projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
video: replace remaining __FUNCTION__ occurrences
[deliverable/linux.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 */
29ac4aa3 198#define for_each_mddev(mddev,tmp) \
1da177e4
LT		 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
d089c6af 314 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 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
d089c6af 326 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 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 945 sb->disks[0].state = (1<<MD_DISK_REMOVED);
d089c6af 946 rdev_for_each(rdev2, tmp, mddev) {
1da177e4 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 1107 if (rdev->sb_size & bmask)
a1801f85
N		 1108 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1109
1110 if (minor_version
1111 && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1112 return -EINVAL;
0002b271 1113
31b65a0d
N		 1114 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1115 rdev->desc_nr = -1;
1116 else
1117 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1118
1da177e4 1119 if (refdev == 0)
8ed75463 1120 ret = 1;
1da177e4
LT		 1121 else {
1122 __u64 ev1, ev2;
1123 struct mdp_superblock_1 *refsb =
1124 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1125
1126 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1127 sb->level != refsb->level ||
1128 sb->layout != refsb->layout ||
1129 sb->chunksize != refsb->chunksize) {
1130 printk(KERN_WARNING "md: %s has strangely different"
1131 " superblock to %s\n",
1132 bdevname(rdev->bdev,b),
1133 bdevname(refdev->bdev,b2));
1134 return -EINVAL;
1135 }
1136 ev1 = le64_to_cpu(sb->events);
1137 ev2 = le64_to_cpu(refsb->events);
1138
1139 if (ev1 > ev2)
8ed75463
N		 1140 ret = 1;
1141 else
1142 ret = 0;
1da177e4 1143 }
a1801f85 1144 if (minor_version)
1da177e4
LT		 1145 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1146 else
1147 rdev->size = rdev->sb_offset;
1148 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1149 return -EINVAL;
1150 rdev->size = le64_to_cpu(sb->data_size)/2;
1151 if (le32_to_cpu(sb->chunksize))
1152 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
2bf071bf 1153
1c05b4bc 1154 if (le64_to_cpu(sb->size) > rdev->size*2)
2bf071bf 1155 return -EINVAL;
8ed75463 1156 return ret;
1da177e4
LT		 1157}
1158
1159static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1160{
1161 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
07d84d10 1162 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1163
41158c7e 1164 rdev->raid_disk = -1;
c5d79adb
N		 1165 clear_bit(Faulty, &rdev->flags);
1166 clear_bit(In_sync, &rdev->flags);
1167 clear_bit(WriteMostly, &rdev->flags);
1168 clear_bit(BarriersNotsupp, &rdev->flags);
1169
1da177e4
LT		 1170 if (mddev->raid_disks == 0) {
1171 mddev->major_version = 1;
1172 mddev->patch_version = 0;
e691063a 1173 mddev->external = 0;
1da177e4
LT		 1174 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1175 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1176 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1177 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1178 mddev->clevel[0] = 0;
1da177e4
LT		 1179 mddev->layout = le32_to_cpu(sb->layout);
1180 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1181 mddev->size = le64_to_cpu(sb->size)/2;
07d84d10 1182 mddev->events = ev1;
9223214e 1183 mddev->bitmap_offset = 0;
29fc7e3e 1184 mddev->default_bitmap_offset = 1024 >> 9;
1da177e4
LT		 1185
1186 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1187 memcpy(mddev->uuid, sb->set_uuid, 16);
1188
1189 mddev->max_disks = (4096-256)/2;
a654b9d8 1190
71c0805c 1191 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
e11e93fa 1192 mddev->bitmap_file == NULL )
a654b9d8 1193 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1194
f6705578
N		 1195 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1196 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1197 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1198 mddev->new_level = le32_to_cpu(sb->new_level);
1199 mddev->new_layout = le32_to_cpu(sb->new_layout);
1200 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1201 } else {
1202 mddev->reshape_position = MaxSector;
1203 mddev->delta_disks = 0;
1204 mddev->new_level = mddev->level;
1205 mddev->new_layout = mddev->layout;
1206 mddev->new_chunk = mddev->chunk_size;
1207 }
1208
41158c7e
N		 1209 } else if (mddev->pers == NULL) {
1210 /* Insist of good event counter while assembling */
1da177e4
LT		 1211 ++ev1;
1212 if (ev1 < mddev->events)
1213 return -EINVAL;
41158c7e
N		 1214 } else if (mddev->bitmap) {
1215 /* If adding to array with a bitmap, then we can accept an
1216 * older device, but not too old.
1217 */
41158c7e
N		 1218 if (ev1 < mddev->bitmap->events_cleared)
1219 return 0;
07d84d10
N		 1220 } else {
1221 if (ev1 < mddev->events)
1222 /* just a hot-add of a new device, leave raid_disk at -1 */
1223 return 0;
1224 }
1da177e4
LT		 1225 if (mddev->level != LEVEL_MULTIPATH) {
1226 int role;
1da177e4
LT		 1227 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1228 switch(role) {
1229 case 0xffff: /* spare */
1da177e4
LT		 1230 break;
1231 case 0xfffe: /* faulty */
b2d444d7 1232 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1233 break;
1234 default:
5fd6c1dc
N		 1235 if ((le32_to_cpu(sb->feature_map) &
1236 MD_FEATURE_RECOVERY_OFFSET))
1237 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1238 else
1239 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1240 rdev->raid_disk = role;
1241 break;
1242 }
8ddf9efe
N		 1243 if (sb->devflags & WriteMostly1)
1244 set_bit(WriteMostly, &rdev->flags);
41158c7e 1245 } else /* MULTIPATH are always insync */
b2d444d7 1246 set_bit(In_sync, &rdev->flags);
41158c7e 1247
1da177e4
LT		 1248 return 0;
1249}
1250
1251static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1252{
1253 struct mdp_superblock_1 *sb;
1254 struct list_head *tmp;
1255 mdk_rdev_t *rdev2;
1256 int max_dev, i;
1257 /* make rdev->sb match mddev and rdev data. */
1258
1259 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1260
1261 sb->feature_map = 0;
1262 sb->pad0 = 0;
5fd6c1dc 1263 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1264 memset(sb->pad1, 0, sizeof(sb->pad1));
1265 memset(sb->pad2, 0, sizeof(sb->pad2));
1266 memset(sb->pad3, 0, sizeof(sb->pad3));
1267
1268 sb->utime = cpu_to_le64((__u64)mddev->utime);
1269 sb->events = cpu_to_le64(mddev->events);
1270 if (mddev->in_sync)
1271 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1272 else
1273 sb->resync_offset = cpu_to_le64(0);
1274
1c05b4bc 1275 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1276
f0ca340c 1277 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
29fc7e3e 1278 sb->size = cpu_to_le64(mddev->size<<1);
f0ca340c 1279
a654b9d8
N		 1280 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1281 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1282 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1283 }
5fd6c1dc
N		 1284
1285 if (rdev->raid_disk >= 0 &&
1286 !test_bit(In_sync, &rdev->flags) &&
1287 rdev->recovery_offset > 0) {
1288 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1289 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1290 }
1291
f6705578
N		 1292 if (mddev->reshape_position != MaxSector) {
1293 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1294 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1295 sb->new_layout = cpu_to_le32(mddev->new_layout);
1296 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1297 sb->new_level = cpu_to_le32(mddev->new_level);
1298 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1299 }
a654b9d8 1300
1da177e4 1301 max_dev = 0;
d089c6af 1302 rdev_for_each(rdev2, tmp, mddev)
1da177e4
LT		 1303 if (rdev2->desc_nr+1 > max_dev)
1304 max_dev = rdev2->desc_nr+1;
a778b73f
N		 1305
1306 if (max_dev > le32_to_cpu(sb->max_dev))
1307 sb->max_dev = cpu_to_le32(max_dev);
1da177e4
LT		 1308 for (i=0; i<max_dev;i++)
1309 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1310
d089c6af 1311 rdev_for_each(rdev2, tmp, mddev) {
1da177e4 1312 i = rdev2->desc_nr;
b2d444d7 1313 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1314 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1315 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1316 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
5fd6c1dc
N		 1317 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1318 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1319 else
1320 sb->dev_roles[i] = cpu_to_le16(0xffff);
1321 }
1322
1da177e4
LT		 1323 sb->sb_csum = calc_sb_1_csum(sb);
1324}
1325
1326
75c96f85 1327static struct super_type super_types[] = {
1da177e4
LT		 1328 [0] = {
1329 .name = "0.90.0",
1330 .owner = THIS_MODULE,
1331 .load_super = super_90_load,
1332 .validate_super = super_90_validate,
1333 .sync_super = super_90_sync,
1334 },
1335 [1] = {
1336 .name = "md-1",
1337 .owner = THIS_MODULE,
1338 .load_super = super_1_load,
1339 .validate_super = super_1_validate,
1340 .sync_super = super_1_sync,
1341 },
1342};
1da177e4
LT		 1343
1344static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1345{
7dd5e7c3
N		 1346 struct list_head *tmp, *tmp2;
1347 mdk_rdev_t *rdev, *rdev2;
1da177e4 1348
d089c6af
N		 1349 rdev_for_each(rdev, tmp, mddev1)
1350 rdev_for_each(rdev2, tmp2, mddev2)
7dd5e7c3
N		 1351 if (rdev->bdev->bd_contains ==
1352 rdev2->bdev->bd_contains)
1353 return 1;
1da177e4
LT		 1354
1355 return 0;
1356}
1357
1358static LIST_HEAD(pending_raid_disks);
1359
1360static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1361{
7dd5e7c3 1362 char b[BDEVNAME_SIZE];
f637b9f9 1363 struct kobject *ko;
1edf80d3 1364 char *s;
5e55e2f5 1365 int err;
1da177e4
LT		 1366
1367 if (rdev->mddev) {
1368 MD_BUG();
1369 return -EINVAL;
1370 }
2bf071bf
N		 1371 /* make sure rdev->size exceeds mddev->size */
1372 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
a778b73f
N		 1373 if (mddev->pers) {
1374 /* Cannot change size, so fail
1375 * If mddev->level <= 0, then we don't care
1376 * about aligning sizes (e.g. linear)
1377 */
1378 if (mddev->level > 0)
1379 return -ENOSPC;
1380 } else
2bf071bf
N		 1381 mddev->size = rdev->size;
1382 }
1da177e4
LT		 1383
1384 /* Verify rdev->desc_nr is unique.
1385 * If it is -1, assign a free number, else
1386 * check number is not in use
1387 */
1388 if (rdev->desc_nr < 0) {
1389 int choice = 0;
1390 if (mddev->pers) choice = mddev->raid_disks;
1391 while (find_rdev_nr(mddev, choice))
1392 choice++;
1393 rdev->desc_nr = choice;
1394 } else {
1395 if (find_rdev_nr(mddev, rdev->desc_nr))
1396 return -EBUSY;
1397 }
19133a42 1398 bdevname(rdev->bdev,b);
649316b2 1399 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 1400 *s = '!';
649316b2 1401
1da177e4 1402 rdev->mddev = mddev;
19133a42 1403 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1404
b2d6db58 1405 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 1406 goto fail;
86e6ffdd 1407
f637b9f9 1408 if (rdev->bdev->bd_part)
edfaa7c3 1409 ko = &rdev->bdev->bd_part->dev.kobj;
f637b9f9 1410 else
edfaa7c3 1411 ko = &rdev->bdev->bd_disk->dev.kobj;
5e55e2f5
N		 1412 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1413 kobject_del(&rdev->kobj);
1414 goto fail;
1415 }
1416 list_add(&rdev->same_set, &mddev->disks);
c5d79adb 1417 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1da177e4 1418 return 0;
5e55e2f5
N		 1419
1420 fail:
1421 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1422 b, mdname(mddev));
1423 return err;
1da177e4
LT		 1424}
1425
177a99b2 1426static void md_delayed_delete(struct work_struct *ws)
5792a285
N		 1427{
1428 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1429 kobject_del(&rdev->kobj);
177a99b2 1430 kobject_put(&rdev->kobj);
5792a285
N		 1431}
1432
1da177e4
LT		 1433static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1434{
1435 char b[BDEVNAME_SIZE];
1436 if (!rdev->mddev) {
1437 MD_BUG();
1438 return;
1439 }
5463c790 1440 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1da177e4
LT		 1441 list_del_init(&rdev->same_set);
1442 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1443 rdev->mddev = NULL;
86e6ffdd 1444 sysfs_remove_link(&rdev->kobj, "block");
5792a285
N		 1445
1446 /* We need to delay this, otherwise we can deadlock when
1447 * writing to 'remove' to "dev/state"
1448 */
177a99b2
N		 1449 INIT_WORK(&rdev->del_work, md_delayed_delete);
1450 kobject_get(&rdev->kobj);
5792a285 1451 schedule_work(&rdev->del_work);
1da177e4
LT		 1452}
1453
1454/*
1455 * prevent the device from being mounted, repartitioned or
1456 * otherwise reused by a RAID array (or any other kernel
1457 * subsystem), by bd_claiming the device.
1458 */
c5d79adb 1459static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1da177e4
LT		 1460{
1461 int err = 0;
1462 struct block_device *bdev;
1463 char b[BDEVNAME_SIZE];
1464
2e7b651d 1465 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1466 if (IS_ERR(bdev)) {
1467 printk(KERN_ERR "md: could not open %s.\n",
1468 __bdevname(dev, b));
1469 return PTR_ERR(bdev);
1470 }
c5d79adb 1471 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1da177e4
LT		 1472 if (err) {
1473 printk(KERN_ERR "md: could not bd_claim %s.\n",
1474 bdevname(bdev, b));
2e7b651d 1475 blkdev_put(bdev);
1da177e4
LT		 1476 return err;
1477 }
c5d79adb
N		 1478 if (!shared)
1479 set_bit(AllReserved, &rdev->flags);
1da177e4
LT		 1480 rdev->bdev = bdev;
1481 return err;
1482}
1483
1484static void unlock_rdev(mdk_rdev_t *rdev)
1485{
1486 struct block_device *bdev = rdev->bdev;
1487 rdev->bdev = NULL;
1488 if (!bdev)
1489 MD_BUG();
1490 bd_release(bdev);
2e7b651d 1491 blkdev_put(bdev);
1da177e4
LT		 1492}
1493
1494void md_autodetect_dev(dev_t dev);
1495
1496static void export_rdev(mdk_rdev_t * rdev)
1497{
1498 char b[BDEVNAME_SIZE];
1499 printk(KERN_INFO "md: export_rdev(%s)\n",
1500 bdevname(rdev->bdev,b));
1501 if (rdev->mddev)
1502 MD_BUG();
1503 free_disk_sb(rdev);
1504 list_del_init(&rdev->same_set);
1505#ifndef MODULE
d0fae18f
N		 1506 if (test_bit(AutoDetected, &rdev->flags))
1507 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 1508#endif
1509 unlock_rdev(rdev);
86e6ffdd 1510 kobject_put(&rdev->kobj);
1da177e4
LT		 1511}
1512
1513static void kick_rdev_from_array(mdk_rdev_t * rdev)
1514{
1515 unbind_rdev_from_array(rdev);
1516 export_rdev(rdev);
1517}
1518
1519static void export_array(mddev_t *mddev)
1520{
1521 struct list_head *tmp;
1522 mdk_rdev_t *rdev;
1523
d089c6af 1524 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 1525 if (!rdev->mddev) {
1526 MD_BUG();
1527 continue;
1528 }
1529 kick_rdev_from_array(rdev);
1530 }
1531 if (!list_empty(&mddev->disks))
1532 MD_BUG();
1533 mddev->raid_disks = 0;
1534 mddev->major_version = 0;
1535}
1536
1537static void print_desc(mdp_disk_t *desc)
1538{
1539 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1540 desc->major,desc->minor,desc->raid_disk,desc->state);
1541}
1542
1543static void print_sb(mdp_super_t *sb)
1544{
1545 int i;
1546
1547 printk(KERN_INFO
1548 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1549 sb->major_version, sb->minor_version, sb->patch_version,
1550 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1551 sb->ctime);
1552 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1553 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1554 sb->md_minor, sb->layout, sb->chunk_size);
1555 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1556 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1557 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1558 sb->failed_disks, sb->spare_disks,
1559 sb->sb_csum, (unsigned long)sb->events_lo);
1560
1561 printk(KERN_INFO);
1562 for (i = 0; i < MD_SB_DISKS; i++) {
1563 mdp_disk_t *desc;
1564
1565 desc = sb->disks + i;
1566 if (desc->number || desc->major || desc->minor ||
1567 desc->raid_disk || (desc->state && (desc->state != 4))) {
1568 printk(" D %2d: ", i);
1569 print_desc(desc);
1570 }
1571 }
1572 printk(KERN_INFO "md: THIS: ");
1573 print_desc(&sb->this_disk);
1574
1575}
1576
1577static void print_rdev(mdk_rdev_t *rdev)
1578{
1579 char b[BDEVNAME_SIZE];
1580 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1581 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
b2d444d7
N		 1582 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1583 rdev->desc_nr);
1da177e4
LT		 1584 if (rdev->sb_loaded) {
1585 printk(KERN_INFO "md: rdev superblock:\n");
1586 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1587 } else
1588 printk(KERN_INFO "md: no rdev superblock!\n");
1589}
1590
5e56341d 1591static void md_print_devices(void)
1da177e4
LT		 1592{
1593 struct list_head *tmp, *tmp2;
1594 mdk_rdev_t *rdev;
1595 mddev_t *mddev;
1596 char b[BDEVNAME_SIZE];
1597
1598 printk("\n");
1599 printk("md: **********************************\n");
1600 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1601 printk("md: **********************************\n");
29ac4aa3 1602 for_each_mddev(mddev, tmp) {
1da177e4 1603
32a7627c
N		 1604 if (mddev->bitmap)
1605 bitmap_print_sb(mddev->bitmap);
1606 else
1607 printk("%s: ", mdname(mddev));
d089c6af 1608 rdev_for_each(rdev, tmp2, mddev)
1da177e4
LT		 1609 printk("<%s>", bdevname(rdev->bdev,b));
1610 printk("\n");
1611
d089c6af 1612 rdev_for_each(rdev, tmp2, mddev)
1da177e4
LT		 1613 print_rdev(rdev);
1614 }
1615 printk("md: **********************************\n");
1616 printk("\n");
1617}
1618
1619
42543769 1620static void sync_sbs(mddev_t * mddev, int nospares)
1da177e4 1621{
42543769
N		 1622 /* Update each superblock (in-memory image), but
1623 * if we are allowed to, skip spares which already
1624 * have the right event counter, or have one earlier
1625 * (which would mean they aren't being marked as dirty
1626 * with the rest of the array)
1627 */
1da177e4
LT		 1628 mdk_rdev_t *rdev;
1629 struct list_head *tmp;
1630
d089c6af 1631 rdev_for_each(rdev, tmp, mddev) {
42543769
N		 1632 if (rdev->sb_events == mddev->events ||
1633 (nospares &&
1634 rdev->raid_disk < 0 &&
1635 (rdev->sb_events&1)==0 &&
1636 rdev->sb_events+1 == mddev->events)) {
1637 /* Don't update this superblock */
1638 rdev->sb_loaded = 2;
1639 } else {
1640 super_types[mddev->major_version].
1641 sync_super(mddev, rdev);
1642 rdev->sb_loaded = 1;
1643 }
1da177e4
LT		 1644 }
1645}
1646
850b2b42 1647static void md_update_sb(mddev_t * mddev, int force_change)
1da177e4 1648{
1da177e4
LT		 1649 struct list_head *tmp;
1650 mdk_rdev_t *rdev;
06d91a5f 1651 int sync_req;
42543769 1652 int nospares = 0;
1da177e4 1653
1da177e4 1654repeat:
a9701a30 1655 spin_lock_irq(&mddev->write_lock);
84692195 1656
850b2b42
N		 1657 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1658 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1659 force_change = 1;
1660 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1661 /* just a clean<-> dirty transition, possibly leave spares alone,
1662 * though if events isn't the right even/odd, we will have to do
1663 * spares after all
1664 */
1665 nospares = 1;
1666 if (force_change)
1667 nospares = 0;
1668 if (mddev->degraded)
84692195
N		 1669 /* If the array is degraded, then skipping spares is both
1670 * dangerous and fairly pointless.
1671 * Dangerous because a device that was removed from the array
1672 * might have a event_count that still looks up-to-date,
1673 * so it can be re-added without a resync.
1674 * Pointless because if there are any spares to skip,
1675 * then a recovery will happen and soon that array won't
1676 * be degraded any more and the spare can go back to sleep then.
