projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: change a few 'int' to 'size_t' in md
[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
35#include <linux/module.h>
50511da3 36#include <linux/kernel.h>
a6fb0934 37#include <linux/kthread.h>
1da177e4
LT		 38#include <linux/linkage.h>
39#include <linux/raid/md.h>
32a7627c 40#include <linux/raid/bitmap.h>
1da177e4 41#include <linux/sysctl.h>
1da177e4 42#include <linux/buffer_head.h> /* for invalidate_bdev */
d7603b7e 43#include <linux/poll.h>
48c9c27b 44#include <linux/mutex.h>
16f17b39 45#include <linux/ctype.h>
7dfb7103 46#include <linux/freezer.h>
1da177e4
LT		 47
48#include <linux/init.h>
49
32a7627c
N		 50#include <linux/file.h>
51
1da177e4
LT		 52#ifdef CONFIG_KMOD
53#include <linux/kmod.h>
54#endif
55
56#include <asm/unaligned.h>
57
58#define MAJOR_NR MD_MAJOR
59#define MD_DRIVER
60
61/* 63 partitions with the alternate major number (mdp) */
62#define MdpMinorShift 6
63
64#define DEBUG 0
65#define dprintk(x...) ((void)(DEBUG && printk(x)))
66
67
68#ifndef MODULE
69static void autostart_arrays (int part);
70#endif
71
2604b703 72static LIST_HEAD(pers_list);
1da177e4
LT		 73static DEFINE_SPINLOCK(pers_lock);
74
5e56341d
AB		 75static void md_print_devices(void);
76
77#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78
1da177e4
LT		 79/*
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 83 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
86 * idle IO detection.
87 *
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 89 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 90 */
91
92static int sysctl_speed_limit_min = 1000;
93static int sysctl_speed_limit_max = 200000;
88202a0c
N		 94static inline int speed_min(mddev_t *mddev)
95{
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
98}
99
100static inline int speed_max(mddev_t *mddev)
101{
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
104}
1da177e4
LT		 105
106static struct ctl_table_header *raid_table_header;
107
108static ctl_table raid_table[] = {
109 {
110 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
80ca3a44 114 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 115 .proc_handler = &proc_dointvec,
116 },
117 {
118 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
80ca3a44 122 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 123 .proc_handler = &proc_dointvec,
124 },
125 { .ctl_name = 0 }
126};
127
128static ctl_table raid_dir_table[] = {
129 {
130 .ctl_name = DEV_RAID,
131 .procname = "raid",
132 .maxlen = 0,
80ca3a44 133 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 134 .child = raid_table,
135 },
136 { .ctl_name = 0 }
137};
138
139static ctl_table raid_root_table[] = {
140 {
141 .ctl_name = CTL_DEV,
142 .procname = "dev",
143 .maxlen = 0,
144 .mode = 0555,
145 .child = raid_dir_table,
146 },
147 { .ctl_name = 0 }
148};
149
150static struct block_device_operations md_fops;
151
f91de92e
N		 152static int start_readonly;
153
d7603b7e
N		 154/*
155 * We have a system wide 'event count' that is incremented
156 * on any 'interesting' event, and readers of /proc/mdstat
157 * can use 'poll' or 'select' to find out when the event
158 * count increases.
159 *
160 * Events are:
161 * start array, stop array, error, add device, remove device,
162 * start build, activate spare
163 */
2989ddbd 164static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 165static atomic_t md_event_count;
29269553 166void md_new_event(mddev_t *mddev)
d7603b7e
N		 167{
168 atomic_inc(&md_event_count);
169 wake_up(&md_event_waiters);
4508a7a7 170 sysfs_notify(&mddev->kobj, NULL, "sync_action");
d7603b7e 171}
29269553 172EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 173
c331eb04
N		 174/* Alternate version that can be called from interrupts
175 * when calling sysfs_notify isn't needed.
176 */
05381954 177static void md_new_event_inintr(mddev_t *mddev)
c331eb04
N		 178{
179 atomic_inc(&md_event_count);
180 wake_up(&md_event_waiters);
181}
182
1da177e4
LT		 183/*
184 * Enables to iterate over all existing md arrays
185 * all_mddevs_lock protects this list.
186 */
187static LIST_HEAD(all_mddevs);
188static DEFINE_SPINLOCK(all_mddevs_lock);
189
190
191/*
192 * iterates through all used mddevs in the system.
193 * We take care to grab the all_mddevs_lock whenever navigating
194 * the list, and to always hold a refcount when unlocked.
195 * Any code which breaks out of this loop while own
196 * a reference to the current mddev and must mddev_put it.
197 */
198#define ITERATE_MDDEV(mddev,tmp) \
199 \
200 for (({ spin_lock(&all_mddevs_lock); \
201 tmp = all_mddevs.next; \
202 mddev = NULL;}); \
203 ({ if (tmp != &all_mddevs) \
204 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205 spin_unlock(&all_mddevs_lock); \
206 if (mddev) mddev_put(mddev); \
207 mddev = list_entry(tmp, mddev_t, all_mddevs); \
208 tmp != &all_mddevs;}); \
209 ({ spin_lock(&all_mddevs_lock); \
210 tmp = tmp->next;}) \
211 )
212
213
165125e1 214static int md_fail_request (struct request_queue *q, struct bio *bio)
1da177e4 215{
6712ecf8 216 bio_io_error(bio);
1da177e4
LT		 217 return 0;
218}
219
220static inline mddev_t *mddev_get(mddev_t *mddev)
221{
222 atomic_inc(&mddev->active);
223 return mddev;
224}
225
226static void mddev_put(mddev_t *mddev)
227{
228 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
229 return;
230 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
231 list_del(&mddev->all_mddevs);
926ce2d8 232 spin_unlock(&all_mddevs_lock);
1312f40e 233 blk_cleanup_queue(mddev->queue);
c10997f6 234 kobject_put(&mddev->kobj);
926ce2d8
N		 235 } else
236 spin_unlock(&all_mddevs_lock);
1da177e4
LT		 237}
238
239static mddev_t * mddev_find(dev_t unit)
240{
241 mddev_t *mddev, *new = NULL;
242
243 retry:
244 spin_lock(&all_mddevs_lock);
245 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
246 if (mddev->unit == unit) {
247 mddev_get(mddev);
248 spin_unlock(&all_mddevs_lock);
990a8baf 249 kfree(new);
1da177e4
LT		 250 return mddev;
251 }
252
253 if (new) {
254 list_add(&new->all_mddevs, &all_mddevs);
255 spin_unlock(&all_mddevs_lock);
256 return new;
257 }
258 spin_unlock(&all_mddevs_lock);
259
9ffae0cf 260 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 261 if (!new)
262 return NULL;
263
1da177e4
LT		 264 new->unit = unit;
265 if (MAJOR(unit) == MD_MAJOR)
266 new->md_minor = MINOR(unit);
267 else
268 new->md_minor = MINOR(unit) >> MdpMinorShift;
269
df5b89b3 270 mutex_init(&new->reconfig_mutex);
1da177e4
LT		 271 INIT_LIST_HEAD(&new->disks);
272 INIT_LIST_HEAD(&new->all_mddevs);
273 init_timer(&new->safemode_timer);
274 atomic_set(&new->active, 1);
06d91a5f 275 spin_lock_init(&new->write_lock);
3d310eb7 276 init_waitqueue_head(&new->sb_wait);
08a02ecd 277 new->reshape_position = MaxSector;
c6207277 278 new->resync_max = MaxSector;
1da177e4
LT		 279
280 new->queue = blk_alloc_queue(GFP_KERNEL);
281 if (!new->queue) {
282 kfree(new);
283 return NULL;
284 }
89e5c8b5 285 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
1da177e4
LT		 286
287 blk_queue_make_request(new->queue, md_fail_request);
288
289 goto retry;
290}
291
292static inline int mddev_lock(mddev_t * mddev)
293{
df5b89b3 294 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 295}
296
1da177e4
LT		 297static inline int mddev_trylock(mddev_t * mddev)
298{
df5b89b3 299 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 300}
301
302static inline void mddev_unlock(mddev_t * mddev)
303{
df5b89b3 304 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 305
005eca5e 306 md_wakeup_thread(mddev->thread);
1da177e4
LT		 307}
308
2989ddbd 309static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4
LT		 310{
311 mdk_rdev_t * rdev;
312 struct list_head *tmp;
313
314 ITERATE_RDEV(mddev,rdev,tmp) {
315 if (rdev->desc_nr == nr)
316 return rdev;
317 }
318 return NULL;
319}
320
321static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
322{
323 struct list_head *tmp;
324 mdk_rdev_t *rdev;
325
326 ITERATE_RDEV(mddev,rdev,tmp) {
327 if (rdev->bdev->bd_dev == dev)
328 return rdev;
329 }
330 return NULL;
331}
332
d9d166c2 333static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 334{
335 struct mdk_personality *pers;
d9d166c2
N		 336 list_for_each_entry(pers, &pers_list, list) {
337 if (level != LEVEL_NONE && pers->level == level)
2604b703 338 return pers;
d9d166c2
N		 339 if (strcmp(pers->name, clevel)==0)
340 return pers;
341 }
2604b703
N		 342 return NULL;
343}
344
77933d72 345static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4
LT		 346{
347 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
348 return MD_NEW_SIZE_BLOCKS(size);
349}
350
351static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
352{
353 sector_t size;
354
355 size = rdev->sb_offset;
356
357 if (chunk_size)
358 size &= ~((sector_t)chunk_size/1024 - 1);
359 return size;
360}
361
362static int alloc_disk_sb(mdk_rdev_t * rdev)
363{
364 if (rdev->sb_page)
365 MD_BUG();
366
367 rdev->sb_page = alloc_page(GFP_KERNEL);
368 if (!rdev->sb_page) {
369 printk(KERN_ALERT "md: out of memory.\n");
370 return -EINVAL;
371 }
372
373 return 0;
374}
375
376static void free_disk_sb(mdk_rdev_t * rdev)
377{
378 if (rdev->sb_page) {
2d1f3b5d 379 put_page(rdev->sb_page);
1da177e4
LT		 380 rdev->sb_loaded = 0;
381 rdev->sb_page = NULL;
382 rdev->sb_offset = 0;
383 rdev->size = 0;
384 }
385}
386
387
6712ecf8 388static void super_written(struct bio *bio, int error)
7bfa19f2
N		 389{
390 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 391 mddev_t *mddev = rdev->mddev;
7bfa19f2 392
3a0f5bbb
N		 393 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
394 printk("md: super_written gets error=%d, uptodate=%d\n",
395 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
396 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 397 md_error(mddev, rdev);
3a0f5bbb 398 }
7bfa19f2 399
a9701a30
N		 400 if (atomic_dec_and_test(&mddev->pending_writes))
401 wake_up(&mddev->sb_wait);
f8b58edf 402 bio_put(bio);
7bfa19f2
N		 403}
404
6712ecf8 405static void super_written_barrier(struct bio *bio, int error)
a9701a30
N		 406{
407 struct bio *bio2 = bio->bi_private;
408 mdk_rdev_t *rdev = bio2->bi_private;
409 mddev_t *mddev = rdev->mddev;
a9701a30
N		 410
411 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
412 error == -EOPNOTSUPP) {
413 unsigned long flags;
414 /* barriers don't appear to be supported :-( */
415 set_bit(BarriersNotsupp, &rdev->flags);
416 mddev->barriers_work = 0;
417 spin_lock_irqsave(&mddev->write_lock, flags);
418 bio2->bi_next = mddev->biolist;
419 mddev->biolist = bio2;
420 spin_unlock_irqrestore(&mddev->write_lock, flags);
421 wake_up(&mddev->sb_wait);
422 bio_put(bio);
6712ecf8
N		 423 } else {
424 bio_put(bio2);
425 bio->bi_private = rdev;
426 super_written(bio, error);
a9701a30 427 }
a9701a30
N		 428}
429
7bfa19f2
N		 430void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
431 sector_t sector, int size, struct page *page)
432{
433 /* write first size bytes of page to sector of rdev
434 * Increment mddev->pending_writes before returning
435 * and decrement it on completion, waking up sb_wait
436 * if zero is reached.
437 * If an error occurred, call md_error
a9701a30
N		 438 *
439 * As we might need to resubmit the request if BIO_RW_BARRIER
440 * causes ENOTSUPP, we allocate a spare bio...
7bfa19f2
N		 441 */
442 struct bio *bio = bio_alloc(GFP_NOIO, 1);
a9701a30 443 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
7bfa19f2
N		 444
445 bio->bi_bdev = rdev->bdev;
446 bio->bi_sector = sector;
447 bio_add_page(bio, page, size, 0);
448 bio->bi_private = rdev;
449 bio->bi_end_io = super_written;
a9701a30
N		 450 bio->bi_rw = rw;
451
7bfa19f2 452 atomic_inc(&mddev->pending_writes);
a9701a30
N		 453 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
454 struct bio *rbio;
455 rw |= (1<<BIO_RW_BARRIER);
456 rbio = bio_clone(bio, GFP_NOIO);
457 rbio->bi_private = bio;
458 rbio->bi_end_io = super_written_barrier;
459 submit_bio(rw, rbio);
460 } else
461 submit_bio(rw, bio);
462}
463
464void md_super_wait(mddev_t *mddev)
465{
466 /* wait for all superblock writes that were scheduled to complete.
467 * if any had to be retried (due to BARRIER problems), retry them
468 */
469 DEFINE_WAIT(wq);
470 for(;;) {
471 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
472 if (atomic_read(&mddev->pending_writes)==0)
473 break;
474 while (mddev->biolist) {
475 struct bio *bio;
476 spin_lock_irq(&mddev->write_lock);
477 bio = mddev->biolist;
478 mddev->biolist = bio->bi_next ;
479 bio->bi_next = NULL;
480 spin_unlock_irq(&mddev->write_lock);
481 submit_bio(bio->bi_rw, bio);
482 }
483 schedule();
484 }
485 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 486}
487
6712ecf8 488static void bi_complete(struct bio *bio, int error)
1da177e4 489{
1da177e4 490 complete((struct completion*)bio->bi_private);
1da177e4
LT		 491}
492
a654b9d8 493int sync_page_io(struct block_device *bdev, sector_t sector, int size,
1da177e4
LT		 494 struct page *page, int rw)
495{
baaa2c51 496 struct bio *bio = bio_alloc(GFP_NOIO, 1);
1da177e4
LT		 497 struct completion event;
498 int ret;
499
500 rw |= (1 << BIO_RW_SYNC);
501
502 bio->bi_bdev = bdev;
503 bio->bi_sector = sector;
504 bio_add_page(bio, page, size, 0);
505 init_completion(&event);
506 bio->bi_private = &event;
507 bio->bi_end_io = bi_complete;
508 submit_bio(rw, bio);
509 wait_for_completion(&event);
510
511 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
512 bio_put(bio);
513 return ret;
514}
a8745db2 515EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 516
0002b271 517static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 518{
519 char b[BDEVNAME_SIZE];
520 if (!rdev->sb_page) {
521 MD_BUG();
522 return -EINVAL;
523 }
524 if (rdev->sb_loaded)
525 return 0;
526
527
0002b271 528 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
1da177e4
LT		 529 goto fail;
530 rdev->sb_loaded = 1;
531 return 0;
532
533fail:
534 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
535 bdevname(rdev->bdev,b));
536 return -EINVAL;
537}
538
539static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
540{
541 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
542 (sb1->set_uuid1 == sb2->set_uuid1) &&
543 (sb1->set_uuid2 == sb2->set_uuid2) &&
544 (sb1->set_uuid3 == sb2->set_uuid3))
545
546 return 1;
547
548 return 0;
549}
550
551
552static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
553{
554 int ret;
555 mdp_super_t *tmp1, *tmp2;
556
557 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
558 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
559
560 if (!tmp1 || !tmp2) {
561 ret = 0;
562 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
563 goto abort;
564 }
565
566 *tmp1 = *sb1;
567 *tmp2 = *sb2;
568
569 /*
570 * nr_disks is not constant
571 */
572 tmp1->nr_disks = 0;
573 tmp2->nr_disks = 0;
574
575 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
576 ret = 0;
577 else
578 ret = 1;
579
580abort:
990a8baf
JJ		 581 kfree(tmp1);
582 kfree(tmp2);
1da177e4
LT		 583 return ret;
584}
585
4d167f09
N		 586
587static u32 md_csum_fold(u32 csum)
588{
589 csum = (csum & 0xffff) + (csum >> 16);
590 return (csum & 0xffff) + (csum >> 16);
591}
592
1da177e4
LT		 593static unsigned int calc_sb_csum(mdp_super_t * sb)
594{
4d167f09
N		 595 u64 newcsum = 0;
596 u32 *sb32 = (u32*)sb;
597 int i;
1da177e4
LT		 598 unsigned int disk_csum, csum;
599
600 disk_csum = sb->sb_csum;
601 sb->sb_csum = 0;
4d167f09
N		 602
603 for (i = 0; i < MD_SB_BYTES/4 ; i++)
604 newcsum += sb32[i];
605 csum = (newcsum & 0xffffffff) + (newcsum>>32);
606
607
608#ifdef CONFIG_ALPHA
609 /* This used to use csum_partial, which was wrong for several
610 * reasons including that different results are returned on
611 * different architectures. It isn't critical that we get exactly
612 * the same return value as before (we always csum_fold before
613 * testing, and that removes any differences). However as we
614 * know that csum_partial always returned a 16bit value on
615 * alphas, do a fold to maximise conformity to previous behaviour.
616 */
617 sb->sb_csum = md_csum_fold(disk_csum);
618#else
1da177e4 619 sb->sb_csum = disk_csum;
4d167f09 620#endif
1da177e4
LT		 621 return csum;
622}
623
624
625/*
626 * Handle superblock details.
627 * We want to be able to handle multiple superblock formats
628 * so we have a common interface to them all, and an array of
629 * different handlers.
630 * We rely on user-space to write the initial superblock, and support
631 * reading and updating of superblocks.
632 * Interface methods are:
633 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
634 * loads and validates a superblock on dev.
635 * if refdev != NULL, compare superblocks on both devices
636 * Return:
637 * 0 - dev has a superblock that is compatible with refdev
638 * 1 - dev has a superblock that is compatible and newer than refdev
639 * so dev should be used as the refdev in future
640 * -EINVAL superblock incompatible or invalid
641 * -othererror e.g. -EIO
642 *
643 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
644 * Verify that dev is acceptable into mddev.
645 * The first time, mddev->raid_disks will be 0, and data from
646 * dev should be merged in. Subsequent calls check that dev
647 * is new enough. Return 0 or -EINVAL
648 *
649 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
650 * Update the superblock for rdev with data in mddev
651 * This does not write to disc.
652 *
653 */
654
655struct super_type {
656 char *name;
657 struct module *owner;
658 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
659 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
660 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
661};
662
663/*
664 * load_super for 0.90.0
665 */
666static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
667{
668 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
669 mdp_super_t *sb;
670 int ret;
671 sector_t sb_offset;
672
673 /*
674 * Calculate the position of the superblock,
675 * it's at the end of the disk.
676 *
677 * It also happens to be a multiple of 4Kb.
678 */
679 sb_offset = calc_dev_sboffset(rdev->bdev);
680 rdev->sb_offset = sb_offset;
681
0002b271 682 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 683 if (ret) return ret;
684
685 ret = -EINVAL;
686
687 bdevname(rdev->bdev, b);
688 sb = (mdp_super_t*)page_address(rdev->sb_page);
689
690 if (sb->md_magic != MD_SB_MAGIC) {
691 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
692 b);
693 goto abort;
694 }
695
696 if (sb->major_version != 0 ||
f6705578
N		 697 sb->minor_version < 90 ||
698 sb->minor_version > 91) {
1da177e4
LT		 699 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
700 sb->major_version, sb->minor_version,
701 b);
702 goto abort;
703 }
704
705 if (sb->raid_disks <= 0)
706 goto abort;
707
4d167f09 708 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 709 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
710 b);
711 goto abort;
712 }
713
714 rdev->preferred_minor = sb->md_minor;
715 rdev->data_offset = 0;
0002b271 716 rdev->sb_size = MD_SB_BYTES;
1da177e4 717
e11e93fa
N		 718 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
719 if (sb->level != 1 && sb->level != 4
720 && sb->level != 5 && sb->level != 6
721 && sb->level != 10) {
722 /* FIXME use a better test */
723 printk(KERN_WARNING
724 "md: bitmaps not supported for this level.\n");
725 goto abort;
726 }
727 }
728
1da177e4
LT		 729 if (sb->level == LEVEL_MULTIPATH)
730 rdev->desc_nr = -1;
731 else
732 rdev->desc_nr = sb->this_disk.number;
733
734 if (refdev == 0)
735 ret = 1;
736 else {
737 __u64 ev1, ev2;
738 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
739 if (!uuid_equal(refsb, sb)) {
740 printk(KERN_WARNING "md: %s has different UUID to %s\n",
741 b, bdevname(refdev->bdev,b2));
742 goto abort;
743 }
744 if (!sb_equal(refsb, sb)) {
745 printk(KERN_WARNING "md: %s has same UUID"
746 " but different superblock to %s\n",
747 b, bdevname(refdev->bdev, b2));
748 goto abort;
749 }
750 ev1 = md_event(sb);
751 ev2 = md_event(refsb);
752 if (ev1 > ev2)
753 ret = 1;
754 else
755 ret = 0;
756 }
757 rdev->size = calc_dev_size(rdev, sb->chunk_size);
758
2bf071bf
N		 759 if (rdev->size < sb->size && sb->level > 1)
760 /* "this cannot possibly happen" ... */
761 ret = -EINVAL;
762
1da177e4
LT		 763 abort:
764 return ret;
765}
766
767/*
768 * validate_super for 0.90.0
769 */
770static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
771{
772 mdp_disk_t *desc;
773 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
07d84d10 774 __u64 ev1 = md_event(sb);
1da177e4 775
41158c7e 776 rdev->raid_disk = -1;
c5d79adb
N		 777 clear_bit(Faulty, &rdev->flags);
778 clear_bit(In_sync, &rdev->flags);
779 clear_bit(WriteMostly, &rdev->flags);
780 clear_bit(BarriersNotsupp, &rdev->flags);
781
1da177e4
LT		 782 if (mddev->raid_disks == 0) {
783 mddev->major_version = 0;
784 mddev->minor_version = sb->minor_version;
785 mddev->patch_version = sb->patch_version;
e691063a 786 mddev->external = 0;
1da177e4
LT		 787 mddev->chunk_size = sb->chunk_size;
788 mddev->ctime = sb->ctime;
789 mddev->utime = sb->utime;
790 mddev->level = sb->level;
d9d166c2 791 mddev->clevel[0] = 0;
1da177e4
LT		 792 mddev->layout = sb->layout;
793 mddev->raid_disks = sb->raid_disks;
794 mddev->size = sb->size;
07d84d10 795 mddev->events = ev1;
9223214e 796 mddev->bitmap_offset = 0;
36fa3063 797 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
1da177e4 798
f6705578
N		 799 if (mddev->minor_version >= 91) {
800 mddev->reshape_position = sb->reshape_position;
801 mddev->delta_disks = sb->delta_disks;
802 mddev->new_level = sb->new_level;
803 mddev->new_layout = sb->new_layout;
804 mddev->new_chunk = sb->new_chunk;
805 } else {
806 mddev->reshape_position = MaxSector;
807 mddev->delta_disks = 0;
808 mddev->new_level = mddev->level;
809 mddev->new_layout = mddev->layout;
810 mddev->new_chunk = mddev->chunk_size;
811 }
812
1da177e4
LT		 813 if (sb->state & (1<<MD_SB_CLEAN))
814 mddev->recovery_cp = MaxSector;
815 else {
816 if (sb->events_hi == sb->cp_events_hi &&
817 sb->events_lo == sb->cp_events_lo) {
818 mddev->recovery_cp = sb->recovery_cp;
819 } else
820 mddev->recovery_cp = 0;
821 }
822
823 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
824 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
825 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
826 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
827
828 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 829
830 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
e11e93fa 831 mddev->bitmap_file == NULL)
36fa3063 832 mddev->bitmap_offset = mddev->default_bitmap_offset;
a654b9d8 833
41158c7e
N		 834 } else if (mddev->pers == NULL) {
835 /* Insist on good event counter while assembling */
1da177e4
LT		 836 ++ev1;
837 if (ev1 < mddev->events)
838 return -EINVAL;
41158c7e
N		 839 } else if (mddev->bitmap) {
840 /* if adding to array with a bitmap, then we can accept an
841 * older device ... but not too old.
