projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: lock address when changing attributes of component devices
[linux-2.6-block.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
35#include <linux/module.h>
50511da3 36#include <linux/kernel.h>
a6fb0934 37#include <linux/kthread.h>
1da177e4
LT		 38#include <linux/linkage.h>
39#include <linux/raid/md.h>
32a7627c 40#include <linux/raid/bitmap.h>
1da177e4 41#include <linux/sysctl.h>
1da177e4 42#include <linux/buffer_head.h> /* for invalidate_bdev */
d7603b7e 43#include <linux/poll.h>
48c9c27b 44#include <linux/mutex.h>
16f17b39 45#include <linux/ctype.h>
7dfb7103 46#include <linux/freezer.h>
1da177e4
LT		 47
48#include <linux/init.h>
49
32a7627c
N		 50#include <linux/file.h>
51
1da177e4
LT		 52#ifdef CONFIG_KMOD
53#include <linux/kmod.h>
54#endif
55
56#include <asm/unaligned.h>
57
58#define MAJOR_NR MD_MAJOR
59#define MD_DRIVER
60
61/* 63 partitions with the alternate major number (mdp) */
62#define MdpMinorShift 6
63
64#define DEBUG 0
65#define dprintk(x...) ((void)(DEBUG && printk(x)))
66
67
68#ifndef MODULE
69static void autostart_arrays (int part);
70#endif
71
2604b703 72static LIST_HEAD(pers_list);
1da177e4
LT		 73static DEFINE_SPINLOCK(pers_lock);
74
5e56341d
AB		 75static void md_print_devices(void);
76
77#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78
1da177e4
LT		 79/*
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 83 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
86 * idle IO detection.
87 *
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 89 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 90 */
91
92static int sysctl_speed_limit_min = 1000;
93static int sysctl_speed_limit_max = 200000;
88202a0c
N		 94static inline int speed_min(mddev_t *mddev)
95{
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
98}
99
100static inline int speed_max(mddev_t *mddev)
101{
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
104}
1da177e4
LT		 105
106static struct ctl_table_header *raid_table_header;
107
108static ctl_table raid_table[] = {
109 {
110 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
80ca3a44 114 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 115 .proc_handler = &proc_dointvec,
116 },
117 {
118 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
80ca3a44 122 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 123 .proc_handler = &proc_dointvec,
124 },
125 { .ctl_name = 0 }
126};
127
128static ctl_table raid_dir_table[] = {
129 {
130 .ctl_name = DEV_RAID,
131 .procname = "raid",
132 .maxlen = 0,
80ca3a44 133 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 134 .child = raid_table,
135 },
136 { .ctl_name = 0 }
137};
138
139static ctl_table raid_root_table[] = {
140 {
141 .ctl_name = CTL_DEV,
142 .procname = "dev",
143 .maxlen = 0,
144 .mode = 0555,
145 .child = raid_dir_table,
146 },
147 { .ctl_name = 0 }
148};
149
150static struct block_device_operations md_fops;
151
f91de92e
N		 152static int start_readonly;
153
d7603b7e
N		 154/*
155 * We have a system wide 'event count' that is incremented
156 * on any 'interesting' event, and readers of /proc/mdstat
157 * can use 'poll' or 'select' to find out when the event
158 * count increases.
159 *
160 * Events are:
161 * start array, stop array, error, add device, remove device,
162 * start build, activate spare
163 */
2989ddbd 164static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 165static atomic_t md_event_count;
29269553 166void md_new_event(mddev_t *mddev)
d7603b7e
N		 167{
168 atomic_inc(&md_event_count);
169 wake_up(&md_event_waiters);
4508a7a7 170 sysfs_notify(&mddev->kobj, NULL, "sync_action");
d7603b7e 171}
29269553 172EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 173
c331eb04
N		 174/* Alternate version that can be called from interrupts
175 * when calling sysfs_notify isn't needed.
176 */
05381954 177static void md_new_event_inintr(mddev_t *mddev)
c331eb04
N		 178{
179 atomic_inc(&md_event_count);
180 wake_up(&md_event_waiters);
181}
182
1da177e4
LT		 183/*
184 * Enables to iterate over all existing md arrays
185 * all_mddevs_lock protects this list.
186 */
187static LIST_HEAD(all_mddevs);
188static DEFINE_SPINLOCK(all_mddevs_lock);
189
190
191/*
192 * iterates through all used mddevs in the system.
193 * We take care to grab the all_mddevs_lock whenever navigating
194 * the list, and to always hold a refcount when unlocked.
195 * Any code which breaks out of this loop while own
196 * a reference to the current mddev and must mddev_put it.
197 */
198#define ITERATE_MDDEV(mddev,tmp) \
199 \
200 for (({ spin_lock(&all_mddevs_lock); \
201 tmp = all_mddevs.next; \
202 mddev = NULL;}); \
203 ({ if (tmp != &all_mddevs) \
204 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205 spin_unlock(&all_mddevs_lock); \
206 if (mddev) mddev_put(mddev); \
207 mddev = list_entry(tmp, mddev_t, all_mddevs); \
208 tmp != &all_mddevs;}); \
209 ({ spin_lock(&all_mddevs_lock); \
210 tmp = tmp->next;}) \
211 )
212
213
165125e1 214static int md_fail_request (struct request_queue *q, struct bio *bio)
1da177e4 215{
6712ecf8 216 bio_io_error(bio);
1da177e4
LT		 217 return 0;
218}
219
220static inline mddev_t *mddev_get(mddev_t *mddev)
221{
222 atomic_inc(&mddev->active);
223 return mddev;
224}
225
226static void mddev_put(mddev_t *mddev)
227{
228 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
229 return;
230 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
231 list_del(&mddev->all_mddevs);
926ce2d8 232 spin_unlock(&all_mddevs_lock);
1312f40e 233 blk_cleanup_queue(mddev->queue);
c10997f6 234 kobject_put(&mddev->kobj);
926ce2d8
N		 235 } else
236 spin_unlock(&all_mddevs_lock);
1da177e4
LT		 237}
238
239static mddev_t * mddev_find(dev_t unit)
240{
241 mddev_t *mddev, *new = NULL;
242
243 retry:
244 spin_lock(&all_mddevs_lock);
245 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
246 if (mddev->unit == unit) {
247 mddev_get(mddev);
248 spin_unlock(&all_mddevs_lock);
990a8baf 249 kfree(new);
1da177e4
LT		 250 return mddev;
251 }
252
253 if (new) {
254 list_add(&new->all_mddevs, &all_mddevs);
255 spin_unlock(&all_mddevs_lock);
256 return new;
257 }
258 spin_unlock(&all_mddevs_lock);
259
9ffae0cf 260 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 261 if (!new)
262 return NULL;
263
1da177e4
LT		 264 new->unit = unit;
265 if (MAJOR(unit) == MD_MAJOR)
266 new->md_minor = MINOR(unit);
267 else
268 new->md_minor = MINOR(unit) >> MdpMinorShift;
269
df5b89b3 270 mutex_init(&new->reconfig_mutex);
1da177e4
LT		 271 INIT_LIST_HEAD(&new->disks);
272 INIT_LIST_HEAD(&new->all_mddevs);
273 init_timer(&new->safemode_timer);
274 atomic_set(&new->active, 1);
06d91a5f 275 spin_lock_init(&new->write_lock);
3d310eb7 276 init_waitqueue_head(&new->sb_wait);
08a02ecd 277 new->reshape_position = MaxSector;
c6207277 278 new->resync_max = MaxSector;
1da177e4
LT		 279
280 new->queue = blk_alloc_queue(GFP_KERNEL);
281 if (!new->queue) {
282 kfree(new);
283 return NULL;
284 }
89e5c8b5 285 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
1da177e4
LT		 286
287 blk_queue_make_request(new->queue, md_fail_request);
288
289 goto retry;
290}
291
292static inline int mddev_lock(mddev_t * mddev)
293{
df5b89b3 294 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 295}
296
1da177e4
LT		 297static inline int mddev_trylock(mddev_t * mddev)
298{
df5b89b3 299 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 300}
301
302static inline void mddev_unlock(mddev_t * mddev)
303{
df5b89b3 304 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 305
005eca5e 306 md_wakeup_thread(mddev->thread);
1da177e4
LT		 307}
308
2989ddbd 309static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4
LT		 310{
311 mdk_rdev_t * rdev;
312 struct list_head *tmp;
313
314 ITERATE_RDEV(mddev,rdev,tmp) {
315 if (rdev->desc_nr == nr)
316 return rdev;
317 }
318 return NULL;
319}
320
321static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
322{
323 struct list_head *tmp;
324 mdk_rdev_t *rdev;
325
326 ITERATE_RDEV(mddev,rdev,tmp) {
327 if (rdev->bdev->bd_dev == dev)
328 return rdev;
329 }
330 return NULL;
331}
332
d9d166c2 333static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 334{
335 struct mdk_personality *pers;
d9d166c2
N		 336 list_for_each_entry(pers, &pers_list, list) {
337 if (level != LEVEL_NONE && pers->level == level)
2604b703 338 return pers;
d9d166c2
N		 339 if (strcmp(pers->name, clevel)==0)
340 return pers;
341 }
2604b703
N		 342 return NULL;
343}
344
77933d72 345static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4
LT		 346{
347 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
348 return MD_NEW_SIZE_BLOCKS(size);
349}
350
351static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
352{
353 sector_t size;
354
355 size = rdev->sb_offset;
356
357 if (chunk_size)
358 size &= ~((sector_t)chunk_size/1024 - 1);
359 return size;
360}
361
362static int alloc_disk_sb(mdk_rdev_t * rdev)
363{
364 if (rdev->sb_page)
365 MD_BUG();
366
367 rdev->sb_page = alloc_page(GFP_KERNEL);
368 if (!rdev->sb_page) {
369 printk(KERN_ALERT "md: out of memory.\n");
370 return -EINVAL;
371 }
372
373 return 0;
374}
375
376static void free_disk_sb(mdk_rdev_t * rdev)
377{
378 if (rdev->sb_page) {
2d1f3b5d 379 put_page(rdev->sb_page);
1da177e4
LT		 380 rdev->sb_loaded = 0;
381 rdev->sb_page = NULL;
382 rdev->sb_offset = 0;
383 rdev->size = 0;
384 }
385}
386
387
6712ecf8 388static void super_written(struct bio *bio, int error)
7bfa19f2
N		 389{
390 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 391 mddev_t *mddev = rdev->mddev;
7bfa19f2 392
3a0f5bbb
N		 393 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
394 printk("md: super_written gets error=%d, uptodate=%d\n",
395 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
396 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 397 md_error(mddev, rdev);
3a0f5bbb 398 }
7bfa19f2 399
a9701a30
N		 400 if (atomic_dec_and_test(&mddev->pending_writes))
401 wake_up(&mddev->sb_wait);
f8b58edf 402 bio_put(bio);
7bfa19f2
N		 403}
404
6712ecf8 405static void super_written_barrier(struct bio *bio, int error)
a9701a30
N		 406{
407 struct bio *bio2 = bio->bi_private;
408 mdk_rdev_t *rdev = bio2->bi_private;
409 mddev_t *mddev = rdev->mddev;
a9701a30
N		 410
411 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
412 error == -EOPNOTSUPP) {
413 unsigned long flags;
414 /* barriers don't appear to be supported :-( */
415 set_bit(BarriersNotsupp, &rdev->flags);
416 mddev->barriers_work = 0;
417 spin_lock_irqsave(&mddev->write_lock, flags);
418 bio2->bi_next = mddev->biolist;
419 mddev->biolist = bio2;
420 spin_unlock_irqrestore(&mddev->write_lock, flags);
421 wake_up(&mddev->sb_wait);
422 bio_put(bio);
6712ecf8
N		 423 } else {
424 bio_put(bio2);
425 bio->bi_private = rdev;
426 super_written(bio, error);
a9701a30 427 }
a9701a30
N		 428}
429
7bfa19f2
N		 430void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
431 sector_t sector, int size, struct page *page)
432{
433 /* write first size bytes of page to sector of rdev
434 * Increment mddev->pending_writes before returning
435 * and decrement it on completion, waking up sb_wait
436 * if zero is reached.
437 * If an error occurred, call md_error
a9701a30
N		 438 *
439 * As we might need to resubmit the request if BIO_RW_BARRIER
440 * causes ENOTSUPP, we allocate a spare bio...
7bfa19f2
N		 441 */
442 struct bio *bio = bio_alloc(GFP_NOIO, 1);
a9701a30 443 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
7bfa19f2
N		 444
445 bio->bi_bdev = rdev->bdev;
446 bio->bi_sector = sector;
447 bio_add_page(bio, page, size, 0);
448 bio->bi_private = rdev;
449 bio->bi_end_io = super_written;
a9701a30
N		 450 bio->bi_rw = rw;
451
7bfa19f2 452 atomic_inc(&mddev->pending_writes);
a9701a30
N		 453 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
454 struct bio *rbio;
455 rw |= (1<<BIO_RW_BARRIER);
456 rbio = bio_clone(bio, GFP_NOIO);
457 rbio->bi_private = bio;
458 rbio->bi_end_io = super_written_barrier;
459 submit_bio(rw, rbio);
460 } else
461 submit_bio(rw, bio);
462}
463
464void md_super_wait(mddev_t *mddev)
465{
466 /* wait for all superblock writes that were scheduled to complete.
467 * if any had to be retried (due to BARRIER problems), retry them
468 */
469 DEFINE_WAIT(wq);
470 for(;;) {
471 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
472 if (atomic_read(&mddev->pending_writes)==0)
473 break;
474 while (mddev->biolist) {
475 struct bio *bio;
476 spin_lock_irq(&mddev->write_lock);
477 bio = mddev->biolist;
478 mddev->biolist = bio->bi_next ;
479 bio->bi_next = NULL;
480 spin_unlock_irq(&mddev->write_lock);
481 submit_bio(bio->bi_rw, bio);
482 }
483 schedule();
484 }
485 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 486}
487
6712ecf8 488static void bi_complete(struct bio *bio, int error)
1da177e4 489{
1da177e4 490 complete((struct completion*)bio->bi_private);
1da177e4
LT		 491}
492
a654b9d8 493int sync_page_io(struct block_device *bdev, sector_t sector, int size,
1da177e4
LT		 494 struct page *page, int rw)
495{
baaa2c51 496 struct bio *bio = bio_alloc(GFP_NOIO, 1);
1da177e4
LT		 497 struct completion event;
498 int ret;
499
500 rw |= (1 << BIO_RW_SYNC);
501
502 bio->bi_bdev = bdev;
503 bio->bi_sector = sector;
504 bio_add_page(bio, page, size, 0);
505 init_completion(&event);
506 bio->bi_private = &event;
507 bio->bi_end_io = bi_complete;
508 submit_bio(rw, bio);
509 wait_for_completion(&event);
510
511 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
512 bio_put(bio);
513 return ret;
514}
a8745db2 515EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 516
0002b271 517static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 518{
519 char b[BDEVNAME_SIZE];
520 if (!rdev->sb_page) {
521 MD_BUG();
522 return -EINVAL;
523 }
524 if (rdev->sb_loaded)
525 return 0;
526
527
0002b271 528 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
1da177e4
LT		 529 goto fail;
530 rdev->sb_loaded = 1;
531 return 0;
532
533fail:
534 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
535 bdevname(rdev->bdev,b));
536 return -EINVAL;
537}
538
539static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
540{
541 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
542 (sb1->set_uuid1 == sb2->set_uuid1) &&
543 (sb1->set_uuid2 == sb2->set_uuid2) &&
544 (sb1->set_uuid3 == sb2->set_uuid3))
545
546 return 1;
547
548 return 0;
549}
550
551
552static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
553{
554 int ret;
555 mdp_super_t *tmp1, *tmp2;
556
557 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
558 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
559
560 if (!tmp1 || !tmp2) {
561 ret = 0;
562 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
563 goto abort;
564 }
565
566 *tmp1 = *sb1;
567 *tmp2 = *sb2;
568
569 /*
570 * nr_disks is not constant
571 */
572 tmp1->nr_disks = 0;
573 tmp2->nr_disks = 0;
574
575 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
576 ret = 0;
577 else
578 ret = 1;
579
580abort:
990a8baf
JJ		 581 kfree(tmp1);
582 kfree(tmp2);
1da177e4
LT		 583 return ret;
584}
585
4d167f09
N		 586
587static u32 md_csum_fold(u32 csum)
588{
589 csum = (csum & 0xffff) + (csum >> 16);
590 return (csum & 0xffff) + (csum >> 16);
591}
592
1da177e4
LT		 593static unsigned int calc_sb_csum(mdp_super_t * sb)
594{
4d167f09
N		 595 u64 newcsum = 0;
596 u32 *sb32 = (u32*)sb;
597 int i;
1da177e4
LT		 598 unsigned int disk_csum, csum;
599
600 disk_csum = sb->sb_csum;
601 sb->sb_csum = 0;
4d167f09
N		 602
603 for (i = 0; i < MD_SB_BYTES/4 ; i++)
604 newcsum += sb32[i];
605 csum = (newcsum & 0xffffffff) + (newcsum>>32);
606
607
608#ifdef CONFIG_ALPHA
609 /* This used to use csum_partial, which was wrong for several
610 * reasons including that different results are returned on
611 * different architectures. It isn't critical that we get exactly
612 * the same return value as before (we always csum_fold before
613 * testing, and that removes any differences). However as we
614 * know that csum_partial always returned a 16bit value on
615 * alphas, do a fold to maximise conformity to previous behaviour.
616 */
617 sb->sb_csum = md_csum_fold(disk_csum);
618#else
1da177e4 619 sb->sb_csum = disk_csum;
4d167f09 620#endif
1da177e4
LT		 621 return csum;
622}
623
624
625/*
626 * Handle superblock details.
627 * We want to be able to handle multiple superblock formats
628 * so we have a common interface to them all, and an array of
629 * different handlers.
630 * We rely on user-space to write the initial superblock, and support
631 * reading and updating of superblocks.
632 * Interface methods are:
633 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
634 * loads and validates a superblock on dev.
635 * if refdev != NULL, compare superblocks on both devices
636 * Return:
637 * 0 - dev has a superblock that is compatible with refdev
638 * 1 - dev has a superblock that is compatible and newer than refdev
639 * so dev should be used as the refdev in future
640 * -EINVAL superblock incompatible or invalid
641 * -othererror e.g. -EIO
642 *
643 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
644 * Verify that dev is acceptable into mddev.
645 * The first time, mddev->raid_disks will be 0, and data from
646 * dev should be merged in. Subsequent calls check that dev
647 * is new enough. Return 0 or -EINVAL
648 *
649 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
650 * Update the superblock for rdev with data in mddev
651 * This does not write to disc.
652 *
653 */
654
655struct super_type {
656 char *name;
657 struct module *owner;
658 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
659 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
660 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
661};
662
663/*
664 * load_super for 0.90.0
665 */
666static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
667{
668 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
669 mdp_super_t *sb;
670 int ret;
671 sector_t sb_offset;
672
673 /*
674 * Calculate the position of the superblock,
675 * it's at the end of the disk.
676 *
677 * It also happens to be a multiple of 4Kb.
678 */
679 sb_offset = calc_dev_sboffset(rdev->bdev);
680 rdev->sb_offset = sb_offset;
681
0002b271 682 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 683 if (ret) return ret;
684
685 ret = -EINVAL;
686
687 bdevname(rdev->bdev, b);
688 sb = (mdp_super_t*)page_address(rdev->sb_page);
689
690 if (sb->md_magic != MD_SB_MAGIC) {
691 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
692 b);
693 goto abort;
694 }
695
696 if (sb->major_version != 0 ||
f6705578
N		 697 sb->minor_version < 90 ||
698 sb->minor_version > 91) {
1da177e4
LT		 699 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
700 sb->major_version, sb->minor_version,
701 b);
702 goto abort;
703 }
704
705 if (sb->raid_disks <= 0)
706 goto abort;
707
4d167f09 708 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 709 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
710 b);
711 goto abort;
712 }
713
714 rdev->preferred_minor = sb->md_minor;
715 rdev->data_offset = 0;
0002b271 716 rdev->sb_size = MD_SB_BYTES;
1da177e4 717
e11e93fa
N		 718 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
719 if (sb->level != 1 && sb->level != 4
720 && sb->level != 5 && sb->level != 6
721 && sb->level != 10) {
722 /* FIXME use a better test */
723 printk(KERN_WARNING
724 "md: bitmaps not supported for this level.\n");
725 goto abort;
726 }
727 }
728
1da177e4
LT		 729 if (sb->level == LEVEL_MULTIPATH)
730 rdev->desc_nr = -1;
731 else
732 rdev->desc_nr = sb->this_disk.number;
733
734 if (refdev == 0)
735 ret = 1;
736 else {
737 __u64 ev1, ev2;
738 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
739 if (!uuid_equal(refsb, sb)) {
740 printk(KERN_WARNING "md: %s has different UUID to %s\n",
741 b, bdevname(refdev->bdev,b2));
742 goto abort;
743 }
744 if (!sb_equal(refsb, sb)) {
745 printk(KERN_WARNING "md: %s has same UUID"
746 " but different superblock to %s\n",
747 b, bdevname(refdev->bdev, b2));
748 goto abort;
749 }
750 ev1 = md_event(sb);
751 ev2 = md_event(refsb);
752 if (ev1 > ev2)
753 ret = 1;
754 else
755 ret = 0;
756 }
757 rdev->size = calc_dev_size(rdev, sb->chunk_size);
758
2bf071bf
N		 759 if (rdev->size < sb->size && sb->level > 1)
760 /* "this cannot possibly happen" ... */
761 ret = -EINVAL;
762
1da177e4
LT		 763 abort:
764 return ret;
765}
766
767/*
768 * validate_super for 0.90.0
769 */
770static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
771{
772 mdp_disk_t *desc;
773 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
07d84d10 774 __u64 ev1 = md_event(sb);
1da177e4 775
41158c7e 776 rdev->raid_disk = -1;
c5d79adb
N		 777 clear_bit(Faulty, &rdev->flags);
778 clear_bit(In_sync, &rdev->flags);
779 clear_bit(WriteMostly, &rdev->flags);
780 clear_bit(BarriersNotsupp, &rdev->flags);
781
1da177e4
LT		 782 if (mddev->raid_disks == 0) {
783 mddev->major_version = 0;
784 mddev->minor_version = sb->minor_version;
785 mddev->patch_version = sb->patch_version;
e691063a 786 mddev->external = 0;
1da177e4
LT		 787 mddev->chunk_size = sb->chunk_size;
788 mddev->ctime = sb->ctime;
789 mddev->utime = sb->utime;
790 mddev->level = sb->level;
d9d166c2 791 mddev->clevel[0] = 0;
1da177e4
LT		 792 mddev->layout = sb->layout;
793 mddev->raid_disks = sb->raid_disks;
794 mddev->size = sb->size;
07d84d10 795 mddev->events = ev1;
9223214e 796 mddev->bitmap_offset = 0;
36fa3063 797 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
1da177e4 798
f6705578
N		 799 if (mddev->minor_version >= 91) {
800 mddev->reshape_position = sb->reshape_position;
801 mddev->delta_disks = sb->delta_disks;
802 mddev->new_level = sb->new_level;
803 mddev->new_layout = sb->new_layout;
804 mddev->new_chunk = sb->new_chunk;
805 } else {
806 mddev->reshape_position = MaxSector;
807 mddev->delta_disks = 0;
808 mddev->new_level = mddev->level;
809 mddev->new_layout = mddev->layout;
810 mddev->new_chunk = mddev->chunk_size;
811 }
812
1da177e4
LT		 813 if (sb->state & (1<<MD_SB_CLEAN))
814 mddev->recovery_cp = MaxSector;
815 else {
816 if (sb->events_hi == sb->cp_events_hi &&
817 sb->events_lo == sb->cp_events_lo) {
818 mddev->recovery_cp = sb->recovery_cp;
819 } else
820 mddev->recovery_cp = 0;
821 }
822
823 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
824 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
825 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
826 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
827
828 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 829
830 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
e11e93fa 831 mddev->bitmap_file == NULL)
36fa3063 832 mddev->bitmap_offset = mddev->default_bitmap_offset;
a654b9d8 833
41158c7e
N		 834 } else if (mddev->pers == NULL) {
835 /* Insist on good event counter while assembling */
1da177e4
LT		 836 ++ev1;
837 if (ev1 < mddev->events)
838 return -EINVAL;
41158c7e
N		 839 } else if (mddev->bitmap) {
840 /* if adding to array with a bitmap, then we can accept an
841 * older device ... but not too old.
