projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
sysfs: drop kobj_ns_type handling
[linux-2.6-block.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;
c4a39551 524 mddev->last_sync_action = "none";
fafd7fb0
N		 525 mddev->resync_min = 0;
526 mddev->resync_max = MaxSector;
527 mddev->level = LEVEL_NONE;
528}
390ee602 529EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 530
fd01b88c 531static struct mddev * mddev_find(dev_t unit)
1da177e4 532{
fd01b88c 533 struct mddev *mddev, *new = NULL;
1da177e4 534
8f5f02c4
N		 535 if (unit && MAJOR(unit) != MD_MAJOR)
536 unit &= ~((1<<MdpMinorShift)-1);
537
1da177e4
LT		 538 retry:
539 spin_lock(&all_mddevs_lock);
efeb53c0
N		 540
541 if (unit) {
542 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
543 if (mddev->unit == unit) {
544 mddev_get(mddev);
545 spin_unlock(&all_mddevs_lock);
546 kfree(new);
547 return mddev;
548 }
549
550 if (new) {
551 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 552 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 553 new->hold_active = UNTIL_IOCTL;
554 return new;
1da177e4 555 }
efeb53c0
N		 556 } else if (new) {
557 /* find an unused unit number */
558 static int next_minor = 512;
559 int start = next_minor;
560 int is_free = 0;
561 int dev = 0;
562 while (!is_free) {
563 dev = MKDEV(MD_MAJOR, next_minor);
564 next_minor++;
565 if (next_minor > MINORMASK)
566 next_minor = 0;
567 if (next_minor == start) {
568 /* Oh dear, all in use. */
569 spin_unlock(&all_mddevs_lock);
570 kfree(new);
571 return NULL;
572 }
573
574 is_free = 1;
575 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
576 if (mddev->unit == dev) {
577 is_free = 0;
578 break;
579 }
580 }
581 new->unit = dev;
582 new->md_minor = MINOR(dev);
583 new->hold_active = UNTIL_STOP;
1da177e4
LT		 584 list_add(&new->all_mddevs, &all_mddevs);
585 spin_unlock(&all_mddevs_lock);
586 return new;
587 }
588 spin_unlock(&all_mddevs_lock);
589
9ffae0cf 590 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 591 if (!new)
592 return NULL;
593
1da177e4
LT		 594 new->unit = unit;
595 if (MAJOR(unit) == MD_MAJOR)
596 new->md_minor = MINOR(unit);
597 else
598 new->md_minor = MINOR(unit) >> MdpMinorShift;
599
fafd7fb0 600 mddev_init(new);
1da177e4 601
1da177e4
LT		 602 goto retry;
603}
604
fd01b88c 605static inline int mddev_lock(struct mddev * mddev)
1da177e4 606{
df5b89b3 607 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 608}
609
fd01b88c 610static inline int mddev_is_locked(struct mddev *mddev)
b522adcd
DW		 611{
612 return mutex_is_locked(&mddev->reconfig_mutex);
613}
614
fd01b88c 615static inline int mddev_trylock(struct mddev * mddev)
1da177e4 616{
df5b89b3 617 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 618}
619
b6eb127d
N		 620static struct attribute_group md_redundancy_group;
621
fd01b88c 622static void mddev_unlock(struct mddev * mddev)
1da177e4 623{
a64c876f 624 if (mddev->to_remove) {
b6eb127d
N		 625 /* These cannot be removed under reconfig_mutex as
626 * an access to the files will try to take reconfig_mutex
627 * while holding the file unremovable, which leads to
628 * a deadlock.
bb4f1e9d
N		 629 * So hold set sysfs_active while the remove in happeing,
630 * and anything else which might set ->to_remove or my
631 * otherwise change the sysfs namespace will fail with
632 * -EBUSY if sysfs_active is still set.
633 * We set sysfs_active under reconfig_mutex and elsewhere
634 * test it under the same mutex to ensure its correct value
635 * is seen.
b6eb127d 636 */
a64c876f
N		 637 struct attribute_group *to_remove = mddev->to_remove;
638 mddev->to_remove = NULL;
bb4f1e9d 639 mddev->sysfs_active = 1;
b6eb127d
N		 640 mutex_unlock(&mddev->reconfig_mutex);
641
00bcb4ac
N		 642 if (mddev->kobj.sd) {
643 if (to_remove != &md_redundancy_group)
644 sysfs_remove_group(&mddev->kobj, to_remove);
645 if (mddev->pers == NULL ||
646 mddev->pers->sync_request == NULL) {
647 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
648 if (mddev->sysfs_action)
649 sysfs_put(mddev->sysfs_action);
650 mddev->sysfs_action = NULL;
651 }
a64c876f 652 }
bb4f1e9d 653 mddev->sysfs_active = 0;
b6eb127d
N		 654 } else
655 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 656
751e67ca
CD		 657 /* As we've dropped the mutex we need a spinlock to
658 * make sure the thread doesn't disappear
01f96c0a
N		 659 */
660 spin_lock(&pers_lock);
005eca5e 661 md_wakeup_thread(mddev->thread);
01f96c0a 662 spin_unlock(&pers_lock);
1da177e4
LT		 663}
664
fd01b88c 665static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
1da177e4 666{
3cb03002 667 struct md_rdev *rdev;
1da177e4 668
dafb20fa 669 rdev_for_each(rdev, mddev)
1da177e4
LT		 670 if (rdev->desc_nr == nr)
671 return rdev;
159ec1fc 672
1da177e4
LT		 673 return NULL;
674}
675
1ca69c4b
N		 676static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
677{
678 struct md_rdev *rdev;
679
680 rdev_for_each_rcu(rdev, mddev)
681 if (rdev->desc_nr == nr)
682 return rdev;
683
684 return NULL;
685}
686
687static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
1da177e4 688{
3cb03002 689 struct md_rdev *rdev;
1da177e4 690
dafb20fa 691 rdev_for_each(rdev, mddev)
1da177e4
LT		 692 if (rdev->bdev->bd_dev == dev)
693 return rdev;
159ec1fc 694
1da177e4
LT		 695 return NULL;
696}
697
1ca69c4b
N		 698static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
699{
700 struct md_rdev *rdev;
701
702 rdev_for_each_rcu(rdev, mddev)
703 if (rdev->bdev->bd_dev == dev)
704 return rdev;
705
706 return NULL;
707}
708
84fc4b56 709static struct md_personality *find_pers(int level, char *clevel)
2604b703 710{
84fc4b56 711 struct md_personality *pers;
d9d166c2
N		 712 list_for_each_entry(pers, &pers_list, list) {
713 if (level != LEVEL_NONE && pers->level == level)
2604b703 714 return pers;
d9d166c2
N		 715 if (strcmp(pers->name, clevel)==0)
716 return pers;
717 }
2604b703
N		 718 return NULL;
719}
720
b73df2d3 721/* return the offset of the super block in 512byte sectors */
3cb03002 722static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 723{
57b2caa3 724 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 725 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 726}
727
3cb03002 728static int alloc_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 729{
730 if (rdev->sb_page)
731 MD_BUG();
732
733 rdev->sb_page = alloc_page(GFP_KERNEL);
734 if (!rdev->sb_page) {
735 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 736 return -ENOMEM;
1da177e4
LT		 737 }
738
739 return 0;
740}
741
545c8795 742void md_rdev_clear(struct md_rdev *rdev)
1da177e4
LT		 743{
744 if (rdev->sb_page) {
2d1f3b5d 745 put_page(rdev->sb_page);
1da177e4
LT		 746 rdev->sb_loaded = 0;
747 rdev->sb_page = NULL;
0f420358 748 rdev->sb_start = 0;
dd8ac336 749 rdev->sectors = 0;
1da177e4 750 }
2699b672
N		 751 if (rdev->bb_page) {
752 put_page(rdev->bb_page);
753 rdev->bb_page = NULL;
754 }
4fa2f327
N		 755 kfree(rdev->badblocks.page);
756 rdev->badblocks.page = NULL;
1da177e4 757}
545c8795 758EXPORT_SYMBOL_GPL(md_rdev_clear);
1da177e4 759
6712ecf8 760static void super_written(struct bio *bio, int error)
7bfa19f2 761{
3cb03002 762 struct md_rdev *rdev = bio->bi_private;
fd01b88c 763 struct mddev *mddev = rdev->mddev;
7bfa19f2 764
3a0f5bbb
N		 765 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
766 printk("md: super_written gets error=%d, uptodate=%d\n",
767 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
768 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 769 md_error(mddev, rdev);
3a0f5bbb 770 }
7bfa19f2 771
a9701a30
N		 772 if (atomic_dec_and_test(&mddev->pending_writes))
773 wake_up(&mddev->sb_wait);
f8b58edf 774 bio_put(bio);
7bfa19f2
N		 775}
776
fd01b88c 777void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 778 sector_t sector, int size, struct page *page)
779{
780 /* write first size bytes of page to sector of rdev
781 * Increment mddev->pending_writes before returning
782 * and decrement it on completion, waking up sb_wait
783 * if zero is reached.
784 * If an error occurred, call md_error
785 */
a167f663 786 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 787
a6ff7e08 788 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
7bfa19f2
N		 789 bio->bi_sector = sector;
790 bio_add_page(bio, page, size, 0);
791 bio->bi_private = rdev;
792 bio->bi_end_io = super_written;
a9701a30 793
7bfa19f2 794 atomic_inc(&mddev->pending_writes);
a5bf4df0 795 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 796}
797
fd01b88c 798void md_super_wait(struct mddev *mddev)
a9701a30 799{
e9c7469b 800 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N		 801 DEFINE_WAIT(wq);
802 for(;;) {
803 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
804 if (atomic_read(&mddev->pending_writes)==0)
805 break;
a9701a30
N		 806 schedule();
807 }
808 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 809}
810
6712ecf8 811static void bi_complete(struct bio *bio, int error)
1da177e4 812{
1da177e4 813 complete((struct completion*)bio->bi_private);
1da177e4
LT		 814}
815
3cb03002 816int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 817 struct page *page, int rw, bool metadata_op)
1da177e4 818{
a167f663 819 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 820 struct completion event;
821 int ret;
822
721a9602 823 rw |= REQ_SYNC;
1da177e4 824
a6ff7e08
JB		 825 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
826 rdev->meta_bdev : rdev->bdev;
ccebd4c4
JB		 827 if (metadata_op)
828 bio->bi_sector = sector + rdev->sb_start;
1fdd6fc9
N		 829 else if (rdev->mddev->reshape_position != MaxSector &&
830 (rdev->mddev->reshape_backwards ==
831 (sector >= rdev->mddev->reshape_position)))
832 bio->bi_sector = sector + rdev->new_data_offset;
ccebd4c4
JB		 833 else
834 bio->bi_sector = sector + rdev->data_offset;
1da177e4
LT		 835 bio_add_page(bio, page, size, 0);
836 init_completion(&event);
837 bio->bi_private = &event;
838 bio->bi_end_io = bi_complete;
839 submit_bio(rw, bio);
840 wait_for_completion(&event);
841
842 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
843 bio_put(bio);
844 return ret;
845}
a8745db2 846EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 847
3cb03002 848static int read_disk_sb(struct md_rdev * rdev, int size)
1da177e4
LT		 849{
850 char b[BDEVNAME_SIZE];
851 if (!rdev->sb_page) {
852 MD_BUG();
853 return -EINVAL;
854 }
855 if (rdev->sb_loaded)
856 return 0;
857
858
ccebd4c4 859 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 860 goto fail;
861 rdev->sb_loaded = 1;
862 return 0;
863
864fail:
865 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
866 bdevname(rdev->bdev,b));
867 return -EINVAL;
868}
869
870static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
871{
05710466
AN		 872 return sb1->set_uuid0 == sb2->set_uuid0 &&
873 sb1->set_uuid1 == sb2->set_uuid1 &&
874 sb1->set_uuid2 == sb2->set_uuid2 &&
875 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 876}
877
1da177e4
LT		 878static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
879{
880 int ret;
881 mdp_super_t *tmp1, *tmp2;
882
883 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
884 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
885
886 if (!tmp1 || !tmp2) {
887 ret = 0;
35020f1a 888 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 889 goto abort;
890 }
891
892 *tmp1 = *sb1;
893 *tmp2 = *sb2;
894
895 /*
896 * nr_disks is not constant
897 */
898 tmp1->nr_disks = 0;
899 tmp2->nr_disks = 0;
900
ce0c8e05 901 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 902abort:
990a8baf
JJ		 903 kfree(tmp1);
904 kfree(tmp2);
1da177e4
LT		 905 return ret;
906}
907
4d167f09
N		 908
909static u32 md_csum_fold(u32 csum)
910{
911 csum = (csum & 0xffff) + (csum >> 16);
912 return (csum & 0xffff) + (csum >> 16);
913}
914
1da177e4
LT		 915static unsigned int calc_sb_csum(mdp_super_t * sb)
916{
4d167f09
N		 917 u64 newcsum = 0;
918 u32 *sb32 = (u32*)sb;
919 int i;
1da177e4
LT		 920 unsigned int disk_csum, csum;
921
922 disk_csum = sb->sb_csum;
923 sb->sb_csum = 0;
4d167f09
N		 924
925 for (i = 0; i < MD_SB_BYTES/4 ; i++)
926 newcsum += sb32[i];
927 csum = (newcsum & 0xffffffff) + (newcsum>>32);
928
929
930#ifdef CONFIG_ALPHA
931 /* This used to use csum_partial, which was wrong for several
932 * reasons including that different results are returned on
933 * different architectures. It isn't critical that we get exactly
934 * the same return value as before (we always csum_fold before
935 * testing, and that removes any differences). However as we
936 * know that csum_partial always returned a 16bit value on
937 * alphas, do a fold to maximise conformity to previous behaviour.
938 */
939 sb->sb_csum = md_csum_fold(disk_csum);
940#else
1da177e4 941 sb->sb_csum = disk_csum;
4d167f09 942#endif
1da177e4
LT		 943 return csum;
944}
945
946
947/*
948 * Handle superblock details.
949 * We want to be able to handle multiple superblock formats
950 * so we have a common interface to them all, and an array of
951 * different handlers.
952 * We rely on user-space to write the initial superblock, and support
953 * reading and updating of superblocks.
954 * Interface methods are:
3cb03002 955 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 956 * loads and validates a superblock on dev.
957 * if refdev != NULL, compare superblocks on both devices
958 * Return:
959 * 0 - dev has a superblock that is compatible with refdev
960 * 1 - dev has a superblock that is compatible and newer than refdev
961 * so dev should be used as the refdev in future
962 * -EINVAL superblock incompatible or invalid
963 * -othererror e.g. -EIO
964 *
fd01b88c 965 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 966 * Verify that dev is acceptable into mddev.
967 * The first time, mddev->raid_disks will be 0, and data from
968 * dev should be merged in. Subsequent calls check that dev
969 * is new enough. Return 0 or -EINVAL
970 *
fd01b88c 971 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 972 * Update the superblock for rdev with data in mddev
973 * This does not write to disc.
974 *
975 */
976
977struct super_type {
0cd17fec
CW		 978 char *name;
979 struct module *owner;
c6563a8c
N		 980 int (*load_super)(struct md_rdev *rdev,
981 struct md_rdev *refdev,
0cd17fec 982 int minor_version);
c6563a8c
N		 983 int (*validate_super)(struct mddev *mddev,
984 struct md_rdev *rdev);
985 void (*sync_super)(struct mddev *mddev,
986 struct md_rdev *rdev);
3cb03002 987 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 988 sector_t num_sectors);
c6563a8c
N		 989 int (*allow_new_offset)(struct md_rdev *rdev,
990 unsigned long long new_offset);
1da177e4
LT		 991};
992
0894cc30
AN		 993/*
994 * Check that the given mddev has no bitmap.
995 *
996 * This function is called from the run method of all personalities that do not
997 * support bitmaps. It prints an error message and returns non-zero if mddev
998 * has a bitmap. Otherwise, it returns 0.
999 *
1000 */
fd01b88c 1001int md_check_no_bitmap(struct mddev *mddev)
0894cc30 1002{
c3d9714e 1003 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 1004 return 0;
1005 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1006 mdname(mddev), mddev->pers->name);
1007 return 1;
1008}
1009EXPORT_SYMBOL(md_check_no_bitmap);
1010
1da177e4
LT		 1011/*
1012 * load_super for 0.90.0
1013 */
3cb03002 1014static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1015{
1016 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1017 mdp_super_t *sb;
1018 int ret;
1da177e4
LT		 1019
1020 /*
0f420358 1021 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 1022 * it's at the end of the disk.
1023 *
1024 * It also happens to be a multiple of 4Kb.
1025 */
57b2caa3 1026 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 1027
0002b271 1028 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 1029 if (ret) return ret;
1030
1031 ret = -EINVAL;
1032
1033 bdevname(rdev->bdev, b);
65a06f06 1034 sb = page_address(rdev->sb_page);
1da177e4
LT		 1035
1036 if (sb->md_magic != MD_SB_MAGIC) {
1037 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1038 b);
1039 goto abort;
1040 }
1041
1042 if (sb->major_version != 0 ||
f6705578
N		 1043 sb->minor_version < 90 ||
1044 sb->minor_version > 91) {
1da177e4
LT		 1045 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1046 sb->major_version, sb->minor_version,
1047 b);
1048 goto abort;
1049 }
1050
1051 if (sb->raid_disks <= 0)
1052 goto abort;
1053
4d167f09 1054 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1055 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1056 b);
1057 goto abort;
1058 }
1059
1060 rdev->preferred_minor = sb->md_minor;
1061 rdev->data_offset = 0;
c6563a8c 1062 rdev->new_data_offset = 0;
0002b271 1063 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1064 rdev->badblocks.shift = -1;
1da177e4
LT		 1065
1066 if (sb->level == LEVEL_MULTIPATH)
1067 rdev->desc_nr = -1;
1068 else
1069 rdev->desc_nr = sb->this_disk.number;
1070
9a7b2b0f 1071 if (!refdev) {
1da177e4 1072 ret = 1;
9a7b2b0f 1073 } else {
1da177e4 1074 __u64 ev1, ev2;
65a06f06 1075 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1076 if (!uuid_equal(refsb, sb)) {
1077 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1078 b, bdevname(refdev->bdev,b2));
1079 goto abort;
1080 }
1081 if (!sb_equal(refsb, sb)) {
1082 printk(KERN_WARNING "md: %s has same UUID"
1083 " but different superblock to %s\n",
1084 b, bdevname(refdev->bdev, b2));
1085 goto abort;
1086 }
1087 ev1 = md_event(sb);
1088 ev2 = md_event(refsb);
1089 if (ev1 > ev2)
1090 ret = 1;
1091 else
1092 ret = 0;
1093 }
8190e754 1094 rdev->sectors = rdev->sb_start;
667a5313
N		 1095 /* Limit to 4TB as metadata cannot record more than that.
1096 * (not needed for Linear and RAID0 as metadata doesn't
1097 * record this size)
1098 */
1099 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
27a7b260 1100 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1101
27a7b260 1102 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1103 /* "this cannot possibly happen" ... */
1104 ret = -EINVAL;
1105
1da177e4
LT		 1106 abort:
1107 return ret;
1108}
1109
1110/*
1111 * validate_super for 0.90.0
1112 */
fd01b88c 1113static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1114{
1115 mdp_disk_t *desc;
65a06f06 1116 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1117 __u64 ev1 = md_event(sb);
1da177e4 1118
41158c7e 1119 rdev->raid_disk = -1;
c5d79adb
N		 1120 clear_bit(Faulty, &rdev->flags);
1121 clear_bit(In_sync, &rdev->flags);
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 1182 mddev->bitmap_info.space =
c9ad020f 1183 mddev->bitmap_info.default_space;
6409bb05 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;
07d84d10
N		 1200 } else {
1201 if (ev1 < mddev->events)
1202 /* just a hot-add of a new device, leave raid_disk at -1 */
1203 return 0;
1204 }
41158c7e 1205
1da177e4 1206 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1207 desc = sb->disks + rdev->desc_nr;
1208
1209 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1210 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1211 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1212 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1213 set_bit(In_sync, &rdev->flags);
1da177e4 1214 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1215 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1216 /* active but not in sync implies recovery up to
1217 * reshape position. We don't know exactly where
1218 * that is, so set to zero for now */
1219 if (mddev->minor_version >= 91) {
1220 rdev->recovery_offset = 0;
1221 rdev->raid_disk = desc->raid_disk;
1222 }
1da177e4 1223 }
8ddf9efe
N		 1224 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1225 set_bit(WriteMostly, &rdev->flags);
41158c7e 1226 } else /* MULTIPATH are always insync */
b2d444d7 1227 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1228 return 0;
1229}
1230
1231/*
1232 * sync_super for 0.90.0
1233 */
fd01b88c 1234static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1235{
1236 mdp_super_t *sb;
3cb03002 1237 struct md_rdev *rdev2;
1da177e4 1238 int next_spare = mddev->raid_disks;
19133a42 1239
1da177e4
LT		 1240
1241 /* make rdev->sb match mddev data..
1242 *
1243 * 1/ zero out disks
1244 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1245 * 3/ any empty disks < next_spare become removed
1246 *
1247 * disks[0] gets initialised to REMOVED because
1248 * we cannot be sure from other fields if it has
1249 * been initialised or not.
1250 */
1251 int i;
1252 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1253
61181565
N		 1254 rdev->sb_size = MD_SB_BYTES;
1255
65a06f06 1256 sb = page_address(rdev->sb_page);
1da177e4
LT		 1257
1258 memset(sb, 0, sizeof(*sb));
1259
1260 sb->md_magic = MD_SB_MAGIC;
1261 sb->major_version = mddev->major_version;
1da177e4
LT		 1262 sb->patch_version = mddev->patch_version;
1263 sb->gvalid_words = 0; /* ignored */
1264 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1265 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1266 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1267 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1268
1269 sb->ctime = mddev->ctime;
1270 sb->level = mddev->level;
58c0fed4 1271 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1272 sb->raid_disks = mddev->raid_disks;
1273 sb->md_minor = mddev->md_minor;
e691063a 1274 sb->not_persistent = 0;
1da177e4
LT		 1275 sb->utime = mddev->utime;
1276 sb->state = 0;
1277 sb->events_hi = (mddev->events>>32);
1278 sb->events_lo = (u32)mddev->events;
1279
f6705578
N		 1280 if (mddev->reshape_position == MaxSector)
1281 sb->minor_version = 90;
1282 else {
1283 sb->minor_version = 91;
1284 sb->reshape_position = mddev->reshape_position;
1285 sb->new_level = mddev->new_level;
1286 sb->delta_disks = mddev->delta_disks;
1287 sb->new_layout = mddev->new_layout;
664e7c41 1288 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1289 }
1290 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1291 if (mddev->in_sync)
1292 {
1293 sb->recovery_cp = mddev->recovery_cp;
1294 sb->cp_events_hi = (mddev->events>>32);
1295 sb->cp_events_lo = (u32)mddev->events;
1296 if (mddev->recovery_cp == MaxSector)
1297 sb->state = (1<< MD_SB_CLEAN);
1298 } else
1299 sb->recovery_cp = 0;
1300
1301 sb->layout = mddev->layout;
9d8f0363 1302 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1303
c3d9714e 1304 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1305 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1306
1da177e4 1307 sb->disks[0].state = (1<<MD_DISK_REMOVED);
dafb20fa 1308 rdev_for_each(rdev2, mddev) {
1da177e4 1309 mdp_disk_t *d;
86e6ffdd 1310 int desc_nr;
0261cd9f
N		 1311 int is_active = test_bit(In_sync, &rdev2->flags);
1312
1313 if (rdev2->raid_disk >= 0 &&
1314 sb->minor_version >= 91)
1315 /* we have nowhere to store the recovery_offset,
1316 * but if it is not below the reshape_position,
1317 * we can piggy-back on that.
1318 */
1319 is_active = 1;
1320 if (rdev2->raid_disk < 0 ||
1321 test_bit(Faulty, &rdev2->flags))
1322 is_active = 0;
1323 if (is_active)
86e6ffdd 1324 desc_nr = rdev2->raid_disk;
1da177e4 1325 else
86e6ffdd 1326 desc_nr = next_spare++;
19133a42 1327 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1328 d = &sb->disks[rdev2->desc_nr];
1329 nr_disks++;
1330 d->number = rdev2->desc_nr;
1331 d->major = MAJOR(rdev2->bdev->bd_dev);
1332 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1333 if (is_active)
1da177e4
LT		 1334 d->raid_disk = rdev2->raid_disk;
1335 else
1336 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1337 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1338 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1339 else if (is_active) {
1da177e4 1340 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1341 if (test_bit(In_sync, &rdev2->flags))
1342 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1343 active++;
1344 working++;
1345 } else {
1346 d->state = 0;
1347 spare++;
1348 working++;
1349 }
8ddf9efe
N		 1350 if (test_bit(WriteMostly, &rdev2->flags))
1351 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1352 }
1da177e4
LT		 1353 /* now set the "removed" and "faulty" bits on any missing devices */
1354 for (i=0 ; i < mddev->raid_disks ; i++) {
1355 mdp_disk_t *d = &sb->disks[i];
1356 if (d->state == 0 && d->number == 0) {
1357 d->number = i;
1358 d->raid_disk = i;
1359 d->state = (1<<MD_DISK_REMOVED);
1360 d->state |= (1<<MD_DISK_FAULTY);
1361 failed++;
1362 }
1363 }
1364 sb->nr_disks = nr_disks;
1365 sb->active_disks = active;
1366 sb->working_disks = working;
1367 sb->failed_disks = failed;
1368 sb->spare_disks = spare;
1369
1370 sb->this_disk = sb->disks[rdev->desc_nr];
1371 sb->sb_csum = calc_sb_csum(sb);
1372}
1373
0cd17fec
CW		 1374/*
1375 * rdev_size_change for 0.90.0
1376 */
1377static unsigned long long
3cb03002 1378super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1379{
58c0fed4 1380 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1381 return 0; /* component must fit device */
c3d9714e 1382 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1383 return 0; /* can't move bitmap */
57b2caa3 1384 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1385 if (!num_sectors || num_sectors > rdev->sb_start)
1386 num_sectors = rdev->sb_start;
27a7b260
N		 1387 /* Limit to 4TB as metadata cannot record more than that.
1388 * 4TB == 2^32 KB, or 2*2^32 sectors.
1389 */
667a5313 1390 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
27a7b260 1391 num_sectors = (2ULL << 32) - 2;
0f420358 1392 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1393 rdev->sb_page);
1394 md_super_wait(rdev->mddev);
c26a44ed 1395 return num_sectors;
0cd17fec
CW		 1396}
1397
c6563a8c
N		 1398static int
1399super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1400{
1401 /* non-zero offset changes not possible with v0.90 */
1402 return new_offset == 0;
1403}
0cd17fec 1404
1da177e4
LT		 1405/*
1406 * version 1 superblock
1407 */
1408
1c05b4bc 1409static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1410{
1c05b4bc
N		 1411 __le32 disk_csum;
1412 u32 csum;
1da177e4
LT		 1413 unsigned long long newcsum;
1414 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1415 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1416
1417 disk_csum = sb->sb_csum;
1418 sb->sb_csum = 0;
1419 newcsum = 0;
1f3c9907 1420 for (; size >= 4; size -= 4)
1da177e4
LT		 1421 newcsum += le32_to_cpu(*isuper++);
1422
1423 if (size == 2)
1c05b4bc 1424 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1425
1426 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1427 sb->sb_csum = disk_csum;
1428 return cpu_to_le32(csum);
1429}
1430
2699b672
N		 1431static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1432 int acknowledged);
3cb03002 1433static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1434{
1435 struct mdp_superblock_1 *sb;
1436 int ret;
0f420358 1437 sector_t sb_start;
c6563a8c 1438 sector_t sectors;
1da177e4 1439 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1440 int bmask;
1da177e4
LT		 1441
1442 /*
0f420358 1443 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1444 * It is always aligned to a 4K boundary and
1445 * depeding on minor_version, it can be:
1446 * 0: At least 8K, but less than 12K, from end of device
1447 * 1: At start of device
1448 * 2: 4K from start of device.
1449 */
1450 switch(minor_version) {
1451 case 0:
77304d2a 1452 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1453 sb_start -= 8*2;
1454 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1455 break;
1456 case 1:
0f420358 1457 sb_start = 0;
1da177e4
LT		 1458 break;
1459 case 2:
0f420358 1460 sb_start = 8;
1da177e4
LT		 1461 break;
1462 default:
1463 return -EINVAL;
1464 }
0f420358 1465 rdev->sb_start = sb_start;
1da177e4 1466
0002b271
N		 1467 /* superblock is rarely larger than 1K, but it can be larger,
1468 * and it is safe to read 4k, so we do that
1469 */
1470 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1471 if (ret) return ret;
1472
1473
65a06f06 1474 sb = page_address(rdev->sb_page);
1da177e4
LT		 1475
1476 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1477 sb->major_version != cpu_to_le32(1) ||
1478 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1479 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1480 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1481 return -EINVAL;
1482
1483 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1484 printk("md: invalid superblock checksum on %s\n",
1485 bdevname(rdev->bdev,b));
1486 return -EINVAL;
1487 }
1488 if (le64_to_cpu(sb->data_size) < 10) {
1489 printk("md: data_size too small on %s\n",
1490 bdevname(rdev->bdev,b));
1491 return -EINVAL;
1492 }
c6563a8c
N		 1493 if (sb->pad0 ||
1494 sb->pad3[0] ||
1495 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1496 /* Some padding is non-zero, might be a new feature */
1497 return -EINVAL;
e11e93fa 1498
1da177e4
LT		 1499 rdev->preferred_minor = 0xffff;
1500 rdev->data_offset = le64_to_cpu(sb->data_offset);
c6563a8c
N		 1501 rdev->new_data_offset = rdev->data_offset;
1502 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1503 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1504 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
4dbcdc75 1505 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1506
0002b271 1507 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1508 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1509 if (rdev->sb_size & bmask)
a1801f85
N		 1510 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1511
1512 if (minor_version
0f420358 1513 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1514 return -EINVAL;
c6563a8c
N		 1515 if (minor_version
1516 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1517 return -EINVAL;
0002b271 1518
31b65a0d
N		 1519 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1520 rdev->desc_nr = -1;
1521 else
1522 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1523
2699b672
N		 1524 if (!rdev->bb_page) {
1525 rdev->bb_page = alloc_page(GFP_KERNEL);
1526 if (!rdev->bb_page)
1527 return -ENOMEM;
1528 }
1529 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1530 rdev->badblocks.count == 0) {
1531 /* need to load the bad block list.
1532 * Currently we limit it to one page.
1533 */
1534 s32 offset;
1535 sector_t bb_sector;
1536 u64 *bbp;
1537 int i;
1538 int sectors = le16_to_cpu(sb->bblog_size);
1539 if (sectors > (PAGE_SIZE / 512))
1540 return -EINVAL;
1541 offset = le32_to_cpu(sb->bblog_offset);
1542 if (offset == 0)
1543 return -EINVAL;
1544 bb_sector = (long long)offset;
1545 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1546 rdev->bb_page, READ, true))
1547 return -EIO;
1548 bbp = (u64 *)page_address(rdev->bb_page);
1549 rdev->badblocks.shift = sb->bblog_shift;
1550 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1551 u64 bb = le64_to_cpu(*bbp);
1552 int count = bb & (0x3ff);
1553 u64 sector = bb >> 10;
1554 sector <<= sb->bblog_shift;
1555 count <<= sb->bblog_shift;
1556 if (bb + 1 == 0)
1557 break;
1558 if (md_set_badblocks(&rdev->badblocks,
1559 sector, count, 1) == 0)
1560 return -EINVAL;
1561 }
486adf72
N		 1562 } else if (sb->bblog_offset != 0)
1563 rdev->badblocks.shift = 0;
2699b672 1564
9a7b2b0f 1565 if (!refdev) {
8ed75463 1566 ret = 1;
9a7b2b0f 1567 } else {
1da177e4 1568 __u64 ev1, ev2;
65a06f06 1569 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1570
1571 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1572 sb->level != refsb->level ||
1573 sb->layout != refsb->layout ||
1574 sb->chunksize != refsb->chunksize) {
1575 printk(KERN_WARNING "md: %s has strangely different"
1576 " superblock to %s\n",
1577 bdevname(rdev->bdev,b),
1578 bdevname(refdev->bdev,b2));
1579 return -EINVAL;
1580 }
1581 ev1 = le64_to_cpu(sb->events);
1582 ev2 = le64_to_cpu(refsb->events);
1583
1584 if (ev1 > ev2)
8ed75463
N		 1585 ret = 1;
1586 else
1587 ret = 0;
1da177e4 1588 }
c6563a8c
N		 1589 if (minor_version) {
1590 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1591 sectors -= rdev->data_offset;
1592 } else
1593 sectors = rdev->sb_start;
1594 if (sectors < le64_to_cpu(sb->data_size))
1da177e4 1595 return -EINVAL;
dd8ac336 1596 rdev->sectors = le64_to_cpu(sb->data_size);
8ed75463 1597 return ret;
1da177e4
LT		 1598}
1599
fd01b88c 1600static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1601{
65a06f06 1602 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1603 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1604
41158c7e 1605 rdev->raid_disk = -1;
c5d79adb
N		 1606 clear_bit(Faulty, &rdev->flags);
1607 clear_bit(In_sync, &rdev->flags);
1608 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1609
1da177e4
LT		 1610 if (mddev->raid_disks == 0) {
1611 mddev->major_version = 1;
1612 mddev->patch_version = 0;
e691063a 1613 mddev->external = 0;
9d8f0363 1614 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1615 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1616 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1617 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1618 mddev->clevel[0] = 0;
1da177e4
LT		 1619 mddev->layout = le32_to_cpu(sb->layout);
1620 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1621 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1622 mddev->events = ev1;
c3d9714e 1623 mddev->bitmap_info.offset = 0;
6409bb05
N		 1624 mddev->bitmap_info.space = 0;
1625 /* Default location for bitmap is 1K after superblock
1626 * using 3K - total of 4K
1627 */
c3d9714e 1628 mddev->bitmap_info.default_offset = 1024 >> 9;
6409bb05 1629 mddev->bitmap_info.default_space = (4096-1024) >> 9;
2c810cdd
N		 1630 mddev->reshape_backwards = 0;
1631
1da177e4
LT		 1632 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1633 memcpy(mddev->uuid, sb->set_uuid, 16);
1634
1635 mddev->max_disks = (4096-256)/2;
a654b9d8 1636
71c0805c 1637 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
6409bb05 1638 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1639 mddev->bitmap_info.offset =
1640 (__s32)le32_to_cpu(sb->bitmap_offset);
6409bb05
N		 1641 /* Metadata doesn't record how much space is available.
