projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: clear mddev->private when it has been freed.
[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;
1edf80d3 2027 char *s;
5e55e2f5 2028 int err;
1da177e4 2029
11e2ede0
DW		 2030 /* prevent duplicates */
2031 if (find_rdev(mddev, rdev->bdev->bd_dev))
2032 return -EEXIST;
2033
dd8ac336
AN		 2034 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2035 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2036 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2037 if (mddev->pers) {
2038 /* Cannot change size, so fail
2039 * If mddev->level <= 0, then we don't care
2040 * about aligning sizes (e.g. linear)
2041 */
2042 if (mddev->level > 0)
2043 return -ENOSPC;
2044 } else
dd8ac336 2045 mddev->dev_sectors = rdev->sectors;
2bf071bf 2046 }
1da177e4
LT		 2047
2048 /* Verify rdev->desc_nr is unique.
2049 * If it is -1, assign a free number, else
2050 * check number is not in use
2051 */
4878e9eb 2052 rcu_read_lock();
1da177e4
LT		 2053 if (rdev->desc_nr < 0) {
2054 int choice = 0;
4878e9eb
N		 2055 if (mddev->pers)
2056 choice = mddev->raid_disks;
57d051dc 2057 while (md_find_rdev_nr_rcu(mddev, choice))
1da177e4
LT		 2058 choice++;
2059 rdev->desc_nr = choice;
2060 } else {
57d051dc 2061 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
4878e9eb 2062 rcu_read_unlock();
1da177e4 2063 return -EBUSY;
4878e9eb 2064 }
1da177e4 2065 }
4878e9eb 2066 rcu_read_unlock();
de01dfad
N		 2067 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2068 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2069 mdname(mddev), mddev->max_disks);
2070 return -EBUSY;
2071 }
19133a42 2072 bdevname(rdev->bdev,b);
649316b2 2073 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2074 *s = '!';
649316b2 2075
1da177e4 2076 rdev->mddev = mddev;
19133a42 2077 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2078
b2d6db58 2079 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2080 goto fail;
86e6ffdd 2081
0762b8bd 2082 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2083 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2084 /* failure here is OK */;
2085 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2086
4b80991c 2087 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2088 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2089
2090 /* May as well allow recovery to be retried once */
5389042f 2091 mddev->recovery_disabled++;
3f9d99c1 2092
1da177e4 2093 return 0;
5e55e2f5
N		 2094
2095 fail:
2096 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2097 b, mdname(mddev));
2098 return err;
1da177e4
LT		 2099}
2100
177a99b2 2101static void md_delayed_delete(struct work_struct *ws)
5792a285 2102{
3cb03002 2103 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2104 kobject_del(&rdev->kobj);
177a99b2 2105 kobject_put(&rdev->kobj);
5792a285
N		 2106}
2107
f72ffdd6 2108static void unbind_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2109{
2110 char b[BDEVNAME_SIZE];
403df478 2111
49731baa 2112 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2113 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2114 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2115 rdev->mddev = NULL;
86e6ffdd 2116 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2117 sysfs_put(rdev->sysfs_state);
2118 rdev->sysfs_state = NULL;
2230dfe4 2119 rdev->badblocks.count = 0;
5792a285 2120 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2121 * writing to 'remove' to "dev/state". We also need
2122 * to delay it due to rcu usage.
5792a285 2123 */
4b80991c 2124 synchronize_rcu();
177a99b2
N		 2125 INIT_WORK(&rdev->del_work, md_delayed_delete);
2126 kobject_get(&rdev->kobj);
e804ac78 2127 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2128}
2129
2130/*
2131 * prevent the device from being mounted, repartitioned or
2132 * otherwise reused by a RAID array (or any other kernel
2133 * subsystem), by bd_claiming the device.
2134 */
3cb03002 2135static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2136{
2137 int err = 0;
2138 struct block_device *bdev;
2139 char b[BDEVNAME_SIZE];
2140
d4d77629 2141 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2142 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2143 if (IS_ERR(bdev)) {
2144 printk(KERN_ERR "md: could not open %s.\n",
2145 __bdevname(dev, b));
2146 return PTR_ERR(bdev);
2147 }
1da177e4
LT		 2148 rdev->bdev = bdev;
2149 return err;
2150}
2151
3cb03002 2152static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2153{
2154 struct block_device *bdev = rdev->bdev;
2155 rdev->bdev = NULL;
e525fd89 2156 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2157}
2158
2159void md_autodetect_dev(dev_t dev);
2160
f72ffdd6 2161static void export_rdev(struct md_rdev *rdev)
1da177e4
LT		 2162{
2163 char b[BDEVNAME_SIZE];
403df478 2164
1da177e4
LT		 2165 printk(KERN_INFO "md: export_rdev(%s)\n",
2166 bdevname(rdev->bdev,b));
545c8795 2167 md_rdev_clear(rdev);
1da177e4 2168#ifndef MODULE
d0fae18f
N		 2169 if (test_bit(AutoDetected, &rdev->flags))
2170 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2171#endif
2172 unlock_rdev(rdev);
86e6ffdd 2173 kobject_put(&rdev->kobj);
1da177e4
LT		 2174}
2175
fb56dfef 2176void md_kick_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2177{
2178 unbind_rdev_from_array(rdev);
2179 export_rdev(rdev);
2180}
fb56dfef 2181EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
1da177e4 2182
fd01b88c 2183static void export_array(struct mddev *mddev)
1da177e4 2184{
0638bb0e 2185 struct md_rdev *rdev;
1da177e4 2186
0638bb0e
N		 2187 while (!list_empty(&mddev->disks)) {
2188 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2189 same_set);
fb56dfef 2190 md_kick_rdev_from_array(rdev);
1da177e4 2191 }
1da177e4
LT		 2192 mddev->raid_disks = 0;
2193 mddev->major_version = 0;
2194}
2195
f72ffdd6 2196static void sync_sbs(struct mddev *mddev, int nospares)
1da177e4 2197{
42543769
N		 2198 /* Update each superblock (in-memory image), but
2199 * if we are allowed to, skip spares which already
2200 * have the right event counter, or have one earlier
2201 * (which would mean they aren't being marked as dirty
2202 * with the rest of the array)
2203 */
3cb03002 2204 struct md_rdev *rdev;
dafb20fa 2205 rdev_for_each(rdev, mddev) {
42543769
N		 2206 if (rdev->sb_events == mddev->events ||
2207 (nospares &&
2208 rdev->raid_disk < 0 &&
42543769
N		 2209 rdev->sb_events+1 == mddev->events)) {
2210 /* Don't update this superblock */
2211 rdev->sb_loaded = 2;
2212 } else {
076f968b 2213 sync_super(mddev, rdev);
42543769
N		 2214 rdev->sb_loaded = 1;
2215 }
1da177e4
LT		 2216 }
2217}
2218
1aee41f6 2219void md_update_sb(struct mddev *mddev, int force_change)
1da177e4 2220{
3cb03002 2221 struct md_rdev *rdev;
06d91a5f 2222 int sync_req;
42543769 2223 int nospares = 0;
2699b672 2224 int any_badblocks_changed = 0;
1da177e4 2225
d87f064f
N		 2226 if (mddev->ro) {
2227 if (force_change)
2228 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2229 return;
2230 }
1da177e4 2231repeat:
3a3a5ddb 2232 /* First make sure individual recovery_offsets are correct */
dafb20fa 2233 rdev_for_each(rdev, mddev) {
3a3a5ddb
N		 2234 if (rdev->raid_disk >= 0 &&
2235 mddev->delta_disks >= 0 &&
2236 !test_bit(In_sync, &rdev->flags) &&
2237 mddev->curr_resync_completed > rdev->recovery_offset)
2238 rdev->recovery_offset = mddev->curr_resync_completed;
2239
f72ffdd6 2240 }
bd52b746 2241 if (!mddev->persistent) {
070dc6dd 2242 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2243 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2244 if (!mddev->external) {
d97a41dc 2245 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
dafb20fa 2246 rdev_for_each(rdev, mddev) {
de393cde 2247 if (rdev->badblocks.changed) {
d0962936 2248 rdev->badblocks.changed = 0;
de393cde
N		 2249 md_ack_all_badblocks(&rdev->badblocks);
2250 md_error(mddev, rdev);
2251 }
2252 clear_bit(Blocked, &rdev->flags);
2253 clear_bit(BlockedBadBlocks, &rdev->flags);
2254 wake_up(&rdev->blocked_wait);
2255 }
2256 }
3a3a5ddb
N		 2257 wake_up(&mddev->sb_wait);
2258 return;
2259 }
2260
85572d7c 2261 spin_lock(&mddev->lock);
84692195 2262
3a3a5ddb
N		 2263 mddev->utime = get_seconds();
2264
850b2b42
N		 2265 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2266 force_change = 1;
2267 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2268 /* just a clean<-> dirty transition, possibly leave spares alone,
2269 * though if events isn't the right even/odd, we will have to do
2270 * spares after all
2271 */
2272 nospares = 1;
2273 if (force_change)
2274 nospares = 0;
2275 if (mddev->degraded)
84692195
N		 2276 /* If the array is degraded, then skipping spares is both
2277 * dangerous and fairly pointless.
2278 * Dangerous because a device that was removed from the array
2279 * might have a event_count that still looks up-to-date,
2280 * so it can be re-added without a resync.
2281 * Pointless because if there are any spares to skip,
2282 * then a recovery will happen and soon that array won't
2283 * be degraded any more and the spare can go back to sleep then.
2284 */
850b2b42 2285 nospares = 0;
84692195 2286
06d91a5f 2287 sync_req = mddev->in_sync;
42543769
N		 2288
2289 /* If this is just a dirty<->clean transition, and the array is clean
2290 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2291 if (nospares
42543769 2292 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2293 && mddev->can_decrease_events
2294 && mddev->events != 1) {
42543769 2295 mddev->events--;
a8707c08
N		 2296 mddev->can_decrease_events = 0;
2297 } else {
42543769
N		 2298 /* otherwise we have to go forward and ... */
2299 mddev->events ++;
a8707c08 2300 mddev->can_decrease_events = nospares;
42543769 2301 }
1da177e4 2302
403df478
N		 2303 /*
2304 * This 64-bit counter should never wrap.
2305 * Either we are in around ~1 trillion A.C., assuming
2306 * 1 reboot per second, or we have a bug...
2307 */
2308 WARN_ON(mddev->events == 0);
2699b672 2309
dafb20fa 2310 rdev_for_each(rdev, mddev) {
2699b672
N		 2311 if (rdev->badblocks.changed)
2312 any_badblocks_changed++;
de393cde
N		 2313 if (test_bit(Faulty, &rdev->flags))
2314 set_bit(FaultRecorded, &rdev->flags);
2315 }
2699b672 2316
e691063a 2317 sync_sbs(mddev, nospares);
85572d7c 2318 spin_unlock(&mddev->lock);
1da177e4 2319
36a4e1fe
N		 2320 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2321 mdname(mddev), mddev->in_sync);
1da177e4 2322
4ad13663 2323 bitmap_update_sb(mddev->bitmap);
dafb20fa 2324 rdev_for_each(rdev, mddev) {
1da177e4 2325 char b[BDEVNAME_SIZE];
36a4e1fe 2326
42543769
N		 2327 if (rdev->sb_loaded != 1)
2328 continue; /* no noise on spare devices */
1da177e4 2329
f466722c 2330 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2331 md_super_write(mddev,rdev,
0f420358 2332 rdev->sb_start, rdev->sb_size,
7bfa19f2 2333 rdev->sb_page);
36a4e1fe
N		 2334 pr_debug("md: (write) %s's sb offset: %llu\n",
2335 bdevname(rdev->bdev, b),
2336 (unsigned long long)rdev->sb_start);
42543769 2337 rdev->sb_events = mddev->events;
2699b672
N		 2338 if (rdev->badblocks.size) {
2339 md_super_write(mddev, rdev,
2340 rdev->badblocks.sector,
2341 rdev->badblocks.size << 9,
2342 rdev->bb_page);
2343 rdev->badblocks.size = 0;
2344 }
7bfa19f2 2345
f466722c 2346 } else
36a4e1fe
N		 2347 pr_debug("md: %s (skipping faulty)\n",
2348 bdevname(rdev->bdev, b));
d70ed2e4 2349
7bfa19f2 2350 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2351 /* only need to write one superblock... */
2352 break;
2353 }
a9701a30 2354 md_super_wait(mddev);
850b2b42 2355 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2356
85572d7c 2357 spin_lock(&mddev->lock);
850b2b42
N		 2358 if (mddev->in_sync != sync_req ||
2359 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2360 /* have to write it out again */
85572d7c 2361 spin_unlock(&mddev->lock);
06d91a5f
N		 2362 goto repeat;
2363 }
850b2b42 2364 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
85572d7c 2365 spin_unlock(&mddev->lock);
3d310eb7 2366 wake_up(&mddev->sb_wait);
acb180b0
N		 2367 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2368 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2369
dafb20fa 2370 rdev_for_each(rdev, mddev) {
de393cde
N		 2371 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2372 clear_bit(Blocked, &rdev->flags);
2373
2374 if (any_badblocks_changed)
2699b672 2375 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2376 clear_bit(BlockedBadBlocks, &rdev->flags);
2377 wake_up(&rdev->blocked_wait);
2378 }
1da177e4 2379}
1aee41f6 2380EXPORT_SYMBOL(md_update_sb);
1da177e4 2381
a6da4ef8
GR		 2382static int add_bound_rdev(struct md_rdev *rdev)
2383{
2384 struct mddev *mddev = rdev->mddev;
2385 int err = 0;
2386
2387 if (!mddev->pers->hot_remove_disk) {
2388 /* If there is hot_add_disk but no hot_remove_disk
2389 * then added disks for geometry changes,
2390 * and should be added immediately.
2391 */
2392 super_types[mddev->major_version].
2393 validate_super(mddev, rdev);
2394 err = mddev->pers->hot_add_disk(mddev, rdev);
2395 if (err) {
2396 unbind_rdev_from_array(rdev);
2397 export_rdev(rdev);
2398 return err;
2399 }
2400 }
2401 sysfs_notify_dirent_safe(rdev->sysfs_state);
2402
2403 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2404 if (mddev->degraded)
2405 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2406 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2407 md_new_event(mddev);
2408 md_wakeup_thread(mddev->thread);
2409 return 0;
2410}
1da177e4 2411
7f6ce769 2412/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2413 * We want to accept with case. For this we use cmd_match.
2414 */
2415static int cmd_match(const char *cmd, const char *str)
2416{
2417 /* See if cmd, written into a sysfs file, matches
2418 * str. They must either be the same, or cmd can
2419 * have a trailing newline
2420 */
2421 while (*cmd && *str && *cmd == *str) {
2422 cmd++;
2423 str++;
2424 }
2425 if (*cmd == '\n')
2426 cmd++;
2427 if (*str || *cmd)
2428 return 0;
2429 return 1;
2430}
2431
86e6ffdd
N		 2432struct rdev_sysfs_entry {
2433 struct attribute attr;
3cb03002
N		 2434 ssize_t (*show)(struct md_rdev *, char *);
2435 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2436};
2437
2438static ssize_t
3cb03002 2439state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2440{
2441 char *sep = "";
20a49ff6 2442 size_t len = 0;
758bfc8a 2443 unsigned long flags = ACCESS_ONCE(rdev->flags);
86e6ffdd 2444
758bfc8a 2445 if (test_bit(Faulty, &flags) ||
de393cde 2446 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2447 len+= sprintf(page+len, "%sfaulty",sep);
2448 sep = ",";
2449 }
758bfc8a 2450 if (test_bit(In_sync, &flags)) {
86e6ffdd
N		 2451 len += sprintf(page+len, "%sin_sync",sep);
2452 sep = ",";
2453 }
758bfc8a 2454 if (test_bit(WriteMostly, &flags)) {
f655675b
N		 2455 len += sprintf(page+len, "%swrite_mostly",sep);
2456 sep = ",";
2457 }
758bfc8a 2458 if (test_bit(Blocked, &flags) ||
52c64152 2459 (rdev->badblocks.unacked_exist
758bfc8a 2460 && !test_bit(Faulty, &flags))) {
6bfe0b49
DW		 2461 len += sprintf(page+len, "%sblocked", sep);
2462 sep = ",";
2463 }
758bfc8a
N		 2464 if (!test_bit(Faulty, &flags) &&
2465 !test_bit(In_sync, &flags)) {
86e6ffdd
N		 2466 len += sprintf(page+len, "%sspare", sep);
2467 sep = ",";
2468 }
758bfc8a 2469 if (test_bit(WriteErrorSeen, &flags)) {
d7a9d443
N		 2470 len += sprintf(page+len, "%swrite_error", sep);
2471 sep = ",";
2472 }
758bfc8a 2473 if (test_bit(WantReplacement, &flags)) {
2d78f8c4
N		 2474 len += sprintf(page+len, "%swant_replacement", sep);
2475 sep = ",";
2476 }
758bfc8a 2477 if (test_bit(Replacement, &flags)) {
2d78f8c4
N		 2478 len += sprintf(page+len, "%sreplacement", sep);
2479 sep = ",";
2480 }
2481
86e6ffdd
N		 2482 return len+sprintf(page+len, "\n");
2483}
2484
45dc2de1 2485static ssize_t
3cb03002 2486state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2487{
2488 /* can write
de393cde 2489 * faulty - simulates an error
45dc2de1 2490 * remove - disconnects the device
f655675b
N		 2491 * writemostly - sets write_mostly
2492 * -writemostly - clears write_mostly
de393cde
N		 2493 * blocked - sets the Blocked flags
2494 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2495 * insync - sets Insync providing device isn't active
f466722c
N		 2496 * -insync - clear Insync for a device with a slot assigned,
2497 * so that it gets rebuilt based on bitmap
d7a9d443
N		 2498 * write_error - sets WriteErrorSeen
2499 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2500 */
2501 int err = -EINVAL;
2502 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2503 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2504 if (test_bit(Faulty, &rdev->flags))
2505 err = 0;
2506 else
2507 err = -EBUSY;
45dc2de1
N		 2508 } else if (cmd_match(buf, "remove")) {
2509 if (rdev->raid_disk >= 0)
2510 err = -EBUSY;
2511 else {
fd01b88c 2512 struct mddev *mddev = rdev->mddev;
293467aa 2513 if (mddev_is_clustered(mddev))
88bcfef7 2514 md_cluster_ops->remove_disk(mddev, rdev);
fb56dfef 2515 md_kick_rdev_from_array(rdev);
88bcfef7
GR		 2516 if (mddev_is_clustered(mddev))
2517 md_cluster_ops->metadata_update_start(mddev);
3f9d7b0d
N		 2518 if (mddev->pers)
2519 md_update_sb(mddev, 1);
45dc2de1 2520 md_new_event(mddev);
293467aa
GR		 2521 if (mddev_is_clustered(mddev))
2522 md_cluster_ops->metadata_update_finish(mddev);
45dc2de1
N		 2523 err = 0;
2524 }
f655675b
N		 2525 } else if (cmd_match(buf, "writemostly")) {
2526 set_bit(WriteMostly, &rdev->flags);
2527 err = 0;
2528 } else if (cmd_match(buf, "-writemostly")) {
2529 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2530 err = 0;
2531 } else if (cmd_match(buf, "blocked")) {
2532 set_bit(Blocked, &rdev->flags);
2533 err = 0;
2534 } else if (cmd_match(buf, "-blocked")) {
de393cde 2535 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2536 rdev->badblocks.unacked_exist) {
de393cde
N		 2537 /* metadata handler doesn't understand badblocks,
2538 * so we need to fail the device
2539 */
2540 md_error(rdev->mddev, rdev);
2541 }
6bfe0b49 2542 clear_bit(Blocked, &rdev->flags);
de393cde 2543 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2544 wake_up(&rdev->blocked_wait);
2545 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2546 md_wakeup_thread(rdev->mddev->thread);
2547
6d56e278
N		 2548 err = 0;
2549 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2550 set_bit(In_sync, &rdev->flags);
f655675b 2551 err = 0;
f466722c 2552 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
e1960f8c
N		 2553 if (rdev->mddev->pers == NULL) {
2554 clear_bit(In_sync, &rdev->flags);
2555 rdev->saved_raid_disk = rdev->raid_disk;
2556 rdev->raid_disk = -1;
2557 err = 0;
2558 }
d7a9d443
N		 2559 } else if (cmd_match(buf, "write_error")) {
2560 set_bit(WriteErrorSeen, &rdev->flags);
2561 err = 0;
2562 } else if (cmd_match(buf, "-write_error")) {
2563 clear_bit(WriteErrorSeen, &rdev->flags);
2564 err = 0;
2d78f8c4
N		 2565 } else if (cmd_match(buf, "want_replacement")) {
2566 /* Any non-spare device that is not a replacement can
2567 * become want_replacement at any time, but we then need to
2568 * check if recovery is needed.
2569 */
2570 if (rdev->raid_disk >= 0 &&
2571 !test_bit(Replacement, &rdev->flags))
2572 set_bit(WantReplacement, &rdev->flags);
2573 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2574 md_wakeup_thread(rdev->mddev->thread);
2575 err = 0;
2576 } else if (cmd_match(buf, "-want_replacement")) {
2577 /* Clearing 'want_replacement' is always allowed.
2578 * Once replacements starts it is too late though.
2579 */
2580 err = 0;
2581 clear_bit(WantReplacement, &rdev->flags);
2582 } else if (cmd_match(buf, "replacement")) {
2583 /* Can only set a device as a replacement when array has not
2584 * yet been started. Once running, replacement is automatic
2585 * from spares, or by assigning 'slot'.
2586 */
2587 if (rdev->mddev->pers)
2588 err = -EBUSY;
2589 else {
2590 set_bit(Replacement, &rdev->flags);
2591 err = 0;
2592 }
2593 } else if (cmd_match(buf, "-replacement")) {
2594 /* Similarly, can only clear Replacement before start */
2595 if (rdev->mddev->pers)
2596 err = -EBUSY;
2597 else {
2598 clear_bit(Replacement, &rdev->flags);
2599 err = 0;
2600 }
a6da4ef8
GR		 2601 } else if (cmd_match(buf, "re-add")) {
2602 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
97f6cd39
GR		 2603 /* clear_bit is performed _after_ all the devices
2604 * have their local Faulty bit cleared. If any writes
2605 * happen in the meantime in the local node, they
2606 * will land in the local bitmap, which will be synced
2607 * by this node eventually
2608 */
2609 if (!mddev_is_clustered(rdev->mddev) ||
2610 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2611 clear_bit(Faulty, &rdev->flags);
2612 err = add_bound_rdev(rdev);
2613 }
a6da4ef8
GR		 2614 } else
2615 err = -EBUSY;
45dc2de1 2616 }
00bcb4ac
N		 2617 if (!err)
2618 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2619 return err ? err : len;
2620}
80ca3a44 2621static struct rdev_sysfs_entry rdev_state =
750f199e 2622__ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2623
4dbcdc75 2624static ssize_t
3cb03002 2625errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2626{
2627 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2628}
2629
2630static ssize_t
3cb03002 2631errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75 2632{
4c9309c0
AD		 2633 unsigned int n;
2634 int rv;
2635
2636 rv = kstrtouint(buf, 10, &n);
2637 if (rv < 0)
2638 return rv;
2639 atomic_set(&rdev->corrected_errors, n);
2640 return len;
4dbcdc75
N		 2641}
2642static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2643__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2644
014236d2 2645static ssize_t
3cb03002 2646slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2647{
2648 if (rdev->raid_disk < 0)
2649 return sprintf(page, "none\n");
2650 else
2651 return sprintf(page, "%d\n", rdev->raid_disk);
2652}
2653
2654static ssize_t
3cb03002 2655slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2 2656{
4c9309c0 2657 int slot;
c303da6d 2658 int err;
4c9309c0 2659
014236d2
N		 2660 if (strncmp(buf, "none", 4)==0)
2661 slot = -1;
4c9309c0
AD		 2662 else {
2663 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2664 if (err < 0)
2665 return err;
2666 }
6c2fce2e 2667 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2668 /* Setting 'slot' on an active array requires also
2669 * updating the 'rd%d' link, and communicating
2670 * with the personality with ->hot_*_disk.
2671 * For now we only support removing
2672 * failed/spare devices. This normally happens automatically,
2673 * but not when the metadata is externally managed.
2674 */
c303da6d
N		 2675 if (rdev->raid_disk == -1)
2676 return -EEXIST;
2677 /* personality does all needed checks */
01393f3d 2678 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d 2679 return -EINVAL;
746d3207
N		 2680 clear_bit(Blocked, &rdev->flags);
2681 remove_and_add_spares(rdev->mddev, rdev);
2682 if (rdev->raid_disk >= 0)
2683 return -EBUSY;
c303da6d
N		 2684 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2685 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2686 } else if (rdev->mddev->pers) {
6c2fce2e 2687 /* Activating a spare .. or possibly reactivating
6d56e278 2688 * if we ever get bitmaps working here.
