projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: raid1/raid10 md devices leak memory when stopping
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
a6fb0934 35#include <linux/kthread.h>
bff61975 36#include <linux/blkdev.h>
1da177e4 37#include <linux/sysctl.h>
bff61975 38#include <linux/seq_file.h>
ff01bb48 39#include <linux/fs.h>
d7603b7e 40#include <linux/poll.h>
16f17b39 41#include <linux/ctype.h>
e7d2860b 42#include <linux/string.h>
fb4d8c76
N		 43#include <linux/hdreg.h>
44#include <linux/proc_fs.h>
45#include <linux/random.h>
056075c7 46#include <linux/module.h>
fb4d8c76 47#include <linux/reboot.h>
32a7627c 48#include <linux/file.h>
aa98aa31 49#include <linux/compat.h>
25570727 50#include <linux/delay.h>
bff61975
N		 51#include <linux/raid/md_p.h>
52#include <linux/raid/md_u.h>
5a0e3ad6 53#include <linux/slab.h>
43b2e5d8 54#include "md.h"
ef740c37 55#include "bitmap.h"
1da177e4 56
1da177e4 57#ifndef MODULE
d710e138 58static void autostart_arrays(int part);
1da177e4
LT		 59#endif
60
01f96c0a
N		 61/* pers_list is a list of registered personalities protected
62 * by pers_lock.
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
65 */
2604b703 66static LIST_HEAD(pers_list);
1da177e4
LT		 67static DEFINE_SPINLOCK(pers_lock);
68
5e56341d
AB		 69static void md_print_devices(void);
70
90b08710 71static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 72static struct workqueue_struct *md_wq;
73static struct workqueue_struct *md_misc_wq;
90b08710 74
746d3207
N		 75static int remove_and_add_spares(struct mddev *mddev,
76 struct md_rdev *this);
77
5e56341d
AB		 78#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
79
1e50915f
RB		 80/*
81 * Default number of read corrections we'll attempt on an rdev
82 * before ejecting it from the array. We divide the read error
83 * count by 2 for every hour elapsed between read errors.
84 */
85#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 86/*
87 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
88 * is 1000 KB/sec, so the extra system load does not show up that much.
89 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 90 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 91 * subsystem is idle. There is also an 'absolute maximum' reconstruction
92 * speed limit - in case reconstruction slows down your system despite
93 * idle IO detection.
94 *
95 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 96 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 97 */
98
99static int sysctl_speed_limit_min = 1000;
100static int sysctl_speed_limit_max = 200000;
fd01b88c 101static inline int speed_min(struct mddev *mddev)
88202a0c
N		 102{
103 return mddev->sync_speed_min ?
104 mddev->sync_speed_min : sysctl_speed_limit_min;
105}
106
fd01b88c 107static inline int speed_max(struct mddev *mddev)
88202a0c
N		 108{
109 return mddev->sync_speed_max ?
110 mddev->sync_speed_max : sysctl_speed_limit_max;
111}
1da177e4
LT		 112
113static struct ctl_table_header *raid_table_header;
114
115static ctl_table raid_table[] = {
116 {
1da177e4
LT		 117 .procname = "speed_limit_min",
118 .data = &sysctl_speed_limit_min,
119 .maxlen = sizeof(int),
80ca3a44 120 .mode = S_IRUGO|S_IWUSR,
6d456111 121 .proc_handler = proc_dointvec,
1da177e4
LT		 122 },
123 {
1da177e4
LT		 124 .procname = "speed_limit_max",
125 .data = &sysctl_speed_limit_max,
126 .maxlen = sizeof(int),
80ca3a44 127 .mode = S_IRUGO|S_IWUSR,
6d456111 128 .proc_handler = proc_dointvec,
1da177e4 129 },
894d2491 130 { }
1da177e4
LT		 131};
132
133static ctl_table raid_dir_table[] = {
134 {
1da177e4
LT		 135 .procname = "raid",
136 .maxlen = 0,
80ca3a44 137 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 138 .child = raid_table,
139 },
894d2491 140 { }
1da177e4
LT		 141};
142
143static ctl_table raid_root_table[] = {
144 {
1da177e4
LT		 145 .procname = "dev",
146 .maxlen = 0,
147 .mode = 0555,
148 .child = raid_dir_table,
149 },
894d2491 150 { }
1da177e4
LT		 151};
152
83d5cde4 153static const struct block_device_operations md_fops;
1da177e4 154
f91de92e
N		 155static int start_readonly;
156
a167f663
N		 157/* bio_clone_mddev
158 * like bio_clone, but with a local bio set
159 */
160
a167f663 161struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 162 struct mddev *mddev)
a167f663
N		 163{
164 struct bio *b;
a167f663
N		 165
166 if (!mddev || !mddev->bio_set)
167 return bio_alloc(gfp_mask, nr_iovecs);
168
395c72a7 169 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
a167f663
N		 170 if (!b)
171 return NULL;
a167f663
N		 172 return b;
173}
174EXPORT_SYMBOL_GPL(bio_alloc_mddev);
175
176struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
fd01b88c 177 struct mddev *mddev)
a167f663 178{
a167f663
N		 179 if (!mddev || !mddev->bio_set)
180 return bio_clone(bio, gfp_mask);
181
bf800ef1 182 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
a167f663
N		 183}
184EXPORT_SYMBOL_GPL(bio_clone_mddev);
185
d2eb35ac
N		 186void md_trim_bio(struct bio *bio, int offset, int size)
187{
188 /* 'bio' is a cloned bio which we need to trim to match
189 * the given offset and size.
190 * This requires adjusting bi_sector, bi_size, and bi_io_vec
191 */
192 int i;
193 struct bio_vec *bvec;
194 int sofar = 0;
195
196 size <<= 9;
197 if (offset == 0 && size == bio->bi_size)
198 return;
199
200 bio->bi_sector += offset;
201 bio->bi_size = size;
202 offset <<= 9;
203 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
204
205 while (bio->bi_idx < bio->bi_vcnt &&
206 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
207 /* remove this whole bio_vec */
208 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
209 bio->bi_idx++;
210 }
211 if (bio->bi_idx < bio->bi_vcnt) {
212 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
213 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
214 }
215 /* avoid any complications with bi_idx being non-zero*/
216 if (bio->bi_idx) {
217 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
218 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
219 bio->bi_vcnt -= bio->bi_idx;
220 bio->bi_idx = 0;
221 }
222 /* Make sure vcnt and last bv are not too big */
223 bio_for_each_segment(bvec, bio, i) {
224 if (sofar + bvec->bv_len > size)
225 bvec->bv_len = size - sofar;
226 if (bvec->bv_len == 0) {
227 bio->bi_vcnt = i;
228 break;
229 }
230 sofar += bvec->bv_len;
231 }
232}
233EXPORT_SYMBOL_GPL(md_trim_bio);
234
d7603b7e
N		 235/*
236 * We have a system wide 'event count' that is incremented
237 * on any 'interesting' event, and readers of /proc/mdstat
238 * can use 'poll' or 'select' to find out when the event
239 * count increases.
240 *
241 * Events are:
242 * start array, stop array, error, add device, remove device,
243 * start build, activate spare
244 */
2989ddbd 245static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 246static atomic_t md_event_count;
fd01b88c 247void md_new_event(struct mddev *mddev)
d7603b7e
N		 248{
249 atomic_inc(&md_event_count);
250 wake_up(&md_event_waiters);
251}
29269553 252EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 253
c331eb04
N		 254/* Alternate version that can be called from interrupts
255 * when calling sysfs_notify isn't needed.
256 */
fd01b88c 257static void md_new_event_inintr(struct mddev *mddev)
c331eb04
N		 258{
259 atomic_inc(&md_event_count);
260 wake_up(&md_event_waiters);
261}
262
1da177e4
LT		 263/*
264 * Enables to iterate over all existing md arrays
265 * all_mddevs_lock protects this list.
266 */
267static LIST_HEAD(all_mddevs);
268static DEFINE_SPINLOCK(all_mddevs_lock);
269
270
271/*
272 * iterates through all used mddevs in the system.
273 * We take care to grab the all_mddevs_lock whenever navigating
274 * the list, and to always hold a refcount when unlocked.
275 * Any code which breaks out of this loop while own
276 * a reference to the current mddev and must mddev_put it.
277 */
fd01b88c 278#define for_each_mddev(_mddev,_tmp) \
1da177e4
LT		 279 \
280 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N		 281 _tmp = all_mddevs.next; \
282 _mddev = NULL;}); \
283 ({ if (_tmp != &all_mddevs) \
284 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 285 spin_unlock(&all_mddevs_lock); \
fd01b88c
N		 286 if (_mddev) mddev_put(_mddev); \
287 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
288 _tmp != &all_mddevs;}); \
1da177e4 289 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 290 _tmp = _tmp->next;}) \
1da177e4
LT		 291 )
292
293
409c57f3
N		 294/* Rather than calling directly into the personality make_request function,
295 * IO requests come here first so that we can check if the device is
296 * being suspended pending a reconfiguration.
297 * We hold a refcount over the call to ->make_request. By the time that
298 * call has finished, the bio has been linked into some internal structure
299 * and so is visible to ->quiesce(), so we don't need the refcount any more.
300 */
5a7bbad2 301static void md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 302{
49077326 303 const int rw = bio_data_dir(bio);
fd01b88c 304 struct mddev *mddev = q->queuedata;
49077326 305 int cpu;
e91ece55 306 unsigned int sectors;
49077326 307
0ca69886
N		 308 if (mddev == NULL || mddev->pers == NULL
309 || !mddev->ready) {
409c57f3 310 bio_io_error(bio);
5a7bbad2 311 return;
409c57f3 312 }
bbfa57c0
SR		 313 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
314 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
315 return;
316 }
0ca69886 317 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 318 rcu_read_lock();
e9c7469b 319 if (mddev->suspended) {
409c57f3
N		 320 DEFINE_WAIT(__wait);
321 for (;;) {
322 prepare_to_wait(&mddev->sb_wait, &__wait,
323 TASK_UNINTERRUPTIBLE);
e9c7469b 324 if (!mddev->suspended)
409c57f3
N		 325 break;
326 rcu_read_unlock();
327 schedule();
328 rcu_read_lock();
329 }
330 finish_wait(&mddev->sb_wait, &__wait);
331 }
332 atomic_inc(&mddev->active_io);
333 rcu_read_unlock();
49077326 334
e91ece55
CM		 335 /*
336 * save the sectors now since our bio can
337 * go away inside make_request
338 */
339 sectors = bio_sectors(bio);
5a7bbad2 340 mddev->pers->make_request(mddev, bio);
49077326
N		 341
342 cpu = part_stat_lock();
343 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
e91ece55 344 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
49077326
N		 345 part_stat_unlock();
346
409c57f3
N		 347 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
348 wake_up(&mddev->sb_wait);
409c57f3
N		 349}
350
9e35b99c
N		 351/* mddev_suspend makes sure no new requests are submitted
352 * to the device, and that any requests that have been submitted
353 * are completely handled.
354 * Once ->stop is called and completes, the module will be completely
355 * unused.
356 */
fd01b88c 357void mddev_suspend(struct mddev *mddev)
409c57f3
N		 358{
359 BUG_ON(mddev->suspended);
360 mddev->suspended = 1;
361 synchronize_rcu();
362 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
363 mddev->pers->quiesce(mddev, 1);
0d9f4f13
JB		 364
365 del_timer_sync(&mddev->safemode_timer);
409c57f3 366}
390ee602 367EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 368
fd01b88c 369void mddev_resume(struct mddev *mddev)
409c57f3
N		 370{
371 mddev->suspended = 0;
372 wake_up(&mddev->sb_wait);
373 mddev->pers->quiesce(mddev, 0);
0fd018af 374
47525e59 375 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
0fd018af
JB		 376 md_wakeup_thread(mddev->thread);
377 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 378}
390ee602 379EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 380
fd01b88c 381int mddev_congested(struct mddev *mddev, int bits)
3fa841d7
N		 382{
383 return mddev->suspended;
384}
385EXPORT_SYMBOL(mddev_congested);
386
a2826aa9 387/*
e9c7469b 388 * Generic flush handling for md
a2826aa9
N		 389 */
390
e9c7469b 391static void md_end_flush(struct bio *bio, int err)
a2826aa9 392{
3cb03002 393 struct md_rdev *rdev = bio->bi_private;
fd01b88c 394 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 395
396 rdev_dec_pending(rdev, mddev);
397
398 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 399 /* The pre-request flush has finished */
e804ac78 400 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 401 }
402 bio_put(bio);
403}
404
a7a07e69
N		 405static void md_submit_flush_data(struct work_struct *ws);
406
a035fc3e 407static void submit_flushes(struct work_struct *ws)
a2826aa9 408{
fd01b88c 409 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 410 struct md_rdev *rdev;
a2826aa9 411
a7a07e69
N		 412 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
413 atomic_set(&mddev->flush_pending, 1);
a2826aa9 414 rcu_read_lock();
dafb20fa 415 rdev_for_each_rcu(rdev, mddev)
a2826aa9
N		 416 if (rdev->raid_disk >= 0 &&
417 !test_bit(Faulty, &rdev->flags)) {
418 /* Take two references, one is dropped
419 * when request finishes, one after
420 * we reclaim rcu_read_lock
421 */
422 struct bio *bi;
423 atomic_inc(&rdev->nr_pending);
424 atomic_inc(&rdev->nr_pending);
425 rcu_read_unlock();
b5e1b8ce 426 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
e9c7469b 427 bi->bi_end_io = md_end_flush;
a2826aa9
N		 428 bi->bi_private = rdev;
429 bi->bi_bdev = rdev->bdev;
430 atomic_inc(&mddev->flush_pending);
e9c7469b 431 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 432 rcu_read_lock();
433 rdev_dec_pending(rdev, mddev);
434 }
435 rcu_read_unlock();
a7a07e69
N		 436 if (atomic_dec_and_test(&mddev->flush_pending))
437 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 438}
439
e9c7469b 440static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 441{
fd01b88c 442 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 443 struct bio *bio = mddev->flush_bio;
a2826aa9 444
e9c7469b 445 if (bio->bi_size == 0)
a2826aa9
N		 446 /* an empty barrier - all done */
447 bio_endio(bio, 0);
448 else {
e9c7469b 449 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 450 mddev->pers->make_request(mddev, bio);
a2826aa9 451 }
2b74e12e
N		 452
453 mddev->flush_bio = NULL;
454 wake_up(&mddev->sb_wait);
a2826aa9
N		 455}
456
fd01b88c 457void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9
N		 458{
459 spin_lock_irq(&mddev->write_lock);
460 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 461 !mddev->flush_bio,
eed8c02e 462 mddev->write_lock);
e9c7469b 463 mddev->flush_bio = bio;
a2826aa9
N		 464 spin_unlock_irq(&mddev->write_lock);
465
a035fc3e
N		 466 INIT_WORK(&mddev->flush_work, submit_flushes);
467 queue_work(md_wq, &mddev->flush_work);
a2826aa9 468}
e9c7469b 469EXPORT_SYMBOL(md_flush_request);
409c57f3 470
74018dc3 471void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
97658cdd 472{
9cbb1750
N		 473 struct mddev *mddev = cb->data;
474 md_wakeup_thread(mddev->thread);
475 kfree(cb);
97658cdd 476}
9cbb1750 477EXPORT_SYMBOL(md_unplug);
2ac87401 478
fd01b88c 479static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 480{
481 atomic_inc(&mddev->active);
482 return mddev;
483}
484
5fd3a17e 485static void mddev_delayed_delete(struct work_struct *ws);
d3374825 486
fd01b88c 487static void mddev_put(struct mddev *mddev)
1da177e4 488{
a167f663
N		 489 struct bio_set *bs = NULL;
490
1da177e4
LT		 491 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
492 return;
d3374825 493 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 494 mddev->ctime == 0 && !mddev->hold_active) {
495 /* Array is not configured at all, and not held active,
496 * so destroy it */
af8a2434 497 list_del_init(&mddev->all_mddevs);
a167f663
N		 498 bs = mddev->bio_set;
499 mddev->bio_set = NULL;
d3374825 500 if (mddev->gendisk) {
e804ac78
TH		 501 /* We did a probe so need to clean up. Call
502 * queue_work inside the spinlock so that
503 * flush_workqueue() after mddev_find will
504 * succeed in waiting for the work to be done.
d3374825
N		 505 */
506 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 507 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 508 } else
509 kfree(mddev);
510 }
511 spin_unlock(&all_mddevs_lock);
a167f663
N		 512 if (bs)
513 bioset_free(bs);
1da177e4
LT		 514}
515
fd01b88c 516void mddev_init(struct mddev *mddev)
fafd7fb0
N		 517{
518 mutex_init(&mddev->open_mutex);
519 mutex_init(&mddev->reconfig_mutex);
520 mutex_init(&mddev->bitmap_info.mutex);
521 INIT_LIST_HEAD(&mddev->disks);
522 INIT_LIST_HEAD(&mddev->all_mddevs);
523 init_timer(&mddev->safemode_timer);
524 atomic_set(&mddev->active, 1);
525 atomic_set(&mddev->openers, 0);
526 atomic_set(&mddev->active_io, 0);
527 spin_lock_init(&mddev->write_lock);
528 atomic_set(&mddev->flush_pending, 0);
529 init_waitqueue_head(&mddev->sb_wait);
530 init_waitqueue_head(&mddev->recovery_wait);
531 mddev->reshape_position = MaxSector;
2c810cdd 532 mddev->reshape_backwards = 0;
fafd7fb0
N		 533 mddev->resync_min = 0;
534 mddev->resync_max = MaxSector;
535 mddev->level = LEVEL_NONE;
536}
390ee602 537EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 538
fd01b88c 539static struct mddev * mddev_find(dev_t unit)
1da177e4 540{
fd01b88c 541 struct mddev *mddev, *new = NULL;
1da177e4 542
8f5f02c4
N		 543 if (unit && MAJOR(unit) != MD_MAJOR)
544 unit &= ~((1<<MdpMinorShift)-1);
545
1da177e4
LT		 546 retry:
547 spin_lock(&all_mddevs_lock);
efeb53c0
N		 548
549 if (unit) {
550 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
551 if (mddev->unit == unit) {
552 mddev_get(mddev);
553 spin_unlock(&all_mddevs_lock);
554 kfree(new);
555 return mddev;
556 }
557
558 if (new) {
559 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 560 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 561 new->hold_active = UNTIL_IOCTL;
562 return new;
1da177e4 563 }
efeb53c0
N		 564 } else if (new) {
565 /* find an unused unit number */
566 static int next_minor = 512;
567 int start = next_minor;
568 int is_free = 0;
569 int dev = 0;
570 while (!is_free) {
571 dev = MKDEV(MD_MAJOR, next_minor);
572 next_minor++;
573 if (next_minor > MINORMASK)
574 next_minor = 0;
575 if (next_minor == start) {
576 /* Oh dear, all in use. */
577 spin_unlock(&all_mddevs_lock);
578 kfree(new);
579 return NULL;
580 }
581
582 is_free = 1;
583 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
584 if (mddev->unit == dev) {
585 is_free = 0;
586 break;
587 }
588 }
589 new->unit = dev;
590 new->md_minor = MINOR(dev);
591 new->hold_active = UNTIL_STOP;
1da177e4
LT		 592 list_add(&new->all_mddevs, &all_mddevs);
593 spin_unlock(&all_mddevs_lock);
594 return new;
595 }
596 spin_unlock(&all_mddevs_lock);
597
9ffae0cf 598 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 599 if (!new)
600 return NULL;
601
1da177e4
LT		 602 new->unit = unit;
603 if (MAJOR(unit) == MD_MAJOR)
604 new->md_minor = MINOR(unit);
605 else
606 new->md_minor = MINOR(unit) >> MdpMinorShift;
607
fafd7fb0 608 mddev_init(new);
1da177e4 609
1da177e4
LT		 610 goto retry;
611}
612
fd01b88c 613static inline int mddev_lock(struct mddev * mddev)
1da177e4 614{
df5b89b3 615 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 616}
617
fd01b88c 618static inline int mddev_is_locked(struct mddev *mddev)
b522adcd
DW		 619{
620 return mutex_is_locked(&mddev->reconfig_mutex);
621}
622
fd01b88c 623static inline int mddev_trylock(struct mddev * mddev)
1da177e4 624{
df5b89b3 625 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 626}
627
b6eb127d
N		 628static struct attribute_group md_redundancy_group;
629
fd01b88c 630static void mddev_unlock(struct mddev * mddev)
1da177e4 631{
a64c876f 632 if (mddev->to_remove) {
b6eb127d
N		 633 /* These cannot be removed under reconfig_mutex as
634 * an access to the files will try to take reconfig_mutex
635 * while holding the file unremovable, which leads to
636 * a deadlock.
bb4f1e9d
N		 637 * So hold set sysfs_active while the remove in happeing,
638 * and anything else which might set ->to_remove or my
639 * otherwise change the sysfs namespace will fail with
640 * -EBUSY if sysfs_active is still set.
641 * We set sysfs_active under reconfig_mutex and elsewhere
642 * test it under the same mutex to ensure its correct value
643 * is seen.
b6eb127d 644 */
a64c876f
N		 645 struct attribute_group *to_remove = mddev->to_remove;
646 mddev->to_remove = NULL;
bb4f1e9d 647 mddev->sysfs_active = 1;
b6eb127d
N		 648 mutex_unlock(&mddev->reconfig_mutex);
649
00bcb4ac
N		 650 if (mddev->kobj.sd) {
651 if (to_remove != &md_redundancy_group)
652 sysfs_remove_group(&mddev->kobj, to_remove);
653 if (mddev->pers == NULL ||
654 mddev->pers->sync_request == NULL) {
655 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
656 if (mddev->sysfs_action)
657 sysfs_put(mddev->sysfs_action);
658 mddev->sysfs_action = NULL;
659 }
a64c876f 660 }
bb4f1e9d 661 mddev->sysfs_active = 0;
b6eb127d
N		 662 } else
663 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 664
751e67ca
CD		 665 /* As we've dropped the mutex we need a spinlock to
666 * make sure the thread doesn't disappear
01f96c0a
N		 667 */
668 spin_lock(&pers_lock);
005eca5e 669 md_wakeup_thread(mddev->thread);
01f96c0a 670 spin_unlock(&pers_lock);
1da177e4
LT		 671}
672
fd01b88c 673static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
1da177e4 674{
3cb03002 675 struct md_rdev *rdev;
1da177e4 676
dafb20fa 677 rdev_for_each(rdev, mddev)
1da177e4
LT		 678 if (rdev->desc_nr == nr)
679 return rdev;
159ec1fc 680
1da177e4
LT		 681 return NULL;
682}
683
1ca69c4b
N		 684static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
685{
686 struct md_rdev *rdev;
687
688 rdev_for_each_rcu(rdev, mddev)
689 if (rdev->desc_nr == nr)
690 return rdev;
691
692 return NULL;
693}
694
695static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
1da177e4 696{
3cb03002 697 struct md_rdev *rdev;
1da177e4 698
dafb20fa 699 rdev_for_each(rdev, mddev)
1da177e4
LT		 700 if (rdev->bdev->bd_dev == dev)
701 return rdev;
159ec1fc 702
1da177e4
LT		 703 return NULL;
704}
705
1ca69c4b
N		 706static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
707{
708 struct md_rdev *rdev;
709
710 rdev_for_each_rcu(rdev, mddev)
711 if (rdev->bdev->bd_dev == dev)
712 return rdev;
713
714 return NULL;
715}
716
84fc4b56 717static struct md_personality *find_pers(int level, char *clevel)
2604b703 718{
84fc4b56 719 struct md_personality *pers;
d9d166c2
N		 720 list_for_each_entry(pers, &pers_list, list) {
721 if (level != LEVEL_NONE && pers->level == level)
2604b703 722 return pers;
d9d166c2
N		 723 if (strcmp(pers->name, clevel)==0)
724 return pers;
725 }
2604b703
N		 726 return NULL;
727}
728
b73df2d3 729/* return the offset of the super block in 512byte sectors */
3cb03002 730static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 731{
57b2caa3 732 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 733 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 734}
735
3cb03002 736static int alloc_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 737{
738 if (rdev->sb_page)
739 MD_BUG();
740
741 rdev->sb_page = alloc_page(GFP_KERNEL);
742 if (!rdev->sb_page) {
743 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 744 return -ENOMEM;
1da177e4
LT		 745 }
746
747 return 0;
748}
749
545c8795 750void md_rdev_clear(struct md_rdev *rdev)
1da177e4
LT		 751{
752 if (rdev->sb_page) {
2d1f3b5d 753 put_page(rdev->sb_page);
1da177e4
LT		 754 rdev->sb_loaded = 0;
755 rdev->sb_page = NULL;
0f420358 756 rdev->sb_start = 0;
dd8ac336 757 rdev->sectors = 0;
1da177e4 758 }
2699b672
N		 759 if (rdev->bb_page) {
760 put_page(rdev->bb_page);
761 rdev->bb_page = NULL;
762 }
4fa2f327
N		 763 kfree(rdev->badblocks.page);
764 rdev->badblocks.page = NULL;
1da177e4 765}
545c8795 766EXPORT_SYMBOL_GPL(md_rdev_clear);
1da177e4 767
6712ecf8 768static void super_written(struct bio *bio, int error)
7bfa19f2 769{
3cb03002 770 struct md_rdev *rdev = bio->bi_private;
fd01b88c 771 struct mddev *mddev = rdev->mddev;
7bfa19f2 772
3a0f5bbb
N		 773 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
774 printk("md: super_written gets error=%d, uptodate=%d\n",
775 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
776 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 777 md_error(mddev, rdev);
3a0f5bbb 778 }
7bfa19f2 779
a9701a30
N		 780 if (atomic_dec_and_test(&mddev->pending_writes))
781 wake_up(&mddev->sb_wait);
f8b58edf 782 bio_put(bio);
7bfa19f2
N		 783}
784
fd01b88c 785void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 786 sector_t sector, int size, struct page *page)
787{
788 /* write first size bytes of page to sector of rdev
789 * Increment mddev->pending_writes before returning
790 * and decrement it on completion, waking up sb_wait
791 * if zero is reached.
792 * If an error occurred, call md_error
793 */
a167f663 794 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 795
a6ff7e08 796 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
7bfa19f2
N		 797 bio->bi_sector = sector;
798 bio_add_page(bio, page, size, 0);
799 bio->bi_private = rdev;
800 bio->bi_end_io = super_written;
a9701a30 801
7bfa19f2 802 atomic_inc(&mddev->pending_writes);
a5bf4df0 803 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 804}
805
fd01b88c 806void md_super_wait(struct mddev *mddev)
a9701a30 807{
e9c7469b 808 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N		 809 DEFINE_WAIT(wq);
810 for(;;) {
811 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
812 if (atomic_read(&mddev->pending_writes)==0)
813 break;
a9701a30
N		 814 schedule();
815 }
816 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 817}
818
6712ecf8 819static void bi_complete(struct bio *bio, int error)
1da177e4 820{
1da177e4 821 complete((struct completion*)bio->bi_private);
1da177e4
LT		 822}
823
3cb03002 824int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 825 struct page *page, int rw, bool metadata_op)
1da177e4 826{
a167f663 827 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 828 struct completion event;
829 int ret;
830
721a9602 831 rw |= REQ_SYNC;
1da177e4 832
a6ff7e08
JB		 833 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
834 rdev->meta_bdev : rdev->bdev;
ccebd4c4
JB		 835 if (metadata_op)
836 bio->bi_sector = sector + rdev->sb_start;
1fdd6fc9
N		 837 else if (rdev->mddev->reshape_position != MaxSector &&
838 (rdev->mddev->reshape_backwards ==
839 (sector >= rdev->mddev->reshape_position)))
840 bio->bi_sector = sector + rdev->new_data_offset;
ccebd4c4
JB		 841 else
842 bio->bi_sector = sector + rdev->data_offset;
1da177e4
LT		 843 bio_add_page(bio, page, size, 0);
844 init_completion(&event);
845 bio->bi_private = &event;
846 bio->bi_end_io = bi_complete;
847 submit_bio(rw, bio);
848 wait_for_completion(&event);
849
850 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
851 bio_put(bio);
852 return ret;
853}
a8745db2 854EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 855
3cb03002 856static int read_disk_sb(struct md_rdev * rdev, int size)
1da177e4
LT		 857{
858 char b[BDEVNAME_SIZE];
859 if (!rdev->sb_page) {
860 MD_BUG();
861 return -EINVAL;
862 }
863 if (rdev->sb_loaded)
864 return 0;
865
866
ccebd4c4 867 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 868 goto fail;
869 rdev->sb_loaded = 1;
870 return 0;
871
872fail:
873 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
874 bdevname(rdev->bdev,b));
875 return -EINVAL;
876}
877
878static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
879{
05710466
AN		 880 return sb1->set_uuid0 == sb2->set_uuid0 &&
881 sb1->set_uuid1 == sb2->set_uuid1 &&
882 sb1->set_uuid2 == sb2->set_uuid2 &&
883 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 884}
885
1da177e4
LT		 886static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
887{
888 int ret;
889 mdp_super_t *tmp1, *tmp2;
890
891 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
892 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
893
894 if (!tmp1 || !tmp2) {
895 ret = 0;
35020f1a 896 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 897 goto abort;
898 }
899
900 *tmp1 = *sb1;
901 *tmp2 = *sb2;
902
903 /*
904 * nr_disks is not constant
905 */
906 tmp1->nr_disks = 0;
907 tmp2->nr_disks = 0;
908
ce0c8e05 909 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 910abort:
990a8baf
JJ		 911 kfree(tmp1);
912 kfree(tmp2);
1da177e4
LT		 913 return ret;
914}
915
4d167f09
N		 916
917static u32 md_csum_fold(u32 csum)
918{
919 csum = (csum & 0xffff) + (csum >> 16);
920 return (csum & 0xffff) + (csum >> 16);
921}
922
1da177e4
LT		 923static unsigned int calc_sb_csum(mdp_super_t * sb)
924{
4d167f09
N		 925 u64 newcsum = 0;
926 u32 *sb32 = (u32*)sb;
927 int i;
1da177e4
LT		 928 unsigned int disk_csum, csum;
929
930 disk_csum = sb->sb_csum;
931 sb->sb_csum = 0;
4d167f09
N		 932
933 for (i = 0; i < MD_SB_BYTES/4 ; i++)
934 newcsum += sb32[i];
935 csum = (newcsum & 0xffffffff) + (newcsum>>32);
936
937
938#ifdef CONFIG_ALPHA
939 /* This used to use csum_partial, which was wrong for several
940 * reasons including that different results are returned on
941 * different architectures. It isn't critical that we get exactly
942 * the same return value as before (we always csum_fold before
943 * testing, and that removes any differences). However as we
944 * know that csum_partial always returned a 16bit value on
945 * alphas, do a fold to maximise conformity to previous behaviour.
946 */
947 sb->sb_csum = md_csum_fold(disk_csum);
948#else
1da177e4 949 sb->sb_csum = disk_csum;
4d167f09 950#endif
1da177e4
LT		 951 return csum;
952}
953
954
955/*
956 * Handle superblock details.
957 * We want to be able to handle multiple superblock formats
958 * so we have a common interface to them all, and an array of
959 * different handlers.
960 * We rely on user-space to write the initial superblock, and support
961 * reading and updating of superblocks.
962 * Interface methods are:
3cb03002 963 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 964 * loads and validates a superblock on dev.
965 * if refdev != NULL, compare superblocks on both devices
966 * Return:
967 * 0 - dev has a superblock that is compatible with refdev
968 * 1 - dev has a superblock that is compatible and newer than refdev
969 * so dev should be used as the refdev in future
970 * -EINVAL superblock incompatible or invalid
971 * -othererror e.g. -EIO
972 *
fd01b88c 973 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 974 * Verify that dev is acceptable into mddev.
975 * The first time, mddev->raid_disks will be 0, and data from
976 * dev should be merged in. Subsequent calls check that dev
977 * is new enough. Return 0 or -EINVAL
978 *
fd01b88c 979 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 980 * Update the superblock for rdev with data in mddev
981 * This does not write to disc.
982 *
983 */
984
985struct super_type {
0cd17fec
CW		 986 char *name;
987 struct module *owner;
c6563a8c
N		 988 int (*load_super)(struct md_rdev *rdev,
989 struct md_rdev *refdev,
0cd17fec 990 int minor_version);
c6563a8c
N		 991 int (*validate_super)(struct mddev *mddev,
992 struct md_rdev *rdev);
993 void (*sync_super)(struct mddev *mddev,
994 struct md_rdev *rdev);
3cb03002 995 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 996 sector_t num_sectors);
c6563a8c
N		 997 int (*allow_new_offset)(struct md_rdev *rdev,
998 unsigned long long new_offset);
1da177e4
LT		 999};
1000
0894cc30
AN		 1001/*
1002 * Check that the given mddev has no bitmap.
1003 *
1004 * This function is called from the run method of all personalities that do not
1005 * support bitmaps. It prints an error message and returns non-zero if mddev
1006 * has a bitmap. Otherwise, it returns 0.