842 */
41158c7e
N		 843 if (ev1 < mddev->bitmap->events_cleared)
844 return 0;
07d84d10
N		 845 } else {
846 if (ev1 < mddev->events)
847 /* just a hot-add of a new device, leave raid_disk at -1 */
848 return 0;
849 }
41158c7e 850
1da177e4 851 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 852 desc = sb->disks + rdev->desc_nr;
853
854 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 855 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 856 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
857 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 858 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 859 rdev->raid_disk = desc->raid_disk;
860 }
8ddf9efe
N		 861 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
862 set_bit(WriteMostly, &rdev->flags);
41158c7e 863 } else /* MULTIPATH are always insync */
b2d444d7 864 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 865 return 0;
866}
867
868/*
869 * sync_super for 0.90.0
870 */
871static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
872{
873 mdp_super_t *sb;
874 struct list_head *tmp;
875 mdk_rdev_t *rdev2;
876 int next_spare = mddev->raid_disks;
19133a42 877
1da177e4
LT		 878
879 /* make rdev->sb match mddev data..
880 *
881 * 1/ zero out disks
882 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
883 * 3/ any empty disks < next_spare become removed
884 *
885 * disks[0] gets initialised to REMOVED because
886 * we cannot be sure from other fields if it has
887 * been initialised or not.
888 */
889 int i;
890 int active=0, working=0,failed=0,spare=0,nr_disks=0;
891
61181565
N		 892 rdev->sb_size = MD_SB_BYTES;
893
1da177e4
LT		 894 sb = (mdp_super_t*)page_address(rdev->sb_page);
895
896 memset(sb, 0, sizeof(*sb));
897
898 sb->md_magic = MD_SB_MAGIC;
899 sb->major_version = mddev->major_version;
1da177e4
LT		 900 sb->patch_version = mddev->patch_version;
901 sb->gvalid_words = 0; /* ignored */
902 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
903 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
904 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
905 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
906
907 sb->ctime = mddev->ctime;
908 sb->level = mddev->level;
909 sb->size = mddev->size;
910 sb->raid_disks = mddev->raid_disks;
911 sb->md_minor = mddev->md_minor;
e691063a 912 sb->not_persistent = 0;
1da177e4
LT		 913 sb->utime = mddev->utime;
914 sb->state = 0;
915 sb->events_hi = (mddev->events>>32);
916 sb->events_lo = (u32)mddev->events;
917
f6705578
N		 918 if (mddev->reshape_position == MaxSector)
919 sb->minor_version = 90;
920 else {
921 sb->minor_version = 91;
922 sb->reshape_position = mddev->reshape_position;
923 sb->new_level = mddev->new_level;
924 sb->delta_disks = mddev->delta_disks;
925 sb->new_layout = mddev->new_layout;
926 sb->new_chunk = mddev->new_chunk;
927 }
928 mddev->minor_version = sb->minor_version;
1da177e4
LT		 929 if (mddev->in_sync)
930 {
931 sb->recovery_cp = mddev->recovery_cp;
932 sb->cp_events_hi = (mddev->events>>32);
933 sb->cp_events_lo = (u32)mddev->events;
934 if (mddev->recovery_cp == MaxSector)
935 sb->state = (1<< MD_SB_CLEAN);
936 } else
937 sb->recovery_cp = 0;
938
939 sb->layout = mddev->layout;
940 sb->chunk_size = mddev->chunk_size;
941
a654b9d8
N		 942 if (mddev->bitmap && mddev->bitmap_file == NULL)
943 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
944
1da177e4
LT		 945 sb->disks[0].state = (1<<MD_DISK_REMOVED);
946 ITERATE_RDEV(mddev,rdev2,tmp) {
947 mdp_disk_t *d;
86e6ffdd 948 int desc_nr;
b2d444d7
N		 949 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
950 && !test_bit(Faulty, &rdev2->flags))
86e6ffdd 951 desc_nr = rdev2->raid_disk;
1da177e4 952 else
86e6ffdd 953 desc_nr = next_spare++;
19133a42 954 rdev2->desc_nr = desc_nr;
1da177e4
LT		 955 d = &sb->disks[rdev2->desc_nr];
956 nr_disks++;
957 d->number = rdev2->desc_nr;
958 d->major = MAJOR(rdev2->bdev->bd_dev);
959 d->minor = MINOR(rdev2->bdev->bd_dev);
b2d444d7
N		 960 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
961 && !test_bit(Faulty, &rdev2->flags))
1da177e4
LT		 962 d->raid_disk = rdev2->raid_disk;
963 else
964 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 965 if (test_bit(Faulty, &rdev2->flags))
1da177e4 966 d->state = (1<<MD_DISK_FAULTY);
1be7892f 967 else if (test_bit(In_sync, &rdev2->flags)) {
1da177e4
LT		 968 d->state = (1<<MD_DISK_ACTIVE);
969 d->state |= (1<<MD_DISK_SYNC);
970 active++;
971 working++;
972 } else {
973 d->state = 0;
974 spare++;
975 working++;
976 }
8ddf9efe
N		 977 if (test_bit(WriteMostly, &rdev2->flags))
978 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 979 }
1da177e4
LT		 980 /* now set the "removed" and "faulty" bits on any missing devices */
981 for (i=0 ; i < mddev->raid_disks ; i++) {
982 mdp_disk_t *d = &sb->disks[i];
983 if (d->state == 0 && d->number == 0) {
984 d->number = i;
985 d->raid_disk = i;
986 d->state = (1<<MD_DISK_REMOVED);
987 d->state |= (1<<MD_DISK_FAULTY);
988 failed++;
989 }
990 }
991 sb->nr_disks = nr_disks;
992 sb->active_disks = active;
993 sb->working_disks = working;
994 sb->failed_disks = failed;
995 sb->spare_disks = spare;
996
997 sb->this_disk = sb->disks[rdev->desc_nr];
998 sb->sb_csum = calc_sb_csum(sb);
999}
1000
1001/*
1002 * version 1 superblock
1003 */
1004
1c05b4bc 1005static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1006{
1c05b4bc
N		 1007 __le32 disk_csum;
1008 u32 csum;
1da177e4
LT		 1009 unsigned long long newcsum;
1010 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1011 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1012 int i;
1013
1014 disk_csum = sb->sb_csum;
1015 sb->sb_csum = 0;
1016 newcsum = 0;
1017 for (i=0; size>=4; size -= 4 )
1018 newcsum += le32_to_cpu(*isuper++);
1019
1020 if (size == 2)
1c05b4bc 1021 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1022
1023 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1024 sb->sb_csum = disk_csum;
1025 return cpu_to_le32(csum);
1026}
1027
1028static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1029{
1030 struct mdp_superblock_1 *sb;
1031 int ret;
1032 sector_t sb_offset;
1033 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1034 int bmask;
1da177e4
LT		 1035
1036 /*
1037 * Calculate the position of the superblock.
1038 * It is always aligned to a 4K boundary and
1039 * depeding on minor_version, it can be:
1040 * 0: At least 8K, but less than 12K, from end of device
1041 * 1: At start of device
1042 * 2: 4K from start of device.
1043 */
1044 switch(minor_version) {
1045 case 0:
1046 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1047 sb_offset -= 8*2;
39730960 1048 sb_offset &= ~(sector_t)(4*2-1);
1da177e4
LT		 1049 /* convert from sectors to K */
1050 sb_offset /= 2;
1051 break;
1052 case 1:
1053 sb_offset = 0;
1054 break;
1055 case 2:
1056 sb_offset = 4;
1057 break;
1058 default:
1059 return -EINVAL;
1060 }
1061 rdev->sb_offset = sb_offset;
1062
0002b271
N		 1063 /* superblock is rarely larger than 1K, but it can be larger,
1064 * and it is safe to read 4k, so we do that
1065 */
1066 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1067 if (ret) return ret;
1068
1069
1070 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1071
1072 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1073 sb->major_version != cpu_to_le32(1) ||
1074 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1075 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
71c0805c 1076 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1077 return -EINVAL;
1078
1079 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1080 printk("md: invalid superblock checksum on %s\n",
1081 bdevname(rdev->bdev,b));
1082 return -EINVAL;
1083 }
1084 if (le64_to_cpu(sb->data_size) < 10) {
1085 printk("md: data_size too small on %s\n",
1086 bdevname(rdev->bdev,b));
1087 return -EINVAL;
1088 }
e11e93fa
N		 1089 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1090 if (sb->level != cpu_to_le32(1) &&
1091 sb->level != cpu_to_le32(4) &&
1092 sb->level != cpu_to_le32(5) &&
1093 sb->level != cpu_to_le32(6) &&
1094 sb->level != cpu_to_le32(10)) {
1095 printk(KERN_WARNING
1096 "md: bitmaps not supported for this level.\n");
1097 return -EINVAL;
1098 }
1099 }
1100
1da177e4
LT		 1101 rdev->preferred_minor = 0xffff;
1102 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1103 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1104
0002b271 1105 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
720a3dc3 1106 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
0002b271
N		 1107 if (rdev->sb_size & bmask)
1108 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1109
31b65a0d
N		 1110 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1111 rdev->desc_nr = -1;
1112 else
1113 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1114
1da177e4 1115 if (refdev == 0)
8ed75463 1116 ret = 1;
1da177e4
LT		 1117 else {
1118 __u64 ev1, ev2;
1119 struct mdp_superblock_1 *refsb =
1120 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1121
1122 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1123 sb->level != refsb->level ||
1124 sb->layout != refsb->layout ||
1125 sb->chunksize != refsb->chunksize) {
1126 printk(KERN_WARNING "md: %s has strangely different"
1127 " superblock to %s\n",
1128 bdevname(rdev->bdev,b),
1129 bdevname(refdev->bdev,b2));
1130 return -EINVAL;
1131 }
1132 ev1 = le64_to_cpu(sb->events);
1133 ev2 = le64_to_cpu(refsb->events);
1134
1135 if (ev1 > ev2)
8ed75463
N		 1136 ret = 1;
1137 else
1138 ret = 0;
1da177e4
LT		 1139 }
1140 if (minor_version)
1141 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1142 else
1143 rdev->size = rdev->sb_offset;
1144 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1145 return -EINVAL;
1146 rdev->size = le64_to_cpu(sb->data_size)/2;
1147 if (le32_to_cpu(sb->chunksize))
1148 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
2bf071bf 1149
1c05b4bc 1150 if (le64_to_cpu(sb->size) > rdev->size*2)
2bf071bf 1151 return -EINVAL;
8ed75463 1152 return ret;
1da177e4
LT		 1153}
1154
1155static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1156{
1157 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
07d84d10 1158 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1159
41158c7e 1160 rdev->raid_disk = -1;
c5d79adb
N		 1161 clear_bit(Faulty, &rdev->flags);
1162 clear_bit(In_sync, &rdev->flags);
1163 clear_bit(WriteMostly, &rdev->flags);
1164 clear_bit(BarriersNotsupp, &rdev->flags);
1165
1da177e4
LT		 1166 if (mddev->raid_disks == 0) {
1167 mddev->major_version = 1;
1168 mddev->patch_version = 0;
e691063a 1169 mddev->external = 0;
1da177e4
LT		 1170 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1171 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1172 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1173 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1174 mddev->clevel[0] = 0;
1da177e4
LT		 1175 mddev->layout = le32_to_cpu(sb->layout);
1176 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1177 mddev->size = le64_to_cpu(sb->size)/2;
07d84d10 1178 mddev->events = ev1;
9223214e 1179 mddev->bitmap_offset = 0;
29fc7e3e 1180 mddev->default_bitmap_offset = 1024 >> 9;
1da177e4
LT		 1181
1182 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1183 memcpy(mddev->uuid, sb->set_uuid, 16);
1184
1185 mddev->max_disks = (4096-256)/2;
a654b9d8 1186
71c0805c 1187 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
e11e93fa 1188 mddev->bitmap_file == NULL )
a654b9d8 1189 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1190
f6705578
N		 1191 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1192 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1193 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1194 mddev->new_level = le32_to_cpu(sb->new_level);
1195 mddev->new_layout = le32_to_cpu(sb->new_layout);
1196 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1197 } else {
1198 mddev->reshape_position = MaxSector;
1199 mddev->delta_disks = 0;
1200 mddev->new_level = mddev->level;
1201 mddev->new_layout = mddev->layout;
1202 mddev->new_chunk = mddev->chunk_size;
1203 }
1204
41158c7e
N		 1205 } else if (mddev->pers == NULL) {
1206 /* Insist of good event counter while assembling */
1da177e4
LT		 1207 ++ev1;
1208 if (ev1 < mddev->events)
1209 return -EINVAL;
41158c7e
N		 1210 } else if (mddev->bitmap) {
1211 /* If adding to array with a bitmap, then we can accept an
1212 * older device, but not too old.
1213 */
41158c7e
N		 1214 if (ev1 < mddev->bitmap->events_cleared)
1215 return 0;
07d84d10
N		 1216 } else {
1217 if (ev1 < mddev->events)
1218 /* just a hot-add of a new device, leave raid_disk at -1 */
1219 return 0;
1220 }
1da177e4
LT		 1221 if (mddev->level != LEVEL_MULTIPATH) {
1222 int role;
1da177e4
LT		 1223 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1224 switch(role) {
1225 case 0xffff: /* spare */
1da177e4
LT		 1226 break;
1227 case 0xfffe: /* faulty */
b2d444d7 1228 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1229 break;
1230 default:
5fd6c1dc
N		 1231 if ((le32_to_cpu(sb->feature_map) &
1232 MD_FEATURE_RECOVERY_OFFSET))
1233 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1234 else
1235 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1236 rdev->raid_disk = role;
1237 break;
1238 }
8ddf9efe
N		 1239 if (sb->devflags & WriteMostly1)
1240 set_bit(WriteMostly, &rdev->flags);
41158c7e 1241 } else /* MULTIPATH are always insync */
b2d444d7 1242 set_bit(In_sync, &rdev->flags);
41158c7e 1243
1da177e4
LT		 1244 return 0;
1245}
1246
1247static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1248{
1249 struct mdp_superblock_1 *sb;
1250 struct list_head *tmp;
1251 mdk_rdev_t *rdev2;
1252 int max_dev, i;
1253 /* make rdev->sb match mddev and rdev data. */
1254
1255 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1256
1257 sb->feature_map = 0;
1258 sb->pad0 = 0;
5fd6c1dc 1259 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1260 memset(sb->pad1, 0, sizeof(sb->pad1));
1261 memset(sb->pad2, 0, sizeof(sb->pad2));
1262 memset(sb->pad3, 0, sizeof(sb->pad3));
1263
1264 sb->utime = cpu_to_le64((__u64)mddev->utime);
1265 sb->events = cpu_to_le64(mddev->events);
1266 if (mddev->in_sync)
1267 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1268 else
1269 sb->resync_offset = cpu_to_le64(0);
1270
1c05b4bc 1271 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1272
f0ca340c 1273 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
29fc7e3e 1274 sb->size = cpu_to_le64(mddev->size<<1);
f0ca340c 1275
a654b9d8
N		 1276 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1277 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1278 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1279 }
5fd6c1dc
N		 1280
1281 if (rdev->raid_disk >= 0 &&
1282 !test_bit(In_sync, &rdev->flags) &&
1283 rdev->recovery_offset > 0) {
1284 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1285 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1286 }
1287
f6705578
N		 1288 if (mddev->reshape_position != MaxSector) {
1289 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1290 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1291 sb->new_layout = cpu_to_le32(mddev->new_layout);
1292 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1293 sb->new_level = cpu_to_le32(mddev->new_level);
1294 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1295 }
a654b9d8 1296
1da177e4
LT		 1297 max_dev = 0;
1298 ITERATE_RDEV(mddev,rdev2,tmp)
1299 if (rdev2->desc_nr+1 > max_dev)
1300 max_dev = rdev2->desc_nr+1;
a778b73f
N		 1301
1302 if (max_dev > le32_to_cpu(sb->max_dev))
1303 sb->max_dev = cpu_to_le32(max_dev);
1da177e4
LT		 1304 for (i=0; i<max_dev;i++)
1305 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1306
1307 ITERATE_RDEV(mddev,rdev2,tmp) {
1308 i = rdev2->desc_nr;
b2d444d7 1309 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1310 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1311 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1312 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
5fd6c1dc
N		 1313 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1314 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1315 else
1316 sb->dev_roles[i] = cpu_to_le16(0xffff);
1317 }
1318
1da177e4
LT		 1319 sb->sb_csum = calc_sb_1_csum(sb);
1320}
1321
1322
75c96f85 1323static struct super_type super_types[] = {
1da177e4
LT		 1324 [0] = {
1325 .name = "0.90.0",
1326 .owner = THIS_MODULE,
1327 .load_super = super_90_load,
1328 .validate_super = super_90_validate,
1329 .sync_super = super_90_sync,
1330 },
1331 [1] = {
1332 .name = "md-1",
1333 .owner = THIS_MODULE,
1334 .load_super = super_1_load,
1335 .validate_super = super_1_validate,
1336 .sync_super = super_1_sync,
1337 },
1338};
1da177e4
LT		 1339
1340static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1341{
7dd5e7c3
N		 1342 struct list_head *tmp, *tmp2;
1343 mdk_rdev_t *rdev, *rdev2;
1da177e4
LT		 1344
1345 ITERATE_RDEV(mddev1,rdev,tmp)
7dd5e7c3
N		 1346 ITERATE_RDEV(mddev2, rdev2, tmp2)
1347 if (rdev->bdev->bd_contains ==
1348 rdev2->bdev->bd_contains)
1349 return 1;
1da177e4
LT		 1350
1351 return 0;
1352}
1353
1354static LIST_HEAD(pending_raid_disks);
1355
1356static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1357{
7dd5e7c3 1358 char b[BDEVNAME_SIZE];
f637b9f9 1359 struct kobject *ko;
1edf80d3 1360 char *s;
5e55e2f5 1361 int err;
1da177e4
LT		 1362
1363 if (rdev->mddev) {
1364 MD_BUG();
1365 return -EINVAL;
1366 }
2bf071bf
N		 1367 /* make sure rdev->size exceeds mddev->size */
1368 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
a778b73f
N		 1369 if (mddev->pers) {
1370 /* Cannot change size, so fail
1371 * If mddev->level <= 0, then we don't care
1372 * about aligning sizes (e.g. linear)
1373 */
1374 if (mddev->level > 0)
1375 return -ENOSPC;
1376 } else
2bf071bf
N		 1377 mddev->size = rdev->size;
1378 }
1da177e4
LT		 1379
1380 /* Verify rdev->desc_nr is unique.
1381 * If it is -1, assign a free number, else
1382 * check number is not in use
1383 */
1384 if (rdev->desc_nr < 0) {
1385 int choice = 0;
1386 if (mddev->pers) choice = mddev->raid_disks;
1387 while (find_rdev_nr(mddev, choice))
1388 choice++;
1389 rdev->desc_nr = choice;
1390 } else {
1391 if (find_rdev_nr(mddev, rdev->desc_nr))
1392 return -EBUSY;
1393 }
19133a42 1394 bdevname(rdev->bdev,b);
649316b2 1395 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 1396 *s = '!';
649316b2 1397
1da177e4 1398 rdev->mddev = mddev;
19133a42 1399 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1400
b2d6db58 1401 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 1402 goto fail;
86e6ffdd 1403
f637b9f9 1404 if (rdev->bdev->bd_part)
edfaa7c3 1405 ko = &rdev->bdev->bd_part->dev.kobj;
f637b9f9 1406 else
edfaa7c3 1407 ko = &rdev->bdev->bd_disk->dev.kobj;
5e55e2f5
N		 1408 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1409 kobject_del(&rdev->kobj);
1410 goto fail;
1411 }
1412 list_add(&rdev->same_set, &mddev->disks);
c5d79adb 1413 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1da177e4 1414 return 0;
5e55e2f5
N		 1415
1416 fail:
1417 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1418 b, mdname(mddev));
1419 return err;
1da177e4
LT		 1420}
1421
5792a285
N		 1422static void delayed_delete(struct work_struct *ws)
1423{
1424 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1425 kobject_del(&rdev->kobj);
1426}
1427
1da177e4
LT		 1428static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1429{
1430 char b[BDEVNAME_SIZE];
1431 if (!rdev->mddev) {
1432 MD_BUG();
1433 return;
1434 }
5463c790 1435 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1da177e4
LT		 1436 list_del_init(&rdev->same_set);
1437 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1438 rdev->mddev = NULL;
86e6ffdd 1439 sysfs_remove_link(&rdev->kobj, "block");
5792a285
N		 1440
1441 /* We need to delay this, otherwise we can deadlock when
1442 * writing to 'remove' to "dev/state"
1443 */
1444 INIT_WORK(&rdev->del_work, delayed_delete);
1445 schedule_work(&rdev->del_work);
1da177e4
LT		 1446}
1447
1448/*
1449 * prevent the device from being mounted, repartitioned or
1450 * otherwise reused by a RAID array (or any other kernel
1451 * subsystem), by bd_claiming the device.