6c2fce2e
NB		 2689 */
2690
2691 if (rdev->raid_disk != -1)
2692 return -EBUSY;
2693
c6751b2b
N		 2694 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2695 return -EBUSY;
2696
6c2fce2e
NB		 2697 if (rdev->mddev->pers->hot_add_disk == NULL)
2698 return -EINVAL;
2699
ba1b41b6
N		 2700 if (slot >= rdev->mddev->raid_disks &&
2701 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2702 return -ENOSPC;
2703
6c2fce2e
NB		 2704 rdev->raid_disk = slot;
2705 if (test_bit(In_sync, &rdev->flags))
2706 rdev->saved_raid_disk = slot;
2707 else
2708 rdev->saved_raid_disk = -1;
d30519fc 2709 clear_bit(In_sync, &rdev->flags);
8313b8e5 2710 clear_bit(Bitmap_sync, &rdev->flags);
6c2fce2e
NB		 2711 err = rdev->mddev->pers->
2712 hot_add_disk(rdev->mddev, rdev);
199050ea 2713 if (err) {
6c2fce2e 2714 rdev->raid_disk = -1;
6c2fce2e 2715 return err;
52664732 2716 } else
00bcb4ac 2717 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2718 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2719 /* failure here is OK */;
6c2fce2e 2720 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2721 } else {
ba1b41b6
N		 2722 if (slot >= rdev->mddev->raid_disks &&
2723 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2724 return -ENOSPC;
2725 rdev->raid_disk = slot;
2726 /* assume it is working */
c5d79adb
N		 2727 clear_bit(Faulty, &rdev->flags);
2728 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2729 set_bit(In_sync, &rdev->flags);
00bcb4ac 2730 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2731 }
014236d2
N		 2732 return len;
2733}
2734
014236d2 2735static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2736__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2737
93c8cad0 2738static ssize_t
3cb03002 2739offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2740{
6961ece4 2741 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2742}
2743
2744static ssize_t
3cb03002 2745offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2746{
c6563a8c 2747 unsigned long long offset;
b29bebd6 2748 if (kstrtoull(buf, 10, &offset) < 0)
93c8cad0 2749 return -EINVAL;
8ed0a521 2750 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2751 return -EBUSY;
dd8ac336 2752 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2753 /* Must set offset before size, so overlap checks
2754 * can be sane */
2755 return -EBUSY;
93c8cad0 2756 rdev->data_offset = offset;
25f7fd47 2757 rdev->new_data_offset = offset;
93c8cad0
N		 2758 return len;
2759}
2760
2761static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2762__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2763
c6563a8c
N		 2764static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2765{
2766 return sprintf(page, "%llu\n",
2767 (unsigned long long)rdev->new_data_offset);
2768}
2769
2770static ssize_t new_offset_store(struct md_rdev *rdev,
2771 const char *buf, size_t len)
2772{
2773 unsigned long long new_offset;
2774 struct mddev *mddev = rdev->mddev;
2775
b29bebd6 2776 if (kstrtoull(buf, 10, &new_offset) < 0)
c6563a8c
N		 2777 return -EINVAL;
2778
f851b60d
N		 2779 if (mddev->sync_thread ||
2780 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
c6563a8c
N		 2781 return -EBUSY;
2782 if (new_offset == rdev->data_offset)
2783 /* reset is always permitted */
2784 ;
2785 else if (new_offset > rdev->data_offset) {
2786 /* must not push array size beyond rdev_sectors */
2787 if (new_offset - rdev->data_offset
2788 + mddev->dev_sectors > rdev->sectors)
2789 return -E2BIG;
2790 }
2791 /* Metadata worries about other space details. */
2792
2793 /* decreasing the offset is inconsistent with a backwards
2794 * reshape.
2795 */
2796 if (new_offset < rdev->data_offset &&
2797 mddev->reshape_backwards)
2798 return -EINVAL;
2799 /* Increasing offset is inconsistent with forwards
2800 * reshape. reshape_direction should be set to
2801 * 'backwards' first.
2802 */
2803 if (new_offset > rdev->data_offset &&
2804 !mddev->reshape_backwards)
2805 return -EINVAL;
2806
2807 if (mddev->pers && mddev->persistent &&
2808 !super_types[mddev->major_version]
2809 .allow_new_offset(rdev, new_offset))
2810 return -E2BIG;
2811 rdev->new_data_offset = new_offset;
2812 if (new_offset > rdev->data_offset)
2813 mddev->reshape_backwards = 1;
2814 else if (new_offset < rdev->data_offset)
2815 mddev->reshape_backwards = 0;
2816
2817 return len;
2818}
2819static struct rdev_sysfs_entry rdev_new_offset =
2820__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2821
83303b61 2822static ssize_t
3cb03002 2823rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2824{
dd8ac336 2825 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2826}
2827
c5d79adb
N		 2828static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2829{
2830 /* check if two start/length pairs overlap */
2831 if (s1+l1 <= s2)
2832 return 0;
2833 if (s2+l2 <= s1)
2834 return 0;
2835 return 1;
2836}
2837
b522adcd
DW		 2838static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2839{
2840 unsigned long long blocks;
2841 sector_t new;
2842
b29bebd6 2843 if (kstrtoull(buf, 10, &blocks) < 0)
b522adcd
DW		 2844 return -EINVAL;
2845
2846 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2847 return -EINVAL; /* sector conversion overflow */
2848
2849 new = blocks * 2;
2850 if (new != blocks * 2)
2851 return -EINVAL; /* unsigned long long to sector_t overflow */
2852
2853 *sectors = new;
2854 return 0;
2855}
2856
83303b61 2857static ssize_t
3cb03002 2858rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2859{
fd01b88c 2860 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2861 sector_t oldsectors = rdev->sectors;
b522adcd 2862 sector_t sectors;
27c529bb 2863
b522adcd 2864 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2865 return -EINVAL;
c6563a8c
N		 2866 if (rdev->data_offset != rdev->new_data_offset)
2867 return -EINVAL; /* too confusing */
0cd17fec 2868 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2869 if (my_mddev->persistent) {
dd8ac336
AN		 2870 sectors = super_types[my_mddev->major_version].
2871 rdev_size_change(rdev, sectors);
2872 if (!sectors)
0cd17fec 2873 return -EBUSY;
dd8ac336 2874 } else if (!sectors)
77304d2a 2875 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2876 rdev->data_offset;
a6468539
N		 2877 if (!my_mddev->pers->resize)
2878 /* Cannot change size for RAID0 or Linear etc */
2879 return -EINVAL;
0cd17fec 2880 }
dd8ac336 2881 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2882 return -EINVAL; /* component must fit device */
0cd17fec 2883
dd8ac336
AN		 2884 rdev->sectors = sectors;
2885 if (sectors > oldsectors && my_mddev->external) {
8b1afc3d
N		 2886 /* Need to check that all other rdevs with the same
2887 * ->bdev do not overlap. 'rcu' is sufficient to walk
2888 * the rdev lists safely.
2889 * This check does not provide a hard guarantee, it
2890 * just helps avoid dangerous mistakes.
c5d79adb 2891 */
fd01b88c 2892 struct mddev *mddev;
c5d79adb 2893 int overlap = 0;
159ec1fc 2894 struct list_head *tmp;
c5d79adb 2895
8b1afc3d 2896 rcu_read_lock();
29ac4aa3 2897 for_each_mddev(mddev, tmp) {
3cb03002 2898 struct md_rdev *rdev2;
c5d79adb 2899
dafb20fa 2900 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 2901 if (rdev->bdev == rdev2->bdev &&
2902 rdev != rdev2 &&
2903 overlaps(rdev->data_offset, rdev->sectors,
2904 rdev2->data_offset,
2905 rdev2->sectors)) {
c5d79adb
N		 2906 overlap = 1;
2907 break;
2908 }
c5d79adb
N		 2909 if (overlap) {
2910 mddev_put(mddev);
2911 break;
2912 }
2913 }
8b1afc3d 2914 rcu_read_unlock();
c5d79adb
N		 2915 if (overlap) {
2916 /* Someone else could have slipped in a size
2917 * change here, but doing so is just silly.
dd8ac336 2918 * We put oldsectors back because we *know* it is
c5d79adb
N		 2919 * safe, and trust userspace not to race with
2920 * itself
2921 */
dd8ac336 2922 rdev->sectors = oldsectors;
c5d79adb
N		 2923 return -EBUSY;
2924 }
2925 }
83303b61
N		 2926 return len;
2927}
2928
2929static struct rdev_sysfs_entry rdev_size =
80ca3a44 2930__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2931
3cb03002 2932static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 2933{
2934 unsigned long long recovery_start = rdev->recovery_offset;
2935
2936 if (test_bit(In_sync, &rdev->flags) ||
2937 recovery_start == MaxSector)
2938 return sprintf(page, "none\n");
2939
2940 return sprintf(page, "%llu\n", recovery_start);
2941}
2942
3cb03002 2943static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 2944{
2945 unsigned long long recovery_start;
2946
2947 if (cmd_match(buf, "none"))
2948 recovery_start = MaxSector;
b29bebd6 2949 else if (kstrtoull(buf, 10, &recovery_start))
06e3c817
DW		 2950 return -EINVAL;
2951
2952 if (rdev->mddev->pers &&
2953 rdev->raid_disk >= 0)
2954 return -EBUSY;
2955
2956 rdev->recovery_offset = recovery_start;
2957 if (recovery_start == MaxSector)
2958 set_bit(In_sync, &rdev->flags);
2959 else
2960 clear_bit(In_sync, &rdev->flags);
2961 return len;
2962}
2963
2964static struct rdev_sysfs_entry rdev_recovery_start =
2965__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2966
16c791a5
N		 2967static ssize_t
2968badblocks_show(struct badblocks *bb, char *page, int unack);
2969static ssize_t
2970badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2971
3cb03002 2972static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2973{
2974 return badblocks_show(&rdev->badblocks, page, 0);
2975}
3cb03002 2976static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 2977{
de393cde
N		 2978 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2979 /* Maybe that ack was all we needed */
2980 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2981 wake_up(&rdev->blocked_wait);
2982 return rv;
16c791a5
N		 2983}
2984static struct rdev_sysfs_entry rdev_bad_blocks =
2985__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2986
3cb03002 2987static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2988{
2989 return badblocks_show(&rdev->badblocks, page, 1);
2990}
3cb03002 2991static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 2992{
2993 return badblocks_store(&rdev->badblocks, page, len, 1);
2994}
2995static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2996__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2997
86e6ffdd
N		 2998static struct attribute *rdev_default_attrs[] = {
2999 &rdev_state.attr,
4dbcdc75 3000 &rdev_errors.attr,
014236d2 3001 &rdev_slot.attr,
93c8cad0 3002 &rdev_offset.attr,
c6563a8c 3003 &rdev_new_offset.attr,
83303b61 3004 &rdev_size.attr,
06e3c817 3005 &rdev_recovery_start.attr,
16c791a5
N		 3006 &rdev_bad_blocks.attr,
3007 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 3008 NULL,
3009};
3010static ssize_t
3011rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3012{
3013 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3014 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
86e6ffdd
N		 3015
3016 if (!entry->show)
3017 return -EIO;
758bfc8a
N		 3018 if (!rdev->mddev)
3019 return -EBUSY;
3020 return entry->show(rdev, page);
86e6ffdd
N		 3021}
3022
3023static ssize_t
3024rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3025 const char *page, size_t length)
3026{
3027 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3028 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3029 ssize_t rv;
fd01b88c 3030 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3031
3032 if (!entry->store)
3033 return -EIO;
67463acb
N		 3034 if (!capable(CAP_SYS_ADMIN))
3035 return -EACCES;
27c529bb 3036 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3037 if (!rv) {
27c529bb
N		 3038 if (rdev->mddev == NULL)
3039 rv = -EBUSY;
3040 else
3041 rv = entry->store(rdev, page, length);
6a51830e 3042 mddev_unlock(mddev);
ca388059
N		 3043 }
3044 return rv;
86e6ffdd
N		 3045}
3046
3047static void rdev_free(struct kobject *ko)
3048{
3cb03002 3049 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3050 kfree(rdev);
3051}
52cf25d0 3052static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3053 .show = rdev_attr_show,
3054 .store = rdev_attr_store,
3055};
3056static struct kobj_type rdev_ktype = {
3057 .release = rdev_free,
3058 .sysfs_ops = &rdev_sysfs_ops,
3059 .default_attrs = rdev_default_attrs,
3060};
3061
3cb03002 3062int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3063{
3064 rdev->desc_nr = -1;
3065 rdev->saved_raid_disk = -1;
3066 rdev->raid_disk = -1;
3067 rdev->flags = 0;
3068 rdev->data_offset = 0;
c6563a8c 3069 rdev->new_data_offset = 0;
e8bb9a83
N		 3070 rdev->sb_events = 0;
3071 rdev->last_read_error.tv_sec = 0;
3072 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3073 rdev->sb_loaded = 0;
3074 rdev->bb_page = NULL;
e8bb9a83
N		 3075 atomic_set(&rdev->nr_pending, 0);
3076 atomic_set(&rdev->read_errors, 0);
3077 atomic_set(&rdev->corrected_errors, 0);
3078
3079 INIT_LIST_HEAD(&rdev->same_set);
3080 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3081
3082 /* Add space to store bad block list.
3083 * This reserves the space even on arrays where it cannot
3084 * be used - I wonder if that matters
3085 */
3086 rdev->badblocks.count = 0;
486adf72 3087 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
2230dfe4
N		 3088 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3089 seqlock_init(&rdev->badblocks.lock);
3090 if (rdev->badblocks.page == NULL)
3091 return -ENOMEM;
3092
3093 return 0;
e8bb9a83
N		 3094}
3095EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3096/*
3097 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3098 *
3099 * mark the device faulty if:
3100 *
3101 * - the device is nonexistent (zero size)
3102 * - the device has no valid superblock
3103 *
3104 * a faulty rdev _never_ has rdev->sb set.
3105 */
3cb03002 3106static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3107{
3108 char b[BDEVNAME_SIZE];
3109 int err;
3cb03002 3110 struct md_rdev *rdev;
1da177e4
LT		 3111 sector_t size;
3112
9ffae0cf 3113 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3114 if (!rdev) {
3115 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3116 return ERR_PTR(-ENOMEM);
3117 }
1da177e4 3118
2230dfe4
N		 3119 err = md_rdev_init(rdev);
3120 if (err)
3121 goto abort_free;
3122 err = alloc_disk_sb(rdev);
3123 if (err)
1da177e4
LT		 3124 goto abort_free;
3125
c5d79adb 3126 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3127 if (err)
3128 goto abort_free;
3129
f9cb074b 3130 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3131
77304d2a 3132 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4 3133 if (!size) {
f72ffdd6 3134 printk(KERN_WARNING
1da177e4
LT		 3135 "md: %s has zero or unknown size, marking faulty!\n",
3136 bdevname(rdev->bdev,b));
3137 err = -EINVAL;
3138 goto abort_free;
3139 }
3140
3141 if (super_format >= 0) {
3142 err = super_types[super_format].
3143 load_super(rdev, NULL, super_minor);
3144 if (err == -EINVAL) {
df968c4e
N		 3145 printk(KERN_WARNING
3146 "md: %s does not have a valid v%d.%d "
3147 "superblock, not importing!\n",
3148 bdevname(rdev->bdev,b),
3149 super_format, super_minor);
1da177e4
LT		 3150 goto abort_free;
3151 }
3152 if (err < 0) {
f72ffdd6 3153 printk(KERN_WARNING
1da177e4
LT		 3154 "md: could not read %s's sb, not importing!\n",
3155 bdevname(rdev->bdev,b));
3156 goto abort_free;
3157 }
3158 }
6bfe0b49 3159
1da177e4
LT		 3160 return rdev;
3161
3162abort_free:
2699b672
N		 3163 if (rdev->bdev)
3164 unlock_rdev(rdev);
545c8795 3165 md_rdev_clear(rdev);
1da177e4
LT		 3166 kfree(rdev);
3167 return ERR_PTR(err);
3168}
3169
3170/*
3171 * Check a full RAID array for plausibility
3172 */
3173
f72ffdd6 3174static void analyze_sbs(struct mddev *mddev)
1da177e4
LT		 3175{
3176 int i;
3cb03002 3177 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3178 char b[BDEVNAME_SIZE];
3179
3180 freshest = NULL;
dafb20fa 3181 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3182 switch (super_types[mddev->major_version].
3183 load_super(rdev, freshest, mddev->minor_version)) {
3184 case 1:
3185 freshest = rdev;
3186 break;
3187 case 0:
3188 break;
3189 default:
3190 printk( KERN_ERR \
3191 "md: fatal superblock inconsistency in %s"
f72ffdd6 3192 " -- removing from array\n",
1da177e4 3193 bdevname(rdev->bdev,b));
fb56dfef 3194 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3195 }
3196
1da177e4
LT		 3197 super_types[mddev->major_version].
3198 validate_super(mddev, freshest);
3199
3200 i = 0;
dafb20fa 3201 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3202 if (mddev->max_disks &&
3203 (rdev->desc_nr >= mddev->max_disks ||
3204 i > mddev->max_disks)) {
de01dfad
N		 3205 printk(KERN_WARNING
3206 "md: %s: %s: only %d devices permitted\n",
3207 mdname(mddev), bdevname(rdev->bdev, b),
3208 mddev->max_disks);
fb56dfef 3209 md_kick_rdev_from_array(rdev);
de01dfad
N		 3210 continue;
3211 }
1aee41f6 3212 if (rdev != freshest) {
1da177e4
LT		 3213 if (super_types[mddev->major_version].
3214 validate_super(mddev, rdev)) {
3215 printk(KERN_WARNING "md: kicking non-fresh %s"
3216 " from array!\n",
3217 bdevname(rdev->bdev,b));
fb56dfef 3218 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3219 continue;
3220 }
1aee41f6
GR		 3221 /* No device should have a Candidate flag
3222 * when reading devices
3223 */
3224 if (test_bit(Candidate, &rdev->flags)) {
3225 pr_info("md: kicking Cluster Candidate %s from array!\n",
3226 bdevname(rdev->bdev, b));
fb56dfef 3227 md_kick_rdev_from_array(rdev);
1aee41f6
GR		 3228 }
3229 }
1da177e4
LT		 3230 if (mddev->level == LEVEL_MULTIPATH) {
3231 rdev->desc_nr = i++;
3232 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3233 set_bit(In_sync, &rdev->flags);
5e5e3e78 3234 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3235 rdev->raid_disk = -1;
3236 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3237 }
3238 }
1da177e4
LT		 3239}
3240
72e02075
N		 3241/* Read a fixed-point number.
3242 * Numbers in sysfs attributes should be in "standard" units where
3243 * possible, so time should be in seconds.
f72ffdd6 3244 * However we internally use a a much smaller unit such as
72e02075
N		 3245 * milliseconds or jiffies.
3246 * This function takes a decimal number with a possible fractional
3247 * component, and produces an integer which is the result of
3248 * multiplying that number by 10^'scale'.
3249 * all without any floating-point arithmetic.
3250 */
3251int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3252{
3253 unsigned long result = 0;
3254 long decimals = -1;
3255 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3256 if (*cp == '.')
3257 decimals = 0;
3258 else if (decimals < scale) {
3259 unsigned int value;
3260 value = *cp - '0';
3261 result = result * 10 + value;
3262 if (decimals >= 0)
3263 decimals++;
3264 }
3265 cp++;
3266 }
3267 if (*cp == '\n')
3268 cp++;
3269 if (*cp)
3270 return -EINVAL;
3271 if (decimals < 0)
3272 decimals = 0;
3273 while (decimals < scale) {
3274 result *= 10;
3275 decimals ++;
3276 }
3277 *res = result;
3278 return 0;
3279}
3280
19052c0e
N		 3281static void md_safemode_timeout(unsigned long data);
3282
16f17b39 3283static ssize_t
fd01b88c 3284safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3285{
3286 int msec = (mddev->safemode_delay*1000)/HZ;
3287 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3288}
3289static ssize_t
fd01b88c 3290safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3291{
16f17b39 3292 unsigned long msec;
97ce0a7f 3293
72e02075 3294 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3295 return -EINVAL;
16f17b39
N		 3296 if (msec == 0)
3297 mddev->safemode_delay = 0;
3298 else {
19052c0e 3299 unsigned long old_delay = mddev->safemode_delay;
1b30e66f
N		 3300 unsigned long new_delay = (msec*HZ)/1000;
3301
3302 if (new_delay == 0)
3303 new_delay = 1;
3304 mddev->safemode_delay = new_delay;
3305 if (new_delay < old_delay || old_delay == 0)
3306 mod_timer(&mddev->safemode_timer, jiffies+1);
16f17b39
N		 3307 }
3308 return len;
3309}
3310static struct md_sysfs_entry md_safe_delay =
80ca3a44 3311__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3312
eae1701f 3313static ssize_t
fd01b88c 3314level_show(struct mddev *mddev, char *page)
eae1701f 3315{
36d091f4
N		 3316 struct md_personality *p;
3317 int ret;
3318 spin_lock(&mddev->lock);
3319 p = mddev->pers;
d9d166c2 3320 if (p)
36d091f4 3321 ret = sprintf(page, "%s\n", p->name);
d9d166c2 3322 else if (mddev->clevel[0])
36d091f4 3323 ret = sprintf(page, "%s\n", mddev->clevel);
d9d166c2 3324 else if (mddev->level != LEVEL_NONE)
36d091f4 3325 ret = sprintf(page, "%d\n", mddev->level);
d9d166c2 3326 else
36d091f4
N		 3327 ret = 0;
3328 spin_unlock(&mddev->lock);
3329 return ret;
eae1701f
N		 3330}
3331
d9d166c2 3332static ssize_t
fd01b88c 3333level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3334{
f2859af6 3335 char clevel[16];
6791875e
N		 3336 ssize_t rv;
3337 size_t slen = len;
db721d32 3338 struct md_personality *pers, *oldpers;
f2859af6 3339 long level;
db721d32 3340 void *priv, *oldpriv;
3cb03002 3341 struct md_rdev *rdev;
245f46c2 3342
6791875e
N		 3343 if (slen == 0 || slen >= sizeof(clevel))
3344 return -EINVAL;
3345
3346 rv = mddev_lock(mddev);
3347 if (rv)
3348 return rv;
3349
245f46c2 3350 if (mddev->pers == NULL) {
6791875e
N		 3351 strncpy(mddev->clevel, buf, slen);
3352 if (mddev->clevel[slen-1] == '\n')
3353 slen--;
3354 mddev->clevel[slen] = 0;
245f46c2 3355 mddev->level = LEVEL_NONE;
6791875e
N		 3356 rv = len;
3357 goto out_unlock;
245f46c2 3358 }
6791875e 3359 rv = -EROFS;
bd8839e0 3360 if (mddev->ro)
6791875e 3361 goto out_unlock;
245f46c2
N		 3362
3363 /* request to change the personality. Need to ensure:
3364 * - array is not engaged in resync/recovery/reshape
3365 * - old personality can be suspended
3366 * - new personality will access other array.
3367 */
3368
6791875e 3369 rv = -EBUSY;
bb4f1e9d 3370 if (mddev->sync_thread ||
f851b60d 3371 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
bb4f1e9d
N		 3372 mddev->reshape_position != MaxSector ||
3373 mddev->sysfs_active)
6791875e 3374 goto out_unlock;
245f46c2 3375
6791875e 3376 rv = -EINVAL;
245f46c2
N		 3377 if (!mddev->pers->quiesce) {
3378 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3379 mdname(mddev), mddev->pers->name);
6791875e 3380 goto out_unlock;
245f46c2
N		 3381 }
3382
3383 /* Now find the new personality */
6791875e
N		 3384 strncpy(clevel, buf, slen);
3385 if (clevel[slen-1] == '\n')
3386 slen--;
3387 clevel[slen] = 0;
b29bebd6 3388 if (kstrtol(clevel, 10, &level))
f2859af6 3389 level = LEVEL_NONE;
245f46c2 3390
f2859af6
DW		 3391 if (request_module("md-%s", clevel) != 0)
3392 request_module("md-level-%s", clevel);
245f46c2 3393 spin_lock(&pers_lock);
f2859af6 3394 pers = find_pers(level, clevel);
245f46c2
N		 3395 if (!pers || !try_module_get(pers->owner)) {
3396 spin_unlock(&pers_lock);
f2859af6 3397 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
6791875e
N		 3398 rv = -EINVAL;
3399 goto out_unlock;
245f46c2
N		 3400 }
3401 spin_unlock(&pers_lock);
3402
3403 if (pers == mddev->pers) {
3404 /* Nothing to do! */
3405 module_put(pers->owner);
6791875e
N		 3406 rv = len;
3407 goto out_unlock;
245f46c2
N		 3408 }
3409 if (!pers->takeover) {
3410 module_put(pers->owner);
3411 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3412 mdname(mddev), clevel);
6791875e
N		 3413 rv = -EINVAL;
3414 goto out_unlock;
245f46c2
N		 3415 }
3416
dafb20fa 3417 rdev_for_each(rdev, mddev)
e93f68a1
N		 3418 rdev->new_raid_disk = rdev->raid_disk;
3419
245f46c2
N		 3420 /* ->takeover must set new_* and/or delta_disks
3421 * if it succeeds, and may set them when it fails.