1007 *
1008 */
fd01b88c 1009int md_check_no_bitmap(struct mddev *mddev)
0894cc30 1010{
c3d9714e 1011 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 1012 return 0;
1013 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1014 mdname(mddev), mddev->pers->name);
1015 return 1;
1016}
1017EXPORT_SYMBOL(md_check_no_bitmap);
1018
1da177e4
LT		 1019/*
1020 * load_super for 0.90.0
1021 */
3cb03002 1022static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1023{
1024 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1025 mdp_super_t *sb;
1026 int ret;
1da177e4
LT		 1027
1028 /*
0f420358 1029 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 1030 * it's at the end of the disk.
1031 *
1032 * It also happens to be a multiple of 4Kb.
1033 */
57b2caa3 1034 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 1035
0002b271 1036 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 1037 if (ret) return ret;
1038
1039 ret = -EINVAL;
1040
1041 bdevname(rdev->bdev, b);
65a06f06 1042 sb = page_address(rdev->sb_page);
1da177e4
LT		 1043
1044 if (sb->md_magic != MD_SB_MAGIC) {
1045 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1046 b);
1047 goto abort;
1048 }
1049
1050 if (sb->major_version != 0 ||
f6705578
N		 1051 sb->minor_version < 90 ||
1052 sb->minor_version > 91) {
1da177e4
LT		 1053 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1054 sb->major_version, sb->minor_version,
1055 b);
1056 goto abort;
1057 }
1058
1059 if (sb->raid_disks <= 0)
1060 goto abort;
1061
4d167f09 1062 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1063 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1064 b);
1065 goto abort;
1066 }
1067
1068 rdev->preferred_minor = sb->md_minor;
1069 rdev->data_offset = 0;
c6563a8c 1070 rdev->new_data_offset = 0;
0002b271 1071 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1072 rdev->badblocks.shift = -1;
1da177e4
LT		 1073
1074 if (sb->level == LEVEL_MULTIPATH)
1075 rdev->desc_nr = -1;
1076 else
1077 rdev->desc_nr = sb->this_disk.number;
1078
9a7b2b0f 1079 if (!refdev) {
1da177e4 1080 ret = 1;
9a7b2b0f 1081 } else {
1da177e4 1082 __u64 ev1, ev2;
65a06f06 1083 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1084 if (!uuid_equal(refsb, sb)) {
1085 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1086 b, bdevname(refdev->bdev,b2));
1087 goto abort;
1088 }
1089 if (!sb_equal(refsb, sb)) {
1090 printk(KERN_WARNING "md: %s has same UUID"
1091 " but different superblock to %s\n",
1092 b, bdevname(refdev->bdev, b2));
1093 goto abort;
1094 }
1095 ev1 = md_event(sb);
1096 ev2 = md_event(refsb);
1097 if (ev1 > ev2)
1098 ret = 1;
1099 else
1100 ret = 0;
1101 }
8190e754 1102 rdev->sectors = rdev->sb_start;
667a5313
N		 1103 /* Limit to 4TB as metadata cannot record more than that.
1104 * (not needed for Linear and RAID0 as metadata doesn't
1105 * record this size)
1106 */
1107 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
27a7b260 1108 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1109
27a7b260 1110 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1111 /* "this cannot possibly happen" ... */
1112 ret = -EINVAL;
1113
1da177e4
LT		 1114 abort:
1115 return ret;
1116}
1117
1118/*
1119 * validate_super for 0.90.0
1120 */
fd01b88c 1121static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1122{
1123 mdp_disk_t *desc;
65a06f06 1124 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1125 __u64 ev1 = md_event(sb);
1da177e4 1126
41158c7e 1127 rdev->raid_disk = -1;
c5d79adb
N		 1128 clear_bit(Faulty, &rdev->flags);
1129 clear_bit(In_sync, &rdev->flags);
1130 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1131
1da177e4
LT		 1132 if (mddev->raid_disks == 0) {
1133 mddev->major_version = 0;
1134 mddev->minor_version = sb->minor_version;
1135 mddev->patch_version = sb->patch_version;
e691063a 1136 mddev->external = 0;
9d8f0363 1137 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1138 mddev->ctime = sb->ctime;
1139 mddev->utime = sb->utime;
1140 mddev->level = sb->level;
d9d166c2 1141 mddev->clevel[0] = 0;
1da177e4
LT		 1142 mddev->layout = sb->layout;
1143 mddev->raid_disks = sb->raid_disks;
27a7b260 1144 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1145 mddev->events = ev1;
c3d9714e 1146 mddev->bitmap_info.offset = 0;
6409bb05
N		 1147 mddev->bitmap_info.space = 0;
1148 /* bitmap can use 60 K after the 4K superblocks */
c3d9714e 1149 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 1150 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
2c810cdd 1151 mddev->reshape_backwards = 0;
1da177e4 1152
f6705578
N		 1153 if (mddev->minor_version >= 91) {
1154 mddev->reshape_position = sb->reshape_position;
1155 mddev->delta_disks = sb->delta_disks;
1156 mddev->new_level = sb->new_level;
1157 mddev->new_layout = sb->new_layout;
664e7c41 1158 mddev->new_chunk_sectors = sb->new_chunk >> 9;
2c810cdd
N		 1159 if (mddev->delta_disks < 0)
1160 mddev->reshape_backwards = 1;
f6705578
N		 1161 } else {
1162 mddev->reshape_position = MaxSector;
1163 mddev->delta_disks = 0;
1164 mddev->new_level = mddev->level;
1165 mddev->new_layout = mddev->layout;
664e7c41 1166 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1167 }
1168
1da177e4
LT		 1169 if (sb->state & (1<<MD_SB_CLEAN))
1170 mddev->recovery_cp = MaxSector;
1171 else {
1172 if (sb->events_hi == sb->cp_events_hi &&
1173 sb->events_lo == sb->cp_events_lo) {
1174 mddev->recovery_cp = sb->recovery_cp;
1175 } else
1176 mddev->recovery_cp = 0;
1177 }
1178
1179 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1180 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1181 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1182 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1183
1184 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1185
1186 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
6409bb05 1187 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1188 mddev->bitmap_info.offset =
1189 mddev->bitmap_info.default_offset;
6409bb05
N		 1190 mddev->bitmap_info.space =
1191 mddev->bitmap_info.space;
1192 }
a654b9d8 1193
41158c7e 1194 } else if (mddev->pers == NULL) {
be6800a7
N		 1195 /* Insist on good event counter while assembling, except
1196 * for spares (which don't need an event count) */
1da177e4 1197 ++ev1;
be6800a7
N		 1198 if (sb->disks[rdev->desc_nr].state & (
1199 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1200 if (ev1 < mddev->events)
1201 return -EINVAL;
41158c7e
N		 1202 } else if (mddev->bitmap) {
1203 /* if adding to array with a bitmap, then we can accept an
1204 * older device ... but not too old.
1205 */
41158c7e
N		 1206 if (ev1 < mddev->bitmap->events_cleared)
1207 return 0;
07d84d10
N		 1208 } else {
1209 if (ev1 < mddev->events)
1210 /* just a hot-add of a new device, leave raid_disk at -1 */
1211 return 0;
1212 }
41158c7e 1213
1da177e4 1214 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1215 desc = sb->disks + rdev->desc_nr;
1216
1217 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1218 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1219 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1220 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1221 set_bit(In_sync, &rdev->flags);
1da177e4 1222 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1223 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1224 /* active but not in sync implies recovery up to
1225 * reshape position. We don't know exactly where
1226 * that is, so set to zero for now */
1227 if (mddev->minor_version >= 91) {
1228 rdev->recovery_offset = 0;
1229 rdev->raid_disk = desc->raid_disk;
1230 }
1da177e4 1231 }
8ddf9efe
N		 1232 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1233 set_bit(WriteMostly, &rdev->flags);
41158c7e 1234 } else /* MULTIPATH are always insync */
b2d444d7 1235 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1236 return 0;
1237}
1238
1239/*
1240 * sync_super for 0.90.0
1241 */
fd01b88c 1242static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1243{
1244 mdp_super_t *sb;
3cb03002 1245 struct md_rdev *rdev2;
1da177e4 1246 int next_spare = mddev->raid_disks;
19133a42 1247
1da177e4
LT		 1248
1249 /* make rdev->sb match mddev data..
1250 *
1251 * 1/ zero out disks
1252 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1253 * 3/ any empty disks < next_spare become removed
1254 *
1255 * disks[0] gets initialised to REMOVED because
1256 * we cannot be sure from other fields if it has
1257 * been initialised or not.
1258 */
1259 int i;
1260 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1261
61181565
N		 1262 rdev->sb_size = MD_SB_BYTES;
1263
65a06f06 1264 sb = page_address(rdev->sb_page);
1da177e4
LT		 1265
1266 memset(sb, 0, sizeof(*sb));
1267
1268 sb->md_magic = MD_SB_MAGIC;
1269 sb->major_version = mddev->major_version;
1da177e4
LT		 1270 sb->patch_version = mddev->patch_version;
1271 sb->gvalid_words = 0; /* ignored */
1272 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1273 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1274 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1275 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1276
1277 sb->ctime = mddev->ctime;
1278 sb->level = mddev->level;
58c0fed4 1279 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1280 sb->raid_disks = mddev->raid_disks;
1281 sb->md_minor = mddev->md_minor;
e691063a 1282 sb->not_persistent = 0;
1da177e4
LT		 1283 sb->utime = mddev->utime;
1284 sb->state = 0;
1285 sb->events_hi = (mddev->events>>32);
1286 sb->events_lo = (u32)mddev->events;
1287
f6705578
N		 1288 if (mddev->reshape_position == MaxSector)
1289 sb->minor_version = 90;
1290 else {
1291 sb->minor_version = 91;
1292 sb->reshape_position = mddev->reshape_position;
1293 sb->new_level = mddev->new_level;
1294 sb->delta_disks = mddev->delta_disks;
1295 sb->new_layout = mddev->new_layout;
664e7c41 1296 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1297 }
1298 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1299 if (mddev->in_sync)
1300 {
1301 sb->recovery_cp = mddev->recovery_cp;
1302 sb->cp_events_hi = (mddev->events>>32);
1303 sb->cp_events_lo = (u32)mddev->events;
1304 if (mddev->recovery_cp == MaxSector)
1305 sb->state = (1<< MD_SB_CLEAN);
1306 } else
1307 sb->recovery_cp = 0;
1308
1309 sb->layout = mddev->layout;
9d8f0363 1310 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1311
c3d9714e 1312 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1313 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1314
1da177e4 1315 sb->disks[0].state = (1<<MD_DISK_REMOVED);
dafb20fa 1316 rdev_for_each(rdev2, mddev) {
1da177e4 1317 mdp_disk_t *d;
86e6ffdd 1318 int desc_nr;
0261cd9f
N		 1319 int is_active = test_bit(In_sync, &rdev2->flags);
1320
1321 if (rdev2->raid_disk >= 0 &&
1322 sb->minor_version >= 91)
1323 /* we have nowhere to store the recovery_offset,
1324 * but if it is not below the reshape_position,
1325 * we can piggy-back on that.
1326 */
1327 is_active = 1;
1328 if (rdev2->raid_disk < 0 ||
1329 test_bit(Faulty, &rdev2->flags))
1330 is_active = 0;
1331 if (is_active)
86e6ffdd 1332 desc_nr = rdev2->raid_disk;
1da177e4 1333 else
86e6ffdd 1334 desc_nr = next_spare++;
19133a42 1335 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1336 d = &sb->disks[rdev2->desc_nr];
1337 nr_disks++;
1338 d->number = rdev2->desc_nr;
1339 d->major = MAJOR(rdev2->bdev->bd_dev);
1340 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1341 if (is_active)
1da177e4
LT		 1342 d->raid_disk = rdev2->raid_disk;
1343 else
1344 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1345 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1346 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1347 else if (is_active) {
1da177e4 1348 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1349 if (test_bit(In_sync, &rdev2->flags))
1350 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1351 active++;
1352 working++;
1353 } else {
1354 d->state = 0;
1355 spare++;
1356 working++;
1357 }
8ddf9efe
N		 1358 if (test_bit(WriteMostly, &rdev2->flags))
1359 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1360 }
1da177e4
LT		 1361 /* now set the "removed" and "faulty" bits on any missing devices */
1362 for (i=0 ; i < mddev->raid_disks ; i++) {
1363 mdp_disk_t *d = &sb->disks[i];
1364 if (d->state == 0 && d->number == 0) {
1365 d->number = i;
1366 d->raid_disk = i;
1367 d->state = (1<<MD_DISK_REMOVED);
1368 d->state |= (1<<MD_DISK_FAULTY);
1369 failed++;
1370 }
1371 }
1372 sb->nr_disks = nr_disks;
1373 sb->active_disks = active;
1374 sb->working_disks = working;
1375 sb->failed_disks = failed;
1376 sb->spare_disks = spare;
1377
1378 sb->this_disk = sb->disks[rdev->desc_nr];
1379 sb->sb_csum = calc_sb_csum(sb);
1380}
1381
0cd17fec
CW		 1382/*
1383 * rdev_size_change for 0.90.0
1384 */
1385static unsigned long long
3cb03002 1386super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1387{
58c0fed4 1388 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1389 return 0; /* component must fit device */
c3d9714e 1390 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1391 return 0; /* can't move bitmap */
57b2caa3 1392 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1393 if (!num_sectors || num_sectors > rdev->sb_start)
1394 num_sectors = rdev->sb_start;
27a7b260
N		 1395 /* Limit to 4TB as metadata cannot record more than that.
1396 * 4TB == 2^32 KB, or 2*2^32 sectors.
1397 */
667a5313 1398 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
27a7b260 1399 num_sectors = (2ULL << 32) - 2;
0f420358 1400 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1401 rdev->sb_page);
1402 md_super_wait(rdev->mddev);
c26a44ed 1403 return num_sectors;
0cd17fec
CW		 1404}
1405
c6563a8c
N		 1406static int
1407super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1408{
1409 /* non-zero offset changes not possible with v0.90 */
1410 return new_offset == 0;
1411}
0cd17fec 1412
1da177e4
LT		 1413/*
1414 * version 1 superblock
1415 */
1416
1c05b4bc 1417static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1418{
1c05b4bc
N		 1419 __le32 disk_csum;
1420 u32 csum;
1da177e4
LT		 1421 unsigned long long newcsum;
1422 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1423 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1424
1425 disk_csum = sb->sb_csum;
1426 sb->sb_csum = 0;
1427 newcsum = 0;
1f3c9907 1428 for (; size >= 4; size -= 4)
1da177e4
LT		 1429 newcsum += le32_to_cpu(*isuper++);
1430
1431 if (size == 2)
1c05b4bc 1432 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1433
1434 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1435 sb->sb_csum = disk_csum;
1436 return cpu_to_le32(csum);
1437}
1438
2699b672
N		 1439static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1440 int acknowledged);
3cb03002 1441static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1442{
1443 struct mdp_superblock_1 *sb;
1444 int ret;
0f420358 1445 sector_t sb_start;
c6563a8c 1446 sector_t sectors;
1da177e4 1447 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1448 int bmask;
1da177e4
LT		 1449
1450 /*
0f420358 1451 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1452 * It is always aligned to a 4K boundary and
1453 * depeding on minor_version, it can be:
1454 * 0: At least 8K, but less than 12K, from end of device
1455 * 1: At start of device
1456 * 2: 4K from start of device.
1457 */
1458 switch(minor_version) {
1459 case 0:
77304d2a 1460 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1461 sb_start -= 8*2;
1462 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1463 break;
1464 case 1:
0f420358 1465 sb_start = 0;
1da177e4
LT		 1466 break;
1467 case 2:
0f420358 1468 sb_start = 8;
1da177e4
LT		 1469 break;
1470 default:
1471 return -EINVAL;
1472 }
0f420358 1473 rdev->sb_start = sb_start;
1da177e4 1474
0002b271
N		 1475 /* superblock is rarely larger than 1K, but it can be larger,
1476 * and it is safe to read 4k, so we do that
1477 */
1478 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1479 if (ret) return ret;
1480
1481
65a06f06 1482 sb = page_address(rdev->sb_page);
1da177e4
LT		 1483
1484 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1485 sb->major_version != cpu_to_le32(1) ||
1486 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1487 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1488 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1489 return -EINVAL;
1490
1491 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1492 printk("md: invalid superblock checksum on %s\n",
1493 bdevname(rdev->bdev,b));
1494 return -EINVAL;
1495 }
1496 if (le64_to_cpu(sb->data_size) < 10) {
1497 printk("md: data_size too small on %s\n",
1498 bdevname(rdev->bdev,b));
1499 return -EINVAL;
1500 }
c6563a8c
N		 1501 if (sb->pad0 ||
1502 sb->pad3[0] ||
1503 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1504 /* Some padding is non-zero, might be a new feature */
1505 return -EINVAL;
e11e93fa 1506
1da177e4
LT		 1507 rdev->preferred_minor = 0xffff;
1508 rdev->data_offset = le64_to_cpu(sb->data_offset);
c6563a8c
N		 1509 rdev->new_data_offset = rdev->data_offset;
1510 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1511 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1512 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
4dbcdc75 1513 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1514
0002b271 1515 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1516 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1517 if (rdev->sb_size & bmask)
a1801f85
N		 1518 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1519
1520 if (minor_version
0f420358 1521 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1522 return -EINVAL;
c6563a8c
N		 1523 if (minor_version
1524 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1525 return -EINVAL;
0002b271 1526
31b65a0d
N		 1527 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1528 rdev->desc_nr = -1;
1529 else
1530 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1531
2699b672
N		 1532 if (!rdev->bb_page) {
1533 rdev->bb_page = alloc_page(GFP_KERNEL);
1534 if (!rdev->bb_page)
1535 return -ENOMEM;
1536 }
1537 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1538 rdev->badblocks.count == 0) {
1539 /* need to load the bad block list.
1540 * Currently we limit it to one page.
1541 */
1542 s32 offset;
1543 sector_t bb_sector;
1544 u64 *bbp;
1545 int i;
1546 int sectors = le16_to_cpu(sb->bblog_size);
1547 if (sectors > (PAGE_SIZE / 512))
1548 return -EINVAL;
1549 offset = le32_to_cpu(sb->bblog_offset);
1550 if (offset == 0)
1551 return -EINVAL;
1552 bb_sector = (long long)offset;
1553 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1554 rdev->bb_page, READ, true))
1555 return -EIO;
1556 bbp = (u64 *)page_address(rdev->bb_page);
1557 rdev->badblocks.shift = sb->bblog_shift;
1558 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1559 u64 bb = le64_to_cpu(*bbp);
1560 int count = bb & (0x3ff);
1561 u64 sector = bb >> 10;
1562 sector <<= sb->bblog_shift;
1563 count <<= sb->bblog_shift;
1564 if (bb + 1 == 0)
1565 break;
1566 if (md_set_badblocks(&rdev->badblocks,
1567 sector, count, 1) == 0)
1568 return -EINVAL;
1569 }
1570 } else if (sb->bblog_offset == 0)
1571 rdev->badblocks.shift = -1;
1572
9a7b2b0f 1573 if (!refdev) {
8ed75463 1574 ret = 1;
9a7b2b0f 1575 } else {
1da177e4 1576 __u64 ev1, ev2;
65a06f06 1577 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1578
1579 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1580 sb->level != refsb->level ||
1581 sb->layout != refsb->layout ||
1582 sb->chunksize != refsb->chunksize) {
1583 printk(KERN_WARNING "md: %s has strangely different"
1584 " superblock to %s\n",
1585 bdevname(rdev->bdev,b),
1586 bdevname(refdev->bdev,b2));
1587 return -EINVAL;
1588 }
1589 ev1 = le64_to_cpu(sb->events);
1590 ev2 = le64_to_cpu(refsb->events);
1591
1592 if (ev1 > ev2)
8ed75463
N		 1593 ret = 1;
1594 else
1595 ret = 0;
1da177e4 1596 }
c6563a8c
N		 1597 if (minor_version) {
1598 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1599 sectors -= rdev->data_offset;
1600 } else
1601 sectors = rdev->sb_start;
1602 if (sectors < le64_to_cpu(sb->data_size))
1da177e4 1603 return -EINVAL;
dd8ac336 1604 rdev->sectors = le64_to_cpu(sb->data_size);
8ed75463 1605 return ret;
1da177e4
LT		 1606}
1607
fd01b88c 1608static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1609{
65a06f06 1610 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1611 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1612
41158c7e 1613 rdev->raid_disk = -1;
c5d79adb
N		 1614 clear_bit(Faulty, &rdev->flags);
1615 clear_bit(In_sync, &rdev->flags);
1616 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1617
1da177e4
LT		 1618 if (mddev->raid_disks == 0) {
1619 mddev->major_version = 1;
1620 mddev->patch_version = 0;
e691063a 1621 mddev->external = 0;
9d8f0363 1622 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1623 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1624 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1625 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1626 mddev->clevel[0] = 0;
1da177e4
LT		 1627 mddev->layout = le32_to_cpu(sb->layout);
1628 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1629 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1630 mddev->events = ev1;
c3d9714e 1631 mddev->bitmap_info.offset = 0;
6409bb05
N		 1632 mddev->bitmap_info.space = 0;
1633 /* Default location for bitmap is 1K after superblock
1634 * using 3K - total of 4K
1635 */
c3d9714e 1636 mddev->bitmap_info.default_offset = 1024 >> 9;
6409bb05 1637 mddev->bitmap_info.default_space = (4096-1024) >> 9;
2c810cdd
N		 1638 mddev->reshape_backwards = 0;
1639
1da177e4
LT		 1640 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1641 memcpy(mddev->uuid, sb->set_uuid, 16);
1642
1643 mddev->max_disks = (4096-256)/2;
a654b9d8 1644
71c0805c 1645 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
6409bb05 1646 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1647 mddev->bitmap_info.offset =
1648 (__s32)le32_to_cpu(sb->bitmap_offset);
6409bb05
N		 1649 /* Metadata doesn't record how much space is available.
1650 * For 1.0, we assume we can use up to the superblock
1651 * if before, else to 4K beyond superblock.
1652 * For others, assume no change is possible.
1653 */
1654 if (mddev->minor_version > 0)
1655 mddev->bitmap_info.space = 0;
1656 else if (mddev->bitmap_info.offset > 0)
1657 mddev->bitmap_info.space =
1658 8 - mddev->bitmap_info.offset;
1659 else
1660 mddev->bitmap_info.space =
1661 -mddev->bitmap_info.offset;
1662 }
e11e93fa 1663
f6705578
N		 1664 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1665 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1666 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1667 mddev->new_level = le32_to_cpu(sb->new_level);
1668 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1669 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
2c810cdd
N		 1670 if (mddev->delta_disks < 0 ||
1671 (mddev->delta_disks == 0 &&
1672 (le32_to_cpu(sb->feature_map)
1673 & MD_FEATURE_RESHAPE_BACKWARDS)))
1674 mddev->reshape_backwards = 1;
f6705578
N		 1675 } else {
1676 mddev->reshape_position = MaxSector;
1677 mddev->delta_disks = 0;
1678 mddev->new_level = mddev->level;
1679 mddev->new_layout = mddev->layout;
664e7c41 1680 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1681 }
1682
41158c7e 1683 } else if (mddev->pers == NULL) {
be6800a7
N		 1684 /* Insist of good event counter while assembling, except for
1685 * spares (which don't need an event count) */
1da177e4 1686 ++ev1;
be6800a7
N		 1687 if (rdev->desc_nr >= 0 &&
1688 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1689 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1690 if (ev1 < mddev->events)
1691 return -EINVAL;
41158c7e
N		 1692 } else if (mddev->bitmap) {
1693 /* If adding to array with a bitmap, then we can accept an
1694 * older device, but not too old.
1695 */
41158c7e
N		 1696 if (ev1 < mddev->bitmap->events_cleared)
1697 return 0;
07d84d10
N		 1698 } else {
1699 if (ev1 < mddev->events)
1700 /* just a hot-add of a new device, leave raid_disk at -1 */
1701 return 0;
1702 }
1da177e4
LT		 1703 if (mddev->level != LEVEL_MULTIPATH) {
1704 int role;
3673f305
N		 1705 if (rdev->desc_nr < 0 ||
1706 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1707 role = 0xffff;
1708 rdev->desc_nr = -1;
1709 } else
1710 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1711 switch(role) {
1712 case 0xffff: /* spare */
1da177e4
LT		 1713 break;
1714 case 0xfffe: /* faulty */
b2d444d7 1715 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1716 break;
1717 default:
5fd6c1dc
N		 1718 if ((le32_to_cpu(sb->feature_map) &
1719 MD_FEATURE_RECOVERY_OFFSET))
1720 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1721 else
1722 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1723 rdev->raid_disk = role;
1724 break;
1725 }
8ddf9efe
N		 1726 if (sb->devflags & WriteMostly1)
1727 set_bit(WriteMostly, &rdev->flags);
2d78f8c4
N		 1728 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1729 set_bit(Replacement, &rdev->flags);
41158c7e 1730 } else /* MULTIPATH are always insync */
b2d444d7 1731 set_bit(In_sync, &rdev->flags);
41158c7e 1732
1da177e4
LT		 1733 return 0;
1734}
1735
fd01b88c 1736static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1737{
1738 struct mdp_superblock_1 *sb;
3cb03002 1739 struct md_rdev *rdev2;
1da177e4
LT		 1740 int max_dev, i;
1741 /* make rdev->sb match mddev and rdev data. */
1742
65a06f06 1743 sb = page_address(rdev->sb_page);
1da177e4
LT		 1744
1745 sb->feature_map = 0;
1746 sb->pad0 = 0;
5fd6c1dc 1747 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1748 memset(sb->pad3, 0, sizeof(sb->pad3));
1749
1750 sb->utime = cpu_to_le64((__u64)mddev->utime);
1751 sb->events = cpu_to_le64(mddev->events);
1752 if (mddev->in_sync)
1753 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1754 else
1755 sb->resync_offset = cpu_to_le64(0);
1756
1c05b4bc 1757 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1758
f0ca340c 1759 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1760 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1761 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1762 sb->level = cpu_to_le32(mddev->level);
1763 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1764
aeb9b211
N		 1765 if (test_bit(WriteMostly, &rdev->flags))
1766 sb->devflags |= WriteMostly1;
1767 else
1768 sb->devflags &= ~WriteMostly1;
c6563a8c
N		 1769 sb->data_offset = cpu_to_le64(rdev->data_offset);
1770 sb->data_size = cpu_to_le64(rdev->sectors);
aeb9b211 1771
c3d9714e
N		 1772 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1773 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1774 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1775 }
5fd6c1dc
N		 1776
1777 if (rdev->raid_disk >= 0 &&
97e4f42d 1778 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1779 sb->feature_map |=
1780 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1781 sb->recovery_offset =
1782 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc 1783 }
2d78f8c4
N		 1784 if (test_bit(Replacement, &rdev->flags))
1785 sb->feature_map |=
1786 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1787
f6705578
N		 1788 if (mddev->reshape_position != MaxSector) {
1789 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1790 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1791 sb->new_layout = cpu_to_le32(mddev->new_layout);
1792 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1793 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1794 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2c810cdd
N		 1795 if (mddev->delta_disks == 0 &&
1796 mddev->reshape_backwards)
1797 sb->feature_map
1798 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
c6563a8c
N		 1799 if (rdev->new_data_offset != rdev->data_offset) {
1800 sb->feature_map
1801 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1802 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1803 - rdev->data_offset));
1804 }
f6705578 1805 }
a654b9d8 1806
2699b672
N		 1807 if (rdev->badblocks.count == 0)
1808 /* Nothing to do for bad blocks*/ ;
1809 else if (sb->bblog_offset == 0)
1810 /* Cannot record bad blocks on this device */
1811 md_error(mddev, rdev);
1812 else {
1813 struct badblocks *bb = &rdev->badblocks;
1814 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1815 u64 *p = bb->page;
1816 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1817 if (bb->changed) {
1818 unsigned seq;
1819
1820retry:
1821 seq = read_seqbegin(&bb->lock);
1822
1823 memset(bbp, 0xff, PAGE_SIZE);
1824
1825 for (i = 0 ; i < bb->count ; i++) {
35f9ac2d 1826 u64 internal_bb = p[i];
2699b672
N		 1827 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1828 | BB_LEN(internal_bb));
35f9ac2d 1829 bbp[i] = cpu_to_le64(store_bb);
2699b672 1830 }
d0962936 1831 bb->changed = 0;
2699b672
N		 1832 if (read_seqretry(&bb->lock, seq))
1833 goto retry;
1834
1835 bb->sector = (rdev->sb_start +
1836 (int)le32_to_cpu(sb->bblog_offset));
1837 bb->size = le16_to_cpu(sb->bblog_size);
2699b672
N		 1838 }
1839 }
1840
1da177e4 1841 max_dev = 0;
dafb20fa 1842 rdev_for_each(rdev2, mddev)
1da177e4
LT		 1843 if (rdev2->desc_nr+1 > max_dev)
1844 max_dev = rdev2->desc_nr+1;
a778b73f 1845
70471daf
N		 1846 if (max_dev > le32_to_cpu(sb->max_dev)) {
1847 int bmask;
a778b73f 1848 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1849 rdev->sb_size = max_dev * 2 + 256;
1850 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1851 if (rdev->sb_size & bmask)
1852 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1853 } else
1854 max_dev = le32_to_cpu(sb->max_dev);
1855
1da177e4
LT		 1856 for (i=0; i<max_dev;i++)
1857 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1858
dafb20fa 1859 rdev_for_each(rdev2, mddev) {
1da177e4 1860 i = rdev2->desc_nr;
b2d444d7 1861 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1862 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1863 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1864 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1865 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1866 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1867 else
1868 sb->dev_roles[i] = cpu_to_le16(0xffff);
1869 }
1870
1da177e4
LT		 1871 sb->sb_csum = calc_sb_1_csum(sb);
1872}
1873
0cd17fec 1874static unsigned long long
3cb03002 1875super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 1876{
1877 struct mdp_superblock_1 *sb;
15f4a5fd 1878 sector_t max_sectors;
58c0fed4 1879 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1880 return 0; /* component must fit device */
c6563a8c
N		 1881 if (rdev->data_offset != rdev->new_data_offset)
1882 return 0; /* too confusing */
0f420358 1883 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1884 /* minor versions 1 and 2; superblock before data */
77304d2a 1885 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1886 max_sectors -= rdev->data_offset;
1887 if (!num_sectors || num_sectors > max_sectors)
1888 num_sectors = max_sectors;
c3d9714e 1889 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1890 /* minor version 0 with bitmap we can't move */
1891 return 0;
1892 } else {
1893 /* minor version 0; superblock after data */
0f420358 1894 sector_t sb_start;
77304d2a 1895 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1896 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1897 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1898 if (!num_sectors || num_sectors > max_sectors)
1899 num_sectors = max_sectors;
0f420358 1900 rdev->sb_start = sb_start;
0cd17fec 1901 }
65a06f06 1902 sb = page_address(rdev->sb_page);
15f4a5fd 1903 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1904 sb->super_offset = rdev->sb_start;
0cd17fec 1905 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1906 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1907 rdev->sb_page);
1908 md_super_wait(rdev->mddev);
c26a44ed 1909 return num_sectors;
c6563a8c
N		 1910
1911}
1912
1913static int
1914super_1_allow_new_offset(struct md_rdev *rdev,
1915 unsigned long long new_offset)
1916{
1917 /* All necessary checks on new >= old have been done */
1918 struct bitmap *bitmap;
1919 if (new_offset >= rdev->data_offset)
1920 return 1;
1921
1922 /* with 1.0 metadata, there is no metadata to tread on
1923 * so we can always move back */
1924 if (rdev->mddev->minor_version == 0)
1925 return 1;
1926
1927 /* otherwise we must be sure not to step on
1928 * any metadata, so stay:
1929 * 36K beyond start of superblock
1930 * beyond end of badblocks
1931 * beyond write-intent bitmap
1932 */
1933 if (rdev->sb_start + (32+4)*2 > new_offset)
1934 return 0;
1935 bitmap = rdev->mddev->bitmap;
1936 if (bitmap && !rdev->mddev->bitmap_info.file &&
1937 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1ec885cd 1938 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
c6563a8c
N		 1939 return 0;
1940 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1941 return 0;
1942
1943 return 1;
0cd17fec 1944}
1da177e4 1945
75c96f85 1946static struct super_type super_types[] = {
1da177e4
LT		 1947 [0] = {
1948 .name = "0.90.0",
1949 .owner = THIS_MODULE,
0cd17fec
CW		 1950 .load_super = super_90_load,
1951 .validate_super = super_90_validate,
1952 .sync_super = super_90_sync,
1953 .rdev_size_change = super_90_rdev_size_change,
c6563a8c 1954 .allow_new_offset = super_90_allow_new_offset,
1da177e4
LT		 1955 },
1956 [1] = {
1957 .name = "md-1",
1958 .owner = THIS_MODULE,
0cd17fec
CW		 1959 .load_super = super_1_load,
1960 .validate_super = super_1_validate,
1961 .sync_super = super_1_sync,
1962 .rdev_size_change = super_1_rdev_size_change,
c6563a8c 1963 .allow_new_offset = super_1_allow_new_offset,
1da177e4
LT		 1964 },
1965};
1da177e4 1966
fd01b88c 1967static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 1968{
1969 if (mddev->sync_super) {
1970 mddev->sync_super(mddev, rdev);
1971 return;
1972 }
1973
1974 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1975
1976 super_types[mddev->major_version].sync_super(mddev, rdev);
1977}
1978
fd01b88c 1979static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1980{
3cb03002 1981 struct md_rdev *rdev, *rdev2;
1da177e4 1982
4b80991c
N		 1983 rcu_read_lock();
1984 rdev_for_each_rcu(rdev, mddev1)
1985 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1986 if (rdev->bdev->bd_contains ==
4b80991c
N		 1987 rdev2->bdev->bd_contains) {
1988 rcu_read_unlock();
7dd5e7c3 1989 return 1;
4b80991c
N		 1990 }
1991 rcu_read_unlock();
1da177e4
LT		 1992 return 0;
1993}
1994
1995static LIST_HEAD(pending_raid_disks);
1996
ac5e7113
AN		 1997/*
1998 * Try to register data integrity profile for an mddev
1999 *
2000 * This is called when an array is started and after a disk has been kicked
2001 * from the array. It only succeeds if all working and active component devices
2002 * are integrity capable with matching profiles.