842 */
41158c7e
N		 843 if (ev1 < mddev->bitmap->events_cleared)
844 return 0;
07d84d10
N		 845 } else {
846 if (ev1 < mddev->events)
847 /* just a hot-add of a new device, leave raid_disk at -1 */
848 return 0;
849 }
41158c7e 850
1da177e4 851 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 852 desc = sb->disks + rdev->desc_nr;
853
854 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 855 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 856 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
857 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 858 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 859 rdev->raid_disk = desc->raid_disk;
860 }
8ddf9efe
N		 861 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
862 set_bit(WriteMostly, &rdev->flags);
41158c7e 863 } else /* MULTIPATH are always insync */
b2d444d7 864 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 865 return 0;
866}
867
868/*
869 * sync_super for 0.90.0
870 */
871static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
872{
873 mdp_super_t *sb;
874 struct list_head *tmp;
875 mdk_rdev_t *rdev2;
876 int next_spare = mddev->raid_disks;
19133a42 877
1da177e4
LT		 878
879 /* make rdev->sb match mddev data..
880 *
881 * 1/ zero out disks
882 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
883 * 3/ any empty disks < next_spare become removed
884 *
885 * disks[0] gets initialised to REMOVED because
886 * we cannot be sure from other fields if it has
887 * been initialised or not.
888 */
889 int i;
890 int active=0, working=0,failed=0,spare=0,nr_disks=0;
891
61181565
N		 892 rdev->sb_size = MD_SB_BYTES;
893
1da177e4
LT		 894 sb = (mdp_super_t*)page_address(rdev->sb_page);
895
896 memset(sb, 0, sizeof(*sb));
897
898 sb->md_magic = MD_SB_MAGIC;
899 sb->major_version = mddev->major_version;
1da177e4
LT		 900 sb->patch_version = mddev->patch_version;
901 sb->gvalid_words = 0; /* ignored */
902 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
903 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
904 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
905 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
906
907 sb->ctime = mddev->ctime;
908 sb->level = mddev->level;
909 sb->size = mddev->size;
910 sb->raid_disks = mddev->raid_disks;
911 sb->md_minor = mddev->md_minor;
e691063a 912 sb->not_persistent = 0;
1da177e4
LT		 913 sb->utime = mddev->utime;
914 sb->state = 0;
915 sb->events_hi = (mddev->events>>32);
916 sb->events_lo = (u32)mddev->events;
917
f6705578
N		 918 if (mddev->reshape_position == MaxSector)
919 sb->minor_version = 90;
920 else {
921 sb->minor_version = 91;
922 sb->reshape_position = mddev->reshape_position;
923 sb->new_level = mddev->new_level;
924 sb->delta_disks = mddev->delta_disks;
925 sb->new_layout = mddev->new_layout;
926 sb->new_chunk = mddev->new_chunk;
927 }
928 mddev->minor_version = sb->minor_version;
1da177e4
LT		 929 if (mddev->in_sync)
930 {
931 sb->recovery_cp = mddev->recovery_cp;
932 sb->cp_events_hi = (mddev->events>>32);
933 sb->cp_events_lo = (u32)mddev->events;
934 if (mddev->recovery_cp == MaxSector)
935 sb->state = (1<< MD_SB_CLEAN);
936 } else
937 sb->recovery_cp = 0;
938
939 sb->layout = mddev->layout;
940 sb->chunk_size = mddev->chunk_size;
941
a654b9d8
N		 942 if (mddev->bitmap && mddev->bitmap_file == NULL)
943 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
944
1da177e4
LT		 945 sb->disks[0].state = (1<<MD_DISK_REMOVED);
946 ITERATE_RDEV(mddev,rdev2,tmp) {
947 mdp_disk_t *d;
86e6ffdd 948 int desc_nr;
b2d444d7
N		 949 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
950 && !test_bit(Faulty, &rdev2->flags))
86e6ffdd 951 desc_nr = rdev2->raid_disk;
1da177e4 952 else
86e6ffdd 953 desc_nr = next_spare++;
19133a42 954 rdev2->desc_nr = desc_nr;
1da177e4
LT		 955 d = &sb->disks[rdev2->desc_nr];
956 nr_disks++;
957 d->number = rdev2->desc_nr;
958 d->major = MAJOR(rdev2->bdev->bd_dev);
959 d->minor = MINOR(rdev2->bdev->bd_dev);
b2d444d7
N		 960 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
961 && !test_bit(Faulty, &rdev2->flags))
1da177e4
LT		 962 d->raid_disk = rdev2->raid_disk;
963 else
964 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 965 if (test_bit(Faulty, &rdev2->flags))
1da177e4 966 d->state = (1<<MD_DISK_FAULTY);
1be7892f 967 else if (test_bit(In_sync, &rdev2->flags)) {
1da177e4
LT		 968 d->state = (1<<MD_DISK_ACTIVE);
969 d->state |= (1<<MD_DISK_SYNC);
970 active++;
971 working++;
972 } else {
973 d->state = 0;
974 spare++;
975 working++;
976 }
8ddf9efe
N		 977 if (test_bit(WriteMostly, &rdev2->flags))
978 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 979 }
1da177e4
LT		 980 /* now set the "removed" and "faulty" bits on any missing devices */
981 for (i=0 ; i < mddev->raid_disks ; i++) {
982 mdp_disk_t *d = &sb->disks[i];
983 if (d->state == 0 && d->number == 0) {
984 d->number = i;
985 d->raid_disk = i;
986 d->state = (1<<MD_DISK_REMOVED);
987 d->state |= (1<<MD_DISK_FAULTY);
988 failed++;
989 }
990 }
991 sb->nr_disks = nr_disks;
992 sb->active_disks = active;
993 sb->working_disks = working;
994 sb->failed_disks = failed;
995 sb->spare_disks = spare;
996
997 sb->this_disk = sb->disks[rdev->desc_nr];
998 sb->sb_csum = calc_sb_csum(sb);
999}
1000
1001/*
1002 * version 1 superblock
1003 */
1004
1c05b4bc 1005static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1006{
1c05b4bc
N		 1007 __le32 disk_csum;
1008 u32 csum;
1da177e4
LT		 1009 unsigned long long newcsum;
1010 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1011 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1012 int i;
1013
1014 disk_csum = sb->sb_csum;
1015 sb->sb_csum = 0;
1016 newcsum = 0;
1017 for (i=0; size>=4; size -= 4 )
1018 newcsum += le32_to_cpu(*isuper++);
1019
1020 if (size == 2)
1c05b4bc 1021 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1022
1023 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1024 sb->sb_csum = disk_csum;
1025 return cpu_to_le32(csum);
1026}
1027
1028static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1029{
1030 struct mdp_superblock_1 *sb;
1031 int ret;
1032 sector_t sb_offset;
1033 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1034 int bmask;
1da177e4
LT		 1035
1036 /*
1037 * Calculate the position of the superblock.
1038 * It is always aligned to a 4K boundary and
1039 * depeding on minor_version, it can be:
1040 * 0: At least 8K, but less than 12K, from end of device
1041 * 1: At start of device
1042 * 2: 4K from start of device.
1043 */
1044 switch(minor_version) {
1045 case 0:
1046 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1047 sb_offset -= 8*2;
39730960 1048 sb_offset &= ~(sector_t)(4*2-1);
1da177e4
LT		 1049 /* convert from sectors to K */
1050 sb_offset /= 2;
1051 break;
1052 case 1:
1053 sb_offset = 0;
1054 break;
1055 case 2:
1056 sb_offset = 4;
1057 break;
1058 default:
1059 return -EINVAL;
1060 }
1061 rdev->sb_offset = sb_offset;
1062
0002b271
N		 1063 /* superblock is rarely larger than 1K, but it can be larger,
1064 * and it is safe to read 4k, so we do that
1065 */
1066 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1067 if (ret) return ret;
1068
1069
1070 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1071
1072 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1073 sb->major_version != cpu_to_le32(1) ||
1074 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1075 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
71c0805c 1076 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1077 return -EINVAL;
1078
1079 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1080 printk("md: invalid superblock checksum on %s\n",
1081 bdevname(rdev->bdev,b));
1082 return -EINVAL;
1083 }
1084 if (le64_to_cpu(sb->data_size) < 10) {
1085 printk("md: data_size too small on %s\n",
1086 bdevname(rdev->bdev,b));
1087 return -EINVAL;
1088 }
e11e93fa
N		 1089 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1090 if (sb->level != cpu_to_le32(1) &&
1091 sb->level != cpu_to_le32(4) &&
1092 sb->level != cpu_to_le32(5) &&
1093 sb->level != cpu_to_le32(6) &&
1094 sb->level != cpu_to_le32(10)) {
1095 printk(KERN_WARNING
1096 "md: bitmaps not supported for this level.\n");
1097 return -EINVAL;
1098 }
1099 }
1100
1da177e4
LT		 1101 rdev->preferred_minor = 0xffff;
1102 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1103 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1104
0002b271 1105 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
720a3dc3 1106 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
0002b271
N		 1107 if (rdev->sb_size & bmask)
1108 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1109
31b65a0d
N		 1110 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1111 rdev->desc_nr = -1;
1112 else
1113 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1114
1da177e4 1115 if (refdev == 0)
8ed75463 1116 ret = 1;
1da177e4
LT		 1117 else {
1118 __u64 ev1, ev2;
1119 struct mdp_superblock_1 *refsb =
1120 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1121
1122 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1123 sb->level != refsb->level ||
1124 sb->layout != refsb->layout ||
1125 sb->chunksize != refsb->chunksize) {
1126 printk(KERN_WARNING "md: %s has strangely different"
1127 " superblock to %s\n",
1128 bdevname(rdev->bdev,b),
1129 bdevname(refdev->bdev,b2));
1130 return -EINVAL;
1131 }
1132 ev1 = le64_to_cpu(sb->events);
1133 ev2 = le64_to_cpu(refsb->events);
1134
1135 if (ev1 > ev2)
8ed75463
N		 1136 ret = 1;
1137 else
1138 ret = 0;
1da177e4
LT		 1139 }
1140 if (minor_version)
1141 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1142 else
1143 rdev->size = rdev->sb_offset;
1144 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1145 return -EINVAL;
1146 rdev->size = le64_to_cpu(sb->data_size)/2;
1147 if (le32_to_cpu(sb->chunksize))
1148 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
2bf071bf 1149
1c05b4bc 1150 if (le64_to_cpu(sb->size) > rdev->size*2)
2bf071bf 1151 return -EINVAL;
8ed75463 1152 return ret;
1da177e4
LT		 1153}
1154
1155static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1156{
1157 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
07d84d10 1158 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1159
41158c7e 1160 rdev->raid_disk = -1;
c5d79adb
N		 1161 clear_bit(Faulty, &rdev->flags);
1162 clear_bit(In_sync, &rdev->flags);
1163 clear_bit(WriteMostly, &rdev->flags);
1164 clear_bit(BarriersNotsupp, &rdev->flags);
1165
1da177e4
LT		 1166 if (mddev->raid_disks == 0) {
1167 mddev->major_version = 1;
1168 mddev->patch_version = 0;
e691063a 1169 mddev->external = 0;
1da177e4
LT		 1170 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1171 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1172 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1173 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1174 mddev->clevel[0] = 0;
1da177e4
LT		 1175 mddev->layout = le32_to_cpu(sb->layout);
1176 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1177 mddev->size = le64_to_cpu(sb->size)/2;
07d84d10 1178 mddev->events = ev1;
9223214e 1179 mddev->bitmap_offset = 0;
29fc7e3e 1180 mddev->default_bitmap_offset = 1024 >> 9;
1da177e4
LT		 1181
1182 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1183 memcpy(mddev->uuid, sb->set_uuid, 16);
1184
1185 mddev->max_disks = (4096-256)/2;
a654b9d8 1186
71c0805c 1187 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
e11e93fa 1188 mddev->bitmap_file == NULL )
a654b9d8 1189 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1190
f6705578
N		 1191 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1192 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1193 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1194 mddev->new_level = le32_to_cpu(sb->new_level);
1195 mddev->new_layout = le32_to_cpu(sb->new_layout);
1196 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1197 } else {
1198 mddev->reshape_position = MaxSector;
1199 mddev->delta_disks = 0;
1200 mddev->new_level = mddev->level;
1201 mddev->new_layout = mddev->layout;
1202 mddev->new_chunk = mddev->chunk_size;
1203 }
1204
41158c7e
N		 1205 } else if (mddev->pers == NULL) {
1206 /* Insist of good event counter while assembling */
1da177e4
LT		 1207 ++ev1;
1208 if (ev1 < mddev->events)
1209 return -EINVAL;
41158c7e
N		 1210 } else if (mddev->bitmap) {
1211 /* If adding to array with a bitmap, then we can accept an
1212 * older device, but not too old.
1213 */
41158c7e
N		 1214 if (ev1 < mddev->bitmap->events_cleared)
1215 return 0;
07d84d10
N		 1216 } else {
1217 if (ev1 < mddev->events)
1218 /* just a hot-add of a new device, leave raid_disk at -1 */
1219 return 0;
1220 }
1da177e4
LT		 1221 if (mddev->level != LEVEL_MULTIPATH) {
1222 int role;
1da177e4
LT		 1223 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1224 switch(role) {
1225 case 0xffff: /* spare */
1da177e4
LT		 1226 break;
1227 case 0xfffe: /* faulty */
b2d444d7 1228 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1229 break;
1230 default:
5fd6c1dc
N		 1231 if ((le32_to_cpu(sb->feature_map) &
1232 MD_FEATURE_RECOVERY_OFFSET))
1233 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1234 else
1235 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1236 rdev->raid_disk = role;
1237 break;
1238 }
8ddf9efe
N		 1239 if (sb->devflags & WriteMostly1)
1240 set_bit(WriteMostly, &rdev->flags);
41158c7e 1241 } else /* MULTIPATH are always insync */
b2d444d7 1242 set_bit(In_sync, &rdev->flags);
41158c7e 1243
1da177e4
LT		 1244 return 0;
1245}
1246
1247static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1248{
1249 struct mdp_superblock_1 *sb;
1250 struct list_head *tmp;
1251 mdk_rdev_t *rdev2;
1252 int max_dev, i;
1253 /* make rdev->sb match mddev and rdev data. */
1254
1255 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1256
1257 sb->feature_map = 0;
1258 sb->pad0 = 0;
5fd6c1dc 1259 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1260 memset(sb->pad1, 0, sizeof(sb->pad1));
1261 memset(sb->pad2, 0, sizeof(sb->pad2));
1262 memset(sb->pad3, 0, sizeof(sb->pad3));
1263
1264 sb->utime = cpu_to_le64((__u64)mddev->utime);
1265 sb->events = cpu_to_le64(mddev->events);
1266 if (mddev->in_sync)
1267 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1268 else
1269 sb->resync_offset = cpu_to_le64(0);
1270
1c05b4bc 1271 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1272
f0ca340c 1273 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
29fc7e3e 1274 sb->size = cpu_to_le64(mddev->size<<1);
f0ca340c 1275
a654b9d8
N		 1276 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1277 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1278 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1279 }
5fd6c1dc
N		 1280
1281 if (rdev->raid_disk >= 0 &&
1282 !test_bit(In_sync, &rdev->flags) &&
1283 rdev->recovery_offset > 0) {
1284 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1285 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1286 }
1287
f6705578
N		 1288 if (mddev->reshape_position != MaxSector) {
1289 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1290 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1291 sb->new_layout = cpu_to_le32(mddev->new_layout);
1292 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1293 sb->new_level = cpu_to_le32(mddev->new_level);
1294 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1295 }
a654b9d8 1296
1da177e4
LT		 1297 max_dev = 0;
1298 ITERATE_RDEV(mddev,rdev2,tmp)
1299 if (rdev2->desc_nr+1 > max_dev)
1300 max_dev = rdev2->desc_nr+1;
a778b73f
N		 1301
1302 if (max_dev > le32_to_cpu(sb->max_dev))
1303 sb->max_dev = cpu_to_le32(max_dev);
1da177e4
LT		 1304 for (i=0; i<max_dev;i++)
1305 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1306
1307 ITERATE_RDEV(mddev,rdev2,tmp) {
1308 i = rdev2->desc_nr;
b2d444d7 1309 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1310 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1311 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1312 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
5fd6c1dc
N		 1313 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1314 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1315 else
1316 sb->dev_roles[i] = cpu_to_le16(0xffff);
1317 }
1318
1da177e4
LT		 1319 sb->sb_csum = calc_sb_1_csum(sb);
1320}
1321
1322
75c96f85 1323static struct super_type super_types[] = {
1da177e4
LT		 1324 [0] = {
1325 .name = "0.90.0",
1326 .owner = THIS_MODULE,
1327 .load_super = super_90_load,
1328 .validate_super = super_90_validate,
1329 .sync_super = super_90_sync,
1330 },
1331 [1] = {
1332 .name = "md-1",
1333 .owner = THIS_MODULE,
1334 .load_super = super_1_load,
1335 .validate_super = super_1_validate,
1336 .sync_super = super_1_sync,
1337 },
1338};
1da177e4
LT		 1339
1340static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1341{
7dd5e7c3
N		 1342 struct list_head *tmp, *tmp2;
1343 mdk_rdev_t *rdev, *rdev2;
1da177e4
LT		 1344
1345 ITERATE_RDEV(mddev1,rdev,tmp)
7dd5e7c3
N		 1346 ITERATE_RDEV(mddev2, rdev2, tmp2)
1347 if (rdev->bdev->bd_contains ==
1348 rdev2->bdev->bd_contains)
1349 return 1;
1da177e4
LT		 1350
1351 return 0;
1352}
1353
1354static LIST_HEAD(pending_raid_disks);
1355
1356static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1357{
7dd5e7c3 1358 char b[BDEVNAME_SIZE];
f637b9f9 1359 struct kobject *ko;
1edf80d3 1360 char *s;
5e55e2f5 1361 int err;
1da177e4
LT		 1362
1363 if (rdev->mddev) {
1364 MD_BUG();
1365 return -EINVAL;
1366 }
2bf071bf
N		 1367 /* make sure rdev->size exceeds mddev->size */
1368 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
a778b73f
N		 1369 if (mddev->pers) {
1370 /* Cannot change size, so fail
1371 * If mddev->level <= 0, then we don't care
1372 * about aligning sizes (e.g. linear)
1373 */
1374 if (mddev->level > 0)
1375 return -ENOSPC;
1376 } else
2bf071bf
N		 1377 mddev->size = rdev->size;
1378 }
1da177e4
LT		 1379
1380 /* Verify rdev->desc_nr is unique.
1381 * If it is -1, assign a free number, else
1382 * check number is not in use
1383 */
1384 if (rdev->desc_nr < 0) {
1385 int choice = 0;
1386 if (mddev->pers) choice = mddev->raid_disks;
1387 while (find_rdev_nr(mddev, choice))
1388 choice++;
1389 rdev->desc_nr = choice;
1390 } else {
1391 if (find_rdev_nr(mddev, rdev->desc_nr))
1392 return -EBUSY;
1393 }
19133a42 1394 bdevname(rdev->bdev,b);
649316b2 1395 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 1396 *s = '!';
649316b2 1397
1da177e4 1398 rdev->mddev = mddev;
19133a42 1399 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1400
b2d6db58 1401 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 1402 goto fail;
86e6ffdd 1403
f637b9f9 1404 if (rdev->bdev->bd_part)
edfaa7c3 1405 ko = &rdev->bdev->bd_part->dev.kobj;
f637b9f9 1406 else
edfaa7c3 1407 ko = &rdev->bdev->bd_disk->dev.kobj;
5e55e2f5
N		 1408 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1409 kobject_del(&rdev->kobj);
1410 goto fail;
1411 }
1412 list_add(&rdev->same_set, &mddev->disks);
c5d79adb 1413 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1da177e4 1414 return 0;
5e55e2f5
N		 1415
1416 fail:
1417 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1418 b, mdname(mddev));
1419 return err;
1da177e4
LT		 1420}
1421
177a99b2 1422static void md_delayed_delete(struct work_struct *ws)
5792a285
N		 1423{
1424 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1425 kobject_del(&rdev->kobj);
177a99b2 1426 kobject_put(&rdev->kobj);
5792a285
N		 1427}
1428
1da177e4
LT		 1429static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1430{
1431 char b[BDEVNAME_SIZE];
1432 if (!rdev->mddev) {
1433 MD_BUG();
1434 return;
1435 }
5463c790 1436 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1da177e4
LT		 1437 list_del_init(&rdev->same_set);
1438 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1439 rdev->mddev = NULL;
86e6ffdd 1440 sysfs_remove_link(&rdev->kobj, "block");
5792a285
N		 1441
1442 /* We need to delay this, otherwise we can deadlock when
1443 * writing to 'remove' to "dev/state"
1444 */
177a99b2
N		 1445 INIT_WORK(&rdev->del_work, md_delayed_delete);
1446 kobject_get(&rdev->kobj);
5792a285 1447 schedule_work(&rdev->del_work);
1da177e4
LT		 1448}
1449
1450/*
1451 * prevent the device from being mounted, repartitioned or
1452 * otherwise reused by a RAID array (or any other kernel
1453 * subsystem), by bd_claiming the device.
1454 */
c5d79adb 1455static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1da177e4
LT		 1456{
1457 int err = 0;
1458 struct block_device *bdev;
1459 char b[BDEVNAME_SIZE];
1460
2e7b651d 1461 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1462 if (IS_ERR(bdev)) {
1463 printk(KERN_ERR "md: could not open %s.\n",
1464 __bdevname(dev, b));
1465 return PTR_ERR(bdev);
1466 }
c5d79adb 1467 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1da177e4
LT		 1468 if (err) {
1469 printk(KERN_ERR "md: could not bd_claim %s.\n",
1470 bdevname(bdev, b));
2e7b651d 1471 blkdev_put(bdev);
1da177e4
LT		 1472 return err;
1473 }
c5d79adb
N		 1474 if (!shared)
1475 set_bit(AllReserved, &rdev->flags);
1da177e4
LT		 1476 rdev->bdev = bdev;
1477 return err;
1478}
1479
1480static void unlock_rdev(mdk_rdev_t *rdev)
1481{
1482 struct block_device *bdev = rdev->bdev;
1483 rdev->bdev = NULL;
1484 if (!bdev)
1485 MD_BUG();
1486 bd_release(bdev);
2e7b651d 1487 blkdev_put(bdev);
1da177e4
LT		 1488}
1489
1490void md_autodetect_dev(dev_t dev);
1491
1492static void export_rdev(mdk_rdev_t * rdev)
1493{
1494 char b[BDEVNAME_SIZE];
1495 printk(KERN_INFO "md: export_rdev(%s)\n",
1496 bdevname(rdev->bdev,b));
1497 if (rdev->mddev)
1498 MD_BUG();
1499 free_disk_sb(rdev);
1500 list_del_init(&rdev->same_set);
1501#ifndef MODULE
1502 md_autodetect_dev(rdev->bdev->bd_dev);
1503#endif
1504 unlock_rdev(rdev);
86e6ffdd 1505 kobject_put(&rdev->kobj);
1da177e4
LT		 1506}
1507
1508static void kick_rdev_from_array(mdk_rdev_t * rdev)
1509{
1510 unbind_rdev_from_array(rdev);
1511 export_rdev(rdev);
1512}
1513
1514static void export_array(mddev_t *mddev)
1515{
1516 struct list_head *tmp;
1517 mdk_rdev_t *rdev;
1518
1519 ITERATE_RDEV(mddev,rdev,tmp) {
1520 if (!rdev->mddev) {
1521 MD_BUG();
1522 continue;
1523 }
1524 kick_rdev_from_array(rdev);
1525 }
1526 if (!list_empty(&mddev->disks))
1527 MD_BUG();
1528 mddev->raid_disks = 0;
1529 mddev->major_version = 0;
1530}
1531
1532static void print_desc(mdp_disk_t *desc)
1533{
1534 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1535 desc->major,desc->minor,desc->raid_disk,desc->state);
1536}
1537
1538static void print_sb(mdp_super_t *sb)
1539{
1540 int i;
1541
1542 printk(KERN_INFO
1543 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1544 sb->major_version, sb->minor_version, sb->patch_version,
1545 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1546 sb->ctime);
1547 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1548 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1549 sb->md_minor, sb->layout, sb->chunk_size);
1550 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1551 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1552 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1553 sb->failed_disks, sb->spare_disks,
1554 sb->sb_csum, (unsigned long)sb->events_lo);
1555
1556 printk(KERN_INFO);
1557 for (i = 0; i < MD_SB_DISKS; i++) {
1558 mdp_disk_t *desc;
1559
1560 desc = sb->disks + i;
1561 if (desc->number || desc->major || desc->minor ||
1562 desc->raid_disk || (desc->state && (desc->state != 4))) {
1563 printk(" D %2d: ", i);
1564 print_desc(desc);
1565 }
1566 }
1567 printk(KERN_INFO "md: THIS: ");
1568 print_desc(&sb->this_disk);
1569
1570}
1571
1572static void print_rdev(mdk_rdev_t *rdev)
1573{
1574 char b[BDEVNAME_SIZE];
1575 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1576 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
b2d444d7
N		 1577 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1578 rdev->desc_nr);
1da177e4
LT		 1579 if (rdev->sb_loaded) {
1580 printk(KERN_INFO "md: rdev superblock:\n");
1581 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1582 } else
1583 printk(KERN_INFO "md: no rdev superblock!\n");
1584}
1585
5e56341d 1586static void md_print_devices(void)
1da177e4
LT		 1587{
1588 struct list_head *tmp, *tmp2;
1589 mdk_rdev_t *rdev;
1590 mddev_t *mddev;
1591 char b[BDEVNAME_SIZE];
1592
1593 printk("\n");
1594 printk("md: **********************************\n");
1595 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1596 printk("md: **********************************\n");
1597 ITERATE_MDDEV(mddev,tmp) {
1da177e4 1598
32a7627c
N		 1599 if (mddev->bitmap)
1600 bitmap_print_sb(mddev->bitmap);
1601 else
1602 printk("%s: ", mdname(mddev));
1da177e4
LT		 1603 ITERATE_RDEV(mddev,rdev,tmp2)
1604 printk("<%s>", bdevname(rdev->bdev,b));
1605 printk("\n");
1606
1607 ITERATE_RDEV(mddev,rdev,tmp2)
1608 print_rdev(rdev);
1609 }
1610 printk("md: **********************************\n");
1611 printk("\n");
1612}
1613
1614
42543769 1615static void sync_sbs(mddev_t * mddev, int nospares)
1da177e4 1616{
42543769
N		 1617 /* Update each superblock (in-memory image), but
1618 * if we are allowed to, skip spares which already
1619 * have the right event counter, or have one earlier
1620 * (which would mean they aren't being marked as dirty
1621 * with the rest of the array)
1622 */
1da177e4
LT		 1623 mdk_rdev_t *rdev;
1624 struct list_head *tmp;
1625
1626 ITERATE_RDEV(mddev,rdev,tmp) {
42543769
N		 1627 if (rdev->sb_events == mddev->events ||
1628 (nospares &&
1629 rdev->raid_disk < 0 &&
1630 (rdev->sb_events&1)==0 &&
1631 rdev->sb_events+1 == mddev->events)) {
1632 /* Don't update this superblock */
1633 rdev->sb_loaded = 2;
1634 } else {
1635 super_types[mddev->major_version].
