projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'for-linus' of git://git.kernel.dk/linux-block
[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
N		 2632{
2633 char *e;
2634 unsigned long n = simple_strtoul(buf, &e, 10);
2635 if (*buf && (*e == 0 || *e == '\n')) {
2636 atomic_set(&rdev->corrected_errors, n);
2637 return len;
2638 }
2639 return -EINVAL;
2640}
2641static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2642__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2643
014236d2 2644static ssize_t
3cb03002 2645slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2646{
2647 if (rdev->raid_disk < 0)
2648 return sprintf(page, "none\n");
2649 else
2650 return sprintf(page, "%d\n", rdev->raid_disk);
2651}
2652
2653static ssize_t
3cb03002 2654slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N		 2655{
2656 char *e;
c303da6d 2657 int err;
014236d2
N		 2658 int slot = simple_strtoul(buf, &e, 10);
2659 if (strncmp(buf, "none", 4)==0)
2660 slot = -1;
2661 else if (e==buf || (*e && *e!= '\n'))
2662 return -EINVAL;
6c2fce2e 2663 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2664 /* Setting 'slot' on an active array requires also
2665 * updating the 'rd%d' link, and communicating
2666 * with the personality with ->hot_*_disk.
2667 * For now we only support removing
2668 * failed/spare devices. This normally happens automatically,
2669 * but not when the metadata is externally managed.
2670 */
c303da6d
N		 2671 if (rdev->raid_disk == -1)
2672 return -EEXIST;
2673 /* personality does all needed checks */
01393f3d 2674 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d 2675 return -EINVAL;
746d3207
N		 2676 clear_bit(Blocked, &rdev->flags);
2677 remove_and_add_spares(rdev->mddev, rdev);
2678 if (rdev->raid_disk >= 0)
2679 return -EBUSY;
c303da6d
N		 2680 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2681 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2682 } else if (rdev->mddev->pers) {
6c2fce2e 2683 /* Activating a spare .. or possibly reactivating
6d56e278 2684 * if we ever get bitmaps working here.
6c2fce2e
NB		 2685 */
2686
2687 if (rdev->raid_disk != -1)
2688 return -EBUSY;
2689
c6751b2b
N		 2690 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2691 return -EBUSY;
2692
6c2fce2e
NB		 2693 if (rdev->mddev->pers->hot_add_disk == NULL)
2694 return -EINVAL;
2695
ba1b41b6
N		 2696 if (slot >= rdev->mddev->raid_disks &&
2697 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2698 return -ENOSPC;
2699
6c2fce2e
NB		 2700 rdev->raid_disk = slot;
2701 if (test_bit(In_sync, &rdev->flags))
2702 rdev->saved_raid_disk = slot;
2703 else
2704 rdev->saved_raid_disk = -1;
d30519fc 2705 clear_bit(In_sync, &rdev->flags);
8313b8e5 2706 clear_bit(Bitmap_sync, &rdev->flags);
6c2fce2e
NB		 2707 err = rdev->mddev->pers->
2708 hot_add_disk(rdev->mddev, rdev);
199050ea 2709 if (err) {
6c2fce2e 2710 rdev->raid_disk = -1;
6c2fce2e 2711 return err;
52664732 2712 } else
00bcb4ac 2713 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2714 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2715 /* failure here is OK */;
6c2fce2e 2716 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2717 } else {
ba1b41b6
N		 2718 if (slot >= rdev->mddev->raid_disks &&
2719 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2720 return -ENOSPC;
2721 rdev->raid_disk = slot;
2722 /* assume it is working */
c5d79adb
N		 2723 clear_bit(Faulty, &rdev->flags);
2724 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2725 set_bit(In_sync, &rdev->flags);
00bcb4ac 2726 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2727 }
014236d2
N		 2728 return len;
2729}
2730
014236d2 2731static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2732__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2733
93c8cad0 2734static ssize_t
3cb03002 2735offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2736{
6961ece4 2737 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2738}
2739
2740static ssize_t
3cb03002 2741offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2742{
c6563a8c 2743 unsigned long long offset;
b29bebd6 2744 if (kstrtoull(buf, 10, &offset) < 0)
93c8cad0 2745 return -EINVAL;
8ed0a521 2746 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2747 return -EBUSY;
dd8ac336 2748 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2749 /* Must set offset before size, so overlap checks
2750 * can be sane */
2751 return -EBUSY;
93c8cad0 2752 rdev->data_offset = offset;
25f7fd47 2753 rdev->new_data_offset = offset;
93c8cad0
N		 2754 return len;
2755}
2756
2757static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2758__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2759
c6563a8c
N		 2760static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2761{
2762 return sprintf(page, "%llu\n",
2763 (unsigned long long)rdev->new_data_offset);
2764}
2765
2766static ssize_t new_offset_store(struct md_rdev *rdev,
2767 const char *buf, size_t len)
2768{
2769 unsigned long long new_offset;
2770 struct mddev *mddev = rdev->mddev;
2771
b29bebd6 2772 if (kstrtoull(buf, 10, &new_offset) < 0)
c6563a8c
N		 2773 return -EINVAL;
2774
f851b60d
N		 2775 if (mddev->sync_thread ||
2776 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
c6563a8c
N		 2777 return -EBUSY;
2778 if (new_offset == rdev->data_offset)
2779 /* reset is always permitted */
2780 ;
2781 else if (new_offset > rdev->data_offset) {
2782 /* must not push array size beyond rdev_sectors */
2783 if (new_offset - rdev->data_offset
2784 + mddev->dev_sectors > rdev->sectors)
2785 return -E2BIG;
2786 }
2787 /* Metadata worries about other space details. */
2788
2789 /* decreasing the offset is inconsistent with a backwards
2790 * reshape.
2791 */
2792 if (new_offset < rdev->data_offset &&
2793 mddev->reshape_backwards)
2794 return -EINVAL;
2795 /* Increasing offset is inconsistent with forwards
2796 * reshape. reshape_direction should be set to
2797 * 'backwards' first.
2798 */
2799 if (new_offset > rdev->data_offset &&
2800 !mddev->reshape_backwards)
2801 return -EINVAL;
2802
2803 if (mddev->pers && mddev->persistent &&
2804 !super_types[mddev->major_version]
2805 .allow_new_offset(rdev, new_offset))
2806 return -E2BIG;
2807 rdev->new_data_offset = new_offset;
2808 if (new_offset > rdev->data_offset)
2809 mddev->reshape_backwards = 1;
2810 else if (new_offset < rdev->data_offset)
2811 mddev->reshape_backwards = 0;
2812
2813 return len;
2814}
2815static struct rdev_sysfs_entry rdev_new_offset =
2816__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2817
83303b61 2818static ssize_t
3cb03002 2819rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2820{
dd8ac336 2821 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2822}
2823
c5d79adb
N		 2824static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2825{
2826 /* check if two start/length pairs overlap */
2827 if (s1+l1 <= s2)
2828 return 0;
2829 if (s2+l2 <= s1)
2830 return 0;
2831 return 1;
2832}
2833
b522adcd
DW		 2834static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2835{
2836 unsigned long long blocks;
2837 sector_t new;
2838
b29bebd6 2839 if (kstrtoull(buf, 10, &blocks) < 0)
b522adcd
DW		 2840 return -EINVAL;
2841
2842 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2843 return -EINVAL; /* sector conversion overflow */
2844
2845 new = blocks * 2;
2846 if (new != blocks * 2)
2847 return -EINVAL; /* unsigned long long to sector_t overflow */
2848
2849 *sectors = new;
2850 return 0;
2851}
2852
83303b61 2853static ssize_t
3cb03002 2854rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2855{
fd01b88c 2856 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2857 sector_t oldsectors = rdev->sectors;
b522adcd 2858 sector_t sectors;
27c529bb 2859
b522adcd 2860 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2861 return -EINVAL;
c6563a8c
N		 2862 if (rdev->data_offset != rdev->new_data_offset)
2863 return -EINVAL; /* too confusing */
0cd17fec 2864 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2865 if (my_mddev->persistent) {
dd8ac336
AN		 2866 sectors = super_types[my_mddev->major_version].
2867 rdev_size_change(rdev, sectors);
2868 if (!sectors)
0cd17fec 2869 return -EBUSY;
dd8ac336 2870 } else if (!sectors)
77304d2a 2871 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2872 rdev->data_offset;
a6468539
N		 2873 if (!my_mddev->pers->resize)
2874 /* Cannot change size for RAID0 or Linear etc */
2875 return -EINVAL;
0cd17fec 2876 }
dd8ac336 2877 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2878 return -EINVAL; /* component must fit device */
0cd17fec 2879
dd8ac336
AN		 2880 rdev->sectors = sectors;
2881 if (sectors > oldsectors && my_mddev->external) {
8b1afc3d
N		 2882 /* Need to check that all other rdevs with the same
2883 * ->bdev do not overlap. 'rcu' is sufficient to walk
2884 * the rdev lists safely.
2885 * This check does not provide a hard guarantee, it
2886 * just helps avoid dangerous mistakes.
c5d79adb 2887 */
fd01b88c 2888 struct mddev *mddev;
c5d79adb 2889 int overlap = 0;
159ec1fc 2890 struct list_head *tmp;
c5d79adb 2891
8b1afc3d 2892 rcu_read_lock();
29ac4aa3 2893 for_each_mddev(mddev, tmp) {
3cb03002 2894 struct md_rdev *rdev2;
c5d79adb 2895
dafb20fa 2896 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 2897 if (rdev->bdev == rdev2->bdev &&
2898 rdev != rdev2 &&
2899 overlaps(rdev->data_offset, rdev->sectors,
2900 rdev2->data_offset,
2901 rdev2->sectors)) {
c5d79adb
N		 2902 overlap = 1;
2903 break;
2904 }
c5d79adb
N		 2905 if (overlap) {
2906 mddev_put(mddev);
2907 break;
2908 }
2909 }
8b1afc3d 2910 rcu_read_unlock();
c5d79adb
N		 2911 if (overlap) {
2912 /* Someone else could have slipped in a size
2913 * change here, but doing so is just silly.
dd8ac336 2914 * We put oldsectors back because we *know* it is
c5d79adb
N		 2915 * safe, and trust userspace not to race with
2916 * itself
2917 */
dd8ac336 2918 rdev->sectors = oldsectors;
c5d79adb
N		 2919 return -EBUSY;
2920 }
2921 }
83303b61
N		 2922 return len;
2923}
2924
2925static struct rdev_sysfs_entry rdev_size =
80ca3a44 2926__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2927
3cb03002 2928static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 2929{
2930 unsigned long long recovery_start = rdev->recovery_offset;
2931
2932 if (test_bit(In_sync, &rdev->flags) ||
2933 recovery_start == MaxSector)
2934 return sprintf(page, "none\n");
2935
2936 return sprintf(page, "%llu\n", recovery_start);
2937}
2938
3cb03002 2939static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 2940{
2941 unsigned long long recovery_start;
2942
2943 if (cmd_match(buf, "none"))
2944 recovery_start = MaxSector;
b29bebd6 2945 else if (kstrtoull(buf, 10, &recovery_start))
06e3c817
DW		 2946 return -EINVAL;
2947
2948 if (rdev->mddev->pers &&
2949 rdev->raid_disk >= 0)
2950 return -EBUSY;
2951
2952 rdev->recovery_offset = recovery_start;
2953 if (recovery_start == MaxSector)
2954 set_bit(In_sync, &rdev->flags);
2955 else
2956 clear_bit(In_sync, &rdev->flags);
2957 return len;
2958}
2959
2960static struct rdev_sysfs_entry rdev_recovery_start =
2961__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2962
16c791a5
N		 2963static ssize_t
2964badblocks_show(struct badblocks *bb, char *page, int unack);
2965static ssize_t
2966badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2967
3cb03002 2968static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2969{
2970 return badblocks_show(&rdev->badblocks, page, 0);
2971}
3cb03002 2972static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 2973{
de393cde
N		 2974 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2975 /* Maybe that ack was all we needed */
2976 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2977 wake_up(&rdev->blocked_wait);
2978 return rv;
16c791a5
N		 2979}
2980static struct rdev_sysfs_entry rdev_bad_blocks =
2981__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2982
3cb03002 2983static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2984{
2985 return badblocks_show(&rdev->badblocks, page, 1);
2986}
3cb03002 2987static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 2988{
2989 return badblocks_store(&rdev->badblocks, page, len, 1);
2990}
2991static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2992__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2993
86e6ffdd
N		 2994static struct attribute *rdev_default_attrs[] = {
2995 &rdev_state.attr,
4dbcdc75 2996 &rdev_errors.attr,
014236d2 2997 &rdev_slot.attr,
93c8cad0 2998 &rdev_offset.attr,
c6563a8c 2999 &rdev_new_offset.attr,
83303b61 3000 &rdev_size.attr,
06e3c817 3001 &rdev_recovery_start.attr,
16c791a5
N		 3002 &rdev_bad_blocks.attr,
3003 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 3004 NULL,
3005};
3006static ssize_t
3007rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3008{
3009 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3010 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
86e6ffdd
N		 3011
3012 if (!entry->show)
3013 return -EIO;
758bfc8a
N		 3014 if (!rdev->mddev)
3015 return -EBUSY;
3016 return entry->show(rdev, page);
86e6ffdd
N		 3017}
3018
3019static ssize_t
3020rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3021 const char *page, size_t length)
3022{
3023 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3024 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3025 ssize_t rv;
fd01b88c 3026 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3027
3028 if (!entry->store)
3029 return -EIO;
67463acb
N		 3030 if (!capable(CAP_SYS_ADMIN))
3031 return -EACCES;
27c529bb 3032 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3033 if (!rv) {
27c529bb
N		 3034 if (rdev->mddev == NULL)
3035 rv = -EBUSY;
3036 else
3037 rv = entry->store(rdev, page, length);
6a51830e 3038 mddev_unlock(mddev);
ca388059
N		 3039 }
3040 return rv;
86e6ffdd
N		 3041}
3042
3043static void rdev_free(struct kobject *ko)
3044{
3cb03002 3045 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3046 kfree(rdev);
3047}
52cf25d0 3048static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3049 .show = rdev_attr_show,
3050 .store = rdev_attr_store,
3051};
3052static struct kobj_type rdev_ktype = {
3053 .release = rdev_free,
3054 .sysfs_ops = &rdev_sysfs_ops,
3055 .default_attrs = rdev_default_attrs,
3056};
3057
3cb03002 3058int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3059{
3060 rdev->desc_nr = -1;
3061 rdev->saved_raid_disk = -1;
3062 rdev->raid_disk = -1;
3063 rdev->flags = 0;
3064 rdev->data_offset = 0;
c6563a8c 3065 rdev->new_data_offset = 0;
e8bb9a83
N		 3066 rdev->sb_events = 0;
3067 rdev->last_read_error.tv_sec = 0;
3068 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3069 rdev->sb_loaded = 0;
3070 rdev->bb_page = NULL;
e8bb9a83
N		 3071 atomic_set(&rdev->nr_pending, 0);
3072 atomic_set(&rdev->read_errors, 0);
3073 atomic_set(&rdev->corrected_errors, 0);
3074
3075 INIT_LIST_HEAD(&rdev->same_set);
3076 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3077
3078 /* Add space to store bad block list.
3079 * This reserves the space even on arrays where it cannot
3080 * be used - I wonder if that matters
3081 */
3082 rdev->badblocks.count = 0;
486adf72 3083 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
2230dfe4
N		 3084 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3085 seqlock_init(&rdev->badblocks.lock);
3086 if (rdev->badblocks.page == NULL)
3087 return -ENOMEM;
3088
3089 return 0;
e8bb9a83
N		 3090}
3091EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3092/*
3093 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3094 *
3095 * mark the device faulty if:
3096 *
3097 * - the device is nonexistent (zero size)
3098 * - the device has no valid superblock
3099 *
3100 * a faulty rdev _never_ has rdev->sb set.
3101 */
3cb03002 3102static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3103{
3104 char b[BDEVNAME_SIZE];
3105 int err;
3cb03002 3106 struct md_rdev *rdev;
1da177e4
LT		 3107 sector_t size;
3108
9ffae0cf 3109 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3110 if (!rdev) {
3111 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3112 return ERR_PTR(-ENOMEM);
3113 }
1da177e4 3114
2230dfe4
N		 3115 err = md_rdev_init(rdev);
3116 if (err)
3117 goto abort_free;
3118 err = alloc_disk_sb(rdev);
3119 if (err)
1da177e4
LT		 3120 goto abort_free;
3121
c5d79adb 3122 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3123 if (err)
3124 goto abort_free;
3125
f9cb074b 3126 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3127
77304d2a 3128 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4 3129 if (!size) {
f72ffdd6 3130 printk(KERN_WARNING
1da177e4
LT		 3131 "md: %s has zero or unknown size, marking faulty!\n",
3132 bdevname(rdev->bdev,b));
3133 err = -EINVAL;
3134 goto abort_free;
3135 }
3136
3137 if (super_format >= 0) {
3138 err = super_types[super_format].
3139 load_super(rdev, NULL, super_minor);
3140 if (err == -EINVAL) {
df968c4e
N		 3141 printk(KERN_WARNING
3142 "md: %s does not have a valid v%d.%d "
3143 "superblock, not importing!\n",
3144 bdevname(rdev->bdev,b),
3145 super_format, super_minor);
1da177e4
LT		 3146 goto abort_free;
3147 }
3148 if (err < 0) {
f72ffdd6 3149 printk(KERN_WARNING
1da177e4
LT		 3150 "md: could not read %s's sb, not importing!\n",
3151 bdevname(rdev->bdev,b));
3152 goto abort_free;
3153 }
3154 }
6bfe0b49 3155
1da177e4
LT		 3156 return rdev;
3157
3158abort_free:
2699b672
N		 3159 if (rdev->bdev)
3160 unlock_rdev(rdev);
545c8795 3161 md_rdev_clear(rdev);
1da177e4
LT		 3162 kfree(rdev);
3163 return ERR_PTR(err);
3164}
3165
3166/*
3167 * Check a full RAID array for plausibility
3168 */
3169
f72ffdd6 3170static void analyze_sbs(struct mddev *mddev)
1da177e4
LT		 3171{
3172 int i;
3cb03002 3173 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3174 char b[BDEVNAME_SIZE];
3175
3176 freshest = NULL;
dafb20fa 3177 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3178 switch (super_types[mddev->major_version].
3179 load_super(rdev, freshest, mddev->minor_version)) {
3180 case 1:
3181 freshest = rdev;
3182 break;
3183 case 0:
3184 break;
3185 default:
3186 printk( KERN_ERR \
3187 "md: fatal superblock inconsistency in %s"
f72ffdd6 3188 " -- removing from array\n",
1da177e4 3189 bdevname(rdev->bdev,b));
fb56dfef 3190 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3191 }
3192
1da177e4
LT		 3193 super_types[mddev->major_version].
3194 validate_super(mddev, freshest);
3195
3196 i = 0;
dafb20fa 3197 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3198 if (mddev->max_disks &&
3199 (rdev->desc_nr >= mddev->max_disks ||
3200 i > mddev->max_disks)) {
de01dfad
N		 3201 printk(KERN_WARNING
3202 "md: %s: %s: only %d devices permitted\n",
3203 mdname(mddev), bdevname(rdev->bdev, b),
3204 mddev->max_disks);
fb56dfef 3205 md_kick_rdev_from_array(rdev);
de01dfad
N		 3206 continue;
3207 }
1aee41f6 3208 if (rdev != freshest) {
1da177e4
LT		 3209 if (super_types[mddev->major_version].
3210 validate_super(mddev, rdev)) {
3211 printk(KERN_WARNING "md: kicking non-fresh %s"
3212 " from array!\n",
3213 bdevname(rdev->bdev,b));
fb56dfef 3214 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3215 continue;
3216 }
1aee41f6
GR		 3217 /* No device should have a Candidate flag
3218 * when reading devices
3219 */
3220 if (test_bit(Candidate, &rdev->flags)) {
3221 pr_info("md: kicking Cluster Candidate %s from array!\n",
3222 bdevname(rdev->bdev, b));
fb56dfef 3223 md_kick_rdev_from_array(rdev);
1aee41f6
GR		 3224 }
3225 }
1da177e4
LT		 3226 if (mddev->level == LEVEL_MULTIPATH) {
3227 rdev->desc_nr = i++;
3228 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3229 set_bit(In_sync, &rdev->flags);
5e5e3e78 3230 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3231 rdev->raid_disk = -1;
3232 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3233 }
3234 }
1da177e4
LT		 3235}
3236
72e02075
N		 3237/* Read a fixed-point number.
3238 * Numbers in sysfs attributes should be in "standard" units where
3239 * possible, so time should be in seconds.
f72ffdd6 3240 * However we internally use a a much smaller unit such as
72e02075
N		 3241 * milliseconds or jiffies.
3242 * This function takes a decimal number with a possible fractional
3243 * component, and produces an integer which is the result of
3244 * multiplying that number by 10^'scale'.
3245 * all without any floating-point arithmetic.
3246 */
3247int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3248{
3249 unsigned long result = 0;
3250 long decimals = -1;
3251 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3252 if (*cp == '.')
3253 decimals = 0;
3254 else if (decimals < scale) {
3255 unsigned int value;
3256 value = *cp - '0';
3257 result = result * 10 + value;
3258 if (decimals >= 0)
3259 decimals++;
3260 }
3261 cp++;
3262 }
3263 if (*cp == '\n')
3264 cp++;
3265 if (*cp)
3266 return -EINVAL;
3267 if (decimals < 0)
3268 decimals = 0;
3269 while (decimals < scale) {
3270 result *= 10;
3271 decimals ++;
3272 }
3273 *res = result;
3274 return 0;
3275}
3276
19052c0e
N		 3277static void md_safemode_timeout(unsigned long data);
3278
16f17b39 3279static ssize_t
fd01b88c 3280safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3281{
3282 int msec = (mddev->safemode_delay*1000)/HZ;
3283 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3284}
3285static ssize_t
fd01b88c 3286safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3287{
16f17b39 3288 unsigned long msec;
97ce0a7f 3289
72e02075 3290 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3291 return -EINVAL;
16f17b39
N		 3292 if (msec == 0)
3293 mddev->safemode_delay = 0;
3294 else {
19052c0e 3295 unsigned long old_delay = mddev->safemode_delay;
1b30e66f
N		 3296 unsigned long new_delay = (msec*HZ)/1000;
3297
3298 if (new_delay == 0)
3299 new_delay = 1;
3300 mddev->safemode_delay = new_delay;
3301 if (new_delay < old_delay || old_delay == 0)
3302 mod_timer(&mddev->safemode_timer, jiffies+1);
16f17b39
N		 3303 }
3304 return len;
3305}
3306static struct md_sysfs_entry md_safe_delay =
80ca3a44 3307__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3308
eae1701f 3309static ssize_t
fd01b88c 3310level_show(struct mddev *mddev, char *page)
eae1701f 3311{
36d091f4
N		 3312 struct md_personality *p;
3313 int ret;
3314 spin_lock(&mddev->lock);
3315 p = mddev->pers;
d9d166c2 3316 if (p)
36d091f4 3317 ret = sprintf(page, "%s\n", p->name);
d9d166c2 3318 else if (mddev->clevel[0])
36d091f4 3319 ret = sprintf(page, "%s\n", mddev->clevel);
d9d166c2 3320 else if (mddev->level != LEVEL_NONE)
36d091f4 3321 ret = sprintf(page, "%d\n", mddev->level);
d9d166c2 3322 else
36d091f4
N		 3323 ret = 0;
3324 spin_unlock(&mddev->lock);
3325 return ret;
eae1701f
N		 3326}
3327
d9d166c2 3328static ssize_t
fd01b88c 3329level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3330{
f2859af6 3331 char clevel[16];
6791875e
N		 3332 ssize_t rv;
3333 size_t slen = len;
db721d32 3334 struct md_personality *pers, *oldpers;
f2859af6 3335 long level;
db721d32 3336 void *priv, *oldpriv;
3cb03002 3337 struct md_rdev *rdev;
245f46c2 3338
6791875e
N		 3339 if (slen == 0 || slen >= sizeof(clevel))
3340 return -EINVAL;
3341
3342 rv = mddev_lock(mddev);
3343 if (rv)
3344 return rv;
3345
245f46c2 3346 if (mddev->pers == NULL) {
6791875e
N		 3347 strncpy(mddev->clevel, buf, slen);
3348 if (mddev->clevel[slen-1] == '\n')
3349 slen--;
3350 mddev->clevel[slen] = 0;
245f46c2 3351 mddev->level = LEVEL_NONE;
6791875e
N		 3352 rv = len;
3353 goto out_unlock;
245f46c2 3354 }
6791875e 3355 rv = -EROFS;
bd8839e0 3356 if (mddev->ro)
6791875e 3357 goto out_unlock;
245f46c2
N		 3358
3359 /* request to change the personality. Need to ensure:
3360 * - array is not engaged in resync/recovery/reshape
3361 * - old personality can be suspended
3362 * - new personality will access other array.
3363 */
3364
6791875e 3365 rv = -EBUSY;
bb4f1e9d 3366 if (mddev->sync_thread ||
f851b60d 3367 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
bb4f1e9d
N		 3368 mddev->reshape_position != MaxSector ||
3369 mddev->sysfs_active)
6791875e 3370 goto out_unlock;
245f46c2 3371
6791875e 3372 rv = -EINVAL;
245f46c2
N		 3373 if (!mddev->pers->quiesce) {
3374 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3375 mdname(mddev), mddev->pers->name);
6791875e 3376 goto out_unlock;
245f46c2
N		 3377 }
3378
3379 /* Now find the new personality */
6791875e
N		 3380 strncpy(clevel, buf, slen);
3381 if (clevel[slen-1] == '\n')
3382 slen--;
3383 clevel[slen] = 0;
b29bebd6 3384 if (kstrtol(clevel, 10, &level))
f2859af6 3385 level = LEVEL_NONE;
245f46c2 3386
f2859af6
DW		 3387 if (request_module("md-%s", clevel) != 0)
3388 request_module("md-level-%s", clevel);
245f46c2 3389 spin_lock(&pers_lock);
f2859af6 3390 pers = find_pers(level, clevel);
245f46c2
N		 3391 if (!pers || !try_module_get(pers->owner)) {
3392 spin_unlock(&pers_lock);
f2859af6 3393 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
6791875e
N		 3394 rv = -EINVAL;
3395 goto out_unlock;
245f46c2
N		 3396 }
3397 spin_unlock(&pers_lock);
3398
3399 if (pers == mddev->pers) {
3400 /* Nothing to do! */
3401 module_put(pers->owner);
6791875e
N		 3402 rv = len;
3403 goto out_unlock;
245f46c2
N		 3404 }
3405 if (!pers->takeover) {
3406 module_put(pers->owner);
3407 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3408 mdname(mddev), clevel);
6791875e
N		 3409 rv = -EINVAL;
3410 goto out_unlock;
245f46c2
N		 3411 }
3412
dafb20fa 3413 rdev_for_each(rdev, mddev)
e93f68a1
N		 3414 rdev->new_raid_disk = rdev->raid_disk;
3415
245f46c2
N		 3416 /* ->takeover must set new_* and/or delta_disks
3417 * if it succeeds, and may set them when it fails.
3418 */
3419 priv = pers->takeover(mddev);
3420 if (IS_ERR(priv)) {
3421 mddev->new_level = mddev->level;
3422 mddev->new_layout = mddev->layout;
664e7c41 3423 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3424 mddev->raid_disks -= mddev->delta_disks;
3425 mddev->delta_disks = 0;
2c810cdd 3426 mddev->reshape_backwards = 0;
245f46c2
N		 3427 module_put(pers->owner);
3428 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3429 mdname(mddev), clevel);
6791875e
N		 3430 rv = PTR_ERR(priv);
3431 goto out_unlock;
245f46c2
N		 3432 }
3433
3434 /* Looks like we have a winner */
3435 mddev_suspend(mddev);
5aa61f42 3436 mddev_detach(mddev);
36d091f4
N		 3437
3438 spin_lock(&mddev->lock);
db721d32
N		 3439 oldpers = mddev->pers;
3440 oldpriv = mddev->private;
3441 mddev->pers = pers;
3442 mddev->private = priv;
3443 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3444 mddev->level = mddev->new_level;
3445 mddev->layout = mddev->new_layout;
3446 mddev->chunk_sectors = mddev->new_chunk_sectors;
3447 mddev->delta_disks = 0;
3448 mddev->reshape_backwards = 0;
3449 mddev->degraded = 0;
36d091f4 3450 spin_unlock(&mddev->lock);
db721d32
N		 3451
3452 if (oldpers->sync_request == NULL &&
3453 mddev->external) {
3454 /* We are converting from a no-redundancy array
3455 * to a redundancy array and metadata is managed
3456 * externally so we need to be sure that writes
3457 * won't block due to a need to transition
3458 * clean->dirty
3459 * until external management is started.
