
2803 return -EINVAL;
2804
2805 if (mddev->pers && mddev->persistent &&
2806 !super_types[mddev->major_version]
2807 .allow_new_offset(rdev, new_offset))
2808 return -E2BIG;
2809 rdev->new_data_offset = new_offset;
2810 if (new_offset > rdev->data_offset)
2811 mddev->reshape_backwards = 1;
2812 else if (new_offset < rdev->data_offset)
2813 mddev->reshape_backwards = 0;
2814
2815 return len;
2816}
2817static struct rdev_sysfs_entry rdev_new_offset =
2818__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2819
83303b61 2820static ssize_t
3cb03002 2821rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2822{
dd8ac336 2823 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2824}
2825
c5d79adb
N		 2826static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2827{
2828 /* check if two start/length pairs overlap */
2829 if (s1+l1 <= s2)
2830 return 0;
2831 if (s2+l2 <= s1)
2832 return 0;
2833 return 1;
2834}
2835
b522adcd
DW		 2836static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2837{
2838 unsigned long long blocks;
2839 sector_t new;
2840
b29bebd6 2841 if (kstrtoull(buf, 10, &blocks) < 0)
b522adcd
DW		 2842 return -EINVAL;
2843
2844 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2845 return -EINVAL; /* sector conversion overflow */
2846
2847 new = blocks * 2;
2848 if (new != blocks * 2)
2849 return -EINVAL; /* unsigned long long to sector_t overflow */
2850
2851 *sectors = new;
2852 return 0;
2853}
2854
83303b61 2855static ssize_t
3cb03002 2856rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2857{
fd01b88c 2858 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2859 sector_t oldsectors = rdev->sectors;
b522adcd 2860 sector_t sectors;
27c529bb 2861
b522adcd 2862 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2863 return -EINVAL;
c6563a8c
N		 2864 if (rdev->data_offset != rdev->new_data_offset)
2865 return -EINVAL; /* too confusing */
0cd17fec 2866 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2867 if (my_mddev->persistent) {
dd8ac336
AN		 2868 sectors = super_types[my_mddev->major_version].
2869 rdev_size_change(rdev, sectors);
2870 if (!sectors)
0cd17fec 2871 return -EBUSY;
dd8ac336 2872 } else if (!sectors)
77304d2a 2873 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2874 rdev->data_offset;
a6468539
N		 2875 if (!my_mddev->pers->resize)
2876 /* Cannot change size for RAID0 or Linear etc */
2877 return -EINVAL;
0cd17fec 2878 }
dd8ac336 2879 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2880 return -EINVAL; /* component must fit device */
0cd17fec 2881
dd8ac336
AN		 2882 rdev->sectors = sectors;
2883 if (sectors > oldsectors && my_mddev->external) {
8b1afc3d
N		 2884 /* Need to check that all other rdevs with the same
2885 * ->bdev do not overlap. 'rcu' is sufficient to walk
2886 * the rdev lists safely.
2887 * This check does not provide a hard guarantee, it
2888 * just helps avoid dangerous mistakes.
c5d79adb 2889 */
fd01b88c 2890 struct mddev *mddev;
c5d79adb 2891 int overlap = 0;
159ec1fc 2892 struct list_head *tmp;
c5d79adb 2893
8b1afc3d 2894 rcu_read_lock();
29ac4aa3 2895 for_each_mddev(mddev, tmp) {
3cb03002 2896 struct md_rdev *rdev2;
c5d79adb 2897
dafb20fa 2898 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 2899 if (rdev->bdev == rdev2->bdev &&
2900 rdev != rdev2 &&
2901 overlaps(rdev->data_offset, rdev->sectors,
2902 rdev2->data_offset,
2903 rdev2->sectors)) {
c5d79adb
N		 2904 overlap = 1;
2905 break;
2906 }
c5d79adb
N		 2907 if (overlap) {
2908 mddev_put(mddev);
2909 break;
2910 }
2911 }
8b1afc3d 2912 rcu_read_unlock();
c5d79adb
N		 2913 if (overlap) {
2914 /* Someone else could have slipped in a size
2915 * change here, but doing so is just silly.
dd8ac336 2916 * We put oldsectors back because we *know* it is
c5d79adb
N		 2917 * safe, and trust userspace not to race with
2918 * itself
2919 */
dd8ac336 2920 rdev->sectors = oldsectors;
c5d79adb
N		 2921 return -EBUSY;
2922 }
2923 }
83303b61
N		 2924 return len;
2925}
2926
2927static struct rdev_sysfs_entry rdev_size =
80ca3a44 2928__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2929
3cb03002 2930static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 2931{
2932 unsigned long long recovery_start = rdev->recovery_offset;
2933
2934 if (test_bit(In_sync, &rdev->flags) ||
2935 recovery_start == MaxSector)
2936 return sprintf(page, "none\n");
2937
2938 return sprintf(page, "%llu\n", recovery_start);
2939}
2940
3cb03002 2941static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 2942{
2943 unsigned long long recovery_start;
2944
2945 if (cmd_match(buf, "none"))
2946 recovery_start = MaxSector;
b29bebd6 2947 else if (kstrtoull(buf, 10, &recovery_start))
06e3c817
DW		 2948 return -EINVAL;
2949
2950 if (rdev->mddev->pers &&
2951 rdev->raid_disk >= 0)
2952 return -EBUSY;
2953
2954 rdev->recovery_offset = recovery_start;
2955 if (recovery_start == MaxSector)
2956 set_bit(In_sync, &rdev->flags);
2957 else
2958 clear_bit(In_sync, &rdev->flags);
2959 return len;
2960}
2961
2962static struct rdev_sysfs_entry rdev_recovery_start =
2963__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2964
16c791a5
N		 2965static ssize_t
2966badblocks_show(struct badblocks *bb, char *page, int unack);
2967static ssize_t
2968badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2969
3cb03002 2970static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2971{
2972 return badblocks_show(&rdev->badblocks, page, 0);
2973}
3cb03002 2974static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 2975{
de393cde
N		 2976 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2977 /* Maybe that ack was all we needed */
2978 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2979 wake_up(&rdev->blocked_wait);
2980 return rv;
16c791a5
N		 2981}
2982static struct rdev_sysfs_entry rdev_bad_blocks =
2983__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2984
3cb03002 2985static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2986{
2987 return badblocks_show(&rdev->badblocks, page, 1);
2988}
3cb03002 2989static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 2990{
2991 return badblocks_store(&rdev->badblocks, page, len, 1);
2992}
2993static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2994__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2995
86e6ffdd
N		 2996static struct attribute *rdev_default_attrs[] = {
2997 &rdev_state.attr,
4dbcdc75 2998 &rdev_errors.attr,
014236d2 2999 &rdev_slot.attr,
93c8cad0 3000 &rdev_offset.attr,
c6563a8c 3001 &rdev_new_offset.attr,
83303b61 3002 &rdev_size.attr,
06e3c817 3003 &rdev_recovery_start.attr,
16c791a5
N		 3004 &rdev_bad_blocks.attr,
3005 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 3006 NULL,
3007};
3008static ssize_t
3009rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3010{
3011 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3012 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
86e6ffdd
N		 3013
3014 if (!entry->show)
3015 return -EIO;
758bfc8a
N		 3016 if (!rdev->mddev)
3017 return -EBUSY;
3018 return entry->show(rdev, page);
86e6ffdd
N		 3019}
3020
3021static ssize_t
3022rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3023 const char *page, size_t length)
3024{
3025 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3026 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3027 ssize_t rv;
fd01b88c 3028 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3029
3030 if (!entry->store)
3031 return -EIO;
67463acb
N		 3032 if (!capable(CAP_SYS_ADMIN))
3033 return -EACCES;
27c529bb 3034 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3035 if (!rv) {
27c529bb
N		 3036 if (rdev->mddev == NULL)
3037 rv = -EBUSY;
3038 else
3039 rv = entry->store(rdev, page, length);
6a51830e 3040 mddev_unlock(mddev);
ca388059
N		 3041 }
3042 return rv;
86e6ffdd
N		 3043}
3044
3045static void rdev_free(struct kobject *ko)
3046{
3cb03002 3047 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3048 kfree(rdev);
3049}
52cf25d0 3050static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3051 .show = rdev_attr_show,
3052 .store = rdev_attr_store,
3053};
3054static struct kobj_type rdev_ktype = {
3055 .release = rdev_free,
3056 .sysfs_ops = &rdev_sysfs_ops,
3057 .default_attrs = rdev_default_attrs,
3058};
3059
3cb03002 3060int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3061{
3062 rdev->desc_nr = -1;
3063 rdev->saved_raid_disk = -1;
3064 rdev->raid_disk = -1;
3065 rdev->flags = 0;
3066 rdev->data_offset = 0;
c6563a8c 3067 rdev->new_data_offset = 0;
e8bb9a83
N		 3068 rdev->sb_events = 0;
3069 rdev->last_read_error.tv_sec = 0;
3070 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3071 rdev->sb_loaded = 0;
3072 rdev->bb_page = NULL;
e8bb9a83
N		 3073 atomic_set(&rdev->nr_pending, 0);
3074 atomic_set(&rdev->read_errors, 0);
3075 atomic_set(&rdev->corrected_errors, 0);
3076
3077 INIT_LIST_HEAD(&rdev->same_set);
3078 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3079
3080 /* Add space to store bad block list.
3081 * This reserves the space even on arrays where it cannot
3082 * be used - I wonder if that matters
3083 */
3084 rdev->badblocks.count = 0;
486adf72 3085 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
2230dfe4
N		 3086 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3087 seqlock_init(&rdev->badblocks.lock);
3088 if (rdev->badblocks.page == NULL)
3089 return -ENOMEM;
3090
3091 return 0;
e8bb9a83
N		 3092}
3093EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3094/*
3095 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3096 *
3097 * mark the device faulty if:
3098 *
3099 * - the device is nonexistent (zero size)
3100 * - the device has no valid superblock
3101 *
3102 * a faulty rdev _never_ has rdev->sb set.
3103 */
3cb03002 3104static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3105{
3106 char b[BDEVNAME_SIZE];
3107 int err;
3cb03002 3108 struct md_rdev *rdev;
1da177e4
LT		 3109 sector_t size;
3110
9ffae0cf 3111 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3112 if (!rdev) {
3113 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3114 return ERR_PTR(-ENOMEM);
3115 }
1da177e4 3116
2230dfe4
N		 3117 err = md_rdev_init(rdev);
3118 if (err)
3119 goto abort_free;
3120 err = alloc_disk_sb(rdev);
3121 if (err)
1da177e4
LT		 3122 goto abort_free;
3123
c5d79adb 3124 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3125 if (err)
3126 goto abort_free;
3127
f9cb074b 3128 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3129
77304d2a 3130 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4 3131 if (!size) {
f72ffdd6 3132 printk(KERN_WARNING
1da177e4
LT		 3133 "md: %s has zero or unknown size, marking faulty!\n",
3134 bdevname(rdev->bdev,b));
3135 err = -EINVAL;
3136 goto abort_free;
3137 }
3138
3139 if (super_format >= 0) {
3140 err = super_types[super_format].
3141 load_super(rdev, NULL, super_minor);
3142 if (err == -EINVAL) {
df968c4e
N		 3143 printk(KERN_WARNING
3144 "md: %s does not have a valid v%d.%d "
3145 "superblock, not importing!\n",
3146 bdevname(rdev->bdev,b),
3147 super_format, super_minor);
1da177e4
LT		 3148 goto abort_free;
3149 }
3150 if (err < 0) {
f72ffdd6 3151 printk(KERN_WARNING
1da177e4
LT		 3152 "md: could not read %s's sb, not importing!\n",
3153 bdevname(rdev->bdev,b));
3154 goto abort_free;
3155 }
3156 }
6bfe0b49 3157
1da177e4
LT		 3158 return rdev;
3159
3160abort_free:
2699b672
N		 3161 if (rdev->bdev)
3162 unlock_rdev(rdev);
545c8795 3163 md_rdev_clear(rdev);
1da177e4
LT		 3164 kfree(rdev);
3165 return ERR_PTR(err);
3166}
3167
3168/*
3169 * Check a full RAID array for plausibility
3170 */
3171
f72ffdd6 3172static void analyze_sbs(struct mddev *mddev)
1da177e4
LT		 3173{
3174 int i;
3cb03002 3175 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3176 char b[BDEVNAME_SIZE];
3177
3178 freshest = NULL;
dafb20fa 3179 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3180 switch (super_types[mddev->major_version].
3181 load_super(rdev, freshest, mddev->minor_version)) {
3182 case 1:
3183 freshest = rdev;
3184 break;
3185 case 0:
3186 break;
3187 default:
3188 printk( KERN_ERR \
3189 "md: fatal superblock inconsistency in %s"
f72ffdd6 3190 " -- removing from array\n",
1da177e4 3191 bdevname(rdev->bdev,b));
fb56dfef 3192 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3193 }
3194
1da177e4
LT		 3195 super_types[mddev->major_version].
3196 validate_super(mddev, freshest);
3197
3198 i = 0;
dafb20fa 3199 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3200 if (mddev->max_disks &&
3201 (rdev->desc_nr >= mddev->max_disks ||
3202 i > mddev->max_disks)) {
de01dfad
N		 3203 printk(KERN_WARNING
3204 "md: %s: %s: only %d devices permitted\n",
3205 mdname(mddev), bdevname(rdev->bdev, b),
3206 mddev->max_disks);
fb56dfef 3207 md_kick_rdev_from_array(rdev);
de01dfad
N		 3208 continue;
3209 }
1aee41f6 3210 if (rdev != freshest) {
1da177e4
LT		 3211 if (super_types[mddev->major_version].
3212 validate_super(mddev, rdev)) {
3213 printk(KERN_WARNING "md: kicking non-fresh %s"
3214 " from array!\n",
3215 bdevname(rdev->bdev,b));
fb56dfef 3216 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3217 continue;
3218 }
1aee41f6
GR		 3219 /* No device should have a Candidate flag
3220 * when reading devices
3221 */
3222 if (test_bit(Candidate, &rdev->flags)) {
3223 pr_info("md: kicking Cluster Candidate %s from array!\n",
3224 bdevname(rdev->bdev, b));
fb56dfef 3225 md_kick_rdev_from_array(rdev);
1aee41f6
GR		 3226 }
3227 }
1da177e4
LT		 3228 if (mddev->level == LEVEL_MULTIPATH) {
3229 rdev->desc_nr = i++;
3230 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3231 set_bit(In_sync, &rdev->flags);
5e5e3e78 3232 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3233 rdev->raid_disk = -1;
3234 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3235 }
3236 }
1da177e4
LT		 3237}
3238
72e02075
N		 3239/* Read a fixed-point number.
3240 * Numbers in sysfs attributes should be in "standard" units where
3241 * possible, so time should be in seconds.
f72ffdd6 3242 * However we internally use a a much smaller unit such as
72e02075
N		 3243 * milliseconds or jiffies.
3244 * This function takes a decimal number with a possible fractional
3245 * component, and produces an integer which is the result of
3246 * multiplying that number by 10^'scale'.
3247 * all without any floating-point arithmetic.
3248 */
3249int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3250{
3251 unsigned long result = 0;
3252 long decimals = -1;
3253 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3254 if (*cp == '.')
3255 decimals = 0;
3256 else if (decimals < scale) {
3257 unsigned int value;
3258 value = *cp - '0';
3259 result = result * 10 + value;
3260 if (decimals >= 0)
3261 decimals++;
3262 }
3263 cp++;
3264 }
3265 if (*cp == '\n')
3266 cp++;
3267 if (*cp)
3268 return -EINVAL;
3269 if (decimals < 0)
3270 decimals = 0;
3271 while (decimals < scale) {
3272 result *= 10;
3273 decimals ++;
3274 }
3275 *res = result;
3276 return 0;
3277}
3278
19052c0e
N		 3279static void md_safemode_timeout(unsigned long data);
3280
16f17b39 3281static ssize_t
fd01b88c 3282safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3283{
3284 int msec = (mddev->safemode_delay*1000)/HZ;
3285 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3286}
3287static ssize_t
fd01b88c 3288safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3289{
16f17b39 3290 unsigned long msec;
97ce0a7f 3291
72e02075 3292 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3293 return -EINVAL;
16f17b39
N		 3294 if (msec == 0)
3295 mddev->safemode_delay = 0;
3296 else {
19052c0e 3297 unsigned long old_delay = mddev->safemode_delay;
1b30e66f
N		 3298 unsigned long new_delay = (msec*HZ)/1000;
3299
3300 if (new_delay == 0)
3301 new_delay = 1;
3302 mddev->safemode_delay = new_delay;
3303 if (new_delay < old_delay || old_delay == 0)
3304 mod_timer(&mddev->safemode_timer, jiffies+1);
16f17b39
N		 3305 }
3306 return len;
3307}
3308static struct md_sysfs_entry md_safe_delay =
80ca3a44 3309__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3310
eae1701f 3311static ssize_t
fd01b88c 3312level_show(struct mddev *mddev, char *page)
eae1701f 3313{
36d091f4
N		 3314 struct md_personality *p;
3315 int ret;
3316 spin_lock(&mddev->lock);
3317 p = mddev->pers;
d9d166c2 3318 if (p)
36d091f4 3319 ret = sprintf(page, "%s\n", p->name);
d9d166c2 3320 else if (mddev->clevel[0])
36d091f4 3321 ret = sprintf(page, "%s\n", mddev->clevel);
d9d166c2 3322 else if (mddev->level != LEVEL_NONE)
36d091f4 3323 ret = sprintf(page, "%d\n", mddev->level);
d9d166c2 3324 else
36d091f4
N		 3325 ret = 0;
3326 spin_unlock(&mddev->lock);
3327 return ret;
eae1701f
N		 3328}
3329
d9d166c2 3330static ssize_t
fd01b88c 3331level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3332{
f2859af6 3333 char clevel[16];
6791875e
N		 3334 ssize_t rv;
3335 size_t slen = len;
db721d32 3336 struct md_personality *pers, *oldpers;
f2859af6 3337 long level;
db721d32 3338 void *priv, *oldpriv;
3cb03002 3339 struct md_rdev *rdev;
245f46c2 3340
6791875e
N		 3341 if (slen == 0 || slen >= sizeof(clevel))
3342 return -EINVAL;
3343
3344 rv = mddev_lock(mddev);
3345 if (rv)
3346 return rv;
3347
245f46c2 3348 if (mddev->pers == NULL) {
6791875e
N		 3349 strncpy(mddev->clevel, buf, slen);
3350 if (mddev->clevel[slen-1] == '\n')
3351 slen--;
3352 mddev->clevel[slen] = 0;
245f46c2 3353 mddev->level = LEVEL_NONE;
6791875e
N		 3354 rv = len;
3355 goto out_unlock;
245f46c2 3356 }
6791875e 3357 rv = -EROFS;
bd8839e0 3358 if (mddev->ro)
6791875e 3359 goto out_unlock;
245f46c2
N		 3360
3361 /* request to change the personality. Need to ensure:
3362 * - array is not engaged in resync/recovery/reshape
3363 * - old personality can be suspended
3364 * - new personality will access other array.
3365 */
3366
6791875e 3367 rv = -EBUSY;
bb4f1e9d 3368 if (mddev->sync_thread ||
f851b60d 3369 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
bb4f1e9d
N		 3370 mddev->reshape_position != MaxSector ||
3371 mddev->sysfs_active)
6791875e 3372 goto out_unlock;
245f46c2 3373
6791875e 3374 rv = -EINVAL;
245f46c2
N		 3375 if (!mddev->pers->quiesce) {
3376 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3377 mdname(mddev), mddev->pers->name);
6791875e 3378 goto out_unlock;
245f46c2
N		 3379 }
3380
3381 /* Now find the new personality */
6791875e
N		 3382 strncpy(clevel, buf, slen);
3383 if (clevel[slen-1] == '\n')
3384 slen--;
3385 clevel[slen] = 0;
b29bebd6 3386 if (kstrtol(clevel, 10, &level))
f2859af6 3387 level = LEVEL_NONE;
245f46c2 3388
f2859af6
DW		 3389 if (request_module("md-%s", clevel) != 0)
3390 request_module("md-level-%s", clevel);
245f46c2 3391 spin_lock(&pers_lock);
f2859af6 3392 pers = find_pers(level, clevel);
245f46c2
N		 3393 if (!pers || !try_module_get(pers->owner)) {
3394 spin_unlock(&pers_lock);
f2859af6 3395 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
6791875e
N		 3396 rv = -EINVAL;
3397 goto out_unlock;
245f46c2
N		 3398 }
3399 spin_unlock(&pers_lock);
3400
3401 if (pers == mddev->pers) {
3402 /* Nothing to do! */
3403 module_put(pers->owner);
6791875e
N		 3404 rv = len;
3405 goto out_unlock;
245f46c2
N		 3406 }
3407 if (!pers->takeover) {
3408 module_put(pers->owner);
3409 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3410 mdname(mddev), clevel);
6791875e
N		 3411 rv = -EINVAL;
3412 goto out_unlock;
245f46c2
N		 3413 }
3414
dafb20fa 3415 rdev_for_each(rdev, mddev)
e93f68a1
N		 3416 rdev->new_raid_disk = rdev->raid_disk;
3417
245f46c2
N		 3418 /* ->takeover must set new_* and/or delta_disks
3419 * if it succeeds, and may set them when it fails.
3420 */
3421 priv = pers->takeover(mddev);
3422 if (IS_ERR(priv)) {
3423 mddev->new_level = mddev->level;
3424 mddev->new_layout = mddev->layout;
664e7c41 3425 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3426 mddev->raid_disks -= mddev->delta_disks;
3427 mddev->delta_disks = 0;
2c810cdd 3428 mddev->reshape_backwards = 0;
245f46c2
N		 3429 module_put(pers->owner);
3430 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3431 mdname(mddev), clevel);
6791875e
N		 3432 rv = PTR_ERR(priv);
3433 goto out_unlock;
245f46c2
N		 3434 }
3435
3436 /* Looks like we have a winner */
3437 mddev_suspend(mddev);
5aa61f42 3438 mddev_detach(mddev);
36d091f4
N		 3439
3440 spin_lock(&mddev->lock);
db721d32
N		 3441 oldpers = mddev->pers;
3442 oldpriv = mddev->private;
3443 mddev->pers = pers;
3444 mddev->private = priv;
3445 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3446 mddev->level = mddev->new_level;
3447 mddev->layout = mddev->new_layout;
3448 mddev->chunk_sectors = mddev->new_chunk_sectors;
3449 mddev->delta_disks = 0;
3450 mddev->reshape_backwards = 0;
3451 mddev->degraded = 0;
36d091f4 3452 spin_unlock(&mddev->lock);
db721d32
N		 3453
3454 if (oldpers->sync_request == NULL &&
3455 mddev->external) {
3456 /* We are converting from a no-redundancy array
3457 * to a redundancy array and metadata is managed
3458 * externally so we need to be sure that writes
3459 * won't block due to a need to transition
3460 * clean->dirty
3461 * until external management is started.