1677 */
850b2b42 1678 nospares = 0;
84692195 1679
06d91a5f 1680 sync_req = mddev->in_sync;
1da177e4 1681 mddev->utime = get_seconds();
42543769
N		 1682
1683 /* If this is just a dirty<->clean transition, and the array is clean
1684 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 1685 if (nospares
42543769 1686 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1031be7a
N		 1687 && (mddev->events & 1)
1688 && mddev->events != 1)
42543769
N		 1689 mddev->events--;
1690 else {
1691 /* otherwise we have to go forward and ... */
1692 mddev->events ++;
1693 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1694 /* .. if the array isn't clean, insist on an odd 'events' */
1695 if ((mddev->events&1)==0) {
1696 mddev->events++;
1697 nospares = 0;
1698 }
1699 } else {
1700 /* otherwise insist on an even 'events' (for clean states) */
1701 if ((mddev->events&1)) {
1702 mddev->events++;
1703 nospares = 0;
1704 }
1705 }
1706 }
1da177e4
LT		 1707
1708 if (!mddev->events) {
1709 /*
1710 * oops, this 64-bit counter should never wrap.
1711 * Either we are in around ~1 trillion A.C., assuming
1712 * 1 reboot per second, or we have a bug:
1713 */
1714 MD_BUG();
1715 mddev->events --;
1716 }
1da177e4
LT		 1717
1718 /*
1719 * do not write anything to disk if using
1720 * nonpersistent superblocks
1721 */
06d91a5f 1722 if (!mddev->persistent) {
e691063a
N		 1723 if (!mddev->external)
1724 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1725
a9701a30 1726 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1727 wake_up(&mddev->sb_wait);
1da177e4 1728 return;
06d91a5f 1729 }
e691063a 1730 sync_sbs(mddev, nospares);
a9701a30 1731 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1732
1733 dprintk(KERN_INFO
1734 "md: updating %s RAID superblock on device (in sync %d)\n",
1735 mdname(mddev),mddev->in_sync);
1736
4ad13663 1737 bitmap_update_sb(mddev->bitmap);
d089c6af 1738 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 1739 char b[BDEVNAME_SIZE];
1740 dprintk(KERN_INFO "md: ");
42543769
N		 1741 if (rdev->sb_loaded != 1)
1742 continue; /* no noise on spare devices */
b2d444d7 1743 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1744 dprintk("(skipping faulty ");
1745
1746 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1747 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1748 md_super_write(mddev,rdev,
0002b271 1749 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1750 rdev->sb_page);
1751 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1752 bdevname(rdev->bdev,b),
1753 (unsigned long long)rdev->sb_offset);
42543769 1754 rdev->sb_events = mddev->events;
7bfa19f2 1755
1da177e4
LT		 1756 } else
1757 dprintk(")\n");
7bfa19f2 1758 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1759 /* only need to write one superblock... */
1760 break;
1761 }
a9701a30 1762 md_super_wait(mddev);
850b2b42 1763 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 1764
a9701a30 1765 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 1766 if (mddev->in_sync != sync_req ||
1767 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 1768 /* have to write it out again */
a9701a30 1769 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1770 goto repeat;
1771 }
850b2b42 1772 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 1773 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1774 wake_up(&mddev->sb_wait);
06d91a5f 1775
1da177e4
LT		 1776}
1777
bce74dac
N		 1778/* words written to sysfs files may, or my not, be \n terminated.
1779 * We want to accept with case. For this we use cmd_match.
1780 */
1781static int cmd_match(const char *cmd, const char *str)
1782{
1783 /* See if cmd, written into a sysfs file, matches
1784 * str. They must either be the same, or cmd can
1785 * have a trailing newline
1786 */
1787 while (*cmd && *str && *cmd == *str) {
1788 cmd++;
1789 str++;
1790 }
1791 if (*cmd == '\n')
1792 cmd++;
1793 if (*str || *cmd)
1794 return 0;
1795 return 1;
1796}
1797
86e6ffdd
N		 1798struct rdev_sysfs_entry {
1799 struct attribute attr;
1800 ssize_t (*show)(mdk_rdev_t *, char *);
1801 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1802};
1803
1804static ssize_t
96de1e66 1805state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1806{
1807 char *sep = "";
20a49ff6 1808 size_t len = 0;
86e6ffdd 1809
b2d444d7 1810 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1811 len+= sprintf(page+len, "%sfaulty",sep);
1812 sep = ",";
1813 }
b2d444d7 1814 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1815 len += sprintf(page+len, "%sin_sync",sep);
1816 sep = ",";
1817 }
f655675b
N		 1818 if (test_bit(WriteMostly, &rdev->flags)) {
1819 len += sprintf(page+len, "%swrite_mostly",sep);
1820 sep = ",";
1821 }
b2d444d7
N		 1822 if (!test_bit(Faulty, &rdev->flags) &&
1823 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1824 len += sprintf(page+len, "%sspare", sep);
1825 sep = ",";
1826 }
1827 return len+sprintf(page+len, "\n");
1828}
1829
45dc2de1
N		 1830static ssize_t
1831state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1832{
1833 /* can write
1834 * faulty - simulates and error
1835 * remove - disconnects the device
f655675b
N		 1836 * writemostly - sets write_mostly
1837 * -writemostly - clears write_mostly
45dc2de1
N		 1838 */
1839 int err = -EINVAL;
1840 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1841 md_error(rdev->mddev, rdev);
1842 err = 0;
1843 } else if (cmd_match(buf, "remove")) {
1844 if (rdev->raid_disk >= 0)
1845 err = -EBUSY;
1846 else {
1847 mddev_t *mddev = rdev->mddev;
1848 kick_rdev_from_array(rdev);
3f9d7b0d
N		 1849 if (mddev->pers)
1850 md_update_sb(mddev, 1);
45dc2de1
N		 1851 md_new_event(mddev);
1852 err = 0;
1853 }
f655675b
N		 1854 } else if (cmd_match(buf, "writemostly")) {
1855 set_bit(WriteMostly, &rdev->flags);
1856 err = 0;
1857 } else if (cmd_match(buf, "-writemostly")) {
1858 clear_bit(WriteMostly, &rdev->flags);
1859 err = 0;
45dc2de1
N		 1860 }
1861 return err ? err : len;
1862}
80ca3a44
N		 1863static struct rdev_sysfs_entry rdev_state =
1864__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 1865
4dbcdc75
N		 1866static ssize_t
1867errors_show(mdk_rdev_t *rdev, char *page)
1868{
1869 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1870}
1871
1872static ssize_t
1873errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1874{
1875 char *e;
1876 unsigned long n = simple_strtoul(buf, &e, 10);
1877 if (*buf && (*e == 0 || *e == '\n')) {
1878 atomic_set(&rdev->corrected_errors, n);
1879 return len;
1880 }
1881 return -EINVAL;
1882}
1883static struct rdev_sysfs_entry rdev_errors =
80ca3a44 1884__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 1885
014236d2
N		 1886static ssize_t
1887slot_show(mdk_rdev_t *rdev, char *page)
1888{
1889 if (rdev->raid_disk < 0)
1890 return sprintf(page, "none\n");
1891 else
1892 return sprintf(page, "%d\n", rdev->raid_disk);
1893}
1894
1895static ssize_t
1896slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1897{
1898 char *e;
c303da6d
N		 1899 int err;
1900 char nm[20];
014236d2
N		 1901 int slot = simple_strtoul(buf, &e, 10);
1902 if (strncmp(buf, "none", 4)==0)
1903 slot = -1;
1904 else if (e==buf || (*e && *e!= '\n'))
1905 return -EINVAL;
c303da6d
N		 1906 if (rdev->mddev->pers) {
1907 /* Setting 'slot' on an active array requires also
1908 * updating the 'rd%d' link, and communicating
1909 * with the personality with ->hot_*_disk.
1910 * For now we only support removing
1911 * failed/spare devices. This normally happens automatically,
1912 * but not when the metadata is externally managed.
1913 */
1914 if (slot != -1)
1915 return -EBUSY;
1916 if (rdev->raid_disk == -1)
1917 return -EEXIST;
1918 /* personality does all needed checks */
1919 if (rdev->mddev->pers->hot_add_disk == NULL)
1920 return -EINVAL;
1921 err = rdev->mddev->pers->
1922 hot_remove_disk(rdev->mddev, rdev->raid_disk);
1923 if (err)
1924 return err;
1925 sprintf(nm, "rd%d", rdev->raid_disk);
1926 sysfs_remove_link(&rdev->mddev->kobj, nm);
1927 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1928 md_wakeup_thread(rdev->mddev->thread);
1929 } else {
1930 if (slot >= rdev->mddev->raid_disks)
1931 return -ENOSPC;
1932 rdev->raid_disk = slot;
1933 /* assume it is working */
c5d79adb
N		 1934 clear_bit(Faulty, &rdev->flags);
1935 clear_bit(WriteMostly, &rdev->flags);
c303da6d
N		 1936 set_bit(In_sync, &rdev->flags);
1937 }
014236d2
N		 1938 return len;
1939}
1940
1941
1942static struct rdev_sysfs_entry rdev_slot =
80ca3a44 1943__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 1944
93c8cad0
N		 1945static ssize_t
1946offset_show(mdk_rdev_t *rdev, char *page)
1947{
6961ece4 1948 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 1949}
1950
1951static ssize_t
1952offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1953{
1954 char *e;
1955 unsigned long long offset = simple_strtoull(buf, &e, 10);
1956 if (e==buf || (*e && *e != '\n'))
1957 return -EINVAL;
1958 if (rdev->mddev->pers)
1959 return -EBUSY;
c5d79adb
N		 1960 if (rdev->size && rdev->mddev->external)
1961 /* Must set offset before size, so overlap checks
1962 * can be sane */
1963 return -EBUSY;
93c8cad0
N		 1964 rdev->data_offset = offset;
1965 return len;
1966}
1967
1968static struct rdev_sysfs_entry rdev_offset =
80ca3a44 1969__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 1970
83303b61
N		 1971static ssize_t
1972rdev_size_show(mdk_rdev_t *rdev, char *page)
1973{
1974 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1975}
1976
c5d79adb
N		 1977static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
1978{
1979 /* check if two start/length pairs overlap */
1980 if (s1+l1 <= s2)
1981 return 0;
1982 if (s2+l2 <= s1)
1983 return 0;
1984 return 1;
1985}
1986
83303b61
N		 1987static ssize_t
1988rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1989{
1990 char *e;
1991 unsigned long long size = simple_strtoull(buf, &e, 10);
c5d79adb 1992 unsigned long long oldsize = rdev->size;
27c529bb
N		 1993 mddev_t *my_mddev = rdev->mddev;
1994
83303b61
N		 1995 if (e==buf || (*e && *e != '\n'))
1996 return -EINVAL;
27c529bb 1997 if (my_mddev->pers)
83303b61
N		 1998 return -EBUSY;
1999 rdev->size = size;
c5d79adb
N		 2000 if (size > oldsize && rdev->mddev->external) {
2001 /* need to check that all other rdevs with the same ->bdev
2002 * do not overlap. We need to unlock the mddev to avoid
2003 * a deadlock. We have already changed rdev->size, and if
2004 * we have to change it back, we will have the lock again.
2005 */
2006 mddev_t *mddev;
2007 int overlap = 0;
2008 struct list_head *tmp, *tmp2;
2009
27c529bb 2010 mddev_unlock(my_mddev);
29ac4aa3 2011 for_each_mddev(mddev, tmp) {
c5d79adb
N		 2012 mdk_rdev_t *rdev2;
2013
2014 mddev_lock(mddev);
d089c6af 2015 rdev_for_each(rdev2, tmp2, mddev)
c5d79adb
N		 2016 if (test_bit(AllReserved, &rdev2->flags) ||
2017 (rdev->bdev == rdev2->bdev &&
2018 rdev != rdev2 &&
2019 overlaps(rdev->data_offset, rdev->size,
2020 rdev2->data_offset, rdev2->size))) {
2021 overlap = 1;
2022 break;
2023 }
2024 mddev_unlock(mddev);
2025 if (overlap) {
2026 mddev_put(mddev);
2027 break;
2028 }
2029 }
27c529bb 2030 mddev_lock(my_mddev);
c5d79adb
N		 2031 if (overlap) {
2032 /* Someone else could have slipped in a size
2033 * change here, but doing so is just silly.
2034 * We put oldsize back because we *know* it is
2035 * safe, and trust userspace not to race with
2036 * itself
2037 */
2038 rdev->size = oldsize;
2039 return -EBUSY;
2040 }
2041 }
27c529bb
N		 2042 if (size < my_mddev->size || my_mddev->size == 0)
2043 my_mddev->size = size;
83303b61
N		 2044 return len;
2045}
2046
2047static struct rdev_sysfs_entry rdev_size =
80ca3a44 2048__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2049
86e6ffdd
N		 2050static struct attribute *rdev_default_attrs[] = {
2051 &rdev_state.attr,
4dbcdc75 2052 &rdev_errors.attr,
014236d2 2053 &rdev_slot.attr,
93c8cad0 2054 &rdev_offset.attr,
83303b61 2055 &rdev_size.attr,
86e6ffdd
N		 2056 NULL,
2057};
2058static ssize_t
2059rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2060{
2061 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2062 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2063 mddev_t *mddev = rdev->mddev;
2064 ssize_t rv;
86e6ffdd
N		 2065
2066 if (!entry->show)
2067 return -EIO;
27c529bb
N		 2068
2069 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2070 if (!rv) {
2071 if (rdev->mddev == NULL)
2072 rv = -EBUSY;
2073 else
2074 rv = entry->show(rdev, page);
2075 mddev_unlock(mddev);
2076 }
2077 return rv;
86e6ffdd
N		 2078}
2079
2080static ssize_t
2081rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2082 const char *page, size_t length)
2083{
2084 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2085 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2086 ssize_t rv;
2087 mddev_t *mddev = rdev->mddev;
86e6ffdd
N		 2088
2089 if (!entry->store)
2090 return -EIO;
67463acb
N		 2091 if (!capable(CAP_SYS_ADMIN))
2092 return -EACCES;
27c529bb 2093 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 2094 if (!rv) {
27c529bb
N		 2095 if (rdev->mddev == NULL)
2096 rv = -EBUSY;
2097 else
2098 rv = entry->store(rdev, page, length);
ca388059
N		 2099 mddev_unlock(rdev->mddev);
2100 }
2101 return rv;
86e6ffdd
N		 2102}
2103
2104static void rdev_free(struct kobject *ko)
2105{
2106 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2107 kfree(rdev);
2108}
2109static struct sysfs_ops rdev_sysfs_ops = {
2110 .show = rdev_attr_show,
2111 .store = rdev_attr_store,
2112};
2113static struct kobj_type rdev_ktype = {
2114 .release = rdev_free,
2115 .sysfs_ops = &rdev_sysfs_ops,
2116 .default_attrs = rdev_default_attrs,
2117};
2118
1da177e4
LT		 2119/*
2120 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2121 *
2122 * mark the device faulty if:
2123 *
2124 * - the device is nonexistent (zero size)
2125 * - the device has no valid superblock
2126 *
2127 * a faulty rdev _never_ has rdev->sb set.
2128 */
2129static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2130{
2131 char b[BDEVNAME_SIZE];
2132 int err;
2133 mdk_rdev_t *rdev;
2134 sector_t size;
2135
9ffae0cf 2136 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 2137 if (!rdev) {
2138 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2139 return ERR_PTR(-ENOMEM);
2140 }
1da177e4
LT		 2141
2142 if ((err = alloc_disk_sb(rdev)))
2143 goto abort_free;
2144
c5d79adb 2145 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 2146 if (err)
2147 goto abort_free;
2148
f9cb074b 2149 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 2150
1da177e4 2151 rdev->desc_nr = -1;
2b6e8459 2152 rdev->saved_raid_disk = -1;
3f9d7b0d 2153 rdev->raid_disk = -1;
b2d444d7 2154 rdev->flags = 0;
1da177e4 2155 rdev->data_offset = 0;
42543769 2156 rdev->sb_events = 0;
1da177e4 2157 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 2158 atomic_set(&rdev->read_errors, 0);
4dbcdc75 2159 atomic_set(&rdev->corrected_errors, 0);
1da177e4
LT		 2160
2161 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2162 if (!size) {
2163 printk(KERN_WARNING
2164 "md: %s has zero or unknown size, marking faulty!\n",
2165 bdevname(rdev->bdev,b));
2166 err = -EINVAL;
2167 goto abort_free;
2168 }
2169
2170 if (super_format >= 0) {
2171 err = super_types[super_format].
2172 load_super(rdev, NULL, super_minor);
2173 if (err == -EINVAL) {
df968c4e
N		 2174 printk(KERN_WARNING
2175 "md: %s does not have a valid v%d.%d "
2176 "superblock, not importing!\n",
2177 bdevname(rdev->bdev,b),
2178 super_format, super_minor);
1da177e4
LT		 2179 goto abort_free;
2180 }
2181 if (err < 0) {
2182 printk(KERN_WARNING
2183 "md: could not read %s's sb, not importing!\n",
2184 bdevname(rdev->bdev,b));
2185 goto abort_free;
2186 }
2187 }
2188 INIT_LIST_HEAD(&rdev->same_set);
2189
2190 return rdev;
2191
2192abort_free:
2193 if (rdev->sb_page) {
2194 if (rdev->bdev)
2195 unlock_rdev(rdev);
2196 free_disk_sb(rdev);
2197 }
2198 kfree(rdev);
2199 return ERR_PTR(err);
2200}
2201
2202/*
2203 * Check a full RAID array for plausibility
2204 */
2205
2206
a757e64c 2207static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2208{
2209 int i;
2210 struct list_head *tmp;
2211 mdk_rdev_t *rdev, *freshest;
2212 char b[BDEVNAME_SIZE];
2213
2214 freshest = NULL;
d089c6af 2215 rdev_for_each(rdev, tmp, mddev)
1da177e4
LT		 2216 switch (super_types[mddev->major_version].
2217 load_super(rdev, freshest, mddev->minor_version)) {
2218 case 1:
2219 freshest = rdev;
2220 break;
2221 case 0:
2222 break;
2223 default:
2224 printk( KERN_ERR \
2225 "md: fatal superblock inconsistency in %s"
2226 " -- removing from array\n",
2227 bdevname(rdev->bdev,b));
2228 kick_rdev_from_array(rdev);
2229 }
2230
2231
2232 super_types[mddev->major_version].
2233 validate_super(mddev, freshest);
2234
2235 i = 0;
d089c6af 2236 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 2237 if (rdev != freshest)
2238 if (super_types[mddev->major_version].
