projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: avoid array overflow with bad v1.x metadata
[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>
1da177e4 39#include <linux/buffer_head.h> /* for invalidate_bdev */
d7603b7e 40#include <linux/poll.h>
16f17b39 41#include <linux/ctype.h>
fb4d8c76
N		 42#include <linux/hdreg.h>
43#include <linux/proc_fs.h>
44#include <linux/random.h>
45#include <linux/reboot.h>
32a7627c 46#include <linux/file.h>
25570727 47#include <linux/delay.h>
bff61975
N		 48#include <linux/raid/md_p.h>
49#include <linux/raid/md_u.h>
43b2e5d8 50#include "md.h"
ef740c37 51#include "bitmap.h"
1da177e4
LT		 52
53#define DEBUG 0
54#define dprintk(x...) ((void)(DEBUG && printk(x)))
55
56
57#ifndef MODULE
d710e138 58static void autostart_arrays(int part);
1da177e4
LT		 59#endif
60
2604b703 61static LIST_HEAD(pers_list);
1da177e4
LT		 62static DEFINE_SPINLOCK(pers_lock);
63
5e56341d
AB		 64static void md_print_devices(void);
65
90b08710
BS		 66static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
67
5e56341d
AB		 68#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
69
1da177e4
LT		 70/*
71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72 * is 1000 KB/sec, so the extra system load does not show up that much.
73 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 74 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 75 * subsystem is idle. There is also an 'absolute maximum' reconstruction
76 * speed limit - in case reconstruction slows down your system despite
77 * idle IO detection.
78 *
79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 80 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 81 */
82
83static int sysctl_speed_limit_min = 1000;
84static int sysctl_speed_limit_max = 200000;
88202a0c
N		 85static inline int speed_min(mddev_t *mddev)
86{
87 return mddev->sync_speed_min ?
88 mddev->sync_speed_min : sysctl_speed_limit_min;
89}
90
91static inline int speed_max(mddev_t *mddev)
92{
93 return mddev->sync_speed_max ?
94 mddev->sync_speed_max : sysctl_speed_limit_max;
95}
1da177e4
LT		 96
97static struct ctl_table_header *raid_table_header;
98
99static ctl_table raid_table[] = {
100 {
101 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
102 .procname = "speed_limit_min",
103 .data = &sysctl_speed_limit_min,
104 .maxlen = sizeof(int),
80ca3a44 105 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 106 .proc_handler = &proc_dointvec,
107 },
108 {
109 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
110 .procname = "speed_limit_max",
111 .data = &sysctl_speed_limit_max,
112 .maxlen = sizeof(int),
80ca3a44 113 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 114 .proc_handler = &proc_dointvec,
115 },
116 { .ctl_name = 0 }
117};
118
119static ctl_table raid_dir_table[] = {
120 {
121 .ctl_name = DEV_RAID,
122 .procname = "raid",
123 .maxlen = 0,
80ca3a44 124 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 125 .child = raid_table,
126 },
127 { .ctl_name = 0 }
128};
129
130static ctl_table raid_root_table[] = {
131 {
132 .ctl_name = CTL_DEV,
133 .procname = "dev",
134 .maxlen = 0,
135 .mode = 0555,
136 .child = raid_dir_table,
137 },
138 { .ctl_name = 0 }
139};
140
141static struct block_device_operations md_fops;
142
f91de92e
N		 143static int start_readonly;
144
d7603b7e
N		 145/*
146 * We have a system wide 'event count' that is incremented
147 * on any 'interesting' event, and readers of /proc/mdstat
148 * can use 'poll' or 'select' to find out when the event
149 * count increases.
150 *
151 * Events are:
152 * start array, stop array, error, add device, remove device,
153 * start build, activate spare
154 */
2989ddbd 155static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 156static atomic_t md_event_count;
29269553 157void md_new_event(mddev_t *mddev)
d7603b7e
N		 158{
159 atomic_inc(&md_event_count);
160 wake_up(&md_event_waiters);
161}
29269553 162EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 163
c331eb04
N		 164/* Alternate version that can be called from interrupts
165 * when calling sysfs_notify isn't needed.
166 */
05381954 167static void md_new_event_inintr(mddev_t *mddev)
c331eb04
N		 168{
169 atomic_inc(&md_event_count);
170 wake_up(&md_event_waiters);
171}
172
1da177e4
LT		 173/*
174 * Enables to iterate over all existing md arrays
175 * all_mddevs_lock protects this list.
176 */
177static LIST_HEAD(all_mddevs);
178static DEFINE_SPINLOCK(all_mddevs_lock);
179
180
181/*
182 * iterates through all used mddevs in the system.
183 * We take care to grab the all_mddevs_lock whenever navigating
184 * the list, and to always hold a refcount when unlocked.
185 * Any code which breaks out of this loop while own
186 * a reference to the current mddev and must mddev_put it.
187 */
29ac4aa3 188#define for_each_mddev(mddev,tmp) \
1da177e4
LT		 189 \
190 for (({ spin_lock(&all_mddevs_lock); \
191 tmp = all_mddevs.next; \
192 mddev = NULL;}); \
193 ({ if (tmp != &all_mddevs) \
194 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
195 spin_unlock(&all_mddevs_lock); \
196 if (mddev) mddev_put(mddev); \
197 mddev = list_entry(tmp, mddev_t, all_mddevs); \
198 tmp != &all_mddevs;}); \
199 ({ spin_lock(&all_mddevs_lock); \
200 tmp = tmp->next;}) \
201 )
202
203
409c57f3
N		 204/* Rather than calling directly into the personality make_request function,
205 * IO requests come here first so that we can check if the device is
206 * being suspended pending a reconfiguration.
207 * We hold a refcount over the call to ->make_request. By the time that
208 * call has finished, the bio has been linked into some internal structure
209 * and so is visible to ->quiesce(), so we don't need the refcount any more.
210 */
211static int md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 212{
409c57f3
N		 213 mddev_t *mddev = q->queuedata;
214 int rv;
215 if (mddev == NULL || mddev->pers == NULL) {
216 bio_io_error(bio);
217 return 0;
218 }
219 rcu_read_lock();
220 if (mddev->suspended) {
221 DEFINE_WAIT(__wait);
222 for (;;) {
223 prepare_to_wait(&mddev->sb_wait, &__wait,
224 TASK_UNINTERRUPTIBLE);
225 if (!mddev->suspended)
226 break;
227 rcu_read_unlock();
228 schedule();
229 rcu_read_lock();
230 }
231 finish_wait(&mddev->sb_wait, &__wait);
232 }
233 atomic_inc(&mddev->active_io);
234 rcu_read_unlock();
235 rv = mddev->pers->make_request(q, bio);
236 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
237 wake_up(&mddev->sb_wait);
238
239 return rv;
240}
241
242static void mddev_suspend(mddev_t *mddev)
243{
244 BUG_ON(mddev->suspended);
245 mddev->suspended = 1;
246 synchronize_rcu();
247 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
248 mddev->pers->quiesce(mddev, 1);
249 md_unregister_thread(mddev->thread);
250 mddev->thread = NULL;
251 /* we now know that no code is executing in the personality module,
252 * except possibly the tail end of a ->bi_end_io function, but that
253 * is certain to complete before the module has a chance to get
254 * unloaded
255 */
256}
257
258static void mddev_resume(mddev_t *mddev)
259{
260 mddev->suspended = 0;
261 wake_up(&mddev->sb_wait);
262 mddev->pers->quiesce(mddev, 0);
1da177e4
LT		 263}
264
409c57f3 265
1da177e4
LT		 266static inline mddev_t *mddev_get(mddev_t *mddev)
267{
268 atomic_inc(&mddev->active);
269 return mddev;
270}
271
5fd3a17e 272static void mddev_delayed_delete(struct work_struct *ws);
d3374825 273
1da177e4
LT		 274static void mddev_put(mddev_t *mddev)
275{
276 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
277 return;
d3374825
N		 278 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
279 !mddev->hold_active) {
1da177e4 280 list_del(&mddev->all_mddevs);
d3374825
N		 281 if (mddev->gendisk) {
282 /* we did a probe so need to clean up.
283 * Call schedule_work inside the spinlock
284 * so that flush_scheduled_work() after
285 * mddev_find will succeed in waiting for the
286 * work to be done.
287 */
288 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
289 schedule_work(&mddev->del_work);
290 } else
291 kfree(mddev);
292 }
293 spin_unlock(&all_mddevs_lock);
1da177e4
LT		 294}
295
296static mddev_t * mddev_find(dev_t unit)
297{
298 mddev_t *mddev, *new = NULL;
299
300 retry:
301 spin_lock(&all_mddevs_lock);
efeb53c0
N		 302
303 if (unit) {
304 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
305 if (mddev->unit == unit) {
306 mddev_get(mddev);
307 spin_unlock(&all_mddevs_lock);
308 kfree(new);
309 return mddev;
310 }
311
312 if (new) {
313 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 314 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 315 new->hold_active = UNTIL_IOCTL;
316 return new;
1da177e4 317 }
efeb53c0
N		 318 } else if (new) {
319 /* find an unused unit number */
320 static int next_minor = 512;
321 int start = next_minor;
322 int is_free = 0;
323 int dev = 0;
324 while (!is_free) {
325 dev = MKDEV(MD_MAJOR, next_minor);
326 next_minor++;
327 if (next_minor > MINORMASK)
328 next_minor = 0;
329 if (next_minor == start) {
330 /* Oh dear, all in use. */
331 spin_unlock(&all_mddevs_lock);
332 kfree(new);
333 return NULL;
334 }
335
336 is_free = 1;
337 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
338 if (mddev->unit == dev) {
339 is_free = 0;
340 break;
341 }
342 }
343 new->unit = dev;
344 new->md_minor = MINOR(dev);
345 new->hold_active = UNTIL_STOP;
1da177e4
LT		 346 list_add(&new->all_mddevs, &all_mddevs);
347 spin_unlock(&all_mddevs_lock);
348 return new;
349 }
350 spin_unlock(&all_mddevs_lock);
351
9ffae0cf 352 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 353 if (!new)
354 return NULL;
355
1da177e4
LT		 356 new->unit = unit;
357 if (MAJOR(unit) == MD_MAJOR)
358 new->md_minor = MINOR(unit);
359 else
360 new->md_minor = MINOR(unit) >> MdpMinorShift;
361
df5b89b3 362 mutex_init(&new->reconfig_mutex);
1da177e4
LT		 363 INIT_LIST_HEAD(&new->disks);
364 INIT_LIST_HEAD(&new->all_mddevs);
365 init_timer(&new->safemode_timer);
366 atomic_set(&new->active, 1);
f2ea68cf 367 atomic_set(&new->openers, 0);
409c57f3 368 atomic_set(&new->active_io, 0);
06d91a5f 369 spin_lock_init(&new->write_lock);
3d310eb7 370 init_waitqueue_head(&new->sb_wait);
a6d8113a 371 init_waitqueue_head(&new->recovery_wait);
08a02ecd 372 new->reshape_position = MaxSector;
5e96ee65 373 new->resync_min = 0;
c6207277 374 new->resync_max = MaxSector;
d897dbf9 375 new->level = LEVEL_NONE;
1da177e4 376
1da177e4
LT		 377 goto retry;
378}
379
380static inline int mddev_lock(mddev_t * mddev)
381{
df5b89b3 382 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 383}
384
b522adcd
DW		 385static inline int mddev_is_locked(mddev_t *mddev)
386{
387 return mutex_is_locked(&mddev->reconfig_mutex);
388}
389
1da177e4
LT		 390static inline int mddev_trylock(mddev_t * mddev)
391{
df5b89b3 392 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 393}
394
395static inline void mddev_unlock(mddev_t * mddev)
396{
df5b89b3 397 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 398
005eca5e 399 md_wakeup_thread(mddev->thread);
1da177e4
LT		 400}
401
2989ddbd 402static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4 403{
159ec1fc 404 mdk_rdev_t *rdev;
1da177e4 405
159ec1fc 406 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 407 if (rdev->desc_nr == nr)
408 return rdev;
159ec1fc 409
1da177e4
LT		 410 return NULL;
411}
412
413static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
414{
1da177e4
LT		 415 mdk_rdev_t *rdev;
416
159ec1fc 417 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 418 if (rdev->bdev->bd_dev == dev)
419 return rdev;
159ec1fc 420
1da177e4
LT		 421 return NULL;
422}
423
d9d166c2 424static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 425{
426 struct mdk_personality *pers;
d9d166c2
N		 427 list_for_each_entry(pers, &pers_list, list) {
428 if (level != LEVEL_NONE && pers->level == level)
2604b703 429 return pers;
d9d166c2
N		 430 if (strcmp(pers->name, clevel)==0)
431 return pers;
432 }
2604b703
N		 433 return NULL;
434}
435
b73df2d3 436/* return the offset of the super block in 512byte sectors */
77933d72 437static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4 438{
b73df2d3
AN		 439 sector_t num_sectors = bdev->bd_inode->i_size / 512;
440 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 441}
442
1da177e4
LT		 443static int alloc_disk_sb(mdk_rdev_t * rdev)
444{
445 if (rdev->sb_page)
446 MD_BUG();
447
448 rdev->sb_page = alloc_page(GFP_KERNEL);
449 if (!rdev->sb_page) {
450 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 451 return -ENOMEM;
1da177e4
LT		 452 }
453
454 return 0;
455}
456
457static void free_disk_sb(mdk_rdev_t * rdev)
458{
459 if (rdev->sb_page) {
2d1f3b5d 460 put_page(rdev->sb_page);
1da177e4
LT		 461 rdev->sb_loaded = 0;
462 rdev->sb_page = NULL;
0f420358 463 rdev->sb_start = 0;
dd8ac336 464 rdev->sectors = 0;
1da177e4
LT		 465 }
466}
467
468
6712ecf8 469static void super_written(struct bio *bio, int error)
7bfa19f2
N		 470{
471 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 472 mddev_t *mddev = rdev->mddev;
7bfa19f2 473
3a0f5bbb
N		 474 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
475 printk("md: super_written gets error=%d, uptodate=%d\n",
476 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
477 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 478 md_error(mddev, rdev);
3a0f5bbb 479 }
7bfa19f2 480
a9701a30
N		 481 if (atomic_dec_and_test(&mddev->pending_writes))
482 wake_up(&mddev->sb_wait);
f8b58edf 483 bio_put(bio);
7bfa19f2
N		 484}
485
6712ecf8 486static void super_written_barrier(struct bio *bio, int error)
a9701a30
N		 487{
488 struct bio *bio2 = bio->bi_private;
489 mdk_rdev_t *rdev = bio2->bi_private;
490 mddev_t *mddev = rdev->mddev;
a9701a30
N		 491
492 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
493 error == -EOPNOTSUPP) {
494 unsigned long flags;
495 /* barriers don't appear to be supported :-( */
496 set_bit(BarriersNotsupp, &rdev->flags);
497 mddev->barriers_work = 0;
498 spin_lock_irqsave(&mddev->write_lock, flags);
499 bio2->bi_next = mddev->biolist;
500 mddev->biolist = bio2;
501 spin_unlock_irqrestore(&mddev->write_lock, flags);
502 wake_up(&mddev->sb_wait);
503 bio_put(bio);
6712ecf8
N		 504 } else {
505 bio_put(bio2);
506 bio->bi_private = rdev;
507 super_written(bio, error);
a9701a30 508 }
a9701a30
N		 509}
510
7bfa19f2
N		 511void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
512 sector_t sector, int size, struct page *page)
513{
514 /* write first size bytes of page to sector of rdev
515 * Increment mddev->pending_writes before returning
516 * and decrement it on completion, waking up sb_wait
517 * if zero is reached.
518 * If an error occurred, call md_error
a9701a30
N		 519 *
520 * As we might need to resubmit the request if BIO_RW_BARRIER
521 * causes ENOTSUPP, we allocate a spare bio...
7bfa19f2
N		 522 */
523 struct bio *bio = bio_alloc(GFP_NOIO, 1);
93dbb393 524 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
7bfa19f2
N		 525
526 bio->bi_bdev = rdev->bdev;
527 bio->bi_sector = sector;
528 bio_add_page(bio, page, size, 0);
529 bio->bi_private = rdev;
530 bio->bi_end_io = super_written;
a9701a30
N		 531 bio->bi_rw = rw;
532
7bfa19f2 533 atomic_inc(&mddev->pending_writes);
a9701a30
N		 534 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
535 struct bio *rbio;
536 rw |= (1<<BIO_RW_BARRIER);
537 rbio = bio_clone(bio, GFP_NOIO);
538 rbio->bi_private = bio;
539 rbio->bi_end_io = super_written_barrier;
540 submit_bio(rw, rbio);
541 } else
542 submit_bio(rw, bio);
543}
544
545void md_super_wait(mddev_t *mddev)
546{
547 /* wait for all superblock writes that were scheduled to complete.
548 * if any had to be retried (due to BARRIER problems), retry them
549 */
550 DEFINE_WAIT(wq);
551 for(;;) {
552 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
553 if (atomic_read(&mddev->pending_writes)==0)
554 break;
555 while (mddev->biolist) {
556 struct bio *bio;
557 spin_lock_irq(&mddev->write_lock);
558 bio = mddev->biolist;
559 mddev->biolist = bio->bi_next ;
560 bio->bi_next = NULL;
561 spin_unlock_irq(&mddev->write_lock);
562 submit_bio(bio->bi_rw, bio);
563 }
564 schedule();
565 }
566 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 567}
568
6712ecf8 569static void bi_complete(struct bio *bio, int error)
1da177e4 570{
1da177e4 571 complete((struct completion*)bio->bi_private);
1da177e4
LT		 572}
573
a654b9d8 574int sync_page_io(struct block_device *bdev, sector_t sector, int size,
1da177e4
LT		 575 struct page *page, int rw)
576{
baaa2c51 577 struct bio *bio = bio_alloc(GFP_NOIO, 1);
1da177e4
LT		 578 struct completion event;
579 int ret;
580
93dbb393 581 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
1da177e4
LT		 582
583 bio->bi_bdev = bdev;
584 bio->bi_sector = sector;
585 bio_add_page(bio, page, size, 0);
586 init_completion(&event);
587 bio->bi_private = &event;
588 bio->bi_end_io = bi_complete;
589 submit_bio(rw, bio);
590 wait_for_completion(&event);
591
592 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
593 bio_put(bio);
594 return ret;
595}
a8745db2 596EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 597
0002b271 598static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 599{
600 char b[BDEVNAME_SIZE];
601 if (!rdev->sb_page) {
602 MD_BUG();
603 return -EINVAL;
604 }
605 if (rdev->sb_loaded)
606 return 0;
607
608
0f420358 609 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
1da177e4
LT		 610 goto fail;
611 rdev->sb_loaded = 1;
612 return 0;
613
614fail:
615 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
616 bdevname(rdev->bdev,b));
617 return -EINVAL;
618}
619
620static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
621{
05710466
AN		 622 return sb1->set_uuid0 == sb2->set_uuid0 &&
623 sb1->set_uuid1 == sb2->set_uuid1 &&
624 sb1->set_uuid2 == sb2->set_uuid2 &&
625 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 626}
627
1da177e4
LT		 628static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
629{
630 int ret;
631 mdp_super_t *tmp1, *tmp2;
632
633 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
634 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
635
636 if (!tmp1 || !tmp2) {
637 ret = 0;
35020f1a 638 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 639 goto abort;
640 }
641
642 *tmp1 = *sb1;
643 *tmp2 = *sb2;
644
645 /*
646 * nr_disks is not constant
647 */
648 tmp1->nr_disks = 0;
649 tmp2->nr_disks = 0;
650
ce0c8e05 651 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 652abort:
990a8baf
JJ		 653 kfree(tmp1);
654 kfree(tmp2);
1da177e4
LT		 655 return ret;
656}
657
4d167f09
N		 658
659static u32 md_csum_fold(u32 csum)
660{
661 csum = (csum & 0xffff) + (csum >> 16);
662 return (csum & 0xffff) + (csum >> 16);
663}
664
1da177e4
LT		 665static unsigned int calc_sb_csum(mdp_super_t * sb)
666{
4d167f09
N		 667 u64 newcsum = 0;
668 u32 *sb32 = (u32*)sb;
669 int i;
1da177e4
LT		 670 unsigned int disk_csum, csum;
671
672 disk_csum = sb->sb_csum;
673 sb->sb_csum = 0;
4d167f09
N		 674
675 for (i = 0; i < MD_SB_BYTES/4 ; i++)
676 newcsum += sb32[i];
677 csum = (newcsum & 0xffffffff) + (newcsum>>32);
678
679
680#ifdef CONFIG_ALPHA
681 /* This used to use csum_partial, which was wrong for several
682 * reasons including that different results are returned on
683 * different architectures. It isn't critical that we get exactly
684 * the same return value as before (we always csum_fold before
685 * testing, and that removes any differences). However as we
686 * know that csum_partial always returned a 16bit value on
687 * alphas, do a fold to maximise conformity to previous behaviour.
688 */
689 sb->sb_csum = md_csum_fold(disk_csum);
690#else
1da177e4 691 sb->sb_csum = disk_csum;
4d167f09 692#endif
1da177e4
LT		 693 return csum;
694}
695
696
697/*
698 * Handle superblock details.
699 * We want to be able to handle multiple superblock formats
700 * so we have a common interface to them all, and an array of
701 * different handlers.
702 * We rely on user-space to write the initial superblock, and support
703 * reading and updating of superblocks.
704 * Interface methods are:
705 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
706 * loads and validates a superblock on dev.
707 * if refdev != NULL, compare superblocks on both devices
708 * Return:
709 * 0 - dev has a superblock that is compatible with refdev
710 * 1 - dev has a superblock that is compatible and newer than refdev
711 * so dev should be used as the refdev in future
712 * -EINVAL superblock incompatible or invalid
713 * -othererror e.g. -EIO
714 *
715 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
716 * Verify that dev is acceptable into mddev.
717 * The first time, mddev->raid_disks will be 0, and data from
718 * dev should be merged in. Subsequent calls check that dev
719 * is new enough. Return 0 or -EINVAL
720 *
721 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
722 * Update the superblock for rdev with data in mddev
723 * This does not write to disc.
724 *
725 */
726
727struct super_type {
0cd17fec
CW		 728 char *name;
729 struct module *owner;
730 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
731 int minor_version);
732 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
733 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
734 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
15f4a5fd 735 sector_t num_sectors);
1da177e4
LT		 736};
737
0894cc30
AN		 738/*
739 * Check that the given mddev has no bitmap.
740 *
741 * This function is called from the run method of all personalities that do not
742 * support bitmaps. It prints an error message and returns non-zero if mddev
743 * has a bitmap. Otherwise, it returns 0.
744 *
745 */
746int md_check_no_bitmap(mddev_t *mddev)
747{
748 if (!mddev->bitmap_file && !mddev->bitmap_offset)
749 return 0;
750 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
751 mdname(mddev), mddev->pers->name);
752 return 1;
753}
754EXPORT_SYMBOL(md_check_no_bitmap);
755
1da177e4
LT		 756/*
757 * load_super for 0.90.0
758 */
759static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
760{
761 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
762 mdp_super_t *sb;
763 int ret;
1da177e4
LT		 764
765 /*
0f420358 766 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 767 * it's at the end of the disk.
768 *
769 * It also happens to be a multiple of 4Kb.
770 */
0f420358 771 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1da177e4 772
0002b271 773 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 774 if (ret) return ret;
775
776 ret = -EINVAL;
777
778 bdevname(rdev->bdev, b);
779 sb = (mdp_super_t*)page_address(rdev->sb_page);
780
781 if (sb->md_magic != MD_SB_MAGIC) {
782 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
783 b);
784 goto abort;
785 }
786
787 if (sb->major_version != 0 ||
f6705578
N		 788 sb->minor_version < 90 ||
789 sb->minor_version > 91) {
1da177e4
LT		 790 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
791 sb->major_version, sb->minor_version,
792 b);
793 goto abort;
794 }
795
796 if (sb->raid_disks <= 0)
797 goto abort;
798
4d167f09 799 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 800 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
801 b);
802 goto abort;
803 }
804
805 rdev->preferred_minor = sb->md_minor;
806 rdev->data_offset = 0;
0002b271 807 rdev->sb_size = MD_SB_BYTES;
1da177e4
LT		 808
809 if (sb->level == LEVEL_MULTIPATH)
810 rdev->desc_nr = -1;
811 else
812 rdev->desc_nr = sb->this_disk.number;
813
9a7b2b0f 814 if (!refdev) {
1da177e4 815 ret = 1;
9a7b2b0f 816 } else {
1da177e4
LT		 817 __u64 ev1, ev2;
818 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
819 if (!uuid_equal(refsb, sb)) {
820 printk(KERN_WARNING "md: %s has different UUID to %s\n",
821 b, bdevname(refdev->bdev,b2));
822 goto abort;
823 }
824 if (!sb_equal(refsb, sb)) {
825 printk(KERN_WARNING "md: %s has same UUID"
826 " but different superblock to %s\n",
827 b, bdevname(refdev->bdev, b2));
828 goto abort;
829 }
830 ev1 = md_event(sb);
831 ev2 = md_event(refsb);
832 if (ev1 > ev2)
833 ret = 1;
834 else
835 ret = 0;
836 }
8190e754 837 rdev->sectors = rdev->sb_start;
1da177e4 838
dd8ac336 839 if (rdev->sectors < sb->size * 2 && sb->level > 1)
2bf071bf
N		 840 /* "this cannot possibly happen" ... */
841 ret = -EINVAL;
842
1da177e4
LT		 843 abort:
844 return ret;
845}
846
847/*
848 * validate_super for 0.90.0
849 */
850static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
851{
852 mdp_disk_t *desc;
853 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
07d84d10 854 __u64 ev1 = md_event(sb);
1da177e4 855
41158c7e 856 rdev->raid_disk = -1;
c5d79adb
N		 857 clear_bit(Faulty, &rdev->flags);
858 clear_bit(In_sync, &rdev->flags);
859 clear_bit(WriteMostly, &rdev->flags);
860 clear_bit(BarriersNotsupp, &rdev->flags);
861
1da177e4
LT		 862 if (mddev->raid_disks == 0) {
863 mddev->major_version = 0;
864 mddev->minor_version = sb->minor_version;
865 mddev->patch_version = sb->patch_version;
e691063a 866 mddev->external = 0;
9d8f0363 867 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 868 mddev->ctime = sb->ctime;
869 mddev->utime = sb->utime;
870 mddev->level = sb->level;
d9d166c2 871 mddev->clevel[0] = 0;
1da177e4
LT		 872 mddev->layout = sb->layout;
873 mddev->raid_disks = sb->raid_disks;
58c0fed4 874 mddev->dev_sectors = sb->size * 2;
07d84d10 875 mddev->events = ev1;
9223214e 876 mddev->bitmap_offset = 0;
36fa3063 877 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
1da177e4 878
f6705578
N		 879 if (mddev->minor_version >= 91) {
880 mddev->reshape_position = sb->reshape_position;
881 mddev->delta_disks = sb->delta_disks;
882 mddev->new_level = sb->new_level;
883 mddev->new_layout = sb->new_layout;
664e7c41 884 mddev->new_chunk_sectors = sb->new_chunk >> 9;
f6705578
N		 885 } else {
886 mddev->reshape_position = MaxSector;
887 mddev->delta_disks = 0;
888 mddev->new_level = mddev->level;
889 mddev->new_layout = mddev->layout;
664e7c41 890 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 891 }
892
1da177e4
LT		 893 if (sb->state & (1<<MD_SB_CLEAN))
894 mddev->recovery_cp = MaxSector;
895 else {
896 if (sb->events_hi == sb->cp_events_hi &&
897 sb->events_lo == sb->cp_events_lo) {
898 mddev->recovery_cp = sb->recovery_cp;
899 } else
900 mddev->recovery_cp = 0;
901 }
902
903 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
904 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
905 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
906 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
907
908 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 909
910 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
e11e93fa 911 mddev->bitmap_file == NULL)
36fa3063 912 mddev->bitmap_offset = mddev->default_bitmap_offset;
a654b9d8 913
41158c7e
N		 914 } else if (mddev->pers == NULL) {
915 /* Insist on good event counter while assembling */
1da177e4
LT		 916 ++ev1;
917 if (ev1 < mddev->events)
918 return -EINVAL;
41158c7e
N		 919 } else if (mddev->bitmap) {
920 /* if adding to array with a bitmap, then we can accept an
921 * older device ... but not too old.
922 */
41158c7e
N		 923 if (ev1 < mddev->bitmap->events_cleared)
924 return 0;
07d84d10
N		 925 } else {
926 if (ev1 < mddev->events)
927 /* just a hot-add of a new device, leave raid_disk at -1 */
928 return 0;
929 }
41158c7e 930
1da177e4 931 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 932 desc = sb->disks + rdev->desc_nr;
933
934 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 935 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 936 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
937 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 938 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 939 rdev->raid_disk = desc->raid_disk;
940 }
8ddf9efe
N		 941 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
942 set_bit(WriteMostly, &rdev->flags);
41158c7e 943 } else /* MULTIPATH are always insync */
b2d444d7 944 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 945 return 0;
946}
947
948/*
949 * sync_super for 0.90.0
950 */
951static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
952{
953 mdp_super_t *sb;
1da177e4
LT		 954 mdk_rdev_t *rdev2;
955 int next_spare = mddev->raid_disks;
19133a42 956
1da177e4
LT		 957
958 /* make rdev->sb match mddev data..