1642 * For 1.0, we assume we can use up to the superblock
1643 * if before, else to 4K beyond superblock.
1644 * For others, assume no change is possible.
1645 */
1646 if (mddev->minor_version > 0)
1647 mddev->bitmap_info.space = 0;
1648 else if (mddev->bitmap_info.offset > 0)
1649 mddev->bitmap_info.space =
1650 8 - mddev->bitmap_info.offset;
1651 else
1652 mddev->bitmap_info.space =
1653 -mddev->bitmap_info.offset;
1654 }
e11e93fa 1655
f6705578
N		 1656 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1657 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1658 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1659 mddev->new_level = le32_to_cpu(sb->new_level);
1660 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1661 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
2c810cdd
N		 1662 if (mddev->delta_disks < 0 ||
1663 (mddev->delta_disks == 0 &&
1664 (le32_to_cpu(sb->feature_map)
1665 & MD_FEATURE_RESHAPE_BACKWARDS)))
1666 mddev->reshape_backwards = 1;
f6705578
N		 1667 } else {
1668 mddev->reshape_position = MaxSector;
1669 mddev->delta_disks = 0;
1670 mddev->new_level = mddev->level;
1671 mddev->new_layout = mddev->layout;
664e7c41 1672 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1673 }
1674
41158c7e 1675 } else if (mddev->pers == NULL) {
be6800a7
N		 1676 /* Insist of good event counter while assembling, except for
1677 * spares (which don't need an event count) */
1da177e4 1678 ++ev1;
be6800a7
N		 1679 if (rdev->desc_nr >= 0 &&
1680 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1681 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1682 if (ev1 < mddev->events)
1683 return -EINVAL;
41158c7e
N		 1684 } else if (mddev->bitmap) {
1685 /* If adding to array with a bitmap, then we can accept an
1686 * older device, but not too old.
1687 */
41158c7e
N		 1688 if (ev1 < mddev->bitmap->events_cleared)
1689 return 0;
07d84d10
N		 1690 } else {
1691 if (ev1 < mddev->events)
1692 /* just a hot-add of a new device, leave raid_disk at -1 */
1693 return 0;
1694 }
1da177e4
LT		 1695 if (mddev->level != LEVEL_MULTIPATH) {
1696 int role;
3673f305
N		 1697 if (rdev->desc_nr < 0 ||
1698 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1699 role = 0xffff;
1700 rdev->desc_nr = -1;
1701 } else
1702 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1703 switch(role) {
1704 case 0xffff: /* spare */
1da177e4
LT		 1705 break;
1706 case 0xfffe: /* faulty */
b2d444d7 1707 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1708 break;
1709 default:
5fd6c1dc
N		 1710 if ((le32_to_cpu(sb->feature_map) &
1711 MD_FEATURE_RECOVERY_OFFSET))
1712 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1713 else
1714 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1715 rdev->raid_disk = role;
1716 break;
1717 }
8ddf9efe
N		 1718 if (sb->devflags & WriteMostly1)
1719 set_bit(WriteMostly, &rdev->flags);
2d78f8c4
N		 1720 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1721 set_bit(Replacement, &rdev->flags);
41158c7e 1722 } else /* MULTIPATH are always insync */
b2d444d7 1723 set_bit(In_sync, &rdev->flags);
41158c7e 1724
1da177e4
LT		 1725 return 0;
1726}
1727
fd01b88c 1728static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1729{
1730 struct mdp_superblock_1 *sb;
3cb03002 1731 struct md_rdev *rdev2;
1da177e4
LT		 1732 int max_dev, i;
1733 /* make rdev->sb match mddev and rdev data. */
1734
65a06f06 1735 sb = page_address(rdev->sb_page);
1da177e4
LT		 1736
1737 sb->feature_map = 0;
1738 sb->pad0 = 0;
5fd6c1dc 1739 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1740 memset(sb->pad3, 0, sizeof(sb->pad3));
1741
1742 sb->utime = cpu_to_le64((__u64)mddev->utime);
1743 sb->events = cpu_to_le64(mddev->events);
1744 if (mddev->in_sync)
1745 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1746 else
1747 sb->resync_offset = cpu_to_le64(0);
1748
1c05b4bc 1749 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1750
f0ca340c 1751 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1752 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1753 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1754 sb->level = cpu_to_le32(mddev->level);
1755 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1756
aeb9b211
N		 1757 if (test_bit(WriteMostly, &rdev->flags))
1758 sb->devflags |= WriteMostly1;
1759 else
1760 sb->devflags &= ~WriteMostly1;
c6563a8c
N		 1761 sb->data_offset = cpu_to_le64(rdev->data_offset);
1762 sb->data_size = cpu_to_le64(rdev->sectors);
aeb9b211 1763
c3d9714e
N		 1764 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1765 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1766 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1767 }
5fd6c1dc
N		 1768
1769 if (rdev->raid_disk >= 0 &&
97e4f42d 1770 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1771 sb->feature_map |=
1772 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1773 sb->recovery_offset =
1774 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc 1775 }
2d78f8c4
N		 1776 if (test_bit(Replacement, &rdev->flags))
1777 sb->feature_map |=
1778 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1779
f6705578
N		 1780 if (mddev->reshape_position != MaxSector) {
1781 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1782 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1783 sb->new_layout = cpu_to_le32(mddev->new_layout);
1784 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1785 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1786 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2c810cdd
N		 1787 if (mddev->delta_disks == 0 &&
1788 mddev->reshape_backwards)
1789 sb->feature_map
1790 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
c6563a8c
N		 1791 if (rdev->new_data_offset != rdev->data_offset) {
1792 sb->feature_map
1793 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1794 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1795 - rdev->data_offset));
1796 }
f6705578 1797 }
a654b9d8 1798
2699b672
N		 1799 if (rdev->badblocks.count == 0)
1800 /* Nothing to do for bad blocks*/ ;
1801 else if (sb->bblog_offset == 0)
1802 /* Cannot record bad blocks on this device */
1803 md_error(mddev, rdev);
1804 else {
1805 struct badblocks *bb = &rdev->badblocks;
1806 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1807 u64 *p = bb->page;
1808 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1809 if (bb->changed) {
1810 unsigned seq;
1811
1812retry:
1813 seq = read_seqbegin(&bb->lock);
1814
1815 memset(bbp, 0xff, PAGE_SIZE);
1816
1817 for (i = 0 ; i < bb->count ; i++) {
35f9ac2d 1818 u64 internal_bb = p[i];
2699b672
N		 1819 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1820 | BB_LEN(internal_bb));
35f9ac2d 1821 bbp[i] = cpu_to_le64(store_bb);
2699b672 1822 }
d0962936 1823 bb->changed = 0;
2699b672
N		 1824 if (read_seqretry(&bb->lock, seq))
1825 goto retry;
1826
1827 bb->sector = (rdev->sb_start +
1828 (int)le32_to_cpu(sb->bblog_offset));
1829 bb->size = le16_to_cpu(sb->bblog_size);
2699b672
N		 1830 }
1831 }
1832
1da177e4 1833 max_dev = 0;
dafb20fa 1834 rdev_for_each(rdev2, mddev)
1da177e4
LT		 1835 if (rdev2->desc_nr+1 > max_dev)
1836 max_dev = rdev2->desc_nr+1;
a778b73f 1837
70471daf
N		 1838 if (max_dev > le32_to_cpu(sb->max_dev)) {
1839 int bmask;
a778b73f 1840 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1841 rdev->sb_size = max_dev * 2 + 256;
1842 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1843 if (rdev->sb_size & bmask)
1844 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1845 } else
1846 max_dev = le32_to_cpu(sb->max_dev);
1847
1da177e4
LT		 1848 for (i=0; i<max_dev;i++)
1849 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1850
dafb20fa 1851 rdev_for_each(rdev2, mddev) {
1da177e4 1852 i = rdev2->desc_nr;
b2d444d7 1853 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1854 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1855 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1856 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1857 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1858 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1859 else
1860 sb->dev_roles[i] = cpu_to_le16(0xffff);
1861 }
1862
1da177e4
LT		 1863 sb->sb_csum = calc_sb_1_csum(sb);
1864}
1865
0cd17fec 1866static unsigned long long
3cb03002 1867super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 1868{
1869 struct mdp_superblock_1 *sb;
15f4a5fd 1870 sector_t max_sectors;
58c0fed4 1871 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1872 return 0; /* component must fit device */
c6563a8c
N		 1873 if (rdev->data_offset != rdev->new_data_offset)
1874 return 0; /* too confusing */
0f420358 1875 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1876 /* minor versions 1 and 2; superblock before data */
77304d2a 1877 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1878 max_sectors -= rdev->data_offset;
1879 if (!num_sectors || num_sectors > max_sectors)
1880 num_sectors = max_sectors;
c3d9714e 1881 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1882 /* minor version 0 with bitmap we can't move */
1883 return 0;
1884 } else {
1885 /* minor version 0; superblock after data */
0f420358 1886 sector_t sb_start;
77304d2a 1887 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1888 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1889 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1890 if (!num_sectors || num_sectors > max_sectors)
1891 num_sectors = max_sectors;
0f420358 1892 rdev->sb_start = sb_start;
0cd17fec 1893 }
65a06f06 1894 sb = page_address(rdev->sb_page);
15f4a5fd 1895 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1896 sb->super_offset = rdev->sb_start;
0cd17fec 1897 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1898 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1899 rdev->sb_page);
1900 md_super_wait(rdev->mddev);
c26a44ed 1901 return num_sectors;
c6563a8c
N		 1902
1903}
1904
1905static int
1906super_1_allow_new_offset(struct md_rdev *rdev,
1907 unsigned long long new_offset)
1908{
1909 /* All necessary checks on new >= old have been done */
1910 struct bitmap *bitmap;
1911 if (new_offset >= rdev->data_offset)
1912 return 1;
1913
1914 /* with 1.0 metadata, there is no metadata to tread on
1915 * so we can always move back */
1916 if (rdev->mddev->minor_version == 0)
1917 return 1;
1918
1919 /* otherwise we must be sure not to step on
1920 * any metadata, so stay:
1921 * 36K beyond start of superblock
1922 * beyond end of badblocks
1923 * beyond write-intent bitmap
1924 */
1925 if (rdev->sb_start + (32+4)*2 > new_offset)
1926 return 0;
1927 bitmap = rdev->mddev->bitmap;
1928 if (bitmap && !rdev->mddev->bitmap_info.file &&
1929 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1ec885cd 1930 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
c6563a8c
N		 1931 return 0;
1932 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1933 return 0;
1934
1935 return 1;
0cd17fec 1936}
1da177e4 1937
75c96f85 1938static struct super_type super_types[] = {
1da177e4
LT		 1939 [0] = {
1940 .name = "0.90.0",
1941 .owner = THIS_MODULE,
0cd17fec
CW		 1942 .load_super = super_90_load,
1943 .validate_super = super_90_validate,
1944 .sync_super = super_90_sync,
1945 .rdev_size_change = super_90_rdev_size_change,
c6563a8c 1946 .allow_new_offset = super_90_allow_new_offset,
1da177e4
LT		 1947 },
1948 [1] = {
1949 .name = "md-1",
1950 .owner = THIS_MODULE,
0cd17fec
CW		 1951 .load_super = super_1_load,
1952 .validate_super = super_1_validate,
1953 .sync_super = super_1_sync,
1954 .rdev_size_change = super_1_rdev_size_change,
c6563a8c 1955 .allow_new_offset = super_1_allow_new_offset,
1da177e4
LT		 1956 },
1957};
1da177e4 1958
fd01b88c 1959static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 1960{
1961 if (mddev->sync_super) {
1962 mddev->sync_super(mddev, rdev);
1963 return;
1964 }
1965
1966 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1967
1968 super_types[mddev->major_version].sync_super(mddev, rdev);
1969}
1970
fd01b88c 1971static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1972{
3cb03002 1973 struct md_rdev *rdev, *rdev2;
1da177e4 1974
4b80991c
N		 1975 rcu_read_lock();
1976 rdev_for_each_rcu(rdev, mddev1)
1977 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1978 if (rdev->bdev->bd_contains ==
4b80991c
N		 1979 rdev2->bdev->bd_contains) {
1980 rcu_read_unlock();
7dd5e7c3 1981 return 1;
4b80991c
N		 1982 }
1983 rcu_read_unlock();
1da177e4
LT		 1984 return 0;
1985}
1986
1987static LIST_HEAD(pending_raid_disks);
1988
ac5e7113
AN		 1989/*
1990 * Try to register data integrity profile for an mddev
1991 *
1992 * This is called when an array is started and after a disk has been kicked
1993 * from the array. It only succeeds if all working and active component devices
1994 * are integrity capable with matching profiles.
1995 */
fd01b88c 1996int md_integrity_register(struct mddev *mddev)
ac5e7113 1997{
3cb03002 1998 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 1999
2000 if (list_empty(&mddev->disks))
2001 return 0; /* nothing to do */
629acb6a
JB		 2002 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2003 return 0; /* shouldn't register, or already is */
dafb20fa 2004 rdev_for_each(rdev, mddev) {
ac5e7113
AN		 2005 /* skip spares and non-functional disks */
2006 if (test_bit(Faulty, &rdev->flags))
2007 continue;
2008 if (rdev->raid_disk < 0)
2009 continue;
ac5e7113
AN		 2010 if (!reference) {
2011 /* Use the first rdev as the reference */
2012 reference = rdev;
2013 continue;
2014 }
2015 /* does this rdev's profile match the reference profile? */
2016 if (blk_integrity_compare(reference->bdev->bd_disk,
2017 rdev->bdev->bd_disk) < 0)
2018 return -EINVAL;
2019 }
89078d57
MP		 2020 if (!reference || !bdev_get_integrity(reference->bdev))
2021 return 0;
ac5e7113
AN		 2022 /*
2023 * All component devices are integrity capable and have matching
2024 * profiles, register the common profile for the md device.
2025 */
2026 if (blk_integrity_register(mddev->gendisk,
2027 bdev_get_integrity(reference->bdev)) != 0) {
2028 printk(KERN_ERR "md: failed to register integrity for %s\n",
2029 mdname(mddev));
2030 return -EINVAL;
2031 }
a91a2785
MP		 2032 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2033 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2034 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2035 mdname(mddev));
2036 return -EINVAL;
2037 }
ac5e7113
AN		 2038 return 0;
2039}
2040EXPORT_SYMBOL(md_integrity_register);
2041
2042/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 2043void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 2044{
2863b9eb
JB		 2045 struct blk_integrity *bi_rdev;
2046 struct blk_integrity *bi_mddev;
2047
2048 if (!mddev->gendisk)
2049 return;
2050
2051 bi_rdev = bdev_get_integrity(rdev->bdev);
2052 bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 2053
ac5e7113 2054 if (!bi_mddev) /* nothing to do */
3f9d99c1 2055 return;
ac5e7113 2056 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 2057 return;
ac5e7113
AN		 2058 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2059 rdev->bdev->bd_disk) >= 0)
2060 return;
2061 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2062 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2063}
ac5e7113 2064EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2065
fd01b88c 2066static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
1da177e4 2067{
7dd5e7c3 2068 char b[BDEVNAME_SIZE];
f637b9f9 2069 struct kobject *ko;
1edf80d3 2070 char *s;
5e55e2f5 2071 int err;
1da177e4
LT		 2072
2073 if (rdev->mddev) {
2074 MD_BUG();
2075 return -EINVAL;
2076 }
11e2ede0
DW		 2077
2078 /* prevent duplicates */
2079 if (find_rdev(mddev, rdev->bdev->bd_dev))
2080 return -EEXIST;
2081
dd8ac336
AN		 2082 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2083 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2084 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2085 if (mddev->pers) {
2086 /* Cannot change size, so fail
2087 * If mddev->level <= 0, then we don't care
2088 * about aligning sizes (e.g. linear)
2089 */
2090 if (mddev->level > 0)
2091 return -ENOSPC;
2092 } else
dd8ac336 2093 mddev->dev_sectors = rdev->sectors;
2bf071bf 2094 }
1da177e4
LT		 2095
2096 /* Verify rdev->desc_nr is unique.
2097 * If it is -1, assign a free number, else
2098 * check number is not in use
2099 */
2100 if (rdev->desc_nr < 0) {
2101 int choice = 0;
2102 if (mddev->pers) choice = mddev->raid_disks;
2103 while (find_rdev_nr(mddev, choice))
2104 choice++;
2105 rdev->desc_nr = choice;
2106 } else {
2107 if (find_rdev_nr(mddev, rdev->desc_nr))
2108 return -EBUSY;
2109 }
de01dfad
N		 2110 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2111 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2112 mdname(mddev), mddev->max_disks);
2113 return -EBUSY;
2114 }
19133a42 2115 bdevname(rdev->bdev,b);
649316b2 2116 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2117 *s = '!';
649316b2 2118
1da177e4 2119 rdev->mddev = mddev;
19133a42 2120 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2121
b2d6db58 2122 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2123 goto fail;
86e6ffdd 2124
0762b8bd 2125 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2126 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2127 /* failure here is OK */;
2128 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2129
4b80991c 2130 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2131 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2132
2133 /* May as well allow recovery to be retried once */
5389042f 2134 mddev->recovery_disabled++;
3f9d99c1 2135
1da177e4 2136 return 0;
5e55e2f5
N		 2137
2138 fail:
2139 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2140 b, mdname(mddev));
2141 return err;
1da177e4
LT		 2142}
2143
177a99b2 2144static void md_delayed_delete(struct work_struct *ws)
5792a285 2145{
3cb03002 2146 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2147 kobject_del(&rdev->kobj);
177a99b2 2148 kobject_put(&rdev->kobj);
5792a285
N		 2149}
2150
3cb03002 2151static void unbind_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2152{
2153 char b[BDEVNAME_SIZE];
2154 if (!rdev->mddev) {
2155 MD_BUG();
2156 return;
2157 }
49731baa 2158 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2159 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2160 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2161 rdev->mddev = NULL;
86e6ffdd 2162 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2163 sysfs_put(rdev->sysfs_state);
2164 rdev->sysfs_state = NULL;
2230dfe4 2165 rdev->badblocks.count = 0;
5792a285 2166 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2167 * writing to 'remove' to "dev/state". We also need
2168 * to delay it due to rcu usage.
5792a285 2169 */
4b80991c 2170 synchronize_rcu();
177a99b2
N		 2171 INIT_WORK(&rdev->del_work, md_delayed_delete);
2172 kobject_get(&rdev->kobj);
e804ac78 2173 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2174}
2175
2176/*
2177 * prevent the device from being mounted, repartitioned or
2178 * otherwise reused by a RAID array (or any other kernel
2179 * subsystem), by bd_claiming the device.
2180 */
3cb03002 2181static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2182{
2183 int err = 0;
2184 struct block_device *bdev;
2185 char b[BDEVNAME_SIZE];
2186
d4d77629 2187 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2188 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2189 if (IS_ERR(bdev)) {
2190 printk(KERN_ERR "md: could not open %s.\n",
2191 __bdevname(dev, b));
2192 return PTR_ERR(bdev);
2193 }
1da177e4
LT		 2194 rdev->bdev = bdev;
2195 return err;
2196}
2197
3cb03002 2198static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2199{
2200 struct block_device *bdev = rdev->bdev;
2201 rdev->bdev = NULL;
2202 if (!bdev)
2203 MD_BUG();
e525fd89 2204 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2205}
2206
2207void md_autodetect_dev(dev_t dev);
2208
3cb03002 2209static void export_rdev(struct md_rdev * rdev)
1da177e4
LT		 2210{
2211 char b[BDEVNAME_SIZE];
2212 printk(KERN_INFO "md: export_rdev(%s)\n",
2213 bdevname(rdev->bdev,b));
2214 if (rdev->mddev)
2215 MD_BUG();
545c8795 2216 md_rdev_clear(rdev);
1da177e4 2217#ifndef MODULE
d0fae18f
N		 2218 if (test_bit(AutoDetected, &rdev->flags))
2219 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2220#endif
2221 unlock_rdev(rdev);
86e6ffdd 2222 kobject_put(&rdev->kobj);
1da177e4
LT		 2223}
2224
3cb03002 2225static void kick_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2226{
2227 unbind_rdev_from_array(rdev);
2228 export_rdev(rdev);
2229}
2230
fd01b88c 2231static void export_array(struct mddev *mddev)
1da177e4 2232{
3cb03002 2233 struct md_rdev *rdev, *tmp;
1da177e4 2234
dafb20fa 2235 rdev_for_each_safe(rdev, tmp, mddev) {
1da177e4
LT		 2236 if (!rdev->mddev) {
2237 MD_BUG();
2238 continue;
2239 }
2240 kick_rdev_from_array(rdev);
2241 }
2242 if (!list_empty(&mddev->disks))
2243 MD_BUG();
2244 mddev->raid_disks = 0;
2245 mddev->major_version = 0;
2246}
2247
2248static void print_desc(mdp_disk_t *desc)
2249{
2250 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2251 desc->major,desc->minor,desc->raid_disk,desc->state);
2252}
2253
cd2ac932 2254static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 2255{
2256 int i;
2257
2258 printk(KERN_INFO
2259 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2260 sb->major_version, sb->minor_version, sb->patch_version,
2261 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2262 sb->ctime);
2263 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2264 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2265 sb->md_minor, sb->layout, sb->chunk_size);
2266 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2267 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2268 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2269 sb->failed_disks, sb->spare_disks,
2270 sb->sb_csum, (unsigned long)sb->events_lo);
2271
2272 printk(KERN_INFO);
2273 for (i = 0; i < MD_SB_DISKS; i++) {
2274 mdp_disk_t *desc;
2275
2276 desc = sb->disks + i;
2277 if (desc->number || desc->major || desc->minor ||
2278 desc->raid_disk || (desc->state && (desc->state != 4))) {
2279 printk(" D %2d: ", i);
2280 print_desc(desc);
2281 }
2282 }
2283 printk(KERN_INFO "md: THIS: ");
2284 print_desc(&sb->this_disk);
cd2ac932 2285}
1da177e4 2286
cd2ac932
CR		 2287static void print_sb_1(struct mdp_superblock_1 *sb)
2288{
2289 __u8 *uuid;
2290
2291 uuid = sb->set_uuid;
ad361c98 2292 printk(KERN_INFO
7b75c2f8 2293 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2294 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 2295 le32_to_cpu(sb->major_version),
2296 le32_to_cpu(sb->feature_map),
7b75c2f8 2297 uuid,
cd2ac932
CR		 2298 sb->set_name,
2299 (unsigned long long)le64_to_cpu(sb->ctime)
2300 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2301
2302 uuid = sb->device_uuid;
ad361c98
JP		 2303 printk(KERN_INFO
2304 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2305 " RO:%llu\n"
7b75c2f8 2306 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP		 2307 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2308 "md: (MaxDev:%u) \n",
cd2ac932
CR		 2309 le32_to_cpu(sb->level),
2310 (unsigned long long)le64_to_cpu(sb->size),
2311 le32_to_cpu(sb->raid_disks),
2312 le32_to_cpu(sb->layout),
2313 le32_to_cpu(sb->chunksize),
2314 (unsigned long long)le64_to_cpu(sb->data_offset),
2315 (unsigned long long)le64_to_cpu(sb->data_size),
2316 (unsigned long long)le64_to_cpu(sb->super_offset),
2317 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2318 le32_to_cpu(sb->dev_number),
7b75c2f8 2319 uuid,
cd2ac932
CR		 2320 sb->devflags,
2321 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2322 (unsigned long long)le64_to_cpu(sb->events),
2323 (unsigned long long)le64_to_cpu(sb->resync_offset),
2324 le32_to_cpu(sb->sb_csum),
2325 le32_to_cpu(sb->max_dev)
2326 );
1da177e4
LT		 2327}
2328
3cb03002 2329static void print_rdev(struct md_rdev *rdev, int major_version)
1da177e4
LT		 2330{
2331 char b[BDEVNAME_SIZE];
dd8ac336
AN		 2332 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2333 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 2334 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2335 rdev->desc_nr);
1da177e4 2336 if (rdev->sb_loaded) {
cd2ac932
CR		 2337 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2338 switch (major_version) {
2339 case 0:
65a06f06 2340 print_sb_90(page_address(rdev->sb_page));
cd2ac932
CR		 2341 break;
2342 case 1:
65a06f06 2343 print_sb_1(page_address(rdev->sb_page));
cd2ac932
CR		 2344 break;
2345 }
1da177e4
LT		 2346 } else
2347 printk(KERN_INFO "md: no rdev superblock!\n");
2348}
2349
5e56341d 2350static void md_print_devices(void)
1da177e4 2351{
159ec1fc 2352 struct list_head *tmp;
3cb03002 2353 struct md_rdev *rdev;
fd01b88c 2354 struct mddev *mddev;
1da177e4
LT		 2355 char b[BDEVNAME_SIZE];
2356
2357 printk("\n");
2358 printk("md: **********************************\n");
2359 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2360 printk("md: **********************************\n");
29ac4aa3 2361 for_each_mddev(mddev, tmp) {
1da177e4 2362
32a7627c
N		 2363 if (mddev->bitmap)
2364 bitmap_print_sb(mddev->bitmap);
2365 else
2366 printk("%s: ", mdname(mddev));
dafb20fa 2367 rdev_for_each(rdev, mddev)
1da177e4
LT		 2368 printk("<%s>", bdevname(rdev->bdev,b));
2369 printk("\n");
2370
dafb20fa 2371 rdev_for_each(rdev, mddev)
cd2ac932 2372 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 2373 }
2374 printk("md: **********************************\n");
2375 printk("\n");
2376}
2377
2378
fd01b88c 2379static void sync_sbs(struct mddev * mddev, int nospares)
1da177e4 2380{
42543769
N		 2381 /* Update each superblock (in-memory image), but
2382 * if we are allowed to, skip spares which already
2383 * have the right event counter, or have one earlier
2384 * (which would mean they aren't being marked as dirty
2385 * with the rest of the array)
2386 */
3cb03002 2387 struct md_rdev *rdev;
dafb20fa 2388 rdev_for_each(rdev, mddev) {
42543769
N		 2389 if (rdev->sb_events == mddev->events ||
2390 (nospares &&
2391 rdev->raid_disk < 0 &&
42543769
N		 2392 rdev->sb_events+1 == mddev->events)) {
2393 /* Don't update this superblock */
2394 rdev->sb_loaded = 2;
2395 } else {
076f968b 2396 sync_super(mddev, rdev);
42543769
N		 2397 rdev->sb_loaded = 1;
2398 }
1da177e4
LT		 2399 }
2400}
2401
fd01b88c 2402static void md_update_sb(struct mddev * mddev, int force_change)
1da177e4 2403{
3cb03002 2404 struct md_rdev *rdev;
06d91a5f 2405 int sync_req;
42543769 2406 int nospares = 0;
2699b672 2407 int any_badblocks_changed = 0;
1da177e4 2408
d87f064f
N		 2409 if (mddev->ro) {
2410 if (force_change)
2411 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2412 return;
2413 }
1da177e4 2414repeat:
3a3a5ddb 2415 /* First make sure individual recovery_offsets are correct */
dafb20fa 2416 rdev_for_each(rdev, mddev) {
3a3a5ddb
N		 2417 if (rdev->raid_disk >= 0 &&
2418 mddev->delta_disks >= 0 &&
2419 !test_bit(In_sync, &rdev->flags) &&
2420 mddev->curr_resync_completed > rdev->recovery_offset)
2421 rdev->recovery_offset = mddev->curr_resync_completed;
2422
2423 }
bd52b746 2424 if (!mddev->persistent) {
070dc6dd 2425 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2426 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2427 if (!mddev->external) {
d97a41dc 2428 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
dafb20fa 2429 rdev_for_each(rdev, mddev) {
de393cde 2430 if (rdev->badblocks.changed) {
d0962936 2431 rdev->badblocks.changed = 0;
de393cde
N		 2432 md_ack_all_badblocks(&rdev->badblocks);
2433 md_error(mddev, rdev);
2434 }
2435 clear_bit(Blocked, &rdev->flags);
2436 clear_bit(BlockedBadBlocks, &rdev->flags);
2437 wake_up(&rdev->blocked_wait);
2438 }
2439 }
3a3a5ddb
N		 2440 wake_up(&mddev->sb_wait);
2441 return;
2442 }
2443
a9701a30 2444 spin_lock_irq(&mddev->write_lock);
84692195 2445
3a3a5ddb
N		 2446 mddev->utime = get_seconds();
2447
850b2b42
N		 2448 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2449 force_change = 1;
2450 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2451 /* just a clean<-> dirty transition, possibly leave spares alone,
2452 * though if events isn't the right even/odd, we will have to do
2453 * spares after all
2454 */
2455 nospares = 1;
2456 if (force_change)
2457 nospares = 0;
2458 if (mddev->degraded)
84692195
N		 2459 /* If the array is degraded, then skipping spares is both
2460 * dangerous and fairly pointless.
2461 * Dangerous because a device that was removed from the array
2462 * might have a event_count that still looks up-to-date,
2463 * so it can be re-added without a resync.
2464 * Pointless because if there are any spares to skip,
2465 * then a recovery will happen and soon that array won't
2466 * be degraded any more and the spare can go back to sleep then.
2467 */
850b2b42 2468 nospares = 0;
84692195 2469
06d91a5f 2470 sync_req = mddev->in_sync;
42543769
N		 2471
2472 /* If this is just a dirty<->clean transition, and the array is clean
2473 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2474 if (nospares
42543769 2475 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2476 && mddev->can_decrease_events
2477 && mddev->events != 1) {
42543769 2478 mddev->events--;
a8707c08
N		 2479 mddev->can_decrease_events = 0;
2480 } else {
42543769
N		 2481 /* otherwise we have to go forward and ... */
2482 mddev->events ++;
a8707c08 2483 mddev->can_decrease_events = nospares;
42543769 2484 }
1da177e4
LT		 2485
2486 if (!mddev->events) {
2487 /*
2488 * oops, this 64-bit counter should never wrap.