3462 */
3463 mddev->in_sync = 0;
3464 mddev->safemode_delay = 0;
3465 mddev->safemode = 0;
3466 }
f72ffdd6 3467
db721d32
N		 3468 oldpers->free(mddev, oldpriv);
3469
3470 if (oldpers->sync_request == NULL &&
a64c876f
N		 3471 pers->sync_request != NULL) {
3472 /* need to add the md_redundancy_group */
3473 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3474 printk(KERN_WARNING
3475 "md: cannot register extra attributes for %s\n",
3476 mdname(mddev));
388975cc 3477 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
f72ffdd6 3478 }
db721d32 3479 if (oldpers->sync_request != NULL &&
a64c876f
N		 3480 pers->sync_request == NULL) {
3481 /* need to remove the md_redundancy_group */
3482 if (mddev->to_remove == NULL)
3483 mddev->to_remove = &md_redundancy_group;
3484 }
3485
dafb20fa 3486 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3487 if (rdev->raid_disk < 0)
3488 continue;
bf2cb0da 3489 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3490 rdev->new_raid_disk = -1;
3491 if (rdev->new_raid_disk == rdev->raid_disk)
3492 continue;
36fad858 3493 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3494 }
dafb20fa 3495 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3496 if (rdev->raid_disk < 0)
3497 continue;
3498 if (rdev->new_raid_disk == rdev->raid_disk)
3499 continue;
3500 rdev->raid_disk = rdev->new_raid_disk;
3501 if (rdev->raid_disk < 0)
3a981b03 3502 clear_bit(In_sync, &rdev->flags);
e93f68a1 3503 else {
36fad858
NK		 3504 if (sysfs_link_rdev(mddev, rdev))
3505 printk(KERN_WARNING "md: cannot register rd%d"
3506 " for %s after level change\n",
3507 rdev->raid_disk, mdname(mddev));
3a981b03 3508 }
e93f68a1
N		 3509 }
3510
db721d32 3511 if (pers->sync_request == NULL) {
9af204cf
TM		 3512 /* this is now an array without redundancy, so
3513 * it must always be in_sync
3514 */
3515 mddev->in_sync = 1;
3516 del_timer_sync(&mddev->safemode_timer);
3517 }
02e5f5c0 3518 blk_set_stacking_limits(&mddev->queue->limits);
245f46c2 3519 pers->run(mddev);
245f46c2 3520 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3521 mddev_resume(mddev);
830778a1
N		 3522 if (!mddev->thread)
3523 md_update_sb(mddev, 1);
5cac7861 3524 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3525 md_new_event(mddev);
6791875e
N		 3526 rv = len;
3527out_unlock:
3528 mddev_unlock(mddev);
d9d166c2
N		 3529 return rv;
3530}
3531
3532static struct md_sysfs_entry md_level =
80ca3a44 3533__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3534
d4dbd025 3535static ssize_t
fd01b88c 3536layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3537{
3538 /* just a number, not meaningful for all levels */
08a02ecd
N		 3539 if (mddev->reshape_position != MaxSector &&
3540 mddev->layout != mddev->new_layout)
3541 return sprintf(page, "%d (%d)\n",
3542 mddev->new_layout, mddev->layout);
d4dbd025
N		 3543 return sprintf(page, "%d\n", mddev->layout);
3544}
3545
3546static ssize_t
fd01b88c 3547layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025 3548{
4c9309c0 3549 unsigned int n;
6791875e 3550 int err;
d4dbd025 3551
4c9309c0
AD		 3552 err = kstrtouint(buf, 10, &n);
3553 if (err < 0)
3554 return err;
6791875e
N		 3555 err = mddev_lock(mddev);
3556 if (err)
3557 return err;
d4dbd025 3558
b3546035 3559 if (mddev->pers) {
50ac168a 3560 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3561 err = -EBUSY;
3562 else if (mddev->ro)
3563 err = -EROFS;
3564 else {
3565 mddev->new_layout = n;
3566 err = mddev->pers->check_reshape(mddev);
3567 if (err)
3568 mddev->new_layout = mddev->layout;
597a711b 3569 }
b3546035 3570 } else {
08a02ecd 3571 mddev->new_layout = n;
b3546035
N		 3572 if (mddev->reshape_position == MaxSector)
3573 mddev->layout = n;
3574 }
6791875e
N		 3575 mddev_unlock(mddev);
3576 return err ?: len;
d4dbd025
N		 3577}
3578static struct md_sysfs_entry md_layout =
80ca3a44 3579__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025 3580
eae1701f 3581static ssize_t
fd01b88c 3582raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3583{
bb636547
N		 3584 if (mddev->raid_disks == 0)
3585 return 0;
08a02ecd
N		 3586 if (mddev->reshape_position != MaxSector &&
3587 mddev->delta_disks != 0)
3588 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3589 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3590 return sprintf(page, "%d\n", mddev->raid_disks);
3591}
3592
fd01b88c 3593static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3594
3595static ssize_t
fd01b88c 3596raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3597{
4c9309c0 3598 unsigned int n;
6791875e 3599 int err;
da943b99 3600
4c9309c0
AD		 3601 err = kstrtouint(buf, 10, &n);
3602 if (err < 0)
3603 return err;
da943b99 3604
6791875e
N		 3605 err = mddev_lock(mddev);
3606 if (err)
3607 return err;
da943b99 3608 if (mddev->pers)
6791875e 3609 err = update_raid_disks(mddev, n);
08a02ecd 3610 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3611 struct md_rdev *rdev;
08a02ecd 3612 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c 3613
6791875e 3614 err = -EINVAL;
c6563a8c
N		 3615 rdev_for_each(rdev, mddev) {
3616 if (olddisks < n &&
3617 rdev->data_offset < rdev->new_data_offset)
6791875e 3618 goto out_unlock;
c6563a8c
N		 3619 if (olddisks > n &&
3620 rdev->data_offset > rdev->new_data_offset)
6791875e 3621 goto out_unlock;
c6563a8c 3622 }
6791875e 3623 err = 0;
08a02ecd
N		 3624 mddev->delta_disks = n - olddisks;
3625 mddev->raid_disks = n;
2c810cdd 3626 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3627 } else
da943b99 3628 mddev->raid_disks = n;
6791875e
N		 3629out_unlock:
3630 mddev_unlock(mddev);
3631 return err ? err : len;
da943b99
N		 3632}
3633static struct md_sysfs_entry md_raid_disks =
80ca3a44 3634__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3635
3b34380a 3636static ssize_t
fd01b88c 3637chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3638{
08a02ecd 3639 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3640 mddev->chunk_sectors != mddev->new_chunk_sectors)
3641 return sprintf(page, "%d (%d)\n",
3642 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3643 mddev->chunk_sectors << 9);
3644 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3645}
3646
3647static ssize_t
fd01b88c 3648chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3649{
4c9309c0 3650 unsigned long n;
6791875e 3651 int err;
3b34380a 3652
4c9309c0
AD		 3653 err = kstrtoul(buf, 10, &n);
3654 if (err < 0)
3655 return err;
3b34380a 3656
6791875e
N		 3657 err = mddev_lock(mddev);
3658 if (err)
3659 return err;
b3546035 3660 if (mddev->pers) {
50ac168a 3661 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3662 err = -EBUSY;
3663 else if (mddev->ro)
3664 err = -EROFS;
3665 else {
3666 mddev->new_chunk_sectors = n >> 9;
3667 err = mddev->pers->check_reshape(mddev);
3668 if (err)
3669 mddev->new_chunk_sectors = mddev->chunk_sectors;
597a711b 3670 }
b3546035 3671 } else {
664e7c41 3672 mddev->new_chunk_sectors = n >> 9;
b3546035 3673 if (mddev->reshape_position == MaxSector)
9d8f0363 3674 mddev->chunk_sectors = n >> 9;
b3546035 3675 }
6791875e
N		 3676 mddev_unlock(mddev);
3677 return err ?: len;
3b34380a
N		 3678}
3679static struct md_sysfs_entry md_chunk_size =
80ca3a44 3680__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3681
a94213b1 3682static ssize_t
fd01b88c 3683resync_start_show(struct mddev *mddev, char *page)
a94213b1 3684{
d1a7c503
N		 3685 if (mddev->recovery_cp == MaxSector)
3686 return sprintf(page, "none\n");
a94213b1
N		 3687 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3688}
3689
3690static ssize_t
fd01b88c 3691resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3692{
4c9309c0 3693 unsigned long long n;
6791875e 3694 int err;
4c9309c0
AD		 3695
3696 if (cmd_match(buf, "none"))
3697 n = MaxSector;
3698 else {
3699 err = kstrtoull(buf, 10, &n);
3700 if (err < 0)
3701 return err;
3702 if (n != (sector_t)n)
3703 return -EINVAL;
3704 }
a94213b1 3705
6791875e
N		 3706 err = mddev_lock(mddev);
3707 if (err)
3708 return err;
b098636c 3709 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6791875e 3710 err = -EBUSY;
a94213b1 3711
6791875e
N		 3712 if (!err) {
3713 mddev->recovery_cp = n;
3714 if (mddev->pers)
3715 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3716 }
3717 mddev_unlock(mddev);
3718 return err ?: len;
a94213b1
N		 3719}
3720static struct md_sysfs_entry md_resync_start =
750f199e
N		 3721__ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3722 resync_start_show, resync_start_store);
a94213b1 3723
9e653b63
N		 3724/*
3725 * The array state can be:
3726 *
3727 * clear
3728 * No devices, no size, no level
3729 * Equivalent to STOP_ARRAY ioctl
3730 * inactive
3731 * May have some settings, but array is not active
3732 * all IO results in error
3733 * When written, doesn't tear down array, but just stops it
3734 * suspended (not supported yet)
3735 * All IO requests will block. The array can be reconfigured.
910d8cb3 3736 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3737 * readonly
3738 * no resync can happen. no superblocks get written.
3739 * write requests fail
3740 * read-auto
3741 * like readonly, but behaves like 'clean' on a write request.
3742 *
3743 * clean - no pending writes, but otherwise active.
3744 * When written to inactive array, starts without resync
3745 * If a write request arrives then
3746 * if metadata is known, mark 'dirty' and switch to 'active'.
3747 * if not known, block and switch to write-pending
3748 * If written to an active array that has pending writes, then fails.
3749 * active
3750 * fully active: IO and resync can be happening.
3751 * When written to inactive array, starts with resync
3752 *
3753 * write-pending
3754 * clean, but writes are blocked waiting for 'active' to be written.
3755 *
3756 * active-idle
3757 * like active, but no writes have been seen for a while (100msec).
3758 *
3759 */
3760enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3761 write_pending, active_idle, bad_word};
05381954 3762static char *array_states[] = {
9e653b63
N		 3763 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3764 "write-pending", "active-idle", NULL };
3765
3766static int match_word(const char *word, char **list)
3767{
3768 int n;
3769 for (n=0; list[n]; n++)
3770 if (cmd_match(word, list[n]))
3771 break;
3772 return n;
3773}
3774
3775static ssize_t
fd01b88c 3776array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3777{
3778 enum array_state st = inactive;
3779
3780 if (mddev->pers)
3781 switch(mddev->ro) {
3782 case 1:
3783 st = readonly;
3784 break;
3785 case 2:
3786 st = read_auto;
3787 break;
3788 case 0:
3789 if (mddev->in_sync)
3790 st = clean;
070dc6dd 3791 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3792 st = write_pending;
9e653b63
N		 3793 else if (mddev->safemode)
3794 st = active_idle;
3795 else
3796 st = active;
3797 }
3798 else {
3799 if (list_empty(&mddev->disks) &&
3800 mddev->raid_disks == 0 &&
58c0fed4 3801 mddev->dev_sectors == 0)
9e653b63
N		 3802 st = clear;
3803 else
3804 st = inactive;
3805 }
3806 return sprintf(page, "%s\n", array_states[st]);
3807}
3808
f72ffdd6
N		 3809static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3810static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3811static int do_md_run(struct mddev *mddev);
fd01b88c 3812static int restart_array(struct mddev *mddev);
9e653b63
N		 3813
3814static ssize_t
fd01b88c 3815array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63 3816{
6791875e 3817 int err;
9e653b63 3818 enum array_state st = match_word(buf, array_states);
6791875e
N		 3819
3820 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3821 /* don't take reconfig_mutex when toggling between
3822 * clean and active
3823 */
3824 spin_lock(&mddev->lock);
3825 if (st == active) {
3826 restart_array(mddev);
3827 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3828 wake_up(&mddev->sb_wait);
3829 err = 0;
3830 } else /* st == clean */ {
3831 restart_array(mddev);
3832 if (atomic_read(&mddev->writes_pending) == 0) {
3833 if (mddev->in_sync == 0) {
3834 mddev->in_sync = 1;
3835 if (mddev->safemode == 1)
3836 mddev->safemode = 0;
3837 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3838 }
3839 err = 0;
3840 } else
3841 err = -EBUSY;
3842 }
3843 spin_unlock(&mddev->lock);
c008f1d3 3844 return err ?: len;
6791875e
N		 3845 }
3846 err = mddev_lock(mddev);
3847 if (err)
3848 return err;
3849 err = -EINVAL;
9e653b63
N		 3850 switch(st) {
3851 case bad_word:
3852 break;
3853 case clear:
3854 /* stopping an active array */
a05b7ea0 3855 err = do_md_stop(mddev, 0, NULL);
9e653b63
N		 3856 break;
3857 case inactive:
3858 /* stopping an active array */
90cf195d 3859 if (mddev->pers)
a05b7ea0 3860 err = do_md_stop(mddev, 2, NULL);
90cf195d 3861 else
e691063a 3862 err = 0; /* already inactive */
9e653b63
N		 3863 break;
3864 case suspended:
3865 break; /* not supported yet */
3866 case readonly:
3867 if (mddev->pers)
a05b7ea0 3868 err = md_set_readonly(mddev, NULL);
9e653b63
N		 3869 else {
3870 mddev->ro = 1;
648b629e 3871 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3872 err = do_md_run(mddev);
3873 }
3874 break;
3875 case read_auto:
9e653b63 3876 if (mddev->pers) {
80268ee9 3877 if (mddev->ro == 0)
a05b7ea0 3878 err = md_set_readonly(mddev, NULL);
80268ee9 3879 else if (mddev->ro == 1)
648b629e
N		 3880 err = restart_array(mddev);
3881 if (err == 0) {
3882 mddev->ro = 2;
3883 set_disk_ro(mddev->gendisk, 0);
3884 }
9e653b63
N		 3885 } else {
3886 mddev->ro = 2;
3887 err = do_md_run(mddev);
3888 }
3889 break;
3890 case clean:
3891 if (mddev->pers) {
3892 restart_array(mddev);
85572d7c 3893 spin_lock(&mddev->lock);
9e653b63 3894 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3895 if (mddev->in_sync == 0) {
3896 mddev->in_sync = 1;
31a59e34
N		 3897 if (mddev->safemode == 1)
3898 mddev->safemode = 0;
070dc6dd 3899 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3900 }
3901 err = 0;
3902 } else
3903 err = -EBUSY;
85572d7c 3904 spin_unlock(&mddev->lock);
5bf29597
N		 3905 } else
3906 err = -EINVAL;
9e653b63
N		 3907 break;
3908 case active:
3909 if (mddev->pers) {
3910 restart_array(mddev);
070dc6dd 3911 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3912 wake_up(&mddev->sb_wait);
3913 err = 0;
3914 } else {
3915 mddev->ro = 0;
648b629e 3916 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3917 err = do_md_run(mddev);
3918 }
3919 break;
3920 case write_pending:
3921 case active_idle:
3922 /* these cannot be set */
3923 break;
3924 }
6791875e
N		 3925
3926 if (!err) {
1d23f178
N		 3927 if (mddev->hold_active == UNTIL_IOCTL)
3928 mddev->hold_active = 0;
00bcb4ac 3929 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 3930 }
6791875e
N		 3931 mddev_unlock(mddev);
3932 return err ?: len;
9e653b63 3933}
80ca3a44 3934static struct md_sysfs_entry md_array_state =
750f199e 3935__ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3936
1e50915f 3937static ssize_t
fd01b88c 3938max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3939 return sprintf(page, "%d\n",
3940 atomic_read(&mddev->max_corr_read_errors));
3941}
3942
3943static ssize_t
fd01b88c 3944max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f 3945{
4c9309c0
AD		 3946 unsigned int n;
3947 int rv;
1e50915f 3948
4c9309c0
AD		 3949 rv = kstrtouint(buf, 10, &n);
3950 if (rv < 0)
3951 return rv;
3952 atomic_set(&mddev->max_corr_read_errors, n);
3953 return len;
1e50915f
RB		 3954}
3955
3956static struct md_sysfs_entry max_corr_read_errors =
3957__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3958 max_corrected_read_errors_store);
3959
6d7ff738 3960static ssize_t
fd01b88c 3961null_show(struct mddev *mddev, char *page)
6d7ff738
N		 3962{
3963 return -EINVAL;
3964}
3965
3966static ssize_t
fd01b88c 3967new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 3968{
3969 /* buf must be %d:%d\n? giving major and minor numbers */
3970 /* The new device is added to the array.
3971 * If the array has a persistent superblock, we read the
3972 * superblock to initialise info and check validity.
3973 * Otherwise, only checking done is that in bind_rdev_to_array,
3974 * which mainly checks size.
3975 */
3976 char *e;
3977 int major = simple_strtoul(buf, &e, 10);
3978 int minor;
3979 dev_t dev;
3cb03002 3980 struct md_rdev *rdev;
6d7ff738
N		 3981 int err;
3982
3983 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3984 return -EINVAL;
3985 minor = simple_strtoul(e+1, &e, 10);
3986 if (*e && *e != '\n')
3987 return -EINVAL;
3988 dev = MKDEV(major, minor);
3989 if (major != MAJOR(dev) ||
3990 minor != MINOR(dev))
3991 return -EOVERFLOW;
3992
6791875e
N		 3993 flush_workqueue(md_misc_wq);
3994
3995 err = mddev_lock(mddev);
3996 if (err)
3997 return err;
6d7ff738
N		 3998 if (mddev->persistent) {
3999 rdev = md_import_device(dev, mddev->major_version,
4000 mddev->minor_version);
4001 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 4002 struct md_rdev *rdev0
4003 = list_entry(mddev->disks.next,
4004 struct md_rdev, same_set);
6d7ff738
N		 4005 err = super_types[mddev->major_version]
4006 .load_super(rdev, rdev0, mddev->minor_version);
4007 if (err < 0)
4008 goto out;
4009 }
c5d79adb
N		 4010 } else if (mddev->external)
4011 rdev = md_import_device(dev, -2, -1);
4012 else
6d7ff738
N		 4013 rdev = md_import_device(dev, -1, -1);
4014
9a8c0fa8
N		 4015 if (IS_ERR(rdev)) {
4016 mddev_unlock(mddev);
6d7ff738 4017 return PTR_ERR(rdev);
9a8c0fa8 4018 }
6d7ff738
N		 4019 err = bind_rdev_to_array(rdev, mddev);
4020 out:
4021 if (err)
4022 export_rdev(rdev);
6791875e 4023 mddev_unlock(mddev);
6d7ff738
N		 4024 return err ? err : len;
4025}
4026
4027static struct md_sysfs_entry md_new_device =
80ca3a44 4028__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 4029
9b1d1dac 4030static ssize_t
fd01b88c 4031bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 4032{
4033 char *end;
4034 unsigned long chunk, end_chunk;
6791875e 4035 int err;
9b1d1dac 4036
6791875e
N		 4037 err = mddev_lock(mddev);
4038 if (err)
4039 return err;
9b1d1dac
PC		 4040 if (!mddev->bitmap)
4041 goto out;
4042 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4043 while (*buf) {
4044 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4045 if (buf == end) break;
4046 if (*end == '-') { /* range */
4047 buf = end + 1;
4048 end_chunk = simple_strtoul(buf, &end, 0);
4049 if (buf == end) break;
4050 }
4051 if (*end && !isspace(*end)) break;
4052 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 4053 buf = skip_spaces(end);
9b1d1dac
PC		 4054 }
4055 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4056out:
6791875e 4057 mddev_unlock(mddev);
9b1d1dac
PC		 4058 return len;
4059}
4060
4061static struct md_sysfs_entry md_bitmap =
4062__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4063
a35b0d69 4064static ssize_t
fd01b88c 4065size_show(struct mddev *mddev, char *page)
a35b0d69 4066{
58c0fed4
AN		 4067 return sprintf(page, "%llu\n",
4068 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 4069}
4070
fd01b88c 4071static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 4072
4073static ssize_t
fd01b88c 4074size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 4075{
4076 /* If array is inactive, we can reduce the component size, but
4077 * not increase it (except from 0).
4078 * If array is active, we can try an on-line resize
4079 */
b522adcd
DW		 4080 sector_t sectors;
4081 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4082
58c0fed4
AN		 4083 if (err < 0)
4084 return err;
6791875e
N		 4085 err = mddev_lock(mddev);
4086 if (err)
4087 return err;
a35b0d69 4088 if (mddev->pers) {
293467aa
GR		 4089 if (mddev_is_clustered(mddev))
4090 md_cluster_ops->metadata_update_start(mddev);
58c0fed4 4091 err = update_size(mddev, sectors);
850b2b42 4092 md_update_sb(mddev, 1);
293467aa
GR		 4093 if (mddev_is_clustered(mddev))
4094 md_cluster_ops->metadata_update_finish(mddev);
a35b0d69 4095 } else {
58c0fed4
AN		 4096 if (mddev->dev_sectors == 0 ||
4097 mddev->dev_sectors > sectors)
4098 mddev->dev_sectors = sectors;
a35b0d69
N		 4099 else
4100 err = -ENOSPC;
4101 }
6791875e 4102 mddev_unlock(mddev);
a35b0d69
N		 4103 return err ? err : len;
4104}
4105
4106static struct md_sysfs_entry md_size =
80ca3a44 4107__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4108
83f0d77a 4109/* Metadata version.
e691063a
N		 4110 * This is one of
4111 * 'none' for arrays with no metadata (good luck...)
4112 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4113 * or N.M for internally known formats
4114 */
4115static ssize_t
fd01b88c 4116metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4117{
4118 if (mddev->persistent)
4119 return sprintf(page, "%d.%d\n",
4120 mddev->major_version, mddev->minor_version);
e691063a
N		 4121 else if (mddev->external)
4122 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4123 else
4124 return sprintf(page, "none\n");
4125}
4126
4127static ssize_t
fd01b88c 4128metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4129{
4130 int major, minor;
4131 char *e;
6791875e 4132 int err;
ea43ddd8
N		 4133 /* Changing the details of 'external' metadata is
4134 * always permitted. Otherwise there must be
4135 * no devices attached to the array.
4136 */
6791875e
N		 4137
4138 err = mddev_lock(mddev);
4139 if (err)
4140 return err;
4141 err = -EBUSY;
ea43ddd8
N		 4142 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4143 ;
4144 else if (!list_empty(&mddev->disks))
6791875e 4145 goto out_unlock;
8bb93aac 4146
6791875e 4147 err = 0;
8bb93aac
N		 4148 if (cmd_match(buf, "none")) {
4149 mddev->persistent = 0;
e691063a
N		 4150 mddev->external = 0;
4151 mddev->major_version = 0;
4152 mddev->minor_version = 90;
6791875e 4153 goto out_unlock;
e691063a
N		 4154 }
4155 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4156 size_t namelen = len-9;
e691063a
N		 4157 if (namelen >= sizeof(mddev->metadata_type))
4158 namelen = sizeof(mddev->metadata_type)-1;
4159 strncpy(mddev->metadata_type, buf+9, namelen);
4160 mddev->metadata_type[namelen] = 0;
4161 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4162 mddev->metadata_type[--namelen] = 0;
4163 mddev->persistent = 0;
4164 mddev->external = 1;
8bb93aac
N		 4165 mddev->major_version = 0;
4166 mddev->minor_version = 90;
6791875e 4167 goto out_unlock;
8bb93aac
N		 4168 }
4169 major = simple_strtoul(buf, &e, 10);
6791875e 4170 err = -EINVAL;
8bb93aac 4171 if (e==buf || *e != '.')
6791875e 4172 goto out_unlock;
8bb93aac
N		 4173 buf = e+1;
4174 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4175 if (e==buf || (*e && *e != '\n') )
6791875e
N		 4176 goto out_unlock;
4177 err = -ENOENT;
50511da3 4178 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
6791875e 4179 goto out_unlock;
8bb93aac
N		 4180 mddev->major_version = major;
4181 mddev->minor_version = minor;
4182 mddev->persistent = 1;
e691063a 4183 mddev->external = 0;
6791875e
N		 4184 err = 0;
4185out_unlock:
4186 mddev_unlock(mddev);
4187 return err ?: len;
8bb93aac
N		 4188}
4189
4190static struct md_sysfs_entry md_metadata =
750f199e 4191__ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4192
24dd469d 4193static ssize_t
fd01b88c 4194action_show(struct mddev *mddev, char *page)
24dd469d 4195{
7eec314d 4196 char *type = "idle";
b7b17c9b
N		 4197 unsigned long recovery = mddev->recovery;
4198 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
b6a9ce68 4199 type = "frozen";
b7b17c9b
N		 4200 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4201 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4202 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
ccfcc3c1 4203 type = "reshape";
b7b17c9b
N		 4204 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4205 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
24dd469d 4206 type = "resync";
b7b17c9b 4207 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
24dd469d
N		 4208 type = "check";
4209 else
4210 type = "repair";
b7b17c9b 4211 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
24dd469d
N		 4212 type = "recover";
4213 }
4214 return sprintf(page, "%s\n", type);
4215}
4216
4217static ssize_t
fd01b88c 4218action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4219{
7eec314d
N		 4220 if (!mddev->pers || !mddev->pers->sync_request)
4221 return -EINVAL;
4222
b6a9ce68
N		 4223
4224 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
56ccc112
N		 4225 if (cmd_match(page, "frozen"))
4226 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4227 else
4228 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
8e8e2518
N		 4229 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4230 mddev_lock(mddev) == 0) {
4231 flush_workqueue(md_misc_wq);
4232 if (mddev->sync_thread) {
4233 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6791875e 4234 md_reap_sync_thread(mddev);
6791875e 4235 }
8e8e2518 4236 mddev_unlock(mddev);
7eec314d 4237 }
03c902e1
N		 4238 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4239 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4240 return -EBUSY;
72a23c21 4241 else if (cmd_match(page, "resync"))
56ccc112 4242 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
72a23c21 4243 else if (cmd_match(page, "recover")) {
56ccc112 4244 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
72a23c21 4245 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
72a23c21 4246 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4247 int err;
4248 if (mddev->pers->start_reshape == NULL)
4249 return -EINVAL;
6791875e
N		 4250 err = mddev_lock(mddev);
4251 if (!err) {
56ccc112 4252 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6791875e
N		 4253 err = mddev->pers->start_reshape(mddev);
4254 mddev_unlock(mddev);
4255 }
16484bf5
N		 4256 if (err)
4257 return err;
a99ac971 4258 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4259 } else {
bce74dac 4260 if (cmd_match(page, "check"))
7eec314d 4261 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4262 else if (!cmd_match(page, "repair"))
7eec314d 4263 return -EINVAL;
56ccc112 4264 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
7eec314d
N		 4265 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4266 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4267 }
