projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md/raid10: submit bio directly to replacement disk
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
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"
1da177e4 56
1da177e4 57#ifndef MODULE
d710e138 58static void autostart_arrays(int part);
1da177e4
LT		 59#endif
60
01f96c0a
N		 61/* pers_list is a list of registered personalities protected
62 * by pers_lock.
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
65 */
2604b703 66static LIST_HEAD(pers_list);
1da177e4
LT		 67static DEFINE_SPINLOCK(pers_lock);
68
5e56341d
AB		 69static void md_print_devices(void);
70
90b08710 71static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 72static struct workqueue_struct *md_wq;
73static struct workqueue_struct *md_misc_wq;
90b08710 74
746d3207
N		 75static int remove_and_add_spares(struct mddev *mddev,
76 struct md_rdev *this);
77
5e56341d
AB		 78#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
79
1e50915f
RB		 80/*
81 * Default number of read corrections we'll attempt on an rdev
82 * before ejecting it from the array. We divide the read error
83 * count by 2 for every hour elapsed between read errors.
84 */
85#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 86/*
87 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
88 * is 1000 KB/sec, so the extra system load does not show up that much.
89 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 90 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 91 * subsystem is idle. There is also an 'absolute maximum' reconstruction
92 * speed limit - in case reconstruction slows down your system despite
93 * idle IO detection.
94 *
95 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 96 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 97 */
98
99static int sysctl_speed_limit_min = 1000;
100static int sysctl_speed_limit_max = 200000;
fd01b88c 101static inline int speed_min(struct mddev *mddev)
88202a0c
N		 102{
103 return mddev->sync_speed_min ?
104 mddev->sync_speed_min : sysctl_speed_limit_min;
105}
106
fd01b88c 107static inline int speed_max(struct mddev *mddev)
88202a0c
N		 108{
109 return mddev->sync_speed_max ?
110 mddev->sync_speed_max : sysctl_speed_limit_max;
111}
1da177e4
LT		 112
113static struct ctl_table_header *raid_table_header;
114
115static ctl_table raid_table[] = {
116 {
1da177e4
LT		 117 .procname = "speed_limit_min",
118 .data = &sysctl_speed_limit_min,
119 .maxlen = sizeof(int),
80ca3a44 120 .mode = S_IRUGO|S_IWUSR,
6d456111 121 .proc_handler = proc_dointvec,
1da177e4
LT		 122 },
123 {
1da177e4
LT		 124 .procname = "speed_limit_max",
125 .data = &sysctl_speed_limit_max,
126 .maxlen = sizeof(int),
80ca3a44 127 .mode = S_IRUGO|S_IWUSR,
6d456111 128 .proc_handler = proc_dointvec,
1da177e4 129 },
894d2491 130 { }
1da177e4
LT		 131};
132
133static ctl_table raid_dir_table[] = {
134 {
1da177e4
LT		 135 .procname = "raid",
136 .maxlen = 0,
80ca3a44 137 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 138 .child = raid_table,
139 },
894d2491 140 { }
1da177e4
LT		 141};
142
143static ctl_table raid_root_table[] = {
144 {
1da177e4
LT		 145 .procname = "dev",
146 .maxlen = 0,
147 .mode = 0555,
148 .child = raid_dir_table,
149 },
894d2491 150 { }
1da177e4
LT		 151};
152
83d5cde4 153static const struct block_device_operations md_fops;
1da177e4 154
f91de92e
N		 155static int start_readonly;
156
a167f663
N		 157/* bio_clone_mddev
158 * like bio_clone, but with a local bio set
159 */
160
a167f663 161struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 162 struct mddev *mddev)
a167f663
N		 163{
164 struct bio *b;
a167f663
N		 165
166 if (!mddev || !mddev->bio_set)
167 return bio_alloc(gfp_mask, nr_iovecs);
168
395c72a7 169 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
a167f663
N		 170 if (!b)
171 return NULL;
a167f663
N		 172 return b;
173}
174EXPORT_SYMBOL_GPL(bio_alloc_mddev);
175
176struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
fd01b88c 177 struct mddev *mddev)
a167f663 178{
a167f663
N		 179 if (!mddev || !mddev->bio_set)
180 return bio_clone(bio, gfp_mask);
181
bf800ef1 182 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
a167f663
N		 183}
184EXPORT_SYMBOL_GPL(bio_clone_mddev);
185
d2eb35ac
N		 186void md_trim_bio(struct bio *bio, int offset, int size)
187{
188 /* 'bio' is a cloned bio which we need to trim to match
189 * the given offset and size.
190 * This requires adjusting bi_sector, bi_size, and bi_io_vec
191 */
192 int i;
193 struct bio_vec *bvec;
194 int sofar = 0;
195
196 size <<= 9;
197 if (offset == 0 && size == bio->bi_size)
198 return;
199
d2eb35ac
N		 200 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
201
fb9e3534
KO		 202 bio_advance(bio, offset << 9);
203
204 bio->bi_size = size;
205
d2eb35ac
N		 206 /* avoid any complications with bi_idx being non-zero*/
207 if (bio->bi_idx) {
208 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
209 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
210 bio->bi_vcnt -= bio->bi_idx;
211 bio->bi_idx = 0;
212 }
213 /* Make sure vcnt and last bv are not too big */
214 bio_for_each_segment(bvec, bio, i) {
215 if (sofar + bvec->bv_len > size)
216 bvec->bv_len = size - sofar;
217 if (bvec->bv_len == 0) {
218 bio->bi_vcnt = i;
219 break;
220 }
221 sofar += bvec->bv_len;
222 }
223}
224EXPORT_SYMBOL_GPL(md_trim_bio);
225
d7603b7e
N		 226/*
227 * We have a system wide 'event count' that is incremented
228 * on any 'interesting' event, and readers of /proc/mdstat
229 * can use 'poll' or 'select' to find out when the event
230 * count increases.
231 *
232 * Events are:
233 * start array, stop array, error, add device, remove device,
234 * start build, activate spare
235 */
2989ddbd 236static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 237static atomic_t md_event_count;
fd01b88c 238void md_new_event(struct mddev *mddev)
d7603b7e
N		 239{
240 atomic_inc(&md_event_count);
241 wake_up(&md_event_waiters);
242}
29269553 243EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 244
c331eb04
N		 245/* Alternate version that can be called from interrupts
246 * when calling sysfs_notify isn't needed.
247 */
fd01b88c 248static void md_new_event_inintr(struct mddev *mddev)
c331eb04
N		 249{
250 atomic_inc(&md_event_count);
251 wake_up(&md_event_waiters);
252}
253
1da177e4
LT		 254/*
255 * Enables to iterate over all existing md arrays
256 * all_mddevs_lock protects this list.
257 */
258static LIST_HEAD(all_mddevs);
259static DEFINE_SPINLOCK(all_mddevs_lock);
260
261
262/*
263 * iterates through all used mddevs in the system.
264 * We take care to grab the all_mddevs_lock whenever navigating
265 * the list, and to always hold a refcount when unlocked.
266 * Any code which breaks out of this loop while own
267 * a reference to the current mddev and must mddev_put it.
268 */
fd01b88c 269#define for_each_mddev(_mddev,_tmp) \
1da177e4
LT		 270 \
271 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N		 272 _tmp = all_mddevs.next; \
273 _mddev = NULL;}); \
274 ({ if (_tmp != &all_mddevs) \
275 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 276 spin_unlock(&all_mddevs_lock); \
fd01b88c
N		 277 if (_mddev) mddev_put(_mddev); \
278 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
279 _tmp != &all_mddevs;}); \
1da177e4 280 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 281 _tmp = _tmp->next;}) \
1da177e4
LT		 282 )
283
284
409c57f3
N		 285/* Rather than calling directly into the personality make_request function,
286 * IO requests come here first so that we can check if the device is
287 * being suspended pending a reconfiguration.
288 * We hold a refcount over the call to ->make_request. By the time that
289 * call has finished, the bio has been linked into some internal structure
290 * and so is visible to ->quiesce(), so we don't need the refcount any more.
291 */
5a7bbad2 292static void md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 293{
49077326 294 const int rw = bio_data_dir(bio);
fd01b88c 295 struct mddev *mddev = q->queuedata;
49077326 296 int cpu;
e91ece55 297 unsigned int sectors;
49077326 298
0ca69886
N		 299 if (mddev == NULL || mddev->pers == NULL
300 || !mddev->ready) {
409c57f3 301 bio_io_error(bio);
5a7bbad2 302 return;
409c57f3 303 }
bbfa57c0
SR		 304 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
305 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
306 return;
307 }
0ca69886 308 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 309 rcu_read_lock();
e9c7469b 310 if (mddev->suspended) {
409c57f3
N		 311 DEFINE_WAIT(__wait);
312 for (;;) {
313 prepare_to_wait(&mddev->sb_wait, &__wait,
314 TASK_UNINTERRUPTIBLE);
e9c7469b 315 if (!mddev->suspended)
409c57f3
N		 316 break;
317 rcu_read_unlock();
318 schedule();
319 rcu_read_lock();
320 }
321 finish_wait(&mddev->sb_wait, &__wait);
322 }
323 atomic_inc(&mddev->active_io);
324 rcu_read_unlock();
49077326 325
e91ece55
CM		 326 /*
327 * save the sectors now since our bio can
328 * go away inside make_request
329 */
330 sectors = bio_sectors(bio);
5a7bbad2 331 mddev->pers->make_request(mddev, bio);
49077326
N		 332
333 cpu = part_stat_lock();
334 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
e91ece55 335 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
49077326
N		 336 part_stat_unlock();
337
409c57f3
N		 338 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
339 wake_up(&mddev->sb_wait);
409c57f3
N		 340}
341
9e35b99c
N		 342/* mddev_suspend makes sure no new requests are submitted
343 * to the device, and that any requests that have been submitted
344 * are completely handled.
345 * Once ->stop is called and completes, the module will be completely
346 * unused.
347 */
fd01b88c 348void mddev_suspend(struct mddev *mddev)
409c57f3
N		 349{
350 BUG_ON(mddev->suspended);
351 mddev->suspended = 1;
352 synchronize_rcu();
353 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
354 mddev->pers->quiesce(mddev, 1);
0d9f4f13
JB		 355
356 del_timer_sync(&mddev->safemode_timer);
409c57f3 357}
390ee602 358EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 359
fd01b88c 360void mddev_resume(struct mddev *mddev)
409c57f3
N		 361{
362 mddev->suspended = 0;
363 wake_up(&mddev->sb_wait);
364 mddev->pers->quiesce(mddev, 0);
0fd018af 365
47525e59 366 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
0fd018af
JB		 367 md_wakeup_thread(mddev->thread);
368 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 369}
390ee602 370EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 371
fd01b88c 372int mddev_congested(struct mddev *mddev, int bits)
3fa841d7
N		 373{
374 return mddev->suspended;
375}
376EXPORT_SYMBOL(mddev_congested);
377
a2826aa9 378/*
e9c7469b 379 * Generic flush handling for md
a2826aa9
N		 380 */
381
e9c7469b 382static void md_end_flush(struct bio *bio, int err)
a2826aa9 383{
3cb03002 384 struct md_rdev *rdev = bio->bi_private;
fd01b88c 385 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 386
387 rdev_dec_pending(rdev, mddev);
388
389 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 390 /* The pre-request flush has finished */
e804ac78 391 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 392 }
393 bio_put(bio);
394}
395
a7a07e69
N		 396static void md_submit_flush_data(struct work_struct *ws);
397
a035fc3e 398static void submit_flushes(struct work_struct *ws)
a2826aa9 399{
fd01b88c 400 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 401 struct md_rdev *rdev;
a2826aa9 402
a7a07e69
N		 403 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
404 atomic_set(&mddev->flush_pending, 1);
a2826aa9 405 rcu_read_lock();
dafb20fa 406 rdev_for_each_rcu(rdev, mddev)
a2826aa9
N		 407 if (rdev->raid_disk >= 0 &&
408 !test_bit(Faulty, &rdev->flags)) {
409 /* Take two references, one is dropped
410 * when request finishes, one after
411 * we reclaim rcu_read_lock
412 */
413 struct bio *bi;
414 atomic_inc(&rdev->nr_pending);
415 atomic_inc(&rdev->nr_pending);
416 rcu_read_unlock();
b5e1b8ce 417 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
e9c7469b 418 bi->bi_end_io = md_end_flush;
a2826aa9
N		 419 bi->bi_private = rdev;
420 bi->bi_bdev = rdev->bdev;
421 atomic_inc(&mddev->flush_pending);
e9c7469b 422 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 423 rcu_read_lock();
424 rdev_dec_pending(rdev, mddev);
425 }
426 rcu_read_unlock();
a7a07e69
N		 427 if (atomic_dec_and_test(&mddev->flush_pending))
428 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 429}
430
e9c7469b 431static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 432{
fd01b88c 433 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 434 struct bio *bio = mddev->flush_bio;
a2826aa9 435
e9c7469b 436 if (bio->bi_size == 0)
a2826aa9
N		 437 /* an empty barrier - all done */
438 bio_endio(bio, 0);
439 else {
e9c7469b 440 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 441 mddev->pers->make_request(mddev, bio);
a2826aa9 442 }
2b74e12e
N		 443
444 mddev->flush_bio = NULL;
445 wake_up(&mddev->sb_wait);
a2826aa9
N		 446}
447
fd01b88c 448void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9
N		 449{
450 spin_lock_irq(&mddev->write_lock);
451 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 452 !mddev->flush_bio,
eed8c02e 453 mddev->write_lock);
e9c7469b 454 mddev->flush_bio = bio;
a2826aa9
N		 455 spin_unlock_irq(&mddev->write_lock);
456
a035fc3e
N		 457 INIT_WORK(&mddev->flush_work, submit_flushes);
458 queue_work(md_wq, &mddev->flush_work);
a2826aa9 459}
e9c7469b 460EXPORT_SYMBOL(md_flush_request);
409c57f3 461
74018dc3 462void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
97658cdd 463{
9cbb1750
N		 464 struct mddev *mddev = cb->data;
465 md_wakeup_thread(mddev->thread);
466 kfree(cb);
97658cdd 467}
9cbb1750 468EXPORT_SYMBOL(md_unplug);
2ac87401 469
fd01b88c 470static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 471{
472 atomic_inc(&mddev->active);
473 return mddev;
474}
475
5fd3a17e 476static void mddev_delayed_delete(struct work_struct *ws);
d3374825 477
fd01b88c 478static void mddev_put(struct mddev *mddev)
1da177e4 479{
a167f663
N		 480 struct bio_set *bs = NULL;
481
1da177e4
LT		 482 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
483 return;
d3374825 484 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 485 mddev->ctime == 0 && !mddev->hold_active) {
486 /* Array is not configured at all, and not held active,
487 * so destroy it */
af8a2434 488 list_del_init(&mddev->all_mddevs);
a167f663
N		 489 bs = mddev->bio_set;
490 mddev->bio_set = NULL;
d3374825 491 if (mddev->gendisk) {
e804ac78
TH		 492 /* We did a probe so need to clean up. Call
493 * queue_work inside the spinlock so that
494 * flush_workqueue() after mddev_find will
495 * succeed in waiting for the work to be done.
d3374825
N		 496 */
497 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 498 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 499 } else
500 kfree(mddev);
501 }
502 spin_unlock(&all_mddevs_lock);
a167f663
N		 503 if (bs)
504 bioset_free(bs);
1da177e4
LT		 505}
506
fd01b88c 507void mddev_init(struct mddev *mddev)
fafd7fb0
N		 508{
509 mutex_init(&mddev->open_mutex);
510 mutex_init(&mddev->reconfig_mutex);
511 mutex_init(&mddev->bitmap_info.mutex);
512 INIT_LIST_HEAD(&mddev->disks);
513 INIT_LIST_HEAD(&mddev->all_mddevs);
514 init_timer(&mddev->safemode_timer);
515 atomic_set(&mddev->active, 1);
516 atomic_set(&mddev->openers, 0);
517 atomic_set(&mddev->active_io, 0);
518 spin_lock_init(&mddev->write_lock);
519 atomic_set(&mddev->flush_pending, 0);
520 init_waitqueue_head(&mddev->sb_wait);
521 init_waitqueue_head(&mddev->recovery_wait);
522 mddev->reshape_position = MaxSector;
2c810cdd 523 mddev->reshape_backwards = 0;
fafd7fb0
N		 524 mddev->resync_min = 0;
525 mddev->resync_max = MaxSector;
526 mddev->level = LEVEL_NONE;
527}
390ee602 528EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 529
fd01b88c 530static struct mddev * mddev_find(dev_t unit)
1da177e4 531{
fd01b88c 532 struct mddev *mddev, *new = NULL;
1da177e4 533
8f5f02c4
N		 534 if (unit && MAJOR(unit) != MD_MAJOR)
535 unit &= ~((1<<MdpMinorShift)-1);
536
1da177e4
LT		 537 retry:
538 spin_lock(&all_mddevs_lock);
efeb53c0
N		 539
540 if (unit) {
541 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
542 if (mddev->unit == unit) {
543 mddev_get(mddev);
544 spin_unlock(&all_mddevs_lock);
545 kfree(new);
546 return mddev;
547 }
548
549 if (new) {
550 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 551 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 552 new->hold_active = UNTIL_IOCTL;
553 return new;
1da177e4 554 }
efeb53c0
N		 555 } else if (new) {
556 /* find an unused unit number */
557 static int next_minor = 512;
558 int start = next_minor;
559 int is_free = 0;
560 int dev = 0;
561 while (!is_free) {
562 dev = MKDEV(MD_MAJOR, next_minor);
563 next_minor++;
564 if (next_minor > MINORMASK)
565 next_minor = 0;
566 if (next_minor == start) {
567 /* Oh dear, all in use. */
568 spin_unlock(&all_mddevs_lock);
569 kfree(new);
570 return NULL;
571 }
572
573 is_free = 1;
574 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
575 if (mddev->unit == dev) {
576 is_free = 0;
577 break;
578 }
579 }
580 new->unit = dev;
581 new->md_minor = MINOR(dev);
582 new->hold_active = UNTIL_STOP;
1da177e4
LT		 583 list_add(&new->all_mddevs, &all_mddevs);
584 spin_unlock(&all_mddevs_lock);
585 return new;
586 }
587 spin_unlock(&all_mddevs_lock);
588
9ffae0cf 589 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 590 if (!new)
591 return NULL;
592
1da177e4
LT		 593 new->unit = unit;
594 if (MAJOR(unit) == MD_MAJOR)
595 new->md_minor = MINOR(unit);
596 else
597 new->md_minor = MINOR(unit) >> MdpMinorShift;
598
fafd7fb0 599 mddev_init(new);
1da177e4 600
1da177e4
LT		 601 goto retry;
602}
603
fd01b88c 604static inline int mddev_lock(struct mddev * mddev)
1da177e4 605{
df5b89b3 606 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 607}
608
fd01b88c 609static inline int mddev_is_locked(struct mddev *mddev)
b522adcd
DW		 610{
611 return mutex_is_locked(&mddev->reconfig_mutex);
612}
613
fd01b88c 614static inline int mddev_trylock(struct mddev * mddev)
1da177e4 615{
df5b89b3 616 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 617}
618
b6eb127d
N		 619static struct attribute_group md_redundancy_group;
620
fd01b88c 621static void mddev_unlock(struct mddev * mddev)
1da177e4 622{
a64c876f 623 if (mddev->to_remove) {
b6eb127d
N		 624 /* These cannot be removed under reconfig_mutex as
625 * an access to the files will try to take reconfig_mutex
626 * while holding the file unremovable, which leads to
627 * a deadlock.
bb4f1e9d
N		 628 * So hold set sysfs_active while the remove in happeing,
629 * and anything else which might set ->to_remove or my
630 * otherwise change the sysfs namespace will fail with
631 * -EBUSY if sysfs_active is still set.
632 * We set sysfs_active under reconfig_mutex and elsewhere
633 * test it under the same mutex to ensure its correct value
634 * is seen.
b6eb127d 635 */
a64c876f
N		 636 struct attribute_group *to_remove = mddev->to_remove;
637 mddev->to_remove = NULL;
bb4f1e9d 638 mddev->sysfs_active = 1;
b6eb127d
N		 639 mutex_unlock(&mddev->reconfig_mutex);
640
00bcb4ac
N		 641 if (mddev->kobj.sd) {
642 if (to_remove != &md_redundancy_group)
643 sysfs_remove_group(&mddev->kobj, to_remove);
644 if (mddev->pers == NULL ||
645 mddev->pers->sync_request == NULL) {
646 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
647 if (mddev->sysfs_action)
648 sysfs_put(mddev->sysfs_action);
649 mddev->sysfs_action = NULL;
650 }
a64c876f 651 }
bb4f1e9d 652 mddev->sysfs_active = 0;
b6eb127d
N		 653 } else
654 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 655
751e67ca
CD		 656 /* As we've dropped the mutex we need a spinlock to
657 * make sure the thread doesn't disappear
01f96c0a
N		 658 */
659 spin_lock(&pers_lock);
005eca5e 660 md_wakeup_thread(mddev->thread);
01f96c0a 661 spin_unlock(&pers_lock);
1da177e4
LT		 662}
663
fd01b88c 664static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
1da177e4 665{
3cb03002 666 struct md_rdev *rdev;
1da177e4 667
dafb20fa 668 rdev_for_each(rdev, mddev)
1da177e4
LT		 669 if (rdev->desc_nr == nr)
670 return rdev;
159ec1fc 671
1da177e4
LT		 672 return NULL;
673}
674
1ca69c4b
N		 675static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
676{
677 struct md_rdev *rdev;
678
679 rdev_for_each_rcu(rdev, mddev)
680 if (rdev->desc_nr == nr)
681 return rdev;
682
683 return NULL;
684}
685
686static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
1da177e4 687{
3cb03002 688 struct md_rdev *rdev;
1da177e4 689
dafb20fa 690 rdev_for_each(rdev, mddev)
1da177e4
LT		 691 if (rdev->bdev->bd_dev == dev)
692 return rdev;
159ec1fc 693
1da177e4
LT		 694 return NULL;
695}
696
1ca69c4b
N		 697static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
698{
699 struct md_rdev *rdev;
700
701 rdev_for_each_rcu(rdev, mddev)
702 if (rdev->bdev->bd_dev == dev)
703 return rdev;
704
705 return NULL;
706}
707
84fc4b56 708static struct md_personality *find_pers(int level, char *clevel)
2604b703 709{
84fc4b56 710 struct md_personality *pers;
d9d166c2
N		 711 list_for_each_entry(pers, &pers_list, list) {
712 if (level != LEVEL_NONE && pers->level == level)
2604b703 713 return pers;
d9d166c2
N		 714 if (strcmp(pers->name, clevel)==0)
715 return pers;
716 }
2604b703
N		 717 return NULL;
718}
719
b73df2d3 720/* return the offset of the super block in 512byte sectors */
3cb03002 721static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 722{
57b2caa3 723 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 724 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 725}
726
3cb03002 727static int alloc_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 728{
729 if (rdev->sb_page)
730 MD_BUG();
731
732 rdev->sb_page = alloc_page(GFP_KERNEL);
733 if (!rdev->sb_page) {
734 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 735 return -ENOMEM;
1da177e4
LT		 736 }
737
738 return 0;
739}
740
545c8795 741void md_rdev_clear(struct md_rdev *rdev)
1da177e4
LT		 742{
743 if (rdev->sb_page) {
2d1f3b5d 744 put_page(rdev->sb_page);
1da177e4
LT		 745 rdev->sb_loaded = 0;
746 rdev->sb_page = NULL;
0f420358 747 rdev->sb_start = 0;
dd8ac336 748 rdev->sectors = 0;
1da177e4 749 }
2699b672
N		 750 if (rdev->bb_page) {
751 put_page(rdev->bb_page);
752 rdev->bb_page = NULL;
753 }
4fa2f327
N		 754 kfree(rdev->badblocks.page);
755 rdev->badblocks.page = NULL;
1da177e4 756}
545c8795 757EXPORT_SYMBOL_GPL(md_rdev_clear);
1da177e4 758
6712ecf8 759static void super_written(struct bio *bio, int error)
7bfa19f2 760{
3cb03002 761 struct md_rdev *rdev = bio->bi_private;
fd01b88c 762 struct mddev *mddev = rdev->mddev;
7bfa19f2 763
3a0f5bbb
N		 764 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
765 printk("md: super_written gets error=%d, uptodate=%d\n",
766 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
767 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 768 md_error(mddev, rdev);
3a0f5bbb 769 }
7bfa19f2 770
a9701a30
N		 771 if (atomic_dec_and_test(&mddev->pending_writes))
772 wake_up(&mddev->sb_wait);
f8b58edf 773 bio_put(bio);
7bfa19f2
N		 774}
775
fd01b88c 776void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 777 sector_t sector, int size, struct page *page)
778{
779 /* write first size bytes of page to sector of rdev
780 * Increment mddev->pending_writes before returning
781 * and decrement it on completion, waking up sb_wait
782 * if zero is reached.
783 * If an error occurred, call md_error
784 */
a167f663 785 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 786
a6ff7e08 787 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
7bfa19f2
N		 788 bio->bi_sector = sector;
789 bio_add_page(bio, page, size, 0);
790 bio->bi_private = rdev;
791 bio->bi_end_io = super_written;
a9701a30 792
7bfa19f2 793 atomic_inc(&mddev->pending_writes);
a5bf4df0 794 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 795}
796
fd01b88c 797void md_super_wait(struct mddev *mddev)
a9701a30 798{
e9c7469b 799 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N		 800 DEFINE_WAIT(wq);
801 for(;;) {
802 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
803 if (atomic_read(&mddev->pending_writes)==0)
804 break;
a9701a30
N		 805 schedule();
806 }
807 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 808}
809
6712ecf8 810static void bi_complete(struct bio *bio, int error)
1da177e4 811{
1da177e4 812 complete((struct completion*)bio->bi_private);
1da177e4
LT		 813}
814
3cb03002 815int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 816 struct page *page, int rw, bool metadata_op)
1da177e4 817{
a167f663 818 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 819 struct completion event;
820 int ret;
821
721a9602 822 rw |= REQ_SYNC;
1da177e4 823
a6ff7e08
JB		 824 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
825 rdev->meta_bdev : rdev->bdev;
ccebd4c4
JB		 826 if (metadata_op)
827 bio->bi_sector = sector + rdev->sb_start;
1fdd6fc9
N		 828 else if (rdev->mddev->reshape_position != MaxSector &&
829 (rdev->mddev->reshape_backwards ==
830 (sector >= rdev->mddev->reshape_position)))
831 bio->bi_sector = sector + rdev->new_data_offset;
ccebd4c4
JB		 832 else
833 bio->bi_sector = sector + rdev->data_offset;
1da177e4
LT		 834 bio_add_page(bio, page, size, 0);
835 init_completion(&event);
836 bio->bi_private = &event;
837 bio->bi_end_io = bi_complete;
838 submit_bio(rw, bio);
839 wait_for_completion(&event);
840
841 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
842 bio_put(bio);
843 return ret;
844}
a8745db2 845EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 846
3cb03002 847static int read_disk_sb(struct md_rdev * rdev, int size)
1da177e4
LT		 848{
849 char b[BDEVNAME_SIZE];
850 if (!rdev->sb_page) {
851 MD_BUG();
852 return -EINVAL;
853 }
854 if (rdev->sb_loaded)
855 return 0;
856
857
ccebd4c4 858 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 859 goto fail;
860 rdev->sb_loaded = 1;
861 return 0;
862
863fail:
864 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
865 bdevname(rdev->bdev,b));
866 return -EINVAL;
867}
868
869static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
870{
05710466
AN		 871 return sb1->set_uuid0 == sb2->set_uuid0 &&
872 sb1->set_uuid1 == sb2->set_uuid1 &&
873 sb1->set_uuid2 == sb2->set_uuid2 &&
874 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 875}
876
1da177e4
LT		 877static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
878{
879 int ret;
880 mdp_super_t *tmp1, *tmp2;
881
882 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
883 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
884
885 if (!tmp1 || !tmp2) {
886 ret = 0;
35020f1a 887 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 888 goto abort;
889 }
890
891 *tmp1 = *sb1;
892 *tmp2 = *sb2;
893
894 /*
895 * nr_disks is not constant
896 */
897 tmp1->nr_disks = 0;
898 tmp2->nr_disks = 0;
899
ce0c8e05 900 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 901abort:
990a8baf
JJ		 902 kfree(tmp1);
903 kfree(tmp2);
1da177e4
LT		 904 return ret;
905}
906
4d167f09
N		 907
908static u32 md_csum_fold(u32 csum)
909{
910 csum = (csum & 0xffff) + (csum >> 16);
911 return (csum & 0xffff) + (csum >> 16);
912}
913
1da177e4
LT		 914static unsigned int calc_sb_csum(mdp_super_t * sb)
915{
4d167f09
N		 916 u64 newcsum = 0;
917 u32 *sb32 = (u32*)sb;
918 int i;
1da177e4
LT		 919 unsigned int disk_csum, csum;
920
921 disk_csum = sb->sb_csum;
922 sb->sb_csum = 0;
4d167f09
N		 923
924 for (i = 0; i < MD_SB_BYTES/4 ; i++)
925 newcsum += sb32[i];
926 csum = (newcsum & 0xffffffff) + (newcsum>>32);
927
928
929#ifdef CONFIG_ALPHA
930 /* This used to use csum_partial, which was wrong for several
931 * reasons including that different results are returned on
932 * different architectures. It isn't critical that we get exactly
933 * the same return value as before (we always csum_fold before
934 * testing, and that removes any differences). However as we
935 * know that csum_partial always returned a 16bit value on
936 * alphas, do a fold to maximise conformity to previous behaviour.
937 */
938 sb->sb_csum = md_csum_fold(disk_csum);
939#else
1da177e4 940 sb->sb_csum = disk_csum;
4d167f09 941#endif
1da177e4
LT		 942 return csum;
943}
944
945
946/*
947 * Handle superblock details.
948 * We want to be able to handle multiple superblock formats
949 * so we have a common interface to them all, and an array of
950 * different handlers.
951 * We rely on user-space to write the initial superblock, and support
952 * reading and updating of superblocks.
953 * Interface methods are:
3cb03002 954 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 955 * loads and validates a superblock on dev.
956 * if refdev != NULL, compare superblocks on both devices
957 * Return:
958 * 0 - dev has a superblock that is compatible with refdev
959 * 1 - dev has a superblock that is compatible and newer than refdev
960 * so dev should be used as the refdev in future
961 * -EINVAL superblock incompatible or invalid
962 * -othererror e.g. -EIO
963 *
fd01b88c 964 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 965 * Verify that dev is acceptable into mddev.
966 * The first time, mddev->raid_disks will be 0, and data from
967 * dev should be merged in. Subsequent calls check that dev
968 * is new enough. Return 0 or -EINVAL
969 *
fd01b88c 970 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 971 * Update the superblock for rdev with data in mddev
972 * This does not write to disc.
973 *
974 */
975
976struct super_type {
0cd17fec
CW		 977 char *name;
978 struct module *owner;
c6563a8c
N		 979 int (*load_super)(struct md_rdev *rdev,
980 struct md_rdev *refdev,
0cd17fec 981 int minor_version);
c6563a8c
N		 982 int (*validate_super)(struct mddev *mddev,
983 struct md_rdev *rdev);
984 void (*sync_super)(struct mddev *mddev,
985 struct md_rdev *rdev);
3cb03002 986 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 987 sector_t num_sectors);
c6563a8c
N		 988 int (*allow_new_offset)(struct md_rdev *rdev,
989 unsigned long long new_offset);
1da177e4
LT		 990};
991
0894cc30
AN		 992/*
993 * Check that the given mddev has no bitmap.
994 *
995 * This function is called from the run method of all personalities that do not
996 * support bitmaps. It prints an error message and returns non-zero if mddev
997 * has a bitmap. Otherwise, it returns 0.
998 *
999 */
fd01b88c 1000int md_check_no_bitmap(struct mddev *mddev)
0894cc30 1001{
c3d9714e 1002 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 1003 return 0;
1004 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1005 mdname(mddev), mddev->pers->name);
1006 return 1;
1007}
1008EXPORT_SYMBOL(md_check_no_bitmap);
1009
1da177e4
LT		 1010/*
1011 * load_super for 0.90.0
1012 */
3cb03002 1013static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1014{
1015 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1016 mdp_super_t *sb;
1017 int ret;
1da177e4
LT		 1018
1019 /*
0f420358 1020 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 1021 * it's at the end of the disk.
1022 *
1023 * It also happens to be a multiple of 4Kb.
1024 */
57b2caa3 1025 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 1026
0002b271 1027 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 1028 if (ret) return ret;
1029
1030 ret = -EINVAL;
1031
1032 bdevname(rdev->bdev, b);
65a06f06 1033 sb = page_address(rdev->sb_page);
1da177e4
LT		 1034
1035 if (sb->md_magic != MD_SB_MAGIC) {
1036 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1037 b);
1038 goto abort;
1039 }
1040
1041 if (sb->major_version != 0 ||
f6705578
N		 1042 sb->minor_version < 90 ||
1043 sb->minor_version > 91) {
1da177e4
LT		 1044 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1045 sb->major_version, sb->minor_version,
1046 b);
1047 goto abort;
1048 }
1049
1050 if (sb->raid_disks <= 0)
1051 goto abort;
1052
4d167f09 1053 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1054 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1055 b);
1056 goto abort;
1057 }
1058
1059 rdev->preferred_minor = sb->md_minor;
1060 rdev->data_offset = 0;
c6563a8c 1061 rdev->new_data_offset = 0;
0002b271 1062 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1063 rdev->badblocks.shift = -1;
1da177e4
LT		 1064
1065 if (sb->level == LEVEL_MULTIPATH)
1066 rdev->desc_nr = -1;
1067 else
1068 rdev->desc_nr = sb->this_disk.number;
1069
9a7b2b0f 1070 if (!refdev) {
1da177e4 1071 ret = 1;
9a7b2b0f 1072 } else {
1da177e4 1073 __u64 ev1, ev2;
65a06f06 1074 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1075 if (!uuid_equal(refsb, sb)) {
1076 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1077 b, bdevname(refdev->bdev,b2));
1078 goto abort;
1079 }
1080 if (!sb_equal(refsb, sb)) {
1081 printk(KERN_WARNING "md: %s has same UUID"
1082 " but different superblock to %s\n",
1083 b, bdevname(refdev->bdev, b2));
1084 goto abort;
1085 }
1086 ev1 = md_event(sb);
1087 ev2 = md_event(refsb);
1088 if (ev1 > ev2)
1089 ret = 1;
1090 else
1091 ret = 0;
1092 }
8190e754 1093 rdev->sectors = rdev->sb_start;
667a5313
N		 1094 /* Limit to 4TB as metadata cannot record more than that.