959 *
960 * 1/ zero out disks
961 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
962 * 3/ any empty disks < next_spare become removed
963 *
964 * disks[0] gets initialised to REMOVED because
965 * we cannot be sure from other fields if it has
966 * been initialised or not.
967 */
968 int i;
969 int active=0, working=0,failed=0,spare=0,nr_disks=0;
970
61181565
N		 971 rdev->sb_size = MD_SB_BYTES;
972
1da177e4
LT		 973 sb = (mdp_super_t*)page_address(rdev->sb_page);
974
975 memset(sb, 0, sizeof(*sb));
976
977 sb->md_magic = MD_SB_MAGIC;
978 sb->major_version = mddev->major_version;
1da177e4
LT		 979 sb->patch_version = mddev->patch_version;
980 sb->gvalid_words = 0; /* ignored */
981 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
982 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
983 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
984 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
985
986 sb->ctime = mddev->ctime;
987 sb->level = mddev->level;
58c0fed4 988 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 989 sb->raid_disks = mddev->raid_disks;
990 sb->md_minor = mddev->md_minor;
e691063a 991 sb->not_persistent = 0;
1da177e4
LT		 992 sb->utime = mddev->utime;
993 sb->state = 0;
994 sb->events_hi = (mddev->events>>32);
995 sb->events_lo = (u32)mddev->events;
996
f6705578
N		 997 if (mddev->reshape_position == MaxSector)
998 sb->minor_version = 90;
999 else {
1000 sb->minor_version = 91;
1001 sb->reshape_position = mddev->reshape_position;
1002 sb->new_level = mddev->new_level;
1003 sb->delta_disks = mddev->delta_disks;
1004 sb->new_layout = mddev->new_layout;
664e7c41 1005 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1006 }
1007 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1008 if (mddev->in_sync)
1009 {
1010 sb->recovery_cp = mddev->recovery_cp;
1011 sb->cp_events_hi = (mddev->events>>32);
1012 sb->cp_events_lo = (u32)mddev->events;
1013 if (mddev->recovery_cp == MaxSector)
1014 sb->state = (1<< MD_SB_CLEAN);
1015 } else
1016 sb->recovery_cp = 0;
1017
1018 sb->layout = mddev->layout;
9d8f0363 1019 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1020
a654b9d8
N		 1021 if (mddev->bitmap && mddev->bitmap_file == NULL)
1022 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1023
1da177e4 1024 sb->disks[0].state = (1<<MD_DISK_REMOVED);
159ec1fc 1025 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1026 mdp_disk_t *d;
86e6ffdd 1027 int desc_nr;
b2d444d7
N		 1028 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1029 && !test_bit(Faulty, &rdev2->flags))
86e6ffdd 1030 desc_nr = rdev2->raid_disk;
1da177e4 1031 else
86e6ffdd 1032 desc_nr = next_spare++;
19133a42 1033 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1034 d = &sb->disks[rdev2->desc_nr];
1035 nr_disks++;
1036 d->number = rdev2->desc_nr;
1037 d->major = MAJOR(rdev2->bdev->bd_dev);
1038 d->minor = MINOR(rdev2->bdev->bd_dev);
b2d444d7
N		 1039 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1040 && !test_bit(Faulty, &rdev2->flags))
1da177e4
LT		 1041 d->raid_disk = rdev2->raid_disk;
1042 else
1043 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1044 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1045 d->state = (1<<MD_DISK_FAULTY);
1be7892f 1046 else if (test_bit(In_sync, &rdev2->flags)) {
1da177e4
LT		 1047 d->state = (1<<MD_DISK_ACTIVE);
1048 d->state |= (1<<MD_DISK_SYNC);
1049 active++;
1050 working++;
1051 } else {
1052 d->state = 0;
1053 spare++;
1054 working++;
1055 }
8ddf9efe
N		 1056 if (test_bit(WriteMostly, &rdev2->flags))
1057 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1058 }
1da177e4
LT		 1059 /* now set the "removed" and "faulty" bits on any missing devices */
1060 for (i=0 ; i < mddev->raid_disks ; i++) {
1061 mdp_disk_t *d = &sb->disks[i];
1062 if (d->state == 0 && d->number == 0) {
1063 d->number = i;
1064 d->raid_disk = i;
1065 d->state = (1<<MD_DISK_REMOVED);
1066 d->state |= (1<<MD_DISK_FAULTY);
1067 failed++;
1068 }
1069 }
1070 sb->nr_disks = nr_disks;
1071 sb->active_disks = active;
1072 sb->working_disks = working;
1073 sb->failed_disks = failed;
1074 sb->spare_disks = spare;
1075
1076 sb->this_disk = sb->disks[rdev->desc_nr];
1077 sb->sb_csum = calc_sb_csum(sb);
1078}
1079
0cd17fec
CW		 1080/*
1081 * rdev_size_change for 0.90.0
1082 */
1083static unsigned long long
15f4a5fd 1084super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
0cd17fec 1085{
58c0fed4 1086 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1087 return 0; /* component must fit device */
0cd17fec
CW		 1088 if (rdev->mddev->bitmap_offset)
1089 return 0; /* can't move bitmap */
0f420358 1090 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
15f4a5fd
AN		 1091 if (!num_sectors || num_sectors > rdev->sb_start)
1092 num_sectors = rdev->sb_start;
0f420358 1093 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1094 rdev->sb_page);
1095 md_super_wait(rdev->mddev);
15f4a5fd 1096 return num_sectors / 2; /* kB for sysfs */
0cd17fec
CW		 1097}
1098
1099
1da177e4
LT		 1100/*
1101 * version 1 superblock
1102 */
1103
1c05b4bc 1104static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1105{
1c05b4bc
N		 1106 __le32 disk_csum;
1107 u32 csum;
1da177e4
LT		 1108 unsigned long long newcsum;
1109 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1110 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1111 int i;
1112
1113 disk_csum = sb->sb_csum;
1114 sb->sb_csum = 0;
1115 newcsum = 0;
1116 for (i=0; size>=4; size -= 4 )
1117 newcsum += le32_to_cpu(*isuper++);
1118
1119 if (size == 2)
1c05b4bc 1120 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1121
1122 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1123 sb->sb_csum = disk_csum;
1124 return cpu_to_le32(csum);
1125}
1126
1127static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1128{
1129 struct mdp_superblock_1 *sb;
1130 int ret;
0f420358 1131 sector_t sb_start;
1da177e4 1132 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1133 int bmask;
1da177e4
LT		 1134
1135 /*
0f420358 1136 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1137 * It is always aligned to a 4K boundary and
1138 * depeding on minor_version, it can be:
1139 * 0: At least 8K, but less than 12K, from end of device
1140 * 1: At start of device
1141 * 2: 4K from start of device.
1142 */
1143 switch(minor_version) {
1144 case 0:
0f420358
AN		 1145 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1146 sb_start -= 8*2;
1147 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1148 break;
1149 case 1:
0f420358 1150 sb_start = 0;
1da177e4
LT		 1151 break;
1152 case 2:
0f420358 1153 sb_start = 8;
1da177e4
LT		 1154 break;
1155 default:
1156 return -EINVAL;
1157 }
0f420358 1158 rdev->sb_start = sb_start;
1da177e4 1159
0002b271
N		 1160 /* superblock is rarely larger than 1K, but it can be larger,
1161 * and it is safe to read 4k, so we do that
1162 */
1163 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1164 if (ret) return ret;
1165
1166
1167 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1168
1169 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1170 sb->major_version != cpu_to_le32(1) ||
1171 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1172 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1173 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1174 return -EINVAL;
1175
1176 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1177 printk("md: invalid superblock checksum on %s\n",
1178 bdevname(rdev->bdev,b));
1179 return -EINVAL;
1180 }
1181 if (le64_to_cpu(sb->data_size) < 10) {
1182 printk("md: data_size too small on %s\n",
1183 bdevname(rdev->bdev,b));
1184 return -EINVAL;
1185 }
e11e93fa 1186
1da177e4
LT		 1187 rdev->preferred_minor = 0xffff;
1188 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1189 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1190
0002b271 1191 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1192 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1193 if (rdev->sb_size & bmask)
a1801f85
N		 1194 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1195
1196 if (minor_version
0f420358 1197 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1198 return -EINVAL;
0002b271 1199
31b65a0d
N		 1200 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1201 rdev->desc_nr = -1;
1202 else
1203 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1204
9a7b2b0f 1205 if (!refdev) {
8ed75463 1206 ret = 1;
9a7b2b0f 1207 } else {
1da177e4
LT		 1208 __u64 ev1, ev2;
1209 struct mdp_superblock_1 *refsb =
1210 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1211
1212 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1213 sb->level != refsb->level ||
1214 sb->layout != refsb->layout ||
1215 sb->chunksize != refsb->chunksize) {
1216 printk(KERN_WARNING "md: %s has strangely different"
1217 " superblock to %s\n",
1218 bdevname(rdev->bdev,b),
1219 bdevname(refdev->bdev,b2));
1220 return -EINVAL;
1221 }
1222 ev1 = le64_to_cpu(sb->events);
1223 ev2 = le64_to_cpu(refsb->events);
1224
1225 if (ev1 > ev2)
8ed75463
N		 1226 ret = 1;
1227 else
1228 ret = 0;
1da177e4 1229 }
a1801f85 1230 if (minor_version)
dd8ac336
AN		 1231 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1232 le64_to_cpu(sb->data_offset);
1da177e4 1233 else
dd8ac336
AN		 1234 rdev->sectors = rdev->sb_start;
1235 if (rdev->sectors < le64_to_cpu(sb->data_size))
1da177e4 1236 return -EINVAL;
dd8ac336 1237 rdev->sectors = le64_to_cpu(sb->data_size);
dd8ac336 1238 if (le64_to_cpu(sb->size) > rdev->sectors)
2bf071bf 1239 return -EINVAL;
8ed75463 1240 return ret;
1da177e4
LT		 1241}
1242
1243static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1244{
1245 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
07d84d10 1246 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1247
41158c7e 1248 rdev->raid_disk = -1;
c5d79adb
N		 1249 clear_bit(Faulty, &rdev->flags);
1250 clear_bit(In_sync, &rdev->flags);
1251 clear_bit(WriteMostly, &rdev->flags);
1252 clear_bit(BarriersNotsupp, &rdev->flags);
1253
1da177e4
LT		 1254 if (mddev->raid_disks == 0) {
1255 mddev->major_version = 1;
1256 mddev->patch_version = 0;
e691063a 1257 mddev->external = 0;
9d8f0363 1258 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1259 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1260 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1261 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1262 mddev->clevel[0] = 0;
1da177e4
LT		 1263 mddev->layout = le32_to_cpu(sb->layout);
1264 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1265 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1266 mddev->events = ev1;
9223214e 1267 mddev->bitmap_offset = 0;
29fc7e3e 1268 mddev->default_bitmap_offset = 1024 >> 9;
1da177e4
LT		 1269
1270 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1271 memcpy(mddev->uuid, sb->set_uuid, 16);
1272
1273 mddev->max_disks = (4096-256)/2;
a654b9d8 1274
71c0805c 1275 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
e11e93fa 1276 mddev->bitmap_file == NULL )
a654b9d8 1277 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1278
f6705578
N		 1279 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1280 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1281 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1282 mddev->new_level = le32_to_cpu(sb->new_level);
1283 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1284 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
f6705578
N		 1285 } else {
1286 mddev->reshape_position = MaxSector;
1287 mddev->delta_disks = 0;
1288 mddev->new_level = mddev->level;
1289 mddev->new_layout = mddev->layout;
664e7c41 1290 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1291 }
1292
41158c7e
N		 1293 } else if (mddev->pers == NULL) {
1294 /* Insist of good event counter while assembling */
1da177e4
LT		 1295 ++ev1;
1296 if (ev1 < mddev->events)
1297 return -EINVAL;
41158c7e
N		 1298 } else if (mddev->bitmap) {
1299 /* If adding to array with a bitmap, then we can accept an
1300 * older device, but not too old.
1301 */
41158c7e
N		 1302 if (ev1 < mddev->bitmap->events_cleared)
1303 return 0;
07d84d10
N		 1304 } else {
1305 if (ev1 < mddev->events)
1306 /* just a hot-add of a new device, leave raid_disk at -1 */
1307 return 0;
1308 }
1da177e4
LT		 1309 if (mddev->level != LEVEL_MULTIPATH) {
1310 int role;
3673f305
N		 1311 if (rdev->desc_nr < 0 ||
1312 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1313 role = 0xffff;
1314 rdev->desc_nr = -1;
1315 } else
1316 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1317 switch(role) {
1318 case 0xffff: /* spare */
1da177e4
LT		 1319 break;
1320 case 0xfffe: /* faulty */
b2d444d7 1321 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1322 break;
1323 default:
5fd6c1dc
N		 1324 if ((le32_to_cpu(sb->feature_map) &
1325 MD_FEATURE_RECOVERY_OFFSET))
1326 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1327 else
1328 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1329 rdev->raid_disk = role;
1330 break;
1331 }
8ddf9efe
N		 1332 if (sb->devflags & WriteMostly1)
1333 set_bit(WriteMostly, &rdev->flags);
41158c7e 1334 } else /* MULTIPATH are always insync */
b2d444d7 1335 set_bit(In_sync, &rdev->flags);
41158c7e 1336
1da177e4
LT		 1337 return 0;
1338}
1339
1340static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1341{
1342 struct mdp_superblock_1 *sb;
1da177e4
LT		 1343 mdk_rdev_t *rdev2;
1344 int max_dev, i;
1345 /* make rdev->sb match mddev and rdev data. */
1346
1347 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1348
1349 sb->feature_map = 0;
1350 sb->pad0 = 0;
5fd6c1dc 1351 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1352 memset(sb->pad1, 0, sizeof(sb->pad1));
1353 memset(sb->pad2, 0, sizeof(sb->pad2));
1354 memset(sb->pad3, 0, sizeof(sb->pad3));
1355
1356 sb->utime = cpu_to_le64((__u64)mddev->utime);
1357 sb->events = cpu_to_le64(mddev->events);
1358 if (mddev->in_sync)
1359 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1360 else
1361 sb->resync_offset = cpu_to_le64(0);
1362
1c05b4bc 1363 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1364
f0ca340c 1365 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1366 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1367 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1368 sb->level = cpu_to_le32(mddev->level);
1369 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1370
a654b9d8
N		 1371 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1372 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1373 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1374 }
5fd6c1dc
N		 1375
1376 if (rdev->raid_disk >= 0 &&
97e4f42d
N		 1377 !test_bit(In_sync, &rdev->flags)) {
1378 if (mddev->curr_resync_completed > rdev->recovery_offset)
1379 rdev->recovery_offset = mddev->curr_resync_completed;
1380 if (rdev->recovery_offset > 0) {
1381 sb->feature_map |=
1382 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1383 sb->recovery_offset =
1384 cpu_to_le64(rdev->recovery_offset);
1385 }
5fd6c1dc
N		 1386 }
1387
f6705578
N		 1388 if (mddev->reshape_position != MaxSector) {
1389 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1390 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1391 sb->new_layout = cpu_to_le32(mddev->new_layout);
1392 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1393 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1394 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
f6705578 1395 }
a654b9d8 1396
1da177e4 1397 max_dev = 0;
159ec1fc 1398 list_for_each_entry(rdev2, &mddev->disks, same_set)
1da177e4
LT		 1399 if (rdev2->desc_nr+1 > max_dev)
1400 max_dev = rdev2->desc_nr+1;
a778b73f
N		 1401
1402 if (max_dev > le32_to_cpu(sb->max_dev))
1403 sb->max_dev = cpu_to_le32(max_dev);
1da177e4
LT		 1404 for (i=0; i<max_dev;i++)
1405 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1406
159ec1fc 1407 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1408 i = rdev2->desc_nr;
b2d444d7 1409 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1410 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1411 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1412 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
5fd6c1dc
N		 1413 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1414 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1415 else
1416 sb->dev_roles[i] = cpu_to_le16(0xffff);
1417 }
1418
1da177e4
LT		 1419 sb->sb_csum = calc_sb_1_csum(sb);
1420}
1421
0cd17fec 1422static unsigned long long
15f4a5fd 1423super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
0cd17fec
CW		 1424{
1425 struct mdp_superblock_1 *sb;
15f4a5fd 1426 sector_t max_sectors;
58c0fed4 1427 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1428 return 0; /* component must fit device */
0f420358 1429 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1430 /* minor versions 1 and 2; superblock before data */
15f4a5fd
AN		 1431 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1432 max_sectors -= rdev->data_offset;
1433 if (!num_sectors || num_sectors > max_sectors)
1434 num_sectors = max_sectors;
0cd17fec
CW		 1435 } else if (rdev->mddev->bitmap_offset) {
1436 /* minor version 0 with bitmap we can't move */
1437 return 0;
1438 } else {
1439 /* minor version 0; superblock after data */
0f420358
AN		 1440 sector_t sb_start;
1441 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1442 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1443 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1444 if (!num_sectors || num_sectors > max_sectors)
1445 num_sectors = max_sectors;
0f420358 1446 rdev->sb_start = sb_start;
0cd17fec
CW		 1447 }
1448 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
15f4a5fd 1449 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1450 sb->super_offset = rdev->sb_start;
0cd17fec 1451 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1452 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1453 rdev->sb_page);
1454 md_super_wait(rdev->mddev);
15f4a5fd 1455 return num_sectors / 2; /* kB for sysfs */
0cd17fec 1456}
1da177e4 1457
75c96f85 1458static struct super_type super_types[] = {
1da177e4
LT		 1459 [0] = {
1460 .name = "0.90.0",
1461 .owner = THIS_MODULE,
0cd17fec
CW		 1462 .load_super = super_90_load,
1463 .validate_super = super_90_validate,
1464 .sync_super = super_90_sync,
1465 .rdev_size_change = super_90_rdev_size_change,
1da177e4
LT		 1466 },
1467 [1] = {
1468 .name = "md-1",
1469 .owner = THIS_MODULE,
0cd17fec
CW		 1470 .load_super = super_1_load,
1471 .validate_super = super_1_validate,
1472 .sync_super = super_1_sync,
1473 .rdev_size_change = super_1_rdev_size_change,
1da177e4
LT		 1474 },
1475};
1da177e4
LT		 1476
1477static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1478{
7dd5e7c3 1479 mdk_rdev_t *rdev, *rdev2;
1da177e4 1480
4b80991c
N		 1481 rcu_read_lock();
1482 rdev_for_each_rcu(rdev, mddev1)
1483 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1484 if (rdev->bdev->bd_contains ==
4b80991c
N		 1485 rdev2->bdev->bd_contains) {
1486 rcu_read_unlock();
7dd5e7c3 1487 return 1;
4b80991c
N		 1488 }
1489 rcu_read_unlock();
1da177e4
LT		 1490 return 0;
1491}
1492
1493static LIST_HEAD(pending_raid_disks);
1494
ac5e7113
AN		 1495/*
1496 * Try to register data integrity profile for an mddev
1497 *
1498 * This is called when an array is started and after a disk has been kicked
1499 * from the array. It only succeeds if all working and active component devices
1500 * are integrity capable with matching profiles.
1501 */
1502int md_integrity_register(mddev_t *mddev)
1503{
1504 mdk_rdev_t *rdev, *reference = NULL;
1505
1506 if (list_empty(&mddev->disks))
1507 return 0; /* nothing to do */
1508 if (blk_get_integrity(mddev->gendisk))
1509 return 0; /* already registered */
1510 list_for_each_entry(rdev, &mddev->disks, same_set) {
1511 /* skip spares and non-functional disks */
1512 if (test_bit(Faulty, &rdev->flags))
1513 continue;
1514 if (rdev->raid_disk < 0)
1515 continue;
1516 /*
1517 * If at least one rdev is not integrity capable, we can not
1518 * enable data integrity for the md device.
1519 */
1520 if (!bdev_get_integrity(rdev->bdev))
1521 return -EINVAL;
1522 if (!reference) {
1523 /* Use the first rdev as the reference */
1524 reference = rdev;
1525 continue;
1526 }
1527 /* does this rdev's profile match the reference profile? */
1528 if (blk_integrity_compare(reference->bdev->bd_disk,
1529 rdev->bdev->bd_disk) < 0)
1530 return -EINVAL;
1531 }
1532 /*
1533 * All component devices are integrity capable and have matching
1534 * profiles, register the common profile for the md device.
1535 */
1536 if (blk_integrity_register(mddev->gendisk,
1537 bdev_get_integrity(reference->bdev)) != 0) {
1538 printk(KERN_ERR "md: failed to register integrity for %s\n",
1539 mdname(mddev));
1540 return -EINVAL;
1541 }
1542 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1543 mdname(mddev));
1544 return 0;
1545}
1546EXPORT_SYMBOL(md_integrity_register);
1547
1548/* Disable data integrity if non-capable/non-matching disk is being added */
1549void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
3f9d99c1 1550{
3f9d99c1 1551 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
ac5e7113 1552 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 1553
ac5e7113 1554 if (!bi_mddev) /* nothing to do */
3f9d99c1 1555 return;
ac5e7113 1556 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 1557 return;
ac5e7113
AN		 1558 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1559 rdev->bdev->bd_disk) >= 0)
1560 return;
1561 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1562 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 1563}
ac5e7113 1564EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 1565
1da177e4
LT		 1566static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1567{
7dd5e7c3 1568 char b[BDEVNAME_SIZE];
f637b9f9 1569 struct kobject *ko;
1edf80d3 1570 char *s;
5e55e2f5 1571 int err;
1da177e4
LT		 1572
1573 if (rdev->mddev) {
1574 MD_BUG();
1575 return -EINVAL;
1576 }
11e2ede0
DW		 1577
1578 /* prevent duplicates */
1579 if (find_rdev(mddev, rdev->bdev->bd_dev))
1580 return -EEXIST;
1581
dd8ac336
AN		 1582 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1583 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1584 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 1585 if (mddev->pers) {
1586 /* Cannot change size, so fail
1587 * If mddev->level <= 0, then we don't care
1588 * about aligning sizes (e.g. linear)
1589 */
1590 if (mddev->level > 0)
1591 return -ENOSPC;
1592 } else
dd8ac336 1593 mddev->dev_sectors = rdev->sectors;
2bf071bf 1594 }
1da177e4
LT		 1595
1596 /* Verify rdev->desc_nr is unique.
1597 * If it is -1, assign a free number, else
1598 * check number is not in use
1599 */
1600 if (rdev->desc_nr < 0) {
1601 int choice = 0;
1602 if (mddev->pers) choice = mddev->raid_disks;
1603 while (find_rdev_nr(mddev, choice))
1604 choice++;
1605 rdev->desc_nr = choice;
1606 } else {
1607 if (find_rdev_nr(mddev, rdev->desc_nr))
1608 return -EBUSY;
1609 }
de01dfad
N		 1610 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1611 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1612 mdname(mddev), mddev->max_disks);
1613 return -EBUSY;
1614 }
19133a42 1615 bdevname(rdev->bdev,b);
649316b2 1616 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 1617 *s = '!';
649316b2 1618
1da177e4 1619 rdev->mddev = mddev;
19133a42 1620 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1621
b2d6db58 1622 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 1623 goto fail;
86e6ffdd 1624
0762b8bd 1625 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
5e55e2f5
N		 1626 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1627 kobject_del(&rdev->kobj);
1628 goto fail;
1629 }
3c0ee63a
N		 1630 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1631
4b80991c 1632 list_add_rcu(&rdev->same_set, &mddev->disks);
c5d79adb 1633 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
4044ba58
N		 1634
1635 /* May as well allow recovery to be retried once */
1636 mddev->recovery_disabled = 0;
3f9d99c1 1637
1da177e4 1638 return 0;
5e55e2f5
N		 1639
1640 fail:
1641 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1642 b, mdname(mddev));
1643 return err;
1da177e4
LT		 1644}
1645
177a99b2 1646static void md_delayed_delete(struct work_struct *ws)
5792a285
N		 1647{
1648 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1649 kobject_del(&rdev->kobj);
177a99b2 1650 kobject_put(&rdev->kobj);
5792a285
N		 1651}
1652
1da177e4
LT		 1653static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1654{
1655 char b[BDEVNAME_SIZE];
1656 if (!rdev->mddev) {
1657 MD_BUG();
1658 return;
1659 }
5463c790 1660 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
4b80991c 1661 list_del_rcu(&rdev->same_set);
1da177e4
LT		 1662 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1663 rdev->mddev = NULL;
86e6ffdd 1664 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 1665 sysfs_put(rdev->sysfs_state);
1666 rdev->sysfs_state = NULL;
5792a285 1667 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 1668 * writing to 'remove' to "dev/state". We also need
1669 * to delay it due to rcu usage.
5792a285 1670 */
4b80991c 1671 synchronize_rcu();
177a99b2
N		 1672 INIT_WORK(&rdev->del_work, md_delayed_delete);
1673 kobject_get(&rdev->kobj);
5792a285 1674 schedule_work(&rdev->del_work);
1da177e4
LT		 1675}
1676
1677/*
1678 * prevent the device from being mounted, repartitioned or
1679 * otherwise reused by a RAID array (or any other kernel
1680 * subsystem), by bd_claiming the device.
1681 */
c5d79adb 1682static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1da177e4
LT		 1683{
1684 int err = 0;
1685 struct block_device *bdev;
1686 char b[BDEVNAME_SIZE];
1687
2e7b651d 1688 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1689 if (IS_ERR(bdev)) {
1690 printk(KERN_ERR "md: could not open %s.\n",
1691 __bdevname(dev, b));
1692 return PTR_ERR(bdev);
1693 }
c5d79adb 1694 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1da177e4
LT		 1695 if (err) {
1696 printk(KERN_ERR "md: could not bd_claim %s.\n",
1697 bdevname(bdev, b));
9a1c3542 1698 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1699 return err;
1700 }
c5d79adb
N		 1701 if (!shared)
1702 set_bit(AllReserved, &rdev->flags);
1da177e4
LT		 1703 rdev->bdev = bdev;
1704 return err;
1705}
1706
1707static void unlock_rdev(mdk_rdev_t *rdev)
1708{
1709 struct block_device *bdev = rdev->bdev;
1710 rdev->bdev = NULL;
1711 if (!bdev)
1712 MD_BUG();
1713 bd_release(bdev);
9a1c3542 1714 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1715}
1716
1717void md_autodetect_dev(dev_t dev);
1718
1719static void export_rdev(mdk_rdev_t * rdev)
1720{
1721 char b[BDEVNAME_SIZE];
1722 printk(KERN_INFO "md: export_rdev(%s)\n",
1723 bdevname(rdev->bdev,b));
1724 if (rdev->mddev)
1725 MD_BUG();
1726 free_disk_sb(rdev);
1da177e4 1727#ifndef MODULE
d0fae18f
N		 1728 if (test_bit(AutoDetected, &rdev->flags))
1729 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 1730#endif
1731 unlock_rdev(rdev);
86e6ffdd 1732 kobject_put(&rdev->kobj);
1da177e4
LT		 1733}
1734
1735static void kick_rdev_from_array(mdk_rdev_t * rdev)
1736{
1737 unbind_rdev_from_array(rdev);
1738 export_rdev(rdev);
1739}
1740
1741static void export_array(mddev_t *mddev)
1742{
159ec1fc 1743 mdk_rdev_t *rdev, *tmp;
1da177e4 1744
d089c6af 1745 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 1746 if (!rdev->mddev) {
1747 MD_BUG();
1748 continue;
1749 }
1750 kick_rdev_from_array(rdev);
1751 }
1752 if (!list_empty(&mddev->disks))
1753 MD_BUG();
1754 mddev->raid_disks = 0;
1755 mddev->major_version = 0;
1756}
1757
1758static void print_desc(mdp_disk_t *desc)
1759{
1760 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1761 desc->major,desc->minor,desc->raid_disk,desc->state);
1762}
1763
cd2ac932 1764static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 1765{
1766 int i;
1767
1768 printk(KERN_INFO
1769 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1770 sb->major_version, sb->minor_version, sb->patch_version,
1771 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1772 sb->ctime);
1773 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1774 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1775 sb->md_minor, sb->layout, sb->chunk_size);
1776 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1777 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1778 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1779 sb->failed_disks, sb->spare_disks,
1780 sb->sb_csum, (unsigned long)sb->events_lo);
1781
1782 printk(KERN_INFO);
1783 for (i = 0; i < MD_SB_DISKS; i++) {
1784 mdp_disk_t *desc;
1785
1786 desc = sb->disks + i;
1787 if (desc->number || desc->major || desc->minor ||
1788 desc->raid_disk || (desc->state && (desc->state != 4))) {
1789 printk(" D %2d: ", i);
1790 print_desc(desc);
1791 }
1792 }
1793 printk(KERN_INFO "md: THIS: ");
1794 print_desc(&sb->this_disk);
cd2ac932 1795}
1da177e4 1796
cd2ac932
CR		 1797static void print_sb_1(struct mdp_superblock_1 *sb)
1798{
1799 __u8 *uuid;
1800
1801 uuid = sb->set_uuid;
ad361c98
JP		 1802 printk(KERN_INFO
1803 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1804 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1805 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 1806 le32_to_cpu(sb->major_version),
1807 le32_to_cpu(sb->feature_map),
1808 uuid[0], uuid[1], uuid[2], uuid[3],
1809 uuid[4], uuid[5], uuid[6], uuid[7],
1810 uuid[8], uuid[9], uuid[10], uuid[11],
1811 uuid[12], uuid[13], uuid[14], uuid[15],
1812 sb->set_name,
1813 (unsigned long long)le64_to_cpu(sb->ctime)
1814 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1815
1816 uuid = sb->device_uuid;
ad361c98
JP		 1817 printk(KERN_INFO
1818 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 1819 " RO:%llu\n"
ad361c98
JP		 1820 "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1821 ":%02x%02x%02x%02x%02x%02x\n"
1822 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1823 "md: (MaxDev:%u) \n",
cd2ac932
CR		 1824 le32_to_cpu(sb->level),
1825 (unsigned long long)le64_to_cpu(sb->size),
1826 le32_to_cpu(sb->raid_disks),
1827 le32_to_cpu(sb->layout),
1828 le32_to_cpu(sb->chunksize),
1829 (unsigned long long)le64_to_cpu(sb->data_offset),
1830 (unsigned long long)le64_to_cpu(sb->data_size),
1831 (unsigned long long)le64_to_cpu(sb->super_offset),
1832 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1833 le32_to_cpu(sb->dev_number),
1834 uuid[0], uuid[1], uuid[2], uuid[3],
1835 uuid[4], uuid[5], uuid[6], uuid[7],
1836 uuid[8], uuid[9], uuid[10], uuid[11],
1837 uuid[12], uuid[13], uuid[14], uuid[15],
1838 sb->devflags,
1839 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1840 (unsigned long long)le64_to_cpu(sb->events),
1841 (unsigned long long)le64_to_cpu(sb->resync_offset),
1842 le32_to_cpu(sb->sb_csum),
1843 le32_to_cpu(sb->max_dev)
1844 );
1da177e4
LT		 1845}
1846
cd2ac932 1847static void print_rdev(mdk_rdev_t *rdev, int major_version)
1da177e4
LT		 1848{
1849 char b[BDEVNAME_SIZE];
dd8ac336
AN		 1850 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1851 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 1852 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1853 rdev->desc_nr);
1da177e4 1854 if (rdev->sb_loaded) {
cd2ac932
CR		 1855 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1856 switch (major_version) {
1857 case 0:
1858 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1859 break;
1860 case 1:
1861 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1862 break;
1863 }
1da177e4
LT		 1864 } else
1865 printk(KERN_INFO "md: no rdev superblock!\n");
1866}
1867
5e56341d 1868static void md_print_devices(void)
1da177e4 1869{
159ec1fc 1870 struct list_head *tmp;
1da177e4
LT		 1871 mdk_rdev_t *rdev;
1872 mddev_t *mddev;
1873 char b[BDEVNAME_SIZE];
1874
1875 printk("\n");
1876 printk("md: **********************************\n");
1877 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1878 printk("md: **********************************\n");
29ac4aa3 1879 for_each_mddev(mddev, tmp) {
1da177e4 1880
32a7627c
N		 1881 if (mddev->bitmap)
1882 bitmap_print_sb(mddev->bitmap);
1883 else
1884 printk("%s: ", mdname(mddev));
159ec1fc 1885 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 1886 printk("<%s>", bdevname(rdev->bdev,b));
1887 printk("\n");
1888
159ec1fc 1889 list_for_each_entry(rdev, &mddev->disks, same_set)
cd2ac932 1890 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 1891 }
1892 printk("md: **********************************\n");
1893 printk("\n");
1894}
1895
1896
42543769 1897static void sync_sbs(mddev_t * mddev, int nospares)
1da177e4 1898{
42543769
N		 1899 /* Update each superblock (in-memory image), but
1900 * if we are allowed to, skip spares which already
1901 * have the right event counter, or have one earlier
1902 * (which would mean they aren't being marked as dirty
1903 * with the rest of the array)
1904 */
1da177e4 1905 mdk_rdev_t *rdev;
1da177e4 1906
159ec1fc 1907 list_for_each_entry(rdev, &mddev->disks, same_set) {
42543769
N		 1908 if (rdev->sb_events == mddev->events ||
1909 (nospares &&
1910 rdev->raid_disk < 0 &&
1911 (rdev->sb_events&1)==0 &&
1912 rdev->sb_events+1 == mddev->events)) {
1913 /* Don't update this superblock */
1914 rdev->sb_loaded = 2;
1915 } else {
1916 super_types[mddev->major_version].