3422 */
3423 priv = pers->takeover(mddev);
3424 if (IS_ERR(priv)) {
3425 mddev->new_level = mddev->level;
3426 mddev->new_layout = mddev->layout;
664e7c41 3427 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3428 mddev->raid_disks -= mddev->delta_disks;
3429 mddev->delta_disks = 0;
2c810cdd 3430 mddev->reshape_backwards = 0;
245f46c2
N		 3431 module_put(pers->owner);
3432 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3433 mdname(mddev), clevel);
6791875e
N		 3434 rv = PTR_ERR(priv);
3435 goto out_unlock;
245f46c2
N		 3436 }
3437
3438 /* Looks like we have a winner */
3439 mddev_suspend(mddev);
5aa61f42 3440 mddev_detach(mddev);
36d091f4
N		 3441
3442 spin_lock(&mddev->lock);
db721d32
N		 3443 oldpers = mddev->pers;
3444 oldpriv = mddev->private;
3445 mddev->pers = pers;
3446 mddev->private = priv;
3447 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3448 mddev->level = mddev->new_level;
3449 mddev->layout = mddev->new_layout;
3450 mddev->chunk_sectors = mddev->new_chunk_sectors;
3451 mddev->delta_disks = 0;
3452 mddev->reshape_backwards = 0;
3453 mddev->degraded = 0;
36d091f4 3454 spin_unlock(&mddev->lock);
db721d32
N		 3455
3456 if (oldpers->sync_request == NULL &&
3457 mddev->external) {
3458 /* We are converting from a no-redundancy array
3459 * to a redundancy array and metadata is managed
3460 * externally so we need to be sure that writes
3461 * won't block due to a need to transition
3462 * clean->dirty
3463 * until external management is started.
3464 */
3465 mddev->in_sync = 0;
3466 mddev->safemode_delay = 0;
3467 mddev->safemode = 0;
3468 }
f72ffdd6 3469
db721d32
N		 3470 oldpers->free(mddev, oldpriv);
3471
3472 if (oldpers->sync_request == NULL &&
a64c876f
N		 3473 pers->sync_request != NULL) {
3474 /* need to add the md_redundancy_group */
3475 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3476 printk(KERN_WARNING
3477 "md: cannot register extra attributes for %s\n",
3478 mdname(mddev));
388975cc 3479 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
f72ffdd6 3480 }
db721d32 3481 if (oldpers->sync_request != NULL &&
a64c876f
N		 3482 pers->sync_request == NULL) {
3483 /* need to remove the md_redundancy_group */
3484 if (mddev->to_remove == NULL)
3485 mddev->to_remove = &md_redundancy_group;
3486 }
3487
dafb20fa 3488 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3489 if (rdev->raid_disk < 0)
3490 continue;
bf2cb0da 3491 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3492 rdev->new_raid_disk = -1;
3493 if (rdev->new_raid_disk == rdev->raid_disk)
3494 continue;
36fad858 3495 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3496 }
dafb20fa 3497 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3498 if (rdev->raid_disk < 0)
3499 continue;
3500 if (rdev->new_raid_disk == rdev->raid_disk)
3501 continue;
3502 rdev->raid_disk = rdev->new_raid_disk;
3503 if (rdev->raid_disk < 0)
3a981b03 3504 clear_bit(In_sync, &rdev->flags);
e93f68a1 3505 else {
36fad858
NK		 3506 if (sysfs_link_rdev(mddev, rdev))
3507 printk(KERN_WARNING "md: cannot register rd%d"
3508 " for %s after level change\n",
3509 rdev->raid_disk, mdname(mddev));
3a981b03 3510 }
e93f68a1
N		 3511 }
3512
db721d32 3513 if (pers->sync_request == NULL) {
9af204cf
TM		 3514 /* this is now an array without redundancy, so
3515 * it must always be in_sync
3516 */
3517 mddev->in_sync = 1;
3518 del_timer_sync(&mddev->safemode_timer);
3519 }
02e5f5c0 3520 blk_set_stacking_limits(&mddev->queue->limits);
245f46c2 3521 pers->run(mddev);
245f46c2 3522 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3523 mddev_resume(mddev);
830778a1
N		 3524 if (!mddev->thread)
3525 md_update_sb(mddev, 1);
5cac7861 3526 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3527 md_new_event(mddev);
6791875e
N		 3528 rv = len;
3529out_unlock:
3530 mddev_unlock(mddev);
d9d166c2
N		 3531 return rv;
3532}
3533
3534static struct md_sysfs_entry md_level =
80ca3a44 3535__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3536
d4dbd025 3537static ssize_t
fd01b88c 3538layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3539{
3540 /* just a number, not meaningful for all levels */
08a02ecd
N		 3541 if (mddev->reshape_position != MaxSector &&
3542 mddev->layout != mddev->new_layout)
3543 return sprintf(page, "%d (%d)\n",
3544 mddev->new_layout, mddev->layout);
d4dbd025
N		 3545 return sprintf(page, "%d\n", mddev->layout);
3546}
3547
3548static ssize_t
fd01b88c 3549layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025 3550{
4c9309c0 3551 unsigned int n;
6791875e 3552 int err;
d4dbd025 3553
4c9309c0
AD		 3554 err = kstrtouint(buf, 10, &n);
3555 if (err < 0)
3556 return err;
6791875e
N		 3557 err = mddev_lock(mddev);
3558 if (err)
3559 return err;
d4dbd025 3560
b3546035 3561 if (mddev->pers) {
50ac168a 3562 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3563 err = -EBUSY;
3564 else if (mddev->ro)
3565 err = -EROFS;
3566 else {
3567 mddev->new_layout = n;
3568 err = mddev->pers->check_reshape(mddev);
3569 if (err)
3570 mddev->new_layout = mddev->layout;
597a711b 3571 }
b3546035 3572 } else {
08a02ecd 3573 mddev->new_layout = n;
b3546035
N		 3574 if (mddev->reshape_position == MaxSector)
3575 mddev->layout = n;
3576 }
6791875e
N		 3577 mddev_unlock(mddev);
3578 return err ?: len;
d4dbd025
N		 3579}
3580static struct md_sysfs_entry md_layout =
80ca3a44 3581__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025 3582
eae1701f 3583static ssize_t
fd01b88c 3584raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3585{
bb636547
N		 3586 if (mddev->raid_disks == 0)
3587 return 0;
08a02ecd
N		 3588 if (mddev->reshape_position != MaxSector &&
3589 mddev->delta_disks != 0)
3590 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3591 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3592 return sprintf(page, "%d\n", mddev->raid_disks);
3593}
3594
fd01b88c 3595static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3596
3597static ssize_t
fd01b88c 3598raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3599{
4c9309c0 3600 unsigned int n;
6791875e 3601 int err;
da943b99 3602
4c9309c0
AD		 3603 err = kstrtouint(buf, 10, &n);
3604 if (err < 0)
3605 return err;
da943b99 3606
6791875e
N		 3607 err = mddev_lock(mddev);
3608 if (err)
3609 return err;
da943b99 3610 if (mddev->pers)
6791875e 3611 err = update_raid_disks(mddev, n);
08a02ecd 3612 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3613 struct md_rdev *rdev;
08a02ecd 3614 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c 3615
6791875e 3616 err = -EINVAL;
c6563a8c
N		 3617 rdev_for_each(rdev, mddev) {
3618 if (olddisks < n &&
3619 rdev->data_offset < rdev->new_data_offset)
6791875e 3620 goto out_unlock;
c6563a8c
N		 3621 if (olddisks > n &&
3622 rdev->data_offset > rdev->new_data_offset)
6791875e 3623 goto out_unlock;
c6563a8c 3624 }
6791875e 3625 err = 0;
08a02ecd
N		 3626 mddev->delta_disks = n - olddisks;
3627 mddev->raid_disks = n;
2c810cdd 3628 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3629 } else
da943b99 3630 mddev->raid_disks = n;
6791875e
N		 3631out_unlock:
3632 mddev_unlock(mddev);
3633 return err ? err : len;
da943b99
N		 3634}
3635static struct md_sysfs_entry md_raid_disks =
80ca3a44 3636__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3637
3b34380a 3638static ssize_t
fd01b88c 3639chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3640{
08a02ecd 3641 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3642 mddev->chunk_sectors != mddev->new_chunk_sectors)
3643 return sprintf(page, "%d (%d)\n",
3644 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3645 mddev->chunk_sectors << 9);
3646 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3647}
3648
3649static ssize_t
fd01b88c 3650chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3651{
4c9309c0 3652 unsigned long n;
6791875e 3653 int err;
3b34380a 3654
4c9309c0
AD		 3655 err = kstrtoul(buf, 10, &n);
3656 if (err < 0)
3657 return err;
3b34380a 3658
6791875e
N		 3659 err = mddev_lock(mddev);
3660 if (err)
3661 return err;
b3546035 3662 if (mddev->pers) {
50ac168a 3663 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3664 err = -EBUSY;
3665 else if (mddev->ro)
3666 err = -EROFS;
3667 else {
3668 mddev->new_chunk_sectors = n >> 9;
3669 err = mddev->pers->check_reshape(mddev);
3670 if (err)
3671 mddev->new_chunk_sectors = mddev->chunk_sectors;
597a711b 3672 }
b3546035 3673 } else {
664e7c41 3674 mddev->new_chunk_sectors = n >> 9;
b3546035 3675 if (mddev->reshape_position == MaxSector)
9d8f0363 3676 mddev->chunk_sectors = n >> 9;
b3546035 3677 }
6791875e
N		 3678 mddev_unlock(mddev);
3679 return err ?: len;
3b34380a
N		 3680}
3681static struct md_sysfs_entry md_chunk_size =
80ca3a44 3682__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3683
a94213b1 3684static ssize_t
fd01b88c 3685resync_start_show(struct mddev *mddev, char *page)
a94213b1 3686{
d1a7c503
N		 3687 if (mddev->recovery_cp == MaxSector)
3688 return sprintf(page, "none\n");
a94213b1
N		 3689 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3690}
3691
3692static ssize_t
fd01b88c 3693resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3694{
4c9309c0 3695 unsigned long long n;
6791875e 3696 int err;
4c9309c0
AD		 3697
3698 if (cmd_match(buf, "none"))
3699 n = MaxSector;
3700 else {
3701 err = kstrtoull(buf, 10, &n);
3702 if (err < 0)
3703 return err;
3704 if (n != (sector_t)n)
3705 return -EINVAL;
3706 }
a94213b1 3707
6791875e
N		 3708 err = mddev_lock(mddev);
3709 if (err)
3710 return err;
b098636c 3711 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6791875e 3712 err = -EBUSY;
a94213b1 3713
6791875e
N		 3714 if (!err) {
3715 mddev->recovery_cp = n;
3716 if (mddev->pers)
3717 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3718 }
3719 mddev_unlock(mddev);
3720 return err ?: len;
a94213b1
N		 3721}
3722static struct md_sysfs_entry md_resync_start =
750f199e
N		 3723__ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3724 resync_start_show, resync_start_store);
a94213b1 3725
9e653b63
N		 3726/*
3727 * The array state can be:
3728 *
3729 * clear
3730 * No devices, no size, no level
3731 * Equivalent to STOP_ARRAY ioctl
3732 * inactive
3733 * May have some settings, but array is not active
3734 * all IO results in error
3735 * When written, doesn't tear down array, but just stops it
3736 * suspended (not supported yet)
3737 * All IO requests will block. The array can be reconfigured.
910d8cb3 3738 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3739 * readonly
3740 * no resync can happen. no superblocks get written.
3741 * write requests fail
3742 * read-auto
3743 * like readonly, but behaves like 'clean' on a write request.
3744 *
3745 * clean - no pending writes, but otherwise active.
3746 * When written to inactive array, starts without resync
3747 * If a write request arrives then
3748 * if metadata is known, mark 'dirty' and switch to 'active'.
3749 * if not known, block and switch to write-pending
3750 * If written to an active array that has pending writes, then fails.
3751 * active
3752 * fully active: IO and resync can be happening.
3753 * When written to inactive array, starts with resync
3754 *
3755 * write-pending
3756 * clean, but writes are blocked waiting for 'active' to be written.
3757 *
3758 * active-idle
3759 * like active, but no writes have been seen for a while (100msec).
3760 *
3761 */
3762enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3763 write_pending, active_idle, bad_word};
05381954 3764static char *array_states[] = {
9e653b63
N		 3765 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3766 "write-pending", "active-idle", NULL };
3767
3768static int match_word(const char *word, char **list)
3769{
3770 int n;
3771 for (n=0; list[n]; n++)
3772 if (cmd_match(word, list[n]))
3773 break;
3774 return n;
3775}
3776
3777static ssize_t
fd01b88c 3778array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3779{
3780 enum array_state st = inactive;
3781
3782 if (mddev->pers)
3783 switch(mddev->ro) {
3784 case 1:
3785 st = readonly;
3786 break;
3787 case 2:
3788 st = read_auto;
3789 break;
3790 case 0:
3791 if (mddev->in_sync)
3792 st = clean;
070dc6dd 3793 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3794 st = write_pending;
9e653b63
N		 3795 else if (mddev->safemode)
3796 st = active_idle;
3797 else
3798 st = active;
3799 }
3800 else {
3801 if (list_empty(&mddev->disks) &&
3802 mddev->raid_disks == 0 &&
58c0fed4 3803 mddev->dev_sectors == 0)
9e653b63
N		 3804 st = clear;
3805 else
3806 st = inactive;
3807 }
3808 return sprintf(page, "%s\n", array_states[st]);
3809}
3810
f72ffdd6
N		 3811static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3812static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3813static int do_md_run(struct mddev *mddev);
fd01b88c 3814static int restart_array(struct mddev *mddev);
9e653b63
N		 3815
3816static ssize_t
fd01b88c 3817array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63 3818{
6791875e 3819 int err;
9e653b63 3820 enum array_state st = match_word(buf, array_states);
6791875e
N		 3821
3822 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3823 /* don't take reconfig_mutex when toggling between
3824 * clean and active
3825 */
3826 spin_lock(&mddev->lock);
3827 if (st == active) {
3828 restart_array(mddev);
3829 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3830 wake_up(&mddev->sb_wait);
3831 err = 0;
3832 } else /* st == clean */ {
3833 restart_array(mddev);
3834 if (atomic_read(&mddev->writes_pending) == 0) {
3835 if (mddev->in_sync == 0) {
3836 mddev->in_sync = 1;
3837 if (mddev->safemode == 1)
3838 mddev->safemode = 0;
3839 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3840 }
3841 err = 0;
3842 } else
3843 err = -EBUSY;
3844 }
3845 spin_unlock(&mddev->lock);
c008f1d3 3846 return err ?: len;
6791875e
N		 3847 }
3848 err = mddev_lock(mddev);
3849 if (err)
3850 return err;
3851 err = -EINVAL;
9e653b63
N		 3852 switch(st) {
3853 case bad_word:
3854 break;
3855 case clear:
3856 /* stopping an active array */
a05b7ea0 3857 err = do_md_stop(mddev, 0, NULL);
9e653b63
N		 3858 break;
3859 case inactive:
3860 /* stopping an active array */
90cf195d 3861 if (mddev->pers)
a05b7ea0 3862 err = do_md_stop(mddev, 2, NULL);
90cf195d 3863 else
e691063a 3864 err = 0; /* already inactive */
9e653b63
N		 3865 break;
3866 case suspended:
3867 break; /* not supported yet */
3868 case readonly:
3869 if (mddev->pers)
a05b7ea0 3870 err = md_set_readonly(mddev, NULL);
9e653b63
N		 3871 else {
3872 mddev->ro = 1;
648b629e 3873 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3874 err = do_md_run(mddev);
3875 }
3876 break;
3877 case read_auto:
9e653b63 3878 if (mddev->pers) {
80268ee9 3879 if (mddev->ro == 0)
a05b7ea0 3880 err = md_set_readonly(mddev, NULL);
80268ee9 3881 else if (mddev->ro == 1)
648b629e
N		 3882 err = restart_array(mddev);
3883 if (err == 0) {
3884 mddev->ro = 2;
3885 set_disk_ro(mddev->gendisk, 0);
3886 }
9e653b63
N		 3887 } else {
3888 mddev->ro = 2;
3889 err = do_md_run(mddev);
3890 }
3891 break;
3892 case clean:
3893 if (mddev->pers) {
3894 restart_array(mddev);
85572d7c 3895 spin_lock(&mddev->lock);
9e653b63 3896 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3897 if (mddev->in_sync == 0) {
3898 mddev->in_sync = 1;
31a59e34
N		 3899 if (mddev->safemode == 1)
3900 mddev->safemode = 0;
070dc6dd 3901 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3902 }
3903 err = 0;
3904 } else
3905 err = -EBUSY;
85572d7c 3906 spin_unlock(&mddev->lock);
5bf29597
N		 3907 } else
3908 err = -EINVAL;
9e653b63
N		 3909 break;
3910 case active:
3911 if (mddev->pers) {
3912 restart_array(mddev);
070dc6dd 3913 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3914 wake_up(&mddev->sb_wait);
3915 err = 0;
3916 } else {
3917 mddev->ro = 0;
648b629e 3918 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3919 err = do_md_run(mddev);
3920 }
3921 break;
3922 case write_pending:
3923 case active_idle:
3924 /* these cannot be set */
3925 break;
3926 }
6791875e
N		 3927
3928 if (!err) {
1d23f178
N		 3929 if (mddev->hold_active == UNTIL_IOCTL)
3930 mddev->hold_active = 0;
00bcb4ac 3931 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 3932 }
6791875e
N		 3933 mddev_unlock(mddev);
3934 return err ?: len;
9e653b63 3935}
80ca3a44 3936static struct md_sysfs_entry md_array_state =
750f199e 3937__ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3938
1e50915f 3939static ssize_t
fd01b88c 3940max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3941 return sprintf(page, "%d\n",
3942 atomic_read(&mddev->max_corr_read_errors));
3943}
3944
3945static ssize_t
fd01b88c 3946max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f 3947{
4c9309c0
AD		 3948 unsigned int n;
3949 int rv;
1e50915f 3950
4c9309c0
AD		 3951 rv = kstrtouint(buf, 10, &n);
3952 if (rv < 0)
3953 return rv;
3954 atomic_set(&mddev->max_corr_read_errors, n);
3955 return len;
1e50915f
RB		 3956}
3957
3958static struct md_sysfs_entry max_corr_read_errors =
3959__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3960 max_corrected_read_errors_store);
3961
6d7ff738 3962static ssize_t
fd01b88c 3963null_show(struct mddev *mddev, char *page)
6d7ff738
N		 3964{
3965 return -EINVAL;
3966}
3967
3968static ssize_t
fd01b88c 3969new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 3970{
3971 /* buf must be %d:%d\n? giving major and minor numbers */
3972 /* The new device is added to the array.
3973 * If the array has a persistent superblock, we read the
3974 * superblock to initialise info and check validity.
3975 * Otherwise, only checking done is that in bind_rdev_to_array,
3976 * which mainly checks size.
3977 */
3978 char *e;
3979 int major = simple_strtoul(buf, &e, 10);
3980 int minor;
3981 dev_t dev;
3cb03002 3982 struct md_rdev *rdev;
6d7ff738
N		 3983 int err;
3984
3985 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3986 return -EINVAL;
3987 minor = simple_strtoul(e+1, &e, 10);
3988 if (*e && *e != '\n')
3989 return -EINVAL;
3990 dev = MKDEV(major, minor);
3991 if (major != MAJOR(dev) ||
3992 minor != MINOR(dev))
3993 return -EOVERFLOW;
3994
6791875e
N		 3995 flush_workqueue(md_misc_wq);
3996
3997 err = mddev_lock(mddev);
3998 if (err)
3999 return err;
6d7ff738
N		 4000 if (mddev->persistent) {
4001 rdev = md_import_device(dev, mddev->major_version,
4002 mddev->minor_version);
4003 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 4004 struct md_rdev *rdev0
4005 = list_entry(mddev->disks.next,
4006 struct md_rdev, same_set);
6d7ff738
N		 4007 err = super_types[mddev->major_version]
4008 .load_super(rdev, rdev0, mddev->minor_version);
4009 if (err < 0)
4010 goto out;
4011 }
c5d79adb
N		 4012 } else if (mddev->external)
4013 rdev = md_import_device(dev, -2, -1);
4014 else
6d7ff738
N		 4015 rdev = md_import_device(dev, -1, -1);
4016
4017 if (IS_ERR(rdev))
4018 return PTR_ERR(rdev);
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);
36d091f4
N		 5385 spin_lock(&mddev->lock);
5386 mddev->ready = 0;
6177b472 5387 mddev->pers = NULL;
36d091f4
N		 5388 spin_unlock(&mddev->lock);
5389 pers->free(mddev, mddev->private);
bd691922 5390 mddev->private = NULL;
36d091f4
N		 5391 if (pers->sync_request && mddev->to_remove == NULL)
5392 mddev->to_remove = &md_redundancy_group;
5393 module_put(pers->owner);
cca9cf90 5394 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5395}
5eff3c43
N		 5396
5397void md_stop(struct mddev *mddev)
5398{
5399 /* stop the array and free an attached data structures.
5400 * This is called from dm-raid
5401 */
5402 __md_stop(mddev);
5403 bitmap_destroy(mddev);
5404 if (mddev->bio_set)
5405 bioset_free(mddev->bio_set);
5406}
5407
390ee602 5408EXPORT_SYMBOL_GPL(md_stop);
6177b472 5409
a05b7ea0 5410static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N		 5411{
5412 int err = 0;
30b8feb7
N		 5413 int did_freeze = 0;
5414
5415 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5416 did_freeze = 1;
5417 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5418 md_wakeup_thread(mddev->thread);
5419 }
f851b60d 5420 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5421 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5422 if (mddev->sync_thread)
30b8feb7
N		 5423 /* Thread might be blocked waiting for metadata update
5424 * which will now never happen */
5425 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5426
30b8feb7 5427 mddev_unlock(mddev);
f851b60d
N		 5428 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5429 &mddev->recovery));
30b8feb7
N		 5430 mddev_lock_nointr(mddev);
5431
a4bd82d0 5432 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5433 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7 5434 mddev->sync_thread ||
f851b60d 5435 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5436 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
a4bd82d0 5437 printk("md: %s still in use.\n",mdname(mddev));
30b8feb7
N		 5438 if (did_freeze) {
5439 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5440 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5441 md_wakeup_thread(mddev->thread);
5442 }
a4bd82d0
N		 5443 err = -EBUSY;
5444 goto out;
5445 }
5446 if (mddev->pers) {
defad61a 5447 __md_stop_writes(mddev);
a4bd82d0
N		 5448
5449 err = -ENXIO;
5450 if (mddev->ro==1)
5451 goto out;
5452 mddev->ro = 1;
5453 set_disk_ro(mddev->gendisk, 1);
5454 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d
N		 5455 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5456 md_wakeup_thread(mddev->thread);
00bcb4ac 5457 sysfs_notify_dirent_safe(mddev->sysfs_state);
30b8feb7 5458 err = 0;
a4bd82d0
N		 5459 }
5460out:
5461 mutex_unlock(&mddev->open_mutex);
5462 return err;
5463}
5464
9e653b63
N		 5465/* mode:
5466 * 0 - completely stop and dis-assemble array
9e653b63
N		 5467 * 2 - stop but do not disassemble array
5468 */
f72ffdd6 5469static int do_md_stop(struct mddev *mddev, int mode,
a05b7ea0 5470 struct block_device *bdev)
1da177e4 5471{
1da177e4 5472 struct gendisk *disk = mddev->gendisk;
3cb03002 5473 struct md_rdev *rdev;
30b8feb7
N		 5474 int did_freeze = 0;
5475
5476 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5477 did_freeze = 1;
5478 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5479 md_wakeup_thread(mddev->thread);
5480 }
f851b60d 5481 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5482 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5483 if (mddev->sync_thread)
30b8feb7
N		 5484 /* Thread might be blocked waiting for metadata update
5485 * which will now never happen */
5486 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5487
30b8feb7 5488 mddev_unlock(mddev);
f851b60d
N		 5489 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5490 !test_bit(MD_RECOVERY_RUNNING,
5491 &mddev->recovery)));
30b8feb7 5492 mddev_lock_nointr(mddev);
1da177e4 5493
c8c00a69 5494 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5495 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7
N		 5496 mddev->sysfs_active ||
5497 mddev->sync_thread ||
f851b60d 5498 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5499 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
df5b20cf 5500 printk("md: %s still in use.\n",mdname(mddev));
6e17b027 5501 mutex_unlock(&mddev->open_mutex);
30b8feb7
N		 5502 if (did_freeze) {
5503 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5504 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5505 md_wakeup_thread(mddev->thread);
5506 }
260fa034
N		 5507 return -EBUSY;
5508 }
6e17b027 5509 if (mddev->pers) {
a4bd82d0
N		 5510 if (mddev->ro)
5511 set_disk_ro(disk, 0);
409c57f3 5512
defad61a 5513 __md_stop_writes(mddev);
5eff3c43 5514 __md_stop(mddev);
a4bd82d0 5515 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5516 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5517
a4bd82d0 5518 /* tell userspace to handle 'inactive' */
00bcb4ac 5519 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5520
dafb20fa 5521 rdev_for_each(rdev, mddev)
36fad858
NK		 5522 if (rdev->raid_disk >= 0)
5523 sysfs_unlink_rdev(mddev, rdev);
c4647292 5524
a4bd82d0 5525 set_capacity(disk, 0);
6e17b027 5526 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5527 mddev->changed = 1;
a4bd82d0 5528 revalidate_disk(disk);
0d4ca600 5529
a4bd82d0
N		 5530 if (mddev->ro)
5531 mddev->ro = 0;
6e17b027
N		 5532 } else
5533 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5534 /*
5535 * Free resources if final stop
5536 */
9e653b63 5537 if (mode == 0) {
1da177e4
LT		 5538 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5539
978f946b 5540 bitmap_destroy(mddev);
c3d9714e 5541 if (mddev->bitmap_info.file) {
4af1a041
N		 5542 struct file *f = mddev->bitmap_info.file;
5543 spin_lock(&mddev->lock);
c3d9714e 5544 mddev->bitmap_info.file = NULL;
4af1a041
N		 5545 spin_unlock(&mddev->lock);
5546 fput(f);
978f946b 5547 }
c3d9714e 5548 mddev->bitmap_info.offset = 0;
978f946b 5549
1da177e4
LT		 5550 export_array(mddev);
5551
6177b472 5552 md_clean(mddev);
934d9c23 5553 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5554 if (mddev->hold_active == UNTIL_STOP)
5555 mddev->hold_active = 0;
a4bd82d0 5556 }
3f9d99c1 5557 blk_integrity_unregister(disk);
d7603b7e 5558 md_new_event(mddev);
00bcb4ac 5559 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5560 return 0;
1da177e4
LT		 5561}
5562
fdee8ae4 5563#ifndef MODULE
fd01b88c 5564static void autorun_array(struct mddev *mddev)
1da177e4 5565{
3cb03002 5566 struct md_rdev *rdev;
1da177e4
LT		 5567 int err;
5568
a757e64c 5569 if (list_empty(&mddev->disks))
1da177e4 5570 return;
1da177e4
LT		 5571
5572 printk(KERN_INFO "md: running: ");
5573
dafb20fa 5574 rdev_for_each(rdev, mddev) {
1da177e4
LT		 5575 char b[BDEVNAME_SIZE];
5576 printk("<%s>", bdevname(rdev->bdev,b));
5577 }
5578 printk("\n");
5579
d710e138 5580 err = do_md_run(mddev);
1da177e4
LT		 5581 if (err) {
5582 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
a05b7ea0 5583 do_md_stop(mddev, 0, NULL);
1da177e4
LT		 5584 }
5585}
5586
5587/*
5588 * lets try to run arrays based on all disks that have arrived
5589 * until now. (those are in pending_raid_disks)
5590 *
5591 * the method: pick the first pending disk, collect all disks with
5592 * the same UUID, remove all from the pending list and put them into
5593 * the 'same_array' list. Then order this list based on superblock
5594 * update time (freshest comes first), kick out 'old' disks and
5595 * compare superblocks. If everything's fine then run it.