1095 * (not needed for Linear and RAID0 as metadata doesn't
1096 * record this size)
1097 */
1098 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
27a7b260 1099 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1100
27a7b260 1101 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1102 /* "this cannot possibly happen" ... */
1103 ret = -EINVAL;
1104
1da177e4
LT		 1105 abort:
1106 return ret;
1107}
1108
1109/*
1110 * validate_super for 0.90.0
1111 */
fd01b88c 1112static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1113{
1114 mdp_disk_t *desc;
65a06f06 1115 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1116 __u64 ev1 = md_event(sb);
1da177e4 1117
41158c7e 1118 rdev->raid_disk = -1;
c5d79adb
N		 1119 clear_bit(Faulty, &rdev->flags);
1120 clear_bit(In_sync, &rdev->flags);
bb4a65df 1121 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1122 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1123
1da177e4
LT		 1124 if (mddev->raid_disks == 0) {
1125 mddev->major_version = 0;
1126 mddev->minor_version = sb->minor_version;
1127 mddev->patch_version = sb->patch_version;
e691063a 1128 mddev->external = 0;
9d8f0363 1129 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1130 mddev->ctime = sb->ctime;
1131 mddev->utime = sb->utime;
1132 mddev->level = sb->level;
d9d166c2 1133 mddev->clevel[0] = 0;
1da177e4
LT		 1134 mddev->layout = sb->layout;
1135 mddev->raid_disks = sb->raid_disks;
27a7b260 1136 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1137 mddev->events = ev1;
c3d9714e 1138 mddev->bitmap_info.offset = 0;
6409bb05
N		 1139 mddev->bitmap_info.space = 0;
1140 /* bitmap can use 60 K after the 4K superblocks */
c3d9714e 1141 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 1142 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
2c810cdd 1143 mddev->reshape_backwards = 0;
1da177e4 1144
f6705578
N		 1145 if (mddev->minor_version >= 91) {
1146 mddev->reshape_position = sb->reshape_position;
1147 mddev->delta_disks = sb->delta_disks;
1148 mddev->new_level = sb->new_level;
1149 mddev->new_layout = sb->new_layout;
664e7c41 1150 mddev->new_chunk_sectors = sb->new_chunk >> 9;
2c810cdd
N		 1151 if (mddev->delta_disks < 0)
1152 mddev->reshape_backwards = 1;
f6705578
N		 1153 } else {
1154 mddev->reshape_position = MaxSector;
1155 mddev->delta_disks = 0;
1156 mddev->new_level = mddev->level;
1157 mddev->new_layout = mddev->layout;
664e7c41 1158 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1159 }
1160
1da177e4
LT		 1161 if (sb->state & (1<<MD_SB_CLEAN))
1162 mddev->recovery_cp = MaxSector;
1163 else {
1164 if (sb->events_hi == sb->cp_events_hi &&
1165 sb->events_lo == sb->cp_events_lo) {
1166 mddev->recovery_cp = sb->recovery_cp;
1167 } else
1168 mddev->recovery_cp = 0;
1169 }
1170
1171 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1172 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1173 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1174 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1175
1176 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1177
1178 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
6409bb05 1179 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1180 mddev->bitmap_info.offset =
1181 mddev->bitmap_info.default_offset;
6409bb05
N		 1182 mddev->bitmap_info.space =
1183 mddev->bitmap_info.space;
1184 }
a654b9d8 1185
41158c7e 1186 } else if (mddev->pers == NULL) {
be6800a7
N		 1187 /* Insist on good event counter while assembling, except
1188 * for spares (which don't need an event count) */
1da177e4 1189 ++ev1;
be6800a7
N		 1190 if (sb->disks[rdev->desc_nr].state & (
1191 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1192 if (ev1 < mddev->events)
1193 return -EINVAL;
41158c7e
N		 1194 } else if (mddev->bitmap) {
1195 /* if adding to array with a bitmap, then we can accept an
1196 * older device ... but not too old.
1197 */
41158c7e
N		 1198 if (ev1 < mddev->bitmap->events_cleared)
1199 return 0;
bb4a65df
N		 1200 if (ev1 < mddev->events)
1201 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1202 } else {
1203 if (ev1 < mddev->events)
1204 /* just a hot-add of a new device, leave raid_disk at -1 */
1205 return 0;
1206 }
41158c7e 1207
1da177e4 1208 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1209 desc = sb->disks + rdev->desc_nr;
1210
1211 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1212 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1213 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1214 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1215 set_bit(In_sync, &rdev->flags);
1da177e4 1216 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1217 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1218 /* active but not in sync implies recovery up to
1219 * reshape position. We don't know exactly where
1220 * that is, so set to zero for now */
1221 if (mddev->minor_version >= 91) {
1222 rdev->recovery_offset = 0;
1223 rdev->raid_disk = desc->raid_disk;
1224 }
1da177e4 1225 }
8ddf9efe
N		 1226 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1227 set_bit(WriteMostly, &rdev->flags);
41158c7e 1228 } else /* MULTIPATH are always insync */
b2d444d7 1229 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1230 return 0;
1231}
1232
1233/*
1234 * sync_super for 0.90.0
1235 */
fd01b88c 1236static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1237{
1238 mdp_super_t *sb;
3cb03002 1239 struct md_rdev *rdev2;
1da177e4 1240 int next_spare = mddev->raid_disks;
19133a42 1241
1da177e4
LT		 1242
1243 /* make rdev->sb match mddev data..
1244 *
1245 * 1/ zero out disks
1246 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1247 * 3/ any empty disks < next_spare become removed
1248 *
1249 * disks[0] gets initialised to REMOVED because
1250 * we cannot be sure from other fields if it has
1251 * been initialised or not.
1252 */
1253 int i;
1254 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1255
61181565
N		 1256 rdev->sb_size = MD_SB_BYTES;
1257
65a06f06 1258 sb = page_address(rdev->sb_page);
1da177e4
LT		 1259
1260 memset(sb, 0, sizeof(*sb));
1261
1262 sb->md_magic = MD_SB_MAGIC;
1263 sb->major_version = mddev->major_version;
1da177e4
LT		 1264 sb->patch_version = mddev->patch_version;
1265 sb->gvalid_words = 0; /* ignored */
1266 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1267 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1268 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1269 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1270
1271 sb->ctime = mddev->ctime;
1272 sb->level = mddev->level;
58c0fed4 1273 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1274 sb->raid_disks = mddev->raid_disks;
1275 sb->md_minor = mddev->md_minor;
e691063a 1276 sb->not_persistent = 0;
1da177e4
LT		 1277 sb->utime = mddev->utime;
1278 sb->state = 0;
1279 sb->events_hi = (mddev->events>>32);
1280 sb->events_lo = (u32)mddev->events;
1281
f6705578
N		 1282 if (mddev->reshape_position == MaxSector)
1283 sb->minor_version = 90;
1284 else {
1285 sb->minor_version = 91;
1286 sb->reshape_position = mddev->reshape_position;
1287 sb->new_level = mddev->new_level;
1288 sb->delta_disks = mddev->delta_disks;
1289 sb->new_layout = mddev->new_layout;
664e7c41 1290 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1291 }
1292 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1293 if (mddev->in_sync)
1294 {
1295 sb->recovery_cp = mddev->recovery_cp;
1296 sb->cp_events_hi = (mddev->events>>32);
1297 sb->cp_events_lo = (u32)mddev->events;
1298 if (mddev->recovery_cp == MaxSector)
1299 sb->state = (1<< MD_SB_CLEAN);
1300 } else
1301 sb->recovery_cp = 0;
1302
1303 sb->layout = mddev->layout;
9d8f0363 1304 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1305
c3d9714e 1306 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1307 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1308
1da177e4 1309 sb->disks[0].state = (1<<MD_DISK_REMOVED);
dafb20fa 1310 rdev_for_each(rdev2, mddev) {
1da177e4 1311 mdp_disk_t *d;
86e6ffdd 1312 int desc_nr;
0261cd9f
N		 1313 int is_active = test_bit(In_sync, &rdev2->flags);
1314
1315 if (rdev2->raid_disk >= 0 &&
1316 sb->minor_version >= 91)
1317 /* we have nowhere to store the recovery_offset,
1318 * but if it is not below the reshape_position,
1319 * we can piggy-back on that.
1320 */
1321 is_active = 1;
1322 if (rdev2->raid_disk < 0 ||
1323 test_bit(Faulty, &rdev2->flags))
1324 is_active = 0;
1325 if (is_active)
86e6ffdd 1326 desc_nr = rdev2->raid_disk;
1da177e4 1327 else
86e6ffdd 1328 desc_nr = next_spare++;
19133a42 1329 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1330 d = &sb->disks[rdev2->desc_nr];
1331 nr_disks++;
1332 d->number = rdev2->desc_nr;
1333 d->major = MAJOR(rdev2->bdev->bd_dev);
1334 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1335 if (is_active)
1da177e4
LT		 1336 d->raid_disk = rdev2->raid_disk;
1337 else
1338 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1339 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1340 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1341 else if (is_active) {
1da177e4 1342 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1343 if (test_bit(In_sync, &rdev2->flags))
1344 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1345 active++;
1346 working++;
1347 } else {
1348 d->state = 0;
1349 spare++;
1350 working++;
1351 }
8ddf9efe
N		 1352 if (test_bit(WriteMostly, &rdev2->flags))
1353 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1354 }
1da177e4
LT		 1355 /* now set the "removed" and "faulty" bits on any missing devices */
1356 for (i=0 ; i < mddev->raid_disks ; i++) {
1357 mdp_disk_t *d = &sb->disks[i];
1358 if (d->state == 0 && d->number == 0) {
1359 d->number = i;
1360 d->raid_disk = i;
1361 d->state = (1<<MD_DISK_REMOVED);
1362 d->state |= (1<<MD_DISK_FAULTY);
1363 failed++;
1364 }
1365 }
1366 sb->nr_disks = nr_disks;
1367 sb->active_disks = active;
1368 sb->working_disks = working;
1369 sb->failed_disks = failed;
1370 sb->spare_disks = spare;
1371
1372 sb->this_disk = sb->disks[rdev->desc_nr];
1373 sb->sb_csum = calc_sb_csum(sb);
1374}
1375
0cd17fec
CW		 1376/*
1377 * rdev_size_change for 0.90.0
1378 */
1379static unsigned long long
3cb03002 1380super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1381{
58c0fed4 1382 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1383 return 0; /* component must fit device */
c3d9714e 1384 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1385 return 0; /* can't move bitmap */
57b2caa3 1386 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1387 if (!num_sectors || num_sectors > rdev->sb_start)
1388 num_sectors = rdev->sb_start;
27a7b260
N		 1389 /* Limit to 4TB as metadata cannot record more than that.
1390 * 4TB == 2^32 KB, or 2*2^32 sectors.
1391 */
667a5313 1392 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
27a7b260 1393 num_sectors = (2ULL << 32) - 2;
0f420358 1394 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1395 rdev->sb_page);
1396 md_super_wait(rdev->mddev);
c26a44ed 1397 return num_sectors;
0cd17fec
CW		 1398}
1399
c6563a8c
N		 1400static int
1401super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1402{
1403 /* non-zero offset changes not possible with v0.90 */
1404 return new_offset == 0;
1405}
0cd17fec 1406
1da177e4
LT		 1407/*
1408 * version 1 superblock
1409 */
1410
1c05b4bc 1411static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1412{
1c05b4bc
N		 1413 __le32 disk_csum;
1414 u32 csum;
1da177e4
LT		 1415 unsigned long long newcsum;
1416 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1417 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1418
1419 disk_csum = sb->sb_csum;
1420 sb->sb_csum = 0;
1421 newcsum = 0;
1f3c9907 1422 for (; size >= 4; size -= 4)
1da177e4
LT		 1423 newcsum += le32_to_cpu(*isuper++);
1424
1425 if (size == 2)
1c05b4bc 1426 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1427
1428 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1429 sb->sb_csum = disk_csum;
1430 return cpu_to_le32(csum);
1431}
1432
2699b672
N		 1433static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1434 int acknowledged);
3cb03002 1435static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1436{
1437 struct mdp_superblock_1 *sb;
1438 int ret;
0f420358 1439 sector_t sb_start;
c6563a8c 1440 sector_t sectors;
1da177e4 1441 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1442 int bmask;
1da177e4
LT		 1443
1444 /*
0f420358 1445 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1446 * It is always aligned to a 4K boundary and
1447 * depeding on minor_version, it can be:
1448 * 0: At least 8K, but less than 12K, from end of device
1449 * 1: At start of device
1450 * 2: 4K from start of device.
1451 */
1452 switch(minor_version) {
1453 case 0:
77304d2a 1454 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1455 sb_start -= 8*2;
1456 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1457 break;
1458 case 1:
0f420358 1459 sb_start = 0;
1da177e4
LT		 1460 break;
1461 case 2:
0f420358 1462 sb_start = 8;
1da177e4
LT		 1463 break;
1464 default:
1465 return -EINVAL;
1466 }
0f420358 1467 rdev->sb_start = sb_start;
1da177e4 1468
0002b271
N		 1469 /* superblock is rarely larger than 1K, but it can be larger,
1470 * and it is safe to read 4k, so we do that
1471 */
1472 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1473 if (ret) return ret;
1474
1475
65a06f06 1476 sb = page_address(rdev->sb_page);
1da177e4
LT		 1477
1478 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1479 sb->major_version != cpu_to_le32(1) ||
1480 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1481 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1482 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1483 return -EINVAL;
1484
1485 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1486 printk("md: invalid superblock checksum on %s\n",
1487 bdevname(rdev->bdev,b));
1488 return -EINVAL;
1489 }
1490 if (le64_to_cpu(sb->data_size) < 10) {
1491 printk("md: data_size too small on %s\n",
1492 bdevname(rdev->bdev,b));
1493 return -EINVAL;
1494 }
c6563a8c
N		 1495 if (sb->pad0 ||
1496 sb->pad3[0] ||
1497 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1498 /* Some padding is non-zero, might be a new feature */
1499 return -EINVAL;
e11e93fa 1500
1da177e4
LT		 1501 rdev->preferred_minor = 0xffff;
1502 rdev->data_offset = le64_to_cpu(sb->data_offset);
c6563a8c
N		 1503 rdev->new_data_offset = rdev->data_offset;
1504 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1505 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1506 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
4dbcdc75 1507 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1508
0002b271 1509 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1510 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1511 if (rdev->sb_size & bmask)
a1801f85
N		 1512 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1513
1514 if (minor_version
0f420358 1515 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1516 return -EINVAL;
c6563a8c
N		 1517 if (minor_version
1518 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1519 return -EINVAL;
0002b271 1520
31b65a0d
N		 1521 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1522 rdev->desc_nr = -1;
1523 else
1524 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1525
2699b672
N		 1526 if (!rdev->bb_page) {
1527 rdev->bb_page = alloc_page(GFP_KERNEL);
1528 if (!rdev->bb_page)
1529 return -ENOMEM;
1530 }
1531 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1532 rdev->badblocks.count == 0) {
1533 /* need to load the bad block list.
1534 * Currently we limit it to one page.
1535 */
1536 s32 offset;
1537 sector_t bb_sector;
1538 u64 *bbp;
1539 int i;
1540 int sectors = le16_to_cpu(sb->bblog_size);
1541 if (sectors > (PAGE_SIZE / 512))
1542 return -EINVAL;
1543 offset = le32_to_cpu(sb->bblog_offset);
1544 if (offset == 0)
1545 return -EINVAL;
1546 bb_sector = (long long)offset;
1547 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1548 rdev->bb_page, READ, true))
1549 return -EIO;
1550 bbp = (u64 *)page_address(rdev->bb_page);
1551 rdev->badblocks.shift = sb->bblog_shift;
1552 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1553 u64 bb = le64_to_cpu(*bbp);
1554 int count = bb & (0x3ff);
1555 u64 sector = bb >> 10;
1556 sector <<= sb->bblog_shift;
1557 count <<= sb->bblog_shift;
1558 if (bb + 1 == 0)
1559 break;
1560 if (md_set_badblocks(&rdev->badblocks,
1561 sector, count, 1) == 0)
1562 return -EINVAL;
1563 }
486adf72
N		 1564 } else if (sb->bblog_offset != 0)
1565 rdev->badblocks.shift = 0;
2699b672 1566
9a7b2b0f 1567 if (!refdev) {
8ed75463 1568 ret = 1;
9a7b2b0f 1569 } else {
1da177e4 1570 __u64 ev1, ev2;
65a06f06 1571 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1572
1573 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1574 sb->level != refsb->level ||
1575 sb->layout != refsb->layout ||
1576 sb->chunksize != refsb->chunksize) {
1577 printk(KERN_WARNING "md: %s has strangely different"
1578 " superblock to %s\n",
1579 bdevname(rdev->bdev,b),
1580 bdevname(refdev->bdev,b2));
1581 return -EINVAL;
1582 }
1583 ev1 = le64_to_cpu(sb->events);
1584 ev2 = le64_to_cpu(refsb->events);
1585
1586 if (ev1 > ev2)
8ed75463
N		 1587 ret = 1;
1588 else
1589 ret = 0;
1da177e4 1590 }
c6563a8c
N		 1591 if (minor_version) {
1592 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1593 sectors -= rdev->data_offset;
1594 } else
1595 sectors = rdev->sb_start;
1596 if (sectors < le64_to_cpu(sb->data_size))
1da177e4 1597 return -EINVAL;
dd8ac336 1598 rdev->sectors = le64_to_cpu(sb->data_size);
8ed75463 1599 return ret;
1da177e4
LT		 1600}
1601
fd01b88c 1602static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1603{
65a06f06 1604 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1605 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1606
41158c7e 1607 rdev->raid_disk = -1;
c5d79adb
N		 1608 clear_bit(Faulty, &rdev->flags);
1609 clear_bit(In_sync, &rdev->flags);
bb4a65df 1610 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1611 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1612
1da177e4
LT		 1613 if (mddev->raid_disks == 0) {
1614 mddev->major_version = 1;
1615 mddev->patch_version = 0;
e691063a 1616 mddev->external = 0;
9d8f0363 1617 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1618 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1619 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1620 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1621 mddev->clevel[0] = 0;
1da177e4
LT		 1622 mddev->layout = le32_to_cpu(sb->layout);
1623 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1624 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1625 mddev->events = ev1;
c3d9714e 1626 mddev->bitmap_info.offset = 0;
6409bb05
N		 1627 mddev->bitmap_info.space = 0;
1628 /* Default location for bitmap is 1K after superblock
1629 * using 3K - total of 4K
1630 */
c3d9714e 1631 mddev->bitmap_info.default_offset = 1024 >> 9;
6409bb05 1632 mddev->bitmap_info.default_space = (4096-1024) >> 9;
2c810cdd
N		 1633 mddev->reshape_backwards = 0;
1634
1da177e4
LT		 1635 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1636 memcpy(mddev->uuid, sb->set_uuid, 16);
1637
1638 mddev->max_disks = (4096-256)/2;
a654b9d8 1639
71c0805c 1640 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
6409bb05 1641 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1642 mddev->bitmap_info.offset =
1643 (__s32)le32_to_cpu(sb->bitmap_offset);
6409bb05
N		 1644 /* Metadata doesn't record how much space is available.
1645 * For 1.0, we assume we can use up to the superblock
1646 * if before, else to 4K beyond superblock.
1647 * For others, assume no change is possible.
1648 */
1649 if (mddev->minor_version > 0)
1650 mddev->bitmap_info.space = 0;
1651 else if (mddev->bitmap_info.offset > 0)
1652 mddev->bitmap_info.space =
1653 8 - mddev->bitmap_info.offset;
1654 else
1655 mddev->bitmap_info.space =
1656 -mddev->bitmap_info.offset;
1657 }
e11e93fa 1658
f6705578
N		 1659 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1660 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1661 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1662 mddev->new_level = le32_to_cpu(sb->new_level);
1663 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1664 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
2c810cdd
N		 1665 if (mddev->delta_disks < 0 ||
1666 (mddev->delta_disks == 0 &&
1667 (le32_to_cpu(sb->feature_map)
1668 & MD_FEATURE_RESHAPE_BACKWARDS)))
1669 mddev->reshape_backwards = 1;
f6705578
N		 1670 } else {
1671 mddev->reshape_position = MaxSector;
1672 mddev->delta_disks = 0;
1673 mddev->new_level = mddev->level;
1674 mddev->new_layout = mddev->layout;
664e7c41 1675 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1676 }
1677
41158c7e 1678 } else if (mddev->pers == NULL) {
be6800a7
N		 1679 /* Insist of good event counter while assembling, except for
1680 * spares (which don't need an event count) */
1da177e4 1681 ++ev1;
be6800a7
N		 1682 if (rdev->desc_nr >= 0 &&
1683 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1684 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1685 if (ev1 < mddev->events)
1686 return -EINVAL;
41158c7e
N		 1687 } else if (mddev->bitmap) {
1688 /* If adding to array with a bitmap, then we can accept an
1689 * older device, but not too old.
1690 */
41158c7e
N		 1691 if (ev1 < mddev->bitmap->events_cleared)
1692 return 0;
bb4a65df
N		 1693 if (ev1 < mddev->events)
1694 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1695 } else {
1696 if (ev1 < mddev->events)
1697 /* just a hot-add of a new device, leave raid_disk at -1 */
1698 return 0;
1699 }
1da177e4
LT		 1700 if (mddev->level != LEVEL_MULTIPATH) {
1701 int role;
3673f305
N		 1702 if (rdev->desc_nr < 0 ||
1703 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1704 role = 0xffff;
1705 rdev->desc_nr = -1;
1706 } else
1707 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1708 switch(role) {
1709 case 0xffff: /* spare */
1da177e4
LT		 1710 break;
1711 case 0xfffe: /* faulty */
b2d444d7 1712 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1713 break;
1714 default:
5fd6c1dc
N		 1715 if ((le32_to_cpu(sb->feature_map) &
1716 MD_FEATURE_RECOVERY_OFFSET))
1717 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1718 else
1719 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1720 rdev->raid_disk = role;
1721 break;
1722 }
8ddf9efe
N		 1723 if (sb->devflags & WriteMostly1)
1724 set_bit(WriteMostly, &rdev->flags);
2d78f8c4
N		 1725 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1726 set_bit(Replacement, &rdev->flags);
41158c7e 1727 } else /* MULTIPATH are always insync */
b2d444d7 1728 set_bit(In_sync, &rdev->flags);
41158c7e 1729
1da177e4
LT		 1730 return 0;
1731}
1732
fd01b88c 1733static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1734{
1735 struct mdp_superblock_1 *sb;
3cb03002 1736 struct md_rdev *rdev2;
1da177e4
LT		 1737 int max_dev, i;
1738 /* make rdev->sb match mddev and rdev data. */
1739
65a06f06 1740 sb = page_address(rdev->sb_page);
1da177e4
LT		 1741
1742 sb->feature_map = 0;
1743 sb->pad0 = 0;
5fd6c1dc 1744 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1745 memset(sb->pad3, 0, sizeof(sb->pad3));
1746
1747 sb->utime = cpu_to_le64((__u64)mddev->utime);
1748 sb->events = cpu_to_le64(mddev->events);
1749 if (mddev->in_sync)
1750 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1751 else
1752 sb->resync_offset = cpu_to_le64(0);
1753
1c05b4bc 1754 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1755
f0ca340c 1756 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1757 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1758 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1759 sb->level = cpu_to_le32(mddev->level);
1760 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1761
aeb9b211
N		 1762 if (test_bit(WriteMostly, &rdev->flags))
1763 sb->devflags |= WriteMostly1;
1764 else
1765 sb->devflags &= ~WriteMostly1;
c6563a8c
N		 1766 sb->data_offset = cpu_to_le64(rdev->data_offset);
1767 sb->data_size = cpu_to_le64(rdev->sectors);
aeb9b211 1768
c3d9714e
N		 1769 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1770 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1771 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1772 }
5fd6c1dc
N		 1773
1774 if (rdev->raid_disk >= 0 &&
97e4f42d 1775 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1776 sb->feature_map |=
1777 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1778 sb->recovery_offset =
1779 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc 1780 }
2d78f8c4
N		 1781 if (test_bit(Replacement, &rdev->flags))
1782 sb->feature_map |=
1783 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1784
f6705578
N		 1785 if (mddev->reshape_position != MaxSector) {
1786 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1787 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1788 sb->new_layout = cpu_to_le32(mddev->new_layout);
1789 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1790 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1791 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2c810cdd
N		 1792 if (mddev->delta_disks == 0 &&
1793 mddev->reshape_backwards)
1794 sb->feature_map
1795 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
c6563a8c
N		 1796 if (rdev->new_data_offset != rdev->data_offset) {
1797 sb->feature_map
1798 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1799 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1800 - rdev->data_offset));
1801 }
f6705578 1802 }
a654b9d8 1803
2699b672
N		 1804 if (rdev->badblocks.count == 0)
1805 /* Nothing to do for bad blocks*/ ;
1806 else if (sb->bblog_offset == 0)
1807 /* Cannot record bad blocks on this device */
1808 md_error(mddev, rdev);
1809 else {
1810 struct badblocks *bb = &rdev->badblocks;
1811 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1812 u64 *p = bb->page;
1813 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1814 if (bb->changed) {
1815 unsigned seq;
1816
1817retry:
1818 seq = read_seqbegin(&bb->lock);
1819
1820 memset(bbp, 0xff, PAGE_SIZE);
1821
1822 for (i = 0 ; i < bb->count ; i++) {
35f9ac2d 1823 u64 internal_bb = p[i];
2699b672
N		 1824 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1825 | BB_LEN(internal_bb));
35f9ac2d 1826 bbp[i] = cpu_to_le64(store_bb);
2699b672 1827 }
d0962936 1828 bb->changed = 0;
2699b672
N		 1829 if (read_seqretry(&bb->lock, seq))
1830 goto retry;
1831
1832 bb->sector = (rdev->sb_start +
1833 (int)le32_to_cpu(sb->bblog_offset));
1834 bb->size = le16_to_cpu(sb->bblog_size);
2699b672
N		 1835 }
1836 }
1837
1da177e4 1838 max_dev = 0;
dafb20fa 1839 rdev_for_each(rdev2, mddev)
1da177e4
LT		 1840 if (rdev2->desc_nr+1 > max_dev)
1841 max_dev = rdev2->desc_nr+1;
a778b73f 1842
70471daf
N		 1843 if (max_dev > le32_to_cpu(sb->max_dev)) {
1844 int bmask;
a778b73f 1845 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1846 rdev->sb_size = max_dev * 2 + 256;
1847 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1848 if (rdev->sb_size & bmask)
1849 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1850 } else
1851 max_dev = le32_to_cpu(sb->max_dev);
1852
1da177e4
LT		 1853 for (i=0; i<max_dev;i++)
1854 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1855
dafb20fa 1856 rdev_for_each(rdev2, mddev) {
1da177e4 1857 i = rdev2->desc_nr;
b2d444d7 1858 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1859 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1860 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1861 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1862 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1863 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1864 else
1865 sb->dev_roles[i] = cpu_to_le16(0xffff);
1866 }
1867
1da177e4
LT		 1868 sb->sb_csum = calc_sb_1_csum(sb);
1869}
1870
0cd17fec 1871static unsigned long long
3cb03002 1872super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 1873{
1874 struct mdp_superblock_1 *sb;
15f4a5fd 1875 sector_t max_sectors;
58c0fed4 1876 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1877 return 0; /* component must fit device */
c6563a8c
N		 1878 if (rdev->data_offset != rdev->new_data_offset)
1879 return 0; /* too confusing */
0f420358 1880 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1881 /* minor versions 1 and 2; superblock before data */
77304d2a 1882 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1883 max_sectors -= rdev->data_offset;
1884 if (!num_sectors || num_sectors > max_sectors)
1885 num_sectors = max_sectors;
c3d9714e 1886 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1887 /* minor version 0 with bitmap we can't move */
1888 return 0;
1889 } else {
1890 /* minor version 0; superblock after data */
0f420358 1891 sector_t sb_start;
77304d2a 1892 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1893 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1894 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1895 if (!num_sectors || num_sectors > max_sectors)
1896 num_sectors = max_sectors;
0f420358 1897 rdev->sb_start = sb_start;
0cd17fec 1898 }
65a06f06 1899 sb = page_address(rdev->sb_page);
15f4a5fd 1900 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1901 sb->super_offset = rdev->sb_start;
0cd17fec 1902 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1903 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1904 rdev->sb_page);
1905 md_super_wait(rdev->mddev);
c26a44ed 1906 return num_sectors;
c6563a8c
N		 1907
1908}
1909
1910static int
1911super_1_allow_new_offset(struct md_rdev *rdev,
1912 unsigned long long new_offset)
1913{
1914 /* All necessary checks on new >= old have been done */
1915 struct bitmap *bitmap;
1916 if (new_offset >= rdev->data_offset)
1917 return 1;
1918
1919 /* with 1.0 metadata, there is no metadata to tread on
1920 * so we can always move back */
1921 if (rdev->mddev->minor_version == 0)
1922 return 1;
1923
1924 /* otherwise we must be sure not to step on
1925 * any metadata, so stay:
1926 * 36K beyond start of superblock
1927 * beyond end of badblocks
1928 * beyond write-intent bitmap
1929 */
1930 if (rdev->sb_start + (32+4)*2 > new_offset)
1931 return 0;
1932 bitmap = rdev->mddev->bitmap;
1933 if (bitmap && !rdev->mddev->bitmap_info.file &&
1934 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1ec885cd 1935 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
c6563a8c
N		 1936 return 0;
1937 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1938 return 0;
1939
1940 return 1;
0cd17fec 1941}
1da177e4 1942
75c96f85 1943static struct super_type super_types[] = {
1da177e4
LT		 1944 [0] = {
1945 .name = "0.90.0",
1946 .owner = THIS_MODULE,
0cd17fec
CW		 1947 .load_super = super_90_load,
1948 .validate_super = super_90_validate,
1949 .sync_super = super_90_sync,
1950 .rdev_size_change = super_90_rdev_size_change,
c6563a8c 1951 .allow_new_offset = super_90_allow_new_offset,
1da177e4
LT		 1952 },
1953 [1] = {
1954 .name = "md-1",
1955 .owner = THIS_MODULE,
0cd17fec
CW		 1956 .load_super = super_1_load,
1957 .validate_super = super_1_validate,
1958 .sync_super = super_1_sync,
1959 .rdev_size_change = super_1_rdev_size_change,
c6563a8c 1960 .allow_new_offset = super_1_allow_new_offset,
1da177e4
LT		 1961 },
1962};
1da177e4 1963
fd01b88c 1964static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 1965{
1966 if (mddev->sync_super) {
1967 mddev->sync_super(mddev, rdev);
1968 return;
1969 }
1970
1971 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1972
1973 super_types[mddev->major_version].sync_super(mddev, rdev);
1974}
1975
fd01b88c 1976static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1977{
3cb03002 1978 struct md_rdev *rdev, *rdev2;
1da177e4 1979
4b80991c
N		 1980 rcu_read_lock();
1981 rdev_for_each_rcu(rdev, mddev1)
1982 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1983 if (rdev->bdev->bd_contains ==
4b80991c
N		 1984 rdev2->bdev->bd_contains) {
1985 rcu_read_unlock();
7dd5e7c3 1986 return 1;
4b80991c
N		 1987 }
1988 rcu_read_unlock();
1da177e4
LT		 1989 return 0;
1990}
1991
1992static LIST_HEAD(pending_raid_disks);
1993
ac5e7113
AN		 1994/*
1995 * Try to register data integrity profile for an mddev
1996 *
1997 * This is called when an array is started and after a disk has been kicked
1998 * from the array. It only succeeds if all working and active component devices
1999 * are integrity capable with matching profiles.
2000 */
fd01b88c 2001int md_integrity_register(struct mddev *mddev)
ac5e7113 2002{
3cb03002 2003 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 2004
2005 if (list_empty(&mddev->disks))
2006 return 0; /* nothing to do */
629acb6a
JB		 2007 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2008 return 0; /* shouldn't register, or already is */
dafb20fa 2009 rdev_for_each(rdev, mddev) {
ac5e7113
AN		 2010 /* skip spares and non-functional disks */
2011 if (test_bit(Faulty, &rdev->flags))
2012 continue;
2013 if (rdev->raid_disk < 0)
2014 continue;
ac5e7113
AN		 2015 if (!reference) {
2016 /* Use the first rdev as the reference */
2017 reference = rdev;
2018 continue;
2019 }
2020 /* does this rdev's profile match the reference profile? */
2021 if (blk_integrity_compare(reference->bdev->bd_disk,
2022 rdev->bdev->bd_disk) < 0)
2023 return -EINVAL;
2024 }
89078d57
MP		 2025 if (!reference || !bdev_get_integrity(reference->bdev))
2026 return 0;
ac5e7113
AN		 2027 /*
2028 * All component devices are integrity capable and have matching
2029 * profiles, register the common profile for the md device.
2030 */
2031 if (blk_integrity_register(mddev->gendisk,
2032 bdev_get_integrity(reference->bdev)) != 0) {
2033 printk(KERN_ERR "md: failed to register integrity for %s\n",
2034 mdname(mddev));
2035 return -EINVAL;
2036 }
a91a2785
MP		 2037 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2038 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2039 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2040 mdname(mddev));
2041 return -EINVAL;
2042 }
ac5e7113
AN		 2043 return 0;
2044}
2045EXPORT_SYMBOL(md_integrity_register);
2046
2047/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 2048void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 2049{
2863b9eb
JB		 2050 struct blk_integrity *bi_rdev;
2051 struct blk_integrity *bi_mddev;
2052
2053 if (!mddev->gendisk)
2054 return;
2055
2056 bi_rdev = bdev_get_integrity(rdev->bdev);
2057 bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 2058
ac5e7113 2059 if (!bi_mddev) /* nothing to do */
3f9d99c1 2060 return;
ac5e7113 2061 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 2062 return;
ac5e7113
AN		 2063 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2064 rdev->bdev->bd_disk) >= 0)
2065 return;
2066 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2067 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2068}
ac5e7113 2069EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2070
fd01b88c 2071static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
1da177e4 2072{
7dd5e7c3 2073 char b[BDEVNAME_SIZE];
f637b9f9 2074 struct kobject *ko;
1edf80d3 2075 char *s;
5e55e2f5 2076 int err;
1da177e4
LT		 2077
2078 if (rdev->mddev) {
2079 MD_BUG();
2080 return -EINVAL;
2081 }
11e2ede0
DW		 2082
2083 /* prevent duplicates */
2084 if (find_rdev(mddev, rdev->bdev->bd_dev))
2085 return -EEXIST;
2086
dd8ac336
AN		 2087 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2088 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2089 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2090 if (mddev->pers) {
2091 /* Cannot change size, so fail
2092 * If mddev->level <= 0, then we don't care
2093 * about aligning sizes (e.g. linear)
2094 */
2095 if (mddev->level > 0)
2096 return -ENOSPC;
2097 } else
dd8ac336 2098 mddev->dev_sectors = rdev->sectors;
2bf071bf 2099 }
1da177e4
LT		 2100
2101 /* Verify rdev->desc_nr is unique.
2102 * If it is -1, assign a free number, else
2103 * check number is not in use
2104 */
2105 if (rdev->desc_nr < 0) {
2106 int choice = 0;
2107 if (mddev->pers) choice = mddev->raid_disks;
2108 while (find_rdev_nr(mddev, choice))
2109 choice++;
2110 rdev->desc_nr = choice;
2111 } else {
2112 if (find_rdev_nr(mddev, rdev->desc_nr))
2113 return -EBUSY;
2114 }
de01dfad
N		 2115 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2116 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2117 mdname(mddev), mddev->max_disks);
2118 return -EBUSY;
2119 }
19133a42 2120 bdevname(rdev->bdev,b);
649316b2 2121 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2122 *s = '!';
649316b2 2123
1da177e4 2124 rdev->mddev = mddev;
19133a42 2125 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2126
b2d6db58 2127 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2128 goto fail;
86e6ffdd 2129
0762b8bd 2130 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2131 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2132 /* failure here is OK */;
2133 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2134
4b80991c 2135 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2136 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2137
2138 /* May as well allow recovery to be retried once */
5389042f 2139 mddev->recovery_disabled++;
3f9d99c1 2140
1da177e4 2141 return 0;
5e55e2f5
N		 2142
2143 fail:
2144 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2145 b, mdname(mddev));
2146 return err;
1da177e4
LT		 2147}
2148
177a99b2 2149static void md_delayed_delete(struct work_struct *ws)
5792a285 2150{
3cb03002 2151 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2152 kobject_del(&rdev->kobj);
177a99b2 2153 kobject_put(&rdev->kobj);
5792a285
N		 2154}
2155
3cb03002 2156static void unbind_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2157{
2158 char b[BDEVNAME_SIZE];
2159 if (!rdev->mddev) {
2160 MD_BUG();
2161 return;
2162 }
49731baa 2163 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2164 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2165 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2166 rdev->mddev = NULL;
86e6ffdd 2167 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2168 sysfs_put(rdev->sysfs_state);
2169 rdev->sysfs_state = NULL;
2230dfe4 2170 rdev->badblocks.count = 0;
5792a285 2171 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2172 * writing to 'remove' to "dev/state". We also need
2173 * to delay it due to rcu usage.
5792a285 2174 */
4b80991c 2175 synchronize_rcu();
177a99b2
N		 2176 INIT_WORK(&rdev->del_work, md_delayed_delete);
2177 kobject_get(&rdev->kobj);
e804ac78 2178 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2179}
2180
2181/*
2182 * prevent the device from being mounted, repartitioned or
2183 * otherwise reused by a RAID array (or any other kernel
2184 * subsystem), by bd_claiming the device.