1452 */
c5d79adb 1453static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1da177e4
LT		 1454{
1455 int err = 0;
1456 struct block_device *bdev;
1457 char b[BDEVNAME_SIZE];
1458
2e7b651d 1459 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1460 if (IS_ERR(bdev)) {
1461 printk(KERN_ERR "md: could not open %s.\n",
1462 __bdevname(dev, b));
1463 return PTR_ERR(bdev);
1464 }
c5d79adb 1465 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1da177e4
LT		 1466 if (err) {
1467 printk(KERN_ERR "md: could not bd_claim %s.\n",
1468 bdevname(bdev, b));
2e7b651d 1469 blkdev_put(bdev);
1da177e4
LT		 1470 return err;
1471 }
c5d79adb
N		 1472 if (!shared)
1473 set_bit(AllReserved, &rdev->flags);
1da177e4
LT		 1474 rdev->bdev = bdev;
1475 return err;
1476}
1477
1478static void unlock_rdev(mdk_rdev_t *rdev)
1479{
1480 struct block_device *bdev = rdev->bdev;
1481 rdev->bdev = NULL;
1482 if (!bdev)
1483 MD_BUG();
1484 bd_release(bdev);
2e7b651d 1485 blkdev_put(bdev);
1da177e4
LT		 1486}
1487
1488void md_autodetect_dev(dev_t dev);
1489
1490static void export_rdev(mdk_rdev_t * rdev)
1491{
1492 char b[BDEVNAME_SIZE];
1493 printk(KERN_INFO "md: export_rdev(%s)\n",
1494 bdevname(rdev->bdev,b));
1495 if (rdev->mddev)
1496 MD_BUG();
1497 free_disk_sb(rdev);
1498 list_del_init(&rdev->same_set);
1499#ifndef MODULE
1500 md_autodetect_dev(rdev->bdev->bd_dev);
1501#endif
1502 unlock_rdev(rdev);
86e6ffdd 1503 kobject_put(&rdev->kobj);
1da177e4
LT		 1504}
1505
1506static void kick_rdev_from_array(mdk_rdev_t * rdev)
1507{
1508 unbind_rdev_from_array(rdev);
1509 export_rdev(rdev);
1510}
1511
1512static void export_array(mddev_t *mddev)
1513{
1514 struct list_head *tmp;
1515 mdk_rdev_t *rdev;
1516
1517 ITERATE_RDEV(mddev,rdev,tmp) {
1518 if (!rdev->mddev) {
1519 MD_BUG();
1520 continue;
1521 }
1522 kick_rdev_from_array(rdev);
1523 }
1524 if (!list_empty(&mddev->disks))
1525 MD_BUG();
1526 mddev->raid_disks = 0;
1527 mddev->major_version = 0;
1528}
1529
1530static void print_desc(mdp_disk_t *desc)
1531{
1532 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1533 desc->major,desc->minor,desc->raid_disk,desc->state);
1534}
1535
1536static void print_sb(mdp_super_t *sb)
1537{
1538 int i;
1539
1540 printk(KERN_INFO
1541 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1542 sb->major_version, sb->minor_version, sb->patch_version,
1543 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1544 sb->ctime);
1545 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1546 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1547 sb->md_minor, sb->layout, sb->chunk_size);
1548 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1549 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1550 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1551 sb->failed_disks, sb->spare_disks,
1552 sb->sb_csum, (unsigned long)sb->events_lo);
1553
1554 printk(KERN_INFO);
1555 for (i = 0; i < MD_SB_DISKS; i++) {
1556 mdp_disk_t *desc;
1557
1558 desc = sb->disks + i;
1559 if (desc->number || desc->major || desc->minor ||
1560 desc->raid_disk || (desc->state && (desc->state != 4))) {
1561 printk(" D %2d: ", i);
1562 print_desc(desc);
1563 }
1564 }
1565 printk(KERN_INFO "md: THIS: ");
1566 print_desc(&sb->this_disk);
1567
1568}
1569
1570static void print_rdev(mdk_rdev_t *rdev)
1571{
1572 char b[BDEVNAME_SIZE];
1573 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1574 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
b2d444d7
N		 1575 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1576 rdev->desc_nr);
1da177e4
LT		 1577 if (rdev->sb_loaded) {
1578 printk(KERN_INFO "md: rdev superblock:\n");
1579 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1580 } else
1581 printk(KERN_INFO "md: no rdev superblock!\n");
1582}
1583
5e56341d 1584static void md_print_devices(void)
1da177e4
LT		 1585{
1586 struct list_head *tmp, *tmp2;
1587 mdk_rdev_t *rdev;
1588 mddev_t *mddev;
1589 char b[BDEVNAME_SIZE];
1590
1591 printk("\n");
1592 printk("md: **********************************\n");
1593 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1594 printk("md: **********************************\n");
1595 ITERATE_MDDEV(mddev,tmp) {
1da177e4 1596
32a7627c
N		 1597 if (mddev->bitmap)
1598 bitmap_print_sb(mddev->bitmap);
1599 else
1600 printk("%s: ", mdname(mddev));
1da177e4
LT		 1601 ITERATE_RDEV(mddev,rdev,tmp2)
1602 printk("<%s>", bdevname(rdev->bdev,b));
1603 printk("\n");
1604
1605 ITERATE_RDEV(mddev,rdev,tmp2)
1606 print_rdev(rdev);
1607 }
1608 printk("md: **********************************\n");
1609 printk("\n");
1610}
1611
1612
42543769 1613static void sync_sbs(mddev_t * mddev, int nospares)
1da177e4 1614{
42543769
N		 1615 /* Update each superblock (in-memory image), but
1616 * if we are allowed to, skip spares which already
1617 * have the right event counter, or have one earlier
1618 * (which would mean they aren't being marked as dirty
1619 * with the rest of the array)
1620 */
1da177e4
LT		 1621 mdk_rdev_t *rdev;
1622 struct list_head *tmp;
1623
1624 ITERATE_RDEV(mddev,rdev,tmp) {
42543769
N		 1625 if (rdev->sb_events == mddev->events ||
1626 (nospares &&
1627 rdev->raid_disk < 0 &&
1628 (rdev->sb_events&1)==0 &&
1629 rdev->sb_events+1 == mddev->events)) {
1630 /* Don't update this superblock */
1631 rdev->sb_loaded = 2;
1632 } else {
1633 super_types[mddev->major_version].
1634 sync_super(mddev, rdev);
1635 rdev->sb_loaded = 1;
1636 }
1da177e4
LT		 1637 }
1638}
1639
850b2b42 1640static void md_update_sb(mddev_t * mddev, int force_change)
1da177e4 1641{
1da177e4
LT		 1642 struct list_head *tmp;
1643 mdk_rdev_t *rdev;
06d91a5f 1644 int sync_req;
42543769 1645 int nospares = 0;
1da177e4 1646
1da177e4 1647repeat:
a9701a30 1648 spin_lock_irq(&mddev->write_lock);
84692195 1649
850b2b42
N		 1650 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1651 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1652 force_change = 1;
1653 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1654 /* just a clean<-> dirty transition, possibly leave spares alone,
1655 * though if events isn't the right even/odd, we will have to do
1656 * spares after all
1657 */
1658 nospares = 1;
1659 if (force_change)
1660 nospares = 0;
1661 if (mddev->degraded)
84692195
N		 1662 /* If the array is degraded, then skipping spares is both
1663 * dangerous and fairly pointless.
1664 * Dangerous because a device that was removed from the array
1665 * might have a event_count that still looks up-to-date,
1666 * so it can be re-added without a resync.
1667 * Pointless because if there are any spares to skip,
1668 * then a recovery will happen and soon that array won't
1669 * be degraded any more and the spare can go back to sleep then.
1670 */
850b2b42 1671 nospares = 0;
84692195 1672
06d91a5f 1673 sync_req = mddev->in_sync;
1da177e4 1674 mddev->utime = get_seconds();
42543769
N		 1675
1676 /* If this is just a dirty<->clean transition, and the array is clean
1677 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 1678 if (nospares
42543769 1679 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1031be7a
N		 1680 && (mddev->events & 1)
1681 && mddev->events != 1)
42543769
N		 1682 mddev->events--;
1683 else {
1684 /* otherwise we have to go forward and ... */
1685 mddev->events ++;
1686 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1687 /* .. if the array isn't clean, insist on an odd 'events' */
1688 if ((mddev->events&1)==0) {
1689 mddev->events++;
1690 nospares = 0;
1691 }
1692 } else {
1693 /* otherwise insist on an even 'events' (for clean states) */
1694 if ((mddev->events&1)) {
1695 mddev->events++;
1696 nospares = 0;
1697 }
1698 }
1699 }
1da177e4
LT		 1700
1701 if (!mddev->events) {
1702 /*
1703 * oops, this 64-bit counter should never wrap.
1704 * Either we are in around ~1 trillion A.C., assuming
1705 * 1 reboot per second, or we have a bug:
1706 */
1707 MD_BUG();
1708 mddev->events --;
1709 }
1da177e4
LT		 1710
1711 /*
1712 * do not write anything to disk if using
1713 * nonpersistent superblocks
1714 */
06d91a5f 1715 if (!mddev->persistent) {
e691063a
N		 1716 if (!mddev->external)
1717 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1718
a9701a30 1719 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1720 wake_up(&mddev->sb_wait);
1da177e4 1721 return;
06d91a5f 1722 }
e691063a 1723 sync_sbs(mddev, nospares);
a9701a30 1724 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1725
1726 dprintk(KERN_INFO
1727 "md: updating %s RAID superblock on device (in sync %d)\n",
1728 mdname(mddev),mddev->in_sync);
1729
4ad13663 1730 bitmap_update_sb(mddev->bitmap);
1da177e4
LT		 1731 ITERATE_RDEV(mddev,rdev,tmp) {
1732 char b[BDEVNAME_SIZE];
1733 dprintk(KERN_INFO "md: ");
42543769
N		 1734 if (rdev->sb_loaded != 1)
1735 continue; /* no noise on spare devices */
b2d444d7 1736 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1737 dprintk("(skipping faulty ");
1738
1739 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1740 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1741 md_super_write(mddev,rdev,
0002b271 1742 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1743 rdev->sb_page);
1744 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1745 bdevname(rdev->bdev,b),
1746 (unsigned long long)rdev->sb_offset);
42543769 1747 rdev->sb_events = mddev->events;
7bfa19f2 1748
1da177e4
LT		 1749 } else
1750 dprintk(")\n");
7bfa19f2 1751 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1752 /* only need to write one superblock... */
1753 break;
1754 }
a9701a30 1755 md_super_wait(mddev);
850b2b42 1756 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 1757
a9701a30 1758 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 1759 if (mddev->in_sync != sync_req ||
1760 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 1761 /* have to write it out again */
a9701a30 1762 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1763 goto repeat;
1764 }
850b2b42 1765 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 1766 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1767 wake_up(&mddev->sb_wait);
06d91a5f 1768
1da177e4
LT		 1769}
1770
bce74dac
N		 1771/* words written to sysfs files may, or my not, be \n terminated.
1772 * We want to accept with case. For this we use cmd_match.
1773 */
1774static int cmd_match(const char *cmd, const char *str)
1775{
1776 /* See if cmd, written into a sysfs file, matches
1777 * str. They must either be the same, or cmd can
1778 * have a trailing newline
1779 */
1780 while (*cmd && *str && *cmd == *str) {
1781 cmd++;
1782 str++;
1783 }
1784 if (*cmd == '\n')
1785 cmd++;
1786 if (*str || *cmd)
1787 return 0;
1788 return 1;
1789}
1790
86e6ffdd
N		 1791struct rdev_sysfs_entry {
1792 struct attribute attr;
1793 ssize_t (*show)(mdk_rdev_t *, char *);
1794 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1795};
1796
1797static ssize_t
96de1e66 1798state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1799{
1800 char *sep = "";
1801 int len=0;
1802
b2d444d7 1803 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1804 len+= sprintf(page+len, "%sfaulty",sep);
1805 sep = ",";
1806 }
b2d444d7 1807 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1808 len += sprintf(page+len, "%sin_sync",sep);
1809 sep = ",";
1810 }
f655675b
N		 1811 if (test_bit(WriteMostly, &rdev->flags)) {
1812 len += sprintf(page+len, "%swrite_mostly",sep);
1813 sep = ",";
1814 }
b2d444d7
N		 1815 if (!test_bit(Faulty, &rdev->flags) &&
1816 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1817 len += sprintf(page+len, "%sspare", sep);
1818 sep = ",";
1819 }
1820 return len+sprintf(page+len, "\n");
1821}
1822
45dc2de1
N		 1823static ssize_t
1824state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1825{
1826 /* can write
1827 * faulty - simulates and error
1828 * remove - disconnects the device
f655675b
N		 1829 * writemostly - sets write_mostly
1830 * -writemostly - clears write_mostly
45dc2de1
N		 1831 */
1832 int err = -EINVAL;
1833 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1834 md_error(rdev->mddev, rdev);
1835 err = 0;
1836 } else if (cmd_match(buf, "remove")) {
1837 if (rdev->raid_disk >= 0)
1838 err = -EBUSY;
1839 else {
1840 mddev_t *mddev = rdev->mddev;
1841 kick_rdev_from_array(rdev);
3f9d7b0d
N		 1842 if (mddev->pers)
1843 md_update_sb(mddev, 1);
45dc2de1
N		 1844 md_new_event(mddev);
1845 err = 0;
1846 }
f655675b
N		 1847 } else if (cmd_match(buf, "writemostly")) {
1848 set_bit(WriteMostly, &rdev->flags);
1849 err = 0;
1850 } else if (cmd_match(buf, "-writemostly")) {
1851 clear_bit(WriteMostly, &rdev->flags);
1852 err = 0;
45dc2de1
N		 1853 }
1854 return err ? err : len;
1855}
80ca3a44
N		 1856static struct rdev_sysfs_entry rdev_state =
1857__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd
N		 1858
1859static ssize_t
96de1e66 1860super_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1861{
1862 if (rdev->sb_loaded && rdev->sb_size) {
1863 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1864 return rdev->sb_size;
1865 } else
1866 return 0;
1867}
96de1e66
N		 1868static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1869
4dbcdc75
N		 1870static ssize_t
1871errors_show(mdk_rdev_t *rdev, char *page)
1872{
1873 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1874}
1875
1876static ssize_t
1877errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1878{
1879 char *e;
1880 unsigned long n = simple_strtoul(buf, &e, 10);
1881 if (*buf && (*e == 0 || *e == '\n')) {
1882 atomic_set(&rdev->corrected_errors, n);
1883 return len;
1884 }
1885 return -EINVAL;
1886}
1887static struct rdev_sysfs_entry rdev_errors =
80ca3a44 1888__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 1889
014236d2
N		 1890static ssize_t
1891slot_show(mdk_rdev_t *rdev, char *page)
1892{
1893 if (rdev->raid_disk < 0)
1894 return sprintf(page, "none\n");
1895 else
1896 return sprintf(page, "%d\n", rdev->raid_disk);
1897}
1898
1899static ssize_t
1900slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1901{
1902 char *e;
c303da6d
N		 1903 int err;
1904 char nm[20];
014236d2
N		 1905 int slot = simple_strtoul(buf, &e, 10);
1906 if (strncmp(buf, "none", 4)==0)
1907 slot = -1;
1908 else if (e==buf || (*e && *e!= '\n'))
1909 return -EINVAL;
c303da6d
N		 1910 if (rdev->mddev->pers) {
1911 /* Setting 'slot' on an active array requires also
1912 * updating the 'rd%d' link, and communicating
1913 * with the personality with ->hot_*_disk.
1914 * For now we only support removing
1915 * failed/spare devices. This normally happens automatically,
1916 * but not when the metadata is externally managed.
1917 */
1918 if (slot != -1)
1919 return -EBUSY;
1920 if (rdev->raid_disk == -1)
1921 return -EEXIST;
1922 /* personality does all needed checks */
1923 if (rdev->mddev->pers->hot_add_disk == NULL)
1924 return -EINVAL;
1925 err = rdev->mddev->pers->
1926 hot_remove_disk(rdev->mddev, rdev->raid_disk);
1927 if (err)
1928 return err;
1929 sprintf(nm, "rd%d", rdev->raid_disk);
1930 sysfs_remove_link(&rdev->mddev->kobj, nm);
1931 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1932 md_wakeup_thread(rdev->mddev->thread);
1933 } else {
1934 if (slot >= rdev->mddev->raid_disks)
1935 return -ENOSPC;
1936 rdev->raid_disk = slot;
1937 /* assume it is working */
c5d79adb
N		 1938 clear_bit(Faulty, &rdev->flags);
1939 clear_bit(WriteMostly, &rdev->flags);
c303da6d
N		 1940 set_bit(In_sync, &rdev->flags);
1941 }
014236d2
N		 1942 return len;
1943}
1944
1945
1946static struct rdev_sysfs_entry rdev_slot =
80ca3a44 1947__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 1948
93c8cad0
N		 1949static ssize_t
1950offset_show(mdk_rdev_t *rdev, char *page)
1951{
6961ece4 1952 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 1953}
1954
1955static ssize_t
1956offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1957{
1958 char *e;
1959 unsigned long long offset = simple_strtoull(buf, &e, 10);
1960 if (e==buf || (*e && *e != '\n'))
1961 return -EINVAL;
1962 if (rdev->mddev->pers)
1963 return -EBUSY;
c5d79adb
N		 1964 if (rdev->size && rdev->mddev->external)
1965 /* Must set offset before size, so overlap checks
1966 * can be sane */
1967 return -EBUSY;
93c8cad0
N		 1968 rdev->data_offset = offset;
1969 return len;
1970}
1971
1972static struct rdev_sysfs_entry rdev_offset =
80ca3a44 1973__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 1974
83303b61
N		 1975static ssize_t
1976rdev_size_show(mdk_rdev_t *rdev, char *page)
1977{
1978 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1979}
1980
c5d79adb
N		 1981static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
1982{
1983 /* check if two start/length pairs overlap */
1984 if (s1+l1 <= s2)
1985 return 0;
1986 if (s2+l2 <= s1)
1987 return 0;
1988 return 1;
1989}
1990
83303b61
N		 1991static ssize_t
1992rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1993{
1994 char *e;
1995 unsigned long long size = simple_strtoull(buf, &e, 10);
c5d79adb 1996 unsigned long long oldsize = rdev->size;
83303b61
N		 1997 if (e==buf || (*e && *e != '\n'))
1998 return -EINVAL;
1999 if (rdev->mddev->pers)
2000 return -EBUSY;
2001 rdev->size = size;
c5d79adb
N		 2002 if (size > oldsize && rdev->mddev->external) {
2003 /* need to check that all other rdevs with the same ->bdev
2004 * do not overlap. We need to unlock the mddev to avoid
2005 * a deadlock. We have already changed rdev->size, and if
2006 * we have to change it back, we will have the lock again.
2007 */
2008 mddev_t *mddev;
2009 int overlap = 0;
2010 struct list_head *tmp, *tmp2;
2011
2012 mddev_unlock(rdev->mddev);
2013 ITERATE_MDDEV(mddev, tmp) {
2014 mdk_rdev_t *rdev2;
2015
2016 mddev_lock(mddev);
2017 ITERATE_RDEV(mddev, rdev2, tmp2)
2018 if (test_bit(AllReserved, &rdev2->flags) ||
2019 (rdev->bdev == rdev2->bdev &&
2020 rdev != rdev2 &&
2021 overlaps(rdev->data_offset, rdev->size,
2022 rdev2->data_offset, rdev2->size))) {
2023 overlap = 1;
2024 break;
2025 }
2026 mddev_unlock(mddev);
2027 if (overlap) {
2028 mddev_put(mddev);
2029 break;
2030 }
2031 }
2032 mddev_lock(rdev->mddev);
2033 if (overlap) {
2034 /* Someone else could have slipped in a size
2035 * change here, but doing so is just silly.
2036 * We put oldsize back because we *know* it is
2037 * safe, and trust userspace not to race with
2038 * itself
2039 */
2040 rdev->size = oldsize;
2041 return -EBUSY;
2042 }
2043 }
83303b61
N		 2044 if (size < rdev->mddev->size || rdev->mddev->size == 0)
2045 rdev->mddev->size = size;
2046 return len;
2047}
2048
2049static struct rdev_sysfs_entry rdev_size =
80ca3a44 2050__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2051
86e6ffdd
N		 2052static struct attribute *rdev_default_attrs[] = {
2053 &rdev_state.attr,
2054 &rdev_super.attr,
4dbcdc75 2055 &rdev_errors.attr,
014236d2 2056 &rdev_slot.attr,
93c8cad0 2057 &rdev_offset.attr,
83303b61 2058 &rdev_size.attr,
86e6ffdd
N		 2059 NULL,
2060};
2061static ssize_t
2062rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2063{
2064 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2065 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2066
2067 if (!entry->show)
2068 return -EIO;
2069 return entry->show(rdev, page);
2070}
2071
2072static ssize_t
2073rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2074 const char *page, size_t length)
2075{
2076 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2077 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
ca388059 2078 int rv;
86e6ffdd
N		 2079
2080 if (!entry->store)
2081 return -EIO;
67463acb
N		 2082 if (!capable(CAP_SYS_ADMIN))
2083 return -EACCES;
ca388059
N		 2084 rv = mddev_lock(rdev->mddev);
2085 if (!rv) {
2086 rv = entry->store(rdev, page, length);
2087 mddev_unlock(rdev->mddev);
2088 }
2089 return rv;
86e6ffdd
N		 2090}
2091
2092static void rdev_free(struct kobject *ko)
2093{
2094 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2095 kfree(rdev);
2096}
2097static struct sysfs_ops rdev_sysfs_ops = {
2098 .show = rdev_attr_show,
2099 .store = rdev_attr_store,
2100};
2101static struct kobj_type rdev_ktype = {
2102 .release = rdev_free,
2103 .sysfs_ops = &rdev_sysfs_ops,
2104 .default_attrs = rdev_default_attrs,
2105};
2106
1da177e4
LT		 2107/*
2108 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2109 *
2110 * mark the device faulty if:
2111 *
2112 * - the device is nonexistent (zero size)
2113 * - the device has no valid superblock
2114 *
2115 * a faulty rdev _never_ has rdev->sb set.
2116 */
2117static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2118{
2119 char b[BDEVNAME_SIZE];
2120 int err;
2121 mdk_rdev_t *rdev;
2122 sector_t size;
2123
9ffae0cf 2124 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 2125 if (!rdev) {
2126 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2127 return ERR_PTR(-ENOMEM);
2128 }
1da177e4
LT		 2129
2130 if ((err = alloc_disk_sb(rdev)))
2131 goto abort_free;
2132
c5d79adb 2133 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 2134 if (err)
2135 goto abort_free;
2136
f9cb074b 2137 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 2138
1da177e4 2139 rdev->desc_nr = -1;
2b6e8459 2140 rdev->saved_raid_disk = -1;
3f9d7b0d 2141 rdev->raid_disk = -1;
b2d444d7 2142 rdev->flags = 0;
1da177e4 2143 rdev->data_offset = 0;
42543769 2144 rdev->sb_events = 0;
1da177e4 2145 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 2146 atomic_set(&rdev->read_errors, 0);
4dbcdc75 2147 atomic_set(&rdev->corrected_errors, 0);
1da177e4
LT		 2148
2149 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2150 if (!size) {
2151 printk(KERN_WARNING
2152 "md: %s has zero or unknown size, marking faulty!\n",
2153 bdevname(rdev->bdev,b));
2154 err = -EINVAL;
2155 goto abort_free;
2156 }
2157
2158 if (super_format >= 0) {
2159 err = super_types[super_format].