3460 */
3461 mddev->in_sync = 0;
3462 mddev->safemode_delay = 0;
3463 mddev->safemode = 0;
3464 }
f72ffdd6 3465
db721d32
N		 3466 oldpers->free(mddev, oldpriv);
3467
3468 if (oldpers->sync_request == NULL &&
a64c876f
N		 3469 pers->sync_request != NULL) {
3470 /* need to add the md_redundancy_group */
3471 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3472 printk(KERN_WARNING
3473 "md: cannot register extra attributes for %s\n",
3474 mdname(mddev));
388975cc 3475 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
f72ffdd6 3476 }
db721d32 3477 if (oldpers->sync_request != NULL &&
a64c876f
N		 3478 pers->sync_request == NULL) {
3479 /* need to remove the md_redundancy_group */
3480 if (mddev->to_remove == NULL)
3481 mddev->to_remove = &md_redundancy_group;
3482 }
3483
dafb20fa 3484 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3485 if (rdev->raid_disk < 0)
3486 continue;
bf2cb0da 3487 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3488 rdev->new_raid_disk = -1;
3489 if (rdev->new_raid_disk == rdev->raid_disk)
3490 continue;
36fad858 3491 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3492 }
dafb20fa 3493 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3494 if (rdev->raid_disk < 0)
3495 continue;
3496 if (rdev->new_raid_disk == rdev->raid_disk)
3497 continue;
3498 rdev->raid_disk = rdev->new_raid_disk;
3499 if (rdev->raid_disk < 0)
3a981b03 3500 clear_bit(In_sync, &rdev->flags);
e93f68a1 3501 else {
36fad858
NK		 3502 if (sysfs_link_rdev(mddev, rdev))
3503 printk(KERN_WARNING "md: cannot register rd%d"
3504 " for %s after level change\n",
3505 rdev->raid_disk, mdname(mddev));
3a981b03 3506 }
e93f68a1
N		 3507 }
3508
db721d32 3509 if (pers->sync_request == NULL) {
9af204cf
TM		 3510 /* this is now an array without redundancy, so
3511 * it must always be in_sync
3512 */
3513 mddev->in_sync = 1;
3514 del_timer_sync(&mddev->safemode_timer);
3515 }
02e5f5c0 3516 blk_set_stacking_limits(&mddev->queue->limits);
245f46c2 3517 pers->run(mddev);
245f46c2 3518 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3519 mddev_resume(mddev);
830778a1
N		 3520 if (!mddev->thread)
3521 md_update_sb(mddev, 1);
5cac7861 3522 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3523 md_new_event(mddev);
6791875e
N		 3524 rv = len;
3525out_unlock:
3526 mddev_unlock(mddev);
d9d166c2
N		 3527 return rv;
3528}
3529
3530static struct md_sysfs_entry md_level =
80ca3a44 3531__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3532
d4dbd025 3533static ssize_t
fd01b88c 3534layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3535{
3536 /* just a number, not meaningful for all levels */
08a02ecd
N		 3537 if (mddev->reshape_position != MaxSector &&
3538 mddev->layout != mddev->new_layout)
3539 return sprintf(page, "%d (%d)\n",
3540 mddev->new_layout, mddev->layout);
d4dbd025
N		 3541 return sprintf(page, "%d\n", mddev->layout);
3542}
3543
3544static ssize_t
fd01b88c 3545layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N		 3546{
3547 char *e;
3548 unsigned long n = simple_strtoul(buf, &e, 10);
6791875e 3549 int err;
d4dbd025
N		 3550
3551 if (!*buf || (*e && *e != '\n'))
3552 return -EINVAL;
6791875e
N		 3553 err = mddev_lock(mddev);
3554 if (err)
3555 return err;
d4dbd025 3556
b3546035 3557 if (mddev->pers) {
50ac168a 3558 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3559 err = -EBUSY;
3560 else if (mddev->ro)
3561 err = -EROFS;
3562 else {
3563 mddev->new_layout = n;
3564 err = mddev->pers->check_reshape(mddev);
3565 if (err)
3566 mddev->new_layout = mddev->layout;
597a711b 3567 }
b3546035 3568 } else {
08a02ecd 3569 mddev->new_layout = n;
b3546035
N		 3570 if (mddev->reshape_position == MaxSector)
3571 mddev->layout = n;
3572 }
6791875e
N		 3573 mddev_unlock(mddev);
3574 return err ?: len;
d4dbd025
N		 3575}
3576static struct md_sysfs_entry md_layout =
80ca3a44 3577__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025 3578
eae1701f 3579static ssize_t
fd01b88c 3580raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3581{
bb636547
N		 3582 if (mddev->raid_disks == 0)
3583 return 0;
08a02ecd
N		 3584 if (mddev->reshape_position != MaxSector &&
3585 mddev->delta_disks != 0)
3586 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3587 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3588 return sprintf(page, "%d\n", mddev->raid_disks);
3589}
3590
fd01b88c 3591static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3592
3593static ssize_t
fd01b88c 3594raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3595{
da943b99 3596 char *e;
6791875e 3597 int err;
da943b99
N		 3598 unsigned long n = simple_strtoul(buf, &e, 10);
3599
3600 if (!*buf || (*e && *e != '\n'))
3601 return -EINVAL;
3602
6791875e
N		 3603 err = mddev_lock(mddev);
3604 if (err)
3605 return err;
da943b99 3606 if (mddev->pers)
6791875e 3607 err = update_raid_disks(mddev, n);
08a02ecd 3608 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3609 struct md_rdev *rdev;
08a02ecd 3610 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c 3611
6791875e 3612 err = -EINVAL;
c6563a8c
N		 3613 rdev_for_each(rdev, mddev) {
3614 if (olddisks < n &&
3615 rdev->data_offset < rdev->new_data_offset)
6791875e 3616 goto out_unlock;
c6563a8c
N		 3617 if (olddisks > n &&
3618 rdev->data_offset > rdev->new_data_offset)
6791875e 3619 goto out_unlock;
c6563a8c 3620 }
6791875e 3621 err = 0;
08a02ecd
N		 3622 mddev->delta_disks = n - olddisks;
3623 mddev->raid_disks = n;
2c810cdd 3624 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3625 } else
da943b99 3626 mddev->raid_disks = n;
6791875e
N		 3627out_unlock:
3628 mddev_unlock(mddev);
3629 return err ? err : len;
da943b99
N		 3630}
3631static struct md_sysfs_entry md_raid_disks =
80ca3a44 3632__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3633
3b34380a 3634static ssize_t
fd01b88c 3635chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3636{
08a02ecd 3637 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3638 mddev->chunk_sectors != mddev->new_chunk_sectors)
3639 return sprintf(page, "%d (%d)\n",
3640 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3641 mddev->chunk_sectors << 9);
3642 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3643}
3644
3645static ssize_t
fd01b88c 3646chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3647{
6791875e 3648 int err;
3b34380a
N		 3649 char *e;
3650 unsigned long n = simple_strtoul(buf, &e, 10);
3651
3b34380a
N		 3652 if (!*buf || (*e && *e != '\n'))
3653 return -EINVAL;
3654
6791875e
N		 3655 err = mddev_lock(mddev);
3656 if (err)
3657 return err;
b3546035 3658 if (mddev->pers) {
50ac168a 3659 if (mddev->pers->check_reshape == NULL)
6791875e
N		 3660 err = -EBUSY;
3661 else if (mddev->ro)
3662 err = -EROFS;
3663 else {
3664 mddev->new_chunk_sectors = n >> 9;
3665 err = mddev->pers->check_reshape(mddev);
3666 if (err)
3667 mddev->new_chunk_sectors = mddev->chunk_sectors;
597a711b 3668 }
b3546035 3669 } else {
664e7c41 3670 mddev->new_chunk_sectors = n >> 9;
b3546035 3671 if (mddev->reshape_position == MaxSector)
9d8f0363 3672 mddev->chunk_sectors = n >> 9;
b3546035 3673 }
6791875e
N		 3674 mddev_unlock(mddev);
3675 return err ?: len;
3b34380a
N		 3676}
3677static struct md_sysfs_entry md_chunk_size =
80ca3a44 3678__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3679
a94213b1 3680static ssize_t
fd01b88c 3681resync_start_show(struct mddev *mddev, char *page)
a94213b1 3682{
d1a7c503
N		 3683 if (mddev->recovery_cp == MaxSector)
3684 return sprintf(page, "none\n");
a94213b1
N		 3685 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3686}
3687
3688static ssize_t
fd01b88c 3689resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3690{
6791875e 3691 int err;
a94213b1
N		 3692 char *e;
3693 unsigned long long n = simple_strtoull(buf, &e, 10);
3694
6791875e
N		 3695 err = mddev_lock(mddev);
3696 if (err)
3697 return err;
b098636c 3698 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6791875e
N		 3699 err = -EBUSY;
3700 else if (cmd_match(buf, "none"))
06e3c817
DW		 3701 n = MaxSector;
3702 else if (!*buf || (*e && *e != '\n'))
6791875e 3703 err = -EINVAL;
a94213b1 3704
6791875e
N		 3705 if (!err) {
3706 mddev->recovery_cp = n;
3707 if (mddev->pers)
3708 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3709 }
3710 mddev_unlock(mddev);
3711 return err ?: len;
a94213b1
N		 3712}
3713static struct md_sysfs_entry md_resync_start =
750f199e
N		 3714__ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3715 resync_start_show, resync_start_store);
a94213b1 3716
9e653b63
N		 3717/*
3718 * The array state can be:
3719 *
3720 * clear
3721 * No devices, no size, no level
3722 * Equivalent to STOP_ARRAY ioctl
3723 * inactive
3724 * May have some settings, but array is not active
3725 * all IO results in error
3726 * When written, doesn't tear down array, but just stops it
3727 * suspended (not supported yet)
3728 * All IO requests will block. The array can be reconfigured.
910d8cb3 3729 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3730 * readonly
3731 * no resync can happen. no superblocks get written.
3732 * write requests fail
3733 * read-auto
3734 * like readonly, but behaves like 'clean' on a write request.
3735 *
3736 * clean - no pending writes, but otherwise active.
3737 * When written to inactive array, starts without resync
3738 * If a write request arrives then
3739 * if metadata is known, mark 'dirty' and switch to 'active'.
3740 * if not known, block and switch to write-pending
3741 * If written to an active array that has pending writes, then fails.
3742 * active
3743 * fully active: IO and resync can be happening.
3744 * When written to inactive array, starts with resync
3745 *
3746 * write-pending
3747 * clean, but writes are blocked waiting for 'active' to be written.
3748 *
3749 * active-idle
3750 * like active, but no writes have been seen for a while (100msec).
3751 *
3752 */
3753enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3754 write_pending, active_idle, bad_word};
05381954 3755static char *array_states[] = {
9e653b63
N		 3756 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3757 "write-pending", "active-idle", NULL };
3758
3759static int match_word(const char *word, char **list)
3760{
3761 int n;
3762 for (n=0; list[n]; n++)
3763 if (cmd_match(word, list[n]))
3764 break;
3765 return n;
3766}
3767
3768static ssize_t
fd01b88c 3769array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3770{
3771 enum array_state st = inactive;
3772
3773 if (mddev->pers)
3774 switch(mddev->ro) {
3775 case 1:
3776 st = readonly;
3777 break;
3778 case 2:
3779 st = read_auto;
3780 break;
3781 case 0:
3782 if (mddev->in_sync)
3783 st = clean;
070dc6dd 3784 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3785 st = write_pending;
9e653b63
N		 3786 else if (mddev->safemode)
3787 st = active_idle;
3788 else
3789 st = active;
3790 }
3791 else {
3792 if (list_empty(&mddev->disks) &&
3793 mddev->raid_disks == 0 &&
58c0fed4 3794 mddev->dev_sectors == 0)
9e653b63
N		 3795 st = clear;
3796 else
3797 st = inactive;
3798 }
3799 return sprintf(page, "%s\n", array_states[st]);
3800}
3801
f72ffdd6
N		 3802static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3803static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3804static int do_md_run(struct mddev *mddev);
fd01b88c 3805static int restart_array(struct mddev *mddev);
9e653b63
N		 3806
3807static ssize_t
fd01b88c 3808array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63 3809{
6791875e 3810 int err;
9e653b63 3811 enum array_state st = match_word(buf, array_states);
6791875e
N		 3812
3813 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3814 /* don't take reconfig_mutex when toggling between
3815 * clean and active
3816 */
3817 spin_lock(&mddev->lock);
3818 if (st == active) {
3819 restart_array(mddev);
3820 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3821 wake_up(&mddev->sb_wait);
3822 err = 0;
3823 } else /* st == clean */ {
3824 restart_array(mddev);
3825 if (atomic_read(&mddev->writes_pending) == 0) {
3826 if (mddev->in_sync == 0) {
3827 mddev->in_sync = 1;
3828 if (mddev->safemode == 1)
3829 mddev->safemode = 0;
3830 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3831 }
3832 err = 0;
3833 } else
3834 err = -EBUSY;
3835 }
3836 spin_unlock(&mddev->lock);
3837 return err;
3838 }
3839 err = mddev_lock(mddev);
3840 if (err)
3841 return err;
3842 err = -EINVAL;
9e653b63
N		 3843 switch(st) {
3844 case bad_word:
3845 break;
3846 case clear:
3847 /* stopping an active array */
a05b7ea0 3848 err = do_md_stop(mddev, 0, NULL);
9e653b63
N		 3849 break;
3850 case inactive:
3851 /* stopping an active array */
90cf195d 3852 if (mddev->pers)
a05b7ea0 3853 err = do_md_stop(mddev, 2, NULL);
90cf195d 3854 else
e691063a 3855 err = 0; /* already inactive */
9e653b63
N		 3856 break;
3857 case suspended:
3858 break; /* not supported yet */
3859 case readonly:
3860 if (mddev->pers)
a05b7ea0 3861 err = md_set_readonly(mddev, NULL);
9e653b63
N		 3862 else {
3863 mddev->ro = 1;
648b629e 3864 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3865 err = do_md_run(mddev);
3866 }
3867 break;
3868 case read_auto:
9e653b63 3869 if (mddev->pers) {
80268ee9 3870 if (mddev->ro == 0)
a05b7ea0 3871 err = md_set_readonly(mddev, NULL);
80268ee9 3872 else if (mddev->ro == 1)
648b629e
N		 3873 err = restart_array(mddev);
3874 if (err == 0) {
3875 mddev->ro = 2;
3876 set_disk_ro(mddev->gendisk, 0);
3877 }
9e653b63
N		 3878 } else {
3879 mddev->ro = 2;
3880 err = do_md_run(mddev);
3881 }
3882 break;
3883 case clean:
3884 if (mddev->pers) {
3885 restart_array(mddev);
85572d7c 3886 spin_lock(&mddev->lock);
9e653b63 3887 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3888 if (mddev->in_sync == 0) {
3889 mddev->in_sync = 1;
31a59e34
N		 3890 if (mddev->safemode == 1)
3891 mddev->safemode = 0;
070dc6dd 3892 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3893 }
3894 err = 0;
3895 } else
3896 err = -EBUSY;
85572d7c 3897 spin_unlock(&mddev->lock);
5bf29597
N		 3898 } else
3899 err = -EINVAL;
9e653b63
N		 3900 break;
3901 case active:
3902 if (mddev->pers) {
3903 restart_array(mddev);
070dc6dd 3904 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3905 wake_up(&mddev->sb_wait);
3906 err = 0;
3907 } else {
3908 mddev->ro = 0;
648b629e 3909 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3910 err = do_md_run(mddev);
3911 }
3912 break;
3913 case write_pending:
3914 case active_idle:
3915 /* these cannot be set */
3916 break;
3917 }
6791875e
N		 3918
3919 if (!err) {
1d23f178
N		 3920 if (mddev->hold_active == UNTIL_IOCTL)
3921 mddev->hold_active = 0;
00bcb4ac 3922 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 3923 }
6791875e
N		 3924 mddev_unlock(mddev);
3925 return err ?: len;
9e653b63 3926}
80ca3a44 3927static struct md_sysfs_entry md_array_state =
750f199e 3928__ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3929
1e50915f 3930static ssize_t
fd01b88c 3931max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3932 return sprintf(page, "%d\n",
3933 atomic_read(&mddev->max_corr_read_errors));
3934}
3935
3936static ssize_t
fd01b88c 3937max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB		 3938{
3939 char *e;
3940 unsigned long n = simple_strtoul(buf, &e, 10);
3941
3942 if (*buf && (*e == 0 || *e == '\n')) {
3943 atomic_set(&mddev->max_corr_read_errors, n);
3944 return len;
3945 }
3946 return -EINVAL;
3947}
3948
3949static struct md_sysfs_entry max_corr_read_errors =
3950__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3951 max_corrected_read_errors_store);
3952
6d7ff738 3953static ssize_t
fd01b88c 3954null_show(struct mddev *mddev, char *page)
6d7ff738
N		 3955{
3956 return -EINVAL;
3957}
3958
3959static ssize_t
fd01b88c 3960new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 3961{
3962 /* buf must be %d:%d\n? giving major and minor numbers */
3963 /* The new device is added to the array.
3964 * If the array has a persistent superblock, we read the
3965 * superblock to initialise info and check validity.
3966 * Otherwise, only checking done is that in bind_rdev_to_array,
3967 * which mainly checks size.
3968 */
3969 char *e;
3970 int major = simple_strtoul(buf, &e, 10);
3971 int minor;
3972 dev_t dev;
3cb03002 3973 struct md_rdev *rdev;
6d7ff738
N		 3974 int err;
3975
3976 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3977 return -EINVAL;
3978 minor = simple_strtoul(e+1, &e, 10);
3979 if (*e && *e != '\n')
3980 return -EINVAL;
3981 dev = MKDEV(major, minor);
3982 if (major != MAJOR(dev) ||
3983 minor != MINOR(dev))
3984 return -EOVERFLOW;
3985
6791875e
N		 3986 flush_workqueue(md_misc_wq);
3987
3988 err = mddev_lock(mddev);
3989 if (err)
3990 return err;
6d7ff738
N		 3991 if (mddev->persistent) {
3992 rdev = md_import_device(dev, mddev->major_version,
3993 mddev->minor_version);
3994 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 3995 struct md_rdev *rdev0
3996 = list_entry(mddev->disks.next,
3997 struct md_rdev, same_set);
6d7ff738
N		 3998 err = super_types[mddev->major_version]
3999 .load_super(rdev, rdev0, mddev->minor_version);
4000 if (err < 0)
4001 goto out;
4002 }
c5d79adb
N		 4003 } else if (mddev->external)
4004 rdev = md_import_device(dev, -2, -1);
4005 else
6d7ff738
N		 4006 rdev = md_import_device(dev, -1, -1);
4007
4008 if (IS_ERR(rdev))
4009 return PTR_ERR(rdev);
4010 err = bind_rdev_to_array(rdev, mddev);
4011 out:
4012 if (err)
4013 export_rdev(rdev);
6791875e 4014 mddev_unlock(mddev);
6d7ff738
N		 4015 return err ? err : len;
4016}
4017
4018static struct md_sysfs_entry md_new_device =
80ca3a44 4019__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 4020
9b1d1dac 4021static ssize_t
fd01b88c 4022bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 4023{
4024 char *end;
4025 unsigned long chunk, end_chunk;
6791875e 4026 int err;
9b1d1dac 4027
6791875e
N		 4028 err = mddev_lock(mddev);
4029 if (err)
4030 return err;
9b1d1dac
PC		 4031 if (!mddev->bitmap)
4032 goto out;
4033 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4034 while (*buf) {
4035 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4036 if (buf == end) break;
4037 if (*end == '-') { /* range */
4038 buf = end + 1;
4039 end_chunk = simple_strtoul(buf, &end, 0);
4040 if (buf == end) break;
4041 }
4042 if (*end && !isspace(*end)) break;
4043 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 4044 buf = skip_spaces(end);
9b1d1dac
PC		 4045 }
4046 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4047out:
6791875e 4048 mddev_unlock(mddev);
9b1d1dac
PC		 4049 return len;
4050}
4051
4052static struct md_sysfs_entry md_bitmap =
4053__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4054
a35b0d69 4055static ssize_t
fd01b88c 4056size_show(struct mddev *mddev, char *page)
a35b0d69 4057{
58c0fed4
AN		 4058 return sprintf(page, "%llu\n",
4059 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 4060}
4061
fd01b88c 4062static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 4063
4064static ssize_t
fd01b88c 4065size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 4066{
4067 /* If array is inactive, we can reduce the component size, but
4068 * not increase it (except from 0).
4069 * If array is active, we can try an on-line resize
4070 */
b522adcd
DW		 4071 sector_t sectors;
4072 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4073
58c0fed4
AN		 4074 if (err < 0)
4075 return err;
6791875e
N		 4076 err = mddev_lock(mddev);
4077 if (err)
4078 return err;
a35b0d69 4079 if (mddev->pers) {
293467aa
GR		 4080 if (mddev_is_clustered(mddev))
4081 md_cluster_ops->metadata_update_start(mddev);
58c0fed4 4082 err = update_size(mddev, sectors);
850b2b42 4083 md_update_sb(mddev, 1);
293467aa
GR		 4084 if (mddev_is_clustered(mddev))
4085 md_cluster_ops->metadata_update_finish(mddev);
a35b0d69 4086 } else {
58c0fed4
AN		 4087 if (mddev->dev_sectors == 0 ||
4088 mddev->dev_sectors > sectors)
4089 mddev->dev_sectors = sectors;
a35b0d69
N		 4090 else
4091 err = -ENOSPC;
4092 }
6791875e 4093 mddev_unlock(mddev);
a35b0d69
N		 4094 return err ? err : len;
4095}
4096
4097static struct md_sysfs_entry md_size =
80ca3a44 4098__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4099
83f0d77a 4100/* Metadata version.
e691063a
N		 4101 * This is one of
4102 * 'none' for arrays with no metadata (good luck...)
4103 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4104 * or N.M for internally known formats
4105 */
4106static ssize_t
fd01b88c 4107metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4108{
4109 if (mddev->persistent)
4110 return sprintf(page, "%d.%d\n",
4111 mddev->major_version, mddev->minor_version);
e691063a
N		 4112 else if (mddev->external)
4113 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4114 else
4115 return sprintf(page, "none\n");
4116}
4117
4118static ssize_t
fd01b88c 4119metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4120{
4121 int major, minor;
4122 char *e;
6791875e 4123 int err;
ea43ddd8
N		 4124 /* Changing the details of 'external' metadata is
4125 * always permitted. Otherwise there must be
4126 * no devices attached to the array.
4127 */
6791875e
N		 4128
4129 err = mddev_lock(mddev);
4130 if (err)
4131 return err;
4132 err = -EBUSY;
ea43ddd8
N		 4133 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4134 ;
4135 else if (!list_empty(&mddev->disks))
6791875e 4136 goto out_unlock;
8bb93aac 4137
6791875e 4138 err = 0;
8bb93aac
N		 4139 if (cmd_match(buf, "none")) {
4140 mddev->persistent = 0;
e691063a
N		 4141 mddev->external = 0;
4142 mddev->major_version = 0;
4143 mddev->minor_version = 90;
6791875e 4144 goto out_unlock;
e691063a
N		 4145 }
4146 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4147 size_t namelen = len-9;
e691063a
N		 4148 if (namelen >= sizeof(mddev->metadata_type))
4149 namelen = sizeof(mddev->metadata_type)-1;
4150 strncpy(mddev->metadata_type, buf+9, namelen);
4151 mddev->metadata_type[namelen] = 0;
4152 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4153 mddev->metadata_type[--namelen] = 0;
4154 mddev->persistent = 0;
4155 mddev->external = 1;
8bb93aac
N		 4156 mddev->major_version = 0;
4157 mddev->minor_version = 90;
6791875e 4158 goto out_unlock;
8bb93aac
N		 4159 }
4160 major = simple_strtoul(buf, &e, 10);
6791875e 4161 err = -EINVAL;
8bb93aac 4162 if (e==buf || *e != '.')