1636 sync_super(mddev, rdev);
1637 rdev->sb_loaded = 1;
1638 }
1da177e4
LT		 1639 }
1640}
1641
850b2b42 1642static void md_update_sb(mddev_t * mddev, int force_change)
1da177e4 1643{
1da177e4
LT		 1644 struct list_head *tmp;
1645 mdk_rdev_t *rdev;
06d91a5f 1646 int sync_req;
42543769 1647 int nospares = 0;
1da177e4 1648
1da177e4 1649repeat:
a9701a30 1650 spin_lock_irq(&mddev->write_lock);
84692195 1651
850b2b42
N		 1652 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1653 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1654 force_change = 1;
1655 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1656 /* just a clean<-> dirty transition, possibly leave spares alone,
1657 * though if events isn't the right even/odd, we will have to do
1658 * spares after all
1659 */
1660 nospares = 1;
1661 if (force_change)
1662 nospares = 0;
1663 if (mddev->degraded)
84692195
N		 1664 /* If the array is degraded, then skipping spares is both
1665 * dangerous and fairly pointless.
1666 * Dangerous because a device that was removed from the array
1667 * might have a event_count that still looks up-to-date,
1668 * so it can be re-added without a resync.
1669 * Pointless because if there are any spares to skip,
1670 * then a recovery will happen and soon that array won't
1671 * be degraded any more and the spare can go back to sleep then.
1672 */
850b2b42 1673 nospares = 0;
84692195 1674
06d91a5f 1675 sync_req = mddev->in_sync;
1da177e4 1676 mddev->utime = get_seconds();
42543769
N		 1677
1678 /* If this is just a dirty<->clean transition, and the array is clean
1679 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 1680 if (nospares
42543769 1681 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1031be7a
N		 1682 && (mddev->events & 1)
1683 && mddev->events != 1)
42543769
N		 1684 mddev->events--;
1685 else {
1686 /* otherwise we have to go forward and ... */
1687 mddev->events ++;
1688 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1689 /* .. if the array isn't clean, insist on an odd 'events' */
1690 if ((mddev->events&1)==0) {
1691 mddev->events++;
1692 nospares = 0;
1693 }
1694 } else {
1695 /* otherwise insist on an even 'events' (for clean states) */
1696 if ((mddev->events&1)) {
1697 mddev->events++;
1698 nospares = 0;
1699 }
1700 }
1701 }
1da177e4
LT		 1702
1703 if (!mddev->events) {
1704 /*
1705 * oops, this 64-bit counter should never wrap.
1706 * Either we are in around ~1 trillion A.C., assuming
1707 * 1 reboot per second, or we have a bug:
1708 */
1709 MD_BUG();
1710 mddev->events --;
1711 }
1da177e4
LT		 1712
1713 /*
1714 * do not write anything to disk if using
1715 * nonpersistent superblocks
1716 */
06d91a5f 1717 if (!mddev->persistent) {
e691063a
N		 1718 if (!mddev->external)
1719 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1720
a9701a30 1721 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1722 wake_up(&mddev->sb_wait);
1da177e4 1723 return;
06d91a5f 1724 }
e691063a 1725 sync_sbs(mddev, nospares);
a9701a30 1726 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1727
1728 dprintk(KERN_INFO
1729 "md: updating %s RAID superblock on device (in sync %d)\n",
1730 mdname(mddev),mddev->in_sync);
1731
4ad13663 1732 bitmap_update_sb(mddev->bitmap);
1da177e4
LT		 1733 ITERATE_RDEV(mddev,rdev,tmp) {
1734 char b[BDEVNAME_SIZE];
1735 dprintk(KERN_INFO "md: ");
42543769
N		 1736 if (rdev->sb_loaded != 1)
1737 continue; /* no noise on spare devices */
b2d444d7 1738 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1739 dprintk("(skipping faulty ");
1740
1741 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1742 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1743 md_super_write(mddev,rdev,
0002b271 1744 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1745 rdev->sb_page);
1746 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1747 bdevname(rdev->bdev,b),
1748 (unsigned long long)rdev->sb_offset);
42543769 1749 rdev->sb_events = mddev->events;
7bfa19f2 1750
1da177e4
LT		 1751 } else
1752 dprintk(")\n");
7bfa19f2 1753 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1754 /* only need to write one superblock... */
1755 break;
1756 }
a9701a30 1757 md_super_wait(mddev);
850b2b42 1758 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 1759
a9701a30 1760 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 1761 if (mddev->in_sync != sync_req ||
1762 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 1763 /* have to write it out again */
a9701a30 1764 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1765 goto repeat;
1766 }
850b2b42 1767 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 1768 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1769 wake_up(&mddev->sb_wait);
06d91a5f 1770
1da177e4
LT		 1771}
1772
bce74dac
N		 1773/* words written to sysfs files may, or my not, be \n terminated.
1774 * We want to accept with case. For this we use cmd_match.
1775 */
1776static int cmd_match(const char *cmd, const char *str)
1777{
1778 /* See if cmd, written into a sysfs file, matches
1779 * str. They must either be the same, or cmd can
1780 * have a trailing newline
1781 */
1782 while (*cmd && *str && *cmd == *str) {
1783 cmd++;
1784 str++;
1785 }
1786 if (*cmd == '\n')
1787 cmd++;
1788 if (*str || *cmd)
1789 return 0;
1790 return 1;
1791}
1792
86e6ffdd
N		 1793struct rdev_sysfs_entry {
1794 struct attribute attr;
1795 ssize_t (*show)(mdk_rdev_t *, char *);
1796 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1797};
1798
1799static ssize_t
96de1e66 1800state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1801{
1802 char *sep = "";
20a49ff6 1803 size_t len = 0;
86e6ffdd 1804
b2d444d7 1805 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1806 len+= sprintf(page+len, "%sfaulty",sep);
1807 sep = ",";
1808 }
b2d444d7 1809 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1810 len += sprintf(page+len, "%sin_sync",sep);
1811 sep = ",";
1812 }
f655675b
N		 1813 if (test_bit(WriteMostly, &rdev->flags)) {
1814 len += sprintf(page+len, "%swrite_mostly",sep);
1815 sep = ",";
1816 }
b2d444d7
N		 1817 if (!test_bit(Faulty, &rdev->flags) &&
1818 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1819 len += sprintf(page+len, "%sspare", sep);
1820 sep = ",";
1821 }
1822 return len+sprintf(page+len, "\n");
1823}
1824
45dc2de1
N		 1825static ssize_t
1826state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1827{
1828 /* can write
1829 * faulty - simulates and error
1830 * remove - disconnects the device
f655675b
N		 1831 * writemostly - sets write_mostly
1832 * -writemostly - clears write_mostly
45dc2de1
N		 1833 */
1834 int err = -EINVAL;
1835 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1836 md_error(rdev->mddev, rdev);
1837 err = 0;
1838 } else if (cmd_match(buf, "remove")) {
1839 if (rdev->raid_disk >= 0)
1840 err = -EBUSY;
1841 else {
1842 mddev_t *mddev = rdev->mddev;
1843 kick_rdev_from_array(rdev);
3f9d7b0d
N		 1844 if (mddev->pers)
1845 md_update_sb(mddev, 1);
45dc2de1
N		 1846 md_new_event(mddev);
1847 err = 0;
1848 }
f655675b
N		 1849 } else if (cmd_match(buf, "writemostly")) {
1850 set_bit(WriteMostly, &rdev->flags);
1851 err = 0;
1852 } else if (cmd_match(buf, "-writemostly")) {
1853 clear_bit(WriteMostly, &rdev->flags);
1854 err = 0;
45dc2de1
N		 1855 }
1856 return err ? err : len;
1857}
80ca3a44
N		 1858static struct rdev_sysfs_entry rdev_state =
1859__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd
N		 1860
1861static ssize_t
96de1e66 1862super_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1863{
1864 if (rdev->sb_loaded && rdev->sb_size) {
1865 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1866 return rdev->sb_size;
1867 } else
1868 return 0;
1869}
96de1e66
N		 1870static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1871
4dbcdc75
N		 1872static ssize_t
1873errors_show(mdk_rdev_t *rdev, char *page)
1874{
1875 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1876}
1877
1878static ssize_t
1879errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1880{
1881 char *e;
1882 unsigned long n = simple_strtoul(buf, &e, 10);
1883 if (*buf && (*e == 0 || *e == '\n')) {
1884 atomic_set(&rdev->corrected_errors, n);
1885 return len;
1886 }
1887 return -EINVAL;
1888}
1889static struct rdev_sysfs_entry rdev_errors =
80ca3a44 1890__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 1891
014236d2
N		 1892static ssize_t
1893slot_show(mdk_rdev_t *rdev, char *page)
1894{
1895 if (rdev->raid_disk < 0)
1896 return sprintf(page, "none\n");
1897 else
1898 return sprintf(page, "%d\n", rdev->raid_disk);
1899}
1900
1901static ssize_t
1902slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1903{
1904 char *e;
c303da6d
N		 1905 int err;
1906 char nm[20];
014236d2
N		 1907 int slot = simple_strtoul(buf, &e, 10);
1908 if (strncmp(buf, "none", 4)==0)
1909 slot = -1;
1910 else if (e==buf || (*e && *e!= '\n'))
1911 return -EINVAL;
c303da6d
N		 1912 if (rdev->mddev->pers) {
1913 /* Setting 'slot' on an active array requires also
1914 * updating the 'rd%d' link, and communicating
1915 * with the personality with ->hot_*_disk.
1916 * For now we only support removing
1917 * failed/spare devices. This normally happens automatically,
1918 * but not when the metadata is externally managed.
1919 */
1920 if (slot != -1)
1921 return -EBUSY;
1922 if (rdev->raid_disk == -1)
1923 return -EEXIST;
1924 /* personality does all needed checks */
1925 if (rdev->mddev->pers->hot_add_disk == NULL)
1926 return -EINVAL;
1927 err = rdev->mddev->pers->
1928 hot_remove_disk(rdev->mddev, rdev->raid_disk);
1929 if (err)
1930 return err;
1931 sprintf(nm, "rd%d", rdev->raid_disk);
1932 sysfs_remove_link(&rdev->mddev->kobj, nm);
1933 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1934 md_wakeup_thread(rdev->mddev->thread);
1935 } else {
1936 if (slot >= rdev->mddev->raid_disks)
1937 return -ENOSPC;
1938 rdev->raid_disk = slot;
1939 /* assume it is working */
c5d79adb
N		 1940 clear_bit(Faulty, &rdev->flags);
1941 clear_bit(WriteMostly, &rdev->flags);
c303da6d
N		 1942 set_bit(In_sync, &rdev->flags);
1943 }
014236d2
N		 1944 return len;
1945}
1946
1947
1948static struct rdev_sysfs_entry rdev_slot =
80ca3a44 1949__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 1950
93c8cad0
N		 1951static ssize_t
1952offset_show(mdk_rdev_t *rdev, char *page)
1953{
6961ece4 1954 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 1955}
1956
1957static ssize_t
1958offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1959{
1960 char *e;
1961 unsigned long long offset = simple_strtoull(buf, &e, 10);
1962 if (e==buf || (*e && *e != '\n'))
1963 return -EINVAL;
1964 if (rdev->mddev->pers)
1965 return -EBUSY;
c5d79adb
N		 1966 if (rdev->size && rdev->mddev->external)
1967 /* Must set offset before size, so overlap checks
1968 * can be sane */
1969 return -EBUSY;
93c8cad0
N		 1970 rdev->data_offset = offset;
1971 return len;
1972}
1973
1974static struct rdev_sysfs_entry rdev_offset =
80ca3a44 1975__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 1976
83303b61
N		 1977static ssize_t
1978rdev_size_show(mdk_rdev_t *rdev, char *page)
1979{
1980 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1981}
1982
c5d79adb
N		 1983static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
1984{
1985 /* check if two start/length pairs overlap */
1986 if (s1+l1 <= s2)
1987 return 0;
1988 if (s2+l2 <= s1)
1989 return 0;
1990 return 1;
1991}
1992
83303b61
N		 1993static ssize_t
1994rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1995{
1996 char *e;
1997 unsigned long long size = simple_strtoull(buf, &e, 10);
c5d79adb 1998 unsigned long long oldsize = rdev->size;
83303b61
N		 1999 if (e==buf || (*e && *e != '\n'))
2000 return -EINVAL;
2001 if (rdev->mddev->pers)
2002 return -EBUSY;
2003 rdev->size = size;
c5d79adb
N		 2004 if (size > oldsize && rdev->mddev->external) {
2005 /* need to check that all other rdevs with the same ->bdev
2006 * do not overlap. We need to unlock the mddev to avoid
2007 * a deadlock. We have already changed rdev->size, and if
2008 * we have to change it back, we will have the lock again.
2009 */
2010 mddev_t *mddev;
2011 int overlap = 0;
2012 struct list_head *tmp, *tmp2;
2013
2014 mddev_unlock(rdev->mddev);
2015 ITERATE_MDDEV(mddev, tmp) {
2016 mdk_rdev_t *rdev2;
2017
2018 mddev_lock(mddev);
2019 ITERATE_RDEV(mddev, rdev2, tmp2)
2020 if (test_bit(AllReserved, &rdev2->flags) ||
2021 (rdev->bdev == rdev2->bdev &&
2022 rdev != rdev2 &&
2023 overlaps(rdev->data_offset, rdev->size,
2024 rdev2->data_offset, rdev2->size))) {
2025 overlap = 1;
2026 break;
2027 }
2028 mddev_unlock(mddev);
2029 if (overlap) {
2030 mddev_put(mddev);
2031 break;
2032 }
2033 }
2034 mddev_lock(rdev->mddev);
2035 if (overlap) {
2036 /* Someone else could have slipped in a size
2037 * change here, but doing so is just silly.
2038 * We put oldsize back because we *know* it is
2039 * safe, and trust userspace not to race with
2040 * itself
2041 */
2042 rdev->size = oldsize;
2043 return -EBUSY;
2044 }
2045 }
83303b61
N		 2046 if (size < rdev->mddev->size || rdev->mddev->size == 0)
2047 rdev->mddev->size = size;
2048 return len;
2049}
2050
2051static struct rdev_sysfs_entry rdev_size =
80ca3a44 2052__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2053
86e6ffdd
N		 2054static struct attribute *rdev_default_attrs[] = {
2055 &rdev_state.attr,
2056 &rdev_super.attr,
4dbcdc75 2057 &rdev_errors.attr,
014236d2 2058 &rdev_slot.attr,
93c8cad0 2059 &rdev_offset.attr,
83303b61 2060 &rdev_size.attr,
86e6ffdd
N		 2061 NULL,
2062};
2063static ssize_t
2064rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2065{
2066 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2067 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2068
2069 if (!entry->show)
2070 return -EIO;
2071 return entry->show(rdev, page);
2072}
2073
2074static ssize_t
2075rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2076 const char *page, size_t length)
2077{
2078 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2079 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
ca388059 2080 int rv;
86e6ffdd
N		 2081
2082 if (!entry->store)
2083 return -EIO;
67463acb
N		 2084 if (!capable(CAP_SYS_ADMIN))
2085 return -EACCES;
ca388059
N		 2086 rv = mddev_lock(rdev->mddev);
2087 if (!rv) {
2088 rv = entry->store(rdev, page, length);
2089 mddev_unlock(rdev->mddev);
2090 }
2091 return rv;
86e6ffdd
N		 2092}
2093
2094static void rdev_free(struct kobject *ko)
2095{
2096 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2097 kfree(rdev);
2098}
2099static struct sysfs_ops rdev_sysfs_ops = {
2100 .show = rdev_attr_show,
2101 .store = rdev_attr_store,
2102};
2103static struct kobj_type rdev_ktype = {
2104 .release = rdev_free,
2105 .sysfs_ops = &rdev_sysfs_ops,
2106 .default_attrs = rdev_default_attrs,
2107};
2108
1da177e4
LT		 2109/*
2110 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2111 *
2112 * mark the device faulty if:
2113 *
2114 * - the device is nonexistent (zero size)
2115 * - the device has no valid superblock
2116 *
2117 * a faulty rdev _never_ has rdev->sb set.
2118 */
2119static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2120{
2121 char b[BDEVNAME_SIZE];
2122 int err;
2123 mdk_rdev_t *rdev;
2124 sector_t size;
2125
9ffae0cf 2126 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 2127 if (!rdev) {
2128 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2129 return ERR_PTR(-ENOMEM);
2130 }
1da177e4
LT		 2131
2132 if ((err = alloc_disk_sb(rdev)))
2133 goto abort_free;
2134
c5d79adb 2135 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 2136 if (err)
2137 goto abort_free;
2138
f9cb074b 2139 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 2140
1da177e4 2141 rdev->desc_nr = -1;
2b6e8459 2142 rdev->saved_raid_disk = -1;
3f9d7b0d 2143 rdev->raid_disk = -1;
b2d444d7 2144 rdev->flags = 0;
1da177e4 2145 rdev->data_offset = 0;
42543769 2146 rdev->sb_events = 0;
1da177e4 2147 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 2148 atomic_set(&rdev->read_errors, 0);
4dbcdc75 2149 atomic_set(&rdev->corrected_errors, 0);
1da177e4
LT		 2150
2151 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2152 if (!size) {
2153 printk(KERN_WARNING
2154 "md: %s has zero or unknown size, marking faulty!\n",
2155 bdevname(rdev->bdev,b));
2156 err = -EINVAL;
2157 goto abort_free;
2158 }
2159
2160 if (super_format >= 0) {
2161 err = super_types[super_format].
2162 load_super(rdev, NULL, super_minor);
2163 if (err == -EINVAL) {
df968c4e
N		 2164 printk(KERN_WARNING
2165 "md: %s does not have a valid v%d.%d "
2166 "superblock, not importing!\n",
2167 bdevname(rdev->bdev,b),
2168 super_format, super_minor);
1da177e4
LT		 2169 goto abort_free;
2170 }
2171 if (err < 0) {
2172 printk(KERN_WARNING
2173 "md: could not read %s's sb, not importing!\n",
2174 bdevname(rdev->bdev,b));
2175 goto abort_free;
2176 }
2177 }
2178 INIT_LIST_HEAD(&rdev->same_set);
2179
2180 return rdev;
2181
2182abort_free:
2183 if (rdev->sb_page) {
2184 if (rdev->bdev)
2185 unlock_rdev(rdev);
2186 free_disk_sb(rdev);
2187 }
2188 kfree(rdev);
2189 return ERR_PTR(err);
2190}
2191
2192/*
2193 * Check a full RAID array for plausibility
2194 */
2195
2196
a757e64c 2197static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2198{
2199 int i;
2200 struct list_head *tmp;
2201 mdk_rdev_t *rdev, *freshest;
2202 char b[BDEVNAME_SIZE];
2203
2204 freshest = NULL;
2205 ITERATE_RDEV(mddev,rdev,tmp)
2206 switch (super_types[mddev->major_version].
2207 load_super(rdev, freshest, mddev->minor_version)) {
2208 case 1:
2209 freshest = rdev;
2210 break;
2211 case 0:
2212 break;
2213 default:
2214 printk( KERN_ERR \
2215 "md: fatal superblock inconsistency in %s"
2216 " -- removing from array\n",
2217 bdevname(rdev->bdev,b));
2218 kick_rdev_from_array(rdev);
2219 }
2220
2221
2222 super_types[mddev->major_version].
2223 validate_super(mddev, freshest);
2224
2225 i = 0;
2226 ITERATE_RDEV(mddev,rdev,tmp) {
2227 if (rdev != freshest)
2228 if (super_types[mddev->major_version].