1917 sync_super(mddev, rdev);
1918 rdev->sb_loaded = 1;
1919 }
1da177e4
LT		 1920 }
1921}
1922
850b2b42 1923static void md_update_sb(mddev_t * mddev, int force_change)
1da177e4 1924{
1da177e4 1925 mdk_rdev_t *rdev;
06d91a5f 1926 int sync_req;
42543769 1927 int nospares = 0;
1da177e4 1928
1b57f132 1929 mddev->utime = get_seconds();
8377bc80
N		 1930 if (mddev->external)
1931 return;
1da177e4 1932repeat:
a9701a30 1933 spin_lock_irq(&mddev->write_lock);
84692195 1934
850b2b42
N		 1935 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1936 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1937 force_change = 1;
1938 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1939 /* just a clean<-> dirty transition, possibly leave spares alone,
1940 * though if events isn't the right even/odd, we will have to do
1941 * spares after all
1942 */
1943 nospares = 1;
1944 if (force_change)
1945 nospares = 0;
1946 if (mddev->degraded)
84692195
N		 1947 /* If the array is degraded, then skipping spares is both
1948 * dangerous and fairly pointless.
1949 * Dangerous because a device that was removed from the array
1950 * might have a event_count that still looks up-to-date,
1951 * so it can be re-added without a resync.
1952 * Pointless because if there are any spares to skip,
1953 * then a recovery will happen and soon that array won't
1954 * be degraded any more and the spare can go back to sleep then.
1955 */
850b2b42 1956 nospares = 0;
84692195 1957
06d91a5f 1958 sync_req = mddev->in_sync;
42543769
N		 1959
1960 /* If this is just a dirty<->clean transition, and the array is clean
1961 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 1962 if (nospares
42543769 1963 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1031be7a
N		 1964 && (mddev->events & 1)
1965 && mddev->events != 1)
42543769
N		 1966 mddev->events--;
1967 else {
1968 /* otherwise we have to go forward and ... */
1969 mddev->events ++;
1970 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1971 /* .. if the array isn't clean, insist on an odd 'events' */
1972 if ((mddev->events&1)==0) {
1973 mddev->events++;
1974 nospares = 0;
1975 }
1976 } else {
1977 /* otherwise insist on an even 'events' (for clean states) */
1978 if ((mddev->events&1)) {
1979 mddev->events++;
1980 nospares = 0;
1981 }
1982 }
1983 }
1da177e4
LT		 1984
1985 if (!mddev->events) {
1986 /*
1987 * oops, this 64-bit counter should never wrap.
1988 * Either we are in around ~1 trillion A.C., assuming
1989 * 1 reboot per second, or we have a bug:
1990 */
1991 MD_BUG();
1992 mddev->events --;
1993 }
1da177e4
LT		 1994
1995 /*
1996 * do not write anything to disk if using
1997 * nonpersistent superblocks
1998 */
06d91a5f 1999 if (!mddev->persistent) {
e691063a
N		 2000 if (!mddev->external)
2001 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2002
a9701a30 2003 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2004 wake_up(&mddev->sb_wait);
1da177e4 2005 return;
06d91a5f 2006 }
e691063a 2007 sync_sbs(mddev, nospares);
a9701a30 2008 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 2009
2010 dprintk(KERN_INFO
2011 "md: updating %s RAID superblock on device (in sync %d)\n",
2012 mdname(mddev),mddev->in_sync);
2013
4ad13663 2014 bitmap_update_sb(mddev->bitmap);
159ec1fc 2015 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 2016 char b[BDEVNAME_SIZE];
2017 dprintk(KERN_INFO "md: ");
42543769
N		 2018 if (rdev->sb_loaded != 1)
2019 continue; /* no noise on spare devices */
b2d444d7 2020 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2021 dprintk("(skipping faulty ");
2022
2023 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 2024 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2025 md_super_write(mddev,rdev,
0f420358 2026 rdev->sb_start, rdev->sb_size,
7bfa19f2
N		 2027 rdev->sb_page);
2028 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2029 bdevname(rdev->bdev,b),
0f420358 2030 (unsigned long long)rdev->sb_start);
42543769 2031 rdev->sb_events = mddev->events;
7bfa19f2 2032
1da177e4
LT		 2033 } else
2034 dprintk(")\n");
7bfa19f2 2035 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2036 /* only need to write one superblock... */
2037 break;
2038 }
a9701a30 2039 md_super_wait(mddev);
850b2b42 2040 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2041
a9701a30 2042 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2043 if (mddev->in_sync != sync_req ||
2044 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2045 /* have to write it out again */
a9701a30 2046 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2047 goto repeat;
2048 }
850b2b42 2049 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2050 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2051 wake_up(&mddev->sb_wait);
acb180b0
N		 2052 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2053 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2054
1da177e4
LT		 2055}
2056
7f6ce769 2057/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2058 * We want to accept with case. For this we use cmd_match.
2059 */
2060static int cmd_match(const char *cmd, const char *str)
2061{
2062 /* See if cmd, written into a sysfs file, matches
2063 * str. They must either be the same, or cmd can
2064 * have a trailing newline
2065 */
2066 while (*cmd && *str && *cmd == *str) {
2067 cmd++;
2068 str++;
2069 }
2070 if (*cmd == '\n')
2071 cmd++;
2072 if (*str || *cmd)
2073 return 0;
2074 return 1;
2075}
2076
86e6ffdd
N		 2077struct rdev_sysfs_entry {
2078 struct attribute attr;
2079 ssize_t (*show)(mdk_rdev_t *, char *);
2080 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2081};
2082
2083static ssize_t
96de1e66 2084state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 2085{
2086 char *sep = "";
20a49ff6 2087 size_t len = 0;
86e6ffdd 2088
b2d444d7 2089 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 2090 len+= sprintf(page+len, "%sfaulty",sep);
2091 sep = ",";
2092 }
b2d444d7 2093 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2094 len += sprintf(page+len, "%sin_sync",sep);
2095 sep = ",";
2096 }
f655675b
N		 2097 if (test_bit(WriteMostly, &rdev->flags)) {
2098 len += sprintf(page+len, "%swrite_mostly",sep);
2099 sep = ",";
2100 }
6bfe0b49
DW		 2101 if (test_bit(Blocked, &rdev->flags)) {
2102 len += sprintf(page+len, "%sblocked", sep);
2103 sep = ",";
2104 }
b2d444d7
N		 2105 if (!test_bit(Faulty, &rdev->flags) &&
2106 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2107 len += sprintf(page+len, "%sspare", sep);
2108 sep = ",";
2109 }
2110 return len+sprintf(page+len, "\n");
2111}
2112
45dc2de1
N		 2113static ssize_t
2114state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2115{
2116 /* can write
2117 * faulty - simulates and error
2118 * remove - disconnects the device
f655675b
N		 2119 * writemostly - sets write_mostly
2120 * -writemostly - clears write_mostly
6bfe0b49
DW		 2121 * blocked - sets the Blocked flag
2122 * -blocked - clears the Blocked flag
6d56e278 2123 * insync - sets Insync providing device isn't active
45dc2de1
N		 2124 */
2125 int err = -EINVAL;
2126 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2127 md_error(rdev->mddev, rdev);
2128 err = 0;
2129 } else if (cmd_match(buf, "remove")) {
2130 if (rdev->raid_disk >= 0)
2131 err = -EBUSY;
2132 else {
2133 mddev_t *mddev = rdev->mddev;
2134 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2135 if (mddev->pers)
2136 md_update_sb(mddev, 1);
45dc2de1
N		 2137 md_new_event(mddev);
2138 err = 0;
2139 }
f655675b
N		 2140 } else if (cmd_match(buf, "writemostly")) {
2141 set_bit(WriteMostly, &rdev->flags);
2142 err = 0;
2143 } else if (cmd_match(buf, "-writemostly")) {
2144 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2145 err = 0;
2146 } else if (cmd_match(buf, "blocked")) {
2147 set_bit(Blocked, &rdev->flags);
2148 err = 0;
2149 } else if (cmd_match(buf, "-blocked")) {
2150 clear_bit(Blocked, &rdev->flags);
2151 wake_up(&rdev->blocked_wait);
2152 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2153 md_wakeup_thread(rdev->mddev->thread);
2154
6d56e278
N		 2155 err = 0;
2156 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2157 set_bit(In_sync, &rdev->flags);
f655675b 2158 err = 0;
45dc2de1 2159 }
3c0ee63a
N		 2160 if (!err && rdev->sysfs_state)
2161 sysfs_notify_dirent(rdev->sysfs_state);
45dc2de1
N		 2162 return err ? err : len;
2163}
80ca3a44
N		 2164static struct rdev_sysfs_entry rdev_state =
2165__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2166
4dbcdc75
N		 2167static ssize_t
2168errors_show(mdk_rdev_t *rdev, char *page)
2169{
2170 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2171}
2172
2173static ssize_t
2174errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2175{
2176 char *e;
2177 unsigned long n = simple_strtoul(buf, &e, 10);
2178 if (*buf && (*e == 0 || *e == '\n')) {
2179 atomic_set(&rdev->corrected_errors, n);
2180 return len;
2181 }
2182 return -EINVAL;
2183}
2184static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2185__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2186
014236d2
N		 2187static ssize_t
2188slot_show(mdk_rdev_t *rdev, char *page)
2189{
2190 if (rdev->raid_disk < 0)
2191 return sprintf(page, "none\n");
2192 else
2193 return sprintf(page, "%d\n", rdev->raid_disk);
2194}
2195
2196static ssize_t
2197slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2198{
2199 char *e;
c303da6d
N		 2200 int err;
2201 char nm[20];
014236d2
N		 2202 int slot = simple_strtoul(buf, &e, 10);
2203 if (strncmp(buf, "none", 4)==0)
2204 slot = -1;
2205 else if (e==buf || (*e && *e!= '\n'))
2206 return -EINVAL;
6c2fce2e 2207 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2208 /* Setting 'slot' on an active array requires also
2209 * updating the 'rd%d' link, and communicating
2210 * with the personality with ->hot_*_disk.
2211 * For now we only support removing
2212 * failed/spare devices. This normally happens automatically,
2213 * but not when the metadata is externally managed.
2214 */
c303da6d
N		 2215 if (rdev->raid_disk == -1)
2216 return -EEXIST;
2217 /* personality does all needed checks */
2218 if (rdev->mddev->pers->hot_add_disk == NULL)
2219 return -EINVAL;
2220 err = rdev->mddev->pers->
2221 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2222 if (err)
2223 return err;
2224 sprintf(nm, "rd%d", rdev->raid_disk);
2225 sysfs_remove_link(&rdev->mddev->kobj, nm);
2226 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2227 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e
NB		 2228 } else if (rdev->mddev->pers) {
2229 mdk_rdev_t *rdev2;
6c2fce2e 2230 /* Activating a spare .. or possibly reactivating
6d56e278 2231 * if we ever get bitmaps working here.
6c2fce2e
NB		 2232 */
2233
2234 if (rdev->raid_disk != -1)
2235 return -EBUSY;
2236
2237 if (rdev->mddev->pers->hot_add_disk == NULL)
2238 return -EINVAL;
2239
159ec1fc 2240 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
6c2fce2e
NB		 2241 if (rdev2->raid_disk == slot)
2242 return -EEXIST;
2243
2244 rdev->raid_disk = slot;
2245 if (test_bit(In_sync, &rdev->flags))
2246 rdev->saved_raid_disk = slot;
2247 else
2248 rdev->saved_raid_disk = -1;
2249 err = rdev->mddev->pers->
2250 hot_add_disk(rdev->mddev, rdev);
199050ea 2251 if (err) {
6c2fce2e 2252 rdev->raid_disk = -1;
6c2fce2e 2253 return err;
52664732 2254 } else
3c0ee63a 2255 sysfs_notify_dirent(rdev->sysfs_state);
6c2fce2e
NB		 2256 sprintf(nm, "rd%d", rdev->raid_disk);
2257 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2258 printk(KERN_WARNING
2259 "md: cannot register "
2260 "%s for %s\n",
2261 nm, mdname(rdev->mddev));
2262
2263 /* don't wakeup anyone, leave that to userspace. */
c303da6d
N		 2264 } else {
2265 if (slot >= rdev->mddev->raid_disks)
2266 return -ENOSPC;
2267 rdev->raid_disk = slot;
2268 /* assume it is working */
c5d79adb
N		 2269 clear_bit(Faulty, &rdev->flags);
2270 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2271 set_bit(In_sync, &rdev->flags);
3c0ee63a 2272 sysfs_notify_dirent(rdev->sysfs_state);
c303da6d 2273 }
014236d2
N		 2274 return len;
2275}
2276
2277
2278static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2279__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2280
93c8cad0
N		 2281static ssize_t
2282offset_show(mdk_rdev_t *rdev, char *page)
2283{
6961ece4 2284 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2285}
2286
2287static ssize_t
2288offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2289{
2290 char *e;
2291 unsigned long long offset = simple_strtoull(buf, &e, 10);
2292 if (e==buf || (*e && *e != '\n'))
2293 return -EINVAL;
8ed0a521 2294 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2295 return -EBUSY;
dd8ac336 2296 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2297 /* Must set offset before size, so overlap checks
2298 * can be sane */
2299 return -EBUSY;
93c8cad0
N		 2300 rdev->data_offset = offset;
2301 return len;
2302}
2303
2304static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2305__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2306
83303b61
N		 2307static ssize_t
2308rdev_size_show(mdk_rdev_t *rdev, char *page)
2309{
dd8ac336 2310 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2311}
2312
c5d79adb
N		 2313static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2314{
2315 /* check if two start/length pairs overlap */
2316 if (s1+l1 <= s2)
2317 return 0;
2318 if (s2+l2 <= s1)
2319 return 0;
2320 return 1;
2321}
2322
b522adcd
DW		 2323static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2324{
2325 unsigned long long blocks;
2326 sector_t new;
2327
2328 if (strict_strtoull(buf, 10, &blocks) < 0)
2329 return -EINVAL;
2330
2331 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2332 return -EINVAL; /* sector conversion overflow */
2333
2334 new = blocks * 2;
2335 if (new != blocks * 2)
2336 return -EINVAL; /* unsigned long long to sector_t overflow */
2337
2338 *sectors = new;
2339 return 0;
2340}
2341
83303b61
N		 2342static ssize_t
2343rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2344{
27c529bb 2345 mddev_t *my_mddev = rdev->mddev;
dd8ac336 2346 sector_t oldsectors = rdev->sectors;
b522adcd 2347 sector_t sectors;
27c529bb 2348
b522adcd 2349 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2350 return -EINVAL;
0cd17fec 2351 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2352 if (my_mddev->persistent) {
dd8ac336
AN		 2353 sectors = super_types[my_mddev->major_version].
2354 rdev_size_change(rdev, sectors);
2355 if (!sectors)
0cd17fec 2356 return -EBUSY;
dd8ac336
AN		 2357 } else if (!sectors)
2358 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2359 rdev->data_offset;
0cd17fec 2360 }
dd8ac336 2361 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2362 return -EINVAL; /* component must fit device */
0cd17fec 2363
dd8ac336
AN		 2364 rdev->sectors = sectors;
2365 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 2366 /* need to check that all other rdevs with the same ->bdev
2367 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 2368 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 2369 * we have to change it back, we will have the lock again.
2370 */
2371 mddev_t *mddev;
2372 int overlap = 0;
159ec1fc 2373 struct list_head *tmp;
c5d79adb 2374
27c529bb 2375 mddev_unlock(my_mddev);
29ac4aa3 2376 for_each_mddev(mddev, tmp) {
c5d79adb
N		 2377 mdk_rdev_t *rdev2;
2378
2379 mddev_lock(mddev);
159ec1fc 2380 list_for_each_entry(rdev2, &mddev->disks, same_set)
c5d79adb
N		 2381 if (test_bit(AllReserved, &rdev2->flags) ||
2382 (rdev->bdev == rdev2->bdev &&
2383 rdev != rdev2 &&
dd8ac336 2384 overlaps(rdev->data_offset, rdev->sectors,
d07bd3bc 2385 rdev2->data_offset,
dd8ac336 2386 rdev2->sectors))) {
c5d79adb
N		 2387 overlap = 1;
2388 break;
2389 }
2390 mddev_unlock(mddev);
2391 if (overlap) {
2392 mddev_put(mddev);
2393 break;
2394 }
2395 }
27c529bb 2396 mddev_lock(my_mddev);
c5d79adb
N		 2397 if (overlap) {
2398 /* Someone else could have slipped in a size
2399 * change here, but doing so is just silly.
dd8ac336 2400 * We put oldsectors back because we *know* it is
c5d79adb
N		 2401 * safe, and trust userspace not to race with
2402 * itself
2403 */
dd8ac336 2404 rdev->sectors = oldsectors;
c5d79adb
N		 2405 return -EBUSY;
2406 }
2407 }
83303b61
N		 2408 return len;
2409}
2410
2411static struct rdev_sysfs_entry rdev_size =
80ca3a44 2412__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2413
86e6ffdd
N		 2414static struct attribute *rdev_default_attrs[] = {
2415 &rdev_state.attr,
4dbcdc75 2416 &rdev_errors.attr,
014236d2 2417 &rdev_slot.attr,
93c8cad0 2418 &rdev_offset.attr,
83303b61 2419 &rdev_size.attr,
86e6ffdd
N		 2420 NULL,
2421};
2422static ssize_t
2423rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2424{
2425 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2426 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2427 mddev_t *mddev = rdev->mddev;
2428 ssize_t rv;
86e6ffdd
N		 2429
2430 if (!entry->show)
2431 return -EIO;
27c529bb
N		 2432
2433 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2434 if (!rv) {
2435 if (rdev->mddev == NULL)
2436 rv = -EBUSY;
2437 else
2438 rv = entry->show(rdev, page);
2439 mddev_unlock(mddev);
2440 }
2441 return rv;
86e6ffdd
N		 2442}
2443
2444static ssize_t
2445rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2446 const char *page, size_t length)
2447{
2448 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2449 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2450 ssize_t rv;
2451 mddev_t *mddev = rdev->mddev;
86e6ffdd
N		 2452
2453 if (!entry->store)
2454 return -EIO;
67463acb
N		 2455 if (!capable(CAP_SYS_ADMIN))
2456 return -EACCES;
27c529bb 2457 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 2458 if (!rv) {
27c529bb
N		 2459 if (rdev->mddev == NULL)
2460 rv = -EBUSY;
2461 else
2462 rv = entry->store(rdev, page, length);
6a51830e 2463 mddev_unlock(mddev);
ca388059
N		 2464 }
2465 return rv;
86e6ffdd
N		 2466}
2467
2468static void rdev_free(struct kobject *ko)
2469{
2470 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2471 kfree(rdev);
2472}
2473static struct sysfs_ops rdev_sysfs_ops = {
2474 .show = rdev_attr_show,
2475 .store = rdev_attr_store,
2476};
2477static struct kobj_type rdev_ktype = {
2478 .release = rdev_free,
2479 .sysfs_ops = &rdev_sysfs_ops,
2480 .default_attrs = rdev_default_attrs,
2481};
2482
1da177e4
LT		 2483/*
2484 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2485 *
2486 * mark the device faulty if:
2487 *
2488 * - the device is nonexistent (zero size)
2489 * - the device has no valid superblock
2490 *
2491 * a faulty rdev _never_ has rdev->sb set.
2492 */
2493static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2494{
2495 char b[BDEVNAME_SIZE];
2496 int err;
2497 mdk_rdev_t *rdev;
2498 sector_t size;
2499
9ffae0cf 2500 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 2501 if (!rdev) {
2502 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2503 return ERR_PTR(-ENOMEM);
2504 }
1da177e4
LT		 2505
2506 if ((err = alloc_disk_sb(rdev)))
2507 goto abort_free;
2508
c5d79adb 2509 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 2510 if (err)
2511 goto abort_free;
2512
f9cb074b 2513 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 2514
1da177e4 2515 rdev->desc_nr = -1;
2b6e8459 2516 rdev->saved_raid_disk = -1;
3f9d7b0d 2517 rdev->raid_disk = -1;
b2d444d7 2518 rdev->flags = 0;
1da177e4 2519 rdev->data_offset = 0;
42543769 2520 rdev->sb_events = 0;
1da177e4 2521 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 2522 atomic_set(&rdev->read_errors, 0);
4dbcdc75 2523 atomic_set(&rdev->corrected_errors, 0);
1da177e4
LT		 2524
2525 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2526 if (!size) {
2527 printk(KERN_WARNING
2528 "md: %s has zero or unknown size, marking faulty!\n",
2529 bdevname(rdev->bdev,b));
2530 err = -EINVAL;
2531 goto abort_free;
2532 }
2533
2534 if (super_format >= 0) {
2535 err = super_types[super_format].
2536 load_super(rdev, NULL, super_minor);
2537 if (err == -EINVAL) {
df968c4e
N		 2538 printk(KERN_WARNING
2539 "md: %s does not have a valid v%d.%d "
2540 "superblock, not importing!\n",
2541 bdevname(rdev->bdev,b),
2542 super_format, super_minor);
1da177e4
LT		 2543 goto abort_free;
2544 }
2545 if (err < 0) {
2546 printk(KERN_WARNING
2547 "md: could not read %s's sb, not importing!\n",
2548 bdevname(rdev->bdev,b));
2549 goto abort_free;
2550 }
2551 }
6bfe0b49 2552
1da177e4 2553 INIT_LIST_HEAD(&rdev->same_set);
6bfe0b49 2554 init_waitqueue_head(&rdev->blocked_wait);
1da177e4
LT		 2555
2556 return rdev;
2557
2558abort_free:
2559 if (rdev->sb_page) {
2560 if (rdev->bdev)
2561 unlock_rdev(rdev);
2562 free_disk_sb(rdev);
2563 }
2564 kfree(rdev);
2565 return ERR_PTR(err);
2566}
2567
2568/*
2569 * Check a full RAID array for plausibility
2570 */
2571
2572
a757e64c 2573static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2574{
2575 int i;
159ec1fc 2576 mdk_rdev_t *rdev, *freshest, *tmp;
1da177e4
LT		 2577 char b[BDEVNAME_SIZE];
2578
2579 freshest = NULL;
d089c6af 2580 rdev_for_each(rdev, tmp, mddev)
1da177e4
LT		 2581 switch (super_types[mddev->major_version].
2582 load_super(rdev, freshest, mddev->minor_version)) {
2583 case 1:
2584 freshest = rdev;
2585 break;
2586 case 0:
2587 break;
2588 default:
2589 printk( KERN_ERR \
2590 "md: fatal superblock inconsistency in %s"
2591 " -- removing from array\n",
2592 bdevname(rdev->bdev,b));
2593 kick_rdev_from_array(rdev);
2594 }
2595
2596
2597 super_types[mddev->major_version].
2598 validate_super(mddev, freshest);
2599
2600 i = 0;
d089c6af 2601 rdev_for_each(rdev, tmp, mddev) {
de01dfad
N		 2602 if (rdev->desc_nr >= mddev->max_disks ||
2603 i > mddev->max_disks) {
2604 printk(KERN_WARNING
2605 "md: %s: %s: only %d devices permitted\n",
2606 mdname(mddev), bdevname(rdev->bdev, b),
2607 mddev->max_disks);
2608 kick_rdev_from_array(rdev);
2609 continue;
2610 }
1da177e4
LT		 2611 if (rdev != freshest)
2612 if (super_types[mddev->major_version].
2613 validate_super(mddev, rdev)) {
2614 printk(KERN_WARNING "md: kicking non-fresh %s"
2615 " from array!\n",
2616 bdevname(rdev->bdev,b));
2617 kick_rdev_from_array(rdev);
2618 continue;
2619 }
2620 if (mddev->level == LEVEL_MULTIPATH) {
2621 rdev->desc_nr = i++;
2622 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2623 set_bit(In_sync, &rdev->flags);
a778b73f
N		 2624 } else if (rdev->raid_disk >= mddev->raid_disks) {
2625 rdev->raid_disk = -1;
2626 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 2627 }
2628 }
1da177e4
LT		 2629}
2630
19052c0e
N		 2631static void md_safemode_timeout(unsigned long data);
2632
16f17b39
N		 2633static ssize_t
2634safe_delay_show(mddev_t *mddev, char *page)
2635{
2636 int msec = (mddev->safemode_delay*1000)/HZ;
2637 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2638}
2639static ssize_t
2640safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2641{
2642 int scale=1;
2643 int dot=0;
2644 int i;
2645 unsigned long msec;
2646 char buf[30];
97ce0a7f 2647
16f17b39
N		 2648 /* remove a period, and count digits after it */
2649 if (len >= sizeof(buf))
2650 return -EINVAL;
97ce0a7f 2651 strlcpy(buf, cbuf, sizeof(buf));
16f17b39
N		 2652 for (i=0; i<len; i++) {
2653 if (dot) {
2654 if (isdigit(buf[i])) {
2655 buf[i-1] = buf[i];
2656 scale *= 10;
2657 }
2658 buf[i] = 0;
2659 } else if (buf[i] == '.') {
2660 dot=1;
2661 buf[i] = 0;
2662 }
2663 }
97ce0a7f 2664 if (strict_strtoul(buf, 10, &msec) < 0)
16f17b39
N		 2665 return -EINVAL;
2666 msec = (msec * 1000) / scale;
2667 if (msec == 0)
2668 mddev->safemode_delay = 0;
2669 else {
19052c0e 2670 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 2671 mddev->safemode_delay = (msec*HZ)/1000;
2672 if (mddev->safemode_delay == 0)
2673 mddev->safemode_delay = 1;
19052c0e
N		 2674 if (mddev->safemode_delay < old_delay)
2675 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 2676 }
2677 return len;
2678}
2679static struct md_sysfs_entry md_safe_delay =
80ca3a44 2680__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2681
eae1701f 2682static ssize_t
96de1e66 2683level_show(mddev_t *mddev, char *page)
eae1701f 2684{
2604b703 2685 struct mdk_personality *p = mddev->pers;
d9d166c2 2686 if (p)
eae1701f 2687 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2688 else if (mddev->clevel[0])
2689 return sprintf(page, "%s\n", mddev->clevel);
2690 else if (mddev->level != LEVEL_NONE)
2691 return sprintf(page, "%d\n", mddev->level);
2692 else
2693 return 0;
eae1701f
N		 2694}
2695
d9d166c2
N		 2696static ssize_t
2697level_store(mddev_t *mddev, const char *buf, size_t len)
2698{
245f46c2 2699 char level[16];
20a49ff6 2700 ssize_t rv = len;
245f46c2
N		 2701 struct mdk_personality *pers;
2702 void *priv;
3a981b03 2703 mdk_rdev_t *rdev;
245f46c2
N		 2704
2705 if (mddev->pers == NULL) {
2706 if (len == 0)
2707 return 0;
2708 if (len >= sizeof(mddev->clevel))
2709 return -ENOSPC;
2710 strncpy(mddev->clevel, buf, len);
2711 if (mddev->clevel[len-1] == '\n')
2712 len--;
2713 mddev->clevel[len] = 0;
2714 mddev->level = LEVEL_NONE;
2715 return rv;
2716 }
2717
2718 /* request to change the personality. Need to ensure:
2719 * - array is not engaged in resync/recovery/reshape
2720 * - old personality can be suspended
2721 * - new personality will access other array.