2489 * Either we are in around ~1 trillion A.C., assuming
2490 * 1 reboot per second, or we have a bug:
2491 */
2492 MD_BUG();
2493 mddev->events --;
2494 }
2699b672 2495
dafb20fa 2496 rdev_for_each(rdev, mddev) {
2699b672
N		 2497 if (rdev->badblocks.changed)
2498 any_badblocks_changed++;
de393cde
N		 2499 if (test_bit(Faulty, &rdev->flags))
2500 set_bit(FaultRecorded, &rdev->flags);
2501 }
2699b672 2502
e691063a 2503 sync_sbs(mddev, nospares);
a9701a30 2504 spin_unlock_irq(&mddev->write_lock);
1da177e4 2505
36a4e1fe
N		 2506 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2507 mdname(mddev), mddev->in_sync);
1da177e4 2508
4ad13663 2509 bitmap_update_sb(mddev->bitmap);
dafb20fa 2510 rdev_for_each(rdev, mddev) {
1da177e4 2511 char b[BDEVNAME_SIZE];
36a4e1fe 2512
42543769
N		 2513 if (rdev->sb_loaded != 1)
2514 continue; /* no noise on spare devices */
1da177e4 2515
d70ed2e4
AW		 2516 if (!test_bit(Faulty, &rdev->flags) &&
2517 rdev->saved_raid_disk == -1) {
7bfa19f2 2518 md_super_write(mddev,rdev,
0f420358 2519 rdev->sb_start, rdev->sb_size,
7bfa19f2 2520 rdev->sb_page);
36a4e1fe
N		 2521 pr_debug("md: (write) %s's sb offset: %llu\n",
2522 bdevname(rdev->bdev, b),
2523 (unsigned long long)rdev->sb_start);
42543769 2524 rdev->sb_events = mddev->events;
2699b672
N		 2525 if (rdev->badblocks.size) {
2526 md_super_write(mddev, rdev,
2527 rdev->badblocks.sector,
2528 rdev->badblocks.size << 9,
2529 rdev->bb_page);
2530 rdev->badblocks.size = 0;
2531 }
7bfa19f2 2532
d70ed2e4 2533 } else if (test_bit(Faulty, &rdev->flags))
36a4e1fe
N		 2534 pr_debug("md: %s (skipping faulty)\n",
2535 bdevname(rdev->bdev, b));
d70ed2e4
AW		 2536 else
2537 pr_debug("(skipping incremental s/r ");
2538
7bfa19f2 2539 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2540 /* only need to write one superblock... */
2541 break;
2542 }
a9701a30 2543 md_super_wait(mddev);
850b2b42 2544 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2545
a9701a30 2546 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2547 if (mddev->in_sync != sync_req ||
2548 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2549 /* have to write it out again */
a9701a30 2550 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2551 goto repeat;
2552 }
850b2b42 2553 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2554 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2555 wake_up(&mddev->sb_wait);
acb180b0
N		 2556 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2557 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2558
dafb20fa 2559 rdev_for_each(rdev, mddev) {
de393cde
N		 2560 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2561 clear_bit(Blocked, &rdev->flags);
2562
2563 if (any_badblocks_changed)
2699b672 2564 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2565 clear_bit(BlockedBadBlocks, &rdev->flags);
2566 wake_up(&rdev->blocked_wait);
2567 }
1da177e4
LT		 2568}
2569
7f6ce769 2570/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2571 * We want to accept with case. For this we use cmd_match.
2572 */
2573static int cmd_match(const char *cmd, const char *str)
2574{
2575 /* See if cmd, written into a sysfs file, matches
2576 * str. They must either be the same, or cmd can
2577 * have a trailing newline
2578 */
2579 while (*cmd && *str && *cmd == *str) {
2580 cmd++;
2581 str++;
2582 }
2583 if (*cmd == '\n')
2584 cmd++;
2585 if (*str || *cmd)
2586 return 0;
2587 return 1;
2588}
2589
86e6ffdd
N		 2590struct rdev_sysfs_entry {
2591 struct attribute attr;
3cb03002
N		 2592 ssize_t (*show)(struct md_rdev *, char *);
2593 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2594};
2595
2596static ssize_t
3cb03002 2597state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2598{
2599 char *sep = "";
20a49ff6 2600 size_t len = 0;
86e6ffdd 2601
de393cde
N		 2602 if (test_bit(Faulty, &rdev->flags) ||
2603 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2604 len+= sprintf(page+len, "%sfaulty",sep);
2605 sep = ",";
2606 }
b2d444d7 2607 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2608 len += sprintf(page+len, "%sin_sync",sep);
2609 sep = ",";
2610 }
f655675b
N		 2611 if (test_bit(WriteMostly, &rdev->flags)) {
2612 len += sprintf(page+len, "%swrite_mostly",sep);
2613 sep = ",";
2614 }
de393cde 2615 if (test_bit(Blocked, &rdev->flags) ||
52c64152
N		 2616 (rdev->badblocks.unacked_exist
2617 && !test_bit(Faulty, &rdev->flags))) {
6bfe0b49
DW		 2618 len += sprintf(page+len, "%sblocked", sep);
2619 sep = ",";
2620 }
b2d444d7
N		 2621 if (!test_bit(Faulty, &rdev->flags) &&
2622 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2623 len += sprintf(page+len, "%sspare", sep);
2624 sep = ",";
2625 }
d7a9d443
N		 2626 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2627 len += sprintf(page+len, "%swrite_error", sep);
2628 sep = ",";
2629 }
2d78f8c4
N		 2630 if (test_bit(WantReplacement, &rdev->flags)) {
2631 len += sprintf(page+len, "%swant_replacement", sep);
2632 sep = ",";
2633 }
2634 if (test_bit(Replacement, &rdev->flags)) {
2635 len += sprintf(page+len, "%sreplacement", sep);
2636 sep = ",";
2637 }
2638
86e6ffdd
N		 2639 return len+sprintf(page+len, "\n");
2640}
2641
45dc2de1 2642static ssize_t
3cb03002 2643state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2644{
2645 /* can write
de393cde 2646 * faulty - simulates an error
45dc2de1 2647 * remove - disconnects the device
f655675b
N		 2648 * writemostly - sets write_mostly
2649 * -writemostly - clears write_mostly
de393cde
N		 2650 * blocked - sets the Blocked flags
2651 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2652 * insync - sets Insync providing device isn't active
d7a9d443
N		 2653 * write_error - sets WriteErrorSeen
2654 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2655 */
2656 int err = -EINVAL;
2657 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2658 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2659 if (test_bit(Faulty, &rdev->flags))
2660 err = 0;
2661 else
2662 err = -EBUSY;
45dc2de1
N		 2663 } else if (cmd_match(buf, "remove")) {
2664 if (rdev->raid_disk >= 0)
2665 err = -EBUSY;
2666 else {
fd01b88c 2667 struct mddev *mddev = rdev->mddev;
45dc2de1 2668 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2669 if (mddev->pers)
2670 md_update_sb(mddev, 1);
45dc2de1
N		 2671 md_new_event(mddev);
2672 err = 0;
2673 }
f655675b
N		 2674 } else if (cmd_match(buf, "writemostly")) {
2675 set_bit(WriteMostly, &rdev->flags);
2676 err = 0;
2677 } else if (cmd_match(buf, "-writemostly")) {
2678 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2679 err = 0;
2680 } else if (cmd_match(buf, "blocked")) {
2681 set_bit(Blocked, &rdev->flags);
2682 err = 0;
2683 } else if (cmd_match(buf, "-blocked")) {
de393cde 2684 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2685 rdev->badblocks.unacked_exist) {
de393cde
N		 2686 /* metadata handler doesn't understand badblocks,
2687 * so we need to fail the device
2688 */
2689 md_error(rdev->mddev, rdev);
2690 }
6bfe0b49 2691 clear_bit(Blocked, &rdev->flags);
de393cde 2692 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2693 wake_up(&rdev->blocked_wait);
2694 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2695 md_wakeup_thread(rdev->mddev->thread);
2696
6d56e278
N		 2697 err = 0;
2698 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2699 set_bit(In_sync, &rdev->flags);
f655675b 2700 err = 0;
d7a9d443
N		 2701 } else if (cmd_match(buf, "write_error")) {
2702 set_bit(WriteErrorSeen, &rdev->flags);
2703 err = 0;
2704 } else if (cmd_match(buf, "-write_error")) {
2705 clear_bit(WriteErrorSeen, &rdev->flags);
2706 err = 0;
2d78f8c4
N		 2707 } else if (cmd_match(buf, "want_replacement")) {
2708 /* Any non-spare device that is not a replacement can
2709 * become want_replacement at any time, but we then need to
2710 * check if recovery is needed.
2711 */
2712 if (rdev->raid_disk >= 0 &&
2713 !test_bit(Replacement, &rdev->flags))
2714 set_bit(WantReplacement, &rdev->flags);
2715 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2716 md_wakeup_thread(rdev->mddev->thread);
2717 err = 0;
2718 } else if (cmd_match(buf, "-want_replacement")) {
2719 /* Clearing 'want_replacement' is always allowed.
2720 * Once replacements starts it is too late though.
2721 */
2722 err = 0;
2723 clear_bit(WantReplacement, &rdev->flags);
2724 } else if (cmd_match(buf, "replacement")) {
2725 /* Can only set a device as a replacement when array has not
2726 * yet been started. Once running, replacement is automatic
2727 * from spares, or by assigning 'slot'.
2728 */
2729 if (rdev->mddev->pers)
2730 err = -EBUSY;
2731 else {
2732 set_bit(Replacement, &rdev->flags);
2733 err = 0;
2734 }
2735 } else if (cmd_match(buf, "-replacement")) {
2736 /* Similarly, can only clear Replacement before start */
2737 if (rdev->mddev->pers)
2738 err = -EBUSY;
2739 else {
2740 clear_bit(Replacement, &rdev->flags);
2741 err = 0;
2742 }
45dc2de1 2743 }
00bcb4ac
N		 2744 if (!err)
2745 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2746 return err ? err : len;
2747}
80ca3a44
N		 2748static struct rdev_sysfs_entry rdev_state =
2749__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2750
4dbcdc75 2751static ssize_t
3cb03002 2752errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2753{
2754 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2755}
2756
2757static ssize_t
3cb03002 2758errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N		 2759{
2760 char *e;
2761 unsigned long n = simple_strtoul(buf, &e, 10);
2762 if (*buf && (*e == 0 || *e == '\n')) {
2763 atomic_set(&rdev->corrected_errors, n);
2764 return len;
2765 }
2766 return -EINVAL;
2767}
2768static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2769__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2770
014236d2 2771static ssize_t
3cb03002 2772slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2773{
2774 if (rdev->raid_disk < 0)
2775 return sprintf(page, "none\n");
2776 else
2777 return sprintf(page, "%d\n", rdev->raid_disk);
2778}
2779
2780static ssize_t
3cb03002 2781slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N		 2782{
2783 char *e;
c303da6d 2784 int err;
014236d2
N		 2785 int slot = simple_strtoul(buf, &e, 10);
2786 if (strncmp(buf, "none", 4)==0)
2787 slot = -1;
2788 else if (e==buf || (*e && *e!= '\n'))
2789 return -EINVAL;
6c2fce2e 2790 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2791 /* Setting 'slot' on an active array requires also
2792 * updating the 'rd%d' link, and communicating
2793 * with the personality with ->hot_*_disk.
2794 * For now we only support removing
2795 * failed/spare devices. This normally happens automatically,
2796 * but not when the metadata is externally managed.
2797 */
c303da6d
N		 2798 if (rdev->raid_disk == -1)
2799 return -EEXIST;
2800 /* personality does all needed checks */
01393f3d 2801 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d 2802 return -EINVAL;
746d3207
N		 2803 clear_bit(Blocked, &rdev->flags);
2804 remove_and_add_spares(rdev->mddev, rdev);
2805 if (rdev->raid_disk >= 0)
2806 return -EBUSY;
c303da6d
N		 2807 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2808 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2809 } else if (rdev->mddev->pers) {
6c2fce2e 2810 /* Activating a spare .. or possibly reactivating
6d56e278 2811 * if we ever get bitmaps working here.
6c2fce2e
NB		 2812 */
2813
2814 if (rdev->raid_disk != -1)
2815 return -EBUSY;
2816
c6751b2b
N		 2817 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2818 return -EBUSY;
2819
6c2fce2e
NB		 2820 if (rdev->mddev->pers->hot_add_disk == NULL)
2821 return -EINVAL;
2822
ba1b41b6
N		 2823 if (slot >= rdev->mddev->raid_disks &&
2824 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2825 return -ENOSPC;
2826
6c2fce2e
NB		 2827 rdev->raid_disk = slot;
2828 if (test_bit(In_sync, &rdev->flags))
2829 rdev->saved_raid_disk = slot;
2830 else
2831 rdev->saved_raid_disk = -1;
d30519fc 2832 clear_bit(In_sync, &rdev->flags);
6c2fce2e
NB		 2833 err = rdev->mddev->pers->
2834 hot_add_disk(rdev->mddev, rdev);
199050ea 2835 if (err) {
6c2fce2e 2836 rdev->raid_disk = -1;
6c2fce2e 2837 return err;
52664732 2838 } else
00bcb4ac 2839 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2840 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2841 /* failure here is OK */;
6c2fce2e 2842 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2843 } else {
ba1b41b6
N		 2844 if (slot >= rdev->mddev->raid_disks &&
2845 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2846 return -ENOSPC;
2847 rdev->raid_disk = slot;
2848 /* assume it is working */
c5d79adb
N		 2849 clear_bit(Faulty, &rdev->flags);
2850 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2851 set_bit(In_sync, &rdev->flags);
00bcb4ac 2852 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2853 }
014236d2
N		 2854 return len;
2855}
2856
2857
2858static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2859__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2860
93c8cad0 2861static ssize_t
3cb03002 2862offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2863{
6961ece4 2864 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2865}
2866
2867static ssize_t
3cb03002 2868offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2869{
c6563a8c 2870 unsigned long long offset;
b29bebd6 2871 if (kstrtoull(buf, 10, &offset) < 0)
93c8cad0 2872 return -EINVAL;
8ed0a521 2873 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2874 return -EBUSY;
dd8ac336 2875 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2876 /* Must set offset before size, so overlap checks
2877 * can be sane */
2878 return -EBUSY;
93c8cad0 2879 rdev->data_offset = offset;
25f7fd47 2880 rdev->new_data_offset = offset;
93c8cad0
N		 2881 return len;
2882}
2883
2884static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2885__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2886
c6563a8c
N		 2887static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2888{
2889 return sprintf(page, "%llu\n",
2890 (unsigned long long)rdev->new_data_offset);
2891}
2892
2893static ssize_t new_offset_store(struct md_rdev *rdev,
2894 const char *buf, size_t len)
2895{
2896 unsigned long long new_offset;
2897 struct mddev *mddev = rdev->mddev;
2898
b29bebd6 2899 if (kstrtoull(buf, 10, &new_offset) < 0)
c6563a8c
N		 2900 return -EINVAL;
2901
2902 if (mddev->sync_thread)
2903 return -EBUSY;
2904 if (new_offset == rdev->data_offset)
2905 /* reset is always permitted */
2906 ;
2907 else if (new_offset > rdev->data_offset) {
2908 /* must not push array size beyond rdev_sectors */
2909 if (new_offset - rdev->data_offset
2910 + mddev->dev_sectors > rdev->sectors)
2911 return -E2BIG;
2912 }
2913 /* Metadata worries about other space details. */
2914
2915 /* decreasing the offset is inconsistent with a backwards
2916 * reshape.
2917 */
2918 if (new_offset < rdev->data_offset &&
2919 mddev->reshape_backwards)
2920 return -EINVAL;
2921 /* Increasing offset is inconsistent with forwards
2922 * reshape. reshape_direction should be set to
2923 * 'backwards' first.
2924 */
2925 if (new_offset > rdev->data_offset &&
2926 !mddev->reshape_backwards)
2927 return -EINVAL;
2928
2929 if (mddev->pers && mddev->persistent &&
2930 !super_types[mddev->major_version]
2931 .allow_new_offset(rdev, new_offset))
2932 return -E2BIG;
2933 rdev->new_data_offset = new_offset;
2934 if (new_offset > rdev->data_offset)
2935 mddev->reshape_backwards = 1;
2936 else if (new_offset < rdev->data_offset)
2937 mddev->reshape_backwards = 0;
2938
2939 return len;
2940}
2941static struct rdev_sysfs_entry rdev_new_offset =
2942__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2943
83303b61 2944static ssize_t
3cb03002 2945rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2946{
dd8ac336 2947 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2948}
2949
c5d79adb
N		 2950static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2951{
2952 /* check if two start/length pairs overlap */
2953 if (s1+l1 <= s2)
2954 return 0;
2955 if (s2+l2 <= s1)
2956 return 0;
2957 return 1;
2958}
2959
b522adcd
DW		 2960static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2961{
2962 unsigned long long blocks;
2963 sector_t new;
2964
b29bebd6 2965 if (kstrtoull(buf, 10, &blocks) < 0)
b522adcd
DW		 2966 return -EINVAL;
2967
2968 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2969 return -EINVAL; /* sector conversion overflow */
2970
2971 new = blocks * 2;
2972 if (new != blocks * 2)
2973 return -EINVAL; /* unsigned long long to sector_t overflow */
2974
2975 *sectors = new;
2976 return 0;
2977}
2978
83303b61 2979static ssize_t
3cb03002 2980rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2981{
fd01b88c 2982 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2983 sector_t oldsectors = rdev->sectors;
b522adcd 2984 sector_t sectors;
27c529bb 2985
b522adcd 2986 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2987 return -EINVAL;
c6563a8c
N		 2988 if (rdev->data_offset != rdev->new_data_offset)
2989 return -EINVAL; /* too confusing */
0cd17fec 2990 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2991 if (my_mddev->persistent) {
dd8ac336
AN		 2992 sectors = super_types[my_mddev->major_version].
2993 rdev_size_change(rdev, sectors);
2994 if (!sectors)
0cd17fec 2995 return -EBUSY;
dd8ac336 2996 } else if (!sectors)
77304d2a 2997 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2998 rdev->data_offset;
a6468539
N		 2999 if (!my_mddev->pers->resize)
3000 /* Cannot change size for RAID0 or Linear etc */
3001 return -EINVAL;
0cd17fec 3002 }
dd8ac336 3003 if (sectors < my_mddev->dev_sectors)
7d3c6f87 3004 return -EINVAL; /* component must fit device */
0cd17fec 3005
dd8ac336
AN		 3006 rdev->sectors = sectors;
3007 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 3008 /* need to check that all other rdevs with the same ->bdev
3009 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 3010 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 3011 * we have to change it back, we will have the lock again.
3012 */
fd01b88c 3013 struct mddev *mddev;
c5d79adb 3014 int overlap = 0;
159ec1fc 3015 struct list_head *tmp;
c5d79adb 3016
27c529bb 3017 mddev_unlock(my_mddev);
29ac4aa3 3018 for_each_mddev(mddev, tmp) {
3cb03002 3019 struct md_rdev *rdev2;
c5d79adb
N		 3020
3021 mddev_lock(mddev);
dafb20fa 3022 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 3023 if (rdev->bdev == rdev2->bdev &&
3024 rdev != rdev2 &&
3025 overlaps(rdev->data_offset, rdev->sectors,
3026 rdev2->data_offset,
3027 rdev2->sectors)) {
c5d79adb
N		 3028 overlap = 1;
3029 break;
3030 }
3031 mddev_unlock(mddev);
3032 if (overlap) {
3033 mddev_put(mddev);
3034 break;
3035 }
3036 }
27c529bb 3037 mddev_lock(my_mddev);
c5d79adb
N		 3038 if (overlap) {
3039 /* Someone else could have slipped in a size
3040 * change here, but doing so is just silly.
dd8ac336 3041 * We put oldsectors back because we *know* it is
c5d79adb
N		 3042 * safe, and trust userspace not to race with
3043 * itself
3044 */
dd8ac336 3045 rdev->sectors = oldsectors;
c5d79adb
N		 3046 return -EBUSY;
3047 }
3048 }
83303b61
N		 3049 return len;
3050}
3051
3052static struct rdev_sysfs_entry rdev_size =
80ca3a44 3053__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 3054
06e3c817 3055
3cb03002 3056static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 3057{
3058 unsigned long long recovery_start = rdev->recovery_offset;
3059
3060 if (test_bit(In_sync, &rdev->flags) ||
3061 recovery_start == MaxSector)
3062 return sprintf(page, "none\n");
3063
3064 return sprintf(page, "%llu\n", recovery_start);
3065}
3066
3cb03002 3067static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 3068{
3069 unsigned long long recovery_start;
3070
3071 if (cmd_match(buf, "none"))
3072 recovery_start = MaxSector;
b29bebd6 3073 else if (kstrtoull(buf, 10, &recovery_start))
06e3c817
DW		 3074 return -EINVAL;
3075
3076 if (rdev->mddev->pers &&
3077 rdev->raid_disk >= 0)
3078 return -EBUSY;
3079
3080 rdev->recovery_offset = recovery_start;
3081 if (recovery_start == MaxSector)
3082 set_bit(In_sync, &rdev->flags);
3083 else
3084 clear_bit(In_sync, &rdev->flags);
3085 return len;
3086}
3087
3088static struct rdev_sysfs_entry rdev_recovery_start =
3089__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3090
16c791a5
N		 3091
3092static ssize_t
3093badblocks_show(struct badblocks *bb, char *page, int unack);
3094static ssize_t
3095badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3096
3cb03002 3097static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3098{
3099 return badblocks_show(&rdev->badblocks, page, 0);
3100}
3cb03002 3101static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 3102{
de393cde
N		 3103 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3104 /* Maybe that ack was all we needed */
3105 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3106 wake_up(&rdev->blocked_wait);
3107 return rv;
16c791a5
N		 3108}
3109static struct rdev_sysfs_entry rdev_bad_blocks =
3110__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3111
3112
3cb03002 3113static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3114{
3115 return badblocks_show(&rdev->badblocks, page, 1);
3116}
3cb03002 3117static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 3118{
3119 return badblocks_store(&rdev->badblocks, page, len, 1);
3120}
3121static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3122__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3123
86e6ffdd
N		 3124static struct attribute *rdev_default_attrs[] = {
3125 &rdev_state.attr,
4dbcdc75 3126 &rdev_errors.attr,
014236d2 3127 &rdev_slot.attr,
93c8cad0 3128 &rdev_offset.attr,
c6563a8c 3129 &rdev_new_offset.attr,
83303b61 3130 &rdev_size.attr,
06e3c817 3131 &rdev_recovery_start.attr,
16c791a5
N		 3132 &rdev_bad_blocks.attr,
3133 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 3134 NULL,
3135};
3136static ssize_t
3137rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3138{
3139 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3140 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
fd01b88c 3141 struct mddev *mddev = rdev->mddev;
27c529bb 3142 ssize_t rv;
86e6ffdd
N		 3143
3144 if (!entry->show)
3145 return -EIO;
27c529bb
N		 3146
3147 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3148 if (!rv) {
3149 if (rdev->mddev == NULL)
3150 rv = -EBUSY;
3151 else
3152 rv = entry->show(rdev, page);
3153 mddev_unlock(mddev);
3154 }
3155 return rv;
86e6ffdd
N		 3156}
3157
3158static ssize_t
3159rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3160 const char *page, size_t length)
3161{
3162 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3163 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3164 ssize_t rv;
fd01b88c 3165 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3166
3167 if (!entry->store)
3168 return -EIO;
67463acb
N		 3169 if (!capable(CAP_SYS_ADMIN))
3170 return -EACCES;
27c529bb 3171 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3172 if (!rv) {
27c529bb
N		 3173 if (rdev->mddev == NULL)
3174 rv = -EBUSY;
3175 else
3176 rv = entry->store(rdev, page, length);
6a51830e 3177 mddev_unlock(mddev);
ca388059
N		 3178 }
3179 return rv;
86e6ffdd
N		 3180}
3181
3182static void rdev_free(struct kobject *ko)
3183{
3cb03002 3184 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3185 kfree(rdev);
3186}
52cf25d0 3187static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3188 .show = rdev_attr_show,
3189 .store = rdev_attr_store,
3190};
3191static struct kobj_type rdev_ktype = {
3192 .release = rdev_free,
3193 .sysfs_ops = &rdev_sysfs_ops,
3194 .default_attrs = rdev_default_attrs,
3195};
3196
3cb03002 3197int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3198{
3199 rdev->desc_nr = -1;
3200 rdev->saved_raid_disk = -1;
3201 rdev->raid_disk = -1;
3202 rdev->flags = 0;
3203 rdev->data_offset = 0;
c6563a8c 3204 rdev->new_data_offset = 0;
e8bb9a83
N		 3205 rdev->sb_events = 0;
3206 rdev->last_read_error.tv_sec = 0;
3207 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3208 rdev->sb_loaded = 0;
3209 rdev->bb_page = NULL;
e8bb9a83
N		 3210 atomic_set(&rdev->nr_pending, 0);
3211 atomic_set(&rdev->read_errors, 0);
3212 atomic_set(&rdev->corrected_errors, 0);
3213
3214 INIT_LIST_HEAD(&rdev->same_set);
3215 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3216
3217 /* Add space to store bad block list.
3218 * This reserves the space even on arrays where it cannot
3219 * be used - I wonder if that matters
3220 */
3221 rdev->badblocks.count = 0;
486adf72 3222 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
2230dfe4
N		 3223 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3224 seqlock_init(&rdev->badblocks.lock);
3225 if (rdev->badblocks.page == NULL)
3226 return -ENOMEM;
3227
3228 return 0;
e8bb9a83
N		 3229}
3230EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3231/*
3232 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3233 *
3234 * mark the device faulty if:
3235 *
3236 * - the device is nonexistent (zero size)
3237 * - the device has no valid superblock
3238 *
3239 * a faulty rdev _never_ has rdev->sb set.
3240 */
3cb03002 3241static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3242{
3243 char b[BDEVNAME_SIZE];
3244 int err;
3cb03002 3245 struct md_rdev *rdev;
1da177e4
LT		 3246 sector_t size;
3247
9ffae0cf 3248 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3249 if (!rdev) {
3250 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3251 return ERR_PTR(-ENOMEM);
3252 }
1da177e4 3253
2230dfe4
N		 3254 err = md_rdev_init(rdev);
3255 if (err)
3256 goto abort_free;
3257 err = alloc_disk_sb(rdev);
3258 if (err)
1da177e4
LT		 3259 goto abort_free;
3260
c5d79adb 3261 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3262 if (err)
3263 goto abort_free;
3264
f9cb074b 3265 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3266
77304d2a 3267 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT		 3268 if (!size) {
3269 printk(KERN_WARNING
3270 "md: %s has zero or unknown size, marking faulty!\n",
3271 bdevname(rdev->bdev,b));
3272 err = -EINVAL;
3273 goto abort_free;
3274 }
3275
3276 if (super_format >= 0) {
3277 err = super_types[super_format].
3278 load_super(rdev, NULL, super_minor);
3279 if (err == -EINVAL) {
df968c4e
N		 3280 printk(KERN_WARNING
3281 "md: %s does not have a valid v%d.%d "
3282 "superblock, not importing!\n",
3283 bdevname(rdev->bdev,b),
3284 super_format, super_minor);
1da177e4
LT		 3285 goto abort_free;
3286 }
3287 if (err < 0) {
3288 printk(KERN_WARNING
3289 "md: could not read %s's sb, not importing!\n",
3290 bdevname(rdev->bdev,b));
3291 goto abort_free;
3292 }
3293 }
6bfe0b49 3294
1da177e4
LT		 3295 return rdev;
3296
3297abort_free:
2699b672
N		 3298 if (rdev->bdev)
3299 unlock_rdev(rdev);
545c8795 3300 md_rdev_clear(rdev);
1da177e4
LT		 3301 kfree(rdev);
3302 return ERR_PTR(err);
3303}
3304
3305/*
3306 * Check a full RAID array for plausibility
3307 */
3308
3309
fd01b88c 3310static void analyze_sbs(struct mddev * mddev)
1da177e4
LT		 3311{
3312 int i;
3cb03002 3313 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3314 char b[BDEVNAME_SIZE];
3315
3316 freshest = NULL;
dafb20fa 3317 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3318 switch (super_types[mddev->major_version].
3319 load_super(rdev, freshest, mddev->minor_version)) {
3320 case 1:
3321 freshest = rdev;
3322 break;
3323 case 0:
3324 break;
3325 default:
3326 printk( KERN_ERR \
3327 "md: fatal superblock inconsistency in %s"
3328 " -- removing from array\n",
3329 bdevname(rdev->bdev,b));
3330 kick_rdev_from_array(rdev);
3331 }
3332
3333
3334 super_types[mddev->major_version].
3335 validate_super(mddev, freshest);
3336
3337 i = 0;
dafb20fa 3338 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3339 if (mddev->max_disks &&
3340 (rdev->desc_nr >= mddev->max_disks ||
3341 i > mddev->max_disks)) {
de01dfad
N		 3342 printk(KERN_WARNING
3343 "md: %s: %s: only %d devices permitted\n",
3344 mdname(mddev), bdevname(rdev->bdev, b),
3345 mddev->max_disks);
3346 kick_rdev_from_array(rdev);
3347 continue;
3348 }
1da177e4
LT		 3349 if (rdev != freshest)
3350 if (super_types[mddev->major_version].
3351 validate_super(mddev, rdev)) {
3352 printk(KERN_WARNING "md: kicking non-fresh %s"
3353 " from array!\n",
3354 bdevname(rdev->bdev,b));
3355 kick_rdev_from_array(rdev);
3356 continue;
3357 }
3358 if (mddev->level == LEVEL_MULTIPATH) {
3359 rdev->desc_nr = i++;
3360 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3361 set_bit(In_sync, &rdev->flags);
5e5e3e78 3362 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3363 rdev->raid_disk = -1;
3364 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3365 }
3366 }
1da177e4
LT		 3367}
3368
72e02075
N		 3369/* Read a fixed-point number.
3370 * Numbers in sysfs attributes should be in "standard" units where
3371 * possible, so time should be in seconds.
3372 * However we internally use a a much smaller unit such as
3373 * milliseconds or jiffies.
3374 * This function takes a decimal number with a possible fractional
3375 * component, and produces an integer which is the result of
3376 * multiplying that number by 10^'scale'.
3377 * all without any floating-point arithmetic.
3378 */
3379int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3380{
3381 unsigned long result = 0;
3382 long decimals = -1;
3383 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3384 if (*cp == '.')
3385 decimals = 0;
3386 else if (decimals < scale) {
3387 unsigned int value;
3388 value = *cp - '0';
3389 result = result * 10 + value;
3390 if (decimals >= 0)
3391 decimals++;
3392 }
3393 cp++;
3394 }
3395 if (*cp == '\n')
3396 cp++;
3397 if (*cp)
3398 return -EINVAL;
3399 if (decimals < 0)
3400 decimals = 0;
3401 while (decimals < scale) {
3402 result *= 10;
3403 decimals ++;
3404 }
3405 *res = result;
3406 return 0;
3407}
3408
3409
19052c0e
N		 3410static void md_safemode_timeout(unsigned long data);
3411
16f17b39 3412static ssize_t
fd01b88c 3413safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3414{
3415 int msec = (mddev->safemode_delay*1000)/HZ;
3416 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3417}
3418static ssize_t
fd01b88c 3419safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3420{
16f17b39 3421 unsigned long msec;
97ce0a7f 3422
72e02075 3423 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3424 return -EINVAL;
16f17b39
N		 3425 if (msec == 0)
3426 mddev->safemode_delay = 0;
3427 else {
19052c0e 3428 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 3429 mddev->safemode_delay = (msec*HZ)/1000;
3430 if (mddev->safemode_delay == 0)
3431 mddev->safemode_delay = 1;
275c51c4 3432 if (mddev->safemode_delay < old_delay || old_delay == 0)
19052c0e 3433 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 3434 }
3435 return len;
3436}
3437static struct md_sysfs_entry md_safe_delay =
80ca3a44 3438__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3439
eae1701f 3440static ssize_t
fd01b88c 3441level_show(struct mddev *mddev, char *page)
eae1701f 3442{
84fc4b56 3443 struct md_personality *p = mddev->pers;
d9d166c2 3444 if (p)
eae1701f 3445 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 3446 else if (mddev->clevel[0])
3447 return sprintf(page, "%s\n", mddev->clevel);
3448 else if (mddev->level != LEVEL_NONE)
3449 return sprintf(page, "%d\n", mddev->level);
3450 else
3451 return 0;
eae1701f
N		 3452}
3453
d9d166c2 3454static ssize_t
fd01b88c 3455level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3456{
f2859af6 3457 char clevel[16];
20a49ff6 3458 ssize_t rv = len;
84fc4b56 3459 struct md_personality *pers;
f2859af6 3460 long level;
245f46c2 3461 void *priv;
3cb03002 3462 struct md_rdev *rdev;
245f46c2
N		 3463
3464 if (mddev->pers == NULL) {
3465 if (len == 0)
3466 return 0;
3467 if (len >= sizeof(mddev->clevel))
3468 return -ENOSPC;
3469 strncpy(mddev->clevel, buf, len);
3470 if (mddev->clevel[len-1] == '\n')
3471 len--;
3472 mddev->clevel[len] = 0;
3473 mddev->level = LEVEL_NONE;
3474 return rv;
3475 }
3476
3477 /* request to change the personality. Need to ensure:
3478 * - array is not engaged in resync/recovery/reshape
3479 * - old personality can be suspended
3480 * - new personality will access other array.