5596 *
5597 * If "unit" is allocated, then bump its reference count
5598 */
5599static void autorun_devices(int part)
5600{
3cb03002 5601 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5602 struct mddev *mddev;
1da177e4
LT		 5603 char b[BDEVNAME_SIZE];
5604
5605 printk(KERN_INFO "md: autorun ...\n");
5606 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5607 int unit;
1da177e4 5608 dev_t dev;
ad01c9e3 5609 LIST_HEAD(candidates);
1da177e4 5610 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5611 struct md_rdev, same_set);
1da177e4
LT		 5612
5613 printk(KERN_INFO "md: considering %s ...\n",
5614 bdevname(rdev0->bdev,b));
5615 INIT_LIST_HEAD(&candidates);
159ec1fc 5616 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5617 if (super_90_load(rdev, rdev0, 0) >= 0) {
5618 printk(KERN_INFO "md: adding %s ...\n",
5619 bdevname(rdev->bdev,b));
5620 list_move(&rdev->same_set, &candidates);
5621 }
5622 /*
5623 * now we have a set of devices, with all of them having
5624 * mostly sane superblocks. It's time to allocate the
5625 * mddev.
5626 */
e8703fe1
N		 5627 if (part) {
5628 dev = MKDEV(mdp_major,
5629 rdev0->preferred_minor << MdpMinorShift);
5630 unit = MINOR(dev) >> MdpMinorShift;
5631 } else {
5632 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5633 unit = MINOR(dev);
5634 }
5635 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5636 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5637 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5638 break;
5639 }
1da177e4
LT		 5640
5641 md_probe(dev, NULL, NULL);
5642 mddev = mddev_find(dev);
9bbbca3a
NB		 5643 if (!mddev || !mddev->gendisk) {
5644 if (mddev)
5645 mddev_put(mddev);
5646 printk(KERN_ERR
1da177e4
LT		 5647 "md: cannot allocate memory for md drive.\n");
5648 break;
5649 }
f72ffdd6 5650 if (mddev_lock(mddev))
1da177e4
LT		 5651 printk(KERN_WARNING "md: %s locked, cannot run\n",
5652 mdname(mddev));
5653 else if (mddev->raid_disks || mddev->major_version
5654 || !list_empty(&mddev->disks)) {
f72ffdd6 5655 printk(KERN_WARNING
1da177e4
LT		 5656 "md: %s already running, cannot run %s\n",
5657 mdname(mddev), bdevname(rdev0->bdev,b));
5658 mddev_unlock(mddev);
5659 } else {
5660 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5661 mddev->persistent = 1;
159ec1fc 5662 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5663 list_del_init(&rdev->same_set);
5664 if (bind_rdev_to_array(rdev, mddev))
5665 export_rdev(rdev);
5666 }
5667 autorun_array(mddev);
5668 mddev_unlock(mddev);
5669 }
5670 /* on success, candidates will be empty, on error
5671 * it won't...
5672 */
159ec1fc 5673 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5674 list_del_init(&rdev->same_set);
1da177e4 5675 export_rdev(rdev);
4b80991c 5676 }
1da177e4
LT		 5677 mddev_put(mddev);
5678 }
5679 printk(KERN_INFO "md: ... autorun DONE.\n");
5680}
fdee8ae4 5681#endif /* !MODULE */
1da177e4 5682
f72ffdd6 5683static int get_version(void __user *arg)
1da177e4
LT		 5684{
5685 mdu_version_t ver;
5686
5687 ver.major = MD_MAJOR_VERSION;
5688 ver.minor = MD_MINOR_VERSION;
5689 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5690
5691 if (copy_to_user(arg, &ver, sizeof(ver)))
5692 return -EFAULT;
5693
5694 return 0;
5695}
5696
f72ffdd6 5697static int get_array_info(struct mddev *mddev, void __user *arg)
1da177e4
LT		 5698{
5699 mdu_array_info_t info;
a9f326eb 5700 int nr,working,insync,failed,spare;
3cb03002 5701 struct md_rdev *rdev;
1da177e4 5702
1ca69c4b
N		 5703 nr = working = insync = failed = spare = 0;
5704 rcu_read_lock();
5705 rdev_for_each_rcu(rdev, mddev) {
1da177e4 5706 nr++;
b2d444d7 5707 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5708 failed++;
5709 else {
5710 working++;
b2d444d7 5711 if (test_bit(In_sync, &rdev->flags))
f72ffdd6 5712 insync++;
1da177e4
LT		 5713 else
5714 spare++;
5715 }
5716 }
1ca69c4b 5717 rcu_read_unlock();
1da177e4
LT		 5718
5719 info.major_version = mddev->major_version;
5720 info.minor_version = mddev->minor_version;
5721 info.patch_version = MD_PATCHLEVEL_VERSION;
5722 info.ctime = mddev->ctime;
5723 info.level = mddev->level;
58c0fed4
AN		 5724 info.size = mddev->dev_sectors / 2;
5725 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5726 info.size = -1;
1da177e4
LT		 5727 info.nr_disks = nr;
5728 info.raid_disks = mddev->raid_disks;
5729 info.md_minor = mddev->md_minor;
5730 info.not_persistent= !mddev->persistent;
5731
5732 info.utime = mddev->utime;
5733 info.state = 0;
5734 if (mddev->in_sync)
5735 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5736 if (mddev->bitmap && mddev->bitmap_info.offset)
9bd35920 5737 info.state |= (1<<MD_SB_BITMAP_PRESENT);
ca8895d9
GR		 5738 if (mddev_is_clustered(mddev))
5739 info.state |= (1<<MD_SB_CLUSTERED);
a9f326eb 5740 info.active_disks = insync;
1da177e4
LT		 5741 info.working_disks = working;
5742 info.failed_disks = failed;
5743 info.spare_disks = spare;
5744
5745 info.layout = mddev->layout;
9d8f0363 5746 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5747
5748 if (copy_to_user(arg, &info, sizeof(info)))
5749 return -EFAULT;
5750
5751 return 0;
5752}
5753
f72ffdd6 5754static int get_bitmap_file(struct mddev *mddev, void __user * arg)
32a7627c
N		 5755{
5756 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
f4ad3d38 5757 char *ptr;
4af1a041 5758 int err;
32a7627c 5759
60559da4 5760 file = kmalloc(sizeof(*file), GFP_NOIO);
32a7627c 5761 if (!file)
4af1a041 5762 return -ENOMEM;
32a7627c 5763
4af1a041
N		 5764 err = 0;
5765 spin_lock(&mddev->lock);
32a7627c 5766 /* bitmap disabled, zero the first byte and copy out */
4af1a041 5767 if (!mddev->bitmap_info.file)
32a7627c 5768 file->pathname[0] = '\0';
4af1a041
N		 5769 else if ((ptr = d_path(&mddev->bitmap_info.file->f_path,
5770 file->pathname, sizeof(file->pathname))),
5771 IS_ERR(ptr))
5772 err = PTR_ERR(ptr);
5773 else
5774 memmove(file->pathname, ptr,
5775 sizeof(file->pathname)-(ptr-file->pathname));
5776 spin_unlock(&mddev->lock);
32a7627c 5777
4af1a041
N		 5778 if (err == 0 &&
5779 copy_to_user(arg, file, sizeof(*file)))
32a7627c 5780 err = -EFAULT;
4af1a041 5781
32a7627c
N		 5782 kfree(file);
5783 return err;
5784}
5785
f72ffdd6 5786static int get_disk_info(struct mddev *mddev, void __user * arg)
1da177e4
LT		 5787{
5788 mdu_disk_info_t info;
3cb03002 5789 struct md_rdev *rdev;
1da177e4
LT		 5790
5791 if (copy_from_user(&info, arg, sizeof(info)))
5792 return -EFAULT;
5793
1ca69c4b 5794 rcu_read_lock();
57d051dc 5795 rdev = md_find_rdev_nr_rcu(mddev, info.number);
1da177e4
LT		 5796 if (rdev) {
5797 info.major = MAJOR(rdev->bdev->bd_dev);
5798 info.minor = MINOR(rdev->bdev->bd_dev);
5799 info.raid_disk = rdev->raid_disk;
5800 info.state = 0;
b2d444d7 5801 if (test_bit(Faulty, &rdev->flags))
1da177e4 5802 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5803 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5804 info.state |= (1<<MD_DISK_ACTIVE);
5805 info.state |= (1<<MD_DISK_SYNC);
5806 }
8ddf9efe
N		 5807 if (test_bit(WriteMostly, &rdev->flags))
5808 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5809 } else {
5810 info.major = info.minor = 0;
5811 info.raid_disk = -1;
5812 info.state = (1<<MD_DISK_REMOVED);
5813 }
1ca69c4b 5814 rcu_read_unlock();
1da177e4
LT		 5815
5816 if (copy_to_user(arg, &info, sizeof(info)))
5817 return -EFAULT;
5818
5819 return 0;
5820}
5821
f72ffdd6 5822static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
1da177e4
LT		 5823{
5824 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5825 struct md_rdev *rdev;
1da177e4
LT		 5826 dev_t dev = MKDEV(info->major,info->minor);
5827
1aee41f6
GR		 5828 if (mddev_is_clustered(mddev) &&
5829 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
fa8259da 5830 pr_err("%s: Cannot add to clustered mddev.\n",
1aee41f6
GR		 5831 mdname(mddev));
5832 return -EINVAL;
5833 }
5834
1da177e4
LT		 5835 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5836 return -EOVERFLOW;
5837
5838 if (!mddev->raid_disks) {
5839 int err;
5840 /* expecting a device which has a superblock */
5841 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5842 if (IS_ERR(rdev)) {
f72ffdd6 5843 printk(KERN_WARNING
1da177e4
LT		 5844 "md: md_import_device returned %ld\n",
5845 PTR_ERR(rdev));
5846 return PTR_ERR(rdev);
5847 }
5848 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5849 struct md_rdev *rdev0
5850 = list_entry(mddev->disks.next,
5851 struct md_rdev, same_set);
a9f326eb 5852 err = super_types[mddev->major_version]
1da177e4
LT		 5853 .load_super(rdev, rdev0, mddev->minor_version);
5854 if (err < 0) {
f72ffdd6 5855 printk(KERN_WARNING
1da177e4 5856 "md: %s has different UUID to %s\n",
f72ffdd6 5857 bdevname(rdev->bdev,b),
1da177e4
LT		 5858 bdevname(rdev0->bdev,b2));
5859 export_rdev(rdev);
5860 return -EINVAL;
5861 }
5862 }
5863 err = bind_rdev_to_array(rdev, mddev);
5864 if (err)
5865 export_rdev(rdev);
5866 return err;
5867 }
5868
5869 /*
5870 * add_new_disk can be used once the array is assembled
5871 * to add "hot spares". They must already have a superblock
5872 * written
5873 */
5874 if (mddev->pers) {
5875 int err;
5876 if (!mddev->pers->hot_add_disk) {
f72ffdd6 5877 printk(KERN_WARNING
1da177e4
LT		 5878 "%s: personality does not support diskops!\n",
5879 mdname(mddev));
5880 return -EINVAL;
5881 }
7b1e35f6
N		 5882 if (mddev->persistent)
5883 rdev = md_import_device(dev, mddev->major_version,
5884 mddev->minor_version);
5885 else
5886 rdev = md_import_device(dev, -1, -1);
1da177e4 5887 if (IS_ERR(rdev)) {
f72ffdd6 5888 printk(KERN_WARNING
1da177e4
LT		 5889 "md: md_import_device returned %ld\n",
5890 PTR_ERR(rdev));
5891 return PTR_ERR(rdev);
5892 }
1a855a06 5893 /* set saved_raid_disk if appropriate */
41158c7e
N		 5894 if (!mddev->persistent) {
5895 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5896 info->raid_disk < mddev->raid_disks) {
41158c7e 5897 rdev->raid_disk = info->raid_disk;
bf572541 5898 set_bit(In_sync, &rdev->flags);
8313b8e5 5899 clear_bit(Bitmap_sync, &rdev->flags);
bf572541 5900 } else
41158c7e 5901 rdev->raid_disk = -1;
f466722c 5902 rdev->saved_raid_disk = rdev->raid_disk;
41158c7e
N		 5903 } else
5904 super_types[mddev->major_version].
5905 validate_super(mddev, rdev);
bedd86b7 5906 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 5907 rdev->raid_disk != info->raid_disk) {
bedd86b7
N		 5908 /* This was a hot-add request, but events doesn't
5909 * match, so reject it.
5910 */
5911 export_rdev(rdev);
5912 return -EINVAL;
5913 }
5914
b2d444d7 5915 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5916 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5917 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5918 else
5919 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5920
1aee41f6
GR		 5921 /*
5922 * check whether the device shows up in other nodes
5923 */
5924 if (mddev_is_clustered(mddev)) {
5925 if (info->state & (1 << MD_DISK_CANDIDATE)) {
5926 /* Through --cluster-confirm */
5927 set_bit(Candidate, &rdev->flags);
fa8259da
GR		 5928 err = md_cluster_ops->new_disk_ack(mddev, true);
5929 if (err) {
5930 export_rdev(rdev);
5931 return err;
5932 }
1aee41f6
GR		 5933 } else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
5934 /* --add initiated by this node */
5935 err = md_cluster_ops->add_new_disk_start(mddev, rdev);
5936 if (err) {
5937 md_cluster_ops->add_new_disk_finish(mddev);
5938 export_rdev(rdev);
5939 return err;
5940 }
5941 }
5942 }
5943
1da177e4
LT		 5944 rdev->raid_disk = -1;
5945 err = bind_rdev_to_array(rdev, mddev);
5946 if (err)
5947 export_rdev(rdev);
52664732 5948 else
a6da4ef8 5949 err = add_bound_rdev(rdev);
1aee41f6
GR		 5950 if (mddev_is_clustered(mddev) &&
5951 (info->state & (1 << MD_DISK_CLUSTER_ADD)))
5952 md_cluster_ops->add_new_disk_finish(mddev);
1da177e4
LT		 5953 return err;
5954 }
5955
5956 /* otherwise, add_new_disk is only allowed
5957 * for major_version==0 superblocks
5958 */
5959 if (mddev->major_version != 0) {
5960 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5961 mdname(mddev));
5962 return -EINVAL;
5963 }
5964
5965 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5966 int err;
d710e138 5967 rdev = md_import_device(dev, -1, 0);
1da177e4 5968 if (IS_ERR(rdev)) {
f72ffdd6 5969 printk(KERN_WARNING
1da177e4
LT		 5970 "md: error, md_import_device() returned %ld\n",
5971 PTR_ERR(rdev));
5972 return PTR_ERR(rdev);
5973 }
5974 rdev->desc_nr = info->number;
5975 if (info->raid_disk < mddev->raid_disks)
5976 rdev->raid_disk = info->raid_disk;
5977 else
5978 rdev->raid_disk = -1;
5979
1da177e4 5980 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5981 if (info->state & (1<<MD_DISK_SYNC))
5982 set_bit(In_sync, &rdev->flags);
1da177e4 5983
8ddf9efe
N		 5984 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5985 set_bit(WriteMostly, &rdev->flags);
5986
1da177e4
LT		 5987 if (!mddev->persistent) {
5988 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5989 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5990 } else
57b2caa3 5991 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5992 rdev->sectors = rdev->sb_start;
1da177e4 5993
2bf071bf
N		 5994 err = bind_rdev_to_array(rdev, mddev);
5995 if (err) {
5996 export_rdev(rdev);
5997 return err;
5998 }
1da177e4
LT		 5999 }
6000
6001 return 0;
6002}
6003
f72ffdd6 6004static int hot_remove_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 6005{
6006 char b[BDEVNAME_SIZE];
3cb03002 6007 struct md_rdev *rdev;
1da177e4 6008
1da177e4
LT		 6009 rdev = find_rdev(mddev, dev);
6010 if (!rdev)
6011 return -ENXIO;
6012
293467aa
GR		 6013 if (mddev_is_clustered(mddev))
6014 md_cluster_ops->metadata_update_start(mddev);
6015
3ea8929d
N		 6016 clear_bit(Blocked, &rdev->flags);
6017 remove_and_add_spares(mddev, rdev);
6018
1da177e4
LT		 6019 if (rdev->raid_disk >= 0)
6020 goto busy;
6021
88bcfef7
GR		 6022 if (mddev_is_clustered(mddev))
6023 md_cluster_ops->remove_disk(mddev, rdev);
6024
fb56dfef 6025 md_kick_rdev_from_array(rdev);
850b2b42 6026 md_update_sb(mddev, 1);
d7603b7e 6027 md_new_event(mddev);
1da177e4 6028
293467aa
GR		 6029 if (mddev_is_clustered(mddev))
6030 md_cluster_ops->metadata_update_finish(mddev);
6031
1da177e4
LT		 6032 return 0;
6033busy:
293467aa
GR		 6034 if (mddev_is_clustered(mddev))
6035 md_cluster_ops->metadata_update_cancel(mddev);
fdefa4d8 6036 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 6037 bdevname(rdev->bdev,b), mdname(mddev));
6038 return -EBUSY;
6039}
6040
f72ffdd6 6041static int hot_add_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 6042{
6043 char b[BDEVNAME_SIZE];
6044 int err;
3cb03002 6045 struct md_rdev *rdev;
1da177e4
LT		 6046
6047 if (!mddev->pers)
6048 return -ENODEV;
6049
6050 if (mddev->major_version != 0) {
6051 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6052 " version-0 superblocks.\n",
6053 mdname(mddev));
6054 return -EINVAL;
6055 }
6056 if (!mddev->pers->hot_add_disk) {
f72ffdd6 6057 printk(KERN_WARNING
1da177e4
LT		 6058 "%s: personality does not support diskops!\n",
6059 mdname(mddev));
6060 return -EINVAL;
6061 }
6062
d710e138 6063 rdev = md_import_device(dev, -1, 0);
1da177e4 6064 if (IS_ERR(rdev)) {
f72ffdd6 6065 printk(KERN_WARNING
1da177e4
LT		 6066 "md: error, md_import_device() returned %ld\n",
6067 PTR_ERR(rdev));
6068 return -EINVAL;
6069 }
6070
6071 if (mddev->persistent)
57b2caa3 6072 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 6073 else
77304d2a 6074 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 6075
8190e754 6076 rdev->sectors = rdev->sb_start;
1da177e4 6077
b2d444d7 6078 if (test_bit(Faulty, &rdev->flags)) {
f72ffdd6 6079 printk(KERN_WARNING
1da177e4
LT		 6080 "md: can not hot-add faulty %s disk to %s!\n",
6081 bdevname(rdev->bdev,b), mdname(mddev));
6082 err = -EINVAL;
6083 goto abort_export;
6084 }
293467aa
GR		 6085
6086 if (mddev_is_clustered(mddev))
6087 md_cluster_ops->metadata_update_start(mddev);
b2d444d7 6088 clear_bit(In_sync, &rdev->flags);
1da177e4 6089 rdev->desc_nr = -1;
5842730d 6090 rdev->saved_raid_disk = -1;
2bf071bf
N		 6091 err = bind_rdev_to_array(rdev, mddev);
6092 if (err)
293467aa 6093 goto abort_clustered;
1da177e4
LT		 6094
6095 /*
6096 * The rest should better be atomic, we can have disk failures
6097 * noticed in interrupt contexts ...
6098 */
6099
1da177e4
LT		 6100 rdev->raid_disk = -1;
6101
850b2b42 6102 md_update_sb(mddev, 1);
1da177e4 6103
293467aa
GR		 6104 if (mddev_is_clustered(mddev))
6105 md_cluster_ops->metadata_update_finish(mddev);
1da177e4
LT		 6106 /*
6107 * Kick recovery, maybe this spare has to be added to the
6108 * array immediately.
6109 */
6110 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6111 md_wakeup_thread(mddev->thread);
d7603b7e 6112 md_new_event(mddev);
1da177e4
LT		 6113 return 0;
6114
293467aa
GR		 6115abort_clustered:
6116 if (mddev_is_clustered(mddev))
6117 md_cluster_ops->metadata_update_cancel(mddev);
1da177e4
LT		 6118abort_export:
6119 export_rdev(rdev);
6120 return err;
6121}
6122
fd01b88c 6123static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c 6124{
035328c2 6125 int err = 0;
32a7627c 6126
36fa3063 6127 if (mddev->pers) {
d66b1b39 6128 if (!mddev->pers->quiesce || !mddev->thread)
36fa3063
N		 6129 return -EBUSY;
6130 if (mddev->recovery || mddev->sync_thread)
6131 return -EBUSY;
6132 /* we should be able to change the bitmap.. */
6133 }
32a7627c 6134
36fa3063 6135 if (fd >= 0) {
035328c2 6136 struct inode *inode;
1e594bb2
N		 6137 struct file *f;
6138
6139 if (mddev->bitmap || mddev->bitmap_info.file)
36fa3063 6140 return -EEXIST; /* cannot add when bitmap is present */
1e594bb2 6141 f = fget(fd);
32a7627c 6142
1e594bb2 6143 if (f == NULL) {
36fa3063
N		 6144 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6145 mdname(mddev));
6146 return -EBADF;
6147 }
6148
1e594bb2 6149 inode = f->f_mapping->host;
035328c2
N		 6150 if (!S_ISREG(inode->i_mode)) {
6151 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6152 mdname(mddev));
6153 err = -EBADF;
1e594bb2 6154 } else if (!(f->f_mode & FMODE_WRITE)) {
035328c2
N		 6155 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6156 mdname(mddev));
6157 err = -EBADF;
6158 } else if (atomic_read(&inode->i_writecount) != 1) {
36fa3063
N		 6159 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6160 mdname(mddev));
035328c2
N		 6161 err = -EBUSY;
6162 }
6163 if (err) {
1e594bb2 6164 fput(f);
36fa3063
N		 6165 return err;
6166 }
1e594bb2 6167 mddev->bitmap_info.file = f;
c3d9714e 6168 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 6169 } else if (mddev->bitmap == NULL)
6170 return -ENOENT; /* cannot remove what isn't there */
6171 err = 0;
6172 if (mddev->pers) {
6173 mddev->pers->quiesce(mddev, 1);
69e51b44 6174 if (fd >= 0) {
f9209a32
GR		 6175 struct bitmap *bitmap;
6176
6177 bitmap = bitmap_create(mddev, -1);
6178 if (!IS_ERR(bitmap)) {
6179 mddev->bitmap = bitmap;
69e51b44 6180 err = bitmap_load(mddev);
ba599aca
N		 6181 } else
6182 err = PTR_ERR(bitmap);
69e51b44 6183 }
d7375ab3 6184 if (fd < 0 || err) {
36fa3063 6185 bitmap_destroy(mddev);
d7375ab3
N		 6186 fd = -1; /* make sure to put the file */
6187 }
36fa3063 6188 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 6189 }
6190 if (fd < 0) {
4af1a041
N		 6191 struct file *f = mddev->bitmap_info.file;
6192 if (f) {
6193 spin_lock(&mddev->lock);
6194 mddev->bitmap_info.file = NULL;
6195 spin_unlock(&mddev->lock);
6196 fput(f);
6197 }
36fa3063
N		 6198 }
6199
32a7627c
N		 6200 return err;
6201}
6202
1da177e4
LT		 6203/*
6204 * set_array_info is used two different ways
6205 * The original usage is when creating a new array.