2722 */
2723
2724 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
d9d166c2 2725 return -EBUSY;
245f46c2
N		 2726
2727 if (!mddev->pers->quiesce) {
2728 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2729 mdname(mddev), mddev->pers->name);
2730 return -EINVAL;
2731 }
2732
2733 /* Now find the new personality */
2734 if (len == 0 || len >= sizeof(level))
2735 return -EINVAL;
2736 strncpy(level, buf, len);
2737 if (level[len-1] == '\n')
d9d166c2 2738 len--;
245f46c2
N		 2739 level[len] = 0;
2740
2741 request_module("md-%s", level);
2742 spin_lock(&pers_lock);
2743 pers = find_pers(LEVEL_NONE, level);
2744 if (!pers || !try_module_get(pers->owner)) {
2745 spin_unlock(&pers_lock);
2746 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2747 return -EINVAL;
2748 }
2749 spin_unlock(&pers_lock);
2750
2751 if (pers == mddev->pers) {
2752 /* Nothing to do! */
2753 module_put(pers->owner);
2754 return rv;
2755 }
2756 if (!pers->takeover) {
2757 module_put(pers->owner);
2758 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2759 mdname(mddev), level);
2760 return -EINVAL;
2761 }
2762
2763 /* ->takeover must set new_* and/or delta_disks
2764 * if it succeeds, and may set them when it fails.
2765 */
2766 priv = pers->takeover(mddev);
2767 if (IS_ERR(priv)) {
2768 mddev->new_level = mddev->level;
2769 mddev->new_layout = mddev->layout;
664e7c41 2770 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 2771 mddev->raid_disks -= mddev->delta_disks;
2772 mddev->delta_disks = 0;
2773 module_put(pers->owner);
2774 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2775 mdname(mddev), level);
2776 return PTR_ERR(priv);
2777 }
2778
2779 /* Looks like we have a winner */
2780 mddev_suspend(mddev);
2781 mddev->pers->stop(mddev);
2782 module_put(mddev->pers->owner);
3a981b03
N		 2783 /* Invalidate devices that are now superfluous */
2784 list_for_each_entry(rdev, &mddev->disks, same_set)
2785 if (rdev->raid_disk >= mddev->raid_disks) {
2786 rdev->raid_disk = -1;
2787 clear_bit(In_sync, &rdev->flags);
2788 }
245f46c2
N		 2789 mddev->pers = pers;
2790 mddev->private = priv;
2791 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2792 mddev->level = mddev->new_level;
2793 mddev->layout = mddev->new_layout;
664e7c41 2794 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2
N		 2795 mddev->delta_disks = 0;
2796 pers->run(mddev);
2797 mddev_resume(mddev);
2798 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2799 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2800 md_wakeup_thread(mddev->thread);
d9d166c2
N		 2801 return rv;
2802}
2803
2804static struct md_sysfs_entry md_level =
80ca3a44 2805__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 2806
d4dbd025
N		 2807
2808static ssize_t
2809layout_show(mddev_t *mddev, char *page)
2810{
2811 /* just a number, not meaningful for all levels */
08a02ecd
N		 2812 if (mddev->reshape_position != MaxSector &&
2813 mddev->layout != mddev->new_layout)
2814 return sprintf(page, "%d (%d)\n",
2815 mddev->new_layout, mddev->layout);
d4dbd025
N		 2816 return sprintf(page, "%d\n", mddev->layout);
2817}
2818
2819static ssize_t
2820layout_store(mddev_t *mddev, const char *buf, size_t len)
2821{
2822 char *e;
2823 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 2824
2825 if (!*buf || (*e && *e != '\n'))
2826 return -EINVAL;
2827
b3546035
N		 2828 if (mddev->pers) {
2829 int err;
50ac168a 2830 if (mddev->pers->check_reshape == NULL)
b3546035 2831 return -EBUSY;
597a711b 2832 mddev->new_layout = n;
50ac168a 2833 err = mddev->pers->check_reshape(mddev);
597a711b
N		 2834 if (err) {
2835 mddev->new_layout = mddev->layout;
b3546035 2836 return err;
597a711b 2837 }
b3546035 2838 } else {
08a02ecd 2839 mddev->new_layout = n;
b3546035
N		 2840 if (mddev->reshape_position == MaxSector)
2841 mddev->layout = n;
2842 }
d4dbd025
N		 2843 return len;
2844}
2845static struct md_sysfs_entry md_layout =
80ca3a44 2846__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 2847
2848
eae1701f 2849static ssize_t
96de1e66 2850raid_disks_show(mddev_t *mddev, char *page)
eae1701f 2851{
bb636547
N		 2852 if (mddev->raid_disks == 0)
2853 return 0;
08a02ecd
N		 2854 if (mddev->reshape_position != MaxSector &&
2855 mddev->delta_disks != 0)
2856 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2857 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 2858 return sprintf(page, "%d\n", mddev->raid_disks);
2859}
2860
da943b99
N		 2861static int update_raid_disks(mddev_t *mddev, int raid_disks);
2862
2863static ssize_t
2864raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2865{
da943b99
N		 2866 char *e;
2867 int rv = 0;
2868 unsigned long n = simple_strtoul(buf, &e, 10);
2869
2870 if (!*buf || (*e && *e != '\n'))
2871 return -EINVAL;
2872
2873 if (mddev->pers)
2874 rv = update_raid_disks(mddev, n);
08a02ecd
N		 2875 else if (mddev->reshape_position != MaxSector) {
2876 int olddisks = mddev->raid_disks - mddev->delta_disks;
2877 mddev->delta_disks = n - olddisks;
2878 mddev->raid_disks = n;
2879 } else
da943b99
N		 2880 mddev->raid_disks = n;
2881 return rv ? rv : len;
2882}
2883static struct md_sysfs_entry md_raid_disks =
80ca3a44 2884__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 2885
3b34380a
N		 2886static ssize_t
2887chunk_size_show(mddev_t *mddev, char *page)
2888{
08a02ecd 2889 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 2890 mddev->chunk_sectors != mddev->new_chunk_sectors)
2891 return sprintf(page, "%d (%d)\n",
2892 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 2893 mddev->chunk_sectors << 9);
2894 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 2895}
2896
2897static ssize_t
2898chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2899{
3b34380a
N		 2900 char *e;
2901 unsigned long n = simple_strtoul(buf, &e, 10);
2902
3b34380a
N		 2903 if (!*buf || (*e && *e != '\n'))
2904 return -EINVAL;
2905
b3546035
N		 2906 if (mddev->pers) {
2907 int err;
50ac168a 2908 if (mddev->pers->check_reshape == NULL)
b3546035 2909 return -EBUSY;
597a711b 2910 mddev->new_chunk_sectors = n >> 9;
50ac168a 2911 err = mddev->pers->check_reshape(mddev);
597a711b
N		 2912 if (err) {
2913 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 2914 return err;
597a711b 2915 }
b3546035 2916 } else {
664e7c41 2917 mddev->new_chunk_sectors = n >> 9;
b3546035 2918 if (mddev->reshape_position == MaxSector)
9d8f0363 2919 mddev->chunk_sectors = n >> 9;
b3546035 2920 }
3b34380a
N		 2921 return len;
2922}
2923static struct md_sysfs_entry md_chunk_size =
80ca3a44 2924__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 2925
a94213b1
N		 2926static ssize_t
2927resync_start_show(mddev_t *mddev, char *page)
2928{
d1a7c503
N		 2929 if (mddev->recovery_cp == MaxSector)
2930 return sprintf(page, "none\n");
a94213b1
N		 2931 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2932}
2933
2934static ssize_t
2935resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2936{
a94213b1
N		 2937 char *e;
2938 unsigned long long n = simple_strtoull(buf, &e, 10);
2939
2940 if (mddev->pers)
2941 return -EBUSY;
2942 if (!*buf || (*e && *e != '\n'))
2943 return -EINVAL;
2944
2945 mddev->recovery_cp = n;
2946 return len;
2947}
2948static struct md_sysfs_entry md_resync_start =
80ca3a44 2949__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 2950
9e653b63
N		 2951/*
2952 * The array state can be:
2953 *
2954 * clear
2955 * No devices, no size, no level
2956 * Equivalent to STOP_ARRAY ioctl
2957 * inactive
2958 * May have some settings, but array is not active
2959 * all IO results in error
2960 * When written, doesn't tear down array, but just stops it
2961 * suspended (not supported yet)
2962 * All IO requests will block. The array can be reconfigured.
910d8cb3 2963 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 2964 * readonly
2965 * no resync can happen. no superblocks get written.
2966 * write requests fail
2967 * read-auto
2968 * like readonly, but behaves like 'clean' on a write request.
2969 *
2970 * clean - no pending writes, but otherwise active.
2971 * When written to inactive array, starts without resync
2972 * If a write request arrives then
2973 * if metadata is known, mark 'dirty' and switch to 'active'.
2974 * if not known, block and switch to write-pending
2975 * If written to an active array that has pending writes, then fails.
2976 * active
2977 * fully active: IO and resync can be happening.
2978 * When written to inactive array, starts with resync
2979 *
2980 * write-pending
2981 * clean, but writes are blocked waiting for 'active' to be written.
2982 *
2983 * active-idle
2984 * like active, but no writes have been seen for a while (100msec).
2985 *
2986 */
2987enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2988 write_pending, active_idle, bad_word};
05381954 2989static char *array_states[] = {
9e653b63
N		 2990 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2991 "write-pending", "active-idle", NULL };
2992
2993static int match_word(const char *word, char **list)
2994{
2995 int n;
2996 for (n=0; list[n]; n++)
2997 if (cmd_match(word, list[n]))
2998 break;
2999 return n;
3000}
3001
3002static ssize_t
3003array_state_show(mddev_t *mddev, char *page)
3004{
3005 enum array_state st = inactive;
3006
3007 if (mddev->pers)
3008 switch(mddev->ro) {
3009 case 1:
3010 st = readonly;
3011 break;
3012 case 2:
3013 st = read_auto;
3014 break;
3015 case 0:
3016 if (mddev->in_sync)
3017 st = clean;
e691063a
N		 3018 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3019 st = write_pending;
9e653b63
N		 3020 else if (mddev->safemode)
3021 st = active_idle;
3022 else
3023 st = active;
3024 }
3025 else {
3026 if (list_empty(&mddev->disks) &&
3027 mddev->raid_disks == 0 &&
58c0fed4 3028 mddev->dev_sectors == 0)
9e653b63
N		 3029 st = clear;
3030 else
3031 st = inactive;
3032 }
3033 return sprintf(page, "%s\n", array_states[st]);
3034}
3035
df5b20cf 3036static int do_md_stop(mddev_t * mddev, int ro, int is_open);
9e653b63
N		 3037static int do_md_run(mddev_t * mddev);
3038static int restart_array(mddev_t *mddev);
3039
3040static ssize_t
3041array_state_store(mddev_t *mddev, const char *buf, size_t len)
3042{
3043 int err = -EINVAL;
3044 enum array_state st = match_word(buf, array_states);
3045 switch(st) {
3046 case bad_word:
3047 break;
3048 case clear:
3049 /* stopping an active array */
f2ea68cf 3050 if (atomic_read(&mddev->openers) > 0)
e691063a 3051 return -EBUSY;
df5b20cf 3052 err = do_md_stop(mddev, 0, 0);
9e653b63
N		 3053 break;
3054 case inactive:
3055 /* stopping an active array */
3056 if (mddev->pers) {
f2ea68cf 3057 if (atomic_read(&mddev->openers) > 0)
9e653b63 3058 return -EBUSY;
df5b20cf 3059 err = do_md_stop(mddev, 2, 0);
e691063a
N		 3060 } else
3061 err = 0; /* already inactive */
9e653b63
N		 3062 break;
3063 case suspended:
3064 break; /* not supported yet */
3065 case readonly:
3066 if (mddev->pers)
df5b20cf 3067 err = do_md_stop(mddev, 1, 0);
9e653b63
N		 3068 else {
3069 mddev->ro = 1;
648b629e 3070 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3071 err = do_md_run(mddev);
3072 }
3073 break;
3074 case read_auto:
9e653b63 3075 if (mddev->pers) {
80268ee9 3076 if (mddev->ro == 0)
df5b20cf 3077 err = do_md_stop(mddev, 1, 0);
80268ee9 3078 else if (mddev->ro == 1)
648b629e
N		 3079 err = restart_array(mddev);
3080 if (err == 0) {
3081 mddev->ro = 2;
3082 set_disk_ro(mddev->gendisk, 0);
3083 }
9e653b63
N		 3084 } else {
3085 mddev->ro = 2;
3086 err = do_md_run(mddev);
3087 }
3088 break;
3089 case clean:
3090 if (mddev->pers) {
3091 restart_array(mddev);
3092 spin_lock_irq(&mddev->write_lock);
3093 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3094 if (mddev->in_sync == 0) {
3095 mddev->in_sync = 1;
31a59e34
N		 3096 if (mddev->safemode == 1)
3097 mddev->safemode = 0;
e691063a
N		 3098 if (mddev->persistent)
3099 set_bit(MD_CHANGE_CLEAN,
3100 &mddev->flags);
3101 }
3102 err = 0;
3103 } else
3104 err = -EBUSY;
9e653b63 3105 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3106 } else
3107 err = -EINVAL;
9e653b63
N		 3108 break;
3109 case active:
3110 if (mddev->pers) {
3111 restart_array(mddev);
e691063a
N		 3112 if (mddev->external)
3113 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
9e653b63
N		 3114 wake_up(&mddev->sb_wait);
3115 err = 0;
3116 } else {
3117 mddev->ro = 0;
648b629e 3118 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3119 err = do_md_run(mddev);
3120 }
3121 break;
3122 case write_pending:
3123 case active_idle:
3124 /* these cannot be set */
3125 break;
3126 }
3127 if (err)
3128 return err;
0fd62b86 3129 else {
b62b7590 3130 sysfs_notify_dirent(mddev->sysfs_state);
9e653b63 3131 return len;
0fd62b86 3132 }
9e653b63 3133}
80ca3a44
N		 3134static struct md_sysfs_entry md_array_state =
3135__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3136
6d7ff738
N		 3137static ssize_t
3138null_show(mddev_t *mddev, char *page)
3139{
3140 return -EINVAL;
3141}
3142
3143static ssize_t
3144new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3145{
3146 /* buf must be %d:%d\n? giving major and minor numbers */
3147 /* The new device is added to the array.
3148 * If the array has a persistent superblock, we read the
3149 * superblock to initialise info and check validity.
3150 * Otherwise, only checking done is that in bind_rdev_to_array,
3151 * which mainly checks size.
3152 */
3153 char *e;
3154 int major = simple_strtoul(buf, &e, 10);
3155 int minor;
3156 dev_t dev;
3157 mdk_rdev_t *rdev;
3158 int err;
3159
3160 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3161 return -EINVAL;
3162 minor = simple_strtoul(e+1, &e, 10);
3163 if (*e && *e != '\n')
3164 return -EINVAL;
3165 dev = MKDEV(major, minor);
3166 if (major != MAJOR(dev) ||
3167 minor != MINOR(dev))
3168 return -EOVERFLOW;
3169
3170
3171 if (mddev->persistent) {
3172 rdev = md_import_device(dev, mddev->major_version,
3173 mddev->minor_version);
3174 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3175 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3176 mdk_rdev_t, same_set);
3177 err = super_types[mddev->major_version]
3178 .load_super(rdev, rdev0, mddev->minor_version);
3179 if (err < 0)
3180 goto out;
3181 }
c5d79adb
N		 3182 } else if (mddev->external)
3183 rdev = md_import_device(dev, -2, -1);
3184 else
6d7ff738
N		 3185 rdev = md_import_device(dev, -1, -1);
3186
3187 if (IS_ERR(rdev))
3188 return PTR_ERR(rdev);
3189 err = bind_rdev_to_array(rdev, mddev);
3190 out:
3191 if (err)
3192 export_rdev(rdev);
3193 return err ? err : len;
3194}
3195
3196static struct md_sysfs_entry md_new_device =
80ca3a44 3197__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3198
9b1d1dac
PC		 3199static ssize_t
3200bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3201{
3202 char *end;
3203 unsigned long chunk, end_chunk;
3204
3205 if (!mddev->bitmap)
3206 goto out;
3207 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3208 while (*buf) {
3209 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3210 if (buf == end) break;
3211 if (*end == '-') { /* range */
3212 buf = end + 1;
3213 end_chunk = simple_strtoul(buf, &end, 0);
3214 if (buf == end) break;
3215 }
3216 if (*end && !isspace(*end)) break;
3217 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3218 buf = end;
3219 while (isspace(*buf)) buf++;
3220 }
3221 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3222out:
3223 return len;
3224}
3225
3226static struct md_sysfs_entry md_bitmap =
3227__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3228
a35b0d69
N		 3229static ssize_t
3230size_show(mddev_t *mddev, char *page)
3231{
58c0fed4
AN		 3232 return sprintf(page, "%llu\n",
3233 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 3234}
3235
d71f9f88 3236static int update_size(mddev_t *mddev, sector_t num_sectors);
a35b0d69
N		 3237
3238static ssize_t
3239size_store(mddev_t *mddev, const char *buf, size_t len)
3240{
3241 /* If array is inactive, we can reduce the component size, but
3242 * not increase it (except from 0).
3243 * If array is active, we can try an on-line resize
3244 */
b522adcd
DW		 3245 sector_t sectors;
3246 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 3247
58c0fed4
AN		 3248 if (err < 0)
3249 return err;
a35b0d69 3250 if (mddev->pers) {
58c0fed4 3251 err = update_size(mddev, sectors);
850b2b42 3252 md_update_sb(mddev, 1);
a35b0d69 3253 } else {
58c0fed4
AN		 3254 if (mddev->dev_sectors == 0 ||
3255 mddev->dev_sectors > sectors)
3256 mddev->dev_sectors = sectors;
a35b0d69
N		 3257 else
3258 err = -ENOSPC;
3259 }
3260 return err ? err : len;
3261}
3262
3263static struct md_sysfs_entry md_size =
80ca3a44 3264__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 3265
8bb93aac
N		 3266
3267/* Metdata version.
e691063a
N		 3268 * This is one of
3269 * 'none' for arrays with no metadata (good luck...)
3270 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 3271 * or N.M for internally known formats
3272 */
3273static ssize_t
3274metadata_show(mddev_t *mddev, char *page)
3275{
3276 if (mddev->persistent)
3277 return sprintf(page, "%d.%d\n",
3278 mddev->major_version, mddev->minor_version);
e691063a
N		 3279 else if (mddev->external)
3280 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 3281 else
3282 return sprintf(page, "none\n");
3283}
3284
3285static ssize_t
3286metadata_store(mddev_t *mddev, const char *buf, size_t len)
3287{
3288 int major, minor;
3289 char *e;
ea43ddd8
N		 3290 /* Changing the details of 'external' metadata is
3291 * always permitted. Otherwise there must be
3292 * no devices attached to the array.
3293 */
3294 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3295 ;
3296 else if (!list_empty(&mddev->disks))
8bb93aac
N		 3297 return -EBUSY;
3298
3299 if (cmd_match(buf, "none")) {
3300 mddev->persistent = 0;
e691063a
N		 3301 mddev->external = 0;
3302 mddev->major_version = 0;
3303 mddev->minor_version = 90;
3304 return len;
3305 }
3306 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 3307 size_t namelen = len-9;
e691063a
N		 3308 if (namelen >= sizeof(mddev->metadata_type))
3309 namelen = sizeof(mddev->metadata_type)-1;
3310 strncpy(mddev->metadata_type, buf+9, namelen);
3311 mddev->metadata_type[namelen] = 0;
3312 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3313 mddev->metadata_type[--namelen] = 0;
3314 mddev->persistent = 0;
3315 mddev->external = 1;
8bb93aac
N		 3316 mddev->major_version = 0;
3317 mddev->minor_version = 90;
3318 return len;
3319 }
3320 major = simple_strtoul(buf, &e, 10);
3321 if (e==buf || *e != '.')