3481 */
3482
bb4f1e9d
N		 3483 if (mddev->sync_thread ||
3484 mddev->reshape_position != MaxSector ||
3485 mddev->sysfs_active)
d9d166c2 3486 return -EBUSY;
245f46c2
N		 3487
3488 if (!mddev->pers->quiesce) {
3489 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3490 mdname(mddev), mddev->pers->name);
3491 return -EINVAL;
3492 }
3493
3494 /* Now find the new personality */
f2859af6 3495 if (len == 0 || len >= sizeof(clevel))
245f46c2 3496 return -EINVAL;
f2859af6
DW		 3497 strncpy(clevel, buf, len);
3498 if (clevel[len-1] == '\n')
d9d166c2 3499 len--;
f2859af6 3500 clevel[len] = 0;
b29bebd6 3501 if (kstrtol(clevel, 10, &level))
f2859af6 3502 level = LEVEL_NONE;
245f46c2 3503
f2859af6
DW		 3504 if (request_module("md-%s", clevel) != 0)
3505 request_module("md-level-%s", clevel);
245f46c2 3506 spin_lock(&pers_lock);
f2859af6 3507 pers = find_pers(level, clevel);
245f46c2
N		 3508 if (!pers || !try_module_get(pers->owner)) {
3509 spin_unlock(&pers_lock);
f2859af6 3510 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3511 return -EINVAL;
3512 }
3513 spin_unlock(&pers_lock);
3514
3515 if (pers == mddev->pers) {
3516 /* Nothing to do! */
3517 module_put(pers->owner);
3518 return rv;
3519 }
3520 if (!pers->takeover) {
3521 module_put(pers->owner);
3522 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3523 mdname(mddev), clevel);
245f46c2
N		 3524 return -EINVAL;
3525 }
3526
dafb20fa 3527 rdev_for_each(rdev, mddev)
e93f68a1
N		 3528 rdev->new_raid_disk = rdev->raid_disk;
3529
245f46c2
N		 3530 /* ->takeover must set new_* and/or delta_disks
3531 * if it succeeds, and may set them when it fails.
3532 */
3533 priv = pers->takeover(mddev);
3534 if (IS_ERR(priv)) {
3535 mddev->new_level = mddev->level;
3536 mddev->new_layout = mddev->layout;
664e7c41 3537 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3538 mddev->raid_disks -= mddev->delta_disks;
3539 mddev->delta_disks = 0;
2c810cdd 3540 mddev->reshape_backwards = 0;
245f46c2
N		 3541 module_put(pers->owner);
3542 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3543 mdname(mddev), clevel);
245f46c2
N		 3544 return PTR_ERR(priv);
3545 }
3546
3547 /* Looks like we have a winner */
3548 mddev_suspend(mddev);
3549 mddev->pers->stop(mddev);
a64c876f
N		 3550
3551 if (mddev->pers->sync_request == NULL &&
3552 pers->sync_request != NULL) {
3553 /* need to add the md_redundancy_group */
3554 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3555 printk(KERN_WARNING
3556 "md: cannot register extra attributes for %s\n",
3557 mdname(mddev));
19fdb9ee 3558 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3559 }
3560 if (mddev->pers->sync_request != NULL &&
3561 pers->sync_request == NULL) {
3562 /* need to remove the md_redundancy_group */
3563 if (mddev->to_remove == NULL)
3564 mddev->to_remove = &md_redundancy_group;
3565 }
3566
54071b38
TM		 3567 if (mddev->pers->sync_request == NULL &&
3568 mddev->external) {
3569 /* We are converting from a no-redundancy array
3570 * to a redundancy array and metadata is managed
3571 * externally so we need to be sure that writes
3572 * won't block due to a need to transition
3573 * clean->dirty
3574 * until external management is started.
3575 */
3576 mddev->in_sync = 0;
3577 mddev->safemode_delay = 0;
3578 mddev->safemode = 0;
3579 }
3580
dafb20fa 3581 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3582 if (rdev->raid_disk < 0)
3583 continue;
bf2cb0da 3584 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3585 rdev->new_raid_disk = -1;
3586 if (rdev->new_raid_disk == rdev->raid_disk)
3587 continue;
36fad858 3588 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3589 }
dafb20fa 3590 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3591 if (rdev->raid_disk < 0)
3592 continue;
3593 if (rdev->new_raid_disk == rdev->raid_disk)
3594 continue;
3595 rdev->raid_disk = rdev->new_raid_disk;
3596 if (rdev->raid_disk < 0)
3a981b03 3597 clear_bit(In_sync, &rdev->flags);
e93f68a1 3598 else {
36fad858
NK		 3599 if (sysfs_link_rdev(mddev, rdev))
3600 printk(KERN_WARNING "md: cannot register rd%d"
3601 " for %s after level change\n",
3602 rdev->raid_disk, mdname(mddev));
3a981b03 3603 }
e93f68a1
N		 3604 }
3605
3606 module_put(mddev->pers->owner);
245f46c2
N		 3607 mddev->pers = pers;
3608 mddev->private = priv;
3609 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3610 mddev->level = mddev->new_level;
3611 mddev->layout = mddev->new_layout;
664e7c41 3612 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3613 mddev->delta_disks = 0;
2c810cdd 3614 mddev->reshape_backwards = 0;
fee68723 3615 mddev->degraded = 0;
9af204cf
TM		 3616 if (mddev->pers->sync_request == NULL) {
3617 /* this is now an array without redundancy, so
3618 * it must always be in_sync
3619 */
3620 mddev->in_sync = 1;
3621 del_timer_sync(&mddev->safemode_timer);
3622 }
245f46c2 3623 pers->run(mddev);
245f46c2 3624 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3625 mddev_resume(mddev);
5cac7861 3626 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3627 md_new_event(mddev);
d9d166c2
N		 3628 return rv;
3629}
3630
3631static struct md_sysfs_entry md_level =
80ca3a44 3632__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3633
d4dbd025
N		 3634
3635static ssize_t
fd01b88c 3636layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3637{
3638 /* just a number, not meaningful for all levels */
08a02ecd
N		 3639 if (mddev->reshape_position != MaxSector &&
3640 mddev->layout != mddev->new_layout)
3641 return sprintf(page, "%d (%d)\n",
3642 mddev->new_layout, mddev->layout);
d4dbd025
N		 3643 return sprintf(page, "%d\n", mddev->layout);
3644}
3645
3646static ssize_t
fd01b88c 3647layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N		 3648{
3649 char *e;
3650 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3651
3652 if (!*buf || (*e && *e != '\n'))
3653 return -EINVAL;
3654
b3546035
N		 3655 if (mddev->pers) {
3656 int err;
50ac168a 3657 if (mddev->pers->check_reshape == NULL)
b3546035 3658 return -EBUSY;
597a711b 3659 mddev->new_layout = n;
50ac168a 3660 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3661 if (err) {
3662 mddev->new_layout = mddev->layout;
b3546035 3663 return err;
597a711b 3664 }
b3546035 3665 } else {
08a02ecd 3666 mddev->new_layout = n;
b3546035
N		 3667 if (mddev->reshape_position == MaxSector)
3668 mddev->layout = n;
3669 }
d4dbd025
N		 3670 return len;
3671}
3672static struct md_sysfs_entry md_layout =
80ca3a44 3673__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3674
3675
eae1701f 3676static ssize_t
fd01b88c 3677raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3678{
bb636547
N		 3679 if (mddev->raid_disks == 0)
3680 return 0;
08a02ecd
N		 3681 if (mddev->reshape_position != MaxSector &&
3682 mddev->delta_disks != 0)
3683 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3684 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3685 return sprintf(page, "%d\n", mddev->raid_disks);
3686}
3687
fd01b88c 3688static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3689
3690static ssize_t
fd01b88c 3691raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3692{
da943b99
N		 3693 char *e;
3694 int rv = 0;
3695 unsigned long n = simple_strtoul(buf, &e, 10);
3696
3697 if (!*buf || (*e && *e != '\n'))
3698 return -EINVAL;
3699
3700 if (mddev->pers)
3701 rv = update_raid_disks(mddev, n);
08a02ecd 3702 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3703 struct md_rdev *rdev;
08a02ecd 3704 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c
N		 3705
3706 rdev_for_each(rdev, mddev) {
3707 if (olddisks < n &&
3708 rdev->data_offset < rdev->new_data_offset)
3709 return -EINVAL;
3710 if (olddisks > n &&
3711 rdev->data_offset > rdev->new_data_offset)
3712 return -EINVAL;
3713 }
08a02ecd
N		 3714 mddev->delta_disks = n - olddisks;
3715 mddev->raid_disks = n;
2c810cdd 3716 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3717 } else
da943b99
N		 3718 mddev->raid_disks = n;
3719 return rv ? rv : len;
3720}
3721static struct md_sysfs_entry md_raid_disks =
80ca3a44 3722__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3723
3b34380a 3724static ssize_t
fd01b88c 3725chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3726{
08a02ecd 3727 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3728 mddev->chunk_sectors != mddev->new_chunk_sectors)
3729 return sprintf(page, "%d (%d)\n",
3730 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3731 mddev->chunk_sectors << 9);
3732 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3733}
3734
3735static ssize_t
fd01b88c 3736chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3737{
3b34380a
N		 3738 char *e;
3739 unsigned long n = simple_strtoul(buf, &e, 10);
3740
3b34380a
N		 3741 if (!*buf || (*e && *e != '\n'))
3742 return -EINVAL;
3743
b3546035
N		 3744 if (mddev->pers) {
3745 int err;
50ac168a 3746 if (mddev->pers->check_reshape == NULL)
b3546035 3747 return -EBUSY;
597a711b 3748 mddev->new_chunk_sectors = n >> 9;
50ac168a 3749 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3750 if (err) {
3751 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3752 return err;
597a711b 3753 }
b3546035 3754 } else {
664e7c41 3755 mddev->new_chunk_sectors = n >> 9;
b3546035 3756 if (mddev->reshape_position == MaxSector)
9d8f0363 3757 mddev->chunk_sectors = n >> 9;
b3546035 3758 }
3b34380a
N		 3759 return len;
3760}
3761static struct md_sysfs_entry md_chunk_size =
80ca3a44 3762__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3763
a94213b1 3764static ssize_t
fd01b88c 3765resync_start_show(struct mddev *mddev, char *page)
a94213b1 3766{
d1a7c503
N		 3767 if (mddev->recovery_cp == MaxSector)
3768 return sprintf(page, "none\n");
a94213b1
N		 3769 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3770}
3771
3772static ssize_t
fd01b88c 3773resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3774{
a94213b1
N		 3775 char *e;
3776 unsigned long long n = simple_strtoull(buf, &e, 10);
3777
b098636c 3778 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
a94213b1 3779 return -EBUSY;
06e3c817
DW		 3780 if (cmd_match(buf, "none"))
3781 n = MaxSector;
3782 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3783 return -EINVAL;
3784
3785 mddev->recovery_cp = n;
db07d85e
N		 3786 if (mddev->pers)
3787 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
a94213b1
N		 3788 return len;
3789}
3790static struct md_sysfs_entry md_resync_start =
80ca3a44 3791__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3792
9e653b63
N		 3793/*
3794 * The array state can be:
3795 *
3796 * clear
3797 * No devices, no size, no level
3798 * Equivalent to STOP_ARRAY ioctl
3799 * inactive
3800 * May have some settings, but array is not active
3801 * all IO results in error
3802 * When written, doesn't tear down array, but just stops it
3803 * suspended (not supported yet)
3804 * All IO requests will block. The array can be reconfigured.
910d8cb3 3805 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3806 * readonly
3807 * no resync can happen. no superblocks get written.
3808 * write requests fail
3809 * read-auto
3810 * like readonly, but behaves like 'clean' on a write request.
3811 *
3812 * clean - no pending writes, but otherwise active.
3813 * When written to inactive array, starts without resync
3814 * If a write request arrives then
3815 * if metadata is known, mark 'dirty' and switch to 'active'.
3816 * if not known, block and switch to write-pending
3817 * If written to an active array that has pending writes, then fails.
3818 * active
3819 * fully active: IO and resync can be happening.
3820 * When written to inactive array, starts with resync
3821 *
3822 * write-pending
3823 * clean, but writes are blocked waiting for 'active' to be written.
3824 *
3825 * active-idle
3826 * like active, but no writes have been seen for a while (100msec).
3827 *
3828 */
3829enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3830 write_pending, active_idle, bad_word};
05381954 3831static char *array_states[] = {
9e653b63
N		 3832 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3833 "write-pending", "active-idle", NULL };
3834
3835static int match_word(const char *word, char **list)
3836{
3837 int n;
3838 for (n=0; list[n]; n++)
3839 if (cmd_match(word, list[n]))
3840 break;
3841 return n;
3842}
3843
3844static ssize_t
fd01b88c 3845array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3846{
3847 enum array_state st = inactive;
3848
3849 if (mddev->pers)
3850 switch(mddev->ro) {
3851 case 1:
3852 st = readonly;
3853 break;
3854 case 2:
3855 st = read_auto;
3856 break;
3857 case 0:
3858 if (mddev->in_sync)
3859 st = clean;
070dc6dd 3860 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3861 st = write_pending;
9e653b63
N		 3862 else if (mddev->safemode)
3863 st = active_idle;
3864 else
3865 st = active;
3866 }
3867 else {
3868 if (list_empty(&mddev->disks) &&
3869 mddev->raid_disks == 0 &&
58c0fed4 3870 mddev->dev_sectors == 0)
9e653b63
N		 3871 st = clear;
3872 else
3873 st = inactive;
3874 }
3875 return sprintf(page, "%s\n", array_states[st]);
3876}
3877
a05b7ea0
N		 3878static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3879static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
fd01b88c
N		 3880static int do_md_run(struct mddev * mddev);
3881static int restart_array(struct mddev *mddev);
9e653b63
N		 3882
3883static ssize_t
fd01b88c 3884array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63
N		 3885{
3886 int err = -EINVAL;
3887 enum array_state st = match_word(buf, array_states);
3888 switch(st) {
3889 case bad_word:
3890 break;
3891 case clear:
3892 /* stopping an active array */
a05b7ea0 3893 err = do_md_stop(mddev, 0, NULL);
9e653b63
N		 3894 break;
3895 case inactive:
3896 /* stopping an active array */
90cf195d 3897 if (mddev->pers)
a05b7ea0 3898 err = do_md_stop(mddev, 2, NULL);
90cf195d 3899 else
e691063a 3900 err = 0; /* already inactive */
9e653b63
N		 3901 break;
3902 case suspended:
3903 break; /* not supported yet */
3904 case readonly:
3905 if (mddev->pers)
a05b7ea0 3906 err = md_set_readonly(mddev, NULL);
9e653b63
N		 3907 else {
3908 mddev->ro = 1;
648b629e 3909 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3910 err = do_md_run(mddev);
3911 }
3912 break;
3913 case read_auto:
9e653b63 3914 if (mddev->pers) {
80268ee9 3915 if (mddev->ro == 0)
a05b7ea0 3916 err = md_set_readonly(mddev, NULL);
80268ee9 3917 else if (mddev->ro == 1)
648b629e
N		 3918 err = restart_array(mddev);
3919 if (err == 0) {
3920 mddev->ro = 2;
3921 set_disk_ro(mddev->gendisk, 0);
3922 }
9e653b63
N		 3923 } else {
3924 mddev->ro = 2;
3925 err = do_md_run(mddev);
3926 }
3927 break;
3928 case clean:
3929 if (mddev->pers) {
3930 restart_array(mddev);
3931 spin_lock_irq(&mddev->write_lock);
3932 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3933 if (mddev->in_sync == 0) {
3934 mddev->in_sync = 1;
31a59e34
N		 3935 if (mddev->safemode == 1)
3936 mddev->safemode = 0;
070dc6dd 3937 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3938 }
3939 err = 0;
3940 } else
3941 err = -EBUSY;
9e653b63 3942 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3943 } else
3944 err = -EINVAL;
9e653b63
N		 3945 break;
3946 case active:
3947 if (mddev->pers) {
3948 restart_array(mddev);
070dc6dd 3949 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3950 wake_up(&mddev->sb_wait);
3951 err = 0;
3952 } else {
3953 mddev->ro = 0;
648b629e 3954 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3955 err = do_md_run(mddev);
3956 }
3957 break;
3958 case write_pending:
3959 case active_idle:
3960 /* these cannot be set */
3961 break;
3962 }
3963 if (err)
3964 return err;
0fd62b86 3965 else {
1d23f178
N		 3966 if (mddev->hold_active == UNTIL_IOCTL)
3967 mddev->hold_active = 0;
00bcb4ac 3968 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3969 return len;
0fd62b86 3970 }
9e653b63 3971}
80ca3a44
N		 3972static struct md_sysfs_entry md_array_state =
3973__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3974
1e50915f 3975static ssize_t
fd01b88c 3976max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3977 return sprintf(page, "%d\n",
3978 atomic_read(&mddev->max_corr_read_errors));
3979}
3980
3981static ssize_t
fd01b88c 3982max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB		 3983{
3984 char *e;
3985 unsigned long n = simple_strtoul(buf, &e, 10);
3986
3987 if (*buf && (*e == 0 || *e == '\n')) {
3988 atomic_set(&mddev->max_corr_read_errors, n);
3989 return len;
3990 }
3991 return -EINVAL;
3992}
3993
3994static struct md_sysfs_entry max_corr_read_errors =
3995__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3996 max_corrected_read_errors_store);
3997
6d7ff738 3998static ssize_t
fd01b88c 3999null_show(struct mddev *mddev, char *page)
6d7ff738
N		 4000{
4001 return -EINVAL;
4002}
4003
4004static ssize_t
fd01b88c 4005new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 4006{
4007 /* buf must be %d:%d\n? giving major and minor numbers */
4008 /* The new device is added to the array.
4009 * If the array has a persistent superblock, we read the
4010 * superblock to initialise info and check validity.
4011 * Otherwise, only checking done is that in bind_rdev_to_array,
4012 * which mainly checks size.
4013 */
4014 char *e;
4015 int major = simple_strtoul(buf, &e, 10);
4016 int minor;
4017 dev_t dev;
3cb03002 4018 struct md_rdev *rdev;
6d7ff738
N		 4019 int err;
4020
4021 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4022 return -EINVAL;
4023 minor = simple_strtoul(e+1, &e, 10);
4024 if (*e && *e != '\n')
4025 return -EINVAL;
4026 dev = MKDEV(major, minor);
4027 if (major != MAJOR(dev) ||
4028 minor != MINOR(dev))
4029 return -EOVERFLOW;
4030
4031
4032 if (mddev->persistent) {
4033 rdev = md_import_device(dev, mddev->major_version,
4034 mddev->minor_version);
4035 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 4036 struct md_rdev *rdev0
4037 = list_entry(mddev->disks.next,
4038 struct md_rdev, same_set);
6d7ff738
N		 4039 err = super_types[mddev->major_version]
4040 .load_super(rdev, rdev0, mddev->minor_version);
4041 if (err < 0)
4042 goto out;
4043 }
c5d79adb
N		 4044 } else if (mddev->external)
4045 rdev = md_import_device(dev, -2, -1);
4046 else
6d7ff738
N		 4047 rdev = md_import_device(dev, -1, -1);
4048
4049 if (IS_ERR(rdev))
4050 return PTR_ERR(rdev);
4051 err = bind_rdev_to_array(rdev, mddev);
4052 out:
4053 if (err)
4054 export_rdev(rdev);
4055 return err ? err : len;
4056}
4057
4058static struct md_sysfs_entry md_new_device =
80ca3a44 4059__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 4060
9b1d1dac 4061static ssize_t
fd01b88c 4062bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 4063{
4064 char *end;
4065 unsigned long chunk, end_chunk;
4066
4067 if (!mddev->bitmap)
4068 goto out;
4069 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4070 while (*buf) {
4071 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4072 if (buf == end) break;
4073 if (*end == '-') { /* range */
4074 buf = end + 1;
4075 end_chunk = simple_strtoul(buf, &end, 0);
4076 if (buf == end) break;
4077 }
4078 if (*end && !isspace(*end)) break;
4079 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 4080 buf = skip_spaces(end);
9b1d1dac
PC		 4081 }
4082 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4083out:
4084 return len;
4085}
4086
4087static struct md_sysfs_entry md_bitmap =
4088__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4089
a35b0d69 4090static ssize_t
fd01b88c 4091size_show(struct mddev *mddev, char *page)
a35b0d69 4092{
58c0fed4
AN		 4093 return sprintf(page, "%llu\n",
4094 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 4095}
4096
fd01b88c 4097static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 4098
4099static ssize_t
fd01b88c 4100size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 4101{
4102 /* If array is inactive, we can reduce the component size, but
4103 * not increase it (except from 0).
4104 * If array is active, we can try an on-line resize
4105 */
b522adcd
DW		 4106 sector_t sectors;
4107 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4108
58c0fed4
AN		 4109 if (err < 0)
4110 return err;
a35b0d69 4111 if (mddev->pers) {
58c0fed4 4112 err = update_size(mddev, sectors);
850b2b42 4113 md_update_sb(mddev, 1);
a35b0d69 4114 } else {
58c0fed4
AN		 4115 if (mddev->dev_sectors == 0 ||
4116 mddev->dev_sectors > sectors)
4117 mddev->dev_sectors = sectors;
a35b0d69
N		 4118 else
4119 err = -ENOSPC;
4120 }
4121 return err ? err : len;
4122}
4123
4124static struct md_sysfs_entry md_size =
80ca3a44 4125__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4126
8bb93aac 4127
83f0d77a 4128/* Metadata version.
e691063a
N		 4129 * This is one of
4130 * 'none' for arrays with no metadata (good luck...)
4131 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4132 * or N.M for internally known formats
4133 */
4134static ssize_t
fd01b88c 4135metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4136{
4137 if (mddev->persistent)
4138 return sprintf(page, "%d.%d\n",
4139 mddev->major_version, mddev->minor_version);
e691063a
N		 4140 else if (mddev->external)
4141 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4142 else
4143 return sprintf(page, "none\n");
4144}
4145
4146static ssize_t
fd01b88c 4147metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4148{
4149 int major, minor;
4150 char *e;
ea43ddd8
N		 4151 /* Changing the details of 'external' metadata is
4152 * always permitted. Otherwise there must be
4153 * no devices attached to the array.
4154 */
4155 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4156 ;
4157 else if (!list_empty(&mddev->disks))
8bb93aac
N		 4158 return -EBUSY;
4159
4160 if (cmd_match(buf, "none")) {
4161 mddev->persistent = 0;
e691063a
N		 4162 mddev->external = 0;
4163 mddev->major_version = 0;
4164 mddev->minor_version = 90;
4165 return len;
4166 }
4167 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4168 size_t namelen = len-9;
e691063a
N		 4169 if (namelen >= sizeof(mddev->metadata_type))
4170 namelen = sizeof(mddev->metadata_type)-1;
4171 strncpy(mddev->metadata_type, buf+9, namelen);
4172 mddev->metadata_type[namelen] = 0;
4173 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4174 mddev->metadata_type[--namelen] = 0;
4175 mddev->persistent = 0;
4176 mddev->external = 1;
8bb93aac
N		 4177 mddev->major_version = 0;
4178 mddev->minor_version = 90;
4179 return len;
4180 }
4181 major = simple_strtoul(buf, &e, 10);
4182 if (e==buf || *e != '.')
4183 return -EINVAL;
4184 buf = e+1;
4185 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4186 if (e==buf || (*e && *e != '\n') )
8bb93aac 4187 return -EINVAL;
50511da3 4188 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 4189 return -ENOENT;
4190 mddev->major_version = major;
4191 mddev->minor_version = minor;
4192 mddev->persistent = 1;
e691063a 4193 mddev->external = 0;
8bb93aac
N		 4194 return len;
4195}
4196
4197static struct md_sysfs_entry md_metadata =
80ca3a44 4198__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4199
24dd469d 4200static ssize_t
fd01b88c 4201action_show(struct mddev *mddev, char *page)
24dd469d 4202{
7eec314d 4203 char *type = "idle";
b6a9ce68
N		 4204 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4205 type = "frozen";
4206 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 4207 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 4208 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4209 type = "reshape";
4210 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 4211 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4212 type = "resync";
4213 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4214 type = "check";
4215 else
4216 type = "repair";
72a23c21 4217 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 4218 type = "recover";
4219 }
4220 return sprintf(page, "%s\n", type);
4221}
4222
4223static ssize_t
fd01b88c 4224action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4225{
7eec314d
N		 4226 if (!mddev->pers || !mddev->pers->sync_request)
4227 return -EINVAL;
4228
b6a9ce68
N		 4229 if (cmd_match(page, "frozen"))
4230 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4231 else
4232 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4233
4234 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 4235 if (mddev->sync_thread) {
4236 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 4237 md_reap_sync_thread(mddev);
7eec314d 4238 }
03c902e1
N		 4239 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4240 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4241 return -EBUSY;
72a23c21
NB		 4242 else if (cmd_match(page, "resync"))
4243 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4244 else if (cmd_match(page, "recover")) {
4245 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4246 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4247 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4248 int err;
4249 if (mddev->pers->start_reshape == NULL)
4250 return -EINVAL;
4251 err = mddev->pers->start_reshape(mddev);
4252 if (err)
4253 return err;
a99ac971 4254 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4255 } else {
bce74dac 4256 if (cmd_match(page, "check"))
7eec314d 4257 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4258 else if (!cmd_match(page, "repair"))
7eec314d
N		 4259 return -EINVAL;
4260 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4261 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4262 }
48c26ddc
N		 4263 if (mddev->ro == 2) {
4264 /* A write to sync_action is enough to justify
4265 * canceling read-auto mode
4266 */
4267 mddev->ro = 0;
4268 md_wakeup_thread(mddev->sync_thread);
4269 }
03c902e1 4270 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4271 md_wakeup_thread(mddev->thread);
00bcb4ac 4272 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4273 return len;
4274}
4275
c4a39551
JB		 4276static struct md_sysfs_entry md_scan_mode =
4277__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4278
4279static ssize_t
4280last_sync_action_show(struct mddev *mddev, char *page)
4281{
4282 return sprintf(page, "%s\n", mddev->last_sync_action);
4283}
4284
4285static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4286
9d88883e 4287static ssize_t
fd01b88c 4288mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4289{
4290 return sprintf(page, "%llu\n",
7f7583d4
JM		 4291 (unsigned long long)
4292 atomic64_read(&mddev->resync_mismatches));
9d88883e
N		 4293}
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
b29bebd6 4367 if (kstrtol(buf, 10, &n))
90b08710
BS		 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;
b29bebd6 4435 if (kstrtoull(buf, 10, &min))
5e96ee65
NB		 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 4471 unsigned long long max;
b29bebd6 4472 if (kstrtoull(buf, 10, &max))
5e96ee65
NB		 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,
c4a39551 4697 &md_last_scan_mode.attr,
9d88883e 4698 &md_mismatches.attr,
88202a0c
N		 4699 &md_sync_min.attr,
4700 &md_sync_max.attr,
4701 &md_sync_speed.attr,
90b08710 4702 &md_sync_force_parallel.attr,
88202a0c 4703 &md_sync_completed.attr,
5e96ee65 4704 &md_min_sync.attr,
c6207277 4705 &md_max_sync.attr,
e464eafd
N		 4706 &md_suspend_lo.attr,
4707 &md_suspend_hi.attr,
9b1d1dac 4708 &md_bitmap.attr,
d7f3d291 4709 &md_degraded.attr,
eae1701f
N		 4710 NULL,
4711};
411036fa
N		 4712static struct attribute_group md_redundancy_group = {
4713 .name = NULL,
4714 .attrs = md_redundancy_attrs,
4715};
4716
eae1701f
N		 4717
4718static ssize_t
4719md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4720{
4721 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4722 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4723 ssize_t rv;
eae1701f
N		 4724
4725 if (!entry->show)
4726 return -EIO;
af8a2434
N		 4727 spin_lock(&all_mddevs_lock);
4728 if (list_empty(&mddev->all_mddevs)) {
4729 spin_unlock(&all_mddevs_lock);
4730 return -EBUSY;
4731 }
4732 mddev_get(mddev);
4733 spin_unlock(&all_mddevs_lock);
4734
5dc5cf7d
IM		 4735 rv = mddev_lock(mddev);
4736 if (!rv) {
4737 rv = entry->show(mddev, page);
4738 mddev_unlock(mddev);
4739 }
af8a2434 4740 mddev_put(mddev);
96de1e66 4741 return rv;
eae1701f
N		 4742}
4743
4744static ssize_t
4745md_attr_store(struct kobject *kobj, struct attribute *attr,
4746 const char *page, size_t length)
4747{
4748 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4749 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4750 ssize_t rv;
eae1701f
N		 4751
4752 if (!entry->store)
4753 return -EIO;
67463acb
N		 4754 if (!capable(CAP_SYS_ADMIN))
4755 return -EACCES;
af8a2434
N		 4756 spin_lock(&all_mddevs_lock);
4757 if (list_empty(&mddev->all_mddevs)) {
4758 spin_unlock(&all_mddevs_lock);
4759 return -EBUSY;
4760 }
4761 mddev_get(mddev);
4762 spin_unlock(&all_mddevs_lock);
a7a3f08d
N		 4763 if (entry->store == new_dev_store)
4764 flush_workqueue(md_misc_wq);
5dc5cf7d
IM		 4765 rv = mddev_lock(mddev);
4766 if (!rv) {
4767 rv = entry->store(mddev, page, length);
4768 mddev_unlock(mddev);
4769 }
af8a2434 4770 mddev_put(mddev);
96de1e66 4771 return rv;
eae1701f
N		 4772}
4773
4774static void md_free(struct kobject *ko)
4775{
fd01b88c 4776 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4777
4778 if (mddev->sysfs_state)
4779 sysfs_put(mddev->sysfs_state);
4780
4781 if (mddev->gendisk) {
4782 del_gendisk(mddev->gendisk);
4783 put_disk(mddev->gendisk);
4784 }
4785 if (mddev->queue)
4786 blk_cleanup_queue(mddev->queue);
4787
eae1701f
N		 4788 kfree(mddev);
4789}
4790
52cf25d0 4791static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4792 .show = md_attr_show,
4793 .store = md_attr_store,
4794};
4795static struct kobj_type md_ktype = {
4796 .release = md_free,
4797 .sysfs_ops = &md_sysfs_ops,
4798 .default_attrs = md_default_attrs,
4799};
4800
1da177e4
LT		 4801int mdp_major = 0;
4802
5fd3a17e
DW		 4803static void mddev_delayed_delete(struct work_struct *ws)
4804{
fd01b88c 4805 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4806
43a70507 4807 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4808 kobject_del(&mddev->kobj);
4809 kobject_put(&mddev->kobj);
4810}
4811
efeb53c0 4812static int md_alloc(dev_t dev, char *name)
1da177e4 4813{
48c9c27b 4814 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4815 struct mddev *mddev = mddev_find(dev);
1da177e4 4816 struct gendisk *disk;
efeb53c0
N		 4817 int partitioned;
4818 int shift;
4819 int unit;
3830c62f 4820 int error;
1da177e4
LT		 4821
4822 if (!mddev)
efeb53c0
N		 4823 return -ENODEV;
4824
4825 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4826 shift = partitioned ? MdpMinorShift : 0;
4827 unit = MINOR(mddev->unit) >> shift;
1da177e4 4828
e804ac78
TH		 4829 /* wait for any previous instance of this device to be
4830 * completely removed (mddev_delayed_delete).
d3374825 4831 */
e804ac78 4832 flush_workqueue(md_misc_wq);
d3374825 4833
48c9c27b 4834 mutex_lock(&disks_mutex);
0909dc44
N		 4835 error = -EEXIST;
4836 if (mddev->gendisk)
4837 goto abort;
efeb53c0
N		 4838
4839 if (name) {
4840 /* Need to ensure that 'name' is not a duplicate.
4841 */
fd01b88c 4842 struct mddev *mddev2;
efeb53c0
N		 4843 spin_lock(&all_mddevs_lock);
4844
4845 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4846 if (mddev2->gendisk &&
4847 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4848 spin_unlock(&all_mddevs_lock);
0909dc44 4849 goto abort;
efeb53c0
N		 4850 }
4851 spin_unlock(&all_mddevs_lock);
1da177e4 4852 }
8b765398 4853
0909dc44 4854 error = -ENOMEM;
8b765398 4855 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4856 if (!mddev->queue)
4857 goto abort;
409c57f3
N		 4858 mddev->queue->queuedata = mddev;
4859
409c57f3 4860 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 4861 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 4862
1da177e4
LT		 4863 disk = alloc_disk(1 << shift);
4864 if (!disk) {
8b765398
N		 4865 blk_cleanup_queue(mddev->queue);
4866 mddev->queue = NULL;
0909dc44 4867 goto abort;
1da177e4 4868 }
efeb53c0 4869 disk->major = MAJOR(mddev->unit);
1da177e4 4870 disk->first_minor = unit << shift;
efeb53c0
N		 4871 if (name)
4872 strcpy(disk->disk_name, name);
4873 else if (partitioned)
1da177e4 4874 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4875 else
1da177e4 4876 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4877 disk->fops = &md_fops;
4878 disk->private_data = mddev;
4879 disk->queue = mddev->queue;
b0140891 4880 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4881 /* Allow extended partitions. This makes the
d3374825 4882 * 'mdp' device redundant, but we can't really
92850bbd
N		 4883 * remove it now.