48c26ddc
N		 4268 if (mddev->ro == 2) {
4269 /* A write to sync_action is enough to justify
4270 * canceling read-auto mode
4271 */
4272 mddev->ro = 0;
4273 md_wakeup_thread(mddev->sync_thread);
4274 }
03c902e1 4275 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4276 md_wakeup_thread(mddev->thread);
00bcb4ac 4277 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4278 return len;
4279}
4280
c4a39551 4281static struct md_sysfs_entry md_scan_mode =
750f199e 4282__ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
c4a39551
JB		 4283
4284static ssize_t
4285last_sync_action_show(struct mddev *mddev, char *page)
4286{
4287 return sprintf(page, "%s\n", mddev->last_sync_action);
4288}
4289
4290static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4291
9d88883e 4292static ssize_t
fd01b88c 4293mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4294{
4295 return sprintf(page, "%llu\n",
7f7583d4
JM		 4296 (unsigned long long)
4297 atomic64_read(&mddev->resync_mismatches));
9d88883e
N		 4298}
4299
80ca3a44 4300static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4301
88202a0c 4302static ssize_t
fd01b88c 4303sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4304{
4305 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4306 mddev->sync_speed_min ? "local": "system");
4307}
4308
4309static ssize_t
fd01b88c 4310sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c 4311{
4c9309c0
AD		 4312 unsigned int min;
4313 int rv;
4314
88202a0c 4315 if (strncmp(buf, "system", 6)==0) {
4c9309c0
AD		 4316 min = 0;
4317 } else {
4318 rv = kstrtouint(buf, 10, &min);
4319 if (rv < 0)
4320 return rv;
4321 if (min == 0)
4322 return -EINVAL;
88202a0c 4323 }
88202a0c
N		 4324 mddev->sync_speed_min = min;
4325 return len;
4326}
4327
4328static struct md_sysfs_entry md_sync_min =
4329__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4330
4331static ssize_t
fd01b88c 4332sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4333{
4334 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4335 mddev->sync_speed_max ? "local": "system");
4336}
4337
4338static ssize_t
fd01b88c 4339sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c 4340{
4c9309c0
AD		 4341 unsigned int max;
4342 int rv;
4343
88202a0c 4344 if (strncmp(buf, "system", 6)==0) {
4c9309c0
AD		 4345 max = 0;
4346 } else {
4347 rv = kstrtouint(buf, 10, &max);
4348 if (rv < 0)
4349 return rv;
4350 if (max == 0)
4351 return -EINVAL;
88202a0c 4352 }
88202a0c
N		 4353 mddev->sync_speed_max = max;
4354 return len;
4355}
4356
4357static struct md_sysfs_entry md_sync_max =
4358__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4359
d7f3d291 4360static ssize_t
fd01b88c 4361degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4362{
4363 return sprintf(page, "%d\n", mddev->degraded);
4364}
4365static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4366
90b08710 4367static ssize_t
fd01b88c 4368sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4369{
4370 return sprintf(page, "%d\n", mddev->parallel_resync);
4371}
4372
4373static ssize_t
fd01b88c 4374sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4375{
4376 long n;
4377
b29bebd6 4378 if (kstrtol(buf, 10, &n))
90b08710
BS		 4379 return -EINVAL;
4380
4381 if (n != 0 && n != 1)
4382 return -EINVAL;
4383
4384 mddev->parallel_resync = n;
4385
4386 if (mddev->sync_thread)
4387 wake_up(&resync_wait);
4388
4389 return len;
4390}
4391
4392/* force parallel resync, even with shared block devices */
4393static struct md_sysfs_entry md_sync_force_parallel =
4394__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4395 sync_force_parallel_show, sync_force_parallel_store);
4396
88202a0c 4397static ssize_t
fd01b88c 4398sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4399{
4400 unsigned long resync, dt, db;
d1a7c503
N		 4401 if (mddev->curr_resync == 0)
4402 return sprintf(page, "none\n");
9687a60c
AN		 4403 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4404 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4405 if (!dt) dt++;
9687a60c
AN		 4406 db = resync - mddev->resync_mark_cnt;
4407 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4408}
4409
80ca3a44 4410static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4411
4412static ssize_t
fd01b88c 4413sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4414{
13ae864b 4415 unsigned long long max_sectors, resync;
88202a0c 4416
acb180b0
N		 4417 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4418 return sprintf(page, "none\n");
4419
72f36d59
N		 4420 if (mddev->curr_resync == 1 ||
4421 mddev->curr_resync == 2)
4422 return sprintf(page, "delayed\n");
4423
c804cdec
N		 4424 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4425 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 4426 max_sectors = mddev->resync_max_sectors;
88202a0c 4427 else
58c0fed4 4428 max_sectors = mddev->dev_sectors;
88202a0c 4429
acb180b0 4430 resync = mddev->curr_resync_completed;
13ae864b 4431 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4432}
4433
750f199e
N		 4434static struct md_sysfs_entry md_sync_completed =
4435 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
88202a0c 4436
5e96ee65 4437static ssize_t
fd01b88c 4438min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4439{
4440 return sprintf(page, "%llu\n",
4441 (unsigned long long)mddev->resync_min);
4442}
4443static ssize_t
fd01b88c 4444min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4445{
4446 unsigned long long min;
23da422b 4447 int err;
23da422b 4448
b29bebd6 4449 if (kstrtoull(buf, 10, &min))
5e96ee65 4450 return -EINVAL;
23da422b
N		 4451
4452 spin_lock(&mddev->lock);
4453 err = -EINVAL;
5e96ee65 4454 if (min > mddev->resync_max)
23da422b
N		 4455 goto out_unlock;
4456
4457 err = -EBUSY;
5e96ee65 4458 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4459 goto out_unlock;
5e96ee65 4460
50c37b13
N		 4461 /* Round down to multiple of 4K for safety */
4462 mddev->resync_min = round_down(min, 8);
23da422b 4463 err = 0;
5e96ee65 4464
23da422b
N		 4465out_unlock:
4466 spin_unlock(&mddev->lock);
4467 return err ?: len;
5e96ee65
NB		 4468}
4469
4470static struct md_sysfs_entry md_min_sync =
4471__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4472
c6207277 4473static ssize_t
fd01b88c 4474max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4475{
4476 if (mddev->resync_max == MaxSector)
4477 return sprintf(page, "max\n");
4478 else
4479 return sprintf(page, "%llu\n",
4480 (unsigned long long)mddev->resync_max);
4481}
4482static ssize_t
fd01b88c 4483max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277 4484{
23da422b
N		 4485 int err;
4486 spin_lock(&mddev->lock);
c6207277
N		 4487 if (strncmp(buf, "max", 3) == 0)
4488 mddev->resync_max = MaxSector;
4489 else {
5e96ee65 4490 unsigned long long max;
23da422b
N		 4491 int chunk;
4492
4493 err = -EINVAL;
b29bebd6 4494 if (kstrtoull(buf, 10, &max))
23da422b 4495 goto out_unlock;
5e96ee65 4496 if (max < mddev->resync_min)
23da422b
N		 4497 goto out_unlock;
4498
4499 err = -EBUSY;
c6207277 4500 if (max < mddev->resync_max &&
4d484a4a 4501 mddev->ro == 0 &&
c6207277 4502 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4503 goto out_unlock;
c6207277
N		 4504
4505 /* Must be a multiple of chunk_size */
23da422b
N		 4506 chunk = mddev->chunk_sectors;
4507 if (chunk) {
2ac06c33 4508 sector_t temp = max;
23da422b
N		 4509
4510 err = -EINVAL;
4511 if (sector_div(temp, chunk))
4512 goto out_unlock;
c6207277
N		 4513 }
4514 mddev->resync_max = max;
4515 }
4516 wake_up(&mddev->recovery_wait);
23da422b
N		 4517 err = 0;
4518out_unlock:
4519 spin_unlock(&mddev->lock);
4520 return err ?: len;
c6207277
N		 4521}
4522
4523static struct md_sysfs_entry md_max_sync =
4524__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4525
e464eafd 4526static ssize_t
fd01b88c 4527suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4528{
4529 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4530}
4531
4532static ssize_t
fd01b88c 4533suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd 4534{
4c9309c0 4535 unsigned long long old, new;
6791875e 4536 int err;
e464eafd 4537
4c9309c0
AD		 4538 err = kstrtoull(buf, 10, &new);
4539 if (err < 0)
4540 return err;
4541 if (new != (sector_t)new)
e464eafd 4542 return -EINVAL;
23ddff37 4543
6791875e
N		 4544 err = mddev_lock(mddev);
4545 if (err)
4546 return err;
4547 err = -EINVAL;
4548 if (mddev->pers == NULL ||
4549 mddev->pers->quiesce == NULL)
4550 goto unlock;
4551 old = mddev->suspend_lo;
23ddff37
N		 4552 mddev->suspend_lo = new;
4553 if (new >= old)
4554 /* Shrinking suspended region */
e464eafd 4555 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4556 else {
4557 /* Expanding suspended region - need to wait */
4558 mddev->pers->quiesce(mddev, 1);
4559 mddev->pers->quiesce(mddev, 0);
4560 }
6791875e
N		 4561 err = 0;
4562unlock:
4563 mddev_unlock(mddev);
4564 return err ?: len;
e464eafd
N		 4565}
4566static struct md_sysfs_entry md_suspend_lo =
4567__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4568
e464eafd 4569static ssize_t
fd01b88c 4570suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4571{
4572 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4573}
4574
4575static ssize_t
fd01b88c 4576suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd 4577{
4c9309c0 4578 unsigned long long old, new;
6791875e 4579 int err;
e464eafd 4580
4c9309c0
AD		 4581 err = kstrtoull(buf, 10, &new);
4582 if (err < 0)
4583 return err;
4584 if (new != (sector_t)new)
e464eafd 4585 return -EINVAL;
23ddff37 4586
6791875e
N		 4587 err = mddev_lock(mddev);
4588 if (err)
4589 return err;
4590 err = -EINVAL;
4591 if (mddev->pers == NULL ||
4592 mddev->pers->quiesce == NULL)
4593 goto unlock;
4594 old = mddev->suspend_hi;
23ddff37
N		 4595 mddev->suspend_hi = new;
4596 if (new <= old)
4597 /* Shrinking suspended region */
4598 mddev->pers->quiesce(mddev, 2);
4599 else {
4600 /* Expanding suspended region - need to wait */
e464eafd
N		 4601 mddev->pers->quiesce(mddev, 1);
4602 mddev->pers->quiesce(mddev, 0);
23ddff37 4603 }
6791875e
N		 4604 err = 0;
4605unlock:
4606 mddev_unlock(mddev);
4607 return err ?: len;
e464eafd
N		 4608}
4609static struct md_sysfs_entry md_suspend_hi =
4610__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4611
08a02ecd 4612static ssize_t
fd01b88c 4613reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4614{
4615 if (mddev->reshape_position != MaxSector)
4616 return sprintf(page, "%llu\n",
4617 (unsigned long long)mddev->reshape_position);
4618 strcpy(page, "none\n");
4619 return 5;
4620}
4621
4622static ssize_t
fd01b88c 4623reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd 4624{
c6563a8c 4625 struct md_rdev *rdev;
4c9309c0 4626 unsigned long long new;
6791875e 4627 int err;
6791875e 4628
4c9309c0
AD		 4629 err = kstrtoull(buf, 10, &new);
4630 if (err < 0)
4631 return err;
4632 if (new != (sector_t)new)
08a02ecd 4633 return -EINVAL;
6791875e
N		 4634 err = mddev_lock(mddev);
4635 if (err)
4636 return err;
4637 err = -EBUSY;
4638 if (mddev->pers)
4639 goto unlock;
08a02ecd
N		 4640 mddev->reshape_position = new;
4641 mddev->delta_disks = 0;
2c810cdd 4642 mddev->reshape_backwards = 0;
08a02ecd
N		 4643 mddev->new_level = mddev->level;
4644 mddev->new_layout = mddev->layout;
664e7c41 4645 mddev->new_chunk_sectors = mddev->chunk_sectors;
c6563a8c
N		 4646 rdev_for_each(rdev, mddev)
4647 rdev->new_data_offset = rdev->data_offset;
6791875e
N		 4648 err = 0;
4649unlock:
4650 mddev_unlock(mddev);
4651 return err ?: len;
08a02ecd
N		 4652}
4653
4654static struct md_sysfs_entry md_reshape_position =
4655__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4656 reshape_position_store);
4657
2c810cdd
N		 4658static ssize_t
4659reshape_direction_show(struct mddev *mddev, char *page)
4660{
4661 return sprintf(page, "%s\n",
4662 mddev->reshape_backwards ? "backwards" : "forwards");
4663}
4664
4665static ssize_t
4666reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4667{
4668 int backwards = 0;
6791875e
N		 4669 int err;
4670
2c810cdd
N		 4671 if (cmd_match(buf, "forwards"))
4672 backwards = 0;
4673 else if (cmd_match(buf, "backwards"))
4674 backwards = 1;
4675 else
4676 return -EINVAL;
4677 if (mddev->reshape_backwards == backwards)
4678 return len;
4679
6791875e
N		 4680 err = mddev_lock(mddev);
4681 if (err)
4682 return err;
2c810cdd
N		 4683 /* check if we are allowed to change */
4684 if (mddev->delta_disks)
6791875e
N		 4685 err = -EBUSY;
4686 else if (mddev->persistent &&
2c810cdd 4687 mddev->major_version == 0)
6791875e
N		 4688 err = -EINVAL;
4689 else
4690 mddev->reshape_backwards = backwards;
4691 mddev_unlock(mddev);
4692 return err ?: len;
2c810cdd
N		 4693}
4694
4695static struct md_sysfs_entry md_reshape_direction =
4696__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4697 reshape_direction_store);
4698
b522adcd 4699static ssize_t
fd01b88c 4700array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4701{
4702 if (mddev->external_size)
4703 return sprintf(page, "%llu\n",
4704 (unsigned long long)mddev->array_sectors/2);
4705 else
4706 return sprintf(page, "default\n");
4707}
4708
4709static ssize_t
fd01b88c 4710array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4711{
4712 sector_t sectors;
6791875e
N		 4713 int err;
4714
4715 err = mddev_lock(mddev);
4716 if (err)
4717 return err;
b522adcd
DW		 4718
4719 if (strncmp(buf, "default", 7) == 0) {
4720 if (mddev->pers)
4721 sectors = mddev->pers->size(mddev, 0, 0);
4722 else
4723 sectors = mddev->array_sectors;
4724
4725 mddev->external_size = 0;
4726 } else {
4727 if (strict_blocks_to_sectors(buf, &sectors) < 0)
6791875e
N		 4728 err = -EINVAL;
4729 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4730 err = -E2BIG;
4731 else
4732 mddev->external_size = 1;
b522adcd
DW		 4733 }
4734
6791875e
N		 4735 if (!err) {
4736 mddev->array_sectors = sectors;
4737 if (mddev->pers) {
4738 set_capacity(mddev->gendisk, mddev->array_sectors);
4739 revalidate_disk(mddev->gendisk);
4740 }
cbe6ef1d 4741 }
6791875e
N		 4742 mddev_unlock(mddev);
4743 return err ?: len;
b522adcd
DW		 4744}
4745
4746static struct md_sysfs_entry md_array_size =
4747__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4748 array_size_store);
e464eafd 4749
eae1701f
N		 4750static struct attribute *md_default_attrs[] = {
4751 &md_level.attr,
d4dbd025 4752 &md_layout.attr,
eae1701f 4753 &md_raid_disks.attr,
3b34380a 4754 &md_chunk_size.attr,
a35b0d69 4755 &md_size.attr,
a94213b1 4756 &md_resync_start.attr,
8bb93aac 4757 &md_metadata.attr,
6d7ff738 4758 &md_new_device.attr,
16f17b39 4759 &md_safe_delay.attr,
9e653b63 4760 &md_array_state.attr,
08a02ecd 4761 &md_reshape_position.attr,
2c810cdd 4762 &md_reshape_direction.attr,
b522adcd 4763 &md_array_size.attr,
1e50915f 4764 &max_corr_read_errors.attr,
411036fa
N		 4765 NULL,
4766};
4767
4768static struct attribute *md_redundancy_attrs[] = {
24dd469d 4769 &md_scan_mode.attr,
c4a39551 4770 &md_last_scan_mode.attr,
9d88883e 4771 &md_mismatches.attr,
88202a0c
N		 4772 &md_sync_min.attr,
4773 &md_sync_max.attr,
4774 &md_sync_speed.attr,
90b08710 4775 &md_sync_force_parallel.attr,
88202a0c 4776 &md_sync_completed.attr,
5e96ee65 4777 &md_min_sync.attr,
c6207277 4778 &md_max_sync.attr,
e464eafd
N		 4779 &md_suspend_lo.attr,
4780 &md_suspend_hi.attr,
9b1d1dac 4781 &md_bitmap.attr,
d7f3d291 4782 &md_degraded.attr,
eae1701f
N		 4783 NULL,
4784};
411036fa
N		 4785static struct attribute_group md_redundancy_group = {
4786 .name = NULL,
4787 .attrs = md_redundancy_attrs,
4788};
4789
eae1701f
N		 4790static ssize_t
4791md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4792{
4793 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4794 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4795 ssize_t rv;
eae1701f
N		 4796
4797 if (!entry->show)
4798 return -EIO;
af8a2434
N		 4799 spin_lock(&all_mddevs_lock);
4800 if (list_empty(&mddev->all_mddevs)) {
4801 spin_unlock(&all_mddevs_lock);
4802 return -EBUSY;
4803 }
4804 mddev_get(mddev);
4805 spin_unlock(&all_mddevs_lock);
4806
b7b17c9b 4807 rv = entry->show(mddev, page);
af8a2434 4808 mddev_put(mddev);
96de1e66 4809 return rv;
eae1701f
N		 4810}
4811
4812static ssize_t
4813md_attr_store(struct kobject *kobj, struct attribute *attr,
4814 const char *page, size_t length)
4815{
4816 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4817 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4818 ssize_t rv;
eae1701f
N		 4819
4820 if (!entry->store)
4821 return -EIO;
67463acb
N		 4822 if (!capable(CAP_SYS_ADMIN))
4823 return -EACCES;
af8a2434
N		 4824 spin_lock(&all_mddevs_lock);
4825 if (list_empty(&mddev->all_mddevs)) {
4826 spin_unlock(&all_mddevs_lock);
4827 return -EBUSY;
4828 }
4829 mddev_get(mddev);
4830 spin_unlock(&all_mddevs_lock);
6791875e 4831 rv = entry->store(mddev, page, length);
af8a2434 4832 mddev_put(mddev);
96de1e66 4833 return rv;
eae1701f
N		 4834}
4835
4836static void md_free(struct kobject *ko)
4837{
fd01b88c 4838 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4839
4840 if (mddev->sysfs_state)
4841 sysfs_put(mddev->sysfs_state);
4842
6cd18e71
N		 4843 if (mddev->queue)
4844 blk_cleanup_queue(mddev->queue);
a21d1504
N		 4845 if (mddev->gendisk) {
4846 del_gendisk(mddev->gendisk);
4847 put_disk(mddev->gendisk);
4848 }
a21d1504 4849
eae1701f
N		 4850 kfree(mddev);
4851}
4852
52cf25d0 4853static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4854 .show = md_attr_show,
4855 .store = md_attr_store,
4856};
4857static struct kobj_type md_ktype = {
4858 .release = md_free,
4859 .sysfs_ops = &md_sysfs_ops,
4860 .default_attrs = md_default_attrs,
4861};
4862
1da177e4
LT		 4863int mdp_major = 0;
4864
5fd3a17e
DW		 4865static void mddev_delayed_delete(struct work_struct *ws)
4866{
fd01b88c 4867 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4868
43a70507 4869 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4870 kobject_del(&mddev->kobj);
4871 kobject_put(&mddev->kobj);
4872}
4873
efeb53c0 4874static int md_alloc(dev_t dev, char *name)
1da177e4 4875{
48c9c27b 4876 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4877 struct mddev *mddev = mddev_find(dev);
1da177e4 4878 struct gendisk *disk;
efeb53c0
N		 4879 int partitioned;
4880 int shift;
4881 int unit;
3830c62f 4882 int error;
1da177e4
LT		 4883
4884 if (!mddev)
efeb53c0
N		 4885 return -ENODEV;
4886
4887 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4888 shift = partitioned ? MdpMinorShift : 0;
4889 unit = MINOR(mddev->unit) >> shift;
1da177e4 4890
e804ac78
TH		 4891 /* wait for any previous instance of this device to be
4892 * completely removed (mddev_delayed_delete).
d3374825 4893 */
e804ac78 4894 flush_workqueue(md_misc_wq);
d3374825 4895
48c9c27b 4896 mutex_lock(&disks_mutex);
0909dc44
N		 4897 error = -EEXIST;
4898 if (mddev->gendisk)
4899 goto abort;
efeb53c0
N		 4900
4901 if (name) {
4902 /* Need to ensure that 'name' is not a duplicate.
4903 */
fd01b88c 4904 struct mddev *mddev2;
efeb53c0
N		 4905 spin_lock(&all_mddevs_lock);
4906
4907 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4908 if (mddev2->gendisk &&
4909 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4910 spin_unlock(&all_mddevs_lock);
0909dc44 4911 goto abort;
efeb53c0
N		 4912 }
4913 spin_unlock(&all_mddevs_lock);
1da177e4 4914 }
8b765398 4915
0909dc44 4916 error = -ENOMEM;
8b765398 4917 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4918 if (!mddev->queue)
4919 goto abort;
409c57f3
N		 4920 mddev->queue->queuedata = mddev;
4921
409c57f3 4922 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 4923 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 4924
1da177e4
LT		 4925 disk = alloc_disk(1 << shift);
4926 if (!disk) {
8b765398
N		 4927 blk_cleanup_queue(mddev->queue);
4928 mddev->queue = NULL;
0909dc44 4929 goto abort;
1da177e4 4930 }
efeb53c0 4931 disk->major = MAJOR(mddev->unit);
1da177e4 4932 disk->first_minor = unit << shift;
efeb53c0
N		 4933 if (name)
4934 strcpy(disk->disk_name, name);
4935 else if (partitioned)
1da177e4 4936 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4937 else
1da177e4 4938 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4939 disk->fops = &md_fops;
4940 disk->private_data = mddev;
4941 disk->queue = mddev->queue;
b0140891 4942 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4943 /* Allow extended partitions. This makes the
d3374825 4944 * 'mdp' device redundant, but we can't really
92850bbd
N		 4945 * remove it now.
4946 */
4947 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4948 mddev->gendisk = disk;
b0140891
N		 4949 /* As soon as we call add_disk(), another thread could get
4950 * through to md_open, so make sure it doesn't get too far
4951 */
4952 mutex_lock(&mddev->open_mutex);
4953 add_disk(disk);
4954
ed9e1982
TH		 4955 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4956 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4957 if (error) {
4958 /* This isn't possible, but as kobject_init_and_add is marked
4959 * __must_check, we must do something with the result
4960 */
5e55e2f5
N		 4961 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4962 disk->disk_name);
0909dc44
N		 4963 error = 0;
4964 }
00bcb4ac
N		 4965 if (mddev->kobj.sd &&
4966 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4967 printk(KERN_DEBUG "pointless warning\n");
b0140891 4968 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4969 abort:
4970 mutex_unlock(&disks_mutex);
00bcb4ac 4971 if (!error && mddev->kobj.sd) {
3830c62f 4972 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4973 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4974 }
d3374825 4975 mddev_put(mddev);
0909dc44 4976 return error;
efeb53c0
N		 4977}
4978
4979static struct kobject *md_probe(dev_t dev, int *part, void *data)
4980{
4981 md_alloc(dev, NULL);
1da177e4
LT		 4982 return NULL;
4983}
4984
efeb53c0
N		 4985static int add_named_array(const char *val, struct kernel_param *kp)
4986{
4987 /* val must be "md_*" where * is not all digits.
4988 * We allocate an array with a large free minor number, and
4989 * set the name to val. val must not already be an active name.
4990 */
4991 int len = strlen(val);
4992 char buf[DISK_NAME_LEN];
4993
4994 while (len && val[len-1] == '\n')
4995 len--;
4996 if (len >= DISK_NAME_LEN)
4997 return -E2BIG;
4998 strlcpy(buf, val, len+1);
4999 if (strncmp(buf, "md_", 3) != 0)
5000 return -EINVAL;
5001 return md_alloc(0, buf);
5002}
5003
1da177e4
LT		 5004static void md_safemode_timeout(unsigned long data)
5005{
fd01b88c 5006 struct mddev *mddev = (struct mddev *) data;
1da177e4 5007
0fd62b86
NB		 5008 if (!atomic_read(&mddev->writes_pending)) {
5009 mddev->safemode = 1;
5010 if (mddev->external)
00bcb4ac 5011 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 5012 }
1da177e4
LT		 5013 md_wakeup_thread(mddev->thread);
5014}
5015
6ff8d8ec 5016static int start_dirty_degraded;
1da177e4 5017
fd01b88c 5018int md_run(struct mddev *mddev)
1da177e4 5019{
2604b703 5020 int err;
3cb03002 5021 struct md_rdev *rdev;
84fc4b56 5022 struct md_personality *pers;
1da177e4 5023
a757e64c
N		 5024 if (list_empty(&mddev->disks))
5025 /* cannot run an array with no devices.. */
1da177e4 5026 return -EINVAL;
1da177e4
LT		 5027
5028 if (mddev->pers)
5029 return -EBUSY;
bb4f1e9d
N		 5030 /* Cannot run until previous stop completes properly */
5031 if (mddev->sysfs_active)
5032 return -EBUSY;
b6eb127d 5033
1da177e4
LT		 5034 /*
5035 * Analyze all RAID superblock(s)
5036 */
1ec4a939
N		 5037 if (!mddev->raid_disks) {
5038 if (!mddev->persistent)
5039 return -EINVAL;
a757e64c 5040 analyze_sbs(mddev);
1ec4a939 5041 }
1da177e4 5042
d9d166c2
N		 5043 if (mddev->level != LEVEL_NONE)
5044 request_module("md-level-%d", mddev->level);
5045 else if (mddev->clevel[0])
5046 request_module("md-%s", mddev->clevel);
1da177e4
LT		 5047
5048 /*
5049 * Drop all container device buffers, from now on
5050 * the only valid external interface is through the md
5051 * device.
1da177e4 5052 */
dafb20fa 5053 rdev_for_each(rdev, mddev) {
b2d444d7 5054 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5055 continue;
5056 sync_blockdev(rdev->bdev);
f98393a6 5057 invalidate_bdev(rdev->bdev);
f0d76d70
N		 5058
5059 /* perform some consistency tests on the device.
5060 * We don't want the data to overlap the metadata,
58c0fed4 5061 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 5062 */
a6ff7e08
JB		 5063 if (rdev->meta_bdev) {
5064 /* Nothing to check */;
5065 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 5066 if (mddev->dev_sectors &&
5067 rdev->data_offset + mddev->dev_sectors
0f420358 5068 > rdev->sb_start) {
f0d76d70
N		 5069 printk("md: %s: data overlaps metadata\n",
5070 mdname(mddev));
5071 return -EINVAL;
5072 }
5073 } else {
0f420358 5074 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 5075 > rdev->data_offset) {
5076 printk("md: %s: metadata overlaps data\n",
5077 mdname(mddev));
5078 return -EINVAL;
5079 }
5080 }
00bcb4ac 5081 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 5082 }
5083
a167f663 5084 if (mddev->bio_set == NULL)
395c72a7 5085 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
a167f663 5086
1da177e4 5087 spin_lock(&pers_lock);
d9d166c2 5088 pers = find_pers(mddev->level, mddev->clevel);
2604b703 5089 if (!pers || !try_module_get(pers->owner)) {
1da177e4 5090 spin_unlock(&pers_lock);
d9d166c2
N		 5091 if (mddev->level != LEVEL_NONE)
5092 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5093 mddev->level);
5094 else
5095 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5096 mddev->clevel);
1da177e4
LT		 5097 return -EINVAL;
5098 }
1da177e4 5099 spin_unlock(&pers_lock);
34817e8c
N		 5100 if (mddev->level != pers->level) {
5101 mddev->level = pers->level;
5102 mddev->new_level = pers->level;
5103 }
d9d166c2 5104 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 5105
f6705578 5106 if (mddev->reshape_position != MaxSector &&
63c70c4f 5107 pers->start_reshape == NULL) {
f6705578 5108 /* This personality cannot handle reshaping... */
f6705578
N		 5109 module_put(pers->owner);
5110 return -EINVAL;
5111 }
5112
7dd5e7c3
N		 5113 if (pers->sync_request) {
5114 /* Warn if this is a potentially silly
5115 * configuration.