2160 load_super(rdev, NULL, super_minor);
2161 if (err == -EINVAL) {
df968c4e
N		 2162 printk(KERN_WARNING
2163 "md: %s does not have a valid v%d.%d "
2164 "superblock, not importing!\n",
2165 bdevname(rdev->bdev,b),
2166 super_format, super_minor);
1da177e4
LT		 2167 goto abort_free;
2168 }
2169 if (err < 0) {
2170 printk(KERN_WARNING
2171 "md: could not read %s's sb, not importing!\n",
2172 bdevname(rdev->bdev,b));
2173 goto abort_free;
2174 }
2175 }
2176 INIT_LIST_HEAD(&rdev->same_set);
2177
2178 return rdev;
2179
2180abort_free:
2181 if (rdev->sb_page) {
2182 if (rdev->bdev)
2183 unlock_rdev(rdev);
2184 free_disk_sb(rdev);
2185 }
2186 kfree(rdev);
2187 return ERR_PTR(err);
2188}
2189
2190/*
2191 * Check a full RAID array for plausibility
2192 */
2193
2194
a757e64c 2195static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2196{
2197 int i;
2198 struct list_head *tmp;
2199 mdk_rdev_t *rdev, *freshest;
2200 char b[BDEVNAME_SIZE];
2201
2202 freshest = NULL;
2203 ITERATE_RDEV(mddev,rdev,tmp)
2204 switch (super_types[mddev->major_version].
2205 load_super(rdev, freshest, mddev->minor_version)) {
2206 case 1:
2207 freshest = rdev;
2208 break;
2209 case 0:
2210 break;
2211 default:
2212 printk( KERN_ERR \
2213 "md: fatal superblock inconsistency in %s"
2214 " -- removing from array\n",
2215 bdevname(rdev->bdev,b));
2216 kick_rdev_from_array(rdev);
2217 }
2218
2219
2220 super_types[mddev->major_version].
2221 validate_super(mddev, freshest);
2222
2223 i = 0;
2224 ITERATE_RDEV(mddev,rdev,tmp) {
2225 if (rdev != freshest)
2226 if (super_types[mddev->major_version].
2227 validate_super(mddev, rdev)) {
2228 printk(KERN_WARNING "md: kicking non-fresh %s"
2229 " from array!\n",
2230 bdevname(rdev->bdev,b));
2231 kick_rdev_from_array(rdev);
2232 continue;
2233 }
2234 if (mddev->level == LEVEL_MULTIPATH) {
2235 rdev->desc_nr = i++;
2236 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2237 set_bit(In_sync, &rdev->flags);
a778b73f
N		 2238 } else if (rdev->raid_disk >= mddev->raid_disks) {
2239 rdev->raid_disk = -1;
2240 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 2241 }
2242 }
2243
2244
2245
2246 if (mddev->recovery_cp != MaxSector &&
2247 mddev->level >= 1)
2248 printk(KERN_ERR "md: %s: raid array is not clean"
2249 " -- starting background reconstruction\n",
2250 mdname(mddev));
2251
1da177e4
LT		 2252}
2253
16f17b39
N		 2254static ssize_t
2255safe_delay_show(mddev_t *mddev, char *page)
2256{
2257 int msec = (mddev->safemode_delay*1000)/HZ;
2258 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2259}
2260static ssize_t
2261safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2262{
2263 int scale=1;
2264 int dot=0;
2265 int i;
2266 unsigned long msec;
2267 char buf[30];
2268 char *e;
2269 /* remove a period, and count digits after it */
2270 if (len >= sizeof(buf))
2271 return -EINVAL;
2272 strlcpy(buf, cbuf, len);
2273 buf[len] = 0;
2274 for (i=0; i<len; i++) {
2275 if (dot) {
2276 if (isdigit(buf[i])) {
2277 buf[i-1] = buf[i];
2278 scale *= 10;
2279 }
2280 buf[i] = 0;
2281 } else if (buf[i] == '.') {
2282 dot=1;
2283 buf[i] = 0;
2284 }
2285 }
2286 msec = simple_strtoul(buf, &e, 10);
2287 if (e == buf || (*e && *e != '\n'))
2288 return -EINVAL;
2289 msec = (msec * 1000) / scale;
2290 if (msec == 0)
2291 mddev->safemode_delay = 0;
2292 else {
2293 mddev->safemode_delay = (msec*HZ)/1000;
2294 if (mddev->safemode_delay == 0)
2295 mddev->safemode_delay = 1;
2296 }
2297 return len;
2298}
2299static struct md_sysfs_entry md_safe_delay =
80ca3a44 2300__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2301
eae1701f 2302static ssize_t
96de1e66 2303level_show(mddev_t *mddev, char *page)
eae1701f 2304{
2604b703 2305 struct mdk_personality *p = mddev->pers;
d9d166c2 2306 if (p)
eae1701f 2307 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2308 else if (mddev->clevel[0])
2309 return sprintf(page, "%s\n", mddev->clevel);
2310 else if (mddev->level != LEVEL_NONE)
2311 return sprintf(page, "%d\n", mddev->level);
2312 else
2313 return 0;
eae1701f
N		 2314}
2315
d9d166c2
N		 2316static ssize_t
2317level_store(mddev_t *mddev, const char *buf, size_t len)
2318{
2319 int rv = len;
2320 if (mddev->pers)
2321 return -EBUSY;
2322 if (len == 0)
2323 return 0;
2324 if (len >= sizeof(mddev->clevel))
2325 return -ENOSPC;
2326 strncpy(mddev->clevel, buf, len);
2327 if (mddev->clevel[len-1] == '\n')
2328 len--;
2329 mddev->clevel[len] = 0;
2330 mddev->level = LEVEL_NONE;
2331 return rv;
2332}
2333
2334static struct md_sysfs_entry md_level =
80ca3a44 2335__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 2336
d4dbd025
N		 2337
2338static ssize_t
2339layout_show(mddev_t *mddev, char *page)
2340{
2341 /* just a number, not meaningful for all levels */
08a02ecd
N		 2342 if (mddev->reshape_position != MaxSector &&
2343 mddev->layout != mddev->new_layout)
2344 return sprintf(page, "%d (%d)\n",
2345 mddev->new_layout, mddev->layout);
d4dbd025
N		 2346 return sprintf(page, "%d\n", mddev->layout);
2347}
2348
2349static ssize_t
2350layout_store(mddev_t *mddev, const char *buf, size_t len)
2351{
2352 char *e;
2353 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 2354
2355 if (!*buf || (*e && *e != '\n'))
2356 return -EINVAL;
2357
08a02ecd
N		 2358 if (mddev->pers)
2359 return -EBUSY;
2360 if (mddev->reshape_position != MaxSector)
2361 mddev->new_layout = n;
2362 else
2363 mddev->layout = n;
d4dbd025
N		 2364 return len;
2365}
2366static struct md_sysfs_entry md_layout =
80ca3a44 2367__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 2368
2369
eae1701f 2370static ssize_t
96de1e66 2371raid_disks_show(mddev_t *mddev, char *page)
eae1701f 2372{
bb636547
N		 2373 if (mddev->raid_disks == 0)
2374 return 0;
08a02ecd
N		 2375 if (mddev->reshape_position != MaxSector &&
2376 mddev->delta_disks != 0)
2377 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2378 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 2379 return sprintf(page, "%d\n", mddev->raid_disks);
2380}
2381
da943b99
N		 2382static int update_raid_disks(mddev_t *mddev, int raid_disks);
2383
2384static ssize_t
2385raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2386{
da943b99
N		 2387 char *e;
2388 int rv = 0;
2389 unsigned long n = simple_strtoul(buf, &e, 10);
2390
2391 if (!*buf || (*e && *e != '\n'))
2392 return -EINVAL;
2393
2394 if (mddev->pers)
2395 rv = update_raid_disks(mddev, n);
08a02ecd
N		 2396 else if (mddev->reshape_position != MaxSector) {
2397 int olddisks = mddev->raid_disks - mddev->delta_disks;
2398 mddev->delta_disks = n - olddisks;
2399 mddev->raid_disks = n;
2400 } else
da943b99
N		 2401 mddev->raid_disks = n;
2402 return rv ? rv : len;
2403}
2404static struct md_sysfs_entry md_raid_disks =
80ca3a44 2405__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 2406
3b34380a
N		 2407static ssize_t
2408chunk_size_show(mddev_t *mddev, char *page)
2409{
08a02ecd
N		 2410 if (mddev->reshape_position != MaxSector &&
2411 mddev->chunk_size != mddev->new_chunk)
2412 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2413 mddev->chunk_size);
3b34380a
N		 2414 return sprintf(page, "%d\n", mddev->chunk_size);
2415}
2416
2417static ssize_t
2418chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2419{
2420 /* can only set chunk_size if array is not yet active */
2421 char *e;
2422 unsigned long n = simple_strtoul(buf, &e, 10);
2423
3b34380a
N		 2424 if (!*buf || (*e && *e != '\n'))
2425 return -EINVAL;
2426
08a02ecd
N		 2427 if (mddev->pers)
2428 return -EBUSY;
2429 else if (mddev->reshape_position != MaxSector)
2430 mddev->new_chunk = n;
2431 else
2432 mddev->chunk_size = n;
3b34380a
N		 2433 return len;
2434}
2435static struct md_sysfs_entry md_chunk_size =
80ca3a44 2436__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 2437
a94213b1
N		 2438static ssize_t
2439resync_start_show(mddev_t *mddev, char *page)
2440{
2441 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2442}
2443
2444static ssize_t
2445resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2446{
2447 /* can only set chunk_size if array is not yet active */
2448 char *e;
2449 unsigned long long n = simple_strtoull(buf, &e, 10);
2450
2451 if (mddev->pers)
2452 return -EBUSY;
2453 if (!*buf || (*e && *e != '\n'))
2454 return -EINVAL;
2455
2456 mddev->recovery_cp = n;
2457 return len;
2458}
2459static struct md_sysfs_entry md_resync_start =
80ca3a44 2460__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 2461
9e653b63
N		 2462/*
2463 * The array state can be:
2464 *
2465 * clear
2466 * No devices, no size, no level
2467 * Equivalent to STOP_ARRAY ioctl
2468 * inactive
2469 * May have some settings, but array is not active
2470 * all IO results in error
2471 * When written, doesn't tear down array, but just stops it
2472 * suspended (not supported yet)
2473 * All IO requests will block. The array can be reconfigured.
2474 * Writing this, if accepted, will block until array is quiessent
2475 * readonly
2476 * no resync can happen. no superblocks get written.
2477 * write requests fail
2478 * read-auto
2479 * like readonly, but behaves like 'clean' on a write request.
2480 *
2481 * clean - no pending writes, but otherwise active.
2482 * When written to inactive array, starts without resync
2483 * If a write request arrives then
2484 * if metadata is known, mark 'dirty' and switch to 'active'.
2485 * if not known, block and switch to write-pending
2486 * If written to an active array that has pending writes, then fails.
2487 * active
2488 * fully active: IO and resync can be happening.
2489 * When written to inactive array, starts with resync
2490 *
2491 * write-pending
2492 * clean, but writes are blocked waiting for 'active' to be written.
2493 *
2494 * active-idle
2495 * like active, but no writes have been seen for a while (100msec).
2496 *
2497 */
2498enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2499 write_pending, active_idle, bad_word};
05381954 2500static char *array_states[] = {
9e653b63
N		 2501 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2502 "write-pending", "active-idle", NULL };
2503
2504static int match_word(const char *word, char **list)
2505{
2506 int n;
2507 for (n=0; list[n]; n++)
2508 if (cmd_match(word, list[n]))
2509 break;
2510 return n;
2511}
2512
2513static ssize_t
2514array_state_show(mddev_t *mddev, char *page)
2515{
2516 enum array_state st = inactive;
2517
2518 if (mddev->pers)
2519 switch(mddev->ro) {
2520 case 1:
2521 st = readonly;
2522 break;
2523 case 2:
2524 st = read_auto;
2525 break;
2526 case 0:
2527 if (mddev->in_sync)
2528 st = clean;
e691063a
N		 2529 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2530 st = write_pending;
9e653b63
N		 2531 else if (mddev->safemode)
2532 st = active_idle;
2533 else
2534 st = active;
2535 }
2536 else {
2537 if (list_empty(&mddev->disks) &&
2538 mddev->raid_disks == 0 &&
2539 mddev->size == 0)
2540 st = clear;
2541 else
2542 st = inactive;
2543 }
2544 return sprintf(page, "%s\n", array_states[st]);
2545}
2546
2547static int do_md_stop(mddev_t * mddev, int ro);
2548static int do_md_run(mddev_t * mddev);
2549static int restart_array(mddev_t *mddev);
2550
2551static ssize_t
2552array_state_store(mddev_t *mddev, const char *buf, size_t len)
2553{
2554 int err = -EINVAL;
2555 enum array_state st = match_word(buf, array_states);
2556 switch(st) {
2557 case bad_word:
2558 break;
2559 case clear:
2560 /* stopping an active array */
e691063a
N		 2561 if (atomic_read(&mddev->active) > 1)
2562 return -EBUSY;
2563 err = do_md_stop(mddev, 0);
9e653b63
N		 2564 break;
2565 case inactive:
2566 /* stopping an active array */
2567 if (mddev->pers) {
2568 if (atomic_read(&mddev->active) > 1)
2569 return -EBUSY;
2570 err = do_md_stop(mddev, 2);
e691063a
N		 2571 } else
2572 err = 0; /* already inactive */
9e653b63
N		 2573 break;
2574 case suspended:
2575 break; /* not supported yet */
2576 case readonly:
2577 if (mddev->pers)
2578 err = do_md_stop(mddev, 1);
2579 else {
2580 mddev->ro = 1;
2581 err = do_md_run(mddev);
2582 }
2583 break;
2584 case read_auto:
2585 /* stopping an active array */
2586 if (mddev->pers) {
2587 err = do_md_stop(mddev, 1);
2588 if (err == 0)
2589 mddev->ro = 2; /* FIXME mark devices writable */
2590 } else {
2591 mddev->ro = 2;
2592 err = do_md_run(mddev);
2593 }
2594 break;
2595 case clean:
2596 if (mddev->pers) {
2597 restart_array(mddev);
2598 spin_lock_irq(&mddev->write_lock);
2599 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 2600 if (mddev->in_sync == 0) {
2601 mddev->in_sync = 1;
2602 if (mddev->persistent)
2603 set_bit(MD_CHANGE_CLEAN,
2604 &mddev->flags);
2605 }
2606 err = 0;
2607 } else
2608 err = -EBUSY;
9e653b63
N		 2609 spin_unlock_irq(&mddev->write_lock);
2610 } else {
2611 mddev->ro = 0;
2612 mddev->recovery_cp = MaxSector;
2613 err = do_md_run(mddev);
2614 }
2615 break;
2616 case active:
2617 if (mddev->pers) {
2618 restart_array(mddev);
e691063a
N		 2619 if (mddev->external)
2620 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
9e653b63
N		 2621 wake_up(&mddev->sb_wait);
2622 err = 0;
2623 } else {
2624 mddev->ro = 0;
2625 err = do_md_run(mddev);
2626 }
2627 break;
2628 case write_pending:
2629 case active_idle:
2630 /* these cannot be set */
2631 break;
2632 }
2633 if (err)
2634 return err;
2635 else
2636 return len;
2637}
80ca3a44
N		 2638static struct md_sysfs_entry md_array_state =
2639__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 2640
6d7ff738
N		 2641static ssize_t
2642null_show(mddev_t *mddev, char *page)
2643{
2644 return -EINVAL;
2645}
2646
2647static ssize_t
2648new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2649{
2650 /* buf must be %d:%d\n? giving major and minor numbers */
2651 /* The new device is added to the array.
2652 * If the array has a persistent superblock, we read the
2653 * superblock to initialise info and check validity.
2654 * Otherwise, only checking done is that in bind_rdev_to_array,
2655 * which mainly checks size.
2656 */
2657 char *e;
2658 int major = simple_strtoul(buf, &e, 10);
2659 int minor;
2660 dev_t dev;
2661 mdk_rdev_t *rdev;
2662 int err;
2663
2664 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2665 return -EINVAL;
2666 minor = simple_strtoul(e+1, &e, 10);
2667 if (*e && *e != '\n')
2668 return -EINVAL;
2669 dev = MKDEV(major, minor);
2670 if (major != MAJOR(dev) ||
2671 minor != MINOR(dev))
2672 return -EOVERFLOW;
2673
2674
2675 if (mddev->persistent) {
2676 rdev = md_import_device(dev, mddev->major_version,
2677 mddev->minor_version);
2678 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2679 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2680 mdk_rdev_t, same_set);
2681 err = super_types[mddev->major_version]
2682 .load_super(rdev, rdev0, mddev->minor_version);
2683 if (err < 0)
2684 goto out;
2685 }
c5d79adb
N		 2686 } else if (mddev->external)
2687 rdev = md_import_device(dev, -2, -1);
2688 else
6d7ff738
N		 2689 rdev = md_import_device(dev, -1, -1);
2690
2691 if (IS_ERR(rdev))
2692 return PTR_ERR(rdev);
2693 err = bind_rdev_to_array(rdev, mddev);
2694 out:
2695 if (err)
2696 export_rdev(rdev);
2697 return err ? err : len;
2698}
2699
2700static struct md_sysfs_entry md_new_device =
80ca3a44 2701__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 2702
9b1d1dac
PC		 2703static ssize_t
2704bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2705{
2706 char *end;
2707 unsigned long chunk, end_chunk;
2708
2709 if (!mddev->bitmap)
2710 goto out;
2711 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2712 while (*buf) {
2713 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2714 if (buf == end) break;
2715 if (*end == '-') { /* range */
2716 buf = end + 1;
2717 end_chunk = simple_strtoul(buf, &end, 0);
2718 if (buf == end) break;
2719 }
2720 if (*end && !isspace(*end)) break;
2721 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2722 buf = end;
2723 while (isspace(*buf)) buf++;
2724 }
2725 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2726out:
2727 return len;
2728}
2729
2730static struct md_sysfs_entry md_bitmap =
2731__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2732
a35b0d69
N		 2733static ssize_t
2734size_show(mddev_t *mddev, char *page)
2735{
2736 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2737}
2738
2739static int update_size(mddev_t *mddev, unsigned long size);
2740
2741static ssize_t
2742size_store(mddev_t *mddev, const char *buf, size_t len)
2743{
2744 /* If array is inactive, we can reduce the component size, but
2745 * not increase it (except from 0).
2746 * If array is active, we can try an on-line resize
2747 */
2748 char *e;
2749 int err = 0;
2750 unsigned long long size = simple_strtoull(buf, &e, 10);
2751 if (!*buf || *buf == '\n' ||
2752 (*e && *e != '\n'))
2753 return -EINVAL;
2754
2755 if (mddev->pers) {
2756 err = update_size(mddev, size);
850b2b42 2757 md_update_sb(mddev, 1);
a35b0d69
N		 2758 } else {
2759 if (mddev->size == 0 ||
2760 mddev->size > size)
2761 mddev->size = size;
2762 else
2763 err = -ENOSPC;
2764 }
2765 return err ? err : len;
2766}
2767
2768static struct md_sysfs_entry md_size =
80ca3a44 2769__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 2770
8bb93aac
N		 2771
2772/* Metdata version.
e691063a
N		 2773 * This is one of
2774 * 'none' for arrays with no metadata (good luck...)
2775 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 2776 * or N.M for internally known formats
2777 */
2778static ssize_t
2779metadata_show(mddev_t *mddev, char *page)
2780{
2781 if (mddev->persistent)
2782 return sprintf(page, "%d.%d\n",
2783 mddev->major_version, mddev->minor_version);
e691063a
N		 2784 else if (mddev->external)
2785 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 2786 else
2787 return sprintf(page, "none\n");
2788}
2789
2790static ssize_t
2791metadata_store(mddev_t *mddev, const char *buf, size_t len)
2792{
2793 int major, minor;
2794 char *e;
2795 if (!list_empty(&mddev->disks))
2796 return -EBUSY;
2797
2798 if (cmd_match(buf, "none")) {
2799 mddev->persistent = 0;
e691063a
N		 2800 mddev->external = 0;
2801 mddev->major_version = 0;
2802 mddev->minor_version = 90;
2803 return len;
2804 }
2805 if (strncmp(buf, "external:", 9) == 0) {
2806 int namelen = len-9;
2807 if (namelen >= sizeof(mddev->metadata_type))
2808 namelen = sizeof(mddev->metadata_type)-1;
2809 strncpy(mddev->metadata_type, buf+9, namelen);
2810 mddev->metadata_type[namelen] = 0;
2811 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2812 mddev->metadata_type[--namelen] = 0;
2813 mddev->persistent = 0;
2814 mddev->external = 1;
8bb93aac
N		 2815 mddev->major_version = 0;
2816 mddev->minor_version = 90;
2817 return len;
2818 }
2819 major = simple_strtoul(buf, &e, 10);
2820 if (e==buf || *e != '.')