4884 */
4885 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4886 mddev->gendisk = disk;
b0140891
N		 4887 /* As soon as we call add_disk(), another thread could get
4888 * through to md_open, so make sure it doesn't get too far
4889 */
4890 mutex_lock(&mddev->open_mutex);
4891 add_disk(disk);
4892
ed9e1982
TH		 4893 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4894 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4895 if (error) {
4896 /* This isn't possible, but as kobject_init_and_add is marked
4897 * __must_check, we must do something with the result
4898 */
5e55e2f5
N		 4899 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4900 disk->disk_name);
0909dc44
N		 4901 error = 0;
4902 }
00bcb4ac
N		 4903 if (mddev->kobj.sd &&
4904 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4905 printk(KERN_DEBUG "pointless warning\n");
b0140891 4906 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4907 abort:
4908 mutex_unlock(&disks_mutex);
00bcb4ac 4909 if (!error && mddev->kobj.sd) {
3830c62f 4910 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4911 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4912 }
d3374825 4913 mddev_put(mddev);
0909dc44 4914 return error;
efeb53c0
N		 4915}
4916
4917static struct kobject *md_probe(dev_t dev, int *part, void *data)
4918{
4919 md_alloc(dev, NULL);
1da177e4
LT		 4920 return NULL;
4921}
4922
efeb53c0
N		 4923static int add_named_array(const char *val, struct kernel_param *kp)
4924{
4925 /* val must be "md_*" where * is not all digits.
4926 * We allocate an array with a large free minor number, and
4927 * set the name to val. val must not already be an active name.
4928 */
4929 int len = strlen(val);
4930 char buf[DISK_NAME_LEN];
4931
4932 while (len && val[len-1] == '\n')
4933 len--;
4934 if (len >= DISK_NAME_LEN)
4935 return -E2BIG;
4936 strlcpy(buf, val, len+1);
4937 if (strncmp(buf, "md_", 3) != 0)
4938 return -EINVAL;
4939 return md_alloc(0, buf);
4940}
4941
1da177e4
LT		 4942static void md_safemode_timeout(unsigned long data)
4943{
fd01b88c 4944 struct mddev *mddev = (struct mddev *) data;
1da177e4 4945
0fd62b86
NB		 4946 if (!atomic_read(&mddev->writes_pending)) {
4947 mddev->safemode = 1;
4948 if (mddev->external)
00bcb4ac 4949 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4950 }
1da177e4
LT		 4951 md_wakeup_thread(mddev->thread);
4952}
4953
6ff8d8ec 4954static int start_dirty_degraded;
1da177e4 4955
fd01b88c 4956int md_run(struct mddev *mddev)
1da177e4 4957{
2604b703 4958 int err;
3cb03002 4959 struct md_rdev *rdev;
84fc4b56 4960 struct md_personality *pers;
1da177e4 4961
a757e64c
N		 4962 if (list_empty(&mddev->disks))
4963 /* cannot run an array with no devices.. */
1da177e4 4964 return -EINVAL;
1da177e4
LT		 4965
4966 if (mddev->pers)
4967 return -EBUSY;
bb4f1e9d
N		 4968 /* Cannot run until previous stop completes properly */
4969 if (mddev->sysfs_active)
4970 return -EBUSY;
b6eb127d 4971
1da177e4
LT		 4972 /*
4973 * Analyze all RAID superblock(s)
4974 */
1ec4a939
N		 4975 if (!mddev->raid_disks) {
4976 if (!mddev->persistent)
4977 return -EINVAL;
a757e64c 4978 analyze_sbs(mddev);
1ec4a939 4979 }
1da177e4 4980
d9d166c2
N		 4981 if (mddev->level != LEVEL_NONE)
4982 request_module("md-level-%d", mddev->level);
4983 else if (mddev->clevel[0])
4984 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4985
4986 /*
4987 * Drop all container device buffers, from now on
4988 * the only valid external interface is through the md
4989 * device.
1da177e4 4990 */
dafb20fa 4991 rdev_for_each(rdev, mddev) {
b2d444d7 4992 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4993 continue;
4994 sync_blockdev(rdev->bdev);
f98393a6 4995 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4996
4997 /* perform some consistency tests on the device.
4998 * We don't want the data to overlap the metadata,
58c0fed4 4999 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 5000 */
a6ff7e08
JB		 5001 if (rdev->meta_bdev) {
5002 /* Nothing to check */;
5003 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 5004 if (mddev->dev_sectors &&
5005 rdev->data_offset + mddev->dev_sectors
0f420358 5006 > rdev->sb_start) {
f0d76d70
N		 5007 printk("md: %s: data overlaps metadata\n",
5008 mdname(mddev));
5009 return -EINVAL;
5010 }
5011 } else {
0f420358 5012 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 5013 > rdev->data_offset) {
5014 printk("md: %s: metadata overlaps data\n",
5015 mdname(mddev));
5016 return -EINVAL;
5017 }
5018 }
00bcb4ac 5019 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 5020 }
5021
a167f663 5022 if (mddev->bio_set == NULL)
395c72a7 5023 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
a167f663 5024
1da177e4 5025 spin_lock(&pers_lock);
d9d166c2 5026 pers = find_pers(mddev->level, mddev->clevel);
2604b703 5027 if (!pers || !try_module_get(pers->owner)) {
1da177e4 5028 spin_unlock(&pers_lock);
d9d166c2
N		 5029 if (mddev->level != LEVEL_NONE)
5030 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5031 mddev->level);
5032 else
5033 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5034 mddev->clevel);
1da177e4
LT		 5035 return -EINVAL;
5036 }
2604b703 5037 mddev->pers = pers;
1da177e4 5038 spin_unlock(&pers_lock);
34817e8c
N		 5039 if (mddev->level != pers->level) {
5040 mddev->level = pers->level;
5041 mddev->new_level = pers->level;
5042 }
d9d166c2 5043 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 5044
f6705578 5045 if (mddev->reshape_position != MaxSector &&
63c70c4f 5046 pers->start_reshape == NULL) {
f6705578
N		 5047 /* This personality cannot handle reshaping... */
5048 mddev->pers = NULL;
5049 module_put(pers->owner);
5050 return -EINVAL;
5051 }
5052
7dd5e7c3
N		 5053 if (pers->sync_request) {
5054 /* Warn if this is a potentially silly
5055 * configuration.
5056 */
5057 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5058 struct md_rdev *rdev2;
7dd5e7c3 5059 int warned = 0;
159ec1fc 5060
dafb20fa
N		 5061 rdev_for_each(rdev, mddev)
5062 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N		 5063 if (rdev < rdev2 &&
5064 rdev->bdev->bd_contains ==
5065 rdev2->bdev->bd_contains) {
5066 printk(KERN_WARNING
5067 "%s: WARNING: %s appears to be"
5068 " on the same physical disk as"
5069 " %s.\n",
5070 mdname(mddev),
5071 bdevname(rdev->bdev,b),
5072 bdevname(rdev2->bdev,b2));
5073 warned = 1;
5074 }
5075 }
159ec1fc 5076
7dd5e7c3
N		 5077 if (warned)
5078 printk(KERN_WARNING
5079 "True protection against single-disk"
5080 " failure might be compromised.\n");
5081 }
5082
657390d2 5083 mddev->recovery = 0;
58c0fed4
AN		 5084 /* may be over-ridden by personality */
5085 mddev->resync_max_sectors = mddev->dev_sectors;
5086
6ff8d8ec 5087 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5088
0f9552b5 5089 if (start_readonly && mddev->ro == 0)
f91de92e
N		 5090 mddev->ro = 2; /* read-only, but switch on first write */
5091
b15c2e57 5092 err = mddev->pers->run(mddev);
13e53df3
AN		 5093 if (err)
5094 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 5095 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5096 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5097 " but 'external_size' not in effect?\n", __func__);
5098 printk(KERN_ERR
5099 "md: invalid array_size %llu > default size %llu\n",
5100 (unsigned long long)mddev->array_sectors / 2,
5101 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5102 err = -EINVAL;
5103 mddev->pers->stop(mddev);
5104 }
ef99bf48
N		 5105 if (err == 0 && mddev->pers->sync_request &&
5106 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
b15c2e57
N		 5107 err = bitmap_create(mddev);
5108 if (err) {
5109 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5110 mdname(mddev), err);
5111 mddev->pers->stop(mddev);
5112 }
5113 }
1da177e4 5114 if (err) {
1da177e4
LT		 5115 module_put(mddev->pers->owner);
5116 mddev->pers = NULL;
32a7627c
N		 5117 bitmap_destroy(mddev);
5118 return err;
1da177e4 5119 }
5e55e2f5 5120 if (mddev->pers->sync_request) {
00bcb4ac
N		 5121 if (mddev->kobj.sd &&
5122 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 5123 printk(KERN_WARNING
5124 "md: cannot register extra attributes for %s\n",
5125 mdname(mddev));
00bcb4ac 5126 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5127 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 5128 mddev->ro = 0;
5129
1da177e4 5130 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 5131 atomic_set(&mddev->max_corr_read_errors,
5132 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 5133 mddev->safemode = 0;
5134 mddev->safemode_timer.function = md_safemode_timeout;
5135 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 5136 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5137 mddev->in_sync = 1;
0ca69886
N		 5138 smp_wmb();
5139 mddev->ready = 1;
dafb20fa 5140 rdev_for_each(rdev, mddev)
36fad858
NK		 5141 if (rdev->raid_disk >= 0)
5142 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 5143 /* failure here is OK */;
1da177e4
LT		 5144
5145 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5146
7a0a5355 5147 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 5148 md_update_sb(mddev, 0);
1da177e4 5149
d7603b7e 5150 md_new_event(mddev);
00bcb4ac
N		 5151 sysfs_notify_dirent_safe(mddev->sysfs_state);
5152 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 5153 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 5154 return 0;
5155}
390ee602 5156EXPORT_SYMBOL_GPL(md_run);
1da177e4 5157
fd01b88c 5158static int do_md_run(struct mddev *mddev)
fe60b014
N		 5159{
5160 int err;
5161
5162 err = md_run(mddev);
5163 if (err)
5164 goto out;
69e51b44
N		 5165 err = bitmap_load(mddev);
5166 if (err) {
5167 bitmap_destroy(mddev);
5168 goto out;
5169 }
0fd018af
JB		 5170
5171 md_wakeup_thread(mddev->thread);
5172 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5173
fe60b014
N		 5174 set_capacity(mddev->gendisk, mddev->array_sectors);
5175 revalidate_disk(mddev->gendisk);
f0b4f7e2 5176 mddev->changed = 1;
fe60b014
N		 5177 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5178out:
5179 return err;
5180}
5181
fd01b88c 5182static int restart_array(struct mddev *mddev)
1da177e4
LT		 5183{
5184 struct gendisk *disk = mddev->gendisk;
1da177e4 5185
80fab1d7 5186 /* Complain if it has no devices */
1da177e4 5187 if (list_empty(&mddev->disks))
80fab1d7
AN		 5188 return -ENXIO;
5189 if (!mddev->pers)
5190 return -EINVAL;
5191 if (!mddev->ro)
5192 return -EBUSY;
5193 mddev->safemode = 0;
5194 mddev->ro = 0;
5195 set_disk_ro(disk, 0);
5196 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5197 mdname(mddev));
5198 /* Kick recovery or resync if necessary */
5199 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5200 md_wakeup_thread(mddev->thread);
5201 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5202 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5203 return 0;
1da177e4
LT		 5204}
5205
acc55e22
N		 5206/* similar to deny_write_access, but accounts for our holding a reference
5207 * to the file ourselves */
5208static int deny_bitmap_write_access(struct file * file)
5209{
5210 struct inode *inode = file->f_mapping->host;
5211
5212 spin_lock(&inode->i_lock);
5213 if (atomic_read(&inode->i_writecount) > 1) {
5214 spin_unlock(&inode->i_lock);
5215 return -ETXTBSY;
5216 }
5217 atomic_set(&inode->i_writecount, -1);
5218 spin_unlock(&inode->i_lock);
5219
5220 return 0;
5221}
5222
43a70507 5223void restore_bitmap_write_access(struct file *file)
acc55e22
N		 5224{
5225 struct inode *inode = file->f_mapping->host;
5226
5227 spin_lock(&inode->i_lock);
5228 atomic_set(&inode->i_writecount, 1);
5229 spin_unlock(&inode->i_lock);
5230}
5231
fd01b88c 5232static void md_clean(struct mddev *mddev)
6177b472
N		 5233{
5234 mddev->array_sectors = 0;
5235 mddev->external_size = 0;
5236 mddev->dev_sectors = 0;
5237 mddev->raid_disks = 0;
5238 mddev->recovery_cp = 0;
5239 mddev->resync_min = 0;
5240 mddev->resync_max = MaxSector;
5241 mddev->reshape_position = MaxSector;
5242 mddev->external = 0;
5243 mddev->persistent = 0;
5244 mddev->level = LEVEL_NONE;
5245 mddev->clevel[0] = 0;
5246 mddev->flags = 0;
5247 mddev->ro = 0;
5248 mddev->metadata_type[0] = 0;
5249 mddev->chunk_sectors = 0;
5250 mddev->ctime = mddev->utime = 0;
5251 mddev->layout = 0;
5252 mddev->max_disks = 0;
5253 mddev->events = 0;
a8707c08 5254 mddev->can_decrease_events = 0;
6177b472 5255 mddev->delta_disks = 0;
2c810cdd 5256 mddev->reshape_backwards = 0;
6177b472
N		 5257 mddev->new_level = LEVEL_NONE;
5258 mddev->new_layout = 0;
5259 mddev->new_chunk_sectors = 0;
5260 mddev->curr_resync = 0;
7f7583d4 5261 atomic64_set(&mddev->resync_mismatches, 0);
6177b472
N		 5262 mddev->suspend_lo = mddev->suspend_hi = 0;
5263 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5264 mddev->recovery = 0;
5265 mddev->in_sync = 0;
f0b4f7e2 5266 mddev->changed = 0;
6177b472 5267 mddev->degraded = 0;
6177b472 5268 mddev->safemode = 0;
050b6615 5269 mddev->merge_check_needed = 0;
6177b472
N		 5270 mddev->bitmap_info.offset = 0;
5271 mddev->bitmap_info.default_offset = 0;
6409bb05 5272 mddev->bitmap_info.default_space = 0;
6177b472
N		 5273 mddev->bitmap_info.chunksize = 0;
5274 mddev->bitmap_info.daemon_sleep = 0;
5275 mddev->bitmap_info.max_write_behind = 0;
5276}
5277
fd01b88c 5278static void __md_stop_writes(struct mddev *mddev)
a047e125 5279{
6b6204ee 5280 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
a047e125 5281 if (mddev->sync_thread) {
a047e125 5282 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 5283 md_reap_sync_thread(mddev);
a047e125
N		 5284 }
5285
5286 del_timer_sync(&mddev->safemode_timer);
5287
5288 bitmap_flush(mddev);
5289 md_super_wait(mddev);
5290
b6d428c6 5291 if (mddev->ro == 0 &&
7a0a5355 5292 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
a047e125
N		 5293 /* mark array as shutdown cleanly */
5294 mddev->in_sync = 1;
5295 md_update_sb(mddev, 1);
5296 }
5297}
defad61a 5298
fd01b88c 5299void md_stop_writes(struct mddev *mddev)
defad61a
N		 5300{
5301 mddev_lock(mddev);
5302 __md_stop_writes(mddev);
5303 mddev_unlock(mddev);
5304}
390ee602 5305EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5306
5eff3c43 5307static void __md_stop(struct mddev *mddev)
6177b472 5308{
0ca69886 5309 mddev->ready = 0;
6177b472
N		 5310 mddev->pers->stop(mddev);
5311 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5312 mddev->to_remove = &md_redundancy_group;
5313 module_put(mddev->pers->owner);
5314 mddev->pers = NULL;
cca9cf90 5315 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5316}
5eff3c43
N		 5317
5318void md_stop(struct mddev *mddev)
5319{
5320 /* stop the array and free an attached data structures.
5321 * This is called from dm-raid
5322 */
5323 __md_stop(mddev);
5324 bitmap_destroy(mddev);
5325 if (mddev->bio_set)
5326 bioset_free(mddev->bio_set);
5327}
5328
390ee602 5329EXPORT_SYMBOL_GPL(md_stop);
6177b472 5330
a05b7ea0 5331static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N		 5332{
5333 int err = 0;
5334 mutex_lock(&mddev->open_mutex);
a05b7ea0 5335 if (atomic_read(&mddev->openers) > !!bdev) {
a4bd82d0
N		 5336 printk("md: %s still in use.\n",mdname(mddev));
5337 err = -EBUSY;
5338 goto out;
5339 }
260fa034
N		 5340 if (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags)) {
5341 /* Someone opened the device since we flushed it
5342 * so page cache could be dirty and it is too late
5343 * to flush. So abort
5344 */
5345 mutex_unlock(&mddev->open_mutex);
5346 return -EBUSY;
5347 }
a4bd82d0 5348 if (mddev->pers) {
defad61a 5349 __md_stop_writes(mddev);
a4bd82d0
N		 5350
5351 err = -ENXIO;
5352 if (mddev->ro==1)
5353 goto out;
5354 mddev->ro = 1;
5355 set_disk_ro(mddev->gendisk, 1);
5356 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 5357 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 5358 err = 0;
5359 }
5360out:
5361 mutex_unlock(&mddev->open_mutex);
5362 return err;
5363}
5364
9e653b63
N		 5365/* mode:
5366 * 0 - completely stop and dis-assemble array
9e653b63
N		 5367 * 2 - stop but do not disassemble array
5368 */
a05b7ea0
N		 5369static int do_md_stop(struct mddev * mddev, int mode,
5370 struct block_device *bdev)
1da177e4 5371{
1da177e4 5372 struct gendisk *disk = mddev->gendisk;
3cb03002 5373 struct md_rdev *rdev;
1da177e4 5374
c8c00a69 5375 mutex_lock(&mddev->open_mutex);
a05b7ea0 5376 if (atomic_read(&mddev->openers) > !!bdev ||
bb4f1e9d 5377 mddev->sysfs_active) {
df5b20cf 5378 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 5379 mutex_unlock(&mddev->open_mutex);
5380 return -EBUSY;
5381 }
260fa034
N		 5382 if (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags)) {
5383 /* Someone opened the device since we flushed it
5384 * so page cache could be dirty and it is too late
5385 * to flush. So abort
a05b7ea0 5386 */
260fa034
N		 5387 mutex_unlock(&mddev->open_mutex);
5388 return -EBUSY;
5389 }
6e17b027 5390 if (mddev->pers) {
a4bd82d0
N		 5391 if (mddev->ro)
5392 set_disk_ro(disk, 0);
409c57f3 5393
defad61a 5394 __md_stop_writes(mddev);
5eff3c43 5395 __md_stop(mddev);
a4bd82d0 5396 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5397 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5398
a4bd82d0 5399 /* tell userspace to handle 'inactive' */
00bcb4ac 5400 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5401
dafb20fa 5402 rdev_for_each(rdev, mddev)
36fad858
NK		 5403 if (rdev->raid_disk >= 0)
5404 sysfs_unlink_rdev(mddev, rdev);
c4647292 5405
a4bd82d0 5406 set_capacity(disk, 0);
6e17b027 5407 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5408 mddev->changed = 1;
a4bd82d0 5409 revalidate_disk(disk);
0d4ca600 5410
a4bd82d0
N		 5411 if (mddev->ro)
5412 mddev->ro = 0;
6e17b027
N		 5413 } else
5414 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5415 /*
5416 * Free resources if final stop
5417 */
9e653b63 5418 if (mode == 0) {
1da177e4
LT		 5419 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5420
978f946b 5421 bitmap_destroy(mddev);
c3d9714e
N		 5422 if (mddev->bitmap_info.file) {
5423 restore_bitmap_write_access(mddev->bitmap_info.file);
5424 fput(mddev->bitmap_info.file);
5425 mddev->bitmap_info.file = NULL;
978f946b 5426 }
c3d9714e 5427 mddev->bitmap_info.offset = 0;
978f946b 5428
1da177e4
LT		 5429 export_array(mddev);
5430
6177b472 5431 md_clean(mddev);
934d9c23 5432 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5433 if (mddev->hold_active == UNTIL_STOP)
5434 mddev->hold_active = 0;
a4bd82d0 5435 }
3f9d99c1 5436 blk_integrity_unregister(disk);
d7603b7e 5437 md_new_event(mddev);
00bcb4ac 5438 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5439 return 0;
1da177e4
LT		 5440}
5441
fdee8ae4 5442#ifndef MODULE
fd01b88c 5443static void autorun_array(struct mddev *mddev)
1da177e4 5444{
3cb03002 5445 struct md_rdev *rdev;
1da177e4
LT		 5446 int err;
5447
a757e64c 5448 if (list_empty(&mddev->disks))
1da177e4 5449 return;
1da177e4
LT		 5450
5451 printk(KERN_INFO "md: running: ");
5452
dafb20fa 5453 rdev_for_each(rdev, mddev) {
1da177e4
LT		 5454 char b[BDEVNAME_SIZE];
5455 printk("<%s>", bdevname(rdev->bdev,b));
5456 }
5457 printk("\n");
5458
d710e138 5459 err = do_md_run(mddev);
1da177e4
LT		 5460 if (err) {
5461 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
a05b7ea0 5462 do_md_stop(mddev, 0, NULL);
1da177e4
LT		 5463 }
5464}
5465
5466/*
5467 * lets try to run arrays based on all disks that have arrived
5468 * until now. (those are in pending_raid_disks)
5469 *
5470 * the method: pick the first pending disk, collect all disks with
5471 * the same UUID, remove all from the pending list and put them into
5472 * the 'same_array' list. Then order this list based on superblock
5473 * update time (freshest comes first), kick out 'old' disks and
5474 * compare superblocks. If everything's fine then run it.
5475 *
5476 * If "unit" is allocated, then bump its reference count
5477 */
5478static void autorun_devices(int part)
5479{
3cb03002 5480 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5481 struct mddev *mddev;
1da177e4
LT		 5482 char b[BDEVNAME_SIZE];
5483
5484 printk(KERN_INFO "md: autorun ...\n");
5485 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5486 int unit;
1da177e4 5487 dev_t dev;
ad01c9e3 5488 LIST_HEAD(candidates);
1da177e4 5489 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5490 struct md_rdev, same_set);
1da177e4
LT		 5491
5492 printk(KERN_INFO "md: considering %s ...\n",
5493 bdevname(rdev0->bdev,b));
5494 INIT_LIST_HEAD(&candidates);
159ec1fc 5495 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5496 if (super_90_load(rdev, rdev0, 0) >= 0) {
5497 printk(KERN_INFO "md: adding %s ...\n",
5498 bdevname(rdev->bdev,b));
5499 list_move(&rdev->same_set, &candidates);
5500 }
5501 /*
5502 * now we have a set of devices, with all of them having
5503 * mostly sane superblocks. It's time to allocate the
5504 * mddev.
5505 */
e8703fe1
N		 5506 if (part) {
5507 dev = MKDEV(mdp_major,
5508 rdev0->preferred_minor << MdpMinorShift);
5509 unit = MINOR(dev) >> MdpMinorShift;
5510 } else {
5511 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5512 unit = MINOR(dev);
5513 }
5514 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5515 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5516 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5517 break;
5518 }
1da177e4
LT		 5519
5520 md_probe(dev, NULL, NULL);
5521 mddev = mddev_find(dev);
9bbbca3a
NB		 5522 if (!mddev || !mddev->gendisk) {
5523 if (mddev)
5524 mddev_put(mddev);
5525 printk(KERN_ERR
1da177e4
LT		 5526 "md: cannot allocate memory for md drive.\n");
5527 break;
5528 }
5529 if (mddev_lock(mddev))
5530 printk(KERN_WARNING "md: %s locked, cannot run\n",
5531 mdname(mddev));
5532 else if (mddev->raid_disks || mddev->major_version
5533 || !list_empty(&mddev->disks)) {
5534 printk(KERN_WARNING
5535 "md: %s already running, cannot run %s\n",
5536 mdname(mddev), bdevname(rdev0->bdev,b));
5537 mddev_unlock(mddev);
5538 } else {
5539 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5540 mddev->persistent = 1;
159ec1fc 5541 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5542 list_del_init(&rdev->same_set);
5543 if (bind_rdev_to_array(rdev, mddev))
5544 export_rdev(rdev);
5545 }
5546 autorun_array(mddev);
5547 mddev_unlock(mddev);
5548 }
5549 /* on success, candidates will be empty, on error
5550 * it won't...
5551 */
159ec1fc 5552 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5553 list_del_init(&rdev->same_set);
1da177e4 5554 export_rdev(rdev);
4b80991c 5555 }
1da177e4
LT		 5556 mddev_put(mddev);
5557 }
5558 printk(KERN_INFO "md: ... autorun DONE.\n");
5559}
fdee8ae4 5560#endif /* !MODULE */
1da177e4 5561
1da177e4
LT		 5562static int get_version(void __user * arg)
5563{
5564 mdu_version_t ver;
5565
5566 ver.major = MD_MAJOR_VERSION;
5567 ver.minor = MD_MINOR_VERSION;
5568 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5569
5570 if (copy_to_user(arg, &ver, sizeof(ver)))
5571 return -EFAULT;
5572
5573 return 0;
5574}
5575
fd01b88c 5576static int get_array_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5577{
5578 mdu_array_info_t info;
a9f326eb 5579 int nr,working,insync,failed,spare;
3cb03002 5580 struct md_rdev *rdev;
1da177e4 5581
1ca69c4b
N		 5582 nr = working = insync = failed = spare = 0;
5583 rcu_read_lock();
5584 rdev_for_each_rcu(rdev, mddev) {
1da177e4 5585 nr++;
b2d444d7 5586 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5587 failed++;
5588 else {
5589 working++;
b2d444d7 5590 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5591 insync++;
1da177e4
LT		 5592 else
5593 spare++;
5594 }
5595 }
1ca69c4b 5596 rcu_read_unlock();
1da177e4
LT		 5597
5598 info.major_version = mddev->major_version;
5599 info.minor_version = mddev->minor_version;
5600 info.patch_version = MD_PATCHLEVEL_VERSION;
5601 info.ctime = mddev->ctime;
5602 info.level = mddev->level;
58c0fed4
AN		 5603 info.size = mddev->dev_sectors / 2;
5604 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5605 info.size = -1;
1da177e4
LT		 5606 info.nr_disks = nr;
5607 info.raid_disks = mddev->raid_disks;
5608 info.md_minor = mddev->md_minor;
5609 info.not_persistent= !mddev->persistent;
5610
5611 info.utime = mddev->utime;
5612 info.state = 0;
5613 if (mddev->in_sync)
5614 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5615 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5616 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5617 info.active_disks = insync;
1da177e4
LT		 5618 info.working_disks = working;
5619 info.failed_disks = failed;
5620 info.spare_disks = spare;
5621
5622 info.layout = mddev->layout;
9d8f0363 5623 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5624
5625 if (copy_to_user(arg, &info, sizeof(info)))
5626 return -EFAULT;
5627
5628 return 0;
5629}
5630
fd01b88c 5631static int get_bitmap_file(struct mddev * mddev, void __user * arg)
32a7627c
N		 5632{
5633 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5634 char *ptr, *buf = NULL;
5635 int err = -ENOMEM;
5636
60559da4 5637 file = kmalloc(sizeof(*file), GFP_NOIO);
2a2275d6 5638
32a7627c
N		 5639 if (!file)
5640 goto out;
5641
5642 /* bitmap disabled, zero the first byte and copy out */
1ec885cd 5643 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
32a7627c
N		 5644 file->pathname[0] = '\0';
5645 goto copy_out;
5646 }
5647
5648 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5649 if (!buf)
5650 goto out;
5651
1ec885cd
N		 5652 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5653 buf, sizeof(file->pathname));
6bcfd601 5654 if (IS_ERR(ptr))
32a7627c
N		 5655 goto out;
5656
5657 strcpy(file->pathname, ptr);
5658
5659copy_out:
5660 err = 0;
5661 if (copy_to_user(arg, file, sizeof(*file)))
5662 err = -EFAULT;
5663out:
5664 kfree(buf);
5665 kfree(file);
5666 return err;
5667}
5668
fd01b88c 5669static int get_disk_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5670{
5671 mdu_disk_info_t info;
3cb03002 5672 struct md_rdev *rdev;
1da177e4
LT		 5673
5674 if (copy_from_user(&info, arg, sizeof(info)))
5675 return -EFAULT;
5676
1ca69c4b
N		 5677 rcu_read_lock();
5678 rdev = find_rdev_nr_rcu(mddev, info.number);
1da177e4
LT		 5679 if (rdev) {
5680 info.major = MAJOR(rdev->bdev->bd_dev);
5681 info.minor = MINOR(rdev->bdev->bd_dev);
5682 info.raid_disk = rdev->raid_disk;
5683 info.state = 0;
b2d444d7 5684 if (test_bit(Faulty, &rdev->flags))
1da177e4 5685 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5686 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5687 info.state |= (1<<MD_DISK_ACTIVE);
5688 info.state |= (1<<MD_DISK_SYNC);
5689 }
8ddf9efe
N		 5690 if (test_bit(WriteMostly, &rdev->flags))
5691 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5692 } else {
5693 info.major = info.minor = 0;
5694 info.raid_disk = -1;
5695 info.state = (1<<MD_DISK_REMOVED);
5696 }
1ca69c4b 5697 rcu_read_unlock();
1da177e4
LT		 5698
5699 if (copy_to_user(arg, &info, sizeof(info)))
5700 return -EFAULT;
5701
5702 return 0;
5703}
5704
fd01b88c 5705static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
1da177e4
LT		 5706{
5707 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5708 struct md_rdev *rdev;
1da177e4
LT		 5709 dev_t dev = MKDEV(info->major,info->minor);
5710
5711 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5712 return -EOVERFLOW;
5713
5714 if (!mddev->raid_disks) {
5715 int err;
5716 /* expecting a device which has a superblock */
5717 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5718 if (IS_ERR(rdev)) {
5719 printk(KERN_WARNING
5720 "md: md_import_device returned %ld\n",
5721 PTR_ERR(rdev));
5722 return PTR_ERR(rdev);
5723 }
5724 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5725 struct md_rdev *rdev0
5726 = list_entry(mddev->disks.next,
5727 struct md_rdev, same_set);
a9f326eb 5728 err = super_types[mddev->major_version]
1da177e4
LT		 5729 .load_super(rdev, rdev0, mddev->minor_version);
5730 if (err < 0) {
5731 printk(KERN_WARNING
5732 "md: %s has different UUID to %s\n",
5733 bdevname(rdev->bdev,b),
5734 bdevname(rdev0->bdev,b2));
5735 export_rdev(rdev);
5736 return -EINVAL;
5737 }
5738 }
5739 err = bind_rdev_to_array(rdev, mddev);
5740 if (err)
5741 export_rdev(rdev);
5742 return err;
5743 }
5744
5745 /*
5746 * add_new_disk can be used once the array is assembled
5747 * to add "hot spares". They must already have a superblock
5748 * written
5749 */
5750 if (mddev->pers) {
5751 int err;
5752 if (!mddev->pers->hot_add_disk) {
5753 printk(KERN_WARNING
5754 "%s: personality does not support diskops!\n",
5755 mdname(mddev));
5756 return -EINVAL;
5757 }
7b1e35f6
N		 5758 if (mddev->persistent)
5759 rdev = md_import_device(dev, mddev->major_version,
5760 mddev->minor_version);
5761 else
5762 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5763 if (IS_ERR(rdev)) {
5764 printk(KERN_WARNING
5765 "md: md_import_device returned %ld\n",
5766 PTR_ERR(rdev));
5767 return PTR_ERR(rdev);
5768 }
1a855a06 5769 /* set saved_raid_disk if appropriate */
41158c7e
N		 5770 if (!mddev->persistent) {
5771 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5772 info->raid_disk < mddev->raid_disks) {
41158c7e 5773 rdev->raid_disk = info->raid_disk;
bf572541
N		 5774 set_bit(In_sync, &rdev->flags);
5775 } else
41158c7e
N		 5776 rdev->raid_disk = -1;
5777 } else
5778 super_types[mddev->major_version].