6206 * In this usage, raid_disks is > 0 and it together with
6207 * level, size, not_persistent,layout,chunksize determine the
6208 * shape of the array.
6209 * This will always create an array with a type-0.90.0 superblock.
6210 * The newer usage is when assembling an array.
6211 * In this case raid_disks will be 0, and the major_version field is
6212 * use to determine which style super-blocks are to be found on the devices.
6213 * The minor and patch _version numbers are also kept incase the
6214 * super_block handler wishes to interpret them.
6215 */
f72ffdd6 6216static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6217{
6218
6219 if (info->raid_disks == 0) {
6220 /* just setting version number for superblock loading */
6221 if (info->major_version < 0 ||
50511da3 6222 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 6223 super_types[info->major_version].name == NULL) {
6224 /* maybe try to auto-load a module? */
f72ffdd6 6225 printk(KERN_INFO
1da177e4
LT		 6226 "md: superblock version %d not known\n",
6227 info->major_version);
6228 return -EINVAL;
6229 }
6230 mddev->major_version = info->major_version;
6231 mddev->minor_version = info->minor_version;
6232 mddev->patch_version = info->patch_version;
3f9d7b0d 6233 mddev->persistent = !info->not_persistent;
cbd19983
N		 6234 /* ensure mddev_put doesn't delete this now that there
6235 * is some minimal configuration.
6236 */
6237 mddev->ctime = get_seconds();
1da177e4
LT		 6238 return 0;
6239 }
6240 mddev->major_version = MD_MAJOR_VERSION;
6241 mddev->minor_version = MD_MINOR_VERSION;
6242 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6243 mddev->ctime = get_seconds();
6244
6245 mddev->level = info->level;
17115e03 6246 mddev->clevel[0] = 0;
58c0fed4 6247 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 6248 mddev->raid_disks = info->raid_disks;
6249 /* don't set md_minor, it is determined by which /dev/md* was
6250 * openned
6251 */
6252 if (info->state & (1<<MD_SB_CLEAN))
6253 mddev->recovery_cp = MaxSector;
6254 else
6255 mddev->recovery_cp = 0;
6256 mddev->persistent = ! info->not_persistent;
e691063a 6257 mddev->external = 0;
1da177e4
LT		 6258
6259 mddev->layout = info->layout;
9d8f0363 6260 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 6261
6262 mddev->max_disks = MD_SB_DISKS;
6263
e691063a
N		 6264 if (mddev->persistent)
6265 mddev->flags = 0;
850b2b42 6266 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6267
c3d9714e 6268 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6269 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6270 mddev->bitmap_info.offset = 0;
b2a2703c 6271
f6705578
N		 6272 mddev->reshape_position = MaxSector;
6273
1da177e4
LT		 6274 /*
6275 * Generate a 128 bit UUID
6276 */
6277 get_random_bytes(mddev->uuid, 16);
6278
f6705578 6279 mddev->new_level = mddev->level;
664e7c41 6280 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 6281 mddev->new_layout = mddev->layout;
6282 mddev->delta_disks = 0;
2c810cdd 6283 mddev->reshape_backwards = 0;
f6705578 6284
1da177e4
LT		 6285 return 0;
6286}
6287
fd01b88c 6288void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6289{
b522adcd
DW		 6290 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6291
6292 if (mddev->external_size)
6293 return;
6294
1f403624
DW		 6295 mddev->array_sectors = array_sectors;
6296}
6297EXPORT_SYMBOL(md_set_array_sectors);
6298
fd01b88c 6299static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6300{
3cb03002 6301 struct md_rdev *rdev;
a35b0d69 6302 int rv;
d71f9f88 6303 int fit = (num_sectors == 0);
a35b0d69
N		 6304
6305 if (mddev->pers->resize == NULL)
6306 return -EINVAL;
d71f9f88
AN		 6307 /* The "num_sectors" is the number of sectors of each device that
6308 * is used. This can only make sense for arrays with redundancy.
6309 * linear and raid0 always use whatever space is available. We can only
6310 * consider changing this number if no resync or reconstruction is
6311 * happening, and if the new size is acceptable. It must fit before the
0f420358 6312 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 6313 * of each device. If num_sectors is zero, we find the largest size
6314 * that fits.
a35b0d69 6315 */
f851b60d
N		 6316 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6317 mddev->sync_thread)
a35b0d69 6318 return -EBUSY;
bd8839e0
N		 6319 if (mddev->ro)
6320 return -EROFS;
a4a6125a 6321
dafb20fa 6322 rdev_for_each(rdev, mddev) {
dd8ac336 6323 sector_t avail = rdev->sectors;
01ab5662 6324
d71f9f88
AN		 6325 if (fit && (num_sectors == 0 || num_sectors > avail))
6326 num_sectors = avail;
6327 if (avail < num_sectors)
a35b0d69
N		 6328 return -ENOSPC;
6329 }
d71f9f88 6330 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 6331 if (!rv)
6332 revalidate_disk(mddev->gendisk);
a35b0d69
N		 6333 return rv;
6334}
6335
fd01b88c 6336static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 6337{
6338 int rv;
c6563a8c 6339 struct md_rdev *rdev;
da943b99 6340 /* change the number of raid disks */
63c70c4f 6341 if (mddev->pers->check_reshape == NULL)
da943b99 6342 return -EINVAL;
bd8839e0
N		 6343 if (mddev->ro)
6344 return -EROFS;
da943b99 6345 if (raid_disks <= 0 ||
233fca36 6346 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 6347 return -EINVAL;
f851b60d
N		 6348 if (mddev->sync_thread ||
6349 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6350 mddev->reshape_position != MaxSector)
da943b99 6351 return -EBUSY;
c6563a8c
N		 6352
6353 rdev_for_each(rdev, mddev) {
6354 if (mddev->raid_disks < raid_disks &&
6355 rdev->data_offset < rdev->new_data_offset)
6356 return -EINVAL;
6357 if (mddev->raid_disks > raid_disks &&
6358 rdev->data_offset > rdev->new_data_offset)
6359 return -EINVAL;
6360 }
6361
63c70c4f 6362 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N		 6363 if (mddev->delta_disks < 0)
6364 mddev->reshape_backwards = 1;
6365 else if (mddev->delta_disks > 0)
6366 mddev->reshape_backwards = 0;
63c70c4f
N		 6367
6368 rv = mddev->pers->check_reshape(mddev);
2c810cdd 6369 if (rv < 0) {
de171cb9 6370 mddev->delta_disks = 0;
2c810cdd
N		 6371 mddev->reshape_backwards = 0;
6372 }
da943b99
N		 6373 return rv;
6374}
6375
1da177e4
LT		 6376/*
6377 * update_array_info is used to change the configuration of an
6378 * on-line array.
6379 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6380 * fields in the info are checked against the array.
6381 * Any differences that cannot be handled will cause an error.
6382 * Normally, only one change can be managed at a time.
6383 */
fd01b88c 6384static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6385{
6386 int rv = 0;
6387 int cnt = 0;
36fa3063
N		 6388 int state = 0;
6389
6390 /* calculate expected state,ignoring low bits */
c3d9714e 6391 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 6392 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 6393
6394 if (mddev->major_version != info->major_version ||
6395 mddev->minor_version != info->minor_version ||
6396/* mddev->patch_version != info->patch_version || */
6397 mddev->ctime != info->ctime ||
6398 mddev->level != info->level ||
6399/* mddev->layout != info->layout || */
4e023612 6400 mddev->persistent != !info->not_persistent ||
9d8f0363 6401 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 6402 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6403 ((state^info->state) & 0xfffffe00)
6404 )
1da177e4
LT		 6405 return -EINVAL;
6406 /* Check there is only one change */
58c0fed4
AN		 6407 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6408 cnt++;
6409 if (mddev->raid_disks != info->raid_disks)
6410 cnt++;
6411 if (mddev->layout != info->layout)
6412 cnt++;
6413 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6414 cnt++;
6415 if (cnt == 0)
6416 return 0;
6417 if (cnt > 1)
6418 return -EINVAL;
1da177e4
LT		 6419
6420 if (mddev->layout != info->layout) {
6421 /* Change layout
6422 * we don't need to do anything at the md level, the
6423 * personality will take care of it all.
6424 */
50ac168a 6425 if (mddev->pers->check_reshape == NULL)
1da177e4 6426 return -EINVAL;
597a711b
N		 6427 else {
6428 mddev->new_layout = info->layout;
50ac168a 6429 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 6430 if (rv)
6431 mddev->new_layout = mddev->layout;
6432 return rv;
6433 }
1da177e4 6434 }
293467aa
GR		 6435 if (mddev_is_clustered(mddev))
6436 md_cluster_ops->metadata_update_start(mddev);
58c0fed4 6437 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6438 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6439
da943b99
N		 6440 if (mddev->raid_disks != info->raid_disks)
6441 rv = update_raid_disks(mddev, info->raid_disks);
6442
36fa3063 6443 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
293467aa
GR		 6444 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6445 rv = -EINVAL;
6446 goto err;
6447 }
6448 if (mddev->recovery || mddev->sync_thread) {
6449 rv = -EBUSY;
6450 goto err;
6451 }
36fa3063 6452 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
f9209a32 6453 struct bitmap *bitmap;
36fa3063 6454 /* add the bitmap */
293467aa
GR		 6455 if (mddev->bitmap) {
6456 rv = -EEXIST;
6457 goto err;
6458 }
6459 if (mddev->bitmap_info.default_offset == 0) {
6460 rv = -EINVAL;
6461 goto err;
6462 }
c3d9714e
N		 6463 mddev->bitmap_info.offset =
6464 mddev->bitmap_info.default_offset;
6409bb05
N		 6465 mddev->bitmap_info.space =
6466 mddev->bitmap_info.default_space;
36fa3063 6467 mddev->pers->quiesce(mddev, 1);
f9209a32
GR		 6468 bitmap = bitmap_create(mddev, -1);
6469 if (!IS_ERR(bitmap)) {
6470 mddev->bitmap = bitmap;
69e51b44 6471 rv = bitmap_load(mddev);
ba599aca
N		 6472 } else
6473 rv = PTR_ERR(bitmap);
36fa3063
N		 6474 if (rv)
6475 bitmap_destroy(mddev);
6476 mddev->pers->quiesce(mddev, 0);
6477 } else {
6478 /* remove the bitmap */
293467aa
GR		 6479 if (!mddev->bitmap) {
6480 rv = -ENOENT;
6481 goto err;
6482 }
6483 if (mddev->bitmap->storage.file) {
6484 rv = -EINVAL;
6485 goto err;
6486 }
36fa3063
N		 6487 mddev->pers->quiesce(mddev, 1);
6488 bitmap_destroy(mddev);
6489 mddev->pers->quiesce(mddev, 0);
c3d9714e 6490 mddev->bitmap_info.offset = 0;
36fa3063
N		 6491 }
6492 }
850b2b42 6493 md_update_sb(mddev, 1);
293467aa
GR		 6494 if (mddev_is_clustered(mddev))
6495 md_cluster_ops->metadata_update_finish(mddev);
6496 return rv;
6497err:
6498 if (mddev_is_clustered(mddev))
6499 md_cluster_ops->metadata_update_cancel(mddev);
1da177e4
LT		 6500 return rv;
6501}
6502
fd01b88c 6503static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6504{
3cb03002 6505 struct md_rdev *rdev;
1ca69c4b 6506 int err = 0;
1da177e4
LT		 6507
6508 if (mddev->pers == NULL)
6509 return -ENODEV;
6510
1ca69c4b
N		 6511 rcu_read_lock();
6512 rdev = find_rdev_rcu(mddev, dev);
1da177e4 6513 if (!rdev)
1ca69c4b
N		 6514 err = -ENODEV;
6515 else {
6516 md_error(mddev, rdev);
6517 if (!test_bit(Faulty, &rdev->flags))
6518 err = -EBUSY;
6519 }
6520 rcu_read_unlock();
6521 return err;
1da177e4
LT		 6522}
6523
2f9618ce
AN		 6524/*
6525 * We have a problem here : there is no easy way to give a CHS
6526 * virtual geometry. We currently pretend that we have a 2 heads
6527 * 4 sectors (with a BIG number of cylinders...). This drives
6528 * dosfs just mad... ;-)
6529 */
a885c8c4
CH		 6530static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6531{
fd01b88c 6532 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6533
6534 geo->heads = 2;
6535 geo->sectors = 4;
49ce6cea 6536 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6537 return 0;
6538}
6539
cb335f88
NS		 6540static inline bool md_ioctl_valid(unsigned int cmd)
6541{
6542 switch (cmd) {
6543 case ADD_NEW_DISK:
6544 case BLKROSET:
6545 case GET_ARRAY_INFO:
6546 case GET_BITMAP_FILE:
6547 case GET_DISK_INFO:
6548 case HOT_ADD_DISK:
6549 case HOT_REMOVE_DISK:
cb335f88
NS		 6550 case RAID_AUTORUN:
6551 case RAID_VERSION:
6552 case RESTART_ARRAY_RW:
6553 case RUN_ARRAY:
6554 case SET_ARRAY_INFO:
6555 case SET_BITMAP_FILE:
6556 case SET_DISK_FAULTY:
6557 case STOP_ARRAY:
6558 case STOP_ARRAY_RO:
1aee41f6 6559 case CLUSTERED_DISK_NACK:
cb335f88
NS		 6560 return true;
6561 default:
6562 return false;
6563 }
6564}
6565
a39907fa 6566static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6567 unsigned int cmd, unsigned long arg)
6568{
6569 int err = 0;
6570 void __user *argp = (void __user *)arg;
fd01b88c 6571 struct mddev *mddev = NULL;
e2218350 6572 int ro;
1da177e4 6573
cb335f88
NS		 6574 if (!md_ioctl_valid(cmd))
6575 return -ENOTTY;
6576
506c9e44
N		 6577 switch (cmd) {
6578 case RAID_VERSION:
6579 case GET_ARRAY_INFO:
6580 case GET_DISK_INFO:
6581 break;
6582 default:
6583 if (!capable(CAP_SYS_ADMIN))
6584 return -EACCES;
6585 }
1da177e4
LT		 6586
6587 /*
6588 * Commands dealing with the RAID driver but not any
6589 * particular array:
6590 */
c02c0aeb
N		 6591 switch (cmd) {
6592 case RAID_VERSION:
6593 err = get_version(argp);
3adc28d8 6594 goto out;
1da177e4 6595
1da177e4 6596#ifndef MODULE
c02c0aeb
N		 6597 case RAID_AUTORUN:
6598 err = 0;
6599 autostart_arrays(arg);
3adc28d8 6600 goto out;
1da177e4 6601#endif
c02c0aeb 6602 default:;
1da177e4
LT		 6603 }
6604
6605 /*
6606 * Commands creating/starting a new array:
6607 */
6608
a39907fa 6609 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6610
6611 if (!mddev) {
6612 BUG();
3adc28d8 6613 goto out;
1da177e4
LT		 6614 }
6615
1ca69c4b
N		 6616 /* Some actions do not requires the mutex */
6617 switch (cmd) {
6618 case GET_ARRAY_INFO:
6619 if (!mddev->raid_disks && !mddev->external)
6620 err = -ENODEV;
6621 else
6622 err = get_array_info(mddev, argp);
3adc28d8 6623 goto out;
1ca69c4b
N		 6624
6625 case GET_DISK_INFO:
6626 if (!mddev->raid_disks && !mddev->external)
6627 err = -ENODEV;
6628 else
6629 err = get_disk_info(mddev, argp);
3adc28d8 6630 goto out;
1ca69c4b
N		 6631
6632 case SET_DISK_FAULTY:
6633 err = set_disk_faulty(mddev, new_decode_dev(arg));
3adc28d8 6634 goto out;
4af1a041
N		 6635
6636 case GET_BITMAP_FILE:
6637 err = get_bitmap_file(mddev, argp);
6638 goto out;
6639
1ca69c4b
N		 6640 }
6641
a7a3f08d
N		 6642 if (cmd == ADD_NEW_DISK)
6643 /* need to ensure md_delayed_delete() has completed */
6644 flush_workqueue(md_misc_wq);
6645
90f5f7ad
HR		 6646 if (cmd == HOT_REMOVE_DISK)
6647 /* need to ensure recovery thread has run */
6648 wait_event_interruptible_timeout(mddev->sb_wait,
6649 !test_bit(MD_RECOVERY_NEEDED,
6650 &mddev->flags),
6651 msecs_to_jiffies(5000));
260fa034
N		 6652 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6653 /* Need to flush page cache, and ensure no-one else opens
6654 * and writes
6655 */
6656 mutex_lock(&mddev->open_mutex);
9ba3b7f5 6657 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
260fa034
N		 6658 mutex_unlock(&mddev->open_mutex);
6659 err = -EBUSY;
3adc28d8 6660 goto out;
260fa034
N		 6661 }
6662 set_bit(MD_STILL_CLOSED, &mddev->flags);
6663 mutex_unlock(&mddev->open_mutex);
6664 sync_blockdev(bdev);
6665 }
1da177e4
LT		 6666 err = mddev_lock(mddev);
6667 if (err) {
f72ffdd6 6668 printk(KERN_INFO
1da177e4
LT		 6669 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6670 err, cmd);
3adc28d8 6671 goto out;
1da177e4
LT		 6672 }
6673
c02c0aeb
N		 6674 if (cmd == SET_ARRAY_INFO) {
6675 mdu_array_info_t info;
6676 if (!arg)
6677 memset(&info, 0, sizeof(info));
6678 else if (copy_from_user(&info, argp, sizeof(info))) {
6679 err = -EFAULT;
3adc28d8 6680 goto unlock;
c02c0aeb
N		 6681 }
6682 if (mddev->pers) {
6683 err = update_array_info(mddev, &info);
6684 if (err) {
6685 printk(KERN_WARNING "md: couldn't update"
6686 " array info. %d\n", err);
3adc28d8 6687 goto unlock;
1da177e4 6688 }
3adc28d8 6689 goto unlock;
c02c0aeb
N		 6690 }
6691 if (!list_empty(&mddev->disks)) {
6692 printk(KERN_WARNING
6693 "md: array %s already has disks!\n",
6694 mdname(mddev));
6695 err = -EBUSY;
3adc28d8 6696 goto unlock;
c02c0aeb
N		 6697 }
6698 if (mddev->raid_disks) {
6699 printk(KERN_WARNING
6700 "md: array %s already initialised!\n",
6701 mdname(mddev));
6702 err = -EBUSY;
3adc28d8 6703 goto unlock;
c02c0aeb
N		 6704 }
6705 err = set_array_info(mddev, &info);
6706 if (err) {
6707 printk(KERN_WARNING "md: couldn't set"
6708 " array info. %d\n", err);
3adc28d8 6709 goto unlock;
c02c0aeb 6710 }
3adc28d8 6711 goto unlock;
1da177e4
LT		 6712 }
6713
6714 /*
6715 * Commands querying/configuring an existing array:
6716 */
32a7627c 6717 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6718 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6719 if ((!mddev->raid_disks && !mddev->external)
6720 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6721 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6722 && cmd != GET_BITMAP_FILE) {
1da177e4 6723 err = -ENODEV;
3adc28d8 6724 goto unlock;
1da177e4
LT		 6725 }
6726
6727 /*
6728 * Commands even a read-only array can execute:
6729 */
c02c0aeb 6730 switch (cmd) {
c02c0aeb
N		 6731 case RESTART_ARRAY_RW:
6732 err = restart_array(mddev);
3adc28d8 6733 goto unlock;
1da177e4 6734
c02c0aeb
N		 6735 case STOP_ARRAY:
6736 err = do_md_stop(mddev, 0, bdev);
3adc28d8 6737 goto unlock;
1da177e4 6738
c02c0aeb
N		 6739 case STOP_ARRAY_RO:
6740 err = md_set_readonly(mddev, bdev);
3adc28d8 6741 goto unlock;
1da177e4 6742
3ea8929d
N		 6743 case HOT_REMOVE_DISK:
6744 err = hot_remove_disk(mddev, new_decode_dev(arg));
3adc28d8 6745 goto unlock;
3ea8929d 6746
7ceb17e8
N		 6747 case ADD_NEW_DISK:
6748 /* We can support ADD_NEW_DISK on read-only arrays
6749 * on if we are re-adding a preexisting device.
6750 * So require mddev->pers and MD_DISK_SYNC.
6751 */
6752 if (mddev->pers) {
6753 mdu_disk_info_t info;
6754 if (copy_from_user(&info, argp, sizeof(info)))
6755 err = -EFAULT;
6756 else if (!(info.state & (1<<MD_DISK_SYNC)))
6757 /* Need to clear read-only for this */
6758 break;
6759 else
6760 err = add_new_disk(mddev, &info);
3adc28d8 6761 goto unlock;
7ceb17e8
N		 6762 }
6763 break;
6764
c02c0aeb
N		 6765 case BLKROSET:
6766 if (get_user(ro, (int __user *)(arg))) {
6767 err = -EFAULT;
3adc28d8 6768 goto unlock;
c02c0aeb
N		 6769 }
6770 err = -EINVAL;
e2218350 6771
c02c0aeb
N		 6772 /* if the bdev is going readonly the value of mddev->ro
6773 * does not matter, no writes are coming
6774 */
6775 if (ro)
3adc28d8 6776 goto unlock;
e2218350 6777
c02c0aeb
N		 6778 /* are we are already prepared for writes? */
6779 if (mddev->ro != 1)
3adc28d8 6780 goto unlock;
e2218350 6781
c02c0aeb
N		 6782 /* transitioning to readauto need only happen for
6783 * arrays that call md_write_start
6784 */
6785 if (mddev->pers) {
6786 err = restart_array(mddev);
6787 if (err == 0) {
6788 mddev->ro = 2;
6789 set_disk_ro(mddev->gendisk, 0);
e2218350 6790 }
c02c0aeb 6791 }
3adc28d8 6792 goto unlock;
1da177e4
LT		 6793 }
6794
6795 /*
6796 * The remaining ioctls are changing the state of the
f91de92e 6797 * superblock, so we do not allow them on read-only arrays.
1da177e4 6798 */
326eb17d 6799 if (mddev->ro && mddev->pers) {
f91de92e
N		 6800 if (mddev->ro == 2) {
6801 mddev->ro = 0;
00bcb4ac 6802 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 6803 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N		 6804 /* mddev_unlock will wake thread */
6805 /* If a device failed while we were read-only, we
6806 * need to make sure the metadata is updated now.
6807 */
6808 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6809 mddev_unlock(mddev);
6810 wait_event(mddev->sb_wait,
6811 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6812 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
29f097c4 6813 mddev_lock_nointr(mddev);
f3378b48 6814 }
f91de92e
N		 6815 } else {
6816 err = -EROFS;
3adc28d8 6817 goto unlock;
f91de92e 6818 }
1da177e4
LT		 6819 }
6820
c02c0aeb
N		 6821 switch (cmd) {
6822 case ADD_NEW_DISK:
1da177e4 6823 {
c02c0aeb
N		 6824 mdu_disk_info_t info;
6825 if (copy_from_user(&info, argp, sizeof(info)))
6826 err = -EFAULT;
6827 else
6828 err = add_new_disk(mddev, &info);
3adc28d8 6829 goto unlock;
c02c0aeb 6830 }
1da177e4 6831
1aee41f6
GR		 6832 case CLUSTERED_DISK_NACK:
6833 if (mddev_is_clustered(mddev))
6834 md_cluster_ops->new_disk_ack(mddev, false);
6835 else
6836 err = -EINVAL;
6837 goto unlock;
6838
c02c0aeb
N		 6839 case HOT_ADD_DISK:
6840 err = hot_add_disk(mddev, new_decode_dev(arg));
3adc28d8 6841 goto unlock;
1da177e4 6842
c02c0aeb
N		 6843 case RUN_ARRAY:
6844 err = do_md_run(mddev);
3adc28d8 6845 goto unlock;
1da177e4 6846
c02c0aeb
N		 6847 case SET_BITMAP_FILE:
6848 err = set_bitmap_file(mddev, (int)arg);
3adc28d8 6849 goto unlock;
32a7627c 6850
c02c0aeb
N		 6851 default:
6852 err = -EINVAL;
3adc28d8 6853 goto unlock;
1da177e4
LT		 6854 }
6855
3adc28d8 6856unlock:
d3374825
N		 6857 if (mddev->hold_active == UNTIL_IOCTL &&
6858 err != -EINVAL)
6859 mddev->hold_active = 0;
1da177e4 6860 mddev_unlock(mddev);
3adc28d8 6861out:
1da177e4
LT		 6862 return err;
6863}
aa98aa31
AB		 6864#ifdef CONFIG_COMPAT
6865static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6866 unsigned int cmd, unsigned long arg)
6867{
6868 switch (cmd) {
6869 case HOT_REMOVE_DISK:
6870 case HOT_ADD_DISK:
6871 case SET_DISK_FAULTY:
6872 case SET_BITMAP_FILE:
6873 /* These take in integer arg, do not convert */
6874 break;
6875 default:
6876 arg = (unsigned long)compat_ptr(arg);
6877 break;
6878 }
6879
6880 return md_ioctl(bdev, mode, cmd, arg);
6881}
6882#endif /* CONFIG_COMPAT */
1da177e4 6883
a39907fa 6884static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6885{
6886 /*
6887 * Succeed if we can lock the mddev, which confirms that
6888 * it isn't being stopped right now.