5116 */
5117 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5118 struct md_rdev *rdev2;
7dd5e7c3 5119 int warned = 0;
159ec1fc 5120
dafb20fa
N		 5121 rdev_for_each(rdev, mddev)
5122 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N		 5123 if (rdev < rdev2 &&
5124 rdev->bdev->bd_contains ==
5125 rdev2->bdev->bd_contains) {
5126 printk(KERN_WARNING
5127 "%s: WARNING: %s appears to be"
5128 " on the same physical disk as"
5129 " %s.\n",
5130 mdname(mddev),
5131 bdevname(rdev->bdev,b),
5132 bdevname(rdev2->bdev,b2));
5133 warned = 1;
5134 }
5135 }
159ec1fc 5136
7dd5e7c3
N		 5137 if (warned)
5138 printk(KERN_WARNING
5139 "True protection against single-disk"
5140 " failure might be compromised.\n");
5141 }
5142
657390d2 5143 mddev->recovery = 0;
58c0fed4
AN		 5144 /* may be over-ridden by personality */
5145 mddev->resync_max_sectors = mddev->dev_sectors;
5146
6ff8d8ec 5147 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5148
0f9552b5 5149 if (start_readonly && mddev->ro == 0)
f91de92e
N		 5150 mddev->ro = 2; /* read-only, but switch on first write */
5151
36d091f4 5152 err = pers->run(mddev);
13e53df3
AN		 5153 if (err)
5154 printk(KERN_ERR "md: pers->run() failed ...\n");
36d091f4 5155 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
b522adcd
DW		 5156 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5157 " but 'external_size' not in effect?\n", __func__);
5158 printk(KERN_ERR
5159 "md: invalid array_size %llu > default size %llu\n",
5160 (unsigned long long)mddev->array_sectors / 2,
36d091f4 5161 (unsigned long long)pers->size(mddev, 0, 0) / 2);
b522adcd 5162 err = -EINVAL;
b522adcd 5163 }
36d091f4 5164 if (err == 0 && pers->sync_request &&
ef99bf48 5165 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
f9209a32
GR		 5166 struct bitmap *bitmap;
5167
5168 bitmap = bitmap_create(mddev, -1);
5169 if (IS_ERR(bitmap)) {
5170 err = PTR_ERR(bitmap);
b15c2e57
N		 5171 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5172 mdname(mddev), err);
f9209a32
GR		 5173 } else
5174 mddev->bitmap = bitmap;
5175
b15c2e57 5176 }
1da177e4 5177 if (err) {
5aa61f42 5178 mddev_detach(mddev);
0c35bd47
N		 5179 if (mddev->private)
5180 pers->free(mddev, mddev->private);
bd691922 5181 mddev->private = NULL;
36d091f4 5182 module_put(pers->owner);
32a7627c
N		 5183 bitmap_destroy(mddev);
5184 return err;
1da177e4 5185 }
5c675f83
N		 5186 if (mddev->queue) {
5187 mddev->queue->backing_dev_info.congested_data = mddev;
5188 mddev->queue->backing_dev_info.congested_fn = md_congested;
64590f45 5189 blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec);
5c675f83 5190 }
36d091f4 5191 if (pers->sync_request) {
00bcb4ac
N		 5192 if (mddev->kobj.sd &&
5193 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 5194 printk(KERN_WARNING
5195 "md: cannot register extra attributes for %s\n",
5196 mdname(mddev));
00bcb4ac 5197 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5198 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 5199 mddev->ro = 0;
5200
f72ffdd6 5201 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 5202 atomic_set(&mddev->max_corr_read_errors,
5203 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 5204 mddev->safemode = 0;
5205 mddev->safemode_timer.function = md_safemode_timeout;
5206 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 5207 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5208 mddev->in_sync = 1;
0ca69886 5209 smp_wmb();
36d091f4
N		 5210 spin_lock(&mddev->lock);
5211 mddev->pers = pers;
0ca69886 5212 mddev->ready = 1;
36d091f4 5213 spin_unlock(&mddev->lock);
dafb20fa 5214 rdev_for_each(rdev, mddev)
36fad858
NK		 5215 if (rdev->raid_disk >= 0)
5216 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 5217 /* failure here is OK */;
f72ffdd6 5218
1da177e4 5219 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f72ffdd6 5220
7a0a5355 5221 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 5222 md_update_sb(mddev, 0);
1da177e4 5223
d7603b7e 5224 md_new_event(mddev);
00bcb4ac
N		 5225 sysfs_notify_dirent_safe(mddev->sysfs_state);
5226 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 5227 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 5228 return 0;
5229}
390ee602 5230EXPORT_SYMBOL_GPL(md_run);
1da177e4 5231
fd01b88c 5232static int do_md_run(struct mddev *mddev)
fe60b014
N		 5233{
5234 int err;
5235
5236 err = md_run(mddev);
5237 if (err)
5238 goto out;
69e51b44
N		 5239 err = bitmap_load(mddev);
5240 if (err) {
5241 bitmap_destroy(mddev);
5242 goto out;
5243 }
0fd018af
JB		 5244
5245 md_wakeup_thread(mddev->thread);
5246 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5247
fe60b014
N		 5248 set_capacity(mddev->gendisk, mddev->array_sectors);
5249 revalidate_disk(mddev->gendisk);
f0b4f7e2 5250 mddev->changed = 1;
fe60b014
N		 5251 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5252out:
5253 return err;
5254}
5255
fd01b88c 5256static int restart_array(struct mddev *mddev)
1da177e4
LT		 5257{
5258 struct gendisk *disk = mddev->gendisk;
1da177e4 5259
80fab1d7 5260 /* Complain if it has no devices */
1da177e4 5261 if (list_empty(&mddev->disks))
80fab1d7
AN		 5262 return -ENXIO;
5263 if (!mddev->pers)
5264 return -EINVAL;
5265 if (!mddev->ro)
5266 return -EBUSY;
5267 mddev->safemode = 0;
5268 mddev->ro = 0;
5269 set_disk_ro(disk, 0);
5270 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5271 mdname(mddev));
5272 /* Kick recovery or resync if necessary */
5273 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5274 md_wakeup_thread(mddev->thread);
5275 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5276 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5277 return 0;
1da177e4
LT		 5278}
5279
fd01b88c 5280static void md_clean(struct mddev *mddev)
6177b472
N		 5281{
5282 mddev->array_sectors = 0;
5283 mddev->external_size = 0;
5284 mddev->dev_sectors = 0;
5285 mddev->raid_disks = 0;
5286 mddev->recovery_cp = 0;
5287 mddev->resync_min = 0;
5288 mddev->resync_max = MaxSector;
5289 mddev->reshape_position = MaxSector;
5290 mddev->external = 0;
5291 mddev->persistent = 0;
5292 mddev->level = LEVEL_NONE;
5293 mddev->clevel[0] = 0;
5294 mddev->flags = 0;
5295 mddev->ro = 0;
5296 mddev->metadata_type[0] = 0;
5297 mddev->chunk_sectors = 0;
5298 mddev->ctime = mddev->utime = 0;
5299 mddev->layout = 0;
5300 mddev->max_disks = 0;
5301 mddev->events = 0;
a8707c08 5302 mddev->can_decrease_events = 0;
6177b472 5303 mddev->delta_disks = 0;
2c810cdd 5304 mddev->reshape_backwards = 0;
6177b472
N		 5305 mddev->new_level = LEVEL_NONE;
5306 mddev->new_layout = 0;
5307 mddev->new_chunk_sectors = 0;
5308 mddev->curr_resync = 0;
7f7583d4 5309 atomic64_set(&mddev->resync_mismatches, 0);
6177b472
N		 5310 mddev->suspend_lo = mddev->suspend_hi = 0;
5311 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5312 mddev->recovery = 0;
5313 mddev->in_sync = 0;
f0b4f7e2 5314 mddev->changed = 0;
6177b472 5315 mddev->degraded = 0;
6177b472 5316 mddev->safemode = 0;
bd691922 5317 mddev->private = NULL;
050b6615 5318 mddev->merge_check_needed = 0;
6177b472
N		 5319 mddev->bitmap_info.offset = 0;
5320 mddev->bitmap_info.default_offset = 0;
6409bb05 5321 mddev->bitmap_info.default_space = 0;
6177b472
N		 5322 mddev->bitmap_info.chunksize = 0;
5323 mddev->bitmap_info.daemon_sleep = 0;
5324 mddev->bitmap_info.max_write_behind = 0;
5325}
5326
fd01b88c 5327static void __md_stop_writes(struct mddev *mddev)
a047e125 5328{
293467aa
GR		 5329 if (mddev_is_clustered(mddev))
5330 md_cluster_ops->metadata_update_start(mddev);
6b6204ee 5331 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
f851b60d 5332 flush_workqueue(md_misc_wq);
a047e125 5333 if (mddev->sync_thread) {
a047e125 5334 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 5335 md_reap_sync_thread(mddev);
a047e125
N		 5336 }
5337
5338 del_timer_sync(&mddev->safemode_timer);
5339
5340 bitmap_flush(mddev);
5341 md_super_wait(mddev);
5342
b6d428c6 5343 if (mddev->ro == 0 &&
7a0a5355 5344 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
a047e125
N		 5345 /* mark array as shutdown cleanly */
5346 mddev->in_sync = 1;
5347 md_update_sb(mddev, 1);
5348 }
293467aa
GR		 5349 if (mddev_is_clustered(mddev))
5350 md_cluster_ops->metadata_update_finish(mddev);
a047e125 5351}
defad61a 5352
fd01b88c 5353void md_stop_writes(struct mddev *mddev)
defad61a 5354{
29f097c4 5355 mddev_lock_nointr(mddev);
defad61a
N		 5356 __md_stop_writes(mddev);
5357 mddev_unlock(mddev);
5358}
390ee602 5359EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5360
5aa61f42
N		 5361static void mddev_detach(struct mddev *mddev)
5362{
5363 struct bitmap *bitmap = mddev->bitmap;
5364 /* wait for behind writes to complete */
5365 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5366 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5367 mdname(mddev));
5368 /* need to kick something here to make sure I/O goes? */
5369 wait_event(bitmap->behind_wait,
5370 atomic_read(&bitmap->behind_writes) == 0);
5371 }
36d091f4 5372 if (mddev->pers && mddev->pers->quiesce) {
5aa61f42
N		 5373 mddev->pers->quiesce(mddev, 1);
5374 mddev->pers->quiesce(mddev, 0);
5375 }
5376 md_unregister_thread(&mddev->thread);
5377 if (mddev->queue)
5378 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5379}
5380
5eff3c43 5381static void __md_stop(struct mddev *mddev)
6177b472 5382{
36d091f4 5383 struct md_personality *pers = mddev->pers;
5aa61f42 5384 mddev_detach(mddev);
ee5d004f
N		 5385 /* Ensure ->event_work is done */
5386 flush_workqueue(md_misc_wq);
36d091f4
N		 5387 spin_lock(&mddev->lock);
5388 mddev->ready = 0;
6177b472 5389 mddev->pers = NULL;
36d091f4
N		 5390 spin_unlock(&mddev->lock);
5391 pers->free(mddev, mddev->private);
bd691922 5392 mddev->private = NULL;
36d091f4
N		 5393 if (pers->sync_request && mddev->to_remove == NULL)
5394 mddev->to_remove = &md_redundancy_group;
5395 module_put(pers->owner);
cca9cf90 5396 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5397}
5eff3c43
N		 5398
5399void md_stop(struct mddev *mddev)
5400{
5401 /* stop the array and free an attached data structures.
5402 * This is called from dm-raid
5403 */
5404 __md_stop(mddev);
5405 bitmap_destroy(mddev);
5406 if (mddev->bio_set)
5407 bioset_free(mddev->bio_set);
5408}
5409
390ee602 5410EXPORT_SYMBOL_GPL(md_stop);
6177b472 5411
a05b7ea0 5412static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N		 5413{
5414 int err = 0;
30b8feb7
N		 5415 int did_freeze = 0;
5416
5417 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5418 did_freeze = 1;
5419 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5420 md_wakeup_thread(mddev->thread);
5421 }
f851b60d 5422 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5423 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5424 if (mddev->sync_thread)
30b8feb7
N		 5425 /* Thread might be blocked waiting for metadata update
5426 * which will now never happen */
5427 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5428
30b8feb7 5429 mddev_unlock(mddev);
f851b60d
N		 5430 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5431 &mddev->recovery));
30b8feb7
N		 5432 mddev_lock_nointr(mddev);
5433
a4bd82d0 5434 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5435 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7 5436 mddev->sync_thread ||
f851b60d 5437 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5438 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
a4bd82d0 5439 printk("md: %s still in use.\n",mdname(mddev));
30b8feb7
N		 5440 if (did_freeze) {
5441 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5442 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5443 md_wakeup_thread(mddev->thread);
5444 }
a4bd82d0
N		 5445 err = -EBUSY;
5446 goto out;
5447 }
5448 if (mddev->pers) {
defad61a 5449 __md_stop_writes(mddev);
a4bd82d0
N		 5450
5451 err = -ENXIO;
5452 if (mddev->ro==1)
5453 goto out;
5454 mddev->ro = 1;
5455 set_disk_ro(mddev->gendisk, 1);
5456 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d
N		 5457 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5458 md_wakeup_thread(mddev->thread);
00bcb4ac 5459 sysfs_notify_dirent_safe(mddev->sysfs_state);
30b8feb7 5460 err = 0;
a4bd82d0
N		 5461 }
5462out:
5463 mutex_unlock(&mddev->open_mutex);
5464 return err;
5465}
5466
9e653b63
N		 5467/* mode:
5468 * 0 - completely stop and dis-assemble array
9e653b63
N		 5469 * 2 - stop but do not disassemble array
5470 */
f72ffdd6 5471static int do_md_stop(struct mddev *mddev, int mode,
a05b7ea0 5472 struct block_device *bdev)
1da177e4 5473{
1da177e4 5474 struct gendisk *disk = mddev->gendisk;
3cb03002 5475 struct md_rdev *rdev;
30b8feb7
N		 5476 int did_freeze = 0;
5477
5478 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5479 did_freeze = 1;
5480 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5481 md_wakeup_thread(mddev->thread);
5482 }
f851b60d 5483 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5484 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5485 if (mddev->sync_thread)
30b8feb7
N		 5486 /* Thread might be blocked waiting for metadata update
5487 * which will now never happen */
5488 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5489
30b8feb7 5490 mddev_unlock(mddev);
f851b60d
N		 5491 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5492 !test_bit(MD_RECOVERY_RUNNING,
5493 &mddev->recovery)));
30b8feb7 5494 mddev_lock_nointr(mddev);
1da177e4 5495
c8c00a69 5496 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5497 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7
N		 5498 mddev->sysfs_active ||
5499 mddev->sync_thread ||
f851b60d 5500 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5501 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
df5b20cf 5502 printk("md: %s still in use.\n",mdname(mddev));
6e17b027 5503 mutex_unlock(&mddev->open_mutex);
30b8feb7
N		 5504 if (did_freeze) {
5505 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5506 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5507 md_wakeup_thread(mddev->thread);
5508 }
260fa034
N		 5509 return -EBUSY;
5510 }
6e17b027 5511 if (mddev->pers) {
a4bd82d0
N		 5512 if (mddev->ro)
5513 set_disk_ro(disk, 0);
409c57f3 5514
defad61a 5515 __md_stop_writes(mddev);
5eff3c43 5516 __md_stop(mddev);
a4bd82d0 5517 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5518 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5519
a4bd82d0 5520 /* tell userspace to handle 'inactive' */
00bcb4ac 5521 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5522
dafb20fa 5523 rdev_for_each(rdev, mddev)
36fad858
NK		 5524 if (rdev->raid_disk >= 0)
5525 sysfs_unlink_rdev(mddev, rdev);
c4647292 5526
a4bd82d0 5527 set_capacity(disk, 0);
6e17b027 5528 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5529 mddev->changed = 1;
a4bd82d0 5530 revalidate_disk(disk);
0d4ca600 5531
a4bd82d0
N		 5532 if (mddev->ro)
5533 mddev->ro = 0;
6e17b027
N		 5534 } else
5535 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5536 /*
5537 * Free resources if final stop
5538 */
9e653b63 5539 if (mode == 0) {
1da177e4
LT		 5540 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5541
978f946b 5542 bitmap_destroy(mddev);
c3d9714e 5543 if (mddev->bitmap_info.file) {
4af1a041
N		 5544 struct file *f = mddev->bitmap_info.file;
5545 spin_lock(&mddev->lock);
c3d9714e 5546 mddev->bitmap_info.file = NULL;
4af1a041
N		 5547 spin_unlock(&mddev->lock);
5548 fput(f);
978f946b 5549 }
c3d9714e 5550 mddev->bitmap_info.offset = 0;
978f946b 5551
1da177e4
LT		 5552 export_array(mddev);
5553
6177b472 5554 md_clean(mddev);
934d9c23 5555 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5556 if (mddev->hold_active == UNTIL_STOP)
5557 mddev->hold_active = 0;
a4bd82d0 5558 }
3f9d99c1 5559 blk_integrity_unregister(disk);
d7603b7e 5560 md_new_event(mddev);
00bcb4ac 5561 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5562 return 0;
1da177e4
LT		 5563}
5564
fdee8ae4 5565#ifndef MODULE
fd01b88c 5566static void autorun_array(struct mddev *mddev)
1da177e4 5567{
3cb03002 5568 struct md_rdev *rdev;
1da177e4
LT		 5569 int err;
5570
a757e64c 5571 if (list_empty(&mddev->disks))
1da177e4 5572 return;
1da177e4
LT		 5573
5574 printk(KERN_INFO "md: running: ");
5575
dafb20fa 5576 rdev_for_each(rdev, mddev) {
1da177e4
LT		 5577 char b[BDEVNAME_SIZE];
5578 printk("<%s>", bdevname(rdev->bdev,b));
5579 }
5580 printk("\n");
5581
d710e138 5582 err = do_md_run(mddev);
1da177e4
LT		 5583 if (err) {
5584 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
a05b7ea0 5585 do_md_stop(mddev, 0, NULL);
1da177e4
LT		 5586 }
5587}
5588
5589/*
5590 * lets try to run arrays based on all disks that have arrived
5591 * until now. (those are in pending_raid_disks)
5592 *
5593 * the method: pick the first pending disk, collect all disks with
5594 * the same UUID, remove all from the pending list and put them into
5595 * the 'same_array' list. Then order this list based on superblock
5596 * update time (freshest comes first), kick out 'old' disks and
5597 * compare superblocks. If everything's fine then run it.
5598 *
5599 * If "unit" is allocated, then bump its reference count
5600 */
5601static void autorun_devices(int part)
5602{
3cb03002 5603 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5604 struct mddev *mddev;
1da177e4
LT		 5605 char b[BDEVNAME_SIZE];
5606
5607 printk(KERN_INFO "md: autorun ...\n");
5608 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5609 int unit;
1da177e4 5610 dev_t dev;
ad01c9e3 5611 LIST_HEAD(candidates);
1da177e4 5612 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5613 struct md_rdev, same_set);
1da177e4
LT		 5614
5615 printk(KERN_INFO "md: considering %s ...\n",
5616 bdevname(rdev0->bdev,b));
5617 INIT_LIST_HEAD(&candidates);
159ec1fc 5618 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5619 if (super_90_load(rdev, rdev0, 0) >= 0) {
5620 printk(KERN_INFO "md: adding %s ...\n",
5621 bdevname(rdev->bdev,b));
5622 list_move(&rdev->same_set, &candidates);
5623 }
5624 /*
5625 * now we have a set of devices, with all of them having
5626 * mostly sane superblocks. It's time to allocate the
5627 * mddev.
5628 */
e8703fe1
N		 5629 if (part) {
5630 dev = MKDEV(mdp_major,
5631 rdev0->preferred_minor << MdpMinorShift);
5632 unit = MINOR(dev) >> MdpMinorShift;
5633 } else {
5634 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5635 unit = MINOR(dev);
5636 }
5637 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5638 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5639 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5640 break;
5641 }
1da177e4
LT		 5642
5643 md_probe(dev, NULL, NULL);
5644 mddev = mddev_find(dev);
9bbbca3a
NB		 5645 if (!mddev || !mddev->gendisk) {
5646 if (mddev)
5647 mddev_put(mddev);
5648 printk(KERN_ERR
1da177e4
LT		 5649 "md: cannot allocate memory for md drive.\n");
5650 break;
5651 }
f72ffdd6 5652 if (mddev_lock(mddev))
1da177e4
LT		 5653 printk(KERN_WARNING "md: %s locked, cannot run\n",
5654 mdname(mddev));
5655 else if (mddev->raid_disks || mddev->major_version
5656 || !list_empty(&mddev->disks)) {
f72ffdd6 5657 printk(KERN_WARNING
1da177e4
LT		 5658 "md: %s already running, cannot run %s\n",
5659 mdname(mddev), bdevname(rdev0->bdev,b));
5660 mddev_unlock(mddev);
5661 } else {
5662 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5663 mddev->persistent = 1;
159ec1fc 5664 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5665 list_del_init(&rdev->same_set);
5666 if (bind_rdev_to_array(rdev, mddev))
5667 export_rdev(rdev);
5668 }
5669 autorun_array(mddev);
5670 mddev_unlock(mddev);
5671 }
5672 /* on success, candidates will be empty, on error
5673 * it won't...
5674 */
159ec1fc 5675 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5676 list_del_init(&rdev->same_set);
1da177e4 5677 export_rdev(rdev);
4b80991c 5678 }
1da177e4
LT		 5679 mddev_put(mddev);
5680 }
5681 printk(KERN_INFO "md: ... autorun DONE.\n");
5682}
fdee8ae4 5683#endif /* !MODULE */
1da177e4 5684
f72ffdd6 5685static int get_version(void __user *arg)
1da177e4
LT		 5686{
5687 mdu_version_t ver;
5688
5689 ver.major = MD_MAJOR_VERSION;
5690 ver.minor = MD_MINOR_VERSION;
5691 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5692
5693 if (copy_to_user(arg, &ver, sizeof(ver)))
5694 return -EFAULT;
5695
5696 return 0;
5697}
5698
f72ffdd6 5699static int get_array_info(struct mddev *mddev, void __user *arg)
1da177e4
LT		 5700{
5701 mdu_array_info_t info;
a9f326eb 5702 int nr,working,insync,failed,spare;
3cb03002 5703 struct md_rdev *rdev;
1da177e4 5704
1ca69c4b
N		 5705 nr = working = insync = failed = spare = 0;
5706 rcu_read_lock();
5707 rdev_for_each_rcu(rdev, mddev) {
1da177e4 5708 nr++;
b2d444d7 5709 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5710 failed++;
5711 else {
5712 working++;
b2d444d7 5713 if (test_bit(In_sync, &rdev->flags))
f72ffdd6 5714 insync++;
1da177e4
LT		 5715 else
5716 spare++;
5717 }
5718 }
1ca69c4b 5719 rcu_read_unlock();
1da177e4
LT		 5720
5721 info.major_version = mddev->major_version;
5722 info.minor_version = mddev->minor_version;
5723 info.patch_version = MD_PATCHLEVEL_VERSION;
5724 info.ctime = mddev->ctime;
5725 info.level = mddev->level;
58c0fed4
AN		 5726 info.size = mddev->dev_sectors / 2;
5727 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5728 info.size = -1;
1da177e4
LT		 5729 info.nr_disks = nr;
5730 info.raid_disks = mddev->raid_disks;
5731 info.md_minor = mddev->md_minor;
5732 info.not_persistent= !mddev->persistent;
5733
5734 info.utime = mddev->utime;
5735 info.state = 0;
5736 if (mddev->in_sync)
5737 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5738 if (mddev->bitmap && mddev->bitmap_info.offset)
9bd35920 5739 info.state |= (1<<MD_SB_BITMAP_PRESENT);
ca8895d9
GR		 5740 if (mddev_is_clustered(mddev))
5741 info.state |= (1<<MD_SB_CLUSTERED);
a9f326eb 5742 info.active_disks = insync;
1da177e4
LT		 5743 info.working_disks = working;
5744 info.failed_disks = failed;
5745 info.spare_disks = spare;
5746
5747 info.layout = mddev->layout;
9d8f0363 5748 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5749
5750 if (copy_to_user(arg, &info, sizeof(info)))
5751 return -EFAULT;
5752
5753 return 0;
5754}
5755
f72ffdd6 5756static int get_bitmap_file(struct mddev *mddev, void __user * arg)
32a7627c
N		 5757{
5758 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
f4ad3d38 5759 char *ptr;
4af1a041 5760 int err;
32a7627c 5761
60559da4 5762 file = kmalloc(sizeof(*file), GFP_NOIO);
32a7627c 5763 if (!file)
4af1a041 5764 return -ENOMEM;
32a7627c 5765
4af1a041
N		 5766 err = 0;
5767 spin_lock(&mddev->lock);
32a7627c 5768 /* bitmap disabled, zero the first byte and copy out */
4af1a041 5769 if (!mddev->bitmap_info.file)
32a7627c 5770 file->pathname[0] = '\0';
4af1a041
N		 5771 else if ((ptr = d_path(&mddev->bitmap_info.file->f_path,
5772 file->pathname, sizeof(file->pathname))),
5773 IS_ERR(ptr))
5774 err = PTR_ERR(ptr);
5775 else
5776 memmove(file->pathname, ptr,
5777 sizeof(file->pathname)-(ptr-file->pathname));
5778 spin_unlock(&mddev->lock);
32a7627c 5779
4af1a041
N		 5780 if (err == 0 &&
5781 copy_to_user(arg, file, sizeof(*file)))
32a7627c 5782 err = -EFAULT;
4af1a041 5783
32a7627c
N		 5784 kfree(file);
5785 return err;
5786}
5787
f72ffdd6 5788static int get_disk_info(struct mddev *mddev, void __user * arg)
1da177e4
LT		 5789{
5790 mdu_disk_info_t info;
3cb03002 5791 struct md_rdev *rdev;
1da177e4
LT		 5792
5793 if (copy_from_user(&info, arg, sizeof(info)))
5794 return -EFAULT;
5795
1ca69c4b 5796 rcu_read_lock();
57d051dc 5797 rdev = md_find_rdev_nr_rcu(mddev, info.number);
1da177e4
LT		 5798 if (rdev) {
5799 info.major = MAJOR(rdev->bdev->bd_dev);
5800 info.minor = MINOR(rdev->bdev->bd_dev);
5801 info.raid_disk = rdev->raid_disk;
5802 info.state = 0;
b2d444d7 5803 if (test_bit(Faulty, &rdev->flags))
1da177e4 5804 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5805 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5806 info.state |= (1<<MD_DISK_ACTIVE);
5807 info.state |= (1<<MD_DISK_SYNC);
5808 }
8ddf9efe
N		 5809 if (test_bit(WriteMostly, &rdev->flags))
5810 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5811 } else {
5812 info.major = info.minor = 0;
5813 info.raid_disk = -1;
5814 info.state = (1<<MD_DISK_REMOVED);
5815 }
1ca69c4b 5816 rcu_read_unlock();
1da177e4
LT		 5817
5818 if (copy_to_user(arg, &info, sizeof(info)))
5819 return -EFAULT;
5820
5821 return 0;
5822}
5823
f72ffdd6 5824static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
1da177e4
LT		 5825{
5826 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5827 struct md_rdev *rdev;
1da177e4
LT		 5828 dev_t dev = MKDEV(info->major,info->minor);
5829
1aee41f6
GR		 5830 if (mddev_is_clustered(mddev) &&
5831 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
fa8259da 5832 pr_err("%s: Cannot add to clustered mddev.\n",
1aee41f6
GR		 5833 mdname(mddev));
5834 return -EINVAL;
5835 }
5836
1da177e4
LT		 5837 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5838 return -EOVERFLOW;
5839
5840 if (!mddev->raid_disks) {
5841 int err;
5842 /* expecting a device which has a superblock */
5843 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5844 if (IS_ERR(rdev)) {
f72ffdd6 5845 printk(KERN_WARNING
1da177e4
LT		 5846 "md: md_import_device returned %ld\n",
5847 PTR_ERR(rdev));
5848 return PTR_ERR(rdev);
5849 }
5850 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5851 struct md_rdev *rdev0
5852 = list_entry(mddev->disks.next,
5853 struct md_rdev, same_set);
a9f326eb 5854 err = super_types[mddev->major_version]
1da177e4
LT		 5855 .load_super(rdev, rdev0, mddev->minor_version);
5856 if (err < 0) {
f72ffdd6 5857 printk(KERN_WARNING
1da177e4 5858 "md: %s has different UUID to %s\n",
f72ffdd6 5859 bdevname(rdev->bdev,b),
1da177e4
LT		 5860 bdevname(rdev0->bdev,b2));
5861 export_rdev(rdev);
5862 return -EINVAL;
5863 }
5864 }
5865 err = bind_rdev_to_array(rdev, mddev);
5866 if (err)
5867 export_rdev(rdev);
5868 return err;
5869 }
5870
5871 /*
5872 * add_new_disk can be used once the array is assembled
5873 * to add "hot spares". They must already have a superblock
5874 * written
5875 */
5876 if (mddev->pers) {
5877 int err;
5878 if (!mddev->pers->hot_add_disk) {
f72ffdd6 5879 printk(KERN_WARNING
1da177e4
LT		 5880 "%s: personality does not support diskops!\n",
5881 mdname(mddev));
5882 return -EINVAL;
5883 }
7b1e35f6
N		 5884 if (mddev->persistent)
5885 rdev = md_import_device(dev, mddev->major_version,
5886 mddev->minor_version);
5887 else
5888 rdev = md_import_device(dev, -1, -1);
1da177e4 5889 if (IS_ERR(rdev)) {
f72ffdd6 5890 printk(KERN_WARNING
1da177e4
LT		 5891 "md: md_import_device returned %ld\n",
5892 PTR_ERR(rdev));
5893 return PTR_ERR(rdev);
5894 }
1a855a06 5895 /* set saved_raid_disk if appropriate */
41158c7e
N		 5896 if (!mddev->persistent) {
5897 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5898 info->raid_disk < mddev->raid_disks) {
41158c7e 5899 rdev->raid_disk = info->raid_disk;
bf572541 5900 set_bit(In_sync, &rdev->flags);
8313b8e5 5901 clear_bit(Bitmap_sync, &rdev->flags);
bf572541 5902 } else
41158c7e 5903 rdev->raid_disk = -1;
f466722c 5904 rdev->saved_raid_disk = rdev->raid_disk;
41158c7e
N		 5905 } else
5906 super_types[mddev->major_version].
5907 validate_super(mddev, rdev);
bedd86b7 5908 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 5909 rdev->raid_disk != info->raid_disk) {
bedd86b7
N		 5910 /* This was a hot-add request, but events doesn't
5911 * match, so reject it.