2239 validate_super(mddev, rdev)) {
2240 printk(KERN_WARNING "md: kicking non-fresh %s"
2241 " from array!\n",
2242 bdevname(rdev->bdev,b));
2243 kick_rdev_from_array(rdev);
2244 continue;
2245 }
2246 if (mddev->level == LEVEL_MULTIPATH) {
2247 rdev->desc_nr = i++;
2248 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2249 set_bit(In_sync, &rdev->flags);
a778b73f
N		 2250 } else if (rdev->raid_disk >= mddev->raid_disks) {
2251 rdev->raid_disk = -1;
2252 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 2253 }
2254 }
2255
2256
2257
2258 if (mddev->recovery_cp != MaxSector &&
2259 mddev->level >= 1)
2260 printk(KERN_ERR "md: %s: raid array is not clean"
2261 " -- starting background reconstruction\n",
2262 mdname(mddev));
2263
1da177e4
LT		 2264}
2265
16f17b39
N		 2266static ssize_t
2267safe_delay_show(mddev_t *mddev, char *page)
2268{
2269 int msec = (mddev->safemode_delay*1000)/HZ;
2270 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2271}
2272static ssize_t
2273safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2274{
2275 int scale=1;
2276 int dot=0;
2277 int i;
2278 unsigned long msec;
2279 char buf[30];
2280 char *e;
2281 /* remove a period, and count digits after it */
2282 if (len >= sizeof(buf))
2283 return -EINVAL;
2284 strlcpy(buf, cbuf, len);
2285 buf[len] = 0;
2286 for (i=0; i<len; i++) {
2287 if (dot) {
2288 if (isdigit(buf[i])) {
2289 buf[i-1] = buf[i];
2290 scale *= 10;
2291 }
2292 buf[i] = 0;
2293 } else if (buf[i] == '.') {
2294 dot=1;
2295 buf[i] = 0;
2296 }
2297 }
2298 msec = simple_strtoul(buf, &e, 10);
2299 if (e == buf || (*e && *e != '\n'))
2300 return -EINVAL;
2301 msec = (msec * 1000) / scale;
2302 if (msec == 0)
2303 mddev->safemode_delay = 0;
2304 else {
2305 mddev->safemode_delay = (msec*HZ)/1000;
2306 if (mddev->safemode_delay == 0)
2307 mddev->safemode_delay = 1;
2308 }
2309 return len;
2310}
2311static struct md_sysfs_entry md_safe_delay =
80ca3a44 2312__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2313
eae1701f 2314static ssize_t
96de1e66 2315level_show(mddev_t *mddev, char *page)
eae1701f 2316{
2604b703 2317 struct mdk_personality *p = mddev->pers;
d9d166c2 2318 if (p)
eae1701f 2319 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2320 else if (mddev->clevel[0])
2321 return sprintf(page, "%s\n", mddev->clevel);
2322 else if (mddev->level != LEVEL_NONE)
2323 return sprintf(page, "%d\n", mddev->level);
2324 else
2325 return 0;
eae1701f
N		 2326}
2327
d9d166c2
N		 2328static ssize_t
2329level_store(mddev_t *mddev, const char *buf, size_t len)
2330{
20a49ff6 2331 ssize_t rv = len;
d9d166c2
N		 2332 if (mddev->pers)
2333 return -EBUSY;
2334 if (len == 0)
2335 return 0;
2336 if (len >= sizeof(mddev->clevel))
2337 return -ENOSPC;
2338 strncpy(mddev->clevel, buf, len);
2339 if (mddev->clevel[len-1] == '\n')
2340 len--;
2341 mddev->clevel[len] = 0;
2342 mddev->level = LEVEL_NONE;
2343 return rv;
2344}
2345
2346static struct md_sysfs_entry md_level =
80ca3a44 2347__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 2348
d4dbd025
N		 2349
2350static ssize_t
2351layout_show(mddev_t *mddev, char *page)
2352{
2353 /* just a number, not meaningful for all levels */
08a02ecd
N		 2354 if (mddev->reshape_position != MaxSector &&
2355 mddev->layout != mddev->new_layout)
2356 return sprintf(page, "%d (%d)\n",
2357 mddev->new_layout, mddev->layout);
d4dbd025
N		 2358 return sprintf(page, "%d\n", mddev->layout);
2359}
2360
2361static ssize_t
2362layout_store(mddev_t *mddev, const char *buf, size_t len)
2363{
2364 char *e;
2365 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 2366
2367 if (!*buf || (*e && *e != '\n'))
2368 return -EINVAL;
2369
08a02ecd
N		 2370 if (mddev->pers)
2371 return -EBUSY;
2372 if (mddev->reshape_position != MaxSector)
2373 mddev->new_layout = n;
2374 else
2375 mddev->layout = n;
d4dbd025
N		 2376 return len;
2377}
2378static struct md_sysfs_entry md_layout =
80ca3a44 2379__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 2380
2381
eae1701f 2382static ssize_t
96de1e66 2383raid_disks_show(mddev_t *mddev, char *page)
eae1701f 2384{
bb636547
N		 2385 if (mddev->raid_disks == 0)
2386 return 0;
08a02ecd
N		 2387 if (mddev->reshape_position != MaxSector &&
2388 mddev->delta_disks != 0)
2389 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2390 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 2391 return sprintf(page, "%d\n", mddev->raid_disks);
2392}
2393
da943b99
N		 2394static int update_raid_disks(mddev_t *mddev, int raid_disks);
2395
2396static ssize_t
2397raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2398{
da943b99
N		 2399 char *e;
2400 int rv = 0;
2401 unsigned long n = simple_strtoul(buf, &e, 10);
2402
2403 if (!*buf || (*e && *e != '\n'))
2404 return -EINVAL;
2405
2406 if (mddev->pers)
2407 rv = update_raid_disks(mddev, n);
08a02ecd
N		 2408 else if (mddev->reshape_position != MaxSector) {
2409 int olddisks = mddev->raid_disks - mddev->delta_disks;
2410 mddev->delta_disks = n - olddisks;
2411 mddev->raid_disks = n;
2412 } else
da943b99
N		 2413 mddev->raid_disks = n;
2414 return rv ? rv : len;
2415}
2416static struct md_sysfs_entry md_raid_disks =
80ca3a44 2417__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 2418
3b34380a
N		 2419static ssize_t
2420chunk_size_show(mddev_t *mddev, char *page)
2421{
08a02ecd
N		 2422 if (mddev->reshape_position != MaxSector &&
2423 mddev->chunk_size != mddev->new_chunk)
2424 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2425 mddev->chunk_size);
3b34380a
N		 2426 return sprintf(page, "%d\n", mddev->chunk_size);
2427}
2428
2429static ssize_t
2430chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2431{
2432 /* can only set chunk_size if array is not yet active */
2433 char *e;
2434 unsigned long n = simple_strtoul(buf, &e, 10);
2435
3b34380a
N		 2436 if (!*buf || (*e && *e != '\n'))
2437 return -EINVAL;
2438
08a02ecd
N		 2439 if (mddev->pers)
2440 return -EBUSY;
2441 else if (mddev->reshape_position != MaxSector)
2442 mddev->new_chunk = n;
2443 else
2444 mddev->chunk_size = n;
3b34380a
N		 2445 return len;
2446}
2447static struct md_sysfs_entry md_chunk_size =
80ca3a44 2448__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 2449
a94213b1
N		 2450static ssize_t
2451resync_start_show(mddev_t *mddev, char *page)
2452{
2453 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2454}
2455
2456static ssize_t
2457resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2458{
2459 /* can only set chunk_size if array is not yet active */
2460 char *e;
2461 unsigned long long n = simple_strtoull(buf, &e, 10);
2462
2463 if (mddev->pers)
2464 return -EBUSY;
2465 if (!*buf || (*e && *e != '\n'))
2466 return -EINVAL;
2467
2468 mddev->recovery_cp = n;
2469 return len;
2470}
2471static struct md_sysfs_entry md_resync_start =
80ca3a44 2472__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 2473
9e653b63
N		 2474/*
2475 * The array state can be:
2476 *
2477 * clear
2478 * No devices, no size, no level
2479 * Equivalent to STOP_ARRAY ioctl
2480 * inactive
2481 * May have some settings, but array is not active
2482 * all IO results in error
2483 * When written, doesn't tear down array, but just stops it
2484 * suspended (not supported yet)
2485 * All IO requests will block. The array can be reconfigured.
2486 * Writing this, if accepted, will block until array is quiessent
2487 * readonly
2488 * no resync can happen. no superblocks get written.
2489 * write requests fail
2490 * read-auto
2491 * like readonly, but behaves like 'clean' on a write request.
2492 *
2493 * clean - no pending writes, but otherwise active.
2494 * When written to inactive array, starts without resync
2495 * If a write request arrives then
2496 * if metadata is known, mark 'dirty' and switch to 'active'.
2497 * if not known, block and switch to write-pending
2498 * If written to an active array that has pending writes, then fails.
2499 * active
2500 * fully active: IO and resync can be happening.
2501 * When written to inactive array, starts with resync
2502 *
2503 * write-pending
2504 * clean, but writes are blocked waiting for 'active' to be written.
2505 *
2506 * active-idle
2507 * like active, but no writes have been seen for a while (100msec).
2508 *
2509 */
2510enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2511 write_pending, active_idle, bad_word};
05381954 2512static char *array_states[] = {
9e653b63
N		 2513 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2514 "write-pending", "active-idle", NULL };
2515
2516static int match_word(const char *word, char **list)
2517{
2518 int n;
2519 for (n=0; list[n]; n++)
2520 if (cmd_match(word, list[n]))
2521 break;
2522 return n;
2523}
2524
2525static ssize_t
2526array_state_show(mddev_t *mddev, char *page)
2527{
2528 enum array_state st = inactive;
2529
2530 if (mddev->pers)
2531 switch(mddev->ro) {
2532 case 1:
2533 st = readonly;
2534 break;
2535 case 2:
2536 st = read_auto;
2537 break;
2538 case 0:
2539 if (mddev->in_sync)
2540 st = clean;
e691063a
N		 2541 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2542 st = write_pending;
9e653b63
N		 2543 else if (mddev->safemode)
2544 st = active_idle;
2545 else
2546 st = active;
2547 }
2548 else {
2549 if (list_empty(&mddev->disks) &&
2550 mddev->raid_disks == 0 &&
2551 mddev->size == 0)
2552 st = clear;
2553 else
2554 st = inactive;
2555 }
2556 return sprintf(page, "%s\n", array_states[st]);
2557}
2558
2559static int do_md_stop(mddev_t * mddev, int ro);
2560static int do_md_run(mddev_t * mddev);
2561static int restart_array(mddev_t *mddev);
2562
2563static ssize_t
2564array_state_store(mddev_t *mddev, const char *buf, size_t len)
2565{
2566 int err = -EINVAL;
2567 enum array_state st = match_word(buf, array_states);
2568 switch(st) {
2569 case bad_word:
2570 break;
2571 case clear:
2572 /* stopping an active array */
e691063a
N		 2573 if (atomic_read(&mddev->active) > 1)
2574 return -EBUSY;
2575 err = do_md_stop(mddev, 0);
9e653b63
N		 2576 break;
2577 case inactive:
2578 /* stopping an active array */
2579 if (mddev->pers) {
2580 if (atomic_read(&mddev->active) > 1)
2581 return -EBUSY;
2582 err = do_md_stop(mddev, 2);
e691063a
N		 2583 } else
2584 err = 0; /* already inactive */
9e653b63
N		 2585 break;
2586 case suspended:
2587 break; /* not supported yet */
2588 case readonly:
2589 if (mddev->pers)
2590 err = do_md_stop(mddev, 1);
2591 else {
2592 mddev->ro = 1;
2593 err = do_md_run(mddev);
2594 }
2595 break;
2596 case read_auto:
2597 /* stopping an active array */
2598 if (mddev->pers) {
2599 err = do_md_stop(mddev, 1);
2600 if (err == 0)
2601 mddev->ro = 2; /* FIXME mark devices writable */
2602 } else {
2603 mddev->ro = 2;
2604 err = do_md_run(mddev);
2605 }
2606 break;
2607 case clean:
2608 if (mddev->pers) {
2609 restart_array(mddev);
2610 spin_lock_irq(&mddev->write_lock);
2611 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 2612 if (mddev->in_sync == 0) {
2613 mddev->in_sync = 1;
2614 if (mddev->persistent)
2615 set_bit(MD_CHANGE_CLEAN,
2616 &mddev->flags);
2617 }
2618 err = 0;
2619 } else
2620 err = -EBUSY;
9e653b63
N		 2621 spin_unlock_irq(&mddev->write_lock);
2622 } else {
2623 mddev->ro = 0;
2624 mddev->recovery_cp = MaxSector;
2625 err = do_md_run(mddev);
2626 }
2627 break;
2628 case active:
2629 if (mddev->pers) {
2630 restart_array(mddev);
e691063a
N		 2631 if (mddev->external)
2632 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
9e653b63
N		 2633 wake_up(&mddev->sb_wait);
2634 err = 0;
2635 } else {
2636 mddev->ro = 0;
2637 err = do_md_run(mddev);
2638 }
2639 break;
2640 case write_pending:
2641 case active_idle:
2642 /* these cannot be set */
2643 break;
2644 }
2645 if (err)
2646 return err;
2647 else
2648 return len;
2649}
80ca3a44
N		 2650static struct md_sysfs_entry md_array_state =
2651__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 2652
6d7ff738
N		 2653static ssize_t
2654null_show(mddev_t *mddev, char *page)
2655{
2656 return -EINVAL;
2657}
2658
2659static ssize_t
2660new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2661{
2662 /* buf must be %d:%d\n? giving major and minor numbers */
2663 /* The new device is added to the array.
2664 * If the array has a persistent superblock, we read the
2665 * superblock to initialise info and check validity.
2666 * Otherwise, only checking done is that in bind_rdev_to_array,
2667 * which mainly checks size.
2668 */
2669 char *e;
2670 int major = simple_strtoul(buf, &e, 10);
2671 int minor;
2672 dev_t dev;
2673 mdk_rdev_t *rdev;
2674 int err;
2675
2676 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2677 return -EINVAL;
2678 minor = simple_strtoul(e+1, &e, 10);
2679 if (*e && *e != '\n')
2680 return -EINVAL;
2681 dev = MKDEV(major, minor);
2682 if (major != MAJOR(dev) ||
2683 minor != MINOR(dev))
2684 return -EOVERFLOW;
2685
2686
2687 if (mddev->persistent) {
2688 rdev = md_import_device(dev, mddev->major_version,
2689 mddev->minor_version);
2690 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2691 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2692 mdk_rdev_t, same_set);
2693 err = super_types[mddev->major_version]
2694 .load_super(rdev, rdev0, mddev->minor_version);
2695 if (err < 0)
2696 goto out;
2697 }
c5d79adb
N		 2698 } else if (mddev->external)
2699 rdev = md_import_device(dev, -2, -1);
2700 else
6d7ff738
N		 2701 rdev = md_import_device(dev, -1, -1);
2702
2703 if (IS_ERR(rdev))
2704 return PTR_ERR(rdev);
2705 err = bind_rdev_to_array(rdev, mddev);
2706 out:
2707 if (err)
2708 export_rdev(rdev);
2709 return err ? err : len;
2710}
2711
2712static struct md_sysfs_entry md_new_device =
80ca3a44 2713__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 2714
9b1d1dac
PC		 2715static ssize_t
2716bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2717{
2718 char *end;
2719 unsigned long chunk, end_chunk;
2720
2721 if (!mddev->bitmap)
2722 goto out;
2723 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2724 while (*buf) {
2725 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2726 if (buf == end) break;
2727 if (*end == '-') { /* range */
2728 buf = end + 1;
2729 end_chunk = simple_strtoul(buf, &end, 0);
2730 if (buf == end) break;
2731 }
2732 if (*end && !isspace(*end)) break;
2733 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2734 buf = end;
2735 while (isspace(*buf)) buf++;
2736 }
2737 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2738out:
2739 return len;
2740}
2741
2742static struct md_sysfs_entry md_bitmap =
2743__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2744
a35b0d69
N		 2745static ssize_t
2746size_show(mddev_t *mddev, char *page)
2747{
2748 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2749}
2750
2751static int update_size(mddev_t *mddev, unsigned long size);
2752
2753static ssize_t
2754size_store(mddev_t *mddev, const char *buf, size_t len)
2755{
2756 /* If array is inactive, we can reduce the component size, but
2757 * not increase it (except from 0).
2758 * If array is active, we can try an on-line resize
2759 */
2760 char *e;
2761 int err = 0;
2762 unsigned long long size = simple_strtoull(buf, &e, 10);
2763 if (!*buf || *buf == '\n' ||
2764 (*e && *e != '\n'))
2765 return -EINVAL;
2766
2767 if (mddev->pers) {
2768 err = update_size(mddev, size);
850b2b42 2769 md_update_sb(mddev, 1);
a35b0d69
N		 2770 } else {
2771 if (mddev->size == 0 ||
2772 mddev->size > size)
2773 mddev->size = size;
2774 else
2775 err = -ENOSPC;
2776 }
2777 return err ? err : len;
2778}
2779
2780static struct md_sysfs_entry md_size =
80ca3a44 2781__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 2782
8bb93aac
N		 2783
2784/* Metdata version.
e691063a
N		 2785 * This is one of
2786 * 'none' for arrays with no metadata (good luck...)
2787 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 2788 * or N.M for internally known formats
2789 */
2790static ssize_t
2791metadata_show(mddev_t *mddev, char *page)
2792{
2793 if (mddev->persistent)
2794 return sprintf(page, "%d.%d\n",
2795 mddev->major_version, mddev->minor_version);
e691063a
N		 2796 else if (mddev->external)
2797 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 2798 else
2799 return sprintf(page, "none\n");
2800}
2801
2802static ssize_t
2803metadata_store(mddev_t *mddev, const char *buf, size_t len)
2804{
2805 int major, minor;
2806 char *e;
2807 if (!list_empty(&mddev->disks))
2808 return -EBUSY;
2809
2810 if (cmd_match(buf, "none")) {
2811 mddev->persistent = 0;
e691063a
N		 2812 mddev->external = 0;
2813 mddev->major_version = 0;
2814 mddev->minor_version = 90;
2815 return len;
2816 }
2817 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 2818 size_t namelen = len-9;
e691063a
N		 2819 if (namelen >= sizeof(mddev->metadata_type))
2820 namelen = sizeof(mddev->metadata_type)-1;
2821 strncpy(mddev->metadata_type, buf+9, namelen);
2822 mddev->metadata_type[namelen] = 0;
2823 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2824 mddev->metadata_type[--namelen] = 0;
2825 mddev->persistent = 0;
2826 mddev->external = 1;
8bb93aac
N		 2827 mddev->major_version = 0;
2828 mddev->minor_version = 90;
2829 return len;
2830 }
2831 major = simple_strtoul(buf, &e, 10);
2832 if (e==buf || *e != '.')