6889 */
fd01b88c 6890 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6891 int err;
6892
0c098220
YL		 6893 if (!mddev)
6894 return -ENODEV;
6895
d3374825
N		 6896 if (mddev->gendisk != bdev->bd_disk) {
6897 /* we are racing with mddev_put which is discarding this
6898 * bd_disk.
6899 */
6900 mddev_put(mddev);
6901 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6902 flush_workqueue(md_misc_wq);
d3374825
N		 6903 /* Then retry the open from the top */
6904 return -ERESTARTSYS;
6905 }
6906 BUG_ON(mddev != bdev->bd_disk->private_data);
6907
c8c00a69 6908 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6909 goto out;
6910
6911 err = 0;
f2ea68cf 6912 atomic_inc(&mddev->openers);
260fa034 6913 clear_bit(MD_STILL_CLOSED, &mddev->flags);
c8c00a69 6914 mutex_unlock(&mddev->open_mutex);
1da177e4 6915
f0b4f7e2 6916 check_disk_change(bdev);
1da177e4
LT		 6917 out:
6918 return err;
6919}
6920
db2a144b 6921static void md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6922{
f72ffdd6 6923 struct mddev *mddev = disk->private_data;
1da177e4 6924
52e5f9d1 6925 BUG_ON(!mddev);
f2ea68cf 6926 atomic_dec(&mddev->openers);
1da177e4 6927 mddev_put(mddev);
1da177e4 6928}
f0b4f7e2
N		 6929
6930static int md_media_changed(struct gendisk *disk)
6931{
fd01b88c 6932 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6933
6934 return mddev->changed;
6935}
6936
6937static int md_revalidate(struct gendisk *disk)
6938{
fd01b88c 6939 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6940
6941 mddev->changed = 0;
6942 return 0;
6943}
83d5cde4 6944static const struct block_device_operations md_fops =
1da177e4
LT		 6945{
6946 .owner = THIS_MODULE,
a39907fa
AV		 6947 .open = md_open,
6948 .release = md_release,
b492b852 6949 .ioctl = md_ioctl,
aa98aa31
AB		 6950#ifdef CONFIG_COMPAT
6951 .compat_ioctl = md_compat_ioctl,
6952#endif
a885c8c4 6953 .getgeo = md_getgeo,
f0b4f7e2
N		 6954 .media_changed = md_media_changed,
6955 .revalidate_disk= md_revalidate,
1da177e4
LT		 6956};
6957
f72ffdd6 6958static int md_thread(void *arg)
1da177e4 6959{
2b8bf345 6960 struct md_thread *thread = arg;
1da177e4 6961
1da177e4
LT		 6962 /*
6963 * md_thread is a 'system-thread', it's priority should be very
6964 * high. We avoid resource deadlocks individually in each
6965 * raid personality. (RAID5 does preallocation) We also use RR and
6966 * the very same RT priority as kswapd, thus we will never get
6967 * into a priority inversion deadlock.
6968 *
6969 * we definitely have to have equal or higher priority than
6970 * bdflush, otherwise bdflush will deadlock if there are too
6971 * many dirty RAID5 blocks.
6972 */
1da177e4 6973
6985c43f 6974 allow_signal(SIGKILL);
a6fb0934 6975 while (!kthread_should_stop()) {
1da177e4 6976
93588e22
N		 6977 /* We need to wait INTERRUPTIBLE so that
6978 * we don't add to the load-average.
6979 * That means we need to be sure no signals are
6980 * pending
6981 */
6982 if (signal_pending(current))
6983 flush_signals(current);
6984
6985 wait_event_interruptible_timeout
6986 (thread->wqueue,
6987 test_bit(THREAD_WAKEUP, &thread->flags)
6988 || kthread_should_stop(),
6989 thread->timeout);
1da177e4 6990
6c987910
N		 6991 clear_bit(THREAD_WAKEUP, &thread->flags);
6992 if (!kthread_should_stop())
4ed8731d 6993 thread->run(thread);
1da177e4 6994 }
a6fb0934 6995
1da177e4
LT		 6996 return 0;
6997}
6998
2b8bf345 6999void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 7000{
7001 if (thread) {
36a4e1fe 7002 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 7003 set_bit(THREAD_WAKEUP, &thread->flags);
7004 wake_up(&thread->wqueue);
7005 }
7006}
6c144d31 7007EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 7008
4ed8731d
SL		 7009struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7010 struct mddev *mddev, const char *name)
1da177e4 7011{
2b8bf345 7012 struct md_thread *thread;
1da177e4 7013
2b8bf345 7014 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 7015 if (!thread)
7016 return NULL;
7017
1da177e4
LT		 7018 init_waitqueue_head(&thread->wqueue);
7019
1da177e4
LT		 7020 thread->run = run;
7021 thread->mddev = mddev;
32a7627c 7022 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 7023 thread->tsk = kthread_run(md_thread, thread,
7024 "%s_%s",
7025 mdname(thread->mddev),
0232605d 7026 name);
a6fb0934 7027 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 7028 kfree(thread);
7029 return NULL;
7030 }
1da177e4
LT		 7031 return thread;
7032}
6c144d31 7033EXPORT_SYMBOL(md_register_thread);
1da177e4 7034
2b8bf345 7035void md_unregister_thread(struct md_thread **threadp)
1da177e4 7036{
2b8bf345 7037 struct md_thread *thread = *threadp;
e0cf8f04
N		 7038 if (!thread)
7039 return;
36a4e1fe 7040 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 7041 /* Locking ensures that mddev_unlock does not wake_up a
7042 * non-existent thread
7043 */
7044 spin_lock(&pers_lock);
7045 *threadp = NULL;
7046 spin_unlock(&pers_lock);
a6fb0934
N		 7047
7048 kthread_stop(thread->tsk);
1da177e4
LT		 7049 kfree(thread);
7050}
6c144d31 7051EXPORT_SYMBOL(md_unregister_thread);
1da177e4 7052
fd01b88c 7053void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7054{
b2d444d7 7055 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 7056 return;
6bfe0b49 7057
de393cde 7058 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 7059 return;
7060 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 7061 if (mddev->degraded)
7062 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 7063 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 7064 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7065 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7066 md_wakeup_thread(mddev->thread);
768a418d 7067 if (mddev->event_work.func)
e804ac78 7068 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 7069 md_new_event_inintr(mddev);
1da177e4 7070}
6c144d31 7071EXPORT_SYMBOL(md_error);
1da177e4
LT		 7072
7073/* seq_file implementation /proc/mdstat */
7074
7075static void status_unused(struct seq_file *seq)
7076{
7077 int i = 0;
3cb03002 7078 struct md_rdev *rdev;
1da177e4
LT		 7079
7080 seq_printf(seq, "unused devices: ");
7081
159ec1fc 7082 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 7083 char b[BDEVNAME_SIZE];
7084 i++;
7085 seq_printf(seq, "%s ",
7086 bdevname(rdev->bdev,b));
7087 }
7088 if (!i)
7089 seq_printf(seq, "<none>");
7090
7091 seq_printf(seq, "\n");
7092}
7093
f72ffdd6 7094static void status_resync(struct seq_file *seq, struct mddev *mddev)
1da177e4 7095{
dd71cf6b
N		 7096 sector_t max_sectors, resync, res;
7097 unsigned long dt, db;
7098 sector_t rt;
4588b42e
N		 7099 int scale;
7100 unsigned int per_milli;
1da177e4 7101
72f36d59
N		 7102 if (mddev->curr_resync <= 3)
7103 resync = 0;
7104 else
7105 resync = mddev->curr_resync
7106 - atomic_read(&mddev->recovery_active);
1da177e4 7107
c804cdec
N		 7108 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7109 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 7110 max_sectors = mddev->resync_max_sectors;
1da177e4 7111 else
dd71cf6b 7112 max_sectors = mddev->dev_sectors;
1da177e4 7113
403df478 7114 WARN_ON(max_sectors == 0);
4588b42e 7115 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 7116 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 7117 * u32, as those are the requirements for sector_div.
7118 * Thus 'scale' must be at least 10
7119 */
7120 scale = 10;
7121 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 7122 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 7123 scale++;
7124 }
7125 res = (resync>>scale)*1000;
dd71cf6b 7126 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 7127
7128 per_milli = res;
1da177e4 7129 {
4588b42e 7130 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 7131 seq_printf(seq, "[");
7132 for (i = 0; i < x; i++)
7133 seq_printf(seq, "=");
7134 seq_printf(seq, ">");
7135 for (i = 0; i < y; i++)
7136 seq_printf(seq, ".");
7137 seq_printf(seq, "] ");
7138 }
4588b42e 7139 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 7140 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7141 "reshape" :
61df9d91
N		 7142 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7143 "check" :
7144 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7145 "resync" : "recovery"))),
7146 per_milli/10, per_milli % 10,
dd71cf6b
N		 7147 (unsigned long long) resync/2,
7148 (unsigned long long) max_sectors/2);
1da177e4
LT		 7149
7150 /*
1da177e4
LT		 7151 * dt: time from mark until now
7152 * db: blocks written from mark until now
7153 * rt: remaining time
dd71cf6b
N		 7154 *
7155 * rt is a sector_t, so could be 32bit or 64bit.
7156 * So we divide before multiply in case it is 32bit and close
7157 * to the limit.
25985edc 7158 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 7159 * near the end of resync when the number of remaining sectors
7160 * is close to 'db'.
7161 * We then divide rt by 32 after multiplying by db to compensate.
7162 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 7163 */
7164 dt = ((jiffies - mddev->resync_mark) / HZ);
7165 if (!dt) dt++;
ff4e8d9a
N		 7166 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7167 - mddev->resync_mark_cnt;
1da177e4 7168
dd71cf6b
N		 7169 rt = max_sectors - resync; /* number of remaining sectors */
7170 sector_div(rt, db/32+1);
7171 rt *= dt;
7172 rt >>= 5;
7173
7174 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7175 ((unsigned long)rt % 60)/6);
1da177e4 7176
ff4e8d9a 7177 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 7178}
7179
7180static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7181{
7182 struct list_head *tmp;
7183 loff_t l = *pos;
fd01b88c 7184 struct mddev *mddev;
1da177e4
LT		 7185
7186 if (l >= 0x10000)
7187 return NULL;
7188 if (!l--)
7189 /* header */
7190 return (void*)1;
7191
7192 spin_lock(&all_mddevs_lock);
7193 list_for_each(tmp,&all_mddevs)
7194 if (!l--) {
fd01b88c 7195 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 7196 mddev_get(mddev);
7197 spin_unlock(&all_mddevs_lock);
7198 return mddev;
7199 }
7200 spin_unlock(&all_mddevs_lock);
7201 if (!l--)
7202 return (void*)2;/* tail */
7203 return NULL;
7204}
7205
7206static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7207{
7208 struct list_head *tmp;
fd01b88c 7209 struct mddev *next_mddev, *mddev = v;
f72ffdd6 7210
1da177e4
LT		 7211 ++*pos;
7212 if (v == (void*)2)
7213 return NULL;
7214
7215 spin_lock(&all_mddevs_lock);
7216 if (v == (void*)1)
7217 tmp = all_mddevs.next;
7218 else
7219 tmp = mddev->all_mddevs.next;
7220 if (tmp != &all_mddevs)
fd01b88c 7221 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 7222 else {
7223 next_mddev = (void*)2;
7224 *pos = 0x10000;
f72ffdd6 7225 }
1da177e4
LT		 7226 spin_unlock(&all_mddevs_lock);
7227
7228 if (v != (void*)1)
7229 mddev_put(mddev);
7230 return next_mddev;
7231
7232}
7233
7234static void md_seq_stop(struct seq_file *seq, void *v)
7235{
fd01b88c 7236 struct mddev *mddev = v;
1da177e4
LT		 7237
7238 if (mddev && v != (void*)1 && v != (void*)2)
7239 mddev_put(mddev);
7240}
7241
7242static int md_seq_show(struct seq_file *seq, void *v)
7243{
fd01b88c 7244 struct mddev *mddev = v;
dd8ac336 7245 sector_t sectors;
3cb03002 7246 struct md_rdev *rdev;
1da177e4
LT		 7247
7248 if (v == (void*)1) {
84fc4b56 7249 struct md_personality *pers;
1da177e4
LT		 7250 seq_printf(seq, "Personalities : ");
7251 spin_lock(&pers_lock);
2604b703
N		 7252 list_for_each_entry(pers, &pers_list, list)
7253 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 7254
7255 spin_unlock(&pers_lock);
7256 seq_printf(seq, "\n");
f1514638 7257 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7258 return 0;
7259 }
7260 if (v == (void*)2) {
7261 status_unused(seq);
7262 return 0;
7263 }
7264
36d091f4 7265 spin_lock(&mddev->lock);
1da177e4
LT		 7266 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7267 seq_printf(seq, "%s : %sactive", mdname(mddev),
7268 mddev->pers ? "" : "in");
7269 if (mddev->pers) {
f91de92e 7270 if (mddev->ro==1)
1da177e4 7271 seq_printf(seq, " (read-only)");
f91de92e 7272 if (mddev->ro==2)
52720ae7 7273 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 7274 seq_printf(seq, " %s", mddev->pers->name);
7275 }
7276
dd8ac336 7277 sectors = 0;
f97fcad3
N		 7278 rcu_read_lock();
7279 rdev_for_each_rcu(rdev, mddev) {
1da177e4
LT		 7280 char b[BDEVNAME_SIZE];
7281 seq_printf(seq, " %s[%d]",
7282 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 7283 if (test_bit(WriteMostly, &rdev->flags))
7284 seq_printf(seq, "(W)");
b2d444d7 7285 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7286 seq_printf(seq, "(F)");
7287 continue;
2d78f8c4
N		 7288 }
7289 if (rdev->raid_disk < 0)
b325a32e 7290 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 7291 if (test_bit(Replacement, &rdev->flags))
7292 seq_printf(seq, "(R)");
dd8ac336 7293 sectors += rdev->sectors;
1da177e4 7294 }
f97fcad3 7295 rcu_read_unlock();
1da177e4
LT		 7296
7297 if (!list_empty(&mddev->disks)) {
7298 if (mddev->pers)
7299 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 7300 (unsigned long long)
7301 mddev->array_sectors / 2);
1da177e4
LT		 7302 else
7303 seq_printf(seq, "\n %llu blocks",
dd8ac336 7304 (unsigned long long)sectors / 2);
1da177e4 7305 }
1cd6bf19
N		 7306 if (mddev->persistent) {
7307 if (mddev->major_version != 0 ||
7308 mddev->minor_version != 90) {
7309 seq_printf(seq," super %d.%d",
7310 mddev->major_version,
7311 mddev->minor_version);
7312 }
e691063a
N		 7313 } else if (mddev->external)
7314 seq_printf(seq, " super external:%s",
7315 mddev->metadata_type);
7316 else
1cd6bf19 7317 seq_printf(seq, " super non-persistent");
1da177e4
LT		 7318
7319 if (mddev->pers) {
d710e138 7320 mddev->pers->status(seq, mddev);
f72ffdd6 7321 seq_printf(seq, "\n ");
8e1b39d6
N		 7322 if (mddev->pers->sync_request) {
7323 if (mddev->curr_resync > 2) {
d710e138 7324 status_resync(seq, mddev);
8e1b39d6 7325 seq_printf(seq, "\n ");
72f36d59 7326 } else if (mddev->curr_resync >= 1)
8e1b39d6
N		 7327 seq_printf(seq, "\tresync=DELAYED\n ");
7328 else if (mddev->recovery_cp < MaxSector)
7329 seq_printf(seq, "\tresync=PENDING\n ");
7330 }
32a7627c
N		 7331 } else
7332 seq_printf(seq, "\n ");
7333
57148964 7334 bitmap_status(seq, mddev->bitmap);
1da177e4
LT		 7335
7336 seq_printf(seq, "\n");
7337 }
36d091f4 7338 spin_unlock(&mddev->lock);
f72ffdd6 7339
1da177e4
LT		 7340 return 0;
7341}
7342
110518bc 7343static const struct seq_operations md_seq_ops = {
1da177e4
LT		 7344 .start = md_seq_start,
7345 .next = md_seq_next,
7346 .stop = md_seq_stop,
7347 .show = md_seq_show,
7348};
7349
7350static int md_seq_open(struct inode *inode, struct file *file)
7351{
f1514638 7352 struct seq_file *seq;
1da177e4
LT		 7353 int error;
7354
7355 error = seq_open(file, &md_seq_ops);
d7603b7e 7356 if (error)
f1514638
KS		 7357 return error;
7358
7359 seq = file->private_data;
7360 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7361 return error;
7362}
7363
e2f23b60 7364static int md_unloading;
d7603b7e
N		 7365static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7366{
f1514638 7367 struct seq_file *seq = filp->private_data;
d7603b7e
N		 7368 int mask;
7369
e2f23b60 7370 if (md_unloading)
7d7e64f2 7371 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
d7603b7e
N		 7372 poll_wait(filp, &md_event_waiters, wait);
7373
7374 /* always allow read */
7375 mask = POLLIN | POLLRDNORM;
7376
f1514638 7377 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 7378 mask |= POLLERR | POLLPRI;
7379 return mask;
7380}
7381
fa027c2a 7382static const struct file_operations md_seq_fops = {
e24650c2 7383 .owner = THIS_MODULE,
1da177e4
LT		 7384 .open = md_seq_open,
7385 .read = seq_read,
7386 .llseek = seq_lseek,
c3f94b40 7387 .release = seq_release_private,
d7603b7e 7388 .poll = mdstat_poll,
1da177e4
LT		 7389};
7390
84fc4b56 7391int register_md_personality(struct md_personality *p)
1da177e4 7392{
50bd3774
CY		 7393 printk(KERN_INFO "md: %s personality registered for level %d\n",
7394 p->name, p->level);
1da177e4 7395 spin_lock(&pers_lock);
2604b703 7396 list_add_tail(&p->list, &pers_list);
1da177e4
LT		 7397 spin_unlock(&pers_lock);
7398 return 0;
7399}
6c144d31 7400EXPORT_SYMBOL(register_md_personality);
1da177e4 7401
84fc4b56 7402int unregister_md_personality(struct md_personality *p)
1da177e4 7403{
2604b703 7404 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7405 spin_lock(&pers_lock);
2604b703 7406 list_del_init(&p->list);
1da177e4
LT		 7407 spin_unlock(&pers_lock);
7408 return 0;
7409}
6c144d31 7410EXPORT_SYMBOL(unregister_md_personality);
1da177e4 7411
edb39c9d
GR		 7412int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module)
7413{
7414 if (md_cluster_ops != NULL)
7415 return -EALREADY;
7416 spin_lock(&pers_lock);
7417 md_cluster_ops = ops;
7418 md_cluster_mod = module;
7419 spin_unlock(&pers_lock);
7420 return 0;
7421}
7422EXPORT_SYMBOL(register_md_cluster_operations);
7423
7424int unregister_md_cluster_operations(void)
7425{
7426 spin_lock(&pers_lock);
7427 md_cluster_ops = NULL;
7428 spin_unlock(&pers_lock);
7429 return 0;
7430}
7431EXPORT_SYMBOL(unregister_md_cluster_operations);
7432
7433int md_setup_cluster(struct mddev *mddev, int nodes)
7434{
7435 int err;
7436
7437 err = request_module("md-cluster");
7438 if (err) {
7439 pr_err("md-cluster module not found.\n");
7440 return err;
7441 }
7442
7443 spin_lock(&pers_lock);
7444 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7445 spin_unlock(&pers_lock);
7446 return -ENOENT;
7447 }
7448 spin_unlock(&pers_lock);
7449
cf921cc1 7450 return md_cluster_ops->join(mddev, nodes);
edb39c9d
GR		 7451}
7452
7453void md_cluster_stop(struct mddev *mddev)
7454{
c4ce867f
GR		 7455 if (!md_cluster_ops)
7456 return;
edb39c9d
GR		 7457 md_cluster_ops->leave(mddev);
7458 module_put(md_cluster_mod);
7459}
7460
fd01b88c 7461static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7462{
f72ffdd6 7463 struct md_rdev *rdev;
1da177e4 7464 int idle;
eea1bf38 7465 int curr_events;
1da177e4
LT		 7466
7467 idle = 1;
4b80991c
N		 7468 rcu_read_lock();
7469 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7470 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 7471 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7472 (int)part_stat_read(&disk->part0, sectors[1]) -
7473 atomic_read(&disk->sync_io);
713f6ab1
N		 7474 /* sync IO will cause sync_io to increase before the disk_stats
7475 * as sync_io is counted when a request starts, and
7476 * disk_stats is counted when it completes.
7477 * So resync activity will cause curr_events to be smaller than
7478 * when there was no such activity.
7479 * non-sync IO will cause disk_stat to increase without
7480 * increasing sync_io so curr_events will (eventually)
7481 * be larger than it was before. Once it becomes
7482 * substantially larger, the test below will cause
7483 * the array to appear non-idle, and resync will slow
7484 * down.
7485 * If there is a lot of outstanding resync activity when
7486 * we set last_event to curr_events, then all that activity
7487 * completing might cause the array to appear non-idle
7488 * and resync will be slowed down even though there might
7489 * not have been non-resync activity. This will only
7490 * happen once though. 'last_events' will soon reflect
7491 * the state where there is little or no outstanding
7492 * resync requests, and further resync activity will
7493 * always make curr_events less than last_events.
c0e48521 7494 *
1da177e4 7495 */
eea1bf38 7496 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 7497 rdev->last_events = curr_events;
7498 idle = 0;
7499 }
7500 }
4b80991c 7501 rcu_read_unlock();
1da177e4
LT		 7502 return idle;
7503}
7504
fd01b88c 7505void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 7506{
7507 /* another "blocks" (512byte) blocks have been synced */
7508 atomic_sub(blocks, &mddev->recovery_active);
7509 wake_up(&mddev->recovery_wait);
7510 if (!ok) {
dfc70645 7511 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 7512 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT		 7513 md_wakeup_thread(mddev->thread);
7514 // stop recovery, signal do_sync ....
7515 }
7516}
6c144d31 7517EXPORT_SYMBOL(md_done_sync);
1da177e4 7518
06d91a5f
N		 7519/* md_write_start(mddev, bi)
7520 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 7521 * in superblock) before writing, schedule a superblock update
7522 * and wait for it to complete.