5912 */
5913 export_rdev(rdev);
5914 return -EINVAL;
5915 }
5916
b2d444d7 5917 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5918 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5919 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5920 else
5921 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5922
1aee41f6
GR		 5923 /*
5924 * check whether the device shows up in other nodes
5925 */
5926 if (mddev_is_clustered(mddev)) {
5927 if (info->state & (1 << MD_DISK_CANDIDATE)) {
5928 /* Through --cluster-confirm */
5929 set_bit(Candidate, &rdev->flags);
fa8259da
GR		 5930 err = md_cluster_ops->new_disk_ack(mddev, true);
5931 if (err) {
5932 export_rdev(rdev);
5933 return err;
5934 }
1aee41f6
GR		 5935 } else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
5936 /* --add initiated by this node */
5937 err = md_cluster_ops->add_new_disk_start(mddev, rdev);
5938 if (err) {
5939 md_cluster_ops->add_new_disk_finish(mddev);
5940 export_rdev(rdev);
5941 return err;
5942 }
5943 }
5944 }
5945
1da177e4
LT		 5946 rdev->raid_disk = -1;
5947 err = bind_rdev_to_array(rdev, mddev);
5948 if (err)
5949 export_rdev(rdev);
52664732 5950 else
a6da4ef8 5951 err = add_bound_rdev(rdev);
1aee41f6
GR		 5952 if (mddev_is_clustered(mddev) &&
5953 (info->state & (1 << MD_DISK_CLUSTER_ADD)))
5954 md_cluster_ops->add_new_disk_finish(mddev);
1da177e4
LT		 5955 return err;
5956 }
5957
5958 /* otherwise, add_new_disk is only allowed
5959 * for major_version==0 superblocks
5960 */
5961 if (mddev->major_version != 0) {
5962 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5963 mdname(mddev));
5964 return -EINVAL;
5965 }
5966
5967 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5968 int err;
d710e138 5969 rdev = md_import_device(dev, -1, 0);
1da177e4 5970 if (IS_ERR(rdev)) {
f72ffdd6 5971 printk(KERN_WARNING
1da177e4
LT		 5972 "md: error, md_import_device() returned %ld\n",
5973 PTR_ERR(rdev));
5974 return PTR_ERR(rdev);
5975 }
5976 rdev->desc_nr = info->number;
5977 if (info->raid_disk < mddev->raid_disks)
5978 rdev->raid_disk = info->raid_disk;
5979 else
5980 rdev->raid_disk = -1;
5981
1da177e4 5982 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5983 if (info->state & (1<<MD_DISK_SYNC))
5984 set_bit(In_sync, &rdev->flags);
1da177e4 5985
8ddf9efe
N		 5986 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5987 set_bit(WriteMostly, &rdev->flags);
5988
1da177e4
LT		 5989 if (!mddev->persistent) {
5990 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5991 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5992 } else
57b2caa3 5993 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5994 rdev->sectors = rdev->sb_start;
1da177e4 5995
2bf071bf
N		 5996 err = bind_rdev_to_array(rdev, mddev);
5997 if (err) {
5998 export_rdev(rdev);
5999 return err;
6000 }
1da177e4
LT		 6001 }
6002
6003 return 0;
6004}
6005
f72ffdd6 6006static int hot_remove_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 6007{
6008 char b[BDEVNAME_SIZE];
3cb03002 6009 struct md_rdev *rdev;
1da177e4 6010
1da177e4
LT		 6011 rdev = find_rdev(mddev, dev);
6012 if (!rdev)
6013 return -ENXIO;
6014
293467aa
GR		 6015 if (mddev_is_clustered(mddev))
6016 md_cluster_ops->metadata_update_start(mddev);
6017
3ea8929d
N		 6018 clear_bit(Blocked, &rdev->flags);
6019 remove_and_add_spares(mddev, rdev);
6020
1da177e4
LT		 6021 if (rdev->raid_disk >= 0)
6022 goto busy;
6023
88bcfef7
GR		 6024 if (mddev_is_clustered(mddev))
6025 md_cluster_ops->remove_disk(mddev, rdev);
6026
fb56dfef 6027 md_kick_rdev_from_array(rdev);
850b2b42 6028 md_update_sb(mddev, 1);
d7603b7e 6029 md_new_event(mddev);
1da177e4 6030
293467aa
GR		 6031 if (mddev_is_clustered(mddev))
6032 md_cluster_ops->metadata_update_finish(mddev);
6033
1da177e4
LT		 6034 return 0;
6035busy:
293467aa
GR		 6036 if (mddev_is_clustered(mddev))
6037 md_cluster_ops->metadata_update_cancel(mddev);
fdefa4d8 6038 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 6039 bdevname(rdev->bdev,b), mdname(mddev));
6040 return -EBUSY;
6041}
6042
f72ffdd6 6043static int hot_add_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 6044{
6045 char b[BDEVNAME_SIZE];
6046 int err;
3cb03002 6047 struct md_rdev *rdev;
1da177e4
LT		 6048
6049 if (!mddev->pers)
6050 return -ENODEV;
6051
6052 if (mddev->major_version != 0) {
6053 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6054 " version-0 superblocks.\n",
6055 mdname(mddev));
6056 return -EINVAL;
6057 }
6058 if (!mddev->pers->hot_add_disk) {
f72ffdd6 6059 printk(KERN_WARNING
1da177e4
LT		 6060 "%s: personality does not support diskops!\n",
6061 mdname(mddev));
6062 return -EINVAL;
6063 }
6064
d710e138 6065 rdev = md_import_device(dev, -1, 0);
1da177e4 6066 if (IS_ERR(rdev)) {
f72ffdd6 6067 printk(KERN_WARNING
1da177e4
LT		 6068 "md: error, md_import_device() returned %ld\n",
6069 PTR_ERR(rdev));
6070 return -EINVAL;
6071 }
6072
6073 if (mddev->persistent)
57b2caa3 6074 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 6075 else
77304d2a 6076 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 6077
8190e754 6078 rdev->sectors = rdev->sb_start;
1da177e4 6079
b2d444d7 6080 if (test_bit(Faulty, &rdev->flags)) {
f72ffdd6 6081 printk(KERN_WARNING
1da177e4
LT		 6082 "md: can not hot-add faulty %s disk to %s!\n",
6083 bdevname(rdev->bdev,b), mdname(mddev));
6084 err = -EINVAL;
6085 goto abort_export;
6086 }
293467aa
GR		 6087
6088 if (mddev_is_clustered(mddev))
6089 md_cluster_ops->metadata_update_start(mddev);
b2d444d7 6090 clear_bit(In_sync, &rdev->flags);
1da177e4 6091 rdev->desc_nr = -1;
5842730d 6092 rdev->saved_raid_disk = -1;
2bf071bf
N		 6093 err = bind_rdev_to_array(rdev, mddev);
6094 if (err)
293467aa 6095 goto abort_clustered;
1da177e4
LT		 6096
6097 /*
6098 * The rest should better be atomic, we can have disk failures
6099 * noticed in interrupt contexts ...
6100 */
6101
1da177e4
LT		 6102 rdev->raid_disk = -1;
6103
850b2b42 6104 md_update_sb(mddev, 1);
1da177e4 6105
293467aa
GR		 6106 if (mddev_is_clustered(mddev))
6107 md_cluster_ops->metadata_update_finish(mddev);
1da177e4
LT		 6108 /*
6109 * Kick recovery, maybe this spare has to be added to the
6110 * array immediately.
6111 */
6112 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6113 md_wakeup_thread(mddev->thread);
d7603b7e 6114 md_new_event(mddev);
1da177e4
LT		 6115 return 0;
6116
293467aa
GR		 6117abort_clustered:
6118 if (mddev_is_clustered(mddev))
6119 md_cluster_ops->metadata_update_cancel(mddev);
1da177e4
LT		 6120abort_export:
6121 export_rdev(rdev);
6122 return err;
6123}
6124
fd01b88c 6125static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c 6126{
035328c2 6127 int err = 0;
32a7627c 6128
36fa3063 6129 if (mddev->pers) {
d66b1b39 6130 if (!mddev->pers->quiesce || !mddev->thread)
36fa3063
N		 6131 return -EBUSY;
6132 if (mddev->recovery || mddev->sync_thread)
6133 return -EBUSY;
6134 /* we should be able to change the bitmap.. */
6135 }
32a7627c 6136
36fa3063 6137 if (fd >= 0) {
035328c2 6138 struct inode *inode;
1e594bb2
N		 6139 struct file *f;
6140
6141 if (mddev->bitmap || mddev->bitmap_info.file)
36fa3063 6142 return -EEXIST; /* cannot add when bitmap is present */
1e594bb2 6143 f = fget(fd);
32a7627c 6144
1e594bb2 6145 if (f == NULL) {
36fa3063
N		 6146 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6147 mdname(mddev));
6148 return -EBADF;
6149 }
6150
1e594bb2 6151 inode = f->f_mapping->host;
035328c2
N		 6152 if (!S_ISREG(inode->i_mode)) {
6153 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6154 mdname(mddev));
6155 err = -EBADF;
1e594bb2 6156 } else if (!(f->f_mode & FMODE_WRITE)) {
035328c2
N		 6157 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6158 mdname(mddev));
6159 err = -EBADF;
6160 } else if (atomic_read(&inode->i_writecount) != 1) {
36fa3063
N		 6161 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6162 mdname(mddev));
035328c2
N		 6163 err = -EBUSY;
6164 }
6165 if (err) {
1e594bb2 6166 fput(f);
36fa3063
N		 6167 return err;
6168 }
1e594bb2 6169 mddev->bitmap_info.file = f;
c3d9714e 6170 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 6171 } else if (mddev->bitmap == NULL)
6172 return -ENOENT; /* cannot remove what isn't there */
6173 err = 0;
6174 if (mddev->pers) {
6175 mddev->pers->quiesce(mddev, 1);
69e51b44 6176 if (fd >= 0) {
f9209a32
GR		 6177 struct bitmap *bitmap;
6178
6179 bitmap = bitmap_create(mddev, -1);
6180 if (!IS_ERR(bitmap)) {
6181 mddev->bitmap = bitmap;
69e51b44 6182 err = bitmap_load(mddev);
ba599aca
N		 6183 } else
6184 err = PTR_ERR(bitmap);
69e51b44 6185 }
d7375ab3 6186 if (fd < 0 || err) {
36fa3063 6187 bitmap_destroy(mddev);
d7375ab3
N		 6188 fd = -1; /* make sure to put the file */
6189 }
36fa3063 6190 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 6191 }
6192 if (fd < 0) {
4af1a041
N		 6193 struct file *f = mddev->bitmap_info.file;
6194 if (f) {
6195 spin_lock(&mddev->lock);
6196 mddev->bitmap_info.file = NULL;
6197 spin_unlock(&mddev->lock);
6198 fput(f);
6199 }
36fa3063
N		 6200 }
6201
32a7627c
N		 6202 return err;
6203}
6204
1da177e4
LT		 6205/*
6206 * set_array_info is used two different ways
6207 * The original usage is when creating a new array.
6208 * In this usage, raid_disks is > 0 and it together with
6209 * level, size, not_persistent,layout,chunksize determine the
6210 * shape of the array.
6211 * This will always create an array with a type-0.90.0 superblock.
6212 * The newer usage is when assembling an array.
6213 * In this case raid_disks will be 0, and the major_version field is
6214 * use to determine which style super-blocks are to be found on the devices.
6215 * The minor and patch _version numbers are also kept incase the
6216 * super_block handler wishes to interpret them.
6217 */
f72ffdd6 6218static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6219{
6220
6221 if (info->raid_disks == 0) {
6222 /* just setting version number for superblock loading */
6223 if (info->major_version < 0 ||
50511da3 6224 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 6225 super_types[info->major_version].name == NULL) {
6226 /* maybe try to auto-load a module? */
f72ffdd6 6227 printk(KERN_INFO
1da177e4
LT		 6228 "md: superblock version %d not known\n",
6229 info->major_version);
6230 return -EINVAL;
6231 }
6232 mddev->major_version = info->major_version;
6233 mddev->minor_version = info->minor_version;
6234 mddev->patch_version = info->patch_version;
3f9d7b0d 6235 mddev->persistent = !info->not_persistent;
cbd19983
N		 6236 /* ensure mddev_put doesn't delete this now that there
6237 * is some minimal configuration.
6238 */
6239 mddev->ctime = get_seconds();
1da177e4
LT		 6240 return 0;
6241 }
6242 mddev->major_version = MD_MAJOR_VERSION;
6243 mddev->minor_version = MD_MINOR_VERSION;
6244 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6245 mddev->ctime = get_seconds();
6246
6247 mddev->level = info->level;
17115e03 6248 mddev->clevel[0] = 0;
58c0fed4 6249 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 6250 mddev->raid_disks = info->raid_disks;
6251 /* don't set md_minor, it is determined by which /dev/md* was
6252 * openned
6253 */
6254 if (info->state & (1<<MD_SB_CLEAN))
6255 mddev->recovery_cp = MaxSector;
6256 else
6257 mddev->recovery_cp = 0;
6258 mddev->persistent = ! info->not_persistent;
e691063a 6259 mddev->external = 0;
1da177e4
LT		 6260
6261 mddev->layout = info->layout;
9d8f0363 6262 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 6263
6264 mddev->max_disks = MD_SB_DISKS;
6265
e691063a
N		 6266 if (mddev->persistent)
6267 mddev->flags = 0;
850b2b42 6268 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6269
c3d9714e 6270 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6271 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6272 mddev->bitmap_info.offset = 0;
b2a2703c 6273
f6705578
N		 6274 mddev->reshape_position = MaxSector;
6275
1da177e4
LT		 6276 /*
6277 * Generate a 128 bit UUID
6278 */
6279 get_random_bytes(mddev->uuid, 16);
6280
f6705578 6281 mddev->new_level = mddev->level;
664e7c41 6282 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 6283 mddev->new_layout = mddev->layout;
6284 mddev->delta_disks = 0;
2c810cdd 6285 mddev->reshape_backwards = 0;
f6705578 6286
1da177e4
LT		 6287 return 0;
6288}
6289
fd01b88c 6290void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6291{
b522adcd
DW		 6292 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6293
6294 if (mddev->external_size)
6295 return;
6296
1f403624
DW		 6297 mddev->array_sectors = array_sectors;
6298}
6299EXPORT_SYMBOL(md_set_array_sectors);
6300
fd01b88c 6301static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6302{
3cb03002 6303 struct md_rdev *rdev;
a35b0d69 6304 int rv;
d71f9f88 6305 int fit = (num_sectors == 0);
a35b0d69
N		 6306
6307 if (mddev->pers->resize == NULL)
6308 return -EINVAL;
d71f9f88
AN		 6309 /* The "num_sectors" is the number of sectors of each device that
6310 * is used. This can only make sense for arrays with redundancy.
6311 * linear and raid0 always use whatever space is available. We can only
6312 * consider changing this number if no resync or reconstruction is
6313 * happening, and if the new size is acceptable. It must fit before the
0f420358 6314 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 6315 * of each device. If num_sectors is zero, we find the largest size
6316 * that fits.
a35b0d69 6317 */
f851b60d
N		 6318 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6319 mddev->sync_thread)
a35b0d69 6320 return -EBUSY;
bd8839e0
N		 6321 if (mddev->ro)
6322 return -EROFS;
a4a6125a 6323
dafb20fa 6324 rdev_for_each(rdev, mddev) {
dd8ac336 6325 sector_t avail = rdev->sectors;
01ab5662 6326
d71f9f88
AN		 6327 if (fit && (num_sectors == 0 || num_sectors > avail))
6328 num_sectors = avail;
6329 if (avail < num_sectors)
a35b0d69
N		 6330 return -ENOSPC;
6331 }
d71f9f88 6332 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 6333 if (!rv)
6334 revalidate_disk(mddev->gendisk);
a35b0d69
N		 6335 return rv;
6336}
6337
fd01b88c 6338static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 6339{
6340 int rv;
c6563a8c 6341 struct md_rdev *rdev;
da943b99 6342 /* change the number of raid disks */
63c70c4f 6343 if (mddev->pers->check_reshape == NULL)
da943b99 6344 return -EINVAL;
bd8839e0
N		 6345 if (mddev->ro)
6346 return -EROFS;
da943b99 6347 if (raid_disks <= 0 ||
233fca36 6348 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 6349 return -EINVAL;
f851b60d
N		 6350 if (mddev->sync_thread ||
6351 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6352 mddev->reshape_position != MaxSector)
da943b99 6353 return -EBUSY;
c6563a8c
N		 6354
6355 rdev_for_each(rdev, mddev) {
6356 if (mddev->raid_disks < raid_disks &&
6357 rdev->data_offset < rdev->new_data_offset)
6358 return -EINVAL;
6359 if (mddev->raid_disks > raid_disks &&
6360 rdev->data_offset > rdev->new_data_offset)
6361 return -EINVAL;
6362 }
6363
63c70c4f 6364 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N		 6365 if (mddev->delta_disks < 0)
6366 mddev->reshape_backwards = 1;
6367 else if (mddev->delta_disks > 0)
6368 mddev->reshape_backwards = 0;
63c70c4f
N		 6369
6370 rv = mddev->pers->check_reshape(mddev);
2c810cdd 6371 if (rv < 0) {
de171cb9 6372 mddev->delta_disks = 0;
2c810cdd
N		 6373 mddev->reshape_backwards = 0;
6374 }
da943b99
N		 6375 return rv;
6376}
6377
1da177e4
LT		 6378/*
6379 * update_array_info is used to change the configuration of an
6380 * on-line array.
6381 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6382 * fields in the info are checked against the array.
6383 * Any differences that cannot be handled will cause an error.
6384 * Normally, only one change can be managed at a time.
6385 */
fd01b88c 6386static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6387{
6388 int rv = 0;
6389 int cnt = 0;
36fa3063
N		 6390 int state = 0;
6391
6392 /* calculate expected state,ignoring low bits */
c3d9714e 6393 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 6394 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 6395
6396 if (mddev->major_version != info->major_version ||
6397 mddev->minor_version != info->minor_version ||
6398/* mddev->patch_version != info->patch_version || */
6399 mddev->ctime != info->ctime ||
6400 mddev->level != info->level ||
6401/* mddev->layout != info->layout || */
4e023612 6402 mddev->persistent != !info->not_persistent ||
9d8f0363 6403 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 6404 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6405 ((state^info->state) & 0xfffffe00)
6406 )
1da177e4
LT		 6407 return -EINVAL;
6408 /* Check there is only one change */
58c0fed4
AN		 6409 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6410 cnt++;
6411 if (mddev->raid_disks != info->raid_disks)
6412 cnt++;
6413 if (mddev->layout != info->layout)
6414 cnt++;
6415 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6416 cnt++;
6417 if (cnt == 0)
6418 return 0;
6419 if (cnt > 1)
6420 return -EINVAL;
1da177e4
LT		 6421
6422 if (mddev->layout != info->layout) {
6423 /* Change layout
6424 * we don't need to do anything at the md level, the
6425 * personality will take care of it all.
6426 */
50ac168a 6427 if (mddev->pers->check_reshape == NULL)
1da177e4 6428 return -EINVAL;
597a711b
N		 6429 else {
6430 mddev->new_layout = info->layout;
50ac168a 6431 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 6432 if (rv)
6433 mddev->new_layout = mddev->layout;
6434 return rv;
6435 }
1da177e4 6436 }
293467aa
GR		 6437 if (mddev_is_clustered(mddev))
6438 md_cluster_ops->metadata_update_start(mddev);
58c0fed4 6439 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6440 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6441
da943b99
N		 6442 if (mddev->raid_disks != info->raid_disks)
6443 rv = update_raid_disks(mddev, info->raid_disks);
6444
36fa3063 6445 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
293467aa
GR		 6446 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6447 rv = -EINVAL;
6448 goto err;
6449 }
6450 if (mddev->recovery || mddev->sync_thread) {
6451 rv = -EBUSY;
6452 goto err;
6453 }
36fa3063 6454 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
f9209a32 6455 struct bitmap *bitmap;
36fa3063 6456 /* add the bitmap */
293467aa
GR		 6457 if (mddev->bitmap) {
6458 rv = -EEXIST;
6459 goto err;
6460 }
6461 if (mddev->bitmap_info.default_offset == 0) {
6462 rv = -EINVAL;
6463 goto err;
6464 }
c3d9714e
N		 6465 mddev->bitmap_info.offset =
6466 mddev->bitmap_info.default_offset;
6409bb05
N		 6467 mddev->bitmap_info.space =
6468 mddev->bitmap_info.default_space;
36fa3063 6469 mddev->pers->quiesce(mddev, 1);
f9209a32
GR		 6470 bitmap = bitmap_create(mddev, -1);
6471 if (!IS_ERR(bitmap)) {
6472 mddev->bitmap = bitmap;
69e51b44 6473 rv = bitmap_load(mddev);
ba599aca
N		 6474 } else
6475 rv = PTR_ERR(bitmap);
36fa3063
N		 6476 if (rv)
6477 bitmap_destroy(mddev);
6478 mddev->pers->quiesce(mddev, 0);
6479 } else {
6480 /* remove the bitmap */
293467aa
GR		 6481 if (!mddev->bitmap) {
6482 rv = -ENOENT;
6483 goto err;
6484 }
6485 if (mddev->bitmap->storage.file) {
6486 rv = -EINVAL;
6487 goto err;
6488 }
36fa3063
N		 6489 mddev->pers->quiesce(mddev, 1);
6490 bitmap_destroy(mddev);
6491 mddev->pers->quiesce(mddev, 0);
c3d9714e 6492 mddev->bitmap_info.offset = 0;
36fa3063
N		 6493 }
6494 }
850b2b42 6495 md_update_sb(mddev, 1);
293467aa
GR		 6496 if (mddev_is_clustered(mddev))
6497 md_cluster_ops->metadata_update_finish(mddev);
6498 return rv;
6499err:
6500 if (mddev_is_clustered(mddev))
6501 md_cluster_ops->metadata_update_cancel(mddev);
1da177e4
LT		 6502 return rv;
6503}
6504
fd01b88c 6505static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6506{
3cb03002 6507 struct md_rdev *rdev;
1ca69c4b 6508 int err = 0;
1da177e4
LT		 6509
6510 if (mddev->pers == NULL)
6511 return -ENODEV;
6512
1ca69c4b
N		 6513 rcu_read_lock();
6514 rdev = find_rdev_rcu(mddev, dev);
1da177e4 6515 if (!rdev)
1ca69c4b
N		 6516 err = -ENODEV;
6517 else {
6518 md_error(mddev, rdev);
6519 if (!test_bit(Faulty, &rdev->flags))
6520 err = -EBUSY;
6521 }
6522 rcu_read_unlock();
6523 return err;
1da177e4
LT		 6524}
6525
2f9618ce
AN		 6526/*
6527 * We have a problem here : there is no easy way to give a CHS
6528 * virtual geometry. We currently pretend that we have a 2 heads
6529 * 4 sectors (with a BIG number of cylinders...). This drives
6530 * dosfs just mad... ;-)
6531 */
a885c8c4
CH		 6532static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6533{
fd01b88c 6534 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6535
6536 geo->heads = 2;
6537 geo->sectors = 4;
49ce6cea 6538 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6539 return 0;
6540}
6541
cb335f88
NS		 6542static inline bool md_ioctl_valid(unsigned int cmd)
6543{
6544 switch (cmd) {
6545 case ADD_NEW_DISK:
6546 case BLKROSET:
6547 case GET_ARRAY_INFO:
6548 case GET_BITMAP_FILE:
6549 case GET_DISK_INFO:
6550 case HOT_ADD_DISK:
6551 case HOT_REMOVE_DISK:
cb335f88
NS		 6552 case RAID_AUTORUN:
6553 case RAID_VERSION:
6554 case RESTART_ARRAY_RW:
6555 case RUN_ARRAY:
6556 case SET_ARRAY_INFO:
6557 case SET_BITMAP_FILE:
6558 case SET_DISK_FAULTY:
6559 case STOP_ARRAY:
6560 case STOP_ARRAY_RO:
1aee41f6 6561 case CLUSTERED_DISK_NACK:
cb335f88
NS		 6562 return true;
6563 default:
6564 return false;
6565 }
6566}
6567
a39907fa 6568static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6569 unsigned int cmd, unsigned long arg)
6570{
6571 int err = 0;
6572 void __user *argp = (void __user *)arg;
fd01b88c 6573 struct mddev *mddev = NULL;
e2218350 6574 int ro;
1da177e4 6575
cb335f88
NS		 6576 if (!md_ioctl_valid(cmd))
6577 return -ENOTTY;
6578
506c9e44
N		 6579 switch (cmd) {
6580 case RAID_VERSION:
6581 case GET_ARRAY_INFO:
6582 case GET_DISK_INFO:
6583 break;
6584 default:
6585 if (!capable(CAP_SYS_ADMIN))
6586 return -EACCES;
6587 }
1da177e4
LT		 6588
6589 /*
6590 * Commands dealing with the RAID driver but not any
6591 * particular array:
6592 */
c02c0aeb
N		 6593 switch (cmd) {
6594 case RAID_VERSION:
6595 err = get_version(argp);
3adc28d8 6596 goto out;
1da177e4 6597
1da177e4 6598#ifndef MODULE
c02c0aeb
N		 6599 case RAID_AUTORUN:
6600 err = 0;
6601 autostart_arrays(arg);
3adc28d8 6602 goto out;
1da177e4 6603#endif
c02c0aeb 6604 default:;
1da177e4
LT		 6605 }
6606
6607 /*
6608 * Commands creating/starting a new array:
6609 */
6610
a39907fa 6611 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6612
6613 if (!mddev) {
6614 BUG();
3adc28d8 6615 goto out;
1da177e4
LT		 6616 }
6617
1ca69c4b
N		 6618 /* Some actions do not requires the mutex */
6619 switch (cmd) {
6620 case GET_ARRAY_INFO:
6621 if (!mddev->raid_disks && !mddev->external)
6622 err = -ENODEV;
6623 else
6624 err = get_array_info(mddev, argp);
3adc28d8 6625 goto out;
1ca69c4b
N		 6626
6627 case GET_DISK_INFO:
6628 if (!mddev->raid_disks && !mddev->external)
6629 err = -ENODEV;
6630 else
6631 err = get_disk_info(mddev, argp);
3adc28d8 6632 goto out;
1ca69c4b
N		 6633
6634 case SET_DISK_FAULTY:
6635 err = set_disk_faulty(mddev, new_decode_dev(arg));
3adc28d8 6636 goto out;
4af1a041
N		 6637
6638 case GET_BITMAP_FILE:
6639 err = get_bitmap_file(mddev, argp);
6640 goto out;
6641
1ca69c4b
N		 6642 }
6643
a7a3f08d
N		 6644 if (cmd == ADD_NEW_DISK)
6645 /* need to ensure md_delayed_delete() has completed */
6646 flush_workqueue(md_misc_wq);
6647
90f5f7ad
HR		 6648 if (cmd == HOT_REMOVE_DISK)
6649 /* need to ensure recovery thread has run */
6650 wait_event_interruptible_timeout(mddev->sb_wait,
6651 !test_bit(MD_RECOVERY_NEEDED,
6652 &mddev->flags),
6653 msecs_to_jiffies(5000));
260fa034
N		 6654 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6655 /* Need to flush page cache, and ensure no-one else opens
6656 * and writes
6657 */
6658 mutex_lock(&mddev->open_mutex);
9ba3b7f5 6659 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
260fa034
N		 6660 mutex_unlock(&mddev->open_mutex);
6661 err = -EBUSY;
3adc28d8 6662 goto out;
260fa034
N		 6663 }
6664 set_bit(MD_STILL_CLOSED, &mddev->flags);
6665 mutex_unlock(&mddev->open_mutex);
6666 sync_blockdev(bdev);
6667 }
1da177e4
LT		 6668 err = mddev_lock(mddev);
6669 if (err) {
f72ffdd6 6670 printk(KERN_INFO
1da177e4
LT		 6671 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6672 err, cmd);
3adc28d8 6673 goto out;
1da177e4
LT		 6674 }
6675
c02c0aeb
N		 6676 if (cmd == SET_ARRAY_INFO) {
6677 mdu_array_info_t info;
6678 if (!arg)
6679 memset(&info, 0, sizeof(info));
6680 else if (copy_from_user(&info, argp, sizeof(info))) {
6681 err = -EFAULT;
3adc28d8 6682 goto unlock;
c02c0aeb
N		 6683 }
6684 if (mddev->pers) {
6685 err = update_array_info(mddev, &info);
6686 if (err) {
6687 printk(KERN_WARNING "md: couldn't update"
6688 " array info. %d\n", err);
3adc28d8 6689 goto unlock;
1da177e4 6690 }
3adc28d8 6691 goto unlock;
c02c0aeb
N		 6692 }
6693 if (!list_empty(&mddev->disks)) {
6694 printk(KERN_WARNING
6695 "md: array %s already has disks!\n",
6696 mdname(mddev));
6697 err = -EBUSY;
3adc28d8 6698 goto unlock;
c02c0aeb
N		 6699 }
6700 if (mddev->raid_disks) {
6701 printk(KERN_WARNING
6702 "md: array %s already initialised!\n",
6703 mdname(mddev));
6704 err = -EBUSY;
3adc28d8 6705 goto unlock;
c02c0aeb
N		 6706 }
6707 err = set_array_info(mddev, &info);
6708 if (err) {
6709 printk(KERN_WARNING "md: couldn't set"
6710 " array info. %d\n", err);
3adc28d8 6711 goto unlock;
c02c0aeb 6712 }
3adc28d8 6713 goto unlock;
1da177e4
LT		 6714 }
6715
6716 /*
6717 * Commands querying/configuring an existing array:
6718 */
32a7627c 6719 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6720 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6721 if ((!mddev->raid_disks && !mddev->external)
6722 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6723 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6724 && cmd != GET_BITMAP_FILE) {
1da177e4 6725 err = -ENODEV;
3adc28d8 6726 goto unlock;
1da177e4
LT		 6727 }
6728
6729 /*
6730 * Commands even a read-only array can execute:
6731 */
c02c0aeb 6732 switch (cmd) {
c02c0aeb
N		 6733 case RESTART_ARRAY_RW:
6734 err = restart_array(mddev);
3adc28d8 6735 goto unlock;
1da177e4 6736
c02c0aeb
N		 6737 case STOP_ARRAY:
6738 err = do_md_stop(mddev, 0, bdev);
3adc28d8 6739 goto unlock;
1da177e4 6740
c02c0aeb
N		 6741 case STOP_ARRAY_RO:
6742 err = md_set_readonly(mddev, bdev);
3adc28d8 6743 goto unlock;
1da177e4 6744
3ea8929d
N		 6745 case HOT_REMOVE_DISK:
6746 err = hot_remove_disk(mddev, new_decode_dev(arg));
3adc28d8 6747 goto unlock;
3ea8929d 6748
7ceb17e8
N		 6749 case ADD_NEW_DISK:
6750 /* We can support ADD_NEW_DISK on read-only arrays
6751 * on if we are re-adding a preexisting device.