5779 validate_super(mddev, rdev);
bedd86b7 5780 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 5781 rdev->raid_disk != info->raid_disk) {
bedd86b7
N		 5782 /* This was a hot-add request, but events doesn't
5783 * match, so reject it.
5784 */
5785 export_rdev(rdev);
5786 return -EINVAL;
5787 }
5788
1a855a06
N		 5789 if (test_bit(In_sync, &rdev->flags))
5790 rdev->saved_raid_disk = rdev->raid_disk;
5791 else
5792 rdev->saved_raid_disk = -1;
41158c7e 5793
b2d444d7 5794 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5795 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5796 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5797 else
5798 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5799
1da177e4
LT		 5800 rdev->raid_disk = -1;
5801 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5802 if (!err && !mddev->pers->hot_remove_disk) {
5803 /* If there is hot_add_disk but no hot_remove_disk
5804 * then added disks for geometry changes,
5805 * and should be added immediately.
5806 */
5807 super_types[mddev->major_version].
5808 validate_super(mddev, rdev);
5809 err = mddev->pers->hot_add_disk(mddev, rdev);
5810 if (err)
5811 unbind_rdev_from_array(rdev);
5812 }
1da177e4
LT		 5813 if (err)
5814 export_rdev(rdev);
52664732 5815 else
00bcb4ac 5816 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5817
7ceb17e8 5818 set_bit(MD_CHANGE_DEVS, &mddev->flags);
72a23c21
NB		 5819 if (mddev->degraded)
5820 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5821 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5822 if (!err)
5823 md_new_event(mddev);
005eca5e 5824 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5825 return err;
5826 }
5827
5828 /* otherwise, add_new_disk is only allowed
5829 * for major_version==0 superblocks
5830 */
5831 if (mddev->major_version != 0) {
5832 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5833 mdname(mddev));
5834 return -EINVAL;
5835 }
5836
5837 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5838 int err;
d710e138 5839 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5840 if (IS_ERR(rdev)) {
5841 printk(KERN_WARNING
5842 "md: error, md_import_device() returned %ld\n",
5843 PTR_ERR(rdev));
5844 return PTR_ERR(rdev);
5845 }
5846 rdev->desc_nr = info->number;
5847 if (info->raid_disk < mddev->raid_disks)
5848 rdev->raid_disk = info->raid_disk;
5849 else
5850 rdev->raid_disk = -1;
5851
1da177e4 5852 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5853 if (info->state & (1<<MD_DISK_SYNC))
5854 set_bit(In_sync, &rdev->flags);
1da177e4 5855
8ddf9efe
N		 5856 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5857 set_bit(WriteMostly, &rdev->flags);
5858
1da177e4
LT		 5859 if (!mddev->persistent) {
5860 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5861 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5862 } else
57b2caa3 5863 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5864 rdev->sectors = rdev->sb_start;
1da177e4 5865
2bf071bf
N		 5866 err = bind_rdev_to_array(rdev, mddev);
5867 if (err) {
5868 export_rdev(rdev);
5869 return err;
5870 }
1da177e4
LT		 5871 }
5872
5873 return 0;
5874}
5875
fd01b88c 5876static int hot_remove_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5877{
5878 char b[BDEVNAME_SIZE];
3cb03002 5879 struct md_rdev *rdev;
1da177e4 5880
1da177e4
LT		 5881 rdev = find_rdev(mddev, dev);
5882 if (!rdev)
5883 return -ENXIO;
5884
3ea8929d
N		 5885 clear_bit(Blocked, &rdev->flags);
5886 remove_and_add_spares(mddev, rdev);
5887
1da177e4
LT		 5888 if (rdev->raid_disk >= 0)
5889 goto busy;
5890
5891 kick_rdev_from_array(rdev);
850b2b42 5892 md_update_sb(mddev, 1);
d7603b7e 5893 md_new_event(mddev);
1da177e4
LT		 5894
5895 return 0;
5896busy:
fdefa4d8 5897 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5898 bdevname(rdev->bdev,b), mdname(mddev));
5899 return -EBUSY;
5900}
5901
fd01b88c 5902static int hot_add_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5903{
5904 char b[BDEVNAME_SIZE];
5905 int err;
3cb03002 5906 struct md_rdev *rdev;
1da177e4
LT		 5907
5908 if (!mddev->pers)
5909 return -ENODEV;
5910
5911 if (mddev->major_version != 0) {
5912 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5913 " version-0 superblocks.\n",
5914 mdname(mddev));
5915 return -EINVAL;
5916 }
5917 if (!mddev->pers->hot_add_disk) {
5918 printk(KERN_WARNING
5919 "%s: personality does not support diskops!\n",
5920 mdname(mddev));
5921 return -EINVAL;
5922 }
5923
d710e138 5924 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5925 if (IS_ERR(rdev)) {
5926 printk(KERN_WARNING
5927 "md: error, md_import_device() returned %ld\n",
5928 PTR_ERR(rdev));
5929 return -EINVAL;
5930 }
5931
5932 if (mddev->persistent)
57b2caa3 5933 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5934 else
77304d2a 5935 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5936
8190e754 5937 rdev->sectors = rdev->sb_start;
1da177e4 5938
b2d444d7 5939 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5940 printk(KERN_WARNING
5941 "md: can not hot-add faulty %s disk to %s!\n",
5942 bdevname(rdev->bdev,b), mdname(mddev));
5943 err = -EINVAL;
5944 goto abort_export;
5945 }
b2d444d7 5946 clear_bit(In_sync, &rdev->flags);
1da177e4 5947 rdev->desc_nr = -1;
5842730d 5948 rdev->saved_raid_disk = -1;
2bf071bf
N		 5949 err = bind_rdev_to_array(rdev, mddev);
5950 if (err)
5951 goto abort_export;
1da177e4
LT		 5952
5953 /*
5954 * The rest should better be atomic, we can have disk failures
5955 * noticed in interrupt contexts ...
5956 */
5957
1da177e4
LT		 5958 rdev->raid_disk = -1;
5959
850b2b42 5960 md_update_sb(mddev, 1);
1da177e4
LT		 5961
5962 /*
5963 * Kick recovery, maybe this spare has to be added to the
5964 * array immediately.
5965 */
5966 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5967 md_wakeup_thread(mddev->thread);
d7603b7e 5968 md_new_event(mddev);
1da177e4
LT		 5969 return 0;
5970
1da177e4
LT		 5971abort_export:
5972 export_rdev(rdev);
5973 return err;
5974}
5975
fd01b88c 5976static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c
N		 5977{
5978 int err;
5979
36fa3063
N		 5980 if (mddev->pers) {
5981 if (!mddev->pers->quiesce)
5982 return -EBUSY;
5983 if (mddev->recovery || mddev->sync_thread)
5984 return -EBUSY;
5985 /* we should be able to change the bitmap.. */
5986 }
32a7627c 5987
32a7627c 5988
36fa3063
N		 5989 if (fd >= 0) {
5990 if (mddev->bitmap)
5991 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5992 mddev->bitmap_info.file = fget(fd);
32a7627c 5993
c3d9714e 5994 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5995 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5996 mdname(mddev));
5997 return -EBADF;
5998 }
5999
c3d9714e 6000 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 6001 if (err) {
6002 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6003 mdname(mddev));
c3d9714e
N		 6004 fput(mddev->bitmap_info.file);
6005 mddev->bitmap_info.file = NULL;
36fa3063
N		 6006 return err;
6007 }
c3d9714e 6008 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 6009 } else if (mddev->bitmap == NULL)
6010 return -ENOENT; /* cannot remove what isn't there */
6011 err = 0;
6012 if (mddev->pers) {
6013 mddev->pers->quiesce(mddev, 1);
69e51b44 6014 if (fd >= 0) {
36fa3063 6015 err = bitmap_create(mddev);
69e51b44
N		 6016 if (!err)
6017 err = bitmap_load(mddev);
6018 }
d7375ab3 6019 if (fd < 0 || err) {
36fa3063 6020 bitmap_destroy(mddev);
d7375ab3
N		 6021 fd = -1; /* make sure to put the file */
6022 }
36fa3063 6023 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 6024 }
6025 if (fd < 0) {
c3d9714e
N		 6026 if (mddev->bitmap_info.file) {
6027 restore_bitmap_write_access(mddev->bitmap_info.file);
6028 fput(mddev->bitmap_info.file);
acc55e22 6029 }
c3d9714e 6030 mddev->bitmap_info.file = NULL;
36fa3063
N		 6031 }
6032
32a7627c
N		 6033 return err;
6034}
6035
1da177e4
LT		 6036/*
6037 * set_array_info is used two different ways
6038 * The original usage is when creating a new array.
6039 * In this usage, raid_disks is > 0 and it together with
6040 * level, size, not_persistent,layout,chunksize determine the
6041 * shape of the array.
6042 * This will always create an array with a type-0.90.0 superblock.
6043 * The newer usage is when assembling an array.
6044 * In this case raid_disks will be 0, and the major_version field is
6045 * use to determine which style super-blocks are to be found on the devices.
6046 * The minor and patch _version numbers are also kept incase the
6047 * super_block handler wishes to interpret them.
6048 */
fd01b88c 6049static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
1da177e4
LT		 6050{
6051
6052 if (info->raid_disks == 0) {
6053 /* just setting version number for superblock loading */
6054 if (info->major_version < 0 ||
50511da3 6055 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 6056 super_types[info->major_version].name == NULL) {
6057 /* maybe try to auto-load a module? */
6058 printk(KERN_INFO
6059 "md: superblock version %d not known\n",
6060 info->major_version);
6061 return -EINVAL;
6062 }
6063 mddev->major_version = info->major_version;
6064 mddev->minor_version = info->minor_version;
6065 mddev->patch_version = info->patch_version;
3f9d7b0d 6066 mddev->persistent = !info->not_persistent;
cbd19983
N		 6067 /* ensure mddev_put doesn't delete this now that there
6068 * is some minimal configuration.
6069 */
6070 mddev->ctime = get_seconds();
1da177e4
LT		 6071 return 0;
6072 }
6073 mddev->major_version = MD_MAJOR_VERSION;
6074 mddev->minor_version = MD_MINOR_VERSION;
6075 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6076 mddev->ctime = get_seconds();
6077
6078 mddev->level = info->level;
17115e03 6079 mddev->clevel[0] = 0;
58c0fed4 6080 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 6081 mddev->raid_disks = info->raid_disks;
6082 /* don't set md_minor, it is determined by which /dev/md* was
6083 * openned
6084 */
6085 if (info->state & (1<<MD_SB_CLEAN))
6086 mddev->recovery_cp = MaxSector;
6087 else
6088 mddev->recovery_cp = 0;
6089 mddev->persistent = ! info->not_persistent;
e691063a 6090 mddev->external = 0;
1da177e4
LT		 6091
6092 mddev->layout = info->layout;
9d8f0363 6093 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 6094
6095 mddev->max_disks = MD_SB_DISKS;
6096
e691063a
N		 6097 if (mddev->persistent)
6098 mddev->flags = 0;
850b2b42 6099 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6100
c3d9714e 6101 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6102 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6103 mddev->bitmap_info.offset = 0;
b2a2703c 6104
f6705578
N		 6105 mddev->reshape_position = MaxSector;
6106
1da177e4
LT		 6107 /*
6108 * Generate a 128 bit UUID
6109 */
6110 get_random_bytes(mddev->uuid, 16);
6111
f6705578 6112 mddev->new_level = mddev->level;
664e7c41 6113 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 6114 mddev->new_layout = mddev->layout;
6115 mddev->delta_disks = 0;
2c810cdd 6116 mddev->reshape_backwards = 0;
f6705578 6117
1da177e4
LT		 6118 return 0;
6119}
6120
fd01b88c 6121void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6122{
b522adcd
DW		 6123 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6124
6125 if (mddev->external_size)
6126 return;
6127
1f403624
DW		 6128 mddev->array_sectors = array_sectors;
6129}
6130EXPORT_SYMBOL(md_set_array_sectors);
6131
fd01b88c 6132static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6133{
3cb03002 6134 struct md_rdev *rdev;
a35b0d69 6135 int rv;
d71f9f88 6136 int fit = (num_sectors == 0);
a35b0d69
N		 6137
6138 if (mddev->pers->resize == NULL)
6139 return -EINVAL;
d71f9f88
AN		 6140 /* The "num_sectors" is the number of sectors of each device that
6141 * is used. This can only make sense for arrays with redundancy.
6142 * linear and raid0 always use whatever space is available. We can only
6143 * consider changing this number if no resync or reconstruction is
6144 * happening, and if the new size is acceptable. It must fit before the
0f420358 6145 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 6146 * of each device. If num_sectors is zero, we find the largest size
6147 * that fits.
a35b0d69
N		 6148 */
6149 if (mddev->sync_thread)
6150 return -EBUSY;
a4a6125a 6151
dafb20fa 6152 rdev_for_each(rdev, mddev) {
dd8ac336 6153 sector_t avail = rdev->sectors;
01ab5662 6154
d71f9f88
AN		 6155 if (fit && (num_sectors == 0 || num_sectors > avail))
6156 num_sectors = avail;
6157 if (avail < num_sectors)
a35b0d69
N		 6158 return -ENOSPC;
6159 }
d71f9f88 6160 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 6161 if (!rv)
6162 revalidate_disk(mddev->gendisk);
a35b0d69
N		 6163 return rv;
6164}
6165
fd01b88c 6166static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 6167{
6168 int rv;
c6563a8c 6169 struct md_rdev *rdev;
da943b99 6170 /* change the number of raid disks */
63c70c4f 6171 if (mddev->pers->check_reshape == NULL)
da943b99
N		 6172 return -EINVAL;
6173 if (raid_disks <= 0 ||
233fca36 6174 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 6175 return -EINVAL;
63c70c4f 6176 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 6177 return -EBUSY;
c6563a8c
N		 6178
6179 rdev_for_each(rdev, mddev) {
6180 if (mddev->raid_disks < raid_disks &&
6181 rdev->data_offset < rdev->new_data_offset)
6182 return -EINVAL;
6183 if (mddev->raid_disks > raid_disks &&
6184 rdev->data_offset > rdev->new_data_offset)
6185 return -EINVAL;
6186 }
6187
63c70c4f 6188 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N		 6189 if (mddev->delta_disks < 0)
6190 mddev->reshape_backwards = 1;
6191 else if (mddev->delta_disks > 0)
6192 mddev->reshape_backwards = 0;
63c70c4f
N		 6193
6194 rv = mddev->pers->check_reshape(mddev);
2c810cdd 6195 if (rv < 0) {
de171cb9 6196 mddev->delta_disks = 0;
2c810cdd
N		 6197 mddev->reshape_backwards = 0;
6198 }
da943b99
N		 6199 return rv;
6200}
6201
6202
1da177e4
LT		 6203/*
6204 * update_array_info is used to change the configuration of an
6205 * on-line array.
6206 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6207 * fields in the info are checked against the array.
6208 * Any differences that cannot be handled will cause an error.
6209 * Normally, only one change can be managed at a time.
6210 */
fd01b88c 6211static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6212{
6213 int rv = 0;
6214 int cnt = 0;
36fa3063
N		 6215 int state = 0;
6216
6217 /* calculate expected state,ignoring low bits */
c3d9714e 6218 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 6219 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 6220
6221 if (mddev->major_version != info->major_version ||
6222 mddev->minor_version != info->minor_version ||
6223/* mddev->patch_version != info->patch_version || */
6224 mddev->ctime != info->ctime ||
6225 mddev->level != info->level ||
6226/* mddev->layout != info->layout || */
6227 !mddev->persistent != info->not_persistent||
9d8f0363 6228 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 6229 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6230 ((state^info->state) & 0xfffffe00)
6231 )
1da177e4
LT		 6232 return -EINVAL;
6233 /* Check there is only one change */
58c0fed4
AN		 6234 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6235 cnt++;
6236 if (mddev->raid_disks != info->raid_disks)
6237 cnt++;
6238 if (mddev->layout != info->layout)
6239 cnt++;
6240 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6241 cnt++;
6242 if (cnt == 0)
6243 return 0;
6244 if (cnt > 1)
6245 return -EINVAL;
1da177e4
LT		 6246
6247 if (mddev->layout != info->layout) {
6248 /* Change layout
6249 * we don't need to do anything at the md level, the
6250 * personality will take care of it all.
6251 */
50ac168a 6252 if (mddev->pers->check_reshape == NULL)
1da177e4 6253 return -EINVAL;
597a711b
N		 6254 else {
6255 mddev->new_layout = info->layout;
50ac168a 6256 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 6257 if (rv)
6258 mddev->new_layout = mddev->layout;
6259 return rv;
6260 }
1da177e4 6261 }
58c0fed4 6262 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6263 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6264
da943b99
N		 6265 if (mddev->raid_disks != info->raid_disks)
6266 rv = update_raid_disks(mddev, info->raid_disks);
6267
36fa3063
N		 6268 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6269 if (mddev->pers->quiesce == NULL)
6270 return -EINVAL;
6271 if (mddev->recovery || mddev->sync_thread)
6272 return -EBUSY;
6273 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6274 /* add the bitmap */
6275 if (mddev->bitmap)
6276 return -EEXIST;
c3d9714e 6277 if (mddev->bitmap_info.default_offset == 0)
36fa3063 6278 return -EINVAL;
c3d9714e
N		 6279 mddev->bitmap_info.offset =
6280 mddev->bitmap_info.default_offset;
6409bb05
N		 6281 mddev->bitmap_info.space =
6282 mddev->bitmap_info.default_space;
36fa3063
N		 6283 mddev->pers->quiesce(mddev, 1);
6284 rv = bitmap_create(mddev);
69e51b44
N		 6285 if (!rv)
6286 rv = bitmap_load(mddev);
36fa3063
N		 6287 if (rv)
6288 bitmap_destroy(mddev);
6289 mddev->pers->quiesce(mddev, 0);
6290 } else {
6291 /* remove the bitmap */
6292 if (!mddev->bitmap)
6293 return -ENOENT;
1ec885cd 6294 if (mddev->bitmap->storage.file)
36fa3063
N		 6295 return -EINVAL;
6296 mddev->pers->quiesce(mddev, 1);
6297 bitmap_destroy(mddev);
6298 mddev->pers->quiesce(mddev, 0);
c3d9714e 6299 mddev->bitmap_info.offset = 0;
36fa3063
N		 6300 }
6301 }
850b2b42 6302 md_update_sb(mddev, 1);
1da177e4
LT		 6303 return rv;
6304}
6305
fd01b88c 6306static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6307{
3cb03002 6308 struct md_rdev *rdev;
1ca69c4b 6309 int err = 0;
1da177e4
LT		 6310
6311 if (mddev->pers == NULL)
6312 return -ENODEV;
6313
1ca69c4b
N		 6314 rcu_read_lock();
6315 rdev = find_rdev_rcu(mddev, dev);
1da177e4 6316 if (!rdev)
1ca69c4b
N		 6317 err = -ENODEV;
6318 else {
6319 md_error(mddev, rdev);
6320 if (!test_bit(Faulty, &rdev->flags))
6321 err = -EBUSY;
6322 }
6323 rcu_read_unlock();
6324 return err;
1da177e4
LT		 6325}
6326
2f9618ce
AN		 6327/*
6328 * We have a problem here : there is no easy way to give a CHS
6329 * virtual geometry. We currently pretend that we have a 2 heads
6330 * 4 sectors (with a BIG number of cylinders...). This drives
6331 * dosfs just mad... ;-)
6332 */
a885c8c4
CH		 6333static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6334{
fd01b88c 6335 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6336
6337 geo->heads = 2;
6338 geo->sectors = 4;
49ce6cea 6339 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6340 return 0;
6341}
6342
a39907fa 6343static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6344 unsigned int cmd, unsigned long arg)
6345{
6346 int err = 0;
6347 void __user *argp = (void __user *)arg;
fd01b88c 6348 struct mddev *mddev = NULL;
e2218350 6349 int ro;
1da177e4 6350
506c9e44
N		 6351 switch (cmd) {
6352 case RAID_VERSION:
6353 case GET_ARRAY_INFO:
6354 case GET_DISK_INFO:
6355 break;
6356 default:
6357 if (!capable(CAP_SYS_ADMIN))
6358 return -EACCES;
6359 }
1da177e4
LT		 6360
6361 /*
6362 * Commands dealing with the RAID driver but not any
6363 * particular array:
6364 */
c02c0aeb
N		 6365 switch (cmd) {
6366 case RAID_VERSION:
6367 err = get_version(argp);
6368 goto done;
1da177e4 6369
c02c0aeb
N		 6370 case PRINT_RAID_DEBUG:
6371 err = 0;
6372 md_print_devices();
6373 goto done;
1da177e4
LT		 6374
6375#ifndef MODULE
c02c0aeb
N		 6376 case RAID_AUTORUN:
6377 err = 0;
6378 autostart_arrays(arg);
6379 goto done;
1da177e4 6380#endif
c02c0aeb 6381 default:;
1da177e4
LT		 6382 }
6383
6384 /*
6385 * Commands creating/starting a new array:
6386 */
6387
a39907fa 6388 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6389
6390 if (!mddev) {
6391 BUG();
6392 goto abort;
6393 }
6394
1ca69c4b
N		 6395 /* Some actions do not requires the mutex */
6396 switch (cmd) {
6397 case GET_ARRAY_INFO:
6398 if (!mddev->raid_disks && !mddev->external)
6399 err = -ENODEV;
6400 else
6401 err = get_array_info(mddev, argp);
6402 goto abort;
6403
6404 case GET_DISK_INFO:
6405 if (!mddev->raid_disks && !mddev->external)
6406 err = -ENODEV;
6407 else
6408 err = get_disk_info(mddev, argp);
6409 goto abort;
6410
6411 case SET_DISK_FAULTY:
6412 err = set_disk_faulty(mddev, new_decode_dev(arg));
6413 goto abort;
6414 }
6415
a7a3f08d
N		 6416 if (cmd == ADD_NEW_DISK)
6417 /* need to ensure md_delayed_delete() has completed */
6418 flush_workqueue(md_misc_wq);
6419
90f5f7ad
HR		 6420 if (cmd == HOT_REMOVE_DISK)
6421 /* need to ensure recovery thread has run */
6422 wait_event_interruptible_timeout(mddev->sb_wait,
6423 !test_bit(MD_RECOVERY_NEEDED,
6424 &mddev->flags),
6425 msecs_to_jiffies(5000));
260fa034
N		 6426 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6427 /* Need to flush page cache, and ensure no-one else opens
6428 * and writes
6429 */
6430 mutex_lock(&mddev->open_mutex);
6431 if (atomic_read(&mddev->openers) > 1) {
6432 mutex_unlock(&mddev->open_mutex);
6433 err = -EBUSY;
6434 goto abort;
6435 }
6436 set_bit(MD_STILL_CLOSED, &mddev->flags);
6437 mutex_unlock(&mddev->open_mutex);
6438 sync_blockdev(bdev);
6439 }
1da177e4
LT		 6440 err = mddev_lock(mddev);
6441 if (err) {
6442 printk(KERN_INFO
6443 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6444 err, cmd);
6445 goto abort;
6446 }
6447
c02c0aeb
N		 6448 if (cmd == SET_ARRAY_INFO) {
6449 mdu_array_info_t info;
6450 if (!arg)
6451 memset(&info, 0, sizeof(info));
6452 else if (copy_from_user(&info, argp, sizeof(info))) {
6453 err = -EFAULT;
6454 goto abort_unlock;
6455 }
6456 if (mddev->pers) {
6457 err = update_array_info(mddev, &info);
6458 if (err) {
6459 printk(KERN_WARNING "md: couldn't update"
6460 " array info. %d\n", err);
6461 goto abort_unlock;
1da177e4
LT		 6462 }
6463 goto done_unlock;
c02c0aeb
N		 6464 }
6465 if (!list_empty(&mddev->disks)) {
6466 printk(KERN_WARNING
6467 "md: array %s already has disks!\n",
6468 mdname(mddev));
6469 err = -EBUSY;
6470 goto abort_unlock;
6471 }
6472 if (mddev->raid_disks) {
6473 printk(KERN_WARNING
6474 "md: array %s already initialised!\n",
6475 mdname(mddev));
6476 err = -EBUSY;
6477 goto abort_unlock;
6478 }
6479 err = set_array_info(mddev, &info);
6480 if (err) {
6481 printk(KERN_WARNING "md: couldn't set"
6482 " array info. %d\n", err);
6483 goto abort_unlock;
6484 }
6485 goto done_unlock;
1da177e4
LT		 6486 }
6487
6488 /*
6489 * Commands querying/configuring an existing array:
6490 */
32a7627c 6491 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6492 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6493 if ((!mddev->raid_disks && !mddev->external)
6494 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6495 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6496 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 6497 err = -ENODEV;
6498 goto abort_unlock;
6499 }
6500
6501 /*
6502 * Commands even a read-only array can execute:
6503 */
c02c0aeb
N		 6504 switch (cmd) {
6505 case GET_BITMAP_FILE:
6506 err = get_bitmap_file(mddev, argp);
6507 goto done_unlock;
32a7627c 6508
c02c0aeb
N		 6509 case RESTART_ARRAY_RW:
6510 err = restart_array(mddev);
6511 goto done_unlock;
1da177e4 6512
c02c0aeb
N		 6513 case STOP_ARRAY:
6514 err = do_md_stop(mddev, 0, bdev);
6515 goto done_unlock;
1da177e4 6516
c02c0aeb
N		 6517 case STOP_ARRAY_RO:
6518 err = md_set_readonly(mddev, bdev);
6519 goto done_unlock;
1da177e4 6520
3ea8929d
N		 6521 case HOT_REMOVE_DISK:
6522 err = hot_remove_disk(mddev, new_decode_dev(arg));
6523 goto done_unlock;
6524
7ceb17e8
N		 6525 case ADD_NEW_DISK:
6526 /* We can support ADD_NEW_DISK on read-only arrays
6527 * on if we are re-adding a preexisting device.
6528 * So require mddev->pers and MD_DISK_SYNC.
6529 */
6530 if (mddev->pers) {
6531 mdu_disk_info_t info;
6532 if (copy_from_user(&info, argp, sizeof(info)))
6533 err = -EFAULT;
6534 else if (!(info.state & (1<<MD_DISK_SYNC)))
6535 /* Need to clear read-only for this */
6536 break;
6537 else
6538 err = add_new_disk(mddev, &info);
6539 goto done_unlock;
6540 }
6541 break;
6542
c02c0aeb
N		 6543 case BLKROSET:
6544 if (get_user(ro, (int __user *)(arg))) {
6545 err = -EFAULT;
6546 goto done_unlock;
6547 }
6548 err = -EINVAL;
e2218350 6549
c02c0aeb
N		 6550 /* if the bdev is going readonly the value of mddev->ro
6551 * does not matter, no writes are coming
6552 */
6553 if (ro)
6554 goto done_unlock;
e2218350 6555
c02c0aeb
N		 6556 /* are we are already prepared for writes? */
6557 if (mddev->ro != 1)
6558 goto done_unlock;
e2218350 6559
c02c0aeb
N		 6560 /* transitioning to readauto need only happen for
6561 * arrays that call md_write_start
6562 */
6563 if (mddev->pers) {
6564 err = restart_array(mddev);
6565 if (err == 0) {
6566 mddev->ro = 2;
6567 set_disk_ro(mddev->gendisk, 0);
e2218350 6568 }
c02c0aeb
N		 6569 }
6570 goto done_unlock;
1da177e4
LT		 6571 }
6572
6573 /*
6574 * The remaining ioctls are changing the state of the
f91de92e
N		 6575 * superblock, so we do not allow them on read-only arrays.
6576 * However non-MD ioctls (e.g. get-size) will still come through
6577 * here and hit the 'default' below, so only disallow
6578 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 6579 */
bb57fc64 6580 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 6581 if (mddev->ro == 2) {
6582 mddev->ro = 0;
00bcb4ac 6583 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 6584 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N		 6585 /* mddev_unlock will wake thread */
6586 /* If a device failed while we were read-only, we
6587 * need to make sure the metadata is updated now.
6588 */
6589 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6590 mddev_unlock(mddev);
6591 wait_event(mddev->sb_wait,
6592 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6593 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6594 mddev_lock(mddev);
6595 }
f91de92e
N		 6596 } else {
6597 err = -EROFS;
6598 goto abort_unlock;
6599 }
1da177e4
LT		 6600 }
6601
c02c0aeb
N		 6602 switch (cmd) {
6603 case ADD_NEW_DISK:
1da177e4 6604 {
c02c0aeb
N		 6605 mdu_disk_info_t info;
6606 if (copy_from_user(&info, argp, sizeof(info)))
6607 err = -EFAULT;
6608 else
6609 err = add_new_disk(mddev, &info);
6610 goto done_unlock;
6611 }
1da177e4 6612
c02c0aeb
N		 6613 case HOT_ADD_DISK:
6614 err = hot_add_disk(mddev, new_decode_dev(arg));
6615 goto done_unlock;
1da177e4 6616
c02c0aeb
N		 6617 case RUN_ARRAY:
6618 err = do_md_run(mddev);
6619 goto done_unlock;
1da177e4 6620
c02c0aeb
N		 6621 case SET_BITMAP_FILE:
6622 err = set_bitmap_file(mddev, (int)arg);
6623 goto done_unlock;
32a7627c 6624
c02c0aeb
N		 6625 default:
6626 err = -EINVAL;
6627 goto abort_unlock;
1da177e4
LT		 6628 }
6629
6630done_unlock:
6631abort_unlock:
d3374825
N		 6632 if (mddev->hold_active == UNTIL_IOCTL &&
6633 err != -EINVAL)
6634 mddev->hold_active = 0;
1da177e4
LT		 6635 mddev_unlock(mddev);
6636
6637 return err;
6638done:
6639 if (err)
6640 MD_BUG();
6641abort:
6642 return err;
6643}
aa98aa31
AB		 6644#ifdef CONFIG_COMPAT
6645static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6646 unsigned int cmd, unsigned long arg)
6647{
6648 switch (cmd) {
6649 case HOT_REMOVE_DISK:
6650 case HOT_ADD_DISK:
6651 case SET_DISK_FAULTY:
6652 case SET_BITMAP_FILE:
6653 /* These take in integer arg, do not convert */
6654 break;
6655 default:
6656 arg = (unsigned long)compat_ptr(arg);
6657 break;
6658 }
6659
6660 return md_ioctl(bdev, mode, cmd, arg);
6661}
6662#endif /* CONFIG_COMPAT */
1da177e4 6663
a39907fa 6664static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6665{
6666 /*
6667 * Succeed if we can lock the mddev, which confirms that
6668 * it isn't being stopped right now.
6669 */
fd01b88c 6670 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6671 int err;
6672
0c098220
YL		 6673 if (!mddev)
6674 return -ENODEV;
6675
d3374825
N		 6676 if (mddev->gendisk != bdev->bd_disk) {
6677 /* we are racing with mddev_put which is discarding this
6678 * bd_disk.
6679 */
6680 mddev_put(mddev);
6681 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6682 flush_workqueue(md_misc_wq);
d3374825
N		 6683 /* Then retry the open from the top */
6684 return -ERESTARTSYS;
6685 }
6686 BUG_ON(mddev != bdev->bd_disk->private_data);
6687
c8c00a69 6688 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6689 goto out;
6690
6691 err = 0;
f2ea68cf 6692 atomic_inc(&mddev->openers);
260fa034 6693 clear_bit(MD_STILL_CLOSED, &mddev->flags);
c8c00a69 6694 mutex_unlock(&mddev->open_mutex);
1da177e4 6695
f0b4f7e2 6696 check_disk_change(bdev);
1da177e4
LT		 6697 out:
6698 return err;
6699}
6700
db2a144b 6701static void md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6702{
fd01b88c 6703 struct mddev *mddev = disk->private_data;
1da177e4 6704
52e5f9d1 6705 BUG_ON(!mddev);
f2ea68cf 6706 atomic_dec(&mddev->openers);
1da177e4 6707 mddev_put(mddev);
1da177e4 6708}
f0b4f7e2
N		 6709
6710static int md_media_changed(struct gendisk *disk)
6711{
fd01b88c 6712 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6713
6714 return mddev->changed;
6715}
6716
6717static int md_revalidate(struct gendisk *disk)
6718{
fd01b88c 6719 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6720
6721 mddev->changed = 0;
6722 return 0;
6723}
83d5cde4 6724static const struct block_device_operations md_fops =
1da177e4
LT		 6725{
6726 .owner = THIS_MODULE,
a39907fa
AV		 6727 .open = md_open,
6728 .release = md_release,
b492b852 6729 .ioctl = md_ioctl,
aa98aa31
AB		 6730#ifdef CONFIG_COMPAT
6731 .compat_ioctl = md_compat_ioctl,
6732#endif
a885c8c4 6733 .getgeo = md_getgeo,
f0b4f7e2
N		 6734 .media_changed = md_media_changed,
6735 .revalidate_disk= md_revalidate,
1da177e4
LT		 6736};
6737
75c96f85 6738static int md_thread(void * arg)
1da177e4 6739{
2b8bf345 6740 struct md_thread *thread = arg;
1da177e4 6741
1da177e4
LT		 6742 /*
6743 * md_thread is a 'system-thread', it's priority should be very
6744 * high. We avoid resource deadlocks individually in each
6745 * raid personality. (RAID5 does preallocation) We also use RR and
6746 * the very same RT priority as kswapd, thus we will never get
6747 * into a priority inversion deadlock.