2821 return -EINVAL;
2822 buf = e+1;
2823 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 2824 if (e==buf || (*e && *e != '\n') )
8bb93aac 2825 return -EINVAL;
50511da3 2826 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 2827 return -ENOENT;
2828 mddev->major_version = major;
2829 mddev->minor_version = minor;
2830 mddev->persistent = 1;
e691063a 2831 mddev->external = 0;
8bb93aac
N		 2832 return len;
2833}
2834
2835static struct md_sysfs_entry md_metadata =
80ca3a44 2836__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 2837
24dd469d 2838static ssize_t
7eec314d 2839action_show(mddev_t *mddev, char *page)
24dd469d 2840{
7eec314d 2841 char *type = "idle";
31399d9e 2842 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 2843 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 2844 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2845 type = "reshape";
2846 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 2847 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2848 type = "resync";
2849 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2850 type = "check";
2851 else
2852 type = "repair";
2853 } else
2854 type = "recover";
2855 }
2856 return sprintf(page, "%s\n", type);
2857}
2858
2859static ssize_t
7eec314d 2860action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 2861{
7eec314d
N		 2862 if (!mddev->pers || !mddev->pers->sync_request)
2863 return -EINVAL;
2864
bce74dac 2865 if (cmd_match(page, "idle")) {
7eec314d
N		 2866 if (mddev->sync_thread) {
2867 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2868 md_unregister_thread(mddev->sync_thread);
2869 mddev->sync_thread = NULL;
2870 mddev->recovery = 0;
2871 }
03c902e1
N		 2872 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2873 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 2874 return -EBUSY;
03c902e1 2875 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
7eec314d 2876 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
16484bf5
N		 2877 else if (cmd_match(page, "reshape")) {
2878 int err;
2879 if (mddev->pers->start_reshape == NULL)
2880 return -EINVAL;
2881 err = mddev->pers->start_reshape(mddev);
2882 if (err)
2883 return err;
2884 } else {
bce74dac 2885 if (cmd_match(page, "check"))
7eec314d 2886 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 2887 else if (!cmd_match(page, "repair"))
7eec314d
N		 2888 return -EINVAL;
2889 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2890 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 2891 }
03c902e1 2892 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 2893 md_wakeup_thread(mddev->thread);
2894 return len;
2895}
2896
9d88883e 2897static ssize_t
96de1e66 2898mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 2899{
2900 return sprintf(page, "%llu\n",
2901 (unsigned long long) mddev->resync_mismatches);
2902}
2903
80ca3a44
N		 2904static struct md_sysfs_entry md_scan_mode =
2905__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 2906
96de1e66 2907
80ca3a44 2908static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 2909
88202a0c
N		 2910static ssize_t
2911sync_min_show(mddev_t *mddev, char *page)
2912{
2913 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2914 mddev->sync_speed_min ? "local": "system");
2915}
2916
2917static ssize_t
2918sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2919{
2920 int min;
2921 char *e;
2922 if (strncmp(buf, "system", 6)==0) {
2923 mddev->sync_speed_min = 0;
2924 return len;
2925 }
2926 min = simple_strtoul(buf, &e, 10);
2927 if (buf == e || (*e && *e != '\n') || min <= 0)
2928 return -EINVAL;
2929 mddev->sync_speed_min = min;
2930 return len;
2931}
2932
2933static struct md_sysfs_entry md_sync_min =
2934__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2935
2936static ssize_t
2937sync_max_show(mddev_t *mddev, char *page)
2938{
2939 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2940 mddev->sync_speed_max ? "local": "system");
2941}
2942
2943static ssize_t
2944sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2945{
2946 int max;
2947 char *e;
2948 if (strncmp(buf, "system", 6)==0) {
2949 mddev->sync_speed_max = 0;
2950 return len;
2951 }
2952 max = simple_strtoul(buf, &e, 10);
2953 if (buf == e || (*e && *e != '\n') || max <= 0)
2954 return -EINVAL;
2955 mddev->sync_speed_max = max;
2956 return len;
2957}
2958
2959static struct md_sysfs_entry md_sync_max =
2960__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2961
d7f3d291
IP		 2962static ssize_t
2963degraded_show(mddev_t *mddev, char *page)
2964{
2965 return sprintf(page, "%d\n", mddev->degraded);
2966}
2967static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c
N		 2968
2969static ssize_t
2970sync_speed_show(mddev_t *mddev, char *page)
2971{
2972 unsigned long resync, dt, db;
ff4e8d9a 2973 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
88202a0c
N		 2974 dt = ((jiffies - mddev->resync_mark) / HZ);
2975 if (!dt) dt++;
2976 db = resync - (mddev->resync_mark_cnt);
2977 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2978}
2979
80ca3a44 2980static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 2981
2982static ssize_t
2983sync_completed_show(mddev_t *mddev, char *page)
2984{
2985 unsigned long max_blocks, resync;
2986
2987 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2988 max_blocks = mddev->resync_max_sectors;
2989 else
2990 max_blocks = mddev->size << 1;
2991
2992 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2993 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2994}
2995
80ca3a44 2996static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 2997
c6207277
N		 2998static ssize_t
2999max_sync_show(mddev_t *mddev, char *page)
3000{
3001 if (mddev->resync_max == MaxSector)
3002 return sprintf(page, "max\n");
3003 else
3004 return sprintf(page, "%llu\n",
3005 (unsigned long long)mddev->resync_max);
3006}
3007static ssize_t
3008max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3009{
3010 if (strncmp(buf, "max", 3) == 0)
3011 mddev->resync_max = MaxSector;
3012 else {
3013 char *ep;
3014 unsigned long long max = simple_strtoull(buf, &ep, 10);
3015 if (ep == buf || (*ep != 0 && *ep != '\n'))
3016 return -EINVAL;
3017 if (max < mddev->resync_max &&
3018 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3019 return -EBUSY;
3020
3021 /* Must be a multiple of chunk_size */
3022 if (mddev->chunk_size) {
3023 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3024 return -EINVAL;
3025 }
3026 mddev->resync_max = max;
3027 }
3028 wake_up(&mddev->recovery_wait);
3029 return len;
3030}
3031
3032static struct md_sysfs_entry md_max_sync =
3033__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3034
e464eafd
N		 3035static ssize_t
3036suspend_lo_show(mddev_t *mddev, char *page)
3037{
3038 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3039}
3040
3041static ssize_t
3042suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3043{
3044 char *e;
3045 unsigned long long new = simple_strtoull(buf, &e, 10);
3046
3047 if (mddev->pers->quiesce == NULL)
3048 return -EINVAL;
3049 if (buf == e || (*e && *e != '\n'))
3050 return -EINVAL;
3051 if (new >= mddev->suspend_hi ||
3052 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3053 mddev->suspend_lo = new;
3054 mddev->pers->quiesce(mddev, 2);
3055 return len;
3056 } else
3057 return -EINVAL;
3058}
3059static struct md_sysfs_entry md_suspend_lo =
3060__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3061
3062
3063static ssize_t
3064suspend_hi_show(mddev_t *mddev, char *page)
3065{
3066 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3067}
3068
3069static ssize_t
3070suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3071{
3072 char *e;
3073 unsigned long long new = simple_strtoull(buf, &e, 10);
3074
3075 if (mddev->pers->quiesce == NULL)
3076 return -EINVAL;
3077 if (buf == e || (*e && *e != '\n'))
3078 return -EINVAL;
3079 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3080 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3081 mddev->suspend_hi = new;
3082 mddev->pers->quiesce(mddev, 1);
3083 mddev->pers->quiesce(mddev, 0);
3084 return len;
3085 } else
3086 return -EINVAL;
3087}
3088static struct md_sysfs_entry md_suspend_hi =
3089__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3090
08a02ecd
N		 3091static ssize_t
3092reshape_position_show(mddev_t *mddev, char *page)
3093{
3094 if (mddev->reshape_position != MaxSector)
3095 return sprintf(page, "%llu\n",
3096 (unsigned long long)mddev->reshape_position);
3097 strcpy(page, "none\n");
3098 return 5;
3099}
3100
3101static ssize_t
3102reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3103{
3104 char *e;
3105 unsigned long long new = simple_strtoull(buf, &e, 10);
3106 if (mddev->pers)
3107 return -EBUSY;
3108 if (buf == e || (*e && *e != '\n'))
3109 return -EINVAL;
3110 mddev->reshape_position = new;
3111 mddev->delta_disks = 0;
3112 mddev->new_level = mddev->level;
3113 mddev->new_layout = mddev->layout;
3114 mddev->new_chunk = mddev->chunk_size;
3115 return len;
3116}
3117
3118static struct md_sysfs_entry md_reshape_position =
3119__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3120 reshape_position_store);
3121
e464eafd 3122
eae1701f
N		 3123static struct attribute *md_default_attrs[] = {
3124 &md_level.attr,
d4dbd025 3125 &md_layout.attr,
eae1701f 3126 &md_raid_disks.attr,
3b34380a 3127 &md_chunk_size.attr,
a35b0d69 3128 &md_size.attr,
a94213b1 3129 &md_resync_start.attr,
8bb93aac 3130 &md_metadata.attr,
6d7ff738 3131 &md_new_device.attr,
16f17b39 3132 &md_safe_delay.attr,
9e653b63 3133 &md_array_state.attr,
08a02ecd 3134 &md_reshape_position.attr,
411036fa
N		 3135 NULL,
3136};
3137
3138static struct attribute *md_redundancy_attrs[] = {
24dd469d 3139 &md_scan_mode.attr,
9d88883e 3140 &md_mismatches.attr,
88202a0c
N		 3141 &md_sync_min.attr,
3142 &md_sync_max.attr,
3143 &md_sync_speed.attr,
3144 &md_sync_completed.attr,
c6207277 3145 &md_max_sync.attr,
e464eafd
N		 3146 &md_suspend_lo.attr,
3147 &md_suspend_hi.attr,
9b1d1dac 3148 &md_bitmap.attr,
d7f3d291 3149 &md_degraded.attr,
eae1701f
N		 3150 NULL,
3151};
411036fa
N		 3152static struct attribute_group md_redundancy_group = {
3153 .name = NULL,
3154 .attrs = md_redundancy_attrs,
3155};
3156
eae1701f
N		 3157
3158static ssize_t
3159md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3160{
3161 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3162 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 3163 ssize_t rv;
eae1701f
N		 3164
3165 if (!entry->show)
3166 return -EIO;
5dc5cf7d
IM		 3167 rv = mddev_lock(mddev);
3168 if (!rv) {
3169 rv = entry->show(mddev, page);
3170 mddev_unlock(mddev);
3171 }
96de1e66 3172 return rv;
eae1701f
N		 3173}
3174
3175static ssize_t
3176md_attr_store(struct kobject *kobj, struct attribute *attr,
3177 const char *page, size_t length)
3178{
3179 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3180 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 3181 ssize_t rv;
eae1701f
N		 3182
3183 if (!entry->store)
3184 return -EIO;
67463acb
N		 3185 if (!capable(CAP_SYS_ADMIN))
3186 return -EACCES;
5dc5cf7d
IM		 3187 rv = mddev_lock(mddev);
3188 if (!rv) {
3189 rv = entry->store(mddev, page, length);
3190 mddev_unlock(mddev);
3191 }
96de1e66 3192 return rv;
eae1701f
N		 3193}
3194
3195static void md_free(struct kobject *ko)
3196{
3197 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3198 kfree(mddev);
3199}
3200
3201static struct sysfs_ops md_sysfs_ops = {
3202 .show = md_attr_show,
3203 .store = md_attr_store,
3204};
3205static struct kobj_type md_ktype = {
3206 .release = md_free,
3207 .sysfs_ops = &md_sysfs_ops,
3208 .default_attrs = md_default_attrs,
3209};
3210
1da177e4
LT		 3211int mdp_major = 0;
3212
3213static struct kobject *md_probe(dev_t dev, int *part, void *data)
3214{
48c9c27b 3215 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 3216 mddev_t *mddev = mddev_find(dev);
3217 struct gendisk *disk;
3218 int partitioned = (MAJOR(dev) != MD_MAJOR);
3219 int shift = partitioned ? MdpMinorShift : 0;
3220 int unit = MINOR(dev) >> shift;
3830c62f 3221 int error;
1da177e4
LT		 3222
3223 if (!mddev)
3224 return NULL;
3225
48c9c27b 3226 mutex_lock(&disks_mutex);
1da177e4 3227 if (mddev->gendisk) {
48c9c27b 3228 mutex_unlock(&disks_mutex);
1da177e4
LT		 3229 mddev_put(mddev);
3230 return NULL;
3231 }
3232 disk = alloc_disk(1 << shift);
3233 if (!disk) {
48c9c27b 3234 mutex_unlock(&disks_mutex);
1da177e4
LT		 3235 mddev_put(mddev);
3236 return NULL;
3237 }
3238 disk->major = MAJOR(dev);
3239 disk->first_minor = unit << shift;
ce7b0f46 3240 if (partitioned)
1da177e4 3241 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 3242 else
1da177e4 3243 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 3244 disk->fops = &md_fops;
3245 disk->private_data = mddev;
3246 disk->queue = mddev->queue;
3247 add_disk(disk);
3248 mddev->gendisk = disk;
48c9c27b 3249 mutex_unlock(&disks_mutex);
edfaa7c3 3250 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3830c62f
GKH		 3251 "%s", "md");
3252 if (error)
5e55e2f5
N		 3253 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3254 disk->disk_name);
3830c62f
GKH		 3255 else
3256 kobject_uevent(&mddev->kobj, KOBJ_ADD);
1da177e4
LT		 3257 return NULL;
3258}
3259
1da177e4
LT		 3260static void md_safemode_timeout(unsigned long data)
3261{
3262 mddev_t *mddev = (mddev_t *) data;
3263
3264 mddev->safemode = 1;
3265 md_wakeup_thread(mddev->thread);
3266}
3267
6ff8d8ec 3268static int start_dirty_degraded;
1da177e4
LT		 3269
3270static int do_md_run(mddev_t * mddev)
3271{
2604b703 3272 int err;
1da177e4
LT		 3273 int chunk_size;
3274 struct list_head *tmp;
3275 mdk_rdev_t *rdev;
3276 struct gendisk *disk;
2604b703 3277 struct mdk_personality *pers;
1da177e4
LT		 3278 char b[BDEVNAME_SIZE];
3279
a757e64c
N		 3280 if (list_empty(&mddev->disks))
3281 /* cannot run an array with no devices.. */
1da177e4 3282 return -EINVAL;
1da177e4
LT		 3283
3284 if (mddev->pers)
3285 return -EBUSY;
3286
3287 /*
3288 * Analyze all RAID superblock(s)
3289 */
1ec4a939
N		 3290 if (!mddev->raid_disks) {
3291 if (!mddev->persistent)
3292 return -EINVAL;
a757e64c 3293 analyze_sbs(mddev);
1ec4a939 3294 }
1da177e4
LT		 3295
3296 chunk_size = mddev->chunk_size;
2604b703
N		 3297
3298 if (chunk_size) {
1da177e4
LT		 3299 if (chunk_size > MAX_CHUNK_SIZE) {
3300 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3301 chunk_size, MAX_CHUNK_SIZE);
3302 return -EINVAL;
3303 }
3304 /*
3305 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3306 */
3307 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 3308 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 3309 return -EINVAL;
3310 }
3311 if (chunk_size < PAGE_SIZE) {
3312 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3313 chunk_size, PAGE_SIZE);
3314 return -EINVAL;
3315 }
3316
3317 /* devices must have minimum size of one chunk */
3318 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 3319 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3320 continue;
3321 if (rdev->size < chunk_size / 1024) {
3322 printk(KERN_WARNING
3323 "md: Dev %s smaller than chunk_size:"
3324 " %lluk < %dk\n",
3325 bdevname(rdev->bdev,b),
3326 (unsigned long long)rdev->size,
3327 chunk_size / 1024);
3328 return -EINVAL;
3329 }
3330 }
3331 }
3332
1da177e4 3333#ifdef CONFIG_KMOD
d9d166c2
N		 3334 if (mddev->level != LEVEL_NONE)
3335 request_module("md-level-%d", mddev->level);
3336 else if (mddev->clevel[0])
3337 request_module("md-%s", mddev->clevel);
1da177e4
LT		 3338#endif
3339
3340 /*
3341 * Drop all container device buffers, from now on
3342 * the only valid external interface is through the md
3343 * device.
1da177e4
LT		 3344 */
3345 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 3346 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3347 continue;
3348 sync_blockdev(rdev->bdev);
f98393a6 3349 invalidate_bdev(rdev->bdev);
f0d76d70
N		 3350
3351 /* perform some consistency tests on the device.
3352 * We don't want the data to overlap the metadata,
3353 * Internal Bitmap issues has handled elsewhere.
3354 */
3355 if (rdev->data_offset < rdev->sb_offset) {
3356 if (mddev->size &&
3357 rdev->data_offset + mddev->size*2
3358 > rdev->sb_offset*2) {
3359 printk("md: %s: data overlaps metadata\n",
3360 mdname(mddev));
3361 return -EINVAL;
3362 }
3363 } else {
3364 if (rdev->sb_offset*2 + rdev->sb_size/512
3365 > rdev->data_offset) {
3366 printk("md: %s: metadata overlaps data\n",
3367 mdname(mddev));
3368 return -EINVAL;
3369 }
3370 }
1da177e4
LT		 3371 }
3372
3373 md_probe(mddev->unit, NULL, NULL);
3374 disk = mddev->gendisk;
3375 if (!disk)
3376 return -ENOMEM;
3377
3378 spin_lock(&pers_lock);
d9d166c2 3379 pers = find_pers(mddev->level, mddev->clevel);
2604b703 3380 if (!pers || !try_module_get(pers->owner)) {
1da177e4 3381 spin_unlock(&pers_lock);
d9d166c2
N		 3382 if (mddev->level != LEVEL_NONE)
3383 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3384 mddev->level);
3385 else
3386 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3387 mddev->clevel);
1da177e4
LT		 3388 return -EINVAL;
3389 }
2604b703 3390 mddev->pers = pers;
1da177e4 3391 spin_unlock(&pers_lock);
d9d166c2
N		 3392 mddev->level = pers->level;
3393 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 3394
f6705578 3395 if (mddev->reshape_position != MaxSector &&
63c70c4f 3396 pers->start_reshape == NULL) {
f6705578
N		 3397 /* This personality cannot handle reshaping... */
3398 mddev->pers = NULL;
3399 module_put(pers->owner);
3400 return -EINVAL;
3401 }
3402
7dd5e7c3
N		 3403 if (pers->sync_request) {
3404 /* Warn if this is a potentially silly
3405 * configuration.
3406 */
3407 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3408 mdk_rdev_t *rdev2;
3409 struct list_head *tmp2;
3410 int warned = 0;
3411 ITERATE_RDEV(mddev, rdev, tmp) {
3412 ITERATE_RDEV(mddev, rdev2, tmp2) {
3413 if (rdev < rdev2 &&
3414 rdev->bdev->bd_contains ==
3415 rdev2->bdev->bd_contains) {
3416 printk(KERN_WARNING
3417 "%s: WARNING: %s appears to be"
3418 " on the same physical disk as"
3419 " %s.\n",
3420 mdname(mddev),
3421 bdevname(rdev->bdev,b),
3422 bdevname(rdev2->bdev,b2));
3423 warned = 1;
3424 }
3425 }
3426 }
3427 if (warned)
3428 printk(KERN_WARNING
3429 "True protection against single-disk"
3430 " failure might be compromised.\n");
3431 }
3432
657390d2 3433 mddev->recovery = 0;
1da177e4 3434 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 3435 mddev->barriers_work = 1;
6ff8d8ec 3436 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 3437
f91de92e
N		 3438 if (start_readonly)
3439 mddev->ro = 2; /* read-only, but switch on first write */
3440
b15c2e57
N		 3441 err = mddev->pers->run(mddev);
3442 if (!err && mddev->pers->sync_request) {
3443 err = bitmap_create(mddev);
3444 if (err) {
3445 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3446 mdname(mddev), err);
3447 mddev->pers->stop(mddev);
3448 }
3449 }
1da177e4
LT		 3450 if (err) {
3451 printk(KERN_ERR "md: pers->run() failed ...\n");
3452 module_put(mddev->pers->owner);
3453 mddev->pers = NULL;
32a7627c
N		 3454 bitmap_destroy(mddev);
3455 return err;
1da177e4 3456 }
5e55e2f5
N		 3457 if (mddev->pers->sync_request) {
3458 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3459 printk(KERN_WARNING
3460 "md: cannot register extra attributes for %s\n",
3461 mdname(mddev));
3462 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 3463 mddev->ro = 0;
3464
1da177e4
LT		 3465 atomic_set(&mddev->writes_pending,0);
3466 mddev->safemode = 0;
3467 mddev->safemode_timer.function = md_safemode_timeout;
3468 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 3469 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 3470 mddev->in_sync = 1;
86e6ffdd
N		 3471
3472 ITERATE_RDEV(mddev,rdev,tmp)
3473 if (rdev->raid_disk >= 0) {
3474 char nm[20];
3475 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 3476 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3477 printk("md: cannot register %s for %s\n",
3478 nm, mdname(mddev));
86e6ffdd 3479 }
1da177e4
LT		 3480
3481 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3482
850b2b42
N		 3483 if (mddev->flags)
3484 md_update_sb(mddev, 0);
1da177e4
LT		 3485
3486 set_capacity(disk, mddev->array_size<<1);
3487
3488 /* If we call blk_queue_make_request here, it will
3489 * re-initialise max_sectors etc which may have been
3490 * refined inside -> run. So just set the bits we need to set.
3491 * Most initialisation happended when we called
3492 * blk_queue_make_request(..., md_fail_request)
3493 * earlier.