6752 * So require mddev->pers and MD_DISK_SYNC.
6753 */
6754 if (mddev->pers) {
6755 mdu_disk_info_t info;
6756 if (copy_from_user(&info, argp, sizeof(info)))
6757 err = -EFAULT;
6758 else if (!(info.state & (1<<MD_DISK_SYNC)))
6759 /* Need to clear read-only for this */
6760 break;
6761 else
6762 err = add_new_disk(mddev, &info);
3adc28d8 6763 goto unlock;
7ceb17e8
N		 6764 }
6765 break;
6766
c02c0aeb
N		 6767 case BLKROSET:
6768 if (get_user(ro, (int __user *)(arg))) {
6769 err = -EFAULT;
3adc28d8 6770 goto unlock;
c02c0aeb
N		 6771 }
6772 err = -EINVAL;
e2218350 6773
c02c0aeb
N		 6774 /* if the bdev is going readonly the value of mddev->ro
6775 * does not matter, no writes are coming
6776 */
6777 if (ro)
3adc28d8 6778 goto unlock;
e2218350 6779
c02c0aeb
N		 6780 /* are we are already prepared for writes? */
6781 if (mddev->ro != 1)
3adc28d8 6782 goto unlock;
e2218350 6783
c02c0aeb
N		 6784 /* transitioning to readauto need only happen for
6785 * arrays that call md_write_start
6786 */
6787 if (mddev->pers) {
6788 err = restart_array(mddev);
6789 if (err == 0) {
6790 mddev->ro = 2;
6791 set_disk_ro(mddev->gendisk, 0);
e2218350 6792 }
c02c0aeb 6793 }
3adc28d8 6794 goto unlock;
1da177e4
LT		 6795 }
6796
6797 /*
6798 * The remaining ioctls are changing the state of the
f91de92e 6799 * superblock, so we do not allow them on read-only arrays.
1da177e4 6800 */
326eb17d 6801 if (mddev->ro && mddev->pers) {
f91de92e
N		 6802 if (mddev->ro == 2) {
6803 mddev->ro = 0;
00bcb4ac 6804 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 6805 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N		 6806 /* mddev_unlock will wake thread */
6807 /* If a device failed while we were read-only, we
6808 * need to make sure the metadata is updated now.
6809 */
6810 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6811 mddev_unlock(mddev);
6812 wait_event(mddev->sb_wait,
6813 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6814 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
29f097c4 6815 mddev_lock_nointr(mddev);
f3378b48 6816 }
f91de92e
N		 6817 } else {
6818 err = -EROFS;
3adc28d8 6819 goto unlock;
f91de92e 6820 }
1da177e4
LT		 6821 }
6822
c02c0aeb
N		 6823 switch (cmd) {
6824 case ADD_NEW_DISK:
1da177e4 6825 {
c02c0aeb
N		 6826 mdu_disk_info_t info;
6827 if (copy_from_user(&info, argp, sizeof(info)))
6828 err = -EFAULT;
6829 else
6830 err = add_new_disk(mddev, &info);
3adc28d8 6831 goto unlock;
c02c0aeb 6832 }
1da177e4 6833
1aee41f6
GR		 6834 case CLUSTERED_DISK_NACK:
6835 if (mddev_is_clustered(mddev))
6836 md_cluster_ops->new_disk_ack(mddev, false);
6837 else
6838 err = -EINVAL;
6839 goto unlock;
6840
c02c0aeb
N		 6841 case HOT_ADD_DISK:
6842 err = hot_add_disk(mddev, new_decode_dev(arg));
3adc28d8 6843 goto unlock;
1da177e4 6844
c02c0aeb
N		 6845 case RUN_ARRAY:
6846 err = do_md_run(mddev);
3adc28d8 6847 goto unlock;
1da177e4 6848
c02c0aeb
N		 6849 case SET_BITMAP_FILE:
6850 err = set_bitmap_file(mddev, (int)arg);
3adc28d8 6851 goto unlock;
32a7627c 6852
c02c0aeb
N		 6853 default:
6854 err = -EINVAL;
3adc28d8 6855 goto unlock;
1da177e4
LT		 6856 }
6857
3adc28d8 6858unlock:
d3374825
N		 6859 if (mddev->hold_active == UNTIL_IOCTL &&
6860 err != -EINVAL)
6861 mddev->hold_active = 0;
1da177e4 6862 mddev_unlock(mddev);
3adc28d8 6863out:
1da177e4
LT		 6864 return err;
6865}
aa98aa31
AB		 6866#ifdef CONFIG_COMPAT
6867static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6868 unsigned int cmd, unsigned long arg)
6869{
6870 switch (cmd) {
6871 case HOT_REMOVE_DISK:
6872 case HOT_ADD_DISK:
6873 case SET_DISK_FAULTY:
6874 case SET_BITMAP_FILE:
6875 /* These take in integer arg, do not convert */
6876 break;
6877 default:
6878 arg = (unsigned long)compat_ptr(arg);
6879 break;
6880 }
6881
6882 return md_ioctl(bdev, mode, cmd, arg);
6883}
6884#endif /* CONFIG_COMPAT */
1da177e4 6885
a39907fa 6886static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6887{
6888 /*
6889 * Succeed if we can lock the mddev, which confirms that
6890 * it isn't being stopped right now.
6891 */
fd01b88c 6892 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6893 int err;
6894
0c098220
YL		 6895 if (!mddev)
6896 return -ENODEV;
6897
d3374825
N		 6898 if (mddev->gendisk != bdev->bd_disk) {
6899 /* we are racing with mddev_put which is discarding this
6900 * bd_disk.
6901 */
6902 mddev_put(mddev);
6903 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6904 flush_workqueue(md_misc_wq);
d3374825
N		 6905 /* Then retry the open from the top */
6906 return -ERESTARTSYS;
6907 }
6908 BUG_ON(mddev != bdev->bd_disk->private_data);
6909
c8c00a69 6910 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6911 goto out;
6912
6913 err = 0;
f2ea68cf 6914 atomic_inc(&mddev->openers);
260fa034 6915 clear_bit(MD_STILL_CLOSED, &mddev->flags);
c8c00a69 6916 mutex_unlock(&mddev->open_mutex);
1da177e4 6917
f0b4f7e2 6918 check_disk_change(bdev);
1da177e4
LT		 6919 out:
6920 return err;
6921}
6922
db2a144b 6923static void md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6924{
f72ffdd6 6925 struct mddev *mddev = disk->private_data;
1da177e4 6926
52e5f9d1 6927 BUG_ON(!mddev);
f2ea68cf 6928 atomic_dec(&mddev->openers);
1da177e4 6929 mddev_put(mddev);
1da177e4 6930}
f0b4f7e2
N		 6931
6932static int md_media_changed(struct gendisk *disk)
6933{
fd01b88c 6934 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6935
6936 return mddev->changed;
6937}
6938
6939static int md_revalidate(struct gendisk *disk)
6940{
fd01b88c 6941 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6942
6943 mddev->changed = 0;
6944 return 0;
6945}
83d5cde4 6946static const struct block_device_operations md_fops =
1da177e4
LT		 6947{
6948 .owner = THIS_MODULE,
a39907fa
AV		 6949 .open = md_open,
6950 .release = md_release,
b492b852 6951 .ioctl = md_ioctl,
aa98aa31
AB		 6952#ifdef CONFIG_COMPAT
6953 .compat_ioctl = md_compat_ioctl,
6954#endif
a885c8c4 6955 .getgeo = md_getgeo,
f0b4f7e2
N		 6956 .media_changed = md_media_changed,
6957 .revalidate_disk= md_revalidate,
1da177e4
LT		 6958};
6959
f72ffdd6 6960static int md_thread(void *arg)
1da177e4 6961{
2b8bf345 6962 struct md_thread *thread = arg;
1da177e4 6963
1da177e4
LT		 6964 /*
6965 * md_thread is a 'system-thread', it's priority should be very
6966 * high. We avoid resource deadlocks individually in each
6967 * raid personality. (RAID5 does preallocation) We also use RR and
6968 * the very same RT priority as kswapd, thus we will never get
6969 * into a priority inversion deadlock.
6970 *
6971 * we definitely have to have equal or higher priority than
6972 * bdflush, otherwise bdflush will deadlock if there are too
6973 * many dirty RAID5 blocks.
6974 */
1da177e4 6975
6985c43f 6976 allow_signal(SIGKILL);
a6fb0934 6977 while (!kthread_should_stop()) {
1da177e4 6978
93588e22
N		 6979 /* We need to wait INTERRUPTIBLE so that
6980 * we don't add to the load-average.
6981 * That means we need to be sure no signals are
6982 * pending
6983 */
6984 if (signal_pending(current))
6985 flush_signals(current);
6986
6987 wait_event_interruptible_timeout
6988 (thread->wqueue,
6989 test_bit(THREAD_WAKEUP, &thread->flags)
6990 || kthread_should_stop(),
6991 thread->timeout);
1da177e4 6992
6c987910
N		 6993 clear_bit(THREAD_WAKEUP, &thread->flags);
6994 if (!kthread_should_stop())
4ed8731d 6995 thread->run(thread);
1da177e4 6996 }
a6fb0934 6997
1da177e4
LT		 6998 return 0;
6999}
7000
2b8bf345 7001void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 7002{
7003 if (thread) {
36a4e1fe 7004 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 7005 set_bit(THREAD_WAKEUP, &thread->flags);
7006 wake_up(&thread->wqueue);
7007 }
7008}
6c144d31 7009EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 7010
4ed8731d
SL		 7011struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7012 struct mddev *mddev, const char *name)
1da177e4 7013{
2b8bf345 7014 struct md_thread *thread;
1da177e4 7015
2b8bf345 7016 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 7017 if (!thread)
7018 return NULL;
7019
1da177e4
LT		 7020 init_waitqueue_head(&thread->wqueue);
7021
1da177e4
LT		 7022 thread->run = run;
7023 thread->mddev = mddev;
32a7627c 7024 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 7025 thread->tsk = kthread_run(md_thread, thread,
7026 "%s_%s",
7027 mdname(thread->mddev),
0232605d 7028 name);
a6fb0934 7029 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 7030 kfree(thread);
7031 return NULL;
7032 }
1da177e4
LT		 7033 return thread;
7034}
6c144d31 7035EXPORT_SYMBOL(md_register_thread);
1da177e4 7036
2b8bf345 7037void md_unregister_thread(struct md_thread **threadp)
1da177e4 7038{
2b8bf345 7039 struct md_thread *thread = *threadp;
e0cf8f04
N		 7040 if (!thread)
7041 return;
36a4e1fe 7042 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 7043 /* Locking ensures that mddev_unlock does not wake_up a
7044 * non-existent thread
7045 */
7046 spin_lock(&pers_lock);
7047 *threadp = NULL;
7048 spin_unlock(&pers_lock);
a6fb0934
N		 7049
7050 kthread_stop(thread->tsk);
1da177e4
LT		 7051 kfree(thread);
7052}
6c144d31 7053EXPORT_SYMBOL(md_unregister_thread);
1da177e4 7054
fd01b88c 7055void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7056{
b2d444d7 7057 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 7058 return;
6bfe0b49 7059
de393cde 7060 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 7061 return;
7062 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 7063 if (mddev->degraded)
7064 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 7065 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 7066 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7067 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7068 md_wakeup_thread(mddev->thread);
768a418d 7069 if (mddev->event_work.func)
e804ac78 7070 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 7071 md_new_event_inintr(mddev);
1da177e4 7072}
6c144d31 7073EXPORT_SYMBOL(md_error);
1da177e4
LT		 7074
7075/* seq_file implementation /proc/mdstat */
7076
7077static void status_unused(struct seq_file *seq)
7078{
7079 int i = 0;
3cb03002 7080 struct md_rdev *rdev;
1da177e4
LT		 7081
7082 seq_printf(seq, "unused devices: ");
7083
159ec1fc 7084 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 7085 char b[BDEVNAME_SIZE];
7086 i++;
7087 seq_printf(seq, "%s ",
7088 bdevname(rdev->bdev,b));
7089 }
7090 if (!i)
7091 seq_printf(seq, "<none>");
7092
7093 seq_printf(seq, "\n");
7094}
7095
f72ffdd6 7096static void status_resync(struct seq_file *seq, struct mddev *mddev)
1da177e4 7097{
dd71cf6b
N		 7098 sector_t max_sectors, resync, res;
7099 unsigned long dt, db;
7100 sector_t rt;
4588b42e
N		 7101 int scale;
7102 unsigned int per_milli;
1da177e4 7103
72f36d59
N		 7104 if (mddev->curr_resync <= 3)
7105 resync = 0;
7106 else
7107 resync = mddev->curr_resync
7108 - atomic_read(&mddev->recovery_active);
1da177e4 7109
c804cdec
N		 7110 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7111 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 7112 max_sectors = mddev->resync_max_sectors;
1da177e4 7113 else
dd71cf6b 7114 max_sectors = mddev->dev_sectors;
1da177e4 7115
403df478 7116 WARN_ON(max_sectors == 0);
4588b42e 7117 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 7118 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 7119 * u32, as those are the requirements for sector_div.
7120 * Thus 'scale' must be at least 10
7121 */
7122 scale = 10;
7123 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 7124 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 7125 scale++;
7126 }
7127 res = (resync>>scale)*1000;
dd71cf6b 7128 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 7129
7130 per_milli = res;
1da177e4 7131 {
4588b42e 7132 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 7133 seq_printf(seq, "[");
7134 for (i = 0; i < x; i++)
7135 seq_printf(seq, "=");
7136 seq_printf(seq, ">");
7137 for (i = 0; i < y; i++)
7138 seq_printf(seq, ".");
7139 seq_printf(seq, "] ");
7140 }
4588b42e 7141 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 7142 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7143 "reshape" :
61df9d91
N		 7144 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7145 "check" :
7146 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7147 "resync" : "recovery"))),
7148 per_milli/10, per_milli % 10,
dd71cf6b
N		 7149 (unsigned long long) resync/2,
7150 (unsigned long long) max_sectors/2);
1da177e4
LT		 7151
7152 /*
1da177e4
LT		 7153 * dt: time from mark until now
7154 * db: blocks written from mark until now
7155 * rt: remaining time
dd71cf6b
N		 7156 *
7157 * rt is a sector_t, so could be 32bit or 64bit.
7158 * So we divide before multiply in case it is 32bit and close
7159 * to the limit.
25985edc 7160 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 7161 * near the end of resync when the number of remaining sectors
7162 * is close to 'db'.
7163 * We then divide rt by 32 after multiplying by db to compensate.
7164 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 7165 */
7166 dt = ((jiffies - mddev->resync_mark) / HZ);
7167 if (!dt) dt++;
ff4e8d9a
N		 7168 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7169 - mddev->resync_mark_cnt;
1da177e4 7170
dd71cf6b
N		 7171 rt = max_sectors - resync; /* number of remaining sectors */
7172 sector_div(rt, db/32+1);
7173 rt *= dt;
7174 rt >>= 5;
7175
7176 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7177 ((unsigned long)rt % 60)/6);
1da177e4 7178
ff4e8d9a 7179 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 7180}
7181
7182static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7183{
7184 struct list_head *tmp;
7185 loff_t l = *pos;
fd01b88c 7186 struct mddev *mddev;
1da177e4
LT		 7187
7188 if (l >= 0x10000)
7189 return NULL;
7190 if (!l--)
7191 /* header */
7192 return (void*)1;
7193
7194 spin_lock(&all_mddevs_lock);
7195 list_for_each(tmp,&all_mddevs)
7196 if (!l--) {
fd01b88c 7197 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 7198 mddev_get(mddev);
7199 spin_unlock(&all_mddevs_lock);
7200 return mddev;
7201 }
7202 spin_unlock(&all_mddevs_lock);
7203 if (!l--)
7204 return (void*)2;/* tail */
7205 return NULL;
7206}
7207
7208static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7209{
7210 struct list_head *tmp;
fd01b88c 7211 struct mddev *next_mddev, *mddev = v;
f72ffdd6 7212
1da177e4
LT		 7213 ++*pos;
7214 if (v == (void*)2)
7215 return NULL;
7216
7217 spin_lock(&all_mddevs_lock);
7218 if (v == (void*)1)
7219 tmp = all_mddevs.next;
7220 else
7221 tmp = mddev->all_mddevs.next;
7222 if (tmp != &all_mddevs)
fd01b88c 7223 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 7224 else {
7225 next_mddev = (void*)2;
7226 *pos = 0x10000;
f72ffdd6 7227 }
1da177e4
LT		 7228 spin_unlock(&all_mddevs_lock);
7229
7230 if (v != (void*)1)
7231 mddev_put(mddev);
7232 return next_mddev;
7233
7234}
7235
7236static void md_seq_stop(struct seq_file *seq, void *v)
7237{
fd01b88c 7238 struct mddev *mddev = v;
1da177e4
LT		 7239
7240 if (mddev && v != (void*)1 && v != (void*)2)
7241 mddev_put(mddev);
7242}
7243
7244static int md_seq_show(struct seq_file *seq, void *v)
7245{
fd01b88c 7246 struct mddev *mddev = v;
dd8ac336 7247 sector_t sectors;
3cb03002 7248 struct md_rdev *rdev;
1da177e4
LT		 7249
7250 if (v == (void*)1) {
84fc4b56 7251 struct md_personality *pers;
1da177e4
LT		 7252 seq_printf(seq, "Personalities : ");
7253 spin_lock(&pers_lock);
2604b703
N		 7254 list_for_each_entry(pers, &pers_list, list)
7255 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 7256
7257 spin_unlock(&pers_lock);
7258 seq_printf(seq, "\n");
f1514638 7259 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7260 return 0;
7261 }
7262 if (v == (void*)2) {
7263 status_unused(seq);
7264 return 0;
7265 }
7266
36d091f4 7267 spin_lock(&mddev->lock);
1da177e4
LT		 7268 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7269 seq_printf(seq, "%s : %sactive", mdname(mddev),
7270 mddev->pers ? "" : "in");
7271 if (mddev->pers) {
f91de92e 7272 if (mddev->ro==1)
1da177e4 7273 seq_printf(seq, " (read-only)");
f91de92e 7274 if (mddev->ro==2)
52720ae7 7275 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 7276 seq_printf(seq, " %s", mddev->pers->name);
7277 }
7278
dd8ac336 7279 sectors = 0;
f97fcad3
N		 7280 rcu_read_lock();
7281 rdev_for_each_rcu(rdev, mddev) {
1da177e4
LT		 7282 char b[BDEVNAME_SIZE];
7283 seq_printf(seq, " %s[%d]",
7284 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 7285 if (test_bit(WriteMostly, &rdev->flags))
7286 seq_printf(seq, "(W)");
b2d444d7 7287 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7288 seq_printf(seq, "(F)");
7289 continue;
2d78f8c4
N		 7290 }
7291 if (rdev->raid_disk < 0)
b325a32e 7292 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 7293 if (test_bit(Replacement, &rdev->flags))
7294 seq_printf(seq, "(R)");
dd8ac336 7295 sectors += rdev->sectors;
1da177e4 7296 }
f97fcad3 7297 rcu_read_unlock();
1da177e4
LT		 7298
7299 if (!list_empty(&mddev->disks)) {
7300 if (mddev->pers)
7301 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 7302 (unsigned long long)
7303 mddev->array_sectors / 2);
1da177e4
LT		 7304 else
7305 seq_printf(seq, "\n %llu blocks",
dd8ac336 7306 (unsigned long long)sectors / 2);
1da177e4 7307 }
1cd6bf19
N		 7308 if (mddev->persistent) {
7309 if (mddev->major_version != 0 ||
7310 mddev->minor_version != 90) {
7311 seq_printf(seq," super %d.%d",
7312 mddev->major_version,
7313 mddev->minor_version);
7314 }
e691063a
N		 7315 } else if (mddev->external)
7316 seq_printf(seq, " super external:%s",
7317 mddev->metadata_type);
7318 else
1cd6bf19 7319 seq_printf(seq, " super non-persistent");
1da177e4
LT		 7320
7321 if (mddev->pers) {
d710e138 7322 mddev->pers->status(seq, mddev);
f72ffdd6 7323 seq_printf(seq, "\n ");
8e1b39d6
N		 7324 if (mddev->pers->sync_request) {
7325 if (mddev->curr_resync > 2) {
d710e138 7326 status_resync(seq, mddev);
8e1b39d6 7327 seq_printf(seq, "\n ");
72f36d59 7328 } else if (mddev->curr_resync >= 1)
8e1b39d6
N		 7329 seq_printf(seq, "\tresync=DELAYED\n ");
7330 else if (mddev->recovery_cp < MaxSector)
7331 seq_printf(seq, "\tresync=PENDING\n ");
7332 }
32a7627c
N		 7333 } else
7334 seq_printf(seq, "\n ");
7335
57148964 7336 bitmap_status(seq, mddev->bitmap);
1da177e4
LT		 7337
7338 seq_printf(seq, "\n");
7339 }
36d091f4 7340 spin_unlock(&mddev->lock);
f72ffdd6 7341
1da177e4
LT		 7342 return 0;
7343}
7344
110518bc 7345static const struct seq_operations md_seq_ops = {
1da177e4
LT		 7346 .start = md_seq_start,
7347 .next = md_seq_next,
7348 .stop = md_seq_stop,
7349 .show = md_seq_show,
7350};
7351
7352static int md_seq_open(struct inode *inode, struct file *file)
7353{
f1514638 7354 struct seq_file *seq;
1da177e4
LT		 7355 int error;
7356
7357 error = seq_open(file, &md_seq_ops);
d7603b7e 7358 if (error)
f1514638
KS		 7359 return error;
7360
7361 seq = file->private_data;
7362 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7363 return error;
7364}
7365
e2f23b60 7366static int md_unloading;
d7603b7e
N		 7367static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7368{
f1514638 7369 struct seq_file *seq = filp->private_data;
d7603b7e
N		 7370 int mask;
7371
e2f23b60 7372 if (md_unloading)
7d7e64f2 7373 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
d7603b7e
N		 7374 poll_wait(filp, &md_event_waiters, wait);
7375
7376 /* always allow read */
7377 mask = POLLIN | POLLRDNORM;
7378
f1514638 7379 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 7380 mask |= POLLERR | POLLPRI;
7381 return mask;
7382}
7383
fa027c2a 7384static const struct file_operations md_seq_fops = {
e24650c2 7385 .owner = THIS_MODULE,
1da177e4
LT		 7386 .open = md_seq_open,
7387 .read = seq_read,
7388 .llseek = seq_lseek,
c3f94b40 7389 .release = seq_release_private,
d7603b7e 7390 .poll = mdstat_poll,
1da177e4
LT		 7391};
7392
84fc4b56 7393int register_md_personality(struct md_personality *p)
1da177e4 7394{
50bd3774
CY		 7395 printk(KERN_INFO "md: %s personality registered for level %d\n",
7396 p->name, p->level);
1da177e4 7397 spin_lock(&pers_lock);
2604b703 7398 list_add_tail(&p->list, &pers_list);
1da177e4
LT		 7399 spin_unlock(&pers_lock);
7400 return 0;
7401}
6c144d31 7402EXPORT_SYMBOL(register_md_personality);
1da177e4 7403
84fc4b56 7404int unregister_md_personality(struct md_personality *p)
1da177e4 7405{
2604b703 7406 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7407 spin_lock(&pers_lock);
2604b703 7408 list_del_init(&p->list);
1da177e4
LT		 7409 spin_unlock(&pers_lock);
7410 return 0;
7411}
6c144d31 7412EXPORT_SYMBOL(unregister_md_personality);
1da177e4 7413
edb39c9d
GR		 7414int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module)
7415{
7416 if (md_cluster_ops != NULL)
7417 return -EALREADY;
7418 spin_lock(&pers_lock);
7419 md_cluster_ops = ops;
7420 md_cluster_mod = module;
7421 spin_unlock(&pers_lock);
7422 return 0;
7423}
7424EXPORT_SYMBOL(register_md_cluster_operations);
7425
7426int unregister_md_cluster_operations(void)
7427{
7428 spin_lock(&pers_lock);
7429 md_cluster_ops = NULL;
7430 spin_unlock(&pers_lock);
7431 return 0;
7432}
7433EXPORT_SYMBOL(unregister_md_cluster_operations);
7434
7435int md_setup_cluster(struct mddev *mddev, int nodes)
7436{
7437 int err;
7438
7439 err = request_module("md-cluster");
7440 if (err) {
7441 pr_err("md-cluster module not found.\n");
7442 return err;
7443 }
7444
7445 spin_lock(&pers_lock);
7446 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7447 spin_unlock(&pers_lock);
7448 return -ENOENT;
7449 }
7450 spin_unlock(&pers_lock);
7451
cf921cc1 7452 return md_cluster_ops->join(mddev, nodes);
edb39c9d
GR		 7453}
7454
7455void md_cluster_stop(struct mddev *mddev)
7456{
c4ce867f
GR		 7457 if (!md_cluster_ops)
7458 return;
edb39c9d
GR		 7459 md_cluster_ops->leave(mddev);
7460 module_put(md_cluster_mod);
7461}
7462
fd01b88c 7463static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7464{
f72ffdd6 7465 struct md_rdev *rdev;
1da177e4 7466 int idle;
eea1bf38 7467 int curr_events;
1da177e4
LT		 7468
7469 idle = 1;
4b80991c
N		 7470 rcu_read_lock();
7471 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7472 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 7473 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7474 (int)part_stat_read(&disk->part0, sectors[1]) -
7475 atomic_read(&disk->sync_io);
713f6ab1
N		 7476 /* sync IO will cause sync_io to increase before the disk_stats
7477 * as sync_io is counted when a request starts, and
7478 * disk_stats is counted when it completes.