2833 return -EINVAL;
2834 buf = e+1;
2835 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 2836 if (e==buf || (*e && *e != '\n') )
8bb93aac 2837 return -EINVAL;
50511da3 2838 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 2839 return -ENOENT;
2840 mddev->major_version = major;
2841 mddev->minor_version = minor;
2842 mddev->persistent = 1;
e691063a 2843 mddev->external = 0;
8bb93aac
N		 2844 return len;
2845}
2846
2847static struct md_sysfs_entry md_metadata =
80ca3a44 2848__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 2849
24dd469d 2850static ssize_t
7eec314d 2851action_show(mddev_t *mddev, char *page)
24dd469d 2852{
7eec314d 2853 char *type = "idle";
31399d9e 2854 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 2855 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 2856 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2857 type = "reshape";
2858 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 2859 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2860 type = "resync";
2861 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2862 type = "check";
2863 else
2864 type = "repair";
2865 } else
2866 type = "recover";
2867 }
2868 return sprintf(page, "%s\n", type);
2869}
2870
2871static ssize_t
7eec314d 2872action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 2873{
7eec314d
N		 2874 if (!mddev->pers || !mddev->pers->sync_request)
2875 return -EINVAL;
2876
bce74dac 2877 if (cmd_match(page, "idle")) {
7eec314d
N		 2878 if (mddev->sync_thread) {
2879 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2880 md_unregister_thread(mddev->sync_thread);
2881 mddev->sync_thread = NULL;
2882 mddev->recovery = 0;
2883 }
03c902e1
N		 2884 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2885 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 2886 return -EBUSY;
03c902e1 2887 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
7eec314d 2888 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
16484bf5
N		 2889 else if (cmd_match(page, "reshape")) {
2890 int err;
2891 if (mddev->pers->start_reshape == NULL)
2892 return -EINVAL;
2893 err = mddev->pers->start_reshape(mddev);
2894 if (err)
2895 return err;
2896 } else {
bce74dac 2897 if (cmd_match(page, "check"))
7eec314d 2898 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 2899 else if (!cmd_match(page, "repair"))
7eec314d
N		 2900 return -EINVAL;
2901 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2902 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 2903 }
03c902e1 2904 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 2905 md_wakeup_thread(mddev->thread);
2906 return len;
2907}
2908
9d88883e 2909static ssize_t
96de1e66 2910mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 2911{
2912 return sprintf(page, "%llu\n",
2913 (unsigned long long) mddev->resync_mismatches);
2914}
2915
80ca3a44
N		 2916static struct md_sysfs_entry md_scan_mode =
2917__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 2918
96de1e66 2919
80ca3a44 2920static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 2921
88202a0c
N		 2922static ssize_t
2923sync_min_show(mddev_t *mddev, char *page)
2924{
2925 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2926 mddev->sync_speed_min ? "local": "system");
2927}
2928
2929static ssize_t
2930sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2931{
2932 int min;
2933 char *e;
2934 if (strncmp(buf, "system", 6)==0) {
2935 mddev->sync_speed_min = 0;
2936 return len;
2937 }
2938 min = simple_strtoul(buf, &e, 10);
2939 if (buf == e || (*e && *e != '\n') || min <= 0)
2940 return -EINVAL;
2941 mddev->sync_speed_min = min;
2942 return len;
2943}
2944
2945static struct md_sysfs_entry md_sync_min =
2946__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2947
2948static ssize_t
2949sync_max_show(mddev_t *mddev, char *page)
2950{
2951 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2952 mddev->sync_speed_max ? "local": "system");
2953}
2954
2955static ssize_t
2956sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2957{
2958 int max;
2959 char *e;
2960 if (strncmp(buf, "system", 6)==0) {
2961 mddev->sync_speed_max = 0;
2962 return len;
2963 }
2964 max = simple_strtoul(buf, &e, 10);
2965 if (buf == e || (*e && *e != '\n') || max <= 0)
2966 return -EINVAL;
2967 mddev->sync_speed_max = max;
2968 return len;
2969}
2970
2971static struct md_sysfs_entry md_sync_max =
2972__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2973
d7f3d291
IP		 2974static ssize_t
2975degraded_show(mddev_t *mddev, char *page)
2976{
2977 return sprintf(page, "%d\n", mddev->degraded);
2978}
2979static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c
N		 2980
2981static ssize_t
2982sync_speed_show(mddev_t *mddev, char *page)
2983{
2984 unsigned long resync, dt, db;
ff4e8d9a 2985 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
88202a0c
N		 2986 dt = ((jiffies - mddev->resync_mark) / HZ);
2987 if (!dt) dt++;
2988 db = resync - (mddev->resync_mark_cnt);
2989 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2990}
2991
80ca3a44 2992static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 2993
2994static ssize_t
2995sync_completed_show(mddev_t *mddev, char *page)
2996{
2997 unsigned long max_blocks, resync;
2998
2999 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3000 max_blocks = mddev->resync_max_sectors;
3001 else
3002 max_blocks = mddev->size << 1;
3003
3004 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3005 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3006}
3007
80ca3a44 3008static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 3009
c6207277
N		 3010static ssize_t
3011max_sync_show(mddev_t *mddev, char *page)
3012{
3013 if (mddev->resync_max == MaxSector)
3014 return sprintf(page, "max\n");
3015 else
3016 return sprintf(page, "%llu\n",
3017 (unsigned long long)mddev->resync_max);
3018}
3019static ssize_t
3020max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3021{
3022 if (strncmp(buf, "max", 3) == 0)
3023 mddev->resync_max = MaxSector;
3024 else {
3025 char *ep;
3026 unsigned long long max = simple_strtoull(buf, &ep, 10);
3027 if (ep == buf || (*ep != 0 && *ep != '\n'))
3028 return -EINVAL;
3029 if (max < mddev->resync_max &&
3030 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3031 return -EBUSY;
3032
3033 /* Must be a multiple of chunk_size */
3034 if (mddev->chunk_size) {
3035 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3036 return -EINVAL;
3037 }
3038 mddev->resync_max = max;
3039 }
3040 wake_up(&mddev->recovery_wait);
3041 return len;
3042}
3043
3044static struct md_sysfs_entry md_max_sync =
3045__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3046
e464eafd
N		 3047static ssize_t
3048suspend_lo_show(mddev_t *mddev, char *page)
3049{
3050 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3051}
3052
3053static ssize_t
3054suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3055{
3056 char *e;
3057 unsigned long long new = simple_strtoull(buf, &e, 10);
3058
3059 if (mddev->pers->quiesce == NULL)
3060 return -EINVAL;
3061 if (buf == e || (*e && *e != '\n'))
3062 return -EINVAL;
3063 if (new >= mddev->suspend_hi ||
3064 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3065 mddev->suspend_lo = new;
3066 mddev->pers->quiesce(mddev, 2);
3067 return len;
3068 } else
3069 return -EINVAL;
3070}
3071static struct md_sysfs_entry md_suspend_lo =
3072__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3073
3074
3075static ssize_t
3076suspend_hi_show(mddev_t *mddev, char *page)
3077{
3078 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3079}
3080
3081static ssize_t
3082suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3083{
3084 char *e;
3085 unsigned long long new = simple_strtoull(buf, &e, 10);
3086
3087 if (mddev->pers->quiesce == NULL)
3088 return -EINVAL;
3089 if (buf == e || (*e && *e != '\n'))
3090 return -EINVAL;
3091 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3092 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3093 mddev->suspend_hi = new;
3094 mddev->pers->quiesce(mddev, 1);
3095 mddev->pers->quiesce(mddev, 0);
3096 return len;
3097 } else
3098 return -EINVAL;
3099}
3100static struct md_sysfs_entry md_suspend_hi =
3101__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3102
08a02ecd
N		 3103static ssize_t
3104reshape_position_show(mddev_t *mddev, char *page)
3105{
3106 if (mddev->reshape_position != MaxSector)
3107 return sprintf(page, "%llu\n",
3108 (unsigned long long)mddev->reshape_position);
3109 strcpy(page, "none\n");
3110 return 5;
3111}
3112
3113static ssize_t
3114reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3115{
3116 char *e;
3117 unsigned long long new = simple_strtoull(buf, &e, 10);
3118 if (mddev->pers)
3119 return -EBUSY;
3120 if (buf == e || (*e && *e != '\n'))
3121 return -EINVAL;
3122 mddev->reshape_position = new;
3123 mddev->delta_disks = 0;
3124 mddev->new_level = mddev->level;
3125 mddev->new_layout = mddev->layout;
3126 mddev->new_chunk = mddev->chunk_size;
3127 return len;
3128}
3129
3130static struct md_sysfs_entry md_reshape_position =
3131__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3132 reshape_position_store);
3133
e464eafd 3134
eae1701f
N		 3135static struct attribute *md_default_attrs[] = {
3136 &md_level.attr,
d4dbd025 3137 &md_layout.attr,
eae1701f 3138 &md_raid_disks.attr,
3b34380a 3139 &md_chunk_size.attr,
a35b0d69 3140 &md_size.attr,
a94213b1 3141 &md_resync_start.attr,
8bb93aac 3142 &md_metadata.attr,
6d7ff738 3143 &md_new_device.attr,
16f17b39 3144 &md_safe_delay.attr,
9e653b63 3145 &md_array_state.attr,
08a02ecd 3146 &md_reshape_position.attr,
411036fa
N		 3147 NULL,
3148};
3149
3150static struct attribute *md_redundancy_attrs[] = {
24dd469d 3151 &md_scan_mode.attr,
9d88883e 3152 &md_mismatches.attr,
88202a0c
N		 3153 &md_sync_min.attr,
3154 &md_sync_max.attr,
3155 &md_sync_speed.attr,
3156 &md_sync_completed.attr,
c6207277 3157 &md_max_sync.attr,
e464eafd
N		 3158 &md_suspend_lo.attr,
3159 &md_suspend_hi.attr,
9b1d1dac 3160 &md_bitmap.attr,
d7f3d291 3161 &md_degraded.attr,
eae1701f
N		 3162 NULL,
3163};
411036fa
N		 3164static struct attribute_group md_redundancy_group = {
3165 .name = NULL,
3166 .attrs = md_redundancy_attrs,
3167};
3168
eae1701f
N		 3169
3170static ssize_t
3171md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3172{
3173 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3174 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 3175 ssize_t rv;
eae1701f
N		 3176
3177 if (!entry->show)
3178 return -EIO;
5dc5cf7d
IM		 3179 rv = mddev_lock(mddev);
3180 if (!rv) {
3181 rv = entry->show(mddev, page);
3182 mddev_unlock(mddev);
3183 }
96de1e66 3184 return rv;
eae1701f
N		 3185}
3186
3187static ssize_t
3188md_attr_store(struct kobject *kobj, struct attribute *attr,
3189 const char *page, size_t length)
3190{
3191 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3192 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 3193 ssize_t rv;
eae1701f
N		 3194
3195 if (!entry->store)
3196 return -EIO;
67463acb
N		 3197 if (!capable(CAP_SYS_ADMIN))
3198 return -EACCES;
5dc5cf7d
IM		 3199 rv = mddev_lock(mddev);
3200 if (!rv) {
3201 rv = entry->store(mddev, page, length);
3202 mddev_unlock(mddev);
3203 }
96de1e66 3204 return rv;
eae1701f
N		 3205}
3206
3207static void md_free(struct kobject *ko)
3208{
3209 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3210 kfree(mddev);
3211}
3212
3213static struct sysfs_ops md_sysfs_ops = {
3214 .show = md_attr_show,
3215 .store = md_attr_store,
3216};
3217static struct kobj_type md_ktype = {
3218 .release = md_free,
3219 .sysfs_ops = &md_sysfs_ops,
3220 .default_attrs = md_default_attrs,
3221};
3222
1da177e4
LT		 3223int mdp_major = 0;
3224
3225static struct kobject *md_probe(dev_t dev, int *part, void *data)
3226{
48c9c27b 3227 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 3228 mddev_t *mddev = mddev_find(dev);
3229 struct gendisk *disk;
3230 int partitioned = (MAJOR(dev) != MD_MAJOR);
3231 int shift = partitioned ? MdpMinorShift : 0;
3232 int unit = MINOR(dev) >> shift;
3830c62f 3233 int error;
1da177e4
LT		 3234
3235 if (!mddev)
3236 return NULL;
3237
48c9c27b 3238 mutex_lock(&disks_mutex);
1da177e4 3239 if (mddev->gendisk) {
48c9c27b 3240 mutex_unlock(&disks_mutex);
1da177e4
LT		 3241 mddev_put(mddev);
3242 return NULL;
3243 }
3244 disk = alloc_disk(1 << shift);
3245 if (!disk) {
48c9c27b 3246 mutex_unlock(&disks_mutex);
1da177e4
LT		 3247 mddev_put(mddev);
3248 return NULL;
3249 }
3250 disk->major = MAJOR(dev);
3251 disk->first_minor = unit << shift;
ce7b0f46 3252 if (partitioned)
1da177e4 3253 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 3254 else
1da177e4 3255 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 3256 disk->fops = &md_fops;
3257 disk->private_data = mddev;
3258 disk->queue = mddev->queue;
3259 add_disk(disk);
3260 mddev->gendisk = disk;
48c9c27b 3261 mutex_unlock(&disks_mutex);
edfaa7c3 3262 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3830c62f
GKH		 3263 "%s", "md");
3264 if (error)
5e55e2f5
N		 3265 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3266 disk->disk_name);
3830c62f
GKH		 3267 else
3268 kobject_uevent(&mddev->kobj, KOBJ_ADD);
1da177e4
LT		 3269 return NULL;
3270}
3271
1da177e4
LT		 3272static void md_safemode_timeout(unsigned long data)
3273{
3274 mddev_t *mddev = (mddev_t *) data;
3275
3276 mddev->safemode = 1;
3277 md_wakeup_thread(mddev->thread);
3278}
3279
6ff8d8ec 3280static int start_dirty_degraded;
1da177e4
LT		 3281
3282static int do_md_run(mddev_t * mddev)
3283{
2604b703 3284 int err;
1da177e4
LT		 3285 int chunk_size;
3286 struct list_head *tmp;
3287 mdk_rdev_t *rdev;
3288 struct gendisk *disk;
2604b703 3289 struct mdk_personality *pers;
1da177e4
LT		 3290 char b[BDEVNAME_SIZE];
3291
a757e64c
N		 3292 if (list_empty(&mddev->disks))
3293 /* cannot run an array with no devices.. */
1da177e4 3294 return -EINVAL;
1da177e4
LT		 3295
3296 if (mddev->pers)
3297 return -EBUSY;
3298
3299 /*
3300 * Analyze all RAID superblock(s)
3301 */
1ec4a939
N		 3302 if (!mddev->raid_disks) {
3303 if (!mddev->persistent)
3304 return -EINVAL;
a757e64c 3305 analyze_sbs(mddev);
1ec4a939 3306 }
1da177e4
LT		 3307
3308 chunk_size = mddev->chunk_size;
2604b703
N		 3309
3310 if (chunk_size) {
1da177e4
LT		 3311 if (chunk_size > MAX_CHUNK_SIZE) {
3312 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3313 chunk_size, MAX_CHUNK_SIZE);
3314 return -EINVAL;
3315 }
3316 /*
3317 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3318 */
3319 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 3320 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 3321 return -EINVAL;
3322 }
3323 if (chunk_size < PAGE_SIZE) {
3324 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3325 chunk_size, PAGE_SIZE);
3326 return -EINVAL;
3327 }
3328
3329 /* devices must have minimum size of one chunk */
d089c6af 3330 rdev_for_each(rdev, tmp, mddev) {
b2d444d7 3331 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3332 continue;
3333 if (rdev->size < chunk_size / 1024) {
3334 printk(KERN_WARNING
3335 "md: Dev %s smaller than chunk_size:"
3336 " %lluk < %dk\n",
3337 bdevname(rdev->bdev,b),
3338 (unsigned long long)rdev->size,
3339 chunk_size / 1024);
3340 return -EINVAL;
3341 }
3342 }
3343 }
3344
1da177e4 3345#ifdef CONFIG_KMOD
d9d166c2
N		 3346 if (mddev->level != LEVEL_NONE)
3347 request_module("md-level-%d", mddev->level);
3348 else if (mddev->clevel[0])
3349 request_module("md-%s", mddev->clevel);
1da177e4
LT		 3350#endif
3351
3352 /*
3353 * Drop all container device buffers, from now on
3354 * the only valid external interface is through the md
3355 * device.
1da177e4 3356 */
d089c6af 3357 rdev_for_each(rdev, tmp, mddev) {
b2d444d7 3358 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3359 continue;
3360 sync_blockdev(rdev->bdev);
f98393a6 3361 invalidate_bdev(rdev->bdev);
f0d76d70
N		 3362
3363 /* perform some consistency tests on the device.
3364 * We don't want the data to overlap the metadata,
3365 * Internal Bitmap issues has handled elsewhere.
3366 */
3367 if (rdev->data_offset < rdev->sb_offset) {
3368 if (mddev->size &&
3369 rdev->data_offset + mddev->size*2
3370 > rdev->sb_offset*2) {
3371 printk("md: %s: data overlaps metadata\n",
3372 mdname(mddev));
3373 return -EINVAL;
3374 }
3375 } else {
3376 if (rdev->sb_offset*2 + rdev->sb_size/512
3377 > rdev->data_offset) {
3378 printk("md: %s: metadata overlaps data\n",
3379 mdname(mddev));
3380 return -EINVAL;
3381 }
3382 }
1da177e4
LT		 3383 }
3384
3385 md_probe(mddev->unit, NULL, NULL);
3386 disk = mddev->gendisk;
3387 if (!disk)
3388 return -ENOMEM;
3389
3390 spin_lock(&pers_lock);
d9d166c2 3391 pers = find_pers(mddev->level, mddev->clevel);
2604b703 3392 if (!pers || !try_module_get(pers->owner)) {
1da177e4 3393 spin_unlock(&pers_lock);
d9d166c2
N		 3394 if (mddev->level != LEVEL_NONE)
3395 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3396 mddev->level);
3397 else
3398 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3399 mddev->clevel);
1da177e4
LT		 3400 return -EINVAL;
3401 }
2604b703 3402 mddev->pers = pers;
1da177e4 3403 spin_unlock(&pers_lock);
d9d166c2
N		 3404 mddev->level = pers->level;
3405 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 3406
f6705578 3407 if (mddev->reshape_position != MaxSector &&
63c70c4f 3408 pers->start_reshape == NULL) {
f6705578
N		 3409 /* This personality cannot handle reshaping... */
3410 mddev->pers = NULL;
3411 module_put(pers->owner);
3412 return -EINVAL;
3413 }
3414
7dd5e7c3
N		 3415 if (pers->sync_request) {
3416 /* Warn if this is a potentially silly
3417 * configuration.
3418 */
3419 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3420 mdk_rdev_t *rdev2;
3421 struct list_head *tmp2;
3422 int warned = 0;
d089c6af
N		 3423 rdev_for_each(rdev, tmp, mddev) {
3424 rdev_for_each(rdev2, tmp2, mddev) {
7dd5e7c3
N		 3425 if (rdev < rdev2 &&
3426 rdev->bdev->bd_contains ==
3427 rdev2->bdev->bd_contains) {
3428 printk(KERN_WARNING
3429 "%s: WARNING: %s appears to be"
3430 " on the same physical disk as"
3431 " %s.\n",
3432 mdname(mddev),
3433 bdevname(rdev->bdev,b),
3434 bdevname(rdev2->bdev,b2));
3435 warned = 1;
3436 }
3437 }
3438 }
3439 if (warned)
3440 printk(KERN_WARNING
3441 "True protection against single-disk"
3442 " failure might be compromised.\n");
3443 }
3444
657390d2 3445 mddev->recovery = 0;
1da177e4 3446 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 3447 mddev->barriers_work = 1;
6ff8d8ec 3448 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 3449
f91de92e
N		 3450 if (start_readonly)
3451 mddev->ro = 2; /* read-only, but switch on first write */
3452
b15c2e57
N		 3453 err = mddev->pers->run(mddev);
3454 if (!err && mddev->pers->sync_request) {
3455 err = bitmap_create(mddev);
3456 if (err) {
3457 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3458 mdname(mddev), err);
3459 mddev->pers->stop(mddev);
3460 }
3461 }
1da177e4
LT		 3462 if (err) {
3463 printk(KERN_ERR "md: pers->run() failed ...\n");
3464 module_put(mddev->pers->owner);
3465 mddev->pers = NULL;
32a7627c
N		 3466 bitmap_destroy(mddev);
3467 return err;
1da177e4 3468 }
5e55e2f5
N		 3469 if (mddev->pers->sync_request) {
3470 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3471 printk(KERN_WARNING
3472 "md: cannot register extra attributes for %s\n",
3473 mdname(mddev));
3474 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 3475 mddev->ro = 0;
3476
1da177e4
LT		 3477 atomic_set(&mddev->writes_pending,0);
3478 mddev->safemode = 0;
3479 mddev->safemode_timer.function = md_safemode_timeout;
3480 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 3481 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 3482 mddev->in_sync = 1;
86e6ffdd 3483
d089c6af 3484 rdev_for_each(rdev, tmp, mddev)
86e6ffdd
N		 3485 if (rdev->raid_disk >= 0) {
3486 char nm[20];
3487 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 3488 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3489 printk("md: cannot register %s for %s\n",
3490 nm, mdname(mddev));
86e6ffdd 3491 }
1da177e4
LT		 3492
3493 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3494
850b2b42
N		 3495 if (mddev->flags)
3496 md_update_sb(mddev, 0);
1da177e4
LT		 3497
3498 set_capacity(disk, mddev->array_size<<1);
3499
3500 /* If we call blk_queue_make_request here, it will
3501 * re-initialise max_sectors etc which may have been
3502 * refined inside -> run. So just set the bits we need to set.
3503 * Most initialisation happended when we called
3504 * blk_queue_make_request(..., md_fail_request)
3505 * earlier.