6791875e 4163 goto out_unlock;
8bb93aac
N		 4164 buf = e+1;
4165 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4166 if (e==buf || (*e && *e != '\n') )
6791875e
N		 4167 goto out_unlock;
4168 err = -ENOENT;
50511da3 4169 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
6791875e 4170 goto out_unlock;
8bb93aac
N		 4171 mddev->major_version = major;
4172 mddev->minor_version = minor;
4173 mddev->persistent = 1;
e691063a 4174 mddev->external = 0;
6791875e
N		 4175 err = 0;
4176out_unlock:
4177 mddev_unlock(mddev);
4178 return err ?: len;
8bb93aac
N		 4179}
4180
4181static struct md_sysfs_entry md_metadata =
750f199e 4182__ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4183
24dd469d 4184static ssize_t
fd01b88c 4185action_show(struct mddev *mddev, char *page)
24dd469d 4186{
7eec314d 4187 char *type = "idle";
b7b17c9b
N		 4188 unsigned long recovery = mddev->recovery;
4189 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
b6a9ce68 4190 type = "frozen";
b7b17c9b
N		 4191 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4192 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4193 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
ccfcc3c1 4194 type = "reshape";
b7b17c9b
N		 4195 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4196 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
24dd469d 4197 type = "resync";
b7b17c9b 4198 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
24dd469d
N		 4199 type = "check";
4200 else
4201 type = "repair";
b7b17c9b 4202 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
24dd469d
N		 4203 type = "recover";
4204 }
4205 return sprintf(page, "%s\n", type);
4206}
4207
4208static ssize_t
fd01b88c 4209action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4210{
7eec314d
N		 4211 if (!mddev->pers || !mddev->pers->sync_request)
4212 return -EINVAL;
4213
b6a9ce68
N		 4214 if (cmd_match(page, "frozen"))
4215 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4216 else
4217 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4218
4219 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
f851b60d 4220 flush_workqueue(md_misc_wq);
7eec314d
N		 4221 if (mddev->sync_thread) {
4222 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6791875e
N		 4223 if (mddev_lock(mddev) == 0) {
4224 md_reap_sync_thread(mddev);
4225 mddev_unlock(mddev);
4226 }
7eec314d 4227 }
03c902e1
N		 4228 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4229 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4230 return -EBUSY;
72a23c21
NB		 4231 else if (cmd_match(page, "resync"))
4232 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4233 else if (cmd_match(page, "recover")) {
4234 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4235 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4236 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4237 int err;
4238 if (mddev->pers->start_reshape == NULL)
4239 return -EINVAL;
6791875e
N		 4240 err = mddev_lock(mddev);
4241 if (!err) {
4242 err = mddev->pers->start_reshape(mddev);
4243 mddev_unlock(mddev);
4244 }
16484bf5
N		 4245 if (err)
4246 return err;
a99ac971 4247 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4248 } else {
bce74dac 4249 if (cmd_match(page, "check"))
7eec314d 4250 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4251 else if (!cmd_match(page, "repair"))
7eec314d
N		 4252 return -EINVAL;
4253 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4254 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4255 }
48c26ddc
N		 4256 if (mddev->ro == 2) {
4257 /* A write to sync_action is enough to justify
4258 * canceling read-auto mode
4259 */
4260 mddev->ro = 0;
4261 md_wakeup_thread(mddev->sync_thread);
4262 }
03c902e1 4263 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4264 md_wakeup_thread(mddev->thread);
00bcb4ac 4265 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4266 return len;
4267}
4268
c4a39551 4269static struct md_sysfs_entry md_scan_mode =
750f199e 4270__ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
c4a39551
JB		 4271
4272static ssize_t
4273last_sync_action_show(struct mddev *mddev, char *page)
4274{
4275 return sprintf(page, "%s\n", mddev->last_sync_action);
4276}
4277
4278static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4279
9d88883e 4280static ssize_t
fd01b88c 4281mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4282{
4283 return sprintf(page, "%llu\n",
7f7583d4
JM		 4284 (unsigned long long)
4285 atomic64_read(&mddev->resync_mismatches));
9d88883e
N		 4286}
4287
80ca3a44 4288static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4289
88202a0c 4290static ssize_t
fd01b88c 4291sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4292{
4293 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4294 mddev->sync_speed_min ? "local": "system");
4295}
4296
4297static ssize_t
fd01b88c 4298sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4299{
4300 int min;
4301 char *e;
4302 if (strncmp(buf, "system", 6)==0) {
4303 mddev->sync_speed_min = 0;
4304 return len;
4305 }
4306 min = simple_strtoul(buf, &e, 10);
4307 if (buf == e || (*e && *e != '\n') || min <= 0)
4308 return -EINVAL;
4309 mddev->sync_speed_min = min;
4310 return len;
4311}
4312
4313static struct md_sysfs_entry md_sync_min =
4314__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4315
4316static ssize_t
fd01b88c 4317sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4318{
4319 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4320 mddev->sync_speed_max ? "local": "system");
4321}
4322
4323static ssize_t
fd01b88c 4324sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4325{
4326 int max;
4327 char *e;
4328 if (strncmp(buf, "system", 6)==0) {
4329 mddev->sync_speed_max = 0;
4330 return len;
4331 }
4332 max = simple_strtoul(buf, &e, 10);
4333 if (buf == e || (*e && *e != '\n') || max <= 0)
4334 return -EINVAL;
4335 mddev->sync_speed_max = max;
4336 return len;
4337}
4338
4339static struct md_sysfs_entry md_sync_max =
4340__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4341
d7f3d291 4342static ssize_t
fd01b88c 4343degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4344{
4345 return sprintf(page, "%d\n", mddev->degraded);
4346}
4347static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4348
90b08710 4349static ssize_t
fd01b88c 4350sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4351{
4352 return sprintf(page, "%d\n", mddev->parallel_resync);
4353}
4354
4355static ssize_t
fd01b88c 4356sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4357{
4358 long n;
4359
b29bebd6 4360 if (kstrtol(buf, 10, &n))
90b08710
BS		 4361 return -EINVAL;
4362
4363 if (n != 0 && n != 1)
4364 return -EINVAL;
4365
4366 mddev->parallel_resync = n;
4367
4368 if (mddev->sync_thread)
4369 wake_up(&resync_wait);
4370
4371 return len;
4372}
4373
4374/* force parallel resync, even with shared block devices */
4375static struct md_sysfs_entry md_sync_force_parallel =
4376__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4377 sync_force_parallel_show, sync_force_parallel_store);
4378
88202a0c 4379static ssize_t
fd01b88c 4380sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4381{
4382 unsigned long resync, dt, db;
d1a7c503
N		 4383 if (mddev->curr_resync == 0)
4384 return sprintf(page, "none\n");
9687a60c
AN		 4385 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4386 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4387 if (!dt) dt++;
9687a60c
AN		 4388 db = resync - mddev->resync_mark_cnt;
4389 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4390}
4391
80ca3a44 4392static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4393
4394static ssize_t
fd01b88c 4395sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4396{
13ae864b 4397 unsigned long long max_sectors, resync;
88202a0c 4398
acb180b0
N		 4399 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4400 return sprintf(page, "none\n");
4401
72f36d59
N		 4402 if (mddev->curr_resync == 1 ||
4403 mddev->curr_resync == 2)
4404 return sprintf(page, "delayed\n");
4405
c804cdec
N		 4406 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4407 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 4408 max_sectors = mddev->resync_max_sectors;
88202a0c 4409 else
58c0fed4 4410 max_sectors = mddev->dev_sectors;
88202a0c 4411
acb180b0 4412 resync = mddev->curr_resync_completed;
13ae864b 4413 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4414}
4415
750f199e
N		 4416static struct md_sysfs_entry md_sync_completed =
4417 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
88202a0c 4418
5e96ee65 4419static ssize_t
fd01b88c 4420min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4421{
4422 return sprintf(page, "%llu\n",
4423 (unsigned long long)mddev->resync_min);
4424}
4425static ssize_t
fd01b88c 4426min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4427{
4428 unsigned long long min;
23da422b 4429 int err;
23da422b 4430
b29bebd6 4431 if (kstrtoull(buf, 10, &min))
5e96ee65 4432 return -EINVAL;
23da422b
N		 4433
4434 spin_lock(&mddev->lock);
4435 err = -EINVAL;
5e96ee65 4436 if (min > mddev->resync_max)
23da422b
N		 4437 goto out_unlock;
4438
4439 err = -EBUSY;
5e96ee65 4440 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4441 goto out_unlock;
5e96ee65 4442
50c37b13
N		 4443 /* Round down to multiple of 4K for safety */
4444 mddev->resync_min = round_down(min, 8);
23da422b 4445 err = 0;
5e96ee65 4446
23da422b
N		 4447out_unlock:
4448 spin_unlock(&mddev->lock);
4449 return err ?: len;
5e96ee65
NB		 4450}
4451
4452static struct md_sysfs_entry md_min_sync =
4453__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4454
c6207277 4455static ssize_t
fd01b88c 4456max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4457{
4458 if (mddev->resync_max == MaxSector)
4459 return sprintf(page, "max\n");
4460 else
4461 return sprintf(page, "%llu\n",
4462 (unsigned long long)mddev->resync_max);
4463}
4464static ssize_t
fd01b88c 4465max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277 4466{
23da422b
N		 4467 int err;
4468 spin_lock(&mddev->lock);
c6207277
N		 4469 if (strncmp(buf, "max", 3) == 0)
4470 mddev->resync_max = MaxSector;
4471 else {
5e96ee65 4472 unsigned long long max;
23da422b
N		 4473 int chunk;
4474
4475 err = -EINVAL;
b29bebd6 4476 if (kstrtoull(buf, 10, &max))
23da422b 4477 goto out_unlock;
5e96ee65 4478 if (max < mddev->resync_min)
23da422b
N		 4479 goto out_unlock;
4480
4481 err = -EBUSY;
c6207277 4482 if (max < mddev->resync_max &&
4d484a4a 4483 mddev->ro == 0 &&
c6207277 4484 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4485 goto out_unlock;
c6207277
N		 4486
4487 /* Must be a multiple of chunk_size */
23da422b
N		 4488 chunk = mddev->chunk_sectors;
4489 if (chunk) {
2ac06c33 4490 sector_t temp = max;
23da422b
N		 4491
4492 err = -EINVAL;
4493 if (sector_div(temp, chunk))
4494 goto out_unlock;
c6207277
N		 4495 }
4496 mddev->resync_max = max;
4497 }
4498 wake_up(&mddev->recovery_wait);
23da422b
N		 4499 err = 0;
4500out_unlock:
4501 spin_unlock(&mddev->lock);
4502 return err ?: len;
c6207277
N		 4503}
4504
4505static struct md_sysfs_entry md_max_sync =
4506__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4507
e464eafd 4508static ssize_t
fd01b88c 4509suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4510{
4511 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4512}
4513
4514static ssize_t
fd01b88c 4515suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4516{
4517 char *e;
4518 unsigned long long new = simple_strtoull(buf, &e, 10);
6791875e
N		 4519 unsigned long long old;
4520 int err;
e464eafd 4521
e464eafd
N		 4522 if (buf == e || (*e && *e != '\n'))
4523 return -EINVAL;
23ddff37 4524
6791875e
N		 4525 err = mddev_lock(mddev);
4526 if (err)
4527 return err;
4528 err = -EINVAL;
4529 if (mddev->pers == NULL ||
4530 mddev->pers->quiesce == NULL)
4531 goto unlock;
4532 old = mddev->suspend_lo;
23ddff37
N		 4533 mddev->suspend_lo = new;
4534 if (new >= old)
4535 /* Shrinking suspended region */
e464eafd 4536 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4537 else {
4538 /* Expanding suspended region - need to wait */
4539 mddev->pers->quiesce(mddev, 1);
4540 mddev->pers->quiesce(mddev, 0);
4541 }
6791875e
N		 4542 err = 0;
4543unlock:
4544 mddev_unlock(mddev);
4545 return err ?: len;
e464eafd
N		 4546}
4547static struct md_sysfs_entry md_suspend_lo =
4548__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4549
e464eafd 4550static ssize_t
fd01b88c 4551suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4552{
4553 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4554}
4555
4556static ssize_t
fd01b88c 4557suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4558{
4559 char *e;
4560 unsigned long long new = simple_strtoull(buf, &e, 10);
6791875e
N		 4561 unsigned long long old;
4562 int err;
e464eafd 4563
e464eafd
N		 4564 if (buf == e || (*e && *e != '\n'))
4565 return -EINVAL;
23ddff37 4566
6791875e
N		 4567 err = mddev_lock(mddev);
4568 if (err)
4569 return err;
4570 err = -EINVAL;
4571 if (mddev->pers == NULL ||
4572 mddev->pers->quiesce == NULL)
4573 goto unlock;
4574 old = mddev->suspend_hi;
23ddff37
N		 4575 mddev->suspend_hi = new;
4576 if (new <= old)
4577 /* Shrinking suspended region */
4578 mddev->pers->quiesce(mddev, 2);
4579 else {
4580 /* Expanding suspended region - need to wait */
e464eafd
N		 4581 mddev->pers->quiesce(mddev, 1);
4582 mddev->pers->quiesce(mddev, 0);
23ddff37 4583 }
6791875e
N		 4584 err = 0;
4585unlock:
4586 mddev_unlock(mddev);
4587 return err ?: len;
e464eafd
N		 4588}
4589static struct md_sysfs_entry md_suspend_hi =
4590__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4591
08a02ecd 4592static ssize_t
fd01b88c 4593reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4594{
4595 if (mddev->reshape_position != MaxSector)
4596 return sprintf(page, "%llu\n",
4597 (unsigned long long)mddev->reshape_position);
4598 strcpy(page, "none\n");
4599 return 5;
4600}
4601
4602static ssize_t
fd01b88c 4603reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd 4604{
c6563a8c 4605 struct md_rdev *rdev;
08a02ecd 4606 char *e;
6791875e 4607 int err;
08a02ecd 4608 unsigned long long new = simple_strtoull(buf, &e, 10);
6791875e 4609
08a02ecd
N		 4610 if (buf == e || (*e && *e != '\n'))
4611 return -EINVAL;
6791875e
N		 4612 err = mddev_lock(mddev);
4613 if (err)
4614 return err;
4615 err = -EBUSY;
4616 if (mddev->pers)
4617 goto unlock;
08a02ecd
N		 4618 mddev->reshape_position = new;
4619 mddev->delta_disks = 0;
2c810cdd 4620 mddev->reshape_backwards = 0;
08a02ecd
N		 4621 mddev->new_level = mddev->level;
4622 mddev->new_layout = mddev->layout;
664e7c41 4623 mddev->new_chunk_sectors = mddev->chunk_sectors;
c6563a8c
N		 4624 rdev_for_each(rdev, mddev)
4625 rdev->new_data_offset = rdev->data_offset;
6791875e
N		 4626 err = 0;
4627unlock:
4628 mddev_unlock(mddev);
4629 return err ?: len;
08a02ecd
N		 4630}
4631
4632static struct md_sysfs_entry md_reshape_position =
4633__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4634 reshape_position_store);
4635
2c810cdd
N		 4636static ssize_t
4637reshape_direction_show(struct mddev *mddev, char *page)
4638{
4639 return sprintf(page, "%s\n",
4640 mddev->reshape_backwards ? "backwards" : "forwards");
4641}
4642
4643static ssize_t
4644reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4645{
4646 int backwards = 0;
6791875e
N		 4647 int err;
4648
2c810cdd
N		 4649 if (cmd_match(buf, "forwards"))
4650 backwards = 0;
4651 else if (cmd_match(buf, "backwards"))
4652 backwards = 1;
4653 else
4654 return -EINVAL;
4655 if (mddev->reshape_backwards == backwards)
4656 return len;
4657
6791875e
N		 4658 err = mddev_lock(mddev);
4659 if (err)
4660 return err;
2c810cdd
N		 4661 /* check if we are allowed to change */
4662 if (mddev->delta_disks)
6791875e
N		 4663 err = -EBUSY;
4664 else if (mddev->persistent &&
2c810cdd 4665 mddev->major_version == 0)
6791875e
N		 4666 err = -EINVAL;
4667 else
4668 mddev->reshape_backwards = backwards;
4669 mddev_unlock(mddev);
4670 return err ?: len;
2c810cdd
N		 4671}
4672
4673static struct md_sysfs_entry md_reshape_direction =
4674__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4675 reshape_direction_store);
4676
b522adcd 4677static ssize_t
fd01b88c 4678array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4679{
4680 if (mddev->external_size)
4681 return sprintf(page, "%llu\n",
4682 (unsigned long long)mddev->array_sectors/2);
4683 else
4684 return sprintf(page, "default\n");
4685}
4686
4687static ssize_t
fd01b88c 4688array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4689{
4690 sector_t sectors;
6791875e
N		 4691 int err;
4692
4693 err = mddev_lock(mddev);
4694 if (err)
4695 return err;
b522adcd
DW		 4696
4697 if (strncmp(buf, "default", 7) == 0) {
4698 if (mddev->pers)
4699 sectors = mddev->pers->size(mddev, 0, 0);
4700 else
4701 sectors = mddev->array_sectors;
4702
4703 mddev->external_size = 0;
4704 } else {
4705 if (strict_blocks_to_sectors(buf, &sectors) < 0)
6791875e
N		 4706 err = -EINVAL;
4707 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4708 err = -E2BIG;
4709 else
4710 mddev->external_size = 1;
b522adcd
DW		 4711 }
4712
6791875e
N		 4713 if (!err) {
4714 mddev->array_sectors = sectors;
4715 if (mddev->pers) {
4716 set_capacity(mddev->gendisk, mddev->array_sectors);
4717 revalidate_disk(mddev->gendisk);
4718 }
cbe6ef1d 4719 }
6791875e
N		 4720 mddev_unlock(mddev);
4721 return err ?: len;
b522adcd
DW		 4722}
4723
4724static struct md_sysfs_entry md_array_size =
4725__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4726 array_size_store);
e464eafd 4727
eae1701f
N		 4728static struct attribute *md_default_attrs[] = {
4729 &md_level.attr,
d4dbd025 4730 &md_layout.attr,
eae1701f 4731 &md_raid_disks.attr,
3b34380a 4732 &md_chunk_size.attr,
a35b0d69 4733 &md_size.attr,
a94213b1 4734 &md_resync_start.attr,
8bb93aac 4735 &md_metadata.attr,
6d7ff738 4736 &md_new_device.attr,
16f17b39 4737 &md_safe_delay.attr,
9e653b63 4738 &md_array_state.attr,
08a02ecd 4739 &md_reshape_position.attr,
2c810cdd 4740 &md_reshape_direction.attr,
b522adcd 4741 &md_array_size.attr,
1e50915f 4742 &max_corr_read_errors.attr,
411036fa
N		 4743 NULL,
4744};
4745
4746static struct attribute *md_redundancy_attrs[] = {
24dd469d 4747 &md_scan_mode.attr,
c4a39551 4748 &md_last_scan_mode.attr,
9d88883e 4749 &md_mismatches.attr,
88202a0c
N		 4750 &md_sync_min.attr,
4751 &md_sync_max.attr,
4752 &md_sync_speed.attr,
90b08710 4753 &md_sync_force_parallel.attr,
88202a0c 4754 &md_sync_completed.attr,
5e96ee65 4755 &md_min_sync.attr,
c6207277 4756 &md_max_sync.attr,
e464eafd
N		 4757 &md_suspend_lo.attr,
4758 &md_suspend_hi.attr,
9b1d1dac 4759 &md_bitmap.attr,
d7f3d291 4760 &md_degraded.attr,
eae1701f
N		 4761 NULL,
4762};
411036fa
N		 4763static struct attribute_group md_redundancy_group = {
4764 .name = NULL,
4765 .attrs = md_redundancy_attrs,
4766};
4767
eae1701f
N		 4768static ssize_t
4769md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4770{
4771 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4772 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4773 ssize_t rv;
eae1701f
N		 4774
4775 if (!entry->show)
4776 return -EIO;
af8a2434
N		 4777 spin_lock(&all_mddevs_lock);
4778 if (list_empty(&mddev->all_mddevs)) {
4779 spin_unlock(&all_mddevs_lock);
4780 return -EBUSY;
4781 }
4782 mddev_get(mddev);
4783 spin_unlock(&all_mddevs_lock);
4784
b7b17c9b 4785 rv = entry->show(mddev, page);
af8a2434 4786 mddev_put(mddev);
96de1e66 4787 return rv;
eae1701f
N		 4788}
4789
4790static ssize_t
4791md_attr_store(struct kobject *kobj, struct attribute *attr,
4792 const char *page, size_t length)
4793{
4794 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4795 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4796 ssize_t rv;
eae1701f
N		 4797
4798 if (!entry->store)
4799 return -EIO;
67463acb
N		 4800 if (!capable(CAP_SYS_ADMIN))
4801 return -EACCES;
af8a2434
N		 4802 spin_lock(&all_mddevs_lock);
4803 if (list_empty(&mddev->all_mddevs)) {
4804 spin_unlock(&all_mddevs_lock);
4805 return -EBUSY;
4806 }
4807 mddev_get(mddev);
4808 spin_unlock(&all_mddevs_lock);
6791875e 4809 rv = entry->store(mddev, page, length);
af8a2434 4810 mddev_put(mddev);
96de1e66 4811 return rv;
eae1701f
N		 4812}
4813
4814static void md_free(struct kobject *ko)
4815{
fd01b88c 4816 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4817
4818 if (mddev->sysfs_state)
4819 sysfs_put(mddev->sysfs_state);
4820
6cd18e71
N		 4821 if (mddev->queue)
4822 blk_cleanup_queue(mddev->queue);
a21d1504
N		 4823 if (mddev->gendisk) {
4824 del_gendisk(mddev->gendisk);
4825 put_disk(mddev->gendisk);
4826 }
a21d1504 4827
eae1701f
N		 4828 kfree(mddev);
4829}
4830
52cf25d0 4831static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4832 .show = md_attr_show,
4833 .store = md_attr_store,
4834};
4835static struct kobj_type md_ktype = {
4836 .release = md_free,
4837 .sysfs_ops = &md_sysfs_ops,
4838 .default_attrs = md_default_attrs,
4839};
4840
1da177e4
LT		 4841int mdp_major = 0;
4842
5fd3a17e
DW		 4843static void mddev_delayed_delete(struct work_struct *ws)
4844{
fd01b88c 4845 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4846
43a70507 4847 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4848 kobject_del(&mddev->kobj);
4849 kobject_put(&mddev->kobj);
4850}
4851
efeb53c0 4852static int md_alloc(dev_t dev, char *name)
1da177e4 4853{
48c9c27b 4854 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4855 struct mddev *mddev = mddev_find(dev);
1da177e4 4856 struct gendisk *disk;
efeb53c0
N		 4857 int partitioned;
4858 int shift;
4859 int unit;
3830c62f 4860 int error;
1da177e4
LT		 4861
4862 if (!mddev)
efeb53c0
N		 4863 return -ENODEV;
4864
4865 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4866 shift = partitioned ? MdpMinorShift : 0;
4867 unit = MINOR(mddev->unit) >> shift;
1da177e4 4868
e804ac78
TH		 4869 /* wait for any previous instance of this device to be
4870 * completely removed (mddev_delayed_delete).
d3374825 4871 */
e804ac78 4872 flush_workqueue(md_misc_wq);
d3374825 4873
48c9c27b 4874 mutex_lock(&disks_mutex);
0909dc44
N		 4875 error = -EEXIST;
4876 if (mddev->gendisk)
4877 goto abort;
efeb53c0
N		 4878
4879 if (name) {
4880 /* Need to ensure that 'name' is not a duplicate.
4881 */
fd01b88c 4882 struct mddev *mddev2;
efeb53c0
N		 4883 spin_lock(&all_mddevs_lock);
4884
4885 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4886 if (mddev2->gendisk &&
4887 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4888 spin_unlock(&all_mddevs_lock);
0909dc44 4889 goto abort;
efeb53c0
N		 4890 }
4891 spin_unlock(&all_mddevs_lock);
1da177e4 4892 }
8b765398 4893
0909dc44 4894 error = -ENOMEM;
8b765398 4895 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4896 if (!mddev->queue)
4897 goto abort;
409c57f3
N		 4898 mddev->queue->queuedata = mddev;
4899
409c57f3 4900 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 4901 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 4902
1da177e4
LT		 4903 disk = alloc_disk(1 << shift);
4904 if (!disk) {
8b765398
N		 4905 blk_cleanup_queue(mddev->queue);
4906 mddev->queue = NULL;
0909dc44 4907 goto abort;
1da177e4 4908 }
efeb53c0 4909 disk->major = MAJOR(mddev->unit);
1da177e4 4910 disk->first_minor = unit << shift;
efeb53c0
N		 4911 if (name)
4912 strcpy(disk->disk_name, name);
4913 else if (partitioned)
1da177e4 4914 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4915 else
1da177e4 4916 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4917 disk->fops = &md_fops;
4918 disk->private_data = mddev;
4919 disk->queue = mddev->queue;
b0140891 4920 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4921 /* Allow extended partitions. This makes the
d3374825 4922 * 'mdp' device redundant, but we can't really
92850bbd
N		 4923 * remove it now.
4924 */
4925 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4926 mddev->gendisk = disk;
b0140891
N		 4927 /* As soon as we call add_disk(), another thread could get
4928 * through to md_open, so make sure it doesn't get too far
4929 */
4930 mutex_lock(&mddev->open_mutex);
4931 add_disk(disk);
4932
ed9e1982
TH		 4933 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4934 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4935 if (error) {
4936 /* This isn't possible, but as kobject_init_and_add is marked
4937 * __must_check, we must do something with the result
4938 */
5e55e2f5
N		 4939 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4940 disk->disk_name);
0909dc44
N		 4941 error = 0;
4942 }
00bcb4ac
N		 4943 if (mddev->kobj.sd &&
4944 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4945 printk(KERN_DEBUG "pointless warning\n");
b0140891 4946 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4947 abort:
4948 mutex_unlock(&disks_mutex);
00bcb4ac 4949 if (!error && mddev->kobj.sd) {
3830c62f 4950 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4951 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4952 }
d3374825 4953 mddev_put(mddev);
0909dc44 4954 return error;
efeb53c0
N		 4955}
4956
4957static struct kobject *md_probe(dev_t dev, int *part, void *data)
4958{
4959 md_alloc(dev, NULL);
1da177e4
LT		 4960 return NULL;
4961}
4962
efeb53c0
N		 4963static int add_named_array(const char *val, struct kernel_param *kp)
4964{
4965 /* val must be "md_*" where * is not all digits.
4966 * We allocate an array with a large free minor number, and
4967 * set the name to val. val must not already be an active name.
4968 */
4969 int len = strlen(val);
4970 char buf[DISK_NAME_LEN];
4971
4972 while (len && val[len-1] == '\n')
4973 len--;
4974 if (len >= DISK_NAME_LEN)
4975 return -E2BIG;
4976 strlcpy(buf, val, len+1);
4977 if (strncmp(buf, "md_", 3) != 0)
4978 return -EINVAL;
4979 return md_alloc(0, buf);
4980}
4981
1da177e4
LT		 4982static void md_safemode_timeout(unsigned long data)
4983{
fd01b88c 4984 struct mddev *mddev = (struct mddev *) data;
1da177e4 4985
0fd62b86
NB		 4986 if (!atomic_read(&mddev->writes_pending)) {
4987 mddev->safemode = 1;
4988 if (mddev->external)
00bcb4ac 4989 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4990 }
1da177e4
LT		 4991 md_wakeup_thread(mddev->thread);
4992}
4993
6ff8d8ec 4994static int start_dirty_degraded;
1da177e4 4995
fd01b88c 4996int md_run(struct mddev *mddev)
1da177e4 4997{
2604b703 4998 int err;
3cb03002 4999 struct md_rdev *rdev;
84fc4b56 5000 struct md_personality *pers;
1da177e4 5001
a757e64c
N		 5002 if (list_empty(&mddev->disks))
5003 /* cannot run an array with no devices.. */
1da177e4 5004 return -EINVAL;
1da177e4
LT		 5005
5006 if (mddev->pers)
5007 return -EBUSY;
bb4f1e9d
N		 5008 /* Cannot run until previous stop completes properly */
5009 if (mddev->sysfs_active)
5010 return -EBUSY;
b6eb127d 5011
1da177e4
LT		 5012 /*
5013 * Analyze all RAID superblock(s)
5014 */
1ec4a939
N		 5015 if (!mddev->raid_disks) {
5016 if (!mddev->persistent)
5017 return -EINVAL;
a757e64c 5018 analyze_sbs(mddev);
1ec4a939 5019 }
1da177e4 5020
d9d166c2
N		 5021 if (mddev->level != LEVEL_NONE)
5022 request_module("md-level-%d", mddev->level);
5023 else if (mddev->clevel[0])
5024 request_module("md-%s", mddev->clevel);
1da177e4
LT		 5025
5026 /*
5027 * Drop all container device buffers, from now on
5028 * the only valid external interface is through the md
5029 * device.
1da177e4 5030 */
dafb20fa 5031 rdev_for_each(rdev, mddev) {
b2d444d7 5032 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5033 continue;
5034 sync_blockdev(rdev->bdev);
f98393a6 5035 invalidate_bdev(rdev->bdev);
f0d76d70
N		 5036
5037 /* perform some consistency tests on the device.
5038 * We don't want the data to overlap the metadata,
58c0fed4 5039 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 5040 */
a6ff7e08
JB		 5041 if (rdev->meta_bdev) {
5042 /* Nothing to check */;
5043 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 5044 if (mddev->dev_sectors &&
5045 rdev->data_offset + mddev->dev_sectors
0f420358 5046 > rdev->sb_start) {
f0d76d70
N		 5047 printk("md: %s: data overlaps metadata\n",
5048 mdname(mddev));
5049 return -EINVAL;
5050 }
5051 } else {
0f420358 5052 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 5053 > rdev->data_offset) {
5054 printk("md: %s: metadata overlaps data\n",
5055 mdname(mddev));
5056 return -EINVAL;
5057 }
5058 }
00bcb4ac 5059 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 5060 }
5061
a167f663 5062 if (mddev->bio_set == NULL)
395c72a7 5063 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
a167f663 5064
1da177e4 5065 spin_lock(&pers_lock);
d9d166c2 5066 pers = find_pers(mddev->level, mddev->clevel);
2604b703 5067 if (!pers || !try_module_get(pers->owner)) {
1da177e4 5068 spin_unlock(&pers_lock);
d9d166c2
N		 5069 if (mddev->level != LEVEL_NONE)
5070 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5071 mddev->level);
5072 else
5073 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5074 mddev->clevel);
1da177e4
LT		 5075 return -EINVAL;
5076 }
1da177e4 5077 spin_unlock(&pers_lock);
34817e8c
N		 5078 if (mddev->level != pers->level) {
5079 mddev->level = pers->level;
5080 mddev->new_level = pers->level;
5081 }
d9d166c2 5082 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 5083
f6705578 5084 if (mddev->reshape_position != MaxSector &&
63c70c4f 5085 pers->start_reshape == NULL) {
f6705578 5086 /* This personality cannot handle reshaping... */
f6705578
N		 5087 module_put(pers->owner);
5088 return -EINVAL;
5089 }
5090
7dd5e7c3
N		 5091 if (pers->sync_request) {
5092 /* Warn if this is a potentially silly
5093 * configuration.