2003 */
fd01b88c 2004int md_integrity_register(struct mddev *mddev)
ac5e7113 2005{
3cb03002 2006 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 2007
2008 if (list_empty(&mddev->disks))
2009 return 0; /* nothing to do */
629acb6a
JB		 2010 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2011 return 0; /* shouldn't register, or already is */
dafb20fa 2012 rdev_for_each(rdev, mddev) {
ac5e7113
AN		 2013 /* skip spares and non-functional disks */
2014 if (test_bit(Faulty, &rdev->flags))
2015 continue;
2016 if (rdev->raid_disk < 0)
2017 continue;
ac5e7113
AN		 2018 if (!reference) {
2019 /* Use the first rdev as the reference */
2020 reference = rdev;
2021 continue;
2022 }
2023 /* does this rdev's profile match the reference profile? */
2024 if (blk_integrity_compare(reference->bdev->bd_disk,
2025 rdev->bdev->bd_disk) < 0)
2026 return -EINVAL;
2027 }
89078d57
MP		 2028 if (!reference || !bdev_get_integrity(reference->bdev))
2029 return 0;
ac5e7113
AN		 2030 /*
2031 * All component devices are integrity capable and have matching
2032 * profiles, register the common profile for the md device.
2033 */
2034 if (blk_integrity_register(mddev->gendisk,
2035 bdev_get_integrity(reference->bdev)) != 0) {
2036 printk(KERN_ERR "md: failed to register integrity for %s\n",
2037 mdname(mddev));
2038 return -EINVAL;
2039 }
a91a2785
MP		 2040 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2041 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2042 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2043 mdname(mddev));
2044 return -EINVAL;
2045 }
ac5e7113
AN		 2046 return 0;
2047}
2048EXPORT_SYMBOL(md_integrity_register);
2049
2050/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 2051void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 2052{
2863b9eb
JB		 2053 struct blk_integrity *bi_rdev;
2054 struct blk_integrity *bi_mddev;
2055
2056 if (!mddev->gendisk)
2057 return;
2058
2059 bi_rdev = bdev_get_integrity(rdev->bdev);
2060 bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 2061
ac5e7113 2062 if (!bi_mddev) /* nothing to do */
3f9d99c1 2063 return;
ac5e7113 2064 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 2065 return;
ac5e7113
AN		 2066 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2067 rdev->bdev->bd_disk) >= 0)
2068 return;
2069 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2070 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2071}
ac5e7113 2072EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2073
fd01b88c 2074static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
1da177e4 2075{
7dd5e7c3 2076 char b[BDEVNAME_SIZE];
f637b9f9 2077 struct kobject *ko;
1edf80d3 2078 char *s;
5e55e2f5 2079 int err;
1da177e4
LT		 2080
2081 if (rdev->mddev) {
2082 MD_BUG();
2083 return -EINVAL;
2084 }
11e2ede0
DW		 2085
2086 /* prevent duplicates */
2087 if (find_rdev(mddev, rdev->bdev->bd_dev))
2088 return -EEXIST;
2089
dd8ac336
AN		 2090 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2091 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2092 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2093 if (mddev->pers) {
2094 /* Cannot change size, so fail
2095 * If mddev->level <= 0, then we don't care
2096 * about aligning sizes (e.g. linear)
2097 */
2098 if (mddev->level > 0)
2099 return -ENOSPC;
2100 } else
dd8ac336 2101 mddev->dev_sectors = rdev->sectors;
2bf071bf 2102 }
1da177e4
LT		 2103
2104 /* Verify rdev->desc_nr is unique.
2105 * If it is -1, assign a free number, else
2106 * check number is not in use
2107 */
2108 if (rdev->desc_nr < 0) {
2109 int choice = 0;
2110 if (mddev->pers) choice = mddev->raid_disks;
2111 while (find_rdev_nr(mddev, choice))
2112 choice++;
2113 rdev->desc_nr = choice;
2114 } else {
2115 if (find_rdev_nr(mddev, rdev->desc_nr))
2116 return -EBUSY;
2117 }
de01dfad
N		 2118 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2119 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2120 mdname(mddev), mddev->max_disks);
2121 return -EBUSY;
2122 }
19133a42 2123 bdevname(rdev->bdev,b);
649316b2 2124 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2125 *s = '!';
649316b2 2126
1da177e4 2127 rdev->mddev = mddev;
19133a42 2128 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2129
b2d6db58 2130 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2131 goto fail;
86e6ffdd 2132
0762b8bd 2133 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2134 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2135 /* failure here is OK */;
2136 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2137
4b80991c 2138 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2139 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2140
2141 /* May as well allow recovery to be retried once */
5389042f 2142 mddev->recovery_disabled++;
3f9d99c1 2143
1da177e4 2144 return 0;
5e55e2f5
N		 2145
2146 fail:
2147 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2148 b, mdname(mddev));
2149 return err;
1da177e4
LT		 2150}
2151
177a99b2 2152static void md_delayed_delete(struct work_struct *ws)
5792a285 2153{
3cb03002 2154 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2155 kobject_del(&rdev->kobj);
177a99b2 2156 kobject_put(&rdev->kobj);
5792a285
N		 2157}
2158
3cb03002 2159static void unbind_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2160{
2161 char b[BDEVNAME_SIZE];
2162 if (!rdev->mddev) {
2163 MD_BUG();
2164 return;
2165 }
49731baa 2166 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2167 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2168 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2169 rdev->mddev = NULL;
86e6ffdd 2170 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2171 sysfs_put(rdev->sysfs_state);
2172 rdev->sysfs_state = NULL;
2230dfe4 2173 rdev->badblocks.count = 0;
5792a285 2174 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2175 * writing to 'remove' to "dev/state". We also need
2176 * to delay it due to rcu usage.
5792a285 2177 */
4b80991c 2178 synchronize_rcu();
177a99b2
N		 2179 INIT_WORK(&rdev->del_work, md_delayed_delete);
2180 kobject_get(&rdev->kobj);
e804ac78 2181 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2182}
2183
2184/*
2185 * prevent the device from being mounted, repartitioned or
2186 * otherwise reused by a RAID array (or any other kernel
2187 * subsystem), by bd_claiming the device.
2188 */
3cb03002 2189static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2190{
2191 int err = 0;
2192 struct block_device *bdev;
2193 char b[BDEVNAME_SIZE];
2194
d4d77629 2195 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2196 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2197 if (IS_ERR(bdev)) {
2198 printk(KERN_ERR "md: could not open %s.\n",
2199 __bdevname(dev, b));
2200 return PTR_ERR(bdev);
2201 }
1da177e4
LT		 2202 rdev->bdev = bdev;
2203 return err;
2204}
2205
3cb03002 2206static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2207{
2208 struct block_device *bdev = rdev->bdev;
2209 rdev->bdev = NULL;
2210 if (!bdev)
2211 MD_BUG();
e525fd89 2212 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2213}
2214
2215void md_autodetect_dev(dev_t dev);
2216
3cb03002 2217static void export_rdev(struct md_rdev * rdev)
1da177e4
LT		 2218{
2219 char b[BDEVNAME_SIZE];
2220 printk(KERN_INFO "md: export_rdev(%s)\n",
2221 bdevname(rdev->bdev,b));
2222 if (rdev->mddev)
2223 MD_BUG();
545c8795 2224 md_rdev_clear(rdev);
1da177e4 2225#ifndef MODULE
d0fae18f
N		 2226 if (test_bit(AutoDetected, &rdev->flags))
2227 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2228#endif
2229 unlock_rdev(rdev);
86e6ffdd 2230 kobject_put(&rdev->kobj);
1da177e4
LT		 2231}
2232
3cb03002 2233static void kick_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2234{
2235 unbind_rdev_from_array(rdev);
2236 export_rdev(rdev);
2237}
2238
fd01b88c 2239static void export_array(struct mddev *mddev)
1da177e4 2240{
3cb03002 2241 struct md_rdev *rdev, *tmp;
1da177e4 2242
dafb20fa 2243 rdev_for_each_safe(rdev, tmp, mddev) {
1da177e4
LT		 2244 if (!rdev->mddev) {
2245 MD_BUG();
2246 continue;
2247 }
2248 kick_rdev_from_array(rdev);
2249 }
2250 if (!list_empty(&mddev->disks))
2251 MD_BUG();
2252 mddev->raid_disks = 0;
2253 mddev->major_version = 0;
2254}
2255
2256static void print_desc(mdp_disk_t *desc)
2257{
2258 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2259 desc->major,desc->minor,desc->raid_disk,desc->state);
2260}
2261
cd2ac932 2262static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 2263{
2264 int i;
2265
2266 printk(KERN_INFO
2267 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2268 sb->major_version, sb->minor_version, sb->patch_version,
2269 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2270 sb->ctime);
2271 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2272 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2273 sb->md_minor, sb->layout, sb->chunk_size);
2274 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2275 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2276 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2277 sb->failed_disks, sb->spare_disks,
2278 sb->sb_csum, (unsigned long)sb->events_lo);
2279
2280 printk(KERN_INFO);
2281 for (i = 0; i < MD_SB_DISKS; i++) {
2282 mdp_disk_t *desc;
2283
2284 desc = sb->disks + i;
2285 if (desc->number || desc->major || desc->minor ||
2286 desc->raid_disk || (desc->state && (desc->state != 4))) {
2287 printk(" D %2d: ", i);
2288 print_desc(desc);
2289 }
2290 }
2291 printk(KERN_INFO "md: THIS: ");
2292 print_desc(&sb->this_disk);
cd2ac932 2293}
1da177e4 2294
cd2ac932
CR		 2295static void print_sb_1(struct mdp_superblock_1 *sb)
2296{
2297 __u8 *uuid;
2298
2299 uuid = sb->set_uuid;
ad361c98 2300 printk(KERN_INFO
7b75c2f8 2301 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2302 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 2303 le32_to_cpu(sb->major_version),
2304 le32_to_cpu(sb->feature_map),
7b75c2f8 2305 uuid,
cd2ac932
CR		 2306 sb->set_name,
2307 (unsigned long long)le64_to_cpu(sb->ctime)
2308 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2309
2310 uuid = sb->device_uuid;
ad361c98
JP		 2311 printk(KERN_INFO
2312 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2313 " RO:%llu\n"
7b75c2f8 2314 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP		 2315 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2316 "md: (MaxDev:%u) \n",
cd2ac932
CR		 2317 le32_to_cpu(sb->level),
2318 (unsigned long long)le64_to_cpu(sb->size),
2319 le32_to_cpu(sb->raid_disks),
2320 le32_to_cpu(sb->layout),
2321 le32_to_cpu(sb->chunksize),
2322 (unsigned long long)le64_to_cpu(sb->data_offset),
2323 (unsigned long long)le64_to_cpu(sb->data_size),
2324 (unsigned long long)le64_to_cpu(sb->super_offset),
2325 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2326 le32_to_cpu(sb->dev_number),
7b75c2f8 2327 uuid,
cd2ac932
CR		 2328 sb->devflags,
2329 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2330 (unsigned long long)le64_to_cpu(sb->events),
2331 (unsigned long long)le64_to_cpu(sb->resync_offset),
2332 le32_to_cpu(sb->sb_csum),
2333 le32_to_cpu(sb->max_dev)
2334 );
1da177e4
LT		 2335}
2336
3cb03002 2337static void print_rdev(struct md_rdev *rdev, int major_version)
1da177e4
LT		 2338{
2339 char b[BDEVNAME_SIZE];
dd8ac336
AN		 2340 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2341 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 2342 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2343 rdev->desc_nr);
1da177e4 2344 if (rdev->sb_loaded) {
cd2ac932
CR		 2345 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2346 switch (major_version) {
2347 case 0:
65a06f06 2348 print_sb_90(page_address(rdev->sb_page));
cd2ac932
CR		 2349 break;
2350 case 1:
65a06f06 2351 print_sb_1(page_address(rdev->sb_page));
cd2ac932
CR		 2352 break;
2353 }
1da177e4
LT		 2354 } else
2355 printk(KERN_INFO "md: no rdev superblock!\n");
2356}
2357
5e56341d 2358static void md_print_devices(void)
1da177e4 2359{
159ec1fc 2360 struct list_head *tmp;
3cb03002 2361 struct md_rdev *rdev;
fd01b88c 2362 struct mddev *mddev;
1da177e4
LT		 2363 char b[BDEVNAME_SIZE];
2364
2365 printk("\n");
2366 printk("md: **********************************\n");
2367 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2368 printk("md: **********************************\n");
29ac4aa3 2369 for_each_mddev(mddev, tmp) {
1da177e4 2370
32a7627c
N		 2371 if (mddev->bitmap)
2372 bitmap_print_sb(mddev->bitmap);
2373 else
2374 printk("%s: ", mdname(mddev));
dafb20fa 2375 rdev_for_each(rdev, mddev)
1da177e4
LT		 2376 printk("<%s>", bdevname(rdev->bdev,b));
2377 printk("\n");
2378
dafb20fa 2379 rdev_for_each(rdev, mddev)
cd2ac932 2380 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 2381 }
2382 printk("md: **********************************\n");
2383 printk("\n");
2384}
2385
2386
fd01b88c 2387static void sync_sbs(struct mddev * mddev, int nospares)
1da177e4 2388{
42543769
N		 2389 /* Update each superblock (in-memory image), but
2390 * if we are allowed to, skip spares which already
2391 * have the right event counter, or have one earlier
2392 * (which would mean they aren't being marked as dirty
2393 * with the rest of the array)
2394 */
3cb03002 2395 struct md_rdev *rdev;
dafb20fa 2396 rdev_for_each(rdev, mddev) {
42543769
N		 2397 if (rdev->sb_events == mddev->events ||
2398 (nospares &&
2399 rdev->raid_disk < 0 &&
42543769
N		 2400 rdev->sb_events+1 == mddev->events)) {
2401 /* Don't update this superblock */
2402 rdev->sb_loaded = 2;
2403 } else {
076f968b 2404 sync_super(mddev, rdev);
42543769
N		 2405 rdev->sb_loaded = 1;
2406 }
1da177e4
LT		 2407 }
2408}
2409
fd01b88c 2410static void md_update_sb(struct mddev * mddev, int force_change)
1da177e4 2411{
3cb03002 2412 struct md_rdev *rdev;
06d91a5f 2413 int sync_req;
42543769 2414 int nospares = 0;
2699b672 2415 int any_badblocks_changed = 0;
1da177e4 2416
d87f064f
N		 2417 if (mddev->ro) {
2418 if (force_change)
2419 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2420 return;
2421 }
1da177e4 2422repeat:
3a3a5ddb 2423 /* First make sure individual recovery_offsets are correct */
dafb20fa 2424 rdev_for_each(rdev, mddev) {
3a3a5ddb
N		 2425 if (rdev->raid_disk >= 0 &&
2426 mddev->delta_disks >= 0 &&
2427 !test_bit(In_sync, &rdev->flags) &&
2428 mddev->curr_resync_completed > rdev->recovery_offset)
2429 rdev->recovery_offset = mddev->curr_resync_completed;
2430
2431 }
bd52b746 2432 if (!mddev->persistent) {
070dc6dd 2433 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2434 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2435 if (!mddev->external) {
d97a41dc 2436 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
dafb20fa 2437 rdev_for_each(rdev, mddev) {
de393cde 2438 if (rdev->badblocks.changed) {
d0962936 2439 rdev->badblocks.changed = 0;
de393cde
N		 2440 md_ack_all_badblocks(&rdev->badblocks);
2441 md_error(mddev, rdev);
2442 }
2443 clear_bit(Blocked, &rdev->flags);
2444 clear_bit(BlockedBadBlocks, &rdev->flags);
2445 wake_up(&rdev->blocked_wait);
2446 }
2447 }
3a3a5ddb
N		 2448 wake_up(&mddev->sb_wait);
2449 return;
2450 }
2451
a9701a30 2452 spin_lock_irq(&mddev->write_lock);
84692195 2453
3a3a5ddb
N		 2454 mddev->utime = get_seconds();
2455
850b2b42
N		 2456 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2457 force_change = 1;
2458 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2459 /* just a clean<-> dirty transition, possibly leave spares alone,
2460 * though if events isn't the right even/odd, we will have to do
2461 * spares after all
2462 */
2463 nospares = 1;
2464 if (force_change)
2465 nospares = 0;
2466 if (mddev->degraded)
84692195
N		 2467 /* If the array is degraded, then skipping spares is both
2468 * dangerous and fairly pointless.
2469 * Dangerous because a device that was removed from the array
2470 * might have a event_count that still looks up-to-date,
2471 * so it can be re-added without a resync.
2472 * Pointless because if there are any spares to skip,
2473 * then a recovery will happen and soon that array won't
2474 * be degraded any more and the spare can go back to sleep then.
2475 */
850b2b42 2476 nospares = 0;
84692195 2477
06d91a5f 2478 sync_req = mddev->in_sync;
42543769
N		 2479
2480 /* If this is just a dirty<->clean transition, and the array is clean
2481 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2482 if (nospares
42543769 2483 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2484 && mddev->can_decrease_events
2485 && mddev->events != 1) {
42543769 2486 mddev->events--;
a8707c08
N		 2487 mddev->can_decrease_events = 0;
2488 } else {
42543769
N		 2489 /* otherwise we have to go forward and ... */
2490 mddev->events ++;
a8707c08 2491 mddev->can_decrease_events = nospares;
42543769 2492 }
1da177e4
LT		 2493
2494 if (!mddev->events) {
2495 /*
2496 * oops, this 64-bit counter should never wrap.
2497 * Either we are in around ~1 trillion A.C., assuming
2498 * 1 reboot per second, or we have a bug:
2499 */
2500 MD_BUG();
2501 mddev->events --;
2502 }
2699b672 2503
dafb20fa 2504 rdev_for_each(rdev, mddev) {
2699b672
N		 2505 if (rdev->badblocks.changed)
2506 any_badblocks_changed++;
de393cde
N		 2507 if (test_bit(Faulty, &rdev->flags))
2508 set_bit(FaultRecorded, &rdev->flags);
2509 }
2699b672 2510
e691063a 2511 sync_sbs(mddev, nospares);
a9701a30 2512 spin_unlock_irq(&mddev->write_lock);
1da177e4 2513
36a4e1fe
N		 2514 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2515 mdname(mddev), mddev->in_sync);
1da177e4 2516
4ad13663 2517 bitmap_update_sb(mddev->bitmap);
dafb20fa 2518 rdev_for_each(rdev, mddev) {
1da177e4 2519 char b[BDEVNAME_SIZE];
36a4e1fe 2520
42543769
N		 2521 if (rdev->sb_loaded != 1)
2522 continue; /* no noise on spare devices */
1da177e4 2523
d70ed2e4
AW		 2524 if (!test_bit(Faulty, &rdev->flags) &&
2525 rdev->saved_raid_disk == -1) {
7bfa19f2 2526 md_super_write(mddev,rdev,
0f420358 2527 rdev->sb_start, rdev->sb_size,
7bfa19f2 2528 rdev->sb_page);
36a4e1fe
N		 2529 pr_debug("md: (write) %s's sb offset: %llu\n",
2530 bdevname(rdev->bdev, b),
2531 (unsigned long long)rdev->sb_start);
42543769 2532 rdev->sb_events = mddev->events;
2699b672
N		 2533 if (rdev->badblocks.size) {
2534 md_super_write(mddev, rdev,
2535 rdev->badblocks.sector,
2536 rdev->badblocks.size << 9,
2537 rdev->bb_page);
2538 rdev->badblocks.size = 0;
2539 }
7bfa19f2 2540
d70ed2e4 2541 } else if (test_bit(Faulty, &rdev->flags))
36a4e1fe
N		 2542 pr_debug("md: %s (skipping faulty)\n",
2543 bdevname(rdev->bdev, b));
d70ed2e4
AW		 2544 else
2545 pr_debug("(skipping incremental s/r ");
2546
7bfa19f2 2547 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2548 /* only need to write one superblock... */
2549 break;
2550 }
a9701a30 2551 md_super_wait(mddev);
850b2b42 2552 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2553
a9701a30 2554 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2555 if (mddev->in_sync != sync_req ||
2556 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2557 /* have to write it out again */
a9701a30 2558 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2559 goto repeat;
2560 }
850b2b42 2561 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2562 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2563 wake_up(&mddev->sb_wait);
acb180b0
N		 2564 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2565 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2566
dafb20fa 2567 rdev_for_each(rdev, mddev) {
de393cde
N		 2568 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2569 clear_bit(Blocked, &rdev->flags);
2570
2571 if (any_badblocks_changed)
2699b672 2572 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2573 clear_bit(BlockedBadBlocks, &rdev->flags);
2574 wake_up(&rdev->blocked_wait);
2575 }
1da177e4
LT		 2576}
2577
7f6ce769 2578/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2579 * We want to accept with case. For this we use cmd_match.
2580 */
2581static int cmd_match(const char *cmd, const char *str)
2582{
2583 /* See if cmd, written into a sysfs file, matches
2584 * str. They must either be the same, or cmd can
2585 * have a trailing newline
2586 */
2587 while (*cmd && *str && *cmd == *str) {
2588 cmd++;
2589 str++;
2590 }
2591 if (*cmd == '\n')
2592 cmd++;
2593 if (*str || *cmd)
2594 return 0;
2595 return 1;
2596}
2597
86e6ffdd
N		 2598struct rdev_sysfs_entry {
2599 struct attribute attr;
3cb03002
N		 2600 ssize_t (*show)(struct md_rdev *, char *);
2601 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2602};
2603
2604static ssize_t
3cb03002 2605state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2606{
2607 char *sep = "";
20a49ff6 2608 size_t len = 0;
86e6ffdd 2609
de393cde
N		 2610 if (test_bit(Faulty, &rdev->flags) ||
2611 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2612 len+= sprintf(page+len, "%sfaulty",sep);
2613 sep = ",";
2614 }
b2d444d7 2615 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2616 len += sprintf(page+len, "%sin_sync",sep);
2617 sep = ",";
2618 }
f655675b
N		 2619 if (test_bit(WriteMostly, &rdev->flags)) {
2620 len += sprintf(page+len, "%swrite_mostly",sep);
2621 sep = ",";
2622 }
de393cde 2623 if (test_bit(Blocked, &rdev->flags) ||
52c64152
N		 2624 (rdev->badblocks.unacked_exist
2625 && !test_bit(Faulty, &rdev->flags))) {
6bfe0b49
DW		 2626 len += sprintf(page+len, "%sblocked", sep);
2627 sep = ",";
2628 }
b2d444d7
N		 2629 if (!test_bit(Faulty, &rdev->flags) &&
2630 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2631 len += sprintf(page+len, "%sspare", sep);
2632 sep = ",";
2633 }
d7a9d443
N		 2634 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2635 len += sprintf(page+len, "%swrite_error", sep);
2636 sep = ",";
2637 }
2d78f8c4
N		 2638 if (test_bit(WantReplacement, &rdev->flags)) {
2639 len += sprintf(page+len, "%swant_replacement", sep);
2640 sep = ",";
2641 }
2642 if (test_bit(Replacement, &rdev->flags)) {
2643 len += sprintf(page+len, "%sreplacement", sep);
2644 sep = ",";
2645 }
2646
86e6ffdd
N		 2647 return len+sprintf(page+len, "\n");
2648}
2649
45dc2de1 2650static ssize_t
3cb03002 2651state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2652{
2653 /* can write
de393cde 2654 * faulty - simulates an error
45dc2de1 2655 * remove - disconnects the device
f655675b
N		 2656 * writemostly - sets write_mostly
2657 * -writemostly - clears write_mostly
de393cde
N		 2658 * blocked - sets the Blocked flags
2659 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2660 * insync - sets Insync providing device isn't active
d7a9d443
N		 2661 * write_error - sets WriteErrorSeen
2662 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2663 */
2664 int err = -EINVAL;
2665 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2666 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2667 if (test_bit(Faulty, &rdev->flags))
2668 err = 0;
2669 else
2670 err = -EBUSY;
45dc2de1
N		 2671 } else if (cmd_match(buf, "remove")) {
2672 if (rdev->raid_disk >= 0)
2673 err = -EBUSY;
2674 else {
fd01b88c 2675 struct mddev *mddev = rdev->mddev;
45dc2de1 2676 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2677 if (mddev->pers)
2678 md_update_sb(mddev, 1);
45dc2de1
N		 2679 md_new_event(mddev);
2680 err = 0;
2681 }
f655675b
N		 2682 } else if (cmd_match(buf, "writemostly")) {
2683 set_bit(WriteMostly, &rdev->flags);
2684 err = 0;
2685 } else if (cmd_match(buf, "-writemostly")) {
2686 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2687 err = 0;
2688 } else if (cmd_match(buf, "blocked")) {
2689 set_bit(Blocked, &rdev->flags);
2690 err = 0;
2691 } else if (cmd_match(buf, "-blocked")) {
de393cde 2692 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2693 rdev->badblocks.unacked_exist) {
de393cde
N		 2694 /* metadata handler doesn't understand badblocks,
2695 * so we need to fail the device
2696 */
2697 md_error(rdev->mddev, rdev);
2698 }
6bfe0b49 2699 clear_bit(Blocked, &rdev->flags);
de393cde 2700 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2701 wake_up(&rdev->blocked_wait);
2702 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2703 md_wakeup_thread(rdev->mddev->thread);
2704
6d56e278
N		 2705 err = 0;
2706 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2707 set_bit(In_sync, &rdev->flags);
f655675b 2708 err = 0;
d7a9d443
N		 2709 } else if (cmd_match(buf, "write_error")) {
2710 set_bit(WriteErrorSeen, &rdev->flags);
2711 err = 0;
2712 } else if (cmd_match(buf, "-write_error")) {
2713 clear_bit(WriteErrorSeen, &rdev->flags);
2714 err = 0;
2d78f8c4
N		 2715 } else if (cmd_match(buf, "want_replacement")) {
2716 /* Any non-spare device that is not a replacement can
2717 * become want_replacement at any time, but we then need to
2718 * check if recovery is needed.
2719 */
2720 if (rdev->raid_disk >= 0 &&
2721 !test_bit(Replacement, &rdev->flags))
2722 set_bit(WantReplacement, &rdev->flags);
2723 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2724 md_wakeup_thread(rdev->mddev->thread);
2725 err = 0;
2726 } else if (cmd_match(buf, "-want_replacement")) {
2727 /* Clearing 'want_replacement' is always allowed.
2728 * Once replacements starts it is too late though.
2729 */
2730 err = 0;
2731 clear_bit(WantReplacement, &rdev->flags);
2732 } else if (cmd_match(buf, "replacement")) {
2733 /* Can only set a device as a replacement when array has not
2734 * yet been started. Once running, replacement is automatic
2735 * from spares, or by assigning 'slot'.
2736 */
2737 if (rdev->mddev->pers)
2738 err = -EBUSY;
2739 else {
2740 set_bit(Replacement, &rdev->flags);
2741 err = 0;
2742 }
2743 } else if (cmd_match(buf, "-replacement")) {
2744 /* Similarly, can only clear Replacement before start */
2745 if (rdev->mddev->pers)
2746 err = -EBUSY;
2747 else {
2748 clear_bit(Replacement, &rdev->flags);
2749 err = 0;
2750 }
45dc2de1 2751 }
00bcb4ac
N		 2752 if (!err)
2753 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2754 return err ? err : len;
2755}
80ca3a44
N		 2756static struct rdev_sysfs_entry rdev_state =
2757__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2758
4dbcdc75 2759static ssize_t
3cb03002 2760errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2761{
2762 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2763}
2764
2765static ssize_t
3cb03002 2766errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N		 2767{
2768 char *e;
2769 unsigned long n = simple_strtoul(buf, &e, 10);
2770 if (*buf && (*e == 0 || *e == '\n')) {
2771 atomic_set(&rdev->corrected_errors, n);
2772 return len;
2773 }
2774 return -EINVAL;
2775}
2776static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2777__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2778
014236d2 2779static ssize_t
3cb03002 2780slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2781{
2782 if (rdev->raid_disk < 0)
2783 return sprintf(page, "none\n");
2784 else
2785 return sprintf(page, "%d\n", rdev->raid_disk);
2786}
2787
2788static ssize_t
3cb03002 2789slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N		 2790{
2791 char *e;
c303da6d 2792 int err;
014236d2
N		 2793 int slot = simple_strtoul(buf, &e, 10);
2794 if (strncmp(buf, "none", 4)==0)
2795 slot = -1;
2796 else if (e==buf || (*e && *e!= '\n'))
2797 return -EINVAL;
6c2fce2e 2798 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2799 /* Setting 'slot' on an active array requires also
2800 * updating the 'rd%d' link, and communicating
2801 * with the personality with ->hot_*_disk.
2802 * For now we only support removing
2803 * failed/spare devices. This normally happens automatically,
2804 * but not when the metadata is externally managed.
2805 */
c303da6d
N		 2806 if (rdev->raid_disk == -1)
2807 return -EEXIST;
2808 /* personality does all needed checks */
01393f3d 2809 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d 2810 return -EINVAL;
746d3207
N		 2811 clear_bit(Blocked, &rdev->flags);
2812 remove_and_add_spares(rdev->mddev, rdev);
2813 if (rdev->raid_disk >= 0)
2814 return -EBUSY;
c303da6d
N		 2815 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2816 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2817 } else if (rdev->mddev->pers) {
6c2fce2e 2818 /* Activating a spare .. or possibly reactivating
6d56e278 2819 * if we ever get bitmaps working here.
6c2fce2e
NB		 2820 */
2821
2822 if (rdev->raid_disk != -1)
2823 return -EBUSY;
2824
c6751b2b
N		 2825 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2826 return -EBUSY;
2827
6c2fce2e
NB		 2828 if (rdev->mddev->pers->hot_add_disk == NULL)
2829 return -EINVAL;
2830
ba1b41b6
N		 2831 if (slot >= rdev->mddev->raid_disks &&
2832 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2833 return -ENOSPC;
2834
6c2fce2e
NB		 2835 rdev->raid_disk = slot;
2836 if (test_bit(In_sync, &rdev->flags))
2837 rdev->saved_raid_disk = slot;
2838 else
2839 rdev->saved_raid_disk = -1;
d30519fc 2840 clear_bit(In_sync, &rdev->flags);
6c2fce2e
NB		 2841 err = rdev->mddev->pers->
2842 hot_add_disk(rdev->mddev, rdev);
199050ea 2843 if (err) {
6c2fce2e 2844 rdev->raid_disk = -1;
6c2fce2e 2845 return err;
52664732 2846 } else
00bcb4ac 2847 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2848 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2849 /* failure here is OK */;
6c2fce2e 2850 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2851 } else {
ba1b41b6
N		 2852 if (slot >= rdev->mddev->raid_disks &&
2853 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2854 return -ENOSPC;
2855 rdev->raid_disk = slot;
2856 /* assume it is working */
c5d79adb
N		 2857 clear_bit(Faulty, &rdev->flags);
2858 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2859 set_bit(In_sync, &rdev->flags);
00bcb4ac 2860 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2861 }
014236d2
N		 2862 return len;
2863}
2864
2865
2866static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2867__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2868
93c8cad0 2869static ssize_t
3cb03002 2870offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2871{
6961ece4 2872 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2873}
2874
2875static ssize_t
3cb03002 2876offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2877{
c6563a8c
N		 2878 unsigned long long offset;
2879 if (strict_strtoull(buf, 10, &offset) < 0)
93c8cad0 2880 return -EINVAL;
8ed0a521 2881 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2882 return -EBUSY;
dd8ac336 2883 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2884 /* Must set offset before size, so overlap checks
2885 * can be sane */
2886 return -EBUSY;
93c8cad0 2887 rdev->data_offset = offset;
25f7fd47 2888 rdev->new_data_offset = offset;
93c8cad0
N		 2889 return len;
2890}
2891
2892static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2893__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2894
c6563a8c
N		 2895static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2896{
2897 return sprintf(page, "%llu\n",
2898 (unsigned long long)rdev->new_data_offset);
2899}
2900
2901static ssize_t new_offset_store(struct md_rdev *rdev,
2902 const char *buf, size_t len)
2903{
2904 unsigned long long new_offset;
2905 struct mddev *mddev = rdev->mddev;
2906
2907 if (strict_strtoull(buf, 10, &new_offset) < 0)
2908 return -EINVAL;
2909
2910 if (mddev->sync_thread)
2911 return -EBUSY;
2912 if (new_offset == rdev->data_offset)
2913 /* reset is always permitted */
2914 ;
2915 else if (new_offset > rdev->data_offset) {
2916 /* must not push array size beyond rdev_sectors */
2917 if (new_offset - rdev->data_offset
2918 + mddev->dev_sectors > rdev->sectors)
2919 return -E2BIG;
2920 }
2921 /* Metadata worries about other space details. */
2922
2923 /* decreasing the offset is inconsistent with a backwards
2924 * reshape.