2229 validate_super(mddev, rdev)) {
2230 printk(KERN_WARNING "md: kicking non-fresh %s"
2231 " from array!\n",
2232 bdevname(rdev->bdev,b));
2233 kick_rdev_from_array(rdev);
2234 continue;
2235 }
2236 if (mddev->level == LEVEL_MULTIPATH) {
2237 rdev->desc_nr = i++;
2238 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2239 set_bit(In_sync, &rdev->flags);
a778b73f
N		 2240 } else if (rdev->raid_disk >= mddev->raid_disks) {
2241 rdev->raid_disk = -1;
2242 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 2243 }
2244 }
2245
2246
2247
2248 if (mddev->recovery_cp != MaxSector &&
2249 mddev->level >= 1)
2250 printk(KERN_ERR "md: %s: raid array is not clean"
2251 " -- starting background reconstruction\n",
2252 mdname(mddev));
2253
1da177e4
LT		 2254}
2255
16f17b39
N		 2256static ssize_t
2257safe_delay_show(mddev_t *mddev, char *page)
2258{
2259 int msec = (mddev->safemode_delay*1000)/HZ;
2260 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2261}
2262static ssize_t
2263safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2264{
2265 int scale=1;
2266 int dot=0;
2267 int i;
2268 unsigned long msec;
2269 char buf[30];
2270 char *e;
2271 /* remove a period, and count digits after it */
2272 if (len >= sizeof(buf))
2273 return -EINVAL;
2274 strlcpy(buf, cbuf, len);
2275 buf[len] = 0;
2276 for (i=0; i<len; i++) {
2277 if (dot) {
2278 if (isdigit(buf[i])) {
2279 buf[i-1] = buf[i];
2280 scale *= 10;
2281 }
2282 buf[i] = 0;
2283 } else if (buf[i] == '.') {
2284 dot=1;
2285 buf[i] = 0;
2286 }
2287 }
2288 msec = simple_strtoul(buf, &e, 10);
2289 if (e == buf || (*e && *e != '\n'))
2290 return -EINVAL;
2291 msec = (msec * 1000) / scale;
2292 if (msec == 0)
2293 mddev->safemode_delay = 0;
2294 else {
2295 mddev->safemode_delay = (msec*HZ)/1000;
2296 if (mddev->safemode_delay == 0)
2297 mddev->safemode_delay = 1;
2298 }
2299 return len;
2300}
2301static struct md_sysfs_entry md_safe_delay =
80ca3a44 2302__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2303
eae1701f 2304static ssize_t
96de1e66 2305level_show(mddev_t *mddev, char *page)
eae1701f 2306{
2604b703 2307 struct mdk_personality *p = mddev->pers;
d9d166c2 2308 if (p)
eae1701f 2309 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2310 else if (mddev->clevel[0])
2311 return sprintf(page, "%s\n", mddev->clevel);
2312 else if (mddev->level != LEVEL_NONE)
2313 return sprintf(page, "%d\n", mddev->level);
2314 else
2315 return 0;
eae1701f
N		 2316}
2317
d9d166c2
N		 2318static ssize_t
2319level_store(mddev_t *mddev, const char *buf, size_t len)
2320{
20a49ff6 2321 ssize_t rv = len;
d9d166c2
N		 2322 if (mddev->pers)
2323 return -EBUSY;
2324 if (len == 0)
2325 return 0;
2326 if (len >= sizeof(mddev->clevel))
2327 return -ENOSPC;
2328 strncpy(mddev->clevel, buf, len);
2329 if (mddev->clevel[len-1] == '\n')
2330 len--;
2331 mddev->clevel[len] = 0;
2332 mddev->level = LEVEL_NONE;
2333 return rv;
2334}
2335
2336static struct md_sysfs_entry md_level =
80ca3a44 2337__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 2338
d4dbd025
N		 2339
2340static ssize_t
2341layout_show(mddev_t *mddev, char *page)
2342{
2343 /* just a number, not meaningful for all levels */
08a02ecd
N		 2344 if (mddev->reshape_position != MaxSector &&
2345 mddev->layout != mddev->new_layout)
2346 return sprintf(page, "%d (%d)\n",
2347 mddev->new_layout, mddev->layout);
d4dbd025
N		 2348 return sprintf(page, "%d\n", mddev->layout);
2349}
2350
2351static ssize_t
2352layout_store(mddev_t *mddev, const char *buf, size_t len)
2353{
2354 char *e;
2355 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 2356
2357 if (!*buf || (*e && *e != '\n'))
2358 return -EINVAL;
2359
08a02ecd
N		 2360 if (mddev->pers)
2361 return -EBUSY;
2362 if (mddev->reshape_position != MaxSector)
2363 mddev->new_layout = n;
2364 else
2365 mddev->layout = n;
d4dbd025
N		 2366 return len;
2367}
2368static struct md_sysfs_entry md_layout =
80ca3a44 2369__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 2370
2371
eae1701f 2372static ssize_t
96de1e66 2373raid_disks_show(mddev_t *mddev, char *page)
eae1701f 2374{
bb636547
N		 2375 if (mddev->raid_disks == 0)
2376 return 0;
08a02ecd
N		 2377 if (mddev->reshape_position != MaxSector &&
2378 mddev->delta_disks != 0)
2379 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2380 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 2381 return sprintf(page, "%d\n", mddev->raid_disks);
2382}
2383
da943b99
N		 2384static int update_raid_disks(mddev_t *mddev, int raid_disks);
2385
2386static ssize_t
2387raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2388{
da943b99
N		 2389 char *e;
2390 int rv = 0;
2391 unsigned long n = simple_strtoul(buf, &e, 10);
2392
2393 if (!*buf || (*e && *e != '\n'))
2394 return -EINVAL;
2395
2396 if (mddev->pers)
2397 rv = update_raid_disks(mddev, n);
08a02ecd
N		 2398 else if (mddev->reshape_position != MaxSector) {
2399 int olddisks = mddev->raid_disks - mddev->delta_disks;
2400 mddev->delta_disks = n - olddisks;
2401 mddev->raid_disks = n;
2402 } else
da943b99
N		 2403 mddev->raid_disks = n;
2404 return rv ? rv : len;
2405}
2406static struct md_sysfs_entry md_raid_disks =
80ca3a44 2407__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 2408
3b34380a
N		 2409static ssize_t
2410chunk_size_show(mddev_t *mddev, char *page)
2411{
08a02ecd
N		 2412 if (mddev->reshape_position != MaxSector &&
2413 mddev->chunk_size != mddev->new_chunk)
2414 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2415 mddev->chunk_size);
3b34380a
N		 2416 return sprintf(page, "%d\n", mddev->chunk_size);
2417}
2418
2419static ssize_t
2420chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2421{
2422 /* can only set chunk_size if array is not yet active */
2423 char *e;
2424 unsigned long n = simple_strtoul(buf, &e, 10);
2425
3b34380a
N		 2426 if (!*buf || (*e && *e != '\n'))
2427 return -EINVAL;
2428
08a02ecd
N		 2429 if (mddev->pers)
2430 return -EBUSY;
2431 else if (mddev->reshape_position != MaxSector)
2432 mddev->new_chunk = n;
2433 else
2434 mddev->chunk_size = n;
3b34380a
N		 2435 return len;
2436}
2437static struct md_sysfs_entry md_chunk_size =
80ca3a44 2438__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 2439
a94213b1
N		 2440static ssize_t
2441resync_start_show(mddev_t *mddev, char *page)
2442{
2443 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2444}
2445
2446static ssize_t
2447resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2448{
2449 /* can only set chunk_size if array is not yet active */
2450 char *e;
2451 unsigned long long n = simple_strtoull(buf, &e, 10);
2452
2453 if (mddev->pers)
2454 return -EBUSY;
2455 if (!*buf || (*e && *e != '\n'))
2456 return -EINVAL;
2457
2458 mddev->recovery_cp = n;
2459 return len;
2460}
2461static struct md_sysfs_entry md_resync_start =
80ca3a44 2462__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 2463
9e653b63
N		 2464/*
2465 * The array state can be:
2466 *
2467 * clear
2468 * No devices, no size, no level
2469 * Equivalent to STOP_ARRAY ioctl
2470 * inactive
2471 * May have some settings, but array is not active
2472 * all IO results in error
2473 * When written, doesn't tear down array, but just stops it
2474 * suspended (not supported yet)
2475 * All IO requests will block. The array can be reconfigured.
2476 * Writing this, if accepted, will block until array is quiessent
2477 * readonly
2478 * no resync can happen. no superblocks get written.
2479 * write requests fail
2480 * read-auto
2481 * like readonly, but behaves like 'clean' on a write request.
2482 *
2483 * clean - no pending writes, but otherwise active.
2484 * When written to inactive array, starts without resync
2485 * If a write request arrives then
2486 * if metadata is known, mark 'dirty' and switch to 'active'.
2487 * if not known, block and switch to write-pending
2488 * If written to an active array that has pending writes, then fails.
2489 * active
2490 * fully active: IO and resync can be happening.
2491 * When written to inactive array, starts with resync
2492 *
2493 * write-pending
2494 * clean, but writes are blocked waiting for 'active' to be written.
2495 *
2496 * active-idle
2497 * like active, but no writes have been seen for a while (100msec).
2498 *
2499 */
2500enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2501 write_pending, active_idle, bad_word};
05381954 2502static char *array_states[] = {
9e653b63
N		 2503 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2504 "write-pending", "active-idle", NULL };
2505
2506static int match_word(const char *word, char **list)
2507{
2508 int n;
2509 for (n=0; list[n]; n++)
2510 if (cmd_match(word, list[n]))
2511 break;
2512 return n;
2513}
2514
2515static ssize_t
2516array_state_show(mddev_t *mddev, char *page)
2517{
2518 enum array_state st = inactive;
2519
2520 if (mddev->pers)
2521 switch(mddev->ro) {
2522 case 1:
2523 st = readonly;
2524 break;
2525 case 2:
2526 st = read_auto;
2527 break;
2528 case 0:
2529 if (mddev->in_sync)
2530 st = clean;
e691063a
N		 2531 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2532 st = write_pending;
9e653b63
N		 2533 else if (mddev->safemode)
2534 st = active_idle;
2535 else
2536 st = active;
2537 }
2538 else {
2539 if (list_empty(&mddev->disks) &&
2540 mddev->raid_disks == 0 &&
2541 mddev->size == 0)
2542 st = clear;
2543 else
2544 st = inactive;
2545 }
2546 return sprintf(page, "%s\n", array_states[st]);
2547}
2548
2549static int do_md_stop(mddev_t * mddev, int ro);
2550static int do_md_run(mddev_t * mddev);
2551static int restart_array(mddev_t *mddev);
2552
2553static ssize_t
2554array_state_store(mddev_t *mddev, const char *buf, size_t len)
2555{
2556 int err = -EINVAL;
2557 enum array_state st = match_word(buf, array_states);
2558 switch(st) {
2559 case bad_word:
2560 break;
2561 case clear:
2562 /* stopping an active array */
e691063a
N		 2563 if (atomic_read(&mddev->active) > 1)
2564 return -EBUSY;
2565 err = do_md_stop(mddev, 0);
9e653b63
N		 2566 break;
2567 case inactive:
2568 /* stopping an active array */
2569 if (mddev->pers) {
2570 if (atomic_read(&mddev->active) > 1)
2571 return -EBUSY;
2572 err = do_md_stop(mddev, 2);
e691063a
N		 2573 } else
2574 err = 0; /* already inactive */
9e653b63
N		 2575 break;
2576 case suspended:
2577 break; /* not supported yet */
2578 case readonly:
2579 if (mddev->pers)
2580 err = do_md_stop(mddev, 1);
2581 else {
2582 mddev->ro = 1;
2583 err = do_md_run(mddev);
2584 }
2585 break;
2586 case read_auto:
2587 /* stopping an active array */
2588 if (mddev->pers) {
2589 err = do_md_stop(mddev, 1);
2590 if (err == 0)
2591 mddev->ro = 2; /* FIXME mark devices writable */
2592 } else {
2593 mddev->ro = 2;
2594 err = do_md_run(mddev);
2595 }
2596 break;
2597 case clean:
2598 if (mddev->pers) {
2599 restart_array(mddev);
2600 spin_lock_irq(&mddev->write_lock);
2601 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 2602 if (mddev->in_sync == 0) {
2603 mddev->in_sync = 1;
2604 if (mddev->persistent)
2605 set_bit(MD_CHANGE_CLEAN,
2606 &mddev->flags);
2607 }
2608 err = 0;
2609 } else
2610 err = -EBUSY;
9e653b63
N		 2611 spin_unlock_irq(&mddev->write_lock);
2612 } else {
2613 mddev->ro = 0;
2614 mddev->recovery_cp = MaxSector;
2615 err = do_md_run(mddev);
2616 }
2617 break;
2618 case active:
2619 if (mddev->pers) {
2620 restart_array(mddev);
e691063a
N		 2621 if (mddev->external)
2622 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
9e653b63
N		 2623 wake_up(&mddev->sb_wait);
2624 err = 0;
2625 } else {
2626 mddev->ro = 0;
2627 err = do_md_run(mddev);
2628 }
2629 break;
2630 case write_pending:
2631 case active_idle:
2632 /* these cannot be set */
2633 break;
2634 }
2635 if (err)
2636 return err;
2637 else
2638 return len;
2639}
80ca3a44
N		 2640static struct md_sysfs_entry md_array_state =
2641__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 2642
6d7ff738
N		 2643static ssize_t
2644null_show(mddev_t *mddev, char *page)
2645{
2646 return -EINVAL;
2647}
2648
2649static ssize_t
2650new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2651{
2652 /* buf must be %d:%d\n? giving major and minor numbers */
2653 /* The new device is added to the array.
2654 * If the array has a persistent superblock, we read the
2655 * superblock to initialise info and check validity.
2656 * Otherwise, only checking done is that in bind_rdev_to_array,
2657 * which mainly checks size.
2658 */
2659 char *e;
2660 int major = simple_strtoul(buf, &e, 10);
2661 int minor;
2662 dev_t dev;
2663 mdk_rdev_t *rdev;
2664 int err;
2665
2666 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2667 return -EINVAL;
2668 minor = simple_strtoul(e+1, &e, 10);
2669 if (*e && *e != '\n')
2670 return -EINVAL;
2671 dev = MKDEV(major, minor);
2672 if (major != MAJOR(dev) ||
2673 minor != MINOR(dev))
2674 return -EOVERFLOW;
2675
2676
2677 if (mddev->persistent) {
2678 rdev = md_import_device(dev, mddev->major_version,
2679 mddev->minor_version);
2680 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2681 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2682 mdk_rdev_t, same_set);
2683 err = super_types[mddev->major_version]
2684 .load_super(rdev, rdev0, mddev->minor_version);
2685 if (err < 0)
2686 goto out;
2687 }
c5d79adb
N		 2688 } else if (mddev->external)
2689 rdev = md_import_device(dev, -2, -1);
2690 else
6d7ff738
N		 2691 rdev = md_import_device(dev, -1, -1);
2692
2693 if (IS_ERR(rdev))
2694 return PTR_ERR(rdev);
2695 err = bind_rdev_to_array(rdev, mddev);
2696 out:
2697 if (err)
2698 export_rdev(rdev);
2699 return err ? err : len;
2700}
2701
2702static struct md_sysfs_entry md_new_device =
80ca3a44 2703__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 2704
9b1d1dac
PC		 2705static ssize_t
2706bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2707{
2708 char *end;
2709 unsigned long chunk, end_chunk;
2710
2711 if (!mddev->bitmap)
2712 goto out;
2713 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2714 while (*buf) {
2715 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2716 if (buf == end) break;
2717 if (*end == '-') { /* range */
2718 buf = end + 1;
2719 end_chunk = simple_strtoul(buf, &end, 0);
2720 if (buf == end) break;
2721 }
2722 if (*end && !isspace(*end)) break;
2723 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2724 buf = end;
2725 while (isspace(*buf)) buf++;
2726 }
2727 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2728out:
2729 return len;
2730}
2731
2732static struct md_sysfs_entry md_bitmap =
2733__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2734
a35b0d69
N		 2735static ssize_t
2736size_show(mddev_t *mddev, char *page)
2737{
2738 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2739}
2740
2741static int update_size(mddev_t *mddev, unsigned long size);
2742
2743static ssize_t
2744size_store(mddev_t *mddev, const char *buf, size_t len)
2745{
2746 /* If array is inactive, we can reduce the component size, but
2747 * not increase it (except from 0).
2748 * If array is active, we can try an on-line resize
2749 */
2750 char *e;
2751 int err = 0;
2752 unsigned long long size = simple_strtoull(buf, &e, 10);
2753 if (!*buf || *buf == '\n' ||
2754 (*e && *e != '\n'))
2755 return -EINVAL;
2756
2757 if (mddev->pers) {
2758 err = update_size(mddev, size);
850b2b42 2759 md_update_sb(mddev, 1);
a35b0d69
N		 2760 } else {
2761 if (mddev->size == 0 ||
2762 mddev->size > size)
2763 mddev->size = size;
2764 else
2765 err = -ENOSPC;
2766 }
2767 return err ? err : len;
2768}
2769
2770static struct md_sysfs_entry md_size =
80ca3a44 2771__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 2772
8bb93aac
N		 2773
2774/* Metdata version.
e691063a
N		 2775 * This is one of
2776 * 'none' for arrays with no metadata (good luck...)
2777 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 2778 * or N.M for internally known formats
2779 */
2780static ssize_t
2781metadata_show(mddev_t *mddev, char *page)
2782{
2783 if (mddev->persistent)
2784 return sprintf(page, "%d.%d\n",
2785 mddev->major_version, mddev->minor_version);
e691063a
N		 2786 else if (mddev->external)
2787 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 2788 else
2789 return sprintf(page, "none\n");
2790}
2791
2792static ssize_t
2793metadata_store(mddev_t *mddev, const char *buf, size_t len)
2794{
2795 int major, minor;
2796 char *e;
2797 if (!list_empty(&mddev->disks))
2798 return -EBUSY;
2799
2800 if (cmd_match(buf, "none")) {
2801 mddev->persistent = 0;
e691063a
N		 2802 mddev->external = 0;
2803 mddev->major_version = 0;
2804 mddev->minor_version = 90;
2805 return len;
2806 }
2807 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 2808 size_t namelen = len-9;
e691063a
N		 2809 if (namelen >= sizeof(mddev->metadata_type))
2810 namelen = sizeof(mddev->metadata_type)-1;
2811 strncpy(mddev->metadata_type, buf+9, namelen);
2812 mddev->metadata_type[namelen] = 0;
2813 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2814 mddev->metadata_type[--namelen] = 0;
2815 mddev->persistent = 0;
2816 mddev->external = 1;
8bb93aac
N		 2817 mddev->major_version = 0;
2818 mddev->minor_version = 90;
2819 return len;
2820 }
2821 major = simple_strtoul(buf, &e, 10);
2822 if (e==buf || *e != '.')
2823 return -EINVAL;
2824 buf = e+1;
2825 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 2826 if (e==buf || (*e && *e != '\n') )
8bb93aac 2827 return -EINVAL;
50511da3 2828 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 2829 return -ENOENT;
2830 mddev->major_version = major;
2831 mddev->minor_version = minor;
2832 mddev->persistent = 1;
e691063a 2833 mddev->external = 0;
8bb93aac
N		 2834 return len;
2835}
2836
2837static struct md_sysfs_entry md_metadata =
80ca3a44 2838__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 2839
24dd469d 2840static ssize_t
7eec314d 2841action_show(mddev_t *mddev, char *page)
24dd469d 2842{
7eec314d 2843 char *type = "idle";
31399d9e 2844 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 2845 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 2846 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2847 type = "reshape";
2848 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 2849 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2850 type = "resync";
2851 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2852 type = "check";
2853 else
2854 type = "repair";
2855 } else
2856 type = "recover";
2857 }
2858 return sprintf(page, "%s\n", type);
2859}
2860
2861static ssize_t
7eec314d 2862action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 2863{
7eec314d
N		 2864 if (!mddev->pers || !mddev->pers->sync_request)
2865 return -EINVAL;
2866
bce74dac 2867 if (cmd_match(page, "idle")) {
7eec314d
N		 2868 if (mddev->sync_thread) {
2869 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2870 md_unregister_thread(mddev->sync_thread);
2871 mddev->sync_thread = NULL;
2872 mddev->recovery = 0;
2873 }
03c902e1
N		 2874 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2875 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 2876 return -EBUSY;
03c902e1 2877 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
7eec314d 2878 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
16484bf5
N		 2879 else if (cmd_match(page, "reshape")) {
2880 int err;
2881 if (mddev->pers->start_reshape == NULL)
2882 return -EINVAL;
2883 err = mddev->pers->start_reshape(mddev);
2884 if (err)
2885 return err;
2886 } else {
bce74dac 2887 if (cmd_match(page, "check"))
7eec314d 2888 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 2889 else if (!cmd_match(page, "repair"))
7eec314d
N		 2890 return -EINVAL;
2891 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2892 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 2893 }
03c902e1 2894 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 2895 md_wakeup_thread(mddev->thread);
2896 return len;
2897}
2898
9d88883e 2899static ssize_t
96de1e66 2900mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 2901{
2902 return sprintf(page, "%llu\n",
2903 (unsigned long long) mddev->resync_mismatches);
2904}
2905
80ca3a44
N		 2906static struct md_sysfs_entry md_scan_mode =
2907__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 2908
96de1e66 2909
80ca3a44 2910static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 2911
88202a0c
N		 2912static ssize_t
2913sync_min_show(mddev_t *mddev, char *page)
2914{
2915 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2916 mddev->sync_speed_min ? "local": "system");
2917}
2918
2919static ssize_t
2920sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2921{
2922 int min;
2923 char *e;
2924 if (strncmp(buf, "system", 6)==0) {
2925 mddev->sync_speed_min = 0;
2926 return len;
2927 }
2928 min = simple_strtoul(buf, &e, 10);
2929 if (buf == e || (*e && *e != '\n') || min <= 0)
2930 return -EINVAL;
2931 mddev->sync_speed_min = min;
2932 return len;
2933}
2934
2935static struct md_sysfs_entry md_sync_min =
2936__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2937
2938static ssize_t
2939sync_max_show(mddev_t *mddev, char *page)
2940{
2941 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2942 mddev->sync_speed_max ? "local": "system");
2943}
2944
2945static ssize_t
2946sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2947{
2948 int max;
2949 char *e;
2950 if (strncmp(buf, "system", 6)==0) {
2951 mddev->sync_speed_max = 0;
2952 return len;
2953 }
2954 max = simple_strtoul(buf, &e, 10);
2955 if (buf == e || (*e && *e != '\n') || max <= 0)
2956 return -EINVAL;
2957 mddev->sync_speed_max = max;
2958 return len;
2959}
2960
2961static struct md_sysfs_entry md_sync_max =
2962__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2963
d7f3d291
IP		 2964static ssize_t
2965degraded_show(mddev_t *mddev, char *page)
2966{
2967 return sprintf(page, "%d\n", mddev->degraded);
2968}
2969static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c
N		 2970
2971static ssize_t
2972sync_speed_show(mddev_t *mddev, char *page)
2973{
2974 unsigned long resync, dt, db;
ff4e8d9a 2975 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
88202a0c
N		 2976 dt = ((jiffies - mddev->resync_mark) / HZ);
2977 if (!dt) dt++;
2978 db = resync - (mddev->resync_mark_cnt);
2979 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2980}
2981
80ca3a44 2982static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 2983
2984static ssize_t
2985sync_completed_show(mddev_t *mddev, char *page)
2986{
2987 unsigned long max_blocks, resync;
2988
2989 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2990 max_blocks = mddev->resync_max_sectors;
2991 else
2992 max_blocks = mddev->size << 1;
2993
2994 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2995 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2996}
2997
80ca3a44 2998static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 2999
c6207277
N		 3000static ssize_t
3001max_sync_show(mddev_t *mddev, char *page)
3002{
3003 if (mddev->resync_max == MaxSector)
3004 return sprintf(page, "max\n");
3005 else
3006 return sprintf(page, "%llu\n",
3007 (unsigned long long)mddev->resync_max);
3008}
3009static ssize_t
3010max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3011{
3012 if (strncmp(buf, "max", 3) == 0)
3013 mddev->resync_max = MaxSector;
3014 else {
3015 char *ep;
3016 unsigned long long max = simple_strtoull(buf, &ep, 10);
3017 if (ep == buf || (*ep != 0 && *ep != '\n'))
3018 return -EINVAL;
3019 if (max < mddev->resync_max &&
3020 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3021 return -EBUSY;
3022
3023 /* Must be a multiple of chunk_size */
3024 if (mddev->chunk_size) {
3025 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3026 return -EINVAL;
3027 }
3028 mddev->resync_max = max;
3029 }
3030 wake_up(&mddev->recovery_wait);
3031 return len;
3032}
3033
3034static struct md_sysfs_entry md_max_sync =
3035__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3036
e464eafd
N		 3037static ssize_t
3038suspend_lo_show(mddev_t *mddev, char *page)
3039{
3040 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3041}
3042
3043static ssize_t
3044suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3045{
3046 char *e;
3047 unsigned long long new = simple_strtoull(buf, &e, 10);
3048
3049 if (mddev->pers->quiesce == NULL)
3050 return -EINVAL;
3051 if (buf == e || (*e && *e != '\n'))
3052 return -EINVAL;
3053 if (new >= mddev->suspend_hi ||
3054 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3055 mddev->suspend_lo = new;
3056 mddev->pers->quiesce(mddev, 2);
3057 return len;
3058 } else
3059 return -EINVAL;
3060}
3061static struct md_sysfs_entry md_suspend_lo =
3062__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3063
3064
3065static ssize_t
3066suspend_hi_show(mddev_t *mddev, char *page)
3067{
3068 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3069}
3070
3071static ssize_t
3072suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3073{
3074 char *e;
3075 unsigned long long new = simple_strtoull(buf, &e, 10);
3076
3077 if (mddev->pers->quiesce == NULL)
3078 return -EINVAL;
3079 if (buf == e || (*e && *e != '\n'))
3080 return -EINVAL;
3081 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3082 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3083 mddev->suspend_hi = new;
3084 mddev->pers->quiesce(mddev, 1);
3085 mddev->pers->quiesce(mddev, 0);
3086 return len;
3087 } else
3088 return -EINVAL;
3089}
3090static struct md_sysfs_entry md_suspend_hi =
3091__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3092
08a02ecd
N		 3093static ssize_t
3094reshape_position_show(mddev_t *mddev, char *page)
3095{
3096 if (mddev->reshape_position != MaxSector)
3097 return sprintf(page, "%llu\n",
3098 (unsigned long long)mddev->reshape_position);
3099 strcpy(page, "none\n");
3100 return 5;
3101}
3102
3103static ssize_t
3104reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3105{
3106 char *e;
3107 unsigned long long new = simple_strtoull(buf, &e, 10);
3108 if (mddev->pers)
3109 return -EBUSY;
3110 if (buf == e || (*e && *e != '\n'))
3111 return -EINVAL;
3112 mddev->reshape_position = new;
3113 mddev->delta_disks = 0;
3114 mddev->new_level = mddev->level;
3115 mddev->new_layout = mddev->layout;
3116 mddev->new_chunk = mddev->chunk_size;
3117 return len;
3118}
3119
3120static struct md_sysfs_entry md_reshape_position =
3121__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3122 reshape_position_store);
3123
e464eafd 3124
eae1701f
N		 3125static struct attribute *md_default_attrs[] = {
3126 &md_level.attr,
d4dbd025 3127 &md_layout.attr,
eae1701f 3128 &md_raid_disks.attr,
3b34380a 3129 &md_chunk_size.attr,
a35b0d69 3130 &md_size.attr,
a94213b1 3131 &md_resync_start.attr,
8bb93aac 3132 &md_metadata.attr,
6d7ff738 3133 &md_new_device.attr,
16f17b39 3134 &md_safe_delay.attr,
9e653b63 3135 &md_array_state.attr,
08a02ecd 3136 &md_reshape_position.attr,
411036fa
N		 3137 NULL,
3138};
3139
3140static struct attribute *md_redundancy_attrs[] = {
24dd469d 3141 &md_scan_mode.attr,
9d88883e 3142 &md_mismatches.attr,
88202a0c
N		 3143 &md_sync_min.attr,
3144 &md_sync_max.attr,
3145 &md_sync_speed.attr,
3146 &md_sync_completed.attr,
c6207277 3147 &md_max_sync.attr,
e464eafd
N		 3148 &md_suspend_lo.attr,
3149 &md_suspend_hi.attr,
9b1d1dac 3150 &md_bitmap.attr,
d7f3d291 3151 &md_degraded.attr,
eae1701f
N		 3152 NULL,
3153};
411036fa
N		 3154static struct attribute_group md_redundancy_group = {
3155 .name = NULL,
3156 .attrs = md_redundancy_attrs,
3157};
3158
eae1701f
N		 3159
3160static ssize_t
3161md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3162{
3163 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3164 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 3165 ssize_t rv;
eae1701f
N		 3166
3167 if (!entry->show)
3168 return -EIO;
5dc5cf7d
IM		 3169 rv = mddev_lock(mddev);
3170 if (!rv) {
3171 rv = entry->show(mddev, page);
3172 mddev_unlock(mddev);
3173 }
96de1e66 3174 return rv;
eae1701f
N		 3175}
3176
3177static ssize_t
3178md_attr_store(struct kobject *kobj, struct attribute *attr,
3179 const char *page, size_t length)
3180{
3181 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3182 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 3183 ssize_t rv;
eae1701f
N		 3184
3185 if (!entry->store)
3186 return -EIO;
67463acb
N		 3187 if (!capable(CAP_SYS_ADMIN))
3188 return -EACCES;
5dc5cf7d
IM		 3189 rv = mddev_lock(mddev);
3190 if (!rv) {
3191 rv = entry->store(mddev, page, length);
3192 mddev_unlock(mddev);
3193 }
96de1e66 3194 return rv;
eae1701f
N		 3195}
3196
3197static void md_free(struct kobject *ko)
3198{
3199 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3200 kfree(mddev);
3201}
3202
3203static struct sysfs_ops md_sysfs_ops = {
3204 .show = md_attr_show,
3205 .store = md_attr_store,
3206};
3207static struct kobj_type md_ktype = {
3208 .release = md_free,
3209 .sysfs_ops = &md_sysfs_ops,
3210 .default_attrs = md_default_attrs,
3211};
3212
1da177e4
LT		 3213int mdp_major = 0;
3214
3215static struct kobject *md_probe(dev_t dev, int *part, void *data)
3216{
48c9c27b 3217 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 3218 mddev_t *mddev = mddev_find(dev);
3219 struct gendisk *disk;
3220 int partitioned = (MAJOR(dev) != MD_MAJOR);
3221 int shift = partitioned ? MdpMinorShift : 0;
3222 int unit = MINOR(dev) >> shift;
3830c62f 3223 int error;
1da177e4
LT		 3224
3225 if (!mddev)
3226 return NULL;
3227
48c9c27b 3228 mutex_lock(&disks_mutex);
1da177e4 3229 if (mddev->gendisk) {
48c9c27b 3230 mutex_unlock(&disks_mutex);
1da177e4
LT		 3231 mddev_put(mddev);
3232 return NULL;
3233 }
3234 disk = alloc_disk(1 << shift);
3235 if (!disk) {
48c9c27b 3236 mutex_unlock(&disks_mutex);
1da177e4
LT		 3237 mddev_put(mddev);
3238 return NULL;
3239 }
3240 disk->major = MAJOR(dev);
3241 disk->first_minor = unit << shift;
ce7b0f46 3242 if (partitioned)
1da177e4 3243 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 3244 else
1da177e4 3245 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 3246 disk->fops = &md_fops;
3247 disk->private_data = mddev;
3248 disk->queue = mddev->queue;
3249 add_disk(disk);
3250 mddev->gendisk = disk;
48c9c27b 3251 mutex_unlock(&disks_mutex);
edfaa7c3 3252 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3830c62f
GKH		 3253 "%s", "md");
3254 if (error)
5e55e2f5
N		 3255 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3256 disk->disk_name);
3830c62f
GKH		 3257 else
3258 kobject_uevent(&mddev->kobj, KOBJ_ADD);
1da177e4
LT		 3259 return NULL;
3260}
3261
1da177e4
LT		 3262static void md_safemode_timeout(unsigned long data)
3263{
3264 mddev_t *mddev = (mddev_t *) data;
3265
3266 mddev->safemode = 1;
3267 md_wakeup_thread(mddev->thread);
3268}
3269
6ff8d8ec 3270static int start_dirty_degraded;
1da177e4
LT		 3271
3272static int do_md_run(mddev_t * mddev)
3273{
2604b703 3274 int err;
1da177e4
LT		 3275 int chunk_size;
3276 struct list_head *tmp;
3277 mdk_rdev_t *rdev;
3278 struct gendisk *disk;
2604b703 3279 struct mdk_personality *pers;
1da177e4
LT		 3280 char b[BDEVNAME_SIZE];
3281
a757e64c
N		 3282 if (list_empty(&mddev->disks))
3283 /* cannot run an array with no devices.. */
1da177e4 3284 return -EINVAL;
1da177e4
LT		 3285
3286 if (mddev->pers)
3287 return -EBUSY;
3288
3289 /*
3290 * Analyze all RAID superblock(s)
3291 */
1ec4a939
N		 3292 if (!mddev->raid_disks) {
3293 if (!mddev->persistent)
3294 return -EINVAL;
a757e64c 3295 analyze_sbs(mddev);
1ec4a939 3296 }
1da177e4
LT		 3297
3298 chunk_size = mddev->chunk_size;
2604b703
N		 3299
3300 if (chunk_size) {
1da177e4
LT		 3301 if (chunk_size > MAX_CHUNK_SIZE) {
3302 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3303 chunk_size, MAX_CHUNK_SIZE);
3304 return -EINVAL;
3305 }
3306 /*
3307 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3308 */
3309 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 3310 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 3311 return -EINVAL;
3312 }
3313 if (chunk_size < PAGE_SIZE) {
3314 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3315 chunk_size, PAGE_SIZE);
3316 return -EINVAL;
3317 }
3318
3319 /* devices must have minimum size of one chunk */
3320 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 3321 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3322 continue;
3323 if (rdev->size < chunk_size / 1024) {
3324 printk(KERN_WARNING
3325 "md: Dev %s smaller than chunk_size:"
3326 " %lluk < %dk\n",
3327 bdevname(rdev->bdev,b),
3328 (unsigned long long)rdev->size,
3329 chunk_size / 1024);
3330 return -EINVAL;
3331 }
3332 }
3333 }
3334
1da177e4 3335#ifdef CONFIG_KMOD
d9d166c2
N		 3336 if (mddev->level != LEVEL_NONE)
3337 request_module("md-level-%d", mddev->level);
3338 else if (mddev->clevel[0])
3339 request_module("md-%s", mddev->clevel);
1da177e4
LT		 3340#endif
3341
3342 /*
3343 * Drop all container device buffers, from now on
3344 * the only valid external interface is through the md
3345 * device.