3322 return -EINVAL;
3323 buf = e+1;
3324 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 3325 if (e==buf || (*e && *e != '\n') )
8bb93aac 3326 return -EINVAL;
50511da3 3327 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 3328 return -ENOENT;
3329 mddev->major_version = major;
3330 mddev->minor_version = minor;
3331 mddev->persistent = 1;
e691063a 3332 mddev->external = 0;
8bb93aac
N		 3333 return len;
3334}
3335
3336static struct md_sysfs_entry md_metadata =
80ca3a44 3337__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 3338
24dd469d 3339static ssize_t
7eec314d 3340action_show(mddev_t *mddev, char *page)
24dd469d 3341{
7eec314d 3342 char *type = "idle";
b6a9ce68
N		 3343 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3344 type = "frozen";
3345 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 3346 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 3347 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3348 type = "reshape";
3349 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 3350 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3351 type = "resync";
3352 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3353 type = "check";
3354 else
3355 type = "repair";
72a23c21 3356 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 3357 type = "recover";
3358 }
3359 return sprintf(page, "%s\n", type);
3360}
3361
3362static ssize_t
7eec314d 3363action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 3364{
7eec314d
N		 3365 if (!mddev->pers || !mddev->pers->sync_request)
3366 return -EINVAL;
3367
b6a9ce68
N		 3368 if (cmd_match(page, "frozen"))
3369 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3370 else
3371 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3372
3373 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 3374 if (mddev->sync_thread) {
3375 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3376 md_unregister_thread(mddev->sync_thread);
3377 mddev->sync_thread = NULL;
3378 mddev->recovery = 0;
3379 }
03c902e1
N		 3380 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3381 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 3382 return -EBUSY;
72a23c21
NB		 3383 else if (cmd_match(page, "resync"))
3384 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3385 else if (cmd_match(page, "recover")) {
3386 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 3387 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 3388 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 3389 int err;
3390 if (mddev->pers->start_reshape == NULL)
3391 return -EINVAL;
3392 err = mddev->pers->start_reshape(mddev);
3393 if (err)
3394 return err;
a99ac971 3395 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 3396 } else {
bce74dac 3397 if (cmd_match(page, "check"))
7eec314d 3398 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 3399 else if (!cmd_match(page, "repair"))
7eec314d
N		 3400 return -EINVAL;
3401 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3402 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 3403 }
03c902e1 3404 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 3405 md_wakeup_thread(mddev->thread);
0c3573f1 3406 sysfs_notify_dirent(mddev->sysfs_action);
24dd469d
N		 3407 return len;
3408}
3409
9d88883e 3410static ssize_t
96de1e66 3411mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 3412{
3413 return sprintf(page, "%llu\n",
3414 (unsigned long long) mddev->resync_mismatches);
3415}
3416
80ca3a44
N		 3417static struct md_sysfs_entry md_scan_mode =
3418__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 3419
96de1e66 3420
80ca3a44 3421static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 3422
88202a0c
N		 3423static ssize_t
3424sync_min_show(mddev_t *mddev, char *page)
3425{
3426 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3427 mddev->sync_speed_min ? "local": "system");
3428}
3429
3430static ssize_t
3431sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3432{
3433 int min;
3434 char *e;
3435 if (strncmp(buf, "system", 6)==0) {
3436 mddev->sync_speed_min = 0;
3437 return len;
3438 }
3439 min = simple_strtoul(buf, &e, 10);
3440 if (buf == e || (*e && *e != '\n') || min <= 0)
3441 return -EINVAL;
3442 mddev->sync_speed_min = min;
3443 return len;
3444}
3445
3446static struct md_sysfs_entry md_sync_min =
3447__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3448
3449static ssize_t
3450sync_max_show(mddev_t *mddev, char *page)
3451{
3452 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3453 mddev->sync_speed_max ? "local": "system");
3454}
3455
3456static ssize_t
3457sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3458{
3459 int max;
3460 char *e;
3461 if (strncmp(buf, "system", 6)==0) {
3462 mddev->sync_speed_max = 0;
3463 return len;
3464 }
3465 max = simple_strtoul(buf, &e, 10);
3466 if (buf == e || (*e && *e != '\n') || max <= 0)
3467 return -EINVAL;
3468 mddev->sync_speed_max = max;
3469 return len;
3470}
3471
3472static struct md_sysfs_entry md_sync_max =
3473__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3474
d7f3d291
IP		 3475static ssize_t
3476degraded_show(mddev_t *mddev, char *page)
3477{
3478 return sprintf(page, "%d\n", mddev->degraded);
3479}
3480static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 3481
90b08710
BS		 3482static ssize_t
3483sync_force_parallel_show(mddev_t *mddev, char *page)
3484{
3485 return sprintf(page, "%d\n", mddev->parallel_resync);
3486}
3487
3488static ssize_t
3489sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3490{
3491 long n;
3492
3493 if (strict_strtol(buf, 10, &n))
3494 return -EINVAL;
3495
3496 if (n != 0 && n != 1)
3497 return -EINVAL;
3498
3499 mddev->parallel_resync = n;
3500
3501 if (mddev->sync_thread)
3502 wake_up(&resync_wait);
3503
3504 return len;
3505}
3506
3507/* force parallel resync, even with shared block devices */
3508static struct md_sysfs_entry md_sync_force_parallel =
3509__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3510 sync_force_parallel_show, sync_force_parallel_store);
3511
88202a0c
N		 3512static ssize_t
3513sync_speed_show(mddev_t *mddev, char *page)
3514{
3515 unsigned long resync, dt, db;
d1a7c503
N		 3516 if (mddev->curr_resync == 0)
3517 return sprintf(page, "none\n");
9687a60c
AN		 3518 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3519 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 3520 if (!dt) dt++;
9687a60c
AN		 3521 db = resync - mddev->resync_mark_cnt;
3522 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 3523}
3524
80ca3a44 3525static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 3526
3527static ssize_t
3528sync_completed_show(mddev_t *mddev, char *page)
3529{
58c0fed4 3530 unsigned long max_sectors, resync;
88202a0c 3531
acb180b0
N		 3532 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3533 return sprintf(page, "none\n");
3534
88202a0c 3535 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
58c0fed4 3536 max_sectors = mddev->resync_max_sectors;
88202a0c 3537 else
58c0fed4 3538 max_sectors = mddev->dev_sectors;
88202a0c 3539
acb180b0 3540 resync = mddev->curr_resync_completed;
58c0fed4 3541 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
88202a0c
N		 3542}
3543
80ca3a44 3544static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 3545
5e96ee65
NB		 3546static ssize_t
3547min_sync_show(mddev_t *mddev, char *page)
3548{
3549 return sprintf(page, "%llu\n",
3550 (unsigned long long)mddev->resync_min);
3551}
3552static ssize_t
3553min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3554{
3555 unsigned long long min;
3556 if (strict_strtoull(buf, 10, &min))
3557 return -EINVAL;
3558 if (min > mddev->resync_max)
3559 return -EINVAL;
3560 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3561 return -EBUSY;
3562
3563 /* Must be a multiple of chunk_size */
9d8f0363 3564 if (mddev->chunk_sectors) {
2ac06c33 3565 sector_t temp = min;
9d8f0363 3566 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 3567 return -EINVAL;
3568 }
3569 mddev->resync_min = min;
3570
3571 return len;
3572}
3573
3574static struct md_sysfs_entry md_min_sync =
3575__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3576
c6207277
N		 3577static ssize_t
3578max_sync_show(mddev_t *mddev, char *page)
3579{
3580 if (mddev->resync_max == MaxSector)
3581 return sprintf(page, "max\n");
3582 else
3583 return sprintf(page, "%llu\n",
3584 (unsigned long long)mddev->resync_max);
3585}
3586static ssize_t
3587max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3588{
3589 if (strncmp(buf, "max", 3) == 0)
3590 mddev->resync_max = MaxSector;
3591 else {
5e96ee65
NB		 3592 unsigned long long max;
3593 if (strict_strtoull(buf, 10, &max))
3594 return -EINVAL;
3595 if (max < mddev->resync_min)
c6207277
N		 3596 return -EINVAL;
3597 if (max < mddev->resync_max &&
3598 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3599 return -EBUSY;
3600
3601 /* Must be a multiple of chunk_size */
9d8f0363 3602 if (mddev->chunk_sectors) {
2ac06c33 3603 sector_t temp = max;
9d8f0363 3604 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 3605 return -EINVAL;
3606 }
3607 mddev->resync_max = max;
3608 }
3609 wake_up(&mddev->recovery_wait);
3610 return len;
3611}
3612
3613static struct md_sysfs_entry md_max_sync =
3614__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3615
e464eafd
N		 3616static ssize_t
3617suspend_lo_show(mddev_t *mddev, char *page)
3618{
3619 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3620}
3621
3622static ssize_t
3623suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3624{
3625 char *e;
3626 unsigned long long new = simple_strtoull(buf, &e, 10);
3627
b8d966ef
N		 3628 if (mddev->pers == NULL ||
3629 mddev->pers->quiesce == NULL)
e464eafd
N		 3630 return -EINVAL;
3631 if (buf == e || (*e && *e != '\n'))
3632 return -EINVAL;
3633 if (new >= mddev->suspend_hi ||
3634 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3635 mddev->suspend_lo = new;
3636 mddev->pers->quiesce(mddev, 2);
3637 return len;
3638 } else
3639 return -EINVAL;
3640}
3641static struct md_sysfs_entry md_suspend_lo =
3642__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3643
3644
3645static ssize_t
3646suspend_hi_show(mddev_t *mddev, char *page)
3647{
3648 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3649}
3650
3651static ssize_t
3652suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3653{
3654 char *e;
3655 unsigned long long new = simple_strtoull(buf, &e, 10);
3656
b8d966ef
N		 3657 if (mddev->pers == NULL ||
3658 mddev->pers->quiesce == NULL)
e464eafd
N		 3659 return -EINVAL;
3660 if (buf == e || (*e && *e != '\n'))
3661 return -EINVAL;
3662 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3663 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3664 mddev->suspend_hi = new;
3665 mddev->pers->quiesce(mddev, 1);
3666 mddev->pers->quiesce(mddev, 0);
3667 return len;
3668 } else
3669 return -EINVAL;
3670}
3671static struct md_sysfs_entry md_suspend_hi =
3672__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3673
08a02ecd
N		 3674static ssize_t
3675reshape_position_show(mddev_t *mddev, char *page)
3676{
3677 if (mddev->reshape_position != MaxSector)
3678 return sprintf(page, "%llu\n",
3679 (unsigned long long)mddev->reshape_position);
3680 strcpy(page, "none\n");
3681 return 5;
3682}
3683
3684static ssize_t
3685reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3686{
3687 char *e;
3688 unsigned long long new = simple_strtoull(buf, &e, 10);
3689 if (mddev->pers)
3690 return -EBUSY;
3691 if (buf == e || (*e && *e != '\n'))
3692 return -EINVAL;
3693 mddev->reshape_position = new;
3694 mddev->delta_disks = 0;
3695 mddev->new_level = mddev->level;
3696 mddev->new_layout = mddev->layout;
664e7c41 3697 mddev->new_chunk_sectors = mddev->chunk_sectors;
08a02ecd
N		 3698 return len;
3699}
3700
3701static struct md_sysfs_entry md_reshape_position =
3702__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3703 reshape_position_store);
3704
b522adcd
DW		 3705static ssize_t
3706array_size_show(mddev_t *mddev, char *page)
3707{
3708 if (mddev->external_size)
3709 return sprintf(page, "%llu\n",
3710 (unsigned long long)mddev->array_sectors/2);
3711 else
3712 return sprintf(page, "default\n");
3713}
3714
3715static ssize_t
3716array_size_store(mddev_t *mddev, const char *buf, size_t len)
3717{
3718 sector_t sectors;
3719
3720 if (strncmp(buf, "default", 7) == 0) {
3721 if (mddev->pers)
3722 sectors = mddev->pers->size(mddev, 0, 0);
3723 else
3724 sectors = mddev->array_sectors;
3725
3726 mddev->external_size = 0;
3727 } else {
3728 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3729 return -EINVAL;
3730 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 3731 return -E2BIG;
b522adcd
DW		 3732
3733 mddev->external_size = 1;
3734 }
3735
3736 mddev->array_sectors = sectors;
3737 set_capacity(mddev->gendisk, mddev->array_sectors);
3738 if (mddev->pers) {
3739 struct block_device *bdev = bdget_disk(mddev->gendisk, 0);
3740
3741 if (bdev) {
3742 mutex_lock(&bdev->bd_inode->i_mutex);
3743 i_size_write(bdev->bd_inode,
3744 (loff_t)mddev->array_sectors << 9);
3745 mutex_unlock(&bdev->bd_inode->i_mutex);
3746 bdput(bdev);
3747 }
3748 }
3749
3750 return len;
3751}
3752
3753static struct md_sysfs_entry md_array_size =
3754__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3755 array_size_store);
e464eafd 3756
eae1701f
N		 3757static struct attribute *md_default_attrs[] = {
3758 &md_level.attr,
d4dbd025 3759 &md_layout.attr,
eae1701f 3760 &md_raid_disks.attr,
3b34380a 3761 &md_chunk_size.attr,
a35b0d69 3762 &md_size.attr,
a94213b1 3763 &md_resync_start.attr,
8bb93aac 3764 &md_metadata.attr,
6d7ff738 3765 &md_new_device.attr,
16f17b39 3766 &md_safe_delay.attr,
9e653b63 3767 &md_array_state.attr,
08a02ecd 3768 &md_reshape_position.attr,
b522adcd 3769 &md_array_size.attr,
411036fa
N		 3770 NULL,
3771};
3772
3773static struct attribute *md_redundancy_attrs[] = {
24dd469d 3774 &md_scan_mode.attr,
9d88883e 3775 &md_mismatches.attr,
88202a0c
N		 3776 &md_sync_min.attr,
3777 &md_sync_max.attr,
3778 &md_sync_speed.attr,
90b08710 3779 &md_sync_force_parallel.attr,
88202a0c 3780 &md_sync_completed.attr,
5e96ee65 3781 &md_min_sync.attr,
c6207277 3782 &md_max_sync.attr,
e464eafd
N		 3783 &md_suspend_lo.attr,
3784 &md_suspend_hi.attr,
9b1d1dac 3785 &md_bitmap.attr,
d7f3d291 3786 &md_degraded.attr,
eae1701f
N		 3787 NULL,
3788};
411036fa
N		 3789static struct attribute_group md_redundancy_group = {
3790 .name = NULL,
3791 .attrs = md_redundancy_attrs,
3792};
3793
eae1701f
N		 3794
3795static ssize_t
3796md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3797{
3798 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3799 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 3800 ssize_t rv;
eae1701f
N		 3801
3802 if (!entry->show)
3803 return -EIO;
5dc5cf7d
IM		 3804 rv = mddev_lock(mddev);
3805 if (!rv) {
3806 rv = entry->show(mddev, page);
3807 mddev_unlock(mddev);
3808 }
96de1e66 3809 return rv;
eae1701f
N		 3810}
3811
3812static ssize_t
3813md_attr_store(struct kobject *kobj, struct attribute *attr,
3814 const char *page, size_t length)
3815{
3816 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3817 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 3818 ssize_t rv;
eae1701f
N		 3819
3820 if (!entry->store)
3821 return -EIO;
67463acb
N		 3822 if (!capable(CAP_SYS_ADMIN))
3823 return -EACCES;
5dc5cf7d 3824 rv = mddev_lock(mddev);
d3374825
N		 3825 if (mddev->hold_active == UNTIL_IOCTL)
3826 mddev->hold_active = 0;
5dc5cf7d
IM		 3827 if (!rv) {
3828 rv = entry->store(mddev, page, length);
3829 mddev_unlock(mddev);
3830 }
96de1e66 3831 return rv;
eae1701f
N		 3832}
3833
3834static void md_free(struct kobject *ko)
3835{
3836 mddev_t *mddev = container_of(ko, mddev_t, kobj);
a21d1504
N		 3837
3838 if (mddev->sysfs_state)
3839 sysfs_put(mddev->sysfs_state);
3840
3841 if (mddev->gendisk) {
3842 del_gendisk(mddev->gendisk);
3843 put_disk(mddev->gendisk);
3844 }
3845 if (mddev->queue)
3846 blk_cleanup_queue(mddev->queue);
3847
eae1701f
N		 3848 kfree(mddev);
3849}
3850
3851static struct sysfs_ops md_sysfs_ops = {
3852 .show = md_attr_show,
3853 .store = md_attr_store,
3854};
3855static struct kobj_type md_ktype = {
3856 .release = md_free,
3857 .sysfs_ops = &md_sysfs_ops,
3858 .default_attrs = md_default_attrs,
3859};
3860
1da177e4
LT		 3861int mdp_major = 0;
3862
5fd3a17e
DW		 3863static void mddev_delayed_delete(struct work_struct *ws)
3864{
3865 mddev_t *mddev = container_of(ws, mddev_t, del_work);
3866
3867 if (mddev->private == &md_redundancy_group) {
3868 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3869 if (mddev->sysfs_action)
3870 sysfs_put(mddev->sysfs_action);
3871 mddev->sysfs_action = NULL;
3872 mddev->private = NULL;
3873 }
3874 kobject_del(&mddev->kobj);
3875 kobject_put(&mddev->kobj);
3876}
3877
efeb53c0 3878static int md_alloc(dev_t dev, char *name)
1da177e4 3879{
48c9c27b 3880 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 3881 mddev_t *mddev = mddev_find(dev);
3882 struct gendisk *disk;
efeb53c0
N		 3883 int partitioned;
3884 int shift;
3885 int unit;
3830c62f 3886 int error;
1da177e4
LT		 3887
3888 if (!mddev)
efeb53c0
N		 3889 return -ENODEV;
3890
3891 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3892 shift = partitioned ? MdpMinorShift : 0;
3893 unit = MINOR(mddev->unit) >> shift;
1da177e4 3894
d3374825
N		 3895 /* wait for any previous instance if this device
3896 * to be completed removed (mddev_delayed_delete).
3897 */
3898 flush_scheduled_work();
3899
48c9c27b 3900 mutex_lock(&disks_mutex);
0909dc44
N		 3901 error = -EEXIST;
3902 if (mddev->gendisk)
3903 goto abort;
efeb53c0
N		 3904
3905 if (name) {
3906 /* Need to ensure that 'name' is not a duplicate.
3907 */
3908 mddev_t *mddev2;
3909 spin_lock(&all_mddevs_lock);
3910
3911 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3912 if (mddev2->gendisk &&
3913 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3914 spin_unlock(&all_mddevs_lock);
0909dc44 3915 goto abort;
efeb53c0
N		 3916 }
3917 spin_unlock(&all_mddevs_lock);
1da177e4 3918 }
8b765398 3919
0909dc44 3920 error = -ENOMEM;
8b765398 3921 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 3922 if (!mddev->queue)
3923 goto abort;
409c57f3
N		 3924 mddev->queue->queuedata = mddev;
3925
8b765398
N		 3926 /* Can be unlocked because the queue is new: no concurrency */
3927 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3928
409c57f3 3929 blk_queue_make_request(mddev->queue, md_make_request);
8b765398 3930
1da177e4
LT		 3931 disk = alloc_disk(1 << shift);
3932 if (!disk) {
8b765398
N		 3933 blk_cleanup_queue(mddev->queue);
3934 mddev->queue = NULL;
0909dc44 3935 goto abort;
1da177e4 3936 }
efeb53c0 3937 disk->major = MAJOR(mddev->unit);
1da177e4 3938 disk->first_minor = unit << shift;
efeb53c0
N		 3939 if (name)
3940 strcpy(disk->disk_name, name);
3941 else if (partitioned)
1da177e4 3942 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 3943 else
1da177e4 3944 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 3945 disk->fops = &md_fops;
3946 disk->private_data = mddev;
3947 disk->queue = mddev->queue;
92850bbd 3948 /* Allow extended partitions. This makes the
d3374825 3949 * 'mdp' device redundant, but we can't really
92850bbd
N		 3950 * remove it now.
3951 */
3952 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4
LT		 3953 add_disk(disk);
3954 mddev->gendisk = disk;
ed9e1982
TH		 3955 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3956 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 3957 if (error) {
3958 /* This isn't possible, but as kobject_init_and_add is marked
3959 * __must_check, we must do something with the result
3960 */
5e55e2f5
N		 3961 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3962 disk->disk_name);
0909dc44
N		 3963 error = 0;
3964 }
3965 abort:
3966 mutex_unlock(&disks_mutex);
3967 if (!error) {
3830c62f 3968 kobject_uevent(&mddev->kobj, KOBJ_ADD);
b62b7590
N		 3969 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3970 }
d3374825 3971 mddev_put(mddev);
0909dc44 3972 return error;
efeb53c0
N		 3973}
3974
3975static struct kobject *md_probe(dev_t dev, int *part, void *data)
3976{
3977 md_alloc(dev, NULL);
1da177e4
LT		 3978 return NULL;
3979}
3980
efeb53c0
N		 3981static int add_named_array(const char *val, struct kernel_param *kp)
3982{
3983 /* val must be "md_*" where * is not all digits.
3984 * We allocate an array with a large free minor number, and
3985 * set the name to val. val must not already be an active name.
3986 */
3987 int len = strlen(val);
3988 char buf[DISK_NAME_LEN];
3989
3990 while (len && val[len-1] == '\n')
3991 len--;
3992 if (len >= DISK_NAME_LEN)
3993 return -E2BIG;
3994 strlcpy(buf, val, len+1);
3995 if (strncmp(buf, "md_", 3) != 0)
3996 return -EINVAL;
3997 return md_alloc(0, buf);
3998}
3999
1da177e4
LT		 4000static void md_safemode_timeout(unsigned long data)
4001{
4002 mddev_t *mddev = (mddev_t *) data;
4003
0fd62b86
NB		 4004 if (!atomic_read(&mddev->writes_pending)) {
4005 mddev->safemode = 1;
4006 if (mddev->external)
b62b7590 4007 sysfs_notify_dirent(mddev->sysfs_state);
0fd62b86 4008 }
1da177e4
LT		 4009 md_wakeup_thread(mddev->thread);
4010}
4011
6ff8d8ec 4012static int start_dirty_degraded;
1da177e4
LT		 4013
4014static int do_md_run(mddev_t * mddev)
4015{
2604b703 4016 int err;
1da177e4
LT		 4017 mdk_rdev_t *rdev;
4018 struct gendisk *disk;
2604b703 4019 struct mdk_personality *pers;
1da177e4 4020
a757e64c
N		 4021 if (list_empty(&mddev->disks))
4022 /* cannot run an array with no devices.. */
1da177e4 4023 return -EINVAL;
1da177e4
LT		 4024
4025 if (mddev->pers)
4026 return -EBUSY;
4027
4028 /*
4029 * Analyze all RAID superblock(s)
4030 */
1ec4a939
N		 4031 if (!mddev->raid_disks) {
4032 if (!mddev->persistent)
4033 return -EINVAL;
a757e64c 4034 analyze_sbs(mddev);
1ec4a939 4035 }
1da177e4 4036
d9d166c2
N		 4037 if (mddev->level != LEVEL_NONE)
4038 request_module("md-level-%d", mddev->level);
4039 else if (mddev->clevel[0])
4040 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4041
4042 /*
4043 * Drop all container device buffers, from now on
4044 * the only valid external interface is through the md
4045 * device.
1da177e4 4046 */
159ec1fc 4047 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7 4048 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4049 continue;
4050 sync_blockdev(rdev->bdev);
f98393a6 4051 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4052
4053 /* perform some consistency tests on the device.
4054 * We don't want the data to overlap the metadata,
58c0fed4 4055 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4056 */
0f420358 4057 if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4058 if (mddev->dev_sectors &&
4059 rdev->data_offset + mddev->dev_sectors
0f420358 4060 > rdev->sb_start) {
f0d76d70
N		 4061 printk("md: %s: data overlaps metadata\n",
4062 mdname(mddev));
4063 return -EINVAL;
4064 }
4065 } else {
0f420358 4066 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4067 > rdev->data_offset) {
4068 printk("md: %s: metadata overlaps data\n",
4069 mdname(mddev));
4070 return -EINVAL;
4071 }
4072 }
3c0ee63a 4073 sysfs_notify_dirent(rdev->sysfs_state);
1da177e4
LT		 4074 }
4075
4076 md_probe(mddev->unit, NULL, NULL);
4077 disk = mddev->gendisk;
4078 if (!disk)
4079 return -ENOMEM;
4080
4081 spin_lock(&pers_lock);
d9d166c2 4082 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4083 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4084 spin_unlock(&pers_lock);
d9d166c2
N		 4085 if (mddev->level != LEVEL_NONE)
4086 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4087 mddev->level);
4088 else
4089 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4090 mddev->clevel);
1da177e4
LT		 4091 return -EINVAL;
4092 }
2604b703 4093 mddev->pers = pers;
1da177e4 4094 spin_unlock(&pers_lock);
34817e8c
N		 4095 if (mddev->level != pers->level) {
4096 mddev->level = pers->level;
4097 mddev->new_level = pers->level;
4098 }
d9d166c2 4099 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4100
f6705578 4101 if (mddev->reshape_position != MaxSector &&
63c70c4f 4102 pers->start_reshape == NULL) {
f6705578
N		 4103 /* This personality cannot handle reshaping... */
4104 mddev->pers = NULL;
4105 module_put(pers->owner);
4106 return -EINVAL;
4107 }
4108
7dd5e7c3
N		 4109 if (pers->sync_request) {
4110 /* Warn if this is a potentially silly
4111 * configuration.
4112 */
4113 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4114 mdk_rdev_t *rdev2;
7dd5e7c3 4115 int warned = 0;
159ec1fc
CR		 4116
4117 list_for_each_entry(rdev, &mddev->disks, same_set)
4118 list_for_each_entry(rdev2, &mddev->disks, same_set) {
7dd5e7c3
N		 4119 if (rdev < rdev2 &&
4120 rdev->bdev->bd_contains ==
4121 rdev2->bdev->bd_contains) {
4122 printk(KERN_WARNING
4123 "%s: WARNING: %s appears to be"
4124 " on the same physical disk as"
4125 " %s.\n",
4126 mdname(mddev),
4127 bdevname(rdev->bdev,b),
4128 bdevname(rdev2->bdev,b2));
4129 warned = 1;
4130 }
4131 }
159ec1fc 4132
7dd5e7c3
N		 4133 if (warned)
4134 printk(KERN_WARNING
4135 "True protection against single-disk"
4136 " failure might be compromised.\n");
4137 }
4138
657390d2 4139 mddev->recovery = 0;
58c0fed4
AN		 4140 /* may be over-ridden by personality */
4141 mddev->resync_max_sectors = mddev->dev_sectors;
4142
a9701a30 4143 mddev->barriers_work = 1;
6ff8d8ec 4144 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 4145
f91de92e
N		 4146 if (start_readonly)
4147 mddev->ro = 2; /* read-only, but switch on first write */
4148
b15c2e57 4149 err = mddev->pers->run(mddev);
13e53df3
AN		 4150 if (err)
4151 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 4152 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4153 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4154 " but 'external_size' not in effect?\n", __func__);
4155 printk(KERN_ERR
4156 "md: invalid array_size %llu > default size %llu\n",
4157 (unsigned long long)mddev->array_sectors / 2,
4158 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4159 err = -EINVAL;
4160 mddev->pers->stop(mddev);
4161 }
4162 if (err == 0 && mddev->pers->sync_request) {
b15c2e57
N		 4163 err = bitmap_create(mddev);
4164 if (err) {
4165 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4166 mdname(mddev), err);
4167 mddev->pers->stop(mddev);
4168 }
4169 }
1da177e4 4170 if (err) {
1da177e4
LT		 4171 module_put(mddev->pers->owner);
4172 mddev->pers = NULL;
32a7627c
N		 4173 bitmap_destroy(mddev);
4174 return err;
1da177e4 4175 }
5e55e2f5
N		 4176 if (mddev->pers->sync_request) {
4177 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4178 printk(KERN_WARNING
4179 "md: cannot register extra attributes for %s\n",
4180 mdname(mddev));
0c3573f1 4181 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
5e55e2f5 4182 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 4183 mddev->ro = 0;
4184
1da177e4
LT		 4185 atomic_set(&mddev->writes_pending,0);
4186 mddev->safemode = 0;
4187 mddev->safemode_timer.function = md_safemode_timeout;
4188 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 4189 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 4190 mddev->in_sync = 1;
86e6ffdd 4191
159ec1fc 4192 list_for_each_entry(rdev, &mddev->disks, same_set)
86e6ffdd
N		 4193 if (rdev->raid_disk >= 0) {
4194 char nm[20];
4195 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 4196 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4197 printk("md: cannot register %s for %s\n",
4198 nm, mdname(mddev));
86e6ffdd 4199 }
1da177e4
LT		 4200
4201 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4202
850b2b42
N		 4203 if (mddev->flags)
4204 md_update_sb(mddev, 0);
1da177e4 4205
f233ea5c 4206 set_capacity(disk, mddev->array_sectors);
1da177e4 4207
5fd6c1dc
N		 4208 /* If there is a partially-recovered drive we need to
4209 * start recovery here. If we leave it to md_check_recovery,
4210 * it will remove the drives and not do the right thing
4211 */
0b8c9de0 4212 if (mddev->degraded && !mddev->sync_thread) {
5fd6c1dc 4213 int spares = 0;
159ec1fc 4214 list_for_each_entry(rdev, &mddev->disks, same_set)
5fd6c1dc
N		 4215 if (rdev->raid_disk >= 0 &&
4216 !test_bit(In_sync, &rdev->flags) &&
4217 !test_bit(Faulty, &rdev->flags))
4218 /* complete an interrupted recovery */
4219 spares++;
4220 if (spares && mddev->pers->sync_request) {
4221 mddev->recovery = 0;
4222 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4223 mddev->sync_thread = md_register_thread(md_do_sync,
4224 mddev,
4225 "%s_resync");
4226 if (!mddev->sync_thread) {
4227 printk(KERN_ERR "%s: could not start resync"
4228 " thread...\n",
4229 mdname(mddev));
4230 /* leave the spares where they are, it shouldn't hurt */
4231 mddev->recovery = 0;
0b8c9de0 4232 }
5fd6c1dc
N		 4233 }
4234 }
0b8c9de0
N		 4235 md_wakeup_thread(mddev->thread);
4236 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 4237
44ce6294 4238 mddev->changed = 1;
d7603b7e 4239 md_new_event(mddev);
b62b7590 4240 sysfs_notify_dirent(mddev->sysfs_state);
0c3573f1
N		 4241 if (mddev->sysfs_action)
4242 sysfs_notify_dirent(mddev->sysfs_action);
a99ac971 4243 sysfs_notify(&mddev->kobj, NULL, "degraded");
ed9e1982 4244 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
1da177e4
LT		 4245 return 0;
4246}
4247
4248static int restart_array(mddev_t *mddev)
4249{
4250 struct gendisk *disk = mddev->gendisk;
1da177e4 4251
80fab1d7 4252 /* Complain if it has no devices */
1da177e4 4253 if (list_empty(&mddev->disks))
80fab1d7
AN		 4254 return -ENXIO;
4255 if (!mddev->pers)
4256 return -EINVAL;
4257 if (!mddev->ro)
4258 return -EBUSY;
4259 mddev->safemode = 0;
4260 mddev->ro = 0;
4261 set_disk_ro(disk, 0);
4262 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4263 mdname(mddev));
4264 /* Kick recovery or resync if necessary */
4265 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4266 md_wakeup_thread(mddev->thread);
4267 md_wakeup_thread(mddev->sync_thread);
b62b7590 4268 sysfs_notify_dirent(mddev->sysfs_state);
80fab1d7 4269 return 0;
1da177e4
LT		 4270}
4271
acc55e22
N		 4272/* similar to deny_write_access, but accounts for our holding a reference
4273 * to the file ourselves */
4274static int deny_bitmap_write_access(struct file * file)
4275{
4276 struct inode *inode = file->f_mapping->host;
4277
4278 spin_lock(&inode->i_lock);
4279 if (atomic_read(&inode->i_writecount) > 1) {
4280 spin_unlock(&inode->i_lock);
4281 return -ETXTBSY;
4282 }
4283 atomic_set(&inode->i_writecount, -1);
4284 spin_unlock(&inode->i_lock);
4285
4286 return 0;
4287}
4288
4289static void restore_bitmap_write_access(struct file *file)
4290{
4291 struct inode *inode = file->f_mapping->host;
4292
4293 spin_lock(&inode->i_lock);
4294 atomic_set(&inode->i_writecount, 1);
4295 spin_unlock(&inode->i_lock);
4296}
4297
9e653b63
N		 4298/* mode:
4299 * 0 - completely stop and dis-assemble array
4300 * 1 - switch to readonly
4301 * 2 - stop but do not disassemble array
4302 */
df5b20cf 4303static int do_md_stop(mddev_t * mddev, int mode, int is_open)
1da177e4
LT		 4304{
4305 int err = 0;
4306 struct gendisk *disk = mddev->gendisk;
c4647292 4307 mdk_rdev_t *rdev;
1da177e4 4308
f2ea68cf 4309 if (atomic_read(&mddev->openers) > is_open) {
df5b20cf
NB		 4310 printk("md: %s still in use.\n",mdname(mddev));
4311 return -EBUSY;
4312 }
4313
1da177e4 4314 if (mddev->pers) {
1da177e4
LT		 4315
4316 if (mddev->sync_thread) {
5fd6c1dc 4317 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4
LT		 4318 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4319 md_unregister_thread(mddev->sync_thread);
4320 mddev->sync_thread = NULL;
4321 }
4322
4323 del_timer_sync(&mddev->safemode_timer);
4324
9e653b63
N		 4325 switch(mode) {
4326 case 1: /* readonly */
1da177e4 4327 err = -ENXIO;
f91de92e 4328 if (mddev->ro==1)
1da177e4
LT		 4329 goto out;
4330 mddev->ro = 1;
9e653b63
N		 4331 break;
4332 case 0: /* disassemble */
4333 case 2: /* stop */
6b8b3e8a 4334 bitmap_flush(mddev);
a9701a30 4335 md_super_wait(mddev);
1da177e4
LT		 4336 if (mddev->ro)
4337 set_disk_ro(disk, 0);
409c57f3 4338
1da177e4 4339 mddev->pers->stop(mddev);
d1b5380c
N		 4340 mddev->queue->merge_bvec_fn = NULL;
4341 mddev->queue->unplug_fn = NULL;
041ae52e 4342 mddev->queue->backing_dev_info.congested_fn = NULL;
1da177e4 4343 module_put(mddev->pers->owner);
5fd3a17e
DW		 4344 if (mddev->pers->sync_request)
4345 mddev->private = &md_redundancy_group;
1da177e4 4346 mddev->pers = NULL;
4f54b0e9 4347 /* tell userspace to handle 'inactive' */
b62b7590 4348 sysfs_notify_dirent(mddev->sysfs_state);
0d4ca600 4349
c4647292
N		 4350 list_for_each_entry(rdev, &mddev->disks, same_set)
4351 if (rdev->raid_disk >= 0) {
4352 char nm[20];
4353 sprintf(nm, "rd%d", rdev->raid_disk);
4354 sysfs_remove_link(&mddev->kobj, nm);
4355 }
4356
0d4ca600 4357 set_capacity(disk, 0);
44ce6294 4358 mddev->changed = 1;
0d4ca600 4359
1da177e4
LT		 4360 if (mddev->ro)
4361 mddev->ro = 0;
4362 }
850b2b42 4363 if (!mddev->in_sync || mddev->flags) {
1da177e4
LT		 4364 /* mark array as shutdown cleanly */
4365 mddev->in_sync = 1;
850b2b42 4366 md_update_sb(mddev, 1);
1da177e4 4367 }
9e653b63 4368 if (mode == 1)
1da177e4 4369 set_disk_ro(disk, 1);
5fd6c1dc 4370 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4 4371 }
32a7627c 4372
1da177e4
LT		 4373 /*
4374 * Free resources if final stop
4375 */
9e653b63 4376 if (mode == 0) {
0d4ca600 4377
1da177e4
LT		 4378 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4379
978f946b
N		 4380 bitmap_destroy(mddev);
4381 if (mddev->bitmap_file) {
acc55e22 4382 restore_bitmap_write_access(mddev->bitmap_file);
978f946b
N		 4383 fput(mddev->bitmap_file);
4384 mddev->bitmap_file = NULL;
4385 }
4386 mddev->bitmap_offset = 0;
4387
177a99b2 4388 /* make sure all md_delayed_delete calls have finished */
5792a285
N		 4389 flush_scheduled_work();
4390
1da177e4
LT		 4391 export_array(mddev);
4392
f233ea5c 4393 mddev->array_sectors = 0;
b522adcd 4394 mddev->external_size = 0;
58c0fed4 4395 mddev->dev_sectors = 0;
9e653b63 4396 mddev->raid_disks = 0;
a94213b1 4397 mddev->recovery_cp = 0;
5e96ee65 4398 mddev->resync_min = 0;
c6207277 4399 mddev->resync_max = MaxSector;
08a02ecd 4400 mddev->reshape_position = MaxSector;
e691063a 4401 mddev->external = 0;
1ec4a939 4402 mddev->persistent = 0;
d897dbf9
N		 4403 mddev->level = LEVEL_NONE;
4404 mddev->clevel[0] = 0;
4405 mddev->flags = 0;
4406 mddev->ro = 0;
4407 mddev->metadata_type[0] = 0;
9d8f0363 4408 mddev->chunk_sectors = 0;
d897dbf9
N		 4409 mddev->ctime = mddev->utime = 0;
4410 mddev->layout = 0;
4411 mddev->max_disks = 0;
4412 mddev->events = 0;
4413 mddev->delta_disks = 0;
4414 mddev->new_level = LEVEL_NONE;
4415 mddev->new_layout = 0;
664e7c41 4416 mddev->new_chunk_sectors = 0;
d897dbf9
N		 4417 mddev->curr_resync = 0;
4418 mddev->resync_mismatches = 0;
4419 mddev->suspend_lo = mddev->suspend_hi = 0;
4420 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4421 mddev->recovery = 0;
4422 mddev->in_sync = 0;
4423 mddev->changed = 0;
4424 mddev->degraded = 0;
4425 mddev->barriers_work = 0;
4426 mddev->safemode = 0;
934d9c23 4427 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 4428 if (mddev->hold_active == UNTIL_STOP)
4429 mddev->hold_active = 0;
9e653b63 4430
a8a55c38 4431 } else if (mddev->pers)
1da177e4
LT		 4432 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4433 mdname(mddev));
4434 err = 0;
3f9d99c1 4435 blk_integrity_unregister(disk);
d7603b7e 4436 md_new_event(mddev);
b62b7590 4437 sysfs_notify_dirent(mddev->sysfs_state);
1da177e4
LT		 4438out:
4439 return err;
4440}
4441
fdee8ae4 4442#ifndef MODULE
1da177e4
LT		 4443static void autorun_array(mddev_t *mddev)
4444{
4445 mdk_rdev_t *rdev;
1da177e4
LT		 4446 int err;
4447
a757e64c 4448 if (list_empty(&mddev->disks))
1da177e4 4449 return;
1da177e4
LT		 4450
4451 printk(KERN_INFO "md: running: ");
4452
159ec1fc 4453 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 4454 char b[BDEVNAME_SIZE];
4455 printk("<%s>", bdevname(rdev->bdev,b));
4456 }
4457 printk("\n");
4458
d710e138 4459 err = do_md_run(mddev);
1da177e4
LT		 4460 if (err) {
4461 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 4462 do_md_stop(mddev, 0, 0);
1da177e4
LT		 4463 }
4464}
4465
4466/*
4467 * lets try to run arrays based on all disks that have arrived
4468 * until now. (those are in pending_raid_disks)
4469 *
4470 * the method: pick the first pending disk, collect all disks with
4471 * the same UUID, remove all from the pending list and put them into
4472 * the 'same_array' list. Then order this list based on superblock
4473 * update time (freshest comes first), kick out 'old' disks and
4474 * compare superblocks. If everything's fine then run it.