06d91a5f 7523 */
fd01b88c 7524void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7525{
0fd62b86 7526 int did_change = 0;
06d91a5f 7527 if (bio_data_dir(bi) != WRITE)
3d310eb7 7528 return;
06d91a5f 7529
f91de92e
N		 7530 BUG_ON(mddev->ro == 1);
7531 if (mddev->ro == 2) {
7532 /* need to switch to read/write */
7533 mddev->ro = 0;
7534 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7535 md_wakeup_thread(mddev->thread);
25156198 7536 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7537 did_change = 1;
f91de92e 7538 }
06d91a5f 7539 atomic_inc(&mddev->writes_pending);
31a59e34
N		 7540 if (mddev->safemode == 1)
7541 mddev->safemode = 0;
06d91a5f 7542 if (mddev->in_sync) {
85572d7c 7543 spin_lock(&mddev->lock);
3d310eb7
N		 7544 if (mddev->in_sync) {
7545 mddev->in_sync = 0;
850b2b42 7546 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7547 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7548 md_wakeup_thread(mddev->thread);
0fd62b86 7549 did_change = 1;
3d310eb7 7550 }
85572d7c 7551 spin_unlock(&mddev->lock);
06d91a5f 7552 }
0fd62b86 7553 if (did_change)
00bcb4ac 7554 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7555 wait_event(mddev->sb_wait,
09a44cc1 7556 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4 7557}
6c144d31 7558EXPORT_SYMBOL(md_write_start);
1da177e4 7559
fd01b88c 7560void md_write_end(struct mddev *mddev)
1da177e4
LT		 7561{
7562 if (atomic_dec_and_test(&mddev->writes_pending)) {
7563 if (mddev->safemode == 2)
7564 md_wakeup_thread(mddev->thread);
16f17b39 7565 else if (mddev->safemode_delay)
1da177e4
LT		 7566 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7567 }
7568}
6c144d31 7569EXPORT_SYMBOL(md_write_end);
1da177e4 7570
2a2275d6
N		 7571/* md_allow_write(mddev)
7572 * Calling this ensures that the array is marked 'active' so that writes
7573 * may proceed without blocking. It is important to call this before
7574 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7575 * Must be called with mddev_lock held.
b5470dc5
DW		 7576 *
7577 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7578 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7579 */
fd01b88c 7580int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7581{
7582 if (!mddev->pers)
b5470dc5 7583 return 0;
2a2275d6 7584 if (mddev->ro)
b5470dc5 7585 return 0;
1a0fd497 7586 if (!mddev->pers->sync_request)
b5470dc5 7587 return 0;
2a2275d6 7588
85572d7c 7589 spin_lock(&mddev->lock);
2a2275d6
N		 7590 if (mddev->in_sync) {
7591 mddev->in_sync = 0;
7592 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7593 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7594 if (mddev->safemode_delay &&
7595 mddev->safemode == 0)
7596 mddev->safemode = 1;
85572d7c 7597 spin_unlock(&mddev->lock);
293467aa
GR		 7598 if (mddev_is_clustered(mddev))
7599 md_cluster_ops->metadata_update_start(mddev);
2a2275d6 7600 md_update_sb(mddev, 0);
293467aa
GR		 7601 if (mddev_is_clustered(mddev))
7602 md_cluster_ops->metadata_update_finish(mddev);
00bcb4ac 7603 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6 7604 } else
85572d7c 7605 spin_unlock(&mddev->lock);
b5470dc5 7606
070dc6dd 7607 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7608 return -EAGAIN;
7609 else
7610 return 0;
2a2275d6
N		 7611}
7612EXPORT_SYMBOL_GPL(md_allow_write);
7613
1da177e4
LT		 7614#define SYNC_MARKS 10
7615#define SYNC_MARK_STEP (3*HZ)
54f89341 7616#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 7617void md_do_sync(struct md_thread *thread)
1da177e4 7618{
4ed8731d 7619 struct mddev *mddev = thread->mddev;
fd01b88c 7620 struct mddev *mddev2;
1da177e4
LT		 7621 unsigned int currspeed = 0,
7622 window;
ac7e50a3 7623 sector_t max_sectors,j, io_sectors, recovery_done;
1da177e4 7624 unsigned long mark[SYNC_MARKS];
54f89341 7625 unsigned long update_time;
1da177e4
LT		 7626 sector_t mark_cnt[SYNC_MARKS];
7627 int last_mark,m;
7628 struct list_head *tmp;
7629 sector_t last_check;
57afd89f 7630 int skipped = 0;
3cb03002 7631 struct md_rdev *rdev;
c4a39551 7632 char *desc, *action = NULL;
7c2c57c9 7633 struct blk_plug plug;
1da177e4
LT		 7634
7635 /* just incase thread restarts... */
7636 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7637 return;
3991b31e
N		 7638 if (mddev->ro) {/* never try to sync a read-only array */
7639 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5fd6c1dc 7640 return;
3991b31e 7641 }
1da177e4 7642
61df9d91 7643 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
c4a39551 7644 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
61df9d91 7645 desc = "data-check";
c4a39551
JB		 7646 action = "check";
7647 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
61df9d91 7648 desc = "requested-resync";
c4a39551
JB		 7649 action = "repair";
7650 } else
61df9d91
N		 7651 desc = "resync";
7652 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7653 desc = "reshape";
7654 else
7655 desc = "recovery";
7656
c4a39551
JB		 7657 mddev->last_sync_action = action ?: desc;
7658
1da177e4
LT		 7659 /* we overload curr_resync somewhat here.
7660 * 0 == not engaged in resync at all
7661 * 2 == checking that there is no conflict with another sync
7662 * 1 == like 2, but have yielded to allow conflicting resync to
7663 * commense
7664 * other == active in resync - this many blocks
7665 *
7666 * Before starting a resync we must have set curr_resync to
7667 * 2, and then checked that every "conflicting" array has curr_resync
7668 * less than ours. When we find one that is the same or higher
7669 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7670 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7671 * This will mean we have to start checking from the beginning again.
7672 *
7673 */
7674
7675 do {
7676 mddev->curr_resync = 2;
7677
7678 try_again:
404e4b43 7679 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7680 goto skip;
29ac4aa3 7681 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7682 if (mddev2 == mddev)
7683 continue;
90b08710
BS		 7684 if (!mddev->parallel_resync
7685 && mddev2->curr_resync
7686 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7687 DEFINE_WAIT(wq);
7688 if (mddev < mddev2 && mddev->curr_resync == 2) {
7689 /* arbitrarily yield */
7690 mddev->curr_resync = 1;
7691 wake_up(&resync_wait);
7692 }
7693 if (mddev > mddev2 && mddev->curr_resync == 1)
7694 /* no need to wait here, we can wait the next
7695 * time 'round when curr_resync == 2
7696 */
7697 continue;
9744197c
N		 7698 /* We need to wait 'interruptible' so as not to
7699 * contribute to the load average, and not to
7700 * be caught by 'softlockup'
7701 */
7702 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
c91abf5a 7703 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8712e553 7704 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7705 printk(KERN_INFO "md: delaying %s of %s"
7706 " until %s has finished (they"
1da177e4 7707 " share one or more physical units)\n",
61df9d91 7708 desc, mdname(mddev), mdname(mddev2));
1da177e4 7709 mddev_put(mddev2);
9744197c
N		 7710 if (signal_pending(current))
7711 flush_signals(current);
1da177e4
LT		 7712 schedule();
7713 finish_wait(&resync_wait, &wq);
7714 goto try_again;
7715 }
7716 finish_wait(&resync_wait, &wq);
7717 }
7718 }
7719 } while (mddev->curr_resync < 2);
7720
5fd6c1dc 7721 j = 0;
9d88883e 7722 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7723 /* resync follows the size requested by the personality,
57afd89f 7724 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7725 */
7726 max_sectors = mddev->resync_max_sectors;
7f7583d4 7727 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 7728 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7729 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7730 j = mddev->resync_min;
7731 else if (!mddev->bitmap)
5fd6c1dc 7732 j = mddev->recovery_cp;
5e96ee65 7733
ccfcc3c1 7734 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7735 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7736 else {
1da177e4 7737 /* recovery follows the physical size of devices */
58c0fed4 7738 max_sectors = mddev->dev_sectors;
5fd6c1dc 7739 j = MaxSector;
4e59ca7d 7740 rcu_read_lock();
dafb20fa 7741 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc
N		 7742 if (rdev->raid_disk >= 0 &&
7743 !test_bit(Faulty, &rdev->flags) &&
7744 !test_bit(In_sync, &rdev->flags) &&
7745 rdev->recovery_offset < j)
7746 j = rdev->recovery_offset;
4e59ca7d 7747 rcu_read_unlock();
133d4527
N		 7748
7749 /* If there is a bitmap, we need to make sure all
7750 * writes that started before we added a spare
7751 * complete before we start doing a recovery.
7752 * Otherwise the write might complete and (via
7753 * bitmap_endwrite) set a bit in the bitmap after the
7754 * recovery has checked that bit and skipped that
7755 * region.
7756 */
7757 if (mddev->bitmap) {
7758 mddev->pers->quiesce(mddev, 1);
7759 mddev->pers->quiesce(mddev, 0);
7760 }
5fd6c1dc 7761 }
1da177e4 7762
61df9d91
N		 7763 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7764 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7765 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7766 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7767 "(but not more than %d KB/sec) for %s.\n",
7768 speed_max(mddev), desc);
1da177e4 7769
eea1bf38 7770 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7771
57afd89f 7772 io_sectors = 0;
1da177e4
LT		 7773 for (m = 0; m < SYNC_MARKS; m++) {
7774 mark[m] = jiffies;
57afd89f 7775 mark_cnt[m] = io_sectors;
1da177e4
LT		 7776 }
7777 last_mark = 0;
7778 mddev->resync_mark = mark[last_mark];
7779 mddev->resync_mark_cnt = mark_cnt[last_mark];
7780
7781 /*
7782 * Tune reconstruction:
7783 */
7784 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7785 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7786 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7787
7788 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7789 last_check = 0;
7790
7791 if (j>2) {
c91abf5a 7792 printk(KERN_INFO
61df9d91
N		 7793 "md: resuming %s of %s from checkpoint.\n",
7794 desc, mdname(mddev));
1da177e4 7795 mddev->curr_resync = j;
72f36d59
N		 7796 } else
7797 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 7798 mddev->curr_resync_completed = j;
72f36d59
N		 7799 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7800 md_new_event(mddev);
54f89341 7801 update_time = jiffies;
1da177e4 7802
96ae923a 7803 if (mddev_is_clustered(mddev))
965400eb 7804 md_cluster_ops->resync_start(mddev, j, max_sectors);
96ae923a 7805
7c2c57c9 7806 blk_start_plug(&plug);
1da177e4 7807 while (j < max_sectors) {
57afd89f 7808 sector_t sectors;
1da177e4 7809
57afd89f 7810 skipped = 0;
97e4f42d 7811
7a91ee1f
N		 7812 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7813 ((mddev->curr_resync > mddev->curr_resync_completed &&
7814 (mddev->curr_resync - mddev->curr_resync_completed)
7815 > (max_sectors >> 4)) ||
54f89341 7816 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f
N		 7817 (j - mddev->curr_resync_completed)*2
7818 >= mddev->resync_max - mddev->curr_resync_completed
7819 )) {
97e4f42d 7820 /* time to update curr_resync_completed */
97e4f42d
N		 7821 wait_event(mddev->recovery_wait,
7822 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7823 mddev->curr_resync_completed = j;
35d78c66 7824 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7825 j > mddev->recovery_cp)
7826 mddev->recovery_cp = j;
54f89341 7827 update_time = jiffies;
070dc6dd 7828 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7829 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7830 }
acb180b0 7831
c91abf5a
N		 7832 while (j >= mddev->resync_max &&
7833 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
e62e58a5
N		 7834 /* As this condition is controlled by user-space,
7835 * we can block indefinitely, so use '_interruptible'
7836 * to avoid triggering warnings.
7837 */
7838 flush_signals(current); /* just in case */
7839 wait_event_interruptible(mddev->recovery_wait,
7840 mddev->resync_max > j
c91abf5a
N		 7841 || test_bit(MD_RECOVERY_INTR,
7842 &mddev->recovery));
e62e58a5 7843 }
acb180b0 7844
c91abf5a
N		 7845 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7846 break;
acb180b0 7847
09314799 7848 sectors = mddev->pers->sync_request(mddev, j, &skipped);
57afd89f 7849 if (sectors == 0) {
dfc70645 7850 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
c91abf5a 7851 break;
1da177e4 7852 }
57afd89f
N		 7853
7854 if (!skipped) { /* actual IO requested */
7855 io_sectors += sectors;
7856 atomic_add(sectors, &mddev->recovery_active);
7857 }
7858
e875ecea
N		 7859 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7860 break;
7861
1da177e4 7862 j += sectors;
72f36d59
N		 7863 if (j > 2)
7864 mddev->curr_resync = j;
96ae923a
GR		 7865 if (mddev_is_clustered(mddev))
7866 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
ff4e8d9a 7867 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7868 if (last_check == 0)
e875ecea 7869 /* this is the earliest that rebuild will be
d7603b7e
N		 7870 * visible in /proc/mdstat
7871 */
7872 md_new_event(mddev);
57afd89f
N		 7873
7874 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7875 continue;
7876
57afd89f 7877 last_check = io_sectors;
1da177e4
LT		 7878 repeat:
7879 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7880 /* step marks */
7881 int next = (last_mark+1) % SYNC_MARKS;
7882
7883 mddev->resync_mark = mark[next];
7884 mddev->resync_mark_cnt = mark_cnt[next];
7885 mark[next] = jiffies;
57afd89f 7886 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7887 last_mark = next;
7888 }
7889
c91abf5a
N		 7890 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7891 break;
1da177e4
LT		 7892
7893 /*
7894 * this loop exits only if either when we are slower than
7895 * the 'hard' speed limit, or the system was IO-idle for
7896 * a jiffy.
7897 * the system might be non-idle CPU-wise, but we only care
7898 * about not overloading the IO subsystem. (things like an
7899 * e2fsck being done on the RAID array should execute fast)
7900 */
1da177e4
LT		 7901 cond_resched();
7902
ac7e50a3
XN		 7903 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7904 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
57afd89f 7905 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7906
88202a0c 7907 if (currspeed > speed_min(mddev)) {
ac8fa419 7908 if (currspeed > speed_max(mddev)) {
c0e48521 7909 msleep(500);
1da177e4
LT		 7910 goto repeat;
7911 }
ac8fa419
N		 7912 if (!is_mddev_idle(mddev, 0)) {
7913 /*
7914 * Give other IO more of a chance.
7915 * The faster the devices, the less we wait.
7916 */
7917 wait_event(mddev->recovery_wait,
7918 !atomic_read(&mddev->recovery_active));
7919 }
1da177e4
LT		 7920 }
7921 }
c91abf5a
N		 7922 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7923 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7924 ? "interrupted" : "done");
1da177e4
LT		 7925 /*
7926 * this also signals 'finished resyncing' to md_stop
7927 */
7c2c57c9 7928 blk_finish_plug(&plug);
1da177e4
LT		 7929 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7930
7931 /* tell personality that we are finished */
09314799 7932 mddev->pers->sync_request(mddev, max_sectors, &skipped);
1da177e4 7933
96ae923a 7934 if (mddev_is_clustered(mddev))
965400eb 7935 md_cluster_ops->resync_finish(mddev);
96ae923a 7936
dfc70645 7937 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7938 mddev->curr_resync > 2) {
7939 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7940 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7941 if (mddev->curr_resync >= mddev->recovery_cp) {
7942 printk(KERN_INFO
61df9d91
N		 7943 "md: checkpointing %s of %s.\n",
7944 desc, mdname(mddev));
0a19caab 7945 if (test_bit(MD_RECOVERY_ERROR,
7946 &mddev->recovery))
7947 mddev->recovery_cp =
7948 mddev->curr_resync_completed;
7949 else
7950 mddev->recovery_cp =
7951 mddev->curr_resync;
5fd6c1dc
N		 7952 }
7953 } else
7954 mddev->recovery_cp = MaxSector;
7955 } else {
7956 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7957 mddev->curr_resync = MaxSector;
4e59ca7d 7958 rcu_read_lock();
dafb20fa 7959 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7960 if (rdev->raid_disk >= 0 &&
70fffd0b 7961 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7962 !test_bit(Faulty, &rdev->flags) &&
7963 !test_bit(In_sync, &rdev->flags) &&
7964 rdev->recovery_offset < mddev->curr_resync)
7965 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7966 rcu_read_unlock();
5fd6c1dc 7967 }
1da177e4 7968 }
db91ff55 7969 skip:
17571284 7970 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7971
23da422b 7972 spin_lock(&mddev->lock);
c07b70ad
N		 7973 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7974 /* We completed so min/max setting can be forgotten if used. */
7975 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7976 mddev->resync_min = 0;
7977 mddev->resync_max = MaxSector;
7978 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7979 mddev->resync_min = mddev->curr_resync_completed;
1da177e4 7980 mddev->curr_resync = 0;
23da422b
N		 7981 spin_unlock(&mddev->lock);
7982
1da177e4
LT		 7983 wake_up(&resync_wait);
7984 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7985 md_wakeup_thread(mddev->thread);
c6207277 7986 return;
1da177e4 7987}
29269553 7988EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7989
746d3207
N		 7990static int remove_and_add_spares(struct mddev *mddev,
7991 struct md_rdev *this)
b4c4c7b8 7992{
3cb03002 7993 struct md_rdev *rdev;
b4c4c7b8 7994 int spares = 0;
f2a371c5 7995 int removed = 0;
b4c4c7b8 7996
dafb20fa 7997 rdev_for_each(rdev, mddev)
746d3207
N		 7998 if ((this == NULL || rdev == this) &&
7999 rdev->raid_disk >= 0 &&
6bfe0b49 8000 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 8001 (test_bit(Faulty, &rdev->flags) ||
8002 ! test_bit(In_sync, &rdev->flags)) &&
8003 atomic_read(&rdev->nr_pending)==0) {
8004 if (mddev->pers->hot_remove_disk(
b8321b68 8005 mddev, rdev) == 0) {
36fad858 8006 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 8007 rdev->raid_disk = -1;
f2a371c5 8008 removed++;
b4c4c7b8
N		 8009 }
8010 }
90584fc9
JB		 8011 if (removed && mddev->kobj.sd)
8012 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 8013
746d3207
N		 8014 if (this)
8015 goto no_add;
8016
dafb20fa 8017 rdev_for_each(rdev, mddev) {
7bfec5f3
N		 8018 if (rdev->raid_disk >= 0 &&
8019 !test_bit(In_sync, &rdev->flags) &&
8020 !test_bit(Faulty, &rdev->flags))
8021 spares++;
7ceb17e8
N		 8022 if (rdev->raid_disk >= 0)
8023 continue;
8024 if (test_bit(Faulty, &rdev->flags))
8025 continue;
8026 if (mddev->ro &&
8313b8e5
N		 8027 ! (rdev->saved_raid_disk >= 0 &&
8028 !test_bit(Bitmap_sync, &rdev->flags)))
7ceb17e8
N		 8029 continue;
8030
7eb41885
N		 8031 if (rdev->saved_raid_disk < 0)
8032 rdev->recovery_offset = 0;
7ceb17e8
N		 8033 if (mddev->pers->
8034 hot_add_disk(mddev, rdev) == 0) {
8035 if (sysfs_link_rdev(mddev, rdev))
8036 /* failure here is OK */;
8037 spares++;
8038 md_new_event(mddev);
8039 set_bit(MD_CHANGE_DEVS, &mddev->flags);
dfc70645 8040 }
b4c4c7b8 8041 }
746d3207 8042no_add:
6dafab6b
N		 8043 if (removed)
8044 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 8045 return spares;
8046}
7ebc0be7 8047
ac05f256
N		 8048static void md_start_sync(struct work_struct *ws)
8049{
8050 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8051
8052 mddev->sync_thread = md_register_thread(md_do_sync,
8053 mddev,
8054 "resync");
8055 if (!mddev->sync_thread) {
8056 printk(KERN_ERR "%s: could not start resync"
8057 " thread...\n",
8058 mdname(mddev));
8059 /* leave the spares where they are, it shouldn't hurt */
8060 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8061 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8062 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8063 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8064 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8065 wake_up(&resync_wait);
ac05f256
N		 8066 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8067 &mddev->recovery))
8068 if (mddev->sysfs_action)
8069 sysfs_notify_dirent_safe(mddev->sysfs_action);
8070 } else
8071 md_wakeup_thread(mddev->sync_thread);
8072 sysfs_notify_dirent_safe(mddev->sysfs_action);
8073 md_new_event(mddev);
8074}
8075
1da177e4
LT		 8076/*
8077 * This routine is regularly called by all per-raid-array threads to
8078 * deal with generic issues like resync and super-block update.
8079 * Raid personalities that don't have a thread (linear/raid0) do not
8080 * need this as they never do any recovery or update the superblock.
8081 *
8082 * It does not do any resync itself, but rather "forks" off other threads
8083 * to do that as needed.
8084 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8085 * "->recovery" and create a thread at ->sync_thread.
dfc70645 8086 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 8087 * and wakeups up this thread which will reap the thread and finish up.
8088 * This thread also removes any faulty devices (with nr_pending == 0).
8089 *
8090 * The overall approach is:
8091 * 1/ if the superblock needs updating, update it.
8092 * 2/ If a recovery thread is running, don't do anything else.
8093 * 3/ If recovery has finished, clean up, possibly marking spares active.
8094 * 4/ If there are any faulty devices, remove them.
8095 * 5/ If array is degraded, try to add spares devices
8096 * 6/ If array has spares or is not in-sync, start a resync thread.
8097 */
fd01b88c 8098void md_check_recovery(struct mddev *mddev)
1da177e4 8099{
68866e42
JB		 8100 if (mddev->suspended)
8101 return;
8102
5f40402d 8103 if (mddev->bitmap)
aa5cbd10 8104 bitmap_daemon_work(mddev);
1da177e4 8105
fca4d848 8106 if (signal_pending(current)) {
31a59e34 8107 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 8108 printk(KERN_INFO "md: %s in immediate safe mode\n",
8109 mdname(mddev));
8110 mddev->safemode = 2;
8111 }
8112 flush_signals(current);
8113 }
8114
c89a8eee
N		 8115 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8116 return;
1da177e4 8117 if ( ! (
142d44c3 8118 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 8119 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 8120 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 8121 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 8122 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8123 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 8124 ))
8125 return;
fca4d848 8126
df5b89b3 8127 if (mddev_trylock(mddev)) {
b4c4c7b8 8128 int spares = 0;
fca4d848 8129
c89a8eee 8130 if (mddev->ro) {
7ceb17e8
N		 8131 /* On a read-only array we can:
8132 * - remove failed devices
8133 * - add already-in_sync devices if the array itself
8134 * is in-sync.
8135 * As we only add devices that are already in-sync,
8136 * we can activate the spares immediately.
c89a8eee 8137 */
7ceb17e8 8138 remove_and_add_spares(mddev, NULL);
8313b8e5
N		 8139 /* There is no thread, but we need to call
8140 * ->spare_active and clear saved_raid_disk
8141 */
2ac295a5 8142 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8313b8e5
N		 8143 md_reap_sync_thread(mddev);
8144 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
c89a8eee
N		 8145 goto unlock;
8146 }
8147
31a59e34 8148 if (!mddev->external) {
0fd62b86 8149 int did_change = 0;
85572d7c 8150 spin_lock(&mddev->lock);
31a59e34
N		 8151 if (mddev->safemode &&
8152 !atomic_read(&mddev->writes_pending) &&
8153 !mddev->in_sync &&
8154 mddev->recovery_cp == MaxSector) {
8155 mddev->in_sync = 1;
0fd62b86 8156 did_change = 1;
070dc6dd 8157 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 8158 }
8159 if (mddev->safemode == 1)
8160 mddev->safemode = 0;
85572d7c 8161 spin_unlock(&mddev->lock);
0fd62b86 8162 if (did_change)
00bcb4ac 8163 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 8164 }
fca4d848 8165
293467aa
GR		 8166 if (mddev->flags & MD_UPDATE_SB_FLAGS) {
8167 if (mddev_is_clustered(mddev))
8168 md_cluster_ops->metadata_update_start(mddev);
850b2b42 8169 md_update_sb(mddev, 0);
293467aa
GR		 8170 if (mddev_is_clustered(mddev))
8171 md_cluster_ops->metadata_update_finish(mddev);
8172 }
06d91a5f 8173
1da177e4
LT		 8174 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8175 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8176 /* resync/recovery still happening */
8177 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8178 goto unlock;
8179 }
8180 if (mddev->sync_thread) {
a91d5ac0 8181 md_reap_sync_thread(mddev);
1da177e4
LT		 8182 goto unlock;
8183 }
72a23c21
NB		 8184 /* Set RUNNING before clearing NEEDED to avoid
8185 * any transients in the value of "sync_action".
8186 */
72f36d59 8187 mddev->curr_resync_completed = 0;
23da422b 8188 spin_lock(&mddev->lock);
72a23c21 8189 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
23da422b 8190 spin_unlock(&mddev->lock);
24dd469d
N		 8191 /* Clear some bits that don't mean anything, but
8192 * might be left set
8193 */
24dd469d
N		 8194 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8195 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 8196
ed209584
N		 8197 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8198 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
ac05f256 8199 goto not_running;
1da177e4
LT		 8200 /* no recovery is running.
8201 * remove any failed drives, then
8202 * add spares if possible.
72f36d59 8203 * Spares are also removed and re-added, to allow
1da177e4
LT		 8204 * the personality to fail the re-add.
8205 */
1da177e4 8206
b4c4c7b8 8207 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 8208 if (mddev->pers->check_reshape == NULL ||
8209 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8 8210 /* Cannot proceed */
ac05f256 8211 goto not_running;
b4c4c7b8 8212 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 8213 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 8214 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N		 8215 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8216 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 8217 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 8218 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8219 } else if (mddev->recovery_cp < MaxSector) {
8220 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 8221 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8222 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8223 /* nothing to be done ... */
ac05f256 8224 goto not_running;
24dd469d 8225
1da177e4 8226 if (mddev->pers->sync_request) {
ef99bf48 8227 if (spares) {
a654b9d8
N		 8228 /* We are adding a device or devices to an array
8229 * which has the bitmap stored on all devices.