2185 */
3cb03002 2186static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2187{
2188 int err = 0;
2189 struct block_device *bdev;
2190 char b[BDEVNAME_SIZE];
2191
d4d77629 2192 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2193 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2194 if (IS_ERR(bdev)) {
2195 printk(KERN_ERR "md: could not open %s.\n",
2196 __bdevname(dev, b));
2197 return PTR_ERR(bdev);
2198 }
1da177e4
LT		 2199 rdev->bdev = bdev;
2200 return err;
2201}
2202
3cb03002 2203static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2204{
2205 struct block_device *bdev = rdev->bdev;
2206 rdev->bdev = NULL;
2207 if (!bdev)
2208 MD_BUG();
e525fd89 2209 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2210}
2211
2212void md_autodetect_dev(dev_t dev);
2213
3cb03002 2214static void export_rdev(struct md_rdev * rdev)
1da177e4
LT		 2215{
2216 char b[BDEVNAME_SIZE];
2217 printk(KERN_INFO "md: export_rdev(%s)\n",
2218 bdevname(rdev->bdev,b));
2219 if (rdev->mddev)
2220 MD_BUG();
545c8795 2221 md_rdev_clear(rdev);
1da177e4 2222#ifndef MODULE
d0fae18f
N		 2223 if (test_bit(AutoDetected, &rdev->flags))
2224 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2225#endif
2226 unlock_rdev(rdev);
86e6ffdd 2227 kobject_put(&rdev->kobj);
1da177e4
LT		 2228}
2229
3cb03002 2230static void kick_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2231{
2232 unbind_rdev_from_array(rdev);
2233 export_rdev(rdev);
2234}
2235
fd01b88c 2236static void export_array(struct mddev *mddev)
1da177e4 2237{
3cb03002 2238 struct md_rdev *rdev, *tmp;
1da177e4 2239
dafb20fa 2240 rdev_for_each_safe(rdev, tmp, mddev) {
1da177e4
LT		 2241 if (!rdev->mddev) {
2242 MD_BUG();
2243 continue;
2244 }
2245 kick_rdev_from_array(rdev);
2246 }
2247 if (!list_empty(&mddev->disks))
2248 MD_BUG();
2249 mddev->raid_disks = 0;
2250 mddev->major_version = 0;
2251}
2252
2253static void print_desc(mdp_disk_t *desc)
2254{
2255 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2256 desc->major,desc->minor,desc->raid_disk,desc->state);
2257}
2258
cd2ac932 2259static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 2260{
2261 int i;
2262
2263 printk(KERN_INFO
2264 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2265 sb->major_version, sb->minor_version, sb->patch_version,
2266 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2267 sb->ctime);
2268 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2269 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2270 sb->md_minor, sb->layout, sb->chunk_size);
2271 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2272 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2273 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2274 sb->failed_disks, sb->spare_disks,
2275 sb->sb_csum, (unsigned long)sb->events_lo);
2276
2277 printk(KERN_INFO);
2278 for (i = 0; i < MD_SB_DISKS; i++) {
2279 mdp_disk_t *desc;
2280
2281 desc = sb->disks + i;
2282 if (desc->number || desc->major || desc->minor ||
2283 desc->raid_disk || (desc->state && (desc->state != 4))) {
2284 printk(" D %2d: ", i);
2285 print_desc(desc);
2286 }
2287 }
2288 printk(KERN_INFO "md: THIS: ");
2289 print_desc(&sb->this_disk);
cd2ac932 2290}
1da177e4 2291
cd2ac932
CR		 2292static void print_sb_1(struct mdp_superblock_1 *sb)
2293{
2294 __u8 *uuid;
2295
2296 uuid = sb->set_uuid;
ad361c98 2297 printk(KERN_INFO
7b75c2f8 2298 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2299 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 2300 le32_to_cpu(sb->major_version),
2301 le32_to_cpu(sb->feature_map),
7b75c2f8 2302 uuid,
cd2ac932
CR		 2303 sb->set_name,
2304 (unsigned long long)le64_to_cpu(sb->ctime)
2305 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2306
2307 uuid = sb->device_uuid;
ad361c98
JP		 2308 printk(KERN_INFO
2309 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2310 " RO:%llu\n"
7b75c2f8 2311 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP		 2312 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2313 "md: (MaxDev:%u) \n",
cd2ac932
CR		 2314 le32_to_cpu(sb->level),
2315 (unsigned long long)le64_to_cpu(sb->size),
2316 le32_to_cpu(sb->raid_disks),
2317 le32_to_cpu(sb->layout),
2318 le32_to_cpu(sb->chunksize),
2319 (unsigned long long)le64_to_cpu(sb->data_offset),
2320 (unsigned long long)le64_to_cpu(sb->data_size),
2321 (unsigned long long)le64_to_cpu(sb->super_offset),
2322 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2323 le32_to_cpu(sb->dev_number),
7b75c2f8 2324 uuid,
cd2ac932
CR		 2325 sb->devflags,
2326 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2327 (unsigned long long)le64_to_cpu(sb->events),
2328 (unsigned long long)le64_to_cpu(sb->resync_offset),
2329 le32_to_cpu(sb->sb_csum),
2330 le32_to_cpu(sb->max_dev)
2331 );
1da177e4
LT		 2332}
2333
3cb03002 2334static void print_rdev(struct md_rdev *rdev, int major_version)
1da177e4
LT		 2335{
2336 char b[BDEVNAME_SIZE];
dd8ac336
AN		 2337 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2338 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 2339 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2340 rdev->desc_nr);
1da177e4 2341 if (rdev->sb_loaded) {
cd2ac932
CR		 2342 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2343 switch (major_version) {
2344 case 0:
65a06f06 2345 print_sb_90(page_address(rdev->sb_page));
cd2ac932
CR		 2346 break;
2347 case 1:
65a06f06 2348 print_sb_1(page_address(rdev->sb_page));
cd2ac932
CR		 2349 break;
2350 }
1da177e4
LT		 2351 } else
2352 printk(KERN_INFO "md: no rdev superblock!\n");
2353}
2354
5e56341d 2355static void md_print_devices(void)
1da177e4 2356{
159ec1fc 2357 struct list_head *tmp;
3cb03002 2358 struct md_rdev *rdev;
fd01b88c 2359 struct mddev *mddev;
1da177e4
LT		 2360 char b[BDEVNAME_SIZE];
2361
2362 printk("\n");
2363 printk("md: **********************************\n");
2364 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2365 printk("md: **********************************\n");
29ac4aa3 2366 for_each_mddev(mddev, tmp) {
1da177e4 2367
32a7627c
N		 2368 if (mddev->bitmap)
2369 bitmap_print_sb(mddev->bitmap);
2370 else
2371 printk("%s: ", mdname(mddev));
dafb20fa 2372 rdev_for_each(rdev, mddev)
1da177e4
LT		 2373 printk("<%s>", bdevname(rdev->bdev,b));
2374 printk("\n");
2375
dafb20fa 2376 rdev_for_each(rdev, mddev)
cd2ac932 2377 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 2378 }
2379 printk("md: **********************************\n");
2380 printk("\n");
2381}
2382
2383
fd01b88c 2384static void sync_sbs(struct mddev * mddev, int nospares)
1da177e4 2385{
42543769
N		 2386 /* Update each superblock (in-memory image), but
2387 * if we are allowed to, skip spares which already
2388 * have the right event counter, or have one earlier
2389 * (which would mean they aren't being marked as dirty
2390 * with the rest of the array)
2391 */
3cb03002 2392 struct md_rdev *rdev;
dafb20fa 2393 rdev_for_each(rdev, mddev) {
42543769
N		 2394 if (rdev->sb_events == mddev->events ||
2395 (nospares &&
2396 rdev->raid_disk < 0 &&
42543769
N		 2397 rdev->sb_events+1 == mddev->events)) {
2398 /* Don't update this superblock */
2399 rdev->sb_loaded = 2;
2400 } else {
076f968b 2401 sync_super(mddev, rdev);
42543769
N		 2402 rdev->sb_loaded = 1;
2403 }
1da177e4
LT		 2404 }
2405}
2406
fd01b88c 2407static void md_update_sb(struct mddev * mddev, int force_change)
1da177e4 2408{
3cb03002 2409 struct md_rdev *rdev;
06d91a5f 2410 int sync_req;
42543769 2411 int nospares = 0;
2699b672 2412 int any_badblocks_changed = 0;
1da177e4 2413
d87f064f
N		 2414 if (mddev->ro) {
2415 if (force_change)
2416 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2417 return;
2418 }
1da177e4 2419repeat:
3a3a5ddb 2420 /* First make sure individual recovery_offsets are correct */
dafb20fa 2421 rdev_for_each(rdev, mddev) {
3a3a5ddb
N		 2422 if (rdev->raid_disk >= 0 &&
2423 mddev->delta_disks >= 0 &&
2424 !test_bit(In_sync, &rdev->flags) &&
2425 mddev->curr_resync_completed > rdev->recovery_offset)
2426 rdev->recovery_offset = mddev->curr_resync_completed;
2427
2428 }
bd52b746 2429 if (!mddev->persistent) {
070dc6dd 2430 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2431 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2432 if (!mddev->external) {
d97a41dc 2433 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
dafb20fa 2434 rdev_for_each(rdev, mddev) {
de393cde 2435 if (rdev->badblocks.changed) {
d0962936 2436 rdev->badblocks.changed = 0;
de393cde
N		 2437 md_ack_all_badblocks(&rdev->badblocks);
2438 md_error(mddev, rdev);
2439 }
2440 clear_bit(Blocked, &rdev->flags);
2441 clear_bit(BlockedBadBlocks, &rdev->flags);
2442 wake_up(&rdev->blocked_wait);
2443 }
2444 }
3a3a5ddb
N		 2445 wake_up(&mddev->sb_wait);
2446 return;
2447 }
2448
a9701a30 2449 spin_lock_irq(&mddev->write_lock);
84692195 2450
3a3a5ddb
N		 2451 mddev->utime = get_seconds();
2452
850b2b42
N		 2453 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2454 force_change = 1;
2455 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2456 /* just a clean<-> dirty transition, possibly leave spares alone,
2457 * though if events isn't the right even/odd, we will have to do
2458 * spares after all
2459 */
2460 nospares = 1;
2461 if (force_change)
2462 nospares = 0;
2463 if (mddev->degraded)
84692195
N		 2464 /* If the array is degraded, then skipping spares is both
2465 * dangerous and fairly pointless.
2466 * Dangerous because a device that was removed from the array
2467 * might have a event_count that still looks up-to-date,
2468 * so it can be re-added without a resync.
2469 * Pointless because if there are any spares to skip,
2470 * then a recovery will happen and soon that array won't
2471 * be degraded any more and the spare can go back to sleep then.
2472 */
850b2b42 2473 nospares = 0;
84692195 2474
06d91a5f 2475 sync_req = mddev->in_sync;
42543769
N		 2476
2477 /* If this is just a dirty<->clean transition, and the array is clean
2478 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2479 if (nospares
42543769 2480 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2481 && mddev->can_decrease_events
2482 && mddev->events != 1) {
42543769 2483 mddev->events--;
a8707c08
N		 2484 mddev->can_decrease_events = 0;
2485 } else {
42543769
N		 2486 /* otherwise we have to go forward and ... */
2487 mddev->events ++;
a8707c08 2488 mddev->can_decrease_events = nospares;
42543769 2489 }
1da177e4
LT		 2490
2491 if (!mddev->events) {
2492 /*
2493 * oops, this 64-bit counter should never wrap.
2494 * Either we are in around ~1 trillion A.C., assuming
2495 * 1 reboot per second, or we have a bug:
2496 */
2497 MD_BUG();
2498 mddev->events --;
2499 }
2699b672 2500
dafb20fa 2501 rdev_for_each(rdev, mddev) {
2699b672
N		 2502 if (rdev->badblocks.changed)
2503 any_badblocks_changed++;
de393cde
N		 2504 if (test_bit(Faulty, &rdev->flags))
2505 set_bit(FaultRecorded, &rdev->flags);
2506 }
2699b672 2507
e691063a 2508 sync_sbs(mddev, nospares);
a9701a30 2509 spin_unlock_irq(&mddev->write_lock);
1da177e4 2510
36a4e1fe
N		 2511 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2512 mdname(mddev), mddev->in_sync);
1da177e4 2513
4ad13663 2514 bitmap_update_sb(mddev->bitmap);
dafb20fa 2515 rdev_for_each(rdev, mddev) {
1da177e4 2516 char b[BDEVNAME_SIZE];
36a4e1fe 2517
42543769
N		 2518 if (rdev->sb_loaded != 1)
2519 continue; /* no noise on spare devices */
1da177e4 2520
d70ed2e4
AW		 2521 if (!test_bit(Faulty, &rdev->flags) &&
2522 rdev->saved_raid_disk == -1) {
7bfa19f2 2523 md_super_write(mddev,rdev,
0f420358 2524 rdev->sb_start, rdev->sb_size,
7bfa19f2 2525 rdev->sb_page);
36a4e1fe
N		 2526 pr_debug("md: (write) %s's sb offset: %llu\n",
2527 bdevname(rdev->bdev, b),
2528 (unsigned long long)rdev->sb_start);
42543769 2529 rdev->sb_events = mddev->events;
2699b672
N		 2530 if (rdev->badblocks.size) {
2531 md_super_write(mddev, rdev,
2532 rdev->badblocks.sector,
2533 rdev->badblocks.size << 9,
2534 rdev->bb_page);
2535 rdev->badblocks.size = 0;
2536 }
7bfa19f2 2537
d70ed2e4 2538 } else if (test_bit(Faulty, &rdev->flags))
36a4e1fe
N		 2539 pr_debug("md: %s (skipping faulty)\n",
2540 bdevname(rdev->bdev, b));
d70ed2e4
AW		 2541 else
2542 pr_debug("(skipping incremental s/r ");
2543
7bfa19f2 2544 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2545 /* only need to write one superblock... */
2546 break;
2547 }
a9701a30 2548 md_super_wait(mddev);
850b2b42 2549 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2550
a9701a30 2551 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2552 if (mddev->in_sync != sync_req ||
2553 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2554 /* have to write it out again */
a9701a30 2555 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2556 goto repeat;
2557 }
850b2b42 2558 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2559 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2560 wake_up(&mddev->sb_wait);
acb180b0
N		 2561 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2562 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2563
dafb20fa 2564 rdev_for_each(rdev, mddev) {
de393cde
N		 2565 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2566 clear_bit(Blocked, &rdev->flags);
2567
2568 if (any_badblocks_changed)
2699b672 2569 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2570 clear_bit(BlockedBadBlocks, &rdev->flags);
2571 wake_up(&rdev->blocked_wait);
2572 }
1da177e4
LT		 2573}
2574
7f6ce769 2575/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2576 * We want to accept with case. For this we use cmd_match.
2577 */
2578static int cmd_match(const char *cmd, const char *str)
2579{
2580 /* See if cmd, written into a sysfs file, matches
2581 * str. They must either be the same, or cmd can
2582 * have a trailing newline
2583 */
2584 while (*cmd && *str && *cmd == *str) {
2585 cmd++;
2586 str++;
2587 }
2588 if (*cmd == '\n')
2589 cmd++;
2590 if (*str || *cmd)
2591 return 0;
2592 return 1;
2593}
2594
86e6ffdd
N		 2595struct rdev_sysfs_entry {
2596 struct attribute attr;
3cb03002
N		 2597 ssize_t (*show)(struct md_rdev *, char *);
2598 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2599};
2600
2601static ssize_t
3cb03002 2602state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2603{
2604 char *sep = "";
20a49ff6 2605 size_t len = 0;
86e6ffdd 2606
de393cde
N		 2607 if (test_bit(Faulty, &rdev->flags) ||
2608 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2609 len+= sprintf(page+len, "%sfaulty",sep);
2610 sep = ",";
2611 }
b2d444d7 2612 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2613 len += sprintf(page+len, "%sin_sync",sep);
2614 sep = ",";
2615 }
f655675b
N		 2616 if (test_bit(WriteMostly, &rdev->flags)) {
2617 len += sprintf(page+len, "%swrite_mostly",sep);
2618 sep = ",";
2619 }
de393cde 2620 if (test_bit(Blocked, &rdev->flags) ||
52c64152
N		 2621 (rdev->badblocks.unacked_exist
2622 && !test_bit(Faulty, &rdev->flags))) {
6bfe0b49
DW		 2623 len += sprintf(page+len, "%sblocked", sep);
2624 sep = ",";
2625 }
b2d444d7
N		 2626 if (!test_bit(Faulty, &rdev->flags) &&
2627 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2628 len += sprintf(page+len, "%sspare", sep);
2629 sep = ",";
2630 }
d7a9d443
N		 2631 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2632 len += sprintf(page+len, "%swrite_error", sep);
2633 sep = ",";
2634 }
2d78f8c4
N		 2635 if (test_bit(WantReplacement, &rdev->flags)) {
2636 len += sprintf(page+len, "%swant_replacement", sep);
2637 sep = ",";
2638 }
2639 if (test_bit(Replacement, &rdev->flags)) {
2640 len += sprintf(page+len, "%sreplacement", sep);
2641 sep = ",";
2642 }
2643
86e6ffdd
N		 2644 return len+sprintf(page+len, "\n");
2645}
2646
45dc2de1 2647static ssize_t
3cb03002 2648state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2649{
2650 /* can write
de393cde 2651 * faulty - simulates an error
45dc2de1 2652 * remove - disconnects the device
f655675b
N		 2653 * writemostly - sets write_mostly
2654 * -writemostly - clears write_mostly
de393cde
N		 2655 * blocked - sets the Blocked flags
2656 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2657 * insync - sets Insync providing device isn't active
d7a9d443
N		 2658 * write_error - sets WriteErrorSeen
2659 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2660 */
2661 int err = -EINVAL;
2662 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2663 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2664 if (test_bit(Faulty, &rdev->flags))
2665 err = 0;
2666 else
2667 err = -EBUSY;
45dc2de1
N		 2668 } else if (cmd_match(buf, "remove")) {
2669 if (rdev->raid_disk >= 0)
2670 err = -EBUSY;
2671 else {
fd01b88c 2672 struct mddev *mddev = rdev->mddev;
45dc2de1 2673 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2674 if (mddev->pers)
2675 md_update_sb(mddev, 1);
45dc2de1
N		 2676 md_new_event(mddev);
2677 err = 0;
2678 }
f655675b
N		 2679 } else if (cmd_match(buf, "writemostly")) {
2680 set_bit(WriteMostly, &rdev->flags);
2681 err = 0;
2682 } else if (cmd_match(buf, "-writemostly")) {
2683 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2684 err = 0;
2685 } else if (cmd_match(buf, "blocked")) {
2686 set_bit(Blocked, &rdev->flags);
2687 err = 0;
2688 } else if (cmd_match(buf, "-blocked")) {
de393cde 2689 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2690 rdev->badblocks.unacked_exist) {
de393cde
N		 2691 /* metadata handler doesn't understand badblocks,
2692 * so we need to fail the device
2693 */
2694 md_error(rdev->mddev, rdev);
2695 }
6bfe0b49 2696 clear_bit(Blocked, &rdev->flags);
de393cde 2697 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2698 wake_up(&rdev->blocked_wait);
2699 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2700 md_wakeup_thread(rdev->mddev->thread);
2701
6d56e278
N		 2702 err = 0;
2703 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2704 set_bit(In_sync, &rdev->flags);
f655675b 2705 err = 0;
d7a9d443
N		 2706 } else if (cmd_match(buf, "write_error")) {
2707 set_bit(WriteErrorSeen, &rdev->flags);
2708 err = 0;
2709 } else if (cmd_match(buf, "-write_error")) {
2710 clear_bit(WriteErrorSeen, &rdev->flags);
2711 err = 0;
2d78f8c4
N		 2712 } else if (cmd_match(buf, "want_replacement")) {
2713 /* Any non-spare device that is not a replacement can
2714 * become want_replacement at any time, but we then need to
2715 * check if recovery is needed.
2716 */
2717 if (rdev->raid_disk >= 0 &&
2718 !test_bit(Replacement, &rdev->flags))
2719 set_bit(WantReplacement, &rdev->flags);
2720 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2721 md_wakeup_thread(rdev->mddev->thread);
2722 err = 0;
2723 } else if (cmd_match(buf, "-want_replacement")) {
2724 /* Clearing 'want_replacement' is always allowed.
2725 * Once replacements starts it is too late though.
2726 */
2727 err = 0;
2728 clear_bit(WantReplacement, &rdev->flags);
2729 } else if (cmd_match(buf, "replacement")) {
2730 /* Can only set a device as a replacement when array has not
2731 * yet been started. Once running, replacement is automatic
2732 * from spares, or by assigning 'slot'.
2733 */
2734 if (rdev->mddev->pers)
2735 err = -EBUSY;
2736 else {
2737 set_bit(Replacement, &rdev->flags);
2738 err = 0;
2739 }
2740 } else if (cmd_match(buf, "-replacement")) {
2741 /* Similarly, can only clear Replacement before start */
2742 if (rdev->mddev->pers)
2743 err = -EBUSY;
2744 else {
2745 clear_bit(Replacement, &rdev->flags);
2746 err = 0;
2747 }
45dc2de1 2748 }
00bcb4ac
N		 2749 if (!err)
2750 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2751 return err ? err : len;
2752}
80ca3a44
N		 2753static struct rdev_sysfs_entry rdev_state =
2754__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2755
4dbcdc75 2756static ssize_t
3cb03002 2757errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2758{
2759 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2760}
2761
2762static ssize_t
3cb03002 2763errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N		 2764{
2765 char *e;
2766 unsigned long n = simple_strtoul(buf, &e, 10);
2767 if (*buf && (*e == 0 || *e == '\n')) {
2768 atomic_set(&rdev->corrected_errors, n);
2769 return len;
2770 }
2771 return -EINVAL;
2772}
2773static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2774__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2775
014236d2 2776static ssize_t
3cb03002 2777slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2778{
2779 if (rdev->raid_disk < 0)
2780 return sprintf(page, "none\n");
2781 else
2782 return sprintf(page, "%d\n", rdev->raid_disk);
2783}
2784
2785static ssize_t
3cb03002 2786slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N		 2787{
2788 char *e;
c303da6d 2789 int err;
014236d2
N		 2790 int slot = simple_strtoul(buf, &e, 10);
2791 if (strncmp(buf, "none", 4)==0)
2792 slot = -1;
2793 else if (e==buf || (*e && *e!= '\n'))
2794 return -EINVAL;
6c2fce2e 2795 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2796 /* Setting 'slot' on an active array requires also
2797 * updating the 'rd%d' link, and communicating
2798 * with the personality with ->hot_*_disk.
2799 * For now we only support removing
2800 * failed/spare devices. This normally happens automatically,
2801 * but not when the metadata is externally managed.
2802 */
c303da6d
N		 2803 if (rdev->raid_disk == -1)
2804 return -EEXIST;
2805 /* personality does all needed checks */
01393f3d 2806 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d 2807 return -EINVAL;
746d3207
N		 2808 clear_bit(Blocked, &rdev->flags);
2809 remove_and_add_spares(rdev->mddev, rdev);
2810 if (rdev->raid_disk >= 0)
2811 return -EBUSY;
c303da6d
N		 2812 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2813 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2814 } else if (rdev->mddev->pers) {
6c2fce2e 2815 /* Activating a spare .. or possibly reactivating
6d56e278 2816 * if we ever get bitmaps working here.
6c2fce2e
NB		 2817 */
2818
2819 if (rdev->raid_disk != -1)
2820 return -EBUSY;
2821
c6751b2b
N		 2822 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2823 return -EBUSY;
2824
6c2fce2e
NB		 2825 if (rdev->mddev->pers->hot_add_disk == NULL)
2826 return -EINVAL;
2827
ba1b41b6
N		 2828 if (slot >= rdev->mddev->raid_disks &&
2829 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2830 return -ENOSPC;
2831
6c2fce2e
NB		 2832 rdev->raid_disk = slot;
2833 if (test_bit(In_sync, &rdev->flags))
2834 rdev->saved_raid_disk = slot;
2835 else
2836 rdev->saved_raid_disk = -1;
d30519fc 2837 clear_bit(In_sync, &rdev->flags);
bb4a65df 2838 clear_bit(Bitmap_sync, &rdev->flags);
6c2fce2e
NB		 2839 err = rdev->mddev->pers->
2840 hot_add_disk(rdev->mddev, rdev);
199050ea 2841 if (err) {
6c2fce2e 2842 rdev->raid_disk = -1;
6c2fce2e 2843 return err;
52664732 2844 } else
00bcb4ac 2845 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2846 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2847 /* failure here is OK */;
6c2fce2e 2848 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2849 } else {
ba1b41b6
N		 2850 if (slot >= rdev->mddev->raid_disks &&
2851 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2852 return -ENOSPC;
2853 rdev->raid_disk = slot;
2854 /* assume it is working */
c5d79adb
N		 2855 clear_bit(Faulty, &rdev->flags);
2856 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2857 set_bit(In_sync, &rdev->flags);
00bcb4ac 2858 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2859 }
014236d2
N		 2860 return len;
2861}
2862
2863
2864static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2865__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2866
93c8cad0 2867static ssize_t
3cb03002 2868offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2869{
6961ece4 2870 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2871}
2872
2873static ssize_t
3cb03002 2874offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2875{
c6563a8c
N		 2876 unsigned long long offset;
2877 if (strict_strtoull(buf, 10, &offset) < 0)
93c8cad0 2878 return -EINVAL;
8ed0a521 2879 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2880 return -EBUSY;
dd8ac336 2881 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2882 /* Must set offset before size, so overlap checks
2883 * can be sane */
2884 return -EBUSY;
93c8cad0 2885 rdev->data_offset = offset;
25f7fd47 2886 rdev->new_data_offset = offset;
93c8cad0
N		 2887 return len;
2888}
2889
2890static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2891__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2892
c6563a8c
N		 2893static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2894{
2895 return sprintf(page, "%llu\n",
2896 (unsigned long long)rdev->new_data_offset);
2897}
2898
2899static ssize_t new_offset_store(struct md_rdev *rdev,
2900 const char *buf, size_t len)
2901{
2902 unsigned long long new_offset;
2903 struct mddev *mddev = rdev->mddev;
2904
2905 if (strict_strtoull(buf, 10, &new_offset) < 0)
2906 return -EINVAL;
2907
2908 if (mddev->sync_thread)
2909 return -EBUSY;
2910 if (new_offset == rdev->data_offset)
2911 /* reset is always permitted */
2912 ;
2913 else if (new_offset > rdev->data_offset) {
2914 /* must not push array size beyond rdev_sectors */
2915 if (new_offset - rdev->data_offset
2916 + mddev->dev_sectors > rdev->sectors)
2917 return -E2BIG;
2918 }
2919 /* Metadata worries about other space details. */
2920
2921 /* decreasing the offset is inconsistent with a backwards
2922 * reshape.
2923 */
2924 if (new_offset < rdev->data_offset &&
2925 mddev->reshape_backwards)
2926 return -EINVAL;
2927 /* Increasing offset is inconsistent with forwards
2928 * reshape. reshape_direction should be set to
2929 * 'backwards' first.
2930 */
2931 if (new_offset > rdev->data_offset &&
2932 !mddev->reshape_backwards)
2933 return -EINVAL;
2934
2935 if (mddev->pers && mddev->persistent &&
2936 !super_types[mddev->major_version]
2937 .allow_new_offset(rdev, new_offset))
2938 return -E2BIG;
2939 rdev->new_data_offset = new_offset;
2940 if (new_offset > rdev->data_offset)
2941 mddev->reshape_backwards = 1;
2942 else if (new_offset < rdev->data_offset)
2943 mddev->reshape_backwards = 0;
2944
2945 return len;
2946}
2947static struct rdev_sysfs_entry rdev_new_offset =
2948__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2949
83303b61 2950static ssize_t
3cb03002 2951rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2952{
dd8ac336 2953 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2954}
2955
c5d79adb
N		 2956static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2957{
2958 /* check if two start/length pairs overlap */
2959 if (s1+l1 <= s2)
2960 return 0;
2961 if (s2+l2 <= s1)
2962 return 0;
2963 return 1;
2964}
2965
b522adcd
DW		 2966static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2967{
2968 unsigned long long blocks;
2969 sector_t new;
2970
2971 if (strict_strtoull(buf, 10, &blocks) < 0)
2972 return -EINVAL;
2973
2974 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2975 return -EINVAL; /* sector conversion overflow */
2976
2977 new = blocks * 2;
2978 if (new != blocks * 2)
2979 return -EINVAL; /* unsigned long long to sector_t overflow */
2980
2981 *sectors = new;
2982 return 0;
2983}
2984
83303b61 2985static ssize_t
3cb03002 2986rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2987{
fd01b88c 2988 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2989 sector_t oldsectors = rdev->sectors;
b522adcd 2990 sector_t sectors;
27c529bb 2991
b522adcd 2992 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2993 return -EINVAL;
c6563a8c
N		 2994 if (rdev->data_offset != rdev->new_data_offset)
2995 return -EINVAL; /* too confusing */
0cd17fec 2996 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2997 if (my_mddev->persistent) {
dd8ac336
AN		 2998 sectors = super_types[my_mddev->major_version].
2999 rdev_size_change(rdev, sectors);
3000 if (!sectors)
0cd17fec 3001 return -EBUSY;
dd8ac336 3002 } else if (!sectors)
77304d2a 3003 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 3004 rdev->data_offset;
a6468539
N		 3005 if (!my_mddev->pers->resize)
3006 /* Cannot change size for RAID0 or Linear etc */
3007 return -EINVAL;
0cd17fec 3008 }
dd8ac336 3009 if (sectors < my_mddev->dev_sectors)
7d3c6f87 3010 return -EINVAL; /* component must fit device */
0cd17fec 3011
dd8ac336
AN		 3012 rdev->sectors = sectors;
3013 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 3014 /* need to check that all other rdevs with the same ->bdev
3015 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 3016 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 3017 * we have to change it back, we will have the lock again.
3018 */
fd01b88c 3019 struct mddev *mddev;
c5d79adb 3020 int overlap = 0;
159ec1fc 3021 struct list_head *tmp;
c5d79adb 3022
27c529bb 3023 mddev_unlock(my_mddev);
29ac4aa3 3024 for_each_mddev(mddev, tmp) {
3cb03002 3025 struct md_rdev *rdev2;
c5d79adb
N		 3026
3027 mddev_lock(mddev);
dafb20fa 3028 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 3029 if (rdev->bdev == rdev2->bdev &&
3030 rdev != rdev2 &&
3031 overlaps(rdev->data_offset, rdev->sectors,
3032 rdev2->data_offset,
3033 rdev2->sectors)) {
c5d79adb
N		 3034 overlap = 1;
3035 break;
3036 }
3037 mddev_unlock(mddev);
3038 if (overlap) {
3039 mddev_put(mddev);
3040 break;
3041 }
3042 }
27c529bb 3043 mddev_lock(my_mddev);
c5d79adb
N		 3044 if (overlap) {
3045 /* Someone else could have slipped in a size
3046 * change here, but doing so is just silly.
dd8ac336 3047 * We put oldsectors back because we *know* it is
c5d79adb
N		 3048 * safe, and trust userspace not to race with
3049 * itself
3050 */
dd8ac336 3051 rdev->sectors = oldsectors;
c5d79adb
N		 3052 return -EBUSY;
3053 }
3054 }
83303b61
N		 3055 return len;
3056}
3057
3058static struct rdev_sysfs_entry rdev_size =
80ca3a44 3059__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 3060
06e3c817 3061
3cb03002 3062static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 3063{
3064 unsigned long long recovery_start = rdev->recovery_offset;
3065
3066 if (test_bit(In_sync, &rdev->flags) ||
3067 recovery_start == MaxSector)
3068 return sprintf(page, "none\n");
3069
3070 return sprintf(page, "%llu\n", recovery_start);
3071}
3072
3cb03002 3073static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 3074{
3075 unsigned long long recovery_start;
3076
3077 if (cmd_match(buf, "none"))
3078 recovery_start = MaxSector;
3079 else if (strict_strtoull(buf, 10, &recovery_start))
3080 return -EINVAL;
3081
3082 if (rdev->mddev->pers &&
3083 rdev->raid_disk >= 0)
3084 return -EBUSY;
3085
3086 rdev->recovery_offset = recovery_start;
3087 if (recovery_start == MaxSector)
3088 set_bit(In_sync, &rdev->flags);
3089 else
3090 clear_bit(In_sync, &rdev->flags);
3091 return len;
3092}
3093
3094static struct rdev_sysfs_entry rdev_recovery_start =
3095__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3096
16c791a5
N		 3097
3098static ssize_t
3099badblocks_show(struct badblocks *bb, char *page, int unack);
3100static ssize_t
3101badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3102
3cb03002 3103static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3104{
3105 return badblocks_show(&rdev->badblocks, page, 0);
3106}
3cb03002 3107static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 3108{
de393cde
N		 3109 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3110 /* Maybe that ack was all we needed */
3111 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3112 wake_up(&rdev->blocked_wait);
3113 return rv;
16c791a5
N		 3114}
3115static struct rdev_sysfs_entry rdev_bad_blocks =
3116__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3117
3118
3cb03002 3119static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3120{
3121 return badblocks_show(&rdev->badblocks, page, 1);
3122}
3cb03002 3123static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 3124{
3125 return badblocks_store(&rdev->badblocks, page, len, 1);
3126}
3127static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3128__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3129
86e6ffdd
N		 3130static struct attribute *rdev_default_attrs[] = {
3131 &rdev_state.attr,
4dbcdc75 3132 &rdev_errors.attr,
014236d2 3133 &rdev_slot.attr,
93c8cad0 3134 &rdev_offset.attr,
c6563a8c 3135 &rdev_new_offset.attr,
83303b61 3136 &rdev_size.attr,
06e3c817 3137 &rdev_recovery_start.attr,
16c791a5
N		 3138 &rdev_bad_blocks.attr,
3139 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 3140 NULL,
3141};
3142static ssize_t
3143rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3144{
3145 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3146 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
fd01b88c 3147 struct mddev *mddev = rdev->mddev;
27c529bb 3148 ssize_t rv;
86e6ffdd
N		 3149
3150 if (!entry->show)
3151 return -EIO;
27c529bb
N		 3152
3153 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3154 if (!rv) {
3155 if (rdev->mddev == NULL)
3156 rv = -EBUSY;
3157 else
3158 rv = entry->show(rdev, page);
3159 mddev_unlock(mddev);
3160 }
3161 return rv;
86e6ffdd
N		 3162}
3163
3164static ssize_t
3165rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3166 const char *page, size_t length)
3167{
3168 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3169 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3170 ssize_t rv;
fd01b88c 3171 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3172
3173 if (!entry->store)
3174 return -EIO;
67463acb
N		 3175 if (!capable(CAP_SYS_ADMIN))
3176 return -EACCES;
27c529bb 3177 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3178 if (!rv) {
27c529bb
N		 3179 if (rdev->mddev == NULL)
3180 rv = -EBUSY;
3181 else
3182 rv = entry->store(rdev, page, length);
6a51830e 3183 mddev_unlock(mddev);
ca388059
N		 3184 }
3185 return rv;
86e6ffdd
N		 3186}
3187
3188static void rdev_free(struct kobject *ko)
3189{
3cb03002 3190 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3191 kfree(rdev);
3192}
52cf25d0 3193static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3194 .show = rdev_attr_show,
3195 .store = rdev_attr_store,
3196};
3197static struct kobj_type rdev_ktype = {
3198 .release = rdev_free,
3199 .sysfs_ops = &rdev_sysfs_ops,
3200 .default_attrs = rdev_default_attrs,
3201};
3202
3cb03002 3203int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3204{
3205 rdev->desc_nr = -1;
3206 rdev->saved_raid_disk = -1;
3207 rdev->raid_disk = -1;
3208 rdev->flags = 0;
3209 rdev->data_offset = 0;
c6563a8c 3210 rdev->new_data_offset = 0;
e8bb9a83
N		 3211 rdev->sb_events = 0;
3212 rdev->last_read_error.tv_sec = 0;
3213 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3214 rdev->sb_loaded = 0;
3215 rdev->bb_page = NULL;
e8bb9a83
N		 3216 atomic_set(&rdev->nr_pending, 0);
3217 atomic_set(&rdev->read_errors, 0);
3218 atomic_set(&rdev->corrected_errors, 0);
3219
3220 INIT_LIST_HEAD(&rdev->same_set);
3221 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3222
3223 /* Add space to store bad block list.
3224 * This reserves the space even on arrays where it cannot
3225 * be used - I wonder if that matters
3226 */
3227 rdev->badblocks.count = 0;
486adf72 3228 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
2230dfe4
N		 3229 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3230 seqlock_init(&rdev->badblocks.lock);
3231 if (rdev->badblocks.page == NULL)
3232 return -ENOMEM;
3233
3234 return 0;
e8bb9a83
N		 3235}
3236EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3237/*
3238 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3239 *
3240 * mark the device faulty if:
3241 *
3242 * - the device is nonexistent (zero size)
3243 * - the device has no valid superblock
3244 *
3245 * a faulty rdev _never_ has rdev->sb set.