4475 *
4476 * If "unit" is allocated, then bump its reference count
4477 */
4478static void autorun_devices(int part)
4479{
159ec1fc 4480 mdk_rdev_t *rdev0, *rdev, *tmp;
1da177e4
LT		 4481 mddev_t *mddev;
4482 char b[BDEVNAME_SIZE];
4483
4484 printk(KERN_INFO "md: autorun ...\n");
4485 while (!list_empty(&pending_raid_disks)) {
e8703fe1 4486 int unit;
1da177e4 4487 dev_t dev;
ad01c9e3 4488 LIST_HEAD(candidates);
1da177e4
LT		 4489 rdev0 = list_entry(pending_raid_disks.next,
4490 mdk_rdev_t, same_set);
4491
4492 printk(KERN_INFO "md: considering %s ...\n",
4493 bdevname(rdev0->bdev,b));
4494 INIT_LIST_HEAD(&candidates);
159ec1fc 4495 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 4496 if (super_90_load(rdev, rdev0, 0) >= 0) {
4497 printk(KERN_INFO "md: adding %s ...\n",
4498 bdevname(rdev->bdev,b));
4499 list_move(&rdev->same_set, &candidates);
4500 }
4501 /*
4502 * now we have a set of devices, with all of them having
4503 * mostly sane superblocks. It's time to allocate the
4504 * mddev.
4505 */
e8703fe1
N		 4506 if (part) {
4507 dev = MKDEV(mdp_major,
4508 rdev0->preferred_minor << MdpMinorShift);
4509 unit = MINOR(dev) >> MdpMinorShift;
4510 } else {
4511 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4512 unit = MINOR(dev);
4513 }
4514 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 4515 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4516 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4517 break;
4518 }
1da177e4
LT		 4519
4520 md_probe(dev, NULL, NULL);
4521 mddev = mddev_find(dev);
9bbbca3a
NB		 4522 if (!mddev || !mddev->gendisk) {
4523 if (mddev)
4524 mddev_put(mddev);
4525 printk(KERN_ERR
1da177e4
LT		 4526 "md: cannot allocate memory for md drive.\n");
4527 break;
4528 }
4529 if (mddev_lock(mddev))
4530 printk(KERN_WARNING "md: %s locked, cannot run\n",
4531 mdname(mddev));
4532 else if (mddev->raid_disks || mddev->major_version
4533 || !list_empty(&mddev->disks)) {
4534 printk(KERN_WARNING
4535 "md: %s already running, cannot run %s\n",
4536 mdname(mddev), bdevname(rdev0->bdev,b));
4537 mddev_unlock(mddev);
4538 } else {
4539 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 4540 mddev->persistent = 1;
159ec1fc 4541 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 4542 list_del_init(&rdev->same_set);
4543 if (bind_rdev_to_array(rdev, mddev))
4544 export_rdev(rdev);
4545 }
4546 autorun_array(mddev);
4547 mddev_unlock(mddev);
4548 }
4549 /* on success, candidates will be empty, on error
4550 * it won't...
4551 */
159ec1fc 4552 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 4553 list_del_init(&rdev->same_set);
1da177e4 4554 export_rdev(rdev);
4b80991c 4555 }
1da177e4
LT		 4556 mddev_put(mddev);
4557 }
4558 printk(KERN_INFO "md: ... autorun DONE.\n");
4559}
fdee8ae4 4560#endif /* !MODULE */
1da177e4 4561
1da177e4
LT		 4562static int get_version(void __user * arg)
4563{
4564 mdu_version_t ver;
4565
4566 ver.major = MD_MAJOR_VERSION;
4567 ver.minor = MD_MINOR_VERSION;
4568 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4569
4570 if (copy_to_user(arg, &ver, sizeof(ver)))
4571 return -EFAULT;
4572
4573 return 0;
4574}
4575
4576static int get_array_info(mddev_t * mddev, void __user * arg)
4577{
4578 mdu_array_info_t info;
4579 int nr,working,active,failed,spare;
4580 mdk_rdev_t *rdev;
1da177e4
LT		 4581
4582 nr=working=active=failed=spare=0;
159ec1fc 4583 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 4584 nr++;
b2d444d7 4585 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4586 failed++;
4587 else {
4588 working++;
b2d444d7 4589 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 4590 active++;
4591 else
4592 spare++;
4593 }
4594 }
4595
4596 info.major_version = mddev->major_version;
4597 info.minor_version = mddev->minor_version;
4598 info.patch_version = MD_PATCHLEVEL_VERSION;
4599 info.ctime = mddev->ctime;
4600 info.level = mddev->level;
58c0fed4
AN		 4601 info.size = mddev->dev_sectors / 2;
4602 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 4603 info.size = -1;
1da177e4
LT		 4604 info.nr_disks = nr;
4605 info.raid_disks = mddev->raid_disks;
4606 info.md_minor = mddev->md_minor;
4607 info.not_persistent= !mddev->persistent;
4608
4609 info.utime = mddev->utime;
4610 info.state = 0;
4611 if (mddev->in_sync)
4612 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 4613 if (mddev->bitmap && mddev->bitmap_offset)
4614 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 4615 info.active_disks = active;
4616 info.working_disks = working;
4617 info.failed_disks = failed;
4618 info.spare_disks = spare;
4619
4620 info.layout = mddev->layout;
9d8f0363 4621 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 4622
4623 if (copy_to_user(arg, &info, sizeof(info)))
4624 return -EFAULT;
4625
4626 return 0;
4627}
4628
87162a28 4629static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 4630{
4631 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4632 char *ptr, *buf = NULL;
4633 int err = -ENOMEM;
4634
b5470dc5
DW		 4635 if (md_allow_write(mddev))
4636 file = kmalloc(sizeof(*file), GFP_NOIO);
4637 else
4638 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 4639
32a7627c
N		 4640 if (!file)
4641 goto out;
4642
4643 /* bitmap disabled, zero the first byte and copy out */
4644 if (!mddev->bitmap || !mddev->bitmap->file) {
4645 file->pathname[0] = '\0';
4646 goto copy_out;
4647 }
4648
4649 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4650 if (!buf)
4651 goto out;
4652
6bcfd601
CH		 4653 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4654 if (IS_ERR(ptr))
32a7627c
N		 4655 goto out;
4656
4657 strcpy(file->pathname, ptr);
4658
4659copy_out:
4660 err = 0;
4661 if (copy_to_user(arg, file, sizeof(*file)))
4662 err = -EFAULT;
4663out:
4664 kfree(buf);
4665 kfree(file);
4666 return err;
4667}
4668
1da177e4
LT		 4669static int get_disk_info(mddev_t * mddev, void __user * arg)
4670{
4671 mdu_disk_info_t info;
1da177e4
LT		 4672 mdk_rdev_t *rdev;
4673
4674 if (copy_from_user(&info, arg, sizeof(info)))
4675 return -EFAULT;
4676
26ef379f 4677 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT		 4678 if (rdev) {
4679 info.major = MAJOR(rdev->bdev->bd_dev);
4680 info.minor = MINOR(rdev->bdev->bd_dev);
4681 info.raid_disk = rdev->raid_disk;
4682 info.state = 0;
b2d444d7 4683 if (test_bit(Faulty, &rdev->flags))
1da177e4 4684 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 4685 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 4686 info.state |= (1<<MD_DISK_ACTIVE);
4687 info.state |= (1<<MD_DISK_SYNC);
4688 }
8ddf9efe
N		 4689 if (test_bit(WriteMostly, &rdev->flags))
4690 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 4691 } else {
4692 info.major = info.minor = 0;
4693 info.raid_disk = -1;
4694 info.state = (1<<MD_DISK_REMOVED);
4695 }
4696
4697 if (copy_to_user(arg, &info, sizeof(info)))
4698 return -EFAULT;
4699
4700 return 0;
4701}
4702
4703static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4704{
4705 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4706 mdk_rdev_t *rdev;
4707 dev_t dev = MKDEV(info->major,info->minor);
4708
4709 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4710 return -EOVERFLOW;
4711
4712 if (!mddev->raid_disks) {
4713 int err;
4714 /* expecting a device which has a superblock */
4715 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4716 if (IS_ERR(rdev)) {
4717 printk(KERN_WARNING
4718 "md: md_import_device returned %ld\n",
4719 PTR_ERR(rdev));
4720 return PTR_ERR(rdev);
4721 }
4722 if (!list_empty(&mddev->disks)) {
4723 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4724 mdk_rdev_t, same_set);
4725 int err = super_types[mddev->major_version]
4726 .load_super(rdev, rdev0, mddev->minor_version);
4727 if (err < 0) {
4728 printk(KERN_WARNING
4729 "md: %s has different UUID to %s\n",
4730 bdevname(rdev->bdev,b),
4731 bdevname(rdev0->bdev,b2));
4732 export_rdev(rdev);
4733 return -EINVAL;
4734 }
4735 }
4736 err = bind_rdev_to_array(rdev, mddev);
4737 if (err)
4738 export_rdev(rdev);
4739 return err;
4740 }
4741
4742 /*
4743 * add_new_disk can be used once the array is assembled
4744 * to add "hot spares". They must already have a superblock
4745 * written
4746 */
4747 if (mddev->pers) {
4748 int err;
4749 if (!mddev->pers->hot_add_disk) {
4750 printk(KERN_WARNING
4751 "%s: personality does not support diskops!\n",
4752 mdname(mddev));
4753 return -EINVAL;
4754 }
7b1e35f6
N		 4755 if (mddev->persistent)
4756 rdev = md_import_device(dev, mddev->major_version,
4757 mddev->minor_version);
4758 else
4759 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 4760 if (IS_ERR(rdev)) {
4761 printk(KERN_WARNING
4762 "md: md_import_device returned %ld\n",
4763 PTR_ERR(rdev));
4764 return PTR_ERR(rdev);
4765 }
41158c7e
N		 4766 /* set save_raid_disk if appropriate */
4767 if (!mddev->persistent) {
4768 if (info->state & (1<<MD_DISK_SYNC) &&
4769 info->raid_disk < mddev->raid_disks)
4770 rdev->raid_disk = info->raid_disk;
4771 else
4772 rdev->raid_disk = -1;
4773 } else
4774 super_types[mddev->major_version].
4775 validate_super(mddev, rdev);
4776 rdev->saved_raid_disk = rdev->raid_disk;
4777
b2d444d7 4778 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 4779 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4780 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 4781 else
4782 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 4783
1da177e4
LT		 4784 rdev->raid_disk = -1;
4785 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 4786 if (!err && !mddev->pers->hot_remove_disk) {
4787 /* If there is hot_add_disk but no hot_remove_disk
4788 * then added disks for geometry changes,
4789 * and should be added immediately.
4790 */
4791 super_types[mddev->major_version].
4792 validate_super(mddev, rdev);
4793 err = mddev->pers->hot_add_disk(mddev, rdev);
4794 if (err)
4795 unbind_rdev_from_array(rdev);
4796 }
1da177e4
LT		 4797 if (err)
4798 export_rdev(rdev);
52664732 4799 else
3c0ee63a 4800 sysfs_notify_dirent(rdev->sysfs_state);
c361777f 4801
17571284 4802 md_update_sb(mddev, 1);
72a23c21
NB		 4803 if (mddev->degraded)
4804 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 4805 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 4806 md_wakeup_thread(mddev->thread);
1da177e4
LT		 4807 return err;
4808 }
4809
4810 /* otherwise, add_new_disk is only allowed
4811 * for major_version==0 superblocks
4812 */
4813 if (mddev->major_version != 0) {
4814 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4815 mdname(mddev));
4816 return -EINVAL;
4817 }
4818
4819 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4820 int err;
d710e138 4821 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 4822 if (IS_ERR(rdev)) {
4823 printk(KERN_WARNING
4824 "md: error, md_import_device() returned %ld\n",
4825 PTR_ERR(rdev));
4826 return PTR_ERR(rdev);
4827 }
4828 rdev->desc_nr = info->number;
4829 if (info->raid_disk < mddev->raid_disks)
4830 rdev->raid_disk = info->raid_disk;
4831 else
4832 rdev->raid_disk = -1;
4833
1da177e4 4834 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 4835 if (info->state & (1<<MD_DISK_SYNC))
4836 set_bit(In_sync, &rdev->flags);
1da177e4 4837
8ddf9efe
N		 4838 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4839 set_bit(WriteMostly, &rdev->flags);
4840
1da177e4
LT		 4841 if (!mddev->persistent) {
4842 printk(KERN_INFO "md: nonpersistent superblock ...\n");
0f420358 4843 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
1da177e4 4844 } else
0f420358 4845 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
8190e754 4846 rdev->sectors = rdev->sb_start;
1da177e4 4847
2bf071bf
N		 4848 err = bind_rdev_to_array(rdev, mddev);
4849 if (err) {
4850 export_rdev(rdev);
4851 return err;
4852 }
1da177e4
LT		 4853 }
4854
4855 return 0;
4856}
4857
4858static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4859{
4860 char b[BDEVNAME_SIZE];
4861 mdk_rdev_t *rdev;
4862
1da177e4
LT		 4863 rdev = find_rdev(mddev, dev);
4864 if (!rdev)
4865 return -ENXIO;
4866
4867 if (rdev->raid_disk >= 0)
4868 goto busy;
4869
4870 kick_rdev_from_array(rdev);
850b2b42 4871 md_update_sb(mddev, 1);
d7603b7e 4872 md_new_event(mddev);
1da177e4
LT		 4873
4874 return 0;
4875busy:
fdefa4d8 4876 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 4877 bdevname(rdev->bdev,b), mdname(mddev));
4878 return -EBUSY;
4879}
4880
4881static int hot_add_disk(mddev_t * mddev, dev_t dev)
4882{
4883 char b[BDEVNAME_SIZE];
4884 int err;
1da177e4
LT		 4885 mdk_rdev_t *rdev;
4886
4887 if (!mddev->pers)
4888 return -ENODEV;
4889
4890 if (mddev->major_version != 0) {
4891 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4892 " version-0 superblocks.\n",
4893 mdname(mddev));
4894 return -EINVAL;
4895 }
4896 if (!mddev->pers->hot_add_disk) {
4897 printk(KERN_WARNING
4898 "%s: personality does not support diskops!\n",
4899 mdname(mddev));
4900 return -EINVAL;
4901 }
4902
d710e138 4903 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 4904 if (IS_ERR(rdev)) {
4905 printk(KERN_WARNING
4906 "md: error, md_import_device() returned %ld\n",
4907 PTR_ERR(rdev));
4908 return -EINVAL;
4909 }
4910
4911 if (mddev->persistent)
0f420358 4912 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1da177e4 4913 else
0f420358 4914 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
1da177e4 4915
8190e754 4916 rdev->sectors = rdev->sb_start;
1da177e4 4917
b2d444d7 4918 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4919 printk(KERN_WARNING
4920 "md: can not hot-add faulty %s disk to %s!\n",
4921 bdevname(rdev->bdev,b), mdname(mddev));
4922 err = -EINVAL;
4923 goto abort_export;
4924 }
b2d444d7 4925 clear_bit(In_sync, &rdev->flags);
1da177e4 4926 rdev->desc_nr = -1;
5842730d 4927 rdev->saved_raid_disk = -1;
2bf071bf
N		 4928 err = bind_rdev_to_array(rdev, mddev);
4929 if (err)
4930 goto abort_export;
1da177e4
LT		 4931
4932 /*
4933 * The rest should better be atomic, we can have disk failures
4934 * noticed in interrupt contexts ...
4935 */
4936
1da177e4
LT		 4937 rdev->raid_disk = -1;
4938
850b2b42 4939 md_update_sb(mddev, 1);
1da177e4
LT		 4940
4941 /*
4942 * Kick recovery, maybe this spare has to be added to the
4943 * array immediately.
4944 */
4945 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4946 md_wakeup_thread(mddev->thread);
d7603b7e 4947 md_new_event(mddev);
1da177e4
LT		 4948 return 0;
4949
1da177e4
LT		 4950abort_export:
4951 export_rdev(rdev);
4952 return err;
4953}
4954
32a7627c
N		 4955static int set_bitmap_file(mddev_t *mddev, int fd)
4956{
4957 int err;
4958
36fa3063
N		 4959 if (mddev->pers) {
4960 if (!mddev->pers->quiesce)
4961 return -EBUSY;
4962 if (mddev->recovery || mddev->sync_thread)
4963 return -EBUSY;
4964 /* we should be able to change the bitmap.. */
4965 }
32a7627c 4966
32a7627c 4967
36fa3063
N		 4968 if (fd >= 0) {
4969 if (mddev->bitmap)
4970 return -EEXIST; /* cannot add when bitmap is present */
4971 mddev->bitmap_file = fget(fd);
32a7627c 4972
36fa3063
N		 4973 if (mddev->bitmap_file == NULL) {
4974 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4975 mdname(mddev));
4976 return -EBADF;
4977 }
4978
4979 err = deny_bitmap_write_access(mddev->bitmap_file);
4980 if (err) {
4981 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4982 mdname(mddev));
4983 fput(mddev->bitmap_file);
4984 mddev->bitmap_file = NULL;
4985 return err;
4986 }
a654b9d8 4987 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 4988 } else if (mddev->bitmap == NULL)
4989 return -ENOENT; /* cannot remove what isn't there */
4990 err = 0;
4991 if (mddev->pers) {
4992 mddev->pers->quiesce(mddev, 1);
4993 if (fd >= 0)
4994 err = bitmap_create(mddev);
d7375ab3 4995 if (fd < 0 || err) {
36fa3063 4996 bitmap_destroy(mddev);
d7375ab3
N		 4997 fd = -1; /* make sure to put the file */
4998 }
36fa3063 4999 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 5000 }
5001 if (fd < 0) {
acc55e22
N		 5002 if (mddev->bitmap_file) {
5003 restore_bitmap_write_access(mddev->bitmap_file);
36fa3063 5004 fput(mddev->bitmap_file);
acc55e22 5005 }
36fa3063
N		 5006 mddev->bitmap_file = NULL;
5007 }
5008
32a7627c
N		 5009 return err;
5010}
5011
1da177e4
LT		 5012/*
5013 * set_array_info is used two different ways
5014 * The original usage is when creating a new array.
5015 * In this usage, raid_disks is > 0 and it together with
5016 * level, size, not_persistent,layout,chunksize determine the
5017 * shape of the array.
5018 * This will always create an array with a type-0.90.0 superblock.
5019 * The newer usage is when assembling an array.
5020 * In this case raid_disks will be 0, and the major_version field is
5021 * use to determine which style super-blocks are to be found on the devices.
5022 * The minor and patch _version numbers are also kept incase the
5023 * super_block handler wishes to interpret them.
5024 */
5025static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5026{
5027
5028 if (info->raid_disks == 0) {
5029 /* just setting version number for superblock loading */
5030 if (info->major_version < 0 ||
50511da3 5031 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 5032 super_types[info->major_version].name == NULL) {
5033 /* maybe try to auto-load a module? */
5034 printk(KERN_INFO
5035 "md: superblock version %d not known\n",
5036 info->major_version);
5037 return -EINVAL;
5038 }
5039 mddev->major_version = info->major_version;
5040 mddev->minor_version = info->minor_version;
5041 mddev->patch_version = info->patch_version;
3f9d7b0d 5042 mddev->persistent = !info->not_persistent;
1da177e4
LT		 5043 return 0;
5044 }
5045 mddev->major_version = MD_MAJOR_VERSION;
5046 mddev->minor_version = MD_MINOR_VERSION;
5047 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5048 mddev->ctime = get_seconds();
5049
5050 mddev->level = info->level;
17115e03 5051 mddev->clevel[0] = 0;
58c0fed4 5052 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 5053 mddev->raid_disks = info->raid_disks;
5054 /* don't set md_minor, it is determined by which /dev/md* was
5055 * openned
5056 */
5057 if (info->state & (1<<MD_SB_CLEAN))
5058 mddev->recovery_cp = MaxSector;
5059 else
5060 mddev->recovery_cp = 0;
5061 mddev->persistent = ! info->not_persistent;
e691063a 5062 mddev->external = 0;
1da177e4
LT		 5063
5064 mddev->layout = info->layout;
9d8f0363 5065 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 5066
5067 mddev->max_disks = MD_SB_DISKS;
5068
e691063a
N		 5069 if (mddev->persistent)
5070 mddev->flags = 0;
850b2b42 5071 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5072
b2a2703c
N		 5073 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5074 mddev->bitmap_offset = 0;
5075
f6705578
N		 5076 mddev->reshape_position = MaxSector;
5077
1da177e4
LT		 5078 /*
5079 * Generate a 128 bit UUID
5080 */
5081 get_random_bytes(mddev->uuid, 16);
5082
f6705578 5083 mddev->new_level = mddev->level;
664e7c41 5084 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 5085 mddev->new_layout = mddev->layout;
5086 mddev->delta_disks = 0;
5087
1da177e4
LT		 5088 return 0;
5089}
5090
1f403624
DW		 5091void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5092{
b522adcd
DW		 5093 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5094
5095 if (mddev->external_size)
5096 return;
5097
1f403624
DW		 5098 mddev->array_sectors = array_sectors;
5099}
5100EXPORT_SYMBOL(md_set_array_sectors);
5101
d71f9f88 5102static int update_size(mddev_t *mddev, sector_t num_sectors)
a35b0d69 5103{
159ec1fc 5104 mdk_rdev_t *rdev;
a35b0d69 5105 int rv;
d71f9f88 5106 int fit = (num_sectors == 0);
a35b0d69
N		 5107
5108 if (mddev->pers->resize == NULL)
5109 return -EINVAL;
d71f9f88
AN		 5110 /* The "num_sectors" is the number of sectors of each device that
5111 * is used. This can only make sense for arrays with redundancy.
5112 * linear and raid0 always use whatever space is available. We can only
5113 * consider changing this number if no resync or reconstruction is
5114 * happening, and if the new size is acceptable. It must fit before the
0f420358 5115 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 5116 * of each device. If num_sectors is zero, we find the largest size
5117 * that fits.
5118
a35b0d69
N		 5119 */
5120 if (mddev->sync_thread)
5121 return -EBUSY;
dba034ee
N		 5122 if (mddev->bitmap)
5123 /* Sorry, cannot grow a bitmap yet, just remove it,
5124 * grow, and re-add.
5125 */
5126 return -EBUSY;
159ec1fc 5127 list_for_each_entry(rdev, &mddev->disks, same_set) {
dd8ac336 5128 sector_t avail = rdev->sectors;
01ab5662 5129
d71f9f88
AN		 5130 if (fit && (num_sectors == 0 || num_sectors > avail))
5131 num_sectors = avail;
5132 if (avail < num_sectors)
a35b0d69
N		 5133 return -ENOSPC;
5134 }
d71f9f88 5135 rv = mddev->pers->resize(mddev, num_sectors);
a35b0d69
N		 5136 if (!rv) {
5137 struct block_device *bdev;
5138
5139 bdev = bdget_disk(mddev->gendisk, 0);
5140 if (bdev) {
1b1dcc1b 5141 mutex_lock(&bdev->bd_inode->i_mutex);
f233ea5c
AN		 5142 i_size_write(bdev->bd_inode,
5143 (loff_t)mddev->array_sectors << 9);
1b1dcc1b 5144 mutex_unlock(&bdev->bd_inode->i_mutex);
a35b0d69
N		 5145 bdput(bdev);
5146 }
5147 }
5148 return rv;
5149}
5150
da943b99
N		 5151static int update_raid_disks(mddev_t *mddev, int raid_disks)
5152{
5153 int rv;
5154 /* change the number of raid disks */
63c70c4f 5155 if (mddev->pers->check_reshape == NULL)
da943b99
N		 5156 return -EINVAL;
5157 if (raid_disks <= 0 ||
5158 raid_disks >= mddev->max_disks)
5159 return -EINVAL;
63c70c4f 5160 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 5161 return -EBUSY;
63c70c4f
N		 5162 mddev->delta_disks = raid_disks - mddev->raid_disks;
5163
5164 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 5165 return rv;
5166}
5167
5168
1da177e4
LT		 5169/*
5170 * update_array_info is used to change the configuration of an
5171 * on-line array.
5172 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5173 * fields in the info are checked against the array.
5174 * Any differences that cannot be handled will cause an error.
5175 * Normally, only one change can be managed at a time.
5176 */
5177static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5178{
5179 int rv = 0;
5180 int cnt = 0;
36fa3063
N		 5181 int state = 0;
5182
5183 /* calculate expected state,ignoring low bits */
5184 if (mddev->bitmap && mddev->bitmap_offset)
5185 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 5186
5187 if (mddev->major_version != info->major_version ||
5188 mddev->minor_version != info->minor_version ||
5189/* mddev->patch_version != info->patch_version || */
5190 mddev->ctime != info->ctime ||
5191 mddev->level != info->level ||
5192/* mddev->layout != info->layout || */
5193 !mddev->persistent != info->not_persistent||
9d8f0363 5194 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 5195 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5196 ((state^info->state) & 0xfffffe00)
5197 )
1da177e4
LT		 5198 return -EINVAL;
5199 /* Check there is only one change */
58c0fed4
AN		 5200 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5201 cnt++;
5202 if (mddev->raid_disks != info->raid_disks)
5203 cnt++;
5204 if (mddev->layout != info->layout)
5205 cnt++;
5206 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5207 cnt++;
5208 if (cnt == 0)
5209 return 0;
5210 if (cnt > 1)
5211 return -EINVAL;
1da177e4
LT		 5212
5213 if (mddev->layout != info->layout) {
5214 /* Change layout
5215 * we don't need to do anything at the md level, the
5216 * personality will take care of it all.