1da177e4
LT		 3346 */
3347 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 3348 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3349 continue;
3350 sync_blockdev(rdev->bdev);
f98393a6 3351 invalidate_bdev(rdev->bdev);
f0d76d70
N		 3352
3353 /* perform some consistency tests on the device.
3354 * We don't want the data to overlap the metadata,
3355 * Internal Bitmap issues has handled elsewhere.
3356 */
3357 if (rdev->data_offset < rdev->sb_offset) {
3358 if (mddev->size &&
3359 rdev->data_offset + mddev->size*2
3360 > rdev->sb_offset*2) {
3361 printk("md: %s: data overlaps metadata\n",
3362 mdname(mddev));
3363 return -EINVAL;
3364 }
3365 } else {
3366 if (rdev->sb_offset*2 + rdev->sb_size/512
3367 > rdev->data_offset) {
3368 printk("md: %s: metadata overlaps data\n",
3369 mdname(mddev));
3370 return -EINVAL;
3371 }
3372 }
1da177e4
LT		 3373 }
3374
3375 md_probe(mddev->unit, NULL, NULL);
3376 disk = mddev->gendisk;
3377 if (!disk)
3378 return -ENOMEM;
3379
3380 spin_lock(&pers_lock);
d9d166c2 3381 pers = find_pers(mddev->level, mddev->clevel);
2604b703 3382 if (!pers || !try_module_get(pers->owner)) {
1da177e4 3383 spin_unlock(&pers_lock);
d9d166c2
N		 3384 if (mddev->level != LEVEL_NONE)
3385 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3386 mddev->level);
3387 else
3388 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3389 mddev->clevel);
1da177e4
LT		 3390 return -EINVAL;
3391 }
2604b703 3392 mddev->pers = pers;
1da177e4 3393 spin_unlock(&pers_lock);
d9d166c2
N		 3394 mddev->level = pers->level;
3395 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 3396
f6705578 3397 if (mddev->reshape_position != MaxSector &&
63c70c4f 3398 pers->start_reshape == NULL) {
f6705578
N		 3399 /* This personality cannot handle reshaping... */
3400 mddev->pers = NULL;
3401 module_put(pers->owner);
3402 return -EINVAL;
3403 }
3404
7dd5e7c3
N		 3405 if (pers->sync_request) {
3406 /* Warn if this is a potentially silly
3407 * configuration.
3408 */
3409 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3410 mdk_rdev_t *rdev2;
3411 struct list_head *tmp2;
3412 int warned = 0;
3413 ITERATE_RDEV(mddev, rdev, tmp) {
3414 ITERATE_RDEV(mddev, rdev2, tmp2) {
3415 if (rdev < rdev2 &&
3416 rdev->bdev->bd_contains ==
3417 rdev2->bdev->bd_contains) {
3418 printk(KERN_WARNING
3419 "%s: WARNING: %s appears to be"
3420 " on the same physical disk as"
3421 " %s.\n",
3422 mdname(mddev),
3423 bdevname(rdev->bdev,b),
3424 bdevname(rdev2->bdev,b2));
3425 warned = 1;
3426 }
3427 }
3428 }
3429 if (warned)
3430 printk(KERN_WARNING
3431 "True protection against single-disk"
3432 " failure might be compromised.\n");
3433 }
3434
657390d2 3435 mddev->recovery = 0;
1da177e4 3436 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 3437 mddev->barriers_work = 1;
6ff8d8ec 3438 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 3439
f91de92e
N		 3440 if (start_readonly)
3441 mddev->ro = 2; /* read-only, but switch on first write */
3442
b15c2e57
N		 3443 err = mddev->pers->run(mddev);
3444 if (!err && mddev->pers->sync_request) {
3445 err = bitmap_create(mddev);
3446 if (err) {
3447 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3448 mdname(mddev), err);
3449 mddev->pers->stop(mddev);
3450 }
3451 }
1da177e4
LT		 3452 if (err) {
3453 printk(KERN_ERR "md: pers->run() failed ...\n");
3454 module_put(mddev->pers->owner);
3455 mddev->pers = NULL;
32a7627c
N		 3456 bitmap_destroy(mddev);
3457 return err;
1da177e4 3458 }
5e55e2f5
N		 3459 if (mddev->pers->sync_request) {
3460 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3461 printk(KERN_WARNING
3462 "md: cannot register extra attributes for %s\n",
3463 mdname(mddev));
3464 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 3465 mddev->ro = 0;
3466
1da177e4
LT		 3467 atomic_set(&mddev->writes_pending,0);
3468 mddev->safemode = 0;
3469 mddev->safemode_timer.function = md_safemode_timeout;
3470 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 3471 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 3472 mddev->in_sync = 1;
86e6ffdd
N		 3473
3474 ITERATE_RDEV(mddev,rdev,tmp)
3475 if (rdev->raid_disk >= 0) {
3476 char nm[20];
3477 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 3478 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3479 printk("md: cannot register %s for %s\n",
3480 nm, mdname(mddev));
86e6ffdd 3481 }
1da177e4
LT		 3482
3483 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3484
850b2b42
N		 3485 if (mddev->flags)
3486 md_update_sb(mddev, 0);
1da177e4
LT		 3487
3488 set_capacity(disk, mddev->array_size<<1);
3489
3490 /* If we call blk_queue_make_request here, it will
3491 * re-initialise max_sectors etc which may have been
3492 * refined inside -> run. So just set the bits we need to set.
3493 * Most initialisation happended when we called
3494 * blk_queue_make_request(..., md_fail_request)
3495 * earlier.
3496 */
3497 mddev->queue->queuedata = mddev;
3498 mddev->queue->make_request_fn = mddev->pers->make_request;
3499
5fd6c1dc
N		 3500 /* If there is a partially-recovered drive we need to
3501 * start recovery here. If we leave it to md_check_recovery,
3502 * it will remove the drives and not do the right thing
3503 */
0b8c9de0 3504 if (mddev->degraded && !mddev->sync_thread) {
5fd6c1dc
N		 3505 struct list_head *rtmp;
3506 int spares = 0;
3507 ITERATE_RDEV(mddev,rdev,rtmp)
3508 if (rdev->raid_disk >= 0 &&
3509 !test_bit(In_sync, &rdev->flags) &&
3510 !test_bit(Faulty, &rdev->flags))
3511 /* complete an interrupted recovery */
3512 spares++;
3513 if (spares && mddev->pers->sync_request) {
3514 mddev->recovery = 0;
3515 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3516 mddev->sync_thread = md_register_thread(md_do_sync,
3517 mddev,
3518 "%s_resync");
3519 if (!mddev->sync_thread) {
3520 printk(KERN_ERR "%s: could not start resync"
3521 " thread...\n",
3522 mdname(mddev));
3523 /* leave the spares where they are, it shouldn't hurt */
3524 mddev->recovery = 0;
0b8c9de0 3525 }
5fd6c1dc
N		 3526 }
3527 }
0b8c9de0
N		 3528 md_wakeup_thread(mddev->thread);
3529 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 3530
44ce6294 3531 mddev->changed = 1;
d7603b7e 3532 md_new_event(mddev);
edfaa7c3 3533 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
1da177e4
LT		 3534 return 0;
3535}
3536
3537static int restart_array(mddev_t *mddev)
3538{
3539 struct gendisk *disk = mddev->gendisk;
3540 int err;
3541
3542 /*
3543 * Complain if it has no devices
3544 */
3545 err = -ENXIO;
3546 if (list_empty(&mddev->disks))
3547 goto out;
3548
3549 if (mddev->pers) {
3550 err = -EBUSY;
3551 if (!mddev->ro)
3552 goto out;
3553
3554 mddev->safemode = 0;
3555 mddev->ro = 0;
3556 set_disk_ro(disk, 0);
3557
3558 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3559 mdname(mddev));
3560 /*
3561 * Kick recovery or resync if necessary
3562 */
3563 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3564 md_wakeup_thread(mddev->thread);
5fd6c1dc 3565 md_wakeup_thread(mddev->sync_thread);
1da177e4 3566 err = 0;
9e653b63 3567 } else
1da177e4 3568 err = -EINVAL;
1da177e4
LT		 3569
3570out:
3571 return err;
3572}
3573
acc55e22
N		 3574/* similar to deny_write_access, but accounts for our holding a reference
3575 * to the file ourselves */
3576static int deny_bitmap_write_access(struct file * file)
3577{
3578 struct inode *inode = file->f_mapping->host;
3579
3580 spin_lock(&inode->i_lock);
3581 if (atomic_read(&inode->i_writecount) > 1) {
3582 spin_unlock(&inode->i_lock);
3583 return -ETXTBSY;
3584 }
3585 atomic_set(&inode->i_writecount, -1);
3586 spin_unlock(&inode->i_lock);
3587
3588 return 0;
3589}
3590
3591static void restore_bitmap_write_access(struct file *file)
3592{
3593 struct inode *inode = file->f_mapping->host;
3594
3595 spin_lock(&inode->i_lock);
3596 atomic_set(&inode->i_writecount, 1);
3597 spin_unlock(&inode->i_lock);
3598}
3599
9e653b63
N		 3600/* mode:
3601 * 0 - completely stop and dis-assemble array
3602 * 1 - switch to readonly
3603 * 2 - stop but do not disassemble array
3604 */
3605static int do_md_stop(mddev_t * mddev, int mode)
1da177e4
LT		 3606{
3607 int err = 0;
3608 struct gendisk *disk = mddev->gendisk;
3609
3610 if (mddev->pers) {
3611 if (atomic_read(&mddev->active)>2) {
3612 printk("md: %s still in use.\n",mdname(mddev));
3613 return -EBUSY;
3614 }
3615
3616 if (mddev->sync_thread) {
5fd6c1dc 3617 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4
LT		 3618 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3619 md_unregister_thread(mddev->sync_thread);
3620 mddev->sync_thread = NULL;
3621 }
3622
3623 del_timer_sync(&mddev->safemode_timer);
3624
3625 invalidate_partition(disk, 0);
3626
9e653b63
N		 3627 switch(mode) {
3628 case 1: /* readonly */
1da177e4 3629 err = -ENXIO;
f91de92e 3630 if (mddev->ro==1)
1da177e4
LT		 3631 goto out;
3632 mddev->ro = 1;
9e653b63
N		 3633 break;
3634 case 0: /* disassemble */
3635 case 2: /* stop */
6b8b3e8a 3636 bitmap_flush(mddev);
a9701a30 3637 md_super_wait(mddev);
1da177e4
LT		 3638 if (mddev->ro)
3639 set_disk_ro(disk, 0);
3640 blk_queue_make_request(mddev->queue, md_fail_request);
3641 mddev->pers->stop(mddev);
d1b5380c
N		 3642 mddev->queue->merge_bvec_fn = NULL;
3643 mddev->queue->unplug_fn = NULL;
041ae52e 3644 mddev->queue->backing_dev_info.congested_fn = NULL;
411036fa
N		 3645 if (mddev->pers->sync_request)
3646 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3647
1da177e4
LT		 3648 module_put(mddev->pers->owner);
3649 mddev->pers = NULL;
0d4ca600
N		 3650
3651 set_capacity(disk, 0);
44ce6294 3652 mddev->changed = 1;
0d4ca600 3653
1da177e4
LT		 3654 if (mddev->ro)
3655 mddev->ro = 0;
3656 }
850b2b42 3657 if (!mddev->in_sync || mddev->flags) {
1da177e4
LT		 3658 /* mark array as shutdown cleanly */
3659 mddev->in_sync = 1;
850b2b42 3660 md_update_sb(mddev, 1);
1da177e4 3661 }
9e653b63 3662 if (mode == 1)
1da177e4 3663 set_disk_ro(disk, 1);
5fd6c1dc 3664 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4 3665 }
32a7627c 3666
1da177e4
LT		 3667 /*
3668 * Free resources if final stop
3669 */
9e653b63 3670 if (mode == 0) {
86e6ffdd
N		 3671 mdk_rdev_t *rdev;
3672 struct list_head *tmp;
0d4ca600 3673
1da177e4
LT		 3674 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3675
978f946b
N		 3676 bitmap_destroy(mddev);
3677 if (mddev->bitmap_file) {
acc55e22 3678 restore_bitmap_write_access(mddev->bitmap_file);
978f946b
N		 3679 fput(mddev->bitmap_file);
3680 mddev->bitmap_file = NULL;
3681 }
3682 mddev->bitmap_offset = 0;
3683
86e6ffdd
N		 3684 ITERATE_RDEV(mddev,rdev,tmp)
3685 if (rdev->raid_disk >= 0) {
3686 char nm[20];
3687 sprintf(nm, "rd%d", rdev->raid_disk);
3688 sysfs_remove_link(&mddev->kobj, nm);
3689 }
3690
177a99b2 3691 /* make sure all md_delayed_delete calls have finished */
5792a285
N		 3692 flush_scheduled_work();
3693
1da177e4
LT		 3694 export_array(mddev);
3695
3696 mddev->array_size = 0;
9e653b63
N		 3697 mddev->size = 0;
3698 mddev->raid_disks = 0;
a94213b1 3699 mddev->recovery_cp = 0;
c6207277 3700 mddev->resync_max = MaxSector;
08a02ecd 3701 mddev->reshape_position = MaxSector;
e691063a 3702 mddev->external = 0;
1ec4a939 3703 mddev->persistent = 0;
9e653b63 3704
a8a55c38 3705 } else if (mddev->pers)
1da177e4
LT		 3706 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3707 mdname(mddev));
3708 err = 0;
d7603b7e 3709 md_new_event(mddev);
1da177e4
LT		 3710out:
3711 return err;
3712}
3713
fdee8ae4 3714#ifndef MODULE
1da177e4
LT		 3715static void autorun_array(mddev_t *mddev)
3716{
3717 mdk_rdev_t *rdev;
3718 struct list_head *tmp;
3719 int err;
3720
a757e64c 3721 if (list_empty(&mddev->disks))
1da177e4 3722 return;
1da177e4
LT		 3723
3724 printk(KERN_INFO "md: running: ");
3725
3726 ITERATE_RDEV(mddev,rdev,tmp) {
3727 char b[BDEVNAME_SIZE];
3728 printk("<%s>", bdevname(rdev->bdev,b));
3729 }
3730 printk("\n");
3731
3732 err = do_md_run (mddev);
3733 if (err) {
3734 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3735 do_md_stop (mddev, 0);
3736 }
3737}
3738
3739/*
3740 * lets try to run arrays based on all disks that have arrived
3741 * until now. (those are in pending_raid_disks)
3742 *
3743 * the method: pick the first pending disk, collect all disks with
3744 * the same UUID, remove all from the pending list and put them into
3745 * the 'same_array' list. Then order this list based on superblock
3746 * update time (freshest comes first), kick out 'old' disks and
3747 * compare superblocks. If everything's fine then run it.
3748 *
3749 * If "unit" is allocated, then bump its reference count
3750 */
3751static void autorun_devices(int part)
3752{
1da177e4
LT		 3753 struct list_head *tmp;
3754 mdk_rdev_t *rdev0, *rdev;
3755 mddev_t *mddev;
3756 char b[BDEVNAME_SIZE];
3757
3758 printk(KERN_INFO "md: autorun ...\n");
3759 while (!list_empty(&pending_raid_disks)) {
e8703fe1 3760 int unit;
1da177e4 3761 dev_t dev;
ad01c9e3 3762 LIST_HEAD(candidates);
1da177e4
LT		 3763 rdev0 = list_entry(pending_raid_disks.next,
3764 mdk_rdev_t, same_set);
3765
3766 printk(KERN_INFO "md: considering %s ...\n",
3767 bdevname(rdev0->bdev,b));
3768 INIT_LIST_HEAD(&candidates);
3769 ITERATE_RDEV_PENDING(rdev,tmp)
3770 if (super_90_load(rdev, rdev0, 0) >= 0) {
3771 printk(KERN_INFO "md: adding %s ...\n",
3772 bdevname(rdev->bdev,b));
3773 list_move(&rdev->same_set, &candidates);
3774 }
3775 /*
3776 * now we have a set of devices, with all of them having
3777 * mostly sane superblocks. It's time to allocate the
3778 * mddev.
3779 */
e8703fe1
N		 3780 if (part) {
3781 dev = MKDEV(mdp_major,
3782 rdev0->preferred_minor << MdpMinorShift);
3783 unit = MINOR(dev) >> MdpMinorShift;
3784 } else {
3785 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3786 unit = MINOR(dev);
3787 }
3788 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 3789 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3790 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3791 break;
3792 }
1da177e4
LT		 3793
3794 md_probe(dev, NULL, NULL);
3795 mddev = mddev_find(dev);
3796 if (!mddev) {
3797 printk(KERN_ERR
3798 "md: cannot allocate memory for md drive.\n");
3799 break;
3800 }
3801 if (mddev_lock(mddev))
3802 printk(KERN_WARNING "md: %s locked, cannot run\n",
3803 mdname(mddev));
3804 else if (mddev->raid_disks || mddev->major_version
3805 || !list_empty(&mddev->disks)) {
3806 printk(KERN_WARNING
3807 "md: %s already running, cannot run %s\n",
3808 mdname(mddev), bdevname(rdev0->bdev,b));
3809 mddev_unlock(mddev);
3810 } else {
3811 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 3812 mddev->persistent = 1;
1da177e4
LT		 3813 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3814 list_del_init(&rdev->same_set);
3815 if (bind_rdev_to_array(rdev, mddev))
3816 export_rdev(rdev);
3817 }
3818 autorun_array(mddev);
3819 mddev_unlock(mddev);
3820 }
3821 /* on success, candidates will be empty, on error
3822 * it won't...