7479 * So resync activity will cause curr_events to be smaller than
7480 * when there was no such activity.
7481 * non-sync IO will cause disk_stat to increase without
7482 * increasing sync_io so curr_events will (eventually)
7483 * be larger than it was before. Once it becomes
7484 * substantially larger, the test below will cause
7485 * the array to appear non-idle, and resync will slow
7486 * down.
7487 * If there is a lot of outstanding resync activity when
7488 * we set last_event to curr_events, then all that activity
7489 * completing might cause the array to appear non-idle
7490 * and resync will be slowed down even though there might
7491 * not have been non-resync activity. This will only
7492 * happen once though. 'last_events' will soon reflect
7493 * the state where there is little or no outstanding
7494 * resync requests, and further resync activity will
7495 * always make curr_events less than last_events.
c0e48521 7496 *
1da177e4 7497 */
eea1bf38 7498 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 7499 rdev->last_events = curr_events;
7500 idle = 0;
7501 }
7502 }
4b80991c 7503 rcu_read_unlock();
1da177e4
LT		 7504 return idle;
7505}
7506
fd01b88c 7507void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 7508{
7509 /* another "blocks" (512byte) blocks have been synced */
7510 atomic_sub(blocks, &mddev->recovery_active);
7511 wake_up(&mddev->recovery_wait);
7512 if (!ok) {
dfc70645 7513 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 7514 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT		 7515 md_wakeup_thread(mddev->thread);
7516 // stop recovery, signal do_sync ....
7517 }
7518}
6c144d31 7519EXPORT_SYMBOL(md_done_sync);
1da177e4 7520
06d91a5f
N		 7521/* md_write_start(mddev, bi)
7522 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 7523 * in superblock) before writing, schedule a superblock update
7524 * and wait for it to complete.
06d91a5f 7525 */
fd01b88c 7526void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7527{
0fd62b86 7528 int did_change = 0;
06d91a5f 7529 if (bio_data_dir(bi) != WRITE)
3d310eb7 7530 return;
06d91a5f 7531
f91de92e
N		 7532 BUG_ON(mddev->ro == 1);
7533 if (mddev->ro == 2) {
7534 /* need to switch to read/write */
7535 mddev->ro = 0;
7536 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7537 md_wakeup_thread(mddev->thread);
25156198 7538 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7539 did_change = 1;
f91de92e 7540 }
06d91a5f 7541 atomic_inc(&mddev->writes_pending);
31a59e34
N		 7542 if (mddev->safemode == 1)
7543 mddev->safemode = 0;
06d91a5f 7544 if (mddev->in_sync) {
85572d7c 7545 spin_lock(&mddev->lock);
3d310eb7
N		 7546 if (mddev->in_sync) {
7547 mddev->in_sync = 0;
850b2b42 7548 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7549 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7550 md_wakeup_thread(mddev->thread);
0fd62b86 7551 did_change = 1;
3d310eb7 7552 }
85572d7c 7553 spin_unlock(&mddev->lock);
06d91a5f 7554 }
0fd62b86 7555 if (did_change)
00bcb4ac 7556 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7557 wait_event(mddev->sb_wait,
09a44cc1 7558 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4 7559}
6c144d31 7560EXPORT_SYMBOL(md_write_start);
1da177e4 7561
fd01b88c 7562void md_write_end(struct mddev *mddev)
1da177e4
LT		 7563{
7564 if (atomic_dec_and_test(&mddev->writes_pending)) {
7565 if (mddev->safemode == 2)
7566 md_wakeup_thread(mddev->thread);
16f17b39 7567 else if (mddev->safemode_delay)
1da177e4
LT		 7568 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7569 }
7570}
6c144d31 7571EXPORT_SYMBOL(md_write_end);
1da177e4 7572
2a2275d6
N		 7573/* md_allow_write(mddev)
7574 * Calling this ensures that the array is marked 'active' so that writes
7575 * may proceed without blocking. It is important to call this before
7576 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7577 * Must be called with mddev_lock held.
b5470dc5
DW		 7578 *
7579 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7580 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7581 */
fd01b88c 7582int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7583{
7584 if (!mddev->pers)
b5470dc5 7585 return 0;
2a2275d6 7586 if (mddev->ro)
b5470dc5 7587 return 0;
1a0fd497 7588 if (!mddev->pers->sync_request)
b5470dc5 7589 return 0;
2a2275d6 7590
85572d7c 7591 spin_lock(&mddev->lock);
2a2275d6
N		 7592 if (mddev->in_sync) {
7593 mddev->in_sync = 0;
7594 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7595 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7596 if (mddev->safemode_delay &&
7597 mddev->safemode == 0)
7598 mddev->safemode = 1;
85572d7c 7599 spin_unlock(&mddev->lock);
293467aa
GR		 7600 if (mddev_is_clustered(mddev))
7601 md_cluster_ops->metadata_update_start(mddev);
2a2275d6 7602 md_update_sb(mddev, 0);
293467aa
GR		 7603 if (mddev_is_clustered(mddev))
7604 md_cluster_ops->metadata_update_finish(mddev);
00bcb4ac 7605 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6 7606 } else
85572d7c 7607 spin_unlock(&mddev->lock);
b5470dc5 7608
070dc6dd 7609 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7610 return -EAGAIN;
7611 else
7612 return 0;
2a2275d6
N		 7613}
7614EXPORT_SYMBOL_GPL(md_allow_write);
7615
1da177e4
LT		 7616#define SYNC_MARKS 10
7617#define SYNC_MARK_STEP (3*HZ)
54f89341 7618#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 7619void md_do_sync(struct md_thread *thread)
1da177e4 7620{
4ed8731d 7621 struct mddev *mddev = thread->mddev;
fd01b88c 7622 struct mddev *mddev2;
1da177e4
LT		 7623 unsigned int currspeed = 0,
7624 window;
ac7e50a3 7625 sector_t max_sectors,j, io_sectors, recovery_done;
1da177e4 7626 unsigned long mark[SYNC_MARKS];
54f89341 7627 unsigned long update_time;
1da177e4
LT		 7628 sector_t mark_cnt[SYNC_MARKS];
7629 int last_mark,m;
7630 struct list_head *tmp;
7631 sector_t last_check;
57afd89f 7632 int skipped = 0;
3cb03002 7633 struct md_rdev *rdev;
c4a39551 7634 char *desc, *action = NULL;
7c2c57c9 7635 struct blk_plug plug;
1da177e4
LT		 7636
7637 /* just incase thread restarts... */
7638 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7639 return;
3991b31e
N		 7640 if (mddev->ro) {/* never try to sync a read-only array */
7641 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5fd6c1dc 7642 return;
3991b31e 7643 }
1da177e4 7644
61df9d91 7645 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
c4a39551 7646 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
61df9d91 7647 desc = "data-check";
c4a39551
JB		 7648 action = "check";
7649 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
61df9d91 7650 desc = "requested-resync";
c4a39551
JB		 7651 action = "repair";
7652 } else
61df9d91
N		 7653 desc = "resync";
7654 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7655 desc = "reshape";
7656 else
7657 desc = "recovery";
7658
c4a39551
JB		 7659 mddev->last_sync_action = action ?: desc;
7660
1da177e4
LT		 7661 /* we overload curr_resync somewhat here.
7662 * 0 == not engaged in resync at all
7663 * 2 == checking that there is no conflict with another sync
7664 * 1 == like 2, but have yielded to allow conflicting resync to
7665 * commense
7666 * other == active in resync - this many blocks
7667 *
7668 * Before starting a resync we must have set curr_resync to
7669 * 2, and then checked that every "conflicting" array has curr_resync
7670 * less than ours. When we find one that is the same or higher
7671 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7672 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7673 * This will mean we have to start checking from the beginning again.
7674 *
7675 */
7676
7677 do {
7678 mddev->curr_resync = 2;
7679
7680 try_again:
404e4b43 7681 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7682 goto skip;
29ac4aa3 7683 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7684 if (mddev2 == mddev)
7685 continue;
90b08710
BS		 7686 if (!mddev->parallel_resync
7687 && mddev2->curr_resync
7688 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7689 DEFINE_WAIT(wq);
7690 if (mddev < mddev2 && mddev->curr_resync == 2) {
7691 /* arbitrarily yield */
7692 mddev->curr_resync = 1;
7693 wake_up(&resync_wait);
7694 }
7695 if (mddev > mddev2 && mddev->curr_resync == 1)
7696 /* no need to wait here, we can wait the next
7697 * time 'round when curr_resync == 2
7698 */
7699 continue;
9744197c
N		 7700 /* We need to wait 'interruptible' so as not to
7701 * contribute to the load average, and not to
7702 * be caught by 'softlockup'
7703 */
7704 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
c91abf5a 7705 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8712e553 7706 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7707 printk(KERN_INFO "md: delaying %s of %s"
7708 " until %s has finished (they"
1da177e4 7709 " share one or more physical units)\n",
61df9d91 7710 desc, mdname(mddev), mdname(mddev2));
1da177e4 7711 mddev_put(mddev2);
9744197c
N		 7712 if (signal_pending(current))
7713 flush_signals(current);
1da177e4
LT		 7714 schedule();
7715 finish_wait(&resync_wait, &wq);
7716 goto try_again;
7717 }
7718 finish_wait(&resync_wait, &wq);
7719 }
7720 }
7721 } while (mddev->curr_resync < 2);
7722
5fd6c1dc 7723 j = 0;
9d88883e 7724 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7725 /* resync follows the size requested by the personality,
57afd89f 7726 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7727 */
7728 max_sectors = mddev->resync_max_sectors;
7f7583d4 7729 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 7730 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7731 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7732 j = mddev->resync_min;
7733 else if (!mddev->bitmap)
5fd6c1dc 7734 j = mddev->recovery_cp;
5e96ee65 7735
ccfcc3c1 7736 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7737 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7738 else {
1da177e4 7739 /* recovery follows the physical size of devices */
58c0fed4 7740 max_sectors = mddev->dev_sectors;
5fd6c1dc 7741 j = MaxSector;
4e59ca7d 7742 rcu_read_lock();
dafb20fa 7743 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc
N		 7744 if (rdev->raid_disk >= 0 &&
7745 !test_bit(Faulty, &rdev->flags) &&
7746 !test_bit(In_sync, &rdev->flags) &&
7747 rdev->recovery_offset < j)
7748 j = rdev->recovery_offset;
4e59ca7d 7749 rcu_read_unlock();
133d4527
N		 7750
7751 /* If there is a bitmap, we need to make sure all
7752 * writes that started before we added a spare
7753 * complete before we start doing a recovery.
7754 * Otherwise the write might complete and (via
7755 * bitmap_endwrite) set a bit in the bitmap after the
7756 * recovery has checked that bit and skipped that
7757 * region.
7758 */
7759 if (mddev->bitmap) {
7760 mddev->pers->quiesce(mddev, 1);
7761 mddev->pers->quiesce(mddev, 0);
7762 }
5fd6c1dc 7763 }
1da177e4 7764
61df9d91
N		 7765 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7766 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7767 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7768 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7769 "(but not more than %d KB/sec) for %s.\n",
7770 speed_max(mddev), desc);
1da177e4 7771
eea1bf38 7772 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7773
57afd89f 7774 io_sectors = 0;
1da177e4
LT		 7775 for (m = 0; m < SYNC_MARKS; m++) {
7776 mark[m] = jiffies;
57afd89f 7777 mark_cnt[m] = io_sectors;
1da177e4
LT		 7778 }
7779 last_mark = 0;
7780 mddev->resync_mark = mark[last_mark];
7781 mddev->resync_mark_cnt = mark_cnt[last_mark];
7782
7783 /*
7784 * Tune reconstruction:
7785 */
7786 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7787 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7788 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7789
7790 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7791 last_check = 0;
7792
7793 if (j>2) {
c91abf5a 7794 printk(KERN_INFO
61df9d91
N		 7795 "md: resuming %s of %s from checkpoint.\n",
7796 desc, mdname(mddev));
1da177e4 7797 mddev->curr_resync = j;
72f36d59
N		 7798 } else
7799 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 7800 mddev->curr_resync_completed = j;
72f36d59
N		 7801 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7802 md_new_event(mddev);
54f89341 7803 update_time = jiffies;
1da177e4 7804
96ae923a 7805 if (mddev_is_clustered(mddev))
965400eb 7806 md_cluster_ops->resync_start(mddev, j, max_sectors);
96ae923a 7807
7c2c57c9 7808 blk_start_plug(&plug);
1da177e4 7809 while (j < max_sectors) {
57afd89f 7810 sector_t sectors;
1da177e4 7811
57afd89f 7812 skipped = 0;
97e4f42d 7813
7a91ee1f
N		 7814 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7815 ((mddev->curr_resync > mddev->curr_resync_completed &&
7816 (mddev->curr_resync - mddev->curr_resync_completed)
7817 > (max_sectors >> 4)) ||
54f89341 7818 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f
N		 7819 (j - mddev->curr_resync_completed)*2
7820 >= mddev->resync_max - mddev->curr_resync_completed
7821 )) {
97e4f42d 7822 /* time to update curr_resync_completed */
97e4f42d
N		 7823 wait_event(mddev->recovery_wait,
7824 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7825 mddev->curr_resync_completed = j;
35d78c66 7826 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7827 j > mddev->recovery_cp)
7828 mddev->recovery_cp = j;
54f89341 7829 update_time = jiffies;
070dc6dd 7830 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7831 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7832 }
acb180b0 7833
c91abf5a
N		 7834 while (j >= mddev->resync_max &&
7835 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
e62e58a5
N		 7836 /* As this condition is controlled by user-space,
7837 * we can block indefinitely, so use '_interruptible'
7838 * to avoid triggering warnings.
7839 */
7840 flush_signals(current); /* just in case */
7841 wait_event_interruptible(mddev->recovery_wait,
7842 mddev->resync_max > j
c91abf5a
N		 7843 || test_bit(MD_RECOVERY_INTR,
7844 &mddev->recovery));
e62e58a5 7845 }
acb180b0 7846
c91abf5a
N		 7847 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7848 break;
acb180b0 7849
09314799 7850 sectors = mddev->pers->sync_request(mddev, j, &skipped);
57afd89f 7851 if (sectors == 0) {
dfc70645 7852 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
c91abf5a 7853 break;
1da177e4 7854 }
57afd89f
N		 7855
7856 if (!skipped) { /* actual IO requested */
7857 io_sectors += sectors;
7858 atomic_add(sectors, &mddev->recovery_active);
7859 }
7860
e875ecea
N		 7861 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7862 break;
7863
1da177e4 7864 j += sectors;
72f36d59
N		 7865 if (j > 2)
7866 mddev->curr_resync = j;
96ae923a
GR		 7867 if (mddev_is_clustered(mddev))
7868 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
ff4e8d9a 7869 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7870 if (last_check == 0)
e875ecea 7871 /* this is the earliest that rebuild will be
d7603b7e
N		 7872 * visible in /proc/mdstat
7873 */
7874 md_new_event(mddev);
57afd89f
N		 7875
7876 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7877 continue;
7878
57afd89f 7879 last_check = io_sectors;
1da177e4
LT		 7880 repeat:
7881 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7882 /* step marks */
7883 int next = (last_mark+1) % SYNC_MARKS;
7884
7885 mddev->resync_mark = mark[next];
7886 mddev->resync_mark_cnt = mark_cnt[next];
7887 mark[next] = jiffies;
57afd89f 7888 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7889 last_mark = next;
7890 }
7891
c91abf5a
N		 7892 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7893 break;
1da177e4
LT		 7894
7895 /*
7896 * this loop exits only if either when we are slower than
7897 * the 'hard' speed limit, or the system was IO-idle for
7898 * a jiffy.
7899 * the system might be non-idle CPU-wise, but we only care
7900 * about not overloading the IO subsystem. (things like an
7901 * e2fsck being done on the RAID array should execute fast)
7902 */
1da177e4
LT		 7903 cond_resched();
7904
ac7e50a3
XN		 7905 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7906 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
57afd89f 7907 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7908
88202a0c 7909 if (currspeed > speed_min(mddev)) {
ac8fa419 7910 if (currspeed > speed_max(mddev)) {
c0e48521 7911 msleep(500);
1da177e4
LT		 7912 goto repeat;
7913 }
ac8fa419
N		 7914 if (!is_mddev_idle(mddev, 0)) {
7915 /*
7916 * Give other IO more of a chance.
7917 * The faster the devices, the less we wait.
7918 */
7919 wait_event(mddev->recovery_wait,
7920 !atomic_read(&mddev->recovery_active));
7921 }
1da177e4
LT		 7922 }
7923 }
c91abf5a
N		 7924 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7925 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7926 ? "interrupted" : "done");
1da177e4
LT		 7927 /*
7928 * this also signals 'finished resyncing' to md_stop
7929 */
7c2c57c9 7930 blk_finish_plug(&plug);
1da177e4
LT		 7931 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7932
7933 /* tell personality that we are finished */
09314799 7934 mddev->pers->sync_request(mddev, max_sectors, &skipped);
1da177e4 7935
96ae923a 7936 if (mddev_is_clustered(mddev))
965400eb 7937 md_cluster_ops->resync_finish(mddev);
96ae923a 7938
dfc70645 7939 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7940 mddev->curr_resync > 2) {
7941 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7942 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7943 if (mddev->curr_resync >= mddev->recovery_cp) {
7944 printk(KERN_INFO
61df9d91
N		 7945 "md: checkpointing %s of %s.\n",
7946 desc, mdname(mddev));
0a19caab 7947 if (test_bit(MD_RECOVERY_ERROR,
7948 &mddev->recovery))
7949 mddev->recovery_cp =
7950 mddev->curr_resync_completed;
7951 else
7952 mddev->recovery_cp =
7953 mddev->curr_resync;
5fd6c1dc
N		 7954 }
7955 } else
7956 mddev->recovery_cp = MaxSector;
7957 } else {
7958 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7959 mddev->curr_resync = MaxSector;
4e59ca7d 7960 rcu_read_lock();
dafb20fa 7961 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7962 if (rdev->raid_disk >= 0 &&
70fffd0b 7963 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7964 !test_bit(Faulty, &rdev->flags) &&
7965 !test_bit(In_sync, &rdev->flags) &&
7966 rdev->recovery_offset < mddev->curr_resync)
7967 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7968 rcu_read_unlock();
5fd6c1dc 7969 }
1da177e4 7970 }
db91ff55 7971 skip:
17571284 7972 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7973
23da422b 7974 spin_lock(&mddev->lock);
c07b70ad
N		 7975 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7976 /* We completed so min/max setting can be forgotten if used. */
7977 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7978 mddev->resync_min = 0;
7979 mddev->resync_max = MaxSector;
7980 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7981 mddev->resync_min = mddev->curr_resync_completed;
1da177e4 7982 mddev->curr_resync = 0;
23da422b
N		 7983 spin_unlock(&mddev->lock);
7984
1da177e4
LT		 7985 wake_up(&resync_wait);
7986 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7987 md_wakeup_thread(mddev->thread);
c6207277 7988 return;
1da177e4 7989}
29269553 7990EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7991
746d3207
N		 7992static int remove_and_add_spares(struct mddev *mddev,
7993 struct md_rdev *this)
b4c4c7b8 7994{
3cb03002 7995 struct md_rdev *rdev;
b4c4c7b8 7996 int spares = 0;
f2a371c5 7997 int removed = 0;
b4c4c7b8 7998
dafb20fa 7999 rdev_for_each(rdev, mddev)
746d3207
N		 8000 if ((this == NULL || rdev == this) &&
8001 rdev->raid_disk >= 0 &&
6bfe0b49 8002 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 8003 (test_bit(Faulty, &rdev->flags) ||
8004 ! test_bit(In_sync, &rdev->flags)) &&
8005 atomic_read(&rdev->nr_pending)==0) {
8006 if (mddev->pers->hot_remove_disk(
b8321b68 8007 mddev, rdev) == 0) {
36fad858 8008 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 8009 rdev->raid_disk = -1;
f2a371c5 8010 removed++;
b4c4c7b8
N		 8011 }
8012 }
90584fc9
JB		 8013 if (removed && mddev->kobj.sd)
8014 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 8015
746d3207
N		 8016 if (this)
8017 goto no_add;
8018
dafb20fa 8019 rdev_for_each(rdev, mddev) {
7bfec5f3
N		 8020 if (rdev->raid_disk >= 0 &&
8021 !test_bit(In_sync, &rdev->flags) &&
8022 !test_bit(Faulty, &rdev->flags))
8023 spares++;
7ceb17e8
N		 8024 if (rdev->raid_disk >= 0)
8025 continue;
8026 if (test_bit(Faulty, &rdev->flags))
8027 continue;
8028 if (mddev->ro &&
8313b8e5
N		 8029 ! (rdev->saved_raid_disk >= 0 &&
8030 !test_bit(Bitmap_sync, &rdev->flags)))
7ceb17e8
N		 8031 continue;
8032
7eb41885
N		 8033 if (rdev->saved_raid_disk < 0)
8034 rdev->recovery_offset = 0;
7ceb17e8
N		 8035 if (mddev->pers->
8036 hot_add_disk(mddev, rdev) == 0) {
8037 if (sysfs_link_rdev(mddev, rdev))
8038 /* failure here is OK */;
8039 spares++;
8040 md_new_event(mddev);
8041 set_bit(MD_CHANGE_DEVS, &mddev->flags);
dfc70645 8042 }
b4c4c7b8 8043 }
746d3207 8044no_add:
6dafab6b
N		 8045 if (removed)
8046 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 8047 return spares;
8048}
7ebc0be7 8049
ac05f256
N		 8050static void md_start_sync(struct work_struct *ws)
8051{
8052 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8053
8054 mddev->sync_thread = md_register_thread(md_do_sync,
8055 mddev,
8056 "resync");
8057 if (!mddev->sync_thread) {
8058 printk(KERN_ERR "%s: could not start resync"
8059 " thread...\n",
8060 mdname(mddev));
8061 /* leave the spares where they are, it shouldn't hurt */
8062 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8063 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8064 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8065 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8066 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8067 wake_up(&resync_wait);
ac05f256
N		 8068 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8069 &mddev->recovery))
8070 if (mddev->sysfs_action)
8071 sysfs_notify_dirent_safe(mddev->sysfs_action);
8072 } else
8073 md_wakeup_thread(mddev->sync_thread);
8074 sysfs_notify_dirent_safe(mddev->sysfs_action);
8075 md_new_event(mddev);
8076}
8077
1da177e4
LT		 8078/*
8079 * This routine is regularly called by all per-raid-array threads to
8080 * deal with generic issues like resync and super-block update.
8081 * Raid personalities that don't have a thread (linear/raid0) do not
8082 * need this as they never do any recovery or update the superblock.
8083 *
8084 * It does not do any resync itself, but rather "forks" off other threads
8085 * to do that as needed.
8086 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8087 * "->recovery" and create a thread at ->sync_thread.
dfc70645 8088 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 8089 * and wakeups up this thread which will reap the thread and finish up.
8090 * This thread also removes any faulty devices (with nr_pending == 0).
8091 *
8092 * The overall approach is:
8093 * 1/ if the superblock needs updating, update it.
8094 * 2/ If a recovery thread is running, don't do anything else.
8095 * 3/ If recovery has finished, clean up, possibly marking spares active.
8096 * 4/ If there are any faulty devices, remove them.
8097 * 5/ If array is degraded, try to add spares devices
8098 * 6/ If array has spares or is not in-sync, start a resync thread.
8099 */
fd01b88c 8100void md_check_recovery(struct mddev *mddev)
1da177e4 8101{
68866e42
JB		 8102 if (mddev->suspended)
8103 return;
8104
5f40402d 8105 if (mddev->bitmap)
aa5cbd10 8106 bitmap_daemon_work(mddev);
1da177e4 8107
fca4d848 8108 if (signal_pending(current)) {
31a59e34 8109 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 8110 printk(KERN_INFO "md: %s in immediate safe mode\n",
8111 mdname(mddev));
8112 mddev->safemode = 2;
8113 }
8114 flush_signals(current);
8115 }
8116
c89a8eee
N		 8117 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8118 return;
1da177e4 8119 if ( ! (
142d44c3 8120 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 8121 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 8122 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 8123 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 8124 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8125 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 8126 ))
8127 return;
fca4d848 8128
df5b89b3 8129 if (mddev_trylock(mddev)) {
b4c4c7b8 8130 int spares = 0;
fca4d848 8131
c89a8eee 8132 if (mddev->ro) {
ab16bfc7
NB		 8133 struct md_rdev *rdev;
8134 if (!mddev->external && mddev->in_sync)
8135 /* 'Blocked' flag not needed as failed devices
8136 * will be recorded if array switched to read/write.
8137 * Leaving it set will prevent the device
8138 * from being removed.
8139 */
8140 rdev_for_each(rdev, mddev)
8141 clear_bit(Blocked, &rdev->flags);
7ceb17e8
N		 8142 /* On a read-only array we can:
8143 * - remove failed devices
8144 * - add already-in_sync devices if the array itself
8145 * is in-sync.
8146 * As we only add devices that are already in-sync,
8147 * we can activate the spares immediately.
c89a8eee 8148 */
7ceb17e8 8149 remove_and_add_spares(mddev, NULL);
8313b8e5
N		 8150 /* There is no thread, but we need to call
8151 * ->spare_active and clear saved_raid_disk
8152 */
2ac295a5 8153 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8313b8e5
N		 8154 md_reap_sync_thread(mddev);
8155 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
c89a8eee
N		 8156 goto unlock;
8157 }
8158
31a59e34 8159 if (!mddev->external) {
0fd62b86 8160 int did_change = 0;
85572d7c 8161 spin_lock(&mddev->lock);
31a59e34
N		 8162 if (mddev->safemode &&
8163 !atomic_read(&mddev->writes_pending) &&
8164 !mddev->in_sync &&
8165 mddev->recovery_cp == MaxSector) {
8166 mddev->in_sync = 1;
0fd62b86 8167 did_change = 1;
070dc6dd 8168 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 8169 }
8170 if (mddev->safemode == 1)
8171 mddev->safemode = 0;
85572d7c 8172 spin_unlock(&mddev->lock);
0fd62b86 8173 if (did_change)
00bcb4ac 8174 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 8175 }
fca4d848 8176
293467aa
GR		 8177 if (mddev->flags & MD_UPDATE_SB_FLAGS) {
8178 if (mddev_is_clustered(mddev))
8179 md_cluster_ops->metadata_update_start(mddev);
850b2b42 8180 md_update_sb(mddev, 0);
293467aa
GR		 8181 if (mddev_is_clustered(mddev))
8182 md_cluster_ops->metadata_update_finish(mddev);
8183 }
06d91a5f 8184
1da177e4
LT		 8185 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8186 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8187 /* resync/recovery still happening */
8188 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8189 goto unlock;
8190 }
8191 if (mddev->sync_thread) {
a91d5ac0 8192 md_reap_sync_thread(mddev);
1da177e4
LT		 8193 goto unlock;
8194 }
72a23c21
NB		 8195 /* Set RUNNING before clearing NEEDED to avoid
8196 * any transients in the value of "sync_action".
8197 */
72f36d59 8198 mddev->curr_resync_completed = 0;
23da422b 8199 spin_lock(&mddev->lock);
72a23c21 8200 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
23da422b 8201 spin_unlock(&mddev->lock);
24dd469d
N		 8202 /* Clear some bits that don't mean anything, but
8203 * might be left set
8204 */
24dd469d
N		 8205 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8206 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 8207
ed209584
N		 8208 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8209 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
ac05f256 8210 goto not_running;
1da177e4
LT		 8211 /* no recovery is running.