5094 */
5095 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5096 struct md_rdev *rdev2;
7dd5e7c3 5097 int warned = 0;
159ec1fc 5098
dafb20fa
N		 5099 rdev_for_each(rdev, mddev)
5100 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N		 5101 if (rdev < rdev2 &&
5102 rdev->bdev->bd_contains ==
5103 rdev2->bdev->bd_contains) {
5104 printk(KERN_WARNING
5105 "%s: WARNING: %s appears to be"
5106 " on the same physical disk as"
5107 " %s.\n",
5108 mdname(mddev),
5109 bdevname(rdev->bdev,b),
5110 bdevname(rdev2->bdev,b2));
5111 warned = 1;
5112 }
5113 }
159ec1fc 5114
7dd5e7c3
N		 5115 if (warned)
5116 printk(KERN_WARNING
5117 "True protection against single-disk"
5118 " failure might be compromised.\n");
5119 }
5120
657390d2 5121 mddev->recovery = 0;
58c0fed4
AN		 5122 /* may be over-ridden by personality */
5123 mddev->resync_max_sectors = mddev->dev_sectors;
5124
6ff8d8ec 5125 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5126
0f9552b5 5127 if (start_readonly && mddev->ro == 0)
f91de92e
N		 5128 mddev->ro = 2; /* read-only, but switch on first write */
5129
36d091f4 5130 err = pers->run(mddev);
13e53df3
AN		 5131 if (err)
5132 printk(KERN_ERR "md: pers->run() failed ...\n");
36d091f4 5133 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
b522adcd
DW		 5134 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5135 " but 'external_size' not in effect?\n", __func__);
5136 printk(KERN_ERR
5137 "md: invalid array_size %llu > default size %llu\n",
5138 (unsigned long long)mddev->array_sectors / 2,
36d091f4 5139 (unsigned long long)pers->size(mddev, 0, 0) / 2);
b522adcd 5140 err = -EINVAL;
b522adcd 5141 }
36d091f4 5142 if (err == 0 && pers->sync_request &&
ef99bf48 5143 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
f9209a32
GR		 5144 struct bitmap *bitmap;
5145
5146 bitmap = bitmap_create(mddev, -1);
5147 if (IS_ERR(bitmap)) {
5148 err = PTR_ERR(bitmap);
b15c2e57
N		 5149 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5150 mdname(mddev), err);
f9209a32
GR		 5151 } else
5152 mddev->bitmap = bitmap;
5153
b15c2e57 5154 }
1da177e4 5155 if (err) {
5aa61f42 5156 mddev_detach(mddev);
0c35bd47
N		 5157 if (mddev->private)
5158 pers->free(mddev, mddev->private);
36d091f4 5159 module_put(pers->owner);
32a7627c
N		 5160 bitmap_destroy(mddev);
5161 return err;
1da177e4 5162 }
5c675f83
N		 5163 if (mddev->queue) {
5164 mddev->queue->backing_dev_info.congested_data = mddev;
5165 mddev->queue->backing_dev_info.congested_fn = md_congested;
64590f45 5166 blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec);
5c675f83 5167 }
36d091f4 5168 if (pers->sync_request) {
00bcb4ac
N		 5169 if (mddev->kobj.sd &&
5170 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 5171 printk(KERN_WARNING
5172 "md: cannot register extra attributes for %s\n",
5173 mdname(mddev));
00bcb4ac 5174 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5175 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 5176 mddev->ro = 0;
5177
f72ffdd6 5178 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 5179 atomic_set(&mddev->max_corr_read_errors,
5180 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 5181 mddev->safemode = 0;
5182 mddev->safemode_timer.function = md_safemode_timeout;
5183 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 5184 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5185 mddev->in_sync = 1;
0ca69886 5186 smp_wmb();
36d091f4
N		 5187 spin_lock(&mddev->lock);
5188 mddev->pers = pers;
0ca69886 5189 mddev->ready = 1;
36d091f4 5190 spin_unlock(&mddev->lock);
dafb20fa 5191 rdev_for_each(rdev, mddev)
36fad858
NK		 5192 if (rdev->raid_disk >= 0)
5193 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 5194 /* failure here is OK */;
f72ffdd6 5195
1da177e4 5196 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f72ffdd6 5197
7a0a5355 5198 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 5199 md_update_sb(mddev, 0);
1da177e4 5200
d7603b7e 5201 md_new_event(mddev);
00bcb4ac
N		 5202 sysfs_notify_dirent_safe(mddev->sysfs_state);
5203 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 5204 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 5205 return 0;
5206}
390ee602 5207EXPORT_SYMBOL_GPL(md_run);
1da177e4 5208
fd01b88c 5209static int do_md_run(struct mddev *mddev)
fe60b014
N		 5210{
5211 int err;
5212
5213 err = md_run(mddev);
5214 if (err)
5215 goto out;
69e51b44
N		 5216 err = bitmap_load(mddev);
5217 if (err) {
5218 bitmap_destroy(mddev);
5219 goto out;
5220 }
0fd018af
JB		 5221
5222 md_wakeup_thread(mddev->thread);
5223 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5224
fe60b014
N		 5225 set_capacity(mddev->gendisk, mddev->array_sectors);
5226 revalidate_disk(mddev->gendisk);
f0b4f7e2 5227 mddev->changed = 1;
fe60b014
N		 5228 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5229out:
5230 return err;
5231}
5232
fd01b88c 5233static int restart_array(struct mddev *mddev)
1da177e4
LT		 5234{
5235 struct gendisk *disk = mddev->gendisk;
1da177e4 5236
80fab1d7 5237 /* Complain if it has no devices */
1da177e4 5238 if (list_empty(&mddev->disks))
80fab1d7
AN		 5239 return -ENXIO;
5240 if (!mddev->pers)
5241 return -EINVAL;
5242 if (!mddev->ro)
5243 return -EBUSY;
5244 mddev->safemode = 0;
5245 mddev->ro = 0;
5246 set_disk_ro(disk, 0);
5247 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5248 mdname(mddev));
5249 /* Kick recovery or resync if necessary */
5250 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5251 md_wakeup_thread(mddev->thread);
5252 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5253 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5254 return 0;
1da177e4
LT		 5255}
5256
fd01b88c 5257static void md_clean(struct mddev *mddev)
6177b472
N		 5258{
5259 mddev->array_sectors = 0;
5260 mddev->external_size = 0;
5261 mddev->dev_sectors = 0;
5262 mddev->raid_disks = 0;
5263 mddev->recovery_cp = 0;
5264 mddev->resync_min = 0;
5265 mddev->resync_max = MaxSector;
5266 mddev->reshape_position = MaxSector;
5267 mddev->external = 0;
5268 mddev->persistent = 0;
5269 mddev->level = LEVEL_NONE;
5270 mddev->clevel[0] = 0;
5271 mddev->flags = 0;
5272 mddev->ro = 0;
5273 mddev->metadata_type[0] = 0;
5274 mddev->chunk_sectors = 0;
5275 mddev->ctime = mddev->utime = 0;
5276 mddev->layout = 0;
5277 mddev->max_disks = 0;
5278 mddev->events = 0;
a8707c08 5279 mddev->can_decrease_events = 0;
6177b472 5280 mddev->delta_disks = 0;
2c810cdd 5281 mddev->reshape_backwards = 0;
6177b472
N		 5282 mddev->new_level = LEVEL_NONE;
5283 mddev->new_layout = 0;
5284 mddev->new_chunk_sectors = 0;
5285 mddev->curr_resync = 0;
7f7583d4 5286 atomic64_set(&mddev->resync_mismatches, 0);
6177b472
N		 5287 mddev->suspend_lo = mddev->suspend_hi = 0;
5288 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5289 mddev->recovery = 0;
5290 mddev->in_sync = 0;
f0b4f7e2 5291 mddev->changed = 0;
6177b472 5292 mddev->degraded = 0;
6177b472 5293 mddev->safemode = 0;
050b6615 5294 mddev->merge_check_needed = 0;
6177b472
N		 5295 mddev->bitmap_info.offset = 0;
5296 mddev->bitmap_info.default_offset = 0;
6409bb05 5297 mddev->bitmap_info.default_space = 0;
6177b472
N		 5298 mddev->bitmap_info.chunksize = 0;
5299 mddev->bitmap_info.daemon_sleep = 0;
5300 mddev->bitmap_info.max_write_behind = 0;
5301}
5302
fd01b88c 5303static void __md_stop_writes(struct mddev *mddev)
a047e125 5304{
293467aa
GR		 5305 if (mddev_is_clustered(mddev))
5306 md_cluster_ops->metadata_update_start(mddev);
6b6204ee 5307 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
f851b60d 5308 flush_workqueue(md_misc_wq);
a047e125 5309 if (mddev->sync_thread) {
a047e125 5310 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 5311 md_reap_sync_thread(mddev);
a047e125
N		 5312 }
5313
5314 del_timer_sync(&mddev->safemode_timer);
5315
5316 bitmap_flush(mddev);
5317 md_super_wait(mddev);
5318
b6d428c6 5319 if (mddev->ro == 0 &&
7a0a5355 5320 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
a047e125
N		 5321 /* mark array as shutdown cleanly */
5322 mddev->in_sync = 1;
5323 md_update_sb(mddev, 1);
5324 }
293467aa
GR		 5325 if (mddev_is_clustered(mddev))
5326 md_cluster_ops->metadata_update_finish(mddev);
a047e125 5327}
defad61a 5328
fd01b88c 5329void md_stop_writes(struct mddev *mddev)
defad61a 5330{
29f097c4 5331 mddev_lock_nointr(mddev);
defad61a
N		 5332 __md_stop_writes(mddev);
5333 mddev_unlock(mddev);
5334}
390ee602 5335EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5336
5aa61f42
N		 5337static void mddev_detach(struct mddev *mddev)
5338{
5339 struct bitmap *bitmap = mddev->bitmap;
5340 /* wait for behind writes to complete */
5341 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5342 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5343 mdname(mddev));
5344 /* need to kick something here to make sure I/O goes? */
5345 wait_event(bitmap->behind_wait,
5346 atomic_read(&bitmap->behind_writes) == 0);
5347 }
36d091f4 5348 if (mddev->pers && mddev->pers->quiesce) {
5aa61f42
N		 5349 mddev->pers->quiesce(mddev, 1);
5350 mddev->pers->quiesce(mddev, 0);
5351 }
5352 md_unregister_thread(&mddev->thread);
5353 if (mddev->queue)
5354 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5355}
5356
5eff3c43 5357static void __md_stop(struct mddev *mddev)
6177b472 5358{
36d091f4 5359 struct md_personality *pers = mddev->pers;
5aa61f42 5360 mddev_detach(mddev);
36d091f4
N		 5361 spin_lock(&mddev->lock);
5362 mddev->ready = 0;
6177b472 5363 mddev->pers = NULL;
36d091f4
N		 5364 spin_unlock(&mddev->lock);
5365 pers->free(mddev, mddev->private);
5366 if (pers->sync_request && mddev->to_remove == NULL)
5367 mddev->to_remove = &md_redundancy_group;
5368 module_put(pers->owner);
cca9cf90 5369 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5370}
5eff3c43
N		 5371
5372void md_stop(struct mddev *mddev)
5373{
5374 /* stop the array and free an attached data structures.
5375 * This is called from dm-raid
5376 */
5377 __md_stop(mddev);
5378 bitmap_destroy(mddev);
5379 if (mddev->bio_set)
5380 bioset_free(mddev->bio_set);
5381}
5382
390ee602 5383EXPORT_SYMBOL_GPL(md_stop);
6177b472 5384
a05b7ea0 5385static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N		 5386{
5387 int err = 0;
30b8feb7
N		 5388 int did_freeze = 0;
5389
5390 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5391 did_freeze = 1;
5392 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5393 md_wakeup_thread(mddev->thread);
5394 }
f851b60d 5395 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5396 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5397 if (mddev->sync_thread)
30b8feb7
N		 5398 /* Thread might be blocked waiting for metadata update
5399 * which will now never happen */
5400 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5401
30b8feb7 5402 mddev_unlock(mddev);
f851b60d
N		 5403 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5404 &mddev->recovery));
30b8feb7
N		 5405 mddev_lock_nointr(mddev);
5406
a4bd82d0 5407 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5408 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7 5409 mddev->sync_thread ||
f851b60d 5410 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5411 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
a4bd82d0 5412 printk("md: %s still in use.\n",mdname(mddev));
30b8feb7
N		 5413 if (did_freeze) {
5414 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5415 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5416 md_wakeup_thread(mddev->thread);
5417 }
a4bd82d0
N		 5418 err = -EBUSY;
5419 goto out;
5420 }
5421 if (mddev->pers) {
defad61a 5422 __md_stop_writes(mddev);
a4bd82d0
N		 5423
5424 err = -ENXIO;
5425 if (mddev->ro==1)
5426 goto out;
5427 mddev->ro = 1;
5428 set_disk_ro(mddev->gendisk, 1);
5429 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d
N		 5430 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5431 md_wakeup_thread(mddev->thread);
00bcb4ac 5432 sysfs_notify_dirent_safe(mddev->sysfs_state);
30b8feb7 5433 err = 0;
a4bd82d0
N		 5434 }
5435out:
5436 mutex_unlock(&mddev->open_mutex);
5437 return err;
5438}
5439
9e653b63
N		 5440/* mode:
5441 * 0 - completely stop and dis-assemble array
9e653b63
N		 5442 * 2 - stop but do not disassemble array
5443 */
f72ffdd6 5444static int do_md_stop(struct mddev *mddev, int mode,
a05b7ea0 5445 struct block_device *bdev)
1da177e4 5446{
1da177e4 5447 struct gendisk *disk = mddev->gendisk;
3cb03002 5448 struct md_rdev *rdev;
30b8feb7
N		 5449 int did_freeze = 0;
5450
5451 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5452 did_freeze = 1;
5453 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5454 md_wakeup_thread(mddev->thread);
5455 }
f851b60d 5456 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 5457 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 5458 if (mddev->sync_thread)
30b8feb7
N		 5459 /* Thread might be blocked waiting for metadata update
5460 * which will now never happen */
5461 wake_up_process(mddev->sync_thread->tsk);
f851b60d 5462
30b8feb7 5463 mddev_unlock(mddev);
f851b60d
N		 5464 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5465 !test_bit(MD_RECOVERY_RUNNING,
5466 &mddev->recovery)));
30b8feb7 5467 mddev_lock_nointr(mddev);
1da177e4 5468
c8c00a69 5469 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5470 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7
N		 5471 mddev->sysfs_active ||
5472 mddev->sync_thread ||
f851b60d 5473 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
30b8feb7 5474 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
df5b20cf 5475 printk("md: %s still in use.\n",mdname(mddev));
6e17b027 5476 mutex_unlock(&mddev->open_mutex);
30b8feb7
N		 5477 if (did_freeze) {
5478 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 5479 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 5480 md_wakeup_thread(mddev->thread);
5481 }
260fa034
N		 5482 return -EBUSY;
5483 }
6e17b027 5484 if (mddev->pers) {
a4bd82d0
N		 5485 if (mddev->ro)
5486 set_disk_ro(disk, 0);
409c57f3 5487
defad61a 5488 __md_stop_writes(mddev);
5eff3c43 5489 __md_stop(mddev);
a4bd82d0 5490 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5491 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5492
a4bd82d0 5493 /* tell userspace to handle 'inactive' */
00bcb4ac 5494 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5495
dafb20fa 5496 rdev_for_each(rdev, mddev)
36fad858
NK		 5497 if (rdev->raid_disk >= 0)
5498 sysfs_unlink_rdev(mddev, rdev);
c4647292 5499
a4bd82d0 5500 set_capacity(disk, 0);
6e17b027 5501 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5502 mddev->changed = 1;
a4bd82d0 5503 revalidate_disk(disk);
0d4ca600 5504
a4bd82d0
N		 5505 if (mddev->ro)
5506 mddev->ro = 0;
6e17b027
N		 5507 } else
5508 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5509 /*
5510 * Free resources if final stop
5511 */
9e653b63 5512 if (mode == 0) {
1da177e4
LT		 5513 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5514
978f946b 5515 bitmap_destroy(mddev);
c3d9714e 5516 if (mddev->bitmap_info.file) {
4af1a041
N		 5517 struct file *f = mddev->bitmap_info.file;
5518 spin_lock(&mddev->lock);
c3d9714e 5519 mddev->bitmap_info.file = NULL;
4af1a041
N		 5520 spin_unlock(&mddev->lock);
5521 fput(f);
978f946b 5522 }
c3d9714e 5523 mddev->bitmap_info.offset = 0;
978f946b 5524
1da177e4
LT		 5525 export_array(mddev);
5526
6177b472 5527 md_clean(mddev);
934d9c23 5528 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5529 if (mddev->hold_active == UNTIL_STOP)
5530 mddev->hold_active = 0;
a4bd82d0 5531 }
3f9d99c1 5532 blk_integrity_unregister(disk);
d7603b7e 5533 md_new_event(mddev);
00bcb4ac 5534 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5535 return 0;
1da177e4
LT		 5536}
5537
fdee8ae4 5538#ifndef MODULE
fd01b88c 5539static void autorun_array(struct mddev *mddev)
1da177e4 5540{
3cb03002 5541 struct md_rdev *rdev;
1da177e4
LT		 5542 int err;
5543
a757e64c 5544 if (list_empty(&mddev->disks))
1da177e4 5545 return;
1da177e4
LT		 5546
5547 printk(KERN_INFO "md: running: ");
5548
dafb20fa 5549 rdev_for_each(rdev, mddev) {
1da177e4
LT		 5550 char b[BDEVNAME_SIZE];
5551 printk("<%s>", bdevname(rdev->bdev,b));
5552 }
5553 printk("\n");
5554
d710e138 5555 err = do_md_run(mddev);
1da177e4
LT		 5556 if (err) {
5557 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
a05b7ea0 5558 do_md_stop(mddev, 0, NULL);
1da177e4
LT		 5559 }
5560}
5561
5562/*
5563 * lets try to run arrays based on all disks that have arrived
5564 * until now. (those are in pending_raid_disks)
5565 *
5566 * the method: pick the first pending disk, collect all disks with
5567 * the same UUID, remove all from the pending list and put them into
5568 * the 'same_array' list. Then order this list based on superblock
5569 * update time (freshest comes first), kick out 'old' disks and
5570 * compare superblocks. If everything's fine then run it.
5571 *
5572 * If "unit" is allocated, then bump its reference count
5573 */
5574static void autorun_devices(int part)
5575{
3cb03002 5576 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5577 struct mddev *mddev;
1da177e4
LT		 5578 char b[BDEVNAME_SIZE];
5579
5580 printk(KERN_INFO "md: autorun ...\n");
5581 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5582 int unit;
1da177e4 5583 dev_t dev;
ad01c9e3 5584 LIST_HEAD(candidates);
1da177e4 5585 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5586 struct md_rdev, same_set);
1da177e4
LT		 5587
5588 printk(KERN_INFO "md: considering %s ...\n",
5589 bdevname(rdev0->bdev,b));
5590 INIT_LIST_HEAD(&candidates);
159ec1fc 5591 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5592 if (super_90_load(rdev, rdev0, 0) >= 0) {
5593 printk(KERN_INFO "md: adding %s ...\n",
5594 bdevname(rdev->bdev,b));
5595 list_move(&rdev->same_set, &candidates);
5596 }
5597 /*
5598 * now we have a set of devices, with all of them having
5599 * mostly sane superblocks. It's time to allocate the
5600 * mddev.
5601 */
e8703fe1
N		 5602 if (part) {
5603 dev = MKDEV(mdp_major,
5604 rdev0->preferred_minor << MdpMinorShift);
5605 unit = MINOR(dev) >> MdpMinorShift;
5606 } else {
5607 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5608 unit = MINOR(dev);
5609 }
5610 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5611 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5612 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5613 break;
5614 }
1da177e4
LT		 5615
5616 md_probe(dev, NULL, NULL);
5617 mddev = mddev_find(dev);
9bbbca3a
NB		 5618 if (!mddev || !mddev->gendisk) {
5619 if (mddev)
5620 mddev_put(mddev);
5621 printk(KERN_ERR
1da177e4
LT		 5622 "md: cannot allocate memory for md drive.\n");
5623 break;
5624 }
f72ffdd6 5625 if (mddev_lock(mddev))
1da177e4
LT		 5626 printk(KERN_WARNING "md: %s locked, cannot run\n",
5627 mdname(mddev));
5628 else if (mddev->raid_disks || mddev->major_version
5629 || !list_empty(&mddev->disks)) {
f72ffdd6 5630 printk(KERN_WARNING
1da177e4
LT		 5631 "md: %s already running, cannot run %s\n",
5632 mdname(mddev), bdevname(rdev0->bdev,b));
5633 mddev_unlock(mddev);
5634 } else {
5635 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5636 mddev->persistent = 1;
159ec1fc 5637 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5638 list_del_init(&rdev->same_set);
5639 if (bind_rdev_to_array(rdev, mddev))
5640 export_rdev(rdev);
5641 }
5642 autorun_array(mddev);
5643 mddev_unlock(mddev);
5644 }
5645 /* on success, candidates will be empty, on error
5646 * it won't...
5647 */
159ec1fc 5648 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5649 list_del_init(&rdev->same_set);
1da177e4 5650 export_rdev(rdev);
4b80991c 5651 }
1da177e4
LT		 5652 mddev_put(mddev);
5653 }
5654 printk(KERN_INFO "md: ... autorun DONE.\n");
5655}
fdee8ae4 5656#endif /* !MODULE */
1da177e4 5657
f72ffdd6 5658static int get_version(void __user *arg)
1da177e4
LT		 5659{
5660 mdu_version_t ver;
5661
5662 ver.major = MD_MAJOR_VERSION;
5663 ver.minor = MD_MINOR_VERSION;
5664 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5665
5666 if (copy_to_user(arg, &ver, sizeof(ver)))
5667 return -EFAULT;
5668
5669 return 0;
5670}
5671
f72ffdd6 5672static int get_array_info(struct mddev *mddev, void __user *arg)
1da177e4
LT		 5673{
5674 mdu_array_info_t info;
a9f326eb 5675 int nr,working,insync,failed,spare;
3cb03002 5676 struct md_rdev *rdev;
1da177e4 5677
1ca69c4b
N		 5678 nr = working = insync = failed = spare = 0;
5679 rcu_read_lock();
5680 rdev_for_each_rcu(rdev, mddev) {
1da177e4 5681 nr++;
b2d444d7 5682 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5683 failed++;
5684 else {
5685 working++;
b2d444d7 5686 if (test_bit(In_sync, &rdev->flags))
f72ffdd6 5687 insync++;
1da177e4
LT		 5688 else
5689 spare++;
5690 }
5691 }
1ca69c4b 5692 rcu_read_unlock();
1da177e4
LT		 5693
5694 info.major_version = mddev->major_version;
5695 info.minor_version = mddev->minor_version;
5696 info.patch_version = MD_PATCHLEVEL_VERSION;
5697 info.ctime = mddev->ctime;
5698 info.level = mddev->level;
58c0fed4
AN		 5699 info.size = mddev->dev_sectors / 2;
5700 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5701 info.size = -1;
1da177e4
LT		 5702 info.nr_disks = nr;
5703 info.raid_disks = mddev->raid_disks;
5704 info.md_minor = mddev->md_minor;
5705 info.not_persistent= !mddev->persistent;
5706
5707 info.utime = mddev->utime;
5708 info.state = 0;
5709 if (mddev->in_sync)
5710 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5711 if (mddev->bitmap && mddev->bitmap_info.offset)
9bd35920 5712 info.state |= (1<<MD_SB_BITMAP_PRESENT);
ca8895d9
GR		 5713 if (mddev_is_clustered(mddev))
5714 info.state |= (1<<MD_SB_CLUSTERED);
a9f326eb 5715 info.active_disks = insync;
1da177e4
LT		 5716 info.working_disks = working;
5717 info.failed_disks = failed;
5718 info.spare_disks = spare;
5719
5720 info.layout = mddev->layout;
9d8f0363 5721 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5722
5723 if (copy_to_user(arg, &info, sizeof(info)))
5724 return -EFAULT;
5725
5726 return 0;
5727}
5728
f72ffdd6 5729static int get_bitmap_file(struct mddev *mddev, void __user * arg)
32a7627c
N		 5730{
5731 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
f4ad3d38 5732 char *ptr;
4af1a041 5733 int err;
32a7627c 5734
60559da4 5735 file = kmalloc(sizeof(*file), GFP_NOIO);
32a7627c 5736 if (!file)
4af1a041 5737 return -ENOMEM;
32a7627c 5738
4af1a041
N		 5739 err = 0;
5740 spin_lock(&mddev->lock);
32a7627c 5741 /* bitmap disabled, zero the first byte and copy out */
4af1a041 5742 if (!mddev->bitmap_info.file)
32a7627c 5743 file->pathname[0] = '\0';
4af1a041
N		 5744 else if ((ptr = d_path(&mddev->bitmap_info.file->f_path,
5745 file->pathname, sizeof(file->pathname))),
5746 IS_ERR(ptr))
5747 err = PTR_ERR(ptr);
5748 else
5749 memmove(file->pathname, ptr,
5750 sizeof(file->pathname)-(ptr-file->pathname));
5751 spin_unlock(&mddev->lock);
32a7627c 5752
4af1a041
N		 5753 if (err == 0 &&
5754 copy_to_user(arg, file, sizeof(*file)))
32a7627c 5755 err = -EFAULT;
4af1a041 5756
32a7627c
N		 5757 kfree(file);
5758 return err;
5759}
5760
f72ffdd6 5761static int get_disk_info(struct mddev *mddev, void __user * arg)
1da177e4
LT		 5762{
5763 mdu_disk_info_t info;
3cb03002 5764 struct md_rdev *rdev;
1da177e4
LT		 5765
5766 if (copy_from_user(&info, arg, sizeof(info)))
5767 return -EFAULT;
5768
1ca69c4b 5769 rcu_read_lock();
57d051dc 5770 rdev = md_find_rdev_nr_rcu(mddev, info.number);
1da177e4
LT		 5771 if (rdev) {
5772 info.major = MAJOR(rdev->bdev->bd_dev);
5773 info.minor = MINOR(rdev->bdev->bd_dev);
5774 info.raid_disk = rdev->raid_disk;
5775 info.state = 0;
b2d444d7 5776 if (test_bit(Faulty, &rdev->flags))
1da177e4 5777 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5778 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5779 info.state |= (1<<MD_DISK_ACTIVE);
5780 info.state |= (1<<MD_DISK_SYNC);
5781 }
8ddf9efe
N		 5782 if (test_bit(WriteMostly, &rdev->flags))
5783 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5784 } else {
5785 info.major = info.minor = 0;
5786 info.raid_disk = -1;
5787 info.state = (1<<MD_DISK_REMOVED);
5788 }
1ca69c4b 5789 rcu_read_unlock();
1da177e4
LT		 5790
5791 if (copy_to_user(arg, &info, sizeof(info)))
5792 return -EFAULT;
5793
5794 return 0;
5795}
5796
f72ffdd6 5797static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
1da177e4
LT		 5798{
5799 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5800 struct md_rdev *rdev;
1da177e4
LT		 5801 dev_t dev = MKDEV(info->major,info->minor);
5802
1aee41f6
GR		 5803 if (mddev_is_clustered(mddev) &&
5804 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
fa8259da 5805 pr_err("%s: Cannot add to clustered mddev.\n",
1aee41f6
GR		 5806 mdname(mddev));
5807 return -EINVAL;
5808 }
5809
1da177e4
LT		 5810 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5811 return -EOVERFLOW;
5812
5813 if (!mddev->raid_disks) {
5814 int err;
5815 /* expecting a device which has a superblock */
5816 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5817 if (IS_ERR(rdev)) {
f72ffdd6 5818 printk(KERN_WARNING
1da177e4
LT		 5819 "md: md_import_device returned %ld\n",
5820 PTR_ERR(rdev));
5821 return PTR_ERR(rdev);
5822 }
5823 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5824 struct md_rdev *rdev0
5825 = list_entry(mddev->disks.next,
5826 struct md_rdev, same_set);
a9f326eb 5827 err = super_types[mddev->major_version]
1da177e4
LT		 5828 .load_super(rdev, rdev0, mddev->minor_version);
5829 if (err < 0) {
f72ffdd6 5830 printk(KERN_WARNING
1da177e4 5831 "md: %s has different UUID to %s\n",
f72ffdd6 5832 bdevname(rdev->bdev,b),
1da177e4
LT		 5833 bdevname(rdev0->bdev,b2));
5834 export_rdev(rdev);
5835 return -EINVAL;
5836 }
5837 }
5838 err = bind_rdev_to_array(rdev, mddev);
5839 if (err)
5840 export_rdev(rdev);
5841 return err;
5842 }
5843
5844 /*
5845 * add_new_disk can be used once the array is assembled
5846 * to add "hot spares". They must already have a superblock
5847 * written
5848 */
5849 if (mddev->pers) {
5850 int err;
5851 if (!mddev->pers->hot_add_disk) {
f72ffdd6 5852 printk(KERN_WARNING
1da177e4
LT		 5853 "%s: personality does not support diskops!\n",
5854 mdname(mddev));
5855 return -EINVAL;
5856 }
7b1e35f6
N		 5857 if (mddev->persistent)
5858 rdev = md_import_device(dev, mddev->major_version,
5859 mddev->minor_version);
5860 else
5861 rdev = md_import_device(dev, -1, -1);
1da177e4 5862 if (IS_ERR(rdev)) {
f72ffdd6 5863 printk(KERN_WARNING
1da177e4
LT		 5864 "md: md_import_device returned %ld\n",
5865 PTR_ERR(rdev));
5866 return PTR_ERR(rdev);
5867 }
1a855a06 5868 /* set saved_raid_disk if appropriate */
41158c7e
N		 5869 if (!mddev->persistent) {
5870 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5871 info->raid_disk < mddev->raid_disks) {
41158c7e 5872 rdev->raid_disk = info->raid_disk;
bf572541 5873 set_bit(In_sync, &rdev->flags);
8313b8e5 5874 clear_bit(Bitmap_sync, &rdev->flags);
bf572541 5875 } else
41158c7e 5876 rdev->raid_disk = -1;
f466722c 5877 rdev->saved_raid_disk = rdev->raid_disk;
41158c7e
N		 5878 } else
5879 super_types[mddev->major_version].