5217 */
50ac168a 5218 if (mddev->pers->check_reshape == NULL)
1da177e4 5219 return -EINVAL;
597a711b
N		 5220 else {
5221 mddev->new_layout = info->layout;
50ac168a 5222 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 5223 if (rv)
5224 mddev->new_layout = mddev->layout;
5225 return rv;
5226 }
1da177e4 5227 }
58c0fed4 5228 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 5229 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 5230
da943b99
N		 5231 if (mddev->raid_disks != info->raid_disks)
5232 rv = update_raid_disks(mddev, info->raid_disks);
5233
36fa3063
N		 5234 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5235 if (mddev->pers->quiesce == NULL)
5236 return -EINVAL;
5237 if (mddev->recovery || mddev->sync_thread)
5238 return -EBUSY;
5239 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5240 /* add the bitmap */
5241 if (mddev->bitmap)
5242 return -EEXIST;
5243 if (mddev->default_bitmap_offset == 0)
5244 return -EINVAL;
5245 mddev->bitmap_offset = mddev->default_bitmap_offset;
5246 mddev->pers->quiesce(mddev, 1);
5247 rv = bitmap_create(mddev);
5248 if (rv)
5249 bitmap_destroy(mddev);
5250 mddev->pers->quiesce(mddev, 0);
5251 } else {
5252 /* remove the bitmap */
5253 if (!mddev->bitmap)
5254 return -ENOENT;
5255 if (mddev->bitmap->file)
5256 return -EINVAL;
5257 mddev->pers->quiesce(mddev, 1);
5258 bitmap_destroy(mddev);
5259 mddev->pers->quiesce(mddev, 0);
5260 mddev->bitmap_offset = 0;
5261 }
5262 }
850b2b42 5263 md_update_sb(mddev, 1);
1da177e4
LT		 5264 return rv;
5265}
5266
5267static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5268{
5269 mdk_rdev_t *rdev;
5270
5271 if (mddev->pers == NULL)
5272 return -ENODEV;
5273
5274 rdev = find_rdev(mddev, dev);
5275 if (!rdev)
5276 return -ENODEV;
5277
5278 md_error(mddev, rdev);
5279 return 0;
5280}
5281
2f9618ce
AN		 5282/*
5283 * We have a problem here : there is no easy way to give a CHS
5284 * virtual geometry. We currently pretend that we have a 2 heads
5285 * 4 sectors (with a BIG number of cylinders...). This drives
5286 * dosfs just mad... ;-)
5287 */
a885c8c4
CH		 5288static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5289{
5290 mddev_t *mddev = bdev->bd_disk->private_data;
5291
5292 geo->heads = 2;
5293 geo->sectors = 4;
5294 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5295 return 0;
5296}
5297
a39907fa 5298static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 5299 unsigned int cmd, unsigned long arg)
5300{
5301 int err = 0;
5302 void __user *argp = (void __user *)arg;
1da177e4
LT		 5303 mddev_t *mddev = NULL;
5304
5305 if (!capable(CAP_SYS_ADMIN))
5306 return -EACCES;
5307
5308 /*
5309 * Commands dealing with the RAID driver but not any
5310 * particular array:
5311 */
5312 switch (cmd)
5313 {
5314 case RAID_VERSION:
5315 err = get_version(argp);
5316 goto done;
5317
5318 case PRINT_RAID_DEBUG:
5319 err = 0;
5320 md_print_devices();
5321 goto done;
5322
5323#ifndef MODULE
5324 case RAID_AUTORUN:
5325 err = 0;
5326 autostart_arrays(arg);
5327 goto done;
5328#endif
5329 default:;
5330 }
5331
5332 /*
5333 * Commands creating/starting a new array:
5334 */
5335
a39907fa 5336 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 5337
5338 if (!mddev) {
5339 BUG();
5340 goto abort;
5341 }
5342
1da177e4
LT		 5343 err = mddev_lock(mddev);
5344 if (err) {
5345 printk(KERN_INFO
5346 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5347 err, cmd);
5348 goto abort;
5349 }
5350
5351 switch (cmd)
5352 {
5353 case SET_ARRAY_INFO:
5354 {
5355 mdu_array_info_t info;
5356 if (!arg)
5357 memset(&info, 0, sizeof(info));
5358 else if (copy_from_user(&info, argp, sizeof(info))) {
5359 err = -EFAULT;
5360 goto abort_unlock;
5361 }
5362 if (mddev->pers) {
5363 err = update_array_info(mddev, &info);
5364 if (err) {
5365 printk(KERN_WARNING "md: couldn't update"
5366 " array info. %d\n", err);
5367 goto abort_unlock;
5368 }
5369 goto done_unlock;
5370 }
5371 if (!list_empty(&mddev->disks)) {
5372 printk(KERN_WARNING
5373 "md: array %s already has disks!\n",
5374 mdname(mddev));
5375 err = -EBUSY;
5376 goto abort_unlock;
5377 }
5378 if (mddev->raid_disks) {
5379 printk(KERN_WARNING
5380 "md: array %s already initialised!\n",
5381 mdname(mddev));
5382 err = -EBUSY;
5383 goto abort_unlock;
5384 }
5385 err = set_array_info(mddev, &info);
5386 if (err) {
5387 printk(KERN_WARNING "md: couldn't set"
5388 " array info. %d\n", err);
5389 goto abort_unlock;
5390 }
5391 }
5392 goto done_unlock;
5393
5394 default:;
5395 }
5396
5397 /*
5398 * Commands querying/configuring an existing array:
5399 */
32a7627c 5400 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 5401 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 5402 if ((!mddev->raid_disks && !mddev->external)
5403 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5404 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5405 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 5406 err = -ENODEV;
5407 goto abort_unlock;
5408 }
5409
5410 /*
5411 * Commands even a read-only array can execute:
5412 */
5413 switch (cmd)
5414 {
5415 case GET_ARRAY_INFO:
5416 err = get_array_info(mddev, argp);
5417 goto done_unlock;
5418
32a7627c 5419 case GET_BITMAP_FILE:
87162a28 5420 err = get_bitmap_file(mddev, argp);
32a7627c
N		 5421 goto done_unlock;
5422
1da177e4
LT		 5423 case GET_DISK_INFO:
5424 err = get_disk_info(mddev, argp);
5425 goto done_unlock;
5426
5427 case RESTART_ARRAY_RW:
5428 err = restart_array(mddev);
5429 goto done_unlock;
5430
5431 case STOP_ARRAY:
d710e138 5432 err = do_md_stop(mddev, 0, 1);
1da177e4
LT		 5433 goto done_unlock;
5434
5435 case STOP_ARRAY_RO:
d710e138 5436 err = do_md_stop(mddev, 1, 1);
1da177e4
LT		 5437 goto done_unlock;
5438
1da177e4
LT		 5439 }
5440
5441 /*
5442 * The remaining ioctls are changing the state of the
f91de92e
N		 5443 * superblock, so we do not allow them on read-only arrays.
5444 * However non-MD ioctls (e.g. get-size) will still come through
5445 * here and hit the 'default' below, so only disallow
5446 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 5447 */
bb57fc64 5448 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 5449 if (mddev->ro == 2) {
5450 mddev->ro = 0;
b62b7590 5451 sysfs_notify_dirent(mddev->sysfs_state);
0fd62b86
NB		 5452 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5453 md_wakeup_thread(mddev->thread);
f91de92e
N		 5454 } else {
5455 err = -EROFS;
5456 goto abort_unlock;
5457 }
1da177e4
LT		 5458 }
5459
5460 switch (cmd)
5461 {
5462 case ADD_NEW_DISK:
5463 {
5464 mdu_disk_info_t info;
5465 if (copy_from_user(&info, argp, sizeof(info)))
5466 err = -EFAULT;
5467 else
5468 err = add_new_disk(mddev, &info);
5469 goto done_unlock;
5470 }
5471
5472 case HOT_REMOVE_DISK:
5473 err = hot_remove_disk(mddev, new_decode_dev(arg));
5474 goto done_unlock;
5475
5476 case HOT_ADD_DISK:
5477 err = hot_add_disk(mddev, new_decode_dev(arg));
5478 goto done_unlock;
5479
5480 case SET_DISK_FAULTY:
5481 err = set_disk_faulty(mddev, new_decode_dev(arg));
5482 goto done_unlock;
5483
5484 case RUN_ARRAY:
d710e138 5485 err = do_md_run(mddev);
1da177e4
LT		 5486 goto done_unlock;
5487
32a7627c
N		 5488 case SET_BITMAP_FILE:
5489 err = set_bitmap_file(mddev, (int)arg);
5490 goto done_unlock;
5491
1da177e4 5492 default:
1da177e4
LT		 5493 err = -EINVAL;
5494 goto abort_unlock;
5495 }
5496
5497done_unlock:
5498abort_unlock:
d3374825
N		 5499 if (mddev->hold_active == UNTIL_IOCTL &&
5500 err != -EINVAL)
5501 mddev->hold_active = 0;
1da177e4
LT		 5502 mddev_unlock(mddev);
5503
5504 return err;
5505done:
5506 if (err)
5507 MD_BUG();
5508abort:
5509 return err;
5510}
5511
a39907fa 5512static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 5513{
5514 /*
5515 * Succeed if we can lock the mddev, which confirms that
5516 * it isn't being stopped right now.
5517 */
d3374825 5518 mddev_t *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 5519 int err;
5520
d3374825
N		 5521 if (mddev->gendisk != bdev->bd_disk) {
5522 /* we are racing with mddev_put which is discarding this
5523 * bd_disk.
5524 */
5525 mddev_put(mddev);
5526 /* Wait until bdev->bd_disk is definitely gone */
5527 flush_scheduled_work();
5528 /* Then retry the open from the top */
5529 return -ERESTARTSYS;
5530 }
5531 BUG_ON(mddev != bdev->bd_disk->private_data);
5532
d63a5a74 5533 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
1da177e4
LT		 5534 goto out;
5535
5536 err = 0;
f2ea68cf 5537 atomic_inc(&mddev->openers);
1da177e4
LT		 5538 mddev_unlock(mddev);
5539
a39907fa 5540 check_disk_change(bdev);
1da177e4
LT		 5541 out:
5542 return err;
5543}
5544
a39907fa 5545static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 5546{
a39907fa 5547 mddev_t *mddev = disk->private_data;
1da177e4 5548
52e5f9d1 5549 BUG_ON(!mddev);
f2ea68cf 5550 atomic_dec(&mddev->openers);
1da177e4
LT		 5551 mddev_put(mddev);
5552
5553 return 0;
5554}
5555
44ce6294
LT		 5556static int md_media_changed(struct gendisk *disk)
5557{
5558 mddev_t *mddev = disk->private_data;
5559
5560 return mddev->changed;
5561}
5562
5563static int md_revalidate(struct gendisk *disk)
5564{
5565 mddev_t *mddev = disk->private_data;
5566
5567 mddev->changed = 0;
5568 return 0;
5569}
1da177e4
LT		 5570static struct block_device_operations md_fops =
5571{
5572 .owner = THIS_MODULE,
a39907fa
AV		 5573 .open = md_open,
5574 .release = md_release,
b492b852 5575 .ioctl = md_ioctl,
a885c8c4 5576 .getgeo = md_getgeo,
44ce6294
LT		 5577 .media_changed = md_media_changed,
5578 .revalidate_disk= md_revalidate,
1da177e4
LT		 5579};
5580
75c96f85 5581static int md_thread(void * arg)
1da177e4
LT		 5582{
5583 mdk_thread_t *thread = arg;
5584
1da177e4
LT		 5585 /*
5586 * md_thread is a 'system-thread', it's priority should be very
5587 * high. We avoid resource deadlocks individually in each
5588 * raid personality. (RAID5 does preallocation) We also use RR and
5589 * the very same RT priority as kswapd, thus we will never get
5590 * into a priority inversion deadlock.
5591 *
5592 * we definitely have to have equal or higher priority than
5593 * bdflush, otherwise bdflush will deadlock if there are too
5594 * many dirty RAID5 blocks.
5595 */
1da177e4 5596
6985c43f 5597 allow_signal(SIGKILL);
a6fb0934 5598 while (!kthread_should_stop()) {
1da177e4 5599
93588e22
N		 5600 /* We need to wait INTERRUPTIBLE so that
5601 * we don't add to the load-average.
5602 * That means we need to be sure no signals are
5603 * pending
5604 */
5605 if (signal_pending(current))
5606 flush_signals(current);
5607
5608 wait_event_interruptible_timeout
5609 (thread->wqueue,
5610 test_bit(THREAD_WAKEUP, &thread->flags)
5611 || kthread_should_stop(),
5612 thread->timeout);
1da177e4
LT		 5613
5614 clear_bit(THREAD_WAKEUP, &thread->flags);
5615
787453c2 5616 thread->run(thread->mddev);
1da177e4 5617 }
a6fb0934 5618
1da177e4
LT		 5619 return 0;
5620}
5621
5622void md_wakeup_thread(mdk_thread_t *thread)
5623{
5624 if (thread) {
5625 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5626 set_bit(THREAD_WAKEUP, &thread->flags);
5627 wake_up(&thread->wqueue);
5628 }
5629}
5630
5631mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5632 const char *name)
5633{
5634 mdk_thread_t *thread;
1da177e4 5635
9ffae0cf 5636 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 5637 if (!thread)
5638 return NULL;
5639
1da177e4
LT		 5640 init_waitqueue_head(&thread->wqueue);
5641
1da177e4
LT		 5642 thread->run = run;
5643 thread->mddev = mddev;
32a7627c 5644 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 5645 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 5646 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 5647 kfree(thread);
5648 return NULL;
5649 }
1da177e4
LT		 5650 return thread;
5651}
5652
1da177e4
LT		 5653void md_unregister_thread(mdk_thread_t *thread)
5654{
e0cf8f04
N		 5655 if (!thread)
5656 return;
ba25f9dc 5657 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
a6fb0934
N		 5658
5659 kthread_stop(thread->tsk);
1da177e4
LT		 5660 kfree(thread);
5661}
5662
5663void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5664{
5665 if (!mddev) {
5666 MD_BUG();
5667 return;
5668 }
5669
b2d444d7 5670 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 5671 return;
6bfe0b49
DW		 5672
5673 if (mddev->external)
5674 set_bit(Blocked, &rdev->flags);
32a7627c 5675/*
1da177e4
LT		 5676 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5677 mdname(mddev),
5678 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5679 __builtin_return_address(0),__builtin_return_address(1),
5680 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 5681*/
d0a0a5ee
AM		 5682 if (!mddev->pers)
5683 return;
1da177e4
LT		 5684 if (!mddev->pers->error_handler)
5685 return;
5686 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 5687 if (mddev->degraded)
5688 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
52664732 5689 set_bit(StateChanged, &rdev->flags);
1da177e4
LT		 5690 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5691 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5692 md_wakeup_thread(mddev->thread);
c331eb04 5693 md_new_event_inintr(mddev);
1da177e4
LT		 5694}
5695
5696/* seq_file implementation /proc/mdstat */
5697
5698static void status_unused(struct seq_file *seq)
5699{
5700 int i = 0;
5701 mdk_rdev_t *rdev;
1da177e4
LT		 5702
5703 seq_printf(seq, "unused devices: ");
5704
159ec1fc 5705 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 5706 char b[BDEVNAME_SIZE];
5707 i++;
5708 seq_printf(seq, "%s ",
5709 bdevname(rdev->bdev,b));
5710 }
5711 if (!i)
5712 seq_printf(seq, "<none>");
5713
5714 seq_printf(seq, "\n");
5715}
5716
5717
5718static void status_resync(struct seq_file *seq, mddev_t * mddev)
5719{
dd71cf6b
N		 5720 sector_t max_sectors, resync, res;
5721 unsigned long dt, db;
5722 sector_t rt;
4588b42e
N		 5723 int scale;
5724 unsigned int per_milli;
1da177e4 5725
dd71cf6b 5726 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
1da177e4
LT		 5727
5728 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
dd71cf6b 5729 max_sectors = mddev->resync_max_sectors;
1da177e4 5730 else
dd71cf6b 5731 max_sectors = mddev->dev_sectors;
1da177e4
LT		 5732
5733 /*
5734 * Should not happen.
5735 */
dd71cf6b 5736 if (!max_sectors) {
1da177e4
LT		 5737 MD_BUG();
5738 return;
5739 }
4588b42e 5740 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 5741 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 5742 * u32, as those are the requirements for sector_div.
5743 * Thus 'scale' must be at least 10
5744 */
5745 scale = 10;
5746 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 5747 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 5748 scale++;
5749 }
5750 res = (resync>>scale)*1000;
dd71cf6b 5751 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 5752
5753 per_milli = res;
1da177e4 5754 {
4588b42e 5755 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 5756 seq_printf(seq, "[");
5757 for (i = 0; i < x; i++)
5758 seq_printf(seq, "=");
5759 seq_printf(seq, ">");
5760 for (i = 0; i < y; i++)
5761 seq_printf(seq, ".");
5762 seq_printf(seq, "] ");
5763 }
4588b42e 5764 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 5765 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5766 "reshape" :
61df9d91
N		 5767 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5768 "check" :
5769 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5770 "resync" : "recovery"))),
5771 per_milli/10, per_milli % 10,
dd71cf6b
N		 5772 (unsigned long long) resync/2,
5773 (unsigned long long) max_sectors/2);
1da177e4
LT		 5774
5775 /*
1da177e4
LT		 5776 * dt: time from mark until now
5777 * db: blocks written from mark until now
5778 * rt: remaining time
dd71cf6b
N		 5779 *
5780 * rt is a sector_t, so could be 32bit or 64bit.
5781 * So we divide before multiply in case it is 32bit and close
5782 * to the limit.
5783 * We scale the divisor (db) by 32 to avoid loosing precision
5784 * near the end of resync when the number of remaining sectors
5785 * is close to 'db'.
5786 * We then divide rt by 32 after multiplying by db to compensate.
5787 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 5788 */
5789 dt = ((jiffies - mddev->resync_mark) / HZ);
5790 if (!dt) dt++;
ff4e8d9a
N		 5791 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5792 - mddev->resync_mark_cnt;
1da177e4 5793
dd71cf6b
N		 5794 rt = max_sectors - resync; /* number of remaining sectors */
5795 sector_div(rt, db/32+1);
5796 rt *= dt;
5797 rt >>= 5;
5798
5799 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5800 ((unsigned long)rt % 60)/6);
1da177e4 5801
ff4e8d9a 5802 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 5803}
5804
5805static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5806{
5807 struct list_head *tmp;
5808 loff_t l = *pos;
5809 mddev_t *mddev;
5810
5811 if (l >= 0x10000)
5812 return NULL;
5813 if (!l--)
5814 /* header */
5815 return (void*)1;
5816
5817 spin_lock(&all_mddevs_lock);
5818 list_for_each(tmp,&all_mddevs)
5819 if (!l--) {
5820 mddev = list_entry(tmp, mddev_t, all_mddevs);
5821 mddev_get(mddev);
5822 spin_unlock(&all_mddevs_lock);
5823 return mddev;
5824 }
5825 spin_unlock(&all_mddevs_lock);
5826 if (!l--)
5827 return (void*)2;/* tail */
5828 return NULL;
5829}
5830
5831static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5832{
5833 struct list_head *tmp;
5834 mddev_t *next_mddev, *mddev = v;
5835
5836 ++*pos;
5837 if (v == (void*)2)
5838 return NULL;
5839
5840 spin_lock(&all_mddevs_lock);
5841 if (v == (void*)1)
5842 tmp = all_mddevs.next;
5843 else
5844 tmp = mddev->all_mddevs.next;
5845 if (tmp != &all_mddevs)
5846 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5847 else {
5848 next_mddev = (void*)2;
5849 *pos = 0x10000;
5850 }
5851 spin_unlock(&all_mddevs_lock);
5852
5853 if (v != (void*)1)
5854 mddev_put(mddev);
5855 return next_mddev;
5856
5857}
5858
5859static void md_seq_stop(struct seq_file *seq, void *v)
5860{
5861 mddev_t *mddev = v;
5862
5863 if (mddev && v != (void*)1 && v != (void*)2)
5864 mddev_put(mddev);
5865}
5866
d7603b7e
N		 5867struct mdstat_info {
5868 int event;
5869};
5870
1da177e4
LT		 5871static int md_seq_show(struct seq_file *seq, void *v)
5872{
5873 mddev_t *mddev = v;
dd8ac336 5874 sector_t sectors;
1da177e4 5875 mdk_rdev_t *rdev;
d7603b7e 5876 struct mdstat_info *mi = seq->private;
32a7627c 5877 struct bitmap *bitmap;
1da177e4
LT		 5878
5879 if (v == (void*)1) {
2604b703 5880 struct mdk_personality *pers;
1da177e4
LT		 5881 seq_printf(seq, "Personalities : ");
5882 spin_lock(&pers_lock);
2604b703
N		 5883 list_for_each_entry(pers, &pers_list, list)
5884 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 5885
5886 spin_unlock(&pers_lock);
5887 seq_printf(seq, "\n");
d7603b7e 5888 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 5889 return 0;
5890 }
5891 if (v == (void*)2) {
5892 status_unused(seq);
5893 return 0;
5894 }
5895
5dc5cf7d 5896 if (mddev_lock(mddev) < 0)
1da177e4 5897 return -EINTR;
5dc5cf7d 5898
1da177e4
LT		 5899 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5900 seq_printf(seq, "%s : %sactive", mdname(mddev),
5901 mddev->pers ? "" : "in");
5902 if (mddev->pers) {
f91de92e 5903 if (mddev->ro==1)
1da177e4 5904 seq_printf(seq, " (read-only)");
f91de92e 5905 if (mddev->ro==2)
52720ae7 5906 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 5907 seq_printf(seq, " %s", mddev->pers->name);
5908 }
5909
dd8ac336 5910 sectors = 0;
159ec1fc 5911 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 5912 char b[BDEVNAME_SIZE];
5913 seq_printf(seq, " %s[%d]",
5914 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 5915 if (test_bit(WriteMostly, &rdev->flags))
5916 seq_printf(seq, "(W)");
b2d444d7 5917 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5918 seq_printf(seq, "(F)");
5919 continue;
b325a32e
N		 5920 } else if (rdev->raid_disk < 0)
5921 seq_printf(seq, "(S)"); /* spare */
dd8ac336 5922 sectors += rdev->sectors;
1da177e4
LT		 5923 }
5924
5925 if (!list_empty(&mddev->disks)) {
5926 if (mddev->pers)
5927 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 5928 (unsigned long long)
5929 mddev->array_sectors / 2);
1da177e4
LT		 5930 else
5931 seq_printf(seq, "\n %llu blocks",
dd8ac336 5932 (unsigned long long)sectors / 2);
1da177e4 5933 }
1cd6bf19
N		 5934 if (mddev->persistent) {
5935 if (mddev->major_version != 0 ||
5936 mddev->minor_version != 90) {
5937 seq_printf(seq," super %d.%d",
5938 mddev->major_version,
5939 mddev->minor_version);
5940 }
e691063a
N		 5941 } else if (mddev->external)
5942 seq_printf(seq, " super external:%s",
5943 mddev->metadata_type);
5944 else
1cd6bf19 5945 seq_printf(seq, " super non-persistent");
1da177e4
LT		 5946
5947 if (mddev->pers) {
d710e138 5948 mddev->pers->status(seq, mddev);
1da177e4 5949 seq_printf(seq, "\n ");
8e1b39d6
N		 5950 if (mddev->pers->sync_request) {
5951 if (mddev->curr_resync > 2) {
d710e138 5952 status_resync(seq, mddev);
8e1b39d6
N		 5953 seq_printf(seq, "\n ");
5954 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5955 seq_printf(seq, "\tresync=DELAYED\n ");
5956 else if (mddev->recovery_cp < MaxSector)
5957 seq_printf(seq, "\tresync=PENDING\n ");
5958 }
32a7627c
N		 5959 } else
5960 seq_printf(seq, "\n ");
5961
5962 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 5963 unsigned long chunk_kb;
5964 unsigned long flags;
32a7627c
N		 5965 spin_lock_irqsave(&bitmap->lock, flags);
5966 chunk_kb = bitmap->chunksize >> 10;
5967 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5968 "%lu%s chunk",
5969 bitmap->pages - bitmap->missing_pages,
5970 bitmap->pages,
5971 (bitmap->pages - bitmap->missing_pages)
5972 << (PAGE_SHIFT - 10),
5973 chunk_kb ? chunk_kb : bitmap->chunksize,
5974 chunk_kb ? "KB" : "B");
78d742d8
N		 5975 if (bitmap->file) {
5976 seq_printf(seq, ", file: ");
c32c2f63 5977 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 5978 }
78d742d8 5979
32a7627c
N		 5980 seq_printf(seq, "\n");
5981 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 5982 }
5983
5984 seq_printf(seq, "\n");
5985 }
5986 mddev_unlock(mddev);
5987
5988 return 0;
5989}
5990
110518bc 5991static const struct seq_operations md_seq_ops = {
1da177e4
LT		 5992 .start = md_seq_start,
5993 .next = md_seq_next,
5994 .stop = md_seq_stop,
5995 .show = md_seq_show,
5996};
5997
5998static int md_seq_open(struct inode *inode, struct file *file)
5999{
6000 int error;
d7603b7e
N		 6001 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6002 if (mi == NULL)
6003 return -ENOMEM;
1da177e4
LT		 6004
6005 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 6006 if (error)
6007 kfree(mi);
6008 else {
6009 struct seq_file *p = file->private_data;
6010 p->private = mi;
6011 mi->event = atomic_read(&md_event_count);
6012 }
1da177e4
LT		 6013 return error;
6014}
6015
d7603b7e
N		 6016static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6017{
6018 struct seq_file *m = filp->private_data;
6019 struct mdstat_info *mi = m->private;
6020 int mask;
6021
6022 poll_wait(filp, &md_event_waiters, wait);
6023
6024 /* always allow read */
6025 mask = POLLIN | POLLRDNORM;
6026
6027 if (mi->event != atomic_read(&md_event_count))
6028 mask |= POLLERR | POLLPRI;
6029 return mask;
6030}
6031
fa027c2a 6032static const struct file_operations md_seq_fops = {
e24650c2 6033 .owner = THIS_MODULE,
1da177e4
LT		 6034 .open = md_seq_open,
6035 .read = seq_read,
6036 .llseek = seq_lseek,
c3f94b40 6037 .release = seq_release_private,
d7603b7e 6038 .poll = mdstat_poll,
1da177e4
LT		 6039};
6040
2604b703 6041int register_md_personality(struct mdk_personality *p)
1da177e4 6042{
1da177e4 6043 spin_lock(&pers_lock);
2604b703
N		 6044 list_add_tail(&p->list, &pers_list);
6045 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 6046 spin_unlock(&pers_lock);
6047 return 0;
6048}
6049
2604b703 6050int unregister_md_personality(struct mdk_personality *p)
1da177e4 6051{
2604b703 6052 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 6053 spin_lock(&pers_lock);
2604b703 6054 list_del_init(&p->list);
1da177e4
LT		 6055 spin_unlock(&pers_lock);
6056 return 0;
6057}
6058
eea1bf38 6059static int is_mddev_idle(mddev_t *mddev, int init)
1da177e4
LT		 6060{
6061 mdk_rdev_t * rdev;
1da177e4 6062 int idle;
eea1bf38 6063 int curr_events;
1da177e4
LT		 6064
6065 idle = 1;
4b80991c
N		 6066 rcu_read_lock();
6067 rdev_for_each_rcu(rdev, mddev) {
1da177e4 6068 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 6069 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6070 (int)part_stat_read(&disk->part0, sectors[1]) -
6071 atomic_read(&disk->sync_io);
713f6ab1
N		 6072 /* sync IO will cause sync_io to increase before the disk_stats
6073 * as sync_io is counted when a request starts, and
6074 * disk_stats is counted when it completes.
6075 * So resync activity will cause curr_events to be smaller than
6076 * when there was no such activity.
6077 * non-sync IO will cause disk_stat to increase without
6078 * increasing sync_io so curr_events will (eventually)
6079 * be larger than it was before. Once it becomes
6080 * substantially larger, the test below will cause
6081 * the array to appear non-idle, and resync will slow
6082 * down.
6083 * If there is a lot of outstanding resync activity when
6084 * we set last_event to curr_events, then all that activity
6085 * completing might cause the array to appear non-idle
6086 * and resync will be slowed down even though there might
6087 * not have been non-resync activity. This will only
6088 * happen once though. 'last_events' will soon reflect
6089 * the state where there is little or no outstanding
6090 * resync requests, and further resync activity will
6091 * always make curr_events less than last_events.
c0e48521 6092 *
1da177e4 6093 */
eea1bf38 6094 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 6095 rdev->last_events = curr_events;
6096 idle = 0;
6097 }
6098 }
4b80991c 6099 rcu_read_unlock();
1da177e4
LT		 6100 return idle;
6101}
6102
6103void md_done_sync(mddev_t *mddev, int blocks, int ok)
6104{
6105 /* another "blocks" (512byte) blocks have been synced */
6106 atomic_sub(blocks, &mddev->recovery_active);
6107 wake_up(&mddev->recovery_wait);
6108 if (!ok) {
dfc70645 6109 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6110 md_wakeup_thread(mddev->thread);
6111 // stop recovery, signal do_sync ....
6112 }
6113}
6114
6115
06d91a5f
N		 6116/* md_write_start(mddev, bi)
6117 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 6118 * in superblock) before writing, schedule a superblock update
6119 * and wait for it to complete.