3506 */
3507 mddev->queue->queuedata = mddev;
3508 mddev->queue->make_request_fn = mddev->pers->make_request;
3509
5fd6c1dc
N		 3510 /* If there is a partially-recovered drive we need to
3511 * start recovery here. If we leave it to md_check_recovery,
3512 * it will remove the drives and not do the right thing
3513 */
0b8c9de0 3514 if (mddev->degraded && !mddev->sync_thread) {
5fd6c1dc
N		 3515 struct list_head *rtmp;
3516 int spares = 0;
d089c6af 3517 rdev_for_each(rdev, rtmp, mddev)
5fd6c1dc
N		 3518 if (rdev->raid_disk >= 0 &&
3519 !test_bit(In_sync, &rdev->flags) &&
3520 !test_bit(Faulty, &rdev->flags))
3521 /* complete an interrupted recovery */
3522 spares++;
3523 if (spares && mddev->pers->sync_request) {
3524 mddev->recovery = 0;
3525 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3526 mddev->sync_thread = md_register_thread(md_do_sync,
3527 mddev,
3528 "%s_resync");
3529 if (!mddev->sync_thread) {
3530 printk(KERN_ERR "%s: could not start resync"
3531 " thread...\n",
3532 mdname(mddev));
3533 /* leave the spares where they are, it shouldn't hurt */
3534 mddev->recovery = 0;
0b8c9de0 3535 }
5fd6c1dc
N		 3536 }
3537 }
0b8c9de0
N		 3538 md_wakeup_thread(mddev->thread);
3539 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 3540
44ce6294 3541 mddev->changed = 1;
d7603b7e 3542 md_new_event(mddev);
edfaa7c3 3543 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
1da177e4
LT		 3544 return 0;
3545}
3546
3547static int restart_array(mddev_t *mddev)
3548{
3549 struct gendisk *disk = mddev->gendisk;
3550 int err;
3551
3552 /*
3553 * Complain if it has no devices
3554 */
3555 err = -ENXIO;
3556 if (list_empty(&mddev->disks))
3557 goto out;
3558
3559 if (mddev->pers) {
3560 err = -EBUSY;
3561 if (!mddev->ro)
3562 goto out;
3563
3564 mddev->safemode = 0;
3565 mddev->ro = 0;
3566 set_disk_ro(disk, 0);
3567
3568 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3569 mdname(mddev));
3570 /*
3571 * Kick recovery or resync if necessary
3572 */
3573 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3574 md_wakeup_thread(mddev->thread);
5fd6c1dc 3575 md_wakeup_thread(mddev->sync_thread);
1da177e4 3576 err = 0;
9e653b63 3577 } else
1da177e4 3578 err = -EINVAL;
1da177e4
LT		 3579
3580out:
3581 return err;
3582}
3583
acc55e22
N		 3584/* similar to deny_write_access, but accounts for our holding a reference
3585 * to the file ourselves */
3586static int deny_bitmap_write_access(struct file * file)
3587{
3588 struct inode *inode = file->f_mapping->host;
3589
3590 spin_lock(&inode->i_lock);
3591 if (atomic_read(&inode->i_writecount) > 1) {
3592 spin_unlock(&inode->i_lock);
3593 return -ETXTBSY;
3594 }
3595 atomic_set(&inode->i_writecount, -1);
3596 spin_unlock(&inode->i_lock);
3597
3598 return 0;
3599}
3600
3601static void restore_bitmap_write_access(struct file *file)
3602{
3603 struct inode *inode = file->f_mapping->host;
3604
3605 spin_lock(&inode->i_lock);
3606 atomic_set(&inode->i_writecount, 1);
3607 spin_unlock(&inode->i_lock);
3608}
3609
9e653b63
N		 3610/* mode:
3611 * 0 - completely stop and dis-assemble array
3612 * 1 - switch to readonly
3613 * 2 - stop but do not disassemble array
3614 */
3615static int do_md_stop(mddev_t * mddev, int mode)
1da177e4
LT		 3616{
3617 int err = 0;
3618 struct gendisk *disk = mddev->gendisk;
3619
3620 if (mddev->pers) {
3621 if (atomic_read(&mddev->active)>2) {
3622 printk("md: %s still in use.\n",mdname(mddev));
3623 return -EBUSY;
3624 }
3625
3626 if (mddev->sync_thread) {
5fd6c1dc 3627 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4
LT		 3628 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3629 md_unregister_thread(mddev->sync_thread);
3630 mddev->sync_thread = NULL;
3631 }
3632
3633 del_timer_sync(&mddev->safemode_timer);
3634
3635 invalidate_partition(disk, 0);
3636
9e653b63
N		 3637 switch(mode) {
3638 case 1: /* readonly */
1da177e4 3639 err = -ENXIO;
f91de92e 3640 if (mddev->ro==1)
1da177e4
LT		 3641 goto out;
3642 mddev->ro = 1;
9e653b63
N		 3643 break;
3644 case 0: /* disassemble */
3645 case 2: /* stop */
6b8b3e8a 3646 bitmap_flush(mddev);
a9701a30 3647 md_super_wait(mddev);
1da177e4
LT		 3648 if (mddev->ro)
3649 set_disk_ro(disk, 0);
3650 blk_queue_make_request(mddev->queue, md_fail_request);
3651 mddev->pers->stop(mddev);
d1b5380c
N		 3652 mddev->queue->merge_bvec_fn = NULL;
3653 mddev->queue->unplug_fn = NULL;
041ae52e 3654 mddev->queue->backing_dev_info.congested_fn = NULL;
411036fa
N		 3655 if (mddev->pers->sync_request)
3656 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3657
1da177e4
LT		 3658 module_put(mddev->pers->owner);
3659 mddev->pers = NULL;
0d4ca600
N		 3660
3661 set_capacity(disk, 0);
44ce6294 3662 mddev->changed = 1;
0d4ca600 3663
1da177e4
LT		 3664 if (mddev->ro)
3665 mddev->ro = 0;
3666 }
850b2b42 3667 if (!mddev->in_sync || mddev->flags) {
1da177e4
LT		 3668 /* mark array as shutdown cleanly */
3669 mddev->in_sync = 1;
850b2b42 3670 md_update_sb(mddev, 1);
1da177e4 3671 }
9e653b63 3672 if (mode == 1)
1da177e4 3673 set_disk_ro(disk, 1);
5fd6c1dc 3674 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4 3675 }
32a7627c 3676
1da177e4
LT		 3677 /*
3678 * Free resources if final stop
3679 */
9e653b63 3680 if (mode == 0) {
86e6ffdd
N		 3681 mdk_rdev_t *rdev;
3682 struct list_head *tmp;
0d4ca600 3683
1da177e4
LT		 3684 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3685
978f946b
N		 3686 bitmap_destroy(mddev);
3687 if (mddev->bitmap_file) {
acc55e22 3688 restore_bitmap_write_access(mddev->bitmap_file);
978f946b
N		 3689 fput(mddev->bitmap_file);
3690 mddev->bitmap_file = NULL;
3691 }
3692 mddev->bitmap_offset = 0;
3693
d089c6af 3694 rdev_for_each(rdev, tmp, mddev)
86e6ffdd
N		 3695 if (rdev->raid_disk >= 0) {
3696 char nm[20];
3697 sprintf(nm, "rd%d", rdev->raid_disk);
3698 sysfs_remove_link(&mddev->kobj, nm);
3699 }
3700
177a99b2 3701 /* make sure all md_delayed_delete calls have finished */
5792a285
N		 3702 flush_scheduled_work();
3703
1da177e4
LT		 3704 export_array(mddev);
3705
3706 mddev->array_size = 0;
9e653b63
N		 3707 mddev->size = 0;
3708 mddev->raid_disks = 0;
a94213b1 3709 mddev->recovery_cp = 0;
c6207277 3710 mddev->resync_max = MaxSector;
08a02ecd 3711 mddev->reshape_position = MaxSector;
e691063a 3712 mddev->external = 0;
1ec4a939 3713 mddev->persistent = 0;
9e653b63 3714
a8a55c38 3715 } else if (mddev->pers)
1da177e4
LT		 3716 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3717 mdname(mddev));
3718 err = 0;
d7603b7e 3719 md_new_event(mddev);
1da177e4
LT		 3720out:
3721 return err;
3722}
3723
fdee8ae4 3724#ifndef MODULE
1da177e4
LT		 3725static void autorun_array(mddev_t *mddev)
3726{
3727 mdk_rdev_t *rdev;
3728 struct list_head *tmp;
3729 int err;
3730
a757e64c 3731 if (list_empty(&mddev->disks))
1da177e4 3732 return;
1da177e4
LT		 3733
3734 printk(KERN_INFO "md: running: ");
3735
d089c6af 3736 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 3737 char b[BDEVNAME_SIZE];
3738 printk("<%s>", bdevname(rdev->bdev,b));
3739 }
3740 printk("\n");
3741
3742 err = do_md_run (mddev);
3743 if (err) {
3744 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3745 do_md_stop (mddev, 0);
3746 }
3747}
3748
3749/*
3750 * lets try to run arrays based on all disks that have arrived
3751 * until now. (those are in pending_raid_disks)
3752 *
3753 * the method: pick the first pending disk, collect all disks with
3754 * the same UUID, remove all from the pending list and put them into
3755 * the 'same_array' list. Then order this list based on superblock
3756 * update time (freshest comes first), kick out 'old' disks and
3757 * compare superblocks. If everything's fine then run it.
3758 *
3759 * If "unit" is allocated, then bump its reference count
3760 */
3761static void autorun_devices(int part)
3762{
1da177e4
LT		 3763 struct list_head *tmp;
3764 mdk_rdev_t *rdev0, *rdev;
3765 mddev_t *mddev;
3766 char b[BDEVNAME_SIZE];
3767
3768 printk(KERN_INFO "md: autorun ...\n");
3769 while (!list_empty(&pending_raid_disks)) {
e8703fe1 3770 int unit;
1da177e4 3771 dev_t dev;
ad01c9e3 3772 LIST_HEAD(candidates);
1da177e4
LT		 3773 rdev0 = list_entry(pending_raid_disks.next,
3774 mdk_rdev_t, same_set);
3775
3776 printk(KERN_INFO "md: considering %s ...\n",
3777 bdevname(rdev0->bdev,b));
3778 INIT_LIST_HEAD(&candidates);
73c34431 3779 rdev_for_each_list(rdev, tmp, pending_raid_disks)
1da177e4
LT		 3780 if (super_90_load(rdev, rdev0, 0) >= 0) {
3781 printk(KERN_INFO "md: adding %s ...\n",
3782 bdevname(rdev->bdev,b));
3783 list_move(&rdev->same_set, &candidates);
3784 }
3785 /*
3786 * now we have a set of devices, with all of them having
3787 * mostly sane superblocks. It's time to allocate the
3788 * mddev.
3789 */
e8703fe1
N		 3790 if (part) {
3791 dev = MKDEV(mdp_major,
3792 rdev0->preferred_minor << MdpMinorShift);
3793 unit = MINOR(dev) >> MdpMinorShift;
3794 } else {
3795 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3796 unit = MINOR(dev);
3797 }
3798 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 3799 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3800 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3801 break;
3802 }
1da177e4
LT		 3803
3804 md_probe(dev, NULL, NULL);
3805 mddev = mddev_find(dev);
3806 if (!mddev) {
3807 printk(KERN_ERR
3808 "md: cannot allocate memory for md drive.\n");
3809 break;
3810 }
3811 if (mddev_lock(mddev))
3812 printk(KERN_WARNING "md: %s locked, cannot run\n",
3813 mdname(mddev));
3814 else if (mddev->raid_disks || mddev->major_version
3815 || !list_empty(&mddev->disks)) {
3816 printk(KERN_WARNING
3817 "md: %s already running, cannot run %s\n",
3818 mdname(mddev), bdevname(rdev0->bdev,b));
3819 mddev_unlock(mddev);
3820 } else {
3821 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 3822 mddev->persistent = 1;
73c34431 3823 rdev_for_each_list(rdev, tmp, candidates) {
1da177e4
LT		 3824 list_del_init(&rdev->same_set);
3825 if (bind_rdev_to_array(rdev, mddev))
3826 export_rdev(rdev);
3827 }
3828 autorun_array(mddev);
3829 mddev_unlock(mddev);
3830 }
3831 /* on success, candidates will be empty, on error
3832 * it won't...
3833 */
73c34431 3834 rdev_for_each_list(rdev, tmp, candidates)
1da177e4
LT		 3835 export_rdev(rdev);
3836 mddev_put(mddev);
3837 }
3838 printk(KERN_INFO "md: ... autorun DONE.\n");
3839}
fdee8ae4 3840#endif /* !MODULE */
1da177e4 3841
1da177e4
LT		 3842static int get_version(void __user * arg)
3843{
3844 mdu_version_t ver;
3845
3846 ver.major = MD_MAJOR_VERSION;
3847 ver.minor = MD_MINOR_VERSION;
3848 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3849
3850 if (copy_to_user(arg, &ver, sizeof(ver)))
3851 return -EFAULT;
3852
3853 return 0;
3854}
3855
3856static int get_array_info(mddev_t * mddev, void __user * arg)
3857{
3858 mdu_array_info_t info;
3859 int nr,working,active,failed,spare;
3860 mdk_rdev_t *rdev;
3861 struct list_head *tmp;
3862
3863 nr=working=active=failed=spare=0;
d089c6af 3864 rdev_for_each(rdev, tmp, mddev) {
1da177e4 3865 nr++;
b2d444d7 3866 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3867 failed++;
3868 else {
3869 working++;
b2d444d7 3870 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 3871 active++;
3872 else
3873 spare++;
3874 }
3875 }
3876
3877 info.major_version = mddev->major_version;
3878 info.minor_version = mddev->minor_version;
3879 info.patch_version = MD_PATCHLEVEL_VERSION;
3880 info.ctime = mddev->ctime;
3881 info.level = mddev->level;
3882 info.size = mddev->size;
284ae7ca
N		 3883 if (info.size != mddev->size) /* overflow */
3884 info.size = -1;
1da177e4
LT		 3885 info.nr_disks = nr;
3886 info.raid_disks = mddev->raid_disks;
3887 info.md_minor = mddev->md_minor;
3888 info.not_persistent= !mddev->persistent;
3889
3890 info.utime = mddev->utime;
3891 info.state = 0;
3892 if (mddev->in_sync)
3893 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 3894 if (mddev->bitmap && mddev->bitmap_offset)
3895 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 3896 info.active_disks = active;
3897 info.working_disks = working;
3898 info.failed_disks = failed;
3899 info.spare_disks = spare;
3900
3901 info.layout = mddev->layout;
3902 info.chunk_size = mddev->chunk_size;
3903
3904 if (copy_to_user(arg, &info, sizeof(info)))
3905 return -EFAULT;
3906
3907 return 0;
3908}
3909
87162a28 3910static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 3911{
3912 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3913 char *ptr, *buf = NULL;
3914 int err = -ENOMEM;
3915
2a2275d6
N		 3916 md_allow_write(mddev);
3917
32a7627c
N		 3918 file = kmalloc(sizeof(*file), GFP_KERNEL);
3919 if (!file)
3920 goto out;
3921
3922 /* bitmap disabled, zero the first byte and copy out */
3923 if (!mddev->bitmap || !mddev->bitmap->file) {
3924 file->pathname[0] = '\0';
3925 goto copy_out;
3926 }
3927
3928 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3929 if (!buf)
3930 goto out;
3931
3932 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3933 if (!ptr)
3934 goto out;
3935
3936 strcpy(file->pathname, ptr);
3937
3938copy_out:
3939 err = 0;
3940 if (copy_to_user(arg, file, sizeof(*file)))
3941 err = -EFAULT;
3942out:
3943 kfree(buf);
3944 kfree(file);
3945 return err;
3946}
3947
1da177e4
LT		 3948static int get_disk_info(mddev_t * mddev, void __user * arg)
3949{
3950 mdu_disk_info_t info;
3951 unsigned int nr;
3952 mdk_rdev_t *rdev;
3953
3954 if (copy_from_user(&info, arg, sizeof(info)))
3955 return -EFAULT;
3956
3957 nr = info.number;
3958
3959 rdev = find_rdev_nr(mddev, nr);
3960 if (rdev) {
3961 info.major = MAJOR(rdev->bdev->bd_dev);
3962 info.minor = MINOR(rdev->bdev->bd_dev);
3963 info.raid_disk = rdev->raid_disk;
3964 info.state = 0;
b2d444d7 3965 if (test_bit(Faulty, &rdev->flags))
1da177e4 3966 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 3967 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 3968 info.state |= (1<<MD_DISK_ACTIVE);
3969 info.state |= (1<<MD_DISK_SYNC);
3970 }
8ddf9efe
N		 3971 if (test_bit(WriteMostly, &rdev->flags))
3972 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 3973 } else {
3974 info.major = info.minor = 0;
3975 info.raid_disk = -1;
3976 info.state = (1<<MD_DISK_REMOVED);
3977 }
3978
3979 if (copy_to_user(arg, &info, sizeof(info)))
3980 return -EFAULT;
3981
3982 return 0;
3983}
3984
3985static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3986{
3987 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3988 mdk_rdev_t *rdev;
3989 dev_t dev = MKDEV(info->major,info->minor);
3990
3991 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3992 return -EOVERFLOW;
3993
3994 if (!mddev->raid_disks) {
3995 int err;
3996 /* expecting a device which has a superblock */
3997 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3998 if (IS_ERR(rdev)) {
3999 printk(KERN_WARNING
4000 "md: md_import_device returned %ld\n",
4001 PTR_ERR(rdev));
4002 return PTR_ERR(rdev);
4003 }
4004 if (!list_empty(&mddev->disks)) {
4005 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4006 mdk_rdev_t, same_set);
4007 int err = super_types[mddev->major_version]
4008 .load_super(rdev, rdev0, mddev->minor_version);
4009 if (err < 0) {
4010 printk(KERN_WARNING
4011 "md: %s has different UUID to %s\n",
4012 bdevname(rdev->bdev,b),
4013 bdevname(rdev0->bdev,b2));
4014 export_rdev(rdev);
4015 return -EINVAL;
4016 }
4017 }
4018 err = bind_rdev_to_array(rdev, mddev);
4019 if (err)
4020 export_rdev(rdev);
4021 return err;
4022 }
4023
4024 /*
4025 * add_new_disk can be used once the array is assembled
4026 * to add "hot spares". They must already have a superblock
4027 * written
4028 */
4029 if (mddev->pers) {
4030 int err;
4031 if (!mddev->pers->hot_add_disk) {
4032 printk(KERN_WARNING
4033 "%s: personality does not support diskops!\n",
4034 mdname(mddev));
4035 return -EINVAL;
4036 }
7b1e35f6
N		 4037 if (mddev->persistent)
4038 rdev = md_import_device(dev, mddev->major_version,
4039 mddev->minor_version);
4040 else
4041 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 4042 if (IS_ERR(rdev)) {
4043 printk(KERN_WARNING
4044 "md: md_import_device returned %ld\n",
4045 PTR_ERR(rdev));
4046 return PTR_ERR(rdev);
4047 }
41158c7e
N		 4048 /* set save_raid_disk if appropriate */
4049 if (!mddev->persistent) {
4050 if (info->state & (1<<MD_DISK_SYNC) &&
4051 info->raid_disk < mddev->raid_disks)
4052 rdev->raid_disk = info->raid_disk;
4053 else
4054 rdev->raid_disk = -1;
4055 } else
4056 super_types[mddev->major_version].
4057 validate_super(mddev, rdev);
4058 rdev->saved_raid_disk = rdev->raid_disk;
4059
b2d444d7 4060 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 4061 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4062 set_bit(WriteMostly, &rdev->flags);
4063
1da177e4
LT		 4064 rdev->raid_disk = -1;
4065 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 4066 if (!err && !mddev->pers->hot_remove_disk) {
4067 /* If there is hot_add_disk but no hot_remove_disk
4068 * then added disks for geometry changes,
4069 * and should be added immediately.
4070 */
4071 super_types[mddev->major_version].
4072 validate_super(mddev, rdev);
4073 err = mddev->pers->hot_add_disk(mddev, rdev);
4074 if (err)
4075 unbind_rdev_from_array(rdev);
4076 }
1da177e4
LT		 4077 if (err)
4078 export_rdev(rdev);
c361777f 4079
17571284 4080 md_update_sb(mddev, 1);
c361777f 4081 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 4082 md_wakeup_thread(mddev->thread);
1da177e4
LT		 4083 return err;
4084 }
4085
4086 /* otherwise, add_new_disk is only allowed
4087 * for major_version==0 superblocks
4088 */
4089 if (mddev->major_version != 0) {
4090 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4091 mdname(mddev));
4092 return -EINVAL;
4093 }
4094
4095 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4096 int err;
4097 rdev = md_import_device (dev, -1, 0);
4098 if (IS_ERR(rdev)) {
4099 printk(KERN_WARNING
4100 "md: error, md_import_device() returned %ld\n",
4101 PTR_ERR(rdev));
4102 return PTR_ERR(rdev);
4103 }
4104 rdev->desc_nr = info->number;
4105 if (info->raid_disk < mddev->raid_disks)
4106 rdev->raid_disk = info->raid_disk;
4107 else
4108 rdev->raid_disk = -1;
4109
1da177e4 4110 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 4111 if (info->state & (1<<MD_DISK_SYNC))
4112 set_bit(In_sync, &rdev->flags);
1da177e4 4113
8ddf9efe
N		 4114 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4115 set_bit(WriteMostly, &rdev->flags);
4116
1da177e4
LT		 4117 if (!mddev->persistent) {
4118 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4119 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4120 } else
4121 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4122 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4123
2bf071bf
N		 4124 err = bind_rdev_to_array(rdev, mddev);
4125 if (err) {
4126 export_rdev(rdev);
4127 return err;
4128 }
1da177e4
LT		 4129 }
4130
4131 return 0;
4132}
4133
4134static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4135{
4136 char b[BDEVNAME_SIZE];
4137 mdk_rdev_t *rdev;
4138
4139 if (!mddev->pers)
4140 return -ENODEV;
4141
4142 rdev = find_rdev(mddev, dev);
4143 if (!rdev)
4144 return -ENXIO;
4145
4146 if (rdev->raid_disk >= 0)
4147 goto busy;
4148
4149 kick_rdev_from_array(rdev);
850b2b42 4150 md_update_sb(mddev, 1);
d7603b7e 4151 md_new_event(mddev);
1da177e4
LT		 4152
4153 return 0;
4154busy:
fdefa4d8 4155 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 4156 bdevname(rdev->bdev,b), mdname(mddev));
4157 return -EBUSY;
4158}
4159
4160static int hot_add_disk(mddev_t * mddev, dev_t dev)
4161{
4162 char b[BDEVNAME_SIZE];
4163 int err;
4164 unsigned int size;
4165 mdk_rdev_t *rdev;
4166
4167 if (!mddev->pers)
4168 return -ENODEV;
4169
4170 if (mddev->major_version != 0) {
4171 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4172 " version-0 superblocks.\n",
4173 mdname(mddev));
4174 return -EINVAL;
4175 }
4176 if (!mddev->pers->hot_add_disk) {
4177 printk(KERN_WARNING
4178 "%s: personality does not support diskops!\n",
4179 mdname(mddev));
4180 return -EINVAL;
4181 }
4182
4183 rdev = md_import_device (dev, -1, 0);
4184 if (IS_ERR(rdev)) {
4185 printk(KERN_WARNING
4186 "md: error, md_import_device() returned %ld\n",
4187 PTR_ERR(rdev));
4188 return -EINVAL;
4189 }
4190
4191 if (mddev->persistent)
4192 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4193 else
4194 rdev->sb_offset =
4195 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4196
4197 size = calc_dev_size(rdev, mddev->chunk_size);
4198 rdev->size = size;
4199
b2d444d7 4200 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4201 printk(KERN_WARNING
4202 "md: can not hot-add faulty %s disk to %s!\n",
4203 bdevname(rdev->bdev,b), mdname(mddev));
4204 err = -EINVAL;
4205 goto abort_export;
4206 }
b2d444d7 4207 clear_bit(In_sync, &rdev->flags);
1da177e4 4208 rdev->desc_nr = -1;
5842730d 4209 rdev->saved_raid_disk = -1;
2bf071bf
N		 4210 err = bind_rdev_to_array(rdev, mddev);
4211 if (err)
4212 goto abort_export;
1da177e4
LT		 4213
4214 /*
4215 * The rest should better be atomic, we can have disk failures
4216 * noticed in interrupt contexts ...
4217 */
4218
4219 if (rdev->desc_nr == mddev->max_disks) {
4220 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4221 mdname(mddev));
4222 err = -EBUSY;
4223 goto abort_unbind_export;
4224 }
4225
4226 rdev->raid_disk = -1;
4227
850b2b42 4228 md_update_sb(mddev, 1);
1da177e4
LT		 4229
4230 /*
4231 * Kick recovery, maybe this spare has to be added to the
4232 * array immediately.