6748 *
6749 * we definitely have to have equal or higher priority than
6750 * bdflush, otherwise bdflush will deadlock if there are too
6751 * many dirty RAID5 blocks.
6752 */
1da177e4 6753
6985c43f 6754 allow_signal(SIGKILL);
a6fb0934 6755 while (!kthread_should_stop()) {
1da177e4 6756
93588e22
N		 6757 /* We need to wait INTERRUPTIBLE so that
6758 * we don't add to the load-average.
6759 * That means we need to be sure no signals are
6760 * pending
6761 */
6762 if (signal_pending(current))
6763 flush_signals(current);
6764
6765 wait_event_interruptible_timeout
6766 (thread->wqueue,
6767 test_bit(THREAD_WAKEUP, &thread->flags)
6768 || kthread_should_stop(),
6769 thread->timeout);
1da177e4 6770
6c987910
N		 6771 clear_bit(THREAD_WAKEUP, &thread->flags);
6772 if (!kthread_should_stop())
4ed8731d 6773 thread->run(thread);
1da177e4 6774 }
a6fb0934 6775
1da177e4
LT		 6776 return 0;
6777}
6778
2b8bf345 6779void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6780{
6781 if (thread) {
36a4e1fe 6782 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6783 set_bit(THREAD_WAKEUP, &thread->flags);
6784 wake_up(&thread->wqueue);
6785 }
6786}
6787
4ed8731d
SL		 6788struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6789 struct mddev *mddev, const char *name)
1da177e4 6790{
2b8bf345 6791 struct md_thread *thread;
1da177e4 6792
2b8bf345 6793 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6794 if (!thread)
6795 return NULL;
6796
1da177e4
LT		 6797 init_waitqueue_head(&thread->wqueue);
6798
1da177e4
LT		 6799 thread->run = run;
6800 thread->mddev = mddev;
32a7627c 6801 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6802 thread->tsk = kthread_run(md_thread, thread,
6803 "%s_%s",
6804 mdname(thread->mddev),
0232605d 6805 name);
a6fb0934 6806 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6807 kfree(thread);
6808 return NULL;
6809 }
1da177e4
LT		 6810 return thread;
6811}
6812
2b8bf345 6813void md_unregister_thread(struct md_thread **threadp)
1da177e4 6814{
2b8bf345 6815 struct md_thread *thread = *threadp;
e0cf8f04
N		 6816 if (!thread)
6817 return;
36a4e1fe 6818 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 6819 /* Locking ensures that mddev_unlock does not wake_up a
6820 * non-existent thread
6821 */
6822 spin_lock(&pers_lock);
6823 *threadp = NULL;
6824 spin_unlock(&pers_lock);
a6fb0934
N		 6825
6826 kthread_stop(thread->tsk);
1da177e4
LT		 6827 kfree(thread);
6828}
6829
fd01b88c 6830void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 6831{
6832 if (!mddev) {
6833 MD_BUG();
6834 return;
6835 }
6836
b2d444d7 6837 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6838 return;
6bfe0b49 6839
de393cde 6840 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 6841 return;
6842 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6843 if (mddev->degraded)
6844 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6845 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6846 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6847 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6848 md_wakeup_thread(mddev->thread);
768a418d 6849 if (mddev->event_work.func)
e804ac78 6850 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6851 md_new_event_inintr(mddev);
1da177e4
LT		 6852}
6853
6854/* seq_file implementation /proc/mdstat */
6855
6856static void status_unused(struct seq_file *seq)
6857{
6858 int i = 0;
3cb03002 6859 struct md_rdev *rdev;
1da177e4
LT		 6860
6861 seq_printf(seq, "unused devices: ");
6862
159ec1fc 6863 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6864 char b[BDEVNAME_SIZE];
6865 i++;
6866 seq_printf(seq, "%s ",
6867 bdevname(rdev->bdev,b));
6868 }
6869 if (!i)
6870 seq_printf(seq, "<none>");
6871
6872 seq_printf(seq, "\n");
6873}
6874
6875
fd01b88c 6876static void status_resync(struct seq_file *seq, struct mddev * mddev)
1da177e4 6877{
dd71cf6b
N		 6878 sector_t max_sectors, resync, res;
6879 unsigned long dt, db;
6880 sector_t rt;
4588b42e
N		 6881 int scale;
6882 unsigned int per_milli;
1da177e4 6883
72f36d59
N		 6884 if (mddev->curr_resync <= 3)
6885 resync = 0;
6886 else
6887 resync = mddev->curr_resync
6888 - atomic_read(&mddev->recovery_active);
1da177e4 6889
c804cdec
N		 6890 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6891 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 6892 max_sectors = mddev->resync_max_sectors;
1da177e4 6893 else
dd71cf6b 6894 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6895
6896 /*
6897 * Should not happen.
6898 */
dd71cf6b 6899 if (!max_sectors) {
1da177e4
LT		 6900 MD_BUG();
6901 return;
6902 }
4588b42e 6903 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6904 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6905 * u32, as those are the requirements for sector_div.
6906 * Thus 'scale' must be at least 10
6907 */
6908 scale = 10;
6909 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6910 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6911 scale++;
6912 }
6913 res = (resync>>scale)*1000;
dd71cf6b 6914 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6915
6916 per_milli = res;
1da177e4 6917 {
4588b42e 6918 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6919 seq_printf(seq, "[");
6920 for (i = 0; i < x; i++)
6921 seq_printf(seq, "=");
6922 seq_printf(seq, ">");
6923 for (i = 0; i < y; i++)
6924 seq_printf(seq, ".");
6925 seq_printf(seq, "] ");
6926 }
4588b42e 6927 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6928 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6929 "reshape" :
61df9d91
N		 6930 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6931 "check" :
6932 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6933 "resync" : "recovery"))),
6934 per_milli/10, per_milli % 10,
dd71cf6b
N		 6935 (unsigned long long) resync/2,
6936 (unsigned long long) max_sectors/2);
1da177e4
LT		 6937
6938 /*
1da177e4
LT		 6939 * dt: time from mark until now
6940 * db: blocks written from mark until now
6941 * rt: remaining time
dd71cf6b
N		 6942 *
6943 * rt is a sector_t, so could be 32bit or 64bit.
6944 * So we divide before multiply in case it is 32bit and close
6945 * to the limit.
25985edc 6946 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 6947 * near the end of resync when the number of remaining sectors
6948 * is close to 'db'.
6949 * We then divide rt by 32 after multiplying by db to compensate.
6950 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6951 */
6952 dt = ((jiffies - mddev->resync_mark) / HZ);
6953 if (!dt) dt++;
ff4e8d9a
N		 6954 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6955 - mddev->resync_mark_cnt;
1da177e4 6956
dd71cf6b
N		 6957 rt = max_sectors - resync; /* number of remaining sectors */
6958 sector_div(rt, db/32+1);
6959 rt *= dt;
6960 rt >>= 5;
6961
6962 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6963 ((unsigned long)rt % 60)/6);
1da177e4 6964
ff4e8d9a 6965 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6966}
6967
6968static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6969{
6970 struct list_head *tmp;
6971 loff_t l = *pos;
fd01b88c 6972 struct mddev *mddev;
1da177e4
LT		 6973
6974 if (l >= 0x10000)
6975 return NULL;
6976 if (!l--)
6977 /* header */
6978 return (void*)1;
6979
6980 spin_lock(&all_mddevs_lock);
6981 list_for_each(tmp,&all_mddevs)
6982 if (!l--) {
fd01b88c 6983 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 6984 mddev_get(mddev);
6985 spin_unlock(&all_mddevs_lock);
6986 return mddev;
6987 }
6988 spin_unlock(&all_mddevs_lock);
6989 if (!l--)
6990 return (void*)2;/* tail */
6991 return NULL;
6992}
6993
6994static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6995{
6996 struct list_head *tmp;
fd01b88c 6997 struct mddev *next_mddev, *mddev = v;
1da177e4
LT		 6998
6999 ++*pos;
7000 if (v == (void*)2)
7001 return NULL;
7002
7003 spin_lock(&all_mddevs_lock);
7004 if (v == (void*)1)
7005 tmp = all_mddevs.next;
7006 else
7007 tmp = mddev->all_mddevs.next;
7008 if (tmp != &all_mddevs)
fd01b88c 7009 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 7010 else {
7011 next_mddev = (void*)2;
7012 *pos = 0x10000;
7013 }
7014 spin_unlock(&all_mddevs_lock);
7015
7016 if (v != (void*)1)
7017 mddev_put(mddev);
7018 return next_mddev;
7019
7020}
7021
7022static void md_seq_stop(struct seq_file *seq, void *v)
7023{
fd01b88c 7024 struct mddev *mddev = v;
1da177e4
LT		 7025
7026 if (mddev && v != (void*)1 && v != (void*)2)
7027 mddev_put(mddev);
7028}
7029
7030static int md_seq_show(struct seq_file *seq, void *v)
7031{
fd01b88c 7032 struct mddev *mddev = v;
dd8ac336 7033 sector_t sectors;
3cb03002 7034 struct md_rdev *rdev;
1da177e4
LT		 7035
7036 if (v == (void*)1) {
84fc4b56 7037 struct md_personality *pers;
1da177e4
LT		 7038 seq_printf(seq, "Personalities : ");
7039 spin_lock(&pers_lock);
2604b703
N		 7040 list_for_each_entry(pers, &pers_list, list)
7041 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 7042
7043 spin_unlock(&pers_lock);
7044 seq_printf(seq, "\n");
f1514638 7045 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7046 return 0;
7047 }
7048 if (v == (void*)2) {
7049 status_unused(seq);
7050 return 0;
7051 }
7052
5dc5cf7d 7053 if (mddev_lock(mddev) < 0)
1da177e4 7054 return -EINTR;
5dc5cf7d 7055
1da177e4
LT		 7056 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7057 seq_printf(seq, "%s : %sactive", mdname(mddev),
7058 mddev->pers ? "" : "in");
7059 if (mddev->pers) {
f91de92e 7060 if (mddev->ro==1)
1da177e4 7061 seq_printf(seq, " (read-only)");
f91de92e 7062 if (mddev->ro==2)
52720ae7 7063 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 7064 seq_printf(seq, " %s", mddev->pers->name);
7065 }
7066
dd8ac336 7067 sectors = 0;
dafb20fa 7068 rdev_for_each(rdev, mddev) {
1da177e4
LT		 7069 char b[BDEVNAME_SIZE];
7070 seq_printf(seq, " %s[%d]",
7071 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 7072 if (test_bit(WriteMostly, &rdev->flags))
7073 seq_printf(seq, "(W)");
b2d444d7 7074 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7075 seq_printf(seq, "(F)");
7076 continue;
2d78f8c4
N		 7077 }
7078 if (rdev->raid_disk < 0)
b325a32e 7079 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 7080 if (test_bit(Replacement, &rdev->flags))
7081 seq_printf(seq, "(R)");
dd8ac336 7082 sectors += rdev->sectors;
1da177e4
LT		 7083 }
7084
7085 if (!list_empty(&mddev->disks)) {
7086 if (mddev->pers)
7087 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 7088 (unsigned long long)
7089 mddev->array_sectors / 2);
1da177e4
LT		 7090 else
7091 seq_printf(seq, "\n %llu blocks",
dd8ac336 7092 (unsigned long long)sectors / 2);
1da177e4 7093 }
1cd6bf19
N		 7094 if (mddev->persistent) {
7095 if (mddev->major_version != 0 ||
7096 mddev->minor_version != 90) {
7097 seq_printf(seq," super %d.%d",
7098 mddev->major_version,
7099 mddev->minor_version);
7100 }
e691063a
N		 7101 } else if (mddev->external)
7102 seq_printf(seq, " super external:%s",
7103 mddev->metadata_type);
7104 else
1cd6bf19 7105 seq_printf(seq, " super non-persistent");
1da177e4
LT		 7106
7107 if (mddev->pers) {
d710e138 7108 mddev->pers->status(seq, mddev);
1da177e4 7109 seq_printf(seq, "\n ");
8e1b39d6
N		 7110 if (mddev->pers->sync_request) {
7111 if (mddev->curr_resync > 2) {
d710e138 7112 status_resync(seq, mddev);
8e1b39d6 7113 seq_printf(seq, "\n ");
72f36d59 7114 } else if (mddev->curr_resync >= 1)
8e1b39d6
N		 7115 seq_printf(seq, "\tresync=DELAYED\n ");
7116 else if (mddev->recovery_cp < MaxSector)
7117 seq_printf(seq, "\tresync=PENDING\n ");
7118 }
32a7627c
N		 7119 } else
7120 seq_printf(seq, "\n ");
7121
57148964 7122 bitmap_status(seq, mddev->bitmap);
1da177e4
LT		 7123
7124 seq_printf(seq, "\n");
7125 }
7126 mddev_unlock(mddev);
7127
7128 return 0;
7129}
7130
110518bc 7131static const struct seq_operations md_seq_ops = {
1da177e4
LT		 7132 .start = md_seq_start,
7133 .next = md_seq_next,
7134 .stop = md_seq_stop,
7135 .show = md_seq_show,
7136};
7137
7138static int md_seq_open(struct inode *inode, struct file *file)
7139{
f1514638 7140 struct seq_file *seq;
1da177e4
LT		 7141 int error;
7142
7143 error = seq_open(file, &md_seq_ops);
d7603b7e 7144 if (error)
f1514638
KS		 7145 return error;
7146
7147 seq = file->private_data;
7148 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7149 return error;
7150}
7151
d7603b7e
N		 7152static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7153{
f1514638 7154 struct seq_file *seq = filp->private_data;
d7603b7e
N		 7155 int mask;
7156
7157 poll_wait(filp, &md_event_waiters, wait);
7158
7159 /* always allow read */
7160 mask = POLLIN | POLLRDNORM;
7161
f1514638 7162 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 7163 mask |= POLLERR | POLLPRI;
7164 return mask;
7165}
7166
fa027c2a 7167static const struct file_operations md_seq_fops = {
e24650c2 7168 .owner = THIS_MODULE,
1da177e4
LT		 7169 .open = md_seq_open,
7170 .read = seq_read,
7171 .llseek = seq_lseek,
c3f94b40 7172 .release = seq_release_private,
d7603b7e 7173 .poll = mdstat_poll,
1da177e4
LT		 7174};
7175
84fc4b56 7176int register_md_personality(struct md_personality *p)
1da177e4 7177{
1da177e4 7178 spin_lock(&pers_lock);
2604b703
N		 7179 list_add_tail(&p->list, &pers_list);
7180 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 7181 spin_unlock(&pers_lock);
7182 return 0;
7183}
7184
84fc4b56 7185int unregister_md_personality(struct md_personality *p)
1da177e4 7186{
2604b703 7187 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7188 spin_lock(&pers_lock);
2604b703 7189 list_del_init(&p->list);
1da177e4
LT		 7190 spin_unlock(&pers_lock);
7191 return 0;
7192}
7193
fd01b88c 7194static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7195{
3cb03002 7196 struct md_rdev * rdev;
1da177e4 7197 int idle;
eea1bf38 7198 int curr_events;
1da177e4
LT		 7199
7200 idle = 1;
4b80991c
N		 7201 rcu_read_lock();
7202 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7203 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 7204 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7205 (int)part_stat_read(&disk->part0, sectors[1]) -
7206 atomic_read(&disk->sync_io);
713f6ab1
N		 7207 /* sync IO will cause sync_io to increase before the disk_stats
7208 * as sync_io is counted when a request starts, and
7209 * disk_stats is counted when it completes.
7210 * So resync activity will cause curr_events to be smaller than
7211 * when there was no such activity.
7212 * non-sync IO will cause disk_stat to increase without
7213 * increasing sync_io so curr_events will (eventually)
7214 * be larger than it was before. Once it becomes
7215 * substantially larger, the test below will cause
7216 * the array to appear non-idle, and resync will slow
7217 * down.
7218 * If there is a lot of outstanding resync activity when
7219 * we set last_event to curr_events, then all that activity
7220 * completing might cause the array to appear non-idle
7221 * and resync will be slowed down even though there might
7222 * not have been non-resync activity. This will only
7223 * happen once though. 'last_events' will soon reflect
7224 * the state where there is little or no outstanding
7225 * resync requests, and further resync activity will
7226 * always make curr_events less than last_events.
c0e48521 7227 *
1da177e4 7228 */
eea1bf38 7229 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 7230 rdev->last_events = curr_events;
7231 idle = 0;
7232 }
7233 }
4b80991c 7234 rcu_read_unlock();
1da177e4
LT		 7235 return idle;
7236}
7237
fd01b88c 7238void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 7239{
7240 /* another "blocks" (512byte) blocks have been synced */
7241 atomic_sub(blocks, &mddev->recovery_active);
7242 wake_up(&mddev->recovery_wait);
7243 if (!ok) {
dfc70645 7244 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 7245 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT		 7246 md_wakeup_thread(mddev->thread);
7247 // stop recovery, signal do_sync ....
7248 }
7249}
7250
7251
06d91a5f
N		 7252/* md_write_start(mddev, bi)
7253 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 7254 * in superblock) before writing, schedule a superblock update
7255 * and wait for it to complete.
06d91a5f 7256 */
fd01b88c 7257void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7258{
0fd62b86 7259 int did_change = 0;
06d91a5f 7260 if (bio_data_dir(bi) != WRITE)
3d310eb7 7261 return;
06d91a5f 7262
f91de92e
N		 7263 BUG_ON(mddev->ro == 1);
7264 if (mddev->ro == 2) {
7265 /* need to switch to read/write */
7266 mddev->ro = 0;
7267 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7268 md_wakeup_thread(mddev->thread);
25156198 7269 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7270 did_change = 1;
f91de92e 7271 }
06d91a5f 7272 atomic_inc(&mddev->writes_pending);
31a59e34
N		 7273 if (mddev->safemode == 1)
7274 mddev->safemode = 0;
06d91a5f 7275 if (mddev->in_sync) {
a9701a30 7276 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 7277 if (mddev->in_sync) {
7278 mddev->in_sync = 0;
850b2b42 7279 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7280 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7281 md_wakeup_thread(mddev->thread);
0fd62b86 7282 did_change = 1;
3d310eb7 7283 }
a9701a30 7284 spin_unlock_irq(&mddev->write_lock);
06d91a5f 7285 }
0fd62b86 7286 if (did_change)
00bcb4ac 7287 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7288 wait_event(mddev->sb_wait,
09a44cc1 7289 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 7290}
7291
fd01b88c 7292void md_write_end(struct mddev *mddev)
1da177e4
LT		 7293{
7294 if (atomic_dec_and_test(&mddev->writes_pending)) {
7295 if (mddev->safemode == 2)
7296 md_wakeup_thread(mddev->thread);
16f17b39 7297 else if (mddev->safemode_delay)
1da177e4
LT		 7298 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7299 }
7300}
7301
2a2275d6
N		 7302/* md_allow_write(mddev)
7303 * Calling this ensures that the array is marked 'active' so that writes
7304 * may proceed without blocking. It is important to call this before
7305 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7306 * Must be called with mddev_lock held.
b5470dc5
DW		 7307 *
7308 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7309 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7310 */
fd01b88c 7311int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7312{
7313 if (!mddev->pers)
b5470dc5 7314 return 0;
2a2275d6 7315 if (mddev->ro)
b5470dc5 7316 return 0;
1a0fd497 7317 if (!mddev->pers->sync_request)
b5470dc5 7318 return 0;
2a2275d6
N		 7319
7320 spin_lock_irq(&mddev->write_lock);
7321 if (mddev->in_sync) {
7322 mddev->in_sync = 0;
7323 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7324 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7325 if (mddev->safemode_delay &&
7326 mddev->safemode == 0)
7327 mddev->safemode = 1;
7328 spin_unlock_irq(&mddev->write_lock);
7329 md_update_sb(mddev, 0);
00bcb4ac 7330 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 7331 } else
7332 spin_unlock_irq(&mddev->write_lock);
b5470dc5 7333
070dc6dd 7334 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7335 return -EAGAIN;
7336 else
7337 return 0;
2a2275d6
N		 7338}
7339EXPORT_SYMBOL_GPL(md_allow_write);
7340
1da177e4
LT		 7341#define SYNC_MARKS 10
7342#define SYNC_MARK_STEP (3*HZ)
54f89341 7343#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 7344void md_do_sync(struct md_thread *thread)
1da177e4 7345{
4ed8731d 7346 struct mddev *mddev = thread->mddev;
fd01b88c 7347 struct mddev *mddev2;
1da177e4
LT		 7348 unsigned int currspeed = 0,
7349 window;
57afd89f 7350 sector_t max_sectors,j, io_sectors;
1da177e4 7351 unsigned long mark[SYNC_MARKS];
54f89341 7352 unsigned long update_time;
1da177e4
LT		 7353 sector_t mark_cnt[SYNC_MARKS];
7354 int last_mark,m;
7355 struct list_head *tmp;
7356 sector_t last_check;
57afd89f 7357 int skipped = 0;
3cb03002 7358 struct md_rdev *rdev;
c4a39551 7359 char *desc, *action = NULL;
7c2c57c9 7360 struct blk_plug plug;
1da177e4
LT		 7361
7362 /* just incase thread restarts... */
7363 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7364 return;
5fd6c1dc
N		 7365 if (mddev->ro) /* never try to sync a read-only array */
7366 return;
1da177e4 7367
61df9d91 7368 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
c4a39551 7369 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
61df9d91 7370 desc = "data-check";
c4a39551
JB		 7371 action = "check";
7372 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
61df9d91 7373 desc = "requested-resync";
c4a39551
JB		 7374 action = "repair";
7375 } else
61df9d91
N		 7376 desc = "resync";
7377 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7378 desc = "reshape";
7379 else
7380 desc = "recovery";
7381
c4a39551
JB		 7382 mddev->last_sync_action = action ?: desc;
7383
1da177e4
LT		 7384 /* we overload curr_resync somewhat here.
7385 * 0 == not engaged in resync at all
7386 * 2 == checking that there is no conflict with another sync
7387 * 1 == like 2, but have yielded to allow conflicting resync to
7388 * commense
7389 * other == active in resync - this many blocks
7390 *
7391 * Before starting a resync we must have set curr_resync to
7392 * 2, and then checked that every "conflicting" array has curr_resync
7393 * less than ours. When we find one that is the same or higher
7394 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7395 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7396 * This will mean we have to start checking from the beginning again.
7397 *
7398 */
7399
7400 do {
7401 mddev->curr_resync = 2;
7402
7403 try_again:
404e4b43 7404 if (kthread_should_stop())
6985c43f 7405 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 7406
7407 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7408 goto skip;
29ac4aa3 7409 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7410 if (mddev2 == mddev)
7411 continue;
90b08710
BS		 7412 if (!mddev->parallel_resync
7413 && mddev2->curr_resync
7414 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7415 DEFINE_WAIT(wq);
7416 if (mddev < mddev2 && mddev->curr_resync == 2) {
7417 /* arbitrarily yield */
7418 mddev->curr_resync = 1;
7419 wake_up(&resync_wait);
7420 }
7421 if (mddev > mddev2 && mddev->curr_resync == 1)
7422 /* no need to wait here, we can wait the next
7423 * time 'round when curr_resync == 2
7424 */
7425 continue;
9744197c
N		 7426 /* We need to wait 'interruptible' so as not to
7427 * contribute to the load average, and not to
7428 * be caught by 'softlockup'
7429 */
7430 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 7431 if (!kthread_should_stop() &&
8712e553 7432 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7433 printk(KERN_INFO "md: delaying %s of %s"
7434 " until %s has finished (they"
1da177e4 7435 " share one or more physical units)\n",
61df9d91 7436 desc, mdname(mddev), mdname(mddev2));
1da177e4 7437 mddev_put(mddev2);
9744197c
N		 7438 if (signal_pending(current))
7439 flush_signals(current);
1da177e4
LT		 7440 schedule();
7441 finish_wait(&resync_wait, &wq);
7442 goto try_again;
7443 }
7444 finish_wait(&resync_wait, &wq);
7445 }
7446 }
7447 } while (mddev->curr_resync < 2);
7448
5fd6c1dc 7449 j = 0;
9d88883e 7450 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7451 /* resync follows the size requested by the personality,
57afd89f 7452 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7453 */
7454 max_sectors = mddev->resync_max_sectors;
7f7583d4 7455 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 7456 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7457 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7458 j = mddev->resync_min;
7459 else if (!mddev->bitmap)
5fd6c1dc 7460 j = mddev->recovery_cp;
5e96ee65 7461
ccfcc3c1 7462 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7463 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7464 else {
1da177e4 7465 /* recovery follows the physical size of devices */
58c0fed4 7466 max_sectors = mddev->dev_sectors;
5fd6c1dc 7467 j = MaxSector;
4e59ca7d 7468 rcu_read_lock();
dafb20fa 7469 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc
N		 7470 if (rdev->raid_disk >= 0 &&
7471 !test_bit(Faulty, &rdev->flags) &&
7472 !test_bit(In_sync, &rdev->flags) &&
7473 rdev->recovery_offset < j)
7474 j = rdev->recovery_offset;
4e59ca7d 7475 rcu_read_unlock();
5fd6c1dc 7476 }
1da177e4 7477
61df9d91
N		 7478 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7479 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7480 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7481 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7482 "(but not more than %d KB/sec) for %s.\n",
7483 speed_max(mddev), desc);
1da177e4 7484
eea1bf38 7485 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7486
57afd89f 7487 io_sectors = 0;
1da177e4
LT		 7488 for (m = 0; m < SYNC_MARKS; m++) {
7489 mark[m] = jiffies;
57afd89f 7490 mark_cnt[m] = io_sectors;
1da177e4
LT		 7491 }
7492 last_mark = 0;
7493 mddev->resync_mark = mark[last_mark];
7494 mddev->resync_mark_cnt = mark_cnt[last_mark];
7495
7496 /*
7497 * Tune reconstruction:
7498 */
7499 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7500 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7501 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7502
7503 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7504 last_check = 0;
7505
7506 if (j>2) {
7507 printk(KERN_INFO
61df9d91
N		 7508 "md: resuming %s of %s from checkpoint.\n",
7509 desc, mdname(mddev));
1da177e4 7510 mddev->curr_resync = j;
72f36d59
N		 7511 } else
7512 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 7513 mddev->curr_resync_completed = j;
72f36d59
N		 7514 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7515 md_new_event(mddev);
54f89341 7516 update_time = jiffies;
1da177e4 7517
7c2c57c9 7518 blk_start_plug(&plug);
1da177e4 7519 while (j < max_sectors) {
57afd89f 7520 sector_t sectors;
1da177e4 7521
57afd89f 7522 skipped = 0;
97e4f42d 7523
7a91ee1f
N		 7524 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7525 ((mddev->curr_resync > mddev->curr_resync_completed &&
7526 (mddev->curr_resync - mddev->curr_resync_completed)
7527 > (max_sectors >> 4)) ||
54f89341 7528 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f
N		 7529 (j - mddev->curr_resync_completed)*2
7530 >= mddev->resync_max - mddev->curr_resync_completed
7531 )) {
97e4f42d 7532 /* time to update curr_resync_completed */
97e4f42d
N		 7533 wait_event(mddev->recovery_wait,
7534 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7535 mddev->curr_resync_completed = j;
35d78c66 7536 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7537 j > mddev->recovery_cp)
7538 mddev->recovery_cp = j;
54f89341 7539 update_time = jiffies;
070dc6dd 7540 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7541 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7542 }
acb180b0 7543
e62e58a5
N		 7544 while (j >= mddev->resync_max && !kthread_should_stop()) {
7545 /* As this condition is controlled by user-space,
7546 * we can block indefinitely, so use '_interruptible'
7547 * to avoid triggering warnings.
7548 */
7549 flush_signals(current); /* just in case */
7550 wait_event_interruptible(mddev->recovery_wait,
7551 mddev->resync_max > j
7552 || kthread_should_stop());
7553 }
acb180b0
N		 7554
7555 if (kthread_should_stop())
7556 goto interrupted;
7557
57afd89f 7558 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7559 currspeed < speed_min(mddev));
57afd89f 7560 if (sectors == 0) {
dfc70645 7561 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 7562 goto out;
7563 }
57afd89f
N		 7564
7565 if (!skipped) { /* actual IO requested */
7566 io_sectors += sectors;
7567 atomic_add(sectors, &mddev->recovery_active);
7568 }
7569
e875ecea
N		 7570 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7571 break;
7572
1da177e4 7573 j += sectors;
72f36d59
N		 7574 if (j > 2)
7575 mddev->curr_resync = j;
ff4e8d9a 7576 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7577 if (last_check == 0)
e875ecea 7578 /* this is the earliest that rebuild will be
d7603b7e
N		 7579 * visible in /proc/mdstat
7580 */
7581 md_new_event(mddev);
57afd89f
N		 7582
7583 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7584 continue;
7585
57afd89f 7586 last_check = io_sectors;
1da177e4
LT		 7587 repeat:
7588 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7589 /* step marks */
7590 int next = (last_mark+1) % SYNC_MARKS;
7591
7592 mddev->resync_mark = mark[next];
7593 mddev->resync_mark_cnt = mark_cnt[next];
7594 mark[next] = jiffies;
57afd89f 7595 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7596 last_mark = next;
7597 }
7598
7599
c6207277
N		 7600 if (kthread_should_stop())
7601 goto interrupted;
7602
1da177e4
LT		 7603
7604 /*
7605 * this loop exits only if either when we are slower than
7606 * the 'hard' speed limit, or the system was IO-idle for
7607 * a jiffy.
7608 * the system might be non-idle CPU-wise, but we only care
7609 * about not overloading the IO subsystem. (things like an
7610 * e2fsck being done on the RAID array should execute fast)
7611 */
1da177e4
LT		 7612 cond_resched();
7613
57afd89f
N		 7614 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7615 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7616
88202a0c
N		 7617 if (currspeed > speed_min(mddev)) {
7618 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7619 !is_mddev_idle(mddev, 0)) {
c0e48521 7620 msleep(500);
1da177e4
LT		 7621 goto repeat;
7622 }
7623 }
7624 }
61df9d91 7625 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 7626 /*
7627 * this also signals 'finished resyncing' to md_stop
7628 */
7629 out:
7c2c57c9 7630 blk_finish_plug(&plug);
1da177e4
LT		 7631 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7632
7633 /* tell personality that we are finished */
57afd89f 7634 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7635
dfc70645 7636 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7637 mddev->curr_resync > 2) {
7638 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7639 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7640 if (mddev->curr_resync >= mddev->recovery_cp) {
7641 printk(KERN_INFO
61df9d91
N		 7642 "md: checkpointing %s of %s.\n",
7643 desc, mdname(mddev));
0a19caab 7644 if (test_bit(MD_RECOVERY_ERROR,
7645 &mddev->recovery))
7646 mddev->recovery_cp =
7647 mddev->curr_resync_completed;
7648 else
7649 mddev->recovery_cp =
7650 mddev->curr_resync;
5fd6c1dc
N		 7651 }
7652 } else
7653 mddev->recovery_cp = MaxSector;
7654 } else {
7655 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7656 mddev->curr_resync = MaxSector;
4e59ca7d 7657 rcu_read_lock();
dafb20fa 7658 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7659 if (rdev->raid_disk >= 0 &&
70fffd0b 7660 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7661 !test_bit(Faulty, &rdev->flags) &&
7662 !test_bit(In_sync, &rdev->flags) &&
7663 rdev->recovery_offset < mddev->curr_resync)
7664 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7665 rcu_read_unlock();
5fd6c1dc 7666 }
1da177e4 7667 }
db91ff55 7668 skip:
17571284 7669 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7670
c07b70ad
N		 7671 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7672 /* We completed so min/max setting can be forgotten if used. */
7673 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7674 mddev->resync_min = 0;
7675 mddev->resync_max = MaxSector;
7676 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7677 mddev->resync_min = mddev->curr_resync_completed;
1da177e4
LT		 7678 mddev->curr_resync = 0;
7679 wake_up(&resync_wait);
7680 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7681 md_wakeup_thread(mddev->thread);
c6207277
N		 7682 return;
7683
7684 interrupted:
7685 /*
7686 * got a signal, exit.