3246 */
3cb03002 3247static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3248{
3249 char b[BDEVNAME_SIZE];
3250 int err;
3cb03002 3251 struct md_rdev *rdev;
1da177e4
LT		 3252 sector_t size;
3253
9ffae0cf 3254 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3255 if (!rdev) {
3256 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3257 return ERR_PTR(-ENOMEM);
3258 }
1da177e4 3259
2230dfe4
N		 3260 err = md_rdev_init(rdev);
3261 if (err)
3262 goto abort_free;
3263 err = alloc_disk_sb(rdev);
3264 if (err)
1da177e4
LT		 3265 goto abort_free;
3266
c5d79adb 3267 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3268 if (err)
3269 goto abort_free;
3270
f9cb074b 3271 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3272
77304d2a 3273 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT		 3274 if (!size) {
3275 printk(KERN_WARNING
3276 "md: %s has zero or unknown size, marking faulty!\n",
3277 bdevname(rdev->bdev,b));
3278 err = -EINVAL;
3279 goto abort_free;
3280 }
3281
3282 if (super_format >= 0) {
3283 err = super_types[super_format].
3284 load_super(rdev, NULL, super_minor);
3285 if (err == -EINVAL) {
df968c4e
N		 3286 printk(KERN_WARNING
3287 "md: %s does not have a valid v%d.%d "
3288 "superblock, not importing!\n",
3289 bdevname(rdev->bdev,b),
3290 super_format, super_minor);
1da177e4
LT		 3291 goto abort_free;
3292 }
3293 if (err < 0) {
3294 printk(KERN_WARNING
3295 "md: could not read %s's sb, not importing!\n",
3296 bdevname(rdev->bdev,b));
3297 goto abort_free;
3298 }
3299 }
6bfe0b49 3300
1da177e4
LT		 3301 return rdev;
3302
3303abort_free:
2699b672
N		 3304 if (rdev->bdev)
3305 unlock_rdev(rdev);
545c8795 3306 md_rdev_clear(rdev);
1da177e4
LT		 3307 kfree(rdev);
3308 return ERR_PTR(err);
3309}
3310
3311/*
3312 * Check a full RAID array for plausibility
3313 */
3314
3315
fd01b88c 3316static void analyze_sbs(struct mddev * mddev)
1da177e4
LT		 3317{
3318 int i;
3cb03002 3319 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3320 char b[BDEVNAME_SIZE];
3321
3322 freshest = NULL;
dafb20fa 3323 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3324 switch (super_types[mddev->major_version].
3325 load_super(rdev, freshest, mddev->minor_version)) {
3326 case 1:
3327 freshest = rdev;
3328 break;
3329 case 0:
3330 break;
3331 default:
3332 printk( KERN_ERR \
3333 "md: fatal superblock inconsistency in %s"
3334 " -- removing from array\n",
3335 bdevname(rdev->bdev,b));
3336 kick_rdev_from_array(rdev);
3337 }
3338
3339
3340 super_types[mddev->major_version].
3341 validate_super(mddev, freshest);
3342
3343 i = 0;
dafb20fa 3344 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3345 if (mddev->max_disks &&
3346 (rdev->desc_nr >= mddev->max_disks ||
3347 i > mddev->max_disks)) {
de01dfad
N		 3348 printk(KERN_WARNING
3349 "md: %s: %s: only %d devices permitted\n",
3350 mdname(mddev), bdevname(rdev->bdev, b),
3351 mddev->max_disks);
3352 kick_rdev_from_array(rdev);
3353 continue;
3354 }
1da177e4
LT		 3355 if (rdev != freshest)
3356 if (super_types[mddev->major_version].
3357 validate_super(mddev, rdev)) {
3358 printk(KERN_WARNING "md: kicking non-fresh %s"
3359 " from array!\n",
3360 bdevname(rdev->bdev,b));
3361 kick_rdev_from_array(rdev);
3362 continue;
3363 }
3364 if (mddev->level == LEVEL_MULTIPATH) {
3365 rdev->desc_nr = i++;
3366 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3367 set_bit(In_sync, &rdev->flags);
5e5e3e78 3368 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3369 rdev->raid_disk = -1;
3370 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3371 }
3372 }
1da177e4
LT		 3373}
3374
72e02075
N		 3375/* Read a fixed-point number.
3376 * Numbers in sysfs attributes should be in "standard" units where
3377 * possible, so time should be in seconds.
3378 * However we internally use a a much smaller unit such as
3379 * milliseconds or jiffies.
3380 * This function takes a decimal number with a possible fractional
3381 * component, and produces an integer which is the result of
3382 * multiplying that number by 10^'scale'.
3383 * all without any floating-point arithmetic.
3384 */
3385int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3386{
3387 unsigned long result = 0;
3388 long decimals = -1;
3389 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3390 if (*cp == '.')
3391 decimals = 0;
3392 else if (decimals < scale) {
3393 unsigned int value;
3394 value = *cp - '0';
3395 result = result * 10 + value;
3396 if (decimals >= 0)
3397 decimals++;
3398 }
3399 cp++;
3400 }
3401 if (*cp == '\n')
3402 cp++;
3403 if (*cp)
3404 return -EINVAL;
3405 if (decimals < 0)
3406 decimals = 0;
3407 while (decimals < scale) {
3408 result *= 10;
3409 decimals ++;
3410 }
3411 *res = result;
3412 return 0;
3413}
3414
3415
19052c0e
N		 3416static void md_safemode_timeout(unsigned long data);
3417
16f17b39 3418static ssize_t
fd01b88c 3419safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3420{
3421 int msec = (mddev->safemode_delay*1000)/HZ;
3422 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3423}
3424static ssize_t
fd01b88c 3425safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3426{
16f17b39 3427 unsigned long msec;
97ce0a7f 3428
72e02075 3429 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3430 return -EINVAL;
16f17b39
N		 3431 if (msec == 0)
3432 mddev->safemode_delay = 0;
3433 else {
19052c0e 3434 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 3435 mddev->safemode_delay = (msec*HZ)/1000;
3436 if (mddev->safemode_delay == 0)
3437 mddev->safemode_delay = 1;
19052c0e
N		 3438 if (mddev->safemode_delay < old_delay)
3439 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 3440 }
3441 return len;
3442}
3443static struct md_sysfs_entry md_safe_delay =
80ca3a44 3444__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3445
eae1701f 3446static ssize_t
fd01b88c 3447level_show(struct mddev *mddev, char *page)
eae1701f 3448{
84fc4b56 3449 struct md_personality *p = mddev->pers;
d9d166c2 3450 if (p)
eae1701f 3451 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 3452 else if (mddev->clevel[0])
3453 return sprintf(page, "%s\n", mddev->clevel);
3454 else if (mddev->level != LEVEL_NONE)
3455 return sprintf(page, "%d\n", mddev->level);
3456 else
3457 return 0;
eae1701f
N		 3458}
3459
d9d166c2 3460static ssize_t
fd01b88c 3461level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3462{
f2859af6 3463 char clevel[16];
20a49ff6 3464 ssize_t rv = len;
84fc4b56 3465 struct md_personality *pers;
f2859af6 3466 long level;
245f46c2 3467 void *priv;
3cb03002 3468 struct md_rdev *rdev;
245f46c2
N		 3469
3470 if (mddev->pers == NULL) {
3471 if (len == 0)
3472 return 0;
3473 if (len >= sizeof(mddev->clevel))
3474 return -ENOSPC;
3475 strncpy(mddev->clevel, buf, len);
3476 if (mddev->clevel[len-1] == '\n')
3477 len--;
3478 mddev->clevel[len] = 0;
3479 mddev->level = LEVEL_NONE;
3480 return rv;
3481 }
3482
3483 /* request to change the personality. Need to ensure:
3484 * - array is not engaged in resync/recovery/reshape
3485 * - old personality can be suspended
3486 * - new personality will access other array.
3487 */
3488
bb4f1e9d
N		 3489 if (mddev->sync_thread ||
3490 mddev->reshape_position != MaxSector ||
3491 mddev->sysfs_active)
d9d166c2 3492 return -EBUSY;
245f46c2
N		 3493
3494 if (!mddev->pers->quiesce) {
3495 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3496 mdname(mddev), mddev->pers->name);
3497 return -EINVAL;
3498 }
3499
3500 /* Now find the new personality */
f2859af6 3501 if (len == 0 || len >= sizeof(clevel))
245f46c2 3502 return -EINVAL;
f2859af6
DW		 3503 strncpy(clevel, buf, len);
3504 if (clevel[len-1] == '\n')
d9d166c2 3505 len--;
f2859af6
DW		 3506 clevel[len] = 0;
3507 if (strict_strtol(clevel, 10, &level))
3508 level = LEVEL_NONE;
245f46c2 3509
f2859af6
DW		 3510 if (request_module("md-%s", clevel) != 0)
3511 request_module("md-level-%s", clevel);
245f46c2 3512 spin_lock(&pers_lock);
f2859af6 3513 pers = find_pers(level, clevel);
245f46c2
N		 3514 if (!pers || !try_module_get(pers->owner)) {
3515 spin_unlock(&pers_lock);
f2859af6 3516 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3517 return -EINVAL;
3518 }
3519 spin_unlock(&pers_lock);
3520
3521 if (pers == mddev->pers) {
3522 /* Nothing to do! */
3523 module_put(pers->owner);
3524 return rv;
3525 }
3526 if (!pers->takeover) {
3527 module_put(pers->owner);
3528 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3529 mdname(mddev), clevel);
245f46c2
N		 3530 return -EINVAL;
3531 }
3532
dafb20fa 3533 rdev_for_each(rdev, mddev)
e93f68a1
N		 3534 rdev->new_raid_disk = rdev->raid_disk;
3535
245f46c2
N		 3536 /* ->takeover must set new_* and/or delta_disks
3537 * if it succeeds, and may set them when it fails.
3538 */
3539 priv = pers->takeover(mddev);
3540 if (IS_ERR(priv)) {
3541 mddev->new_level = mddev->level;
3542 mddev->new_layout = mddev->layout;
664e7c41 3543 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3544 mddev->raid_disks -= mddev->delta_disks;
3545 mddev->delta_disks = 0;
2c810cdd 3546 mddev->reshape_backwards = 0;
245f46c2
N		 3547 module_put(pers->owner);
3548 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3549 mdname(mddev), clevel);
245f46c2
N		 3550 return PTR_ERR(priv);
3551 }
3552
3553 /* Looks like we have a winner */
3554 mddev_suspend(mddev);
3555 mddev->pers->stop(mddev);
a64c876f
N		 3556
3557 if (mddev->pers->sync_request == NULL &&
3558 pers->sync_request != NULL) {
3559 /* need to add the md_redundancy_group */
3560 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3561 printk(KERN_WARNING
3562 "md: cannot register extra attributes for %s\n",
3563 mdname(mddev));
19fdb9ee 3564 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3565 }
3566 if (mddev->pers->sync_request != NULL &&
3567 pers->sync_request == NULL) {
3568 /* need to remove the md_redundancy_group */
3569 if (mddev->to_remove == NULL)
3570 mddev->to_remove = &md_redundancy_group;
3571 }
3572
54071b38
TM		 3573 if (mddev->pers->sync_request == NULL &&
3574 mddev->external) {
3575 /* We are converting from a no-redundancy array
3576 * to a redundancy array and metadata is managed
3577 * externally so we need to be sure that writes
3578 * won't block due to a need to transition
3579 * clean->dirty
3580 * until external management is started.
3581 */
3582 mddev->in_sync = 0;
3583 mddev->safemode_delay = 0;
3584 mddev->safemode = 0;
3585 }
3586
dafb20fa 3587 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3588 if (rdev->raid_disk < 0)
3589 continue;
bf2cb0da 3590 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3591 rdev->new_raid_disk = -1;
3592 if (rdev->new_raid_disk == rdev->raid_disk)
3593 continue;
36fad858 3594 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3595 }
dafb20fa 3596 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3597 if (rdev->raid_disk < 0)
3598 continue;
3599 if (rdev->new_raid_disk == rdev->raid_disk)
3600 continue;
3601 rdev->raid_disk = rdev->new_raid_disk;
3602 if (rdev->raid_disk < 0)
3a981b03 3603 clear_bit(In_sync, &rdev->flags);
e93f68a1 3604 else {
36fad858
NK		 3605 if (sysfs_link_rdev(mddev, rdev))
3606 printk(KERN_WARNING "md: cannot register rd%d"
3607 " for %s after level change\n",
3608 rdev->raid_disk, mdname(mddev));
3a981b03 3609 }
e93f68a1
N		 3610 }
3611
3612 module_put(mddev->pers->owner);
245f46c2
N		 3613 mddev->pers = pers;
3614 mddev->private = priv;
3615 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3616 mddev->level = mddev->new_level;
3617 mddev->layout = mddev->new_layout;
664e7c41 3618 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3619 mddev->delta_disks = 0;
2c810cdd 3620 mddev->reshape_backwards = 0;
fee68723 3621 mddev->degraded = 0;
9af204cf
TM		 3622 if (mddev->pers->sync_request == NULL) {
3623 /* this is now an array without redundancy, so
3624 * it must always be in_sync
3625 */
3626 mddev->in_sync = 1;
3627 del_timer_sync(&mddev->safemode_timer);
3628 }
8e43aac1 3629 blk_set_stacking_limits(&mddev->queue->limits);
245f46c2 3630 pers->run(mddev);
245f46c2 3631 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3632 mddev_resume(mddev);
5cac7861 3633 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3634 md_new_event(mddev);
d9d166c2
N		 3635 return rv;
3636}
3637
3638static struct md_sysfs_entry md_level =
80ca3a44 3639__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3640
d4dbd025
N		 3641
3642static ssize_t
fd01b88c 3643layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3644{
3645 /* just a number, not meaningful for all levels */
08a02ecd
N		 3646 if (mddev->reshape_position != MaxSector &&
3647 mddev->layout != mddev->new_layout)
3648 return sprintf(page, "%d (%d)\n",
3649 mddev->new_layout, mddev->layout);
d4dbd025
N		 3650 return sprintf(page, "%d\n", mddev->layout);
3651}
3652
3653static ssize_t
fd01b88c 3654layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N		 3655{
3656 char *e;
3657 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3658
3659 if (!*buf || (*e && *e != '\n'))
3660 return -EINVAL;
3661
b3546035
N		 3662 if (mddev->pers) {
3663 int err;
50ac168a 3664 if (mddev->pers->check_reshape == NULL)
b3546035 3665 return -EBUSY;
597a711b 3666 mddev->new_layout = n;
50ac168a 3667 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3668 if (err) {
3669 mddev->new_layout = mddev->layout;
b3546035 3670 return err;
597a711b 3671 }
b3546035 3672 } else {
08a02ecd 3673 mddev->new_layout = n;
b3546035
N		 3674 if (mddev->reshape_position == MaxSector)
3675 mddev->layout = n;
3676 }
d4dbd025
N		 3677 return len;
3678}
3679static struct md_sysfs_entry md_layout =
80ca3a44 3680__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3681
3682
eae1701f 3683static ssize_t
fd01b88c 3684raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3685{
bb636547
N		 3686 if (mddev->raid_disks == 0)
3687 return 0;
08a02ecd
N		 3688 if (mddev->reshape_position != MaxSector &&
3689 mddev->delta_disks != 0)
3690 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3691 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3692 return sprintf(page, "%d\n", mddev->raid_disks);
3693}
3694
fd01b88c 3695static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3696
3697static ssize_t
fd01b88c 3698raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3699{
da943b99
N		 3700 char *e;
3701 int rv = 0;
3702 unsigned long n = simple_strtoul(buf, &e, 10);
3703
3704 if (!*buf || (*e && *e != '\n'))
3705 return -EINVAL;
3706
3707 if (mddev->pers)
3708 rv = update_raid_disks(mddev, n);
08a02ecd 3709 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3710 struct md_rdev *rdev;
08a02ecd 3711 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c
N		 3712
3713 rdev_for_each(rdev, mddev) {
3714 if (olddisks < n &&
3715 rdev->data_offset < rdev->new_data_offset)
3716 return -EINVAL;
3717 if (olddisks > n &&
3718 rdev->data_offset > rdev->new_data_offset)
3719 return -EINVAL;
3720 }
08a02ecd
N		 3721 mddev->delta_disks = n - olddisks;
3722 mddev->raid_disks = n;
2c810cdd 3723 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3724 } else
da943b99
N		 3725 mddev->raid_disks = n;
3726 return rv ? rv : len;
3727}
3728static struct md_sysfs_entry md_raid_disks =
80ca3a44 3729__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3730
3b34380a 3731static ssize_t
fd01b88c 3732chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3733{
08a02ecd 3734 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3735 mddev->chunk_sectors != mddev->new_chunk_sectors)
3736 return sprintf(page, "%d (%d)\n",
3737 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3738 mddev->chunk_sectors << 9);
3739 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3740}
3741
3742static ssize_t
fd01b88c 3743chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3744{
3b34380a
N		 3745 char *e;
3746 unsigned long n = simple_strtoul(buf, &e, 10);
3747
3b34380a
N		 3748 if (!*buf || (*e && *e != '\n'))
3749 return -EINVAL;
3750
b3546035
N		 3751 if (mddev->pers) {
3752 int err;
50ac168a 3753 if (mddev->pers->check_reshape == NULL)
b3546035 3754 return -EBUSY;
597a711b 3755 mddev->new_chunk_sectors = n >> 9;
50ac168a 3756 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3757 if (err) {
3758 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3759 return err;
597a711b 3760 }
b3546035 3761 } else {
664e7c41 3762 mddev->new_chunk_sectors = n >> 9;
b3546035 3763 if (mddev->reshape_position == MaxSector)
9d8f0363 3764 mddev->chunk_sectors = n >> 9;
b3546035 3765 }
3b34380a
N		 3766 return len;
3767}
3768static struct md_sysfs_entry md_chunk_size =
80ca3a44 3769__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3770
a94213b1 3771static ssize_t
fd01b88c 3772resync_start_show(struct mddev *mddev, char *page)
a94213b1 3773{
d1a7c503
N		 3774 if (mddev->recovery_cp == MaxSector)
3775 return sprintf(page, "none\n");
a94213b1
N		 3776 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3777}
3778
3779static ssize_t
fd01b88c 3780resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3781{
a94213b1
N		 3782 char *e;
3783 unsigned long long n = simple_strtoull(buf, &e, 10);
3784
b098636c 3785 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
a94213b1 3786 return -EBUSY;
06e3c817
DW		 3787 if (cmd_match(buf, "none"))
3788 n = MaxSector;
3789 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3790 return -EINVAL;
3791
3792 mddev->recovery_cp = n;
db07d85e
N		 3793 if (mddev->pers)
3794 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
a94213b1
N		 3795 return len;
3796}
3797static struct md_sysfs_entry md_resync_start =
80ca3a44 3798__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3799
9e653b63
N		 3800/*
3801 * The array state can be:
3802 *
3803 * clear
3804 * No devices, no size, no level
3805 * Equivalent to STOP_ARRAY ioctl
3806 * inactive
3807 * May have some settings, but array is not active
3808 * all IO results in error
3809 * When written, doesn't tear down array, but just stops it
3810 * suspended (not supported yet)
3811 * All IO requests will block. The array can be reconfigured.
910d8cb3 3812 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3813 * readonly
3814 * no resync can happen. no superblocks get written.
3815 * write requests fail
3816 * read-auto
3817 * like readonly, but behaves like 'clean' on a write request.
3818 *
3819 * clean - no pending writes, but otherwise active.
3820 * When written to inactive array, starts without resync
3821 * If a write request arrives then
3822 * if metadata is known, mark 'dirty' and switch to 'active'.
3823 * if not known, block and switch to write-pending
3824 * If written to an active array that has pending writes, then fails.
3825 * active
3826 * fully active: IO and resync can be happening.
3827 * When written to inactive array, starts with resync
3828 *
3829 * write-pending
3830 * clean, but writes are blocked waiting for 'active' to be written.
3831 *
3832 * active-idle
3833 * like active, but no writes have been seen for a while (100msec).
3834 *
3835 */
3836enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3837 write_pending, active_idle, bad_word};
05381954 3838static char *array_states[] = {
9e653b63
N		 3839 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3840 "write-pending", "active-idle", NULL };
3841
3842static int match_word(const char *word, char **list)
3843{
3844 int n;
3845 for (n=0; list[n]; n++)
3846 if (cmd_match(word, list[n]))
3847 break;
3848 return n;
3849}
3850
3851static ssize_t
fd01b88c 3852array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3853{
3854 enum array_state st = inactive;
3855
3856 if (mddev->pers)
3857 switch(mddev->ro) {
3858 case 1:
3859 st = readonly;
3860 break;
3861 case 2:
3862 st = read_auto;
3863 break;
3864 case 0:
3865 if (mddev->in_sync)
3866 st = clean;
070dc6dd 3867 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3868 st = write_pending;
9e653b63
N		 3869 else if (mddev->safemode)
3870 st = active_idle;
3871 else
3872 st = active;
3873 }
3874 else {
3875 if (list_empty(&mddev->disks) &&
3876 mddev->raid_disks == 0 &&
58c0fed4 3877 mddev->dev_sectors == 0)
9e653b63
N		 3878 st = clear;
3879 else
3880 st = inactive;
3881 }
3882 return sprintf(page, "%s\n", array_states[st]);
3883}
3884
a05b7ea0
N		 3885static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3886static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
fd01b88c
N		 3887static int do_md_run(struct mddev * mddev);
3888static int restart_array(struct mddev *mddev);
9e653b63
N		 3889
3890static ssize_t
fd01b88c 3891array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63
N		 3892{
3893 int err = -EINVAL;
3894 enum array_state st = match_word(buf, array_states);
3895 switch(st) {
3896 case bad_word:
3897 break;
3898 case clear:
3899 /* stopping an active array */
a05b7ea0 3900 err = do_md_stop(mddev, 0, NULL);
9e653b63
N		 3901 break;
3902 case inactive:
3903 /* stopping an active array */
90cf195d 3904 if (mddev->pers)
a05b7ea0 3905 err = do_md_stop(mddev, 2, NULL);
90cf195d 3906 else
e691063a 3907 err = 0; /* already inactive */
9e653b63
N		 3908 break;
3909 case suspended:
3910 break; /* not supported yet */
3911 case readonly:
3912 if (mddev->pers)
a05b7ea0 3913 err = md_set_readonly(mddev, NULL);
9e653b63
N		 3914 else {
3915 mddev->ro = 1;
648b629e 3916 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3917 err = do_md_run(mddev);
3918 }
3919 break;
3920 case read_auto:
9e653b63 3921 if (mddev->pers) {
80268ee9 3922 if (mddev->ro == 0)
a05b7ea0 3923 err = md_set_readonly(mddev, NULL);
80268ee9 3924 else if (mddev->ro == 1)
648b629e
N		 3925 err = restart_array(mddev);
3926 if (err == 0) {
3927 mddev->ro = 2;
3928 set_disk_ro(mddev->gendisk, 0);
3929 }
9e653b63
N		 3930 } else {
3931 mddev->ro = 2;
3932 err = do_md_run(mddev);
3933 }
3934 break;
3935 case clean:
3936 if (mddev->pers) {
3937 restart_array(mddev);
3938 spin_lock_irq(&mddev->write_lock);
3939 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3940 if (mddev->in_sync == 0) {
3941 mddev->in_sync = 1;
31a59e34
N		 3942 if (mddev->safemode == 1)
3943 mddev->safemode = 0;
070dc6dd 3944 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3945 }
3946 err = 0;
3947 } else
3948 err = -EBUSY;
9e653b63 3949 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3950 } else
3951 err = -EINVAL;
9e653b63
N		 3952 break;
3953 case active:
3954 if (mddev->pers) {
3955 restart_array(mddev);
070dc6dd 3956 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3957 wake_up(&mddev->sb_wait);
3958 err = 0;
3959 } else {
3960 mddev->ro = 0;
648b629e 3961 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3962 err = do_md_run(mddev);
3963 }
3964 break;
3965 case write_pending:
3966 case active_idle:
3967 /* these cannot be set */
3968 break;
3969 }
3970 if (err)
3971 return err;
0fd62b86 3972 else {
1d23f178
N		 3973 if (mddev->hold_active == UNTIL_IOCTL)
3974 mddev->hold_active = 0;
00bcb4ac 3975 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3976 return len;
0fd62b86 3977 }
9e653b63 3978}
80ca3a44
N		 3979static struct md_sysfs_entry md_array_state =
3980__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3981
1e50915f 3982static ssize_t
fd01b88c 3983max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3984 return sprintf(page, "%d\n",
3985 atomic_read(&mddev->max_corr_read_errors));
3986}
3987
3988static ssize_t
fd01b88c 3989max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB		 3990{
3991 char *e;
3992 unsigned long n = simple_strtoul(buf, &e, 10);
3993
3994 if (*buf && (*e == 0 || *e == '\n')) {
3995 atomic_set(&mddev->max_corr_read_errors, n);
3996 return len;
3997 }
3998 return -EINVAL;
3999}
4000
4001static struct md_sysfs_entry max_corr_read_errors =
4002__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4003 max_corrected_read_errors_store);
4004
6d7ff738 4005static ssize_t
fd01b88c 4006null_show(struct mddev *mddev, char *page)
6d7ff738
N		 4007{
4008 return -EINVAL;
4009}
4010
4011static ssize_t
fd01b88c 4012new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 4013{
4014 /* buf must be %d:%d\n? giving major and minor numbers */
4015 /* The new device is added to the array.
4016 * If the array has a persistent superblock, we read the
4017 * superblock to initialise info and check validity.
4018 * Otherwise, only checking done is that in bind_rdev_to_array,
4019 * which mainly checks size.
4020 */
4021 char *e;
4022 int major = simple_strtoul(buf, &e, 10);
4023 int minor;
4024 dev_t dev;
3cb03002 4025 struct md_rdev *rdev;
6d7ff738
N		 4026 int err;
4027
4028 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4029 return -EINVAL;
4030 minor = simple_strtoul(e+1, &e, 10);
4031 if (*e && *e != '\n')
4032 return -EINVAL;
4033 dev = MKDEV(major, minor);
4034 if (major != MAJOR(dev) ||
4035 minor != MINOR(dev))
4036 return -EOVERFLOW;
4037
4038
4039 if (mddev->persistent) {
4040 rdev = md_import_device(dev, mddev->major_version,
4041 mddev->minor_version);
4042 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 4043 struct md_rdev *rdev0
4044 = list_entry(mddev->disks.next,
4045 struct md_rdev, same_set);
6d7ff738
N		 4046 err = super_types[mddev->major_version]
4047 .load_super(rdev, rdev0, mddev->minor_version);
4048 if (err < 0)
4049 goto out;
4050 }
c5d79adb
N		 4051 } else if (mddev->external)
4052 rdev = md_import_device(dev, -2, -1);
4053 else
6d7ff738
N		 4054 rdev = md_import_device(dev, -1, -1);
4055
4056 if (IS_ERR(rdev))
4057 return PTR_ERR(rdev);
4058 err = bind_rdev_to_array(rdev, mddev);
4059 out:
4060 if (err)
4061 export_rdev(rdev);
4062 return err ? err : len;
4063}
4064
4065static struct md_sysfs_entry md_new_device =
80ca3a44 4066__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 4067
9b1d1dac 4068static ssize_t
fd01b88c 4069bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 4070{
4071 char *end;
4072 unsigned long chunk, end_chunk;
4073
4074 if (!mddev->bitmap)
4075 goto out;
4076 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4077 while (*buf) {
4078 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4079 if (buf == end) break;
4080 if (*end == '-') { /* range */
4081 buf = end + 1;
4082 end_chunk = simple_strtoul(buf, &end, 0);
4083 if (buf == end) break;
4084 }
4085 if (*end && !isspace(*end)) break;
4086 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 4087 buf = skip_spaces(end);
9b1d1dac
PC		 4088 }
4089 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4090out:
4091 return len;
4092}
4093
4094static struct md_sysfs_entry md_bitmap =
4095__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4096
a35b0d69 4097static ssize_t
fd01b88c 4098size_show(struct mddev *mddev, char *page)
a35b0d69 4099{
58c0fed4
AN		 4100 return sprintf(page, "%llu\n",
4101 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 4102}
4103
fd01b88c 4104static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 4105
4106static ssize_t
fd01b88c 4107size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 4108{
4109 /* If array is inactive, we can reduce the component size, but
4110 * not increase it (except from 0).
4111 * If array is active, we can try an on-line resize
4112 */
b522adcd
DW		 4113 sector_t sectors;
4114 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4115
58c0fed4
AN		 4116 if (err < 0)
4117 return err;
a35b0d69 4118 if (mddev->pers) {
58c0fed4 4119 err = update_size(mddev, sectors);
850b2b42 4120 md_update_sb(mddev, 1);
a35b0d69 4121 } else {
58c0fed4
AN		 4122 if (mddev->dev_sectors == 0 ||
4123 mddev->dev_sectors > sectors)
4124 mddev->dev_sectors = sectors;
a35b0d69
N		 4125 else
4126 err = -ENOSPC;
4127 }
4128 return err ? err : len;
4129}
4130
4131static struct md_sysfs_entry md_size =
80ca3a44 4132__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4133
8bb93aac 4134
83f0d77a 4135/* Metadata version.
e691063a
N		 4136 * This is one of
4137 * 'none' for arrays with no metadata (good luck...)
4138 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4139 * or N.M for internally known formats
4140 */
4141static ssize_t
fd01b88c 4142metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4143{
4144 if (mddev->persistent)
4145 return sprintf(page, "%d.%d\n",
4146 mddev->major_version, mddev->minor_version);
e691063a
N		 4147 else if (mddev->external)
4148 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4149 else
4150 return sprintf(page, "none\n");
4151}
4152
4153static ssize_t
fd01b88c 4154metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4155{
4156 int major, minor;
4157 char *e;
ea43ddd8
N		 4158 /* Changing the details of 'external' metadata is
4159 * always permitted. Otherwise there must be
4160 * no devices attached to the array.
4161 */
4162 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4163 ;
4164 else if (!list_empty(&mddev->disks))
8bb93aac
N		 4165 return -EBUSY;
4166
4167 if (cmd_match(buf, "none")) {
4168 mddev->persistent = 0;
e691063a
N		 4169 mddev->external = 0;
4170 mddev->major_version = 0;
4171 mddev->minor_version = 90;
4172 return len;
4173 }
4174 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4175 size_t namelen = len-9;
e691063a
N		 4176 if (namelen >= sizeof(mddev->metadata_type))
4177 namelen = sizeof(mddev->metadata_type)-1;
4178 strncpy(mddev->metadata_type, buf+9, namelen);
4179 mddev->metadata_type[namelen] = 0;
4180 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4181 mddev->metadata_type[--namelen] = 0;
4182 mddev->persistent = 0;
4183 mddev->external = 1;
8bb93aac
N		 4184 mddev->major_version = 0;
4185 mddev->minor_version = 90;
4186 return len;
4187 }
4188 major = simple_strtoul(buf, &e, 10);
4189 if (e==buf || *e != '.')