3823 */
3824 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3825 export_rdev(rdev);
3826 mddev_put(mddev);
3827 }
3828 printk(KERN_INFO "md: ... autorun DONE.\n");
3829}
fdee8ae4 3830#endif /* !MODULE */
1da177e4 3831
1da177e4
LT		 3832static int get_version(void __user * arg)
3833{
3834 mdu_version_t ver;
3835
3836 ver.major = MD_MAJOR_VERSION;
3837 ver.minor = MD_MINOR_VERSION;
3838 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3839
3840 if (copy_to_user(arg, &ver, sizeof(ver)))
3841 return -EFAULT;
3842
3843 return 0;
3844}
3845
3846static int get_array_info(mddev_t * mddev, void __user * arg)
3847{
3848 mdu_array_info_t info;
3849 int nr,working,active,failed,spare;
3850 mdk_rdev_t *rdev;
3851 struct list_head *tmp;
3852
3853 nr=working=active=failed=spare=0;
3854 ITERATE_RDEV(mddev,rdev,tmp) {
3855 nr++;
b2d444d7 3856 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3857 failed++;
3858 else {
3859 working++;
b2d444d7 3860 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 3861 active++;
3862 else
3863 spare++;
3864 }
3865 }
3866
3867 info.major_version = mddev->major_version;
3868 info.minor_version = mddev->minor_version;
3869 info.patch_version = MD_PATCHLEVEL_VERSION;
3870 info.ctime = mddev->ctime;
3871 info.level = mddev->level;
3872 info.size = mddev->size;
284ae7ca
N		 3873 if (info.size != mddev->size) /* overflow */
3874 info.size = -1;
1da177e4
LT		 3875 info.nr_disks = nr;
3876 info.raid_disks = mddev->raid_disks;
3877 info.md_minor = mddev->md_minor;
3878 info.not_persistent= !mddev->persistent;
3879
3880 info.utime = mddev->utime;
3881 info.state = 0;
3882 if (mddev->in_sync)
3883 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 3884 if (mddev->bitmap && mddev->bitmap_offset)
3885 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 3886 info.active_disks = active;
3887 info.working_disks = working;
3888 info.failed_disks = failed;
3889 info.spare_disks = spare;
3890
3891 info.layout = mddev->layout;
3892 info.chunk_size = mddev->chunk_size;
3893
3894 if (copy_to_user(arg, &info, sizeof(info)))
3895 return -EFAULT;
3896
3897 return 0;
3898}
3899
87162a28 3900static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 3901{
3902 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3903 char *ptr, *buf = NULL;
3904 int err = -ENOMEM;
3905
2a2275d6
N		 3906 md_allow_write(mddev);
3907
32a7627c
N		 3908 file = kmalloc(sizeof(*file), GFP_KERNEL);
3909 if (!file)
3910 goto out;
3911
3912 /* bitmap disabled, zero the first byte and copy out */
3913 if (!mddev->bitmap || !mddev->bitmap->file) {
3914 file->pathname[0] = '\0';
3915 goto copy_out;
3916 }
3917
3918 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3919 if (!buf)
3920 goto out;
3921
3922 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3923 if (!ptr)
3924 goto out;
3925
3926 strcpy(file->pathname, ptr);
3927
3928copy_out:
3929 err = 0;
3930 if (copy_to_user(arg, file, sizeof(*file)))
3931 err = -EFAULT;
3932out:
3933 kfree(buf);
3934 kfree(file);
3935 return err;
3936}
3937
1da177e4
LT		 3938static int get_disk_info(mddev_t * mddev, void __user * arg)
3939{
3940 mdu_disk_info_t info;
3941 unsigned int nr;
3942 mdk_rdev_t *rdev;
3943
3944 if (copy_from_user(&info, arg, sizeof(info)))
3945 return -EFAULT;
3946
3947 nr = info.number;
3948
3949 rdev = find_rdev_nr(mddev, nr);
3950 if (rdev) {
3951 info.major = MAJOR(rdev->bdev->bd_dev);
3952 info.minor = MINOR(rdev->bdev->bd_dev);
3953 info.raid_disk = rdev->raid_disk;
3954 info.state = 0;
b2d444d7 3955 if (test_bit(Faulty, &rdev->flags))
1da177e4 3956 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 3957 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 3958 info.state |= (1<<MD_DISK_ACTIVE);
3959 info.state |= (1<<MD_DISK_SYNC);
3960 }
8ddf9efe
N		 3961 if (test_bit(WriteMostly, &rdev->flags))
3962 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 3963 } else {
3964 info.major = info.minor = 0;
3965 info.raid_disk = -1;
3966 info.state = (1<<MD_DISK_REMOVED);
3967 }
3968
3969 if (copy_to_user(arg, &info, sizeof(info)))
3970 return -EFAULT;
3971
3972 return 0;
3973}
3974
3975static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3976{
3977 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3978 mdk_rdev_t *rdev;
3979 dev_t dev = MKDEV(info->major,info->minor);
3980
3981 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3982 return -EOVERFLOW;
3983
3984 if (!mddev->raid_disks) {
3985 int err;
3986 /* expecting a device which has a superblock */
3987 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3988 if (IS_ERR(rdev)) {
3989 printk(KERN_WARNING
3990 "md: md_import_device returned %ld\n",
3991 PTR_ERR(rdev));
3992 return PTR_ERR(rdev);
3993 }
3994 if (!list_empty(&mddev->disks)) {
3995 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3996 mdk_rdev_t, same_set);
3997 int err = super_types[mddev->major_version]
3998 .load_super(rdev, rdev0, mddev->minor_version);
3999 if (err < 0) {
4000 printk(KERN_WARNING
4001 "md: %s has different UUID to %s\n",
4002 bdevname(rdev->bdev,b),
4003 bdevname(rdev0->bdev,b2));
4004 export_rdev(rdev);
4005 return -EINVAL;
4006 }
4007 }
4008 err = bind_rdev_to_array(rdev, mddev);
4009 if (err)
4010 export_rdev(rdev);
4011 return err;
4012 }
4013
4014 /*
4015 * add_new_disk can be used once the array is assembled
4016 * to add "hot spares". They must already have a superblock
4017 * written
4018 */
4019 if (mddev->pers) {
4020 int err;
4021 if (!mddev->pers->hot_add_disk) {
4022 printk(KERN_WARNING
4023 "%s: personality does not support diskops!\n",
4024 mdname(mddev));
4025 return -EINVAL;
4026 }
7b1e35f6
N		 4027 if (mddev->persistent)
4028 rdev = md_import_device(dev, mddev->major_version,
4029 mddev->minor_version);
4030 else
4031 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 4032 if (IS_ERR(rdev)) {
4033 printk(KERN_WARNING
4034 "md: md_import_device returned %ld\n",
4035 PTR_ERR(rdev));
4036 return PTR_ERR(rdev);
4037 }
41158c7e
N		 4038 /* set save_raid_disk if appropriate */
4039 if (!mddev->persistent) {
4040 if (info->state & (1<<MD_DISK_SYNC) &&
4041 info->raid_disk < mddev->raid_disks)
4042 rdev->raid_disk = info->raid_disk;
4043 else
4044 rdev->raid_disk = -1;
4045 } else
4046 super_types[mddev->major_version].
4047 validate_super(mddev, rdev);
4048 rdev->saved_raid_disk = rdev->raid_disk;
4049
b2d444d7 4050 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 4051 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4052 set_bit(WriteMostly, &rdev->flags);
4053
1da177e4
LT		 4054 rdev->raid_disk = -1;
4055 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 4056 if (!err && !mddev->pers->hot_remove_disk) {
4057 /* If there is hot_add_disk but no hot_remove_disk
4058 * then added disks for geometry changes,
4059 * and should be added immediately.
4060 */
4061 super_types[mddev->major_version].
4062 validate_super(mddev, rdev);
4063 err = mddev->pers->hot_add_disk(mddev, rdev);
4064 if (err)
4065 unbind_rdev_from_array(rdev);
4066 }
1da177e4
LT		 4067 if (err)
4068 export_rdev(rdev);
c361777f 4069
17571284 4070 md_update_sb(mddev, 1);
c361777f 4071 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 4072 md_wakeup_thread(mddev->thread);
1da177e4
LT		 4073 return err;
4074 }
4075
4076 /* otherwise, add_new_disk is only allowed
4077 * for major_version==0 superblocks
4078 */
4079 if (mddev->major_version != 0) {
4080 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4081 mdname(mddev));
4082 return -EINVAL;
4083 }
4084
4085 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4086 int err;
4087 rdev = md_import_device (dev, -1, 0);
4088 if (IS_ERR(rdev)) {
4089 printk(KERN_WARNING
4090 "md: error, md_import_device() returned %ld\n",
4091 PTR_ERR(rdev));
4092 return PTR_ERR(rdev);
4093 }
4094 rdev->desc_nr = info->number;
4095 if (info->raid_disk < mddev->raid_disks)
4096 rdev->raid_disk = info->raid_disk;
4097 else
4098 rdev->raid_disk = -1;
4099
1da177e4 4100 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 4101 if (info->state & (1<<MD_DISK_SYNC))
4102 set_bit(In_sync, &rdev->flags);
1da177e4 4103
8ddf9efe
N		 4104 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4105 set_bit(WriteMostly, &rdev->flags);
4106
1da177e4
LT		 4107 if (!mddev->persistent) {
4108 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4109 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4110 } else
4111 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4112 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4113
2bf071bf
N		 4114 err = bind_rdev_to_array(rdev, mddev);
4115 if (err) {
4116 export_rdev(rdev);
4117 return err;
4118 }
1da177e4
LT		 4119 }
4120
4121 return 0;
4122}
4123
4124static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4125{
4126 char b[BDEVNAME_SIZE];
4127 mdk_rdev_t *rdev;
4128
4129 if (!mddev->pers)
4130 return -ENODEV;
4131
4132 rdev = find_rdev(mddev, dev);
4133 if (!rdev)
4134 return -ENXIO;
4135
4136 if (rdev->raid_disk >= 0)
4137 goto busy;
4138
4139 kick_rdev_from_array(rdev);
850b2b42 4140 md_update_sb(mddev, 1);
d7603b7e 4141 md_new_event(mddev);
1da177e4
LT		 4142
4143 return 0;
4144busy:
4145 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
4146 bdevname(rdev->bdev,b), mdname(mddev));
4147 return -EBUSY;
4148}
4149
4150static int hot_add_disk(mddev_t * mddev, dev_t dev)
4151{
4152 char b[BDEVNAME_SIZE];
4153 int err;
4154 unsigned int size;
4155 mdk_rdev_t *rdev;
4156
4157 if (!mddev->pers)
4158 return -ENODEV;
4159
4160 if (mddev->major_version != 0) {
4161 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4162 " version-0 superblocks.\n",
4163 mdname(mddev));
4164 return -EINVAL;
4165 }
4166 if (!mddev->pers->hot_add_disk) {
4167 printk(KERN_WARNING
4168 "%s: personality does not support diskops!\n",
4169 mdname(mddev));
4170 return -EINVAL;
4171 }
4172
4173 rdev = md_import_device (dev, -1, 0);
4174 if (IS_ERR(rdev)) {
4175 printk(KERN_WARNING
4176 "md: error, md_import_device() returned %ld\n",
4177 PTR_ERR(rdev));
4178 return -EINVAL;
4179 }
4180
4181 if (mddev->persistent)
4182 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4183 else
4184 rdev->sb_offset =
4185 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4186
4187 size = calc_dev_size(rdev, mddev->chunk_size);
4188 rdev->size = size;
4189
b2d444d7 4190 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4191 printk(KERN_WARNING
4192 "md: can not hot-add faulty %s disk to %s!\n",
4193 bdevname(rdev->bdev,b), mdname(mddev));
4194 err = -EINVAL;
4195 goto abort_export;
4196 }
b2d444d7 4197 clear_bit(In_sync, &rdev->flags);
1da177e4 4198 rdev->desc_nr = -1;
5842730d 4199 rdev->saved_raid_disk = -1;
2bf071bf
N		 4200 err = bind_rdev_to_array(rdev, mddev);
4201 if (err)
4202 goto abort_export;
1da177e4
LT		 4203
4204 /*
4205 * The rest should better be atomic, we can have disk failures
4206 * noticed in interrupt contexts ...
4207 */
4208
4209 if (rdev->desc_nr == mddev->max_disks) {
4210 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4211 mdname(mddev));
4212 err = -EBUSY;
4213 goto abort_unbind_export;
4214 }
4215
4216 rdev->raid_disk = -1;
4217
850b2b42 4218 md_update_sb(mddev, 1);
1da177e4
LT		 4219
4220 /*
4221 * Kick recovery, maybe this spare has to be added to the
4222 * array immediately.
4223 */
4224 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4225 md_wakeup_thread(mddev->thread);
d7603b7e 4226 md_new_event(mddev);
1da177e4
LT		 4227 return 0;
4228
4229abort_unbind_export:
4230 unbind_rdev_from_array(rdev);
4231
4232abort_export:
4233 export_rdev(rdev);
4234 return err;
4235}
4236
32a7627c
N		 4237static int set_bitmap_file(mddev_t *mddev, int fd)
4238{
4239 int err;
4240
36fa3063
N		 4241 if (mddev->pers) {
4242 if (!mddev->pers->quiesce)
4243 return -EBUSY;
4244 if (mddev->recovery || mddev->sync_thread)
4245 return -EBUSY;
4246 /* we should be able to change the bitmap.. */
4247 }
32a7627c 4248
32a7627c 4249
36fa3063
N		 4250 if (fd >= 0) {
4251 if (mddev->bitmap)
4252 return -EEXIST; /* cannot add when bitmap is present */
4253 mddev->bitmap_file = fget(fd);
32a7627c 4254
36fa3063
N		 4255 if (mddev->bitmap_file == NULL) {
4256 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4257 mdname(mddev));
4258 return -EBADF;
4259 }
4260
4261 err = deny_bitmap_write_access(mddev->bitmap_file);
4262 if (err) {
4263 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4264 mdname(mddev));
4265 fput(mddev->bitmap_file);
4266 mddev->bitmap_file = NULL;
4267 return err;
4268 }
a654b9d8 4269 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 4270 } else if (mddev->bitmap == NULL)
4271 return -ENOENT; /* cannot remove what isn't there */
4272 err = 0;
4273 if (mddev->pers) {
4274 mddev->pers->quiesce(mddev, 1);
4275 if (fd >= 0)
4276 err = bitmap_create(mddev);
d7375ab3 4277 if (fd < 0 || err) {
36fa3063 4278 bitmap_destroy(mddev);
d7375ab3
N		 4279 fd = -1; /* make sure to put the file */
4280 }
36fa3063 4281 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 4282 }
4283 if (fd < 0) {
acc55e22
N		 4284 if (mddev->bitmap_file) {
4285 restore_bitmap_write_access(mddev->bitmap_file);
36fa3063 4286 fput(mddev->bitmap_file);
acc55e22 4287 }
36fa3063
N		 4288 mddev->bitmap_file = NULL;
4289 }
4290
32a7627c
N		 4291 return err;
4292}
4293
1da177e4
LT		 4294/*
4295 * set_array_info is used two different ways
4296 * The original usage is when creating a new array.
4297 * In this usage, raid_disks is > 0 and it together with
4298 * level, size, not_persistent,layout,chunksize determine the
4299 * shape of the array.
4300 * This will always create an array with a type-0.90.0 superblock.
4301 * The newer usage is when assembling an array.
4302 * In this case raid_disks will be 0, and the major_version field is
4303 * use to determine which style super-blocks are to be found on the devices.
4304 * The minor and patch _version numbers are also kept incase the
4305 * super_block handler wishes to interpret them.
4306 */
4307static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4308{
4309
4310 if (info->raid_disks == 0) {
4311 /* just setting version number for superblock loading */
4312 if (info->major_version < 0 ||
50511da3 4313 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 4314 super_types[info->major_version].name == NULL) {
4315 /* maybe try to auto-load a module? */
4316 printk(KERN_INFO
4317 "md: superblock version %d not known\n",
4318 info->major_version);
4319 return -EINVAL;
4320 }
4321 mddev->major_version = info->major_version;
4322 mddev->minor_version = info->minor_version;
4323 mddev->patch_version = info->patch_version;
3f9d7b0d 4324 mddev->persistent = !info->not_persistent;
1da177e4
LT		 4325 return 0;
4326 }
4327 mddev->major_version = MD_MAJOR_VERSION;
4328 mddev->minor_version = MD_MINOR_VERSION;
4329 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4330 mddev->ctime = get_seconds();
4331
4332 mddev->level = info->level;
17115e03 4333 mddev->clevel[0] = 0;
1da177e4
LT		 4334 mddev->size = info->size;
4335 mddev->raid_disks = info->raid_disks;
4336 /* don't set md_minor, it is determined by which /dev/md* was
4337 * openned
4338 */
4339 if (info->state & (1<<MD_SB_CLEAN))
4340 mddev->recovery_cp = MaxSector;
4341 else
4342 mddev->recovery_cp = 0;
4343 mddev->persistent = ! info->not_persistent;
e691063a 4344 mddev->external = 0;
1da177e4
LT		 4345
4346 mddev->layout = info->layout;
4347 mddev->chunk_size = info->chunk_size;
4348
4349 mddev->max_disks = MD_SB_DISKS;
4350
e691063a
N		 4351 if (mddev->persistent)
4352 mddev->flags = 0;
850b2b42 4353 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 4354
b2a2703c
N		 4355 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4356 mddev->bitmap_offset = 0;
4357
f6705578
N		 4358 mddev->reshape_position = MaxSector;
4359
1da177e4
LT		 4360 /*
4361 * Generate a 128 bit UUID
4362 */
4363 get_random_bytes(mddev->uuid, 16);
4364
f6705578
N		 4365 mddev->new_level = mddev->level;
4366 mddev->new_chunk = mddev->chunk_size;
4367 mddev->new_layout = mddev->layout;
4368 mddev->delta_disks = 0;
4369
1da177e4
LT		 4370 return 0;
4371}
4372
a35b0d69
N		 4373static int update_size(mddev_t *mddev, unsigned long size)
4374{
4375 mdk_rdev_t * rdev;
4376 int rv;
4377 struct list_head *tmp;
8ddeeae5 4378 int fit = (size == 0);
a35b0d69
N		 4379
4380 if (mddev->pers->resize == NULL)
4381 return -EINVAL;
4382 /* The "size" is the amount of each device that is used.
4383 * This can only make sense for arrays with redundancy.
4384 * linear and raid0 always use whatever space is available
4385 * We can only consider changing the size if no resync
4386 * or reconstruction is happening, and if the new size
4387 * is acceptable. It must fit before the sb_offset or,
4388 * if that is <data_offset, it must fit before the
4389 * size of each device.
4390 * If size is zero, we find the largest size that fits.
4391 */
4392 if (mddev->sync_thread)
4393 return -EBUSY;
4394 ITERATE_RDEV(mddev,rdev,tmp) {
4395 sector_t avail;
01ab5662
N		 4396 avail = rdev->size * 2;
4397
a35b0d69
N		 4398 if (fit && (size == 0 || size > avail/2))
4399 size = avail/2;
4400 if (avail < ((sector_t)size << 1))
4401 return -ENOSPC;
4402 }
4403 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4404 if (!rv) {
4405 struct block_device *bdev;
4406
4407 bdev = bdget_disk(mddev->gendisk, 0);
4408 if (bdev) {
1b1dcc1b 4409 mutex_lock(&bdev->bd_inode->i_mutex);
6d89332b 4410 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
1b1dcc1b 4411 mutex_unlock(&bdev->bd_inode->i_mutex);
a35b0d69
N		 4412 bdput(bdev);
4413 }
4414 }
4415 return rv;
4416}
4417
da943b99
N		 4418static int update_raid_disks(mddev_t *mddev, int raid_disks)
4419{
4420 int rv;
4421 /* change the number of raid disks */
63c70c4f 4422 if (mddev->pers->check_reshape == NULL)
da943b99
N		 4423 return -EINVAL;
4424 if (raid_disks <= 0 ||
4425 raid_disks >= mddev->max_disks)
4426 return -EINVAL;
63c70c4f 4427 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 4428 return -EBUSY;
63c70c4f
N		 4429 mddev->delta_disks = raid_disks - mddev->raid_disks;
4430
4431 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 4432 return rv;
4433}
4434
4435
1da177e4
LT		 4436/*
4437 * update_array_info is used to change the configuration of an
4438 * on-line array.
4439 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4440 * fields in the info are checked against the array.
4441 * Any differences that cannot be handled will cause an error.
4442 * Normally, only one change can be managed at a time.
4443 */
4444static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4445{
4446 int rv = 0;
4447 int cnt = 0;
36fa3063
N		 4448 int state = 0;
4449
4450 /* calculate expected state,ignoring low bits */
4451 if (mddev->bitmap && mddev->bitmap_offset)
4452 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 4453
4454 if (mddev->major_version != info->major_version ||
4455 mddev->minor_version != info->minor_version ||
4456/* mddev->patch_version != info->patch_version || */
4457 mddev->ctime != info->ctime ||
4458 mddev->level != info->level ||
4459/* mddev->layout != info->layout || */
4460 !mddev->persistent != info->not_persistent||
36fa3063
N		 4461 mddev->chunk_size != info->chunk_size ||
4462 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4463 ((state^info->state) & 0xfffffe00)
4464 )
1da177e4
LT		 4465 return -EINVAL;
4466 /* Check there is only one change */
284ae7ca 4467 if (info->size >= 0 && mddev->size != info->size) cnt++;
1da177e4
LT		 4468 if (mddev->raid_disks != info->raid_disks) cnt++;
4469 if (mddev->layout != info->layout) cnt++;
36fa3063 4470 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 4471 if (cnt == 0) return 0;
4472 if (cnt > 1) return -EINVAL;
4473
4474 if (mddev->layout != info->layout) {
4475 /* Change layout
4476 * we don't need to do anything at the md level, the
4477 * personality will take care of it all.
4478 */
4479 if (mddev->pers->reconfig == NULL)
4480 return -EINVAL;
4481 else
4482 return mddev->pers->reconfig(mddev, info->layout, -1);
4483 }
284ae7ca 4484 if (info->size >= 0 && mddev->size != info->size)
a35b0d69
N		 4485 rv = update_size(mddev, info->size);
4486
da943b99
N		 4487 if (mddev->raid_disks != info->raid_disks)
4488 rv = update_raid_disks(mddev, info->raid_disks);
4489
36fa3063
N		 4490 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4491 if (mddev->pers->quiesce == NULL)
4492 return -EINVAL;
4493 if (mddev->recovery || mddev->sync_thread)
4494 return -EBUSY;
4495 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4496 /* add the bitmap */
4497 if (mddev->bitmap)
4498 return -EEXIST;
4499 if (mddev->default_bitmap_offset == 0)
4500 return -EINVAL;
4501 mddev->bitmap_offset = mddev->default_bitmap_offset;
4502 mddev->pers->quiesce(mddev, 1);
4503 rv = bitmap_create(mddev);
4504 if (rv)
4505 bitmap_destroy(mddev);
4506 mddev->pers->quiesce(mddev, 0);
4507 } else {
4508 /* remove the bitmap */
4509 if (!mddev->bitmap)
4510 return -ENOENT;
4511 if (mddev->bitmap->file)
4512 return -EINVAL;
4513 mddev->pers->quiesce(mddev, 1);
4514 bitmap_destroy(mddev);
4515 mddev->pers->quiesce(mddev, 0);
4516 mddev->bitmap_offset = 0;
4517 }
4518 }
850b2b42 4519 md_update_sb(mddev, 1);
1da177e4
LT		 4520 return rv;
4521}
4522
4523static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4524{
4525 mdk_rdev_t *rdev;
4526
4527 if (mddev->pers == NULL)
4528 return -ENODEV;
4529
4530 rdev = find_rdev(mddev, dev);
4531 if (!rdev)
4532 return -ENODEV;
4533
4534 md_error(mddev, rdev);
4535 return 0;
4536}
4537
a885c8c4
CH		 4538static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4539{
4540 mddev_t *mddev = bdev->bd_disk->private_data;
4541
4542 geo->heads = 2;
4543 geo->sectors = 4;
4544 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4545 return 0;
4546}
4547
1da177e4
LT		 4548static int md_ioctl(struct inode *inode, struct file *file,
4549 unsigned int cmd, unsigned long arg)
4550{
4551 int err = 0;
4552 void __user *argp = (void __user *)arg;
1da177e4
LT		 4553 mddev_t *mddev = NULL;
4554
4555 if (!capable(CAP_SYS_ADMIN))
4556 return -EACCES;
4557
4558 /*
4559 * Commands dealing with the RAID driver but not any
4560 * particular array:
4561 */
4562 switch (cmd)
4563 {
4564 case RAID_VERSION:
4565 err = get_version(argp);
4566 goto done;
4567
4568 case PRINT_RAID_DEBUG:
4569 err = 0;
4570 md_print_devices();
4571 goto done;
4572
4573#ifndef MODULE
4574 case RAID_AUTORUN:
4575 err = 0;
4576 autostart_arrays(arg);
4577 goto done;
4578#endif
4579 default:;
4580 }
4581
4582 /*
4583 * Commands creating/starting a new array:
4584 */
4585
4586 mddev = inode->i_bdev->bd_disk->private_data;
4587
4588 if (!mddev) {
4589 BUG();
4590 goto abort;
4591 }
4592
1da177e4
LT		 4593 err = mddev_lock(mddev);
4594 if (err) {
4595 printk(KERN_INFO
4596 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4597 err, cmd);
4598 goto abort;
4599 }
4600
4601 switch (cmd)
4602 {
4603 case SET_ARRAY_INFO:
4604 {
4605 mdu_array_info_t info;
4606 if (!arg)
4607 memset(&info, 0, sizeof(info));
4608 else if (copy_from_user(&info, argp, sizeof(info))) {
4609 err = -EFAULT;
4610 goto abort_unlock;
4611 }
4612 if (mddev->pers) {
4613 err = update_array_info(mddev, &info);
4614 if (err) {
4615 printk(KERN_WARNING "md: couldn't update"
4616 " array info. %d\n", err);
4617 goto abort_unlock;
4618 }
4619 goto done_unlock;
4620 }
4621 if (!list_empty(&mddev->disks)) {
4622 printk(KERN_WARNING
4623 "md: array %s already has disks!\n",
4624 mdname(mddev));
4625 err = -EBUSY;
4626 goto abort_unlock;
4627 }
4628 if (mddev->raid_disks) {
4629 printk(KERN_WARNING
4630 "md: array %s already initialised!\n",
4631 mdname(mddev));
4632 err = -EBUSY;
4633 goto abort_unlock;
4634 }
4635 err = set_array_info(mddev, &info);
4636 if (err) {
4637 printk(KERN_WARNING "md: couldn't set"
4638 " array info. %d\n", err);
4639 goto abort_unlock;
4640 }
4641 }
4642 goto done_unlock;
4643
4644 default:;
4645 }
4646
4647 /*
4648 * Commands querying/configuring an existing array:
4649 */
32a7627c 4650 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 4651 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 4652 if ((!mddev->raid_disks && !mddev->external)
4653 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4654 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4655 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 4656 err = -ENODEV;
4657 goto abort_unlock;
4658 }
4659
4660 /*
4661 * Commands even a read-only array can execute:
4662 */
4663 switch (cmd)
4664 {
4665 case GET_ARRAY_INFO:
4666 err = get_array_info(mddev, argp);
4667 goto done_unlock;
4668
32a7627c 4669 case GET_BITMAP_FILE:
87162a28 4670 err = get_bitmap_file(mddev, argp);
32a7627c
N		 4671 goto done_unlock;
4672
1da177e4
LT		 4673 case GET_DISK_INFO:
4674 err = get_disk_info(mddev, argp);
4675 goto done_unlock;
4676
4677 case RESTART_ARRAY_RW:
4678 err = restart_array(mddev);
4679 goto done_unlock;
4680
4681 case STOP_ARRAY:
4682 err = do_md_stop (mddev, 0);
4683 goto done_unlock;
4684
4685 case STOP_ARRAY_RO:
4686 err = do_md_stop (mddev, 1);
4687 goto done_unlock;
4688
4689 /*
4690 * We have a problem here : there is no easy way to give a CHS
4691 * virtual geometry. We currently pretend that we have a 2 heads
4692 * 4 sectors (with a BIG number of cylinders...). This drives
4693 * dosfs just mad... ;-)
4694 */
1da177e4
LT		 4695 }
4696
4697 /*
4698 * The remaining ioctls are changing the state of the
f91de92e
N		 4699 * superblock, so we do not allow them on read-only arrays.