4233 */
4234 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4235 md_wakeup_thread(mddev->thread);
d7603b7e 4236 md_new_event(mddev);
1da177e4
LT		 4237 return 0;
4238
4239abort_unbind_export:
4240 unbind_rdev_from_array(rdev);
4241
4242abort_export:
4243 export_rdev(rdev);
4244 return err;
4245}
4246
32a7627c
N		 4247static int set_bitmap_file(mddev_t *mddev, int fd)
4248{
4249 int err;
4250
36fa3063
N		 4251 if (mddev->pers) {
4252 if (!mddev->pers->quiesce)
4253 return -EBUSY;
4254 if (mddev->recovery || mddev->sync_thread)
4255 return -EBUSY;
4256 /* we should be able to change the bitmap.. */
4257 }
32a7627c 4258
32a7627c 4259
36fa3063
N		 4260 if (fd >= 0) {
4261 if (mddev->bitmap)
4262 return -EEXIST; /* cannot add when bitmap is present */
4263 mddev->bitmap_file = fget(fd);
32a7627c 4264
36fa3063
N		 4265 if (mddev->bitmap_file == NULL) {
4266 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4267 mdname(mddev));
4268 return -EBADF;
4269 }
4270
4271 err = deny_bitmap_write_access(mddev->bitmap_file);
4272 if (err) {
4273 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4274 mdname(mddev));
4275 fput(mddev->bitmap_file);
4276 mddev->bitmap_file = NULL;
4277 return err;
4278 }
a654b9d8 4279 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 4280 } else if (mddev->bitmap == NULL)
4281 return -ENOENT; /* cannot remove what isn't there */
4282 err = 0;
4283 if (mddev->pers) {
4284 mddev->pers->quiesce(mddev, 1);
4285 if (fd >= 0)
4286 err = bitmap_create(mddev);
d7375ab3 4287 if (fd < 0 || err) {
36fa3063 4288 bitmap_destroy(mddev);
d7375ab3
N		 4289 fd = -1; /* make sure to put the file */
4290 }
36fa3063 4291 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 4292 }
4293 if (fd < 0) {
acc55e22
N		 4294 if (mddev->bitmap_file) {
4295 restore_bitmap_write_access(mddev->bitmap_file);
36fa3063 4296 fput(mddev->bitmap_file);
acc55e22 4297 }
36fa3063
N		 4298 mddev->bitmap_file = NULL;
4299 }
4300
32a7627c
N		 4301 return err;
4302}
4303
1da177e4
LT		 4304/*
4305 * set_array_info is used two different ways
4306 * The original usage is when creating a new array.
4307 * In this usage, raid_disks is > 0 and it together with
4308 * level, size, not_persistent,layout,chunksize determine the
4309 * shape of the array.
4310 * This will always create an array with a type-0.90.0 superblock.
4311 * The newer usage is when assembling an array.
4312 * In this case raid_disks will be 0, and the major_version field is
4313 * use to determine which style super-blocks are to be found on the devices.
4314 * The minor and patch _version numbers are also kept incase the
4315 * super_block handler wishes to interpret them.
4316 */
4317static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4318{
4319
4320 if (info->raid_disks == 0) {
4321 /* just setting version number for superblock loading */
4322 if (info->major_version < 0 ||
50511da3 4323 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 4324 super_types[info->major_version].name == NULL) {
4325 /* maybe try to auto-load a module? */
4326 printk(KERN_INFO
4327 "md: superblock version %d not known\n",
4328 info->major_version);
4329 return -EINVAL;
4330 }
4331 mddev->major_version = info->major_version;
4332 mddev->minor_version = info->minor_version;
4333 mddev->patch_version = info->patch_version;
3f9d7b0d 4334 mddev->persistent = !info->not_persistent;
1da177e4
LT		 4335 return 0;
4336 }
4337 mddev->major_version = MD_MAJOR_VERSION;
4338 mddev->minor_version = MD_MINOR_VERSION;
4339 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4340 mddev->ctime = get_seconds();
4341
4342 mddev->level = info->level;
17115e03 4343 mddev->clevel[0] = 0;
1da177e4
LT		 4344 mddev->size = info->size;
4345 mddev->raid_disks = info->raid_disks;
4346 /* don't set md_minor, it is determined by which /dev/md* was
4347 * openned
4348 */
4349 if (info->state & (1<<MD_SB_CLEAN))
4350 mddev->recovery_cp = MaxSector;
4351 else
4352 mddev->recovery_cp = 0;
4353 mddev->persistent = ! info->not_persistent;
e691063a 4354 mddev->external = 0;
1da177e4
LT		 4355
4356 mddev->layout = info->layout;
4357 mddev->chunk_size = info->chunk_size;
4358
4359 mddev->max_disks = MD_SB_DISKS;
4360
e691063a
N		 4361 if (mddev->persistent)
4362 mddev->flags = 0;
850b2b42 4363 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 4364
b2a2703c
N		 4365 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4366 mddev->bitmap_offset = 0;
4367
f6705578
N		 4368 mddev->reshape_position = MaxSector;
4369
1da177e4
LT		 4370 /*
4371 * Generate a 128 bit UUID
4372 */
4373 get_random_bytes(mddev->uuid, 16);
4374
f6705578
N		 4375 mddev->new_level = mddev->level;
4376 mddev->new_chunk = mddev->chunk_size;
4377 mddev->new_layout = mddev->layout;
4378 mddev->delta_disks = 0;
4379
1da177e4
LT		 4380 return 0;
4381}
4382
a35b0d69
N		 4383static int update_size(mddev_t *mddev, unsigned long size)
4384{
4385 mdk_rdev_t * rdev;
4386 int rv;
4387 struct list_head *tmp;
8ddeeae5 4388 int fit = (size == 0);
a35b0d69
N		 4389
4390 if (mddev->pers->resize == NULL)
4391 return -EINVAL;
4392 /* The "size" is the amount of each device that is used.
4393 * This can only make sense for arrays with redundancy.
4394 * linear and raid0 always use whatever space is available
4395 * We can only consider changing the size if no resync
4396 * or reconstruction is happening, and if the new size
4397 * is acceptable. It must fit before the sb_offset or,
4398 * if that is <data_offset, it must fit before the
4399 * size of each device.
4400 * If size is zero, we find the largest size that fits.
4401 */
4402 if (mddev->sync_thread)
4403 return -EBUSY;
d089c6af 4404 rdev_for_each(rdev, tmp, mddev) {
a35b0d69 4405 sector_t avail;
01ab5662
N		 4406 avail = rdev->size * 2;
4407
a35b0d69
N		 4408 if (fit && (size == 0 || size > avail/2))
4409 size = avail/2;
4410 if (avail < ((sector_t)size << 1))
4411 return -ENOSPC;
4412 }
4413 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4414 if (!rv) {
4415 struct block_device *bdev;
4416
4417 bdev = bdget_disk(mddev->gendisk, 0);
4418 if (bdev) {
1b1dcc1b 4419 mutex_lock(&bdev->bd_inode->i_mutex);
6d89332b 4420 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
1b1dcc1b 4421 mutex_unlock(&bdev->bd_inode->i_mutex);
a35b0d69
N		 4422 bdput(bdev);
4423 }
4424 }
4425 return rv;
4426}
4427
da943b99
N		 4428static int update_raid_disks(mddev_t *mddev, int raid_disks)
4429{
4430 int rv;
4431 /* change the number of raid disks */
63c70c4f 4432 if (mddev->pers->check_reshape == NULL)
da943b99
N		 4433 return -EINVAL;
4434 if (raid_disks <= 0 ||
4435 raid_disks >= mddev->max_disks)
4436 return -EINVAL;
63c70c4f 4437 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 4438 return -EBUSY;
63c70c4f
N		 4439 mddev->delta_disks = raid_disks - mddev->raid_disks;
4440
4441 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 4442 return rv;
4443}
4444
4445
1da177e4
LT		 4446/*
4447 * update_array_info is used to change the configuration of an
4448 * on-line array.
4449 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4450 * fields in the info are checked against the array.
4451 * Any differences that cannot be handled will cause an error.
4452 * Normally, only one change can be managed at a time.
4453 */
4454static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4455{
4456 int rv = 0;
4457 int cnt = 0;
36fa3063
N		 4458 int state = 0;
4459
4460 /* calculate expected state,ignoring low bits */
4461 if (mddev->bitmap && mddev->bitmap_offset)
4462 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 4463
4464 if (mddev->major_version != info->major_version ||
4465 mddev->minor_version != info->minor_version ||
4466/* mddev->patch_version != info->patch_version || */
4467 mddev->ctime != info->ctime ||
4468 mddev->level != info->level ||
4469/* mddev->layout != info->layout || */
4470 !mddev->persistent != info->not_persistent||
36fa3063
N		 4471 mddev->chunk_size != info->chunk_size ||
4472 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4473 ((state^info->state) & 0xfffffe00)
4474 )
1da177e4
LT		 4475 return -EINVAL;
4476 /* Check there is only one change */
284ae7ca 4477 if (info->size >= 0 && mddev->size != info->size) cnt++;
1da177e4
LT		 4478 if (mddev->raid_disks != info->raid_disks) cnt++;
4479 if (mddev->layout != info->layout) cnt++;
36fa3063 4480 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 4481 if (cnt == 0) return 0;
4482 if (cnt > 1) return -EINVAL;
4483
4484 if (mddev->layout != info->layout) {
4485 /* Change layout
4486 * we don't need to do anything at the md level, the
4487 * personality will take care of it all.
4488 */
4489 if (mddev->pers->reconfig == NULL)
4490 return -EINVAL;
4491 else
4492 return mddev->pers->reconfig(mddev, info->layout, -1);
4493 }
284ae7ca 4494 if (info->size >= 0 && mddev->size != info->size)
a35b0d69
N		 4495 rv = update_size(mddev, info->size);
4496
da943b99
N		 4497 if (mddev->raid_disks != info->raid_disks)
4498 rv = update_raid_disks(mddev, info->raid_disks);
4499
36fa3063
N		 4500 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4501 if (mddev->pers->quiesce == NULL)
4502 return -EINVAL;
4503 if (mddev->recovery || mddev->sync_thread)
4504 return -EBUSY;
4505 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4506 /* add the bitmap */
4507 if (mddev->bitmap)
4508 return -EEXIST;
4509 if (mddev->default_bitmap_offset == 0)
4510 return -EINVAL;
4511 mddev->bitmap_offset = mddev->default_bitmap_offset;
4512 mddev->pers->quiesce(mddev, 1);
4513 rv = bitmap_create(mddev);
4514 if (rv)
4515 bitmap_destroy(mddev);
4516 mddev->pers->quiesce(mddev, 0);
4517 } else {
4518 /* remove the bitmap */
4519 if (!mddev->bitmap)
4520 return -ENOENT;
4521 if (mddev->bitmap->file)
4522 return -EINVAL;
4523 mddev->pers->quiesce(mddev, 1);
4524 bitmap_destroy(mddev);
4525 mddev->pers->quiesce(mddev, 0);
4526 mddev->bitmap_offset = 0;
4527 }
4528 }
850b2b42 4529 md_update_sb(mddev, 1);
1da177e4
LT		 4530 return rv;
4531}
4532
4533static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4534{
4535 mdk_rdev_t *rdev;
4536
4537 if (mddev->pers == NULL)
4538 return -ENODEV;
4539
4540 rdev = find_rdev(mddev, dev);
4541 if (!rdev)
4542 return -ENODEV;
4543
4544 md_error(mddev, rdev);
4545 return 0;
4546}
4547
a885c8c4
CH		 4548static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4549{
4550 mddev_t *mddev = bdev->bd_disk->private_data;
4551
4552 geo->heads = 2;
4553 geo->sectors = 4;
4554 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4555 return 0;
4556}
4557
1da177e4
LT		 4558static int md_ioctl(struct inode *inode, struct file *file,
4559 unsigned int cmd, unsigned long arg)
4560{
4561 int err = 0;
4562 void __user *argp = (void __user *)arg;
1da177e4
LT		 4563 mddev_t *mddev = NULL;
4564
4565 if (!capable(CAP_SYS_ADMIN))
4566 return -EACCES;
4567
4568 /*
4569 * Commands dealing with the RAID driver but not any
4570 * particular array:
4571 */
4572 switch (cmd)
4573 {
4574 case RAID_VERSION:
4575 err = get_version(argp);
4576 goto done;
4577
4578 case PRINT_RAID_DEBUG:
4579 err = 0;
4580 md_print_devices();
4581 goto done;
4582
4583#ifndef MODULE
4584 case RAID_AUTORUN:
4585 err = 0;
4586 autostart_arrays(arg);
4587 goto done;
4588#endif
4589 default:;
4590 }
4591
4592 /*
4593 * Commands creating/starting a new array:
4594 */
4595
4596 mddev = inode->i_bdev->bd_disk->private_data;
4597
4598 if (!mddev) {
4599 BUG();
4600 goto abort;
4601 }
4602
1da177e4
LT		 4603 err = mddev_lock(mddev);
4604 if (err) {
4605 printk(KERN_INFO
4606 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4607 err, cmd);
4608 goto abort;
4609 }
4610
4611 switch (cmd)
4612 {
4613 case SET_ARRAY_INFO:
4614 {
4615 mdu_array_info_t info;
4616 if (!arg)
4617 memset(&info, 0, sizeof(info));
4618 else if (copy_from_user(&info, argp, sizeof(info))) {
4619 err = -EFAULT;
4620 goto abort_unlock;
4621 }
4622 if (mddev->pers) {
4623 err = update_array_info(mddev, &info);
4624 if (err) {
4625 printk(KERN_WARNING "md: couldn't update"
4626 " array info. %d\n", err);
4627 goto abort_unlock;
4628 }
4629 goto done_unlock;
4630 }
4631 if (!list_empty(&mddev->disks)) {
4632 printk(KERN_WARNING
4633 "md: array %s already has disks!\n",
4634 mdname(mddev));
4635 err = -EBUSY;
4636 goto abort_unlock;
4637 }
4638 if (mddev->raid_disks) {
4639 printk(KERN_WARNING
4640 "md: array %s already initialised!\n",
4641 mdname(mddev));
4642 err = -EBUSY;
4643 goto abort_unlock;
4644 }
4645 err = set_array_info(mddev, &info);
4646 if (err) {
4647 printk(KERN_WARNING "md: couldn't set"
4648 " array info. %d\n", err);
4649 goto abort_unlock;
4650 }
4651 }
4652 goto done_unlock;
4653
4654 default:;
4655 }
4656
4657 /*
4658 * Commands querying/configuring an existing array:
4659 */
32a7627c 4660 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 4661 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 4662 if ((!mddev->raid_disks && !mddev->external)
4663 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4664 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4665 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 4666 err = -ENODEV;
4667 goto abort_unlock;
4668 }
4669
4670 /*
4671 * Commands even a read-only array can execute:
4672 */
4673 switch (cmd)
4674 {
4675 case GET_ARRAY_INFO:
4676 err = get_array_info(mddev, argp);
4677 goto done_unlock;
4678
32a7627c 4679 case GET_BITMAP_FILE:
87162a28 4680 err = get_bitmap_file(mddev, argp);
32a7627c
N		 4681 goto done_unlock;
4682
1da177e4
LT		 4683 case GET_DISK_INFO:
4684 err = get_disk_info(mddev, argp);
4685 goto done_unlock;
4686
4687 case RESTART_ARRAY_RW:
4688 err = restart_array(mddev);
4689 goto done_unlock;
4690
4691 case STOP_ARRAY:
4692 err = do_md_stop (mddev, 0);
4693 goto done_unlock;
4694
4695 case STOP_ARRAY_RO:
4696 err = do_md_stop (mddev, 1);
4697 goto done_unlock;
4698
4699 /*
4700 * We have a problem here : there is no easy way to give a CHS
4701 * virtual geometry. We currently pretend that we have a 2 heads
4702 * 4 sectors (with a BIG number of cylinders...). This drives
4703 * dosfs just mad... ;-)
4704 */
1da177e4
LT		 4705 }
4706
4707 /*
4708 * The remaining ioctls are changing the state of the
f91de92e
N		 4709 * superblock, so we do not allow them on read-only arrays.
4710 * However non-MD ioctls (e.g. get-size) will still come through
4711 * here and hit the 'default' below, so only disallow
4712 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 4713 */
f91de92e
N		 4714 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4715 mddev->ro && mddev->pers) {
4716 if (mddev->ro == 2) {
4717 mddev->ro = 0;
4718 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4719 md_wakeup_thread(mddev->thread);
4720
4721 } else {
4722 err = -EROFS;
4723 goto abort_unlock;
4724 }
1da177e4
LT		 4725 }
4726
4727 switch (cmd)
4728 {
4729 case ADD_NEW_DISK:
4730 {
4731 mdu_disk_info_t info;
4732 if (copy_from_user(&info, argp, sizeof(info)))
4733 err = -EFAULT;
4734 else
4735 err = add_new_disk(mddev, &info);
4736 goto done_unlock;
4737 }
4738
4739 case HOT_REMOVE_DISK:
4740 err = hot_remove_disk(mddev, new_decode_dev(arg));
4741 goto done_unlock;
4742
4743 case HOT_ADD_DISK:
4744 err = hot_add_disk(mddev, new_decode_dev(arg));
4745 goto done_unlock;
4746
4747 case SET_DISK_FAULTY:
4748 err = set_disk_faulty(mddev, new_decode_dev(arg));
4749 goto done_unlock;
4750
4751 case RUN_ARRAY:
4752 err = do_md_run (mddev);
4753 goto done_unlock;
4754
32a7627c
N		 4755 case SET_BITMAP_FILE:
4756 err = set_bitmap_file(mddev, (int)arg);
4757 goto done_unlock;
4758
1da177e4 4759 default:
1da177e4
LT		 4760 err = -EINVAL;
4761 goto abort_unlock;
4762 }
4763
4764done_unlock:
4765abort_unlock:
4766 mddev_unlock(mddev);
4767
4768 return err;
4769done:
4770 if (err)
4771 MD_BUG();
4772abort:
4773 return err;
4774}
4775
4776static int md_open(struct inode *inode, struct file *file)
4777{
4778 /*
4779 * Succeed if we can lock the mddev, which confirms that
4780 * it isn't being stopped right now.
4781 */
4782 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4783 int err;
4784
d63a5a74 4785 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
1da177e4
LT		 4786 goto out;
4787
4788 err = 0;
4789 mddev_get(mddev);
4790 mddev_unlock(mddev);
4791
4792 check_disk_change(inode->i_bdev);
4793 out:
4794 return err;
4795}
4796
4797static int md_release(struct inode *inode, struct file * file)
4798{
4799 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4800
52e5f9d1 4801 BUG_ON(!mddev);
1da177e4
LT		 4802 mddev_put(mddev);
4803
4804 return 0;
4805}
4806
44ce6294
LT		 4807static int md_media_changed(struct gendisk *disk)
4808{
4809 mddev_t *mddev = disk->private_data;
4810
4811 return mddev->changed;
4812}
4813
4814static int md_revalidate(struct gendisk *disk)
4815{
4816 mddev_t *mddev = disk->private_data;
4817
4818 mddev->changed = 0;
4819 return 0;
4820}
1da177e4
LT		 4821static struct block_device_operations md_fops =
4822{
4823 .owner = THIS_MODULE,
4824 .open = md_open,
4825 .release = md_release,
4826 .ioctl = md_ioctl,
a885c8c4 4827 .getgeo = md_getgeo,
44ce6294
LT		 4828 .media_changed = md_media_changed,
4829 .revalidate_disk= md_revalidate,
1da177e4
LT		 4830};
4831
75c96f85 4832static int md_thread(void * arg)
1da177e4
LT		 4833{
4834 mdk_thread_t *thread = arg;
4835
1da177e4
LT		 4836 /*
4837 * md_thread is a 'system-thread', it's priority should be very
4838 * high. We avoid resource deadlocks individually in each
4839 * raid personality. (RAID5 does preallocation) We also use RR and
4840 * the very same RT priority as kswapd, thus we will never get
4841 * into a priority inversion deadlock.
4842 *
4843 * we definitely have to have equal or higher priority than
4844 * bdflush, otherwise bdflush will deadlock if there are too
4845 * many dirty RAID5 blocks.
4846 */
1da177e4 4847
6985c43f 4848 allow_signal(SIGKILL);
a6fb0934 4849 while (!kthread_should_stop()) {
1da177e4 4850
93588e22
N		 4851 /* We need to wait INTERRUPTIBLE so that
4852 * we don't add to the load-average.