5880 validate_super(mddev, rdev);
bedd86b7 5881 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 5882 rdev->raid_disk != info->raid_disk) {
bedd86b7
N		 5883 /* This was a hot-add request, but events doesn't
5884 * match, so reject it.
5885 */
5886 export_rdev(rdev);
5887 return -EINVAL;
5888 }
5889
b2d444d7 5890 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5891 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5892 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5893 else
5894 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5895
1aee41f6
GR		 5896 /*
5897 * check whether the device shows up in other nodes
5898 */
5899 if (mddev_is_clustered(mddev)) {
5900 if (info->state & (1 << MD_DISK_CANDIDATE)) {
5901 /* Through --cluster-confirm */
5902 set_bit(Candidate, &rdev->flags);
fa8259da
GR		 5903 err = md_cluster_ops->new_disk_ack(mddev, true);
5904 if (err) {
5905 export_rdev(rdev);
5906 return err;
5907 }
1aee41f6
GR		 5908 } else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
5909 /* --add initiated by this node */
5910 err = md_cluster_ops->add_new_disk_start(mddev, rdev);
5911 if (err) {
5912 md_cluster_ops->add_new_disk_finish(mddev);
5913 export_rdev(rdev);
5914 return err;
5915 }
5916 }
5917 }
5918
1da177e4
LT		 5919 rdev->raid_disk = -1;
5920 err = bind_rdev_to_array(rdev, mddev);
5921 if (err)
5922 export_rdev(rdev);
52664732 5923 else
a6da4ef8 5924 err = add_bound_rdev(rdev);
1aee41f6
GR		 5925 if (mddev_is_clustered(mddev) &&
5926 (info->state & (1 << MD_DISK_CLUSTER_ADD)))
5927 md_cluster_ops->add_new_disk_finish(mddev);
1da177e4
LT		 5928 return err;
5929 }
5930
5931 /* otherwise, add_new_disk is only allowed
5932 * for major_version==0 superblocks
5933 */
5934 if (mddev->major_version != 0) {
5935 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5936 mdname(mddev));
5937 return -EINVAL;
5938 }
5939
5940 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5941 int err;
d710e138 5942 rdev = md_import_device(dev, -1, 0);
1da177e4 5943 if (IS_ERR(rdev)) {
f72ffdd6 5944 printk(KERN_WARNING
1da177e4
LT		 5945 "md: error, md_import_device() returned %ld\n",
5946 PTR_ERR(rdev));
5947 return PTR_ERR(rdev);
5948 }
5949 rdev->desc_nr = info->number;
5950 if (info->raid_disk < mddev->raid_disks)
5951 rdev->raid_disk = info->raid_disk;
5952 else
5953 rdev->raid_disk = -1;
5954
1da177e4 5955 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5956 if (info->state & (1<<MD_DISK_SYNC))
5957 set_bit(In_sync, &rdev->flags);
1da177e4 5958
8ddf9efe
N		 5959 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5960 set_bit(WriteMostly, &rdev->flags);
5961
1da177e4
LT		 5962 if (!mddev->persistent) {
5963 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5964 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5965 } else
57b2caa3 5966 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5967 rdev->sectors = rdev->sb_start;
1da177e4 5968
2bf071bf
N		 5969 err = bind_rdev_to_array(rdev, mddev);
5970 if (err) {
5971 export_rdev(rdev);
5972 return err;
5973 }
1da177e4
LT		 5974 }
5975
5976 return 0;
5977}
5978
f72ffdd6 5979static int hot_remove_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 5980{
5981 char b[BDEVNAME_SIZE];
3cb03002 5982 struct md_rdev *rdev;
1da177e4 5983
1da177e4
LT		 5984 rdev = find_rdev(mddev, dev);
5985 if (!rdev)
5986 return -ENXIO;
5987
293467aa
GR		 5988 if (mddev_is_clustered(mddev))
5989 md_cluster_ops->metadata_update_start(mddev);
5990
3ea8929d
N		 5991 clear_bit(Blocked, &rdev->flags);
5992 remove_and_add_spares(mddev, rdev);
5993
1da177e4
LT		 5994 if (rdev->raid_disk >= 0)
5995 goto busy;
5996
88bcfef7
GR		 5997 if (mddev_is_clustered(mddev))
5998 md_cluster_ops->remove_disk(mddev, rdev);
5999
fb56dfef 6000 md_kick_rdev_from_array(rdev);
850b2b42 6001 md_update_sb(mddev, 1);
d7603b7e 6002 md_new_event(mddev);
1da177e4 6003
293467aa
GR		 6004 if (mddev_is_clustered(mddev))
6005 md_cluster_ops->metadata_update_finish(mddev);
6006
1da177e4
LT		 6007 return 0;
6008busy:
293467aa
GR		 6009 if (mddev_is_clustered(mddev))
6010 md_cluster_ops->metadata_update_cancel(mddev);
fdefa4d8 6011 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 6012 bdevname(rdev->bdev,b), mdname(mddev));
6013 return -EBUSY;
6014}
6015
f72ffdd6 6016static int hot_add_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 6017{
6018 char b[BDEVNAME_SIZE];
6019 int err;
3cb03002 6020 struct md_rdev *rdev;
1da177e4
LT		 6021
6022 if (!mddev->pers)
6023 return -ENODEV;
6024
6025 if (mddev->major_version != 0) {
6026 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6027 " version-0 superblocks.\n",
6028 mdname(mddev));
6029 return -EINVAL;
6030 }
6031 if (!mddev->pers->hot_add_disk) {
f72ffdd6 6032 printk(KERN_WARNING
1da177e4
LT		 6033 "%s: personality does not support diskops!\n",
6034 mdname(mddev));
6035 return -EINVAL;
6036 }
6037
d710e138 6038 rdev = md_import_device(dev, -1, 0);
1da177e4 6039 if (IS_ERR(rdev)) {
f72ffdd6 6040 printk(KERN_WARNING
1da177e4
LT		 6041 "md: error, md_import_device() returned %ld\n",
6042 PTR_ERR(rdev));
6043 return -EINVAL;
6044 }
6045
6046 if (mddev->persistent)
57b2caa3 6047 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 6048 else
77304d2a 6049 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 6050
8190e754 6051 rdev->sectors = rdev->sb_start;
1da177e4 6052
b2d444d7 6053 if (test_bit(Faulty, &rdev->flags)) {
f72ffdd6 6054 printk(KERN_WARNING
1da177e4
LT		 6055 "md: can not hot-add faulty %s disk to %s!\n",
6056 bdevname(rdev->bdev,b), mdname(mddev));
6057 err = -EINVAL;
6058 goto abort_export;
6059 }
293467aa
GR		 6060
6061 if (mddev_is_clustered(mddev))
6062 md_cluster_ops->metadata_update_start(mddev);
b2d444d7 6063 clear_bit(In_sync, &rdev->flags);
1da177e4 6064 rdev->desc_nr = -1;
5842730d 6065 rdev->saved_raid_disk = -1;
2bf071bf
N		 6066 err = bind_rdev_to_array(rdev, mddev);
6067 if (err)
293467aa 6068 goto abort_clustered;
1da177e4
LT		 6069
6070 /*
6071 * The rest should better be atomic, we can have disk failures
6072 * noticed in interrupt contexts ...
6073 */
6074
1da177e4
LT		 6075 rdev->raid_disk = -1;
6076
850b2b42 6077 md_update_sb(mddev, 1);
1da177e4 6078
293467aa
GR		 6079 if (mddev_is_clustered(mddev))
6080 md_cluster_ops->metadata_update_finish(mddev);
1da177e4
LT		 6081 /*
6082 * Kick recovery, maybe this spare has to be added to the
6083 * array immediately.
6084 */
6085 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6086 md_wakeup_thread(mddev->thread);
d7603b7e 6087 md_new_event(mddev);
1da177e4
LT		 6088 return 0;
6089
293467aa
GR		 6090abort_clustered:
6091 if (mddev_is_clustered(mddev))
6092 md_cluster_ops->metadata_update_cancel(mddev);
1da177e4
LT		 6093abort_export:
6094 export_rdev(rdev);
6095 return err;
6096}
6097
fd01b88c 6098static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c 6099{
035328c2 6100 int err = 0;
32a7627c 6101
36fa3063 6102 if (mddev->pers) {
d66b1b39 6103 if (!mddev->pers->quiesce || !mddev->thread)
36fa3063
N		 6104 return -EBUSY;
6105 if (mddev->recovery || mddev->sync_thread)
6106 return -EBUSY;
6107 /* we should be able to change the bitmap.. */
6108 }
32a7627c 6109
36fa3063 6110 if (fd >= 0) {
035328c2 6111 struct inode *inode;
1e594bb2
N		 6112 struct file *f;
6113
6114 if (mddev->bitmap || mddev->bitmap_info.file)
36fa3063 6115 return -EEXIST; /* cannot add when bitmap is present */
1e594bb2 6116 f = fget(fd);
32a7627c 6117
1e594bb2 6118 if (f == NULL) {
36fa3063
N		 6119 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6120 mdname(mddev));
6121 return -EBADF;
6122 }
6123
1e594bb2 6124 inode = f->f_mapping->host;
035328c2
N		 6125 if (!S_ISREG(inode->i_mode)) {
6126 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6127 mdname(mddev));
6128 err = -EBADF;
1e594bb2 6129 } else if (!(f->f_mode & FMODE_WRITE)) {
035328c2
N		 6130 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6131 mdname(mddev));
6132 err = -EBADF;
6133 } else if (atomic_read(&inode->i_writecount) != 1) {
36fa3063
N		 6134 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6135 mdname(mddev));
035328c2
N		 6136 err = -EBUSY;
6137 }
6138 if (err) {
1e594bb2 6139 fput(f);
36fa3063
N		 6140 return err;
6141 }
1e594bb2 6142 mddev->bitmap_info.file = f;
c3d9714e 6143 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 6144 } else if (mddev->bitmap == NULL)
6145 return -ENOENT; /* cannot remove what isn't there */
6146 err = 0;
6147 if (mddev->pers) {
6148 mddev->pers->quiesce(mddev, 1);
69e51b44 6149 if (fd >= 0) {
f9209a32
GR		 6150 struct bitmap *bitmap;
6151
6152 bitmap = bitmap_create(mddev, -1);
6153 if (!IS_ERR(bitmap)) {
6154 mddev->bitmap = bitmap;
69e51b44 6155 err = bitmap_load(mddev);
ba599aca
N		 6156 } else
6157 err = PTR_ERR(bitmap);
69e51b44 6158 }
d7375ab3 6159 if (fd < 0 || err) {
36fa3063 6160 bitmap_destroy(mddev);
d7375ab3
N		 6161 fd = -1; /* make sure to put the file */
6162 }
36fa3063 6163 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 6164 }
6165 if (fd < 0) {
4af1a041
N		 6166 struct file *f = mddev->bitmap_info.file;
6167 if (f) {
6168 spin_lock(&mddev->lock);
6169 mddev->bitmap_info.file = NULL;
6170 spin_unlock(&mddev->lock);
6171 fput(f);
6172 }
36fa3063
N		 6173 }
6174
32a7627c
N		 6175 return err;
6176}
6177
1da177e4
LT		 6178/*
6179 * set_array_info is used two different ways
6180 * The original usage is when creating a new array.
6181 * In this usage, raid_disks is > 0 and it together with
6182 * level, size, not_persistent,layout,chunksize determine the
6183 * shape of the array.
6184 * This will always create an array with a type-0.90.0 superblock.
6185 * The newer usage is when assembling an array.
6186 * In this case raid_disks will be 0, and the major_version field is
6187 * use to determine which style super-blocks are to be found on the devices.
6188 * The minor and patch _version numbers are also kept incase the
6189 * super_block handler wishes to interpret them.
6190 */
f72ffdd6 6191static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6192{
6193
6194 if (info->raid_disks == 0) {
6195 /* just setting version number for superblock loading */
6196 if (info->major_version < 0 ||
50511da3 6197 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 6198 super_types[info->major_version].name == NULL) {
6199 /* maybe try to auto-load a module? */
f72ffdd6 6200 printk(KERN_INFO
1da177e4
LT		 6201 "md: superblock version %d not known\n",
6202 info->major_version);
6203 return -EINVAL;
6204 }
6205 mddev->major_version = info->major_version;
6206 mddev->minor_version = info->minor_version;
6207 mddev->patch_version = info->patch_version;
3f9d7b0d 6208 mddev->persistent = !info->not_persistent;
cbd19983
N		 6209 /* ensure mddev_put doesn't delete this now that there
6210 * is some minimal configuration.
6211 */
6212 mddev->ctime = get_seconds();
1da177e4
LT		 6213 return 0;
6214 }
6215 mddev->major_version = MD_MAJOR_VERSION;
6216 mddev->minor_version = MD_MINOR_VERSION;
6217 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6218 mddev->ctime = get_seconds();
6219
6220 mddev->level = info->level;
17115e03 6221 mddev->clevel[0] = 0;
58c0fed4 6222 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 6223 mddev->raid_disks = info->raid_disks;
6224 /* don't set md_minor, it is determined by which /dev/md* was
6225 * openned
6226 */
6227 if (info->state & (1<<MD_SB_CLEAN))
6228 mddev->recovery_cp = MaxSector;
6229 else
6230 mddev->recovery_cp = 0;
6231 mddev->persistent = ! info->not_persistent;
e691063a 6232 mddev->external = 0;
1da177e4
LT		 6233
6234 mddev->layout = info->layout;
9d8f0363 6235 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 6236
6237 mddev->max_disks = MD_SB_DISKS;
6238
e691063a
N		 6239 if (mddev->persistent)
6240 mddev->flags = 0;
850b2b42 6241 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6242
c3d9714e 6243 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6244 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6245 mddev->bitmap_info.offset = 0;
b2a2703c 6246
f6705578
N		 6247 mddev->reshape_position = MaxSector;
6248
1da177e4
LT		 6249 /*
6250 * Generate a 128 bit UUID
6251 */
6252 get_random_bytes(mddev->uuid, 16);
6253
f6705578 6254 mddev->new_level = mddev->level;
664e7c41 6255 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 6256 mddev->new_layout = mddev->layout;
6257 mddev->delta_disks = 0;
2c810cdd 6258 mddev->reshape_backwards = 0;
f6705578 6259
1da177e4
LT		 6260 return 0;
6261}
6262
fd01b88c 6263void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6264{
b522adcd
DW		 6265 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6266
6267 if (mddev->external_size)
6268 return;
6269
1f403624
DW		 6270 mddev->array_sectors = array_sectors;
6271}
6272EXPORT_SYMBOL(md_set_array_sectors);
6273
fd01b88c 6274static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6275{
3cb03002 6276 struct md_rdev *rdev;
a35b0d69 6277 int rv;
d71f9f88 6278 int fit = (num_sectors == 0);
a35b0d69
N		 6279
6280 if (mddev->pers->resize == NULL)
6281 return -EINVAL;
d71f9f88
AN		 6282 /* The "num_sectors" is the number of sectors of each device that
6283 * is used. This can only make sense for arrays with redundancy.
6284 * linear and raid0 always use whatever space is available. We can only
6285 * consider changing this number if no resync or reconstruction is
6286 * happening, and if the new size is acceptable. It must fit before the
0f420358 6287 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 6288 * of each device. If num_sectors is zero, we find the largest size
6289 * that fits.
a35b0d69 6290 */
f851b60d
N		 6291 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6292 mddev->sync_thread)
a35b0d69 6293 return -EBUSY;
bd8839e0
N		 6294 if (mddev->ro)
6295 return -EROFS;
a4a6125a 6296
dafb20fa 6297 rdev_for_each(rdev, mddev) {
dd8ac336 6298 sector_t avail = rdev->sectors;
01ab5662 6299
d71f9f88
AN		 6300 if (fit && (num_sectors == 0 || num_sectors > avail))
6301 num_sectors = avail;
6302 if (avail < num_sectors)
a35b0d69
N		 6303 return -ENOSPC;
6304 }
d71f9f88 6305 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 6306 if (!rv)
6307 revalidate_disk(mddev->gendisk);
a35b0d69
N		 6308 return rv;
6309}
6310
fd01b88c 6311static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 6312{
6313 int rv;
c6563a8c 6314 struct md_rdev *rdev;
da943b99 6315 /* change the number of raid disks */
63c70c4f 6316 if (mddev->pers->check_reshape == NULL)
da943b99 6317 return -EINVAL;
bd8839e0
N		 6318 if (mddev->ro)
6319 return -EROFS;
da943b99 6320 if (raid_disks <= 0 ||
233fca36 6321 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 6322 return -EINVAL;
f851b60d
N		 6323 if (mddev->sync_thread ||
6324 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6325 mddev->reshape_position != MaxSector)
da943b99 6326 return -EBUSY;
c6563a8c
N		 6327
6328 rdev_for_each(rdev, mddev) {
6329 if (mddev->raid_disks < raid_disks &&
6330 rdev->data_offset < rdev->new_data_offset)
6331 return -EINVAL;
6332 if (mddev->raid_disks > raid_disks &&
6333 rdev->data_offset > rdev->new_data_offset)
6334 return -EINVAL;
6335 }
6336
63c70c4f 6337 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N		 6338 if (mddev->delta_disks < 0)
6339 mddev->reshape_backwards = 1;
6340 else if (mddev->delta_disks > 0)
6341 mddev->reshape_backwards = 0;
63c70c4f
N		 6342
6343 rv = mddev->pers->check_reshape(mddev);
2c810cdd 6344 if (rv < 0) {
de171cb9 6345 mddev->delta_disks = 0;
2c810cdd
N		 6346 mddev->reshape_backwards = 0;
6347 }
da943b99
N		 6348 return rv;
6349}
6350
1da177e4
LT		 6351/*
6352 * update_array_info is used to change the configuration of an
6353 * on-line array.
6354 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6355 * fields in the info are checked against the array.
6356 * Any differences that cannot be handled will cause an error.
6357 * Normally, only one change can be managed at a time.
6358 */
fd01b88c 6359static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6360{
6361 int rv = 0;
6362 int cnt = 0;
36fa3063
N		 6363 int state = 0;
6364
6365 /* calculate expected state,ignoring low bits */
c3d9714e 6366 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 6367 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 6368
6369 if (mddev->major_version != info->major_version ||
6370 mddev->minor_version != info->minor_version ||
6371/* mddev->patch_version != info->patch_version || */
6372 mddev->ctime != info->ctime ||
6373 mddev->level != info->level ||
6374/* mddev->layout != info->layout || */
6375 !mddev->persistent != info->not_persistent||
9d8f0363 6376 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 6377 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6378 ((state^info->state) & 0xfffffe00)
6379 )
1da177e4
LT		 6380 return -EINVAL;
6381 /* Check there is only one change */
58c0fed4
AN		 6382 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6383 cnt++;
6384 if (mddev->raid_disks != info->raid_disks)
6385 cnt++;
6386 if (mddev->layout != info->layout)
6387 cnt++;
6388 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6389 cnt++;
6390 if (cnt == 0)
6391 return 0;
6392 if (cnt > 1)
6393 return -EINVAL;
1da177e4
LT		 6394
6395 if (mddev->layout != info->layout) {
6396 /* Change layout
6397 * we don't need to do anything at the md level, the
6398 * personality will take care of it all.
6399 */
50ac168a 6400 if (mddev->pers->check_reshape == NULL)
1da177e4 6401 return -EINVAL;
597a711b
N		 6402 else {
6403 mddev->new_layout = info->layout;
50ac168a 6404 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 6405 if (rv)
6406 mddev->new_layout = mddev->layout;
6407 return rv;
6408 }
1da177e4 6409 }
293467aa
GR		 6410 if (mddev_is_clustered(mddev))
6411 md_cluster_ops->metadata_update_start(mddev);
58c0fed4 6412 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6413 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6414
da943b99
N		 6415 if (mddev->raid_disks != info->raid_disks)
6416 rv = update_raid_disks(mddev, info->raid_disks);
6417
36fa3063 6418 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
293467aa
GR		 6419 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6420 rv = -EINVAL;
6421 goto err;
6422 }
6423 if (mddev->recovery || mddev->sync_thread) {
6424 rv = -EBUSY;
6425 goto err;
6426 }
36fa3063 6427 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
f9209a32 6428 struct bitmap *bitmap;
36fa3063 6429 /* add the bitmap */
293467aa
GR		 6430 if (mddev->bitmap) {
6431 rv = -EEXIST;
6432 goto err;
6433 }
6434 if (mddev->bitmap_info.default_offset == 0) {
6435 rv = -EINVAL;
6436 goto err;
6437 }
c3d9714e
N		 6438 mddev->bitmap_info.offset =
6439 mddev->bitmap_info.default_offset;
6409bb05
N		 6440 mddev->bitmap_info.space =
6441 mddev->bitmap_info.default_space;
36fa3063 6442 mddev->pers->quiesce(mddev, 1);
f9209a32
GR		 6443 bitmap = bitmap_create(mddev, -1);
6444 if (!IS_ERR(bitmap)) {
6445 mddev->bitmap = bitmap;
69e51b44 6446 rv = bitmap_load(mddev);
ba599aca
N		 6447 } else
6448 rv = PTR_ERR(bitmap);
36fa3063
N		 6449 if (rv)
6450 bitmap_destroy(mddev);
6451 mddev->pers->quiesce(mddev, 0);
6452 } else {
6453 /* remove the bitmap */
293467aa
GR		 6454 if (!mddev->bitmap) {
6455 rv = -ENOENT;
6456 goto err;
6457 }
6458 if (mddev->bitmap->storage.file) {
6459 rv = -EINVAL;
6460 goto err;
6461 }
36fa3063
N		 6462 mddev->pers->quiesce(mddev, 1);
6463 bitmap_destroy(mddev);
6464 mddev->pers->quiesce(mddev, 0);
c3d9714e 6465 mddev->bitmap_info.offset = 0;
36fa3063
N		 6466 }
6467 }
850b2b42 6468 md_update_sb(mddev, 1);
293467aa
GR		 6469 if (mddev_is_clustered(mddev))
6470 md_cluster_ops->metadata_update_finish(mddev);
6471 return rv;
6472err:
6473 if (mddev_is_clustered(mddev))
6474 md_cluster_ops->metadata_update_cancel(mddev);
1da177e4
LT		 6475 return rv;
6476}
6477
fd01b88c 6478static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6479{
3cb03002 6480 struct md_rdev *rdev;
1ca69c4b 6481 int err = 0;
1da177e4
LT		 6482
6483 if (mddev->pers == NULL)
6484 return -ENODEV;
6485
1ca69c4b
N		 6486 rcu_read_lock();
6487 rdev = find_rdev_rcu(mddev, dev);
1da177e4 6488 if (!rdev)
1ca69c4b
N		 6489 err = -ENODEV;
6490 else {
6491 md_error(mddev, rdev);
6492 if (!test_bit(Faulty, &rdev->flags))
6493 err = -EBUSY;
6494 }
6495 rcu_read_unlock();
6496 return err;
1da177e4
LT		 6497}
6498
2f9618ce
AN		 6499/*
6500 * We have a problem here : there is no easy way to give a CHS
6501 * virtual geometry. We currently pretend that we have a 2 heads
6502 * 4 sectors (with a BIG number of cylinders...). This drives
6503 * dosfs just mad... ;-)
6504 */
a885c8c4
CH		 6505static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6506{
fd01b88c 6507 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6508
6509 geo->heads = 2;
6510 geo->sectors = 4;
49ce6cea 6511 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6512 return 0;
6513}
6514
cb335f88
NS		 6515static inline bool md_ioctl_valid(unsigned int cmd)
6516{
6517 switch (cmd) {
6518 case ADD_NEW_DISK:
6519 case BLKROSET:
6520 case GET_ARRAY_INFO:
6521 case GET_BITMAP_FILE:
6522 case GET_DISK_INFO:
6523 case HOT_ADD_DISK:
6524 case HOT_REMOVE_DISK:
cb335f88
NS		 6525 case RAID_AUTORUN:
6526 case RAID_VERSION:
6527 case RESTART_ARRAY_RW:
6528 case RUN_ARRAY:
6529 case SET_ARRAY_INFO:
6530 case SET_BITMAP_FILE:
6531 case SET_DISK_FAULTY:
6532 case STOP_ARRAY:
6533 case STOP_ARRAY_RO:
1aee41f6 6534 case CLUSTERED_DISK_NACK:
cb335f88
NS		 6535 return true;
6536 default:
6537 return false;
6538 }
6539}
6540
a39907fa 6541static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6542 unsigned int cmd, unsigned long arg)
6543{
6544 int err = 0;
6545 void __user *argp = (void __user *)arg;
fd01b88c 6546 struct mddev *mddev = NULL;
e2218350 6547 int ro;
1da177e4 6548
cb335f88
NS		 6549 if (!md_ioctl_valid(cmd))
6550 return -ENOTTY;
6551
506c9e44
N		 6552 switch (cmd) {
6553 case RAID_VERSION:
6554 case GET_ARRAY_INFO:
6555 case GET_DISK_INFO:
6556 break;
6557 default:
6558 if (!capable(CAP_SYS_ADMIN))
6559 return -EACCES;
6560 }
1da177e4
LT		 6561
6562 /*
6563 * Commands dealing with the RAID driver but not any
6564 * particular array:
6565 */
c02c0aeb
N		 6566 switch (cmd) {
6567 case RAID_VERSION:
6568 err = get_version(argp);
3adc28d8 6569 goto out;
1da177e4 6570
1da177e4 6571#ifndef MODULE
c02c0aeb
N		 6572 case RAID_AUTORUN:
6573 err = 0;
6574 autostart_arrays(arg);
3adc28d8 6575 goto out;
1da177e4 6576#endif
c02c0aeb 6577 default:;
1da177e4
LT		 6578 }
6579
6580 /*
6581 * Commands creating/starting a new array:
6582 */
6583
a39907fa 6584 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6585
6586 if (!mddev) {
6587 BUG();
3adc28d8 6588 goto out;
1da177e4
LT		 6589 }
6590
1ca69c4b
N		 6591 /* Some actions do not requires the mutex */
6592 switch (cmd) {
6593 case GET_ARRAY_INFO:
6594 if (!mddev->raid_disks && !mddev->external)
6595 err = -ENODEV;
6596 else
6597 err = get_array_info(mddev, argp);
3adc28d8 6598 goto out;
1ca69c4b
N		 6599
6600 case GET_DISK_INFO:
6601 if (!mddev->raid_disks && !mddev->external)
6602 err = -ENODEV;
6603 else
6604 err = get_disk_info(mddev, argp);
3adc28d8 6605 goto out;
1ca69c4b
N		 6606
6607 case SET_DISK_FAULTY:
6608 err = set_disk_faulty(mddev, new_decode_dev(arg));
3adc28d8 6609 goto out;
4af1a041
N		 6610
6611 case GET_BITMAP_FILE:
6612 err = get_bitmap_file(mddev, argp);
6613 goto out;
6614
1ca69c4b
N		 6615 }
6616
a7a3f08d
N		 6617 if (cmd == ADD_NEW_DISK)
6618 /* need to ensure md_delayed_delete() has completed */
6619 flush_workqueue(md_misc_wq);
6620
90f5f7ad
HR		 6621 if (cmd == HOT_REMOVE_DISK)
6622 /* need to ensure recovery thread has run */
6623 wait_event_interruptible_timeout(mddev->sb_wait,
6624 !test_bit(MD_RECOVERY_NEEDED,
6625 &mddev->flags),
6626 msecs_to_jiffies(5000));
260fa034
N		 6627 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6628 /* Need to flush page cache, and ensure no-one else opens
6629 * and writes
6630 */
6631 mutex_lock(&mddev->open_mutex);
9ba3b7f5 6632 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
260fa034
N		 6633 mutex_unlock(&mddev->open_mutex);
6634 err = -EBUSY;
3adc28d8 6635 goto out;
260fa034
N		 6636 }
6637 set_bit(MD_STILL_CLOSED, &mddev->flags);
6638 mutex_unlock(&mddev->open_mutex);
6639 sync_blockdev(bdev);
6640 }
1da177e4
LT		 6641 err = mddev_lock(mddev);
6642 if (err) {
f72ffdd6 6643 printk(KERN_INFO
1da177e4
LT		 6644 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6645 err, cmd);
3adc28d8 6646 goto out;
1da177e4
LT		 6647 }
6648
c02c0aeb
N		 6649 if (cmd == SET_ARRAY_INFO) {
6650 mdu_array_info_t info;
6651 if (!arg)
6652 memset(&info, 0, sizeof(info));
6653 else if (copy_from_user(&info, argp, sizeof(info))) {
6654 err = -EFAULT;
3adc28d8 6655 goto unlock;
c02c0aeb
N		 6656 }
6657 if (mddev->pers) {
6658 err = update_array_info(mddev, &info);
6659 if (err) {
6660 printk(KERN_WARNING "md: couldn't update"
6661 " array info. %d\n", err);
3adc28d8 6662 goto unlock;
1da177e4 6663 }
3adc28d8 6664 goto unlock;
c02c0aeb
N		 6665 }
6666 if (!list_empty(&mddev->disks)) {
6667 printk(KERN_WARNING
6668 "md: array %s already has disks!\n",
6669 mdname(mddev));
6670 err = -EBUSY;
3adc28d8 6671 goto unlock;
c02c0aeb
N		 6672 }
6673 if (mddev->raid_disks) {
6674 printk(KERN_WARNING
6675 "md: array %s already initialised!\n",
6676 mdname(mddev));
6677 err = -EBUSY;
3adc28d8 6678 goto unlock;
c02c0aeb
N		 6679 }
6680 err = set_array_info(mddev, &info);
6681 if (err) {
6682 printk(KERN_WARNING "md: couldn't set"
6683 " array info. %d\n", err);
3adc28d8 6684 goto unlock;
c02c0aeb 6685 }
3adc28d8 6686 goto unlock;
1da177e4
LT		 6687 }
6688
6689 /*
6690 * Commands querying/configuring an existing array:
6691 */
32a7627c 6692 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6693 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6694 if ((!mddev->raid_disks && !mddev->external)
6695 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6696 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6697 && cmd != GET_BITMAP_FILE) {
1da177e4 6698 err = -ENODEV;
3adc28d8 6699 goto unlock;
1da177e4
LT		 6700 }
6701
6702 /*
6703 * Commands even a read-only array can execute:
6704 */
c02c0aeb 6705 switch (cmd) {
c02c0aeb
N		 6706 case RESTART_ARRAY_RW:
6707 err = restart_array(mddev);
3adc28d8 6708 goto unlock;
1da177e4 6709
c02c0aeb
N		 6710 case STOP_ARRAY:
6711 err = do_md_stop(mddev, 0, bdev);
3adc28d8 6712 goto unlock;
1da177e4 6713
c02c0aeb
N		 6714 case STOP_ARRAY_RO:
6715 err = md_set_readonly(mddev, bdev);
3adc28d8 6716 goto unlock;
1da177e4 6717
3ea8929d
N		 6718 case HOT_REMOVE_DISK:
6719 err = hot_remove_disk(mddev, new_decode_dev(arg));
3adc28d8 6720 goto unlock;
3ea8929d 6721
7ceb17e8
N		 6722 case ADD_NEW_DISK:
6723 /* We can support ADD_NEW_DISK on read-only arrays
6724 * on if we are re-adding a preexisting device.