2925 */
2926 if (new_offset < rdev->data_offset &&
2927 mddev->reshape_backwards)
2928 return -EINVAL;
2929 /* Increasing offset is inconsistent with forwards
2930 * reshape. reshape_direction should be set to
2931 * 'backwards' first.
2932 */
2933 if (new_offset > rdev->data_offset &&
2934 !mddev->reshape_backwards)
2935 return -EINVAL;
2936
2937 if (mddev->pers && mddev->persistent &&
2938 !super_types[mddev->major_version]
2939 .allow_new_offset(rdev, new_offset))
2940 return -E2BIG;
2941 rdev->new_data_offset = new_offset;
2942 if (new_offset > rdev->data_offset)
2943 mddev->reshape_backwards = 1;
2944 else if (new_offset < rdev->data_offset)
2945 mddev->reshape_backwards = 0;
2946
2947 return len;
2948}
2949static struct rdev_sysfs_entry rdev_new_offset =
2950__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2951
83303b61 2952static ssize_t
3cb03002 2953rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2954{
dd8ac336 2955 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2956}
2957
c5d79adb
N		 2958static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2959{
2960 /* check if two start/length pairs overlap */
2961 if (s1+l1 <= s2)
2962 return 0;
2963 if (s2+l2 <= s1)
2964 return 0;
2965 return 1;
2966}
2967
b522adcd
DW		 2968static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2969{
2970 unsigned long long blocks;
2971 sector_t new;
2972
2973 if (strict_strtoull(buf, 10, &blocks) < 0)
2974 return -EINVAL;
2975
2976 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2977 return -EINVAL; /* sector conversion overflow */
2978
2979 new = blocks * 2;
2980 if (new != blocks * 2)
2981 return -EINVAL; /* unsigned long long to sector_t overflow */
2982
2983 *sectors = new;
2984 return 0;
2985}
2986
83303b61 2987static ssize_t
3cb03002 2988rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2989{
fd01b88c 2990 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2991 sector_t oldsectors = rdev->sectors;
b522adcd 2992 sector_t sectors;
27c529bb 2993
b522adcd 2994 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2995 return -EINVAL;
c6563a8c
N		 2996 if (rdev->data_offset != rdev->new_data_offset)
2997 return -EINVAL; /* too confusing */
0cd17fec 2998 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2999 if (my_mddev->persistent) {
dd8ac336
AN		 3000 sectors = super_types[my_mddev->major_version].
3001 rdev_size_change(rdev, sectors);
3002 if (!sectors)
0cd17fec 3003 return -EBUSY;
dd8ac336 3004 } else if (!sectors)
77304d2a 3005 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 3006 rdev->data_offset;
a6468539
N		 3007 if (!my_mddev->pers->resize)
3008 /* Cannot change size for RAID0 or Linear etc */
3009 return -EINVAL;
0cd17fec 3010 }
dd8ac336 3011 if (sectors < my_mddev->dev_sectors)
7d3c6f87 3012 return -EINVAL; /* component must fit device */
0cd17fec 3013
dd8ac336
AN		 3014 rdev->sectors = sectors;
3015 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 3016 /* need to check that all other rdevs with the same ->bdev
3017 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 3018 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 3019 * we have to change it back, we will have the lock again.
3020 */
fd01b88c 3021 struct mddev *mddev;
c5d79adb 3022 int overlap = 0;
159ec1fc 3023 struct list_head *tmp;
c5d79adb 3024
27c529bb 3025 mddev_unlock(my_mddev);
29ac4aa3 3026 for_each_mddev(mddev, tmp) {
3cb03002 3027 struct md_rdev *rdev2;
c5d79adb
N		 3028
3029 mddev_lock(mddev);
dafb20fa 3030 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 3031 if (rdev->bdev == rdev2->bdev &&
3032 rdev != rdev2 &&
3033 overlaps(rdev->data_offset, rdev->sectors,
3034 rdev2->data_offset,
3035 rdev2->sectors)) {
c5d79adb
N		 3036 overlap = 1;
3037 break;
3038 }
3039 mddev_unlock(mddev);
3040 if (overlap) {
3041 mddev_put(mddev);
3042 break;
3043 }
3044 }
27c529bb 3045 mddev_lock(my_mddev);
c5d79adb
N		 3046 if (overlap) {
3047 /* Someone else could have slipped in a size
3048 * change here, but doing so is just silly.
dd8ac336 3049 * We put oldsectors back because we *know* it is
c5d79adb
N		 3050 * safe, and trust userspace not to race with
3051 * itself
3052 */
dd8ac336 3053 rdev->sectors = oldsectors;
c5d79adb
N		 3054 return -EBUSY;
3055 }
3056 }
83303b61
N		 3057 return len;
3058}
3059
3060static struct rdev_sysfs_entry rdev_size =
80ca3a44 3061__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 3062
06e3c817 3063
3cb03002 3064static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 3065{
3066 unsigned long long recovery_start = rdev->recovery_offset;
3067
3068 if (test_bit(In_sync, &rdev->flags) ||
3069 recovery_start == MaxSector)
3070 return sprintf(page, "none\n");
3071
3072 return sprintf(page, "%llu\n", recovery_start);
3073}
3074
3cb03002 3075static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 3076{
3077 unsigned long long recovery_start;
3078
3079 if (cmd_match(buf, "none"))
3080 recovery_start = MaxSector;
3081 else if (strict_strtoull(buf, 10, &recovery_start))
3082 return -EINVAL;
3083
3084 if (rdev->mddev->pers &&
3085 rdev->raid_disk >= 0)
3086 return -EBUSY;
3087
3088 rdev->recovery_offset = recovery_start;
3089 if (recovery_start == MaxSector)
3090 set_bit(In_sync, &rdev->flags);
3091 else
3092 clear_bit(In_sync, &rdev->flags);
3093 return len;
3094}
3095
3096static struct rdev_sysfs_entry rdev_recovery_start =
3097__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3098
16c791a5
N		 3099
3100static ssize_t
3101badblocks_show(struct badblocks *bb, char *page, int unack);
3102static ssize_t
3103badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3104
3cb03002 3105static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3106{
3107 return badblocks_show(&rdev->badblocks, page, 0);
3108}
3cb03002 3109static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 3110{
de393cde
N		 3111 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3112 /* Maybe that ack was all we needed */
3113 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3114 wake_up(&rdev->blocked_wait);
3115 return rv;
16c791a5
N		 3116}
3117static struct rdev_sysfs_entry rdev_bad_blocks =
3118__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3119
3120
3cb03002 3121static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3122{
3123 return badblocks_show(&rdev->badblocks, page, 1);
3124}
3cb03002 3125static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 3126{
3127 return badblocks_store(&rdev->badblocks, page, len, 1);
3128}
3129static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3130__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3131
86e6ffdd
N		 3132static struct attribute *rdev_default_attrs[] = {
3133 &rdev_state.attr,
4dbcdc75 3134 &rdev_errors.attr,
014236d2 3135 &rdev_slot.attr,
93c8cad0 3136 &rdev_offset.attr,
c6563a8c 3137 &rdev_new_offset.attr,
83303b61 3138 &rdev_size.attr,
06e3c817 3139 &rdev_recovery_start.attr,
16c791a5
N		 3140 &rdev_bad_blocks.attr,
3141 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 3142 NULL,
3143};
3144static ssize_t
3145rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3146{
3147 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3148 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
fd01b88c 3149 struct mddev *mddev = rdev->mddev;
27c529bb 3150 ssize_t rv;
86e6ffdd
N		 3151
3152 if (!entry->show)
3153 return -EIO;
27c529bb
N		 3154
3155 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3156 if (!rv) {
3157 if (rdev->mddev == NULL)
3158 rv = -EBUSY;
3159 else
3160 rv = entry->show(rdev, page);
3161 mddev_unlock(mddev);
3162 }
3163 return rv;
86e6ffdd
N		 3164}
3165
3166static ssize_t
3167rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3168 const char *page, size_t length)
3169{
3170 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3171 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3172 ssize_t rv;
fd01b88c 3173 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3174
3175 if (!entry->store)
3176 return -EIO;
67463acb
N		 3177 if (!capable(CAP_SYS_ADMIN))
3178 return -EACCES;
27c529bb 3179 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3180 if (!rv) {
27c529bb
N		 3181 if (rdev->mddev == NULL)
3182 rv = -EBUSY;
3183 else
3184 rv = entry->store(rdev, page, length);
6a51830e 3185 mddev_unlock(mddev);
ca388059
N		 3186 }
3187 return rv;
86e6ffdd
N		 3188}
3189
3190static void rdev_free(struct kobject *ko)
3191{
3cb03002 3192 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3193 kfree(rdev);
3194}
52cf25d0 3195static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3196 .show = rdev_attr_show,
3197 .store = rdev_attr_store,
3198};
3199static struct kobj_type rdev_ktype = {
3200 .release = rdev_free,
3201 .sysfs_ops = &rdev_sysfs_ops,
3202 .default_attrs = rdev_default_attrs,
3203};
3204
3cb03002 3205int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3206{
3207 rdev->desc_nr = -1;
3208 rdev->saved_raid_disk = -1;
3209 rdev->raid_disk = -1;
3210 rdev->flags = 0;
3211 rdev->data_offset = 0;
c6563a8c 3212 rdev->new_data_offset = 0;
e8bb9a83
N		 3213 rdev->sb_events = 0;
3214 rdev->last_read_error.tv_sec = 0;
3215 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3216 rdev->sb_loaded = 0;
3217 rdev->bb_page = NULL;
e8bb9a83
N		 3218 atomic_set(&rdev->nr_pending, 0);
3219 atomic_set(&rdev->read_errors, 0);
3220 atomic_set(&rdev->corrected_errors, 0);
3221
3222 INIT_LIST_HEAD(&rdev->same_set);
3223 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3224
3225 /* Add space to store bad block list.
3226 * This reserves the space even on arrays where it cannot
3227 * be used - I wonder if that matters
3228 */
3229 rdev->badblocks.count = 0;
3230 rdev->badblocks.shift = 0;
3231 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3232 seqlock_init(&rdev->badblocks.lock);
3233 if (rdev->badblocks.page == NULL)
3234 return -ENOMEM;
3235
3236 return 0;
e8bb9a83
N		 3237}
3238EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3239/*
3240 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3241 *
3242 * mark the device faulty if:
3243 *
3244 * - the device is nonexistent (zero size)
3245 * - the device has no valid superblock
3246 *
3247 * a faulty rdev _never_ has rdev->sb set.
3248 */
3cb03002 3249static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3250{
3251 char b[BDEVNAME_SIZE];
3252 int err;
3cb03002 3253 struct md_rdev *rdev;
1da177e4
LT		 3254 sector_t size;
3255
9ffae0cf 3256 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3257 if (!rdev) {
3258 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3259 return ERR_PTR(-ENOMEM);
3260 }
1da177e4 3261
2230dfe4
N		 3262 err = md_rdev_init(rdev);
3263 if (err)
3264 goto abort_free;
3265 err = alloc_disk_sb(rdev);
3266 if (err)
1da177e4
LT		 3267 goto abort_free;
3268
c5d79adb 3269 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3270 if (err)
3271 goto abort_free;
3272
f9cb074b 3273 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3274
77304d2a 3275 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT		 3276 if (!size) {
3277 printk(KERN_WARNING
3278 "md: %s has zero or unknown size, marking faulty!\n",
3279 bdevname(rdev->bdev,b));
3280 err = -EINVAL;
3281 goto abort_free;
3282 }
3283
3284 if (super_format >= 0) {
3285 err = super_types[super_format].
3286 load_super(rdev, NULL, super_minor);
3287 if (err == -EINVAL) {
df968c4e
N		 3288 printk(KERN_WARNING
3289 "md: %s does not have a valid v%d.%d "
3290 "superblock, not importing!\n",
3291 bdevname(rdev->bdev,b),
3292 super_format, super_minor);
1da177e4
LT		 3293 goto abort_free;
3294 }
3295 if (err < 0) {
3296 printk(KERN_WARNING
3297 "md: could not read %s's sb, not importing!\n",
3298 bdevname(rdev->bdev,b));
3299 goto abort_free;
3300 }
3301 }
9f2f3830
N		 3302 if (super_format == -1)
3303 /* hot-add for 0.90, or non-persistent: so no badblocks */
3304 rdev->badblocks.shift = -1;
6bfe0b49 3305
1da177e4
LT		 3306 return rdev;
3307
3308abort_free:
2699b672
N		 3309 if (rdev->bdev)
3310 unlock_rdev(rdev);
545c8795 3311 md_rdev_clear(rdev);
1da177e4
LT		 3312 kfree(rdev);
3313 return ERR_PTR(err);
3314}
3315
3316/*
3317 * Check a full RAID array for plausibility
3318 */
3319
3320
fd01b88c 3321static void analyze_sbs(struct mddev * mddev)
1da177e4
LT		 3322{
3323 int i;
3cb03002 3324 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3325 char b[BDEVNAME_SIZE];
3326
3327 freshest = NULL;
dafb20fa 3328 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3329 switch (super_types[mddev->major_version].
3330 load_super(rdev, freshest, mddev->minor_version)) {
3331 case 1:
3332 freshest = rdev;
3333 break;
3334 case 0:
3335 break;
3336 default:
3337 printk( KERN_ERR \
3338 "md: fatal superblock inconsistency in %s"
3339 " -- removing from array\n",
3340 bdevname(rdev->bdev,b));
3341 kick_rdev_from_array(rdev);
3342 }
3343
3344
3345 super_types[mddev->major_version].
3346 validate_super(mddev, freshest);
3347
3348 i = 0;
dafb20fa 3349 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3350 if (mddev->max_disks &&
3351 (rdev->desc_nr >= mddev->max_disks ||
3352 i > mddev->max_disks)) {
de01dfad
N		 3353 printk(KERN_WARNING
3354 "md: %s: %s: only %d devices permitted\n",
3355 mdname(mddev), bdevname(rdev->bdev, b),
3356 mddev->max_disks);
3357 kick_rdev_from_array(rdev);
3358 continue;
3359 }
1da177e4
LT		 3360 if (rdev != freshest)
3361 if (super_types[mddev->major_version].
3362 validate_super(mddev, rdev)) {
3363 printk(KERN_WARNING "md: kicking non-fresh %s"
3364 " from array!\n",
3365 bdevname(rdev->bdev,b));
3366 kick_rdev_from_array(rdev);
3367 continue;
3368 }
3369 if (mddev->level == LEVEL_MULTIPATH) {
3370 rdev->desc_nr = i++;
3371 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3372 set_bit(In_sync, &rdev->flags);
5e5e3e78 3373 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3374 rdev->raid_disk = -1;
3375 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3376 }
3377 }
1da177e4
LT		 3378}
3379
72e02075
N		 3380/* Read a fixed-point number.
3381 * Numbers in sysfs attributes should be in "standard" units where
3382 * possible, so time should be in seconds.
3383 * However we internally use a a much smaller unit such as
3384 * milliseconds or jiffies.
3385 * This function takes a decimal number with a possible fractional
3386 * component, and produces an integer which is the result of
3387 * multiplying that number by 10^'scale'.
3388 * all without any floating-point arithmetic.
3389 */
3390int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3391{
3392 unsigned long result = 0;
3393 long decimals = -1;
3394 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3395 if (*cp == '.')
3396 decimals = 0;
3397 else if (decimals < scale) {
3398 unsigned int value;
3399 value = *cp - '0';
3400 result = result * 10 + value;
3401 if (decimals >= 0)
3402 decimals++;
3403 }
3404 cp++;
3405 }
3406 if (*cp == '\n')
3407 cp++;
3408 if (*cp)
3409 return -EINVAL;
3410 if (decimals < 0)
3411 decimals = 0;
3412 while (decimals < scale) {
3413 result *= 10;
3414 decimals ++;
3415 }
3416 *res = result;
3417 return 0;
3418}
3419
3420
19052c0e
N		 3421static void md_safemode_timeout(unsigned long data);
3422
16f17b39 3423static ssize_t
fd01b88c 3424safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3425{
3426 int msec = (mddev->safemode_delay*1000)/HZ;
3427 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3428}
3429static ssize_t
fd01b88c 3430safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3431{
16f17b39 3432 unsigned long msec;
97ce0a7f 3433
72e02075 3434 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3435 return -EINVAL;
16f17b39
N		 3436 if (msec == 0)
3437 mddev->safemode_delay = 0;
3438 else {
19052c0e 3439 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 3440 mddev->safemode_delay = (msec*HZ)/1000;
3441 if (mddev->safemode_delay == 0)
3442 mddev->safemode_delay = 1;
19052c0e
N		 3443 if (mddev->safemode_delay < old_delay)
3444 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 3445 }
3446 return len;
3447}
3448static struct md_sysfs_entry md_safe_delay =
80ca3a44 3449__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3450
eae1701f 3451static ssize_t
fd01b88c 3452level_show(struct mddev *mddev, char *page)
eae1701f 3453{
84fc4b56 3454 struct md_personality *p = mddev->pers;
d9d166c2 3455 if (p)
eae1701f 3456 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 3457 else if (mddev->clevel[0])
3458 return sprintf(page, "%s\n", mddev->clevel);
3459 else if (mddev->level != LEVEL_NONE)
3460 return sprintf(page, "%d\n", mddev->level);
3461 else
3462 return 0;
eae1701f
N		 3463}
3464
d9d166c2 3465static ssize_t
fd01b88c 3466level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3467{
f2859af6 3468 char clevel[16];
20a49ff6 3469 ssize_t rv = len;
84fc4b56 3470 struct md_personality *pers;
f2859af6 3471 long level;
245f46c2 3472 void *priv;
3cb03002 3473 struct md_rdev *rdev;
245f46c2
N		 3474
3475 if (mddev->pers == NULL) {
3476 if (len == 0)
3477 return 0;
3478 if (len >= sizeof(mddev->clevel))
3479 return -ENOSPC;
3480 strncpy(mddev->clevel, buf, len);
3481 if (mddev->clevel[len-1] == '\n')
3482 len--;
3483 mddev->clevel[len] = 0;
3484 mddev->level = LEVEL_NONE;
3485 return rv;
3486 }
3487
3488 /* request to change the personality. Need to ensure:
3489 * - array is not engaged in resync/recovery/reshape
3490 * - old personality can be suspended
3491 * - new personality will access other array.
3492 */
3493
bb4f1e9d
N		 3494 if (mddev->sync_thread ||
3495 mddev->reshape_position != MaxSector ||
3496 mddev->sysfs_active)
d9d166c2 3497 return -EBUSY;
245f46c2
N		 3498
3499 if (!mddev->pers->quiesce) {
3500 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3501 mdname(mddev), mddev->pers->name);
3502 return -EINVAL;
3503 }
3504
3505 /* Now find the new personality */
f2859af6 3506 if (len == 0 || len >= sizeof(clevel))
245f46c2 3507 return -EINVAL;
f2859af6
DW		 3508 strncpy(clevel, buf, len);
3509 if (clevel[len-1] == '\n')
d9d166c2 3510 len--;
f2859af6
DW		 3511 clevel[len] = 0;
3512 if (strict_strtol(clevel, 10, &level))
3513 level = LEVEL_NONE;
245f46c2 3514
f2859af6
DW		 3515 if (request_module("md-%s", clevel) != 0)
3516 request_module("md-level-%s", clevel);
245f46c2 3517 spin_lock(&pers_lock);
f2859af6 3518 pers = find_pers(level, clevel);
245f46c2
N		 3519 if (!pers || !try_module_get(pers->owner)) {
3520 spin_unlock(&pers_lock);
f2859af6 3521 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3522 return -EINVAL;
3523 }
3524 spin_unlock(&pers_lock);
3525
3526 if (pers == mddev->pers) {
3527 /* Nothing to do! */
3528 module_put(pers->owner);
3529 return rv;
3530 }
3531 if (!pers->takeover) {
3532 module_put(pers->owner);
3533 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3534 mdname(mddev), clevel);
245f46c2
N		 3535 return -EINVAL;
3536 }
3537
dafb20fa 3538 rdev_for_each(rdev, mddev)
e93f68a1
N		 3539 rdev->new_raid_disk = rdev->raid_disk;
3540
245f46c2
N		 3541 /* ->takeover must set new_* and/or delta_disks
3542 * if it succeeds, and may set them when it fails.
3543 */
3544 priv = pers->takeover(mddev);
3545 if (IS_ERR(priv)) {
3546 mddev->new_level = mddev->level;
3547 mddev->new_layout = mddev->layout;
664e7c41 3548 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3549 mddev->raid_disks -= mddev->delta_disks;
3550 mddev->delta_disks = 0;
2c810cdd 3551 mddev->reshape_backwards = 0;
245f46c2
N		 3552 module_put(pers->owner);
3553 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3554 mdname(mddev), clevel);
245f46c2
N		 3555 return PTR_ERR(priv);
3556 }
3557
3558 /* Looks like we have a winner */
3559 mddev_suspend(mddev);
3560 mddev->pers->stop(mddev);
a64c876f
N		 3561
3562 if (mddev->pers->sync_request == NULL &&
3563 pers->sync_request != NULL) {
3564 /* need to add the md_redundancy_group */
3565 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3566 printk(KERN_WARNING
3567 "md: cannot register extra attributes for %s\n",
3568 mdname(mddev));
19fdb9ee 3569 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3570 }
3571 if (mddev->pers->sync_request != NULL &&
3572 pers->sync_request == NULL) {
3573 /* need to remove the md_redundancy_group */
3574 if (mddev->to_remove == NULL)
3575 mddev->to_remove = &md_redundancy_group;
3576 }
3577
54071b38
TM		 3578 if (mddev->pers->sync_request == NULL &&
3579 mddev->external) {
3580 /* We are converting from a no-redundancy array
3581 * to a redundancy array and metadata is managed
3582 * externally so we need to be sure that writes
3583 * won't block due to a need to transition
3584 * clean->dirty
3585 * until external management is started.
3586 */
3587 mddev->in_sync = 0;
3588 mddev->safemode_delay = 0;
3589 mddev->safemode = 0;
3590 }
3591
dafb20fa 3592 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3593 if (rdev->raid_disk < 0)
3594 continue;
bf2cb0da 3595 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3596 rdev->new_raid_disk = -1;
3597 if (rdev->new_raid_disk == rdev->raid_disk)
3598 continue;
36fad858 3599 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3600 }
dafb20fa 3601 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3602 if (rdev->raid_disk < 0)
3603 continue;
3604 if (rdev->new_raid_disk == rdev->raid_disk)
3605 continue;
3606 rdev->raid_disk = rdev->new_raid_disk;
3607 if (rdev->raid_disk < 0)
3a981b03 3608 clear_bit(In_sync, &rdev->flags);
e93f68a1 3609 else {
36fad858
NK		 3610 if (sysfs_link_rdev(mddev, rdev))
3611 printk(KERN_WARNING "md: cannot register rd%d"
3612 " for %s after level change\n",
3613 rdev->raid_disk, mdname(mddev));
3a981b03 3614 }
e93f68a1
N		 3615 }
3616
3617 module_put(mddev->pers->owner);
245f46c2
N		 3618 mddev->pers = pers;
3619 mddev->private = priv;
3620 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3621 mddev->level = mddev->new_level;
3622 mddev->layout = mddev->new_layout;
664e7c41 3623 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3624 mddev->delta_disks = 0;
2c810cdd 3625 mddev->reshape_backwards = 0;
fee68723 3626 mddev->degraded = 0;
9af204cf
TM		 3627 if (mddev->pers->sync_request == NULL) {
3628 /* this is now an array without redundancy, so
3629 * it must always be in_sync
3630 */
3631 mddev->in_sync = 1;
3632 del_timer_sync(&mddev->safemode_timer);
3633 }
245f46c2 3634 pers->run(mddev);
245f46c2 3635 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3636 mddev_resume(mddev);
5cac7861 3637 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3638 md_new_event(mddev);
d9d166c2
N		 3639 return rv;
3640}
3641
3642static struct md_sysfs_entry md_level =
80ca3a44 3643__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3644
d4dbd025
N		 3645
3646static ssize_t
fd01b88c 3647layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3648{
3649 /* just a number, not meaningful for all levels */
08a02ecd
N		 3650 if (mddev->reshape_position != MaxSector &&
3651 mddev->layout != mddev->new_layout)
3652 return sprintf(page, "%d (%d)\n",
3653 mddev->new_layout, mddev->layout);
d4dbd025
N		 3654 return sprintf(page, "%d\n", mddev->layout);
3655}
3656
3657static ssize_t
fd01b88c 3658layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N		 3659{
3660 char *e;
3661 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3662
3663 if (!*buf || (*e && *e != '\n'))
3664 return -EINVAL;
3665
b3546035
N		 3666 if (mddev->pers) {
3667 int err;
50ac168a 3668 if (mddev->pers->check_reshape == NULL)
b3546035 3669 return -EBUSY;
597a711b 3670 mddev->new_layout = n;
50ac168a 3671 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3672 if (err) {
3673 mddev->new_layout = mddev->layout;
b3546035 3674 return err;
597a711b 3675 }
b3546035 3676 } else {
08a02ecd 3677 mddev->new_layout = n;
b3546035
N		 3678 if (mddev->reshape_position == MaxSector)
3679 mddev->layout = n;
3680 }
d4dbd025
N		 3681 return len;
3682}
3683static struct md_sysfs_entry md_layout =
80ca3a44 3684__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3685
3686
eae1701f 3687static ssize_t
fd01b88c 3688raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3689{
bb636547
N		 3690 if (mddev->raid_disks == 0)
3691 return 0;
08a02ecd
N		 3692 if (mddev->reshape_position != MaxSector &&
3693 mddev->delta_disks != 0)
3694 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3695 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3696 return sprintf(page, "%d\n", mddev->raid_disks);
3697}
3698
fd01b88c 3699static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3700
3701static ssize_t
fd01b88c 3702raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3703{
da943b99
N		 3704 char *e;
3705 int rv = 0;
3706 unsigned long n = simple_strtoul(buf, &e, 10);
3707
3708 if (!*buf || (*e && *e != '\n'))
3709 return -EINVAL;
3710
3711 if (mddev->pers)
3712 rv = update_raid_disks(mddev, n);
08a02ecd 3713 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3714 struct md_rdev *rdev;
08a02ecd 3715 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c
N		 3716
3717 rdev_for_each(rdev, mddev) {
3718 if (olddisks < n &&
3719 rdev->data_offset < rdev->new_data_offset)
3720 return -EINVAL;
3721 if (olddisks > n &&
3722 rdev->data_offset > rdev->new_data_offset)
3723 return -EINVAL;
3724 }
08a02ecd
N		 3725 mddev->delta_disks = n - olddisks;
3726 mddev->raid_disks = n;
2c810cdd 3727 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3728 } else
da943b99
N		 3729 mddev->raid_disks = n;
3730 return rv ? rv : len;
3731}
3732static struct md_sysfs_entry md_raid_disks =
80ca3a44 3733__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3734
3b34380a 3735static ssize_t
fd01b88c 3736chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3737{
08a02ecd 3738 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3739 mddev->chunk_sectors != mddev->new_chunk_sectors)
3740 return sprintf(page, "%d (%d)\n",
3741 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3742 mddev->chunk_sectors << 9);
3743 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3744}
3745
3746static ssize_t
fd01b88c 3747chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3748{
3b34380a
N		 3749 char *e;
3750 unsigned long n = simple_strtoul(buf, &e, 10);
3751
3b34380a
N		 3752 if (!*buf || (*e && *e != '\n'))
3753 return -EINVAL;
3754
b3546035
N		 3755 if (mddev->pers) {
3756 int err;
50ac168a 3757 if (mddev->pers->check_reshape == NULL)
b3546035 3758 return -EBUSY;
597a711b 3759 mddev->new_chunk_sectors = n >> 9;
50ac168a 3760 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3761 if (err) {
3762 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3763 return err;
597a711b 3764 }
b3546035 3765 } else {
664e7c41 3766 mddev->new_chunk_sectors = n >> 9;
b3546035 3767 if (mddev->reshape_position == MaxSector)
9d8f0363 3768 mddev->chunk_sectors = n >> 9;
b3546035 3769 }
3b34380a
N		 3770 return len;
3771}
3772static struct md_sysfs_entry md_chunk_size =
80ca3a44 3773__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3774
a94213b1 3775static ssize_t
fd01b88c 3776resync_start_show(struct mddev *mddev, char *page)
a94213b1 3777{
d1a7c503
N		 3778 if (mddev->recovery_cp == MaxSector)
3779 return sprintf(page, "none\n");
a94213b1
N		 3780 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3781}
3782
3783static ssize_t
fd01b88c 3784resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3785{
a94213b1
N		 3786 char *e;
3787 unsigned long long n = simple_strtoull(buf, &e, 10);
3788
b098636c 3789 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
a94213b1 3790 return -EBUSY;
06e3c817
DW		 3791 if (cmd_match(buf, "none"))
3792 n = MaxSector;
3793 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3794 return -EINVAL;
3795
3796 mddev->recovery_cp = n;
db07d85e
N		 3797 if (mddev->pers)
3798 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
a94213b1
N		 3799 return len;
3800}
3801static struct md_sysfs_entry md_resync_start =
80ca3a44 3802__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3803
9e653b63
N		 3804/*
3805 * The array state can be:
3806 *
3807 * clear
3808 * No devices, no size, no level
3809 * Equivalent to STOP_ARRAY ioctl
3810 * inactive
3811 * May have some settings, but array is not active
3812 * all IO results in error
3813 * When written, doesn't tear down array, but just stops it
3814 * suspended (not supported yet)
3815 * All IO requests will block. The array can be reconfigured.
910d8cb3 3816 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3817 * readonly
3818 * no resync can happen. no superblocks get written.
3819 * write requests fail
3820 * read-auto
3821 * like readonly, but behaves like 'clean' on a write request.
3822 *
3823 * clean - no pending writes, but otherwise active.
3824 * When written to inactive array, starts without resync
3825 * If a write request arrives then
3826 * if metadata is known, mark 'dirty' and switch to 'active'.
3827 * if not known, block and switch to write-pending
3828 * If written to an active array that has pending writes, then fails.
3829 * active
3830 * fully active: IO and resync can be happening.
3831 * When written to inactive array, starts with resync
3832 *
3833 * write-pending
3834 * clean, but writes are blocked waiting for 'active' to be written.
3835 *
3836 * active-idle
3837 * like active, but no writes have been seen for a while (100msec).