4190 return -EINVAL;
4191 buf = e+1;
4192 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4193 if (e==buf || (*e && *e != '\n') )
8bb93aac 4194 return -EINVAL;
50511da3 4195 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 4196 return -ENOENT;
4197 mddev->major_version = major;
4198 mddev->minor_version = minor;
4199 mddev->persistent = 1;
e691063a 4200 mddev->external = 0;
8bb93aac
N		 4201 return len;
4202}
4203
4204static struct md_sysfs_entry md_metadata =
80ca3a44 4205__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4206
24dd469d 4207static ssize_t
fd01b88c 4208action_show(struct mddev *mddev, char *page)
24dd469d 4209{
7eec314d 4210 char *type = "idle";
b6a9ce68
N		 4211 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4212 type = "frozen";
4213 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 4214 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 4215 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4216 type = "reshape";
4217 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 4218 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4219 type = "resync";
4220 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4221 type = "check";
4222 else
4223 type = "repair";
72a23c21 4224 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 4225 type = "recover";
4226 }
4227 return sprintf(page, "%s\n", type);
4228}
4229
4230static ssize_t
fd01b88c 4231action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4232{
7eec314d
N		 4233 if (!mddev->pers || !mddev->pers->sync_request)
4234 return -EINVAL;
4235
b6a9ce68
N		 4236 if (cmd_match(page, "frozen"))
4237 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4238 else
4239 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4240
4241 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 4242 if (mddev->sync_thread) {
4243 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 4244 md_reap_sync_thread(mddev);
7eec314d 4245 }
03c902e1
N		 4246 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4247 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4248 return -EBUSY;
72a23c21
NB		 4249 else if (cmd_match(page, "resync"))
4250 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4251 else if (cmd_match(page, "recover")) {
4252 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4253 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4254 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4255 int err;
4256 if (mddev->pers->start_reshape == NULL)
4257 return -EINVAL;
4258 err = mddev->pers->start_reshape(mddev);
4259 if (err)
4260 return err;
a99ac971 4261 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4262 } else {
bce74dac 4263 if (cmd_match(page, "check"))
7eec314d 4264 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4265 else if (!cmd_match(page, "repair"))
7eec314d
N		 4266 return -EINVAL;
4267 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4268 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4269 }
48c26ddc
N		 4270 if (mddev->ro == 2) {
4271 /* A write to sync_action is enough to justify
4272 * canceling read-auto mode
4273 */
4274 mddev->ro = 0;
4275 md_wakeup_thread(mddev->sync_thread);
4276 }
03c902e1 4277 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4278 md_wakeup_thread(mddev->thread);
00bcb4ac 4279 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4280 return len;
4281}
4282
9d88883e 4283static ssize_t
fd01b88c 4284mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4285{
4286 return sprintf(page, "%llu\n",
7f7583d4
JM		 4287 (unsigned long long)
4288 atomic64_read(&mddev->resync_mismatches));
9d88883e
N		 4289}
4290
80ca3a44
N		 4291static struct md_sysfs_entry md_scan_mode =
4292__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 4293
96de1e66 4294
80ca3a44 4295static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4296
88202a0c 4297static ssize_t
fd01b88c 4298sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4299{
4300 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4301 mddev->sync_speed_min ? "local": "system");
4302}
4303
4304static ssize_t
fd01b88c 4305sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4306{
4307 int min;
4308 char *e;
4309 if (strncmp(buf, "system", 6)==0) {
4310 mddev->sync_speed_min = 0;
4311 return len;
4312 }
4313 min = simple_strtoul(buf, &e, 10);
4314 if (buf == e || (*e && *e != '\n') || min <= 0)
4315 return -EINVAL;
4316 mddev->sync_speed_min = min;
4317 return len;
4318}
4319
4320static struct md_sysfs_entry md_sync_min =
4321__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4322
4323static ssize_t
fd01b88c 4324sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4325{
4326 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4327 mddev->sync_speed_max ? "local": "system");
4328}
4329
4330static ssize_t
fd01b88c 4331sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4332{
4333 int max;
4334 char *e;
4335 if (strncmp(buf, "system", 6)==0) {
4336 mddev->sync_speed_max = 0;
4337 return len;
4338 }
4339 max = simple_strtoul(buf, &e, 10);
4340 if (buf == e || (*e && *e != '\n') || max <= 0)
4341 return -EINVAL;
4342 mddev->sync_speed_max = max;
4343 return len;
4344}
4345
4346static struct md_sysfs_entry md_sync_max =
4347__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4348
d7f3d291 4349static ssize_t
fd01b88c 4350degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4351{
4352 return sprintf(page, "%d\n", mddev->degraded);
4353}
4354static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4355
90b08710 4356static ssize_t
fd01b88c 4357sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4358{
4359 return sprintf(page, "%d\n", mddev->parallel_resync);
4360}
4361
4362static ssize_t
fd01b88c 4363sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4364{
4365 long n;
4366
4367 if (strict_strtol(buf, 10, &n))
4368 return -EINVAL;
4369
4370 if (n != 0 && n != 1)
4371 return -EINVAL;
4372
4373 mddev->parallel_resync = n;
4374
4375 if (mddev->sync_thread)
4376 wake_up(&resync_wait);
4377
4378 return len;
4379}
4380
4381/* force parallel resync, even with shared block devices */
4382static struct md_sysfs_entry md_sync_force_parallel =
4383__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4384 sync_force_parallel_show, sync_force_parallel_store);
4385
88202a0c 4386static ssize_t
fd01b88c 4387sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4388{
4389 unsigned long resync, dt, db;
d1a7c503
N		 4390 if (mddev->curr_resync == 0)
4391 return sprintf(page, "none\n");
9687a60c
AN		 4392 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4393 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4394 if (!dt) dt++;
9687a60c
AN		 4395 db = resync - mddev->resync_mark_cnt;
4396 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4397}
4398
80ca3a44 4399static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4400
4401static ssize_t
fd01b88c 4402sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4403{
13ae864b 4404 unsigned long long max_sectors, resync;
88202a0c 4405
acb180b0
N		 4406 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4407 return sprintf(page, "none\n");
4408
72f36d59
N		 4409 if (mddev->curr_resync == 1 ||
4410 mddev->curr_resync == 2)
4411 return sprintf(page, "delayed\n");
4412
c804cdec
N		 4413 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4414 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 4415 max_sectors = mddev->resync_max_sectors;
88202a0c 4416 else
58c0fed4 4417 max_sectors = mddev->dev_sectors;
88202a0c 4418
acb180b0 4419 resync = mddev->curr_resync_completed;
13ae864b 4420 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4421}
4422
80ca3a44 4423static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 4424
5e96ee65 4425static ssize_t
fd01b88c 4426min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4427{
4428 return sprintf(page, "%llu\n",
4429 (unsigned long long)mddev->resync_min);
4430}
4431static ssize_t
fd01b88c 4432min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4433{
4434 unsigned long long min;
4435 if (strict_strtoull(buf, 10, &min))
4436 return -EINVAL;
4437 if (min > mddev->resync_max)
4438 return -EINVAL;
4439 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4440 return -EBUSY;
4441
4442 /* Must be a multiple of chunk_size */
9d8f0363 4443 if (mddev->chunk_sectors) {
2ac06c33 4444 sector_t temp = min;
9d8f0363 4445 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 4446 return -EINVAL;
4447 }
4448 mddev->resync_min = min;
4449
4450 return len;
4451}
4452
4453static struct md_sysfs_entry md_min_sync =
4454__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4455
c6207277 4456static ssize_t
fd01b88c 4457max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4458{
4459 if (mddev->resync_max == MaxSector)
4460 return sprintf(page, "max\n");
4461 else
4462 return sprintf(page, "%llu\n",
4463 (unsigned long long)mddev->resync_max);
4464}
4465static ssize_t
fd01b88c 4466max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277
N		 4467{
4468 if (strncmp(buf, "max", 3) == 0)
4469 mddev->resync_max = MaxSector;
4470 else {
5e96ee65
NB		 4471 unsigned long long max;
4472 if (strict_strtoull(buf, 10, &max))
4473 return -EINVAL;
4474 if (max < mddev->resync_min)
c6207277
N		 4475 return -EINVAL;
4476 if (max < mddev->resync_max &&
4d484a4a 4477 mddev->ro == 0 &&
c6207277
N		 4478 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4479 return -EBUSY;
4480
4481 /* Must be a multiple of chunk_size */
9d8f0363 4482 if (mddev->chunk_sectors) {
2ac06c33 4483 sector_t temp = max;
9d8f0363 4484 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 4485 return -EINVAL;
4486 }
4487 mddev->resync_max = max;
4488 }
4489 wake_up(&mddev->recovery_wait);
4490 return len;
4491}
4492
4493static struct md_sysfs_entry md_max_sync =
4494__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4495
e464eafd 4496static ssize_t
fd01b88c 4497suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4498{
4499 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4500}
4501
4502static ssize_t
fd01b88c 4503suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4504{
4505 char *e;
4506 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4507 unsigned long long old = mddev->suspend_lo;
e464eafd 4508
b8d966ef
N		 4509 if (mddev->pers == NULL ||
4510 mddev->pers->quiesce == NULL)
e464eafd
N		 4511 return -EINVAL;
4512 if (buf == e || (*e && *e != '\n'))
4513 return -EINVAL;
23ddff37
N		 4514
4515 mddev->suspend_lo = new;
4516 if (new >= old)
4517 /* Shrinking suspended region */
e464eafd 4518 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4519 else {
4520 /* Expanding suspended region - need to wait */
4521 mddev->pers->quiesce(mddev, 1);
4522 mddev->pers->quiesce(mddev, 0);
4523 }
4524 return len;
e464eafd
N		 4525}
4526static struct md_sysfs_entry md_suspend_lo =
4527__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4528
4529
4530static ssize_t
fd01b88c 4531suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4532{
4533 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4534}
4535
4536static ssize_t
fd01b88c 4537suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4538{
4539 char *e;
4540 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4541 unsigned long long old = mddev->suspend_hi;
e464eafd 4542
b8d966ef
N		 4543 if (mddev->pers == NULL ||
4544 mddev->pers->quiesce == NULL)
e464eafd
N		 4545 return -EINVAL;
4546 if (buf == e || (*e && *e != '\n'))
4547 return -EINVAL;
23ddff37
N		 4548
4549 mddev->suspend_hi = new;
4550 if (new <= old)
4551 /* Shrinking suspended region */
4552 mddev->pers->quiesce(mddev, 2);
4553 else {
4554 /* Expanding suspended region - need to wait */
e464eafd
N		 4555 mddev->pers->quiesce(mddev, 1);
4556 mddev->pers->quiesce(mddev, 0);
23ddff37
N		 4557 }
4558 return len;
e464eafd
N		 4559}
4560static struct md_sysfs_entry md_suspend_hi =
4561__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4562
08a02ecd 4563static ssize_t
fd01b88c 4564reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4565{
4566 if (mddev->reshape_position != MaxSector)
4567 return sprintf(page, "%llu\n",
4568 (unsigned long long)mddev->reshape_position);
4569 strcpy(page, "none\n");
4570 return 5;
4571}
4572
4573static ssize_t
fd01b88c 4574reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd 4575{
c6563a8c 4576 struct md_rdev *rdev;
08a02ecd
N		 4577 char *e;
4578 unsigned long long new = simple_strtoull(buf, &e, 10);
4579 if (mddev->pers)
4580 return -EBUSY;
4581 if (buf == e || (*e && *e != '\n'))
4582 return -EINVAL;
4583 mddev->reshape_position = new;
4584 mddev->delta_disks = 0;
2c810cdd 4585 mddev->reshape_backwards = 0;
08a02ecd
N		 4586 mddev->new_level = mddev->level;
4587 mddev->new_layout = mddev->layout;
664e7c41 4588 mddev->new_chunk_sectors = mddev->chunk_sectors;
c6563a8c
N		 4589 rdev_for_each(rdev, mddev)
4590 rdev->new_data_offset = rdev->data_offset;
08a02ecd
N		 4591 return len;
4592}
4593
4594static struct md_sysfs_entry md_reshape_position =
4595__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4596 reshape_position_store);
4597
2c810cdd
N		 4598static ssize_t
4599reshape_direction_show(struct mddev *mddev, char *page)
4600{
4601 return sprintf(page, "%s\n",
4602 mddev->reshape_backwards ? "backwards" : "forwards");
4603}
4604
4605static ssize_t
4606reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4607{
4608 int backwards = 0;
4609 if (cmd_match(buf, "forwards"))
4610 backwards = 0;
4611 else if (cmd_match(buf, "backwards"))
4612 backwards = 1;
4613 else
4614 return -EINVAL;
4615 if (mddev->reshape_backwards == backwards)
4616 return len;
4617
4618 /* check if we are allowed to change */
4619 if (mddev->delta_disks)
4620 return -EBUSY;
4621
4622 if (mddev->persistent &&
4623 mddev->major_version == 0)
4624 return -EINVAL;
4625
4626 mddev->reshape_backwards = backwards;
4627 return len;
4628}
4629
4630static struct md_sysfs_entry md_reshape_direction =
4631__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4632 reshape_direction_store);
4633
b522adcd 4634static ssize_t
fd01b88c 4635array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4636{
4637 if (mddev->external_size)
4638 return sprintf(page, "%llu\n",
4639 (unsigned long long)mddev->array_sectors/2);
4640 else
4641 return sprintf(page, "default\n");
4642}
4643
4644static ssize_t
fd01b88c 4645array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4646{
4647 sector_t sectors;
4648
4649 if (strncmp(buf, "default", 7) == 0) {
4650 if (mddev->pers)
4651 sectors = mddev->pers->size(mddev, 0, 0);
4652 else
4653 sectors = mddev->array_sectors;
4654
4655 mddev->external_size = 0;
4656 } else {
4657 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4658 return -EINVAL;
4659 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4660 return -E2BIG;
b522adcd
DW		 4661
4662 mddev->external_size = 1;
4663 }
4664
4665 mddev->array_sectors = sectors;
cbe6ef1d
N		 4666 if (mddev->pers) {
4667 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 4668 revalidate_disk(mddev->gendisk);
cbe6ef1d 4669 }
b522adcd
DW		 4670 return len;
4671}
4672
4673static struct md_sysfs_entry md_array_size =
4674__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4675 array_size_store);
e464eafd 4676
eae1701f
N		 4677static struct attribute *md_default_attrs[] = {
4678 &md_level.attr,
d4dbd025 4679 &md_layout.attr,
eae1701f 4680 &md_raid_disks.attr,
3b34380a 4681 &md_chunk_size.attr,
a35b0d69 4682 &md_size.attr,
a94213b1 4683 &md_resync_start.attr,
8bb93aac 4684 &md_metadata.attr,
6d7ff738 4685 &md_new_device.attr,
16f17b39 4686 &md_safe_delay.attr,
9e653b63 4687 &md_array_state.attr,
08a02ecd 4688 &md_reshape_position.attr,
2c810cdd 4689 &md_reshape_direction.attr,
b522adcd 4690 &md_array_size.attr,
1e50915f 4691 &max_corr_read_errors.attr,
411036fa
N		 4692 NULL,
4693};
4694
4695static struct attribute *md_redundancy_attrs[] = {
24dd469d 4696 &md_scan_mode.attr,
9d88883e 4697 &md_mismatches.attr,
88202a0c
N		 4698 &md_sync_min.attr,
4699 &md_sync_max.attr,
4700 &md_sync_speed.attr,
90b08710 4701 &md_sync_force_parallel.attr,
88202a0c 4702 &md_sync_completed.attr,
5e96ee65 4703 &md_min_sync.attr,
c6207277 4704 &md_max_sync.attr,
e464eafd
N		 4705 &md_suspend_lo.attr,
4706 &md_suspend_hi.attr,
9b1d1dac 4707 &md_bitmap.attr,
d7f3d291 4708 &md_degraded.attr,
eae1701f
N		 4709 NULL,
4710};
411036fa
N		 4711static struct attribute_group md_redundancy_group = {
4712 .name = NULL,
4713 .attrs = md_redundancy_attrs,
4714};
4715
eae1701f
N		 4716
4717static ssize_t
4718md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4719{
4720 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4721 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4722 ssize_t rv;
eae1701f
N		 4723
4724 if (!entry->show)
4725 return -EIO;
af8a2434
N		 4726 spin_lock(&all_mddevs_lock);
4727 if (list_empty(&mddev->all_mddevs)) {
4728 spin_unlock(&all_mddevs_lock);
4729 return -EBUSY;
4730 }
4731 mddev_get(mddev);
4732 spin_unlock(&all_mddevs_lock);
4733
5dc5cf7d
IM		 4734 rv = mddev_lock(mddev);
4735 if (!rv) {
4736 rv = entry->show(mddev, page);
4737 mddev_unlock(mddev);
4738 }
af8a2434 4739 mddev_put(mddev);
96de1e66 4740 return rv;
eae1701f
N		 4741}
4742
4743static ssize_t
4744md_attr_store(struct kobject *kobj, struct attribute *attr,
4745 const char *page, size_t length)
4746{
4747 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4748 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4749 ssize_t rv;
eae1701f
N		 4750
4751 if (!entry->store)
4752 return -EIO;
67463acb
N		 4753 if (!capable(CAP_SYS_ADMIN))
4754 return -EACCES;
af8a2434
N		 4755 spin_lock(&all_mddevs_lock);
4756 if (list_empty(&mddev->all_mddevs)) {
4757 spin_unlock(&all_mddevs_lock);
4758 return -EBUSY;
4759 }
4760 mddev_get(mddev);
4761 spin_unlock(&all_mddevs_lock);
a7a3f08d
N		 4762 if (entry->store == new_dev_store)
4763 flush_workqueue(md_misc_wq);
5dc5cf7d
IM		 4764 rv = mddev_lock(mddev);
4765 if (!rv) {
4766 rv = entry->store(mddev, page, length);
4767 mddev_unlock(mddev);
4768 }
af8a2434 4769 mddev_put(mddev);
96de1e66 4770 return rv;
eae1701f
N		 4771}
4772
4773static void md_free(struct kobject *ko)
4774{
fd01b88c 4775 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4776
4777 if (mddev->sysfs_state)
4778 sysfs_put(mddev->sysfs_state);
4779
4780 if (mddev->gendisk) {
4781 del_gendisk(mddev->gendisk);
4782 put_disk(mddev->gendisk);
4783 }
4784 if (mddev->queue)
4785 blk_cleanup_queue(mddev->queue);
4786
eae1701f
N		 4787 kfree(mddev);
4788}
4789
52cf25d0 4790static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4791 .show = md_attr_show,
4792 .store = md_attr_store,
4793};
4794static struct kobj_type md_ktype = {
4795 .release = md_free,
4796 .sysfs_ops = &md_sysfs_ops,
4797 .default_attrs = md_default_attrs,
4798};
4799
1da177e4
LT		 4800int mdp_major = 0;
4801
5fd3a17e
DW		 4802static void mddev_delayed_delete(struct work_struct *ws)
4803{
fd01b88c 4804 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4805
43a70507 4806 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4807 kobject_del(&mddev->kobj);
4808 kobject_put(&mddev->kobj);
4809}
4810
efeb53c0 4811static int md_alloc(dev_t dev, char *name)
1da177e4 4812{
48c9c27b 4813 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4814 struct mddev *mddev = mddev_find(dev);
1da177e4 4815 struct gendisk *disk;
efeb53c0
N		 4816 int partitioned;
4817 int shift;
4818 int unit;
3830c62f 4819 int error;
1da177e4
LT		 4820
4821 if (!mddev)
efeb53c0
N		 4822 return -ENODEV;
4823
4824 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4825 shift = partitioned ? MdpMinorShift : 0;
4826 unit = MINOR(mddev->unit) >> shift;
1da177e4 4827
e804ac78
TH		 4828 /* wait for any previous instance of this device to be
4829 * completely removed (mddev_delayed_delete).
d3374825 4830 */
e804ac78 4831 flush_workqueue(md_misc_wq);
d3374825 4832
48c9c27b 4833 mutex_lock(&disks_mutex);
0909dc44
N		 4834 error = -EEXIST;
4835 if (mddev->gendisk)
4836 goto abort;
efeb53c0
N		 4837
4838 if (name) {
4839 /* Need to ensure that 'name' is not a duplicate.
4840 */
fd01b88c 4841 struct mddev *mddev2;
efeb53c0
N		 4842 spin_lock(&all_mddevs_lock);
4843
4844 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4845 if (mddev2->gendisk &&
4846 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4847 spin_unlock(&all_mddevs_lock);
0909dc44 4848 goto abort;
efeb53c0
N		 4849 }
4850 spin_unlock(&all_mddevs_lock);
1da177e4 4851 }
8b765398 4852
0909dc44 4853 error = -ENOMEM;
8b765398 4854 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4855 if (!mddev->queue)
4856 goto abort;
409c57f3
N		 4857 mddev->queue->queuedata = mddev;
4858
409c57f3 4859 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 4860 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 4861
1da177e4
LT		 4862 disk = alloc_disk(1 << shift);
4863 if (!disk) {
8b765398
N		 4864 blk_cleanup_queue(mddev->queue);
4865 mddev->queue = NULL;
0909dc44 4866 goto abort;
1da177e4 4867 }
efeb53c0 4868 disk->major = MAJOR(mddev->unit);
1da177e4 4869 disk->first_minor = unit << shift;
efeb53c0
N		 4870 if (name)
4871 strcpy(disk->disk_name, name);
4872 else if (partitioned)
1da177e4 4873 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4874 else
1da177e4 4875 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4876 disk->fops = &md_fops;
4877 disk->private_data = mddev;
4878 disk->queue = mddev->queue;
b0140891 4879 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4880 /* Allow extended partitions. This makes the
d3374825 4881 * 'mdp' device redundant, but we can't really
92850bbd
N		 4882 * remove it now.
4883 */
4884 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4885 mddev->gendisk = disk;
b0140891
N		 4886 /* As soon as we call add_disk(), another thread could get
4887 * through to md_open, so make sure it doesn't get too far
4888 */
4889 mutex_lock(&mddev->open_mutex);
4890 add_disk(disk);
4891
ed9e1982
TH		 4892 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4893 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4894 if (error) {
4895 /* This isn't possible, but as kobject_init_and_add is marked
4896 * __must_check, we must do something with the result
4897 */
5e55e2f5
N		 4898 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4899 disk->disk_name);
0909dc44
N		 4900 error = 0;
4901 }
00bcb4ac
N		 4902 if (mddev->kobj.sd &&
4903 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4904 printk(KERN_DEBUG "pointless warning\n");
b0140891 4905 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4906 abort:
4907 mutex_unlock(&disks_mutex);
00bcb4ac 4908 if (!error && mddev->kobj.sd) {
3830c62f 4909 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4910 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4911 }
d3374825 4912 mddev_put(mddev);
0909dc44 4913 return error;
efeb53c0
N		 4914}
4915
4916static struct kobject *md_probe(dev_t dev, int *part, void *data)
4917{
4918 md_alloc(dev, NULL);
1da177e4
LT		 4919 return NULL;
4920}
4921
efeb53c0
N		 4922static int add_named_array(const char *val, struct kernel_param *kp)
4923{
4924 /* val must be "md_*" where * is not all digits.
4925 * We allocate an array with a large free minor number, and
4926 * set the name to val. val must not already be an active name.
4927 */
4928 int len = strlen(val);
4929 char buf[DISK_NAME_LEN];
4930
4931 while (len && val[len-1] == '\n')
4932 len--;
4933 if (len >= DISK_NAME_LEN)
4934 return -E2BIG;
4935 strlcpy(buf, val, len+1);
4936 if (strncmp(buf, "md_", 3) != 0)
4937 return -EINVAL;
4938 return md_alloc(0, buf);
4939}
4940
1da177e4
LT		 4941static void md_safemode_timeout(unsigned long data)
4942{
fd01b88c 4943 struct mddev *mddev = (struct mddev *) data;
1da177e4 4944
0fd62b86
NB		 4945 if (!atomic_read(&mddev->writes_pending)) {
4946 mddev->safemode = 1;
4947 if (mddev->external)
00bcb4ac 4948 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4949 }
1da177e4
LT		 4950 md_wakeup_thread(mddev->thread);
4951}
4952
6ff8d8ec 4953static int start_dirty_degraded;
1da177e4 4954
fd01b88c 4955int md_run(struct mddev *mddev)
1da177e4 4956{
2604b703 4957 int err;
3cb03002 4958 struct md_rdev *rdev;
84fc4b56 4959 struct md_personality *pers;
1da177e4 4960
a757e64c
N		 4961 if (list_empty(&mddev->disks))
4962 /* cannot run an array with no devices.. */
1da177e4 4963 return -EINVAL;
1da177e4
LT		 4964
4965 if (mddev->pers)
4966 return -EBUSY;
bb4f1e9d
N		 4967 /* Cannot run until previous stop completes properly */
4968 if (mddev->sysfs_active)
4969 return -EBUSY;
b6eb127d 4970
1da177e4
LT		 4971 /*
4972 * Analyze all RAID superblock(s)
4973 */
1ec4a939
N		 4974 if (!mddev->raid_disks) {
4975 if (!mddev->persistent)
4976 return -EINVAL;
a757e64c 4977 analyze_sbs(mddev);
1ec4a939 4978 }
1da177e4 4979
d9d166c2
N		 4980 if (mddev->level != LEVEL_NONE)
4981 request_module("md-level-%d", mddev->level);
4982 else if (mddev->clevel[0])
4983 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4984
4985 /*
4986 * Drop all container device buffers, from now on
4987 * the only valid external interface is through the md
4988 * device.
1da177e4 4989 */
dafb20fa 4990 rdev_for_each(rdev, mddev) {
b2d444d7 4991 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4992 continue;
4993 sync_blockdev(rdev->bdev);
f98393a6 4994 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4995
4996 /* perform some consistency tests on the device.
4997 * We don't want the data to overlap the metadata,
58c0fed4 4998 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4999 */
a6ff7e08
JB		 5000 if (rdev->meta_bdev) {
5001 /* Nothing to check */;
5002 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 5003 if (mddev->dev_sectors &&
5004 rdev->data_offset + mddev->dev_sectors
0f420358 5005 > rdev->sb_start) {
f0d76d70
N		 5006 printk("md: %s: data overlaps metadata\n",
5007 mdname(mddev));
5008 return -EINVAL;
5009 }
5010 } else {
0f420358 5011 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 5012 > rdev->data_offset) {
5013 printk("md: %s: metadata overlaps data\n",
5014 mdname(mddev));
5015 return -EINVAL;
5016 }
5017 }
00bcb4ac 5018 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 5019 }
5020
a167f663 5021 if (mddev->bio_set == NULL)
395c72a7 5022 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
a167f663 5023
1da177e4 5024 spin_lock(&pers_lock);
d9d166c2 5025 pers = find_pers(mddev->level, mddev->clevel);
2604b703 5026 if (!pers || !try_module_get(pers->owner)) {
1da177e4 5027 spin_unlock(&pers_lock);
d9d166c2
N		 5028 if (mddev->level != LEVEL_NONE)
5029 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5030 mddev->level);
5031 else
5032 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5033 mddev->clevel);
1da177e4
LT		 5034 return -EINVAL;
5035 }
2604b703 5036 mddev->pers = pers;
1da177e4 5037 spin_unlock(&pers_lock);
34817e8c
N		 5038 if (mddev->level != pers->level) {
5039 mddev->level = pers->level;
5040 mddev->new_level = pers->level;
5041 }
d9d166c2 5042 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 5043
f6705578 5044 if (mddev->reshape_position != MaxSector &&
63c70c4f 5045 pers->start_reshape == NULL) {
f6705578
N		 5046 /* This personality cannot handle reshaping... */
5047 mddev->pers = NULL;
5048 module_put(pers->owner);
5049 return -EINVAL;
5050 }
5051
7dd5e7c3
N		 5052 if (pers->sync_request) {
5053 /* Warn if this is a potentially silly
5054 * configuration.
5055 */
5056 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5057 struct md_rdev *rdev2;
7dd5e7c3 5058 int warned = 0;
159ec1fc 5059
dafb20fa
N		 5060 rdev_for_each(rdev, mddev)
5061 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N		 5062 if (rdev < rdev2 &&
5063 rdev->bdev->bd_contains ==
5064 rdev2->bdev->bd_contains) {
5065 printk(KERN_WARNING
5066 "%s: WARNING: %s appears to be"
5067 " on the same physical disk as"
5068 " %s.\n",
5069 mdname(mddev),
5070 bdevname(rdev->bdev,b),
5071 bdevname(rdev2->bdev,b2));
5072 warned = 1;
5073 }
5074 }
159ec1fc 5075
7dd5e7c3
N		 5076 if (warned)
5077 printk(KERN_WARNING
5078 "True protection against single-disk"
5079 " failure might be compromised.\n");
5080 }
5081
657390d2 5082 mddev->recovery = 0;
58c0fed4
AN		 5083 /* may be over-ridden by personality */
5084 mddev->resync_max_sectors = mddev->dev_sectors;
5085
6ff8d8ec 5086 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5087
0f9552b5 5088 if (start_readonly && mddev->ro == 0)
f91de92e
N		 5089 mddev->ro = 2; /* read-only, but switch on first write */
5090
b15c2e57 5091 err = mddev->pers->run(mddev);
13e53df3
AN		 5092 if (err)
5093 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 5094 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5095 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5096 " but 'external_size' not in effect?\n", __func__);
5097 printk(KERN_ERR
5098 "md: invalid array_size %llu > default size %llu\n",
5099 (unsigned long long)mddev->array_sectors / 2,
5100 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5101 err = -EINVAL;
5102 mddev->pers->stop(mddev);
5103 }
ef99bf48
N		 5104 if (err == 0 && mddev->pers->sync_request &&
5105 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
b15c2e57
N		 5106 err = bitmap_create(mddev);
5107 if (err) {
5108 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5109 mdname(mddev), err);
5110 mddev->pers->stop(mddev);
5111 }
5112 }
1da177e4 5113 if (err) {
1da177e4
LT		 5114 module_put(mddev->pers->owner);
5115 mddev->pers = NULL;
32a7627c
N		 5116 bitmap_destroy(mddev);
5117 return err;
1da177e4 5118 }
5e55e2f5 5119 if (mddev->pers->sync_request) {
00bcb4ac
N		 5120 if (mddev->kobj.sd &&
5121 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 5122 printk(KERN_WARNING
5123 "md: cannot register extra attributes for %s\n",
5124 mdname(mddev));
00bcb4ac 5125 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5126 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 5127 mddev->ro = 0;
5128
1da177e4 5129 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 5130 atomic_set(&mddev->max_corr_read_errors,
5131 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 5132 mddev->safemode = 0;
5133 mddev->safemode_timer.function = md_safemode_timeout;
5134 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 5135 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5136 mddev->in_sync = 1;
0ca69886
N		 5137 smp_wmb();
5138 mddev->ready = 1;
dafb20fa 5139 rdev_for_each(rdev, mddev)
36fad858
NK		 5140 if (rdev->raid_disk >= 0)
5141 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 5142 /* failure here is OK */;
1da177e4
LT		 5143
5144 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5145
850b2b42
N		 5146 if (mddev->flags)
5147 md_update_sb(mddev, 0);
1da177e4 5148
d7603b7e 5149 md_new_event(mddev);
00bcb4ac
N		 5150 sysfs_notify_dirent_safe(mddev->sysfs_state);
5151 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 5152 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 5153 return 0;
5154}
390ee602 5155EXPORT_SYMBOL_GPL(md_run);
1da177e4 5156
fd01b88c 5157static int do_md_run(struct mddev *mddev)
fe60b014
N		 5158{
5159 int err;
5160
5161 err = md_run(mddev);
5162 if (err)
5163 goto out;
69e51b44
N		 5164 err = bitmap_load(mddev);
5165 if (err) {
5166 bitmap_destroy(mddev);
5167 goto out;
5168 }
0fd018af
JB		 5169
5170 md_wakeup_thread(mddev->thread);
5171 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5172
fe60b014
N		 5173 set_capacity(mddev->gendisk, mddev->array_sectors);
5174 revalidate_disk(mddev->gendisk);
f0b4f7e2 5175 mddev->changed = 1;
fe60b014
N		 5176 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5177out:
5178 return err;
5179}
5180
fd01b88c 5181static int restart_array(struct mddev *mddev)
1da177e4
LT		 5182{
5183 struct gendisk *disk = mddev->gendisk;
1da177e4 5184
80fab1d7 5185 /* Complain if it has no devices */
1da177e4 5186 if (list_empty(&mddev->disks))
80fab1d7
AN		 5187 return -ENXIO;
5188 if (!mddev->pers)
5189 return -EINVAL;
5190 if (!mddev->ro)
5191 return -EBUSY;
5192 mddev->safemode = 0;
5193 mddev->ro = 0;
5194 set_disk_ro(disk, 0);
5195 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5196 mdname(mddev));
5197 /* Kick recovery or resync if necessary */
5198 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5199 md_wakeup_thread(mddev->thread);
5200 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5201 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5202 return 0;
1da177e4
LT		 5203}
5204
acc55e22
N		 5205/* similar to deny_write_access, but accounts for our holding a reference
5206 * to the file ourselves */
5207static int deny_bitmap_write_access(struct file * file)
5208{
5209 struct inode *inode = file->f_mapping->host;
5210
5211 spin_lock(&inode->i_lock);
5212 if (atomic_read(&inode->i_writecount) > 1) {
5213 spin_unlock(&inode->i_lock);
5214 return -ETXTBSY;
5215 }
5216 atomic_set(&inode->i_writecount, -1);
5217 spin_unlock(&inode->i_lock);
5218
5219 return 0;
5220}
5221
43a70507 5222void restore_bitmap_write_access(struct file *file)
acc55e22
N		 5223{
5224 struct inode *inode = file->f_mapping->host;
5225
5226 spin_lock(&inode->i_lock);
5227 atomic_set(&inode->i_writecount, 1);
5228 spin_unlock(&inode->i_lock);
5229}
5230
fd01b88c 5231static void md_clean(struct mddev *mddev)
6177b472
N		 5232{
5233 mddev->array_sectors = 0;
5234 mddev->external_size = 0;
5235 mddev->dev_sectors = 0;
5236 mddev->raid_disks = 0;
5237 mddev->recovery_cp = 0;
5238 mddev->resync_min = 0;
5239 mddev->resync_max = MaxSector;
5240 mddev->reshape_position = MaxSector;
5241 mddev->external = 0;
5242 mddev->persistent = 0;
5243 mddev->level = LEVEL_NONE;
5244 mddev->clevel[0] = 0;
5245 mddev->flags = 0;
5246 mddev->ro = 0;
5247 mddev->metadata_type[0] = 0;
5248 mddev->chunk_sectors = 0;
5249 mddev->ctime = mddev->utime = 0;
5250 mddev->layout = 0;
5251 mddev->max_disks = 0;
5252 mddev->events = 0;
a8707c08 5253 mddev->can_decrease_events = 0;
6177b472 5254 mddev->delta_disks = 0;
2c810cdd 5255 mddev->reshape_backwards = 0;
6177b472
N		 5256 mddev->new_level = LEVEL_NONE;
5257 mddev->new_layout = 0;
5258 mddev->new_chunk_sectors = 0;
5259 mddev->curr_resync = 0;
7f7583d4 5260 atomic64_set(&mddev->resync_mismatches, 0);
6177b472
N		 5261 mddev->suspend_lo = mddev->suspend_hi = 0;
5262 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5263 mddev->recovery = 0;
5264 mddev->in_sync = 0;
f0b4f7e2 5265 mddev->changed = 0;
6177b472 5266 mddev->degraded = 0;
6177b472 5267 mddev->safemode = 0;
050b6615 5268 mddev->merge_check_needed = 0;
6177b472
N		 5269 mddev->bitmap_info.offset = 0;
5270 mddev->bitmap_info.default_offset = 0;
6409bb05 5271 mddev->bitmap_info.default_space = 0;
6177b472
N		 5272 mddev->bitmap_info.chunksize = 0;
5273 mddev->bitmap_info.daemon_sleep = 0;
5274 mddev->bitmap_info.max_write_behind = 0;
5275}
5276
fd01b88c 5277static void __md_stop_writes(struct mddev *mddev)
a047e125 5278{
6b6204ee 5279 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
a047e125 5280 if (mddev->sync_thread) {
a047e125 5281 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 5282 md_reap_sync_thread(mddev);
a047e125
N		 5283 }
5284
5285 del_timer_sync(&mddev->safemode_timer);
5286
5287 bitmap_flush(mddev);
5288 md_super_wait(mddev);
5289
b6d428c6
N		 5290 if (mddev->ro == 0 &&
5291 (!mddev->in_sync || mddev->flags)) {
a047e125
N		 5292 /* mark array as shutdown cleanly */
5293 mddev->in_sync = 1;
5294 md_update_sb(mddev, 1);
5295 }
5296}
defad61a 5297
fd01b88c 5298void md_stop_writes(struct mddev *mddev)
defad61a
N		 5299{
5300 mddev_lock(mddev);
5301 __md_stop_writes(mddev);
5302 mddev_unlock(mddev);
5303}
390ee602 5304EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5305
5eff3c43 5306static void __md_stop(struct mddev *mddev)
6177b472 5307{
0ca69886 5308 mddev->ready = 0;
d7b00aad
N		 5309 /* Ensure ->event_work is done */
5310 flush_workqueue(md_misc_wq);
6177b472
N		 5311 mddev->pers->stop(mddev);
5312 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5313 mddev->to_remove = &md_redundancy_group;
5314 module_put(mddev->pers->owner);
5315 mddev->pers = NULL;
cca9cf90 5316 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5317}
5eff3c43
N		 5318
5319void md_stop(struct mddev *mddev)
5320{
5321 /* stop the array and free an attached data structures.
5322 * This is called from dm-raid
5323 */
5324 __md_stop(mddev);
5325 bitmap_destroy(mddev);
5326 if (mddev->bio_set)
5327 bioset_free(mddev->bio_set);
5328}
5329
390ee602 5330EXPORT_SYMBOL_GPL(md_stop);
6177b472 5331
a05b7ea0 5332static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N		 5333{
5334 int err = 0;
5335 mutex_lock(&mddev->open_mutex);
a05b7ea0 5336 if (atomic_read(&mddev->openers) > !!bdev) {
a4bd82d0
N		 5337 printk("md: %s still in use.\n",mdname(mddev));
5338 err = -EBUSY;
5339 goto out;
5340 }
a05b7ea0
N		 5341 if (bdev)
5342 sync_blockdev(bdev);
a4bd82d0 5343 if (mddev->pers) {
defad61a 5344 __md_stop_writes(mddev);
a4bd82d0
N		 5345
5346 err = -ENXIO;
5347 if (mddev->ro==1)
5348 goto out;
5349 mddev->ro = 1;
5350 set_disk_ro(mddev->gendisk, 1);
5351 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 5352 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 5353 err = 0;
5354 }
5355out:
5356 mutex_unlock(&mddev->open_mutex);
5357 return err;
5358}
5359
9e653b63
N		 5360/* mode:
5361 * 0 - completely stop and dis-assemble array
9e653b63
N		 5362 * 2 - stop but do not disassemble array
5363 */
a05b7ea0
N		 5364static int do_md_stop(struct mddev * mddev, int mode,
5365 struct block_device *bdev)
1da177e4 5366{
1da177e4 5367 struct gendisk *disk = mddev->gendisk;
3cb03002 5368 struct md_rdev *rdev;
1da177e4 5369
c8c00a69 5370 mutex_lock(&mddev->open_mutex);
a05b7ea0 5371 if (atomic_read(&mddev->openers) > !!bdev ||
bb4f1e9d 5372 mddev->sysfs_active) {
df5b20cf 5373 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 5374 mutex_unlock(&mddev->open_mutex);
5375 return -EBUSY;
5376 }
a05b7ea0
N		 5377 if (bdev)
5378 /* It is possible IO was issued on some other
5379 * open file which was closed before we took ->open_mutex.
5380 * As that was not the last close __blkdev_put will not
5381 * have called sync_blockdev, so we must.
5382 */
5383 sync_blockdev(bdev);
1da177e4 5384
6e17b027 5385 if (mddev->pers) {
a4bd82d0
N		 5386 if (mddev->ro)
5387 set_disk_ro(disk, 0);
409c57f3 5388
defad61a 5389 __md_stop_writes(mddev);
5eff3c43 5390 __md_stop(mddev);
a4bd82d0 5391 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5392 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5393
a4bd82d0 5394 /* tell userspace to handle 'inactive' */
00bcb4ac 5395 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5396
dafb20fa 5397 rdev_for_each(rdev, mddev)
36fad858
NK		 5398 if (rdev->raid_disk >= 0)
5399 sysfs_unlink_rdev(mddev, rdev);
c4647292 5400
a4bd82d0 5401 set_capacity(disk, 0);
6e17b027 5402 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5403 mddev->changed = 1;
a4bd82d0 5404 revalidate_disk(disk);
0d4ca600 5405
a4bd82d0
N		 5406 if (mddev->ro)
5407 mddev->ro = 0;
6e17b027
N		 5408 } else
5409 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5410 /*
5411 * Free resources if final stop
5412 */
9e653b63 5413 if (mode == 0) {
1da177e4
LT		 5414 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5415
978f946b 5416 bitmap_destroy(mddev);
c3d9714e
N		 5417 if (mddev->bitmap_info.file) {
5418 restore_bitmap_write_access(mddev->bitmap_info.file);
5419 fput(mddev->bitmap_info.file);
5420 mddev->bitmap_info.file = NULL;
978f946b 5421 }
c3d9714e 5422 mddev->bitmap_info.offset = 0;
978f946b 5423
1da177e4
LT		 5424 export_array(mddev);
5425
6177b472 5426 md_clean(mddev);
934d9c23 5427 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5428 if (mddev->hold_active == UNTIL_STOP)
5429 mddev->hold_active = 0;
a4bd82d0 5430 }
3f9d99c1 5431 blk_integrity_unregister(disk);
d7603b7e 5432 md_new_event(mddev);
00bcb4ac 5433 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5434 return 0;
1da177e4
LT		 5435}
5436
fdee8ae4 5437#ifndef MODULE
fd01b88c 5438static void autorun_array(struct mddev *mddev)
1da177e4 5439{
3cb03002 5440 struct md_rdev *rdev;
1da177e4
LT		 5441 int err;
5442
a757e64c 5443 if (list_empty(&mddev->disks))
1da177e4 5444 return;
1da177e4
LT		 5445
5446 printk(KERN_INFO "md: running: ");
5447
dafb20fa 5448 rdev_for_each(rdev, mddev) {
1da177e4
LT		 5449 char b[BDEVNAME_SIZE];
5450 printk("<%s>", bdevname(rdev->bdev,b));
5451 }
5452 printk("\n");
5453
d710e138 5454 err = do_md_run(mddev);
1da177e4
LT		 5455 if (err) {
5456 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
a05b7ea0 5457 do_md_stop(mddev, 0, NULL);
1da177e4
LT		 5458 }
5459}
5460
5461/*
5462 * lets try to run arrays based on all disks that have arrived
5463 * until now. (those are in pending_raid_disks)
5464 *
5465 * the method: pick the first pending disk, collect all disks with
5466 * the same UUID, remove all from the pending list and put them into
5467 * the 'same_array' list. Then order this list based on superblock
5468 * update time (freshest comes first), kick out 'old' disks and
5469 * compare superblocks. If everything's fine then run it.