6725 * So require mddev->pers and MD_DISK_SYNC.
6726 */
6727 if (mddev->pers) {
6728 mdu_disk_info_t info;
6729 if (copy_from_user(&info, argp, sizeof(info)))
6730 err = -EFAULT;
6731 else if (!(info.state & (1<<MD_DISK_SYNC)))
6732 /* Need to clear read-only for this */
6733 break;
6734 else
6735 err = add_new_disk(mddev, &info);
3adc28d8 6736 goto unlock;
7ceb17e8
N		 6737 }
6738 break;
6739
c02c0aeb
N		 6740 case BLKROSET:
6741 if (get_user(ro, (int __user *)(arg))) {
6742 err = -EFAULT;
3adc28d8 6743 goto unlock;
c02c0aeb
N		 6744 }
6745 err = -EINVAL;
e2218350 6746
c02c0aeb
N		 6747 /* if the bdev is going readonly the value of mddev->ro
6748 * does not matter, no writes are coming
6749 */
6750 if (ro)
3adc28d8 6751 goto unlock;
e2218350 6752
c02c0aeb
N		 6753 /* are we are already prepared for writes? */
6754 if (mddev->ro != 1)
3adc28d8 6755 goto unlock;
e2218350 6756
c02c0aeb
N		 6757 /* transitioning to readauto need only happen for
6758 * arrays that call md_write_start
6759 */
6760 if (mddev->pers) {
6761 err = restart_array(mddev);
6762 if (err == 0) {
6763 mddev->ro = 2;
6764 set_disk_ro(mddev->gendisk, 0);
e2218350 6765 }
c02c0aeb 6766 }
3adc28d8 6767 goto unlock;
1da177e4
LT		 6768 }
6769
6770 /*
6771 * The remaining ioctls are changing the state of the
f91de92e 6772 * superblock, so we do not allow them on read-only arrays.
1da177e4 6773 */
326eb17d 6774 if (mddev->ro && mddev->pers) {
f91de92e
N		 6775 if (mddev->ro == 2) {
6776 mddev->ro = 0;
00bcb4ac 6777 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 6778 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N		 6779 /* mddev_unlock will wake thread */
6780 /* If a device failed while we were read-only, we
6781 * need to make sure the metadata is updated now.
6782 */
6783 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6784 mddev_unlock(mddev);
6785 wait_event(mddev->sb_wait,
6786 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6787 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
29f097c4 6788 mddev_lock_nointr(mddev);
f3378b48 6789 }
f91de92e
N		 6790 } else {
6791 err = -EROFS;
3adc28d8 6792 goto unlock;
f91de92e 6793 }
1da177e4
LT		 6794 }
6795
c02c0aeb
N		 6796 switch (cmd) {
6797 case ADD_NEW_DISK:
1da177e4 6798 {
c02c0aeb
N		 6799 mdu_disk_info_t info;
6800 if (copy_from_user(&info, argp, sizeof(info)))
6801 err = -EFAULT;
6802 else
6803 err = add_new_disk(mddev, &info);
3adc28d8 6804 goto unlock;
c02c0aeb 6805 }
1da177e4 6806
1aee41f6
GR		 6807 case CLUSTERED_DISK_NACK:
6808 if (mddev_is_clustered(mddev))
6809 md_cluster_ops->new_disk_ack(mddev, false);
6810 else
6811 err = -EINVAL;
6812 goto unlock;
6813
c02c0aeb
N		 6814 case HOT_ADD_DISK:
6815 err = hot_add_disk(mddev, new_decode_dev(arg));
3adc28d8 6816 goto unlock;
1da177e4 6817
c02c0aeb
N		 6818 case RUN_ARRAY:
6819 err = do_md_run(mddev);
3adc28d8 6820 goto unlock;
1da177e4 6821
c02c0aeb
N		 6822 case SET_BITMAP_FILE:
6823 err = set_bitmap_file(mddev, (int)arg);
3adc28d8 6824 goto unlock;
32a7627c 6825
c02c0aeb
N		 6826 default:
6827 err = -EINVAL;
3adc28d8 6828 goto unlock;
1da177e4
LT		 6829 }
6830
3adc28d8 6831unlock:
d3374825
N		 6832 if (mddev->hold_active == UNTIL_IOCTL &&
6833 err != -EINVAL)
6834 mddev->hold_active = 0;
1da177e4 6835 mddev_unlock(mddev);
3adc28d8 6836out:
1da177e4
LT		 6837 return err;
6838}
aa98aa31
AB		 6839#ifdef CONFIG_COMPAT
6840static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6841 unsigned int cmd, unsigned long arg)
6842{
6843 switch (cmd) {
6844 case HOT_REMOVE_DISK:
6845 case HOT_ADD_DISK:
6846 case SET_DISK_FAULTY:
6847 case SET_BITMAP_FILE:
6848 /* These take in integer arg, do not convert */
6849 break;
6850 default:
6851 arg = (unsigned long)compat_ptr(arg);
6852 break;
6853 }
6854
6855 return md_ioctl(bdev, mode, cmd, arg);
6856}
6857#endif /* CONFIG_COMPAT */
1da177e4 6858
a39907fa 6859static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6860{
6861 /*
6862 * Succeed if we can lock the mddev, which confirms that
6863 * it isn't being stopped right now.
6864 */
fd01b88c 6865 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6866 int err;
6867
0c098220
YL		 6868 if (!mddev)
6869 return -ENODEV;
6870
d3374825
N		 6871 if (mddev->gendisk != bdev->bd_disk) {
6872 /* we are racing with mddev_put which is discarding this
6873 * bd_disk.
6874 */
6875 mddev_put(mddev);
6876 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6877 flush_workqueue(md_misc_wq);
d3374825
N		 6878 /* Then retry the open from the top */
6879 return -ERESTARTSYS;
6880 }
6881 BUG_ON(mddev != bdev->bd_disk->private_data);
6882
c8c00a69 6883 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6884 goto out;
6885
6886 err = 0;
f2ea68cf 6887 atomic_inc(&mddev->openers);
260fa034 6888 clear_bit(MD_STILL_CLOSED, &mddev->flags);
c8c00a69 6889 mutex_unlock(&mddev->open_mutex);
1da177e4 6890
f0b4f7e2 6891 check_disk_change(bdev);
1da177e4
LT		 6892 out:
6893 return err;
6894}
6895
db2a144b 6896static void md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6897{
f72ffdd6 6898 struct mddev *mddev = disk->private_data;
1da177e4 6899
52e5f9d1 6900 BUG_ON(!mddev);
f2ea68cf 6901 atomic_dec(&mddev->openers);
1da177e4 6902 mddev_put(mddev);
1da177e4 6903}
f0b4f7e2
N		 6904
6905static int md_media_changed(struct gendisk *disk)
6906{
fd01b88c 6907 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6908
6909 return mddev->changed;
6910}
6911
6912static int md_revalidate(struct gendisk *disk)
6913{
fd01b88c 6914 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6915
6916 mddev->changed = 0;
6917 return 0;
6918}
83d5cde4 6919static const struct block_device_operations md_fops =
1da177e4
LT		 6920{
6921 .owner = THIS_MODULE,
a39907fa
AV		 6922 .open = md_open,
6923 .release = md_release,
b492b852 6924 .ioctl = md_ioctl,
aa98aa31
AB		 6925#ifdef CONFIG_COMPAT
6926 .compat_ioctl = md_compat_ioctl,
6927#endif
a885c8c4 6928 .getgeo = md_getgeo,
f0b4f7e2
N		 6929 .media_changed = md_media_changed,
6930 .revalidate_disk= md_revalidate,
1da177e4
LT		 6931};
6932
f72ffdd6 6933static int md_thread(void *arg)
1da177e4 6934{
2b8bf345 6935 struct md_thread *thread = arg;
1da177e4 6936
1da177e4
LT		 6937 /*
6938 * md_thread is a 'system-thread', it's priority should be very
6939 * high. We avoid resource deadlocks individually in each
6940 * raid personality. (RAID5 does preallocation) We also use RR and
6941 * the very same RT priority as kswapd, thus we will never get
6942 * into a priority inversion deadlock.
6943 *
6944 * we definitely have to have equal or higher priority than
6945 * bdflush, otherwise bdflush will deadlock if there are too
6946 * many dirty RAID5 blocks.
6947 */
1da177e4 6948
6985c43f 6949 allow_signal(SIGKILL);
a6fb0934 6950 while (!kthread_should_stop()) {
1da177e4 6951
93588e22
N		 6952 /* We need to wait INTERRUPTIBLE so that
6953 * we don't add to the load-average.
6954 * That means we need to be sure no signals are
6955 * pending
6956 */
6957 if (signal_pending(current))
6958 flush_signals(current);
6959
6960 wait_event_interruptible_timeout
6961 (thread->wqueue,
6962 test_bit(THREAD_WAKEUP, &thread->flags)
6963 || kthread_should_stop(),
6964 thread->timeout);
1da177e4 6965
6c987910
N		 6966 clear_bit(THREAD_WAKEUP, &thread->flags);
6967 if (!kthread_should_stop())
4ed8731d 6968 thread->run(thread);
1da177e4 6969 }
a6fb0934 6970
1da177e4
LT		 6971 return 0;
6972}
6973
2b8bf345 6974void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6975{
6976 if (thread) {
36a4e1fe 6977 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6978 set_bit(THREAD_WAKEUP, &thread->flags);
6979 wake_up(&thread->wqueue);
6980 }
6981}
6c144d31 6982EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 6983
4ed8731d
SL		 6984struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6985 struct mddev *mddev, const char *name)
1da177e4 6986{
2b8bf345 6987 struct md_thread *thread;
1da177e4 6988
2b8bf345 6989 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6990 if (!thread)
6991 return NULL;
6992
1da177e4
LT		 6993 init_waitqueue_head(&thread->wqueue);
6994
1da177e4
LT		 6995 thread->run = run;
6996 thread->mddev = mddev;
32a7627c 6997 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6998 thread->tsk = kthread_run(md_thread, thread,
6999 "%s_%s",
7000 mdname(thread->mddev),
0232605d 7001 name);
a6fb0934 7002 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 7003 kfree(thread);
7004 return NULL;
7005 }
1da177e4
LT		 7006 return thread;
7007}
6c144d31 7008EXPORT_SYMBOL(md_register_thread);
1da177e4 7009
2b8bf345 7010void md_unregister_thread(struct md_thread **threadp)
1da177e4 7011{
2b8bf345 7012 struct md_thread *thread = *threadp;
e0cf8f04
N		 7013 if (!thread)
7014 return;
36a4e1fe 7015 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 7016 /* Locking ensures that mddev_unlock does not wake_up a
7017 * non-existent thread
7018 */
7019 spin_lock(&pers_lock);
7020 *threadp = NULL;
7021 spin_unlock(&pers_lock);
a6fb0934
N		 7022
7023 kthread_stop(thread->tsk);
1da177e4
LT		 7024 kfree(thread);
7025}
6c144d31 7026EXPORT_SYMBOL(md_unregister_thread);
1da177e4 7027
fd01b88c 7028void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7029{
b2d444d7 7030 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 7031 return;
6bfe0b49 7032
de393cde 7033 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 7034 return;
7035 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 7036 if (mddev->degraded)
7037 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 7038 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 7039 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7040 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7041 md_wakeup_thread(mddev->thread);
768a418d 7042 if (mddev->event_work.func)
e804ac78 7043 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 7044 md_new_event_inintr(mddev);
1da177e4 7045}
6c144d31 7046EXPORT_SYMBOL(md_error);
1da177e4
LT		 7047
7048/* seq_file implementation /proc/mdstat */
7049
7050static void status_unused(struct seq_file *seq)
7051{
7052 int i = 0;
3cb03002 7053 struct md_rdev *rdev;
1da177e4
LT		 7054
7055 seq_printf(seq, "unused devices: ");
7056
159ec1fc 7057 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 7058 char b[BDEVNAME_SIZE];
7059 i++;
7060 seq_printf(seq, "%s ",
7061 bdevname(rdev->bdev,b));
7062 }
7063 if (!i)
7064 seq_printf(seq, "<none>");
7065
7066 seq_printf(seq, "\n");
7067}
7068
f72ffdd6 7069static void status_resync(struct seq_file *seq, struct mddev *mddev)
1da177e4 7070{
dd71cf6b
N		 7071 sector_t max_sectors, resync, res;
7072 unsigned long dt, db;
7073 sector_t rt;
4588b42e
N		 7074 int scale;
7075 unsigned int per_milli;
1da177e4 7076
72f36d59
N		 7077 if (mddev->curr_resync <= 3)
7078 resync = 0;
7079 else
7080 resync = mddev->curr_resync
7081 - atomic_read(&mddev->recovery_active);
1da177e4 7082
c804cdec
N		 7083 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7084 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 7085 max_sectors = mddev->resync_max_sectors;
1da177e4 7086 else
dd71cf6b 7087 max_sectors = mddev->dev_sectors;
1da177e4 7088
403df478 7089 WARN_ON(max_sectors == 0);
4588b42e 7090 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 7091 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 7092 * u32, as those are the requirements for sector_div.
7093 * Thus 'scale' must be at least 10
7094 */
7095 scale = 10;
7096 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 7097 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 7098 scale++;
7099 }
7100 res = (resync>>scale)*1000;
dd71cf6b 7101 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 7102
7103 per_milli = res;
1da177e4 7104 {
4588b42e 7105 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 7106 seq_printf(seq, "[");
7107 for (i = 0; i < x; i++)
7108 seq_printf(seq, "=");
7109 seq_printf(seq, ">");
7110 for (i = 0; i < y; i++)
7111 seq_printf(seq, ".");
7112 seq_printf(seq, "] ");
7113 }
4588b42e 7114 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 7115 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7116 "reshape" :
61df9d91
N		 7117 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7118 "check" :
7119 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7120 "resync" : "recovery"))),
7121 per_milli/10, per_milli % 10,
dd71cf6b
N		 7122 (unsigned long long) resync/2,
7123 (unsigned long long) max_sectors/2);
1da177e4
LT		 7124
7125 /*
1da177e4
LT		 7126 * dt: time from mark until now
7127 * db: blocks written from mark until now
7128 * rt: remaining time
dd71cf6b
N		 7129 *
7130 * rt is a sector_t, so could be 32bit or 64bit.
7131 * So we divide before multiply in case it is 32bit and close
7132 * to the limit.
25985edc 7133 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 7134 * near the end of resync when the number of remaining sectors
7135 * is close to 'db'.
7136 * We then divide rt by 32 after multiplying by db to compensate.
7137 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 7138 */
7139 dt = ((jiffies - mddev->resync_mark) / HZ);
7140 if (!dt) dt++;
ff4e8d9a
N		 7141 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7142 - mddev->resync_mark_cnt;
1da177e4 7143
dd71cf6b
N		 7144 rt = max_sectors - resync; /* number of remaining sectors */
7145 sector_div(rt, db/32+1);
7146 rt *= dt;
7147 rt >>= 5;
7148
7149 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7150 ((unsigned long)rt % 60)/6);
1da177e4 7151
ff4e8d9a 7152 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 7153}
7154
7155static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7156{
7157 struct list_head *tmp;
7158 loff_t l = *pos;
fd01b88c 7159 struct mddev *mddev;
1da177e4
LT		 7160
7161 if (l >= 0x10000)
7162 return NULL;
7163 if (!l--)
7164 /* header */
7165 return (void*)1;
7166
7167 spin_lock(&all_mddevs_lock);
7168 list_for_each(tmp,&all_mddevs)
7169 if (!l--) {
fd01b88c 7170 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 7171 mddev_get(mddev);
7172 spin_unlock(&all_mddevs_lock);
7173 return mddev;
7174 }
7175 spin_unlock(&all_mddevs_lock);
7176 if (!l--)
7177 return (void*)2;/* tail */
7178 return NULL;
7179}
7180
7181static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7182{
7183 struct list_head *tmp;
fd01b88c 7184 struct mddev *next_mddev, *mddev = v;
f72ffdd6 7185
1da177e4
LT		 7186 ++*pos;
7187 if (v == (void*)2)
7188 return NULL;
7189
7190 spin_lock(&all_mddevs_lock);
7191 if (v == (void*)1)
7192 tmp = all_mddevs.next;
7193 else
7194 tmp = mddev->all_mddevs.next;
7195 if (tmp != &all_mddevs)
fd01b88c 7196 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 7197 else {
7198 next_mddev = (void*)2;
7199 *pos = 0x10000;
f72ffdd6 7200 }
1da177e4
LT		 7201 spin_unlock(&all_mddevs_lock);
7202
7203 if (v != (void*)1)
7204 mddev_put(mddev);
7205 return next_mddev;
7206
7207}
7208
7209static void md_seq_stop(struct seq_file *seq, void *v)
7210{
fd01b88c 7211 struct mddev *mddev = v;
1da177e4
LT		 7212
7213 if (mddev && v != (void*)1 && v != (void*)2)
7214 mddev_put(mddev);
7215}
7216
7217static int md_seq_show(struct seq_file *seq, void *v)
7218{
fd01b88c 7219 struct mddev *mddev = v;
dd8ac336 7220 sector_t sectors;
3cb03002 7221 struct md_rdev *rdev;
1da177e4
LT		 7222
7223 if (v == (void*)1) {
84fc4b56 7224 struct md_personality *pers;
1da177e4
LT		 7225 seq_printf(seq, "Personalities : ");
7226 spin_lock(&pers_lock);
2604b703
N		 7227 list_for_each_entry(pers, &pers_list, list)
7228 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 7229
7230 spin_unlock(&pers_lock);
7231 seq_printf(seq, "\n");
f1514638 7232 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7233 return 0;
7234 }
7235 if (v == (void*)2) {
7236 status_unused(seq);
7237 return 0;
7238 }
7239
36d091f4 7240 spin_lock(&mddev->lock);
1da177e4
LT		 7241 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7242 seq_printf(seq, "%s : %sactive", mdname(mddev),
7243 mddev->pers ? "" : "in");
7244 if (mddev->pers) {
f91de92e 7245 if (mddev->ro==1)
1da177e4 7246 seq_printf(seq, " (read-only)");
f91de92e 7247 if (mddev->ro==2)
52720ae7 7248 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 7249 seq_printf(seq, " %s", mddev->pers->name);
7250 }
7251
dd8ac336 7252 sectors = 0;
f97fcad3
N		 7253 rcu_read_lock();
7254 rdev_for_each_rcu(rdev, mddev) {
1da177e4
LT		 7255 char b[BDEVNAME_SIZE];
7256 seq_printf(seq, " %s[%d]",
7257 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 7258 if (test_bit(WriteMostly, &rdev->flags))
7259 seq_printf(seq, "(W)");
b2d444d7 7260 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7261 seq_printf(seq, "(F)");
7262 continue;
2d78f8c4
N		 7263 }
7264 if (rdev->raid_disk < 0)
b325a32e 7265 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 7266 if (test_bit(Replacement, &rdev->flags))
7267 seq_printf(seq, "(R)");
dd8ac336 7268 sectors += rdev->sectors;
1da177e4 7269 }
f97fcad3 7270 rcu_read_unlock();
1da177e4
LT		 7271
7272 if (!list_empty(&mddev->disks)) {
7273 if (mddev->pers)
7274 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 7275 (unsigned long long)
7276 mddev->array_sectors / 2);
1da177e4
LT		 7277 else
7278 seq_printf(seq, "\n %llu blocks",
dd8ac336 7279 (unsigned long long)sectors / 2);
1da177e4 7280 }
1cd6bf19
N		 7281 if (mddev->persistent) {
7282 if (mddev->major_version != 0 ||
7283 mddev->minor_version != 90) {
7284 seq_printf(seq," super %d.%d",
7285 mddev->major_version,
7286 mddev->minor_version);
7287 }
e691063a
N		 7288 } else if (mddev->external)
7289 seq_printf(seq, " super external:%s",
7290 mddev->metadata_type);
7291 else
1cd6bf19 7292 seq_printf(seq, " super non-persistent");
1da177e4
LT		 7293
7294 if (mddev->pers) {
d710e138 7295 mddev->pers->status(seq, mddev);
f72ffdd6 7296 seq_printf(seq, "\n ");
8e1b39d6
N		 7297 if (mddev->pers->sync_request) {
7298 if (mddev->curr_resync > 2) {
d710e138 7299 status_resync(seq, mddev);
8e1b39d6 7300 seq_printf(seq, "\n ");
72f36d59 7301 } else if (mddev->curr_resync >= 1)
8e1b39d6
N		 7302 seq_printf(seq, "\tresync=DELAYED\n ");
7303 else if (mddev->recovery_cp < MaxSector)
7304 seq_printf(seq, "\tresync=PENDING\n ");
7305 }
32a7627c
N		 7306 } else
7307 seq_printf(seq, "\n ");
7308
57148964 7309 bitmap_status(seq, mddev->bitmap);
1da177e4
LT		 7310
7311 seq_printf(seq, "\n");
7312 }
36d091f4 7313 spin_unlock(&mddev->lock);
f72ffdd6 7314
1da177e4
LT		 7315 return 0;
7316}
7317
110518bc 7318static const struct seq_operations md_seq_ops = {
1da177e4
LT		 7319 .start = md_seq_start,
7320 .next = md_seq_next,
7321 .stop = md_seq_stop,
7322 .show = md_seq_show,
7323};
7324
7325static int md_seq_open(struct inode *inode, struct file *file)
7326{
f1514638 7327 struct seq_file *seq;
1da177e4
LT		 7328 int error;
7329
7330 error = seq_open(file, &md_seq_ops);
d7603b7e 7331 if (error)
f1514638
KS		 7332 return error;
7333
7334 seq = file->private_data;
7335 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7336 return error;
7337}
7338
e2f23b60 7339static int md_unloading;
d7603b7e
N		 7340static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7341{
f1514638 7342 struct seq_file *seq = filp->private_data;
d7603b7e
N		 7343 int mask;
7344
e2f23b60 7345 if (md_unloading)
7d7e64f2 7346 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
d7603b7e
N		 7347 poll_wait(filp, &md_event_waiters, wait);
7348
7349 /* always allow read */
7350 mask = POLLIN | POLLRDNORM;
7351
f1514638 7352 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 7353 mask |= POLLERR | POLLPRI;
7354 return mask;
7355}
7356
fa027c2a 7357static const struct file_operations md_seq_fops = {
e24650c2 7358 .owner = THIS_MODULE,
1da177e4
LT		 7359 .open = md_seq_open,
7360 .read = seq_read,
7361 .llseek = seq_lseek,
c3f94b40 7362 .release = seq_release_private,
d7603b7e 7363 .poll = mdstat_poll,
1da177e4
LT		 7364};
7365
84fc4b56 7366int register_md_personality(struct md_personality *p)
1da177e4 7367{
50bd3774
CY		 7368 printk(KERN_INFO "md: %s personality registered for level %d\n",
7369 p->name, p->level);
1da177e4 7370 spin_lock(&pers_lock);
2604b703 7371 list_add_tail(&p->list, &pers_list);
1da177e4
LT		 7372 spin_unlock(&pers_lock);
7373 return 0;
7374}
6c144d31 7375EXPORT_SYMBOL(register_md_personality);
1da177e4 7376
84fc4b56 7377int unregister_md_personality(struct md_personality *p)
1da177e4 7378{
2604b703 7379 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7380 spin_lock(&pers_lock);
2604b703 7381 list_del_init(&p->list);
1da177e4
LT		 7382 spin_unlock(&pers_lock);
7383 return 0;
7384}
6c144d31 7385EXPORT_SYMBOL(unregister_md_personality);
1da177e4 7386
edb39c9d
GR		 7387int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module)
7388{
7389 if (md_cluster_ops != NULL)
7390 return -EALREADY;
7391 spin_lock(&pers_lock);
7392 md_cluster_ops = ops;
7393 md_cluster_mod = module;
7394 spin_unlock(&pers_lock);
7395 return 0;
7396}
7397EXPORT_SYMBOL(register_md_cluster_operations);
7398
7399int unregister_md_cluster_operations(void)
7400{
7401 spin_lock(&pers_lock);
7402 md_cluster_ops = NULL;
7403 spin_unlock(&pers_lock);
7404 return 0;
7405}
7406EXPORT_SYMBOL(unregister_md_cluster_operations);
7407
7408int md_setup_cluster(struct mddev *mddev, int nodes)
7409{
7410 int err;
7411
7412 err = request_module("md-cluster");
7413 if (err) {
7414 pr_err("md-cluster module not found.\n");
7415 return err;
7416 }
7417
7418 spin_lock(&pers_lock);
7419 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7420 spin_unlock(&pers_lock);
7421 return -ENOENT;
7422 }
7423 spin_unlock(&pers_lock);
7424
cf921cc1 7425 return md_cluster_ops->join(mddev, nodes);
edb39c9d
GR		 7426}
7427
7428void md_cluster_stop(struct mddev *mddev)
7429{
c4ce867f
GR		 7430 if (!md_cluster_ops)
7431 return;
edb39c9d
GR		 7432 md_cluster_ops->leave(mddev);
7433 module_put(md_cluster_mod);
7434}
7435
fd01b88c 7436static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7437{
f72ffdd6 7438 struct md_rdev *rdev;
1da177e4 7439 int idle;
eea1bf38 7440 int curr_events;
1da177e4
LT		 7441
7442 idle = 1;
4b80991c
N		 7443 rcu_read_lock();
7444 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7445 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 7446 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7447 (int)part_stat_read(&disk->part0, sectors[1]) -
7448 atomic_read(&disk->sync_io);
713f6ab1
N		 7449 /* sync IO will cause sync_io to increase before the disk_stats
7450 * as sync_io is counted when a request starts, and
7451 * disk_stats is counted when it completes.