3838 *
3839 */
3840enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3841 write_pending, active_idle, bad_word};
05381954 3842static char *array_states[] = {
9e653b63
N		 3843 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3844 "write-pending", "active-idle", NULL };
3845
3846static int match_word(const char *word, char **list)
3847{
3848 int n;
3849 for (n=0; list[n]; n++)
3850 if (cmd_match(word, list[n]))
3851 break;
3852 return n;
3853}
3854
3855static ssize_t
fd01b88c 3856array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3857{
3858 enum array_state st = inactive;
3859
3860 if (mddev->pers)
3861 switch(mddev->ro) {
3862 case 1:
3863 st = readonly;
3864 break;
3865 case 2:
3866 st = read_auto;
3867 break;
3868 case 0:
3869 if (mddev->in_sync)
3870 st = clean;
070dc6dd 3871 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3872 st = write_pending;
9e653b63
N		 3873 else if (mddev->safemode)
3874 st = active_idle;
3875 else
3876 st = active;
3877 }
3878 else {
3879 if (list_empty(&mddev->disks) &&
3880 mddev->raid_disks == 0 &&
58c0fed4 3881 mddev->dev_sectors == 0)
9e653b63
N		 3882 st = clear;
3883 else
3884 st = inactive;
3885 }
3886 return sprintf(page, "%s\n", array_states[st]);
3887}
3888
a05b7ea0
N		 3889static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3890static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
fd01b88c
N		 3891static int do_md_run(struct mddev * mddev);
3892static int restart_array(struct mddev *mddev);
9e653b63
N		 3893
3894static ssize_t
fd01b88c 3895array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63
N		 3896{
3897 int err = -EINVAL;
3898 enum array_state st = match_word(buf, array_states);
3899 switch(st) {
3900 case bad_word:
3901 break;
3902 case clear:
3903 /* stopping an active array */
a05b7ea0 3904 err = do_md_stop(mddev, 0, NULL);
9e653b63
N		 3905 break;
3906 case inactive:
3907 /* stopping an active array */
90cf195d 3908 if (mddev->pers)
a05b7ea0 3909 err = do_md_stop(mddev, 2, NULL);
90cf195d 3910 else
e691063a 3911 err = 0; /* already inactive */
9e653b63
N		 3912 break;
3913 case suspended:
3914 break; /* not supported yet */
3915 case readonly:
3916 if (mddev->pers)
a05b7ea0 3917 err = md_set_readonly(mddev, NULL);
9e653b63
N		 3918 else {
3919 mddev->ro = 1;
648b629e 3920 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3921 err = do_md_run(mddev);
3922 }
3923 break;
3924 case read_auto:
9e653b63 3925 if (mddev->pers) {
80268ee9 3926 if (mddev->ro == 0)
a05b7ea0 3927 err = md_set_readonly(mddev, NULL);
80268ee9 3928 else if (mddev->ro == 1)
648b629e
N		 3929 err = restart_array(mddev);
3930 if (err == 0) {
3931 mddev->ro = 2;
3932 set_disk_ro(mddev->gendisk, 0);
3933 }
9e653b63
N		 3934 } else {
3935 mddev->ro = 2;
3936 err = do_md_run(mddev);
3937 }
3938 break;
3939 case clean:
3940 if (mddev->pers) {
3941 restart_array(mddev);
3942 spin_lock_irq(&mddev->write_lock);
3943 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3944 if (mddev->in_sync == 0) {
3945 mddev->in_sync = 1;
31a59e34
N		 3946 if (mddev->safemode == 1)
3947 mddev->safemode = 0;
070dc6dd 3948 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3949 }
3950 err = 0;
3951 } else
3952 err = -EBUSY;
9e653b63 3953 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3954 } else
3955 err = -EINVAL;
9e653b63
N		 3956 break;
3957 case active:
3958 if (mddev->pers) {
3959 restart_array(mddev);
070dc6dd 3960 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3961 wake_up(&mddev->sb_wait);
3962 err = 0;
3963 } else {
3964 mddev->ro = 0;
648b629e 3965 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3966 err = do_md_run(mddev);
3967 }
3968 break;
3969 case write_pending:
3970 case active_idle:
3971 /* these cannot be set */
3972 break;
3973 }
3974 if (err)
3975 return err;
0fd62b86 3976 else {
1d23f178
N		 3977 if (mddev->hold_active == UNTIL_IOCTL)
3978 mddev->hold_active = 0;
00bcb4ac 3979 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3980 return len;
0fd62b86 3981 }
9e653b63 3982}
80ca3a44
N		 3983static struct md_sysfs_entry md_array_state =
3984__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3985
1e50915f 3986static ssize_t
fd01b88c 3987max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3988 return sprintf(page, "%d\n",
3989 atomic_read(&mddev->max_corr_read_errors));
3990}
3991
3992static ssize_t
fd01b88c 3993max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB		 3994{
3995 char *e;
3996 unsigned long n = simple_strtoul(buf, &e, 10);
3997
3998 if (*buf && (*e == 0 || *e == '\n')) {
3999 atomic_set(&mddev->max_corr_read_errors, n);
4000 return len;
4001 }
4002 return -EINVAL;
4003}
4004
4005static struct md_sysfs_entry max_corr_read_errors =
4006__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4007 max_corrected_read_errors_store);
4008
6d7ff738 4009static ssize_t
fd01b88c 4010null_show(struct mddev *mddev, char *page)
6d7ff738
N		 4011{
4012 return -EINVAL;
4013}
4014
4015static ssize_t
fd01b88c 4016new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 4017{
4018 /* buf must be %d:%d\n? giving major and minor numbers */
4019 /* The new device is added to the array.
4020 * If the array has a persistent superblock, we read the
4021 * superblock to initialise info and check validity.
4022 * Otherwise, only checking done is that in bind_rdev_to_array,
4023 * which mainly checks size.
4024 */
4025 char *e;
4026 int major = simple_strtoul(buf, &e, 10);
4027 int minor;
4028 dev_t dev;
3cb03002 4029 struct md_rdev *rdev;
6d7ff738
N		 4030 int err;
4031
4032 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4033 return -EINVAL;
4034 minor = simple_strtoul(e+1, &e, 10);
4035 if (*e && *e != '\n')
4036 return -EINVAL;
4037 dev = MKDEV(major, minor);
4038 if (major != MAJOR(dev) ||
4039 minor != MINOR(dev))
4040 return -EOVERFLOW;
4041
4042
4043 if (mddev->persistent) {
4044 rdev = md_import_device(dev, mddev->major_version,
4045 mddev->minor_version);
4046 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 4047 struct md_rdev *rdev0
4048 = list_entry(mddev->disks.next,
4049 struct md_rdev, same_set);
6d7ff738
N		 4050 err = super_types[mddev->major_version]
4051 .load_super(rdev, rdev0, mddev->minor_version);
4052 if (err < 0)
4053 goto out;
4054 }
c5d79adb
N		 4055 } else if (mddev->external)
4056 rdev = md_import_device(dev, -2, -1);
4057 else
6d7ff738
N		 4058 rdev = md_import_device(dev, -1, -1);
4059
4060 if (IS_ERR(rdev))
4061 return PTR_ERR(rdev);
4062 err = bind_rdev_to_array(rdev, mddev);
4063 out:
4064 if (err)
4065 export_rdev(rdev);
4066 return err ? err : len;
4067}
4068
4069static struct md_sysfs_entry md_new_device =
80ca3a44 4070__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 4071
9b1d1dac 4072static ssize_t
fd01b88c 4073bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 4074{
4075 char *end;
4076 unsigned long chunk, end_chunk;
4077
4078 if (!mddev->bitmap)
4079 goto out;
4080 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4081 while (*buf) {
4082 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4083 if (buf == end) break;
4084 if (*end == '-') { /* range */
4085 buf = end + 1;
4086 end_chunk = simple_strtoul(buf, &end, 0);
4087 if (buf == end) break;
4088 }
4089 if (*end && !isspace(*end)) break;
4090 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 4091 buf = skip_spaces(end);
9b1d1dac
PC		 4092 }
4093 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4094out:
4095 return len;
4096}
4097
4098static struct md_sysfs_entry md_bitmap =
4099__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4100
a35b0d69 4101static ssize_t
fd01b88c 4102size_show(struct mddev *mddev, char *page)
a35b0d69 4103{
58c0fed4
AN		 4104 return sprintf(page, "%llu\n",
4105 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 4106}
4107
fd01b88c 4108static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 4109
4110static ssize_t
fd01b88c 4111size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 4112{
4113 /* If array is inactive, we can reduce the component size, but
4114 * not increase it (except from 0).
4115 * If array is active, we can try an on-line resize
4116 */
b522adcd
DW		 4117 sector_t sectors;
4118 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4119
58c0fed4
AN		 4120 if (err < 0)
4121 return err;
a35b0d69 4122 if (mddev->pers) {
58c0fed4 4123 err = update_size(mddev, sectors);
850b2b42 4124 md_update_sb(mddev, 1);
a35b0d69 4125 } else {
58c0fed4
AN		 4126 if (mddev->dev_sectors == 0 ||
4127 mddev->dev_sectors > sectors)
4128 mddev->dev_sectors = sectors;
a35b0d69
N		 4129 else
4130 err = -ENOSPC;
4131 }
4132 return err ? err : len;
4133}
4134
4135static struct md_sysfs_entry md_size =
80ca3a44 4136__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4137
8bb93aac 4138
83f0d77a 4139/* Metadata version.
e691063a
N		 4140 * This is one of
4141 * 'none' for arrays with no metadata (good luck...)
4142 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4143 * or N.M for internally known formats
4144 */
4145static ssize_t
fd01b88c 4146metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4147{
4148 if (mddev->persistent)
4149 return sprintf(page, "%d.%d\n",
4150 mddev->major_version, mddev->minor_version);
e691063a
N		 4151 else if (mddev->external)
4152 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4153 else
4154 return sprintf(page, "none\n");
4155}
4156
4157static ssize_t
fd01b88c 4158metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4159{
4160 int major, minor;
4161 char *e;
ea43ddd8
N		 4162 /* Changing the details of 'external' metadata is
4163 * always permitted. Otherwise there must be
4164 * no devices attached to the array.
4165 */
4166 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4167 ;
4168 else if (!list_empty(&mddev->disks))
8bb93aac
N		 4169 return -EBUSY;
4170
4171 if (cmd_match(buf, "none")) {
4172 mddev->persistent = 0;
e691063a
N		 4173 mddev->external = 0;
4174 mddev->major_version = 0;
4175 mddev->minor_version = 90;
4176 return len;
4177 }
4178 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4179 size_t namelen = len-9;
e691063a
N		 4180 if (namelen >= sizeof(mddev->metadata_type))
4181 namelen = sizeof(mddev->metadata_type)-1;
4182 strncpy(mddev->metadata_type, buf+9, namelen);
4183 mddev->metadata_type[namelen] = 0;
4184 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4185 mddev->metadata_type[--namelen] = 0;
4186 mddev->persistent = 0;
4187 mddev->external = 1;
8bb93aac
N		 4188 mddev->major_version = 0;
4189 mddev->minor_version = 90;
4190 return len;
4191 }
4192 major = simple_strtoul(buf, &e, 10);
4193 if (e==buf || *e != '.')
4194 return -EINVAL;
4195 buf = e+1;
4196 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4197 if (e==buf || (*e && *e != '\n') )
8bb93aac 4198 return -EINVAL;
50511da3 4199 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 4200 return -ENOENT;
4201 mddev->major_version = major;
4202 mddev->minor_version = minor;
4203 mddev->persistent = 1;
e691063a 4204 mddev->external = 0;
8bb93aac
N		 4205 return len;
4206}
4207
4208static struct md_sysfs_entry md_metadata =
80ca3a44 4209__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4210
24dd469d 4211static ssize_t
fd01b88c 4212action_show(struct mddev *mddev, char *page)
24dd469d 4213{
7eec314d 4214 char *type = "idle";
b6a9ce68
N		 4215 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4216 type = "frozen";
4217 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 4218 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 4219 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4220 type = "reshape";
4221 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 4222 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4223 type = "resync";
4224 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4225 type = "check";
4226 else
4227 type = "repair";
72a23c21 4228 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 4229 type = "recover";
4230 }
4231 return sprintf(page, "%s\n", type);
4232}
4233
4234static ssize_t
fd01b88c 4235action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4236{
7eec314d
N		 4237 if (!mddev->pers || !mddev->pers->sync_request)
4238 return -EINVAL;
4239
b6a9ce68
N		 4240 if (cmd_match(page, "frozen"))
4241 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4242 else
4243 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4244
4245 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 4246 if (mddev->sync_thread) {
4247 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 4248 md_reap_sync_thread(mddev);
7eec314d 4249 }
03c902e1
N		 4250 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4251 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4252 return -EBUSY;
72a23c21
NB		 4253 else if (cmd_match(page, "resync"))
4254 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4255 else if (cmd_match(page, "recover")) {
4256 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4257 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4258 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4259 int err;
4260 if (mddev->pers->start_reshape == NULL)
4261 return -EINVAL;
4262 err = mddev->pers->start_reshape(mddev);
4263 if (err)
4264 return err;
a99ac971 4265 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4266 } else {
bce74dac 4267 if (cmd_match(page, "check"))
7eec314d 4268 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4269 else if (!cmd_match(page, "repair"))
7eec314d
N		 4270 return -EINVAL;
4271 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4272 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4273 }
48c26ddc
N		 4274 if (mddev->ro == 2) {
4275 /* A write to sync_action is enough to justify
4276 * canceling read-auto mode
4277 */
4278 mddev->ro = 0;
4279 md_wakeup_thread(mddev->sync_thread);
4280 }
03c902e1 4281 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4282 md_wakeup_thread(mddev->thread);
00bcb4ac 4283 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4284 return len;
4285}
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
N		 4295static struct md_sysfs_entry md_scan_mode =
4296__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 4297
96de1e66 4298
80ca3a44 4299static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4300
88202a0c 4301static ssize_t
fd01b88c 4302sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4303{
4304 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4305 mddev->sync_speed_min ? "local": "system");
4306}
4307
4308static ssize_t
fd01b88c 4309sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4310{
4311 int min;
4312 char *e;
4313 if (strncmp(buf, "system", 6)==0) {
4314 mddev->sync_speed_min = 0;
4315 return len;
4316 }
4317 min = simple_strtoul(buf, &e, 10);
4318 if (buf == e || (*e && *e != '\n') || min <= 0)
4319 return -EINVAL;
4320 mddev->sync_speed_min = min;
4321 return len;
4322}
4323
4324static struct md_sysfs_entry md_sync_min =
4325__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4326
4327static ssize_t
fd01b88c 4328sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4329{
4330 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4331 mddev->sync_speed_max ? "local": "system");
4332}
4333
4334static ssize_t
fd01b88c 4335sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4336{
4337 int max;
4338 char *e;
4339 if (strncmp(buf, "system", 6)==0) {
4340 mddev->sync_speed_max = 0;
4341 return len;
4342 }
4343 max = simple_strtoul(buf, &e, 10);
4344 if (buf == e || (*e && *e != '\n') || max <= 0)
4345 return -EINVAL;
4346 mddev->sync_speed_max = max;
4347 return len;
4348}
4349
4350static struct md_sysfs_entry md_sync_max =
4351__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4352
d7f3d291 4353static ssize_t
fd01b88c 4354degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4355{
4356 return sprintf(page, "%d\n", mddev->degraded);
4357}
4358static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4359
90b08710 4360static ssize_t
fd01b88c 4361sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4362{
4363 return sprintf(page, "%d\n", mddev->parallel_resync);
4364}
4365
4366static ssize_t
fd01b88c 4367sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4368{
4369 long n;
4370
4371 if (strict_strtol(buf, 10, &n))
4372 return -EINVAL;
4373
4374 if (n != 0 && n != 1)
4375 return -EINVAL;
4376
4377 mddev->parallel_resync = n;
4378
4379 if (mddev->sync_thread)
4380 wake_up(&resync_wait);
4381
4382 return len;
4383}
4384
4385/* force parallel resync, even with shared block devices */
4386static struct md_sysfs_entry md_sync_force_parallel =
4387__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4388 sync_force_parallel_show, sync_force_parallel_store);
4389
88202a0c 4390static ssize_t
fd01b88c 4391sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4392{
4393 unsigned long resync, dt, db;
d1a7c503
N		 4394 if (mddev->curr_resync == 0)
4395 return sprintf(page, "none\n");
9687a60c
AN		 4396 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4397 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4398 if (!dt) dt++;
9687a60c
AN		 4399 db = resync - mddev->resync_mark_cnt;
4400 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4401}
4402
80ca3a44 4403static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4404
4405static ssize_t
fd01b88c 4406sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4407{
13ae864b 4408 unsigned long long max_sectors, resync;
88202a0c 4409
acb180b0
N		 4410 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4411 return sprintf(page, "none\n");
4412
72f36d59
N		 4413 if (mddev->curr_resync == 1 ||
4414 mddev->curr_resync == 2)
4415 return sprintf(page, "delayed\n");
4416
c804cdec
N		 4417 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4418 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 4419 max_sectors = mddev->resync_max_sectors;
88202a0c 4420 else
58c0fed4 4421 max_sectors = mddev->dev_sectors;
88202a0c 4422
acb180b0 4423 resync = mddev->curr_resync_completed;
13ae864b 4424 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4425}
4426
80ca3a44 4427static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 4428
5e96ee65 4429static ssize_t
fd01b88c 4430min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4431{
4432 return sprintf(page, "%llu\n",
4433 (unsigned long long)mddev->resync_min);
4434}
4435static ssize_t
fd01b88c 4436min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4437{
4438 unsigned long long min;
4439 if (strict_strtoull(buf, 10, &min))
4440 return -EINVAL;
4441 if (min > mddev->resync_max)
4442 return -EINVAL;
4443 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4444 return -EBUSY;
4445
4446 /* Must be a multiple of chunk_size */
9d8f0363 4447 if (mddev->chunk_sectors) {
2ac06c33 4448 sector_t temp = min;
9d8f0363 4449 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 4450 return -EINVAL;
4451 }
4452 mddev->resync_min = min;
4453
4454 return len;
4455}
4456
4457static struct md_sysfs_entry md_min_sync =
4458__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4459
c6207277 4460static ssize_t
fd01b88c 4461max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4462{
4463 if (mddev->resync_max == MaxSector)
4464 return sprintf(page, "max\n");
4465 else
4466 return sprintf(page, "%llu\n",
4467 (unsigned long long)mddev->resync_max);
4468}
4469static ssize_t
fd01b88c 4470max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277
N		 4471{
4472 if (strncmp(buf, "max", 3) == 0)
4473 mddev->resync_max = MaxSector;
4474 else {
5e96ee65
NB		 4475 unsigned long long max;
4476 if (strict_strtoull(buf, 10, &max))
4477 return -EINVAL;
4478 if (max < mddev->resync_min)
c6207277
N		 4479 return -EINVAL;
4480 if (max < mddev->resync_max &&
4d484a4a 4481 mddev->ro == 0 &&
c6207277
N		 4482 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4483 return -EBUSY;
4484
4485 /* Must be a multiple of chunk_size */
9d8f0363 4486 if (mddev->chunk_sectors) {
2ac06c33 4487 sector_t temp = max;
9d8f0363 4488 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 4489 return -EINVAL;
4490 }
4491 mddev->resync_max = max;
4492 }
4493 wake_up(&mddev->recovery_wait);
4494 return len;
4495}
4496
4497static struct md_sysfs_entry md_max_sync =
4498__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4499
e464eafd 4500static ssize_t
fd01b88c 4501suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4502{
4503 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4504}
4505
4506static ssize_t
fd01b88c 4507suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4508{
4509 char *e;
4510 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4511 unsigned long long old = mddev->suspend_lo;
e464eafd 4512
b8d966ef
N		 4513 if (mddev->pers == NULL ||
4514 mddev->pers->quiesce == NULL)
e464eafd
N		 4515 return -EINVAL;
4516 if (buf == e || (*e && *e != '\n'))
4517 return -EINVAL;
23ddff37
N		 4518
4519 mddev->suspend_lo = new;
4520 if (new >= old)
4521 /* Shrinking suspended region */
e464eafd 4522 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4523 else {
4524 /* Expanding suspended region - need to wait */
4525 mddev->pers->quiesce(mddev, 1);
4526 mddev->pers->quiesce(mddev, 0);
4527 }
4528 return len;
e464eafd
N		 4529}
4530static struct md_sysfs_entry md_suspend_lo =
4531__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4532
4533
4534static ssize_t
fd01b88c 4535suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4536{
4537 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4538}
4539
4540static ssize_t
fd01b88c 4541suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4542{
4543 char *e;
4544 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4545 unsigned long long old = mddev->suspend_hi;
e464eafd 4546
b8d966ef
N		 4547 if (mddev->pers == NULL ||
4548 mddev->pers->quiesce == NULL)
e464eafd
N		 4549 return -EINVAL;
4550 if (buf == e || (*e && *e != '\n'))
4551 return -EINVAL;
23ddff37
N		 4552
4553 mddev->suspend_hi = new;
4554 if (new <= old)
4555 /* Shrinking suspended region */
4556 mddev->pers->quiesce(mddev, 2);
4557 else {
4558 /* Expanding suspended region - need to wait */
e464eafd
N		 4559 mddev->pers->quiesce(mddev, 1);
4560 mddev->pers->quiesce(mddev, 0);
23ddff37
N		 4561 }
4562 return len;
e464eafd
N		 4563}
4564static struct md_sysfs_entry md_suspend_hi =
4565__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4566
08a02ecd 4567static ssize_t
fd01b88c 4568reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4569{
4570 if (mddev->reshape_position != MaxSector)
4571 return sprintf(page, "%llu\n",
4572 (unsigned long long)mddev->reshape_position);
4573 strcpy(page, "none\n");
4574 return 5;
4575}
4576
4577static ssize_t
fd01b88c 4578reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd 4579{
c6563a8c 4580 struct md_rdev *rdev;
08a02ecd
N		 4581 char *e;
4582 unsigned long long new = simple_strtoull(buf, &e, 10);
4583 if (mddev->pers)
4584 return -EBUSY;
4585 if (buf == e || (*e && *e != '\n'))
4586 return -EINVAL;
4587 mddev->reshape_position = new;
4588 mddev->delta_disks = 0;
2c810cdd 4589 mddev->reshape_backwards = 0;
08a02ecd
N		 4590 mddev->new_level = mddev->level;
4591 mddev->new_layout = mddev->layout;
664e7c41 4592 mddev->new_chunk_sectors = mddev->chunk_sectors;
c6563a8c
N		 4593 rdev_for_each(rdev, mddev)
4594 rdev->new_data_offset = rdev->data_offset;
08a02ecd
N		 4595 return len;
4596}
4597
4598static struct md_sysfs_entry md_reshape_position =
4599__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4600 reshape_position_store);
4601
2c810cdd
N		 4602static ssize_t
4603reshape_direction_show(struct mddev *mddev, char *page)
4604{
4605 return sprintf(page, "%s\n",
4606 mddev->reshape_backwards ? "backwards" : "forwards");
4607}
4608
4609static ssize_t
4610reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4611{
4612 int backwards = 0;
4613 if (cmd_match(buf, "forwards"))
4614 backwards = 0;
4615 else if (cmd_match(buf, "backwards"))
4616 backwards = 1;
4617 else
4618 return -EINVAL;
4619 if (mddev->reshape_backwards == backwards)
4620 return len;
4621
4622 /* check if we are allowed to change */
4623 if (mddev->delta_disks)
4624 return -EBUSY;
4625
4626 if (mddev->persistent &&
4627 mddev->major_version == 0)
4628 return -EINVAL;
4629
4630 mddev->reshape_backwards = backwards;
4631 return len;
4632}
4633
4634static struct md_sysfs_entry md_reshape_direction =
4635__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4636 reshape_direction_store);
4637
b522adcd 4638static ssize_t
fd01b88c 4639array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4640{
4641 if (mddev->external_size)
4642 return sprintf(page, "%llu\n",
4643 (unsigned long long)mddev->array_sectors/2);
4644 else
4645 return sprintf(page, "default\n");
4646}
4647
4648static ssize_t
fd01b88c 4649array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4650{
4651 sector_t sectors;
4652
4653 if (strncmp(buf, "default", 7) == 0) {
4654 if (mddev->pers)
4655 sectors = mddev->pers->size(mddev, 0, 0);
4656 else
4657 sectors = mddev->array_sectors;
4658
4659 mddev->external_size = 0;
4660 } else {
4661 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4662 return -EINVAL;
4663 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4664 return -E2BIG;
b522adcd
DW		 4665
4666 mddev->external_size = 1;
4667 }
4668
4669 mddev->array_sectors = sectors;
cbe6ef1d
N		 4670 if (mddev->pers) {
4671 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 4672 revalidate_disk(mddev->gendisk);
cbe6ef1d 4673 }
b522adcd
DW		 4674 return len;
4675}
4676
4677static struct md_sysfs_entry md_array_size =
4678__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4679 array_size_store);
e464eafd 4680
eae1701f
N		 4681static struct attribute *md_default_attrs[] = {
4682 &md_level.attr,
d4dbd025 4683 &md_layout.attr,
eae1701f 4684 &md_raid_disks.attr,
3b34380a 4685 &md_chunk_size.attr,
a35b0d69 4686 &md_size.attr,
a94213b1 4687 &md_resync_start.attr,
8bb93aac 4688 &md_metadata.attr,
6d7ff738 4689 &md_new_device.attr,
16f17b39 4690 &md_safe_delay.attr,
9e653b63 4691 &md_array_state.attr,
08a02ecd 4692 &md_reshape_position.attr,
2c810cdd 4693 &md_reshape_direction.attr,
b522adcd 4694 &md_array_size.attr,
1e50915f 4695 &max_corr_read_errors.attr,
411036fa
N		 4696 NULL,
4697};
4698
4699static struct attribute *md_redundancy_attrs[] = {
24dd469d 4700 &md_scan_mode.attr,
9d88883e 4701 &md_mismatches.attr,
88202a0c
N		 4702 &md_sync_min.attr,
4703 &md_sync_max.attr,
4704 &md_sync_speed.attr,
90b08710 4705 &md_sync_force_parallel.attr,
88202a0c 4706 &md_sync_completed.attr,
5e96ee65 4707 &md_min_sync.attr,
c6207277 4708 &md_max_sync.attr,
e464eafd
N		 4709 &md_suspend_lo.attr,
4710 &md_suspend_hi.attr,
9b1d1dac 4711 &md_bitmap.attr,
d7f3d291 4712 &md_degraded.attr,
eae1701f
N		 4713 NULL,
4714};
411036fa
N		 4715static struct attribute_group md_redundancy_group = {
4716 .name = NULL,
4717 .attrs = md_redundancy_attrs,
4718};
4719
eae1701f
N		 4720
4721static ssize_t
4722md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4723{
4724 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4725 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4726 ssize_t rv;
eae1701f
N		 4727
4728 if (!entry->show)
4729 return -EIO;
af8a2434
N		 4730 spin_lock(&all_mddevs_lock);
4731 if (list_empty(&mddev->all_mddevs)) {
4732 spin_unlock(&all_mddevs_lock);
4733 return -EBUSY;
4734 }
4735 mddev_get(mddev);
4736 spin_unlock(&all_mddevs_lock);
4737
5dc5cf7d
IM		 4738 rv = mddev_lock(mddev);
4739 if (!rv) {
4740 rv = entry->show(mddev, page);
4741 mddev_unlock(mddev);
4742 }
af8a2434 4743 mddev_put(mddev);
96de1e66 4744 return rv;
eae1701f
N		 4745}
4746
4747static ssize_t
4748md_attr_store(struct kobject *kobj, struct attribute *attr,
4749 const char *page, size_t length)
4750{
4751 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4752 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4753 ssize_t rv;
eae1701f
N		 4754
4755 if (!entry->store)
4756 return -EIO;
67463acb
N		 4757 if (!capable(CAP_SYS_ADMIN))
4758 return -EACCES;
af8a2434
N		 4759 spin_lock(&all_mddevs_lock);
4760 if (list_empty(&mddev->all_mddevs)) {
4761 spin_unlock(&all_mddevs_lock);
4762 return -EBUSY;
4763 }
4764 mddev_get(mddev);
4765 spin_unlock(&all_mddevs_lock);
a7a3f08d
N		 4766 if (entry->store == new_dev_store)
4767 flush_workqueue(md_misc_wq);
5dc5cf7d
IM		 4768 rv = mddev_lock(mddev);
4769 if (!rv) {
4770 rv = entry->store(mddev, page, length);
4771 mddev_unlock(mddev);
4772 }
af8a2434 4773 mddev_put(mddev);
96de1e66 4774 return rv;
eae1701f
N		 4775}
4776
4777static void md_free(struct kobject *ko)
4778{
fd01b88c 4779 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4780
4781 if (mddev->sysfs_state)
4782 sysfs_put(mddev->sysfs_state);
4783
4784 if (mddev->gendisk) {
4785 del_gendisk(mddev->gendisk);
4786 put_disk(mddev->gendisk);
4787 }
4788 if (mddev->queue)
4789 blk_cleanup_queue(mddev->queue);
4790
eae1701f
N		 4791 kfree(mddev);
4792}
4793
52cf25d0 4794static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4795 .show = md_attr_show,
4796 .store = md_attr_store,
4797};
4798static struct kobj_type md_ktype = {
4799 .release = md_free,
4800 .sysfs_ops = &md_sysfs_ops,
4801 .default_attrs = md_default_attrs,
4802};
4803
1da177e4
LT		 4804int mdp_major = 0;
4805
5fd3a17e
DW		 4806static void mddev_delayed_delete(struct work_struct *ws)
4807{
fd01b88c 4808 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4809
43a70507 4810 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4811 kobject_del(&mddev->kobj);
4812 kobject_put(&mddev->kobj);
4813}
4814
efeb53c0 4815static int md_alloc(dev_t dev, char *name)
1da177e4 4816{
48c9c27b 4817 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4818 struct mddev *mddev = mddev_find(dev);
1da177e4 4819 struct gendisk *disk;
efeb53c0
N		 4820 int partitioned;
4821 int shift;
4822 int unit;
3830c62f 4823 int error;
1da177e4
LT		 4824
4825 if (!mddev)
efeb53c0
N		 4826 return -ENODEV;
4827
4828 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4829 shift = partitioned ? MdpMinorShift : 0;
4830 unit = MINOR(mddev->unit) >> shift;
1da177e4 4831
e804ac78
TH		 4832 /* wait for any previous instance of this device to be
4833 * completely removed (mddev_delayed_delete).
d3374825 4834 */
e804ac78 4835 flush_workqueue(md_misc_wq);
d3374825 4836
48c9c27b 4837 mutex_lock(&disks_mutex);
0909dc44
N		 4838 error = -EEXIST;
4839 if (mddev->gendisk)
4840 goto abort;
efeb53c0
N		 4841
4842 if (name) {
4843 /* Need to ensure that 'name' is not a duplicate.
4844 */
fd01b88c 4845 struct mddev *mddev2;
efeb53c0
N		 4846 spin_lock(&all_mddevs_lock);
4847
4848 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4849 if (mddev2->gendisk &&
4850 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4851 spin_unlock(&all_mddevs_lock);
0909dc44 4852 goto abort;
efeb53c0
N		 4853 }
4854 spin_unlock(&all_mddevs_lock);
1da177e4 4855 }
8b765398 4856
0909dc44 4857 error = -ENOMEM;
8b765398 4858 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4859 if (!mddev->queue)
4860 goto abort;
409c57f3
N		 4861 mddev->queue->queuedata = mddev;
4862
409c57f3 4863 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 4864 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 4865
1da177e4
LT		 4866 disk = alloc_disk(1 << shift);
4867 if (!disk) {
8b765398
N		 4868 blk_cleanup_queue(mddev->queue);
4869 mddev->queue = NULL;
0909dc44 4870 goto abort;
1da177e4 4871 }
efeb53c0 4872 disk->major = MAJOR(mddev->unit);
1da177e4 4873 disk->first_minor = unit << shift;
efeb53c0
N		 4874 if (name)
4875 strcpy(disk->disk_name, name);
4876 else if (partitioned)
1da177e4 4877 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4878 else
1da177e4 4879 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4880 disk->fops = &md_fops;
4881 disk->private_data = mddev;
4882 disk->queue = mddev->queue;
b0140891 4883 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4884 /* Allow extended partitions. This makes the
d3374825 4885 * 'mdp' device redundant, but we can't really
92850bbd
N		 4886 * remove it now.
4887 */
4888 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4889 mddev->gendisk = disk;
b0140891
N		 4890 /* As soon as we call add_disk(), another thread could get
4891 * through to md_open, so make sure it doesn't get too far
4892 */
4893 mutex_lock(&mddev->open_mutex);
4894 add_disk(disk);
4895
ed9e1982
TH		 4896 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4897 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4898 if (error) {
4899 /* This isn't possible, but as kobject_init_and_add is marked
4900 * __must_check, we must do something with the result
4901 */
5e55e2f5
N		 4902 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4903 disk->disk_name);
0909dc44
N		 4904 error = 0;
4905 }
00bcb4ac
N		 4906 if (mddev->kobj.sd &&
4907 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4908 printk(KERN_DEBUG "pointless warning\n");
b0140891 4909 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4910 abort:
4911 mutex_unlock(&disks_mutex);
00bcb4ac 4912 if (!error && mddev->kobj.sd) {
3830c62f 4913 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4914 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4915 }
d3374825 4916 mddev_put(mddev);
0909dc44 4917 return error;
efeb53c0
N		 4918}
4919
4920static struct kobject *md_probe(dev_t dev, int *part, void *data)
4921{
4922 md_alloc(dev, NULL);
1da177e4
LT		 4923 return NULL;
4924}
4925
efeb53c0
N		 4926static int add_named_array(const char *val, struct kernel_param *kp)
4927{
4928 /* val must be "md_*" where * is not all digits.
4929 * We allocate an array with a large free minor number, and
4930 * set the name to val. val must not already be an active name.
4931 */
4932 int len = strlen(val);
4933 char buf[DISK_NAME_LEN];
4934
4935 while (len && val[len-1] == '\n')
4936 len--;
4937 if (len >= DISK_NAME_LEN)
4938 return -E2BIG;
4939 strlcpy(buf, val, len+1);
4940 if (strncmp(buf, "md_", 3) != 0)
4941 return -EINVAL;
4942 return md_alloc(0, buf);
4943}
4944
1da177e4
LT		 4945static void md_safemode_timeout(unsigned long data)
4946{
fd01b88c 4947 struct mddev *mddev = (struct mddev *) data;
1da177e4 4948
0fd62b86
NB		 4949 if (!atomic_read(&mddev->writes_pending)) {
4950 mddev->safemode = 1;
4951 if (mddev->external)
00bcb4ac 4952 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4953 }
1da177e4
LT		 4954 md_wakeup_thread(mddev->thread);
4955}
4956
6ff8d8ec 4957static int start_dirty_degraded;
1da177e4 4958
fd01b88c 4959int md_run(struct mddev *mddev)
1da177e4 4960{
2604b703 4961 int err;
3cb03002 4962 struct md_rdev *rdev;
84fc4b56 4963 struct md_personality *pers;
1da177e4 4964
a757e64c
N		 4965 if (list_empty(&mddev->disks))
4966 /* cannot run an array with no devices.. */
1da177e4 4967 return -EINVAL;
1da177e4
LT		 4968
4969 if (mddev->pers)
4970 return -EBUSY;
bb4f1e9d
N		 4971 /* Cannot run until previous stop completes properly */
4972 if (mddev->sysfs_active)
4973 return -EBUSY;
b6eb127d 4974
1da177e4
LT		 4975 /*
4976 * Analyze all RAID superblock(s)
4977 */
1ec4a939
N		 4978 if (!mddev->raid_disks) {
4979 if (!mddev->persistent)
4980 return -EINVAL;
a757e64c 4981 analyze_sbs(mddev);
1ec4a939 4982 }
1da177e4 4983
d9d166c2
N		 4984 if (mddev->level != LEVEL_NONE)
4985 request_module("md-level-%d", mddev->level);
4986 else if (mddev->clevel[0])
4987 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4988
4989 /*
4990 * Drop all container device buffers, from now on
4991 * the only valid external interface is through the md
4992 * device.