5470 *
5471 * If "unit" is allocated, then bump its reference count
5472 */
5473static void autorun_devices(int part)
5474{
3cb03002 5475 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5476 struct mddev *mddev;
1da177e4
LT		 5477 char b[BDEVNAME_SIZE];
5478
5479 printk(KERN_INFO "md: autorun ...\n");
5480 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5481 int unit;
1da177e4 5482 dev_t dev;
ad01c9e3 5483 LIST_HEAD(candidates);
1da177e4 5484 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5485 struct md_rdev, same_set);
1da177e4
LT		 5486
5487 printk(KERN_INFO "md: considering %s ...\n",
5488 bdevname(rdev0->bdev,b));
5489 INIT_LIST_HEAD(&candidates);
159ec1fc 5490 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5491 if (super_90_load(rdev, rdev0, 0) >= 0) {
5492 printk(KERN_INFO "md: adding %s ...\n",
5493 bdevname(rdev->bdev,b));
5494 list_move(&rdev->same_set, &candidates);
5495 }
5496 /*
5497 * now we have a set of devices, with all of them having
5498 * mostly sane superblocks. It's time to allocate the
5499 * mddev.
5500 */
e8703fe1
N		 5501 if (part) {
5502 dev = MKDEV(mdp_major,
5503 rdev0->preferred_minor << MdpMinorShift);
5504 unit = MINOR(dev) >> MdpMinorShift;
5505 } else {
5506 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5507 unit = MINOR(dev);
5508 }
5509 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5510 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5511 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5512 break;
5513 }
1da177e4
LT		 5514
5515 md_probe(dev, NULL, NULL);
5516 mddev = mddev_find(dev);
9bbbca3a
NB		 5517 if (!mddev || !mddev->gendisk) {
5518 if (mddev)
5519 mddev_put(mddev);
5520 printk(KERN_ERR
1da177e4
LT		 5521 "md: cannot allocate memory for md drive.\n");
5522 break;
5523 }
5524 if (mddev_lock(mddev))
5525 printk(KERN_WARNING "md: %s locked, cannot run\n",
5526 mdname(mddev));
5527 else if (mddev->raid_disks || mddev->major_version
5528 || !list_empty(&mddev->disks)) {
5529 printk(KERN_WARNING
5530 "md: %s already running, cannot run %s\n",
5531 mdname(mddev), bdevname(rdev0->bdev,b));
5532 mddev_unlock(mddev);
5533 } else {
5534 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5535 mddev->persistent = 1;
159ec1fc 5536 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5537 list_del_init(&rdev->same_set);
5538 if (bind_rdev_to_array(rdev, mddev))
5539 export_rdev(rdev);
5540 }
5541 autorun_array(mddev);
5542 mddev_unlock(mddev);
5543 }
5544 /* on success, candidates will be empty, on error
5545 * it won't...
5546 */
159ec1fc 5547 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5548 list_del_init(&rdev->same_set);
1da177e4 5549 export_rdev(rdev);
4b80991c 5550 }
1da177e4
LT		 5551 mddev_put(mddev);
5552 }
5553 printk(KERN_INFO "md: ... autorun DONE.\n");
5554}
fdee8ae4 5555#endif /* !MODULE */
1da177e4 5556
1da177e4
LT		 5557static int get_version(void __user * arg)
5558{
5559 mdu_version_t ver;
5560
5561 ver.major = MD_MAJOR_VERSION;
5562 ver.minor = MD_MINOR_VERSION;
5563 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5564
5565 if (copy_to_user(arg, &ver, sizeof(ver)))
5566 return -EFAULT;
5567
5568 return 0;
5569}
5570
fd01b88c 5571static int get_array_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5572{
5573 mdu_array_info_t info;
a9f326eb 5574 int nr,working,insync,failed,spare;
3cb03002 5575 struct md_rdev *rdev;
1da177e4 5576
1ca69c4b
N		 5577 nr = working = insync = failed = spare = 0;
5578 rcu_read_lock();
5579 rdev_for_each_rcu(rdev, mddev) {
1da177e4 5580 nr++;
b2d444d7 5581 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5582 failed++;
5583 else {
5584 working++;
b2d444d7 5585 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5586 insync++;
1da177e4
LT		 5587 else
5588 spare++;
5589 }
5590 }
1ca69c4b 5591 rcu_read_unlock();
1da177e4
LT		 5592
5593 info.major_version = mddev->major_version;
5594 info.minor_version = mddev->minor_version;
5595 info.patch_version = MD_PATCHLEVEL_VERSION;
5596 info.ctime = mddev->ctime;
5597 info.level = mddev->level;
58c0fed4
AN		 5598 info.size = mddev->dev_sectors / 2;
5599 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5600 info.size = -1;
1da177e4
LT		 5601 info.nr_disks = nr;
5602 info.raid_disks = mddev->raid_disks;
5603 info.md_minor = mddev->md_minor;
5604 info.not_persistent= !mddev->persistent;
5605
5606 info.utime = mddev->utime;
5607 info.state = 0;
5608 if (mddev->in_sync)
5609 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5610 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5611 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5612 info.active_disks = insync;
1da177e4
LT		 5613 info.working_disks = working;
5614 info.failed_disks = failed;
5615 info.spare_disks = spare;
5616
5617 info.layout = mddev->layout;
9d8f0363 5618 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5619
5620 if (copy_to_user(arg, &info, sizeof(info)))
5621 return -EFAULT;
5622
5623 return 0;
5624}
5625
fd01b88c 5626static int get_bitmap_file(struct mddev * mddev, void __user * arg)
32a7627c
N		 5627{
5628 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5629 char *ptr, *buf = NULL;
5630 int err = -ENOMEM;
5631
b5470dc5 5632 if (md_allow_write(mddev))
21c7d380 5633 file = kzalloc(sizeof(*file), GFP_NOIO);
b5470dc5 5634 else
21c7d380 5635 file = kzalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5636
32a7627c
N		 5637 if (!file)
5638 goto out;
5639
5640 /* bitmap disabled, zero the first byte and copy out */
1ec885cd 5641 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
32a7627c
N		 5642 file->pathname[0] = '\0';
5643 goto copy_out;
5644 }
5645
5646 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5647 if (!buf)
5648 goto out;
5649
1ec885cd
N		 5650 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5651 buf, sizeof(file->pathname));
6bcfd601 5652 if (IS_ERR(ptr))
32a7627c
N		 5653 goto out;
5654
5655 strcpy(file->pathname, ptr);
5656
5657copy_out:
5658 err = 0;
5659 if (copy_to_user(arg, file, sizeof(*file)))
5660 err = -EFAULT;
5661out:
5662 kfree(buf);
5663 kfree(file);
5664 return err;
5665}
5666
fd01b88c 5667static int get_disk_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5668{
5669 mdu_disk_info_t info;
3cb03002 5670 struct md_rdev *rdev;
1da177e4
LT		 5671
5672 if (copy_from_user(&info, arg, sizeof(info)))
5673 return -EFAULT;
5674
1ca69c4b
N		 5675 rcu_read_lock();
5676 rdev = find_rdev_nr_rcu(mddev, info.number);
1da177e4
LT		 5677 if (rdev) {
5678 info.major = MAJOR(rdev->bdev->bd_dev);
5679 info.minor = MINOR(rdev->bdev->bd_dev);
5680 info.raid_disk = rdev->raid_disk;
5681 info.state = 0;
b2d444d7 5682 if (test_bit(Faulty, &rdev->flags))
1da177e4 5683 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5684 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5685 info.state |= (1<<MD_DISK_ACTIVE);
5686 info.state |= (1<<MD_DISK_SYNC);
5687 }
8ddf9efe
N		 5688 if (test_bit(WriteMostly, &rdev->flags))
5689 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5690 } else {
5691 info.major = info.minor = 0;
5692 info.raid_disk = -1;
5693 info.state = (1<<MD_DISK_REMOVED);
5694 }
1ca69c4b 5695 rcu_read_unlock();
1da177e4
LT		 5696
5697 if (copy_to_user(arg, &info, sizeof(info)))
5698 return -EFAULT;
5699
5700 return 0;
5701}
5702
fd01b88c 5703static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
1da177e4
LT		 5704{
5705 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5706 struct md_rdev *rdev;
1da177e4
LT		 5707 dev_t dev = MKDEV(info->major,info->minor);
5708
5709 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5710 return -EOVERFLOW;
5711
5712 if (!mddev->raid_disks) {
5713 int err;
5714 /* expecting a device which has a superblock */
5715 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5716 if (IS_ERR(rdev)) {
5717 printk(KERN_WARNING
5718 "md: md_import_device returned %ld\n",
5719 PTR_ERR(rdev));
5720 return PTR_ERR(rdev);
5721 }
5722 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5723 struct md_rdev *rdev0
5724 = list_entry(mddev->disks.next,
5725 struct md_rdev, same_set);
a9f326eb 5726 err = super_types[mddev->major_version]
1da177e4
LT		 5727 .load_super(rdev, rdev0, mddev->minor_version);
5728 if (err < 0) {
5729 printk(KERN_WARNING
5730 "md: %s has different UUID to %s\n",
5731 bdevname(rdev->bdev,b),
5732 bdevname(rdev0->bdev,b2));
5733 export_rdev(rdev);
5734 return -EINVAL;
5735 }
5736 }
5737 err = bind_rdev_to_array(rdev, mddev);
5738 if (err)
5739 export_rdev(rdev);
5740 return err;
5741 }
5742
5743 /*
5744 * add_new_disk can be used once the array is assembled
5745 * to add "hot spares". They must already have a superblock
5746 * written
5747 */
5748 if (mddev->pers) {
5749 int err;
5750 if (!mddev->pers->hot_add_disk) {
5751 printk(KERN_WARNING
5752 "%s: personality does not support diskops!\n",
5753 mdname(mddev));
5754 return -EINVAL;
5755 }
7b1e35f6
N		 5756 if (mddev->persistent)
5757 rdev = md_import_device(dev, mddev->major_version,
5758 mddev->minor_version);
5759 else
5760 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5761 if (IS_ERR(rdev)) {
5762 printk(KERN_WARNING
5763 "md: md_import_device returned %ld\n",
5764 PTR_ERR(rdev));
5765 return PTR_ERR(rdev);
5766 }
1a855a06 5767 /* set saved_raid_disk if appropriate */
41158c7e
N		 5768 if (!mddev->persistent) {
5769 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5770 info->raid_disk < mddev->raid_disks) {
41158c7e 5771 rdev->raid_disk = info->raid_disk;
bf572541 5772 set_bit(In_sync, &rdev->flags);
bb4a65df 5773 clear_bit(Bitmap_sync, &rdev->flags);
bf572541 5774 } else
41158c7e
N		 5775 rdev->raid_disk = -1;
5776 } else
5777 super_types[mddev->major_version].
5778 validate_super(mddev, rdev);
bedd86b7 5779 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 5780 rdev->raid_disk != info->raid_disk) {
bedd86b7
N		 5781 /* This was a hot-add request, but events doesn't
5782 * match, so reject it.
5783 */
5784 export_rdev(rdev);
5785 return -EINVAL;
5786 }
5787
1a855a06
N		 5788 if (test_bit(In_sync, &rdev->flags))
5789 rdev->saved_raid_disk = rdev->raid_disk;
5790 else
5791 rdev->saved_raid_disk = -1;
41158c7e 5792
b2d444d7 5793 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5794 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5795 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5796 else
5797 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5798
1da177e4
LT		 5799 rdev->raid_disk = -1;
5800 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5801 if (!err && !mddev->pers->hot_remove_disk) {
5802 /* If there is hot_add_disk but no hot_remove_disk
5803 * then added disks for geometry changes,
5804 * and should be added immediately.
5805 */
5806 super_types[mddev->major_version].
5807 validate_super(mddev, rdev);
5808 err = mddev->pers->hot_add_disk(mddev, rdev);
5809 if (err)
5810 unbind_rdev_from_array(rdev);
5811 }
1da177e4
LT		 5812 if (err)
5813 export_rdev(rdev);
52664732 5814 else
00bcb4ac 5815 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5816
7ceb17e8 5817 set_bit(MD_CHANGE_DEVS, &mddev->flags);
72a23c21
NB		 5818 if (mddev->degraded)
5819 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5820 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5821 if (!err)
5822 md_new_event(mddev);
005eca5e 5823 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5824 return err;
5825 }
5826
5827 /* otherwise, add_new_disk is only allowed
5828 * for major_version==0 superblocks
5829 */
5830 if (mddev->major_version != 0) {
5831 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5832 mdname(mddev));
5833 return -EINVAL;
5834 }
5835
5836 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5837 int err;
d710e138 5838 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5839 if (IS_ERR(rdev)) {
5840 printk(KERN_WARNING
5841 "md: error, md_import_device() returned %ld\n",
5842 PTR_ERR(rdev));
5843 return PTR_ERR(rdev);
5844 }
5845 rdev->desc_nr = info->number;
5846 if (info->raid_disk < mddev->raid_disks)
5847 rdev->raid_disk = info->raid_disk;
5848 else
5849 rdev->raid_disk = -1;
5850
1da177e4 5851 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5852 if (info->state & (1<<MD_DISK_SYNC))
5853 set_bit(In_sync, &rdev->flags);
1da177e4 5854
8ddf9efe
N		 5855 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5856 set_bit(WriteMostly, &rdev->flags);
5857
1da177e4
LT		 5858 if (!mddev->persistent) {
5859 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5860 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5861 } else
57b2caa3 5862 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5863 rdev->sectors = rdev->sb_start;
1da177e4 5864
2bf071bf
N		 5865 err = bind_rdev_to_array(rdev, mddev);
5866 if (err) {
5867 export_rdev(rdev);
5868 return err;
5869 }
1da177e4
LT		 5870 }
5871
5872 return 0;
5873}
5874
fd01b88c 5875static int hot_remove_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5876{
5877 char b[BDEVNAME_SIZE];
3cb03002 5878 struct md_rdev *rdev;
1da177e4 5879
1da177e4
LT		 5880 rdev = find_rdev(mddev, dev);
5881 if (!rdev)
5882 return -ENXIO;
5883
3ea8929d
N		 5884 clear_bit(Blocked, &rdev->flags);
5885 remove_and_add_spares(mddev, rdev);
5886
1da177e4
LT		 5887 if (rdev->raid_disk >= 0)
5888 goto busy;
5889
5890 kick_rdev_from_array(rdev);
850b2b42 5891 md_update_sb(mddev, 1);
d7603b7e 5892 md_new_event(mddev);
1da177e4
LT		 5893
5894 return 0;
5895busy:
fdefa4d8 5896 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5897 bdevname(rdev->bdev,b), mdname(mddev));
5898 return -EBUSY;
5899}
5900
fd01b88c 5901static int hot_add_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5902{
5903 char b[BDEVNAME_SIZE];
5904 int err;
3cb03002 5905 struct md_rdev *rdev;
1da177e4
LT		 5906
5907 if (!mddev->pers)
5908 return -ENODEV;
5909
5910 if (mddev->major_version != 0) {
5911 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5912 " version-0 superblocks.\n",
5913 mdname(mddev));
5914 return -EINVAL;
5915 }
5916 if (!mddev->pers->hot_add_disk) {
5917 printk(KERN_WARNING
5918 "%s: personality does not support diskops!\n",
5919 mdname(mddev));
5920 return -EINVAL;
5921 }
5922
d710e138 5923 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5924 if (IS_ERR(rdev)) {
5925 printk(KERN_WARNING
5926 "md: error, md_import_device() returned %ld\n",
5927 PTR_ERR(rdev));
5928 return -EINVAL;
5929 }
5930
5931 if (mddev->persistent)
57b2caa3 5932 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5933 else
77304d2a 5934 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5935
8190e754 5936 rdev->sectors = rdev->sb_start;
1da177e4 5937
b2d444d7 5938 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5939 printk(KERN_WARNING
5940 "md: can not hot-add faulty %s disk to %s!\n",
5941 bdevname(rdev->bdev,b), mdname(mddev));
5942 err = -EINVAL;
5943 goto abort_export;
5944 }
b2d444d7 5945 clear_bit(In_sync, &rdev->flags);
1da177e4 5946 rdev->desc_nr = -1;
5842730d 5947 rdev->saved_raid_disk = -1;
2bf071bf
N		 5948 err = bind_rdev_to_array(rdev, mddev);
5949 if (err)
5950 goto abort_export;
1da177e4
LT		 5951
5952 /*
5953 * The rest should better be atomic, we can have disk failures
5954 * noticed in interrupt contexts ...
5955 */
5956
1da177e4
LT		 5957 rdev->raid_disk = -1;
5958
850b2b42 5959 md_update_sb(mddev, 1);
1da177e4
LT		 5960
5961 /*
5962 * Kick recovery, maybe this spare has to be added to the
5963 * array immediately.
5964 */
5965 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5966 md_wakeup_thread(mddev->thread);
d7603b7e 5967 md_new_event(mddev);
1da177e4
LT		 5968 return 0;
5969
1da177e4
LT		 5970abort_export:
5971 export_rdev(rdev);
5972 return err;
5973}
5974
fd01b88c 5975static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c
N		 5976{
5977 int err;
5978
36fa3063
N		 5979 if (mddev->pers) {
5980 if (!mddev->pers->quiesce)
5981 return -EBUSY;
5982 if (mddev->recovery || mddev->sync_thread)
5983 return -EBUSY;
5984 /* we should be able to change the bitmap.. */
5985 }
32a7627c 5986
32a7627c 5987
36fa3063
N		 5988 if (fd >= 0) {
5989 if (mddev->bitmap)
5990 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5991 mddev->bitmap_info.file = fget(fd);
32a7627c 5992
c3d9714e 5993 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5994 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5995 mdname(mddev));
5996 return -EBADF;
5997 }
5998
c3d9714e 5999 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 6000 if (err) {
6001 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6002 mdname(mddev));
c3d9714e
N		 6003 fput(mddev->bitmap_info.file);
6004 mddev->bitmap_info.file = NULL;
36fa3063
N		 6005 return err;
6006 }
c3d9714e 6007 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 6008 } else if (mddev->bitmap == NULL)
6009 return -ENOENT; /* cannot remove what isn't there */
6010 err = 0;
6011 if (mddev->pers) {
6012 mddev->pers->quiesce(mddev, 1);
69e51b44 6013 if (fd >= 0) {
36fa3063 6014 err = bitmap_create(mddev);
69e51b44
N		 6015 if (!err)
6016 err = bitmap_load(mddev);
6017 }
d7375ab3 6018 if (fd < 0 || err) {
36fa3063 6019 bitmap_destroy(mddev);
d7375ab3
N		 6020 fd = -1; /* make sure to put the file */
6021 }
36fa3063 6022 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 6023 }
6024 if (fd < 0) {
c3d9714e
N		 6025 if (mddev->bitmap_info.file) {
6026 restore_bitmap_write_access(mddev->bitmap_info.file);
6027 fput(mddev->bitmap_info.file);
acc55e22 6028 }
c3d9714e 6029 mddev->bitmap_info.file = NULL;
36fa3063
N		 6030 }
6031
32a7627c
N		 6032 return err;
6033}
6034
1da177e4
LT		 6035/*
6036 * set_array_info is used two different ways
6037 * The original usage is when creating a new array.
6038 * In this usage, raid_disks is > 0 and it together with
6039 * level, size, not_persistent,layout,chunksize determine the
6040 * shape of the array.
6041 * This will always create an array with a type-0.90.0 superblock.
6042 * The newer usage is when assembling an array.
6043 * In this case raid_disks will be 0, and the major_version field is
6044 * use to determine which style super-blocks are to be found on the devices.
6045 * The minor and patch _version numbers are also kept incase the
6046 * super_block handler wishes to interpret them.
6047 */
fd01b88c 6048static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
1da177e4
LT		 6049{
6050
6051 if (info->raid_disks == 0) {
6052 /* just setting version number for superblock loading */
6053 if (info->major_version < 0 ||
50511da3 6054 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 6055 super_types[info->major_version].name == NULL) {
6056 /* maybe try to auto-load a module? */
6057 printk(KERN_INFO
6058 "md: superblock version %d not known\n",
6059 info->major_version);
6060 return -EINVAL;
6061 }
6062 mddev->major_version = info->major_version;
6063 mddev->minor_version = info->minor_version;
6064 mddev->patch_version = info->patch_version;
3f9d7b0d 6065 mddev->persistent = !info->not_persistent;
cbd19983
N		 6066 /* ensure mddev_put doesn't delete this now that there
6067 * is some minimal configuration.
6068 */
6069 mddev->ctime = get_seconds();
1da177e4
LT		 6070 return 0;
6071 }
6072 mddev->major_version = MD_MAJOR_VERSION;
6073 mddev->minor_version = MD_MINOR_VERSION;
6074 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6075 mddev->ctime = get_seconds();
6076
6077 mddev->level = info->level;
17115e03 6078 mddev->clevel[0] = 0;
58c0fed4 6079 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 6080 mddev->raid_disks = info->raid_disks;
6081 /* don't set md_minor, it is determined by which /dev/md* was
6082 * openned
6083 */
6084 if (info->state & (1<<MD_SB_CLEAN))
6085 mddev->recovery_cp = MaxSector;
6086 else
6087 mddev->recovery_cp = 0;
6088 mddev->persistent = ! info->not_persistent;
e691063a 6089 mddev->external = 0;
1da177e4
LT		 6090
6091 mddev->layout = info->layout;
9d8f0363 6092 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 6093
6094 mddev->max_disks = MD_SB_DISKS;
6095
e691063a
N		 6096 if (mddev->persistent)
6097 mddev->flags = 0;
850b2b42 6098 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6099
c3d9714e 6100 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6101 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6102 mddev->bitmap_info.offset = 0;
b2a2703c 6103
f6705578
N		 6104 mddev->reshape_position = MaxSector;
6105
1da177e4
LT		 6106 /*
6107 * Generate a 128 bit UUID
6108 */
6109 get_random_bytes(mddev->uuid, 16);
6110
f6705578 6111 mddev->new_level = mddev->level;
664e7c41 6112 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 6113 mddev->new_layout = mddev->layout;
6114 mddev->delta_disks = 0;
2c810cdd 6115 mddev->reshape_backwards = 0;
f6705578 6116
1da177e4
LT		 6117 return 0;
6118}
6119
fd01b88c 6120void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6121{
b522adcd
DW		 6122 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6123
6124 if (mddev->external_size)
6125 return;
6126
1f403624
DW		 6127 mddev->array_sectors = array_sectors;
6128}
6129EXPORT_SYMBOL(md_set_array_sectors);
6130
fd01b88c 6131static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6132{
3cb03002 6133 struct md_rdev *rdev;
a35b0d69 6134 int rv;
d71f9f88 6135 int fit = (num_sectors == 0);
a35b0d69
N		 6136
6137 if (mddev->pers->resize == NULL)
6138 return -EINVAL;
d71f9f88
AN		 6139 /* The "num_sectors" is the number of sectors of each device that
6140 * is used. This can only make sense for arrays with redundancy.
6141 * linear and raid0 always use whatever space is available. We can only
6142 * consider changing this number if no resync or reconstruction is
6143 * happening, and if the new size is acceptable. It must fit before the
0f420358 6144 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 6145 * of each device. If num_sectors is zero, we find the largest size
6146 * that fits.
a35b0d69
N		 6147 */
6148 if (mddev->sync_thread)
6149 return -EBUSY;
a4a6125a 6150
dafb20fa 6151 rdev_for_each(rdev, mddev) {
dd8ac336 6152 sector_t avail = rdev->sectors;
01ab5662 6153
d71f9f88
AN		 6154 if (fit && (num_sectors == 0 || num_sectors > avail))
6155 num_sectors = avail;
6156 if (avail < num_sectors)
a35b0d69
N		 6157 return -ENOSPC;
6158 }
d71f9f88 6159 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 6160 if (!rv)
6161 revalidate_disk(mddev->gendisk);
a35b0d69
N		 6162 return rv;
6163}
6164
fd01b88c 6165static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 6166{
6167 int rv;
c6563a8c 6168 struct md_rdev *rdev;
da943b99 6169 /* change the number of raid disks */
63c70c4f 6170 if (mddev->pers->check_reshape == NULL)
da943b99
N		 6171 return -EINVAL;
6172 if (raid_disks <= 0 ||
233fca36 6173 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 6174 return -EINVAL;
63c70c4f 6175 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 6176 return -EBUSY;
c6563a8c
N		 6177
6178 rdev_for_each(rdev, mddev) {
6179 if (mddev->raid_disks < raid_disks &&
6180 rdev->data_offset < rdev->new_data_offset)
6181 return -EINVAL;
6182 if (mddev->raid_disks > raid_disks &&
6183 rdev->data_offset > rdev->new_data_offset)
6184 return -EINVAL;
6185 }
6186
63c70c4f 6187 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N		 6188 if (mddev->delta_disks < 0)
6189 mddev->reshape_backwards = 1;
6190 else if (mddev->delta_disks > 0)
6191 mddev->reshape_backwards = 0;
63c70c4f
N		 6192
6193 rv = mddev->pers->check_reshape(mddev);
2c810cdd 6194 if (rv < 0) {
de171cb9 6195 mddev->delta_disks = 0;
2c810cdd
N		 6196 mddev->reshape_backwards = 0;
6197 }
da943b99
N		 6198 return rv;
6199}
6200
6201
1da177e4
LT		 6202/*
6203 * update_array_info is used to change the configuration of an
6204 * on-line array.
6205 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6206 * fields in the info are checked against the array.
6207 * Any differences that cannot be handled will cause an error.
6208 * Normally, only one change can be managed at a time.
6209 */
fd01b88c 6210static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6211{
6212 int rv = 0;
6213 int cnt = 0;
36fa3063
N		 6214 int state = 0;
6215
6216 /* calculate expected state,ignoring low bits */
c3d9714e 6217 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 6218 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 6219
6220 if (mddev->major_version != info->major_version ||
6221 mddev->minor_version != info->minor_version ||
6222/* mddev->patch_version != info->patch_version || */
6223 mddev->ctime != info->ctime ||
6224 mddev->level != info->level ||
6225/* mddev->layout != info->layout || */
882d00c1 6226 mddev->persistent != !info->not_persistent ||
9d8f0363 6227 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 6228 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6229 ((state^info->state) & 0xfffffe00)
6230 )
1da177e4
LT		 6231 return -EINVAL;
6232 /* Check there is only one change */
58c0fed4
AN		 6233 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6234 cnt++;
6235 if (mddev->raid_disks != info->raid_disks)
6236 cnt++;
6237 if (mddev->layout != info->layout)
6238 cnt++;
6239 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6240 cnt++;
6241 if (cnt == 0)
6242 return 0;
6243 if (cnt > 1)
6244 return -EINVAL;
1da177e4
LT		 6245
6246 if (mddev->layout != info->layout) {
6247 /* Change layout
6248 * we don't need to do anything at the md level, the
6249 * personality will take care of it all.
6250 */
50ac168a 6251 if (mddev->pers->check_reshape == NULL)
1da177e4 6252 return -EINVAL;
597a711b
N		 6253 else {
6254 mddev->new_layout = info->layout;
50ac168a 6255 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 6256 if (rv)
6257 mddev->new_layout = mddev->layout;
6258 return rv;
6259 }
1da177e4 6260 }
58c0fed4 6261 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6262 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6263
da943b99
N		 6264 if (mddev->raid_disks != info->raid_disks)
6265 rv = update_raid_disks(mddev, info->raid_disks);
6266
36fa3063
N		 6267 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6268 if (mddev->pers->quiesce == NULL)
6269 return -EINVAL;
6270 if (mddev->recovery || mddev->sync_thread)
6271 return -EBUSY;
6272 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6273 /* add the bitmap */
6274 if (mddev->bitmap)
6275 return -EEXIST;
c3d9714e 6276 if (mddev->bitmap_info.default_offset == 0)
36fa3063 6277 return -EINVAL;
c3d9714e
N		 6278 mddev->bitmap_info.offset =
6279 mddev->bitmap_info.default_offset;
6409bb05
N		 6280 mddev->bitmap_info.space =
6281 mddev->bitmap_info.default_space;
36fa3063
N		 6282 mddev->pers->quiesce(mddev, 1);
6283 rv = bitmap_create(mddev);
69e51b44
N		 6284 if (!rv)
6285 rv = bitmap_load(mddev);
36fa3063
N		 6286 if (rv)
6287 bitmap_destroy(mddev);
6288 mddev->pers->quiesce(mddev, 0);
6289 } else {
6290 /* remove the bitmap */
6291 if (!mddev->bitmap)
6292 return -ENOENT;
1ec885cd 6293 if (mddev->bitmap->storage.file)
36fa3063
N		 6294 return -EINVAL;
6295 mddev->pers->quiesce(mddev, 1);
6296 bitmap_destroy(mddev);
6297 mddev->pers->quiesce(mddev, 0);
c3d9714e 6298 mddev->bitmap_info.offset = 0;
36fa3063
N		 6299 }
6300 }
850b2b42 6301 md_update_sb(mddev, 1);
1da177e4
LT		 6302 return rv;
6303}
6304
fd01b88c 6305static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6306{
3cb03002 6307 struct md_rdev *rdev;
1ca69c4b 6308 int err = 0;
1da177e4
LT		 6309
6310 if (mddev->pers == NULL)
6311 return -ENODEV;
6312
1ca69c4b
N		 6313 rcu_read_lock();
6314 rdev = find_rdev_rcu(mddev, dev);
1da177e4 6315 if (!rdev)
1ca69c4b
N		 6316 err = -ENODEV;
6317 else {
6318 md_error(mddev, rdev);
6319 if (!test_bit(Faulty, &rdev->flags))
6320 err = -EBUSY;
6321 }
6322 rcu_read_unlock();
6323 return err;
1da177e4
LT		 6324}
6325
2f9618ce
AN		 6326/*
6327 * We have a problem here : there is no easy way to give a CHS
6328 * virtual geometry. We currently pretend that we have a 2 heads
6329 * 4 sectors (with a BIG number of cylinders...). This drives
6330 * dosfs just mad... ;-)
6331 */
a885c8c4
CH		 6332static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6333{
fd01b88c 6334 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6335
6336 geo->heads = 2;
6337 geo->sectors = 4;
49ce6cea 6338 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6339 return 0;
6340}
6341
a39907fa 6342static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6343 unsigned int cmd, unsigned long arg)
6344{
6345 int err = 0;
6346 void __user *argp = (void __user *)arg;
fd01b88c 6347 struct mddev *mddev = NULL;
e2218350 6348 int ro;
1da177e4 6349
506c9e44
N		 6350 switch (cmd) {
6351 case RAID_VERSION:
6352 case GET_ARRAY_INFO:
6353 case GET_DISK_INFO:
6354 break;
6355 default:
6356 if (!capable(CAP_SYS_ADMIN))
6357 return -EACCES;
6358 }
1da177e4
LT		 6359
6360 /*
6361 * Commands dealing with the RAID driver but not any
6362 * particular array:
6363 */
c02c0aeb
N		 6364 switch (cmd) {
6365 case RAID_VERSION:
6366 err = get_version(argp);
6367 goto done;
1da177e4 6368
c02c0aeb
N		 6369 case PRINT_RAID_DEBUG:
6370 err = 0;
6371 md_print_devices();
6372 goto done;
1da177e4
LT		 6373
6374#ifndef MODULE
c02c0aeb
N		 6375 case RAID_AUTORUN:
6376 err = 0;
6377 autostart_arrays(arg);
6378 goto done;
1da177e4 6379#endif
c02c0aeb 6380 default:;
1da177e4
LT		 6381 }
6382
6383 /*
6384 * Commands creating/starting a new array:
6385 */
6386
a39907fa 6387 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6388
6389 if (!mddev) {
6390 BUG();
6391 goto abort;
6392 }
6393
1ca69c4b
N		 6394 /* Some actions do not requires the mutex */
6395 switch (cmd) {
6396 case GET_ARRAY_INFO:
6397 if (!mddev->raid_disks && !mddev->external)
6398 err = -ENODEV;
6399 else
6400 err = get_array_info(mddev, argp);
6401 goto abort;
6402
6403 case GET_DISK_INFO:
6404 if (!mddev->raid_disks && !mddev->external)
6405 err = -ENODEV;
6406 else
6407 err = get_disk_info(mddev, argp);
6408 goto abort;
6409
6410 case SET_DISK_FAULTY:
6411 err = set_disk_faulty(mddev, new_decode_dev(arg));
6412 goto abort;
6413 }
6414
a7a3f08d
N		 6415 if (cmd == ADD_NEW_DISK)
6416 /* need to ensure md_delayed_delete() has completed */
6417 flush_workqueue(md_misc_wq);
6418
1da177e4
LT		 6419 err = mddev_lock(mddev);
6420 if (err) {
6421 printk(KERN_INFO
6422 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6423 err, cmd);
6424 goto abort;
6425 }
6426
c02c0aeb
N		 6427 if (cmd == SET_ARRAY_INFO) {
6428 mdu_array_info_t info;
6429 if (!arg)
6430 memset(&info, 0, sizeof(info));
6431 else if (copy_from_user(&info, argp, sizeof(info))) {
6432 err = -EFAULT;
6433 goto abort_unlock;
6434 }
6435 if (mddev->pers) {
6436 err = update_array_info(mddev, &info);
6437 if (err) {
6438 printk(KERN_WARNING "md: couldn't update"
6439 " array info. %d\n", err);
6440 goto abort_unlock;
1da177e4
LT		 6441 }
6442 goto done_unlock;
c02c0aeb
N		 6443 }
6444 if (!list_empty(&mddev->disks)) {
6445 printk(KERN_WARNING
6446 "md: array %s already has disks!\n",
6447 mdname(mddev));
6448 err = -EBUSY;
6449 goto abort_unlock;
6450 }
6451 if (mddev->raid_disks) {
6452 printk(KERN_WARNING
6453 "md: array %s already initialised!\n",
6454 mdname(mddev));
6455 err = -EBUSY;
6456 goto abort_unlock;
6457 }
6458 err = set_array_info(mddev, &info);
6459 if (err) {
6460 printk(KERN_WARNING "md: couldn't set"
6461 " array info. %d\n", err);
6462 goto abort_unlock;
6463 }
6464 goto done_unlock;
1da177e4
LT		 6465 }
6466
6467 /*
6468 * Commands querying/configuring an existing array:
6469 */
32a7627c 6470 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6471 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6472 if ((!mddev->raid_disks && !mddev->external)
6473 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6474 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6475 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 6476 err = -ENODEV;
6477 goto abort_unlock;
6478 }
6479
6480 /*
6481 * Commands even a read-only array can execute:
6482 */
c02c0aeb
N		 6483 switch (cmd) {
6484 case GET_BITMAP_FILE:
6485 err = get_bitmap_file(mddev, argp);
6486 goto done_unlock;
32a7627c 6487
c02c0aeb
N		 6488 case RESTART_ARRAY_RW:
6489 err = restart_array(mddev);
6490 goto done_unlock;
1da177e4 6491
c02c0aeb
N		 6492 case STOP_ARRAY:
6493 err = do_md_stop(mddev, 0, bdev);
6494 goto done_unlock;
1da177e4 6495
c02c0aeb
N		 6496 case STOP_ARRAY_RO:
6497 err = md_set_readonly(mddev, bdev);
6498 goto done_unlock;
1da177e4 6499
3ea8929d
N		 6500 case HOT_REMOVE_DISK:
6501 err = hot_remove_disk(mddev, new_decode_dev(arg));
6502 goto done_unlock;
6503
7ceb17e8
N		 6504 case ADD_NEW_DISK:
6505 /* We can support ADD_NEW_DISK on read-only arrays
6506 * on if we are re-adding a preexisting device.