3494 */
3495 mddev->queue->queuedata = mddev;
3496 mddev->queue->make_request_fn = mddev->pers->make_request;
3497
5fd6c1dc
N		 3498 /* If there is a partially-recovered drive we need to
3499 * start recovery here. If we leave it to md_check_recovery,
3500 * it will remove the drives and not do the right thing
3501 */
0b8c9de0 3502 if (mddev->degraded && !mddev->sync_thread) {
5fd6c1dc
N		 3503 struct list_head *rtmp;
3504 int spares = 0;
3505 ITERATE_RDEV(mddev,rdev,rtmp)
3506 if (rdev->raid_disk >= 0 &&
3507 !test_bit(In_sync, &rdev->flags) &&
3508 !test_bit(Faulty, &rdev->flags))
3509 /* complete an interrupted recovery */
3510 spares++;
3511 if (spares && mddev->pers->sync_request) {
3512 mddev->recovery = 0;
3513 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3514 mddev->sync_thread = md_register_thread(md_do_sync,
3515 mddev,
3516 "%s_resync");
3517 if (!mddev->sync_thread) {
3518 printk(KERN_ERR "%s: could not start resync"
3519 " thread...\n",
3520 mdname(mddev));
3521 /* leave the spares where they are, it shouldn't hurt */
3522 mddev->recovery = 0;
0b8c9de0 3523 }
5fd6c1dc
N		 3524 }
3525 }
0b8c9de0
N		 3526 md_wakeup_thread(mddev->thread);
3527 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 3528
44ce6294 3529 mddev->changed = 1;
d7603b7e 3530 md_new_event(mddev);
edfaa7c3 3531 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
1da177e4
LT		 3532 return 0;
3533}
3534
3535static int restart_array(mddev_t *mddev)
3536{
3537 struct gendisk *disk = mddev->gendisk;
3538 int err;
3539
3540 /*
3541 * Complain if it has no devices
3542 */
3543 err = -ENXIO;
3544 if (list_empty(&mddev->disks))
3545 goto out;
3546
3547 if (mddev->pers) {
3548 err = -EBUSY;
3549 if (!mddev->ro)
3550 goto out;
3551
3552 mddev->safemode = 0;
3553 mddev->ro = 0;
3554 set_disk_ro(disk, 0);
3555
3556 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3557 mdname(mddev));
3558 /*
3559 * Kick recovery or resync if necessary
3560 */
3561 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3562 md_wakeup_thread(mddev->thread);
5fd6c1dc 3563 md_wakeup_thread(mddev->sync_thread);
1da177e4 3564 err = 0;
9e653b63 3565 } else
1da177e4 3566 err = -EINVAL;
1da177e4
LT		 3567
3568out:
3569 return err;
3570}
3571
acc55e22
N		 3572/* similar to deny_write_access, but accounts for our holding a reference
3573 * to the file ourselves */
3574static int deny_bitmap_write_access(struct file * file)
3575{
3576 struct inode *inode = file->f_mapping->host;
3577
3578 spin_lock(&inode->i_lock);
3579 if (atomic_read(&inode->i_writecount) > 1) {
3580 spin_unlock(&inode->i_lock);
3581 return -ETXTBSY;
3582 }
3583 atomic_set(&inode->i_writecount, -1);
3584 spin_unlock(&inode->i_lock);
3585
3586 return 0;
3587}
3588
3589static void restore_bitmap_write_access(struct file *file)
3590{
3591 struct inode *inode = file->f_mapping->host;
3592
3593 spin_lock(&inode->i_lock);
3594 atomic_set(&inode->i_writecount, 1);
3595 spin_unlock(&inode->i_lock);
3596}
3597
9e653b63
N		 3598/* mode:
3599 * 0 - completely stop and dis-assemble array
3600 * 1 - switch to readonly
3601 * 2 - stop but do not disassemble array
3602 */
3603static int do_md_stop(mddev_t * mddev, int mode)
1da177e4
LT		 3604{
3605 int err = 0;
3606 struct gendisk *disk = mddev->gendisk;
3607
3608 if (mddev->pers) {
3609 if (atomic_read(&mddev->active)>2) {
3610 printk("md: %s still in use.\n",mdname(mddev));
3611 return -EBUSY;
3612 }
3613
3614 if (mddev->sync_thread) {
5fd6c1dc 3615 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4
LT		 3616 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3617 md_unregister_thread(mddev->sync_thread);
3618 mddev->sync_thread = NULL;
3619 }
3620
3621 del_timer_sync(&mddev->safemode_timer);
3622
3623 invalidate_partition(disk, 0);
3624
9e653b63
N		 3625 switch(mode) {
3626 case 1: /* readonly */
1da177e4 3627 err = -ENXIO;
f91de92e 3628 if (mddev->ro==1)
1da177e4
LT		 3629 goto out;
3630 mddev->ro = 1;
9e653b63
N		 3631 break;
3632 case 0: /* disassemble */
3633 case 2: /* stop */
6b8b3e8a 3634 bitmap_flush(mddev);
a9701a30 3635 md_super_wait(mddev);
1da177e4
LT		 3636 if (mddev->ro)
3637 set_disk_ro(disk, 0);
3638 blk_queue_make_request(mddev->queue, md_fail_request);
3639 mddev->pers->stop(mddev);
d1b5380c
N		 3640 mddev->queue->merge_bvec_fn = NULL;
3641 mddev->queue->unplug_fn = NULL;
041ae52e 3642 mddev->queue->backing_dev_info.congested_fn = NULL;
411036fa
N		 3643 if (mddev->pers->sync_request)
3644 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3645
1da177e4
LT		 3646 module_put(mddev->pers->owner);
3647 mddev->pers = NULL;
0d4ca600
N		 3648
3649 set_capacity(disk, 0);
44ce6294 3650 mddev->changed = 1;
0d4ca600 3651
1da177e4
LT		 3652 if (mddev->ro)
3653 mddev->ro = 0;
3654 }
850b2b42 3655 if (!mddev->in_sync || mddev->flags) {
1da177e4
LT		 3656 /* mark array as shutdown cleanly */
3657 mddev->in_sync = 1;
850b2b42 3658 md_update_sb(mddev, 1);
1da177e4 3659 }
9e653b63 3660 if (mode == 1)
1da177e4 3661 set_disk_ro(disk, 1);
5fd6c1dc 3662 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4 3663 }
32a7627c 3664
1da177e4
LT		 3665 /*
3666 * Free resources if final stop
3667 */
9e653b63 3668 if (mode == 0) {
86e6ffdd
N		 3669 mdk_rdev_t *rdev;
3670 struct list_head *tmp;
0d4ca600 3671
1da177e4
LT		 3672 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3673
978f946b
N		 3674 bitmap_destroy(mddev);
3675 if (mddev->bitmap_file) {
acc55e22 3676 restore_bitmap_write_access(mddev->bitmap_file);
978f946b
N		 3677 fput(mddev->bitmap_file);
3678 mddev->bitmap_file = NULL;
3679 }
3680 mddev->bitmap_offset = 0;
3681
86e6ffdd
N		 3682 ITERATE_RDEV(mddev,rdev,tmp)
3683 if (rdev->raid_disk >= 0) {
3684 char nm[20];
3685 sprintf(nm, "rd%d", rdev->raid_disk);
3686 sysfs_remove_link(&mddev->kobj, nm);
3687 }
3688
5792a285
N		 3689 /* make sure all delayed_delete calls have finished */
3690 flush_scheduled_work();
3691
1da177e4
LT		 3692 export_array(mddev);
3693
3694 mddev->array_size = 0;
9e653b63
N		 3695 mddev->size = 0;
3696 mddev->raid_disks = 0;
a94213b1 3697 mddev->recovery_cp = 0;
c6207277 3698 mddev->resync_max = MaxSector;
08a02ecd 3699 mddev->reshape_position = MaxSector;
e691063a 3700 mddev->external = 0;
1ec4a939 3701 mddev->persistent = 0;
9e653b63 3702
a8a55c38 3703 } else if (mddev->pers)
1da177e4
LT		 3704 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3705 mdname(mddev));
3706 err = 0;
d7603b7e 3707 md_new_event(mddev);
1da177e4
LT		 3708out:
3709 return err;
3710}
3711
fdee8ae4 3712#ifndef MODULE
1da177e4
LT		 3713static void autorun_array(mddev_t *mddev)
3714{
3715 mdk_rdev_t *rdev;
3716 struct list_head *tmp;
3717 int err;
3718
a757e64c 3719 if (list_empty(&mddev->disks))
1da177e4 3720 return;
1da177e4
LT		 3721
3722 printk(KERN_INFO "md: running: ");
3723
3724 ITERATE_RDEV(mddev,rdev,tmp) {
3725 char b[BDEVNAME_SIZE];
3726 printk("<%s>", bdevname(rdev->bdev,b));
3727 }
3728 printk("\n");
3729
3730 err = do_md_run (mddev);
3731 if (err) {
3732 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3733 do_md_stop (mddev, 0);
3734 }
3735}
3736
3737/*
3738 * lets try to run arrays based on all disks that have arrived
3739 * until now. (those are in pending_raid_disks)
3740 *
3741 * the method: pick the first pending disk, collect all disks with
3742 * the same UUID, remove all from the pending list and put them into
3743 * the 'same_array' list. Then order this list based on superblock
3744 * update time (freshest comes first), kick out 'old' disks and
3745 * compare superblocks. If everything's fine then run it.
3746 *
3747 * If "unit" is allocated, then bump its reference count
3748 */
3749static void autorun_devices(int part)
3750{
1da177e4
LT		 3751 struct list_head *tmp;
3752 mdk_rdev_t *rdev0, *rdev;
3753 mddev_t *mddev;
3754 char b[BDEVNAME_SIZE];
3755
3756 printk(KERN_INFO "md: autorun ...\n");
3757 while (!list_empty(&pending_raid_disks)) {
e8703fe1 3758 int unit;
1da177e4 3759 dev_t dev;
ad01c9e3 3760 LIST_HEAD(candidates);
1da177e4
LT		 3761 rdev0 = list_entry(pending_raid_disks.next,
3762 mdk_rdev_t, same_set);
3763
3764 printk(KERN_INFO "md: considering %s ...\n",
3765 bdevname(rdev0->bdev,b));
3766 INIT_LIST_HEAD(&candidates);
3767 ITERATE_RDEV_PENDING(rdev,tmp)
3768 if (super_90_load(rdev, rdev0, 0) >= 0) {
3769 printk(KERN_INFO "md: adding %s ...\n",
3770 bdevname(rdev->bdev,b));
3771 list_move(&rdev->same_set, &candidates);
3772 }
3773 /*
3774 * now we have a set of devices, with all of them having
3775 * mostly sane superblocks. It's time to allocate the
3776 * mddev.
3777 */
e8703fe1
N		 3778 if (part) {
3779 dev = MKDEV(mdp_major,
3780 rdev0->preferred_minor << MdpMinorShift);
3781 unit = MINOR(dev) >> MdpMinorShift;
3782 } else {
3783 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3784 unit = MINOR(dev);
3785 }
3786 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 3787 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3788 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3789 break;
3790 }
1da177e4
LT		 3791
3792 md_probe(dev, NULL, NULL);
3793 mddev = mddev_find(dev);
3794 if (!mddev) {
3795 printk(KERN_ERR
3796 "md: cannot allocate memory for md drive.\n");
3797 break;
3798 }
3799 if (mddev_lock(mddev))
3800 printk(KERN_WARNING "md: %s locked, cannot run\n",
3801 mdname(mddev));
3802 else if (mddev->raid_disks || mddev->major_version
3803 || !list_empty(&mddev->disks)) {
3804 printk(KERN_WARNING
3805 "md: %s already running, cannot run %s\n",
3806 mdname(mddev), bdevname(rdev0->bdev,b));
3807 mddev_unlock(mddev);
3808 } else {
3809 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 3810 mddev->persistent = 1;
1da177e4
LT		 3811 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3812 list_del_init(&rdev->same_set);
3813 if (bind_rdev_to_array(rdev, mddev))
3814 export_rdev(rdev);
3815 }
3816 autorun_array(mddev);
3817 mddev_unlock(mddev);
3818 }
3819 /* on success, candidates will be empty, on error
3820 * it won't...
3821 */
3822 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3823 export_rdev(rdev);
3824 mddev_put(mddev);
3825 }
3826 printk(KERN_INFO "md: ... autorun DONE.\n");
3827}
fdee8ae4 3828#endif /* !MODULE */
1da177e4 3829
1da177e4
LT		 3830static int get_version(void __user * arg)
3831{
3832 mdu_version_t ver;
3833
3834 ver.major = MD_MAJOR_VERSION;
3835 ver.minor = MD_MINOR_VERSION;
3836 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3837
3838 if (copy_to_user(arg, &ver, sizeof(ver)))
3839 return -EFAULT;
3840
3841 return 0;
3842}
3843
3844static int get_array_info(mddev_t * mddev, void __user * arg)
3845{
3846 mdu_array_info_t info;
3847 int nr,working,active,failed,spare;
3848 mdk_rdev_t *rdev;
3849 struct list_head *tmp;
3850
3851 nr=working=active=failed=spare=0;
3852 ITERATE_RDEV(mddev,rdev,tmp) {
3853 nr++;
b2d444d7 3854 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3855 failed++;
3856 else {
3857 working++;
b2d444d7 3858 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 3859 active++;
3860 else
3861 spare++;
3862 }
3863 }
3864
3865 info.major_version = mddev->major_version;
3866 info.minor_version = mddev->minor_version;
3867 info.patch_version = MD_PATCHLEVEL_VERSION;
3868 info.ctime = mddev->ctime;
3869 info.level = mddev->level;
3870 info.size = mddev->size;
284ae7ca
N		 3871 if (info.size != mddev->size) /* overflow */
3872 info.size = -1;
1da177e4
LT		 3873 info.nr_disks = nr;
3874 info.raid_disks = mddev->raid_disks;
3875 info.md_minor = mddev->md_minor;
3876 info.not_persistent= !mddev->persistent;
3877
3878 info.utime = mddev->utime;
3879 info.state = 0;
3880 if (mddev->in_sync)
3881 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 3882 if (mddev->bitmap && mddev->bitmap_offset)
3883 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 3884 info.active_disks = active;
3885 info.working_disks = working;
3886 info.failed_disks = failed;
3887 info.spare_disks = spare;
3888
3889 info.layout = mddev->layout;
3890 info.chunk_size = mddev->chunk_size;
3891
3892 if (copy_to_user(arg, &info, sizeof(info)))
3893 return -EFAULT;
3894
3895 return 0;
3896}
3897
87162a28 3898static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 3899{
3900 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3901 char *ptr, *buf = NULL;
3902 int err = -ENOMEM;
3903
2a2275d6
N		 3904 md_allow_write(mddev);
3905
32a7627c
N		 3906 file = kmalloc(sizeof(*file), GFP_KERNEL);
3907 if (!file)
3908 goto out;
3909
3910 /* bitmap disabled, zero the first byte and copy out */
3911 if (!mddev->bitmap || !mddev->bitmap->file) {
3912 file->pathname[0] = '\0';
3913 goto copy_out;
3914 }
3915
3916 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3917 if (!buf)
3918 goto out;
3919
3920 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3921 if (!ptr)
3922 goto out;
3923
3924 strcpy(file->pathname, ptr);
3925
3926copy_out:
3927 err = 0;
3928 if (copy_to_user(arg, file, sizeof(*file)))
3929 err = -EFAULT;
3930out:
3931 kfree(buf);
3932 kfree(file);
3933 return err;
3934}
3935
1da177e4
LT		 3936static int get_disk_info(mddev_t * mddev, void __user * arg)
3937{
3938 mdu_disk_info_t info;
3939 unsigned int nr;
3940 mdk_rdev_t *rdev;
3941
3942 if (copy_from_user(&info, arg, sizeof(info)))
3943 return -EFAULT;
3944
3945 nr = info.number;
3946
3947 rdev = find_rdev_nr(mddev, nr);
3948 if (rdev) {
3949 info.major = MAJOR(rdev->bdev->bd_dev);
3950 info.minor = MINOR(rdev->bdev->bd_dev);
3951 info.raid_disk = rdev->raid_disk;
3952 info.state = 0;
b2d444d7 3953 if (test_bit(Faulty, &rdev->flags))
1da177e4 3954 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 3955 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 3956 info.state |= (1<<MD_DISK_ACTIVE);
3957 info.state |= (1<<MD_DISK_SYNC);
3958 }
8ddf9efe
N		 3959 if (test_bit(WriteMostly, &rdev->flags))
3960 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 3961 } else {
3962 info.major = info.minor = 0;
3963 info.raid_disk = -1;
3964 info.state = (1<<MD_DISK_REMOVED);
3965 }
3966
3967 if (copy_to_user(arg, &info, sizeof(info)))
3968 return -EFAULT;
3969
3970 return 0;
3971}
3972
3973static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3974{
3975 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3976 mdk_rdev_t *rdev;
3977 dev_t dev = MKDEV(info->major,info->minor);
3978
3979 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3980 return -EOVERFLOW;
3981
3982 if (!mddev->raid_disks) {
3983 int err;
3984 /* expecting a device which has a superblock */
3985 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3986 if (IS_ERR(rdev)) {
3987 printk(KERN_WARNING
3988 "md: md_import_device returned %ld\n",
3989 PTR_ERR(rdev));
3990 return PTR_ERR(rdev);
3991 }
3992 if (!list_empty(&mddev->disks)) {
3993 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3994 mdk_rdev_t, same_set);
3995 int err = super_types[mddev->major_version]
3996 .load_super(rdev, rdev0, mddev->minor_version);
3997 if (err < 0) {
3998 printk(KERN_WARNING
3999 "md: %s has different UUID to %s\n",
4000 bdevname(rdev->bdev,b),
4001 bdevname(rdev0->bdev,b2));
4002 export_rdev(rdev);
4003 return -EINVAL;
4004 }
4005 }
4006 err = bind_rdev_to_array(rdev, mddev);
4007 if (err)
4008 export_rdev(rdev);
4009 return err;
4010 }
4011
4012 /*
4013 * add_new_disk can be used once the array is assembled
4014 * to add "hot spares". They must already have a superblock
4015 * written
4016 */
4017 if (mddev->pers) {
4018 int err;
4019 if (!mddev->pers->hot_add_disk) {
4020 printk(KERN_WARNING
4021 "%s: personality does not support diskops!\n",
4022 mdname(mddev));
4023 return -EINVAL;
4024 }
7b1e35f6
N		 4025 if (mddev->persistent)
4026 rdev = md_import_device(dev, mddev->major_version,
4027 mddev->minor_version);
4028 else
4029 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 4030 if (IS_ERR(rdev)) {
4031 printk(KERN_WARNING
4032 "md: md_import_device returned %ld\n",
4033 PTR_ERR(rdev));
4034 return PTR_ERR(rdev);
4035 }
41158c7e
N		 4036 /* set save_raid_disk if appropriate */
4037 if (!mddev->persistent) {
4038 if (info->state & (1<<MD_DISK_SYNC) &&
4039 info->raid_disk < mddev->raid_disks)
4040 rdev->raid_disk = info->raid_disk;
4041 else
4042 rdev->raid_disk = -1;
4043 } else
4044 super_types[mddev->major_version].
4045 validate_super(mddev, rdev);
4046 rdev->saved_raid_disk = rdev->raid_disk;
4047
b2d444d7 4048 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 4049 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4050 set_bit(WriteMostly, &rdev->flags);
4051
1da177e4
LT		 4052 rdev->raid_disk = -1;
4053 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 4054 if (!err && !mddev->pers->hot_remove_disk) {
4055 /* If there is hot_add_disk but no hot_remove_disk
4056 * then added disks for geometry changes,
4057 * and should be added immediately.
4058 */
4059 super_types[mddev->major_version].
4060 validate_super(mddev, rdev);
4061 err = mddev->pers->hot_add_disk(mddev, rdev);
4062 if (err)
4063 unbind_rdev_from_array(rdev);
4064 }
1da177e4
LT		 4065 if (err)
4066 export_rdev(rdev);
c361777f 4067
17571284 4068 md_update_sb(mddev, 1);
c361777f 4069 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 4070 md_wakeup_thread(mddev->thread);
1da177e4
LT		 4071 return err;
4072 }
4073
4074 /* otherwise, add_new_disk is only allowed
4075 * for major_version==0 superblocks
4076 */
4077 if (mddev->major_version != 0) {
4078 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4079 mdname(mddev));
4080 return -EINVAL;
4081 }
4082
4083 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4084 int err;
4085 rdev = md_import_device (dev, -1, 0);
4086 if (IS_ERR(rdev)) {
4087 printk(KERN_WARNING
4088 "md: error, md_import_device() returned %ld\n",
4089 PTR_ERR(rdev));
4090 return PTR_ERR(rdev);
4091 }
4092 rdev->desc_nr = info->number;
4093 if (info->raid_disk < mddev->raid_disks)
4094 rdev->raid_disk = info->raid_disk;
4095 else
4096 rdev->raid_disk = -1;
4097
1da177e4 4098 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 4099 if (info->state & (1<<MD_DISK_SYNC))
4100 set_bit(In_sync, &rdev->flags);
1da177e4 4101
8ddf9efe
N		 4102 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4103 set_bit(WriteMostly, &rdev->flags);
4104
1da177e4
LT		 4105 if (!mddev->persistent) {
4106 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4107 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4108 } else
4109 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4110 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4111
2bf071bf
N		 4112 err = bind_rdev_to_array(rdev, mddev);
4113 if (err) {
4114 export_rdev(rdev);
4115 return err;
4116 }
1da177e4
LT		 4117 }
4118
4119 return 0;
4120}
4121
4122static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4123{
4124 char b[BDEVNAME_SIZE];
4125 mdk_rdev_t *rdev;
4126
4127 if (!mddev->pers)
4128 return -ENODEV;
4129
4130 rdev = find_rdev(mddev, dev);
4131 if (!rdev)
4132 return -ENXIO;
4133
4134 if (rdev->raid_disk >= 0)
4135 goto busy;
4136
4137 kick_rdev_from_array(rdev);
850b2b42 4138 md_update_sb(mddev, 1);
d7603b7e 4139 md_new_event(mddev);
1da177e4
LT		 4140
4141 return 0;
4142busy:
4143 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
4144 bdevname(rdev->bdev,b), mdname(mddev));
4145 return -EBUSY;
4146}
4147
4148static int hot_add_disk(mddev_t * mddev, dev_t dev)
4149{
4150 char b[BDEVNAME_SIZE];
4151 int err;
4152 unsigned int size;
4153 mdk_rdev_t *rdev;
4154
4155 if (!mddev->pers)
4156 return -ENODEV;
4157
4158 if (mddev->major_version != 0) {
4159 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4160 " version-0 superblocks.\n",
4161 mdname(mddev));
4162 return -EINVAL;
4163 }
4164 if (!mddev->pers->hot_add_disk) {
4165 printk(KERN_WARNING
4166 "%s: personality does not support diskops!\n",
4167 mdname(mddev));
4168 return -EINVAL;
4169 }
4170
4171 rdev = md_import_device (dev, -1, 0);
4172 if (IS_ERR(rdev)) {
4173 printk(KERN_WARNING
4174 "md: error, md_import_device() returned %ld\n",
4175 PTR_ERR(rdev));
4176 return -EINVAL;
4177 }
4178
4179 if (mddev->persistent)
4180 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4181 else
4182 rdev->sb_offset =
4183 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4184
4185 size = calc_dev_size(rdev, mddev->chunk_size);
4186 rdev->size = size;
4187
b2d444d7 4188 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4189 printk(KERN_WARNING
4190 "md: can not hot-add faulty %s disk to %s!\n",
4191 bdevname(rdev->bdev,b), mdname(mddev));
4192 err = -EINVAL;
4193 goto abort_export;
4194 }
b2d444d7 4195 clear_bit(In_sync, &rdev->flags);
1da177e4 4196 rdev->desc_nr = -1;
5842730d 4197 rdev->saved_raid_disk = -1;
2bf071bf
N		 4198 err = bind_rdev_to_array(rdev, mddev);
4199 if (err)
4200 goto abort_export;
1da177e4
LT		 4201
4202 /*
4203 * The rest should better be atomic, we can have disk failures
4204 * noticed in interrupt contexts ...
4205 */
4206
4207 if (rdev->desc_nr == mddev->max_disks) {
4208 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4209 mdname(mddev));
4210 err = -EBUSY;
4211 goto abort_unbind_export;
4212 }
4213
4214 rdev->raid_disk = -1;
4215
850b2b42 4216 md_update_sb(mddev, 1);
1da177e4
LT		 4217
4218 /*
4219 * Kick recovery, maybe this spare has to be added to the
4220 * array immediately.
4221 */
4222 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4223 md_wakeup_thread(mddev->thread);
d7603b7e 4224 md_new_event(mddev);
1da177e4
LT		 4225 return 0;
4226
4227abort_unbind_export:
4228 unbind_rdev_from_array(rdev);
4229
4230abort_export:
4231 export_rdev(rdev);
4232 return err;
4233}
4234
32a7627c
N		 4235static int set_bitmap_file(mddev_t *mddev, int fd)
4236{
4237 int err;
4238
36fa3063
N		 4239 if (mddev->pers) {
4240 if (!mddev->pers->quiesce)
4241 return -EBUSY;
4242 if (mddev->recovery || mddev->sync_thread)
4243 return -EBUSY;
4244 /* we should be able to change the bitmap.. */
4245 }
32a7627c 4246
32a7627c 4247
36fa3063
N		 4248 if (fd >= 0) {
4249 if (mddev->bitmap)
4250 return -EEXIST; /* cannot add when bitmap is present */
4251 mddev->bitmap_file = fget(fd);
32a7627c 4252
36fa3063
N		 4253 if (mddev->bitmap_file == NULL) {
4254 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4255 mdname(mddev));
4256 return -EBADF;
4257 }
4258
4259 err = deny_bitmap_write_access(mddev->bitmap_file);
4260 if (err) {
4261 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4262 mdname(mddev));
4263 fput(mddev->bitmap_file);
4264 mddev->bitmap_file = NULL;
4265 return err;
4266 }
a654b9d8 4267 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 4268 } else if (mddev->bitmap == NULL)
4269 return -ENOENT; /* cannot remove what isn't there */
4270 err = 0;
4271 if (mddev->pers) {
4272 mddev->pers->quiesce(mddev, 1);
4273 if (fd >= 0)
4274 err = bitmap_create(mddev);
d7375ab3 4275 if (fd < 0 || err) {
36fa3063 4276 bitmap_destroy(mddev);
d7375ab3
N		 4277 fd = -1; /* make sure to put the file */
4278 }
36fa3063 4279 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 4280 }
4281 if (fd < 0) {
acc55e22
N		 4282 if (mddev->bitmap_file) {
4283 restore_bitmap_write_access(mddev->bitmap_file);
36fa3063 4284 fput(mddev->bitmap_file);
acc55e22 4285 }
36fa3063
N		 4286 mddev->bitmap_file = NULL;
4287 }
4288
32a7627c
N		 4289 return err;
4290}
4291
1da177e4
LT		 4292/*
4293 * set_array_info is used two different ways
4294 * The original usage is when creating a new array.