1da177e4 4993 */
dafb20fa 4994 rdev_for_each(rdev, mddev) {
b2d444d7 4995 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4996 continue;
4997 sync_blockdev(rdev->bdev);
f98393a6 4998 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4999
5000 /* perform some consistency tests on the device.
5001 * We don't want the data to overlap the metadata,
58c0fed4 5002 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 5003 */
a6ff7e08
JB		 5004 if (rdev->meta_bdev) {
5005 /* Nothing to check */;
5006 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 5007 if (mddev->dev_sectors &&
5008 rdev->data_offset + mddev->dev_sectors
0f420358 5009 > rdev->sb_start) {
f0d76d70
N		 5010 printk("md: %s: data overlaps metadata\n",
5011 mdname(mddev));
5012 return -EINVAL;
5013 }
5014 } else {
0f420358 5015 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 5016 > rdev->data_offset) {
5017 printk("md: %s: metadata overlaps data\n",
5018 mdname(mddev));
5019 return -EINVAL;
5020 }
5021 }
00bcb4ac 5022 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 5023 }
5024
a167f663 5025 if (mddev->bio_set == NULL)
395c72a7 5026 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
a167f663 5027
1da177e4 5028 spin_lock(&pers_lock);
d9d166c2 5029 pers = find_pers(mddev->level, mddev->clevel);
2604b703 5030 if (!pers || !try_module_get(pers->owner)) {
1da177e4 5031 spin_unlock(&pers_lock);
d9d166c2
N		 5032 if (mddev->level != LEVEL_NONE)
5033 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5034 mddev->level);
5035 else
5036 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5037 mddev->clevel);
1da177e4
LT		 5038 return -EINVAL;
5039 }
2604b703 5040 mddev->pers = pers;
1da177e4 5041 spin_unlock(&pers_lock);
34817e8c
N		 5042 if (mddev->level != pers->level) {
5043 mddev->level = pers->level;
5044 mddev->new_level = pers->level;
5045 }
d9d166c2 5046 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 5047
f6705578 5048 if (mddev->reshape_position != MaxSector &&
63c70c4f 5049 pers->start_reshape == NULL) {
f6705578
N		 5050 /* This personality cannot handle reshaping... */
5051 mddev->pers = NULL;
5052 module_put(pers->owner);
5053 return -EINVAL;
5054 }
5055
7dd5e7c3
N		 5056 if (pers->sync_request) {
5057 /* Warn if this is a potentially silly
5058 * configuration.
5059 */
5060 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5061 struct md_rdev *rdev2;
7dd5e7c3 5062 int warned = 0;
159ec1fc 5063
dafb20fa
N		 5064 rdev_for_each(rdev, mddev)
5065 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N		 5066 if (rdev < rdev2 &&
5067 rdev->bdev->bd_contains ==
5068 rdev2->bdev->bd_contains) {
5069 printk(KERN_WARNING
5070 "%s: WARNING: %s appears to be"
5071 " on the same physical disk as"
5072 " %s.\n",
5073 mdname(mddev),
5074 bdevname(rdev->bdev,b),
5075 bdevname(rdev2->bdev,b2));
5076 warned = 1;
5077 }
5078 }
159ec1fc 5079
7dd5e7c3
N		 5080 if (warned)
5081 printk(KERN_WARNING
5082 "True protection against single-disk"
5083 " failure might be compromised.\n");
5084 }
5085
657390d2 5086 mddev->recovery = 0;
58c0fed4
AN		 5087 /* may be over-ridden by personality */
5088 mddev->resync_max_sectors = mddev->dev_sectors;
5089
6ff8d8ec 5090 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5091
0f9552b5 5092 if (start_readonly && mddev->ro == 0)
f91de92e
N		 5093 mddev->ro = 2; /* read-only, but switch on first write */
5094
b15c2e57 5095 err = mddev->pers->run(mddev);
13e53df3
AN		 5096 if (err)
5097 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 5098 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5099 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5100 " but 'external_size' not in effect?\n", __func__);
5101 printk(KERN_ERR
5102 "md: invalid array_size %llu > default size %llu\n",
5103 (unsigned long long)mddev->array_sectors / 2,
5104 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5105 err = -EINVAL;
5106 mddev->pers->stop(mddev);
5107 }
ef99bf48
N		 5108 if (err == 0 && mddev->pers->sync_request &&
5109 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
b15c2e57
N		 5110 err = bitmap_create(mddev);
5111 if (err) {
5112 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5113 mdname(mddev), err);
5114 mddev->pers->stop(mddev);
5115 }
5116 }
1da177e4 5117 if (err) {
1da177e4
LT		 5118 module_put(mddev->pers->owner);
5119 mddev->pers = NULL;
32a7627c
N		 5120 bitmap_destroy(mddev);
5121 return err;
1da177e4 5122 }
5e55e2f5 5123 if (mddev->pers->sync_request) {
00bcb4ac
N		 5124 if (mddev->kobj.sd &&
5125 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 5126 printk(KERN_WARNING
5127 "md: cannot register extra attributes for %s\n",
5128 mdname(mddev));
00bcb4ac 5129 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5130 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 5131 mddev->ro = 0;
5132
1da177e4 5133 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 5134 atomic_set(&mddev->max_corr_read_errors,
5135 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 5136 mddev->safemode = 0;
5137 mddev->safemode_timer.function = md_safemode_timeout;
5138 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 5139 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5140 mddev->in_sync = 1;
0ca69886
N		 5141 smp_wmb();
5142 mddev->ready = 1;
dafb20fa 5143 rdev_for_each(rdev, mddev)
36fad858
NK		 5144 if (rdev->raid_disk >= 0)
5145 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 5146 /* failure here is OK */;
1da177e4
LT		 5147
5148 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5149
850b2b42
N		 5150 if (mddev->flags)
5151 md_update_sb(mddev, 0);
1da177e4 5152
d7603b7e 5153 md_new_event(mddev);
00bcb4ac
N		 5154 sysfs_notify_dirent_safe(mddev->sysfs_state);
5155 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 5156 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 5157 return 0;
5158}
390ee602 5159EXPORT_SYMBOL_GPL(md_run);
1da177e4 5160
fd01b88c 5161static int do_md_run(struct mddev *mddev)
fe60b014
N		 5162{
5163 int err;
5164
5165 err = md_run(mddev);
5166 if (err)
5167 goto out;
69e51b44
N		 5168 err = bitmap_load(mddev);
5169 if (err) {
5170 bitmap_destroy(mddev);
5171 goto out;
5172 }
0fd018af
JB		 5173
5174 md_wakeup_thread(mddev->thread);
5175 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5176
fe60b014
N		 5177 set_capacity(mddev->gendisk, mddev->array_sectors);
5178 revalidate_disk(mddev->gendisk);
f0b4f7e2 5179 mddev->changed = 1;
fe60b014
N		 5180 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5181out:
5182 return err;
5183}
5184
fd01b88c 5185static int restart_array(struct mddev *mddev)
1da177e4
LT		 5186{
5187 struct gendisk *disk = mddev->gendisk;
1da177e4 5188
80fab1d7 5189 /* Complain if it has no devices */
1da177e4 5190 if (list_empty(&mddev->disks))
80fab1d7
AN		 5191 return -ENXIO;
5192 if (!mddev->pers)
5193 return -EINVAL;
5194 if (!mddev->ro)
5195 return -EBUSY;
5196 mddev->safemode = 0;
5197 mddev->ro = 0;
5198 set_disk_ro(disk, 0);
5199 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5200 mdname(mddev));
5201 /* Kick recovery or resync if necessary */
5202 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5203 md_wakeup_thread(mddev->thread);
5204 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5205 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5206 return 0;
1da177e4
LT		 5207}
5208
acc55e22
N		 5209/* similar to deny_write_access, but accounts for our holding a reference
5210 * to the file ourselves */
5211static int deny_bitmap_write_access(struct file * file)
5212{
5213 struct inode *inode = file->f_mapping->host;
5214
5215 spin_lock(&inode->i_lock);
5216 if (atomic_read(&inode->i_writecount) > 1) {
5217 spin_unlock(&inode->i_lock);
5218 return -ETXTBSY;
5219 }
5220 atomic_set(&inode->i_writecount, -1);
5221 spin_unlock(&inode->i_lock);
5222
5223 return 0;
5224}
5225
43a70507 5226void restore_bitmap_write_access(struct file *file)
acc55e22
N		 5227{
5228 struct inode *inode = file->f_mapping->host;
5229
5230 spin_lock(&inode->i_lock);
5231 atomic_set(&inode->i_writecount, 1);
5232 spin_unlock(&inode->i_lock);
5233}
5234
fd01b88c 5235static void md_clean(struct mddev *mddev)
6177b472
N		 5236{
5237 mddev->array_sectors = 0;
5238 mddev->external_size = 0;
5239 mddev->dev_sectors = 0;
5240 mddev->raid_disks = 0;
5241 mddev->recovery_cp = 0;
5242 mddev->resync_min = 0;
5243 mddev->resync_max = MaxSector;
5244 mddev->reshape_position = MaxSector;
5245 mddev->external = 0;
5246 mddev->persistent = 0;
5247 mddev->level = LEVEL_NONE;
5248 mddev->clevel[0] = 0;
5249 mddev->flags = 0;
5250 mddev->ro = 0;
5251 mddev->metadata_type[0] = 0;
5252 mddev->chunk_sectors = 0;
5253 mddev->ctime = mddev->utime = 0;
5254 mddev->layout = 0;
5255 mddev->max_disks = 0;
5256 mddev->events = 0;
a8707c08 5257 mddev->can_decrease_events = 0;
6177b472 5258 mddev->delta_disks = 0;
2c810cdd 5259 mddev->reshape_backwards = 0;
6177b472
N		 5260 mddev->new_level = LEVEL_NONE;
5261 mddev->new_layout = 0;
5262 mddev->new_chunk_sectors = 0;
5263 mddev->curr_resync = 0;
7f7583d4 5264 atomic64_set(&mddev->resync_mismatches, 0);
6177b472
N		 5265 mddev->suspend_lo = mddev->suspend_hi = 0;
5266 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5267 mddev->recovery = 0;
5268 mddev->in_sync = 0;
f0b4f7e2 5269 mddev->changed = 0;
6177b472 5270 mddev->degraded = 0;
6177b472 5271 mddev->safemode = 0;
050b6615 5272 mddev->merge_check_needed = 0;
6177b472
N		 5273 mddev->bitmap_info.offset = 0;
5274 mddev->bitmap_info.default_offset = 0;
6409bb05 5275 mddev->bitmap_info.default_space = 0;
6177b472
N		 5276 mddev->bitmap_info.chunksize = 0;
5277 mddev->bitmap_info.daemon_sleep = 0;
5278 mddev->bitmap_info.max_write_behind = 0;
5279}
5280
fd01b88c 5281static void __md_stop_writes(struct mddev *mddev)
a047e125
N		 5282{
5283 if (mddev->sync_thread) {
5284 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5285 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 5286 md_reap_sync_thread(mddev);
a047e125
N		 5287 }
5288
5289 del_timer_sync(&mddev->safemode_timer);
5290
5291 bitmap_flush(mddev);
5292 md_super_wait(mddev);
5293
b6d428c6
N		 5294 if (mddev->ro == 0 &&
5295 (!mddev->in_sync || mddev->flags)) {
a047e125
N		 5296 /* mark array as shutdown cleanly */
5297 mddev->in_sync = 1;
5298 md_update_sb(mddev, 1);
5299 }
5300}
defad61a 5301
fd01b88c 5302void md_stop_writes(struct mddev *mddev)
defad61a
N		 5303{
5304 mddev_lock(mddev);
5305 __md_stop_writes(mddev);
5306 mddev_unlock(mddev);
5307}
390ee602 5308EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5309
5eff3c43 5310static void __md_stop(struct mddev *mddev)
6177b472 5311{
0ca69886 5312 mddev->ready = 0;
6177b472
N		 5313 mddev->pers->stop(mddev);
5314 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5315 mddev->to_remove = &md_redundancy_group;
5316 module_put(mddev->pers->owner);
5317 mddev->pers = NULL;
cca9cf90 5318 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5319}
5eff3c43
N		 5320
5321void md_stop(struct mddev *mddev)
5322{
5323 /* stop the array and free an attached data structures.
5324 * This is called from dm-raid
5325 */
5326 __md_stop(mddev);
5327 bitmap_destroy(mddev);
5328 if (mddev->bio_set)
5329 bioset_free(mddev->bio_set);
5330}
5331
390ee602 5332EXPORT_SYMBOL_GPL(md_stop);
6177b472 5333
a05b7ea0 5334static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N		 5335{
5336 int err = 0;
5337 mutex_lock(&mddev->open_mutex);
a05b7ea0 5338 if (atomic_read(&mddev->openers) > !!bdev) {
a4bd82d0
N		 5339 printk("md: %s still in use.\n",mdname(mddev));
5340 err = -EBUSY;
5341 goto out;
5342 }
a05b7ea0
N		 5343 if (bdev)
5344 sync_blockdev(bdev);
a4bd82d0 5345 if (mddev->pers) {
defad61a 5346 __md_stop_writes(mddev);
a4bd82d0
N		 5347
5348 err = -ENXIO;
5349 if (mddev->ro==1)
5350 goto out;
5351 mddev->ro = 1;
5352 set_disk_ro(mddev->gendisk, 1);
5353 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 5354 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 5355 err = 0;
5356 }
5357out:
5358 mutex_unlock(&mddev->open_mutex);
5359 return err;
5360}
5361
9e653b63
N		 5362/* mode:
5363 * 0 - completely stop and dis-assemble array
9e653b63
N		 5364 * 2 - stop but do not disassemble array
5365 */
a05b7ea0
N		 5366static int do_md_stop(struct mddev * mddev, int mode,
5367 struct block_device *bdev)
1da177e4 5368{
1da177e4 5369 struct gendisk *disk = mddev->gendisk;
3cb03002 5370 struct md_rdev *rdev;
1da177e4 5371
c8c00a69 5372 mutex_lock(&mddev->open_mutex);
a05b7ea0 5373 if (atomic_read(&mddev->openers) > !!bdev ||
bb4f1e9d 5374 mddev->sysfs_active) {
df5b20cf 5375 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 5376 mutex_unlock(&mddev->open_mutex);
5377 return -EBUSY;
5378 }
a05b7ea0
N		 5379 if (bdev)
5380 /* It is possible IO was issued on some other
5381 * open file which was closed before we took ->open_mutex.
5382 * As that was not the last close __blkdev_put will not
5383 * have called sync_blockdev, so we must.
5384 */
5385 sync_blockdev(bdev);
1da177e4 5386
6e17b027 5387 if (mddev->pers) {
a4bd82d0
N		 5388 if (mddev->ro)
5389 set_disk_ro(disk, 0);
409c57f3 5390
defad61a 5391 __md_stop_writes(mddev);
5eff3c43 5392 __md_stop(mddev);
a4bd82d0 5393 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5394 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5395
a4bd82d0 5396 /* tell userspace to handle 'inactive' */
00bcb4ac 5397 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5398
dafb20fa 5399 rdev_for_each(rdev, mddev)
36fad858
NK		 5400 if (rdev->raid_disk >= 0)
5401 sysfs_unlink_rdev(mddev, rdev);
c4647292 5402
a4bd82d0 5403 set_capacity(disk, 0);
6e17b027 5404 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5405 mddev->changed = 1;
a4bd82d0 5406 revalidate_disk(disk);
0d4ca600 5407
a4bd82d0
N		 5408 if (mddev->ro)
5409 mddev->ro = 0;
6e17b027
N		 5410 } else
5411 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5412 /*
5413 * Free resources if final stop
5414 */
9e653b63 5415 if (mode == 0) {
1da177e4
LT		 5416 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5417
978f946b 5418 bitmap_destroy(mddev);
c3d9714e
N		 5419 if (mddev->bitmap_info.file) {
5420 restore_bitmap_write_access(mddev->bitmap_info.file);
5421 fput(mddev->bitmap_info.file);
5422 mddev->bitmap_info.file = NULL;
978f946b 5423 }
c3d9714e 5424 mddev->bitmap_info.offset = 0;
978f946b 5425
1da177e4
LT		 5426 export_array(mddev);
5427
6177b472 5428 md_clean(mddev);
934d9c23 5429 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5430 if (mddev->hold_active == UNTIL_STOP)
5431 mddev->hold_active = 0;
a4bd82d0 5432 }
3f9d99c1 5433 blk_integrity_unregister(disk);
d7603b7e 5434 md_new_event(mddev);
00bcb4ac 5435 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5436 return 0;
1da177e4
LT		 5437}
5438
fdee8ae4 5439#ifndef MODULE
fd01b88c 5440static void autorun_array(struct mddev *mddev)
1da177e4 5441{
3cb03002 5442 struct md_rdev *rdev;
1da177e4
LT		 5443 int err;
5444
a757e64c 5445 if (list_empty(&mddev->disks))
1da177e4 5446 return;
1da177e4
LT		 5447
5448 printk(KERN_INFO "md: running: ");
5449
dafb20fa 5450 rdev_for_each(rdev, mddev) {
1da177e4
LT		 5451 char b[BDEVNAME_SIZE];
5452 printk("<%s>", bdevname(rdev->bdev,b));
5453 }
5454 printk("\n");
5455
d710e138 5456 err = do_md_run(mddev);
1da177e4
LT		 5457 if (err) {
5458 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
a05b7ea0 5459 do_md_stop(mddev, 0, NULL);
1da177e4
LT		 5460 }
5461}
5462
5463/*
5464 * lets try to run arrays based on all disks that have arrived
5465 * until now. (those are in pending_raid_disks)
5466 *
5467 * the method: pick the first pending disk, collect all disks with
5468 * the same UUID, remove all from the pending list and put them into
5469 * the 'same_array' list. Then order this list based on superblock
5470 * update time (freshest comes first), kick out 'old' disks and
5471 * compare superblocks. If everything's fine then run it.
5472 *
5473 * If "unit" is allocated, then bump its reference count
5474 */
5475static void autorun_devices(int part)
5476{
3cb03002 5477 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5478 struct mddev *mddev;
1da177e4
LT		 5479 char b[BDEVNAME_SIZE];
5480
5481 printk(KERN_INFO "md: autorun ...\n");
5482 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5483 int unit;
1da177e4 5484 dev_t dev;
ad01c9e3 5485 LIST_HEAD(candidates);
1da177e4 5486 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5487 struct md_rdev, same_set);
1da177e4
LT		 5488
5489 printk(KERN_INFO "md: considering %s ...\n",
5490 bdevname(rdev0->bdev,b));
5491 INIT_LIST_HEAD(&candidates);
159ec1fc 5492 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5493 if (super_90_load(rdev, rdev0, 0) >= 0) {
5494 printk(KERN_INFO "md: adding %s ...\n",
5495 bdevname(rdev->bdev,b));
5496 list_move(&rdev->same_set, &candidates);
5497 }
5498 /*
5499 * now we have a set of devices, with all of them having
5500 * mostly sane superblocks. It's time to allocate the
5501 * mddev.
5502 */
e8703fe1
N		 5503 if (part) {
5504 dev = MKDEV(mdp_major,
5505 rdev0->preferred_minor << MdpMinorShift);
5506 unit = MINOR(dev) >> MdpMinorShift;
5507 } else {
5508 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5509 unit = MINOR(dev);
5510 }
5511 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5512 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5513 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5514 break;
5515 }
1da177e4
LT		 5516
5517 md_probe(dev, NULL, NULL);
5518 mddev = mddev_find(dev);
9bbbca3a
NB		 5519 if (!mddev || !mddev->gendisk) {
5520 if (mddev)
5521 mddev_put(mddev);
5522 printk(KERN_ERR
1da177e4
LT		 5523 "md: cannot allocate memory for md drive.\n");
5524 break;
5525 }
5526 if (mddev_lock(mddev))
5527 printk(KERN_WARNING "md: %s locked, cannot run\n",
5528 mdname(mddev));
5529 else if (mddev->raid_disks || mddev->major_version
5530 || !list_empty(&mddev->disks)) {
5531 printk(KERN_WARNING
5532 "md: %s already running, cannot run %s\n",
5533 mdname(mddev), bdevname(rdev0->bdev,b));
5534 mddev_unlock(mddev);
5535 } else {
5536 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5537 mddev->persistent = 1;
159ec1fc 5538 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5539 list_del_init(&rdev->same_set);
5540 if (bind_rdev_to_array(rdev, mddev))
5541 export_rdev(rdev);
5542 }
5543 autorun_array(mddev);
5544 mddev_unlock(mddev);
5545 }
5546 /* on success, candidates will be empty, on error
5547 * it won't...
5548 */
159ec1fc 5549 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5550 list_del_init(&rdev->same_set);
1da177e4 5551 export_rdev(rdev);
4b80991c 5552 }
1da177e4
LT		 5553 mddev_put(mddev);
5554 }
5555 printk(KERN_INFO "md: ... autorun DONE.\n");
5556}
fdee8ae4 5557#endif /* !MODULE */
1da177e4 5558
1da177e4
LT		 5559static int get_version(void __user * arg)
5560{
5561 mdu_version_t ver;
5562
5563 ver.major = MD_MAJOR_VERSION;
5564 ver.minor = MD_MINOR_VERSION;
5565 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5566
5567 if (copy_to_user(arg, &ver, sizeof(ver)))
5568 return -EFAULT;
5569
5570 return 0;
5571}
5572
fd01b88c 5573static int get_array_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5574{
5575 mdu_array_info_t info;
a9f326eb 5576 int nr,working,insync,failed,spare;
3cb03002 5577 struct md_rdev *rdev;
1da177e4 5578
1ca69c4b
N		 5579 nr = working = insync = failed = spare = 0;
5580 rcu_read_lock();
5581 rdev_for_each_rcu(rdev, mddev) {
1da177e4 5582 nr++;
b2d444d7 5583 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5584 failed++;
5585 else {
5586 working++;
b2d444d7 5587 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5588 insync++;
1da177e4
LT		 5589 else
5590 spare++;
5591 }
5592 }
1ca69c4b 5593 rcu_read_unlock();
1da177e4
LT		 5594
5595 info.major_version = mddev->major_version;
5596 info.minor_version = mddev->minor_version;
5597 info.patch_version = MD_PATCHLEVEL_VERSION;
5598 info.ctime = mddev->ctime;
5599 info.level = mddev->level;
58c0fed4
AN		 5600 info.size = mddev->dev_sectors / 2;
5601 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5602 info.size = -1;
1da177e4
LT		 5603 info.nr_disks = nr;
5604 info.raid_disks = mddev->raid_disks;
5605 info.md_minor = mddev->md_minor;
5606 info.not_persistent= !mddev->persistent;
5607
5608 info.utime = mddev->utime;
5609 info.state = 0;
5610 if (mddev->in_sync)
5611 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5612 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5613 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5614 info.active_disks = insync;
1da177e4
LT		 5615 info.working_disks = working;
5616 info.failed_disks = failed;
5617 info.spare_disks = spare;
5618
5619 info.layout = mddev->layout;
9d8f0363 5620 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5621
5622 if (copy_to_user(arg, &info, sizeof(info)))
5623 return -EFAULT;
5624
5625 return 0;
5626}
5627
fd01b88c 5628static int get_bitmap_file(struct mddev * mddev, void __user * arg)
32a7627c
N		 5629{
5630 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5631 char *ptr, *buf = NULL;
5632 int err = -ENOMEM;
5633
b5470dc5
DW		 5634 if (md_allow_write(mddev))
5635 file = kmalloc(sizeof(*file), GFP_NOIO);
5636 else
5637 file = kmalloc(sizeof(*file), GFP_KERNEL);
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
1da177e4
LT		 6420 err = mddev_lock(mddev);
6421 if (err) {
6422 printk(KERN_INFO
6423 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6424 err, cmd);
6425 goto abort;
6426 }
6427
c02c0aeb
N		 6428 if (cmd == SET_ARRAY_INFO) {
6429 mdu_array_info_t info;
6430 if (!arg)
6431 memset(&info, 0, sizeof(info));
6432 else if (copy_from_user(&info, argp, sizeof(info))) {
6433 err = -EFAULT;
6434 goto abort_unlock;
6435 }
6436 if (mddev->pers) {
6437 err = update_array_info(mddev, &info);
6438 if (err) {
6439 printk(KERN_WARNING "md: couldn't update"
6440 " array info. %d\n", err);
6441 goto abort_unlock;
1da177e4
LT		 6442 }
6443 goto done_unlock;
c02c0aeb
N		 6444 }
6445 if (!list_empty(&mddev->disks)) {
6446 printk(KERN_WARNING
6447 "md: array %s already has disks!\n",
6448 mdname(mddev));
6449 err = -EBUSY;
6450 goto abort_unlock;
6451 }
6452 if (mddev->raid_disks) {
6453 printk(KERN_WARNING
6454 "md: array %s already initialised!\n",
6455 mdname(mddev));
6456 err = -EBUSY;
6457 goto abort_unlock;
6458 }
6459 err = set_array_info(mddev, &info);
6460 if (err) {
6461 printk(KERN_WARNING "md: couldn't set"
6462 " array info. %d\n", err);
6463 goto abort_unlock;
6464 }
6465 goto done_unlock;
1da177e4
LT		 6466 }
6467
6468 /*
6469 * Commands querying/configuring an existing array:
6470 */
32a7627c 6471 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6472 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6473 if ((!mddev->raid_disks && !mddev->external)
6474 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6475 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6476 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 6477 err = -ENODEV;
6478 goto abort_unlock;
6479 }
6480
6481 /*
6482 * Commands even a read-only array can execute:
6483 */
c02c0aeb
N		 6484 switch (cmd) {
6485 case GET_BITMAP_FILE:
6486 err = get_bitmap_file(mddev, argp);
6487 goto done_unlock;
32a7627c 6488
c02c0aeb
N		 6489 case RESTART_ARRAY_RW:
6490 err = restart_array(mddev);
6491 goto done_unlock;
1da177e4 6492
c02c0aeb
N		 6493 case STOP_ARRAY:
6494 err = do_md_stop(mddev, 0, bdev);
6495 goto done_unlock;
1da177e4 6496
c02c0aeb
N		 6497 case STOP_ARRAY_RO:
6498 err = md_set_readonly(mddev, bdev);
6499 goto done_unlock;
1da177e4 6500
3ea8929d
N		 6501 case HOT_REMOVE_DISK:
6502 err = hot_remove_disk(mddev, new_decode_dev(arg));
6503 goto done_unlock;
6504
7ceb17e8
N		 6505 case ADD_NEW_DISK:
6506 /* We can support ADD_NEW_DISK on read-only arrays
6507 * on if we are re-adding a preexisting device.
6508 * So require mddev->pers and MD_DISK_SYNC.
6509 */
6510 if (mddev->pers) {
6511 mdu_disk_info_t info;
6512 if (copy_from_user(&info, argp, sizeof(info)))
6513 err = -EFAULT;
6514 else if (!(info.state & (1<<MD_DISK_SYNC)))
6515 /* Need to clear read-only for this */
6516 break;
6517 else
6518 err = add_new_disk(mddev, &info);
6519 goto done_unlock;
6520 }
6521 break;
6522
c02c0aeb
N		 6523 case BLKROSET:
6524 if (get_user(ro, (int __user *)(arg))) {
6525 err = -EFAULT;
6526 goto done_unlock;
6527 }
6528 err = -EINVAL;
e2218350 6529
c02c0aeb
N		 6530 /* if the bdev is going readonly the value of mddev->ro
6531 * does not matter, no writes are coming
6532 */
6533 if (ro)
6534 goto done_unlock;
e2218350 6535
c02c0aeb
N		 6536 /* are we are already prepared for writes? */
6537 if (mddev->ro != 1)
6538 goto done_unlock;
e2218350 6539
c02c0aeb
N		 6540 /* transitioning to readauto need only happen for
6541 * arrays that call md_write_start
6542 */
6543 if (mddev->pers) {
6544 err = restart_array(mddev);
6545 if (err == 0) {
6546 mddev->ro = 2;
6547 set_disk_ro(mddev->gendisk, 0);
e2218350 6548 }
c02c0aeb
N		 6549 }
6550 goto done_unlock;
1da177e4
LT		 6551 }
6552
6553 /*
6554 * The remaining ioctls are changing the state of the
f91de92e
N		 6555 * superblock, so we do not allow them on read-only arrays.
6556 * However non-MD ioctls (e.g. get-size) will still come through
6557 * here and hit the 'default' below, so only disallow
6558 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 6559 */
bb57fc64 6560 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 6561 if (mddev->ro == 2) {
6562 mddev->ro = 0;
00bcb4ac 6563 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 6564 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N		 6565 /* mddev_unlock will wake thread */
6566 /* If a device failed while we were read-only, we
6567 * need to make sure the metadata is updated now.
6568 */
6569 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6570 mddev_unlock(mddev);
6571 wait_event(mddev->sb_wait,
6572 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6573 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6574 mddev_lock(mddev);
6575 }
f91de92e
N		 6576 } else {
6577 err = -EROFS;
6578 goto abort_unlock;
6579 }
1da177e4
LT		 6580 }
6581
c02c0aeb
N		 6582 switch (cmd) {
6583 case ADD_NEW_DISK:
1da177e4 6584 {
c02c0aeb
N		 6585 mdu_disk_info_t info;
6586 if (copy_from_user(&info, argp, sizeof(info)))
6587 err = -EFAULT;
6588 else
6589 err = add_new_disk(mddev, &info);
6590 goto done_unlock;
6591 }
1da177e4 6592
c02c0aeb
N		 6593 case HOT_ADD_DISK:
6594 err = hot_add_disk(mddev, new_decode_dev(arg));
6595 goto done_unlock;
1da177e4 6596
c02c0aeb
N		 6597 case RUN_ARRAY:
6598 err = do_md_run(mddev);
6599 goto done_unlock;
1da177e4 6600
c02c0aeb
N		 6601 case SET_BITMAP_FILE:
6602 err = set_bitmap_file(mddev, (int)arg);
6603 goto done_unlock;
32a7627c 6604
c02c0aeb
N		 6605 default:
6606 err = -EINVAL;
6607 goto abort_unlock;
1da177e4
LT		 6608 }
6609
6610done_unlock:
6611abort_unlock:
d3374825
N		 6612 if (mddev->hold_active == UNTIL_IOCTL &&
6613 err != -EINVAL)
6614 mddev->hold_active = 0;
1da177e4
LT		 6615 mddev_unlock(mddev);
6616
6617 return err;
6618done:
6619 if (err)
6620 MD_BUG();
6621abort:
6622 return err;
6623}
aa98aa31
AB		 6624#ifdef CONFIG_COMPAT
6625static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6626 unsigned int cmd, unsigned long arg)
6627{
6628 switch (cmd) {
6629 case HOT_REMOVE_DISK:
6630 case HOT_ADD_DISK:
6631 case SET_DISK_FAULTY:
6632 case SET_BITMAP_FILE:
6633 /* These take in integer arg, do not convert */
6634 break;
6635 default:
6636 arg = (unsigned long)compat_ptr(arg);
6637 break;
6638 }
6639
6640 return md_ioctl(bdev, mode, cmd, arg);
6641}
6642#endif /* CONFIG_COMPAT */
1da177e4 6643
a39907fa 6644static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6645{
6646 /*
6647 * Succeed if we can lock the mddev, which confirms that
6648 * it isn't being stopped right now.
6649 */
fd01b88c 6650 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6651 int err;
6652
0c098220
YL		 6653 if (!mddev)
6654 return -ENODEV;
6655
d3374825
N		 6656 if (mddev->gendisk != bdev->bd_disk) {
6657 /* we are racing with mddev_put which is discarding this
6658 * bd_disk.