4700 * However non-MD ioctls (e.g. get-size) will still come through
4701 * here and hit the 'default' below, so only disallow
4702 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 4703 */
f91de92e
N		 4704 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4705 mddev->ro && mddev->pers) {
4706 if (mddev->ro == 2) {
4707 mddev->ro = 0;
4708 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4709 md_wakeup_thread(mddev->thread);
4710
4711 } else {
4712 err = -EROFS;
4713 goto abort_unlock;
4714 }
1da177e4
LT		 4715 }
4716
4717 switch (cmd)
4718 {
4719 case ADD_NEW_DISK:
4720 {
4721 mdu_disk_info_t info;
4722 if (copy_from_user(&info, argp, sizeof(info)))
4723 err = -EFAULT;
4724 else
4725 err = add_new_disk(mddev, &info);
4726 goto done_unlock;
4727 }
4728
4729 case HOT_REMOVE_DISK:
4730 err = hot_remove_disk(mddev, new_decode_dev(arg));
4731 goto done_unlock;
4732
4733 case HOT_ADD_DISK:
4734 err = hot_add_disk(mddev, new_decode_dev(arg));
4735 goto done_unlock;
4736
4737 case SET_DISK_FAULTY:
4738 err = set_disk_faulty(mddev, new_decode_dev(arg));
4739 goto done_unlock;
4740
4741 case RUN_ARRAY:
4742 err = do_md_run (mddev);
4743 goto done_unlock;
4744
32a7627c
N		 4745 case SET_BITMAP_FILE:
4746 err = set_bitmap_file(mddev, (int)arg);
4747 goto done_unlock;
4748
1da177e4 4749 default:
1da177e4
LT		 4750 err = -EINVAL;
4751 goto abort_unlock;
4752 }
4753
4754done_unlock:
4755abort_unlock:
4756 mddev_unlock(mddev);
4757
4758 return err;
4759done:
4760 if (err)
4761 MD_BUG();
4762abort:
4763 return err;
4764}
4765
4766static int md_open(struct inode *inode, struct file *file)
4767{
4768 /*
4769 * Succeed if we can lock the mddev, which confirms that
4770 * it isn't being stopped right now.
4771 */
4772 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4773 int err;
4774
d63a5a74 4775 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
1da177e4
LT		 4776 goto out;
4777
4778 err = 0;
4779 mddev_get(mddev);
4780 mddev_unlock(mddev);
4781
4782 check_disk_change(inode->i_bdev);
4783 out:
4784 return err;
4785}
4786
4787static int md_release(struct inode *inode, struct file * file)
4788{
4789 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4790
52e5f9d1 4791 BUG_ON(!mddev);
1da177e4
LT		 4792 mddev_put(mddev);
4793
4794 return 0;
4795}
4796
44ce6294
LT		 4797static int md_media_changed(struct gendisk *disk)
4798{
4799 mddev_t *mddev = disk->private_data;
4800
4801 return mddev->changed;
4802}
4803
4804static int md_revalidate(struct gendisk *disk)
4805{
4806 mddev_t *mddev = disk->private_data;
4807
4808 mddev->changed = 0;
4809 return 0;
4810}
1da177e4
LT		 4811static struct block_device_operations md_fops =
4812{
4813 .owner = THIS_MODULE,
4814 .open = md_open,
4815 .release = md_release,
4816 .ioctl = md_ioctl,
a885c8c4 4817 .getgeo = md_getgeo,
44ce6294
LT		 4818 .media_changed = md_media_changed,
4819 .revalidate_disk= md_revalidate,
1da177e4
LT		 4820};
4821
75c96f85 4822static int md_thread(void * arg)
1da177e4
LT		 4823{
4824 mdk_thread_t *thread = arg;
4825
1da177e4
LT		 4826 /*
4827 * md_thread is a 'system-thread', it's priority should be very
4828 * high. We avoid resource deadlocks individually in each
4829 * raid personality. (RAID5 does preallocation) We also use RR and
4830 * the very same RT priority as kswapd, thus we will never get
4831 * into a priority inversion deadlock.
4832 *
4833 * we definitely have to have equal or higher priority than
4834 * bdflush, otherwise bdflush will deadlock if there are too
4835 * many dirty RAID5 blocks.
4836 */
1da177e4 4837
6985c43f 4838 allow_signal(SIGKILL);
a6fb0934 4839 while (!kthread_should_stop()) {
1da177e4 4840
93588e22
N		 4841 /* We need to wait INTERRUPTIBLE so that
4842 * we don't add to the load-average.
4843 * That means we need to be sure no signals are
4844 * pending
4845 */
4846 if (signal_pending(current))
4847 flush_signals(current);
4848
4849 wait_event_interruptible_timeout
4850 (thread->wqueue,
4851 test_bit(THREAD_WAKEUP, &thread->flags)
4852 || kthread_should_stop(),
4853 thread->timeout);
1da177e4
LT		 4854
4855 clear_bit(THREAD_WAKEUP, &thread->flags);
4856
787453c2 4857 thread->run(thread->mddev);
1da177e4 4858 }
a6fb0934 4859
1da177e4
LT		 4860 return 0;
4861}
4862
4863void md_wakeup_thread(mdk_thread_t *thread)
4864{
4865 if (thread) {
4866 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4867 set_bit(THREAD_WAKEUP, &thread->flags);
4868 wake_up(&thread->wqueue);
4869 }
4870}
4871
4872mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4873 const char *name)
4874{
4875 mdk_thread_t *thread;
1da177e4 4876
9ffae0cf 4877 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 4878 if (!thread)
4879 return NULL;
4880
1da177e4
LT		 4881 init_waitqueue_head(&thread->wqueue);
4882
1da177e4
LT		 4883 thread->run = run;
4884 thread->mddev = mddev;
32a7627c 4885 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 4886 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 4887 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 4888 kfree(thread);
4889 return NULL;
4890 }
1da177e4
LT		 4891 return thread;
4892}
4893
1da177e4
LT		 4894void md_unregister_thread(mdk_thread_t *thread)
4895{
ba25f9dc 4896 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
a6fb0934
N		 4897
4898 kthread_stop(thread->tsk);
1da177e4
LT		 4899 kfree(thread);
4900}
4901
4902void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4903{
4904 if (!mddev) {
4905 MD_BUG();
4906 return;
4907 }
4908
b2d444d7 4909 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 4910 return;
32a7627c 4911/*
1da177e4
LT		 4912 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4913 mdname(mddev),
4914 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4915 __builtin_return_address(0),__builtin_return_address(1),
4916 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 4917*/
d0a0a5ee
AM		 4918 if (!mddev->pers)
4919 return;
1da177e4
LT		 4920 if (!mddev->pers->error_handler)
4921 return;
4922 mddev->pers->error_handler(mddev,rdev);
4923 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4924 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4925 md_wakeup_thread(mddev->thread);
c331eb04 4926 md_new_event_inintr(mddev);
1da177e4
LT		 4927}
4928
4929/* seq_file implementation /proc/mdstat */
4930
4931static void status_unused(struct seq_file *seq)
4932{
4933 int i = 0;
4934 mdk_rdev_t *rdev;
4935 struct list_head *tmp;
4936
4937 seq_printf(seq, "unused devices: ");
4938
4939 ITERATE_RDEV_PENDING(rdev,tmp) {
4940 char b[BDEVNAME_SIZE];
4941 i++;
4942 seq_printf(seq, "%s ",
4943 bdevname(rdev->bdev,b));
4944 }
4945 if (!i)
4946 seq_printf(seq, "<none>");
4947
4948 seq_printf(seq, "\n");
4949}
4950
4951
4952static void status_resync(struct seq_file *seq, mddev_t * mddev)
4953{
4588b42e
N		 4954 sector_t max_blocks, resync, res;
4955 unsigned long dt, db, rt;
4956 int scale;
4957 unsigned int per_milli;
1da177e4
LT		 4958
4959 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4960
4961 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4962 max_blocks = mddev->resync_max_sectors >> 1;
4963 else
4964 max_blocks = mddev->size;
4965
4966 /*
4967 * Should not happen.
4968 */
4969 if (!max_blocks) {
4970 MD_BUG();
4971 return;
4972 }
4588b42e
N		 4973 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4974 * in a sector_t, and (max_blocks>>scale) will fit in a
4975 * u32, as those are the requirements for sector_div.
4976 * Thus 'scale' must be at least 10
4977 */
4978 scale = 10;
4979 if (sizeof(sector_t) > sizeof(unsigned long)) {
4980 while ( max_blocks/2 > (1ULL<<(scale+32)))
4981 scale++;
4982 }
4983 res = (resync>>scale)*1000;
4984 sector_div(res, (u32)((max_blocks>>scale)+1));
4985
4986 per_milli = res;
1da177e4 4987 {
4588b42e 4988 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 4989 seq_printf(seq, "[");
4990 for (i = 0; i < x; i++)
4991 seq_printf(seq, "=");
4992 seq_printf(seq, ">");
4993 for (i = 0; i < y; i++)
4994 seq_printf(seq, ".");
4995 seq_printf(seq, "] ");
4996 }
4588b42e 4997 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 4998 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4999 "reshape" :
61df9d91
N		 5000 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5001 "check" :
5002 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5003 "resync" : "recovery"))),
5004 per_milli/10, per_milli % 10,
4588b42e
N		 5005 (unsigned long long) resync,
5006 (unsigned long long) max_blocks);
1da177e4
LT		 5007
5008 /*
5009 * We do not want to overflow, so the order of operands and
5010 * the * 100 / 100 trick are important. We do a +1 to be
5011 * safe against division by zero. We only estimate anyway.
5012 *
5013 * dt: time from mark until now
5014 * db: blocks written from mark until now
5015 * rt: remaining time
5016 */
5017 dt = ((jiffies - mddev->resync_mark) / HZ);
5018 if (!dt) dt++;
ff4e8d9a
N		 5019 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5020 - mddev->resync_mark_cnt;
5021 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
1da177e4
LT		 5022
5023 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5024
ff4e8d9a 5025 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 5026}
5027
5028static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5029{
5030 struct list_head *tmp;
5031 loff_t l = *pos;
5032 mddev_t *mddev;
5033
5034 if (l >= 0x10000)
5035 return NULL;
5036 if (!l--)
5037 /* header */
5038 return (void*)1;
5039
5040 spin_lock(&all_mddevs_lock);
5041 list_for_each(tmp,&all_mddevs)
5042 if (!l--) {
5043 mddev = list_entry(tmp, mddev_t, all_mddevs);
5044 mddev_get(mddev);
5045 spin_unlock(&all_mddevs_lock);
5046 return mddev;
5047 }
5048 spin_unlock(&all_mddevs_lock);
5049 if (!l--)
5050 return (void*)2;/* tail */
5051 return NULL;
5052}
5053
5054static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5055{
5056 struct list_head *tmp;
5057 mddev_t *next_mddev, *mddev = v;
5058
5059 ++*pos;
5060 if (v == (void*)2)
5061 return NULL;
5062
5063 spin_lock(&all_mddevs_lock);
5064 if (v == (void*)1)
5065 tmp = all_mddevs.next;
5066 else
5067 tmp = mddev->all_mddevs.next;
5068 if (tmp != &all_mddevs)
5069 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5070 else {
5071 next_mddev = (void*)2;
5072 *pos = 0x10000;
5073 }
5074 spin_unlock(&all_mddevs_lock);
5075
5076 if (v != (void*)1)
5077 mddev_put(mddev);
5078 return next_mddev;
5079
5080}
5081
5082static void md_seq_stop(struct seq_file *seq, void *v)
5083{
5084 mddev_t *mddev = v;
5085
5086 if (mddev && v != (void*)1 && v != (void*)2)
5087 mddev_put(mddev);
5088}
5089
d7603b7e
N		 5090struct mdstat_info {
5091 int event;
5092};
5093
1da177e4
LT		 5094static int md_seq_show(struct seq_file *seq, void *v)
5095{
5096 mddev_t *mddev = v;
5097 sector_t size;
5098 struct list_head *tmp2;
5099 mdk_rdev_t *rdev;
d7603b7e 5100 struct mdstat_info *mi = seq->private;
32a7627c 5101 struct bitmap *bitmap;
1da177e4
LT		 5102
5103 if (v == (void*)1) {
2604b703 5104 struct mdk_personality *pers;
1da177e4
LT		 5105 seq_printf(seq, "Personalities : ");
5106 spin_lock(&pers_lock);
2604b703
N		 5107 list_for_each_entry(pers, &pers_list, list)
5108 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 5109
5110 spin_unlock(&pers_lock);
5111 seq_printf(seq, "\n");
d7603b7e 5112 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 5113 return 0;
5114 }
5115 if (v == (void*)2) {
5116 status_unused(seq);
5117 return 0;
5118 }
5119
5dc5cf7d 5120 if (mddev_lock(mddev) < 0)
1da177e4 5121 return -EINTR;
5dc5cf7d 5122
1da177e4
LT		 5123 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5124 seq_printf(seq, "%s : %sactive", mdname(mddev),
5125 mddev->pers ? "" : "in");
5126 if (mddev->pers) {
f91de92e 5127 if (mddev->ro==1)
1da177e4 5128 seq_printf(seq, " (read-only)");
f91de92e
N		 5129 if (mddev->ro==2)
5130 seq_printf(seq, "(auto-read-only)");
1da177e4
LT		 5131 seq_printf(seq, " %s", mddev->pers->name);
5132 }
5133
5134 size = 0;
5135 ITERATE_RDEV(mddev,rdev,tmp2) {
5136 char b[BDEVNAME_SIZE];
5137 seq_printf(seq, " %s[%d]",
5138 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 5139 if (test_bit(WriteMostly, &rdev->flags))
5140 seq_printf(seq, "(W)");
b2d444d7 5141 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5142 seq_printf(seq, "(F)");
5143 continue;
b325a32e
N		 5144 } else if (rdev->raid_disk < 0)
5145 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 5146 size += rdev->size;
5147 }
5148
5149 if (!list_empty(&mddev->disks)) {
5150 if (mddev->pers)
5151 seq_printf(seq, "\n %llu blocks",
5152 (unsigned long long)mddev->array_size);
5153 else
5154 seq_printf(seq, "\n %llu blocks",
5155 (unsigned long long)size);
5156 }
1cd6bf19
N		 5157 if (mddev->persistent) {
5158 if (mddev->major_version != 0 ||
5159 mddev->minor_version != 90) {
5160 seq_printf(seq," super %d.%d",
5161 mddev->major_version,
5162 mddev->minor_version);
5163 }
e691063a
N		 5164 } else if (mddev->external)
5165 seq_printf(seq, " super external:%s",
5166 mddev->metadata_type);
5167 else
1cd6bf19 5168 seq_printf(seq, " super non-persistent");
1da177e4
LT		 5169
5170 if (mddev->pers) {
5171 mddev->pers->status (seq, mddev);
5172 seq_printf(seq, "\n ");
8e1b39d6
N		 5173 if (mddev->pers->sync_request) {
5174 if (mddev->curr_resync > 2) {
5175 status_resync (seq, mddev);
5176 seq_printf(seq, "\n ");
5177 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5178 seq_printf(seq, "\tresync=DELAYED\n ");
5179 else if (mddev->recovery_cp < MaxSector)
5180 seq_printf(seq, "\tresync=PENDING\n ");
5181 }
32a7627c
N		 5182 } else
5183 seq_printf(seq, "\n ");
5184
5185 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 5186 unsigned long chunk_kb;
5187 unsigned long flags;
32a7627c
N		 5188 spin_lock_irqsave(&bitmap->lock, flags);
5189 chunk_kb = bitmap->chunksize >> 10;
5190 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5191 "%lu%s chunk",
5192 bitmap->pages - bitmap->missing_pages,
5193 bitmap->pages,
5194 (bitmap->pages - bitmap->missing_pages)
5195 << (PAGE_SHIFT - 10),
5196 chunk_kb ? chunk_kb : bitmap->chunksize,
5197 chunk_kb ? "KB" : "B");
78d742d8
N		 5198 if (bitmap->file) {
5199 seq_printf(seq, ", file: ");
c649bb9c
JS		 5200 seq_path(seq, bitmap->file->f_path.mnt,
5201 bitmap->file->f_path.dentry," \t\n");
32a7627c 5202 }
78d742d8 5203
32a7627c
N		 5204 seq_printf(seq, "\n");
5205 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 5206 }
5207
5208 seq_printf(seq, "\n");
5209 }
5210 mddev_unlock(mddev);
5211
5212 return 0;
5213}
5214
5215static struct seq_operations md_seq_ops = {
5216 .start = md_seq_start,
5217 .next = md_seq_next,
5218 .stop = md_seq_stop,
5219 .show = md_seq_show,
5220};
5221
5222static int md_seq_open(struct inode *inode, struct file *file)
5223{
5224 int error;
d7603b7e
N		 5225 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5226 if (mi == NULL)
5227 return -ENOMEM;
1da177e4
LT		 5228
5229 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 5230 if (error)
5231 kfree(mi);
5232 else {
5233 struct seq_file *p = file->private_data;
5234 p->private = mi;
5235 mi->event = atomic_read(&md_event_count);
5236 }
1da177e4
LT		 5237 return error;
5238}
5239
d7603b7e
N		 5240static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5241{
5242 struct seq_file *m = filp->private_data;
5243 struct mdstat_info *mi = m->private;
5244 int mask;
5245
5246 poll_wait(filp, &md_event_waiters, wait);
5247
5248 /* always allow read */
5249 mask = POLLIN | POLLRDNORM;
5250
5251 if (mi->event != atomic_read(&md_event_count))
5252 mask |= POLLERR | POLLPRI;
5253 return mask;
5254}
5255
fa027c2a 5256static const struct file_operations md_seq_fops = {
e24650c2 5257 .owner = THIS_MODULE,
1da177e4
LT		 5258 .open = md_seq_open,
5259 .read = seq_read,
5260 .llseek = seq_lseek,
c3f94b40 5261 .release = seq_release_private,
d7603b7e 5262 .poll = mdstat_poll,
1da177e4
LT		 5263};
5264
2604b703 5265int register_md_personality(struct mdk_personality *p)
1da177e4 5266{
1da177e4 5267 spin_lock(&pers_lock);
2604b703
N		 5268 list_add_tail(&p->list, &pers_list);
5269 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 5270 spin_unlock(&pers_lock);
5271 return 0;
5272}
5273
2604b703 5274int unregister_md_personality(struct mdk_personality *p)
1da177e4 5275{
2604b703 5276 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 5277 spin_lock(&pers_lock);
2604b703 5278 list_del_init(&p->list);
1da177e4
LT		 5279 spin_unlock(&pers_lock);
5280 return 0;
5281}
5282
5283static int is_mddev_idle(mddev_t *mddev)
5284{
5285 mdk_rdev_t * rdev;
5286 struct list_head *tmp;
5287 int idle;
713f6ab1 5288 long curr_events;
1da177e4
LT		 5289
5290 idle = 1;
5291 ITERATE_RDEV(mddev,rdev,tmp) {
5292 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 5293 curr_events = disk_stat_read(disk, sectors[0]) +
5294 disk_stat_read(disk, sectors[1]) -
1da177e4 5295 atomic_read(&disk->sync_io);
713f6ab1
N		 5296 /* sync IO will cause sync_io to increase before the disk_stats
5297 * as sync_io is counted when a request starts, and
5298 * disk_stats is counted when it completes.
5299 * So resync activity will cause curr_events to be smaller than
5300 * when there was no such activity.
5301 * non-sync IO will cause disk_stat to increase without
5302 * increasing sync_io so curr_events will (eventually)
5303 * be larger than it was before. Once it becomes
5304 * substantially larger, the test below will cause
5305 * the array to appear non-idle, and resync will slow
5306 * down.
5307 * If there is a lot of outstanding resync activity when
5308 * we set last_event to curr_events, then all that activity
5309 * completing might cause the array to appear non-idle
5310 * and resync will be slowed down even though there might
5311 * not have been non-resync activity. This will only
5312 * happen once though. 'last_events' will soon reflect
5313 * the state where there is little or no outstanding
5314 * resync requests, and further resync activity will
5315 * always make curr_events less than last_events.
c0e48521 5316 *
1da177e4 5317 */
713f6ab1 5318 if (curr_events - rdev->last_events > 4096) {
1da177e4
LT		 5319 rdev->last_events = curr_events;
5320 idle = 0;
5321 }
5322 }
5323 return idle;
5324}
5325
5326void md_done_sync(mddev_t *mddev, int blocks, int ok)
5327{
5328 /* another "blocks" (512byte) blocks have been synced */
5329 atomic_sub(blocks, &mddev->recovery_active);
5330 wake_up(&mddev->recovery_wait);
5331 if (!ok) {
5332 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5333 md_wakeup_thread(mddev->thread);
5334 // stop recovery, signal do_sync ....
5335 }
5336}
5337
5338
06d91a5f
N		 5339/* md_write_start(mddev, bi)
5340 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 5341 * in superblock) before writing, schedule a superblock update
5342 * and wait for it to complete.
06d91a5f 5343 */
3d310eb7 5344void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 5345{
06d91a5f 5346 if (bio_data_dir(bi) != WRITE)
3d310eb7 5347 return;
06d91a5f 5348
f91de92e
N		 5349 BUG_ON(mddev->ro == 1);
5350 if (mddev->ro == 2) {
5351 /* need to switch to read/write */
5352 mddev->ro = 0;
5353 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5354 md_wakeup_thread(mddev->thread);
5355 }
06d91a5f 5356 atomic_inc(&mddev->writes_pending);
06d91a5f 5357 if (mddev->in_sync) {
a9701a30 5358 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 5359 if (mddev->in_sync) {
5360 mddev->in_sync = 0;
850b2b42 5361 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3d310eb7
N		 5362 md_wakeup_thread(mddev->thread);
5363 }
a9701a30 5364 spin_unlock_irq(&mddev->write_lock);
06d91a5f 5365 }
850b2b42 5366 wait_event(mddev->sb_wait, mddev->flags==0);
1da177e4
LT		 5367}
5368
5369void md_write_end(mddev_t *mddev)
5370{
5371 if (atomic_dec_and_test(&mddev->writes_pending)) {
5372 if (mddev->safemode == 2)
5373 md_wakeup_thread(mddev->thread);
16f17b39 5374 else if (mddev->safemode_delay)
1da177e4
LT		 5375 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5376 }
5377}
5378
2a2275d6
N		 5379/* md_allow_write(mddev)
5380 * Calling this ensures that the array is marked 'active' so that writes
5381 * may proceed without blocking. It is important to call this before
5382 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5383 * Must be called with mddev_lock held.