8230 * So make sure all bitmap pages get written
8231 */
8232 bitmap_write_all(mddev->bitmap);
8233 }
ac05f256
N		 8234 INIT_WORK(&mddev->del_work, md_start_sync);
8235 queue_work(md_misc_wq, &mddev->del_work);
8236 goto unlock;
1da177e4 8237 }
ac05f256 8238 not_running:
72a23c21
NB		 8239 if (!mddev->sync_thread) {
8240 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8241 wake_up(&resync_wait);
72a23c21
NB		 8242 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8243 &mddev->recovery))
0c3573f1 8244 if (mddev->sysfs_action)
00bcb4ac 8245 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 8246 }
ac05f256
N		 8247 unlock:
8248 wake_up(&mddev->sb_wait);
1da177e4
LT		 8249 mddev_unlock(mddev);
8250 }
8251}
6c144d31 8252EXPORT_SYMBOL(md_check_recovery);
1da177e4 8253
a91d5ac0
JB		 8254void md_reap_sync_thread(struct mddev *mddev)
8255{
8256 struct md_rdev *rdev;
8257
8258 /* resync has finished, collect result */
8259 md_unregister_thread(&mddev->sync_thread);
8260 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8261 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8262 /* success...*/
8263 /* activate any spares */
8264 if (mddev->pers->spare_active(mddev)) {
8265 sysfs_notify(&mddev->kobj, NULL,
8266 "degraded");
8267 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8268 }
8269 }
293467aa
GR		 8270 if (mddev_is_clustered(mddev))
8271 md_cluster_ops->metadata_update_start(mddev);
a91d5ac0
JB		 8272 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8273 mddev->pers->finish_reshape)
8274 mddev->pers->finish_reshape(mddev);
8275
8276 /* If array is no-longer degraded, then any saved_raid_disk
f466722c 8277 * information must be scrapped.
a91d5ac0 8278 */
f466722c
N		 8279 if (!mddev->degraded)
8280 rdev_for_each(rdev, mddev)
a91d5ac0
JB		 8281 rdev->saved_raid_disk = -1;
8282
8283 md_update_sb(mddev, 1);
293467aa
GR		 8284 if (mddev_is_clustered(mddev))
8285 md_cluster_ops->metadata_update_finish(mddev);
a91d5ac0 8286 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
ea358cd0 8287 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
a91d5ac0
JB		 8288 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8289 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8290 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8291 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
f851b60d 8292 wake_up(&resync_wait);
a91d5ac0
JB		 8293 /* flag recovery needed just to double check */
8294 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8295 sysfs_notify_dirent_safe(mddev->sysfs_action);
8296 md_new_event(mddev);
8297 if (mddev->event_work.func)
8298 queue_work(md_misc_wq, &mddev->event_work);
8299}
6c144d31 8300EXPORT_SYMBOL(md_reap_sync_thread);
a91d5ac0 8301
fd01b88c 8302void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 8303{
00bcb4ac 8304 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 8305 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 8306 !test_bit(Blocked, &rdev->flags) &&
8307 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 8308 msecs_to_jiffies(5000));
8309 rdev_dec_pending(rdev, mddev);
8310}
8311EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8312
c6563a8c
N		 8313void md_finish_reshape(struct mddev *mddev)
8314{
8315 /* called be personality module when reshape completes. */
8316 struct md_rdev *rdev;
8317
8318 rdev_for_each(rdev, mddev) {
8319 if (rdev->data_offset > rdev->new_data_offset)
8320 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8321 else
8322 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8323 rdev->data_offset = rdev->new_data_offset;
8324 }
8325}
8326EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N		 8327
8328/* Bad block management.
8329 * We can record which blocks on each device are 'bad' and so just
8330 * fail those blocks, or that stripe, rather than the whole device.
8331 * Entries in the bad-block table are 64bits wide. This comprises:
8332 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8333 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8334 * A 'shift' can be set so that larger blocks are tracked and
8335 * consequently larger devices can be covered.
8336 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8337 *
8338 * Locking of the bad-block table uses a seqlock so md_is_badblock
8339 * might need to retry if it is very unlucky.
8340 * We will sometimes want to check for bad blocks in a bi_end_io function,
8341 * so we use the write_seqlock_irq variant.
8342 *
8343 * When looking for a bad block we specify a range and want to
8344 * know if any block in the range is bad. So we binary-search
8345 * to the last range that starts at-or-before the given endpoint,
8346 * (or "before the sector after the target range")
8347 * then see if it ends after the given start.
8348 * We return
8349 * 0 if there are no known bad blocks in the range
8350 * 1 if there are known bad block which are all acknowledged
8351 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8352 * plus the start/length of the first bad section we overlap.
8353 */
8354int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8355 sector_t *first_bad, int *bad_sectors)
8356{
8357 int hi;
ab05613a 8358 int lo;
2230dfe4 8359 u64 *p = bb->page;
ab05613a 8360 int rv;
2230dfe4
N		 8361 sector_t target = s + sectors;
8362 unsigned seq;
8363
8364 if (bb->shift > 0) {
8365 /* round the start down, and the end up */
8366 s >>= bb->shift;
8367 target += (1<<bb->shift) - 1;
8368 target >>= bb->shift;
8369 sectors = target - s;
8370 }
8371 /* 'target' is now the first block after the bad range */
8372
8373retry:
8374 seq = read_seqbegin(&bb->lock);
ab05613a 8375 lo = 0;
8376 rv = 0;
2230dfe4
N		 8377 hi = bb->count;
8378
8379 /* Binary search between lo and hi for 'target'
8380 * i.e. for the last range that starts before 'target'
8381 */
8382 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8383 * are known not to be the last range before target.
8384 * VARIANT: hi-lo is the number of possible
8385 * ranges, and decreases until it reaches 1
8386 */
8387 while (hi - lo > 1) {
8388 int mid = (lo + hi) / 2;
8389 sector_t a = BB_OFFSET(p[mid]);
8390 if (a < target)
8391 /* This could still be the one, earlier ranges
8392 * could not. */
8393 lo = mid;
8394 else
8395 /* This and later ranges are definitely out. */
8396 hi = mid;
8397 }
8398 /* 'lo' might be the last that started before target, but 'hi' isn't */
8399 if (hi > lo) {
8400 /* need to check all range that end after 's' to see if
8401 * any are unacknowledged.
8402 */
8403 while (lo >= 0 &&
8404 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8405 if (BB_OFFSET(p[lo]) < target) {
8406 /* starts before the end, and finishes after
8407 * the start, so they must overlap
8408 */
8409 if (rv != -1 && BB_ACK(p[lo]))
8410 rv = 1;
8411 else
8412 rv = -1;
8413 *first_bad = BB_OFFSET(p[lo]);
8414 *bad_sectors = BB_LEN(p[lo]);
8415 }
8416 lo--;
8417 }
8418 }
8419
8420 if (read_seqretry(&bb->lock, seq))
8421 goto retry;
8422
8423 return rv;
8424}
8425EXPORT_SYMBOL_GPL(md_is_badblock);
8426
8427/*
8428 * Add a range of bad blocks to the table.
8429 * This might extend the table, or might contract it
8430 * if two adjacent ranges can be merged.
8431 * We binary-search to find the 'insertion' point, then
8432 * decide how best to handle it.
8433 */
8434static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8435 int acknowledged)
8436{
8437 u64 *p;
8438 int lo, hi;
8439 int rv = 1;
905b0297 8440 unsigned long flags;
2230dfe4
N		 8441
8442 if (bb->shift < 0)
8443 /* badblocks are disabled */
8444 return 0;
8445
8446 if (bb->shift) {
8447 /* round the start down, and the end up */
8448 sector_t next = s + sectors;
8449 s >>= bb->shift;
8450 next += (1<<bb->shift) - 1;
8451 next >>= bb->shift;
8452 sectors = next - s;
8453 }
8454
905b0297 8455 write_seqlock_irqsave(&bb->lock, flags);
2230dfe4
N		 8456
8457 p = bb->page;
8458 lo = 0;
8459 hi = bb->count;
8460 /* Find the last range that starts at-or-before 's' */
8461 while (hi - lo > 1) {
8462 int mid = (lo + hi) / 2;
8463 sector_t a = BB_OFFSET(p[mid]);
8464 if (a <= s)
8465 lo = mid;
8466 else
8467 hi = mid;
8468 }
8469 if (hi > lo && BB_OFFSET(p[lo]) > s)
8470 hi = lo;
8471
8472 if (hi > lo) {
8473 /* we found a range that might merge with the start
8474 * of our new range
8475 */
8476 sector_t a = BB_OFFSET(p[lo]);
8477 sector_t e = a + BB_LEN(p[lo]);
8478 int ack = BB_ACK(p[lo]);
8479 if (e >= s) {
8480 /* Yes, we can merge with a previous range */
8481 if (s == a && s + sectors >= e)
8482 /* new range covers old */
8483 ack = acknowledged;
8484 else
8485 ack = ack && acknowledged;
8486
8487 if (e < s + sectors)
8488 e = s + sectors;
8489 if (e - a <= BB_MAX_LEN) {
8490 p[lo] = BB_MAKE(a, e-a, ack);
8491 s = e;
8492 } else {
8493 /* does not all fit in one range,
8494 * make p[lo] maximal
8495 */
8496 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8497 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8498 s = a + BB_MAX_LEN;
8499 }
8500 sectors = e - s;
8501 }
8502 }
8503 if (sectors && hi < bb->count) {
8504 /* 'hi' points to the first range that starts after 's'.
8505 * Maybe we can merge with the start of that range */
8506 sector_t a = BB_OFFSET(p[hi]);
8507 sector_t e = a + BB_LEN(p[hi]);
8508 int ack = BB_ACK(p[hi]);
8509 if (a <= s + sectors) {
8510 /* merging is possible */
8511 if (e <= s + sectors) {
8512 /* full overlap */
8513 e = s + sectors;
8514 ack = acknowledged;
8515 } else
8516 ack = ack && acknowledged;
8517
8518 a = s;
8519 if (e - a <= BB_MAX_LEN) {
8520 p[hi] = BB_MAKE(a, e-a, ack);
8521 s = e;
8522 } else {
8523 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8524 s = a + BB_MAX_LEN;
8525 }
8526 sectors = e - s;
8527 lo = hi;
8528 hi++;
8529 }
8530 }
8531 if (sectors == 0 && hi < bb->count) {
8532 /* we might be able to combine lo and hi */
8533 /* Note: 's' is at the end of 'lo' */
8534 sector_t a = BB_OFFSET(p[hi]);
8535 int lolen = BB_LEN(p[lo]);
8536 int hilen = BB_LEN(p[hi]);
8537 int newlen = lolen + hilen - (s - a);
8538 if (s >= a && newlen < BB_MAX_LEN) {
8539 /* yes, we can combine them */
8540 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8541 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8542 memmove(p + hi, p + hi + 1,
8543 (bb->count - hi - 1) * 8);
8544 bb->count--;
8545 }
8546 }
8547 while (sectors) {
8548 /* didn't merge (it all).
8549 * Need to add a range just before 'hi' */
8550 if (bb->count >= MD_MAX_BADBLOCKS) {
8551 /* No room for more */
8552 rv = 0;
8553 break;
8554 } else {
8555 int this_sectors = sectors;
8556 memmove(p + hi + 1, p + hi,
8557 (bb->count - hi) * 8);
8558 bb->count++;
8559
8560 if (this_sectors > BB_MAX_LEN)
8561 this_sectors = BB_MAX_LEN;
8562 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8563 sectors -= this_sectors;
8564 s += this_sectors;
8565 }
8566 }
8567
8568 bb->changed = 1;
de393cde
N		 8569 if (!acknowledged)
8570 bb->unacked_exist = 1;
905b0297 8571 write_sequnlock_irqrestore(&bb->lock, flags);
2230dfe4
N		 8572
8573 return rv;
8574}
8575
3cb03002 8576int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8577 int is_new)
2230dfe4 8578{
c6563a8c
N		 8579 int rv;
8580 if (is_new)
8581 s += rdev->new_data_offset;
8582 else
8583 s += rdev->data_offset;
8584 rv = md_set_badblocks(&rdev->badblocks,
8585 s, sectors, 0);
2230dfe4
N		 8586 if (rv) {
8587 /* Make sure they get written out promptly */
8bd2f0a0 8588 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 8589 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8590 md_wakeup_thread(rdev->mddev->thread);
8591 }
8592 return rv;
8593}
8594EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8595
8596/*
8597 * Remove a range of bad blocks from the table.
8598 * This may involve extending the table if we spilt a region,
8599 * but it must not fail. So if the table becomes full, we just
8600 * drop the remove request.
8601 */
8602static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8603{
8604 u64 *p;
8605 int lo, hi;
8606 sector_t target = s + sectors;
8607 int rv = 0;
8608
8609 if (bb->shift > 0) {
8610 /* When clearing we round the start up and the end down.
8611 * This should not matter as the shift should align with
8612 * the block size and no rounding should ever be needed.
8613 * However it is better the think a block is bad when it
8614 * isn't than to think a block is not bad when it is.
8615 */
8616 s += (1<<bb->shift) - 1;
8617 s >>= bb->shift;
8618 target >>= bb->shift;
8619 sectors = target - s;
8620 }
8621
8622 write_seqlock_irq(&bb->lock);
8623
8624 p = bb->page;
8625 lo = 0;
8626 hi = bb->count;
8627 /* Find the last range that starts before 'target' */
8628 while (hi - lo > 1) {
8629 int mid = (lo + hi) / 2;
8630 sector_t a = BB_OFFSET(p[mid]);
8631 if (a < target)
8632 lo = mid;
8633 else
8634 hi = mid;
8635 }
8636 if (hi > lo) {
8637 /* p[lo] is the last range that could overlap the
8638 * current range. Earlier ranges could also overlap,
8639 * but only this one can overlap the end of the range.
8640 */
8641 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8642 /* Partial overlap, leave the tail of this range */
8643 int ack = BB_ACK(p[lo]);
8644 sector_t a = BB_OFFSET(p[lo]);
8645 sector_t end = a + BB_LEN(p[lo]);
8646
8647 if (a < s) {
8648 /* we need to split this range */
8649 if (bb->count >= MD_MAX_BADBLOCKS) {
8b32bf5e 8650 rv = -ENOSPC;
2230dfe4
N		 8651 goto out;
8652 }
8653 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8654 bb->count++;
8655 p[lo] = BB_MAKE(a, s-a, ack);
8656 lo++;
8657 }
8658 p[lo] = BB_MAKE(target, end - target, ack);
8659 /* there is no longer an overlap */
8660 hi = lo;
8661 lo--;
8662 }
8663 while (lo >= 0 &&
8664 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8665 /* This range does overlap */
8666 if (BB_OFFSET(p[lo]) < s) {
8667 /* Keep the early parts of this range. */
8668 int ack = BB_ACK(p[lo]);
8669 sector_t start = BB_OFFSET(p[lo]);
8670 p[lo] = BB_MAKE(start, s - start, ack);
8671 /* now low doesn't overlap, so.. */
8672 break;
8673 }
8674 lo--;
8675 }
8676 /* 'lo' is strictly before, 'hi' is strictly after,
8677 * anything between needs to be discarded
8678 */
8679 if (hi - lo > 1) {
8680 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8681 bb->count -= (hi - lo - 1);
8682 }
8683 }
8684
8685 bb->changed = 1;
8686out:
8687 write_sequnlock_irq(&bb->lock);
8688 return rv;
8689}
8690
c6563a8c
N		 8691int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8692 int is_new)
2230dfe4 8693{
c6563a8c
N		 8694 if (is_new)
8695 s += rdev->new_data_offset;
8696 else
8697 s += rdev->data_offset;
2230dfe4 8698 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8699 s, sectors);
2230dfe4
N		 8700}
8701EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8702
8703/*
8704 * Acknowledge all bad blocks in a list.
8705 * This only succeeds if ->changed is clear. It is used by
8706 * in-kernel metadata updates
8707 */
8708void md_ack_all_badblocks(struct badblocks *bb)
8709{
8710 if (bb->page == NULL || bb->changed)
8711 /* no point even trying */
8712 return;
8713 write_seqlock_irq(&bb->lock);
8714
ecb178bb 8715 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N		 8716 u64 *p = bb->page;
8717 int i;
8718 for (i = 0; i < bb->count ; i++) {
8719 if (!BB_ACK(p[i])) {
8720 sector_t start = BB_OFFSET(p[i]);
8721 int len = BB_LEN(p[i]);
8722 p[i] = BB_MAKE(start, len, 1);
8723 }
8724 }
de393cde 8725 bb->unacked_exist = 0;
2230dfe4
N		 8726 }
8727 write_sequnlock_irq(&bb->lock);
8728}
8729EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8730
16c791a5
N		 8731/* sysfs access to bad-blocks list.
8732 * We present two files.
8733 * 'bad-blocks' lists sector numbers and lengths of ranges that
8734 * are recorded as bad. The list is truncated to fit within
8735 * the one-page limit of sysfs.
8736 * Writing "sector length" to this file adds an acknowledged
8737 * bad block list.
8738 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8739 * been acknowledged. Writing to this file adds bad blocks
8740 * without acknowledging them. This is largely for testing.
8741 */
8742
8743static ssize_t
8744badblocks_show(struct badblocks *bb, char *page, int unack)
8745{
8746 size_t len;
8747 int i;
8748 u64 *p = bb->page;
8749 unsigned seq;
8750
8751 if (bb->shift < 0)
8752 return 0;
8753
8754retry:
8755 seq = read_seqbegin(&bb->lock);
8756
8757 len = 0;
8758 i = 0;
8759
8760 while (len < PAGE_SIZE && i < bb->count) {
8761 sector_t s = BB_OFFSET(p[i]);
8762 unsigned int length = BB_LEN(p[i]);
8763 int ack = BB_ACK(p[i]);
8764 i++;
8765
8766 if (unack && ack)
8767 continue;
8768
8769 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8770 (unsigned long long)s << bb->shift,
8771 length << bb->shift);
8772 }
de393cde
N		 8773 if (unack && len == 0)
8774 bb->unacked_exist = 0;
16c791a5
N		 8775
8776 if (read_seqretry(&bb->lock, seq))
8777 goto retry;
8778
8779 return len;
8780}
8781
8782#define DO_DEBUG 1
8783
8784static ssize_t
8785badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8786{
8787 unsigned long long sector;
8788 int length;
8789 char newline;
8790#ifdef DO_DEBUG
8791 /* Allow clearing via sysfs *only* for testing/debugging.
8792 * Normally only a successful write may clear a badblock
8793 */
8794 int clear = 0;
8795 if (page[0] == '-') {
8796 clear = 1;
8797 page++;
8798 }
8799#endif /* DO_DEBUG */
8800
8801 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8802 case 3:
8803 if (newline != '\n')
8804 return -EINVAL;
8805 case 2:
8806 if (length <= 0)
8807 return -EINVAL;
8808 break;
8809 default:
8810 return -EINVAL;
8811 }
8812
8813#ifdef DO_DEBUG
8814 if (clear) {
8815 md_clear_badblocks(bb, sector, length);
8816 return len;
8817 }
8818#endif /* DO_DEBUG */
8819 if (md_set_badblocks(bb, sector, length, !unack))
8820 return len;
8821 else
8822 return -ENOSPC;
8823}
8824
75c96f85
AB		 8825static int md_notify_reboot(struct notifier_block *this,
8826 unsigned long code, void *x)
1da177e4
LT		 8827{
8828 struct list_head *tmp;
fd01b88c 8829 struct mddev *mddev;
2dba6a91 8830 int need_delay = 0;
1da177e4 8831
c744a65c
N		 8832 for_each_mddev(mddev, tmp) {
8833 if (mddev_trylock(mddev)) {
30b8aa91
N		 8834 if (mddev->pers)
8835 __md_stop_writes(mddev);
0f62fb22
N		 8836 if (mddev->persistent)
8837 mddev->safemode = 2;
c744a65c 8838 mddev_unlock(mddev);
2dba6a91 8839 }
c744a65c 8840 need_delay = 1;
1da177e4 8841 }
c744a65c
N		 8842 /*
8843 * certain more exotic SCSI devices are known to be
8844 * volatile wrt too early system reboots. While the
8845 * right place to handle this issue is the given
8846 * driver, we do want to have a safe RAID driver ...
8847 */
8848 if (need_delay)
8849 mdelay(1000*1);
8850
1da177e4
LT		 8851 return NOTIFY_DONE;
8852}
8853
75c96f85 8854static struct notifier_block md_notifier = {
1da177e4
LT		 8855 .notifier_call = md_notify_reboot,
8856 .next = NULL,
8857 .priority = INT_MAX, /* before any real devices */
8858};
8859
8860static void md_geninit(void)
8861{
36a4e1fe 8862 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8863
c7705f34 8864 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8865}
8866
75c96f85 8867static int __init md_init(void)
1da177e4 8868{
e804ac78
TH		 8869 int ret = -ENOMEM;
8870
ada609ee 8871 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8872 if (!md_wq)
8873 goto err_wq;
8874
8875 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8876 if (!md_misc_wq)
8877 goto err_misc_wq;
8878
8879 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8880 goto err_md;
8881
8882 if ((ret = register_blkdev(0, "mdp")) < 0)
8883 goto err_mdp;
8884 mdp_major = ret;
8885
af5628f0 8886 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
e8703fe1
N		 8887 md_probe, NULL, NULL);
8888 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8889 md_probe, NULL, NULL);
8890
1da177e4 8891 register_reboot_notifier(&md_notifier);
0b4d4147 8892 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8893
8894 md_geninit();
d710e138 8895 return 0;
1da177e4 8896
e804ac78
TH		 8897err_mdp:
8898 unregister_blkdev(MD_MAJOR, "md");
8899err_md:
8900 destroy_workqueue(md_misc_wq);
8901err_misc_wq:
8902 destroy_workqueue(md_wq);
8903err_wq:
8904 return ret;
8905}
1da177e4 8906
1d7e3e96
GR		 8907void md_reload_sb(struct mddev *mddev)
8908{
8909 struct md_rdev *rdev, *tmp;
8910
8911 rdev_for_each_safe(rdev, tmp, mddev) {
8912 rdev->sb_loaded = 0;
8913 ClearPageUptodate(rdev->sb_page);
8914 }
8915 mddev->raid_disks = 0;
8916 analyze_sbs(mddev);
8917 rdev_for_each_safe(rdev, tmp, mddev) {
8918 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8919 /* since we don't write to faulty devices, we figure out if the
8920 * disk is faulty by comparing events
8921 */
8922 if (mddev->events > sb->events)
8923 set_bit(Faulty, &rdev->flags);
8924 }
8925
8926}
8927EXPORT_SYMBOL(md_reload_sb);
8928
1da177e4
LT		 8929#ifndef MODULE
8930
8931/*
8932 * Searches all registered partitions for autorun RAID arrays
8933 * at boot time.
8934 */
4d936ec1
ME		 8935
8936static LIST_HEAD(all_detected_devices);
8937struct detected_devices_node {
8938 struct list_head list;
8939 dev_t dev;
8940};
1da177e4
LT		 8941
8942void md_autodetect_dev(dev_t dev)
8943{
4d936ec1
ME		 8944 struct detected_devices_node *node_detected_dev;
8945
8946 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8947 if (node_detected_dev) {
8948 node_detected_dev->dev = dev;
8949 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8950 } else {
8951 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8952 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8953 }
1da177e4
LT		 8954}
8955
1da177e4
LT		 8956static void autostart_arrays(int part)
8957{
3cb03002 8958 struct md_rdev *rdev;
4d936ec1
ME		 8959 struct detected_devices_node *node_detected_dev;
8960 dev_t dev;
8961 int i_scanned, i_passed;
1da177e4 8962
4d936ec1
ME		 8963 i_scanned = 0;
8964 i_passed = 0;
1da177e4 8965
4d936ec1 8966 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8967
4d936ec1
ME		 8968 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8969 i_scanned++;
8970 node_detected_dev = list_entry(all_detected_devices.next,
8971 struct detected_devices_node, list);
8972 list_del(&node_detected_dev->list);
8973 dev = node_detected_dev->dev;
8974 kfree(node_detected_dev);
df968c4e 8975 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8976 if (IS_ERR(rdev))
8977 continue;
8978
403df478 8979 if (test_bit(Faulty, &rdev->flags))
1da177e4 8980 continue;
403df478 8981
d0fae18f 8982 set_bit(AutoDetected, &rdev->flags);
1da177e4 8983 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8984 i_passed++;
1da177e4 8985 }
4d936ec1
ME		 8986
8987 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8988 i_scanned, i_passed);
1da177e4
LT		 8989
8990 autorun_devices(part);
8991}
8992
fdee8ae4 8993#endif /* !MODULE */
1da177e4
LT		 8994
8995static __exit void md_exit(void)
8996{
fd01b88c 8997 struct mddev *mddev;
1da177e4 8998 struct list_head *tmp;
e2f23b60 8999 int delay = 1;
8ab5e4c1 9000
af5628f0 9001 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
e8703fe1 9002 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 9003
3dbd8c2e 9004 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 9005 unregister_blkdev(mdp_major, "mdp");
9006 unregister_reboot_notifier(&md_notifier);
9007 unregister_sysctl_table(raid_table_header);
e2f23b60
N		 9008
9009 /* We cannot unload the modules while some process is
9010 * waiting for us in select() or poll() - wake them up
9011 */
9012 md_unloading = 1;
9013 while (waitqueue_active(&md_event_waiters)) {
9014 /* not safe to leave yet */
9015 wake_up(&md_event_waiters);
9016 msleep(delay);
9017 delay += delay;
9018 }
1da177e4 9019 remove_proc_entry("mdstat", NULL);
e2f23b60 9020
29ac4aa3 9021 for_each_mddev(mddev, tmp) {
1da177e4 9022 export_array(mddev);
d3374825 9023 mddev->hold_active = 0;
1da177e4 9024 }
e804ac78
TH		 9025 destroy_workqueue(md_misc_wq);
9026 destroy_workqueue(md_wq);
1da177e4
LT		 9027}
9028
685784aa 9029subsys_initcall(md_init);
1da177e4
LT		 9030module_exit(md_exit)
9031
f91de92e
N		 9032static int get_ro(char *buffer, struct kernel_param *kp)
9033{
9034 return sprintf(buffer, "%d", start_readonly);
9035}
9036static int set_ro(const char *val, struct kernel_param *kp)
9037{
4c9309c0 9038 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
f91de92e
N		 9039}
9040
80ca3a44
N		 9041module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9042module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
efeb53c0 9043module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 9044
1da177e4 9045MODULE_LICENSE("GPL");
0efb9e61 9046MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 9047MODULE_ALIAS("md");
72008652 9048MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)