6507 * So require mddev->pers and MD_DISK_SYNC.
6508 */
6509 if (mddev->pers) {
6510 mdu_disk_info_t info;
6511 if (copy_from_user(&info, argp, sizeof(info)))
6512 err = -EFAULT;
6513 else if (!(info.state & (1<<MD_DISK_SYNC)))
6514 /* Need to clear read-only for this */
6515 break;
6516 else
6517 err = add_new_disk(mddev, &info);
6518 goto done_unlock;
6519 }
6520 break;
6521
c02c0aeb
N		 6522 case BLKROSET:
6523 if (get_user(ro, (int __user *)(arg))) {
6524 err = -EFAULT;
6525 goto done_unlock;
6526 }
6527 err = -EINVAL;
e2218350 6528
c02c0aeb
N		 6529 /* if the bdev is going readonly the value of mddev->ro
6530 * does not matter, no writes are coming
6531 */
6532 if (ro)
6533 goto done_unlock;
e2218350 6534
c02c0aeb
N		 6535 /* are we are already prepared for writes? */
6536 if (mddev->ro != 1)
6537 goto done_unlock;
e2218350 6538
c02c0aeb
N		 6539 /* transitioning to readauto need only happen for
6540 * arrays that call md_write_start
6541 */
6542 if (mddev->pers) {
6543 err = restart_array(mddev);
6544 if (err == 0) {
6545 mddev->ro = 2;
6546 set_disk_ro(mddev->gendisk, 0);
e2218350 6547 }
c02c0aeb
N		 6548 }
6549 goto done_unlock;
1da177e4
LT		 6550 }
6551
6552 /*
6553 * The remaining ioctls are changing the state of the
f91de92e
N		 6554 * superblock, so we do not allow them on read-only arrays.
6555 * However non-MD ioctls (e.g. get-size) will still come through
6556 * here and hit the 'default' below, so only disallow
6557 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 6558 */
bb57fc64 6559 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 6560 if (mddev->ro == 2) {
6561 mddev->ro = 0;
00bcb4ac 6562 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 6563 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N		 6564 /* mddev_unlock will wake thread */
6565 /* If a device failed while we were read-only, we
6566 * need to make sure the metadata is updated now.
6567 */
6568 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6569 mddev_unlock(mddev);
6570 wait_event(mddev->sb_wait,
6571 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6572 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6573 mddev_lock(mddev);
6574 }
f91de92e
N		 6575 } else {
6576 err = -EROFS;
6577 goto abort_unlock;
6578 }
1da177e4
LT		 6579 }
6580
c02c0aeb
N		 6581 switch (cmd) {
6582 case ADD_NEW_DISK:
1da177e4 6583 {
c02c0aeb
N		 6584 mdu_disk_info_t info;
6585 if (copy_from_user(&info, argp, sizeof(info)))
6586 err = -EFAULT;
6587 else
6588 err = add_new_disk(mddev, &info);
6589 goto done_unlock;
6590 }
1da177e4 6591
c02c0aeb
N		 6592 case HOT_ADD_DISK:
6593 err = hot_add_disk(mddev, new_decode_dev(arg));
6594 goto done_unlock;
1da177e4 6595
c02c0aeb
N		 6596 case RUN_ARRAY:
6597 err = do_md_run(mddev);
6598 goto done_unlock;
1da177e4 6599
c02c0aeb
N		 6600 case SET_BITMAP_FILE:
6601 err = set_bitmap_file(mddev, (int)arg);
6602 goto done_unlock;
32a7627c 6603
c02c0aeb
N		 6604 default:
6605 err = -EINVAL;
6606 goto abort_unlock;
1da177e4
LT		 6607 }
6608
6609done_unlock:
6610abort_unlock:
d3374825
N		 6611 if (mddev->hold_active == UNTIL_IOCTL &&
6612 err != -EINVAL)
6613 mddev->hold_active = 0;
1da177e4
LT		 6614 mddev_unlock(mddev);
6615
6616 return err;
6617done:
6618 if (err)
6619 MD_BUG();
6620abort:
6621 return err;
6622}
aa98aa31
AB		 6623#ifdef CONFIG_COMPAT
6624static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6625 unsigned int cmd, unsigned long arg)
6626{
6627 switch (cmd) {
6628 case HOT_REMOVE_DISK:
6629 case HOT_ADD_DISK:
6630 case SET_DISK_FAULTY:
6631 case SET_BITMAP_FILE:
6632 /* These take in integer arg, do not convert */
6633 break;
6634 default:
6635 arg = (unsigned long)compat_ptr(arg);
6636 break;
6637 }
6638
6639 return md_ioctl(bdev, mode, cmd, arg);
6640}
6641#endif /* CONFIG_COMPAT */
1da177e4 6642
a39907fa 6643static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6644{
6645 /*
6646 * Succeed if we can lock the mddev, which confirms that
6647 * it isn't being stopped right now.
6648 */
fd01b88c 6649 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6650 int err;
6651
0c098220
YL		 6652 if (!mddev)
6653 return -ENODEV;
6654
d3374825
N		 6655 if (mddev->gendisk != bdev->bd_disk) {
6656 /* we are racing with mddev_put which is discarding this
6657 * bd_disk.
6658 */
6659 mddev_put(mddev);
6660 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6661 flush_workqueue(md_misc_wq);
d3374825
N		 6662 /* Then retry the open from the top */
6663 return -ERESTARTSYS;
6664 }
6665 BUG_ON(mddev != bdev->bd_disk->private_data);
6666
c8c00a69 6667 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6668 goto out;
6669
6670 err = 0;
f2ea68cf 6671 atomic_inc(&mddev->openers);
c8c00a69 6672 mutex_unlock(&mddev->open_mutex);
1da177e4 6673
f0b4f7e2 6674 check_disk_change(bdev);
1da177e4
LT		 6675 out:
6676 return err;
6677}
6678
db2a144b 6679static void md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6680{
fd01b88c 6681 struct mddev *mddev = disk->private_data;
1da177e4 6682
52e5f9d1 6683 BUG_ON(!mddev);
f2ea68cf 6684 atomic_dec(&mddev->openers);
1da177e4 6685 mddev_put(mddev);
1da177e4 6686}
f0b4f7e2
N		 6687
6688static int md_media_changed(struct gendisk *disk)
6689{
fd01b88c 6690 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6691
6692 return mddev->changed;
6693}
6694
6695static int md_revalidate(struct gendisk *disk)
6696{
fd01b88c 6697 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6698
6699 mddev->changed = 0;
6700 return 0;
6701}
83d5cde4 6702static const struct block_device_operations md_fops =
1da177e4
LT		 6703{
6704 .owner = THIS_MODULE,
a39907fa
AV		 6705 .open = md_open,
6706 .release = md_release,
b492b852 6707 .ioctl = md_ioctl,
aa98aa31
AB		 6708#ifdef CONFIG_COMPAT
6709 .compat_ioctl = md_compat_ioctl,
6710#endif
a885c8c4 6711 .getgeo = md_getgeo,
f0b4f7e2
N		 6712 .media_changed = md_media_changed,
6713 .revalidate_disk= md_revalidate,
1da177e4
LT		 6714};
6715
75c96f85 6716static int md_thread(void * arg)
1da177e4 6717{
2b8bf345 6718 struct md_thread *thread = arg;
1da177e4 6719
1da177e4
LT		 6720 /*
6721 * md_thread is a 'system-thread', it's priority should be very
6722 * high. We avoid resource deadlocks individually in each
6723 * raid personality. (RAID5 does preallocation) We also use RR and
6724 * the very same RT priority as kswapd, thus we will never get
6725 * into a priority inversion deadlock.
6726 *
6727 * we definitely have to have equal or higher priority than
6728 * bdflush, otherwise bdflush will deadlock if there are too
6729 * many dirty RAID5 blocks.
6730 */
1da177e4 6731
6985c43f 6732 allow_signal(SIGKILL);
a6fb0934 6733 while (!kthread_should_stop()) {
1da177e4 6734
93588e22
N		 6735 /* We need to wait INTERRUPTIBLE so that
6736 * we don't add to the load-average.
6737 * That means we need to be sure no signals are
6738 * pending
6739 */
6740 if (signal_pending(current))
6741 flush_signals(current);
6742
6743 wait_event_interruptible_timeout
6744 (thread->wqueue,
6745 test_bit(THREAD_WAKEUP, &thread->flags)
6746 || kthread_should_stop(),
6747 thread->timeout);
1da177e4 6748
6c987910
N		 6749 clear_bit(THREAD_WAKEUP, &thread->flags);
6750 if (!kthread_should_stop())
4ed8731d 6751 thread->run(thread);
1da177e4 6752 }
a6fb0934 6753
1da177e4
LT		 6754 return 0;
6755}
6756
2b8bf345 6757void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6758{
6759 if (thread) {
36a4e1fe 6760 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6761 set_bit(THREAD_WAKEUP, &thread->flags);
6762 wake_up(&thread->wqueue);
6763 }
6764}
6765
4ed8731d
SL		 6766struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6767 struct mddev *mddev, const char *name)
1da177e4 6768{
2b8bf345 6769 struct md_thread *thread;
1da177e4 6770
2b8bf345 6771 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6772 if (!thread)
6773 return NULL;
6774
1da177e4
LT		 6775 init_waitqueue_head(&thread->wqueue);
6776
1da177e4
LT		 6777 thread->run = run;
6778 thread->mddev = mddev;
32a7627c 6779 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6780 thread->tsk = kthread_run(md_thread, thread,
6781 "%s_%s",
6782 mdname(thread->mddev),
0232605d 6783 name);
a6fb0934 6784 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6785 kfree(thread);
6786 return NULL;
6787 }
1da177e4
LT		 6788 return thread;
6789}
6790
2b8bf345 6791void md_unregister_thread(struct md_thread **threadp)
1da177e4 6792{
2b8bf345 6793 struct md_thread *thread = *threadp;
e0cf8f04
N		 6794 if (!thread)
6795 return;
36a4e1fe 6796 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 6797 /* Locking ensures that mddev_unlock does not wake_up a
6798 * non-existent thread
6799 */
6800 spin_lock(&pers_lock);
6801 *threadp = NULL;
6802 spin_unlock(&pers_lock);
a6fb0934
N		 6803
6804 kthread_stop(thread->tsk);
1da177e4
LT		 6805 kfree(thread);
6806}
6807
fd01b88c 6808void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 6809{
6810 if (!mddev) {
6811 MD_BUG();
6812 return;
6813 }
6814
b2d444d7 6815 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6816 return;
6bfe0b49 6817
de393cde 6818 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 6819 return;
6820 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6821 if (mddev->degraded)
6822 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6823 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6824 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6825 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6826 md_wakeup_thread(mddev->thread);
768a418d 6827 if (mddev->event_work.func)
e804ac78 6828 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6829 md_new_event_inintr(mddev);
1da177e4
LT		 6830}
6831
6832/* seq_file implementation /proc/mdstat */
6833
6834static void status_unused(struct seq_file *seq)
6835{
6836 int i = 0;
3cb03002 6837 struct md_rdev *rdev;
1da177e4
LT		 6838
6839 seq_printf(seq, "unused devices: ");
6840
159ec1fc 6841 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6842 char b[BDEVNAME_SIZE];
6843 i++;
6844 seq_printf(seq, "%s ",
6845 bdevname(rdev->bdev,b));
6846 }
6847 if (!i)
6848 seq_printf(seq, "<none>");
6849
6850 seq_printf(seq, "\n");
6851}
6852
6853
fd01b88c 6854static void status_resync(struct seq_file *seq, struct mddev * mddev)
1da177e4 6855{
dd71cf6b
N		 6856 sector_t max_sectors, resync, res;
6857 unsigned long dt, db;
6858 sector_t rt;
4588b42e
N		 6859 int scale;
6860 unsigned int per_milli;
1da177e4 6861
72f36d59
N		 6862 if (mddev->curr_resync <= 3)
6863 resync = 0;
6864 else
6865 resync = mddev->curr_resync
6866 - atomic_read(&mddev->recovery_active);
1da177e4 6867
c804cdec
N		 6868 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6869 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 6870 max_sectors = mddev->resync_max_sectors;
1da177e4 6871 else
dd71cf6b 6872 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6873
6874 /*
6875 * Should not happen.
6876 */
dd71cf6b 6877 if (!max_sectors) {
1da177e4
LT		 6878 MD_BUG();
6879 return;
6880 }
4588b42e 6881 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6882 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6883 * u32, as those are the requirements for sector_div.
6884 * Thus 'scale' must be at least 10
6885 */
6886 scale = 10;
6887 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6888 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6889 scale++;
6890 }
6891 res = (resync>>scale)*1000;
dd71cf6b 6892 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6893
6894 per_milli = res;
1da177e4 6895 {
4588b42e 6896 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6897 seq_printf(seq, "[");
6898 for (i = 0; i < x; i++)
6899 seq_printf(seq, "=");
6900 seq_printf(seq, ">");
6901 for (i = 0; i < y; i++)
6902 seq_printf(seq, ".");
6903 seq_printf(seq, "] ");
6904 }
4588b42e 6905 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6906 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6907 "reshape" :
61df9d91
N		 6908 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6909 "check" :
6910 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6911 "resync" : "recovery"))),
6912 per_milli/10, per_milli % 10,
dd71cf6b
N		 6913 (unsigned long long) resync/2,
6914 (unsigned long long) max_sectors/2);
1da177e4
LT		 6915
6916 /*
1da177e4
LT		 6917 * dt: time from mark until now
6918 * db: blocks written from mark until now
6919 * rt: remaining time
dd71cf6b
N		 6920 *
6921 * rt is a sector_t, so could be 32bit or 64bit.
6922 * So we divide before multiply in case it is 32bit and close
6923 * to the limit.
25985edc 6924 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 6925 * near the end of resync when the number of remaining sectors
6926 * is close to 'db'.
6927 * We then divide rt by 32 after multiplying by db to compensate.
6928 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6929 */
6930 dt = ((jiffies - mddev->resync_mark) / HZ);
6931 if (!dt) dt++;
ff4e8d9a
N		 6932 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6933 - mddev->resync_mark_cnt;
1da177e4 6934
dd71cf6b
N		 6935 rt = max_sectors - resync; /* number of remaining sectors */
6936 sector_div(rt, db/32+1);
6937 rt *= dt;
6938 rt >>= 5;
6939
6940 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6941 ((unsigned long)rt % 60)/6);
1da177e4 6942
ff4e8d9a 6943 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6944}
6945
6946static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6947{
6948 struct list_head *tmp;
6949 loff_t l = *pos;
fd01b88c 6950 struct mddev *mddev;
1da177e4
LT		 6951
6952 if (l >= 0x10000)
6953 return NULL;
6954 if (!l--)
6955 /* header */
6956 return (void*)1;
6957
6958 spin_lock(&all_mddevs_lock);
6959 list_for_each(tmp,&all_mddevs)
6960 if (!l--) {
fd01b88c 6961 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 6962 mddev_get(mddev);
6963 spin_unlock(&all_mddevs_lock);
6964 return mddev;
6965 }
6966 spin_unlock(&all_mddevs_lock);
6967 if (!l--)
6968 return (void*)2;/* tail */
6969 return NULL;
6970}
6971
6972static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6973{
6974 struct list_head *tmp;
fd01b88c 6975 struct mddev *next_mddev, *mddev = v;
1da177e4
LT		 6976
6977 ++*pos;
6978 if (v == (void*)2)
6979 return NULL;
6980
6981 spin_lock(&all_mddevs_lock);
6982 if (v == (void*)1)
6983 tmp = all_mddevs.next;
6984 else
6985 tmp = mddev->all_mddevs.next;
6986 if (tmp != &all_mddevs)
fd01b88c 6987 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 6988 else {
6989 next_mddev = (void*)2;
6990 *pos = 0x10000;
6991 }
6992 spin_unlock(&all_mddevs_lock);
6993
6994 if (v != (void*)1)
6995 mddev_put(mddev);
6996 return next_mddev;
6997
6998}
6999
7000static void md_seq_stop(struct seq_file *seq, void *v)
7001{
fd01b88c 7002 struct mddev *mddev = v;
1da177e4
LT		 7003
7004 if (mddev && v != (void*)1 && v != (void*)2)
7005 mddev_put(mddev);
7006}
7007
7008static int md_seq_show(struct seq_file *seq, void *v)
7009{
fd01b88c 7010 struct mddev *mddev = v;
dd8ac336 7011 sector_t sectors;
3cb03002 7012 struct md_rdev *rdev;
1da177e4
LT		 7013
7014 if (v == (void*)1) {
84fc4b56 7015 struct md_personality *pers;
1da177e4
LT		 7016 seq_printf(seq, "Personalities : ");
7017 spin_lock(&pers_lock);
2604b703
N		 7018 list_for_each_entry(pers, &pers_list, list)
7019 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 7020
7021 spin_unlock(&pers_lock);
7022 seq_printf(seq, "\n");
f1514638 7023 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7024 return 0;
7025 }
7026 if (v == (void*)2) {
7027 status_unused(seq);
7028 return 0;
7029 }
7030
5dc5cf7d 7031 if (mddev_lock(mddev) < 0)
1da177e4 7032 return -EINTR;
5dc5cf7d 7033
1da177e4
LT		 7034 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7035 seq_printf(seq, "%s : %sactive", mdname(mddev),
7036 mddev->pers ? "" : "in");
7037 if (mddev->pers) {
f91de92e 7038 if (mddev->ro==1)
1da177e4 7039 seq_printf(seq, " (read-only)");
f91de92e 7040 if (mddev->ro==2)
52720ae7 7041 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 7042 seq_printf(seq, " %s", mddev->pers->name);
7043 }
7044
dd8ac336 7045 sectors = 0;
dafb20fa 7046 rdev_for_each(rdev, mddev) {
1da177e4
LT		 7047 char b[BDEVNAME_SIZE];
7048 seq_printf(seq, " %s[%d]",
7049 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 7050 if (test_bit(WriteMostly, &rdev->flags))
7051 seq_printf(seq, "(W)");
b2d444d7 7052 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7053 seq_printf(seq, "(F)");
7054 continue;
2d78f8c4
N		 7055 }
7056 if (rdev->raid_disk < 0)
b325a32e 7057 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 7058 if (test_bit(Replacement, &rdev->flags))
7059 seq_printf(seq, "(R)");
dd8ac336 7060 sectors += rdev->sectors;
1da177e4
LT		 7061 }
7062
7063 if (!list_empty(&mddev->disks)) {
7064 if (mddev->pers)
7065 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 7066 (unsigned long long)
7067 mddev->array_sectors / 2);
1da177e4
LT		 7068 else
7069 seq_printf(seq, "\n %llu blocks",
dd8ac336 7070 (unsigned long long)sectors / 2);
1da177e4 7071 }
1cd6bf19
N		 7072 if (mddev->persistent) {
7073 if (mddev->major_version != 0 ||
7074 mddev->minor_version != 90) {
7075 seq_printf(seq," super %d.%d",
7076 mddev->major_version,
7077 mddev->minor_version);
7078 }
e691063a
N		 7079 } else if (mddev->external)
7080 seq_printf(seq, " super external:%s",
7081 mddev->metadata_type);
7082 else
1cd6bf19 7083 seq_printf(seq, " super non-persistent");
1da177e4
LT		 7084
7085 if (mddev->pers) {
d710e138 7086 mddev->pers->status(seq, mddev);
1da177e4 7087 seq_printf(seq, "\n ");
8e1b39d6
N		 7088 if (mddev->pers->sync_request) {
7089 if (mddev->curr_resync > 2) {
d710e138 7090 status_resync(seq, mddev);
8e1b39d6 7091 seq_printf(seq, "\n ");
72f36d59 7092 } else if (mddev->curr_resync >= 1)
8e1b39d6
N		 7093 seq_printf(seq, "\tresync=DELAYED\n ");
7094 else if (mddev->recovery_cp < MaxSector)
7095 seq_printf(seq, "\tresync=PENDING\n ");
7096 }
32a7627c
N		 7097 } else
7098 seq_printf(seq, "\n ");
7099
57148964 7100 bitmap_status(seq, mddev->bitmap);
1da177e4
LT		 7101
7102 seq_printf(seq, "\n");
7103 }
7104 mddev_unlock(mddev);
7105
7106 return 0;
7107}
7108
110518bc 7109static const struct seq_operations md_seq_ops = {
1da177e4
LT		 7110 .start = md_seq_start,
7111 .next = md_seq_next,
7112 .stop = md_seq_stop,
7113 .show = md_seq_show,
7114};
7115
7116static int md_seq_open(struct inode *inode, struct file *file)
7117{
f1514638 7118 struct seq_file *seq;
1da177e4
LT		 7119 int error;
7120
7121 error = seq_open(file, &md_seq_ops);
d7603b7e 7122 if (error)
f1514638
KS		 7123 return error;
7124
7125 seq = file->private_data;
7126 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7127 return error;
7128}
7129
d7603b7e
N		 7130static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7131{
f1514638 7132 struct seq_file *seq = filp->private_data;
d7603b7e
N		 7133 int mask;
7134
7135 poll_wait(filp, &md_event_waiters, wait);
7136
7137 /* always allow read */
7138 mask = POLLIN | POLLRDNORM;
7139
f1514638 7140 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 7141 mask |= POLLERR | POLLPRI;
7142 return mask;
7143}
7144
fa027c2a 7145static const struct file_operations md_seq_fops = {
e24650c2 7146 .owner = THIS_MODULE,
1da177e4
LT		 7147 .open = md_seq_open,
7148 .read = seq_read,
7149 .llseek = seq_lseek,
c3f94b40 7150 .release = seq_release_private,
d7603b7e 7151 .poll = mdstat_poll,
1da177e4
LT		 7152};
7153
84fc4b56 7154int register_md_personality(struct md_personality *p)
1da177e4 7155{
1da177e4 7156 spin_lock(&pers_lock);
2604b703
N		 7157 list_add_tail(&p->list, &pers_list);
7158 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 7159 spin_unlock(&pers_lock);
7160 return 0;
7161}
7162
84fc4b56 7163int unregister_md_personality(struct md_personality *p)
1da177e4 7164{
2604b703 7165 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7166 spin_lock(&pers_lock);
2604b703 7167 list_del_init(&p->list);
1da177e4
LT		 7168 spin_unlock(&pers_lock);
7169 return 0;
7170}
7171
fd01b88c 7172static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7173{
3cb03002 7174 struct md_rdev * rdev;
1da177e4 7175 int idle;
eea1bf38 7176 int curr_events;
1da177e4
LT		 7177
7178 idle = 1;
4b80991c
N		 7179 rcu_read_lock();
7180 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7181 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 7182 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7183 (int)part_stat_read(&disk->part0, sectors[1]) -
7184 atomic_read(&disk->sync_io);
713f6ab1
N		 7185 /* sync IO will cause sync_io to increase before the disk_stats
7186 * as sync_io is counted when a request starts, and
7187 * disk_stats is counted when it completes.
7188 * So resync activity will cause curr_events to be smaller than
7189 * when there was no such activity.
7190 * non-sync IO will cause disk_stat to increase without
7191 * increasing sync_io so curr_events will (eventually)
7192 * be larger than it was before. Once it becomes
7193 * substantially larger, the test below will cause
7194 * the array to appear non-idle, and resync will slow
7195 * down.
7196 * If there is a lot of outstanding resync activity when
7197 * we set last_event to curr_events, then all that activity
7198 * completing might cause the array to appear non-idle
7199 * and resync will be slowed down even though there might
7200 * not have been non-resync activity. This will only
7201 * happen once though. 'last_events' will soon reflect
7202 * the state where there is little or no outstanding
7203 * resync requests, and further resync activity will
7204 * always make curr_events less than last_events.
c0e48521 7205 *
1da177e4 7206 */
eea1bf38 7207 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 7208 rdev->last_events = curr_events;
7209 idle = 0;
7210 }
7211 }
4b80991c 7212 rcu_read_unlock();
1da177e4
LT		 7213 return idle;
7214}
7215
fd01b88c 7216void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 7217{
7218 /* another "blocks" (512byte) blocks have been synced */
7219 atomic_sub(blocks, &mddev->recovery_active);
7220 wake_up(&mddev->recovery_wait);
7221 if (!ok) {
dfc70645 7222 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 7223 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT		 7224 md_wakeup_thread(mddev->thread);
7225 // stop recovery, signal do_sync ....
7226 }
7227}
7228
7229
06d91a5f
N		 7230/* md_write_start(mddev, bi)
7231 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 7232 * in superblock) before writing, schedule a superblock update
7233 * and wait for it to complete.
06d91a5f 7234 */
fd01b88c 7235void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7236{
0fd62b86 7237 int did_change = 0;
06d91a5f 7238 if (bio_data_dir(bi) != WRITE)
3d310eb7 7239 return;
06d91a5f 7240
f91de92e
N		 7241 BUG_ON(mddev->ro == 1);
7242 if (mddev->ro == 2) {
7243 /* need to switch to read/write */
7244 mddev->ro = 0;
7245 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7246 md_wakeup_thread(mddev->thread);
25156198 7247 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7248 did_change = 1;
f91de92e 7249 }
06d91a5f 7250 atomic_inc(&mddev->writes_pending);
31a59e34
N		 7251 if (mddev->safemode == 1)
7252 mddev->safemode = 0;
06d91a5f 7253 if (mddev->in_sync) {
a9701a30 7254 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 7255 if (mddev->in_sync) {
7256 mddev->in_sync = 0;
850b2b42 7257 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7258 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7259 md_wakeup_thread(mddev->thread);
0fd62b86 7260 did_change = 1;
3d310eb7 7261 }
a9701a30 7262 spin_unlock_irq(&mddev->write_lock);
06d91a5f 7263 }
0fd62b86 7264 if (did_change)
00bcb4ac 7265 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7266 wait_event(mddev->sb_wait,
09a44cc1 7267 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 7268}
7269
fd01b88c 7270void md_write_end(struct mddev *mddev)
1da177e4
LT		 7271{
7272 if (atomic_dec_and_test(&mddev->writes_pending)) {
7273 if (mddev->safemode == 2)
7274 md_wakeup_thread(mddev->thread);
16f17b39 7275 else if (mddev->safemode_delay)
1da177e4
LT		 7276 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7277 }
7278}
7279
2a2275d6
N		 7280/* md_allow_write(mddev)
7281 * Calling this ensures that the array is marked 'active' so that writes
7282 * may proceed without blocking. It is important to call this before
7283 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7284 * Must be called with mddev_lock held.
b5470dc5
DW		 7285 *
7286 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7287 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7288 */
fd01b88c 7289int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7290{
7291 if (!mddev->pers)
b5470dc5 7292 return 0;
2a2275d6 7293 if (mddev->ro)
b5470dc5 7294 return 0;
1a0fd497 7295 if (!mddev->pers->sync_request)
b5470dc5 7296 return 0;
2a2275d6
N		 7297
7298 spin_lock_irq(&mddev->write_lock);
7299 if (mddev->in_sync) {
7300 mddev->in_sync = 0;
7301 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7302 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7303 if (mddev->safemode_delay &&
7304 mddev->safemode == 0)
7305 mddev->safemode = 1;
7306 spin_unlock_irq(&mddev->write_lock);
7307 md_update_sb(mddev, 0);
00bcb4ac 7308 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 7309 } else
7310 spin_unlock_irq(&mddev->write_lock);
b5470dc5 7311
070dc6dd 7312 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7313 return -EAGAIN;
7314 else
7315 return 0;
2a2275d6
N		 7316}
7317EXPORT_SYMBOL_GPL(md_allow_write);
7318
1da177e4
LT		 7319#define SYNC_MARKS 10
7320#define SYNC_MARK_STEP (3*HZ)
54f89341 7321#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 7322void md_do_sync(struct md_thread *thread)
1da177e4 7323{
4ed8731d 7324 struct mddev *mddev = thread->mddev;
fd01b88c 7325 struct mddev *mddev2;
1da177e4
LT		 7326 unsigned int currspeed = 0,
7327 window;
57afd89f 7328 sector_t max_sectors,j, io_sectors;
1da177e4 7329 unsigned long mark[SYNC_MARKS];
54f89341 7330 unsigned long update_time;
1da177e4
LT		 7331 sector_t mark_cnt[SYNC_MARKS];
7332 int last_mark,m;
7333 struct list_head *tmp;
7334 sector_t last_check;
57afd89f 7335 int skipped = 0;
3cb03002 7336 struct md_rdev *rdev;
61df9d91 7337 char *desc;
7c2c57c9 7338 struct blk_plug plug;
1da177e4
LT		 7339
7340 /* just incase thread restarts... */
7341 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7342 return;
d86a5d29
N		 7343 if (mddev->ro) {/* never try to sync a read-only array */
7344 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5fd6c1dc 7345 return;
d86a5d29 7346 }
1da177e4 7347
61df9d91
N		 7348 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7349 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7350 desc = "data-check";
7351 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7352 desc = "requested-resync";
7353 else
7354 desc = "resync";
7355 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7356 desc = "reshape";
7357 else
7358 desc = "recovery";
7359
1da177e4
LT		 7360 /* we overload curr_resync somewhat here.
7361 * 0 == not engaged in resync at all
7362 * 2 == checking that there is no conflict with another sync
7363 * 1 == like 2, but have yielded to allow conflicting resync to
7364 * commense
7365 * other == active in resync - this many blocks
7366 *
7367 * Before starting a resync we must have set curr_resync to
7368 * 2, and then checked that every "conflicting" array has curr_resync
7369 * less than ours. When we find one that is the same or higher
7370 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7371 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7372 * This will mean we have to start checking from the beginning again.
7373 *
7374 */
7375
7376 do {
7377 mddev->curr_resync = 2;
7378
7379 try_again:
404e4b43 7380 if (kthread_should_stop())
6985c43f 7381 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 7382
7383 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7384 goto skip;
29ac4aa3 7385 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7386 if (mddev2 == mddev)
7387 continue;
90b08710
BS		 7388 if (!mddev->parallel_resync
7389 && mddev2->curr_resync
7390 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7391 DEFINE_WAIT(wq);
7392 if (mddev < mddev2 && mddev->curr_resync == 2) {
7393 /* arbitrarily yield */
7394 mddev->curr_resync = 1;
7395 wake_up(&resync_wait);
7396 }
7397 if (mddev > mddev2 && mddev->curr_resync == 1)
7398 /* no need to wait here, we can wait the next
7399 * time 'round when curr_resync == 2
7400 */
7401 continue;
9744197c
N		 7402 /* We need to wait 'interruptible' so as not to
7403 * contribute to the load average, and not to
7404 * be caught by 'softlockup'
7405 */
7406 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 7407 if (!kthread_should_stop() &&
8712e553 7408 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7409 printk(KERN_INFO "md: delaying %s of %s"
7410 " until %s has finished (they"
1da177e4 7411 " share one or more physical units)\n",
61df9d91 7412 desc, mdname(mddev), mdname(mddev2));
1da177e4 7413 mddev_put(mddev2);
9744197c
N		 7414 if (signal_pending(current))
7415 flush_signals(current);
1da177e4
LT		 7416 schedule();
7417 finish_wait(&resync_wait, &wq);
7418 goto try_again;
7419 }
7420 finish_wait(&resync_wait, &wq);
7421 }
7422 }
7423 } while (mddev->curr_resync < 2);
7424
5fd6c1dc 7425 j = 0;
9d88883e 7426 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7427 /* resync follows the size requested by the personality,
57afd89f 7428 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7429 */
7430 max_sectors = mddev->resync_max_sectors;
7f7583d4 7431 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 7432 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7433 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7434 j = mddev->resync_min;
7435 else if (!mddev->bitmap)
5fd6c1dc 7436 j = mddev->recovery_cp;
5e96ee65 7437
ccfcc3c1 7438 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7439 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7440 else {
1da177e4 7441 /* recovery follows the physical size of devices */
58c0fed4 7442 max_sectors = mddev->dev_sectors;
5fd6c1dc 7443 j = MaxSector;
4e59ca7d 7444 rcu_read_lock();
dafb20fa 7445 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc
N		 7446 if (rdev->raid_disk >= 0 &&
7447 !test_bit(Faulty, &rdev->flags) &&
7448 !test_bit(In_sync, &rdev->flags) &&
7449 rdev->recovery_offset < j)
7450 j = rdev->recovery_offset;
4e59ca7d 7451 rcu_read_unlock();
527d0194
N		 7452
7453 /* If there is a bitmap, we need to make sure all
7454 * writes that started before we added a spare
7455 * complete before we start doing a recovery.
7456 * Otherwise the write might complete and (via
7457 * bitmap_endwrite) set a bit in the bitmap after the
7458 * recovery has checked that bit and skipped that
7459 * region.
7460 */
7461 if (mddev->bitmap) {
7462 mddev->pers->quiesce(mddev, 1);
7463 mddev->pers->quiesce(mddev, 0);
7464 }
5fd6c1dc 7465 }
1da177e4 7466
61df9d91
N		 7467 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7468 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7469 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7470 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7471 "(but not more than %d KB/sec) for %s.\n",
7472 speed_max(mddev), desc);
1da177e4 7473
eea1bf38 7474 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7475
57afd89f 7476 io_sectors = 0;
1da177e4
LT		 7477 for (m = 0; m < SYNC_MARKS; m++) {
7478 mark[m] = jiffies;
57afd89f 7479 mark_cnt[m] = io_sectors;
1da177e4
LT		 7480 }
7481 last_mark = 0;
7482 mddev->resync_mark = mark[last_mark];
7483 mddev->resync_mark_cnt = mark_cnt[last_mark];
7484
7485 /*
7486 * Tune reconstruction:
7487 */
7488 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7489 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7490 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7491
7492 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7493 last_check = 0;
7494
7495 if (j>2) {
7496 printk(KERN_INFO
61df9d91
N		 7497 "md: resuming %s of %s from checkpoint.\n",
7498 desc, mdname(mddev));
1da177e4 7499 mddev->curr_resync = j;
72f36d59
N		 7500 } else
7501 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 7502 mddev->curr_resync_completed = j;
72f36d59
N		 7503 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7504 md_new_event(mddev);
54f89341 7505 update_time = jiffies;
1da177e4 7506
7c2c57c9 7507 blk_start_plug(&plug);
1da177e4 7508 while (j < max_sectors) {
57afd89f 7509 sector_t sectors;
1da177e4 7510
57afd89f 7511 skipped = 0;
97e4f42d 7512
7a91ee1f
N		 7513 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7514 ((mddev->curr_resync > mddev->curr_resync_completed &&
7515 (mddev->curr_resync - mddev->curr_resync_completed)
7516 > (max_sectors >> 4)) ||
54f89341 7517 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f
N		 7518 (j - mddev->curr_resync_completed)*2
7519 >= mddev->resync_max - mddev->curr_resync_completed
7520 )) {
97e4f42d 7521 /* time to update curr_resync_completed */
97e4f42d
N		 7522 wait_event(mddev->recovery_wait,
7523 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7524 mddev->curr_resync_completed = j;
35d78c66 7525 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7526 j > mddev->recovery_cp)
7527 mddev->recovery_cp = j;
54f89341 7528 update_time = jiffies;
070dc6dd 7529 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7530 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7531 }
acb180b0 7532
e62e58a5
N		 7533 while (j >= mddev->resync_max && !kthread_should_stop()) {
7534 /* As this condition is controlled by user-space,
7535 * we can block indefinitely, so use '_interruptible'
7536 * to avoid triggering warnings.
7537 */
7538 flush_signals(current); /* just in case */
7539 wait_event_interruptible(mddev->recovery_wait,
7540 mddev->resync_max > j
7541 || kthread_should_stop());
7542 }
acb180b0
N		 7543
7544 if (kthread_should_stop())
7545 goto interrupted;
7546
57afd89f 7547 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7548 currspeed < speed_min(mddev));
57afd89f 7549 if (sectors == 0) {
dfc70645 7550 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 7551 goto out;
7552 }
57afd89f
N		 7553
7554 if (!skipped) { /* actual IO requested */
7555 io_sectors += sectors;
7556 atomic_add(sectors, &mddev->recovery_active);
7557 }
7558
e875ecea
N		 7559 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7560 break;
7561
1da177e4 7562 j += sectors;
72f36d59
N		 7563 if (j > 2)
7564 mddev->curr_resync = j;
ff4e8d9a 7565 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7566 if (last_check == 0)
e875ecea 7567 /* this is the earliest that rebuild will be
d7603b7e
N		 7568 * visible in /proc/mdstat
7569 */
7570 md_new_event(mddev);
57afd89f
N		 7571
7572 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7573 continue;
7574
57afd89f 7575 last_check = io_sectors;
1da177e4
LT		 7576 repeat:
7577 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7578 /* step marks */
7579 int next = (last_mark+1) % SYNC_MARKS;
7580
7581 mddev->resync_mark = mark[next];
7582 mddev->resync_mark_cnt = mark_cnt[next];
7583 mark[next] = jiffies;
57afd89f 7584 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7585 last_mark = next;
7586 }
7587
7588
c6207277
N		 7589 if (kthread_should_stop())
7590 goto interrupted;
7591
1da177e4
LT		 7592
7593 /*
7594 * this loop exits only if either when we are slower than
7595 * the 'hard' speed limit, or the system was IO-idle for
7596 * a jiffy.