6659 */
6660 mddev_put(mddev);
6661 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6662 flush_workqueue(md_misc_wq);
d3374825
N		 6663 /* Then retry the open from the top */
6664 return -ERESTARTSYS;
6665 }
6666 BUG_ON(mddev != bdev->bd_disk->private_data);
6667
c8c00a69 6668 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6669 goto out;
6670
6671 err = 0;
f2ea68cf 6672 atomic_inc(&mddev->openers);
c8c00a69 6673 mutex_unlock(&mddev->open_mutex);
1da177e4 6674
f0b4f7e2 6675 check_disk_change(bdev);
1da177e4
LT		 6676 out:
6677 return err;
6678}
6679
a39907fa 6680static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6681{
fd01b88c 6682 struct mddev *mddev = disk->private_data;
1da177e4 6683
52e5f9d1 6684 BUG_ON(!mddev);
f2ea68cf 6685 atomic_dec(&mddev->openers);
1da177e4
LT		 6686 mddev_put(mddev);
6687
6688 return 0;
6689}
f0b4f7e2
N		 6690
6691static int md_media_changed(struct gendisk *disk)
6692{
fd01b88c 6693 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6694
6695 return mddev->changed;
6696}
6697
6698static int md_revalidate(struct gendisk *disk)
6699{
fd01b88c 6700 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6701
6702 mddev->changed = 0;
6703 return 0;
6704}
83d5cde4 6705static const struct block_device_operations md_fops =
1da177e4
LT		 6706{
6707 .owner = THIS_MODULE,
a39907fa
AV		 6708 .open = md_open,
6709 .release = md_release,
b492b852 6710 .ioctl = md_ioctl,
aa98aa31
AB		 6711#ifdef CONFIG_COMPAT
6712 .compat_ioctl = md_compat_ioctl,
6713#endif
a885c8c4 6714 .getgeo = md_getgeo,
f0b4f7e2
N		 6715 .media_changed = md_media_changed,
6716 .revalidate_disk= md_revalidate,
1da177e4
LT		 6717};
6718
75c96f85 6719static int md_thread(void * arg)
1da177e4 6720{
2b8bf345 6721 struct md_thread *thread = arg;
1da177e4 6722
1da177e4
LT		 6723 /*
6724 * md_thread is a 'system-thread', it's priority should be very
6725 * high. We avoid resource deadlocks individually in each
6726 * raid personality. (RAID5 does preallocation) We also use RR and
6727 * the very same RT priority as kswapd, thus we will never get
6728 * into a priority inversion deadlock.
6729 *
6730 * we definitely have to have equal or higher priority than
6731 * bdflush, otherwise bdflush will deadlock if there are too
6732 * many dirty RAID5 blocks.
6733 */
1da177e4 6734
6985c43f 6735 allow_signal(SIGKILL);
a6fb0934 6736 while (!kthread_should_stop()) {
1da177e4 6737
93588e22
N		 6738 /* We need to wait INTERRUPTIBLE so that
6739 * we don't add to the load-average.
6740 * That means we need to be sure no signals are
6741 * pending
6742 */
6743 if (signal_pending(current))
6744 flush_signals(current);
6745
6746 wait_event_interruptible_timeout
6747 (thread->wqueue,
6748 test_bit(THREAD_WAKEUP, &thread->flags)
6749 || kthread_should_stop(),
6750 thread->timeout);
1da177e4 6751
6c987910
N		 6752 clear_bit(THREAD_WAKEUP, &thread->flags);
6753 if (!kthread_should_stop())
4ed8731d 6754 thread->run(thread);
1da177e4 6755 }
a6fb0934 6756
1da177e4
LT		 6757 return 0;
6758}
6759
2b8bf345 6760void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6761{
6762 if (thread) {
36a4e1fe 6763 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6764 set_bit(THREAD_WAKEUP, &thread->flags);
6765 wake_up(&thread->wqueue);
6766 }
6767}
6768
4ed8731d
SL		 6769struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6770 struct mddev *mddev, const char *name)
1da177e4 6771{
2b8bf345 6772 struct md_thread *thread;
1da177e4 6773
2b8bf345 6774 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6775 if (!thread)
6776 return NULL;
6777
1da177e4
LT		 6778 init_waitqueue_head(&thread->wqueue);
6779
1da177e4
LT		 6780 thread->run = run;
6781 thread->mddev = mddev;
32a7627c 6782 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6783 thread->tsk = kthread_run(md_thread, thread,
6784 "%s_%s",
6785 mdname(thread->mddev),
0232605d 6786 name);
a6fb0934 6787 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6788 kfree(thread);
6789 return NULL;
6790 }
1da177e4
LT		 6791 return thread;
6792}
6793
2b8bf345 6794void md_unregister_thread(struct md_thread **threadp)
1da177e4 6795{
2b8bf345 6796 struct md_thread *thread = *threadp;
e0cf8f04
N		 6797 if (!thread)
6798 return;
36a4e1fe 6799 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 6800 /* Locking ensures that mddev_unlock does not wake_up a
6801 * non-existent thread
6802 */
6803 spin_lock(&pers_lock);
6804 *threadp = NULL;
6805 spin_unlock(&pers_lock);
a6fb0934
N		 6806
6807 kthread_stop(thread->tsk);
1da177e4
LT		 6808 kfree(thread);
6809}
6810
fd01b88c 6811void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 6812{
6813 if (!mddev) {
6814 MD_BUG();
6815 return;
6816 }
6817
b2d444d7 6818 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6819 return;
6bfe0b49 6820
de393cde 6821 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 6822 return;
6823 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6824 if (mddev->degraded)
6825 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6826 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6827 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6828 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6829 md_wakeup_thread(mddev->thread);
768a418d 6830 if (mddev->event_work.func)
e804ac78 6831 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6832 md_new_event_inintr(mddev);
1da177e4
LT		 6833}
6834
6835/* seq_file implementation /proc/mdstat */
6836
6837static void status_unused(struct seq_file *seq)
6838{
6839 int i = 0;
3cb03002 6840 struct md_rdev *rdev;
1da177e4
LT		 6841
6842 seq_printf(seq, "unused devices: ");
6843
159ec1fc 6844 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6845 char b[BDEVNAME_SIZE];
6846 i++;
6847 seq_printf(seq, "%s ",
6848 bdevname(rdev->bdev,b));
6849 }
6850 if (!i)
6851 seq_printf(seq, "<none>");
6852
6853 seq_printf(seq, "\n");
6854}
6855
6856
fd01b88c 6857static void status_resync(struct seq_file *seq, struct mddev * mddev)
1da177e4 6858{
dd71cf6b
N		 6859 sector_t max_sectors, resync, res;
6860 unsigned long dt, db;
6861 sector_t rt;
4588b42e
N		 6862 int scale;
6863 unsigned int per_milli;
1da177e4 6864
72f36d59
N		 6865 if (mddev->curr_resync <= 3)
6866 resync = 0;
6867 else
6868 resync = mddev->curr_resync
6869 - atomic_read(&mddev->recovery_active);
1da177e4 6870
c804cdec
N		 6871 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6872 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 6873 max_sectors = mddev->resync_max_sectors;
1da177e4 6874 else
dd71cf6b 6875 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6876
6877 /*
6878 * Should not happen.
6879 */
dd71cf6b 6880 if (!max_sectors) {
1da177e4
LT		 6881 MD_BUG();
6882 return;
6883 }
4588b42e 6884 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6885 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6886 * u32, as those are the requirements for sector_div.
6887 * Thus 'scale' must be at least 10
6888 */
6889 scale = 10;
6890 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6891 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6892 scale++;
6893 }
6894 res = (resync>>scale)*1000;
dd71cf6b 6895 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6896
6897 per_milli = res;
1da177e4 6898 {
4588b42e 6899 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6900 seq_printf(seq, "[");
6901 for (i = 0; i < x; i++)
6902 seq_printf(seq, "=");
6903 seq_printf(seq, ">");
6904 for (i = 0; i < y; i++)
6905 seq_printf(seq, ".");
6906 seq_printf(seq, "] ");
6907 }
4588b42e 6908 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6909 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6910 "reshape" :
61df9d91
N		 6911 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6912 "check" :
6913 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6914 "resync" : "recovery"))),
6915 per_milli/10, per_milli % 10,
dd71cf6b
N		 6916 (unsigned long long) resync/2,
6917 (unsigned long long) max_sectors/2);
1da177e4
LT		 6918
6919 /*
1da177e4
LT		 6920 * dt: time from mark until now
6921 * db: blocks written from mark until now
6922 * rt: remaining time
dd71cf6b
N		 6923 *
6924 * rt is a sector_t, so could be 32bit or 64bit.
6925 * So we divide before multiply in case it is 32bit and close
6926 * to the limit.
25985edc 6927 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 6928 * near the end of resync when the number of remaining sectors
6929 * is close to 'db'.
6930 * We then divide rt by 32 after multiplying by db to compensate.
6931 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6932 */
6933 dt = ((jiffies - mddev->resync_mark) / HZ);
6934 if (!dt) dt++;
ff4e8d9a
N		 6935 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6936 - mddev->resync_mark_cnt;
1da177e4 6937
dd71cf6b
N		 6938 rt = max_sectors - resync; /* number of remaining sectors */
6939 sector_div(rt, db/32+1);
6940 rt *= dt;
6941 rt >>= 5;
6942
6943 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6944 ((unsigned long)rt % 60)/6);
1da177e4 6945
ff4e8d9a 6946 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6947}
6948
6949static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6950{
6951 struct list_head *tmp;
6952 loff_t l = *pos;
fd01b88c 6953 struct mddev *mddev;
1da177e4
LT		 6954
6955 if (l >= 0x10000)
6956 return NULL;
6957 if (!l--)
6958 /* header */
6959 return (void*)1;
6960
6961 spin_lock(&all_mddevs_lock);
6962 list_for_each(tmp,&all_mddevs)
6963 if (!l--) {
fd01b88c 6964 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 6965 mddev_get(mddev);
6966 spin_unlock(&all_mddevs_lock);
6967 return mddev;
6968 }
6969 spin_unlock(&all_mddevs_lock);
6970 if (!l--)
6971 return (void*)2;/* tail */
6972 return NULL;
6973}
6974
6975static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6976{
6977 struct list_head *tmp;
fd01b88c 6978 struct mddev *next_mddev, *mddev = v;
1da177e4
LT		 6979
6980 ++*pos;
6981 if (v == (void*)2)
6982 return NULL;
6983
6984 spin_lock(&all_mddevs_lock);
6985 if (v == (void*)1)
6986 tmp = all_mddevs.next;
6987 else
6988 tmp = mddev->all_mddevs.next;
6989 if (tmp != &all_mddevs)
fd01b88c 6990 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 6991 else {
6992 next_mddev = (void*)2;
6993 *pos = 0x10000;
6994 }
6995 spin_unlock(&all_mddevs_lock);
6996
6997 if (v != (void*)1)
6998 mddev_put(mddev);
6999 return next_mddev;
7000
7001}
7002
7003static void md_seq_stop(struct seq_file *seq, void *v)
7004{
fd01b88c 7005 struct mddev *mddev = v;
1da177e4
LT		 7006
7007 if (mddev && v != (void*)1 && v != (void*)2)
7008 mddev_put(mddev);
7009}
7010
7011static int md_seq_show(struct seq_file *seq, void *v)
7012{
fd01b88c 7013 struct mddev *mddev = v;
dd8ac336 7014 sector_t sectors;
3cb03002 7015 struct md_rdev *rdev;
1da177e4
LT		 7016
7017 if (v == (void*)1) {
84fc4b56 7018 struct md_personality *pers;
1da177e4
LT		 7019 seq_printf(seq, "Personalities : ");
7020 spin_lock(&pers_lock);
2604b703
N		 7021 list_for_each_entry(pers, &pers_list, list)
7022 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 7023
7024 spin_unlock(&pers_lock);
7025 seq_printf(seq, "\n");
f1514638 7026 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7027 return 0;
7028 }
7029 if (v == (void*)2) {
7030 status_unused(seq);
7031 return 0;
7032 }
7033
5dc5cf7d 7034 if (mddev_lock(mddev) < 0)
1da177e4 7035 return -EINTR;
5dc5cf7d 7036
1da177e4
LT		 7037 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7038 seq_printf(seq, "%s : %sactive", mdname(mddev),
7039 mddev->pers ? "" : "in");
7040 if (mddev->pers) {
f91de92e 7041 if (mddev->ro==1)
1da177e4 7042 seq_printf(seq, " (read-only)");
f91de92e 7043 if (mddev->ro==2)
52720ae7 7044 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 7045 seq_printf(seq, " %s", mddev->pers->name);
7046 }
7047
dd8ac336 7048 sectors = 0;
dafb20fa 7049 rdev_for_each(rdev, mddev) {
1da177e4
LT		 7050 char b[BDEVNAME_SIZE];
7051 seq_printf(seq, " %s[%d]",
7052 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 7053 if (test_bit(WriteMostly, &rdev->flags))
7054 seq_printf(seq, "(W)");
b2d444d7 7055 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7056 seq_printf(seq, "(F)");
7057 continue;
2d78f8c4
N		 7058 }
7059 if (rdev->raid_disk < 0)
b325a32e 7060 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 7061 if (test_bit(Replacement, &rdev->flags))
7062 seq_printf(seq, "(R)");
dd8ac336 7063 sectors += rdev->sectors;
1da177e4
LT		 7064 }
7065
7066 if (!list_empty(&mddev->disks)) {
7067 if (mddev->pers)
7068 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 7069 (unsigned long long)
7070 mddev->array_sectors / 2);
1da177e4
LT		 7071 else
7072 seq_printf(seq, "\n %llu blocks",
dd8ac336 7073 (unsigned long long)sectors / 2);
1da177e4 7074 }
1cd6bf19
N		 7075 if (mddev->persistent) {
7076 if (mddev->major_version != 0 ||
7077 mddev->minor_version != 90) {
7078 seq_printf(seq," super %d.%d",
7079 mddev->major_version,
7080 mddev->minor_version);
7081 }
e691063a
N		 7082 } else if (mddev->external)
7083 seq_printf(seq, " super external:%s",
7084 mddev->metadata_type);
7085 else
1cd6bf19 7086 seq_printf(seq, " super non-persistent");
1da177e4
LT		 7087
7088 if (mddev->pers) {
d710e138 7089 mddev->pers->status(seq, mddev);
1da177e4 7090 seq_printf(seq, "\n ");
8e1b39d6
N		 7091 if (mddev->pers->sync_request) {
7092 if (mddev->curr_resync > 2) {
d710e138 7093 status_resync(seq, mddev);
8e1b39d6 7094 seq_printf(seq, "\n ");
72f36d59 7095 } else if (mddev->curr_resync >= 1)
8e1b39d6
N		 7096 seq_printf(seq, "\tresync=DELAYED\n ");
7097 else if (mddev->recovery_cp < MaxSector)
7098 seq_printf(seq, "\tresync=PENDING\n ");
7099 }
32a7627c
N		 7100 } else
7101 seq_printf(seq, "\n ");
7102
57148964 7103 bitmap_status(seq, mddev->bitmap);
1da177e4
LT		 7104
7105 seq_printf(seq, "\n");
7106 }
7107 mddev_unlock(mddev);
7108
7109 return 0;
7110}
7111
110518bc 7112static const struct seq_operations md_seq_ops = {
1da177e4
LT		 7113 .start = md_seq_start,
7114 .next = md_seq_next,
7115 .stop = md_seq_stop,
7116 .show = md_seq_show,
7117};
7118
7119static int md_seq_open(struct inode *inode, struct file *file)
7120{
f1514638 7121 struct seq_file *seq;
1da177e4
LT		 7122 int error;
7123
7124 error = seq_open(file, &md_seq_ops);
d7603b7e 7125 if (error)
f1514638
KS		 7126 return error;
7127
7128 seq = file->private_data;
7129 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 7130 return error;
7131}
7132
d7603b7e
N		 7133static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7134{
f1514638 7135 struct seq_file *seq = filp->private_data;
d7603b7e
N		 7136 int mask;
7137
7138 poll_wait(filp, &md_event_waiters, wait);
7139
7140 /* always allow read */
7141 mask = POLLIN | POLLRDNORM;
7142
f1514638 7143 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 7144 mask |= POLLERR | POLLPRI;
7145 return mask;
7146}
7147
fa027c2a 7148static const struct file_operations md_seq_fops = {
e24650c2 7149 .owner = THIS_MODULE,
1da177e4
LT		 7150 .open = md_seq_open,
7151 .read = seq_read,
7152 .llseek = seq_lseek,
c3f94b40 7153 .release = seq_release_private,
d7603b7e 7154 .poll = mdstat_poll,
1da177e4
LT		 7155};
7156
84fc4b56 7157int register_md_personality(struct md_personality *p)
1da177e4 7158{
1da177e4 7159 spin_lock(&pers_lock);
2604b703
N		 7160 list_add_tail(&p->list, &pers_list);
7161 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 7162 spin_unlock(&pers_lock);
7163 return 0;
7164}
7165
84fc4b56 7166int unregister_md_personality(struct md_personality *p)
1da177e4 7167{
2604b703 7168 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7169 spin_lock(&pers_lock);
2604b703 7170 list_del_init(&p->list);
1da177e4
LT		 7171 spin_unlock(&pers_lock);
7172 return 0;
7173}
7174
fd01b88c 7175static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7176{
3cb03002 7177 struct md_rdev * rdev;
1da177e4 7178 int idle;
eea1bf38 7179 int curr_events;
1da177e4
LT		 7180
7181 idle = 1;
4b80991c
N		 7182 rcu_read_lock();
7183 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7184 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 7185 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7186 (int)part_stat_read(&disk->part0, sectors[1]) -
7187 atomic_read(&disk->sync_io);
713f6ab1
N		 7188 /* sync IO will cause sync_io to increase before the disk_stats
7189 * as sync_io is counted when a request starts, and
7190 * disk_stats is counted when it completes.
7191 * So resync activity will cause curr_events to be smaller than
7192 * when there was no such activity.
7193 * non-sync IO will cause disk_stat to increase without
7194 * increasing sync_io so curr_events will (eventually)
7195 * be larger than it was before. Once it becomes
7196 * substantially larger, the test below will cause
7197 * the array to appear non-idle, and resync will slow
7198 * down.
7199 * If there is a lot of outstanding resync activity when
7200 * we set last_event to curr_events, then all that activity
7201 * completing might cause the array to appear non-idle
7202 * and resync will be slowed down even though there might
7203 * not have been non-resync activity. This will only
7204 * happen once though. 'last_events' will soon reflect
7205 * the state where there is little or no outstanding
7206 * resync requests, and further resync activity will
7207 * always make curr_events less than last_events.
c0e48521 7208 *
1da177e4 7209 */
eea1bf38 7210 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 7211 rdev->last_events = curr_events;
7212 idle = 0;
7213 }
7214 }
4b80991c 7215 rcu_read_unlock();
1da177e4
LT		 7216 return idle;
7217}
7218
fd01b88c 7219void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 7220{
7221 /* another "blocks" (512byte) blocks have been synced */
7222 atomic_sub(blocks, &mddev->recovery_active);
7223 wake_up(&mddev->recovery_wait);
7224 if (!ok) {
dfc70645 7225 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 7226 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT		 7227 md_wakeup_thread(mddev->thread);
7228 // stop recovery, signal do_sync ....
7229 }
7230}
7231
7232
06d91a5f
N		 7233/* md_write_start(mddev, bi)
7234 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 7235 * in superblock) before writing, schedule a superblock update
7236 * and wait for it to complete.
06d91a5f 7237 */
fd01b88c 7238void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7239{
0fd62b86 7240 int did_change = 0;
06d91a5f 7241 if (bio_data_dir(bi) != WRITE)
3d310eb7 7242 return;
06d91a5f 7243
f91de92e
N		 7244 BUG_ON(mddev->ro == 1);
7245 if (mddev->ro == 2) {
7246 /* need to switch to read/write */
7247 mddev->ro = 0;
7248 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7249 md_wakeup_thread(mddev->thread);
25156198 7250 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7251 did_change = 1;
f91de92e 7252 }
06d91a5f 7253 atomic_inc(&mddev->writes_pending);
31a59e34
N		 7254 if (mddev->safemode == 1)
7255 mddev->safemode = 0;
06d91a5f 7256 if (mddev->in_sync) {
a9701a30 7257 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 7258 if (mddev->in_sync) {
7259 mddev->in_sync = 0;
850b2b42 7260 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7261 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7262 md_wakeup_thread(mddev->thread);
0fd62b86 7263 did_change = 1;
3d310eb7 7264 }
a9701a30 7265 spin_unlock_irq(&mddev->write_lock);
06d91a5f 7266 }
0fd62b86 7267 if (did_change)
00bcb4ac 7268 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7269 wait_event(mddev->sb_wait,
09a44cc1 7270 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 7271}
7272
fd01b88c 7273void md_write_end(struct mddev *mddev)
1da177e4
LT		 7274{
7275 if (atomic_dec_and_test(&mddev->writes_pending)) {
7276 if (mddev->safemode == 2)
7277 md_wakeup_thread(mddev->thread);
16f17b39 7278 else if (mddev->safemode_delay)
1da177e4
LT		 7279 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7280 }
7281}
7282
2a2275d6
N		 7283/* md_allow_write(mddev)
7284 * Calling this ensures that the array is marked 'active' so that writes
7285 * may proceed without blocking. It is important to call this before
7286 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7287 * Must be called with mddev_lock held.
b5470dc5
DW		 7288 *
7289 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7290 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7291 */
fd01b88c 7292int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7293{
7294 if (!mddev->pers)
b5470dc5 7295 return 0;
2a2275d6 7296 if (mddev->ro)
b5470dc5 7297 return 0;
1a0fd497 7298 if (!mddev->pers->sync_request)
b5470dc5 7299 return 0;
2a2275d6
N		 7300
7301 spin_lock_irq(&mddev->write_lock);
7302 if (mddev->in_sync) {
7303 mddev->in_sync = 0;
7304 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7305 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7306 if (mddev->safemode_delay &&
7307 mddev->safemode == 0)
7308 mddev->safemode = 1;
7309 spin_unlock_irq(&mddev->write_lock);
7310 md_update_sb(mddev, 0);
00bcb4ac 7311 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 7312 } else
7313 spin_unlock_irq(&mddev->write_lock);
b5470dc5 7314
070dc6dd 7315 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7316 return -EAGAIN;
7317 else
7318 return 0;
2a2275d6
N		 7319}
7320EXPORT_SYMBOL_GPL(md_allow_write);
7321
1da177e4
LT		 7322#define SYNC_MARKS 10
7323#define SYNC_MARK_STEP (3*HZ)
54f89341 7324#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 7325void md_do_sync(struct md_thread *thread)
1da177e4 7326{
4ed8731d 7327 struct mddev *mddev = thread->mddev;
fd01b88c 7328 struct mddev *mddev2;
1da177e4
LT		 7329 unsigned int currspeed = 0,
7330 window;
57afd89f 7331 sector_t max_sectors,j, io_sectors;
1da177e4 7332 unsigned long mark[SYNC_MARKS];
54f89341 7333 unsigned long update_time;
1da177e4
LT		 7334 sector_t mark_cnt[SYNC_MARKS];
7335 int last_mark,m;
7336 struct list_head *tmp;
7337 sector_t last_check;
57afd89f 7338 int skipped = 0;
3cb03002 7339 struct md_rdev *rdev;
61df9d91 7340 char *desc;
7c2c57c9 7341 struct blk_plug plug;
1da177e4
LT		 7342
7343 /* just incase thread restarts... */
7344 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7345 return;
5fd6c1dc
N		 7346 if (mddev->ro) /* never try to sync a read-only array */
7347 return;
1da177e4 7348
61df9d91
N		 7349 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7350 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7351 desc = "data-check";
7352 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7353 desc = "requested-resync";
7354 else
7355 desc = "resync";
7356 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7357 desc = "reshape";
7358 else
7359 desc = "recovery";
7360
1da177e4
LT		 7361 /* we overload curr_resync somewhat here.
7362 * 0 == not engaged in resync at all
7363 * 2 == checking that there is no conflict with another sync
7364 * 1 == like 2, but have yielded to allow conflicting resync to
7365 * commense
7366 * other == active in resync - this many blocks
7367 *
7368 * Before starting a resync we must have set curr_resync to
7369 * 2, and then checked that every "conflicting" array has curr_resync
7370 * less than ours. When we find one that is the same or higher
7371 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7372 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7373 * This will mean we have to start checking from the beginning again.
7374 *
7375 */
7376
7377 do {
7378 mddev->curr_resync = 2;
7379
7380 try_again:
404e4b43 7381 if (kthread_should_stop())
6985c43f 7382 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 7383
7384 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7385 goto skip;
29ac4aa3 7386 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7387 if (mddev2 == mddev)
7388 continue;
90b08710
BS		 7389 if (!mddev->parallel_resync
7390 && mddev2->curr_resync
7391 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7392 DEFINE_WAIT(wq);
7393 if (mddev < mddev2 && mddev->curr_resync == 2) {
7394 /* arbitrarily yield */
7395 mddev->curr_resync = 1;
7396 wake_up(&resync_wait);
7397 }
7398 if (mddev > mddev2 && mddev->curr_resync == 1)
7399 /* no need to wait here, we can wait the next
7400 * time 'round when curr_resync == 2
7401 */
7402 continue;
9744197c
N		 7403 /* We need to wait 'interruptible' so as not to
7404 * contribute to the load average, and not to
7405 * be caught by 'softlockup'
7406 */
7407 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 7408 if (!kthread_should_stop() &&
8712e553 7409 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7410 printk(KERN_INFO "md: delaying %s of %s"
7411 " until %s has finished (they"
1da177e4 7412 " share one or more physical units)\n",
61df9d91 7413 desc, mdname(mddev), mdname(mddev2));
1da177e4 7414 mddev_put(mddev2);
9744197c
N		 7415 if (signal_pending(current))
7416 flush_signals(current);
1da177e4
LT		 7417 schedule();
7418 finish_wait(&resync_wait, &wq);
7419 goto try_again;
7420 }
7421 finish_wait(&resync_wait, &wq);
7422 }
7423 }
7424 } while (mddev->curr_resync < 2);
7425
5fd6c1dc 7426 j = 0;
9d88883e 7427 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7428 /* resync follows the size requested by the personality,
57afd89f 7429 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7430 */
7431 max_sectors = mddev->resync_max_sectors;
7f7583d4 7432 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 7433 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7434 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7435 j = mddev->resync_min;
7436 else if (!mddev->bitmap)
5fd6c1dc 7437 j = mddev->recovery_cp;
5e96ee65 7438
ccfcc3c1 7439 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7440 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7441 else {
1da177e4 7442 /* recovery follows the physical size of devices */
58c0fed4 7443 max_sectors = mddev->dev_sectors;
5fd6c1dc 7444 j = MaxSector;
4e59ca7d 7445 rcu_read_lock();
dafb20fa 7446 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc
N		 7447 if (rdev->raid_disk >= 0 &&
7448 !test_bit(Faulty, &rdev->flags) &&
7449 !test_bit(In_sync, &rdev->flags) &&
7450 rdev->recovery_offset < j)
7451 j = rdev->recovery_offset;
4e59ca7d 7452 rcu_read_unlock();
5fd6c1dc 7453 }
1da177e4 7454
61df9d91
N		 7455 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7456 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7457 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7458 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7459 "(but not more than %d KB/sec) for %s.\n",
7460 speed_max(mddev), desc);
1da177e4 7461
eea1bf38 7462 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7463
57afd89f 7464 io_sectors = 0;
1da177e4
LT		 7465 for (m = 0; m < SYNC_MARKS; m++) {
7466 mark[m] = jiffies;
57afd89f 7467 mark_cnt[m] = io_sectors;
1da177e4
LT		 7468 }
7469 last_mark = 0;
7470 mddev->resync_mark = mark[last_mark];
7471 mddev->resync_mark_cnt = mark_cnt[last_mark];
7472
7473 /*
7474 * Tune reconstruction:
7475 */
7476 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7477 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7478 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7479
7480 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7481 last_check = 0;
7482
7483 if (j>2) {
7484 printk(KERN_INFO
61df9d91
N		 7485 "md: resuming %s of %s from checkpoint.\n",
7486 desc, mdname(mddev));
1da177e4 7487 mddev->curr_resync = j;
72f36d59
N		 7488 } else
7489 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 7490 mddev->curr_resync_completed = j;
72f36d59
N		 7491 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7492 md_new_event(mddev);
54f89341 7493 update_time = jiffies;
1da177e4 7494
7c2c57c9 7495 blk_start_plug(&plug);
1da177e4 7496 while (j < max_sectors) {
57afd89f 7497 sector_t sectors;
1da177e4 7498
57afd89f 7499 skipped = 0;
97e4f42d 7500
7a91ee1f
N		 7501 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7502 ((mddev->curr_resync > mddev->curr_resync_completed &&
7503 (mddev->curr_resync - mddev->curr_resync_completed)
7504 > (max_sectors >> 4)) ||
54f89341 7505 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f
N		 7506 (j - mddev->curr_resync_completed)*2
7507 >= mddev->resync_max - mddev->curr_resync_completed
7508 )) {
97e4f42d 7509 /* time to update curr_resync_completed */
97e4f42d
N		 7510 wait_event(mddev->recovery_wait,
7511 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7512 mddev->curr_resync_completed = j;
35d78c66 7513 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7514 j > mddev->recovery_cp)
7515 mddev->recovery_cp = j;
54f89341 7516 update_time = jiffies;
070dc6dd 7517 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7518 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7519 }
acb180b0 7520
e62e58a5
N		 7521 while (j >= mddev->resync_max && !kthread_should_stop()) {
7522 /* As this condition is controlled by user-space,
7523 * we can block indefinitely, so use '_interruptible'
7524 * to avoid triggering warnings.
7525 */
7526 flush_signals(current); /* just in case */
7527 wait_event_interruptible(mddev->recovery_wait,
7528 mddev->resync_max > j
7529 || kthread_should_stop());
7530 }
acb180b0
N		 7531
7532 if (kthread_should_stop())
7533 goto interrupted;
7534
57afd89f 7535 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7536 currspeed < speed_min(mddev));
57afd89f 7537 if (sectors == 0) {
dfc70645 7538 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 7539 goto out;
7540 }
57afd89f
N		 7541
7542 if (!skipped) { /* actual IO requested */
7543 io_sectors += sectors;
7544 atomic_add(sectors, &mddev->recovery_active);
7545 }
7546
e875ecea
N		 7547 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7548 break;
7549
1da177e4 7550 j += sectors;
72f36d59
N		 7551 if (j > 2)
7552 mddev->curr_resync = j;
ff4e8d9a 7553 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7554 if (last_check == 0)
e875ecea 7555 /* this is the earliest that rebuild will be
d7603b7e
N		 7556 * visible in /proc/mdstat
7557 */
7558 md_new_event(mddev);
57afd89f
N		 7559
7560 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7561 continue;
7562
57afd89f 7563 last_check = io_sectors;
1da177e4
LT		 7564 repeat:
7565 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7566 /* step marks */
7567 int next = (last_mark+1) % SYNC_MARKS;
7568
7569 mddev->resync_mark = mark[next];
7570 mddev->resync_mark_cnt = mark_cnt[next];
7571 mark[next] = jiffies;
57afd89f 7572 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7573 last_mark = next;
7574 }
7575
7576
c6207277
N		 7577 if (kthread_should_stop())
7578 goto interrupted;
7579
1da177e4
LT		 7580
7581 /*
7582 * this loop exits only if either when we are slower than
7583 * the 'hard' speed limit, or the system was IO-idle for
7584 * a jiffy.
7585 * the system might be non-idle CPU-wise, but we only care
7586 * about not overloading the IO subsystem. (things like an
7587 * e2fsck being done on the RAID array should execute fast)
7588 */
1da177e4
LT		 7589 cond_resched();
7590
57afd89f
N		 7591 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7592 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7593
88202a0c
N		 7594 if (currspeed > speed_min(mddev)) {
7595 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7596 !is_mddev_idle(mddev, 0)) {
c0e48521 7597 msleep(500);
1da177e4
LT		 7598 goto repeat;
7599 }
7600 }
7601 }
61df9d91 7602 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 7603 /*
7604 * this also signals 'finished resyncing' to md_stop
7605 */
7606 out:
7c2c57c9 7607 blk_finish_plug(&plug);
1da177e4
LT		 7608 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7609
7610 /* tell personality that we are finished */
57afd89f 7611 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7612
dfc70645 7613 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7614 mddev->curr_resync > 2) {
7615 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7616 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7617 if (mddev->curr_resync >= mddev->recovery_cp) {
7618 printk(KERN_INFO
61df9d91
N		 7619 "md: checkpointing %s of %s.\n",
7620 desc, mdname(mddev));
0a19caab 7621 if (test_bit(MD_RECOVERY_ERROR,
7622 &mddev->recovery))
7623 mddev->recovery_cp =
7624 mddev->curr_resync_completed;
7625 else
7626 mddev->recovery_cp =
7627 mddev->curr_resync;
5fd6c1dc
N		 7628 }
7629 } else
7630 mddev->recovery_cp = MaxSector;
7631 } else {
7632 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7633 mddev->curr_resync = MaxSector;
4e59ca7d 7634 rcu_read_lock();
dafb20fa 7635 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7636 if (rdev->raid_disk >= 0 &&
70fffd0b 7637 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7638 !test_bit(Faulty, &rdev->flags) &&
7639 !test_bit(In_sync, &rdev->flags) &&
7640 rdev->recovery_offset < mddev->curr_resync)
7641 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7642 rcu_read_unlock();
5fd6c1dc 7643 }
1da177e4 7644 }
db91ff55 7645 skip:
17571284 7646 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7647
c07b70ad
N		 7648 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7649 /* We completed so min/max setting can be forgotten if used. */
7650 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7651 mddev->resync_min = 0;
7652 mddev->resync_max = MaxSector;
7653 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7654 mddev->resync_min = mddev->curr_resync_completed;
1da177e4
LT		 7655 mddev->curr_resync = 0;
7656 wake_up(&resync_wait);
7657 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7658 md_wakeup_thread(mddev->thread);
c6207277
N		 7659 return;
7660
7661 interrupted:
7662 /*
7663 * got a signal, exit.