4853 * That means we need to be sure no signals are
4854 * pending
4855 */
4856 if (signal_pending(current))
4857 flush_signals(current);
4858
4859 wait_event_interruptible_timeout
4860 (thread->wqueue,
4861 test_bit(THREAD_WAKEUP, &thread->flags)
4862 || kthread_should_stop(),
4863 thread->timeout);
1da177e4
LT		 4864
4865 clear_bit(THREAD_WAKEUP, &thread->flags);
4866
787453c2 4867 thread->run(thread->mddev);
1da177e4 4868 }
a6fb0934 4869
1da177e4
LT		 4870 return 0;
4871}
4872
4873void md_wakeup_thread(mdk_thread_t *thread)
4874{
4875 if (thread) {
4876 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4877 set_bit(THREAD_WAKEUP, &thread->flags);
4878 wake_up(&thread->wqueue);
4879 }
4880}
4881
4882mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4883 const char *name)
4884{
4885 mdk_thread_t *thread;
1da177e4 4886
9ffae0cf 4887 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 4888 if (!thread)
4889 return NULL;
4890
1da177e4
LT		 4891 init_waitqueue_head(&thread->wqueue);
4892
1da177e4
LT		 4893 thread->run = run;
4894 thread->mddev = mddev;
32a7627c 4895 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 4896 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 4897 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 4898 kfree(thread);
4899 return NULL;
4900 }
1da177e4
LT		 4901 return thread;
4902}
4903
1da177e4
LT		 4904void md_unregister_thread(mdk_thread_t *thread)
4905{
ba25f9dc 4906 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
a6fb0934
N		 4907
4908 kthread_stop(thread->tsk);
1da177e4
LT		 4909 kfree(thread);
4910}
4911
4912void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4913{
4914 if (!mddev) {
4915 MD_BUG();
4916 return;
4917 }
4918
b2d444d7 4919 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 4920 return;
32a7627c 4921/*
1da177e4
LT		 4922 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4923 mdname(mddev),
4924 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4925 __builtin_return_address(0),__builtin_return_address(1),
4926 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 4927*/
d0a0a5ee
AM		 4928 if (!mddev->pers)
4929 return;
1da177e4
LT		 4930 if (!mddev->pers->error_handler)
4931 return;
4932 mddev->pers->error_handler(mddev,rdev);
4933 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4934 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4935 md_wakeup_thread(mddev->thread);
c331eb04 4936 md_new_event_inintr(mddev);
1da177e4
LT		 4937}
4938
4939/* seq_file implementation /proc/mdstat */
4940
4941static void status_unused(struct seq_file *seq)
4942{
4943 int i = 0;
4944 mdk_rdev_t *rdev;
4945 struct list_head *tmp;
4946
4947 seq_printf(seq, "unused devices: ");
4948
73c34431 4949 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
1da177e4
LT		 4950 char b[BDEVNAME_SIZE];
4951 i++;
4952 seq_printf(seq, "%s ",
4953 bdevname(rdev->bdev,b));
4954 }
4955 if (!i)
4956 seq_printf(seq, "<none>");
4957
4958 seq_printf(seq, "\n");
4959}
4960
4961
4962static void status_resync(struct seq_file *seq, mddev_t * mddev)
4963{
4588b42e
N		 4964 sector_t max_blocks, resync, res;
4965 unsigned long dt, db, rt;
4966 int scale;
4967 unsigned int per_milli;
1da177e4
LT		 4968
4969 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4970
4971 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4972 max_blocks = mddev->resync_max_sectors >> 1;
4973 else
4974 max_blocks = mddev->size;
4975
4976 /*
4977 * Should not happen.
4978 */
4979 if (!max_blocks) {
4980 MD_BUG();
4981 return;
4982 }
4588b42e
N		 4983 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4984 * in a sector_t, and (max_blocks>>scale) will fit in a
4985 * u32, as those are the requirements for sector_div.
4986 * Thus 'scale' must be at least 10
4987 */
4988 scale = 10;
4989 if (sizeof(sector_t) > sizeof(unsigned long)) {
4990 while ( max_blocks/2 > (1ULL<<(scale+32)))
4991 scale++;
4992 }
4993 res = (resync>>scale)*1000;
4994 sector_div(res, (u32)((max_blocks>>scale)+1));
4995
4996 per_milli = res;
1da177e4 4997 {
4588b42e 4998 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 4999 seq_printf(seq, "[");
5000 for (i = 0; i < x; i++)
5001 seq_printf(seq, "=");
5002 seq_printf(seq, ">");
5003 for (i = 0; i < y; i++)
5004 seq_printf(seq, ".");
5005 seq_printf(seq, "] ");
5006 }
4588b42e 5007 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 5008 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5009 "reshape" :
61df9d91
N		 5010 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5011 "check" :
5012 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5013 "resync" : "recovery"))),
5014 per_milli/10, per_milli % 10,
4588b42e
N		 5015 (unsigned long long) resync,
5016 (unsigned long long) max_blocks);
1da177e4
LT		 5017
5018 /*
5019 * We do not want to overflow, so the order of operands and
5020 * the * 100 / 100 trick are important. We do a +1 to be
5021 * safe against division by zero. We only estimate anyway.
5022 *
5023 * dt: time from mark until now
5024 * db: blocks written from mark until now
5025 * rt: remaining time
5026 */
5027 dt = ((jiffies - mddev->resync_mark) / HZ);
5028 if (!dt) dt++;
ff4e8d9a
N		 5029 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5030 - mddev->resync_mark_cnt;
5031 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
1da177e4
LT		 5032
5033 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5034
ff4e8d9a 5035 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 5036}
5037
5038static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5039{
5040 struct list_head *tmp;
5041 loff_t l = *pos;
5042 mddev_t *mddev;
5043
5044 if (l >= 0x10000)
5045 return NULL;
5046 if (!l--)
5047 /* header */
5048 return (void*)1;
5049
5050 spin_lock(&all_mddevs_lock);
5051 list_for_each(tmp,&all_mddevs)
5052 if (!l--) {
5053 mddev = list_entry(tmp, mddev_t, all_mddevs);
5054 mddev_get(mddev);
5055 spin_unlock(&all_mddevs_lock);
5056 return mddev;
5057 }
5058 spin_unlock(&all_mddevs_lock);
5059 if (!l--)
5060 return (void*)2;/* tail */
5061 return NULL;
5062}
5063
5064static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5065{
5066 struct list_head *tmp;
5067 mddev_t *next_mddev, *mddev = v;
5068
5069 ++*pos;
5070 if (v == (void*)2)
5071 return NULL;
5072
5073 spin_lock(&all_mddevs_lock);
5074 if (v == (void*)1)
5075 tmp = all_mddevs.next;
5076 else
5077 tmp = mddev->all_mddevs.next;
5078 if (tmp != &all_mddevs)
5079 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5080 else {
5081 next_mddev = (void*)2;
5082 *pos = 0x10000;
5083 }
5084 spin_unlock(&all_mddevs_lock);
5085
5086 if (v != (void*)1)
5087 mddev_put(mddev);
5088 return next_mddev;
5089
5090}
5091
5092static void md_seq_stop(struct seq_file *seq, void *v)
5093{
5094 mddev_t *mddev = v;
5095
5096 if (mddev && v != (void*)1 && v != (void*)2)
5097 mddev_put(mddev);
5098}
5099
d7603b7e
N		 5100struct mdstat_info {
5101 int event;
5102};
5103
1da177e4
LT		 5104static int md_seq_show(struct seq_file *seq, void *v)
5105{
5106 mddev_t *mddev = v;
5107 sector_t size;
5108 struct list_head *tmp2;
5109 mdk_rdev_t *rdev;
d7603b7e 5110 struct mdstat_info *mi = seq->private;
32a7627c 5111 struct bitmap *bitmap;
1da177e4
LT		 5112
5113 if (v == (void*)1) {
2604b703 5114 struct mdk_personality *pers;
1da177e4
LT		 5115 seq_printf(seq, "Personalities : ");
5116 spin_lock(&pers_lock);
2604b703
N		 5117 list_for_each_entry(pers, &pers_list, list)
5118 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 5119
5120 spin_unlock(&pers_lock);
5121 seq_printf(seq, "\n");
d7603b7e 5122 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 5123 return 0;
5124 }
5125 if (v == (void*)2) {
5126 status_unused(seq);
5127 return 0;
5128 }
5129
5dc5cf7d 5130 if (mddev_lock(mddev) < 0)
1da177e4 5131 return -EINTR;
5dc5cf7d 5132
1da177e4
LT		 5133 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5134 seq_printf(seq, "%s : %sactive", mdname(mddev),
5135 mddev->pers ? "" : "in");
5136 if (mddev->pers) {
f91de92e 5137 if (mddev->ro==1)
1da177e4 5138 seq_printf(seq, " (read-only)");
f91de92e 5139 if (mddev->ro==2)
52720ae7 5140 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 5141 seq_printf(seq, " %s", mddev->pers->name);
5142 }
5143
5144 size = 0;
d089c6af 5145 rdev_for_each(rdev, tmp2, mddev) {
1da177e4
LT		 5146 char b[BDEVNAME_SIZE];
5147 seq_printf(seq, " %s[%d]",
5148 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 5149 if (test_bit(WriteMostly, &rdev->flags))
5150 seq_printf(seq, "(W)");
b2d444d7 5151 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5152 seq_printf(seq, "(F)");
5153 continue;
b325a32e
N		 5154 } else if (rdev->raid_disk < 0)
5155 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 5156 size += rdev->size;
5157 }
5158
5159 if (!list_empty(&mddev->disks)) {
5160 if (mddev->pers)
5161 seq_printf(seq, "\n %llu blocks",
5162 (unsigned long long)mddev->array_size);
5163 else
5164 seq_printf(seq, "\n %llu blocks",
5165 (unsigned long long)size);
5166 }
1cd6bf19
N		 5167 if (mddev->persistent) {
5168 if (mddev->major_version != 0 ||
5169 mddev->minor_version != 90) {
5170 seq_printf(seq," super %d.%d",
5171 mddev->major_version,
5172 mddev->minor_version);
5173 }
e691063a
N		 5174 } else if (mddev->external)
5175 seq_printf(seq, " super external:%s",
5176 mddev->metadata_type);
5177 else
1cd6bf19 5178 seq_printf(seq, " super non-persistent");
1da177e4
LT		 5179
5180 if (mddev->pers) {
5181 mddev->pers->status (seq, mddev);
5182 seq_printf(seq, "\n ");
8e1b39d6
N		 5183 if (mddev->pers->sync_request) {
5184 if (mddev->curr_resync > 2) {
5185 status_resync (seq, mddev);
5186 seq_printf(seq, "\n ");
5187 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5188 seq_printf(seq, "\tresync=DELAYED\n ");
5189 else if (mddev->recovery_cp < MaxSector)
5190 seq_printf(seq, "\tresync=PENDING\n ");
5191 }
32a7627c
N		 5192 } else
5193 seq_printf(seq, "\n ");
5194
5195 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 5196 unsigned long chunk_kb;
5197 unsigned long flags;
32a7627c
N		 5198 spin_lock_irqsave(&bitmap->lock, flags);
5199 chunk_kb = bitmap->chunksize >> 10;
5200 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5201 "%lu%s chunk",
5202 bitmap->pages - bitmap->missing_pages,
5203 bitmap->pages,
5204 (bitmap->pages - bitmap->missing_pages)
5205 << (PAGE_SHIFT - 10),
5206 chunk_kb ? chunk_kb : bitmap->chunksize,
5207 chunk_kb ? "KB" : "B");
78d742d8
N		 5208 if (bitmap->file) {
5209 seq_printf(seq, ", file: ");
c32c2f63 5210 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 5211 }
78d742d8 5212
32a7627c
N		 5213 seq_printf(seq, "\n");
5214 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 5215 }
5216
5217 seq_printf(seq, "\n");
5218 }
5219 mddev_unlock(mddev);
5220
5221 return 0;
5222}
5223
5224static struct seq_operations md_seq_ops = {
5225 .start = md_seq_start,
5226 .next = md_seq_next,
5227 .stop = md_seq_stop,
5228 .show = md_seq_show,
5229};
5230
5231static int md_seq_open(struct inode *inode, struct file *file)
5232{
5233 int error;
d7603b7e
N		 5234 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5235 if (mi == NULL)
5236 return -ENOMEM;
1da177e4
LT		 5237
5238 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 5239 if (error)
5240 kfree(mi);
5241 else {
5242 struct seq_file *p = file->private_data;
5243 p->private = mi;
5244 mi->event = atomic_read(&md_event_count);
5245 }
1da177e4
LT		 5246 return error;
5247}
5248
d7603b7e
N		 5249static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5250{
5251 struct seq_file *m = filp->private_data;
5252 struct mdstat_info *mi = m->private;
5253 int mask;
5254
5255 poll_wait(filp, &md_event_waiters, wait);
5256
5257 /* always allow read */
5258 mask = POLLIN | POLLRDNORM;
5259
5260 if (mi->event != atomic_read(&md_event_count))
5261 mask |= POLLERR | POLLPRI;
5262 return mask;
5263}
5264
fa027c2a 5265static const struct file_operations md_seq_fops = {
e24650c2 5266 .owner = THIS_MODULE,
1da177e4
LT		 5267 .open = md_seq_open,
5268 .read = seq_read,
5269 .llseek = seq_lseek,
c3f94b40 5270 .release = seq_release_private,
d7603b7e 5271 .poll = mdstat_poll,
1da177e4
LT		 5272};
5273
2604b703 5274int register_md_personality(struct mdk_personality *p)
1da177e4 5275{
1da177e4 5276 spin_lock(&pers_lock);
2604b703
N		 5277 list_add_tail(&p->list, &pers_list);
5278 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 5279 spin_unlock(&pers_lock);
5280 return 0;
5281}
5282
2604b703 5283int unregister_md_personality(struct mdk_personality *p)
1da177e4 5284{
2604b703 5285 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 5286 spin_lock(&pers_lock);
2604b703 5287 list_del_init(&p->list);
1da177e4
LT		 5288 spin_unlock(&pers_lock);
5289 return 0;
5290}
5291
5292static int is_mddev_idle(mddev_t *mddev)
5293{
5294 mdk_rdev_t * rdev;
5295 struct list_head *tmp;
5296 int idle;
713f6ab1 5297 long curr_events;
1da177e4
LT		 5298
5299 idle = 1;
d089c6af 5300 rdev_for_each(rdev, tmp, mddev) {
1da177e4 5301 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 5302 curr_events = disk_stat_read(disk, sectors[0]) +
5303 disk_stat_read(disk, sectors[1]) -
1da177e4 5304 atomic_read(&disk->sync_io);
713f6ab1
N		 5305 /* sync IO will cause sync_io to increase before the disk_stats
5306 * as sync_io is counted when a request starts, and
5307 * disk_stats is counted when it completes.
5308 * So resync activity will cause curr_events to be smaller than
5309 * when there was no such activity.
5310 * non-sync IO will cause disk_stat to increase without
5311 * increasing sync_io so curr_events will (eventually)
5312 * be larger than it was before. Once it becomes
5313 * substantially larger, the test below will cause
5314 * the array to appear non-idle, and resync will slow
5315 * down.
5316 * If there is a lot of outstanding resync activity when
5317 * we set last_event to curr_events, then all that activity
5318 * completing might cause the array to appear non-idle
5319 * and resync will be slowed down even though there might
5320 * not have been non-resync activity. This will only
5321 * happen once though. 'last_events' will soon reflect
5322 * the state where there is little or no outstanding
5323 * resync requests, and further resync activity will
5324 * always make curr_events less than last_events.
c0e48521 5325 *
1da177e4 5326 */
713f6ab1 5327 if (curr_events - rdev->last_events > 4096) {
1da177e4
LT		 5328 rdev->last_events = curr_events;
5329 idle = 0;
5330 }
5331 }
5332 return idle;
5333}
5334
5335void md_done_sync(mddev_t *mddev, int blocks, int ok)
5336{
5337 /* another "blocks" (512byte) blocks have been synced */
5338 atomic_sub(blocks, &mddev->recovery_active);
5339 wake_up(&mddev->recovery_wait);
5340 if (!ok) {
5341 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5342 md_wakeup_thread(mddev->thread);
5343 // stop recovery, signal do_sync ....
5344 }
5345}
5346
5347
06d91a5f
N		 5348/* md_write_start(mddev, bi)
5349 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 5350 * in superblock) before writing, schedule a superblock update
5351 * and wait for it to complete.
06d91a5f 5352 */
3d310eb7 5353void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 5354{
06d91a5f 5355 if (bio_data_dir(bi) != WRITE)
3d310eb7 5356 return;
06d91a5f 5357
f91de92e
N		 5358 BUG_ON(mddev->ro == 1);
5359 if (mddev->ro == 2) {
5360 /* need to switch to read/write */
5361 mddev->ro = 0;
5362 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5363 md_wakeup_thread(mddev->thread);
25156198 5364 md_wakeup_thread(mddev->sync_thread);
f91de92e 5365 }
06d91a5f 5366 atomic_inc(&mddev->writes_pending);
06d91a5f 5367 if (mddev->in_sync) {
a9701a30 5368 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 5369 if (mddev->in_sync) {
5370 mddev->in_sync = 0;
850b2b42 5371 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3d310eb7
N		 5372 md_wakeup_thread(mddev->thread);
5373 }
a9701a30 5374 spin_unlock_irq(&mddev->write_lock);
06d91a5f 5375 }
850b2b42 5376 wait_event(mddev->sb_wait, mddev->flags==0);
1da177e4
LT		 5377}
5378
5379void md_write_end(mddev_t *mddev)
5380{
5381 if (atomic_dec_and_test(&mddev->writes_pending)) {
5382 if (mddev->safemode == 2)
5383 md_wakeup_thread(mddev->thread);
16f17b39 5384 else if (mddev->safemode_delay)
1da177e4
LT		 5385 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5386 }
5387}
5388
2a2275d6
N		 5389/* md_allow_write(mddev)
5390 * Calling this ensures that the array is marked 'active' so that writes
5391 * may proceed without blocking. It is important to call this before
5392 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5393 * Must be called with mddev_lock held.
5394 */
5395void md_allow_write(mddev_t *mddev)
5396{
5397 if (!mddev->pers)
5398 return;
5399 if (mddev->ro)
5400 return;
5401
5402 spin_lock_irq(&mddev->write_lock);
5403 if (mddev->in_sync) {
5404 mddev->in_sync = 0;
5405 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5406 if (mddev->safemode_delay &&
5407 mddev->safemode == 0)
5408 mddev->safemode = 1;
5409 spin_unlock_irq(&mddev->write_lock);
5410 md_update_sb(mddev, 0);
5411 } else
5412 spin_unlock_irq(&mddev->write_lock);
5413}
5414EXPORT_SYMBOL_GPL(md_allow_write);
5415
75c96f85 5416static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 5417
5418#define SYNC_MARKS 10
5419#define SYNC_MARK_STEP (3*HZ)
29269553 5420void md_do_sync(mddev_t *mddev)
1da177e4
LT		 5421{
5422 mddev_t *mddev2;
5423 unsigned int currspeed = 0,
5424 window;
57afd89f 5425 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 5426 unsigned long mark[SYNC_MARKS];
5427 sector_t mark_cnt[SYNC_MARKS];
5428 int last_mark,m;
5429 struct list_head *tmp;
5430 sector_t last_check;
57afd89f 5431 int skipped = 0;
5fd6c1dc
N		 5432 struct list_head *rtmp;
5433 mdk_rdev_t *rdev;
61df9d91 5434 char *desc;
1da177e4
LT		 5435
5436 /* just incase thread restarts... */
5437 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5438 return;
5fd6c1dc
N		 5439 if (mddev->ro) /* never try to sync a read-only array */
5440 return;
1da177e4 5441
61df9d91
N		 5442 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5443 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5444 desc = "data-check";
5445 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5446 desc = "requested-resync";
5447 else
5448 desc = "resync";
5449 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5450 desc = "reshape";
5451 else
5452 desc = "recovery";
5453
1da177e4
LT		 5454 /* we overload curr_resync somewhat here.
5455 * 0 == not engaged in resync at all
5456 * 2 == checking that there is no conflict with another sync
5457 * 1 == like 2, but have yielded to allow conflicting resync to
5458 * commense
5459 * other == active in resync - this many blocks
5460 *
5461 * Before starting a resync we must have set curr_resync to
5462 * 2, and then checked that every "conflicting" array has curr_resync
5463 * less than ours. When we find one that is the same or higher
5464 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5465 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5466 * This will mean we have to start checking from the beginning again.