7452 * So resync activity will cause curr_events to be smaller than
7453 * when there was no such activity.
7454 * non-sync IO will cause disk_stat to increase without
7455 * increasing sync_io so curr_events will (eventually)
7456 * be larger than it was before. Once it becomes
7457 * substantially larger, the test below will cause
7458 * the array to appear non-idle, and resync will slow
7459 * down.
7460 * If there is a lot of outstanding resync activity when
7461 * we set last_event to curr_events, then all that activity
7462 * completing might cause the array to appear non-idle
7463 * and resync will be slowed down even though there might
7464 * not have been non-resync activity. This will only
7465 * happen once though. 'last_events' will soon reflect
7466 * the state where there is little or no outstanding
7467 * resync requests, and further resync activity will
7468 * always make curr_events less than last_events.
c0e48521 7469 *
1da177e4 7470 */
eea1bf38 7471 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 7472 rdev->last_events = curr_events;
7473 idle = 0;
7474 }
7475 }
4b80991c 7476 rcu_read_unlock();
1da177e4
LT		 7477 return idle;
7478}
7479
fd01b88c 7480void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 7481{
7482 /* another "blocks" (512byte) blocks have been synced */
7483 atomic_sub(blocks, &mddev->recovery_active);
7484 wake_up(&mddev->recovery_wait);
7485 if (!ok) {
dfc70645 7486 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 7487 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT		 7488 md_wakeup_thread(mddev->thread);
7489 // stop recovery, signal do_sync ....
7490 }
7491}
6c144d31 7492EXPORT_SYMBOL(md_done_sync);
1da177e4 7493
06d91a5f
N		 7494/* md_write_start(mddev, bi)
7495 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 7496 * in superblock) before writing, schedule a superblock update
7497 * and wait for it to complete.
06d91a5f 7498 */
fd01b88c 7499void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7500{
0fd62b86 7501 int did_change = 0;
06d91a5f 7502 if (bio_data_dir(bi) != WRITE)
3d310eb7 7503 return;
06d91a5f 7504
f91de92e
N		 7505 BUG_ON(mddev->ro == 1);
7506 if (mddev->ro == 2) {
7507 /* need to switch to read/write */
7508 mddev->ro = 0;
7509 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7510 md_wakeup_thread(mddev->thread);
25156198 7511 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7512 did_change = 1;
f91de92e 7513 }
06d91a5f 7514 atomic_inc(&mddev->writes_pending);
31a59e34
N		 7515 if (mddev->safemode == 1)
7516 mddev->safemode = 0;
06d91a5f 7517 if (mddev->in_sync) {
85572d7c 7518 spin_lock(&mddev->lock);
3d310eb7
N		 7519 if (mddev->in_sync) {
7520 mddev->in_sync = 0;
850b2b42 7521 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7522 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7523 md_wakeup_thread(mddev->thread);
0fd62b86 7524 did_change = 1;
3d310eb7 7525 }
85572d7c 7526 spin_unlock(&mddev->lock);
06d91a5f 7527 }
0fd62b86 7528 if (did_change)
00bcb4ac 7529 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7530 wait_event(mddev->sb_wait,
09a44cc1 7531 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4 7532}
6c144d31 7533EXPORT_SYMBOL(md_write_start);
1da177e4 7534
fd01b88c 7535void md_write_end(struct mddev *mddev)
1da177e4
LT		 7536{
7537 if (atomic_dec_and_test(&mddev->writes_pending)) {
7538 if (mddev->safemode == 2)
7539 md_wakeup_thread(mddev->thread);
16f17b39 7540 else if (mddev->safemode_delay)
1da177e4
LT		 7541 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7542 }
7543}
6c144d31 7544EXPORT_SYMBOL(md_write_end);
1da177e4 7545
2a2275d6
N		 7546/* md_allow_write(mddev)
7547 * Calling this ensures that the array is marked 'active' so that writes
7548 * may proceed without blocking. It is important to call this before
7549 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7550 * Must be called with mddev_lock held.
b5470dc5
DW		 7551 *
7552 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7553 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7554 */
fd01b88c 7555int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7556{
7557 if (!mddev->pers)
b5470dc5 7558 return 0;
2a2275d6 7559 if (mddev->ro)
b5470dc5 7560 return 0;
1a0fd497 7561 if (!mddev->pers->sync_request)
b5470dc5 7562 return 0;
2a2275d6 7563
85572d7c 7564 spin_lock(&mddev->lock);
2a2275d6
N		 7565 if (mddev->in_sync) {
7566 mddev->in_sync = 0;
7567 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7568 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7569 if (mddev->safemode_delay &&
7570 mddev->safemode == 0)
7571 mddev->safemode = 1;
85572d7c 7572 spin_unlock(&mddev->lock);
293467aa
GR		 7573 if (mddev_is_clustered(mddev))
7574 md_cluster_ops->metadata_update_start(mddev);
2a2275d6 7575 md_update_sb(mddev, 0);
293467aa
GR		 7576 if (mddev_is_clustered(mddev))
7577 md_cluster_ops->metadata_update_finish(mddev);
00bcb4ac 7578 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6 7579 } else
85572d7c 7580 spin_unlock(&mddev->lock);
b5470dc5 7581
070dc6dd 7582 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7583 return -EAGAIN;
7584 else
7585 return 0;
2a2275d6
N		 7586}
7587EXPORT_SYMBOL_GPL(md_allow_write);
7588
1da177e4
LT		 7589#define SYNC_MARKS 10
7590#define SYNC_MARK_STEP (3*HZ)
54f89341 7591#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 7592void md_do_sync(struct md_thread *thread)
1da177e4 7593{
4ed8731d 7594 struct mddev *mddev = thread->mddev;
fd01b88c 7595 struct mddev *mddev2;
1da177e4
LT		 7596 unsigned int currspeed = 0,
7597 window;
ac7e50a3 7598 sector_t max_sectors,j, io_sectors, recovery_done;
1da177e4 7599 unsigned long mark[SYNC_MARKS];
54f89341 7600 unsigned long update_time;
1da177e4
LT		 7601 sector_t mark_cnt[SYNC_MARKS];
7602 int last_mark,m;
7603 struct list_head *tmp;
7604 sector_t last_check;
57afd89f 7605 int skipped = 0;
3cb03002 7606 struct md_rdev *rdev;
c4a39551 7607 char *desc, *action = NULL;
7c2c57c9 7608 struct blk_plug plug;
1da177e4
LT		 7609
7610 /* just incase thread restarts... */
7611 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7612 return;
3991b31e
N		 7613 if (mddev->ro) {/* never try to sync a read-only array */
7614 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5fd6c1dc 7615 return;
3991b31e 7616 }
1da177e4 7617
61df9d91 7618 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
c4a39551 7619 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
61df9d91 7620 desc = "data-check";
c4a39551
JB		 7621 action = "check";
7622 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
61df9d91 7623 desc = "requested-resync";
c4a39551
JB		 7624 action = "repair";
7625 } else
61df9d91
N		 7626 desc = "resync";
7627 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7628 desc = "reshape";
7629 else
7630 desc = "recovery";
7631
c4a39551
JB		 7632 mddev->last_sync_action = action ?: desc;
7633
1da177e4
LT		 7634 /* we overload curr_resync somewhat here.
7635 * 0 == not engaged in resync at all
7636 * 2 == checking that there is no conflict with another sync
7637 * 1 == like 2, but have yielded to allow conflicting resync to
7638 * commense
7639 * other == active in resync - this many blocks
7640 *
7641 * Before starting a resync we must have set curr_resync to
7642 * 2, and then checked that every "conflicting" array has curr_resync
7643 * less than ours. When we find one that is the same or higher
7644 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7645 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7646 * This will mean we have to start checking from the beginning again.
7647 *
7648 */
7649
7650 do {
7651 mddev->curr_resync = 2;
7652
7653 try_again:
404e4b43 7654 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7655 goto skip;
29ac4aa3 7656 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7657 if (mddev2 == mddev)
7658 continue;
90b08710
BS		 7659 if (!mddev->parallel_resync
7660 && mddev2->curr_resync
7661 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7662 DEFINE_WAIT(wq);
7663 if (mddev < mddev2 && mddev->curr_resync == 2) {
7664 /* arbitrarily yield */
7665 mddev->curr_resync = 1;
7666 wake_up(&resync_wait);
7667 }
7668 if (mddev > mddev2 && mddev->curr_resync == 1)
7669 /* no need to wait here, we can wait the next
7670 * time 'round when curr_resync == 2
7671 */
7672 continue;
9744197c
N		 7673 /* We need to wait 'interruptible' so as not to
7674 * contribute to the load average, and not to
7675 * be caught by 'softlockup'
7676 */
7677 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
c91abf5a 7678 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8712e553 7679 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7680 printk(KERN_INFO "md: delaying %s of %s"
7681 " until %s has finished (they"
1da177e4 7682 " share one or more physical units)\n",
61df9d91 7683 desc, mdname(mddev), mdname(mddev2));
1da177e4 7684 mddev_put(mddev2);
9744197c
N		 7685 if (signal_pending(current))
7686 flush_signals(current);
1da177e4
LT		 7687 schedule();
7688 finish_wait(&resync_wait, &wq);
7689 goto try_again;
7690 }
7691 finish_wait(&resync_wait, &wq);
7692 }
7693 }
7694 } while (mddev->curr_resync < 2);
7695
5fd6c1dc 7696 j = 0;
9d88883e 7697 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7698 /* resync follows the size requested by the personality,
57afd89f 7699 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7700 */
7701 max_sectors = mddev->resync_max_sectors;
7f7583d4 7702 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 7703 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7704 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7705 j = mddev->resync_min;
7706 else if (!mddev->bitmap)
5fd6c1dc 7707 j = mddev->recovery_cp;
5e96ee65 7708
ccfcc3c1 7709 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7710 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7711 else {
1da177e4 7712 /* recovery follows the physical size of devices */
58c0fed4 7713 max_sectors = mddev->dev_sectors;
5fd6c1dc 7714 j = MaxSector;
4e59ca7d 7715 rcu_read_lock();
dafb20fa 7716 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc
N		 7717 if (rdev->raid_disk >= 0 &&
7718 !test_bit(Faulty, &rdev->flags) &&
7719 !test_bit(In_sync, &rdev->flags) &&
7720 rdev->recovery_offset < j)
7721 j = rdev->recovery_offset;
4e59ca7d 7722 rcu_read_unlock();
133d4527
N		 7723
7724 /* If there is a bitmap, we need to make sure all
7725 * writes that started before we added a spare
7726 * complete before we start doing a recovery.
7727 * Otherwise the write might complete and (via
7728 * bitmap_endwrite) set a bit in the bitmap after the
7729 * recovery has checked that bit and skipped that
7730 * region.
7731 */
7732 if (mddev->bitmap) {
7733 mddev->pers->quiesce(mddev, 1);
7734 mddev->pers->quiesce(mddev, 0);
7735 }
5fd6c1dc 7736 }
1da177e4 7737
61df9d91
N		 7738 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7739 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7740 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7741 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7742 "(but not more than %d KB/sec) for %s.\n",
7743 speed_max(mddev), desc);
1da177e4 7744
eea1bf38 7745 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7746
57afd89f 7747 io_sectors = 0;
1da177e4
LT		 7748 for (m = 0; m < SYNC_MARKS; m++) {
7749 mark[m] = jiffies;
57afd89f 7750 mark_cnt[m] = io_sectors;
1da177e4
LT		 7751 }
7752 last_mark = 0;
7753 mddev->resync_mark = mark[last_mark];
7754 mddev->resync_mark_cnt = mark_cnt[last_mark];
7755
7756 /*
7757 * Tune reconstruction:
7758 */
7759 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7760 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7761 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7762
7763 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7764 last_check = 0;
7765
7766 if (j>2) {
c91abf5a 7767 printk(KERN_INFO
61df9d91
N		 7768 "md: resuming %s of %s from checkpoint.\n",
7769 desc, mdname(mddev));
1da177e4 7770 mddev->curr_resync = j;
72f36d59
N		 7771 } else
7772 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 7773 mddev->curr_resync_completed = j;
72f36d59
N		 7774 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7775 md_new_event(mddev);
54f89341 7776 update_time = jiffies;
1da177e4 7777
96ae923a 7778 if (mddev_is_clustered(mddev))
965400eb 7779 md_cluster_ops->resync_start(mddev, j, max_sectors);
96ae923a 7780
7c2c57c9 7781 blk_start_plug(&plug);
1da177e4 7782 while (j < max_sectors) {
57afd89f 7783 sector_t sectors;
1da177e4 7784
57afd89f 7785 skipped = 0;
97e4f42d 7786
7a91ee1f
N		 7787 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7788 ((mddev->curr_resync > mddev->curr_resync_completed &&
7789 (mddev->curr_resync - mddev->curr_resync_completed)
7790 > (max_sectors >> 4)) ||
54f89341 7791 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f
N		 7792 (j - mddev->curr_resync_completed)*2
7793 >= mddev->resync_max - mddev->curr_resync_completed
7794 )) {
97e4f42d 7795 /* time to update curr_resync_completed */
97e4f42d
N		 7796 wait_event(mddev->recovery_wait,
7797 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7798 mddev->curr_resync_completed = j;
35d78c66 7799 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7800 j > mddev->recovery_cp)
7801 mddev->recovery_cp = j;
54f89341 7802 update_time = jiffies;
070dc6dd 7803 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7804 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7805 }
acb180b0 7806
c91abf5a
N		 7807 while (j >= mddev->resync_max &&
7808 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
e62e58a5
N		 7809 /* As this condition is controlled by user-space,
7810 * we can block indefinitely, so use '_interruptible'
7811 * to avoid triggering warnings.
7812 */
7813 flush_signals(current); /* just in case */
7814 wait_event_interruptible(mddev->recovery_wait,
7815 mddev->resync_max > j
c91abf5a
N		 7816 || test_bit(MD_RECOVERY_INTR,
7817 &mddev->recovery));
e62e58a5 7818 }
acb180b0 7819
c91abf5a
N		 7820 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7821 break;
acb180b0 7822
09314799 7823 sectors = mddev->pers->sync_request(mddev, j, &skipped);
57afd89f 7824 if (sectors == 0) {
dfc70645 7825 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
c91abf5a 7826 break;
1da177e4 7827 }
57afd89f
N		 7828
7829 if (!skipped) { /* actual IO requested */
7830 io_sectors += sectors;
7831 atomic_add(sectors, &mddev->recovery_active);
7832 }
7833
e875ecea
N		 7834 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7835 break;
7836
1da177e4 7837 j += sectors;
72f36d59
N		 7838 if (j > 2)
7839 mddev->curr_resync = j;
96ae923a
GR		 7840 if (mddev_is_clustered(mddev))
7841 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
ff4e8d9a 7842 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7843 if (last_check == 0)
e875ecea 7844 /* this is the earliest that rebuild will be
d7603b7e
N		 7845 * visible in /proc/mdstat
7846 */
7847 md_new_event(mddev);
57afd89f
N		 7848
7849 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7850 continue;
7851
57afd89f 7852 last_check = io_sectors;
1da177e4
LT		 7853 repeat:
7854 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7855 /* step marks */
7856 int next = (last_mark+1) % SYNC_MARKS;
7857
7858 mddev->resync_mark = mark[next];
7859 mddev->resync_mark_cnt = mark_cnt[next];
7860 mark[next] = jiffies;
57afd89f 7861 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7862 last_mark = next;
7863 }
7864
c91abf5a
N		 7865 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7866 break;
1da177e4
LT		 7867
7868 /*
7869 * this loop exits only if either when we are slower than
7870 * the 'hard' speed limit, or the system was IO-idle for
7871 * a jiffy.
7872 * the system might be non-idle CPU-wise, but we only care
7873 * about not overloading the IO subsystem. (things like an
7874 * e2fsck being done on the RAID array should execute fast)
7875 */
1da177e4
LT		 7876 cond_resched();
7877
ac7e50a3
XN		 7878 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7879 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
57afd89f 7880 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7881
88202a0c 7882 if (currspeed > speed_min(mddev)) {
ac8fa419 7883 if (currspeed > speed_max(mddev)) {
c0e48521 7884 msleep(500);
1da177e4
LT		 7885 goto repeat;
7886 }
ac8fa419
N		 7887 if (!is_mddev_idle(mddev, 0)) {
7888 /*
7889 * Give other IO more of a chance.
7890 * The faster the devices, the less we wait.
7891 */
7892 wait_event(mddev->recovery_wait,
7893 !atomic_read(&mddev->recovery_active));
7894 }
1da177e4
LT		 7895 }
7896 }
c91abf5a
N		 7897 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7898 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7899 ? "interrupted" : "done");
1da177e4
LT		 7900 /*
7901 * this also signals 'finished resyncing' to md_stop
7902 */
7c2c57c9 7903 blk_finish_plug(&plug);
1da177e4
LT		 7904 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7905
7906 /* tell personality that we are finished */
09314799 7907 mddev->pers->sync_request(mddev, max_sectors, &skipped);
1da177e4 7908
96ae923a 7909 if (mddev_is_clustered(mddev))
965400eb 7910 md_cluster_ops->resync_finish(mddev);
96ae923a 7911
dfc70645 7912 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7913 mddev->curr_resync > 2) {
7914 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7915 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7916 if (mddev->curr_resync >= mddev->recovery_cp) {
7917 printk(KERN_INFO
61df9d91
N		 7918 "md: checkpointing %s of %s.\n",
7919 desc, mdname(mddev));
0a19caab 7920 if (test_bit(MD_RECOVERY_ERROR,
7921 &mddev->recovery))
7922 mddev->recovery_cp =
7923 mddev->curr_resync_completed;
7924 else
7925 mddev->recovery_cp =
7926 mddev->curr_resync;
5fd6c1dc
N		 7927 }
7928 } else
7929 mddev->recovery_cp = MaxSector;
7930 } else {
7931 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7932 mddev->curr_resync = MaxSector;
4e59ca7d 7933 rcu_read_lock();
dafb20fa 7934 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7935 if (rdev->raid_disk >= 0 &&
70fffd0b 7936 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7937 !test_bit(Faulty, &rdev->flags) &&
7938 !test_bit(In_sync, &rdev->flags) &&
7939 rdev->recovery_offset < mddev->curr_resync)
7940 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7941 rcu_read_unlock();
5fd6c1dc 7942 }
1da177e4 7943 }
db91ff55 7944 skip:
17571284 7945 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7946
23da422b 7947 spin_lock(&mddev->lock);
c07b70ad
N		 7948 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7949 /* We completed so min/max setting can be forgotten if used. */
7950 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7951 mddev->resync_min = 0;
7952 mddev->resync_max = MaxSector;
7953 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7954 mddev->resync_min = mddev->curr_resync_completed;
1da177e4 7955 mddev->curr_resync = 0;
23da422b
N		 7956 spin_unlock(&mddev->lock);
7957
1da177e4
LT		 7958 wake_up(&resync_wait);
7959 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7960 md_wakeup_thread(mddev->thread);
c6207277 7961 return;
1da177e4 7962}
29269553 7963EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7964
746d3207
N		 7965static int remove_and_add_spares(struct mddev *mddev,
7966 struct md_rdev *this)
b4c4c7b8 7967{
3cb03002 7968 struct md_rdev *rdev;
b4c4c7b8 7969 int spares = 0;
f2a371c5 7970 int removed = 0;
b4c4c7b8 7971
dafb20fa 7972 rdev_for_each(rdev, mddev)
746d3207
N		 7973 if ((this == NULL || rdev == this) &&
7974 rdev->raid_disk >= 0 &&
6bfe0b49 7975 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7976 (test_bit(Faulty, &rdev->flags) ||
7977 ! test_bit(In_sync, &rdev->flags)) &&
7978 atomic_read(&rdev->nr_pending)==0) {
7979 if (mddev->pers->hot_remove_disk(
b8321b68 7980 mddev, rdev) == 0) {
36fad858 7981 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 7982 rdev->raid_disk = -1;
f2a371c5 7983 removed++;
b4c4c7b8
N		 7984 }
7985 }
90584fc9
JB		 7986 if (removed && mddev->kobj.sd)
7987 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 7988
746d3207
N		 7989 if (this)
7990 goto no_add;
7991
dafb20fa 7992 rdev_for_each(rdev, mddev) {
7bfec5f3
N		 7993 if (rdev->raid_disk >= 0 &&
7994 !test_bit(In_sync, &rdev->flags) &&
7995 !test_bit(Faulty, &rdev->flags))
7996 spares++;
7ceb17e8
N		 7997 if (rdev->raid_disk >= 0)
7998 continue;
7999 if (test_bit(Faulty, &rdev->flags))
8000 continue;
8001 if (mddev->ro &&
8313b8e5
N		 8002 ! (rdev->saved_raid_disk >= 0 &&
8003 !test_bit(Bitmap_sync, &rdev->flags)))
7ceb17e8
N		 8004 continue;
8005
7eb41885
N		 8006 if (rdev->saved_raid_disk < 0)
8007 rdev->recovery_offset = 0;
7ceb17e8
N		 8008 if (mddev->pers->
8009 hot_add_disk(mddev, rdev) == 0) {
8010 if (sysfs_link_rdev(mddev, rdev))
8011 /* failure here is OK */;
8012 spares++;
8013 md_new_event(mddev);
8014 set_bit(MD_CHANGE_DEVS, &mddev->flags);
dfc70645 8015 }
b4c4c7b8 8016 }
746d3207 8017no_add:
6dafab6b
N		 8018 if (removed)
8019 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 8020 return spares;
8021}
7ebc0be7 8022
ac05f256
N		 8023static void md_start_sync(struct work_struct *ws)
8024{
8025 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8026
8027 mddev->sync_thread = md_register_thread(md_do_sync,
8028 mddev,
8029 "resync");
8030 if (!mddev->sync_thread) {
8031 printk(KERN_ERR "%s: could not start resync"
8032 " thread...\n",
8033 mdname(mddev));
8034 /* leave the spares where they are, it shouldn't hurt */
8035 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8036 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8037 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8038 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8039 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8040 wake_up(&resync_wait);
ac05f256
N		 8041 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8042 &mddev->recovery))
8043 if (mddev->sysfs_action)
8044 sysfs_notify_dirent_safe(mddev->sysfs_action);
8045 } else
8046 md_wakeup_thread(mddev->sync_thread);
8047 sysfs_notify_dirent_safe(mddev->sysfs_action);
8048 md_new_event(mddev);
8049}
8050
1da177e4
LT		 8051/*
8052 * This routine is regularly called by all per-raid-array threads to
8053 * deal with generic issues like resync and super-block update.
8054 * Raid personalities that don't have a thread (linear/raid0) do not
8055 * need this as they never do any recovery or update the superblock.
8056 *
8057 * It does not do any resync itself, but rather "forks" off other threads
8058 * to do that as needed.
8059 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8060 * "->recovery" and create a thread at ->sync_thread.
dfc70645 8061 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 8062 * and wakeups up this thread which will reap the thread and finish up.
8063 * This thread also removes any faulty devices (with nr_pending == 0).
8064 *
8065 * The overall approach is:
8066 * 1/ if the superblock needs updating, update it.
8067 * 2/ If a recovery thread is running, don't do anything else.
8068 * 3/ If recovery has finished, clean up, possibly marking spares active.
8069 * 4/ If there are any faulty devices, remove them.
8070 * 5/ If array is degraded, try to add spares devices
8071 * 6/ If array has spares or is not in-sync, start a resync thread.
8072 */
fd01b88c 8073void md_check_recovery(struct mddev *mddev)
1da177e4 8074{
68866e42
JB		 8075 if (mddev->suspended)
8076 return;
8077
5f40402d 8078 if (mddev->bitmap)
aa5cbd10 8079 bitmap_daemon_work(mddev);
1da177e4 8080
fca4d848 8081 if (signal_pending(current)) {
31a59e34 8082 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 8083 printk(KERN_INFO "md: %s in immediate safe mode\n",
8084 mdname(mddev));
8085 mddev->safemode = 2;
8086 }
8087 flush_signals(current);
8088 }
8089
c89a8eee
N		 8090 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8091 return;
1da177e4 8092 if ( ! (
142d44c3 8093 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 8094 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 8095 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 8096 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 8097 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8098 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 8099 ))
8100 return;
fca4d848 8101
df5b89b3 8102 if (mddev_trylock(mddev)) {
b4c4c7b8 8103 int spares = 0;
fca4d848 8104
c89a8eee 8105 if (mddev->ro) {
7ceb17e8
N		 8106 /* On a read-only array we can:
8107 * - remove failed devices
8108 * - add already-in_sync devices if the array itself
8109 * is in-sync.
8110 * As we only add devices that are already in-sync,
8111 * we can activate the spares immediately.
c89a8eee 8112 */
7ceb17e8 8113 remove_and_add_spares(mddev, NULL);
8313b8e5
N		 8114 /* There is no thread, but we need to call
8115 * ->spare_active and clear saved_raid_disk
8116 */
2ac295a5 8117 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8313b8e5
N		 8118 md_reap_sync_thread(mddev);
8119 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
c89a8eee
N		 8120 goto unlock;
8121 }
8122
31a59e34 8123 if (!mddev->external) {
0fd62b86 8124 int did_change = 0;
85572d7c 8125 spin_lock(&mddev->lock);
31a59e34
N		 8126 if (mddev->safemode &&
8127 !atomic_read(&mddev->writes_pending) &&
8128 !mddev->in_sync &&
8129 mddev->recovery_cp == MaxSector) {
8130 mddev->in_sync = 1;
0fd62b86 8131 did_change = 1;
070dc6dd 8132 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 8133 }
8134 if (mddev->safemode == 1)
8135 mddev->safemode = 0;
85572d7c 8136 spin_unlock(&mddev->lock);
0fd62b86 8137 if (did_change)
00bcb4ac 8138 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 8139 }
fca4d848 8140
293467aa
GR		 8141 if (mddev->flags & MD_UPDATE_SB_FLAGS) {
8142 if (mddev_is_clustered(mddev))
8143 md_cluster_ops->metadata_update_start(mddev);
850b2b42 8144 md_update_sb(mddev, 0);
293467aa
GR		 8145 if (mddev_is_clustered(mddev))
8146 md_cluster_ops->metadata_update_finish(mddev);
8147 }
06d91a5f 8148
1da177e4
LT		 8149 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8150 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8151 /* resync/recovery still happening */
8152 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8153 goto unlock;
8154 }
8155 if (mddev->sync_thread) {
a91d5ac0 8156 md_reap_sync_thread(mddev);
1da177e4
LT		 8157 goto unlock;
8158 }
72a23c21
NB		 8159 /* Set RUNNING before clearing NEEDED to avoid
8160 * any transients in the value of "sync_action".