8212 * remove any failed drives, then
8213 * add spares if possible.
72f36d59 8214 * Spares are also removed and re-added, to allow
1da177e4
LT		 8215 * the personality to fail the re-add.
8216 */
1da177e4 8217
b4c4c7b8 8218 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 8219 if (mddev->pers->check_reshape == NULL ||
8220 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8 8221 /* Cannot proceed */
ac05f256 8222 goto not_running;
b4c4c7b8 8223 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 8224 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 8225 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N		 8226 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8227 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 8228 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 8229 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8230 } else if (mddev->recovery_cp < MaxSector) {
8231 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 8232 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8233 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8234 /* nothing to be done ... */
ac05f256 8235 goto not_running;
24dd469d 8236
1da177e4 8237 if (mddev->pers->sync_request) {
ef99bf48 8238 if (spares) {
a654b9d8
N		 8239 /* We are adding a device or devices to an array
8240 * which has the bitmap stored on all devices.
8241 * So make sure all bitmap pages get written
8242 */
8243 bitmap_write_all(mddev->bitmap);
8244 }
ac05f256
N		 8245 INIT_WORK(&mddev->del_work, md_start_sync);
8246 queue_work(md_misc_wq, &mddev->del_work);
8247 goto unlock;
1da177e4 8248 }
ac05f256 8249 not_running:
72a23c21
NB		 8250 if (!mddev->sync_thread) {
8251 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8252 wake_up(&resync_wait);
72a23c21
NB		 8253 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8254 &mddev->recovery))
0c3573f1 8255 if (mddev->sysfs_action)
00bcb4ac 8256 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 8257 }
ac05f256
N		 8258 unlock:
8259 wake_up(&mddev->sb_wait);
1da177e4
LT		 8260 mddev_unlock(mddev);
8261 }
8262}
6c144d31 8263EXPORT_SYMBOL(md_check_recovery);
1da177e4 8264
a91d5ac0
JB		 8265void md_reap_sync_thread(struct mddev *mddev)
8266{
8267 struct md_rdev *rdev;
8268
8269 /* resync has finished, collect result */
8270 md_unregister_thread(&mddev->sync_thread);
8271 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8272 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8273 /* success...*/
8274 /* activate any spares */
8275 if (mddev->pers->spare_active(mddev)) {
8276 sysfs_notify(&mddev->kobj, NULL,
8277 "degraded");
8278 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8279 }
8280 }
293467aa
GR		 8281 if (mddev_is_clustered(mddev))
8282 md_cluster_ops->metadata_update_start(mddev);
a91d5ac0
JB		 8283 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8284 mddev->pers->finish_reshape)
8285 mddev->pers->finish_reshape(mddev);
8286
8287 /* If array is no-longer degraded, then any saved_raid_disk
f466722c 8288 * information must be scrapped.
a91d5ac0 8289 */
f466722c
N		 8290 if (!mddev->degraded)
8291 rdev_for_each(rdev, mddev)
a91d5ac0
JB		 8292 rdev->saved_raid_disk = -1;
8293
8294 md_update_sb(mddev, 1);
293467aa
GR		 8295 if (mddev_is_clustered(mddev))
8296 md_cluster_ops->metadata_update_finish(mddev);
a91d5ac0 8297 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
ea358cd0 8298 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
a91d5ac0
JB		 8299 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8300 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8301 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8302 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
f851b60d 8303 wake_up(&resync_wait);
a91d5ac0
JB		 8304 /* flag recovery needed just to double check */
8305 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8306 sysfs_notify_dirent_safe(mddev->sysfs_action);
8307 md_new_event(mddev);
8308 if (mddev->event_work.func)
8309 queue_work(md_misc_wq, &mddev->event_work);
8310}
6c144d31 8311EXPORT_SYMBOL(md_reap_sync_thread);
a91d5ac0 8312
fd01b88c 8313void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 8314{
00bcb4ac 8315 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 8316 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 8317 !test_bit(Blocked, &rdev->flags) &&
8318 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 8319 msecs_to_jiffies(5000));
8320 rdev_dec_pending(rdev, mddev);
8321}
8322EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8323
c6563a8c
N		 8324void md_finish_reshape(struct mddev *mddev)
8325{
8326 /* called be personality module when reshape completes. */
8327 struct md_rdev *rdev;
8328
8329 rdev_for_each(rdev, mddev) {
8330 if (rdev->data_offset > rdev->new_data_offset)
8331 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8332 else
8333 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8334 rdev->data_offset = rdev->new_data_offset;
8335 }
8336}
8337EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N		 8338
8339/* Bad block management.
8340 * We can record which blocks on each device are 'bad' and so just
8341 * fail those blocks, or that stripe, rather than the whole device.
8342 * Entries in the bad-block table are 64bits wide. This comprises:
8343 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8344 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8345 * A 'shift' can be set so that larger blocks are tracked and
8346 * consequently larger devices can be covered.
8347 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8348 *
8349 * Locking of the bad-block table uses a seqlock so md_is_badblock
8350 * might need to retry if it is very unlucky.
8351 * We will sometimes want to check for bad blocks in a bi_end_io function,
8352 * so we use the write_seqlock_irq variant.
8353 *
8354 * When looking for a bad block we specify a range and want to
8355 * know if any block in the range is bad. So we binary-search
8356 * to the last range that starts at-or-before the given endpoint,
8357 * (or "before the sector after the target range")
8358 * then see if it ends after the given start.
8359 * We return
8360 * 0 if there are no known bad blocks in the range
8361 * 1 if there are known bad block which are all acknowledged
8362 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8363 * plus the start/length of the first bad section we overlap.
8364 */
8365int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8366 sector_t *first_bad, int *bad_sectors)
8367{
8368 int hi;
ab05613a 8369 int lo;
2230dfe4 8370 u64 *p = bb->page;
ab05613a 8371 int rv;
2230dfe4
N		 8372 sector_t target = s + sectors;
8373 unsigned seq;
8374
8375 if (bb->shift > 0) {
8376 /* round the start down, and the end up */
8377 s >>= bb->shift;
8378 target += (1<<bb->shift) - 1;
8379 target >>= bb->shift;
8380 sectors = target - s;
8381 }
8382 /* 'target' is now the first block after the bad range */
8383
8384retry:
8385 seq = read_seqbegin(&bb->lock);
ab05613a 8386 lo = 0;
8387 rv = 0;
2230dfe4
N		 8388 hi = bb->count;
8389
8390 /* Binary search between lo and hi for 'target'
8391 * i.e. for the last range that starts before 'target'
8392 */
8393 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8394 * are known not to be the last range before target.
8395 * VARIANT: hi-lo is the number of possible
8396 * ranges, and decreases until it reaches 1
8397 */
8398 while (hi - lo > 1) {
8399 int mid = (lo + hi) / 2;
8400 sector_t a = BB_OFFSET(p[mid]);
8401 if (a < target)
8402 /* This could still be the one, earlier ranges
8403 * could not. */
8404 lo = mid;
8405 else
8406 /* This and later ranges are definitely out. */
8407 hi = mid;
8408 }
8409 /* 'lo' might be the last that started before target, but 'hi' isn't */
8410 if (hi > lo) {
8411 /* need to check all range that end after 's' to see if
8412 * any are unacknowledged.
8413 */
8414 while (lo >= 0 &&
8415 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8416 if (BB_OFFSET(p[lo]) < target) {
8417 /* starts before the end, and finishes after
8418 * the start, so they must overlap
8419 */
8420 if (rv != -1 && BB_ACK(p[lo]))
8421 rv = 1;
8422 else
8423 rv = -1;
8424 *first_bad = BB_OFFSET(p[lo]);
8425 *bad_sectors = BB_LEN(p[lo]);
8426 }
8427 lo--;
8428 }
8429 }
8430
8431 if (read_seqretry(&bb->lock, seq))
8432 goto retry;
8433
8434 return rv;
8435}
8436EXPORT_SYMBOL_GPL(md_is_badblock);
8437
8438/*
8439 * Add a range of bad blocks to the table.
8440 * This might extend the table, or might contract it
8441 * if two adjacent ranges can be merged.
8442 * We binary-search to find the 'insertion' point, then
8443 * decide how best to handle it.
8444 */
8445static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8446 int acknowledged)
8447{
8448 u64 *p;
8449 int lo, hi;
8450 int rv = 1;
905b0297 8451 unsigned long flags;
2230dfe4
N		 8452
8453 if (bb->shift < 0)
8454 /* badblocks are disabled */
8455 return 0;
8456
8457 if (bb->shift) {
8458 /* round the start down, and the end up */
8459 sector_t next = s + sectors;
8460 s >>= bb->shift;
8461 next += (1<<bb->shift) - 1;
8462 next >>= bb->shift;
8463 sectors = next - s;
8464 }
8465
905b0297 8466 write_seqlock_irqsave(&bb->lock, flags);
2230dfe4
N		 8467
8468 p = bb->page;
8469 lo = 0;
8470 hi = bb->count;
8471 /* Find the last range that starts at-or-before 's' */
8472 while (hi - lo > 1) {
8473 int mid = (lo + hi) / 2;
8474 sector_t a = BB_OFFSET(p[mid]);
8475 if (a <= s)
8476 lo = mid;
8477 else
8478 hi = mid;
8479 }
8480 if (hi > lo && BB_OFFSET(p[lo]) > s)
8481 hi = lo;
8482
8483 if (hi > lo) {
8484 /* we found a range that might merge with the start
8485 * of our new range
8486 */
8487 sector_t a = BB_OFFSET(p[lo]);
8488 sector_t e = a + BB_LEN(p[lo]);
8489 int ack = BB_ACK(p[lo]);
8490 if (e >= s) {
8491 /* Yes, we can merge with a previous range */
8492 if (s == a && s + sectors >= e)
8493 /* new range covers old */
8494 ack = acknowledged;
8495 else
8496 ack = ack && acknowledged;
8497
8498 if (e < s + sectors)
8499 e = s + sectors;
8500 if (e - a <= BB_MAX_LEN) {
8501 p[lo] = BB_MAKE(a, e-a, ack);
8502 s = e;
8503 } else {
8504 /* does not all fit in one range,
8505 * make p[lo] maximal
8506 */
8507 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8508 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8509 s = a + BB_MAX_LEN;
8510 }
8511 sectors = e - s;
8512 }
8513 }
8514 if (sectors && hi < bb->count) {
8515 /* 'hi' points to the first range that starts after 's'.
8516 * Maybe we can merge with the start of that range */
8517 sector_t a = BB_OFFSET(p[hi]);
8518 sector_t e = a + BB_LEN(p[hi]);
8519 int ack = BB_ACK(p[hi]);
8520 if (a <= s + sectors) {
8521 /* merging is possible */
8522 if (e <= s + sectors) {
8523 /* full overlap */
8524 e = s + sectors;
8525 ack = acknowledged;
8526 } else
8527 ack = ack && acknowledged;
8528
8529 a = s;
8530 if (e - a <= BB_MAX_LEN) {
8531 p[hi] = BB_MAKE(a, e-a, ack);
8532 s = e;
8533 } else {
8534 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8535 s = a + BB_MAX_LEN;
8536 }
8537 sectors = e - s;
8538 lo = hi;
8539 hi++;
8540 }
8541 }
8542 if (sectors == 0 && hi < bb->count) {
8543 /* we might be able to combine lo and hi */
8544 /* Note: 's' is at the end of 'lo' */
8545 sector_t a = BB_OFFSET(p[hi]);
8546 int lolen = BB_LEN(p[lo]);
8547 int hilen = BB_LEN(p[hi]);
8548 int newlen = lolen + hilen - (s - a);
8549 if (s >= a && newlen < BB_MAX_LEN) {
8550 /* yes, we can combine them */
8551 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8552 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8553 memmove(p + hi, p + hi + 1,
8554 (bb->count - hi - 1) * 8);
8555 bb->count--;
8556 }
8557 }
8558 while (sectors) {
8559 /* didn't merge (it all).
8560 * Need to add a range just before 'hi' */
8561 if (bb->count >= MD_MAX_BADBLOCKS) {
8562 /* No room for more */
8563 rv = 0;
8564 break;
8565 } else {
8566 int this_sectors = sectors;
8567 memmove(p + hi + 1, p + hi,
8568 (bb->count - hi) * 8);
8569 bb->count++;
8570
8571 if (this_sectors > BB_MAX_LEN)
8572 this_sectors = BB_MAX_LEN;
8573 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8574 sectors -= this_sectors;
8575 s += this_sectors;
8576 }
8577 }
8578
8579 bb->changed = 1;
de393cde
N		 8580 if (!acknowledged)
8581 bb->unacked_exist = 1;
905b0297 8582 write_sequnlock_irqrestore(&bb->lock, flags);
2230dfe4
N		 8583
8584 return rv;
8585}
8586
3cb03002 8587int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8588 int is_new)
2230dfe4 8589{
c6563a8c
N		 8590 int rv;
8591 if (is_new)
8592 s += rdev->new_data_offset;
8593 else
8594 s += rdev->data_offset;
8595 rv = md_set_badblocks(&rdev->badblocks,
8596 s, sectors, 0);
2230dfe4
N		 8597 if (rv) {
8598 /* Make sure they get written out promptly */
8bd2f0a0 8599 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 8600 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8601 md_wakeup_thread(rdev->mddev->thread);
8602 }
8603 return rv;
8604}
8605EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8606
8607/*
8608 * Remove a range of bad blocks from the table.
8609 * This may involve extending the table if we spilt a region,
8610 * but it must not fail. So if the table becomes full, we just
8611 * drop the remove request.
8612 */
8613static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8614{
8615 u64 *p;
8616 int lo, hi;
8617 sector_t target = s + sectors;
8618 int rv = 0;
8619
8620 if (bb->shift > 0) {
8621 /* When clearing we round the start up and the end down.
8622 * This should not matter as the shift should align with
8623 * the block size and no rounding should ever be needed.
8624 * However it is better the think a block is bad when it
8625 * isn't than to think a block is not bad when it is.
8626 */
8627 s += (1<<bb->shift) - 1;
8628 s >>= bb->shift;
8629 target >>= bb->shift;
8630 sectors = target - s;
8631 }
8632
8633 write_seqlock_irq(&bb->lock);
8634
8635 p = bb->page;
8636 lo = 0;
8637 hi = bb->count;
8638 /* Find the last range that starts before 'target' */
8639 while (hi - lo > 1) {
8640 int mid = (lo + hi) / 2;
8641 sector_t a = BB_OFFSET(p[mid]);
8642 if (a < target)
8643 lo = mid;
8644 else
8645 hi = mid;
8646 }
8647 if (hi > lo) {
8648 /* p[lo] is the last range that could overlap the
8649 * current range. Earlier ranges could also overlap,
8650 * but only this one can overlap the end of the range.
8651 */
8652 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8653 /* Partial overlap, leave the tail of this range */
8654 int ack = BB_ACK(p[lo]);
8655 sector_t a = BB_OFFSET(p[lo]);
8656 sector_t end = a + BB_LEN(p[lo]);
8657
8658 if (a < s) {
8659 /* we need to split this range */
8660 if (bb->count >= MD_MAX_BADBLOCKS) {
8b32bf5e 8661 rv = -ENOSPC;
2230dfe4
N		 8662 goto out;
8663 }
8664 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8665 bb->count++;
8666 p[lo] = BB_MAKE(a, s-a, ack);
8667 lo++;
8668 }
8669 p[lo] = BB_MAKE(target, end - target, ack);
8670 /* there is no longer an overlap */
8671 hi = lo;
8672 lo--;
8673 }
8674 while (lo >= 0 &&
8675 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8676 /* This range does overlap */
8677 if (BB_OFFSET(p[lo]) < s) {
8678 /* Keep the early parts of this range. */
8679 int ack = BB_ACK(p[lo]);
8680 sector_t start = BB_OFFSET(p[lo]);
8681 p[lo] = BB_MAKE(start, s - start, ack);
8682 /* now low doesn't overlap, so.. */
8683 break;
8684 }
8685 lo--;
8686 }
8687 /* 'lo' is strictly before, 'hi' is strictly after,
8688 * anything between needs to be discarded
8689 */
8690 if (hi - lo > 1) {
8691 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8692 bb->count -= (hi - lo - 1);
8693 }
8694 }
8695
8696 bb->changed = 1;
8697out:
8698 write_sequnlock_irq(&bb->lock);
8699 return rv;
8700}
8701
c6563a8c
N		 8702int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8703 int is_new)
2230dfe4 8704{
c6563a8c
N		 8705 if (is_new)
8706 s += rdev->new_data_offset;
8707 else
8708 s += rdev->data_offset;
2230dfe4 8709 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8710 s, sectors);
2230dfe4
N		 8711}
8712EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8713
8714/*
8715 * Acknowledge all bad blocks in a list.
8716 * This only succeeds if ->changed is clear. It is used by
8717 * in-kernel metadata updates
8718 */
8719void md_ack_all_badblocks(struct badblocks *bb)
8720{
8721 if (bb->page == NULL || bb->changed)
8722 /* no point even trying */
8723 return;
8724 write_seqlock_irq(&bb->lock);
8725
ecb178bb 8726 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N		 8727 u64 *p = bb->page;
8728 int i;
8729 for (i = 0; i < bb->count ; i++) {
8730 if (!BB_ACK(p[i])) {
8731 sector_t start = BB_OFFSET(p[i]);
8732 int len = BB_LEN(p[i]);
8733 p[i] = BB_MAKE(start, len, 1);
8734 }
8735 }
de393cde 8736 bb->unacked_exist = 0;
2230dfe4
N		 8737 }
8738 write_sequnlock_irq(&bb->lock);
8739}
8740EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8741
16c791a5
N		 8742/* sysfs access to bad-blocks list.
8743 * We present two files.
8744 * 'bad-blocks' lists sector numbers and lengths of ranges that
8745 * are recorded as bad. The list is truncated to fit within
8746 * the one-page limit of sysfs.
8747 * Writing "sector length" to this file adds an acknowledged
8748 * bad block list.
8749 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8750 * been acknowledged. Writing to this file adds bad blocks
8751 * without acknowledging them. This is largely for testing.
8752 */
8753
8754static ssize_t
8755badblocks_show(struct badblocks *bb, char *page, int unack)
8756{
8757 size_t len;
8758 int i;
8759 u64 *p = bb->page;
8760 unsigned seq;
8761
8762 if (bb->shift < 0)
8763 return 0;
8764
8765retry:
8766 seq = read_seqbegin(&bb->lock);
8767
8768 len = 0;
8769 i = 0;
8770
8771 while (len < PAGE_SIZE && i < bb->count) {
8772 sector_t s = BB_OFFSET(p[i]);
8773 unsigned int length = BB_LEN(p[i]);
8774 int ack = BB_ACK(p[i]);
8775 i++;
8776
8777 if (unack && ack)
8778 continue;
8779
8780 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8781 (unsigned long long)s << bb->shift,
8782 length << bb->shift);
8783 }
de393cde
N		 8784 if (unack && len == 0)
8785 bb->unacked_exist = 0;
16c791a5
N		 8786
8787 if (read_seqretry(&bb->lock, seq))
8788 goto retry;
8789
8790 return len;
8791}
8792
8793#define DO_DEBUG 1
8794
8795static ssize_t
8796badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8797{
8798 unsigned long long sector;
8799 int length;
8800 char newline;
8801#ifdef DO_DEBUG
8802 /* Allow clearing via sysfs *only* for testing/debugging.
8803 * Normally only a successful write may clear a badblock
8804 */
8805 int clear = 0;
8806 if (page[0] == '-') {
8807 clear = 1;
8808 page++;
8809 }
8810#endif /* DO_DEBUG */
8811
8812 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8813 case 3:
8814 if (newline != '\n')
8815 return -EINVAL;
8816 case 2:
8817 if (length <= 0)
8818 return -EINVAL;
8819 break;
8820 default:
8821 return -EINVAL;
8822 }
8823
8824#ifdef DO_DEBUG
8825 if (clear) {
8826 md_clear_badblocks(bb, sector, length);
8827 return len;
8828 }
8829#endif /* DO_DEBUG */
8830 if (md_set_badblocks(bb, sector, length, !unack))
8831 return len;
8832 else
8833 return -ENOSPC;
8834}
8835
75c96f85
AB		 8836static int md_notify_reboot(struct notifier_block *this,
8837 unsigned long code, void *x)
1da177e4
LT		 8838{
8839 struct list_head *tmp;
fd01b88c 8840 struct mddev *mddev;
2dba6a91 8841 int need_delay = 0;
1da177e4 8842
c744a65c
N		 8843 for_each_mddev(mddev, tmp) {
8844 if (mddev_trylock(mddev)) {
30b8aa91
N		 8845 if (mddev->pers)
8846 __md_stop_writes(mddev);
0f62fb22
N		 8847 if (mddev->persistent)
8848 mddev->safemode = 2;
c744a65c 8849 mddev_unlock(mddev);
2dba6a91 8850 }
c744a65c 8851 need_delay = 1;
1da177e4 8852 }
c744a65c
N		 8853 /*
8854 * certain more exotic SCSI devices are known to be
8855 * volatile wrt too early system reboots. While the
8856 * right place to handle this issue is the given
8857 * driver, we do want to have a safe RAID driver ...
8858 */
8859 if (need_delay)
8860 mdelay(1000*1);
8861
1da177e4
LT		 8862 return NOTIFY_DONE;
8863}
8864
75c96f85 8865static struct notifier_block md_notifier = {
1da177e4
LT		 8866 .notifier_call = md_notify_reboot,
8867 .next = NULL,
8868 .priority = INT_MAX, /* before any real devices */
8869};
8870
8871static void md_geninit(void)
8872{
36a4e1fe 8873 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8874
c7705f34 8875 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8876}
8877
75c96f85 8878static int __init md_init(void)
1da177e4 8879{
e804ac78
TH		 8880 int ret = -ENOMEM;
8881
ada609ee 8882 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8883 if (!md_wq)
8884 goto err_wq;
8885
8886 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8887 if (!md_misc_wq)
8888 goto err_misc_wq;
8889
8890 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8891 goto err_md;
8892
8893 if ((ret = register_blkdev(0, "mdp")) < 0)
8894 goto err_mdp;
8895 mdp_major = ret;
8896
af5628f0 8897 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
e8703fe1
N		 8898 md_probe, NULL, NULL);
8899 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8900 md_probe, NULL, NULL);
8901
1da177e4 8902 register_reboot_notifier(&md_notifier);
0b4d4147 8903 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8904
8905 md_geninit();
d710e138 8906 return 0;
1da177e4 8907
e804ac78
TH		 8908err_mdp:
8909 unregister_blkdev(MD_MAJOR, "md");
8910err_md:
8911 destroy_workqueue(md_misc_wq);
8912err_misc_wq:
8913 destroy_workqueue(md_wq);
8914err_wq:
8915 return ret;
8916}
1da177e4 8917
1d7e3e96
GR		 8918void md_reload_sb(struct mddev *mddev)
8919{
8920 struct md_rdev *rdev, *tmp;
8921
8922 rdev_for_each_safe(rdev, tmp, mddev) {
8923 rdev->sb_loaded = 0;
8924 ClearPageUptodate(rdev->sb_page);
8925 }
8926 mddev->raid_disks = 0;
8927 analyze_sbs(mddev);
8928 rdev_for_each_safe(rdev, tmp, mddev) {
8929 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8930 /* since we don't write to faulty devices, we figure out if the
8931 * disk is faulty by comparing events
8932 */
8933 if (mddev->events > sb->events)
8934 set_bit(Faulty, &rdev->flags);
8935 }
8936
8937}
8938EXPORT_SYMBOL(md_reload_sb);
8939
1da177e4
LT		 8940#ifndef MODULE
8941
8942/*
8943 * Searches all registered partitions for autorun RAID arrays
8944 * at boot time.
8945 */
4d936ec1
ME		 8946
8947static LIST_HEAD(all_detected_devices);
8948struct detected_devices_node {
8949 struct list_head list;
8950 dev_t dev;
8951};
1da177e4
LT		 8952
8953void md_autodetect_dev(dev_t dev)
8954{
4d936ec1
ME		 8955 struct detected_devices_node *node_detected_dev;
8956
8957 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8958 if (node_detected_dev) {
8959 node_detected_dev->dev = dev;
8960 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8961 } else {
8962 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8963 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8964 }
1da177e4
LT		 8965}
8966
1da177e4
LT		 8967static void autostart_arrays(int part)
8968{
3cb03002 8969 struct md_rdev *rdev;
4d936ec1
ME		 8970 struct detected_devices_node *node_detected_dev;
8971 dev_t dev;
8972 int i_scanned, i_passed;
1da177e4 8973
4d936ec1
ME		 8974 i_scanned = 0;
8975 i_passed = 0;
1da177e4 8976
4d936ec1 8977 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8978
4d936ec1
ME		 8979 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8980 i_scanned++;
8981 node_detected_dev = list_entry(all_detected_devices.next,
8982 struct detected_devices_node, list);
8983 list_del(&node_detected_dev->list);
8984 dev = node_detected_dev->dev;
8985 kfree(node_detected_dev);
df968c4e 8986 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8987 if (IS_ERR(rdev))
8988 continue;
8989
403df478 8990 if (test_bit(Faulty, &rdev->flags))
1da177e4 8991 continue;
403df478 8992
d0fae18f 8993 set_bit(AutoDetected, &rdev->flags);
1da177e4 8994 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8995 i_passed++;
1da177e4 8996 }
4d936ec1
ME		 8997
8998 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8999 i_scanned, i_passed);
1da177e4
LT		 9000
9001 autorun_devices(part);
9002}
9003
fdee8ae4 9004#endif /* !MODULE */
1da177e4
LT		 9005
9006static __exit void md_exit(void)
9007{
fd01b88c 9008 struct mddev *mddev;
1da177e4 9009 struct list_head *tmp;
e2f23b60 9010 int delay = 1;
8ab5e4c1 9011
af5628f0 9012 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
e8703fe1 9013 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 9014
3dbd8c2e 9015 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 9016 unregister_blkdev(mdp_major, "mdp");
9017 unregister_reboot_notifier(&md_notifier);
9018 unregister_sysctl_table(raid_table_header);
e2f23b60
N		 9019
9020 /* We cannot unload the modules while some process is
9021 * waiting for us in select() or poll() - wake them up
9022 */
9023 md_unloading = 1;
9024 while (waitqueue_active(&md_event_waiters)) {
9025 /* not safe to leave yet */
9026 wake_up(&md_event_waiters);
9027 msleep(delay);
9028 delay += delay;
9029 }
1da177e4 9030 remove_proc_entry("mdstat", NULL);
e2f23b60 9031
29ac4aa3 9032 for_each_mddev(mddev, tmp) {
1da177e4 9033 export_array(mddev);
d3374825 9034 mddev->hold_active = 0;
1da177e4 9035 }
e804ac78
TH		 9036 destroy_workqueue(md_misc_wq);
9037 destroy_workqueue(md_wq);
1da177e4
LT		 9038}
9039
685784aa 9040subsys_initcall(md_init);
1da177e4
LT		 9041module_exit(md_exit)
9042
f91de92e
N		 9043static int get_ro(char *buffer, struct kernel_param *kp)
9044{
9045 return sprintf(buffer, "%d", start_readonly);
9046}
9047static int set_ro(const char *val, struct kernel_param *kp)
9048{
4c9309c0 9049 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
f91de92e
N		 9050}
9051
80ca3a44
N		 9052module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9053module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
efeb53c0 9054module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 9055
1da177e4 9056MODULE_LICENSE("GPL");
0efb9e61 9057MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 9058MODULE_ALIAS("md");
72008652 9059MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)