5467 *
5468 */
5469
5470 do {
5471 mddev->curr_resync = 2;
5472
5473 try_again:
787453c2 5474 if (kthread_should_stop()) {
6985c43f 5475 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 5476 goto skip;
5477 }
29ac4aa3 5478 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 5479 if (mddev2 == mddev)
5480 continue;
5481 if (mddev2->curr_resync &&
5482 match_mddev_units(mddev,mddev2)) {
5483 DEFINE_WAIT(wq);
5484 if (mddev < mddev2 && mddev->curr_resync == 2) {
5485 /* arbitrarily yield */
5486 mddev->curr_resync = 1;
5487 wake_up(&resync_wait);
5488 }
5489 if (mddev > mddev2 && mddev->curr_resync == 1)
5490 /* no need to wait here, we can wait the next
5491 * time 'round when curr_resync == 2
5492 */
5493 continue;
787453c2
N		 5494 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5495 if (!kthread_should_stop() &&
8712e553 5496 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 5497 printk(KERN_INFO "md: delaying %s of %s"
5498 " until %s has finished (they"
1da177e4 5499 " share one or more physical units)\n",
61df9d91 5500 desc, mdname(mddev), mdname(mddev2));
1da177e4
LT		 5501 mddev_put(mddev2);
5502 schedule();
5503 finish_wait(&resync_wait, &wq);
5504 goto try_again;
5505 }
5506 finish_wait(&resync_wait, &wq);
5507 }
5508 }
5509 } while (mddev->curr_resync < 2);
5510
5fd6c1dc 5511 j = 0;
9d88883e 5512 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 5513 /* resync follows the size requested by the personality,
57afd89f 5514 * which defaults to physical size, but can be virtual size
1da177e4
LT		 5515 */
5516 max_sectors = mddev->resync_max_sectors;
9d88883e 5517 mddev->resync_mismatches = 0;
5fd6c1dc
N		 5518 /* we don't use the checkpoint if there's a bitmap */
5519 if (!mddev->bitmap &&
5520 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5521 j = mddev->recovery_cp;
ccfcc3c1
N		 5522 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5523 max_sectors = mddev->size << 1;
5fd6c1dc 5524 else {
1da177e4
LT		 5525 /* recovery follows the physical size of devices */
5526 max_sectors = mddev->size << 1;
5fd6c1dc 5527 j = MaxSector;
d089c6af 5528 rdev_for_each(rdev, rtmp, mddev)
5fd6c1dc
N		 5529 if (rdev->raid_disk >= 0 &&
5530 !test_bit(Faulty, &rdev->flags) &&
5531 !test_bit(In_sync, &rdev->flags) &&
5532 rdev->recovery_offset < j)
5533 j = rdev->recovery_offset;
5534 }
1da177e4 5535
61df9d91
N		 5536 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5537 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5538 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 5539 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 5540 "(but not more than %d KB/sec) for %s.\n",
5541 speed_max(mddev), desc);
1da177e4
LT		 5542
5543 is_mddev_idle(mddev); /* this also initializes IO event counters */
5fd6c1dc 5544
57afd89f 5545 io_sectors = 0;
1da177e4
LT		 5546 for (m = 0; m < SYNC_MARKS; m++) {
5547 mark[m] = jiffies;
57afd89f 5548 mark_cnt[m] = io_sectors;
1da177e4
LT		 5549 }
5550 last_mark = 0;
5551 mddev->resync_mark = mark[last_mark];
5552 mddev->resync_mark_cnt = mark_cnt[last_mark];
5553
5554 /*
5555 * Tune reconstruction:
5556 */
5557 window = 32*(PAGE_SIZE/512);
5558 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5559 window/2,(unsigned long long) max_sectors/2);
5560
5561 atomic_set(&mddev->recovery_active, 0);
5562 init_waitqueue_head(&mddev->recovery_wait);
5563 last_check = 0;
5564
5565 if (j>2) {
5566 printk(KERN_INFO
61df9d91
N		 5567 "md: resuming %s of %s from checkpoint.\n",
5568 desc, mdname(mddev));
1da177e4
LT		 5569 mddev->curr_resync = j;
5570 }
5571
5572 while (j < max_sectors) {
57afd89f 5573 sector_t sectors;
1da177e4 5574
57afd89f 5575 skipped = 0;
c6207277
N		 5576 if (j >= mddev->resync_max) {
5577 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5578 wait_event(mddev->recovery_wait,
5579 mddev->resync_max > j
5580 || kthread_should_stop());
5581 }
5582 if (kthread_should_stop())
5583 goto interrupted;
57afd89f 5584 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 5585 currspeed < speed_min(mddev));
57afd89f 5586 if (sectors == 0) {
1da177e4
LT		 5587 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5588 goto out;
5589 }
57afd89f
N		 5590
5591 if (!skipped) { /* actual IO requested */
5592 io_sectors += sectors;
5593 atomic_add(sectors, &mddev->recovery_active);
5594 }
5595
1da177e4
LT		 5596 j += sectors;
5597 if (j>1) mddev->curr_resync = j;
ff4e8d9a 5598 mddev->curr_mark_cnt = io_sectors;
d7603b7e
N		 5599 if (last_check == 0)
5600 /* this is the earliers that rebuilt will be
5601 * visible in /proc/mdstat
5602 */
5603 md_new_event(mddev);
57afd89f
N		 5604
5605 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 5606 continue;
5607
57afd89f 5608 last_check = io_sectors;
1da177e4
LT		 5609
5610 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5611 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5612 break;
5613
5614 repeat:
5615 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5616 /* step marks */
5617 int next = (last_mark+1) % SYNC_MARKS;
5618
5619 mddev->resync_mark = mark[next];
5620 mddev->resync_mark_cnt = mark_cnt[next];
5621 mark[next] = jiffies;
57afd89f 5622 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 5623 last_mark = next;
5624 }
5625
5626
c6207277
N		 5627 if (kthread_should_stop())
5628 goto interrupted;
5629
1da177e4
LT		 5630
5631 /*
5632 * this loop exits only if either when we are slower than
5633 * the 'hard' speed limit, or the system was IO-idle for
5634 * a jiffy.
5635 * the system might be non-idle CPU-wise, but we only care
5636 * about not overloading the IO subsystem. (things like an
5637 * e2fsck being done on the RAID array should execute fast)
5638 */
2ad8b1ef 5639 blk_unplug(mddev->queue);
1da177e4
LT		 5640 cond_resched();
5641
57afd89f
N		 5642 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5643 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 5644
88202a0c
N		 5645 if (currspeed > speed_min(mddev)) {
5646 if ((currspeed > speed_max(mddev)) ||
1da177e4 5647 !is_mddev_idle(mddev)) {
c0e48521 5648 msleep(500);
1da177e4
LT		 5649 goto repeat;
5650 }
5651 }
5652 }
61df9d91 5653 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 5654 /*
5655 * this also signals 'finished resyncing' to md_stop
5656 */
5657 out:
2ad8b1ef 5658 blk_unplug(mddev->queue);
1da177e4
LT		 5659
5660 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5661
5662 /* tell personality that we are finished */
57afd89f 5663 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 5664
5665 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
ccfcc3c1 5666 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 5667 mddev->curr_resync > 2) {
5668 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5669 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5670 if (mddev->curr_resync >= mddev->recovery_cp) {
5671 printk(KERN_INFO
61df9d91
N		 5672 "md: checkpointing %s of %s.\n",
5673 desc, mdname(mddev));
5fd6c1dc
N		 5674 mddev->recovery_cp = mddev->curr_resync;
5675 }
5676 } else
5677 mddev->recovery_cp = MaxSector;
5678 } else {
5679 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5680 mddev->curr_resync = MaxSector;
d089c6af 5681 rdev_for_each(rdev, rtmp, mddev)
5fd6c1dc
N		 5682 if (rdev->raid_disk >= 0 &&
5683 !test_bit(Faulty, &rdev->flags) &&
5684 !test_bit(In_sync, &rdev->flags) &&
5685 rdev->recovery_offset < mddev->curr_resync)
5686 rdev->recovery_offset = mddev->curr_resync;
5fd6c1dc 5687 }
1da177e4 5688 }
17571284 5689 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5690
1da177e4
LT		 5691 skip:
5692 mddev->curr_resync = 0;
c6207277
N		 5693 mddev->resync_max = MaxSector;
5694 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
1da177e4
LT		 5695 wake_up(&resync_wait);
5696 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5697 md_wakeup_thread(mddev->thread);
c6207277
N		 5698 return;
5699
5700 interrupted:
5701 /*
5702 * got a signal, exit.
5703 */
5704 printk(KERN_INFO
5705 "md: md_do_sync() got signal ... exiting\n");
5706 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5707 goto out;
5708
1da177e4 5709}
29269553 5710EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 5711
5712
b4c4c7b8
N		 5713static int remove_and_add_spares(mddev_t *mddev)
5714{
5715 mdk_rdev_t *rdev;
5716 struct list_head *rtmp;
5717 int spares = 0;
5718
d089c6af 5719 rdev_for_each(rdev, rtmp, mddev)
b4c4c7b8 5720 if (rdev->raid_disk >= 0 &&
c303da6d 5721 !mddev->external &&
b4c4c7b8
N		 5722 (test_bit(Faulty, &rdev->flags) ||
5723 ! test_bit(In_sync, &rdev->flags)) &&
5724 atomic_read(&rdev->nr_pending)==0) {
5725 if (mddev->pers->hot_remove_disk(
5726 mddev, rdev->raid_disk)==0) {
5727 char nm[20];
5728 sprintf(nm,"rd%d", rdev->raid_disk);
5729 sysfs_remove_link(&mddev->kobj, nm);
5730 rdev->raid_disk = -1;
5731 }
5732 }
5733
5734 if (mddev->degraded) {
d089c6af 5735 rdev_for_each(rdev, rtmp, mddev)
b4c4c7b8
N		 5736 if (rdev->raid_disk < 0
5737 && !test_bit(Faulty, &rdev->flags)) {
5738 rdev->recovery_offset = 0;
5739 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5740 char nm[20];
5741 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 5742 if (sysfs_create_link(&mddev->kobj,
5743 &rdev->kobj, nm))
5744 printk(KERN_WARNING
5745 "md: cannot register "
5746 "%s for %s\n",
5747 nm, mdname(mddev));
b4c4c7b8
N		 5748 spares++;
5749 md_new_event(mddev);
5750 } else
5751 break;
5752 }
5753 }
5754 return spares;
5755}
1da177e4
LT		 5756/*
5757 * This routine is regularly called by all per-raid-array threads to
5758 * deal with generic issues like resync and super-block update.
5759 * Raid personalities that don't have a thread (linear/raid0) do not
5760 * need this as they never do any recovery or update the superblock.
5761 *
5762 * It does not do any resync itself, but rather "forks" off other threads
5763 * to do that as needed.
5764 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5765 * "->recovery" and create a thread at ->sync_thread.
5766 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5767 * and wakeups up this thread which will reap the thread and finish up.
5768 * This thread also removes any faulty devices (with nr_pending == 0).
5769 *
5770 * The overall approach is:
5771 * 1/ if the superblock needs updating, update it.
5772 * 2/ If a recovery thread is running, don't do anything else.
5773 * 3/ If recovery has finished, clean up, possibly marking spares active.
5774 * 4/ If there are any faulty devices, remove them.
5775 * 5/ If array is degraded, try to add spares devices
5776 * 6/ If array has spares or is not in-sync, start a resync thread.
5777 */
5778void md_check_recovery(mddev_t *mddev)
5779{
5780 mdk_rdev_t *rdev;
5781 struct list_head *rtmp;
5782
5783
5f40402d
N		 5784 if (mddev->bitmap)
5785 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 5786
5787 if (mddev->ro)
5788 return;
fca4d848
N		 5789
5790 if (signal_pending(current)) {
5791 if (mddev->pers->sync_request) {
5792 printk(KERN_INFO "md: %s in immediate safe mode\n",
5793 mdname(mddev));
5794 mddev->safemode = 2;
5795 }
5796 flush_signals(current);
5797 }
5798
1da177e4 5799 if ( ! (
e691063a 5800 (mddev->flags && !mddev->external) ||
1da177e4 5801 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 5802 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5803 (mddev->safemode == 1) ||
5804 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5805 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 5806 ))
5807 return;
fca4d848 5808
df5b89b3 5809 if (mddev_trylock(mddev)) {
b4c4c7b8 5810 int spares = 0;
fca4d848 5811
a9701a30 5812 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 5813 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5814 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5815 mddev->in_sync = 1;
e691063a
N		 5816 if (mddev->persistent)
5817 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
fca4d848
N		 5818 }
5819 if (mddev->safemode == 1)
5820 mddev->safemode = 0;
a9701a30 5821 spin_unlock_irq(&mddev->write_lock);
fca4d848 5822
850b2b42
N		 5823 if (mddev->flags)
5824 md_update_sb(mddev, 0);
06d91a5f 5825
06d91a5f 5826
1da177e4
LT		 5827 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5828 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5829 /* resync/recovery still happening */
5830 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5831 goto unlock;
5832 }
5833 if (mddev->sync_thread) {
5834 /* resync has finished, collect result */
5835 md_unregister_thread(mddev->sync_thread);
5836 mddev->sync_thread = NULL;
5837 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5838 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5839 /* success...*/
5840 /* activate any spares */
5841 mddev->pers->spare_active(mddev);
5842 }
850b2b42 5843 md_update_sb(mddev, 1);
41158c7e
N		 5844
5845 /* if array is no-longer degraded, then any saved_raid_disk
5846 * information must be scrapped
5847 */
5848 if (!mddev->degraded)
d089c6af 5849 rdev_for_each(rdev, rtmp, mddev)
41158c7e
N		 5850 rdev->saved_raid_disk = -1;
5851
1da177e4
LT		 5852 mddev->recovery = 0;
5853 /* flag recovery needed just to double check */
5854 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d7603b7e 5855 md_new_event(mddev);
1da177e4
LT		 5856 goto unlock;
5857 }
24dd469d
N		 5858 /* Clear some bits that don't mean anything, but
5859 * might be left set
5860 */
5861 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5862 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5863 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5864 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 5865
5fd6c1dc
N		 5866 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5867 goto unlock;
1da177e4
LT		 5868 /* no recovery is running.
5869 * remove any failed drives, then
5870 * add spares if possible.
5871 * Spare are also removed and re-added, to allow
5872 * the personality to fail the re-add.
5873 */
1da177e4 5874
b4c4c7b8
N		 5875 if (mddev->reshape_position != MaxSector) {
5876 if (mddev->pers->check_reshape(mddev) != 0)
5877 /* Cannot proceed */
5878 goto unlock;
5879 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5880 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 5881 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5882 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5883 } else if (mddev->recovery_cp < MaxSector) {
5884 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5885 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5886 /* nothing to be done ... */
1da177e4 5887 goto unlock;
24dd469d 5888
1da177e4
LT		 5889 if (mddev->pers->sync_request) {
5890 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 5891 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5892 /* We are adding a device or devices to an array
5893 * which has the bitmap stored on all devices.
5894 * So make sure all bitmap pages get written
5895 */
5896 bitmap_write_all(mddev->bitmap);
5897 }
1da177e4
LT		 5898 mddev->sync_thread = md_register_thread(md_do_sync,
5899 mddev,
5900 "%s_resync");
5901 if (!mddev->sync_thread) {
5902 printk(KERN_ERR "%s: could not start resync"
5903 " thread...\n",
5904 mdname(mddev));
5905 /* leave the spares where they are, it shouldn't hurt */
5906 mddev->recovery = 0;
d7603b7e 5907 } else
1da177e4 5908 md_wakeup_thread(mddev->sync_thread);
d7603b7e 5909 md_new_event(mddev);
1da177e4
LT		 5910 }
5911 unlock:
5912 mddev_unlock(mddev);
5913 }
5914}
5915
75c96f85
AB		 5916static int md_notify_reboot(struct notifier_block *this,
5917 unsigned long code, void *x)
1da177e4
LT		 5918{
5919 struct list_head *tmp;
5920 mddev_t *mddev;
5921
5922 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5923
5924 printk(KERN_INFO "md: stopping all md devices.\n");
5925
29ac4aa3 5926 for_each_mddev(mddev, tmp)
c71d4887 5927 if (mddev_trylock(mddev)) {
1da177e4 5928 do_md_stop (mddev, 1);
c71d4887
NB		 5929 mddev_unlock(mddev);
5930 }
1da177e4
LT		 5931 /*
5932 * certain more exotic SCSI devices are known to be
5933 * volatile wrt too early system reboots. While the
5934 * right place to handle this issue is the given
5935 * driver, we do want to have a safe RAID driver ...
5936 */
5937 mdelay(1000*1);
5938 }
5939 return NOTIFY_DONE;
5940}
5941
75c96f85 5942static struct notifier_block md_notifier = {
1da177e4
LT		 5943 .notifier_call = md_notify_reboot,
5944 .next = NULL,
5945 .priority = INT_MAX, /* before any real devices */
5946};
5947
5948static void md_geninit(void)
5949{
5950 struct proc_dir_entry *p;
5951
5952 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5953
5954 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5955 if (p)
5956 p->proc_fops = &md_seq_fops;
5957}
5958
75c96f85 5959static int __init md_init(void)
1da177e4 5960{
1da177e4
LT		 5961 if (register_blkdev(MAJOR_NR, "md"))
5962 return -1;
5963 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5964 unregister_blkdev(MAJOR_NR, "md");
5965 return -1;
5966 }
e8703fe1
N		 5967 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5968 md_probe, NULL, NULL);
5969 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 5970 md_probe, NULL, NULL);
5971
1da177e4 5972 register_reboot_notifier(&md_notifier);
0b4d4147 5973 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 5974
5975 md_geninit();
5976 return (0);
5977}
5978
5979
5980#ifndef MODULE
5981
5982/*
5983 * Searches all registered partitions for autorun RAID arrays
5984 * at boot time.
5985 */
4d936ec1
ME		 5986
5987static LIST_HEAD(all_detected_devices);
5988struct detected_devices_node {
5989 struct list_head list;
5990 dev_t dev;
5991};
1da177e4
LT		 5992
5993void md_autodetect_dev(dev_t dev)
5994{
4d936ec1
ME		 5995 struct detected_devices_node *node_detected_dev;
5996
5997 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
5998 if (node_detected_dev) {
5999 node_detected_dev->dev = dev;
6000 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6001 } else {
6002 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6003 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6004 }
1da177e4
LT		 6005}
6006
6007
6008static void autostart_arrays(int part)
6009{
6010 mdk_rdev_t *rdev;
4d936ec1
ME		 6011 struct detected_devices_node *node_detected_dev;
6012 dev_t dev;
6013 int i_scanned, i_passed;
1da177e4 6014
4d936ec1
ME		 6015 i_scanned = 0;
6016 i_passed = 0;
1da177e4 6017
4d936ec1 6018 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 6019
4d936ec1
ME		 6020 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6021 i_scanned++;
6022 node_detected_dev = list_entry(all_detected_devices.next,
6023 struct detected_devices_node, list);
6024 list_del(&node_detected_dev->list);
6025 dev = node_detected_dev->dev;
6026 kfree(node_detected_dev);
df968c4e 6027 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 6028 if (IS_ERR(rdev))
6029 continue;
6030
b2d444d7 6031 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6032 MD_BUG();
6033 continue;
6034 }
d0fae18f 6035 set_bit(AutoDetected, &rdev->flags);
1da177e4 6036 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 6037 i_passed++;
1da177e4 6038 }
4d936ec1
ME		 6039
6040 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6041 i_scanned, i_passed);
1da177e4
LT		 6042
6043 autorun_devices(part);
6044}
6045
fdee8ae4 6046#endif /* !MODULE */
1da177e4
LT		 6047
6048static __exit void md_exit(void)
6049{
6050 mddev_t *mddev;
6051 struct list_head *tmp;
8ab5e4c1 6052
e8703fe1
N		 6053 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6054 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4
LT		 6055
6056 unregister_blkdev(MAJOR_NR,"md");
6057 unregister_blkdev(mdp_major, "mdp");
6058 unregister_reboot_notifier(&md_notifier);
6059 unregister_sysctl_table(raid_table_header);
6060 remove_proc_entry("mdstat", NULL);
29ac4aa3 6061 for_each_mddev(mddev, tmp) {
1da177e4
LT		 6062 struct gendisk *disk = mddev->gendisk;
6063 if (!disk)
6064 continue;
6065 export_array(mddev);
6066 del_gendisk(disk);
6067 put_disk(disk);
6068 mddev->gendisk = NULL;
6069 mddev_put(mddev);
6070 }
6071}
6072
685784aa 6073subsys_initcall(md_init);
1da177e4
LT		 6074module_exit(md_exit)
6075
f91de92e
N		 6076static int get_ro(char *buffer, struct kernel_param *kp)
6077{
6078 return sprintf(buffer, "%d", start_readonly);
6079}
6080static int set_ro(const char *val, struct kernel_param *kp)
6081{
6082 char *e;
6083 int num = simple_strtoul(val, &e, 10);
6084 if (*val && (*e == '\0' || *e == '\n')) {
6085 start_readonly = num;
4dbcdc75 6086 return 0;
f91de92e
N		 6087 }
6088 return -EINVAL;
6089}
6090
80ca3a44
N		 6091module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6092module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 6093
f91de92e 6094
1da177e4
LT		 6095EXPORT_SYMBOL(register_md_personality);
6096EXPORT_SYMBOL(unregister_md_personality);
6097EXPORT_SYMBOL(md_error);
6098EXPORT_SYMBOL(md_done_sync);
6099EXPORT_SYMBOL(md_write_start);
6100EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 6101EXPORT_SYMBOL(md_register_thread);
6102EXPORT_SYMBOL(md_unregister_thread);
6103EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 6104EXPORT_SYMBOL(md_check_recovery);
6105MODULE_LICENSE("GPL");
aa1595e9 6106MODULE_ALIAS("md");
72008652 6107MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux kernel - Deliverables
This page took 0.787696 seconds and 5 git commands to generate.