5384 */
5385void md_allow_write(mddev_t *mddev)
5386{
5387 if (!mddev->pers)
5388 return;
5389 if (mddev->ro)
5390 return;
5391
5392 spin_lock_irq(&mddev->write_lock);
5393 if (mddev->in_sync) {
5394 mddev->in_sync = 0;
5395 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5396 if (mddev->safemode_delay &&
5397 mddev->safemode == 0)
5398 mddev->safemode = 1;
5399 spin_unlock_irq(&mddev->write_lock);
5400 md_update_sb(mddev, 0);
5401 } else
5402 spin_unlock_irq(&mddev->write_lock);
5403}
5404EXPORT_SYMBOL_GPL(md_allow_write);
5405
75c96f85 5406static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 5407
5408#define SYNC_MARKS 10
5409#define SYNC_MARK_STEP (3*HZ)
29269553 5410void md_do_sync(mddev_t *mddev)
1da177e4
LT		 5411{
5412 mddev_t *mddev2;
5413 unsigned int currspeed = 0,
5414 window;
57afd89f 5415 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 5416 unsigned long mark[SYNC_MARKS];
5417 sector_t mark_cnt[SYNC_MARKS];
5418 int last_mark,m;
5419 struct list_head *tmp;
5420 sector_t last_check;
57afd89f 5421 int skipped = 0;
5fd6c1dc
N		 5422 struct list_head *rtmp;
5423 mdk_rdev_t *rdev;
61df9d91 5424 char *desc;
1da177e4
LT		 5425
5426 /* just incase thread restarts... */
5427 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5428 return;
5fd6c1dc
N		 5429 if (mddev->ro) /* never try to sync a read-only array */
5430 return;
1da177e4 5431
61df9d91
N		 5432 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5433 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5434 desc = "data-check";
5435 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5436 desc = "requested-resync";
5437 else
5438 desc = "resync";
5439 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5440 desc = "reshape";
5441 else
5442 desc = "recovery";
5443
1da177e4
LT		 5444 /* we overload curr_resync somewhat here.
5445 * 0 == not engaged in resync at all
5446 * 2 == checking that there is no conflict with another sync
5447 * 1 == like 2, but have yielded to allow conflicting resync to
5448 * commense
5449 * other == active in resync - this many blocks
5450 *
5451 * Before starting a resync we must have set curr_resync to
5452 * 2, and then checked that every "conflicting" array has curr_resync
5453 * less than ours. When we find one that is the same or higher
5454 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5455 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5456 * This will mean we have to start checking from the beginning again.
5457 *
5458 */
5459
5460 do {
5461 mddev->curr_resync = 2;
5462
5463 try_again:
787453c2 5464 if (kthread_should_stop()) {
6985c43f 5465 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 5466 goto skip;
5467 }
5468 ITERATE_MDDEV(mddev2,tmp) {
1da177e4
LT		 5469 if (mddev2 == mddev)
5470 continue;
5471 if (mddev2->curr_resync &&
5472 match_mddev_units(mddev,mddev2)) {
5473 DEFINE_WAIT(wq);
5474 if (mddev < mddev2 && mddev->curr_resync == 2) {
5475 /* arbitrarily yield */
5476 mddev->curr_resync = 1;
5477 wake_up(&resync_wait);
5478 }
5479 if (mddev > mddev2 && mddev->curr_resync == 1)
5480 /* no need to wait here, we can wait the next
5481 * time 'round when curr_resync == 2
5482 */
5483 continue;
787453c2
N		 5484 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5485 if (!kthread_should_stop() &&
8712e553 5486 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 5487 printk(KERN_INFO "md: delaying %s of %s"
5488 " until %s has finished (they"
1da177e4 5489 " share one or more physical units)\n",
61df9d91 5490 desc, mdname(mddev), mdname(mddev2));
1da177e4
LT		 5491 mddev_put(mddev2);
5492 schedule();
5493 finish_wait(&resync_wait, &wq);
5494 goto try_again;
5495 }
5496 finish_wait(&resync_wait, &wq);
5497 }
5498 }
5499 } while (mddev->curr_resync < 2);
5500
5fd6c1dc 5501 j = 0;
9d88883e 5502 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 5503 /* resync follows the size requested by the personality,
57afd89f 5504 * which defaults to physical size, but can be virtual size
1da177e4
LT		 5505 */
5506 max_sectors = mddev->resync_max_sectors;
9d88883e 5507 mddev->resync_mismatches = 0;
5fd6c1dc
N		 5508 /* we don't use the checkpoint if there's a bitmap */
5509 if (!mddev->bitmap &&
5510 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5511 j = mddev->recovery_cp;
ccfcc3c1
N		 5512 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5513 max_sectors = mddev->size << 1;
5fd6c1dc 5514 else {
1da177e4
LT		 5515 /* recovery follows the physical size of devices */
5516 max_sectors = mddev->size << 1;
5fd6c1dc
N		 5517 j = MaxSector;
5518 ITERATE_RDEV(mddev,rdev,rtmp)
5519 if (rdev->raid_disk >= 0 &&
5520 !test_bit(Faulty, &rdev->flags) &&
5521 !test_bit(In_sync, &rdev->flags) &&
5522 rdev->recovery_offset < j)
5523 j = rdev->recovery_offset;
5524 }
1da177e4 5525
61df9d91
N		 5526 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5527 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5528 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 5529 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 5530 "(but not more than %d KB/sec) for %s.\n",
5531 speed_max(mddev), desc);
1da177e4
LT		 5532
5533 is_mddev_idle(mddev); /* this also initializes IO event counters */
5fd6c1dc 5534
57afd89f 5535 io_sectors = 0;
1da177e4
LT		 5536 for (m = 0; m < SYNC_MARKS; m++) {
5537 mark[m] = jiffies;
57afd89f 5538 mark_cnt[m] = io_sectors;
1da177e4
LT		 5539 }
5540 last_mark = 0;
5541 mddev->resync_mark = mark[last_mark];
5542 mddev->resync_mark_cnt = mark_cnt[last_mark];
5543
5544 /*
5545 * Tune reconstruction:
5546 */
5547 window = 32*(PAGE_SIZE/512);
5548 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5549 window/2,(unsigned long long) max_sectors/2);
5550
5551 atomic_set(&mddev->recovery_active, 0);
5552 init_waitqueue_head(&mddev->recovery_wait);
5553 last_check = 0;
5554
5555 if (j>2) {
5556 printk(KERN_INFO
61df9d91
N		 5557 "md: resuming %s of %s from checkpoint.\n",
5558 desc, mdname(mddev));
1da177e4
LT		 5559 mddev->curr_resync = j;
5560 }
5561
5562 while (j < max_sectors) {
57afd89f 5563 sector_t sectors;
1da177e4 5564
57afd89f 5565 skipped = 0;
c6207277
N		 5566 if (j >= mddev->resync_max) {
5567 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5568 wait_event(mddev->recovery_wait,
5569 mddev->resync_max > j
5570 || kthread_should_stop());
5571 }
5572 if (kthread_should_stop())
5573 goto interrupted;
57afd89f 5574 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 5575 currspeed < speed_min(mddev));
57afd89f 5576 if (sectors == 0) {
1da177e4
LT		 5577 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5578 goto out;
5579 }
57afd89f
N		 5580
5581 if (!skipped) { /* actual IO requested */
5582 io_sectors += sectors;
5583 atomic_add(sectors, &mddev->recovery_active);
5584 }
5585
1da177e4
LT		 5586 j += sectors;
5587 if (j>1) mddev->curr_resync = j;
ff4e8d9a 5588 mddev->curr_mark_cnt = io_sectors;
d7603b7e
N		 5589 if (last_check == 0)
5590 /* this is the earliers that rebuilt will be
5591 * visible in /proc/mdstat
5592 */
5593 md_new_event(mddev);
57afd89f
N		 5594
5595 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 5596 continue;
5597
57afd89f 5598 last_check = io_sectors;
1da177e4
LT		 5599
5600 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5601 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5602 break;
5603
5604 repeat:
5605 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5606 /* step marks */
5607 int next = (last_mark+1) % SYNC_MARKS;
5608
5609 mddev->resync_mark = mark[next];
5610 mddev->resync_mark_cnt = mark_cnt[next];
5611 mark[next] = jiffies;
57afd89f 5612 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 5613 last_mark = next;
5614 }
5615
5616
c6207277
N		 5617 if (kthread_should_stop())
5618 goto interrupted;
5619
1da177e4
LT		 5620
5621 /*
5622 * this loop exits only if either when we are slower than
5623 * the 'hard' speed limit, or the system was IO-idle for
5624 * a jiffy.
5625 * the system might be non-idle CPU-wise, but we only care
5626 * about not overloading the IO subsystem. (things like an
5627 * e2fsck being done on the RAID array should execute fast)
5628 */
2ad8b1ef 5629 blk_unplug(mddev->queue);
1da177e4
LT		 5630 cond_resched();
5631
57afd89f
N		 5632 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5633 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 5634
88202a0c
N		 5635 if (currspeed > speed_min(mddev)) {
5636 if ((currspeed > speed_max(mddev)) ||
1da177e4 5637 !is_mddev_idle(mddev)) {
c0e48521 5638 msleep(500);
1da177e4
LT		 5639 goto repeat;
5640 }
5641 }
5642 }
61df9d91 5643 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 5644 /*
5645 * this also signals 'finished resyncing' to md_stop
5646 */
5647 out:
2ad8b1ef 5648 blk_unplug(mddev->queue);
1da177e4
LT		 5649
5650 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5651
5652 /* tell personality that we are finished */
57afd89f 5653 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 5654
5655 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
ccfcc3c1 5656 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 5657 mddev->curr_resync > 2) {
5658 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5659 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5660 if (mddev->curr_resync >= mddev->recovery_cp) {
5661 printk(KERN_INFO
61df9d91
N		 5662 "md: checkpointing %s of %s.\n",
5663 desc, mdname(mddev));
5fd6c1dc
N		 5664 mddev->recovery_cp = mddev->curr_resync;
5665 }
5666 } else
5667 mddev->recovery_cp = MaxSector;
5668 } else {
5669 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5670 mddev->curr_resync = MaxSector;
5671 ITERATE_RDEV(mddev,rdev,rtmp)
5672 if (rdev->raid_disk >= 0 &&
5673 !test_bit(Faulty, &rdev->flags) &&
5674 !test_bit(In_sync, &rdev->flags) &&
5675 rdev->recovery_offset < mddev->curr_resync)
5676 rdev->recovery_offset = mddev->curr_resync;
5fd6c1dc 5677 }
1da177e4 5678 }
17571284 5679 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5680
1da177e4
LT		 5681 skip:
5682 mddev->curr_resync = 0;
c6207277
N		 5683 mddev->resync_max = MaxSector;
5684 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
1da177e4
LT		 5685 wake_up(&resync_wait);
5686 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5687 md_wakeup_thread(mddev->thread);
c6207277
N		 5688 return;
5689
5690 interrupted:
5691 /*
5692 * got a signal, exit.
5693 */
5694 printk(KERN_INFO
5695 "md: md_do_sync() got signal ... exiting\n");
5696 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5697 goto out;
5698
1da177e4 5699}
29269553 5700EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 5701
5702
b4c4c7b8
N		 5703static int remove_and_add_spares(mddev_t *mddev)
5704{
5705 mdk_rdev_t *rdev;
5706 struct list_head *rtmp;
5707 int spares = 0;
5708
5709 ITERATE_RDEV(mddev,rdev,rtmp)
5710 if (rdev->raid_disk >= 0 &&
c303da6d 5711 !mddev->external &&
b4c4c7b8
N		 5712 (test_bit(Faulty, &rdev->flags) ||
5713 ! test_bit(In_sync, &rdev->flags)) &&
5714 atomic_read(&rdev->nr_pending)==0) {
5715 if (mddev->pers->hot_remove_disk(
5716 mddev, rdev->raid_disk)==0) {
5717 char nm[20];
5718 sprintf(nm,"rd%d", rdev->raid_disk);
5719 sysfs_remove_link(&mddev->kobj, nm);
5720 rdev->raid_disk = -1;
5721 }
5722 }
5723
5724 if (mddev->degraded) {
5725 ITERATE_RDEV(mddev,rdev,rtmp)
5726 if (rdev->raid_disk < 0
5727 && !test_bit(Faulty, &rdev->flags)) {
5728 rdev->recovery_offset = 0;
5729 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5730 char nm[20];
5731 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 5732 if (sysfs_create_link(&mddev->kobj,
5733 &rdev->kobj, nm))
5734 printk(KERN_WARNING
5735 "md: cannot register "
5736 "%s for %s\n",
5737 nm, mdname(mddev));
b4c4c7b8
N		 5738 spares++;
5739 md_new_event(mddev);
5740 } else
5741 break;
5742 }
5743 }
5744 return spares;
5745}
1da177e4
LT		 5746/*
5747 * This routine is regularly called by all per-raid-array threads to
5748 * deal with generic issues like resync and super-block update.
5749 * Raid personalities that don't have a thread (linear/raid0) do not
5750 * need this as they never do any recovery or update the superblock.
5751 *
5752 * It does not do any resync itself, but rather "forks" off other threads
5753 * to do that as needed.
5754 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5755 * "->recovery" and create a thread at ->sync_thread.
5756 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5757 * and wakeups up this thread which will reap the thread and finish up.
5758 * This thread also removes any faulty devices (with nr_pending == 0).
5759 *
5760 * The overall approach is:
5761 * 1/ if the superblock needs updating, update it.
5762 * 2/ If a recovery thread is running, don't do anything else.
5763 * 3/ If recovery has finished, clean up, possibly marking spares active.
5764 * 4/ If there are any faulty devices, remove them.
5765 * 5/ If array is degraded, try to add spares devices
5766 * 6/ If array has spares or is not in-sync, start a resync thread.
5767 */
5768void md_check_recovery(mddev_t *mddev)
5769{
5770 mdk_rdev_t *rdev;
5771 struct list_head *rtmp;
5772
5773
5f40402d
N		 5774 if (mddev->bitmap)
5775 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 5776
5777 if (mddev->ro)
5778 return;
fca4d848
N		 5779
5780 if (signal_pending(current)) {
5781 if (mddev->pers->sync_request) {
5782 printk(KERN_INFO "md: %s in immediate safe mode\n",
5783 mdname(mddev));
5784 mddev->safemode = 2;
5785 }
5786 flush_signals(current);
5787 }
5788
1da177e4 5789 if ( ! (
e691063a 5790 (mddev->flags && !mddev->external) ||
1da177e4 5791 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 5792 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5793 (mddev->safemode == 1) ||
5794 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5795 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 5796 ))
5797 return;
fca4d848 5798
df5b89b3 5799 if (mddev_trylock(mddev)) {
b4c4c7b8 5800 int spares = 0;
fca4d848 5801
a9701a30 5802 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 5803 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5804 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5805 mddev->in_sync = 1;
e691063a
N		 5806 if (mddev->persistent)
5807 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
fca4d848
N		 5808 }
5809 if (mddev->safemode == 1)
5810 mddev->safemode = 0;
a9701a30 5811 spin_unlock_irq(&mddev->write_lock);
fca4d848 5812
850b2b42
N		 5813 if (mddev->flags)
5814 md_update_sb(mddev, 0);
06d91a5f 5815
06d91a5f 5816
1da177e4
LT		 5817 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5818 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5819 /* resync/recovery still happening */
5820 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5821 goto unlock;
5822 }
5823 if (mddev->sync_thread) {
5824 /* resync has finished, collect result */
5825 md_unregister_thread(mddev->sync_thread);
5826 mddev->sync_thread = NULL;
5827 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5828 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5829 /* success...*/
5830 /* activate any spares */
5831 mddev->pers->spare_active(mddev);
5832 }
850b2b42 5833 md_update_sb(mddev, 1);
41158c7e
N		 5834
5835 /* if array is no-longer degraded, then any saved_raid_disk
5836 * information must be scrapped
5837 */
5838 if (!mddev->degraded)
5839 ITERATE_RDEV(mddev,rdev,rtmp)
5840 rdev->saved_raid_disk = -1;
5841
1da177e4
LT		 5842 mddev->recovery = 0;
5843 /* flag recovery needed just to double check */
5844 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d7603b7e 5845 md_new_event(mddev);
1da177e4
LT		 5846 goto unlock;
5847 }
24dd469d
N		 5848 /* Clear some bits that don't mean anything, but
5849 * might be left set
5850 */
5851 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5852 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5853 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5854 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 5855
5fd6c1dc
N		 5856 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5857 goto unlock;
1da177e4
LT		 5858 /* no recovery is running.
5859 * remove any failed drives, then
5860 * add spares if possible.
5861 * Spare are also removed and re-added, to allow
5862 * the personality to fail the re-add.
5863 */
1da177e4 5864
b4c4c7b8
N		 5865 if (mddev->reshape_position != MaxSector) {
5866 if (mddev->pers->check_reshape(mddev) != 0)
5867 /* Cannot proceed */
5868 goto unlock;
5869 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5870 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 5871 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5872 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5873 } else if (mddev->recovery_cp < MaxSector) {
5874 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5875 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5876 /* nothing to be done ... */
1da177e4 5877 goto unlock;
24dd469d 5878
1da177e4
LT		 5879 if (mddev->pers->sync_request) {
5880 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 5881 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5882 /* We are adding a device or devices to an array
5883 * which has the bitmap stored on all devices.
5884 * So make sure all bitmap pages get written
5885 */
5886 bitmap_write_all(mddev->bitmap);
5887 }
1da177e4
LT		 5888 mddev->sync_thread = md_register_thread(md_do_sync,
5889 mddev,
5890 "%s_resync");
5891 if (!mddev->sync_thread) {
5892 printk(KERN_ERR "%s: could not start resync"
5893 " thread...\n",
5894 mdname(mddev));
5895 /* leave the spares where they are, it shouldn't hurt */
5896 mddev->recovery = 0;
d7603b7e 5897 } else
1da177e4 5898 md_wakeup_thread(mddev->sync_thread);
d7603b7e 5899 md_new_event(mddev);
1da177e4
LT		 5900 }
5901 unlock:
5902 mddev_unlock(mddev);
5903 }
5904}
5905
75c96f85
AB		 5906static int md_notify_reboot(struct notifier_block *this,
5907 unsigned long code, void *x)
1da177e4
LT		 5908{
5909 struct list_head *tmp;
5910 mddev_t *mddev;
5911
5912 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5913
5914 printk(KERN_INFO "md: stopping all md devices.\n");
5915
5916 ITERATE_MDDEV(mddev,tmp)
c71d4887 5917 if (mddev_trylock(mddev)) {
1da177e4 5918 do_md_stop (mddev, 1);
c71d4887
NB		 5919 mddev_unlock(mddev);
5920 }
1da177e4
LT		 5921 /*
5922 * certain more exotic SCSI devices are known to be
5923 * volatile wrt too early system reboots. While the
5924 * right place to handle this issue is the given
5925 * driver, we do want to have a safe RAID driver ...
5926 */
5927 mdelay(1000*1);
5928 }
5929 return NOTIFY_DONE;
5930}
5931
75c96f85 5932static struct notifier_block md_notifier = {
1da177e4
LT		 5933 .notifier_call = md_notify_reboot,
5934 .next = NULL,
5935 .priority = INT_MAX, /* before any real devices */
5936};
5937
5938static void md_geninit(void)
5939{
5940 struct proc_dir_entry *p;
5941
5942 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5943
5944 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5945 if (p)
5946 p->proc_fops = &md_seq_fops;
5947}
5948
75c96f85 5949static int __init md_init(void)
1da177e4 5950{
1da177e4
LT		 5951 if (register_blkdev(MAJOR_NR, "md"))
5952 return -1;
5953 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5954 unregister_blkdev(MAJOR_NR, "md");
5955 return -1;
5956 }
e8703fe1
N		 5957 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5958 md_probe, NULL, NULL);
5959 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 5960 md_probe, NULL, NULL);
5961
1da177e4 5962 register_reboot_notifier(&md_notifier);
0b4d4147 5963 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 5964
5965 md_geninit();
5966 return (0);
5967}
5968
5969
5970#ifndef MODULE
5971
5972/*
5973 * Searches all registered partitions for autorun RAID arrays
5974 * at boot time.
5975 */
4d936ec1
ME		 5976
5977static LIST_HEAD(all_detected_devices);
5978struct detected_devices_node {
5979 struct list_head list;
5980 dev_t dev;
5981};
1da177e4
LT		 5982
5983void md_autodetect_dev(dev_t dev)
5984{
4d936ec1
ME		 5985 struct detected_devices_node *node_detected_dev;
5986
5987 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
5988 if (node_detected_dev) {
5989 node_detected_dev->dev = dev;
5990 list_add_tail(&node_detected_dev->list, &all_detected_devices);
5991 } else {
5992 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
5993 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
5994 }
1da177e4
LT		 5995}
5996
5997
5998static void autostart_arrays(int part)
5999{
6000 mdk_rdev_t *rdev;
4d936ec1
ME		 6001 struct detected_devices_node *node_detected_dev;
6002 dev_t dev;
6003 int i_scanned, i_passed;
1da177e4 6004
4d936ec1
ME		 6005 i_scanned = 0;
6006 i_passed = 0;
1da177e4 6007
4d936ec1 6008 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 6009
4d936ec1
ME		 6010 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6011 i_scanned++;
6012 node_detected_dev = list_entry(all_detected_devices.next,
6013 struct detected_devices_node, list);
6014 list_del(&node_detected_dev->list);
6015 dev = node_detected_dev->dev;
6016 kfree(node_detected_dev);
df968c4e 6017 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 6018 if (IS_ERR(rdev))
6019 continue;
6020
b2d444d7 6021 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6022 MD_BUG();
6023 continue;
6024 }
6025 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 6026 i_passed++;
1da177e4 6027 }
4d936ec1
ME		 6028
6029 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6030 i_scanned, i_passed);
1da177e4
LT		 6031
6032 autorun_devices(part);
6033}
6034
fdee8ae4 6035#endif /* !MODULE */
1da177e4
LT		 6036
6037static __exit void md_exit(void)
6038{
6039 mddev_t *mddev;
6040 struct list_head *tmp;
8ab5e4c1 6041
e8703fe1
N		 6042 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6043 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4
LT		 6044
6045 unregister_blkdev(MAJOR_NR,"md");
6046 unregister_blkdev(mdp_major, "mdp");
6047 unregister_reboot_notifier(&md_notifier);
6048 unregister_sysctl_table(raid_table_header);
6049 remove_proc_entry("mdstat", NULL);
6050 ITERATE_MDDEV(mddev,tmp) {
6051 struct gendisk *disk = mddev->gendisk;
6052 if (!disk)
6053 continue;
6054 export_array(mddev);
6055 del_gendisk(disk);
6056 put_disk(disk);
6057 mddev->gendisk = NULL;
6058 mddev_put(mddev);
6059 }
6060}
6061
685784aa 6062subsys_initcall(md_init);
1da177e4
LT		 6063module_exit(md_exit)
6064
f91de92e
N		 6065static int get_ro(char *buffer, struct kernel_param *kp)
6066{
6067 return sprintf(buffer, "%d", start_readonly);
6068}
6069static int set_ro(const char *val, struct kernel_param *kp)
6070{
6071 char *e;
6072 int num = simple_strtoul(val, &e, 10);
6073 if (*val && (*e == '\0' || *e == '\n')) {
6074 start_readonly = num;
4dbcdc75 6075 return 0;
f91de92e
N		 6076 }
6077 return -EINVAL;
6078}
6079
80ca3a44
N		 6080module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6081module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 6082
f91de92e 6083
1da177e4
LT		 6084EXPORT_SYMBOL(register_md_personality);
6085EXPORT_SYMBOL(unregister_md_personality);
6086EXPORT_SYMBOL(md_error);
6087EXPORT_SYMBOL(md_done_sync);
6088EXPORT_SYMBOL(md_write_start);
6089EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 6090EXPORT_SYMBOL(md_register_thread);
6091EXPORT_SYMBOL(md_unregister_thread);
6092EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 6093EXPORT_SYMBOL(md_check_recovery);
6094MODULE_LICENSE("GPL");
aa1595e9 6095MODULE_ALIAS("md");
72008652 6096MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)