7664 */
7665 printk(KERN_INFO
7666 "md: md_do_sync() got signal ... exiting\n");
7667 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7668 goto out;
7669
1da177e4 7670}
29269553 7671EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7672
746d3207
N		 7673static int remove_and_add_spares(struct mddev *mddev,
7674 struct md_rdev *this)
b4c4c7b8 7675{
3cb03002 7676 struct md_rdev *rdev;
b4c4c7b8 7677 int spares = 0;
f2a371c5 7678 int removed = 0;
b4c4c7b8 7679
dafb20fa 7680 rdev_for_each(rdev, mddev)
746d3207
N		 7681 if ((this == NULL || rdev == this) &&
7682 rdev->raid_disk >= 0 &&
6bfe0b49 7683 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7684 (test_bit(Faulty, &rdev->flags) ||
7685 ! test_bit(In_sync, &rdev->flags)) &&
7686 atomic_read(&rdev->nr_pending)==0) {
7687 if (mddev->pers->hot_remove_disk(
b8321b68 7688 mddev, rdev) == 0) {
36fad858 7689 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 7690 rdev->raid_disk = -1;
f2a371c5 7691 removed++;
b4c4c7b8
N		 7692 }
7693 }
90584fc9
JB		 7694 if (removed && mddev->kobj.sd)
7695 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 7696
746d3207
N		 7697 if (this)
7698 goto no_add;
7699
dafb20fa 7700 rdev_for_each(rdev, mddev) {
7bfec5f3
N		 7701 if (rdev->raid_disk >= 0 &&
7702 !test_bit(In_sync, &rdev->flags) &&
7703 !test_bit(Faulty, &rdev->flags))
7704 spares++;
7ceb17e8
N		 7705 if (rdev->raid_disk >= 0)
7706 continue;
7707 if (test_bit(Faulty, &rdev->flags))
7708 continue;
7709 if (mddev->ro &&
7710 rdev->saved_raid_disk < 0)
7711 continue;
7712
7713 rdev->recovery_offset = 0;
7714 if (rdev->saved_raid_disk >= 0 && mddev->in_sync) {
7715 spin_lock_irq(&mddev->write_lock);
7716 if (mddev->in_sync)
7717 /* OK, this device, which is in_sync,
7718 * will definitely be noticed before
7719 * the next write, so recovery isn't
7720 * needed.
7721 */
7722 rdev->recovery_offset = mddev->recovery_cp;
7723 spin_unlock_irq(&mddev->write_lock);
7724 }
7725 if (mddev->ro && rdev->recovery_offset != MaxSector)
7726 /* not safe to add this disk now */
7727 continue;
7728 if (mddev->pers->
7729 hot_add_disk(mddev, rdev) == 0) {
7730 if (sysfs_link_rdev(mddev, rdev))
7731 /* failure here is OK */;
7732 spares++;
7733 md_new_event(mddev);
7734 set_bit(MD_CHANGE_DEVS, &mddev->flags);
dfc70645 7735 }
b4c4c7b8 7736 }
746d3207 7737no_add:
6dafab6b
N		 7738 if (removed)
7739 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 7740 return spares;
7741}
7ebc0be7 7742
1da177e4
LT		 7743/*
7744 * This routine is regularly called by all per-raid-array threads to
7745 * deal with generic issues like resync and super-block update.
7746 * Raid personalities that don't have a thread (linear/raid0) do not
7747 * need this as they never do any recovery or update the superblock.
7748 *
7749 * It does not do any resync itself, but rather "forks" off other threads
7750 * to do that as needed.
7751 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7752 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7753 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7754 * and wakeups up this thread which will reap the thread and finish up.
7755 * This thread also removes any faulty devices (with nr_pending == 0).
7756 *
7757 * The overall approach is:
7758 * 1/ if the superblock needs updating, update it.
7759 * 2/ If a recovery thread is running, don't do anything else.
7760 * 3/ If recovery has finished, clean up, possibly marking spares active.
7761 * 4/ If there are any faulty devices, remove them.
7762 * 5/ If array is degraded, try to add spares devices
7763 * 6/ If array has spares or is not in-sync, start a resync thread.
7764 */
fd01b88c 7765void md_check_recovery(struct mddev *mddev)
1da177e4 7766{
68866e42
JB		 7767 if (mddev->suspended)
7768 return;
7769
5f40402d 7770 if (mddev->bitmap)
aa5cbd10 7771 bitmap_daemon_work(mddev);
1da177e4 7772
fca4d848 7773 if (signal_pending(current)) {
31a59e34 7774 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7775 printk(KERN_INFO "md: %s in immediate safe mode\n",
7776 mdname(mddev));
7777 mddev->safemode = 2;
7778 }
7779 flush_signals(current);
7780 }
7781
c89a8eee
N		 7782 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7783 return;
1da177e4 7784 if ( ! (
126925c0 7785 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7786 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7787 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7788 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7789 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7790 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7791 ))
7792 return;
fca4d848 7793
df5b89b3 7794 if (mddev_trylock(mddev)) {
b4c4c7b8 7795 int spares = 0;
fca4d848 7796
c89a8eee 7797 if (mddev->ro) {
7ceb17e8
N		 7798 /* On a read-only array we can:
7799 * - remove failed devices
7800 * - add already-in_sync devices if the array itself
7801 * is in-sync.
7802 * As we only add devices that are already in-sync,
7803 * we can activate the spares immediately.
c89a8eee 7804 */
c89a8eee 7805 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7ceb17e8
N		 7806 remove_and_add_spares(mddev, NULL);
7807 mddev->pers->spare_active(mddev);
c89a8eee
N		 7808 goto unlock;
7809 }
7810
31a59e34 7811 if (!mddev->external) {
0fd62b86 7812 int did_change = 0;
31a59e34
N		 7813 spin_lock_irq(&mddev->write_lock);
7814 if (mddev->safemode &&
7815 !atomic_read(&mddev->writes_pending) &&
7816 !mddev->in_sync &&
7817 mddev->recovery_cp == MaxSector) {
7818 mddev->in_sync = 1;
0fd62b86 7819 did_change = 1;
070dc6dd 7820 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7821 }
7822 if (mddev->safemode == 1)
7823 mddev->safemode = 0;
7824 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7825 if (did_change)
00bcb4ac 7826 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7827 }
fca4d848 7828
850b2b42
N		 7829 if (mddev->flags)
7830 md_update_sb(mddev, 0);
06d91a5f 7831
1da177e4
LT		 7832 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7833 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7834 /* resync/recovery still happening */
7835 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7836 goto unlock;
7837 }
7838 if (mddev->sync_thread) {
a91d5ac0 7839 md_reap_sync_thread(mddev);
1da177e4
LT		 7840 goto unlock;
7841 }
72a23c21
NB		 7842 /* Set RUNNING before clearing NEEDED to avoid
7843 * any transients in the value of "sync_action".
7844 */
72f36d59 7845 mddev->curr_resync_completed = 0;
72a23c21 7846 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
24dd469d
N		 7847 /* Clear some bits that don't mean anything, but
7848 * might be left set
7849 */
24dd469d
N		 7850 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7851 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7852
ed209584
N		 7853 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7854 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5fd6c1dc 7855 goto unlock;
1da177e4
LT		 7856 /* no recovery is running.
7857 * remove any failed drives, then
7858 * add spares if possible.
72f36d59 7859 * Spares are also removed and re-added, to allow
1da177e4
LT		 7860 * the personality to fail the re-add.
7861 */
1da177e4 7862
b4c4c7b8 7863 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7864 if (mddev->pers->check_reshape == NULL ||
7865 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7866 /* Cannot proceed */
7867 goto unlock;
7868 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7869 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 7870 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N		 7871 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7872 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7873 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7874 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7875 } else if (mddev->recovery_cp < MaxSector) {
7876 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7877 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7878 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7879 /* nothing to be done ... */
1da177e4 7880 goto unlock;
24dd469d 7881
1da177e4 7882 if (mddev->pers->sync_request) {
ef99bf48 7883 if (spares) {
a654b9d8
N		 7884 /* We are adding a device or devices to an array
7885 * which has the bitmap stored on all devices.
7886 * So make sure all bitmap pages get written
7887 */
7888 bitmap_write_all(mddev->bitmap);
7889 }
1da177e4
LT		 7890 mddev->sync_thread = md_register_thread(md_do_sync,
7891 mddev,
0da3c619 7892 "resync");
1da177e4
LT		 7893 if (!mddev->sync_thread) {
7894 printk(KERN_ERR "%s: could not start resync"
7895 " thread...\n",
7896 mdname(mddev));
7897 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7898 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7899 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7900 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7901 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7902 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7903 } else
1da177e4 7904 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7905 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7906 md_new_event(mddev);
1da177e4
LT		 7907 }
7908 unlock:
72a23c21
NB		 7909 if (!mddev->sync_thread) {
7910 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7911 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7912 &mddev->recovery))
0c3573f1 7913 if (mddev->sysfs_action)
00bcb4ac 7914 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7915 }
1da177e4
LT		 7916 mddev_unlock(mddev);
7917 }
7918}
7919
a91d5ac0
JB		 7920void md_reap_sync_thread(struct mddev *mddev)
7921{
7922 struct md_rdev *rdev;
7923
7924 /* resync has finished, collect result */
7925 md_unregister_thread(&mddev->sync_thread);
7926 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7927 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7928 /* success...*/
7929 /* activate any spares */
7930 if (mddev->pers->spare_active(mddev)) {
7931 sysfs_notify(&mddev->kobj, NULL,
7932 "degraded");
7933 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7934 }
7935 }
7936 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7937 mddev->pers->finish_reshape)
7938 mddev->pers->finish_reshape(mddev);
7939
7940 /* If array is no-longer degraded, then any saved_raid_disk
7941 * information must be scrapped. Also if any device is now
7942 * In_sync we must scrape the saved_raid_disk for that device
7943 * do the superblock for an incrementally recovered device
7944 * written out.
7945 */
7946 rdev_for_each(rdev, mddev)
7947 if (!mddev->degraded ||
7948 test_bit(In_sync, &rdev->flags))
7949 rdev->saved_raid_disk = -1;
7950
7951 md_update_sb(mddev, 1);
7952 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7953 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7954 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7955 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7956 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7957 /* flag recovery needed just to double check */
7958 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7959 sysfs_notify_dirent_safe(mddev->sysfs_action);
7960 md_new_event(mddev);
7961 if (mddev->event_work.func)
7962 queue_work(md_misc_wq, &mddev->event_work);
7963}
7964
fd01b88c 7965void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 7966{
00bcb4ac 7967 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 7968 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 7969 !test_bit(Blocked, &rdev->flags) &&
7970 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 7971 msecs_to_jiffies(5000));
7972 rdev_dec_pending(rdev, mddev);
7973}
7974EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7975
c6563a8c
N		 7976void md_finish_reshape(struct mddev *mddev)
7977{
7978 /* called be personality module when reshape completes. */
7979 struct md_rdev *rdev;
7980
7981 rdev_for_each(rdev, mddev) {
7982 if (rdev->data_offset > rdev->new_data_offset)
7983 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7984 else
7985 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7986 rdev->data_offset = rdev->new_data_offset;
7987 }
7988}
7989EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N		 7990
7991/* Bad block management.
7992 * We can record which blocks on each device are 'bad' and so just
7993 * fail those blocks, or that stripe, rather than the whole device.
7994 * Entries in the bad-block table are 64bits wide. This comprises:
7995 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7996 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7997 * A 'shift' can be set so that larger blocks are tracked and
7998 * consequently larger devices can be covered.
7999 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8000 *
8001 * Locking of the bad-block table uses a seqlock so md_is_badblock
8002 * might need to retry if it is very unlucky.
8003 * We will sometimes want to check for bad blocks in a bi_end_io function,
8004 * so we use the write_seqlock_irq variant.
8005 *
8006 * When looking for a bad block we specify a range and want to
8007 * know if any block in the range is bad. So we binary-search
8008 * to the last range that starts at-or-before the given endpoint,
8009 * (or "before the sector after the target range")
8010 * then see if it ends after the given start.
8011 * We return
8012 * 0 if there are no known bad blocks in the range
8013 * 1 if there are known bad block which are all acknowledged
8014 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8015 * plus the start/length of the first bad section we overlap.
8016 */
8017int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8018 sector_t *first_bad, int *bad_sectors)
8019{
8020 int hi;
ab05613a 8021 int lo;
2230dfe4 8022 u64 *p = bb->page;
ab05613a 8023 int rv;
2230dfe4
N		 8024 sector_t target = s + sectors;
8025 unsigned seq;
8026
8027 if (bb->shift > 0) {
8028 /* round the start down, and the end up */
8029 s >>= bb->shift;
8030 target += (1<<bb->shift) - 1;
8031 target >>= bb->shift;
8032 sectors = target - s;
8033 }
8034 /* 'target' is now the first block after the bad range */
8035
8036retry:
8037 seq = read_seqbegin(&bb->lock);
ab05613a 8038 lo = 0;
8039 rv = 0;
2230dfe4
N		 8040 hi = bb->count;
8041
8042 /* Binary search between lo and hi for 'target'
8043 * i.e. for the last range that starts before 'target'
8044 */
8045 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8046 * are known not to be the last range before target.
8047 * VARIANT: hi-lo is the number of possible
8048 * ranges, and decreases until it reaches 1
8049 */
8050 while (hi - lo > 1) {
8051 int mid = (lo + hi) / 2;
8052 sector_t a = BB_OFFSET(p[mid]);
8053 if (a < target)
8054 /* This could still be the one, earlier ranges
8055 * could not. */
8056 lo = mid;
8057 else
8058 /* This and later ranges are definitely out. */
8059 hi = mid;
8060 }
8061 /* 'lo' might be the last that started before target, but 'hi' isn't */
8062 if (hi > lo) {
8063 /* need to check all range that end after 's' to see if
8064 * any are unacknowledged.
8065 */
8066 while (lo >= 0 &&
8067 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8068 if (BB_OFFSET(p[lo]) < target) {
8069 /* starts before the end, and finishes after
8070 * the start, so they must overlap
8071 */
8072 if (rv != -1 && BB_ACK(p[lo]))
8073 rv = 1;
8074 else
8075 rv = -1;
8076 *first_bad = BB_OFFSET(p[lo]);
8077 *bad_sectors = BB_LEN(p[lo]);
8078 }
8079 lo--;
8080 }
8081 }
8082
8083 if (read_seqretry(&bb->lock, seq))
8084 goto retry;
8085
8086 return rv;
8087}
8088EXPORT_SYMBOL_GPL(md_is_badblock);
8089
8090/*
8091 * Add a range of bad blocks to the table.
8092 * This might extend the table, or might contract it
8093 * if two adjacent ranges can be merged.
8094 * We binary-search to find the 'insertion' point, then
8095 * decide how best to handle it.
8096 */
8097static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8098 int acknowledged)
8099{
8100 u64 *p;
8101 int lo, hi;
8102 int rv = 1;
8103
8104 if (bb->shift < 0)
8105 /* badblocks are disabled */
8106 return 0;
8107
8108 if (bb->shift) {
8109 /* round the start down, and the end up */
8110 sector_t next = s + sectors;
8111 s >>= bb->shift;
8112 next += (1<<bb->shift) - 1;
8113 next >>= bb->shift;
8114 sectors = next - s;
8115 }
8116
8117 write_seqlock_irq(&bb->lock);
8118
8119 p = bb->page;
8120 lo = 0;
8121 hi = bb->count;
8122 /* Find the last range that starts at-or-before 's' */
8123 while (hi - lo > 1) {
8124 int mid = (lo + hi) / 2;
8125 sector_t a = BB_OFFSET(p[mid]);
8126 if (a <= s)
8127 lo = mid;
8128 else
8129 hi = mid;
8130 }
8131 if (hi > lo && BB_OFFSET(p[lo]) > s)
8132 hi = lo;
8133
8134 if (hi > lo) {
8135 /* we found a range that might merge with the start
8136 * of our new range
8137 */
8138 sector_t a = BB_OFFSET(p[lo]);
8139 sector_t e = a + BB_LEN(p[lo]);
8140 int ack = BB_ACK(p[lo]);
8141 if (e >= s) {
8142 /* Yes, we can merge with a previous range */
8143 if (s == a && s + sectors >= e)
8144 /* new range covers old */
8145 ack = acknowledged;
8146 else
8147 ack = ack && acknowledged;
8148
8149 if (e < s + sectors)
8150 e = s + sectors;
8151 if (e - a <= BB_MAX_LEN) {
8152 p[lo] = BB_MAKE(a, e-a, ack);
8153 s = e;
8154 } else {
8155 /* does not all fit in one range,
8156 * make p[lo] maximal
8157 */
8158 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8159 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8160 s = a + BB_MAX_LEN;
8161 }
8162 sectors = e - s;
8163 }
8164 }
8165 if (sectors && hi < bb->count) {
8166 /* 'hi' points to the first range that starts after 's'.
8167 * Maybe we can merge with the start of that range */
8168 sector_t a = BB_OFFSET(p[hi]);
8169 sector_t e = a + BB_LEN(p[hi]);
8170 int ack = BB_ACK(p[hi]);
8171 if (a <= s + sectors) {
8172 /* merging is possible */
8173 if (e <= s + sectors) {
8174 /* full overlap */
8175 e = s + sectors;
8176 ack = acknowledged;
8177 } else
8178 ack = ack && acknowledged;
8179
8180 a = s;
8181 if (e - a <= BB_MAX_LEN) {
8182 p[hi] = BB_MAKE(a, e-a, ack);
8183 s = e;
8184 } else {
8185 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8186 s = a + BB_MAX_LEN;
8187 }
8188 sectors = e - s;
8189 lo = hi;
8190 hi++;
8191 }
8192 }
8193 if (sectors == 0 && hi < bb->count) {
8194 /* we might be able to combine lo and hi */
8195 /* Note: 's' is at the end of 'lo' */
8196 sector_t a = BB_OFFSET(p[hi]);
8197 int lolen = BB_LEN(p[lo]);
8198 int hilen = BB_LEN(p[hi]);
8199 int newlen = lolen + hilen - (s - a);
8200 if (s >= a && newlen < BB_MAX_LEN) {
8201 /* yes, we can combine them */
8202 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8203 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8204 memmove(p + hi, p + hi + 1,
8205 (bb->count - hi - 1) * 8);
8206 bb->count--;
8207 }
8208 }
8209 while (sectors) {
8210 /* didn't merge (it all).
8211 * Need to add a range just before 'hi' */
8212 if (bb->count >= MD_MAX_BADBLOCKS) {
8213 /* No room for more */
8214 rv = 0;
8215 break;
8216 } else {
8217 int this_sectors = sectors;
8218 memmove(p + hi + 1, p + hi,
8219 (bb->count - hi) * 8);
8220 bb->count++;
8221
8222 if (this_sectors > BB_MAX_LEN)
8223 this_sectors = BB_MAX_LEN;
8224 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8225 sectors -= this_sectors;
8226 s += this_sectors;
8227 }
8228 }
8229
8230 bb->changed = 1;
de393cde
N		 8231 if (!acknowledged)
8232 bb->unacked_exist = 1;
2230dfe4
N		 8233 write_sequnlock_irq(&bb->lock);
8234
8235 return rv;
8236}
8237
3cb03002 8238int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8239 int is_new)
2230dfe4 8240{
c6563a8c
N		 8241 int rv;
8242 if (is_new)
8243 s += rdev->new_data_offset;
8244 else
8245 s += rdev->data_offset;
8246 rv = md_set_badblocks(&rdev->badblocks,
8247 s, sectors, 0);
2230dfe4
N		 8248 if (rv) {
8249 /* Make sure they get written out promptly */
8bd2f0a0 8250 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 8251 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8252 md_wakeup_thread(rdev->mddev->thread);
8253 }
8254 return rv;
8255}
8256EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8257
8258/*
8259 * Remove a range of bad blocks from the table.
8260 * This may involve extending the table if we spilt a region,
8261 * but it must not fail. So if the table becomes full, we just
8262 * drop the remove request.
8263 */
8264static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8265{
8266 u64 *p;
8267 int lo, hi;
8268 sector_t target = s + sectors;
8269 int rv = 0;
8270
8271 if (bb->shift > 0) {
8272 /* When clearing we round the start up and the end down.
8273 * This should not matter as the shift should align with
8274 * the block size and no rounding should ever be needed.
8275 * However it is better the think a block is bad when it
8276 * isn't than to think a block is not bad when it is.
8277 */
8278 s += (1<<bb->shift) - 1;
8279 s >>= bb->shift;
8280 target >>= bb->shift;
8281 sectors = target - s;
8282 }
8283
8284 write_seqlock_irq(&bb->lock);
8285
8286 p = bb->page;
8287 lo = 0;
8288 hi = bb->count;
8289 /* Find the last range that starts before 'target' */
8290 while (hi - lo > 1) {
8291 int mid = (lo + hi) / 2;
8292 sector_t a = BB_OFFSET(p[mid]);
8293 if (a < target)
8294 lo = mid;
8295 else
8296 hi = mid;
8297 }
8298 if (hi > lo) {
8299 /* p[lo] is the last range that could overlap the
8300 * current range. Earlier ranges could also overlap,
8301 * but only this one can overlap the end of the range.
8302 */
8303 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8304 /* Partial overlap, leave the tail of this range */
8305 int ack = BB_ACK(p[lo]);
8306 sector_t a = BB_OFFSET(p[lo]);
8307 sector_t end = a + BB_LEN(p[lo]);
8308
8309 if (a < s) {
8310 /* we need to split this range */
8311 if (bb->count >= MD_MAX_BADBLOCKS) {
8312 rv = 0;
8313 goto out;
8314 }
8315 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8316 bb->count++;
8317 p[lo] = BB_MAKE(a, s-a, ack);
8318 lo++;
8319 }
8320 p[lo] = BB_MAKE(target, end - target, ack);
8321 /* there is no longer an overlap */
8322 hi = lo;
8323 lo--;
8324 }
8325 while (lo >= 0 &&
8326 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8327 /* This range does overlap */
8328 if (BB_OFFSET(p[lo]) < s) {
8329 /* Keep the early parts of this range. */
8330 int ack = BB_ACK(p[lo]);
8331 sector_t start = BB_OFFSET(p[lo]);
8332 p[lo] = BB_MAKE(start, s - start, ack);
8333 /* now low doesn't overlap, so.. */
8334 break;
8335 }
8336 lo--;
8337 }
8338 /* 'lo' is strictly before, 'hi' is strictly after,
8339 * anything between needs to be discarded
8340 */
8341 if (hi - lo > 1) {
8342 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8343 bb->count -= (hi - lo - 1);
8344 }
8345 }
8346
8347 bb->changed = 1;
8348out:
8349 write_sequnlock_irq(&bb->lock);
8350 return rv;
8351}
8352
c6563a8c
N		 8353int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8354 int is_new)
2230dfe4 8355{
c6563a8c
N		 8356 if (is_new)
8357 s += rdev->new_data_offset;
8358 else
8359 s += rdev->data_offset;
2230dfe4 8360 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8361 s, sectors);
2230dfe4
N		 8362}
8363EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8364
8365/*
8366 * Acknowledge all bad blocks in a list.
8367 * This only succeeds if ->changed is clear. It is used by
8368 * in-kernel metadata updates
8369 */
8370void md_ack_all_badblocks(struct badblocks *bb)
8371{
8372 if (bb->page == NULL || bb->changed)
8373 /* no point even trying */
8374 return;
8375 write_seqlock_irq(&bb->lock);
8376
ecb178bb 8377 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N		 8378 u64 *p = bb->page;
8379 int i;
8380 for (i = 0; i < bb->count ; i++) {
8381 if (!BB_ACK(p[i])) {
8382 sector_t start = BB_OFFSET(p[i]);
8383 int len = BB_LEN(p[i]);
8384 p[i] = BB_MAKE(start, len, 1);
8385 }
8386 }
de393cde 8387 bb->unacked_exist = 0;
2230dfe4
N		 8388 }
8389 write_sequnlock_irq(&bb->lock);
8390}
8391EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8392
16c791a5
N		 8393/* sysfs access to bad-blocks list.
8394 * We present two files.
8395 * 'bad-blocks' lists sector numbers and lengths of ranges that
8396 * are recorded as bad. The list is truncated to fit within
8397 * the one-page limit of sysfs.
8398 * Writing "sector length" to this file adds an acknowledged
8399 * bad block list.
8400 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8401 * been acknowledged. Writing to this file adds bad blocks
8402 * without acknowledging them. This is largely for testing.
8403 */
8404
8405static ssize_t
8406badblocks_show(struct badblocks *bb, char *page, int unack)
8407{
8408 size_t len;
8409 int i;
8410 u64 *p = bb->page;
8411 unsigned seq;
8412
8413 if (bb->shift < 0)
8414 return 0;
8415
8416retry:
8417 seq = read_seqbegin(&bb->lock);
8418
8419 len = 0;
8420 i = 0;
8421
8422 while (len < PAGE_SIZE && i < bb->count) {
8423 sector_t s = BB_OFFSET(p[i]);
8424 unsigned int length = BB_LEN(p[i]);
8425 int ack = BB_ACK(p[i]);
8426 i++;
8427
8428 if (unack && ack)
8429 continue;
8430
8431 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8432 (unsigned long long)s << bb->shift,
8433 length << bb->shift);
8434 }
de393cde
N		 8435 if (unack && len == 0)
8436 bb->unacked_exist = 0;
16c791a5
N		 8437
8438 if (read_seqretry(&bb->lock, seq))
8439 goto retry;
8440
8441 return len;
8442}
8443
8444#define DO_DEBUG 1
8445
8446static ssize_t
8447badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8448{
8449 unsigned long long sector;
8450 int length;
8451 char newline;
8452#ifdef DO_DEBUG
8453 /* Allow clearing via sysfs *only* for testing/debugging.
8454 * Normally only a successful write may clear a badblock
8455 */
8456 int clear = 0;
8457 if (page[0] == '-') {
8458 clear = 1;
8459 page++;
8460 }
8461#endif /* DO_DEBUG */
8462
8463 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8464 case 3:
8465 if (newline != '\n')
8466 return -EINVAL;
8467 case 2:
8468 if (length <= 0)
8469 return -EINVAL;
8470 break;
8471 default:
8472 return -EINVAL;
8473 }
8474
8475#ifdef DO_DEBUG
8476 if (clear) {
8477 md_clear_badblocks(bb, sector, length);
8478 return len;
8479 }
8480#endif /* DO_DEBUG */
8481 if (md_set_badblocks(bb, sector, length, !unack))
8482 return len;
8483 else
8484 return -ENOSPC;
8485}
8486
75c96f85
AB		 8487static int md_notify_reboot(struct notifier_block *this,
8488 unsigned long code, void *x)
1da177e4
LT		 8489{
8490 struct list_head *tmp;
fd01b88c 8491 struct mddev *mddev;
2dba6a91 8492 int need_delay = 0;
1da177e4 8493
c744a65c
N		 8494 for_each_mddev(mddev, tmp) {
8495 if (mddev_trylock(mddev)) {
30b8aa91
N		 8496 if (mddev->pers)
8497 __md_stop_writes(mddev);
c744a65c
N		 8498 mddev->safemode = 2;
8499 mddev_unlock(mddev);
2dba6a91 8500 }
c744a65c 8501 need_delay = 1;
1da177e4 8502 }
c744a65c
N		 8503 /*
8504 * certain more exotic SCSI devices are known to be
8505 * volatile wrt too early system reboots. While the
8506 * right place to handle this issue is the given
8507 * driver, we do want to have a safe RAID driver ...
8508 */
8509 if (need_delay)
8510 mdelay(1000*1);
8511
1da177e4
LT		 8512 return NOTIFY_DONE;
8513}
8514
75c96f85 8515static struct notifier_block md_notifier = {
1da177e4
LT		 8516 .notifier_call = md_notify_reboot,
8517 .next = NULL,
8518 .priority = INT_MAX, /* before any real devices */
8519};
8520
8521static void md_geninit(void)
8522{
36a4e1fe 8523 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8524
c7705f34 8525 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8526}
8527
75c96f85 8528static int __init md_init(void)
1da177e4 8529{
e804ac78
TH		 8530 int ret = -ENOMEM;
8531
ada609ee 8532 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8533 if (!md_wq)
8534 goto err_wq;
8535
8536 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8537 if (!md_misc_wq)
8538 goto err_misc_wq;
8539
8540 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8541 goto err_md;
8542
8543 if ((ret = register_blkdev(0, "mdp")) < 0)
8544 goto err_mdp;
8545 mdp_major = ret;
8546
3dbd8c2e 8547 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 8548 md_probe, NULL, NULL);
8549 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8550 md_probe, NULL, NULL);
8551
1da177e4 8552 register_reboot_notifier(&md_notifier);
0b4d4147 8553 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8554
8555 md_geninit();
d710e138 8556 return 0;
1da177e4 8557
e804ac78
TH		 8558err_mdp:
8559 unregister_blkdev(MD_MAJOR, "md");
8560err_md:
8561 destroy_workqueue(md_misc_wq);
8562err_misc_wq:
8563 destroy_workqueue(md_wq);
8564err_wq:
8565 return ret;
8566}
1da177e4
LT		 8567
8568#ifndef MODULE
8569
8570/*
8571 * Searches all registered partitions for autorun RAID arrays
8572 * at boot time.
8573 */
4d936ec1
ME		 8574
8575static LIST_HEAD(all_detected_devices);
8576struct detected_devices_node {
8577 struct list_head list;
8578 dev_t dev;
8579};
1da177e4
LT		 8580
8581void md_autodetect_dev(dev_t dev)
8582{
4d936ec1
ME		 8583 struct detected_devices_node *node_detected_dev;
8584
8585 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8586 if (node_detected_dev) {
8587 node_detected_dev->dev = dev;
8588 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8589 } else {
8590 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8591 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8592 }
1da177e4
LT		 8593}
8594
8595
8596static void autostart_arrays(int part)
8597{
3cb03002 8598 struct md_rdev *rdev;
4d936ec1
ME		 8599 struct detected_devices_node *node_detected_dev;
8600 dev_t dev;
8601 int i_scanned, i_passed;
1da177e4 8602
4d936ec1
ME		 8603 i_scanned = 0;
8604 i_passed = 0;
1da177e4 8605
4d936ec1 8606 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8607
4d936ec1
ME		 8608 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8609 i_scanned++;
8610 node_detected_dev = list_entry(all_detected_devices.next,
8611 struct detected_devices_node, list);
8612 list_del(&node_detected_dev->list);
8613 dev = node_detected_dev->dev;
8614 kfree(node_detected_dev);
df968c4e 8615 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8616 if (IS_ERR(rdev))
8617 continue;
8618
b2d444d7 8619 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 8620 MD_BUG();
8621 continue;
8622 }
d0fae18f 8623 set_bit(AutoDetected, &rdev->flags);
1da177e4 8624 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8625 i_passed++;
1da177e4 8626 }
4d936ec1
ME		 8627
8628 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8629 i_scanned, i_passed);
1da177e4
LT		 8630
8631 autorun_devices(part);
8632}
8633
fdee8ae4 8634#endif /* !MODULE */
1da177e4
LT		 8635
8636static __exit void md_exit(void)
8637{
fd01b88c 8638 struct mddev *mddev;
1da177e4 8639 struct list_head *tmp;
8ab5e4c1 8640
3dbd8c2e 8641 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 8642 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8643
3dbd8c2e 8644 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8645 unregister_blkdev(mdp_major, "mdp");
8646 unregister_reboot_notifier(&md_notifier);
8647 unregister_sysctl_table(raid_table_header);
8648 remove_proc_entry("mdstat", NULL);
29ac4aa3 8649 for_each_mddev(mddev, tmp) {
1da177e4 8650 export_array(mddev);
d3374825 8651 mddev->hold_active = 0;
1da177e4 8652 }
e804ac78
TH		 8653 destroy_workqueue(md_misc_wq);
8654 destroy_workqueue(md_wq);
1da177e4
LT		 8655}
8656
685784aa 8657subsys_initcall(md_init);
1da177e4
LT		 8658module_exit(md_exit)
8659
f91de92e
N		 8660static int get_ro(char *buffer, struct kernel_param *kp)
8661{
8662 return sprintf(buffer, "%d", start_readonly);
8663}
8664static int set_ro(const char *val, struct kernel_param *kp)
8665{
8666 char *e;
8667 int num = simple_strtoul(val, &e, 10);
8668 if (*val && (*e == '\0' || *e == '\n')) {
8669 start_readonly = num;
4dbcdc75 8670 return 0;
f91de92e
N		 8671 }
8672 return -EINVAL;
8673}
8674
80ca3a44
N		 8675module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8676module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 8677
efeb53c0 8678module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8679
1da177e4
LT		 8680EXPORT_SYMBOL(register_md_personality);
8681EXPORT_SYMBOL(unregister_md_personality);
8682EXPORT_SYMBOL(md_error);
8683EXPORT_SYMBOL(md_done_sync);
8684EXPORT_SYMBOL(md_write_start);
8685EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 8686EXPORT_SYMBOL(md_register_thread);
8687EXPORT_SYMBOL(md_unregister_thread);
8688EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 8689EXPORT_SYMBOL(md_check_recovery);
a91d5ac0 8690EXPORT_SYMBOL(md_reap_sync_thread);
1da177e4 8691MODULE_LICENSE("GPL");
0efb9e61 8692MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8693MODULE_ALIAS("md");
72008652 8694MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)