4295 * In this usage, raid_disks is > 0 and it together with
4296 * level, size, not_persistent,layout,chunksize determine the
4297 * shape of the array.
4298 * This will always create an array with a type-0.90.0 superblock.
4299 * The newer usage is when assembling an array.
4300 * In this case raid_disks will be 0, and the major_version field is
4301 * use to determine which style super-blocks are to be found on the devices.
4302 * The minor and patch _version numbers are also kept incase the
4303 * super_block handler wishes to interpret them.
4304 */
4305static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4306{
4307
4308 if (info->raid_disks == 0) {
4309 /* just setting version number for superblock loading */
4310 if (info->major_version < 0 ||
50511da3 4311 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 4312 super_types[info->major_version].name == NULL) {
4313 /* maybe try to auto-load a module? */
4314 printk(KERN_INFO
4315 "md: superblock version %d not known\n",
4316 info->major_version);
4317 return -EINVAL;
4318 }
4319 mddev->major_version = info->major_version;
4320 mddev->minor_version = info->minor_version;
4321 mddev->patch_version = info->patch_version;
3f9d7b0d 4322 mddev->persistent = !info->not_persistent;
1da177e4
LT		 4323 return 0;
4324 }
4325 mddev->major_version = MD_MAJOR_VERSION;
4326 mddev->minor_version = MD_MINOR_VERSION;
4327 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4328 mddev->ctime = get_seconds();
4329
4330 mddev->level = info->level;
17115e03 4331 mddev->clevel[0] = 0;
1da177e4
LT		 4332 mddev->size = info->size;
4333 mddev->raid_disks = info->raid_disks;
4334 /* don't set md_minor, it is determined by which /dev/md* was
4335 * openned
4336 */
4337 if (info->state & (1<<MD_SB_CLEAN))
4338 mddev->recovery_cp = MaxSector;
4339 else
4340 mddev->recovery_cp = 0;
4341 mddev->persistent = ! info->not_persistent;
e691063a 4342 mddev->external = 0;
1da177e4
LT		 4343
4344 mddev->layout = info->layout;
4345 mddev->chunk_size = info->chunk_size;
4346
4347 mddev->max_disks = MD_SB_DISKS;
4348
e691063a
N		 4349 if (mddev->persistent)
4350 mddev->flags = 0;
850b2b42 4351 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 4352
b2a2703c
N		 4353 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4354 mddev->bitmap_offset = 0;
4355
f6705578
N		 4356 mddev->reshape_position = MaxSector;
4357
1da177e4
LT		 4358 /*
4359 * Generate a 128 bit UUID
4360 */
4361 get_random_bytes(mddev->uuid, 16);
4362
f6705578
N		 4363 mddev->new_level = mddev->level;
4364 mddev->new_chunk = mddev->chunk_size;
4365 mddev->new_layout = mddev->layout;
4366 mddev->delta_disks = 0;
4367
1da177e4
LT		 4368 return 0;
4369}
4370
a35b0d69
N		 4371static int update_size(mddev_t *mddev, unsigned long size)
4372{
4373 mdk_rdev_t * rdev;
4374 int rv;
4375 struct list_head *tmp;
8ddeeae5 4376 int fit = (size == 0);
a35b0d69
N		 4377
4378 if (mddev->pers->resize == NULL)
4379 return -EINVAL;
4380 /* The "size" is the amount of each device that is used.
4381 * This can only make sense for arrays with redundancy.
4382 * linear and raid0 always use whatever space is available
4383 * We can only consider changing the size if no resync
4384 * or reconstruction is happening, and if the new size
4385 * is acceptable. It must fit before the sb_offset or,
4386 * if that is <data_offset, it must fit before the
4387 * size of each device.
4388 * If size is zero, we find the largest size that fits.
4389 */
4390 if (mddev->sync_thread)
4391 return -EBUSY;
4392 ITERATE_RDEV(mddev,rdev,tmp) {
4393 sector_t avail;
01ab5662
N		 4394 avail = rdev->size * 2;
4395
a35b0d69
N		 4396 if (fit && (size == 0 || size > avail/2))
4397 size = avail/2;
4398 if (avail < ((sector_t)size << 1))
4399 return -ENOSPC;
4400 }
4401 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4402 if (!rv) {
4403 struct block_device *bdev;
4404
4405 bdev = bdget_disk(mddev->gendisk, 0);
4406 if (bdev) {
1b1dcc1b 4407 mutex_lock(&bdev->bd_inode->i_mutex);
6d89332b 4408 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
1b1dcc1b 4409 mutex_unlock(&bdev->bd_inode->i_mutex);
a35b0d69
N		 4410 bdput(bdev);
4411 }
4412 }
4413 return rv;
4414}
4415
da943b99
N		 4416static int update_raid_disks(mddev_t *mddev, int raid_disks)
4417{
4418 int rv;
4419 /* change the number of raid disks */
63c70c4f 4420 if (mddev->pers->check_reshape == NULL)
da943b99
N		 4421 return -EINVAL;
4422 if (raid_disks <= 0 ||
4423 raid_disks >= mddev->max_disks)
4424 return -EINVAL;
63c70c4f 4425 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 4426 return -EBUSY;
63c70c4f
N		 4427 mddev->delta_disks = raid_disks - mddev->raid_disks;
4428
4429 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 4430 return rv;
4431}
4432
4433
1da177e4
LT		 4434/*
4435 * update_array_info is used to change the configuration of an
4436 * on-line array.
4437 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4438 * fields in the info are checked against the array.
4439 * Any differences that cannot be handled will cause an error.
4440 * Normally, only one change can be managed at a time.
4441 */
4442static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4443{
4444 int rv = 0;
4445 int cnt = 0;
36fa3063
N		 4446 int state = 0;
4447
4448 /* calculate expected state,ignoring low bits */
4449 if (mddev->bitmap && mddev->bitmap_offset)
4450 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 4451
4452 if (mddev->major_version != info->major_version ||
4453 mddev->minor_version != info->minor_version ||
4454/* mddev->patch_version != info->patch_version || */
4455 mddev->ctime != info->ctime ||
4456 mddev->level != info->level ||
4457/* mddev->layout != info->layout || */
4458 !mddev->persistent != info->not_persistent||
36fa3063
N		 4459 mddev->chunk_size != info->chunk_size ||
4460 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4461 ((state^info->state) & 0xfffffe00)
4462 )
1da177e4
LT		 4463 return -EINVAL;
4464 /* Check there is only one change */
284ae7ca 4465 if (info->size >= 0 && mddev->size != info->size) cnt++;
1da177e4
LT		 4466 if (mddev->raid_disks != info->raid_disks) cnt++;
4467 if (mddev->layout != info->layout) cnt++;
36fa3063 4468 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 4469 if (cnt == 0) return 0;
4470 if (cnt > 1) return -EINVAL;
4471
4472 if (mddev->layout != info->layout) {
4473 /* Change layout
4474 * we don't need to do anything at the md level, the
4475 * personality will take care of it all.
4476 */
4477 if (mddev->pers->reconfig == NULL)
4478 return -EINVAL;
4479 else
4480 return mddev->pers->reconfig(mddev, info->layout, -1);
4481 }
284ae7ca 4482 if (info->size >= 0 && mddev->size != info->size)
a35b0d69
N		 4483 rv = update_size(mddev, info->size);
4484
da943b99
N		 4485 if (mddev->raid_disks != info->raid_disks)
4486 rv = update_raid_disks(mddev, info->raid_disks);
4487
36fa3063
N		 4488 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4489 if (mddev->pers->quiesce == NULL)
4490 return -EINVAL;
4491 if (mddev->recovery || mddev->sync_thread)
4492 return -EBUSY;
4493 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4494 /* add the bitmap */
4495 if (mddev->bitmap)
4496 return -EEXIST;
4497 if (mddev->default_bitmap_offset == 0)
4498 return -EINVAL;
4499 mddev->bitmap_offset = mddev->default_bitmap_offset;
4500 mddev->pers->quiesce(mddev, 1);
4501 rv = bitmap_create(mddev);
4502 if (rv)
4503 bitmap_destroy(mddev);
4504 mddev->pers->quiesce(mddev, 0);
4505 } else {
4506 /* remove the bitmap */
4507 if (!mddev->bitmap)
4508 return -ENOENT;
4509 if (mddev->bitmap->file)
4510 return -EINVAL;
4511 mddev->pers->quiesce(mddev, 1);
4512 bitmap_destroy(mddev);
4513 mddev->pers->quiesce(mddev, 0);
4514 mddev->bitmap_offset = 0;
4515 }
4516 }
850b2b42 4517 md_update_sb(mddev, 1);
1da177e4
LT		 4518 return rv;
4519}
4520
4521static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4522{
4523 mdk_rdev_t *rdev;
4524
4525 if (mddev->pers == NULL)
4526 return -ENODEV;
4527
4528 rdev = find_rdev(mddev, dev);
4529 if (!rdev)
4530 return -ENODEV;
4531
4532 md_error(mddev, rdev);
4533 return 0;
4534}
4535
a885c8c4
CH		 4536static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4537{
4538 mddev_t *mddev = bdev->bd_disk->private_data;
4539
4540 geo->heads = 2;
4541 geo->sectors = 4;
4542 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4543 return 0;
4544}
4545
1da177e4
LT		 4546static int md_ioctl(struct inode *inode, struct file *file,
4547 unsigned int cmd, unsigned long arg)
4548{
4549 int err = 0;
4550 void __user *argp = (void __user *)arg;
1da177e4
LT		 4551 mddev_t *mddev = NULL;
4552
4553 if (!capable(CAP_SYS_ADMIN))
4554 return -EACCES;
4555
4556 /*
4557 * Commands dealing with the RAID driver but not any
4558 * particular array:
4559 */
4560 switch (cmd)
4561 {
4562 case RAID_VERSION:
4563 err = get_version(argp);
4564 goto done;
4565
4566 case PRINT_RAID_DEBUG:
4567 err = 0;
4568 md_print_devices();
4569 goto done;
4570
4571#ifndef MODULE
4572 case RAID_AUTORUN:
4573 err = 0;
4574 autostart_arrays(arg);
4575 goto done;
4576#endif
4577 default:;
4578 }
4579
4580 /*
4581 * Commands creating/starting a new array:
4582 */
4583
4584 mddev = inode->i_bdev->bd_disk->private_data;
4585
4586 if (!mddev) {
4587 BUG();
4588 goto abort;
4589 }
4590
1da177e4
LT		 4591 err = mddev_lock(mddev);
4592 if (err) {
4593 printk(KERN_INFO
4594 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4595 err, cmd);
4596 goto abort;
4597 }
4598
4599 switch (cmd)
4600 {
4601 case SET_ARRAY_INFO:
4602 {
4603 mdu_array_info_t info;
4604 if (!arg)
4605 memset(&info, 0, sizeof(info));
4606 else if (copy_from_user(&info, argp, sizeof(info))) {
4607 err = -EFAULT;
4608 goto abort_unlock;
4609 }
4610 if (mddev->pers) {
4611 err = update_array_info(mddev, &info);
4612 if (err) {
4613 printk(KERN_WARNING "md: couldn't update"
4614 " array info. %d\n", err);
4615 goto abort_unlock;
4616 }
4617 goto done_unlock;
4618 }
4619 if (!list_empty(&mddev->disks)) {
4620 printk(KERN_WARNING
4621 "md: array %s already has disks!\n",
4622 mdname(mddev));
4623 err = -EBUSY;
4624 goto abort_unlock;
4625 }
4626 if (mddev->raid_disks) {
4627 printk(KERN_WARNING
4628 "md: array %s already initialised!\n",
4629 mdname(mddev));
4630 err = -EBUSY;
4631 goto abort_unlock;
4632 }
4633 err = set_array_info(mddev, &info);
4634 if (err) {
4635 printk(KERN_WARNING "md: couldn't set"
4636 " array info. %d\n", err);
4637 goto abort_unlock;
4638 }
4639 }
4640 goto done_unlock;
4641
4642 default:;
4643 }
4644
4645 /*
4646 * Commands querying/configuring an existing array:
4647 */
32a7627c 4648 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 4649 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
32a7627c 4650 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3f9d7b0d
N		 4651 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4652 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 4653 err = -ENODEV;
4654 goto abort_unlock;
4655 }
4656
4657 /*
4658 * Commands even a read-only array can execute:
4659 */
4660 switch (cmd)
4661 {
4662 case GET_ARRAY_INFO:
4663 err = get_array_info(mddev, argp);
4664 goto done_unlock;
4665
32a7627c 4666 case GET_BITMAP_FILE:
87162a28 4667 err = get_bitmap_file(mddev, argp);
32a7627c
N		 4668 goto done_unlock;
4669
1da177e4
LT		 4670 case GET_DISK_INFO:
4671 err = get_disk_info(mddev, argp);
4672 goto done_unlock;
4673
4674 case RESTART_ARRAY_RW:
4675 err = restart_array(mddev);
4676 goto done_unlock;
4677
4678 case STOP_ARRAY:
4679 err = do_md_stop (mddev, 0);
4680 goto done_unlock;
4681
4682 case STOP_ARRAY_RO:
4683 err = do_md_stop (mddev, 1);
4684 goto done_unlock;
4685
4686 /*
4687 * We have a problem here : there is no easy way to give a CHS
4688 * virtual geometry. We currently pretend that we have a 2 heads
4689 * 4 sectors (with a BIG number of cylinders...). This drives
4690 * dosfs just mad... ;-)
4691 */
1da177e4
LT		 4692 }
4693
4694 /*
4695 * The remaining ioctls are changing the state of the
f91de92e
N		 4696 * superblock, so we do not allow them on read-only arrays.
4697 * However non-MD ioctls (e.g. get-size) will still come through
4698 * here and hit the 'default' below, so only disallow
4699 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 4700 */
f91de92e
N		 4701 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4702 mddev->ro && mddev->pers) {
4703 if (mddev->ro == 2) {
4704 mddev->ro = 0;
4705 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4706 md_wakeup_thread(mddev->thread);
4707
4708 } else {
4709 err = -EROFS;
4710 goto abort_unlock;
4711 }
1da177e4
LT		 4712 }
4713
4714 switch (cmd)
4715 {
4716 case ADD_NEW_DISK:
4717 {
4718 mdu_disk_info_t info;
4719 if (copy_from_user(&info, argp, sizeof(info)))
4720 err = -EFAULT;
4721 else
4722 err = add_new_disk(mddev, &info);
4723 goto done_unlock;
4724 }
4725
4726 case HOT_REMOVE_DISK:
4727 err = hot_remove_disk(mddev, new_decode_dev(arg));
4728 goto done_unlock;
4729
4730 case HOT_ADD_DISK:
4731 err = hot_add_disk(mddev, new_decode_dev(arg));
4732 goto done_unlock;
4733
4734 case SET_DISK_FAULTY:
4735 err = set_disk_faulty(mddev, new_decode_dev(arg));
4736 goto done_unlock;
4737
4738 case RUN_ARRAY:
4739 err = do_md_run (mddev);
4740 goto done_unlock;
4741
32a7627c
N		 4742 case SET_BITMAP_FILE:
4743 err = set_bitmap_file(mddev, (int)arg);
4744 goto done_unlock;
4745
1da177e4 4746 default:
1da177e4
LT		 4747 err = -EINVAL;
4748 goto abort_unlock;
4749 }
4750
4751done_unlock:
4752abort_unlock:
4753 mddev_unlock(mddev);
4754
4755 return err;
4756done:
4757 if (err)
4758 MD_BUG();
4759abort:
4760 return err;
4761}
4762
4763static int md_open(struct inode *inode, struct file *file)
4764{
4765 /*
4766 * Succeed if we can lock the mddev, which confirms that
4767 * it isn't being stopped right now.
4768 */
4769 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4770 int err;
4771
d63a5a74 4772 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
1da177e4
LT		 4773 goto out;
4774
4775 err = 0;
4776 mddev_get(mddev);
4777 mddev_unlock(mddev);
4778
4779 check_disk_change(inode->i_bdev);
4780 out:
4781 return err;
4782}
4783
4784static int md_release(struct inode *inode, struct file * file)
4785{
4786 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4787
52e5f9d1 4788 BUG_ON(!mddev);
1da177e4
LT		 4789 mddev_put(mddev);
4790
4791 return 0;
4792}
4793
44ce6294
LT		 4794static int md_media_changed(struct gendisk *disk)
4795{
4796 mddev_t *mddev = disk->private_data;
4797
4798 return mddev->changed;
4799}
4800
4801static int md_revalidate(struct gendisk *disk)
4802{
4803 mddev_t *mddev = disk->private_data;
4804
4805 mddev->changed = 0;
4806 return 0;
4807}
1da177e4
LT		 4808static struct block_device_operations md_fops =
4809{
4810 .owner = THIS_MODULE,
4811 .open = md_open,
4812 .release = md_release,
4813 .ioctl = md_ioctl,
a885c8c4 4814 .getgeo = md_getgeo,
44ce6294
LT		 4815 .media_changed = md_media_changed,
4816 .revalidate_disk= md_revalidate,
1da177e4
LT		 4817};
4818
75c96f85 4819static int md_thread(void * arg)
1da177e4
LT		 4820{
4821 mdk_thread_t *thread = arg;
4822
1da177e4
LT		 4823 /*
4824 * md_thread is a 'system-thread', it's priority should be very
4825 * high. We avoid resource deadlocks individually in each
4826 * raid personality. (RAID5 does preallocation) We also use RR and
4827 * the very same RT priority as kswapd, thus we will never get
4828 * into a priority inversion deadlock.
4829 *
4830 * we definitely have to have equal or higher priority than
4831 * bdflush, otherwise bdflush will deadlock if there are too
4832 * many dirty RAID5 blocks.
4833 */
1da177e4 4834
6985c43f 4835 allow_signal(SIGKILL);
a6fb0934 4836 while (!kthread_should_stop()) {
1da177e4 4837
93588e22
N		 4838 /* We need to wait INTERRUPTIBLE so that
4839 * we don't add to the load-average.
4840 * That means we need to be sure no signals are
4841 * pending
4842 */
4843 if (signal_pending(current))
4844 flush_signals(current);
4845
4846 wait_event_interruptible_timeout
4847 (thread->wqueue,
4848 test_bit(THREAD_WAKEUP, &thread->flags)
4849 || kthread_should_stop(),
4850 thread->timeout);
1da177e4
LT		 4851
4852 clear_bit(THREAD_WAKEUP, &thread->flags);
4853
787453c2 4854 thread->run(thread->mddev);
1da177e4 4855 }
a6fb0934 4856
1da177e4
LT		 4857 return 0;
4858}
4859
4860void md_wakeup_thread(mdk_thread_t *thread)
4861{
4862 if (thread) {
4863 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4864 set_bit(THREAD_WAKEUP, &thread->flags);
4865 wake_up(&thread->wqueue);
4866 }
4867}
4868
4869mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4870 const char *name)
4871{
4872 mdk_thread_t *thread;
1da177e4 4873
9ffae0cf 4874 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 4875 if (!thread)
4876 return NULL;
4877
1da177e4
LT		 4878 init_waitqueue_head(&thread->wqueue);
4879
1da177e4
LT		 4880 thread->run = run;
4881 thread->mddev = mddev;
32a7627c 4882 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 4883 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 4884 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 4885 kfree(thread);
4886 return NULL;
4887 }
1da177e4
LT		 4888 return thread;
4889}
4890
1da177e4
LT		 4891void md_unregister_thread(mdk_thread_t *thread)
4892{
ba25f9dc 4893 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
a6fb0934
N		 4894
4895 kthread_stop(thread->tsk);
1da177e4
LT		 4896 kfree(thread);
4897}
4898
4899void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4900{
4901 if (!mddev) {
4902 MD_BUG();
4903 return;
4904 }
4905
b2d444d7 4906 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 4907 return;
32a7627c 4908/*
1da177e4
LT		 4909 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4910 mdname(mddev),
4911 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4912 __builtin_return_address(0),__builtin_return_address(1),
4913 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 4914*/
d0a0a5ee
AM		 4915 if (!mddev->pers)
4916 return;
1da177e4
LT		 4917 if (!mddev->pers->error_handler)
4918 return;
4919 mddev->pers->error_handler(mddev,rdev);
4920 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4921 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4922 md_wakeup_thread(mddev->thread);
c331eb04 4923 md_new_event_inintr(mddev);
1da177e4
LT		 4924}
4925
4926/* seq_file implementation /proc/mdstat */
4927
4928static void status_unused(struct seq_file *seq)
4929{
4930 int i = 0;
4931 mdk_rdev_t *rdev;
4932 struct list_head *tmp;
4933
4934 seq_printf(seq, "unused devices: ");
4935
4936 ITERATE_RDEV_PENDING(rdev,tmp) {
4937 char b[BDEVNAME_SIZE];
4938 i++;
4939 seq_printf(seq, "%s ",
4940 bdevname(rdev->bdev,b));
4941 }
4942 if (!i)
4943 seq_printf(seq, "<none>");
4944
4945 seq_printf(seq, "\n");
4946}
4947
4948
4949static void status_resync(struct seq_file *seq, mddev_t * mddev)
4950{
4588b42e
N		 4951 sector_t max_blocks, resync, res;
4952 unsigned long dt, db, rt;
4953 int scale;
4954 unsigned int per_milli;
1da177e4
LT		 4955
4956 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4957
4958 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4959 max_blocks = mddev->resync_max_sectors >> 1;
4960 else
4961 max_blocks = mddev->size;
4962
4963 /*
4964 * Should not happen.
4965 */
4966 if (!max_blocks) {
4967 MD_BUG();
4968 return;
4969 }
4588b42e
N		 4970 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4971 * in a sector_t, and (max_blocks>>scale) will fit in a
4972 * u32, as those are the requirements for sector_div.