7687 */
7688 printk(KERN_INFO
7689 "md: md_do_sync() got signal ... exiting\n");
7690 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7691 goto out;
7692
1da177e4 7693}
29269553 7694EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7695
746d3207
N		 7696static int remove_and_add_spares(struct mddev *mddev,
7697 struct md_rdev *this)
b4c4c7b8 7698{
3cb03002 7699 struct md_rdev *rdev;
b4c4c7b8 7700 int spares = 0;
f2a371c5 7701 int removed = 0;
b4c4c7b8 7702
dafb20fa 7703 rdev_for_each(rdev, mddev)
746d3207
N		 7704 if ((this == NULL || rdev == this) &&
7705 rdev->raid_disk >= 0 &&
6bfe0b49 7706 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7707 (test_bit(Faulty, &rdev->flags) ||
7708 ! test_bit(In_sync, &rdev->flags)) &&
7709 atomic_read(&rdev->nr_pending)==0) {
7710 if (mddev->pers->hot_remove_disk(
b8321b68 7711 mddev, rdev) == 0) {
36fad858 7712 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 7713 rdev->raid_disk = -1;
f2a371c5 7714 removed++;
b4c4c7b8
N		 7715 }
7716 }
90584fc9
JB		 7717 if (removed && mddev->kobj.sd)
7718 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 7719
746d3207
N		 7720 if (this)
7721 goto no_add;
7722
dafb20fa 7723 rdev_for_each(rdev, mddev) {
7bfec5f3
N		 7724 if (rdev->raid_disk >= 0 &&
7725 !test_bit(In_sync, &rdev->flags) &&
7726 !test_bit(Faulty, &rdev->flags))
7727 spares++;
7ceb17e8
N		 7728 if (rdev->raid_disk >= 0)
7729 continue;
7730 if (test_bit(Faulty, &rdev->flags))
7731 continue;
7732 if (mddev->ro &&
7733 rdev->saved_raid_disk < 0)
7734 continue;
7735
7736 rdev->recovery_offset = 0;
7ceb17e8
N		 7737 if (mddev->pers->
7738 hot_add_disk(mddev, rdev) == 0) {
7739 if (sysfs_link_rdev(mddev, rdev))
7740 /* failure here is OK */;
7741 spares++;
7742 md_new_event(mddev);
7743 set_bit(MD_CHANGE_DEVS, &mddev->flags);
dfc70645 7744 }
b4c4c7b8 7745 }
746d3207 7746no_add:
6dafab6b
N		 7747 if (removed)
7748 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 7749 return spares;
7750}
7ebc0be7 7751
1da177e4
LT		 7752/*
7753 * This routine is regularly called by all per-raid-array threads to
7754 * deal with generic issues like resync and super-block update.
7755 * Raid personalities that don't have a thread (linear/raid0) do not
7756 * need this as they never do any recovery or update the superblock.
7757 *
7758 * It does not do any resync itself, but rather "forks" off other threads
7759 * to do that as needed.
7760 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7761 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7762 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7763 * and wakeups up this thread which will reap the thread and finish up.
7764 * This thread also removes any faulty devices (with nr_pending == 0).
7765 *
7766 * The overall approach is:
7767 * 1/ if the superblock needs updating, update it.
7768 * 2/ If a recovery thread is running, don't do anything else.
7769 * 3/ If recovery has finished, clean up, possibly marking spares active.
7770 * 4/ If there are any faulty devices, remove them.
7771 * 5/ If array is degraded, try to add spares devices
7772 * 6/ If array has spares or is not in-sync, start a resync thread.
7773 */
fd01b88c 7774void md_check_recovery(struct mddev *mddev)
1da177e4 7775{
68866e42
JB		 7776 if (mddev->suspended)
7777 return;
7778
5f40402d 7779 if (mddev->bitmap)
aa5cbd10 7780 bitmap_daemon_work(mddev);
1da177e4 7781
fca4d848 7782 if (signal_pending(current)) {
31a59e34 7783 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7784 printk(KERN_INFO "md: %s in immediate safe mode\n",
7785 mdname(mddev));
7786 mddev->safemode = 2;
7787 }
7788 flush_signals(current);
7789 }
7790
c89a8eee
N		 7791 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7792 return;
1da177e4 7793 if ( ! (
126925c0 7794 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7795 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7796 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7797 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7798 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7799 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7800 ))
7801 return;
fca4d848 7802
df5b89b3 7803 if (mddev_trylock(mddev)) {
b4c4c7b8 7804 int spares = 0;
fca4d848 7805
c89a8eee 7806 if (mddev->ro) {
7ceb17e8
N		 7807 /* On a read-only array we can:
7808 * - remove failed devices
7809 * - add already-in_sync devices if the array itself
7810 * is in-sync.
7811 * As we only add devices that are already in-sync,
7812 * we can activate the spares immediately.
c89a8eee 7813 */
c89a8eee 7814 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7ceb17e8
N		 7815 remove_and_add_spares(mddev, NULL);
7816 mddev->pers->spare_active(mddev);
c89a8eee
N		 7817 goto unlock;
7818 }
7819
31a59e34 7820 if (!mddev->external) {
0fd62b86 7821 int did_change = 0;
31a59e34
N		 7822 spin_lock_irq(&mddev->write_lock);
7823 if (mddev->safemode &&
7824 !atomic_read(&mddev->writes_pending) &&
7825 !mddev->in_sync &&
7826 mddev->recovery_cp == MaxSector) {
7827 mddev->in_sync = 1;
0fd62b86 7828 did_change = 1;
070dc6dd 7829 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7830 }
7831 if (mddev->safemode == 1)
7832 mddev->safemode = 0;
7833 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7834 if (did_change)
00bcb4ac 7835 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7836 }
fca4d848 7837
7a0a5355 7838 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 7839 md_update_sb(mddev, 0);
06d91a5f 7840
1da177e4
LT		 7841 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7842 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7843 /* resync/recovery still happening */
7844 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7845 goto unlock;
7846 }
7847 if (mddev->sync_thread) {
a91d5ac0 7848 md_reap_sync_thread(mddev);
1da177e4
LT		 7849 goto unlock;
7850 }
72a23c21
NB		 7851 /* Set RUNNING before clearing NEEDED to avoid
7852 * any transients in the value of "sync_action".
7853 */
72f36d59 7854 mddev->curr_resync_completed = 0;
72a23c21 7855 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
24dd469d
N		 7856 /* Clear some bits that don't mean anything, but
7857 * might be left set
7858 */
24dd469d
N		 7859 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7860 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7861
ed209584
N		 7862 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7863 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5fd6c1dc 7864 goto unlock;
1da177e4
LT		 7865 /* no recovery is running.
7866 * remove any failed drives, then
7867 * add spares if possible.
72f36d59 7868 * Spares are also removed and re-added, to allow
1da177e4
LT		 7869 * the personality to fail the re-add.
7870 */
1da177e4 7871
b4c4c7b8 7872 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7873 if (mddev->pers->check_reshape == NULL ||
7874 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7875 /* Cannot proceed */
7876 goto unlock;
7877 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7878 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 7879 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N		 7880 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7881 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7882 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7883 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7884 } else if (mddev->recovery_cp < MaxSector) {
7885 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7886 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7887 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7888 /* nothing to be done ... */
1da177e4 7889 goto unlock;
24dd469d 7890
1da177e4 7891 if (mddev->pers->sync_request) {
ef99bf48 7892 if (spares) {
a654b9d8
N		 7893 /* We are adding a device or devices to an array
7894 * which has the bitmap stored on all devices.
7895 * So make sure all bitmap pages get written
7896 */
7897 bitmap_write_all(mddev->bitmap);
7898 }
1da177e4
LT		 7899 mddev->sync_thread = md_register_thread(md_do_sync,
7900 mddev,
0da3c619 7901 "resync");
1da177e4
LT		 7902 if (!mddev->sync_thread) {
7903 printk(KERN_ERR "%s: could not start resync"
7904 " thread...\n",
7905 mdname(mddev));
7906 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7907 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7908 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7909 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7910 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7911 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7912 } else
1da177e4 7913 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7914 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7915 md_new_event(mddev);
1da177e4
LT		 7916 }
7917 unlock:
90f5f7ad
HR		 7918 wake_up(&mddev->sb_wait);
7919
72a23c21
NB		 7920 if (!mddev->sync_thread) {
7921 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7922 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7923 &mddev->recovery))
0c3573f1 7924 if (mddev->sysfs_action)
00bcb4ac 7925 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7926 }
1da177e4
LT		 7927 mddev_unlock(mddev);
7928 }
7929}
7930
a91d5ac0
JB		 7931void md_reap_sync_thread(struct mddev *mddev)
7932{
7933 struct md_rdev *rdev;
7934
7935 /* resync has finished, collect result */
7936 md_unregister_thread(&mddev->sync_thread);
7937 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7938 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7939 /* success...*/
7940 /* activate any spares */
7941 if (mddev->pers->spare_active(mddev)) {
7942 sysfs_notify(&mddev->kobj, NULL,
7943 "degraded");
7944 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7945 }
7946 }
7947 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7948 mddev->pers->finish_reshape)
7949 mddev->pers->finish_reshape(mddev);
7950
7951 /* If array is no-longer degraded, then any saved_raid_disk
7952 * information must be scrapped. Also if any device is now
7953 * In_sync we must scrape the saved_raid_disk for that device
7954 * do the superblock for an incrementally recovered device
7955 * written out.
7956 */
7957 rdev_for_each(rdev, mddev)
7958 if (!mddev->degraded ||
7959 test_bit(In_sync, &rdev->flags))
7960 rdev->saved_raid_disk = -1;
7961
7962 md_update_sb(mddev, 1);
7963 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7964 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7965 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7966 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7967 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7968 /* flag recovery needed just to double check */
7969 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7970 sysfs_notify_dirent_safe(mddev->sysfs_action);
7971 md_new_event(mddev);
7972 if (mddev->event_work.func)
7973 queue_work(md_misc_wq, &mddev->event_work);
7974}
7975
fd01b88c 7976void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 7977{
00bcb4ac 7978 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 7979 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 7980 !test_bit(Blocked, &rdev->flags) &&
7981 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 7982 msecs_to_jiffies(5000));
7983 rdev_dec_pending(rdev, mddev);
7984}
7985EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7986
c6563a8c
N		 7987void md_finish_reshape(struct mddev *mddev)
7988{
7989 /* called be personality module when reshape completes. */
7990 struct md_rdev *rdev;
7991
7992 rdev_for_each(rdev, mddev) {
7993 if (rdev->data_offset > rdev->new_data_offset)
7994 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7995 else
7996 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7997 rdev->data_offset = rdev->new_data_offset;
7998 }
7999}
8000EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N		 8001
8002/* Bad block management.
8003 * We can record which blocks on each device are 'bad' and so just
8004 * fail those blocks, or that stripe, rather than the whole device.
8005 * Entries in the bad-block table are 64bits wide. This comprises:
8006 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8007 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8008 * A 'shift' can be set so that larger blocks are tracked and
8009 * consequently larger devices can be covered.
8010 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8011 *
8012 * Locking of the bad-block table uses a seqlock so md_is_badblock
8013 * might need to retry if it is very unlucky.
8014 * We will sometimes want to check for bad blocks in a bi_end_io function,
8015 * so we use the write_seqlock_irq variant.
8016 *
8017 * When looking for a bad block we specify a range and want to
8018 * know if any block in the range is bad. So we binary-search
8019 * to the last range that starts at-or-before the given endpoint,
8020 * (or "before the sector after the target range")
8021 * then see if it ends after the given start.
8022 * We return
8023 * 0 if there are no known bad blocks in the range
8024 * 1 if there are known bad block which are all acknowledged
8025 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8026 * plus the start/length of the first bad section we overlap.
8027 */
8028int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8029 sector_t *first_bad, int *bad_sectors)
8030{
8031 int hi;
ab05613a 8032 int lo;
2230dfe4 8033 u64 *p = bb->page;
ab05613a 8034 int rv;
2230dfe4
N		 8035 sector_t target = s + sectors;
8036 unsigned seq;
8037
8038 if (bb->shift > 0) {
8039 /* round the start down, and the end up */
8040 s >>= bb->shift;
8041 target += (1<<bb->shift) - 1;
8042 target >>= bb->shift;
8043 sectors = target - s;
8044 }
8045 /* 'target' is now the first block after the bad range */
8046
8047retry:
8048 seq = read_seqbegin(&bb->lock);
ab05613a 8049 lo = 0;
8050 rv = 0;
2230dfe4
N		 8051 hi = bb->count;
8052
8053 /* Binary search between lo and hi for 'target'
8054 * i.e. for the last range that starts before 'target'
8055 */
8056 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8057 * are known not to be the last range before target.
8058 * VARIANT: hi-lo is the number of possible
8059 * ranges, and decreases until it reaches 1
8060 */
8061 while (hi - lo > 1) {
8062 int mid = (lo + hi) / 2;
8063 sector_t a = BB_OFFSET(p[mid]);
8064 if (a < target)
8065 /* This could still be the one, earlier ranges
8066 * could not. */
8067 lo = mid;
8068 else
8069 /* This and later ranges are definitely out. */
8070 hi = mid;
8071 }
8072 /* 'lo' might be the last that started before target, but 'hi' isn't */
8073 if (hi > lo) {
8074 /* need to check all range that end after 's' to see if
8075 * any are unacknowledged.
8076 */
8077 while (lo >= 0 &&
8078 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8079 if (BB_OFFSET(p[lo]) < target) {
8080 /* starts before the end, and finishes after
8081 * the start, so they must overlap
8082 */
8083 if (rv != -1 && BB_ACK(p[lo]))
8084 rv = 1;
8085 else
8086 rv = -1;
8087 *first_bad = BB_OFFSET(p[lo]);
8088 *bad_sectors = BB_LEN(p[lo]);
8089 }
8090 lo--;
8091 }
8092 }
8093
8094 if (read_seqretry(&bb->lock, seq))
8095 goto retry;
8096
8097 return rv;
8098}
8099EXPORT_SYMBOL_GPL(md_is_badblock);
8100
8101/*
8102 * Add a range of bad blocks to the table.
8103 * This might extend the table, or might contract it
8104 * if two adjacent ranges can be merged.
8105 * We binary-search to find the 'insertion' point, then
8106 * decide how best to handle it.
8107 */
8108static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8109 int acknowledged)
8110{
8111 u64 *p;
8112 int lo, hi;
8113 int rv = 1;
8114
8115 if (bb->shift < 0)
8116 /* badblocks are disabled */
8117 return 0;
8118
8119 if (bb->shift) {
8120 /* round the start down, and the end up */
8121 sector_t next = s + sectors;
8122 s >>= bb->shift;
8123 next += (1<<bb->shift) - 1;
8124 next >>= bb->shift;
8125 sectors = next - s;
8126 }
8127
8128 write_seqlock_irq(&bb->lock);
8129
8130 p = bb->page;
8131 lo = 0;
8132 hi = bb->count;
8133 /* Find the last range that starts at-or-before 's' */
8134 while (hi - lo > 1) {
8135 int mid = (lo + hi) / 2;
8136 sector_t a = BB_OFFSET(p[mid]);
8137 if (a <= s)
8138 lo = mid;
8139 else
8140 hi = mid;
8141 }
8142 if (hi > lo && BB_OFFSET(p[lo]) > s)
8143 hi = lo;
8144
8145 if (hi > lo) {
8146 /* we found a range that might merge with the start
8147 * of our new range
8148 */
8149 sector_t a = BB_OFFSET(p[lo]);
8150 sector_t e = a + BB_LEN(p[lo]);
8151 int ack = BB_ACK(p[lo]);
8152 if (e >= s) {
8153 /* Yes, we can merge with a previous range */
8154 if (s == a && s + sectors >= e)
8155 /* new range covers old */
8156 ack = acknowledged;
8157 else
8158 ack = ack && acknowledged;
8159
8160 if (e < s + sectors)
8161 e = s + sectors;
8162 if (e - a <= BB_MAX_LEN) {
8163 p[lo] = BB_MAKE(a, e-a, ack);
8164 s = e;
8165 } else {
8166 /* does not all fit in one range,
8167 * make p[lo] maximal
8168 */
8169 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8170 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8171 s = a + BB_MAX_LEN;
8172 }
8173 sectors = e - s;
8174 }
8175 }
8176 if (sectors && hi < bb->count) {
8177 /* 'hi' points to the first range that starts after 's'.
8178 * Maybe we can merge with the start of that range */
8179 sector_t a = BB_OFFSET(p[hi]);
8180 sector_t e = a + BB_LEN(p[hi]);
8181 int ack = BB_ACK(p[hi]);
8182 if (a <= s + sectors) {
8183 /* merging is possible */
8184 if (e <= s + sectors) {
8185 /* full overlap */
8186 e = s + sectors;
8187 ack = acknowledged;
8188 } else
8189 ack = ack && acknowledged;
8190
8191 a = s;
8192 if (e - a <= BB_MAX_LEN) {
8193 p[hi] = BB_MAKE(a, e-a, ack);
8194 s = e;
8195 } else {
8196 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8197 s = a + BB_MAX_LEN;
8198 }
8199 sectors = e - s;
8200 lo = hi;
8201 hi++;
8202 }
8203 }
8204 if (sectors == 0 && hi < bb->count) {
8205 /* we might be able to combine lo and hi */
8206 /* Note: 's' is at the end of 'lo' */
8207 sector_t a = BB_OFFSET(p[hi]);
8208 int lolen = BB_LEN(p[lo]);
8209 int hilen = BB_LEN(p[hi]);
8210 int newlen = lolen + hilen - (s - a);
8211 if (s >= a && newlen < BB_MAX_LEN) {
8212 /* yes, we can combine them */
8213 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8214 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8215 memmove(p + hi, p + hi + 1,
8216 (bb->count - hi - 1) * 8);
8217 bb->count--;
8218 }
8219 }
8220 while (sectors) {
8221 /* didn't merge (it all).
8222 * Need to add a range just before 'hi' */
8223 if (bb->count >= MD_MAX_BADBLOCKS) {
8224 /* No room for more */
8225 rv = 0;
8226 break;
8227 } else {
8228 int this_sectors = sectors;
8229 memmove(p + hi + 1, p + hi,
8230 (bb->count - hi) * 8);
8231 bb->count++;
8232
8233 if (this_sectors > BB_MAX_LEN)
8234 this_sectors = BB_MAX_LEN;
8235 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8236 sectors -= this_sectors;
8237 s += this_sectors;
8238 }
8239 }
8240
8241 bb->changed = 1;
de393cde
N		 8242 if (!acknowledged)
8243 bb->unacked_exist = 1;
2230dfe4
N		 8244 write_sequnlock_irq(&bb->lock);
8245
8246 return rv;
8247}
8248
3cb03002 8249int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8250 int is_new)
2230dfe4 8251{
c6563a8c
N		 8252 int rv;
8253 if (is_new)
8254 s += rdev->new_data_offset;
8255 else
8256 s += rdev->data_offset;
8257 rv = md_set_badblocks(&rdev->badblocks,
8258 s, sectors, 0);
2230dfe4
N		 8259 if (rv) {
8260 /* Make sure they get written out promptly */
8bd2f0a0 8261 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 8262 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8263 md_wakeup_thread(rdev->mddev->thread);
8264 }
8265 return rv;
8266}
8267EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8268
8269/*
8270 * Remove a range of bad blocks from the table.
8271 * This may involve extending the table if we spilt a region,
8272 * but it must not fail. So if the table becomes full, we just
8273 * drop the remove request.
8274 */
8275static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8276{
8277 u64 *p;
8278 int lo, hi;
8279 sector_t target = s + sectors;
8280 int rv = 0;
8281
8282 if (bb->shift > 0) {
8283 /* When clearing we round the start up and the end down.
8284 * This should not matter as the shift should align with
8285 * the block size and no rounding should ever be needed.
8286 * However it is better the think a block is bad when it
8287 * isn't than to think a block is not bad when it is.
8288 */
8289 s += (1<<bb->shift) - 1;
8290 s >>= bb->shift;
8291 target >>= bb->shift;
8292 sectors = target - s;
8293 }
8294
8295 write_seqlock_irq(&bb->lock);
8296
8297 p = bb->page;
8298 lo = 0;
8299 hi = bb->count;
8300 /* Find the last range that starts before 'target' */
8301 while (hi - lo > 1) {
8302 int mid = (lo + hi) / 2;
8303 sector_t a = BB_OFFSET(p[mid]);
8304 if (a < target)
8305 lo = mid;
8306 else
8307 hi = mid;
8308 }
8309 if (hi > lo) {
8310 /* p[lo] is the last range that could overlap the
8311 * current range. Earlier ranges could also overlap,
8312 * but only this one can overlap the end of the range.
8313 */
8314 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8315 /* Partial overlap, leave the tail of this range */
8316 int ack = BB_ACK(p[lo]);
8317 sector_t a = BB_OFFSET(p[lo]);
8318 sector_t end = a + BB_LEN(p[lo]);
8319
8320 if (a < s) {
8321 /* we need to split this range */
8322 if (bb->count >= MD_MAX_BADBLOCKS) {
8323 rv = 0;
8324 goto out;
8325 }
8326 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8327 bb->count++;
8328 p[lo] = BB_MAKE(a, s-a, ack);
8329 lo++;
8330 }
8331 p[lo] = BB_MAKE(target, end - target, ack);
8332 /* there is no longer an overlap */
8333 hi = lo;
8334 lo--;
8335 }
8336 while (lo >= 0 &&
8337 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8338 /* This range does overlap */
8339 if (BB_OFFSET(p[lo]) < s) {
8340 /* Keep the early parts of this range. */
8341 int ack = BB_ACK(p[lo]);
8342 sector_t start = BB_OFFSET(p[lo]);
8343 p[lo] = BB_MAKE(start, s - start, ack);
8344 /* now low doesn't overlap, so.. */
8345 break;
8346 }
8347 lo--;
8348 }
8349 /* 'lo' is strictly before, 'hi' is strictly after,
8350 * anything between needs to be discarded
8351 */
8352 if (hi - lo > 1) {
8353 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8354 bb->count -= (hi - lo - 1);
8355 }
8356 }
8357
8358 bb->changed = 1;
8359out:
8360 write_sequnlock_irq(&bb->lock);
8361 return rv;
8362}
8363
c6563a8c
N		 8364int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8365 int is_new)
2230dfe4 8366{
c6563a8c
N		 8367 if (is_new)
8368 s += rdev->new_data_offset;
8369 else
8370 s += rdev->data_offset;
2230dfe4 8371 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8372 s, sectors);
2230dfe4
N		 8373}
8374EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8375
8376/*
8377 * Acknowledge all bad blocks in a list.
8378 * This only succeeds if ->changed is clear. It is used by
8379 * in-kernel metadata updates
8380 */
8381void md_ack_all_badblocks(struct badblocks *bb)
8382{
8383 if (bb->page == NULL || bb->changed)
8384 /* no point even trying */
8385 return;
8386 write_seqlock_irq(&bb->lock);
8387
ecb178bb 8388 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N		 8389 u64 *p = bb->page;
8390 int i;
8391 for (i = 0; i < bb->count ; i++) {
8392 if (!BB_ACK(p[i])) {
8393 sector_t start = BB_OFFSET(p[i]);
8394 int len = BB_LEN(p[i]);
8395 p[i] = BB_MAKE(start, len, 1);
8396 }
8397 }
de393cde 8398 bb->unacked_exist = 0;
2230dfe4
N		 8399 }
8400 write_sequnlock_irq(&bb->lock);
8401}
8402EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8403
16c791a5
N		 8404/* sysfs access to bad-blocks list.
8405 * We present two files.
8406 * 'bad-blocks' lists sector numbers and lengths of ranges that
8407 * are recorded as bad. The list is truncated to fit within
8408 * the one-page limit of sysfs.
8409 * Writing "sector length" to this file adds an acknowledged
8410 * bad block list.
8411 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8412 * been acknowledged. Writing to this file adds bad blocks
8413 * without acknowledging them. This is largely for testing.
8414 */
8415
8416static ssize_t
8417badblocks_show(struct badblocks *bb, char *page, int unack)
8418{
8419 size_t len;
8420 int i;
8421 u64 *p = bb->page;
8422 unsigned seq;
8423
8424 if (bb->shift < 0)
8425 return 0;
8426
8427retry:
8428 seq = read_seqbegin(&bb->lock);
8429
8430 len = 0;
8431 i = 0;
8432
8433 while (len < PAGE_SIZE && i < bb->count) {
8434 sector_t s = BB_OFFSET(p[i]);
8435 unsigned int length = BB_LEN(p[i]);
8436 int ack = BB_ACK(p[i]);
8437 i++;
8438
8439 if (unack && ack)
8440 continue;
8441
8442 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8443 (unsigned long long)s << bb->shift,
8444 length << bb->shift);
8445 }
de393cde
N		 8446 if (unack && len == 0)
8447 bb->unacked_exist = 0;
16c791a5
N		 8448
8449 if (read_seqretry(&bb->lock, seq))
8450 goto retry;
8451
8452 return len;
8453}
8454
8455#define DO_DEBUG 1
8456
8457static ssize_t
8458badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8459{
8460 unsigned long long sector;
8461 int length;
8462 char newline;
8463#ifdef DO_DEBUG
8464 /* Allow clearing via sysfs *only* for testing/debugging.
8465 * Normally only a successful write may clear a badblock
8466 */
8467 int clear = 0;
8468 if (page[0] == '-') {
8469 clear = 1;
8470 page++;
8471 }
8472#endif /* DO_DEBUG */
8473
8474 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8475 case 3:
8476 if (newline != '\n')
8477 return -EINVAL;
8478 case 2:
8479 if (length <= 0)
8480 return -EINVAL;
8481 break;
8482 default:
8483 return -EINVAL;
8484 }
8485
8486#ifdef DO_DEBUG
8487 if (clear) {
8488 md_clear_badblocks(bb, sector, length);
8489 return len;
8490 }
8491#endif /* DO_DEBUG */
8492 if (md_set_badblocks(bb, sector, length, !unack))
8493 return len;
8494 else
8495 return -ENOSPC;
8496}
8497
75c96f85
AB		 8498static int md_notify_reboot(struct notifier_block *this,
8499 unsigned long code, void *x)
1da177e4
LT		 8500{
8501 struct list_head *tmp;
fd01b88c 8502 struct mddev *mddev;
2dba6a91 8503 int need_delay = 0;
1da177e4 8504
c744a65c
N		 8505 for_each_mddev(mddev, tmp) {
8506 if (mddev_trylock(mddev)) {
30b8aa91
N		 8507 if (mddev->pers)
8508 __md_stop_writes(mddev);
c744a65c
N		 8509 mddev->safemode = 2;
8510 mddev_unlock(mddev);
2dba6a91 8511 }
c744a65c 8512 need_delay = 1;
1da177e4 8513 }
c744a65c
N		 8514 /*
8515 * certain more exotic SCSI devices are known to be
8516 * volatile wrt too early system reboots. While the
8517 * right place to handle this issue is the given
8518 * driver, we do want to have a safe RAID driver ...
8519 */
8520 if (need_delay)
8521 mdelay(1000*1);
8522
1da177e4
LT		 8523 return NOTIFY_DONE;
8524}
8525
75c96f85 8526static struct notifier_block md_notifier = {
1da177e4
LT		 8527 .notifier_call = md_notify_reboot,
8528 .next = NULL,
8529 .priority = INT_MAX, /* before any real devices */
8530};
8531
8532static void md_geninit(void)
8533{
36a4e1fe 8534 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8535
c7705f34 8536 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8537}
8538
75c96f85 8539static int __init md_init(void)
1da177e4 8540{
e804ac78
TH		 8541 int ret = -ENOMEM;
8542
ada609ee 8543 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8544 if (!md_wq)
8545 goto err_wq;
8546
8547 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8548 if (!md_misc_wq)
8549 goto err_misc_wq;
8550
8551 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8552 goto err_md;
8553
8554 if ((ret = register_blkdev(0, "mdp")) < 0)
8555 goto err_mdp;
8556 mdp_major = ret;
8557
3dbd8c2e 8558 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 8559 md_probe, NULL, NULL);
8560 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8561 md_probe, NULL, NULL);
8562
1da177e4 8563 register_reboot_notifier(&md_notifier);
0b4d4147 8564 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8565
8566 md_geninit();
d710e138 8567 return 0;
1da177e4 8568
e804ac78
TH		 8569err_mdp:
8570 unregister_blkdev(MD_MAJOR, "md");
8571err_md:
8572 destroy_workqueue(md_misc_wq);
8573err_misc_wq:
8574 destroy_workqueue(md_wq);
8575err_wq:
8576 return ret;
8577}
1da177e4
LT		 8578
8579#ifndef MODULE
8580
8581/*
8582 * Searches all registered partitions for autorun RAID arrays
8583 * at boot time.
8584 */
4d936ec1
ME		 8585
8586static LIST_HEAD(all_detected_devices);
8587struct detected_devices_node {
8588 struct list_head list;
8589 dev_t dev;
8590};
1da177e4
LT		 8591
8592void md_autodetect_dev(dev_t dev)
8593{
4d936ec1
ME		 8594 struct detected_devices_node *node_detected_dev;
8595
8596 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8597 if (node_detected_dev) {
8598 node_detected_dev->dev = dev;
8599 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8600 } else {
8601 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8602 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8603 }
1da177e4
LT		 8604}
8605
8606
8607static void autostart_arrays(int part)
8608{
3cb03002 8609 struct md_rdev *rdev;
4d936ec1
ME		 8610 struct detected_devices_node *node_detected_dev;
8611 dev_t dev;
8612 int i_scanned, i_passed;
1da177e4 8613
4d936ec1
ME		 8614 i_scanned = 0;
8615 i_passed = 0;
1da177e4 8616
4d936ec1 8617 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8618
4d936ec1
ME		 8619 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8620 i_scanned++;
8621 node_detected_dev = list_entry(all_detected_devices.next,
8622 struct detected_devices_node, list);
8623 list_del(&node_detected_dev->list);
8624 dev = node_detected_dev->dev;
8625 kfree(node_detected_dev);
df968c4e 8626 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8627 if (IS_ERR(rdev))
8628 continue;
8629
b2d444d7 8630 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 8631 MD_BUG();
8632 continue;
8633 }
d0fae18f 8634 set_bit(AutoDetected, &rdev->flags);
1da177e4 8635 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8636 i_passed++;
1da177e4 8637 }
4d936ec1
ME		 8638
8639 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8640 i_scanned, i_passed);
1da177e4
LT		 8641
8642 autorun_devices(part);
8643}
8644
fdee8ae4 8645#endif /* !MODULE */
1da177e4
LT		 8646
8647static __exit void md_exit(void)
8648{
fd01b88c 8649 struct mddev *mddev;
1da177e4 8650 struct list_head *tmp;
8ab5e4c1 8651
3dbd8c2e 8652 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 8653 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8654
3dbd8c2e 8655 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8656 unregister_blkdev(mdp_major, "mdp");
8657 unregister_reboot_notifier(&md_notifier);
8658 unregister_sysctl_table(raid_table_header);
8659 remove_proc_entry("mdstat", NULL);
29ac4aa3 8660 for_each_mddev(mddev, tmp) {
1da177e4 8661 export_array(mddev);
d3374825 8662 mddev->hold_active = 0;
1da177e4 8663 }
e804ac78
TH		 8664 destroy_workqueue(md_misc_wq);
8665 destroy_workqueue(md_wq);
1da177e4
LT		 8666}
8667
685784aa 8668subsys_initcall(md_init);
1da177e4
LT		 8669module_exit(md_exit)
8670
f91de92e
N		 8671static int get_ro(char *buffer, struct kernel_param *kp)
8672{
8673 return sprintf(buffer, "%d", start_readonly);
8674}
8675static int set_ro(const char *val, struct kernel_param *kp)
8676{
8677 char *e;
8678 int num = simple_strtoul(val, &e, 10);
8679 if (*val && (*e == '\0' || *e == '\n')) {
8680 start_readonly = num;
4dbcdc75 8681 return 0;
f91de92e
N		 8682 }
8683 return -EINVAL;
8684}
8685
80ca3a44
N		 8686module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8687module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 8688
efeb53c0 8689module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8690
1da177e4
LT		 8691EXPORT_SYMBOL(register_md_personality);
8692EXPORT_SYMBOL(unregister_md_personality);
8693EXPORT_SYMBOL(md_error);
8694EXPORT_SYMBOL(md_done_sync);
8695EXPORT_SYMBOL(md_write_start);
8696EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 8697EXPORT_SYMBOL(md_register_thread);
8698EXPORT_SYMBOL(md_unregister_thread);
8699EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 8700EXPORT_SYMBOL(md_check_recovery);
a91d5ac0 8701EXPORT_SYMBOL(md_reap_sync_thread);
1da177e4 8702MODULE_LICENSE("GPL");
0efb9e61 8703MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8704MODULE_ALIAS("md");
72008652 8705MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)