8161 */
72f36d59 8162 mddev->curr_resync_completed = 0;
23da422b 8163 spin_lock(&mddev->lock);
72a23c21 8164 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
23da422b 8165 spin_unlock(&mddev->lock);
24dd469d
N		 8166 /* Clear some bits that don't mean anything, but
8167 * might be left set
8168 */
24dd469d
N		 8169 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8170 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 8171
ed209584
N		 8172 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8173 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
ac05f256 8174 goto not_running;
1da177e4
LT		 8175 /* no recovery is running.
8176 * remove any failed drives, then
8177 * add spares if possible.
72f36d59 8178 * Spares are also removed and re-added, to allow
1da177e4
LT		 8179 * the personality to fail the re-add.
8180 */
1da177e4 8181
b4c4c7b8 8182 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 8183 if (mddev->pers->check_reshape == NULL ||
8184 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8 8185 /* Cannot proceed */
ac05f256 8186 goto not_running;
b4c4c7b8 8187 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 8188 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 8189 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N		 8190 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8191 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 8192 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 8193 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8194 } else if (mddev->recovery_cp < MaxSector) {
8195 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 8196 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 8197 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8198 /* nothing to be done ... */
ac05f256 8199 goto not_running;
24dd469d 8200
1da177e4 8201 if (mddev->pers->sync_request) {
ef99bf48 8202 if (spares) {
a654b9d8
N		 8203 /* We are adding a device or devices to an array
8204 * which has the bitmap stored on all devices.
8205 * So make sure all bitmap pages get written
8206 */
8207 bitmap_write_all(mddev->bitmap);
8208 }
ac05f256
N		 8209 INIT_WORK(&mddev->del_work, md_start_sync);
8210 queue_work(md_misc_wq, &mddev->del_work);
8211 goto unlock;
1da177e4 8212 }
ac05f256 8213 not_running:
72a23c21
NB		 8214 if (!mddev->sync_thread) {
8215 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8216 wake_up(&resync_wait);
72a23c21
NB		 8217 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8218 &mddev->recovery))
0c3573f1 8219 if (mddev->sysfs_action)
00bcb4ac 8220 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 8221 }
ac05f256
N		 8222 unlock:
8223 wake_up(&mddev->sb_wait);
1da177e4
LT		 8224 mddev_unlock(mddev);
8225 }
8226}
6c144d31 8227EXPORT_SYMBOL(md_check_recovery);
1da177e4 8228
a91d5ac0
JB		 8229void md_reap_sync_thread(struct mddev *mddev)
8230{
8231 struct md_rdev *rdev;
8232
8233 /* resync has finished, collect result */
8234 md_unregister_thread(&mddev->sync_thread);
8235 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8236 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8237 /* success...*/
8238 /* activate any spares */
8239 if (mddev->pers->spare_active(mddev)) {
8240 sysfs_notify(&mddev->kobj, NULL,
8241 "degraded");
8242 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8243 }
8244 }
293467aa
GR		 8245 if (mddev_is_clustered(mddev))
8246 md_cluster_ops->metadata_update_start(mddev);
a91d5ac0
JB		 8247 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8248 mddev->pers->finish_reshape)
8249 mddev->pers->finish_reshape(mddev);
8250
8251 /* If array is no-longer degraded, then any saved_raid_disk
f466722c 8252 * information must be scrapped.
a91d5ac0 8253 */
f466722c
N		 8254 if (!mddev->degraded)
8255 rdev_for_each(rdev, mddev)
a91d5ac0
JB		 8256 rdev->saved_raid_disk = -1;
8257
8258 md_update_sb(mddev, 1);
293467aa
GR		 8259 if (mddev_is_clustered(mddev))
8260 md_cluster_ops->metadata_update_finish(mddev);
a91d5ac0
JB		 8261 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8262 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8263 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8264 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8265 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
f851b60d 8266 wake_up(&resync_wait);
a91d5ac0
JB		 8267 /* flag recovery needed just to double check */
8268 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8269 sysfs_notify_dirent_safe(mddev->sysfs_action);
8270 md_new_event(mddev);
8271 if (mddev->event_work.func)
8272 queue_work(md_misc_wq, &mddev->event_work);
8273}
6c144d31 8274EXPORT_SYMBOL(md_reap_sync_thread);
a91d5ac0 8275
fd01b88c 8276void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 8277{
00bcb4ac 8278 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 8279 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 8280 !test_bit(Blocked, &rdev->flags) &&
8281 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 8282 msecs_to_jiffies(5000));
8283 rdev_dec_pending(rdev, mddev);
8284}
8285EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8286
c6563a8c
N		 8287void md_finish_reshape(struct mddev *mddev)
8288{
8289 /* called be personality module when reshape completes. */
8290 struct md_rdev *rdev;
8291
8292 rdev_for_each(rdev, mddev) {
8293 if (rdev->data_offset > rdev->new_data_offset)
8294 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8295 else
8296 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8297 rdev->data_offset = rdev->new_data_offset;
8298 }
8299}
8300EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N		 8301
8302/* Bad block management.
8303 * We can record which blocks on each device are 'bad' and so just
8304 * fail those blocks, or that stripe, rather than the whole device.
8305 * Entries in the bad-block table are 64bits wide. This comprises:
8306 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8307 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8308 * A 'shift' can be set so that larger blocks are tracked and
8309 * consequently larger devices can be covered.
8310 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8311 *
8312 * Locking of the bad-block table uses a seqlock so md_is_badblock
8313 * might need to retry if it is very unlucky.
8314 * We will sometimes want to check for bad blocks in a bi_end_io function,
8315 * so we use the write_seqlock_irq variant.
8316 *
8317 * When looking for a bad block we specify a range and want to
8318 * know if any block in the range is bad. So we binary-search
8319 * to the last range that starts at-or-before the given endpoint,
8320 * (or "before the sector after the target range")
8321 * then see if it ends after the given start.
8322 * We return
8323 * 0 if there are no known bad blocks in the range
8324 * 1 if there are known bad block which are all acknowledged
8325 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8326 * plus the start/length of the first bad section we overlap.
8327 */
8328int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8329 sector_t *first_bad, int *bad_sectors)
8330{
8331 int hi;
ab05613a 8332 int lo;
2230dfe4 8333 u64 *p = bb->page;
ab05613a 8334 int rv;
2230dfe4
N		 8335 sector_t target = s + sectors;
8336 unsigned seq;
8337
8338 if (bb->shift > 0) {
8339 /* round the start down, and the end up */
8340 s >>= bb->shift;
8341 target += (1<<bb->shift) - 1;
8342 target >>= bb->shift;
8343 sectors = target - s;
8344 }
8345 /* 'target' is now the first block after the bad range */
8346
8347retry:
8348 seq = read_seqbegin(&bb->lock);
ab05613a 8349 lo = 0;
8350 rv = 0;
2230dfe4
N		 8351 hi = bb->count;
8352
8353 /* Binary search between lo and hi for 'target'
8354 * i.e. for the last range that starts before 'target'
8355 */
8356 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8357 * are known not to be the last range before target.
8358 * VARIANT: hi-lo is the number of possible
8359 * ranges, and decreases until it reaches 1
8360 */
8361 while (hi - lo > 1) {
8362 int mid = (lo + hi) / 2;
8363 sector_t a = BB_OFFSET(p[mid]);
8364 if (a < target)
8365 /* This could still be the one, earlier ranges
8366 * could not. */
8367 lo = mid;
8368 else
8369 /* This and later ranges are definitely out. */
8370 hi = mid;
8371 }
8372 /* 'lo' might be the last that started before target, but 'hi' isn't */
8373 if (hi > lo) {
8374 /* need to check all range that end after 's' to see if
8375 * any are unacknowledged.
8376 */
8377 while (lo >= 0 &&
8378 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8379 if (BB_OFFSET(p[lo]) < target) {
8380 /* starts before the end, and finishes after
8381 * the start, so they must overlap
8382 */
8383 if (rv != -1 && BB_ACK(p[lo]))
8384 rv = 1;
8385 else
8386 rv = -1;
8387 *first_bad = BB_OFFSET(p[lo]);
8388 *bad_sectors = BB_LEN(p[lo]);
8389 }
8390 lo--;
8391 }
8392 }
8393
8394 if (read_seqretry(&bb->lock, seq))
8395 goto retry;
8396
8397 return rv;
8398}
8399EXPORT_SYMBOL_GPL(md_is_badblock);
8400
8401/*
8402 * Add a range of bad blocks to the table.
8403 * This might extend the table, or might contract it
8404 * if two adjacent ranges can be merged.
8405 * We binary-search to find the 'insertion' point, then
8406 * decide how best to handle it.
8407 */
8408static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8409 int acknowledged)
8410{
8411 u64 *p;
8412 int lo, hi;
8413 int rv = 1;
905b0297 8414 unsigned long flags;
2230dfe4
N		 8415
8416 if (bb->shift < 0)
8417 /* badblocks are disabled */
8418 return 0;
8419
8420 if (bb->shift) {
8421 /* round the start down, and the end up */
8422 sector_t next = s + sectors;
8423 s >>= bb->shift;
8424 next += (1<<bb->shift) - 1;
8425 next >>= bb->shift;
8426 sectors = next - s;
8427 }
8428
905b0297 8429 write_seqlock_irqsave(&bb->lock, flags);
2230dfe4
N		 8430
8431 p = bb->page;
8432 lo = 0;
8433 hi = bb->count;
8434 /* Find the last range that starts at-or-before 's' */
8435 while (hi - lo > 1) {
8436 int mid = (lo + hi) / 2;
8437 sector_t a = BB_OFFSET(p[mid]);
8438 if (a <= s)
8439 lo = mid;
8440 else
8441 hi = mid;
8442 }
8443 if (hi > lo && BB_OFFSET(p[lo]) > s)
8444 hi = lo;
8445
8446 if (hi > lo) {
8447 /* we found a range that might merge with the start
8448 * of our new range
8449 */
8450 sector_t a = BB_OFFSET(p[lo]);
8451 sector_t e = a + BB_LEN(p[lo]);
8452 int ack = BB_ACK(p[lo]);
8453 if (e >= s) {
8454 /* Yes, we can merge with a previous range */
8455 if (s == a && s + sectors >= e)
8456 /* new range covers old */
8457 ack = acknowledged;
8458 else
8459 ack = ack && acknowledged;
8460
8461 if (e < s + sectors)
8462 e = s + sectors;
8463 if (e - a <= BB_MAX_LEN) {
8464 p[lo] = BB_MAKE(a, e-a, ack);
8465 s = e;
8466 } else {
8467 /* does not all fit in one range,
8468 * make p[lo] maximal
8469 */
8470 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8471 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8472 s = a + BB_MAX_LEN;
8473 }
8474 sectors = e - s;
8475 }
8476 }
8477 if (sectors && hi < bb->count) {
8478 /* 'hi' points to the first range that starts after 's'.
8479 * Maybe we can merge with the start of that range */
8480 sector_t a = BB_OFFSET(p[hi]);
8481 sector_t e = a + BB_LEN(p[hi]);
8482 int ack = BB_ACK(p[hi]);
8483 if (a <= s + sectors) {
8484 /* merging is possible */
8485 if (e <= s + sectors) {
8486 /* full overlap */
8487 e = s + sectors;
8488 ack = acknowledged;
8489 } else
8490 ack = ack && acknowledged;
8491
8492 a = s;
8493 if (e - a <= BB_MAX_LEN) {
8494 p[hi] = BB_MAKE(a, e-a, ack);
8495 s = e;
8496 } else {
8497 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8498 s = a + BB_MAX_LEN;
8499 }
8500 sectors = e - s;
8501 lo = hi;
8502 hi++;
8503 }
8504 }
8505 if (sectors == 0 && hi < bb->count) {
8506 /* we might be able to combine lo and hi */
8507 /* Note: 's' is at the end of 'lo' */
8508 sector_t a = BB_OFFSET(p[hi]);
8509 int lolen = BB_LEN(p[lo]);
8510 int hilen = BB_LEN(p[hi]);
8511 int newlen = lolen + hilen - (s - a);
8512 if (s >= a && newlen < BB_MAX_LEN) {
8513 /* yes, we can combine them */
8514 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8515 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8516 memmove(p + hi, p + hi + 1,
8517 (bb->count - hi - 1) * 8);
8518 bb->count--;
8519 }
8520 }
8521 while (sectors) {
8522 /* didn't merge (it all).
8523 * Need to add a range just before 'hi' */
8524 if (bb->count >= MD_MAX_BADBLOCKS) {
8525 /* No room for more */
8526 rv = 0;
8527 break;
8528 } else {
8529 int this_sectors = sectors;
8530 memmove(p + hi + 1, p + hi,
8531 (bb->count - hi) * 8);
8532 bb->count++;
8533
8534 if (this_sectors > BB_MAX_LEN)
8535 this_sectors = BB_MAX_LEN;
8536 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8537 sectors -= this_sectors;
8538 s += this_sectors;
8539 }
8540 }
8541
8542 bb->changed = 1;
de393cde
N		 8543 if (!acknowledged)
8544 bb->unacked_exist = 1;
905b0297 8545 write_sequnlock_irqrestore(&bb->lock, flags);
2230dfe4
N		 8546
8547 return rv;
8548}
8549
3cb03002 8550int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8551 int is_new)
2230dfe4 8552{
c6563a8c
N		 8553 int rv;
8554 if (is_new)
8555 s += rdev->new_data_offset;
8556 else
8557 s += rdev->data_offset;
8558 rv = md_set_badblocks(&rdev->badblocks,
8559 s, sectors, 0);
2230dfe4
N		 8560 if (rv) {
8561 /* Make sure they get written out promptly */
8bd2f0a0 8562 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 8563 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8564 md_wakeup_thread(rdev->mddev->thread);
8565 }
8566 return rv;
8567}
8568EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8569
8570/*
8571 * Remove a range of bad blocks from the table.
8572 * This may involve extending the table if we spilt a region,
8573 * but it must not fail. So if the table becomes full, we just
8574 * drop the remove request.
8575 */
8576static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8577{
8578 u64 *p;
8579 int lo, hi;
8580 sector_t target = s + sectors;
8581 int rv = 0;
8582
8583 if (bb->shift > 0) {
8584 /* When clearing we round the start up and the end down.
8585 * This should not matter as the shift should align with
8586 * the block size and no rounding should ever be needed.
8587 * However it is better the think a block is bad when it
8588 * isn't than to think a block is not bad when it is.
8589 */
8590 s += (1<<bb->shift) - 1;
8591 s >>= bb->shift;
8592 target >>= bb->shift;
8593 sectors = target - s;
8594 }
8595
8596 write_seqlock_irq(&bb->lock);
8597
8598 p = bb->page;
8599 lo = 0;
8600 hi = bb->count;
8601 /* Find the last range that starts before 'target' */
8602 while (hi - lo > 1) {
8603 int mid = (lo + hi) / 2;
8604 sector_t a = BB_OFFSET(p[mid]);
8605 if (a < target)
8606 lo = mid;
8607 else
8608 hi = mid;
8609 }
8610 if (hi > lo) {
8611 /* p[lo] is the last range that could overlap the
8612 * current range. Earlier ranges could also overlap,
8613 * but only this one can overlap the end of the range.
8614 */
8615 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8616 /* Partial overlap, leave the tail of this range */
8617 int ack = BB_ACK(p[lo]);
8618 sector_t a = BB_OFFSET(p[lo]);
8619 sector_t end = a + BB_LEN(p[lo]);
8620
8621 if (a < s) {
8622 /* we need to split this range */
8623 if (bb->count >= MD_MAX_BADBLOCKS) {
8b32bf5e 8624 rv = -ENOSPC;
2230dfe4
N		 8625 goto out;
8626 }
8627 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8628 bb->count++;
8629 p[lo] = BB_MAKE(a, s-a, ack);
8630 lo++;
8631 }
8632 p[lo] = BB_MAKE(target, end - target, ack);
8633 /* there is no longer an overlap */
8634 hi = lo;
8635 lo--;
8636 }
8637 while (lo >= 0 &&
8638 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8639 /* This range does overlap */
8640 if (BB_OFFSET(p[lo]) < s) {
8641 /* Keep the early parts of this range. */
8642 int ack = BB_ACK(p[lo]);
8643 sector_t start = BB_OFFSET(p[lo]);
8644 p[lo] = BB_MAKE(start, s - start, ack);
8645 /* now low doesn't overlap, so.. */
8646 break;
8647 }
8648 lo--;
8649 }
8650 /* 'lo' is strictly before, 'hi' is strictly after,
8651 * anything between needs to be discarded
8652 */
8653 if (hi - lo > 1) {
8654 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8655 bb->count -= (hi - lo - 1);
8656 }
8657 }
8658
8659 bb->changed = 1;
8660out:
8661 write_sequnlock_irq(&bb->lock);
8662 return rv;
8663}
8664
c6563a8c
N		 8665int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8666 int is_new)
2230dfe4 8667{
c6563a8c
N		 8668 if (is_new)
8669 s += rdev->new_data_offset;
8670 else
8671 s += rdev->data_offset;
2230dfe4 8672 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8673 s, sectors);
2230dfe4
N		 8674}
8675EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8676
8677/*
8678 * Acknowledge all bad blocks in a list.
8679 * This only succeeds if ->changed is clear. It is used by
8680 * in-kernel metadata updates
8681 */
8682void md_ack_all_badblocks(struct badblocks *bb)
8683{
8684 if (bb->page == NULL || bb->changed)
8685 /* no point even trying */
8686 return;
8687 write_seqlock_irq(&bb->lock);
8688
ecb178bb 8689 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N		 8690 u64 *p = bb->page;
8691 int i;
8692 for (i = 0; i < bb->count ; i++) {
8693 if (!BB_ACK(p[i])) {
8694 sector_t start = BB_OFFSET(p[i]);
8695 int len = BB_LEN(p[i]);
8696 p[i] = BB_MAKE(start, len, 1);
8697 }
8698 }
de393cde 8699 bb->unacked_exist = 0;
2230dfe4
N		 8700 }
8701 write_sequnlock_irq(&bb->lock);
8702}
8703EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8704
16c791a5
N		 8705/* sysfs access to bad-blocks list.
8706 * We present two files.
8707 * 'bad-blocks' lists sector numbers and lengths of ranges that
8708 * are recorded as bad. The list is truncated to fit within
8709 * the one-page limit of sysfs.
8710 * Writing "sector length" to this file adds an acknowledged
8711 * bad block list.
8712 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8713 * been acknowledged. Writing to this file adds bad blocks
8714 * without acknowledging them. This is largely for testing.
8715 */
8716
8717static ssize_t
8718badblocks_show(struct badblocks *bb, char *page, int unack)
8719{
8720 size_t len;
8721 int i;
8722 u64 *p = bb->page;
8723 unsigned seq;
8724
8725 if (bb->shift < 0)
8726 return 0;
8727
8728retry:
8729 seq = read_seqbegin(&bb->lock);
8730
8731 len = 0;
8732 i = 0;
8733
8734 while (len < PAGE_SIZE && i < bb->count) {
8735 sector_t s = BB_OFFSET(p[i]);
8736 unsigned int length = BB_LEN(p[i]);
8737 int ack = BB_ACK(p[i]);
8738 i++;
8739
8740 if (unack && ack)
8741 continue;
8742
8743 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8744 (unsigned long long)s << bb->shift,
8745 length << bb->shift);
8746 }
de393cde
N		 8747 if (unack && len == 0)
8748 bb->unacked_exist = 0;
16c791a5
N		 8749
8750 if (read_seqretry(&bb->lock, seq))
8751 goto retry;
8752
8753 return len;
8754}
8755
8756#define DO_DEBUG 1
8757
8758static ssize_t
8759badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8760{
8761 unsigned long long sector;
8762 int length;
8763 char newline;
8764#ifdef DO_DEBUG
8765 /* Allow clearing via sysfs *only* for testing/debugging.
8766 * Normally only a successful write may clear a badblock
8767 */
8768 int clear = 0;
8769 if (page[0] == '-') {
8770 clear = 1;
8771 page++;
8772 }
8773#endif /* DO_DEBUG */
8774
8775 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8776 case 3:
8777 if (newline != '\n')
8778 return -EINVAL;
8779 case 2:
8780 if (length <= 0)
8781 return -EINVAL;
8782 break;
8783 default:
8784 return -EINVAL;
8785 }
8786
8787#ifdef DO_DEBUG
8788 if (clear) {
8789 md_clear_badblocks(bb, sector, length);
8790 return len;
8791 }
8792#endif /* DO_DEBUG */
8793 if (md_set_badblocks(bb, sector, length, !unack))
8794 return len;
8795 else
8796 return -ENOSPC;
8797}
8798
75c96f85
AB		 8799static int md_notify_reboot(struct notifier_block *this,
8800 unsigned long code, void *x)
1da177e4
LT		 8801{
8802 struct list_head *tmp;
fd01b88c 8803 struct mddev *mddev;
2dba6a91 8804 int need_delay = 0;
1da177e4 8805
c744a65c
N		 8806 for_each_mddev(mddev, tmp) {
8807 if (mddev_trylock(mddev)) {
30b8aa91
N		 8808 if (mddev->pers)
8809 __md_stop_writes(mddev);
0f62fb22
N		 8810 if (mddev->persistent)
8811 mddev->safemode = 2;
c744a65c 8812 mddev_unlock(mddev);
2dba6a91 8813 }
c744a65c 8814 need_delay = 1;
1da177e4 8815 }
c744a65c
N		 8816 /*
8817 * certain more exotic SCSI devices are known to be
8818 * volatile wrt too early system reboots. While the
8819 * right place to handle this issue is the given
8820 * driver, we do want to have a safe RAID driver ...
8821 */
8822 if (need_delay)
8823 mdelay(1000*1);
8824
1da177e4
LT		 8825 return NOTIFY_DONE;
8826}
8827
75c96f85 8828static struct notifier_block md_notifier = {
1da177e4
LT		 8829 .notifier_call = md_notify_reboot,
8830 .next = NULL,
8831 .priority = INT_MAX, /* before any real devices */
8832};
8833
8834static void md_geninit(void)
8835{
36a4e1fe 8836 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8837
c7705f34 8838 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8839}
8840
75c96f85 8841static int __init md_init(void)
1da177e4 8842{
e804ac78
TH		 8843 int ret = -ENOMEM;
8844
ada609ee 8845 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8846 if (!md_wq)
8847 goto err_wq;
8848
8849 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8850 if (!md_misc_wq)
8851 goto err_misc_wq;
8852
8853 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8854 goto err_md;
8855
8856 if ((ret = register_blkdev(0, "mdp")) < 0)
8857 goto err_mdp;
8858 mdp_major = ret;
8859
af5628f0 8860 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
e8703fe1
N		 8861 md_probe, NULL, NULL);
8862 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8863 md_probe, NULL, NULL);
8864
1da177e4 8865 register_reboot_notifier(&md_notifier);
0b4d4147 8866 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8867
8868 md_geninit();
d710e138 8869 return 0;
1da177e4 8870
e804ac78
TH		 8871err_mdp:
8872 unregister_blkdev(MD_MAJOR, "md");
8873err_md:
8874 destroy_workqueue(md_misc_wq);
8875err_misc_wq:
8876 destroy_workqueue(md_wq);
8877err_wq:
8878 return ret;
8879}
1da177e4 8880
1d7e3e96
GR		 8881void md_reload_sb(struct mddev *mddev)
8882{
8883 struct md_rdev *rdev, *tmp;
8884
8885 rdev_for_each_safe(rdev, tmp, mddev) {
8886 rdev->sb_loaded = 0;
8887 ClearPageUptodate(rdev->sb_page);
8888 }
8889 mddev->raid_disks = 0;
8890 analyze_sbs(mddev);
8891 rdev_for_each_safe(rdev, tmp, mddev) {
8892 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8893 /* since we don't write to faulty devices, we figure out if the
8894 * disk is faulty by comparing events
8895 */
8896 if (mddev->events > sb->events)
8897 set_bit(Faulty, &rdev->flags);
8898 }
8899
8900}
8901EXPORT_SYMBOL(md_reload_sb);
8902
1da177e4
LT		 8903#ifndef MODULE
8904
8905/*
8906 * Searches all registered partitions for autorun RAID arrays
8907 * at boot time.
8908 */
4d936ec1
ME		 8909
8910static LIST_HEAD(all_detected_devices);
8911struct detected_devices_node {
8912 struct list_head list;
8913 dev_t dev;
8914};
1da177e4
LT		 8915
8916void md_autodetect_dev(dev_t dev)
8917{
4d936ec1
ME		 8918 struct detected_devices_node *node_detected_dev;
8919
8920 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8921 if (node_detected_dev) {
8922 node_detected_dev->dev = dev;
8923 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8924 } else {
8925 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8926 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8927 }
1da177e4
LT		 8928}
8929
1da177e4
LT		 8930static void autostart_arrays(int part)
8931{
3cb03002 8932 struct md_rdev *rdev;
4d936ec1
ME		 8933 struct detected_devices_node *node_detected_dev;
8934 dev_t dev;
8935 int i_scanned, i_passed;
1da177e4 8936
4d936ec1
ME		 8937 i_scanned = 0;
8938 i_passed = 0;
1da177e4 8939
4d936ec1 8940 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8941
4d936ec1
ME		 8942 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8943 i_scanned++;
8944 node_detected_dev = list_entry(all_detected_devices.next,
8945 struct detected_devices_node, list);
8946 list_del(&node_detected_dev->list);
8947 dev = node_detected_dev->dev;
8948 kfree(node_detected_dev);
df968c4e 8949 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8950 if (IS_ERR(rdev))
8951 continue;
8952
403df478 8953 if (test_bit(Faulty, &rdev->flags))
1da177e4 8954 continue;
403df478 8955
d0fae18f 8956 set_bit(AutoDetected, &rdev->flags);
1da177e4 8957 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8958 i_passed++;
1da177e4 8959 }
4d936ec1
ME		 8960
8961 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8962 i_scanned, i_passed);
1da177e4
LT		 8963
8964 autorun_devices(part);
8965}
8966
fdee8ae4 8967#endif /* !MODULE */
1da177e4
LT		 8968
8969static __exit void md_exit(void)
8970{
fd01b88c 8971 struct mddev *mddev;
1da177e4 8972 struct list_head *tmp;
e2f23b60 8973 int delay = 1;
8ab5e4c1 8974
af5628f0 8975 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
e8703fe1 8976 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8977
3dbd8c2e 8978 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8979 unregister_blkdev(mdp_major, "mdp");
8980 unregister_reboot_notifier(&md_notifier);
8981 unregister_sysctl_table(raid_table_header);
e2f23b60
N		 8982
8983 /* We cannot unload the modules while some process is
8984 * waiting for us in select() or poll() - wake them up
8985 */
8986 md_unloading = 1;
8987 while (waitqueue_active(&md_event_waiters)) {
8988 /* not safe to leave yet */
8989 wake_up(&md_event_waiters);
8990 msleep(delay);
8991 delay += delay;
8992 }
1da177e4 8993 remove_proc_entry("mdstat", NULL);
e2f23b60 8994
29ac4aa3 8995 for_each_mddev(mddev, tmp) {
1da177e4 8996 export_array(mddev);
d3374825 8997 mddev->hold_active = 0;
1da177e4 8998 }
e804ac78
TH		 8999 destroy_workqueue(md_misc_wq);
9000 destroy_workqueue(md_wq);
1da177e4
LT		 9001}
9002
685784aa 9003subsys_initcall(md_init);
1da177e4
LT		 9004module_exit(md_exit)
9005
f91de92e
N		 9006static int get_ro(char *buffer, struct kernel_param *kp)
9007{
9008 return sprintf(buffer, "%d", start_readonly);
9009}
9010static int set_ro(const char *val, struct kernel_param *kp)
9011{
9012 char *e;
9013 int num = simple_strtoul(val, &e, 10);
9014 if (*val && (*e == '\0' || *e == '\n')) {
9015 start_readonly = num;
4dbcdc75 9016 return 0;
f91de92e
N		 9017 }
9018 return -EINVAL;
9019}
9020
80ca3a44
N		 9021module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9022module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
efeb53c0 9023module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 9024
1da177e4 9025MODULE_LICENSE("GPL");
0efb9e61 9026MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 9027MODULE_ALIAS("md");
72008652 9028MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)