06d91a5f 6120 */
3d310eb7 6121void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 6122{
0fd62b86 6123 int did_change = 0;
06d91a5f 6124 if (bio_data_dir(bi) != WRITE)
3d310eb7 6125 return;
06d91a5f 6126
f91de92e
N		 6127 BUG_ON(mddev->ro == 1);
6128 if (mddev->ro == 2) {
6129 /* need to switch to read/write */
6130 mddev->ro = 0;
6131 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6132 md_wakeup_thread(mddev->thread);
25156198 6133 md_wakeup_thread(mddev->sync_thread);
0fd62b86 6134 did_change = 1;
f91de92e 6135 }
06d91a5f 6136 atomic_inc(&mddev->writes_pending);
31a59e34
N		 6137 if (mddev->safemode == 1)
6138 mddev->safemode = 0;
06d91a5f 6139 if (mddev->in_sync) {
a9701a30 6140 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 6141 if (mddev->in_sync) {
6142 mddev->in_sync = 0;
850b2b42 6143 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3d310eb7 6144 md_wakeup_thread(mddev->thread);
0fd62b86 6145 did_change = 1;
3d310eb7 6146 }
a9701a30 6147 spin_unlock_irq(&mddev->write_lock);
06d91a5f 6148 }
0fd62b86 6149 if (did_change)
b62b7590 6150 sysfs_notify_dirent(mddev->sysfs_state);
09a44cc1
N		 6151 wait_event(mddev->sb_wait,
6152 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6153 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 6154}
6155
6156void md_write_end(mddev_t *mddev)
6157{
6158 if (atomic_dec_and_test(&mddev->writes_pending)) {
6159 if (mddev->safemode == 2)
6160 md_wakeup_thread(mddev->thread);
16f17b39 6161 else if (mddev->safemode_delay)
1da177e4
LT		 6162 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6163 }
6164}
6165
2a2275d6
N		 6166/* md_allow_write(mddev)
6167 * Calling this ensures that the array is marked 'active' so that writes
6168 * may proceed without blocking. It is important to call this before
6169 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6170 * Must be called with mddev_lock held.
b5470dc5
DW		 6171 *
6172 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6173 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 6174 */
b5470dc5 6175int md_allow_write(mddev_t *mddev)
2a2275d6
N		 6176{
6177 if (!mddev->pers)
b5470dc5 6178 return 0;
2a2275d6 6179 if (mddev->ro)
b5470dc5 6180 return 0;
1a0fd497 6181 if (!mddev->pers->sync_request)
b5470dc5 6182 return 0;
2a2275d6
N		 6183
6184 spin_lock_irq(&mddev->write_lock);
6185 if (mddev->in_sync) {
6186 mddev->in_sync = 0;
6187 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6188 if (mddev->safemode_delay &&
6189 mddev->safemode == 0)
6190 mddev->safemode = 1;
6191 spin_unlock_irq(&mddev->write_lock);
6192 md_update_sb(mddev, 0);
b62b7590 6193 sysfs_notify_dirent(mddev->sysfs_state);
2a2275d6
N		 6194 } else
6195 spin_unlock_irq(&mddev->write_lock);
b5470dc5
DW		 6196
6197 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6198 return -EAGAIN;
6199 else
6200 return 0;
2a2275d6
N		 6201}
6202EXPORT_SYMBOL_GPL(md_allow_write);
6203
1da177e4
LT		 6204#define SYNC_MARKS 10
6205#define SYNC_MARK_STEP (3*HZ)
29269553 6206void md_do_sync(mddev_t *mddev)
1da177e4
LT		 6207{
6208 mddev_t *mddev2;
6209 unsigned int currspeed = 0,
6210 window;
57afd89f 6211 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 6212 unsigned long mark[SYNC_MARKS];
6213 sector_t mark_cnt[SYNC_MARKS];
6214 int last_mark,m;
6215 struct list_head *tmp;
6216 sector_t last_check;
57afd89f 6217 int skipped = 0;
5fd6c1dc 6218 mdk_rdev_t *rdev;
61df9d91 6219 char *desc;
1da177e4
LT		 6220
6221 /* just incase thread restarts... */
6222 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6223 return;
5fd6c1dc
N		 6224 if (mddev->ro) /* never try to sync a read-only array */
6225 return;
1da177e4 6226
61df9d91
N		 6227 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6228 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6229 desc = "data-check";
6230 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6231 desc = "requested-resync";
6232 else
6233 desc = "resync";
6234 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6235 desc = "reshape";
6236 else
6237 desc = "recovery";
6238
1da177e4
LT		 6239 /* we overload curr_resync somewhat here.
6240 * 0 == not engaged in resync at all
6241 * 2 == checking that there is no conflict with another sync
6242 * 1 == like 2, but have yielded to allow conflicting resync to
6243 * commense
6244 * other == active in resync - this many blocks
6245 *
6246 * Before starting a resync we must have set curr_resync to
6247 * 2, and then checked that every "conflicting" array has curr_resync
6248 * less than ours. When we find one that is the same or higher
6249 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6250 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6251 * This will mean we have to start checking from the beginning again.
6252 *
6253 */
6254
6255 do {
6256 mddev->curr_resync = 2;
6257
6258 try_again:
787453c2 6259 if (kthread_should_stop()) {
6985c43f 6260 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6261 goto skip;
6262 }
29ac4aa3 6263 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 6264 if (mddev2 == mddev)
6265 continue;
90b08710
BS		 6266 if (!mddev->parallel_resync
6267 && mddev2->curr_resync
6268 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 6269 DEFINE_WAIT(wq);
6270 if (mddev < mddev2 && mddev->curr_resync == 2) {
6271 /* arbitrarily yield */
6272 mddev->curr_resync = 1;
6273 wake_up(&resync_wait);
6274 }
6275 if (mddev > mddev2 && mddev->curr_resync == 1)
6276 /* no need to wait here, we can wait the next
6277 * time 'round when curr_resync == 2
6278 */
6279 continue;
9744197c
N		 6280 /* We need to wait 'interruptible' so as not to
6281 * contribute to the load average, and not to
6282 * be caught by 'softlockup'
6283 */
6284 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 6285 if (!kthread_should_stop() &&
8712e553 6286 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 6287 printk(KERN_INFO "md: delaying %s of %s"
6288 " until %s has finished (they"
1da177e4 6289 " share one or more physical units)\n",
61df9d91 6290 desc, mdname(mddev), mdname(mddev2));
1da177e4 6291 mddev_put(mddev2);
9744197c
N		 6292 if (signal_pending(current))
6293 flush_signals(current);
1da177e4
LT		 6294 schedule();
6295 finish_wait(&resync_wait, &wq);
6296 goto try_again;
6297 }
6298 finish_wait(&resync_wait, &wq);
6299 }
6300 }
6301 } while (mddev->curr_resync < 2);
6302
5fd6c1dc 6303 j = 0;
9d88883e 6304 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 6305 /* resync follows the size requested by the personality,
57afd89f 6306 * which defaults to physical size, but can be virtual size
1da177e4
LT		 6307 */
6308 max_sectors = mddev->resync_max_sectors;
9d88883e 6309 mddev->resync_mismatches = 0;
5fd6c1dc 6310 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 6311 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6312 j = mddev->resync_min;
6313 else if (!mddev->bitmap)
5fd6c1dc 6314 j = mddev->recovery_cp;
5e96ee65 6315
ccfcc3c1 6316 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 6317 max_sectors = mddev->dev_sectors;
5fd6c1dc 6318 else {
1da177e4 6319 /* recovery follows the physical size of devices */
58c0fed4 6320 max_sectors = mddev->dev_sectors;
5fd6c1dc 6321 j = MaxSector;
159ec1fc 6322 list_for_each_entry(rdev, &mddev->disks, same_set)
5fd6c1dc
N		 6323 if (rdev->raid_disk >= 0 &&
6324 !test_bit(Faulty, &rdev->flags) &&
6325 !test_bit(In_sync, &rdev->flags) &&
6326 rdev->recovery_offset < j)
6327 j = rdev->recovery_offset;
6328 }
1da177e4 6329
61df9d91
N		 6330 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6331 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6332 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 6333 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 6334 "(but not more than %d KB/sec) for %s.\n",
6335 speed_max(mddev), desc);
1da177e4 6336
eea1bf38 6337 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 6338
57afd89f 6339 io_sectors = 0;
1da177e4
LT		 6340 for (m = 0; m < SYNC_MARKS; m++) {
6341 mark[m] = jiffies;
57afd89f 6342 mark_cnt[m] = io_sectors;
1da177e4
LT		 6343 }
6344 last_mark = 0;
6345 mddev->resync_mark = mark[last_mark];
6346 mddev->resync_mark_cnt = mark_cnt[last_mark];
6347
6348 /*
6349 * Tune reconstruction:
6350 */
6351 window = 32*(PAGE_SIZE/512);
6352 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6353 window/2,(unsigned long long) max_sectors/2);
6354
6355 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 6356 last_check = 0;
6357
6358 if (j>2) {
6359 printk(KERN_INFO
61df9d91
N		 6360 "md: resuming %s of %s from checkpoint.\n",
6361 desc, mdname(mddev));
1da177e4
LT		 6362 mddev->curr_resync = j;
6363 }
6364
6365 while (j < max_sectors) {
57afd89f 6366 sector_t sectors;
1da177e4 6367
57afd89f 6368 skipped = 0;
97e4f42d 6369
7a91ee1f
N		 6370 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6371 ((mddev->curr_resync > mddev->curr_resync_completed &&
6372 (mddev->curr_resync - mddev->curr_resync_completed)
6373 > (max_sectors >> 4)) ||
6374 (j - mddev->curr_resync_completed)*2
6375 >= mddev->resync_max - mddev->curr_resync_completed
6376 )) {
97e4f42d
N		 6377 /* time to update curr_resync_completed */
6378 blk_unplug(mddev->queue);
6379 wait_event(mddev->recovery_wait,
6380 atomic_read(&mddev->recovery_active) == 0);
6381 mddev->curr_resync_completed =
6382 mddev->curr_resync;
6383 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 6384 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 6385 }
acb180b0 6386
e62e58a5
N		 6387 while (j >= mddev->resync_max && !kthread_should_stop()) {
6388 /* As this condition is controlled by user-space,
6389 * we can block indefinitely, so use '_interruptible'
6390 * to avoid triggering warnings.
6391 */
6392 flush_signals(current); /* just in case */
6393 wait_event_interruptible(mddev->recovery_wait,
6394 mddev->resync_max > j
6395 || kthread_should_stop());
6396 }
acb180b0
N		 6397
6398 if (kthread_should_stop())
6399 goto interrupted;
6400
57afd89f 6401 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 6402 currspeed < speed_min(mddev));
57afd89f 6403 if (sectors == 0) {
dfc70645 6404 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6405 goto out;
6406 }
57afd89f
N		 6407
6408 if (!skipped) { /* actual IO requested */
6409 io_sectors += sectors;
6410 atomic_add(sectors, &mddev->recovery_active);
6411 }
6412
1da177e4
LT		 6413 j += sectors;
6414 if (j>1) mddev->curr_resync = j;
ff4e8d9a 6415 mddev->curr_mark_cnt = io_sectors;
d7603b7e
N		 6416 if (last_check == 0)
6417 /* this is the earliers that rebuilt will be
6418 * visible in /proc/mdstat
6419 */
6420 md_new_event(mddev);
57afd89f
N		 6421
6422 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 6423 continue;
6424
57afd89f 6425 last_check = io_sectors;
1da177e4 6426
dfc70645 6427 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4
LT		 6428 break;
6429
6430 repeat:
6431 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6432 /* step marks */
6433 int next = (last_mark+1) % SYNC_MARKS;
6434
6435 mddev->resync_mark = mark[next];
6436 mddev->resync_mark_cnt = mark_cnt[next];
6437 mark[next] = jiffies;
57afd89f 6438 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6439 last_mark = next;
6440 }
6441
6442
c6207277
N		 6443 if (kthread_should_stop())
6444 goto interrupted;
6445
1da177e4
LT		 6446
6447 /*
6448 * this loop exits only if either when we are slower than
6449 * the 'hard' speed limit, or the system was IO-idle for
6450 * a jiffy.
6451 * the system might be non-idle CPU-wise, but we only care
6452 * about not overloading the IO subsystem. (things like an
6453 * e2fsck being done on the RAID array should execute fast)
6454 */
2ad8b1ef 6455 blk_unplug(mddev->queue);
1da177e4
LT		 6456 cond_resched();
6457
57afd89f
N		 6458 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6459 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 6460
88202a0c
N		 6461 if (currspeed > speed_min(mddev)) {
6462 if ((currspeed > speed_max(mddev)) ||
eea1bf38 6463 !is_mddev_idle(mddev, 0)) {
c0e48521 6464 msleep(500);
1da177e4
LT		 6465 goto repeat;
6466 }
6467 }
6468 }
61df9d91 6469 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 6470 /*
6471 * this also signals 'finished resyncing' to md_stop
6472 */
6473 out:
2ad8b1ef 6474 blk_unplug(mddev->queue);
1da177e4
LT		 6475
6476 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6477
6478 /* tell personality that we are finished */
57afd89f 6479 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 6480
dfc70645 6481 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 6482 mddev->curr_resync > 2) {
6483 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6484 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6485 if (mddev->curr_resync >= mddev->recovery_cp) {
6486 printk(KERN_INFO
61df9d91
N		 6487 "md: checkpointing %s of %s.\n",
6488 desc, mdname(mddev));
5fd6c1dc
N		 6489 mddev->recovery_cp = mddev->curr_resync;
6490 }
6491 } else
6492 mddev->recovery_cp = MaxSector;
6493 } else {
6494 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6495 mddev->curr_resync = MaxSector;
159ec1fc 6496 list_for_each_entry(rdev, &mddev->disks, same_set)
5fd6c1dc
N		 6497 if (rdev->raid_disk >= 0 &&
6498 !test_bit(Faulty, &rdev->flags) &&
6499 !test_bit(In_sync, &rdev->flags) &&
6500 rdev->recovery_offset < mddev->curr_resync)
6501 rdev->recovery_offset = mddev->curr_resync;
5fd6c1dc 6502 }
1da177e4 6503 }
17571284 6504 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6505
1da177e4
LT		 6506 skip:
6507 mddev->curr_resync = 0;
acb180b0 6508 mddev->curr_resync_completed = 0;
5e96ee65 6509 mddev->resync_min = 0;
c6207277
N		 6510 mddev->resync_max = MaxSector;
6511 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
1da177e4
LT		 6512 wake_up(&resync_wait);
6513 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6514 md_wakeup_thread(mddev->thread);
c6207277
N		 6515 return;
6516
6517 interrupted:
6518 /*
6519 * got a signal, exit.
6520 */
6521 printk(KERN_INFO
6522 "md: md_do_sync() got signal ... exiting\n");
6523 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6524 goto out;
6525
1da177e4 6526}
29269553 6527EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 6528
6529
b4c4c7b8
N		 6530static int remove_and_add_spares(mddev_t *mddev)
6531{
6532 mdk_rdev_t *rdev;
b4c4c7b8
N		 6533 int spares = 0;
6534
97e4f42d
N		 6535 mddev->curr_resync_completed = 0;
6536
159ec1fc 6537 list_for_each_entry(rdev, &mddev->disks, same_set)
b4c4c7b8 6538 if (rdev->raid_disk >= 0 &&
6bfe0b49 6539 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 6540 (test_bit(Faulty, &rdev->flags) ||
6541 ! test_bit(In_sync, &rdev->flags)) &&
6542 atomic_read(&rdev->nr_pending)==0) {
6543 if (mddev->pers->hot_remove_disk(
6544 mddev, rdev->raid_disk)==0) {
6545 char nm[20];
6546 sprintf(nm,"rd%d", rdev->raid_disk);
6547 sysfs_remove_link(&mddev->kobj, nm);
6548 rdev->raid_disk = -1;
6549 }
6550 }
6551
4044ba58 6552 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
159ec1fc 6553 list_for_each_entry(rdev, &mddev->disks, same_set) {
dfc70645 6554 if (rdev->raid_disk >= 0 &&
e5427135
DW		 6555 !test_bit(In_sync, &rdev->flags) &&
6556 !test_bit(Blocked, &rdev->flags))
dfc70645 6557 spares++;
b4c4c7b8
N		 6558 if (rdev->raid_disk < 0
6559 && !test_bit(Faulty, &rdev->flags)) {
6560 rdev->recovery_offset = 0;
199050ea
NB		 6561 if (mddev->pers->
6562 hot_add_disk(mddev, rdev) == 0) {
b4c4c7b8
N		 6563 char nm[20];
6564 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 6565 if (sysfs_create_link(&mddev->kobj,
6566 &rdev->kobj, nm))
6567 printk(KERN_WARNING
6568 "md: cannot register "
6569 "%s for %s\n",
6570 nm, mdname(mddev));
b4c4c7b8
N		 6571 spares++;
6572 md_new_event(mddev);
6573 } else
6574 break;
6575 }
dfc70645 6576 }
b4c4c7b8
N		 6577 }
6578 return spares;
6579}
1da177e4
LT		 6580/*
6581 * This routine is regularly called by all per-raid-array threads to
6582 * deal with generic issues like resync and super-block update.
6583 * Raid personalities that don't have a thread (linear/raid0) do not
6584 * need this as they never do any recovery or update the superblock.
6585 *
6586 * It does not do any resync itself, but rather "forks" off other threads
6587 * to do that as needed.
6588 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6589 * "->recovery" and create a thread at ->sync_thread.
dfc70645 6590 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 6591 * and wakeups up this thread which will reap the thread and finish up.
6592 * This thread also removes any faulty devices (with nr_pending == 0).
6593 *
6594 * The overall approach is:
6595 * 1/ if the superblock needs updating, update it.
6596 * 2/ If a recovery thread is running, don't do anything else.
6597 * 3/ If recovery has finished, clean up, possibly marking spares active.
6598 * 4/ If there are any faulty devices, remove them.
6599 * 5/ If array is degraded, try to add spares devices
6600 * 6/ If array has spares or is not in-sync, start a resync thread.
6601 */
6602void md_check_recovery(mddev_t *mddev)
6603{
6604 mdk_rdev_t *rdev;
1da177e4
LT		 6605
6606
5f40402d
N		 6607 if (mddev->bitmap)
6608 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 6609
6610 if (mddev->ro)
6611 return;
fca4d848
N		 6612
6613 if (signal_pending(current)) {
31a59e34 6614 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 6615 printk(KERN_INFO "md: %s in immediate safe mode\n",
6616 mdname(mddev));
6617 mddev->safemode = 2;
6618 }
6619 flush_signals(current);
6620 }
6621
c89a8eee
N		 6622 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6623 return;
1da177e4 6624 if ( ! (
e691063a 6625 (mddev->flags && !mddev->external) ||
1da177e4 6626 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 6627 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 6628 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 6629 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6630 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 6631 ))
6632 return;
fca4d848 6633
df5b89b3 6634 if (mddev_trylock(mddev)) {
b4c4c7b8 6635 int spares = 0;
fca4d848 6636
c89a8eee
N		 6637 if (mddev->ro) {
6638 /* Only thing we do on a ro array is remove
6639 * failed devices.
6640 */
6641 remove_and_add_spares(mddev);
6642 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6643 goto unlock;
6644 }
6645
31a59e34 6646 if (!mddev->external) {
0fd62b86 6647 int did_change = 0;
31a59e34
N		 6648 spin_lock_irq(&mddev->write_lock);
6649 if (mddev->safemode &&
6650 !atomic_read(&mddev->writes_pending) &&
6651 !mddev->in_sync &&
6652 mddev->recovery_cp == MaxSector) {
6653 mddev->in_sync = 1;
0fd62b86 6654 did_change = 1;
31a59e34
N		 6655 if (mddev->persistent)
6656 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6657 }
6658 if (mddev->safemode == 1)
6659 mddev->safemode = 0;
6660 spin_unlock_irq(&mddev->write_lock);
0fd62b86 6661 if (did_change)
b62b7590 6662 sysfs_notify_dirent(mddev->sysfs_state);
fca4d848 6663 }
fca4d848 6664
850b2b42
N		 6665 if (mddev->flags)
6666 md_update_sb(mddev, 0);
06d91a5f 6667
159ec1fc 6668 list_for_each_entry(rdev, &mddev->disks, same_set)
52664732 6669 if (test_and_clear_bit(StateChanged, &rdev->flags))
3c0ee63a 6670 sysfs_notify_dirent(rdev->sysfs_state);
52664732 6671
06d91a5f 6672
1da177e4
LT		 6673 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6674 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6675 /* resync/recovery still happening */
6676 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6677 goto unlock;
6678 }
6679 if (mddev->sync_thread) {
6680 /* resync has finished, collect result */
6681 md_unregister_thread(mddev->sync_thread);
6682 mddev->sync_thread = NULL;
56ac36d7
DW		 6683 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6684 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1da177e4
LT		 6685 /* success...*/
6686 /* activate any spares */
a99ac971
NB		 6687 if (mddev->pers->spare_active(mddev))
6688 sysfs_notify(&mddev->kobj, NULL,
6689 "degraded");
1da177e4 6690 }
cea9c228
N		 6691 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6692 mddev->pers->finish_reshape)
6693 mddev->pers->finish_reshape(mddev);
850b2b42 6694 md_update_sb(mddev, 1);
41158c7e
N		 6695
6696 /* if array is no-longer degraded, then any saved_raid_disk
6697 * information must be scrapped
6698 */
6699 if (!mddev->degraded)
159ec1fc 6700 list_for_each_entry(rdev, &mddev->disks, same_set)
41158c7e
N		 6701 rdev->saved_raid_disk = -1;
6702
1da177e4
LT		 6703 mddev->recovery = 0;
6704 /* flag recovery needed just to double check */
6705 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
0c3573f1 6706 sysfs_notify_dirent(mddev->sysfs_action);
d7603b7e 6707 md_new_event(mddev);
1da177e4
LT		 6708 goto unlock;
6709 }
72a23c21
NB		 6710 /* Set RUNNING before clearing NEEDED to avoid
6711 * any transients in the value of "sync_action".
6712 */
6713 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6714 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 6715 /* Clear some bits that don't mean anything, but
6716 * might be left set
6717 */
24dd469d
N		 6718 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6719 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 6720
5fd6c1dc
N		 6721 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6722 goto unlock;
1da177e4
LT		 6723 /* no recovery is running.
6724 * remove any failed drives, then
6725 * add spares if possible.
6726 * Spare are also removed and re-added, to allow
6727 * the personality to fail the re-add.
6728 */
1da177e4 6729
b4c4c7b8 6730 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 6731 if (mddev->pers->check_reshape == NULL ||
6732 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 6733 /* Cannot proceed */
6734 goto unlock;
6735 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 6736 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 6737 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 6738 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6739 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 6740 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 6741 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 6742 } else if (mddev->recovery_cp < MaxSector) {
6743 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 6744 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 6745 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6746 /* nothing to be done ... */
1da177e4 6747 goto unlock;
24dd469d 6748
1da177e4 6749 if (mddev->pers->sync_request) {
a654b9d8
N		 6750 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6751 /* We are adding a device or devices to an array
6752 * which has the bitmap stored on all devices.
6753 * So make sure all bitmap pages get written
6754 */
6755 bitmap_write_all(mddev->bitmap);
6756 }
1da177e4
LT		 6757 mddev->sync_thread = md_register_thread(md_do_sync,
6758 mddev,
6759 "%s_resync");
6760 if (!mddev->sync_thread) {
6761 printk(KERN_ERR "%s: could not start resync"
6762 " thread...\n",
6763 mdname(mddev));
6764 /* leave the spares where they are, it shouldn't hurt */
6765 mddev->recovery = 0;
d7603b7e 6766 } else
1da177e4 6767 md_wakeup_thread(mddev->sync_thread);
0c3573f1 6768 sysfs_notify_dirent(mddev->sysfs_action);
d7603b7e 6769 md_new_event(mddev);
1da177e4
LT		 6770 }
6771 unlock:
72a23c21
NB		 6772 if (!mddev->sync_thread) {
6773 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6774 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6775 &mddev->recovery))
0c3573f1
N		 6776 if (mddev->sysfs_action)
6777 sysfs_notify_dirent(mddev->sysfs_action);
72a23c21 6778 }
1da177e4
LT		 6779 mddev_unlock(mddev);
6780 }
6781}
6782
6bfe0b49
DW		 6783void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6784{
3c0ee63a 6785 sysfs_notify_dirent(rdev->sysfs_state);
6bfe0b49
DW		 6786 wait_event_timeout(rdev->blocked_wait,
6787 !test_bit(Blocked, &rdev->flags),
6788 msecs_to_jiffies(5000));
6789 rdev_dec_pending(rdev, mddev);
6790}
6791EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6792
75c96f85
AB		 6793static int md_notify_reboot(struct notifier_block *this,
6794 unsigned long code, void *x)
1da177e4
LT		 6795{
6796 struct list_head *tmp;
6797 mddev_t *mddev;
6798
6799 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6800
6801 printk(KERN_INFO "md: stopping all md devices.\n");
6802
29ac4aa3 6803 for_each_mddev(mddev, tmp)
c71d4887 6804 if (mddev_trylock(mddev)) {
2b25000b
N		 6805 /* Force a switch to readonly even array
6806 * appears to still be in use. Hence
6807 * the '100'.
6808 */
d710e138 6809 do_md_stop(mddev, 1, 100);
c71d4887
NB		 6810 mddev_unlock(mddev);
6811 }
1da177e4
LT		 6812 /*
6813 * certain more exotic SCSI devices are known to be
6814 * volatile wrt too early system reboots. While the
6815 * right place to handle this issue is the given
6816 * driver, we do want to have a safe RAID driver ...
6817 */
6818 mdelay(1000*1);
6819 }
6820 return NOTIFY_DONE;
6821}
6822
75c96f85 6823static struct notifier_block md_notifier = {
1da177e4
LT		 6824 .notifier_call = md_notify_reboot,
6825 .next = NULL,
6826 .priority = INT_MAX, /* before any real devices */
6827};
6828
6829static void md_geninit(void)
6830{
1da177e4
LT		 6831 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6832
c7705f34 6833 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 6834}
6835
75c96f85 6836static int __init md_init(void)
1da177e4 6837{
3dbd8c2e 6838 if (register_blkdev(MD_MAJOR, "md"))
1da177e4
LT		 6839 return -1;
6840 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3dbd8c2e 6841 unregister_blkdev(MD_MAJOR, "md");
1da177e4
LT		 6842 return -1;
6843 }
3dbd8c2e 6844 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 6845 md_probe, NULL, NULL);
6846 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 6847 md_probe, NULL, NULL);
6848
1da177e4 6849 register_reboot_notifier(&md_notifier);
0b4d4147 6850 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 6851
6852 md_geninit();
d710e138 6853 return 0;
1da177e4
LT		 6854}
6855
6856
6857#ifndef MODULE
6858
6859/*
6860 * Searches all registered partitions for autorun RAID arrays
6861 * at boot time.
6862 */
4d936ec1
ME		 6863
6864static LIST_HEAD(all_detected_devices);
6865struct detected_devices_node {
6866 struct list_head list;
6867 dev_t dev;
6868};
1da177e4
LT		 6869
6870void md_autodetect_dev(dev_t dev)
6871{
4d936ec1
ME		 6872 struct detected_devices_node *node_detected_dev;
6873
6874 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6875 if (node_detected_dev) {
6876 node_detected_dev->dev = dev;
6877 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6878 } else {
6879 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6880 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6881 }
1da177e4
LT		 6882}
6883
6884
6885static void autostart_arrays(int part)
6886{
6887 mdk_rdev_t *rdev;
4d936ec1
ME		 6888 struct detected_devices_node *node_detected_dev;
6889 dev_t dev;
6890 int i_scanned, i_passed;
1da177e4 6891
4d936ec1
ME		 6892 i_scanned = 0;
6893 i_passed = 0;
1da177e4 6894
4d936ec1 6895 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 6896
4d936ec1
ME		 6897 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6898 i_scanned++;
6899 node_detected_dev = list_entry(all_detected_devices.next,
6900 struct detected_devices_node, list);
6901 list_del(&node_detected_dev->list);
6902 dev = node_detected_dev->dev;
6903 kfree(node_detected_dev);
df968c4e 6904 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 6905 if (IS_ERR(rdev))
6906 continue;
6907
b2d444d7 6908 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6909 MD_BUG();
6910 continue;
6911 }
d0fae18f 6912 set_bit(AutoDetected, &rdev->flags);
1da177e4 6913 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 6914 i_passed++;
1da177e4 6915 }
4d936ec1
ME		 6916
6917 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6918 i_scanned, i_passed);
1da177e4
LT		 6919
6920 autorun_devices(part);
6921}
6922
fdee8ae4 6923#endif /* !MODULE */
1da177e4
LT		 6924
6925static __exit void md_exit(void)
6926{
6927 mddev_t *mddev;
6928 struct list_head *tmp;
8ab5e4c1 6929
3dbd8c2e 6930 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 6931 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 6932
3dbd8c2e 6933 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 6934 unregister_blkdev(mdp_major, "mdp");
6935 unregister_reboot_notifier(&md_notifier);
6936 unregister_sysctl_table(raid_table_header);
6937 remove_proc_entry("mdstat", NULL);
29ac4aa3 6938 for_each_mddev(mddev, tmp) {
1da177e4 6939 export_array(mddev);
d3374825 6940 mddev->hold_active = 0;
1da177e4
LT		 6941 }
6942}
6943
685784aa 6944subsys_initcall(md_init);
1da177e4
LT		 6945module_exit(md_exit)
6946
f91de92e
N		 6947static int get_ro(char *buffer, struct kernel_param *kp)
6948{
6949 return sprintf(buffer, "%d", start_readonly);
6950}
6951static int set_ro(const char *val, struct kernel_param *kp)
6952{
6953 char *e;
6954 int num = simple_strtoul(val, &e, 10);
6955 if (*val && (*e == '\0' || *e == '\n')) {
6956 start_readonly = num;
4dbcdc75 6957 return 0;
f91de92e
N		 6958 }
6959 return -EINVAL;
6960}
6961
80ca3a44
N		 6962module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6963module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 6964
efeb53c0 6965module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 6966
1da177e4
LT		 6967EXPORT_SYMBOL(register_md_personality);
6968EXPORT_SYMBOL(unregister_md_personality);
6969EXPORT_SYMBOL(md_error);
6970EXPORT_SYMBOL(md_done_sync);
6971EXPORT_SYMBOL(md_write_start);
6972EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 6973EXPORT_SYMBOL(md_register_thread);
6974EXPORT_SYMBOL(md_unregister_thread);
6975EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 6976EXPORT_SYMBOL(md_check_recovery);
6977MODULE_LICENSE("GPL");
aa1595e9 6978MODULE_ALIAS("md");
72008652 6979MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)