7597 * the system might be non-idle CPU-wise, but we only care
7598 * about not overloading the IO subsystem. (things like an
7599 * e2fsck being done on the RAID array should execute fast)
7600 */
1da177e4
LT		 7601 cond_resched();
7602
57afd89f
N		 7603 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7604 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7605
88202a0c
N		 7606 if (currspeed > speed_min(mddev)) {
7607 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7608 !is_mddev_idle(mddev, 0)) {
c0e48521 7609 msleep(500);
1da177e4
LT		 7610 goto repeat;
7611 }
7612 }
7613 }
61df9d91 7614 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 7615 /*
7616 * this also signals 'finished resyncing' to md_stop
7617 */
7618 out:
7c2c57c9 7619 blk_finish_plug(&plug);
1da177e4
LT		 7620 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7621
7622 /* tell personality that we are finished */
57afd89f 7623 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7624
dfc70645 7625 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7626 mddev->curr_resync > 2) {
7627 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7628 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7629 if (mddev->curr_resync >= mddev->recovery_cp) {
7630 printk(KERN_INFO
61df9d91
N		 7631 "md: checkpointing %s of %s.\n",
7632 desc, mdname(mddev));
0a19caab 7633 if (test_bit(MD_RECOVERY_ERROR,
7634 &mddev->recovery))
7635 mddev->recovery_cp =
7636 mddev->curr_resync_completed;
7637 else
7638 mddev->recovery_cp =
7639 mddev->curr_resync;
5fd6c1dc
N		 7640 }
7641 } else
7642 mddev->recovery_cp = MaxSector;
7643 } else {
7644 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7645 mddev->curr_resync = MaxSector;
4e59ca7d 7646 rcu_read_lock();
dafb20fa 7647 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7648 if (rdev->raid_disk >= 0 &&
70fffd0b 7649 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7650 !test_bit(Faulty, &rdev->flags) &&
7651 !test_bit(In_sync, &rdev->flags) &&
7652 rdev->recovery_offset < mddev->curr_resync)
7653 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7654 rcu_read_unlock();
5fd6c1dc 7655 }
1da177e4 7656 }
db91ff55 7657 skip:
17571284 7658 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7659
c07b70ad
N		 7660 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7661 /* We completed so min/max setting can be forgotten if used. */
7662 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7663 mddev->resync_min = 0;
7664 mddev->resync_max = MaxSector;
7665 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7666 mddev->resync_min = mddev->curr_resync_completed;
1da177e4
LT		 7667 mddev->curr_resync = 0;
7668 wake_up(&resync_wait);
7669 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7670 md_wakeup_thread(mddev->thread);
c6207277
N		 7671 return;
7672
7673 interrupted:
7674 /*
7675 * got a signal, exit.
7676 */
7677 printk(KERN_INFO
7678 "md: md_do_sync() got signal ... exiting\n");
7679 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7680 goto out;
7681
1da177e4 7682}
29269553 7683EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7684
746d3207
N		 7685static int remove_and_add_spares(struct mddev *mddev,
7686 struct md_rdev *this)
b4c4c7b8 7687{
3cb03002 7688 struct md_rdev *rdev;
b4c4c7b8 7689 int spares = 0;
f2a371c5 7690 int removed = 0;
b4c4c7b8 7691
dafb20fa 7692 rdev_for_each(rdev, mddev)
746d3207
N		 7693 if ((this == NULL || rdev == this) &&
7694 rdev->raid_disk >= 0 &&
6bfe0b49 7695 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7696 (test_bit(Faulty, &rdev->flags) ||
7697 ! test_bit(In_sync, &rdev->flags)) &&
7698 atomic_read(&rdev->nr_pending)==0) {
7699 if (mddev->pers->hot_remove_disk(
b8321b68 7700 mddev, rdev) == 0) {
36fad858 7701 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 7702 rdev->raid_disk = -1;
f2a371c5 7703 removed++;
b4c4c7b8
N		 7704 }
7705 }
90584fc9
JB		 7706 if (removed && mddev->kobj.sd)
7707 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 7708
746d3207
N		 7709 if (this)
7710 goto no_add;
7711
dafb20fa 7712 rdev_for_each(rdev, mddev) {
7bfec5f3
N		 7713 if (rdev->raid_disk >= 0 &&
7714 !test_bit(In_sync, &rdev->flags) &&
7715 !test_bit(Faulty, &rdev->flags))
7716 spares++;
7ceb17e8
N		 7717 if (rdev->raid_disk >= 0)
7718 continue;
7719 if (test_bit(Faulty, &rdev->flags))
7720 continue;
7721 if (mddev->ro &&
bb4a65df
N		 7722 ! (rdev->saved_raid_disk >= 0 &&
7723 !test_bit(Bitmap_sync, &rdev->flags)))
7ceb17e8
N		 7724 continue;
7725
7726 rdev->recovery_offset = 0;
7ceb17e8
N		 7727 if (mddev->pers->
7728 hot_add_disk(mddev, rdev) == 0) {
7729 if (sysfs_link_rdev(mddev, rdev))
7730 /* failure here is OK */;
7731 spares++;
7732 md_new_event(mddev);
7733 set_bit(MD_CHANGE_DEVS, &mddev->flags);
dfc70645 7734 }
b4c4c7b8 7735 }
746d3207 7736no_add:
6dafab6b
N		 7737 if (removed)
7738 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 7739 return spares;
7740}
7ebc0be7 7741
1da177e4
LT		 7742/*
7743 * This routine is regularly called by all per-raid-array threads to
7744 * deal with generic issues like resync and super-block update.
7745 * Raid personalities that don't have a thread (linear/raid0) do not
7746 * need this as they never do any recovery or update the superblock.
7747 *
7748 * It does not do any resync itself, but rather "forks" off other threads
7749 * to do that as needed.
7750 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7751 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7752 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7753 * and wakeups up this thread which will reap the thread and finish up.
7754 * This thread also removes any faulty devices (with nr_pending == 0).
7755 *
7756 * The overall approach is:
7757 * 1/ if the superblock needs updating, update it.
7758 * 2/ If a recovery thread is running, don't do anything else.
7759 * 3/ If recovery has finished, clean up, possibly marking spares active.
7760 * 4/ If there are any faulty devices, remove them.
7761 * 5/ If array is degraded, try to add spares devices
7762 * 6/ If array has spares or is not in-sync, start a resync thread.
7763 */
fd01b88c 7764void md_check_recovery(struct mddev *mddev)
1da177e4 7765{
68866e42
JB		 7766 if (mddev->suspended)
7767 return;
7768
5f40402d 7769 if (mddev->bitmap)
aa5cbd10 7770 bitmap_daemon_work(mddev);
1da177e4 7771
fca4d848 7772 if (signal_pending(current)) {
31a59e34 7773 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7774 printk(KERN_INFO "md: %s in immediate safe mode\n",
7775 mdname(mddev));
7776 mddev->safemode = 2;
7777 }
7778 flush_signals(current);
7779 }
7780
c89a8eee
N		 7781 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7782 return;
1da177e4 7783 if ( ! (
126925c0 7784 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7785 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7786 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7787 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7788 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7789 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7790 ))
7791 return;
fca4d848 7792
df5b89b3 7793 if (mddev_trylock(mddev)) {
b4c4c7b8 7794 int spares = 0;
fca4d848 7795
c89a8eee 7796 if (mddev->ro) {
7ceb17e8
N		 7797 /* On a read-only array we can:
7798 * - remove failed devices
7799 * - add already-in_sync devices if the array itself
7800 * is in-sync.
7801 * As we only add devices that are already in-sync,
7802 * we can activate the spares immediately.
c89a8eee 7803 */
7ceb17e8 7804 remove_and_add_spares(mddev, NULL);
bb4a65df
N		 7805 /* There is no thread, but we need to call
7806 * ->spare_active and clear saved_raid_disk
7807 */
88580b12 7808 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
bb4a65df
N		 7809 md_reap_sync_thread(mddev);
7810 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
c89a8eee
N		 7811 goto unlock;
7812 }
7813
31a59e34 7814 if (!mddev->external) {
0fd62b86 7815 int did_change = 0;
31a59e34
N		 7816 spin_lock_irq(&mddev->write_lock);
7817 if (mddev->safemode &&
7818 !atomic_read(&mddev->writes_pending) &&
7819 !mddev->in_sync &&
7820 mddev->recovery_cp == MaxSector) {
7821 mddev->in_sync = 1;
0fd62b86 7822 did_change = 1;
070dc6dd 7823 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7824 }
7825 if (mddev->safemode == 1)
7826 mddev->safemode = 0;
7827 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7828 if (did_change)
00bcb4ac 7829 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7830 }
fca4d848 7831
850b2b42
N		 7832 if (mddev->flags)
7833 md_update_sb(mddev, 0);
06d91a5f 7834
1da177e4
LT		 7835 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7836 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7837 /* resync/recovery still happening */
7838 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7839 goto unlock;
7840 }
7841 if (mddev->sync_thread) {
a91d5ac0 7842 md_reap_sync_thread(mddev);
1da177e4
LT		 7843 goto unlock;
7844 }
72a23c21
NB		 7845 /* Set RUNNING before clearing NEEDED to avoid
7846 * any transients in the value of "sync_action".
7847 */
72f36d59 7848 mddev->curr_resync_completed = 0;
72a23c21 7849 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
24dd469d
N		 7850 /* Clear some bits that don't mean anything, but
7851 * might be left set
7852 */
24dd469d
N		 7853 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7854 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7855
ed209584
N		 7856 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7857 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5fd6c1dc 7858 goto unlock;
1da177e4
LT		 7859 /* no recovery is running.
7860 * remove any failed drives, then
7861 * add spares if possible.
72f36d59 7862 * Spares are also removed and re-added, to allow
1da177e4
LT		 7863 * the personality to fail the re-add.
7864 */
1da177e4 7865
b4c4c7b8 7866 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7867 if (mddev->pers->check_reshape == NULL ||
7868 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7869 /* Cannot proceed */
7870 goto unlock;
7871 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7872 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 7873 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N		 7874 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7875 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7876 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7877 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7878 } else if (mddev->recovery_cp < MaxSector) {
7879 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7880 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7881 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7882 /* nothing to be done ... */
1da177e4 7883 goto unlock;
24dd469d 7884
1da177e4 7885 if (mddev->pers->sync_request) {
ef99bf48 7886 if (spares) {
a654b9d8
N		 7887 /* We are adding a device or devices to an array
7888 * which has the bitmap stored on all devices.
7889 * So make sure all bitmap pages get written
7890 */
7891 bitmap_write_all(mddev->bitmap);
7892 }
1da177e4
LT		 7893 mddev->sync_thread = md_register_thread(md_do_sync,
7894 mddev,
0da3c619 7895 "resync");
1da177e4
LT		 7896 if (!mddev->sync_thread) {
7897 printk(KERN_ERR "%s: could not start resync"
7898 " thread...\n",
7899 mdname(mddev));
7900 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7901 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7902 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7903 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7904 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7905 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7906 } else
1da177e4 7907 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7908 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7909 md_new_event(mddev);
1da177e4
LT		 7910 }
7911 unlock:
72a23c21
NB		 7912 if (!mddev->sync_thread) {
7913 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7914 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7915 &mddev->recovery))
0c3573f1 7916 if (mddev->sysfs_action)
00bcb4ac 7917 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7918 }
1da177e4
LT		 7919 mddev_unlock(mddev);
7920 }
7921}
7922
a91d5ac0
JB		 7923void md_reap_sync_thread(struct mddev *mddev)
7924{
7925 struct md_rdev *rdev;
7926
7927 /* resync has finished, collect result */
7928 md_unregister_thread(&mddev->sync_thread);
7929 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7930 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7931 /* success...*/
7932 /* activate any spares */
7933 if (mddev->pers->spare_active(mddev)) {
7934 sysfs_notify(&mddev->kobj, NULL,
7935 "degraded");
7936 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7937 }
7938 }
7939 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7940 mddev->pers->finish_reshape)
7941 mddev->pers->finish_reshape(mddev);
7942
7943 /* If array is no-longer degraded, then any saved_raid_disk
7944 * information must be scrapped. Also if any device is now
7945 * In_sync we must scrape the saved_raid_disk for that device
7946 * do the superblock for an incrementally recovered device
7947 * written out.
7948 */
7949 rdev_for_each(rdev, mddev)
7950 if (!mddev->degraded ||
7951 test_bit(In_sync, &rdev->flags))
7952 rdev->saved_raid_disk = -1;
7953
7954 md_update_sb(mddev, 1);
7955 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7956 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7957 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7958 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7959 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7960 /* flag recovery needed just to double check */
7961 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7962 sysfs_notify_dirent_safe(mddev->sysfs_action);
7963 md_new_event(mddev);
7964 if (mddev->event_work.func)
7965 queue_work(md_misc_wq, &mddev->event_work);
7966}
7967
fd01b88c 7968void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 7969{
00bcb4ac 7970 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 7971 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 7972 !test_bit(Blocked, &rdev->flags) &&
7973 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 7974 msecs_to_jiffies(5000));
7975 rdev_dec_pending(rdev, mddev);
7976}
7977EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7978
c6563a8c
N		 7979void md_finish_reshape(struct mddev *mddev)
7980{
7981 /* called be personality module when reshape completes. */
7982 struct md_rdev *rdev;
7983
7984 rdev_for_each(rdev, mddev) {
7985 if (rdev->data_offset > rdev->new_data_offset)
7986 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7987 else
7988 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7989 rdev->data_offset = rdev->new_data_offset;
7990 }
7991}
7992EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N		 7993
7994/* Bad block management.
7995 * We can record which blocks on each device are 'bad' and so just
7996 * fail those blocks, or that stripe, rather than the whole device.
7997 * Entries in the bad-block table are 64bits wide. This comprises:
7998 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7999 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8000 * A 'shift' can be set so that larger blocks are tracked and
8001 * consequently larger devices can be covered.
8002 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8003 *
8004 * Locking of the bad-block table uses a seqlock so md_is_badblock
8005 * might need to retry if it is very unlucky.
8006 * We will sometimes want to check for bad blocks in a bi_end_io function,
8007 * so we use the write_seqlock_irq variant.
8008 *
8009 * When looking for a bad block we specify a range and want to
8010 * know if any block in the range is bad. So we binary-search
8011 * to the last range that starts at-or-before the given endpoint,
8012 * (or "before the sector after the target range")
8013 * then see if it ends after the given start.
8014 * We return
8015 * 0 if there are no known bad blocks in the range
8016 * 1 if there are known bad block which are all acknowledged
8017 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8018 * plus the start/length of the first bad section we overlap.
8019 */
8020int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8021 sector_t *first_bad, int *bad_sectors)
8022{
8023 int hi;
ab05613a 8024 int lo;
2230dfe4 8025 u64 *p = bb->page;
ab05613a 8026 int rv;
2230dfe4
N		 8027 sector_t target = s + sectors;
8028 unsigned seq;
8029
8030 if (bb->shift > 0) {
8031 /* round the start down, and the end up */
8032 s >>= bb->shift;
8033 target += (1<<bb->shift) - 1;
8034 target >>= bb->shift;
8035 sectors = target - s;
8036 }
8037 /* 'target' is now the first block after the bad range */
8038
8039retry:
8040 seq = read_seqbegin(&bb->lock);
ab05613a 8041 lo = 0;
8042 rv = 0;
2230dfe4
N		 8043 hi = bb->count;
8044
8045 /* Binary search between lo and hi for 'target'
8046 * i.e. for the last range that starts before 'target'
8047 */
8048 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8049 * are known not to be the last range before target.
8050 * VARIANT: hi-lo is the number of possible
8051 * ranges, and decreases until it reaches 1
8052 */
8053 while (hi - lo > 1) {
8054 int mid = (lo + hi) / 2;
8055 sector_t a = BB_OFFSET(p[mid]);
8056 if (a < target)
8057 /* This could still be the one, earlier ranges
8058 * could not. */
8059 lo = mid;
8060 else
8061 /* This and later ranges are definitely out. */
8062 hi = mid;
8063 }
8064 /* 'lo' might be the last that started before target, but 'hi' isn't */
8065 if (hi > lo) {
8066 /* need to check all range that end after 's' to see if
8067 * any are unacknowledged.
8068 */
8069 while (lo >= 0 &&
8070 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8071 if (BB_OFFSET(p[lo]) < target) {
8072 /* starts before the end, and finishes after
8073 * the start, so they must overlap
8074 */
8075 if (rv != -1 && BB_ACK(p[lo]))
8076 rv = 1;
8077 else
8078 rv = -1;
8079 *first_bad = BB_OFFSET(p[lo]);
8080 *bad_sectors = BB_LEN(p[lo]);
8081 }
8082 lo--;
8083 }
8084 }
8085
8086 if (read_seqretry(&bb->lock, seq))
8087 goto retry;
8088
8089 return rv;
8090}
8091EXPORT_SYMBOL_GPL(md_is_badblock);
8092
8093/*
8094 * Add a range of bad blocks to the table.
8095 * This might extend the table, or might contract it
8096 * if two adjacent ranges can be merged.
8097 * We binary-search to find the 'insertion' point, then
8098 * decide how best to handle it.
8099 */
8100static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8101 int acknowledged)
8102{
8103 u64 *p;
8104 int lo, hi;
8105 int rv = 1;
04654966 8106 unsigned long flags;
2230dfe4
N		 8107
8108 if (bb->shift < 0)
8109 /* badblocks are disabled */
8110 return 0;
8111
8112 if (bb->shift) {
8113 /* round the start down, and the end up */
8114 sector_t next = s + sectors;
8115 s >>= bb->shift;
8116 next += (1<<bb->shift) - 1;
8117 next >>= bb->shift;
8118 sectors = next - s;
8119 }
8120
04654966 8121 write_seqlock_irqsave(&bb->lock, flags);
2230dfe4
N		 8122
8123 p = bb->page;
8124 lo = 0;
8125 hi = bb->count;
8126 /* Find the last range that starts at-or-before 's' */
8127 while (hi - lo > 1) {
8128 int mid = (lo + hi) / 2;
8129 sector_t a = BB_OFFSET(p[mid]);
8130 if (a <= s)
8131 lo = mid;
8132 else
8133 hi = mid;
8134 }
8135 if (hi > lo && BB_OFFSET(p[lo]) > s)
8136 hi = lo;
8137
8138 if (hi > lo) {
8139 /* we found a range that might merge with the start
8140 * of our new range
8141 */
8142 sector_t a = BB_OFFSET(p[lo]);
8143 sector_t e = a + BB_LEN(p[lo]);
8144 int ack = BB_ACK(p[lo]);
8145 if (e >= s) {
8146 /* Yes, we can merge with a previous range */
8147 if (s == a && s + sectors >= e)
8148 /* new range covers old */
8149 ack = acknowledged;
8150 else
8151 ack = ack && acknowledged;
8152
8153 if (e < s + sectors)
8154 e = s + sectors;
8155 if (e - a <= BB_MAX_LEN) {
8156 p[lo] = BB_MAKE(a, e-a, ack);
8157 s = e;
8158 } else {
8159 /* does not all fit in one range,
8160 * make p[lo] maximal
8161 */
8162 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8163 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8164 s = a + BB_MAX_LEN;
8165 }
8166 sectors = e - s;
8167 }
8168 }
8169 if (sectors && hi < bb->count) {
8170 /* 'hi' points to the first range that starts after 's'.
8171 * Maybe we can merge with the start of that range */
8172 sector_t a = BB_OFFSET(p[hi]);
8173 sector_t e = a + BB_LEN(p[hi]);
8174 int ack = BB_ACK(p[hi]);
8175 if (a <= s + sectors) {
8176 /* merging is possible */
8177 if (e <= s + sectors) {
8178 /* full overlap */
8179 e = s + sectors;
8180 ack = acknowledged;
8181 } else
8182 ack = ack && acknowledged;
8183
8184 a = s;
8185 if (e - a <= BB_MAX_LEN) {
8186 p[hi] = BB_MAKE(a, e-a, ack);
8187 s = e;
8188 } else {
8189 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8190 s = a + BB_MAX_LEN;
8191 }
8192 sectors = e - s;
8193 lo = hi;
8194 hi++;
8195 }
8196 }
8197 if (sectors == 0 && hi < bb->count) {
8198 /* we might be able to combine lo and hi */
8199 /* Note: 's' is at the end of 'lo' */
8200 sector_t a = BB_OFFSET(p[hi]);
8201 int lolen = BB_LEN(p[lo]);
8202 int hilen = BB_LEN(p[hi]);
8203 int newlen = lolen + hilen - (s - a);
8204 if (s >= a && newlen < BB_MAX_LEN) {
8205 /* yes, we can combine them */
8206 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8207 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8208 memmove(p + hi, p + hi + 1,
8209 (bb->count - hi - 1) * 8);
8210 bb->count--;
8211 }
8212 }
8213 while (sectors) {
8214 /* didn't merge (it all).
8215 * Need to add a range just before 'hi' */
8216 if (bb->count >= MD_MAX_BADBLOCKS) {
8217 /* No room for more */
8218 rv = 0;
8219 break;
8220 } else {
8221 int this_sectors = sectors;
8222 memmove(p + hi + 1, p + hi,
8223 (bb->count - hi) * 8);
8224 bb->count++;
8225
8226 if (this_sectors > BB_MAX_LEN)
8227 this_sectors = BB_MAX_LEN;
8228 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8229 sectors -= this_sectors;
8230 s += this_sectors;
8231 }
8232 }
8233
8234 bb->changed = 1;
de393cde
N		 8235 if (!acknowledged)
8236 bb->unacked_exist = 1;
04654966 8237 write_sequnlock_irqrestore(&bb->lock, flags);
2230dfe4
N		 8238
8239 return rv;
8240}
8241
3cb03002 8242int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8243 int is_new)
2230dfe4 8244{
c6563a8c
N		 8245 int rv;
8246 if (is_new)
8247 s += rdev->new_data_offset;
8248 else
8249 s += rdev->data_offset;
8250 rv = md_set_badblocks(&rdev->badblocks,
8251 s, sectors, 0);
2230dfe4
N		 8252 if (rv) {
8253 /* Make sure they get written out promptly */
8bd2f0a0 8254 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 8255 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8256 md_wakeup_thread(rdev->mddev->thread);
8257 }
8258 return rv;
8259}
8260EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8261
8262/*
8263 * Remove a range of bad blocks from the table.
8264 * This may involve extending the table if we spilt a region,
8265 * but it must not fail. So if the table becomes full, we just
8266 * drop the remove request.
8267 */
8268static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8269{
8270 u64 *p;
8271 int lo, hi;
8272 sector_t target = s + sectors;
8273 int rv = 0;
8274
8275 if (bb->shift > 0) {
8276 /* When clearing we round the start up and the end down.
8277 * This should not matter as the shift should align with
8278 * the block size and no rounding should ever be needed.
8279 * However it is better the think a block is bad when it
8280 * isn't than to think a block is not bad when it is.
8281 */
8282 s += (1<<bb->shift) - 1;
8283 s >>= bb->shift;
8284 target >>= bb->shift;
8285 sectors = target - s;
8286 }
8287
8288 write_seqlock_irq(&bb->lock);
8289
8290 p = bb->page;
8291 lo = 0;
8292 hi = bb->count;
8293 /* Find the last range that starts before 'target' */
8294 while (hi - lo > 1) {
8295 int mid = (lo + hi) / 2;
8296 sector_t a = BB_OFFSET(p[mid]);
8297 if (a < target)
8298 lo = mid;
8299 else
8300 hi = mid;
8301 }
8302 if (hi > lo) {
8303 /* p[lo] is the last range that could overlap the
8304 * current range. Earlier ranges could also overlap,
8305 * but only this one can overlap the end of the range.
8306 */
8307 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8308 /* Partial overlap, leave the tail of this range */
8309 int ack = BB_ACK(p[lo]);
8310 sector_t a = BB_OFFSET(p[lo]);
8311 sector_t end = a + BB_LEN(p[lo]);
8312
8313 if (a < s) {
8314 /* we need to split this range */
8315 if (bb->count >= MD_MAX_BADBLOCKS) {
8316 rv = 0;
8317 goto out;
8318 }
8319 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8320 bb->count++;
8321 p[lo] = BB_MAKE(a, s-a, ack);
8322 lo++;
8323 }
8324 p[lo] = BB_MAKE(target, end - target, ack);
8325 /* there is no longer an overlap */
8326 hi = lo;
8327 lo--;
8328 }
8329 while (lo >= 0 &&
8330 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8331 /* This range does overlap */
8332 if (BB_OFFSET(p[lo]) < s) {
8333 /* Keep the early parts of this range. */
8334 int ack = BB_ACK(p[lo]);
8335 sector_t start = BB_OFFSET(p[lo]);
8336 p[lo] = BB_MAKE(start, s - start, ack);
8337 /* now low doesn't overlap, so.. */
8338 break;
8339 }
8340 lo--;
8341 }
8342 /* 'lo' is strictly before, 'hi' is strictly after,
8343 * anything between needs to be discarded
8344 */
8345 if (hi - lo > 1) {
8346 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8347 bb->count -= (hi - lo - 1);
8348 }
8349 }
8350
8351 bb->changed = 1;
8352out:
8353 write_sequnlock_irq(&bb->lock);
8354 return rv;
8355}
8356
c6563a8c
N		 8357int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8358 int is_new)
2230dfe4 8359{
c6563a8c
N		 8360 if (is_new)
8361 s += rdev->new_data_offset;
8362 else
8363 s += rdev->data_offset;
2230dfe4 8364 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8365 s, sectors);
2230dfe4
N		 8366}
8367EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8368
8369/*
8370 * Acknowledge all bad blocks in a list.
8371 * This only succeeds if ->changed is clear. It is used by
8372 * in-kernel metadata updates
8373 */
8374void md_ack_all_badblocks(struct badblocks *bb)
8375{
8376 if (bb->page == NULL || bb->changed)
8377 /* no point even trying */
8378 return;
8379 write_seqlock_irq(&bb->lock);
8380
ecb178bb 8381 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N		 8382 u64 *p = bb->page;
8383 int i;
8384 for (i = 0; i < bb->count ; i++) {
8385 if (!BB_ACK(p[i])) {
8386 sector_t start = BB_OFFSET(p[i]);
8387 int len = BB_LEN(p[i]);
8388 p[i] = BB_MAKE(start, len, 1);
8389 }
8390 }
de393cde 8391 bb->unacked_exist = 0;
2230dfe4
N		 8392 }
8393 write_sequnlock_irq(&bb->lock);
8394}
8395EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8396
16c791a5
N		 8397/* sysfs access to bad-blocks list.
8398 * We present two files.
8399 * 'bad-blocks' lists sector numbers and lengths of ranges that
8400 * are recorded as bad. The list is truncated to fit within
8401 * the one-page limit of sysfs.
8402 * Writing "sector length" to this file adds an acknowledged
8403 * bad block list.
8404 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8405 * been acknowledged. Writing to this file adds bad blocks
8406 * without acknowledging them. This is largely for testing.
8407 */
8408
8409static ssize_t
8410badblocks_show(struct badblocks *bb, char *page, int unack)
8411{
8412 size_t len;
8413 int i;
8414 u64 *p = bb->page;
8415 unsigned seq;
8416
8417 if (bb->shift < 0)
8418 return 0;
8419
8420retry:
8421 seq = read_seqbegin(&bb->lock);
8422
8423 len = 0;
8424 i = 0;
8425
8426 while (len < PAGE_SIZE && i < bb->count) {
8427 sector_t s = BB_OFFSET(p[i]);
8428 unsigned int length = BB_LEN(p[i]);
8429 int ack = BB_ACK(p[i]);
8430 i++;
8431
8432 if (unack && ack)
8433 continue;
8434
8435 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8436 (unsigned long long)s << bb->shift,
8437 length << bb->shift);
8438 }
de393cde
N		 8439 if (unack && len == 0)
8440 bb->unacked_exist = 0;
16c791a5
N		 8441
8442 if (read_seqretry(&bb->lock, seq))
8443 goto retry;
8444
8445 return len;
8446}
8447
8448#define DO_DEBUG 1
8449
8450static ssize_t
8451badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8452{
8453 unsigned long long sector;
8454 int length;
8455 char newline;
8456#ifdef DO_DEBUG
8457 /* Allow clearing via sysfs *only* for testing/debugging.
8458 * Normally only a successful write may clear a badblock
8459 */
8460 int clear = 0;
8461 if (page[0] == '-') {
8462 clear = 1;
8463 page++;
8464 }
8465#endif /* DO_DEBUG */
8466
8467 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8468 case 3:
8469 if (newline != '\n')
8470 return -EINVAL;
8471 case 2:
8472 if (length <= 0)
8473 return -EINVAL;
8474 break;
8475 default:
8476 return -EINVAL;
8477 }
8478
8479#ifdef DO_DEBUG
8480 if (clear) {
8481 md_clear_badblocks(bb, sector, length);
8482 return len;
8483 }
8484#endif /* DO_DEBUG */
8485 if (md_set_badblocks(bb, sector, length, !unack))
8486 return len;
8487 else
8488 return -ENOSPC;
8489}
8490
75c96f85
AB		 8491static int md_notify_reboot(struct notifier_block *this,
8492 unsigned long code, void *x)
1da177e4
LT		 8493{
8494 struct list_head *tmp;
fd01b88c 8495 struct mddev *mddev;
2dba6a91 8496 int need_delay = 0;
1da177e4 8497
c744a65c
N		 8498 for_each_mddev(mddev, tmp) {
8499 if (mddev_trylock(mddev)) {
30b8aa91
N		 8500 if (mddev->pers)
8501 __md_stop_writes(mddev);
658864cc
N		 8502 if (mddev->persistent)
8503 mddev->safemode = 2;
c744a65c 8504 mddev_unlock(mddev);
2dba6a91 8505 }
c744a65c 8506 need_delay = 1;
1da177e4 8507 }
c744a65c
N		 8508 /*
8509 * certain more exotic SCSI devices are known to be
8510 * volatile wrt too early system reboots. While the
8511 * right place to handle this issue is the given
8512 * driver, we do want to have a safe RAID driver ...
8513 */
8514 if (need_delay)
8515 mdelay(1000*1);
8516
1da177e4
LT		 8517 return NOTIFY_DONE;
8518}
8519
75c96f85 8520static struct notifier_block md_notifier = {
1da177e4
LT		 8521 .notifier_call = md_notify_reboot,
8522 .next = NULL,
8523 .priority = INT_MAX, /* before any real devices */
8524};
8525
8526static void md_geninit(void)
8527{
36a4e1fe 8528 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8529
c7705f34 8530 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8531}
8532
75c96f85 8533static int __init md_init(void)
1da177e4 8534{
e804ac78
TH		 8535 int ret = -ENOMEM;
8536
ada609ee 8537 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8538 if (!md_wq)
8539 goto err_wq;
8540
8541 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8542 if (!md_misc_wq)
8543 goto err_misc_wq;
8544
8545 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8546 goto err_md;
8547
8548 if ((ret = register_blkdev(0, "mdp")) < 0)
8549 goto err_mdp;
8550 mdp_major = ret;
8551
3dbd8c2e 8552 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 8553 md_probe, NULL, NULL);
8554 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8555 md_probe, NULL, NULL);
8556
1da177e4 8557 register_reboot_notifier(&md_notifier);
0b4d4147 8558 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8559
8560 md_geninit();
d710e138 8561 return 0;
1da177e4 8562
e804ac78
TH		 8563err_mdp:
8564 unregister_blkdev(MD_MAJOR, "md");
8565err_md:
8566 destroy_workqueue(md_misc_wq);
8567err_misc_wq:
8568 destroy_workqueue(md_wq);
8569err_wq:
8570 return ret;
8571}
1da177e4
LT		 8572
8573#ifndef MODULE
8574
8575/*
8576 * Searches all registered partitions for autorun RAID arrays
8577 * at boot time.
8578 */
4d936ec1
ME		 8579
8580static LIST_HEAD(all_detected_devices);
8581struct detected_devices_node {
8582 struct list_head list;
8583 dev_t dev;
8584};
1da177e4
LT		 8585
8586void md_autodetect_dev(dev_t dev)
8587{
4d936ec1
ME		 8588 struct detected_devices_node *node_detected_dev;
8589
8590 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8591 if (node_detected_dev) {
8592 node_detected_dev->dev = dev;
8593 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8594 } else {
8595 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8596 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8597 }
1da177e4
LT		 8598}
8599
8600
8601static void autostart_arrays(int part)
8602{
3cb03002 8603 struct md_rdev *rdev;
4d936ec1
ME		 8604 struct detected_devices_node *node_detected_dev;
8605 dev_t dev;
8606 int i_scanned, i_passed;
1da177e4 8607
4d936ec1
ME		 8608 i_scanned = 0;
8609 i_passed = 0;
1da177e4 8610
4d936ec1 8611 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8612
4d936ec1
ME		 8613 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8614 i_scanned++;
8615 node_detected_dev = list_entry(all_detected_devices.next,
8616 struct detected_devices_node, list);
8617 list_del(&node_detected_dev->list);
8618 dev = node_detected_dev->dev;
8619 kfree(node_detected_dev);
df968c4e 8620 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8621 if (IS_ERR(rdev))
8622 continue;
8623
b2d444d7 8624 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 8625 MD_BUG();
8626 continue;
8627 }
d0fae18f 8628 set_bit(AutoDetected, &rdev->flags);
1da177e4 8629 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8630 i_passed++;
1da177e4 8631 }
4d936ec1
ME		 8632
8633 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8634 i_scanned, i_passed);
1da177e4
LT		 8635
8636 autorun_devices(part);
8637}
8638
fdee8ae4 8639#endif /* !MODULE */
1da177e4
LT		 8640
8641static __exit void md_exit(void)
8642{
fd01b88c 8643 struct mddev *mddev;
1da177e4 8644 struct list_head *tmp;
8ab5e4c1 8645
3dbd8c2e 8646 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 8647 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8648
3dbd8c2e 8649 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8650 unregister_blkdev(mdp_major, "mdp");
8651 unregister_reboot_notifier(&md_notifier);
8652 unregister_sysctl_table(raid_table_header);
8653 remove_proc_entry("mdstat", NULL);
29ac4aa3 8654 for_each_mddev(mddev, tmp) {
1da177e4 8655 export_array(mddev);
d3374825 8656 mddev->hold_active = 0;
1da177e4 8657 }
e804ac78
TH		 8658 destroy_workqueue(md_misc_wq);
8659 destroy_workqueue(md_wq);
1da177e4
LT		 8660}
8661
685784aa 8662subsys_initcall(md_init);
1da177e4
LT		 8663module_exit(md_exit)
8664
f91de92e
N		 8665static int get_ro(char *buffer, struct kernel_param *kp)
8666{
8667 return sprintf(buffer, "%d", start_readonly);
8668}
8669static int set_ro(const char *val, struct kernel_param *kp)
8670{
8671 char *e;
8672 int num = simple_strtoul(val, &e, 10);
8673 if (*val && (*e == '\0' || *e == '\n')) {
8674 start_readonly = num;
4dbcdc75 8675 return 0;
f91de92e
N		 8676 }
8677 return -EINVAL;
8678}
8679
80ca3a44
N		 8680module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8681module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 8682
efeb53c0 8683module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8684
1da177e4
LT		 8685EXPORT_SYMBOL(register_md_personality);
8686EXPORT_SYMBOL(unregister_md_personality);
8687EXPORT_SYMBOL(md_error);
8688EXPORT_SYMBOL(md_done_sync);
8689EXPORT_SYMBOL(md_write_start);
8690EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 8691EXPORT_SYMBOL(md_register_thread);
8692EXPORT_SYMBOL(md_unregister_thread);
8693EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 8694EXPORT_SYMBOL(md_check_recovery);
a91d5ac0 8695EXPORT_SYMBOL(md_reap_sync_thread);
1da177e4 8696MODULE_LICENSE("GPL");
0efb9e61 8697MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8698MODULE_ALIAS("md");
72008652 8699MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)