4973 * Thus 'scale' must be at least 10
4974 */
4975 scale = 10;
4976 if (sizeof(sector_t) > sizeof(unsigned long)) {
4977 while ( max_blocks/2 > (1ULL<<(scale+32)))
4978 scale++;
4979 }
4980 res = (resync>>scale)*1000;
4981 sector_div(res, (u32)((max_blocks>>scale)+1));
4982
4983 per_milli = res;
1da177e4 4984 {
4588b42e 4985 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 4986 seq_printf(seq, "[");
4987 for (i = 0; i < x; i++)
4988 seq_printf(seq, "=");
4989 seq_printf(seq, ">");
4990 for (i = 0; i < y; i++)
4991 seq_printf(seq, ".");
4992 seq_printf(seq, "] ");
4993 }
4588b42e 4994 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 4995 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4996 "reshape" :
61df9d91
N		 4997 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4998 "check" :
4999 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5000 "resync" : "recovery"))),
5001 per_milli/10, per_milli % 10,
4588b42e
N		 5002 (unsigned long long) resync,
5003 (unsigned long long) max_blocks);
1da177e4
LT		 5004
5005 /*
5006 * We do not want to overflow, so the order of operands and
5007 * the * 100 / 100 trick are important. We do a +1 to be
5008 * safe against division by zero. We only estimate anyway.
5009 *
5010 * dt: time from mark until now
5011 * db: blocks written from mark until now
5012 * rt: remaining time
5013 */
5014 dt = ((jiffies - mddev->resync_mark) / HZ);
5015 if (!dt) dt++;
ff4e8d9a
N		 5016 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5017 - mddev->resync_mark_cnt;
5018 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
1da177e4
LT		 5019
5020 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5021
ff4e8d9a 5022 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 5023}
5024
5025static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5026{
5027 struct list_head *tmp;
5028 loff_t l = *pos;
5029 mddev_t *mddev;
5030
5031 if (l >= 0x10000)
5032 return NULL;
5033 if (!l--)
5034 /* header */
5035 return (void*)1;
5036
5037 spin_lock(&all_mddevs_lock);
5038 list_for_each(tmp,&all_mddevs)
5039 if (!l--) {
5040 mddev = list_entry(tmp, mddev_t, all_mddevs);
5041 mddev_get(mddev);
5042 spin_unlock(&all_mddevs_lock);
5043 return mddev;
5044 }
5045 spin_unlock(&all_mddevs_lock);
5046 if (!l--)
5047 return (void*)2;/* tail */
5048 return NULL;
5049}
5050
5051static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5052{
5053 struct list_head *tmp;
5054 mddev_t *next_mddev, *mddev = v;
5055
5056 ++*pos;
5057 if (v == (void*)2)
5058 return NULL;
5059
5060 spin_lock(&all_mddevs_lock);
5061 if (v == (void*)1)
5062 tmp = all_mddevs.next;
5063 else
5064 tmp = mddev->all_mddevs.next;
5065 if (tmp != &all_mddevs)
5066 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5067 else {
5068 next_mddev = (void*)2;
5069 *pos = 0x10000;
5070 }
5071 spin_unlock(&all_mddevs_lock);
5072
5073 if (v != (void*)1)
5074 mddev_put(mddev);
5075 return next_mddev;
5076
5077}
5078
5079static void md_seq_stop(struct seq_file *seq, void *v)
5080{
5081 mddev_t *mddev = v;
5082
5083 if (mddev && v != (void*)1 && v != (void*)2)
5084 mddev_put(mddev);
5085}
5086
d7603b7e
N		 5087struct mdstat_info {
5088 int event;
5089};
5090
1da177e4
LT		 5091static int md_seq_show(struct seq_file *seq, void *v)
5092{
5093 mddev_t *mddev = v;
5094 sector_t size;
5095 struct list_head *tmp2;
5096 mdk_rdev_t *rdev;
d7603b7e 5097 struct mdstat_info *mi = seq->private;
32a7627c 5098 struct bitmap *bitmap;
1da177e4
LT		 5099
5100 if (v == (void*)1) {
2604b703 5101 struct mdk_personality *pers;
1da177e4
LT		 5102 seq_printf(seq, "Personalities : ");
5103 spin_lock(&pers_lock);
2604b703
N		 5104 list_for_each_entry(pers, &pers_list, list)
5105 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 5106
5107 spin_unlock(&pers_lock);
5108 seq_printf(seq, "\n");
d7603b7e 5109 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 5110 return 0;
5111 }
5112 if (v == (void*)2) {
5113 status_unused(seq);
5114 return 0;
5115 }
5116
5dc5cf7d 5117 if (mddev_lock(mddev) < 0)
1da177e4 5118 return -EINTR;
5dc5cf7d 5119
1da177e4
LT		 5120 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5121 seq_printf(seq, "%s : %sactive", mdname(mddev),
5122 mddev->pers ? "" : "in");
5123 if (mddev->pers) {
f91de92e 5124 if (mddev->ro==1)
1da177e4 5125 seq_printf(seq, " (read-only)");
f91de92e
N		 5126 if (mddev->ro==2)
5127 seq_printf(seq, "(auto-read-only)");
1da177e4
LT		 5128 seq_printf(seq, " %s", mddev->pers->name);
5129 }
5130
5131 size = 0;
5132 ITERATE_RDEV(mddev,rdev,tmp2) {
5133 char b[BDEVNAME_SIZE];
5134 seq_printf(seq, " %s[%d]",
5135 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 5136 if (test_bit(WriteMostly, &rdev->flags))
5137 seq_printf(seq, "(W)");
b2d444d7 5138 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5139 seq_printf(seq, "(F)");
5140 continue;
b325a32e
N		 5141 } else if (rdev->raid_disk < 0)
5142 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 5143 size += rdev->size;
5144 }
5145
5146 if (!list_empty(&mddev->disks)) {
5147 if (mddev->pers)
5148 seq_printf(seq, "\n %llu blocks",
5149 (unsigned long long)mddev->array_size);
5150 else
5151 seq_printf(seq, "\n %llu blocks",
5152 (unsigned long long)size);
5153 }
1cd6bf19
N		 5154 if (mddev->persistent) {
5155 if (mddev->major_version != 0 ||
5156 mddev->minor_version != 90) {
5157 seq_printf(seq," super %d.%d",
5158 mddev->major_version,
5159 mddev->minor_version);
5160 }
e691063a
N		 5161 } else if (mddev->external)
5162 seq_printf(seq, " super external:%s",
5163 mddev->metadata_type);
5164 else
1cd6bf19 5165 seq_printf(seq, " super non-persistent");
1da177e4
LT		 5166
5167 if (mddev->pers) {
5168 mddev->pers->status (seq, mddev);
5169 seq_printf(seq, "\n ");
8e1b39d6
N		 5170 if (mddev->pers->sync_request) {
5171 if (mddev->curr_resync > 2) {
5172 status_resync (seq, mddev);
5173 seq_printf(seq, "\n ");
5174 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5175 seq_printf(seq, "\tresync=DELAYED\n ");
5176 else if (mddev->recovery_cp < MaxSector)
5177 seq_printf(seq, "\tresync=PENDING\n ");
5178 }
32a7627c
N		 5179 } else
5180 seq_printf(seq, "\n ");
5181
5182 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 5183 unsigned long chunk_kb;
5184 unsigned long flags;
32a7627c
N		 5185 spin_lock_irqsave(&bitmap->lock, flags);
5186 chunk_kb = bitmap->chunksize >> 10;
5187 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5188 "%lu%s chunk",
5189 bitmap->pages - bitmap->missing_pages,
5190 bitmap->pages,
5191 (bitmap->pages - bitmap->missing_pages)
5192 << (PAGE_SHIFT - 10),
5193 chunk_kb ? chunk_kb : bitmap->chunksize,
5194 chunk_kb ? "KB" : "B");
78d742d8
N		 5195 if (bitmap->file) {
5196 seq_printf(seq, ", file: ");
c649bb9c
JS		 5197 seq_path(seq, bitmap->file->f_path.mnt,
5198 bitmap->file->f_path.dentry," \t\n");
32a7627c 5199 }
78d742d8 5200
32a7627c
N		 5201 seq_printf(seq, "\n");
5202 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 5203 }
5204
5205 seq_printf(seq, "\n");
5206 }
5207 mddev_unlock(mddev);
5208
5209 return 0;
5210}
5211
5212static struct seq_operations md_seq_ops = {
5213 .start = md_seq_start,
5214 .next = md_seq_next,
5215 .stop = md_seq_stop,
5216 .show = md_seq_show,
5217};
5218
5219static int md_seq_open(struct inode *inode, struct file *file)
5220{
5221 int error;
d7603b7e
N		 5222 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5223 if (mi == NULL)
5224 return -ENOMEM;
1da177e4
LT		 5225
5226 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 5227 if (error)
5228 kfree(mi);
5229 else {
5230 struct seq_file *p = file->private_data;
5231 p->private = mi;
5232 mi->event = atomic_read(&md_event_count);
5233 }
1da177e4
LT		 5234 return error;
5235}
5236
d7603b7e
N		 5237static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5238{
5239 struct seq_file *m = filp->private_data;
5240 struct mdstat_info *mi = m->private;
5241 int mask;
5242
5243 poll_wait(filp, &md_event_waiters, wait);
5244
5245 /* always allow read */
5246 mask = POLLIN | POLLRDNORM;
5247
5248 if (mi->event != atomic_read(&md_event_count))
5249 mask |= POLLERR | POLLPRI;
5250 return mask;
5251}
5252
fa027c2a 5253static const struct file_operations md_seq_fops = {
e24650c2 5254 .owner = THIS_MODULE,
1da177e4
LT		 5255 .open = md_seq_open,
5256 .read = seq_read,
5257 .llseek = seq_lseek,
c3f94b40 5258 .release = seq_release_private,
d7603b7e 5259 .poll = mdstat_poll,
1da177e4
LT		 5260};
5261
2604b703 5262int register_md_personality(struct mdk_personality *p)
1da177e4 5263{
1da177e4 5264 spin_lock(&pers_lock);
2604b703
N		 5265 list_add_tail(&p->list, &pers_list);
5266 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 5267 spin_unlock(&pers_lock);
5268 return 0;
5269}
5270
2604b703 5271int unregister_md_personality(struct mdk_personality *p)
1da177e4 5272{
2604b703 5273 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 5274 spin_lock(&pers_lock);
2604b703 5275 list_del_init(&p->list);
1da177e4
LT		 5276 spin_unlock(&pers_lock);
5277 return 0;
5278}
5279
5280static int is_mddev_idle(mddev_t *mddev)
5281{
5282 mdk_rdev_t * rdev;
5283 struct list_head *tmp;
5284 int idle;
713f6ab1 5285 long curr_events;
1da177e4
LT		 5286
5287 idle = 1;
5288 ITERATE_RDEV(mddev,rdev,tmp) {
5289 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 5290 curr_events = disk_stat_read(disk, sectors[0]) +
5291 disk_stat_read(disk, sectors[1]) -
1da177e4 5292 atomic_read(&disk->sync_io);
713f6ab1
N		 5293 /* sync IO will cause sync_io to increase before the disk_stats
5294 * as sync_io is counted when a request starts, and
5295 * disk_stats is counted when it completes.
5296 * So resync activity will cause curr_events to be smaller than
5297 * when there was no such activity.
5298 * non-sync IO will cause disk_stat to increase without
5299 * increasing sync_io so curr_events will (eventually)
5300 * be larger than it was before. Once it becomes
5301 * substantially larger, the test below will cause
5302 * the array to appear non-idle, and resync will slow
5303 * down.
5304 * If there is a lot of outstanding resync activity when
5305 * we set last_event to curr_events, then all that activity
5306 * completing might cause the array to appear non-idle
5307 * and resync will be slowed down even though there might
5308 * not have been non-resync activity. This will only
5309 * happen once though. 'last_events' will soon reflect
5310 * the state where there is little or no outstanding
5311 * resync requests, and further resync activity will
5312 * always make curr_events less than last_events.
c0e48521 5313 *
1da177e4 5314 */
713f6ab1 5315 if (curr_events - rdev->last_events > 4096) {
1da177e4
LT		 5316 rdev->last_events = curr_events;
5317 idle = 0;
5318 }
5319 }
5320 return idle;
5321}
5322
5323void md_done_sync(mddev_t *mddev, int blocks, int ok)
5324{
5325 /* another "blocks" (512byte) blocks have been synced */
5326 atomic_sub(blocks, &mddev->recovery_active);
5327 wake_up(&mddev->recovery_wait);
5328 if (!ok) {
5329 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5330 md_wakeup_thread(mddev->thread);
5331 // stop recovery, signal do_sync ....
5332 }
5333}
5334
5335
06d91a5f
N		 5336/* md_write_start(mddev, bi)
5337 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 5338 * in superblock) before writing, schedule a superblock update
5339 * and wait for it to complete.
06d91a5f 5340 */
3d310eb7 5341void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 5342{
06d91a5f 5343 if (bio_data_dir(bi) != WRITE)
3d310eb7 5344 return;
06d91a5f 5345
f91de92e
N		 5346 BUG_ON(mddev->ro == 1);
5347 if (mddev->ro == 2) {
5348 /* need to switch to read/write */
5349 mddev->ro = 0;
5350 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5351 md_wakeup_thread(mddev->thread);
5352 }
06d91a5f 5353 atomic_inc(&mddev->writes_pending);
06d91a5f 5354 if (mddev->in_sync) {
a9701a30 5355 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 5356 if (mddev->in_sync) {
5357 mddev->in_sync = 0;
850b2b42 5358 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3d310eb7
N		 5359 md_wakeup_thread(mddev->thread);
5360 }
a9701a30 5361 spin_unlock_irq(&mddev->write_lock);
06d91a5f 5362 }
850b2b42 5363 wait_event(mddev->sb_wait, mddev->flags==0);
1da177e4
LT		 5364}
5365
5366void md_write_end(mddev_t *mddev)
5367{
5368 if (atomic_dec_and_test(&mddev->writes_pending)) {
5369 if (mddev->safemode == 2)
5370 md_wakeup_thread(mddev->thread);
16f17b39 5371 else if (mddev->safemode_delay)
1da177e4
LT		 5372 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5373 }
5374}
5375
2a2275d6
N		 5376/* md_allow_write(mddev)
5377 * Calling this ensures that the array is marked 'active' so that writes
5378 * may proceed without blocking. It is important to call this before
5379 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5380 * Must be called with mddev_lock held.
5381 */
5382void md_allow_write(mddev_t *mddev)
5383{
5384 if (!mddev->pers)
5385 return;
5386 if (mddev->ro)
5387 return;
5388
5389 spin_lock_irq(&mddev->write_lock);
5390 if (mddev->in_sync) {
5391 mddev->in_sync = 0;
5392 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5393 if (mddev->safemode_delay &&
5394 mddev->safemode == 0)
5395 mddev->safemode = 1;
5396 spin_unlock_irq(&mddev->write_lock);
5397 md_update_sb(mddev, 0);
5398 } else
5399 spin_unlock_irq(&mddev->write_lock);
5400}
5401EXPORT_SYMBOL_GPL(md_allow_write);
5402
75c96f85 5403static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 5404
5405#define SYNC_MARKS 10
5406#define SYNC_MARK_STEP (3*HZ)
29269553 5407void md_do_sync(mddev_t *mddev)
1da177e4
LT		 5408{
5409 mddev_t *mddev2;
5410 unsigned int currspeed = 0,
5411 window;
57afd89f 5412 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 5413 unsigned long mark[SYNC_MARKS];
5414 sector_t mark_cnt[SYNC_MARKS];
5415 int last_mark,m;
5416 struct list_head *tmp;
5417 sector_t last_check;
57afd89f 5418 int skipped = 0;
5fd6c1dc
N		 5419 struct list_head *rtmp;
5420 mdk_rdev_t *rdev;
61df9d91 5421 char *desc;
1da177e4
LT		 5422
5423 /* just incase thread restarts... */
5424 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5425 return;
5fd6c1dc
N		 5426 if (mddev->ro) /* never try to sync a read-only array */
5427 return;
1da177e4 5428
61df9d91
N		 5429 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5430 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5431 desc = "data-check";
5432 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5433 desc = "requested-resync";
5434 else
5435 desc = "resync";
5436 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5437 desc = "reshape";
5438 else
5439 desc = "recovery";
5440
1da177e4
LT		 5441 /* we overload curr_resync somewhat here.
5442 * 0 == not engaged in resync at all
5443 * 2 == checking that there is no conflict with another sync
5444 * 1 == like 2, but have yielded to allow conflicting resync to
5445 * commense
5446 * other == active in resync - this many blocks
5447 *
5448 * Before starting a resync we must have set curr_resync to
5449 * 2, and then checked that every "conflicting" array has curr_resync
5450 * less than ours. When we find one that is the same or higher
5451 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5452 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5453 * This will mean we have to start checking from the beginning again.
5454 *
5455 */
5456
5457 do {
5458 mddev->curr_resync = 2;
5459
5460 try_again:
787453c2 5461 if (kthread_should_stop()) {
6985c43f 5462 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 5463 goto skip;
5464 }
5465 ITERATE_MDDEV(mddev2,tmp) {
1da177e4
LT		 5466 if (mddev2 == mddev)
5467 continue;
5468 if (mddev2->curr_resync &&
5469 match_mddev_units(mddev,mddev2)) {
5470 DEFINE_WAIT(wq);
5471 if (mddev < mddev2 && mddev->curr_resync == 2) {
5472 /* arbitrarily yield */
5473 mddev->curr_resync = 1;
5474 wake_up(&resync_wait);
5475 }
5476 if (mddev > mddev2 && mddev->curr_resync == 1)
5477 /* no need to wait here, we can wait the next
5478 * time 'round when curr_resync == 2
5479 */
5480 continue;
787453c2
N		 5481 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5482 if (!kthread_should_stop() &&
8712e553 5483 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 5484 printk(KERN_INFO "md: delaying %s of %s"
5485 " until %s has finished (they"
1da177e4 5486 " share one or more physical units)\n",
61df9d91 5487 desc, mdname(mddev), mdname(mddev2));
1da177e4
LT		 5488 mddev_put(mddev2);
5489 schedule();
5490 finish_wait(&resync_wait, &wq);
5491 goto try_again;
5492 }
5493 finish_wait(&resync_wait, &wq);
5494 }
5495 }
5496 } while (mddev->curr_resync < 2);
5497
5fd6c1dc 5498 j = 0;
9d88883e 5499 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 5500 /* resync follows the size requested by the personality,
57afd89f 5501 * which defaults to physical size, but can be virtual size
1da177e4
LT		 5502 */
5503 max_sectors = mddev->resync_max_sectors;
9d88883e 5504 mddev->resync_mismatches = 0;
5fd6c1dc
N		 5505 /* we don't use the checkpoint if there's a bitmap */
5506 if (!mddev->bitmap &&
5507 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5508 j = mddev->recovery_cp;
ccfcc3c1
N		 5509 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5510 max_sectors = mddev->size << 1;
5fd6c1dc 5511 else {
1da177e4
LT		 5512 /* recovery follows the physical size of devices */
5513 max_sectors = mddev->size << 1;
5fd6c1dc
N		 5514 j = MaxSector;
5515 ITERATE_RDEV(mddev,rdev,rtmp)
5516 if (rdev->raid_disk >= 0 &&
5517 !test_bit(Faulty, &rdev->flags) &&
5518 !test_bit(In_sync, &rdev->flags) &&
5519 rdev->recovery_offset < j)
5520 j = rdev->recovery_offset;
5521 }
1da177e4 5522
61df9d91
N		 5523 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5524 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5525 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 5526 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 5527 "(but not more than %d KB/sec) for %s.\n",
5528 speed_max(mddev), desc);
1da177e4
LT		 5529
5530 is_mddev_idle(mddev); /* this also initializes IO event counters */
5fd6c1dc 5531
57afd89f 5532 io_sectors = 0;
1da177e4
LT		 5533 for (m = 0; m < SYNC_MARKS; m++) {
5534 mark[m] = jiffies;
57afd89f 5535 mark_cnt[m] = io_sectors;
1da177e4
LT		 5536 }
5537 last_mark = 0;
5538 mddev->resync_mark = mark[last_mark];
5539 mddev->resync_mark_cnt = mark_cnt[last_mark];
5540
5541 /*
5542 * Tune reconstruction:
5543 */
5544 window = 32*(PAGE_SIZE/512);
5545 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5546 window/2,(unsigned long long) max_sectors/2);
5547
5548 atomic_set(&mddev->recovery_active, 0);
5549 init_waitqueue_head(&mddev->recovery_wait);
5550 last_check = 0;
5551
5552 if (j>2) {
5553 printk(KERN_INFO
61df9d91
N		 5554 "md: resuming %s of %s from checkpoint.\n",
5555 desc, mdname(mddev));
1da177e4
LT		 5556 mddev->curr_resync = j;
5557 }
5558
5559 while (j < max_sectors) {
57afd89f 5560 sector_t sectors;
1da177e4 5561
57afd89f 5562 skipped = 0;
c6207277
N		 5563 if (j >= mddev->resync_max) {
5564 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5565 wait_event(mddev->recovery_wait,
5566 mddev->resync_max > j
5567 || kthread_should_stop());
5568 }
5569 if (kthread_should_stop())
5570 goto interrupted;
57afd89f 5571 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 5572 currspeed < speed_min(mddev));
57afd89f 5573 if (sectors == 0) {
1da177e4
LT		 5574 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5575 goto out;
5576 }
57afd89f
N		 5577
5578 if (!skipped) { /* actual IO requested */
5579 io_sectors += sectors;
5580 atomic_add(sectors, &mddev->recovery_active);
5581 }
5582
1da177e4
LT		 5583 j += sectors;
5584 if (j>1) mddev->curr_resync = j;
ff4e8d9a 5585 mddev->curr_mark_cnt = io_sectors;
d7603b7e
N		 5586 if (last_check == 0)
5587 /* this is the earliers that rebuilt will be
5588 * visible in /proc/mdstat
5589 */
5590 md_new_event(mddev);
57afd89f
N		 5591
5592 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 5593 continue;
5594
57afd89f 5595 last_check = io_sectors;
1da177e4
LT		 5596
5597 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5598 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5599 break;
5600
5601 repeat:
5602 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5603 /* step marks */
5604 int next = (last_mark+1) % SYNC_MARKS;
5605
5606 mddev->resync_mark = mark[next];
5607 mddev->resync_mark_cnt = mark_cnt[next];
5608 mark[next] = jiffies;
57afd89f 5609 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 5610 last_mark = next;
5611 }
5612
5613
c6207277
N		 5614 if (kthread_should_stop())
5615 goto interrupted;
5616
1da177e4
LT		 5617
5618 /*
5619 * this loop exits only if either when we are slower than
5620 * the 'hard' speed limit, or the system was IO-idle for
5621 * a jiffy.
5622 * the system might be non-idle CPU-wise, but we only care
5623 * about not overloading the IO subsystem. (things like an
5624 * e2fsck being done on the RAID array should execute fast)
5625 */
2ad8b1ef 5626 blk_unplug(mddev->queue);
1da177e4
LT		 5627 cond_resched();
5628
57afd89f
N		 5629 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5630 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 5631
88202a0c
N		 5632 if (currspeed > speed_min(mddev)) {
5633 if ((currspeed > speed_max(mddev)) ||
1da177e4 5634 !is_mddev_idle(mddev)) {
c0e48521 5635 msleep(500);
1da177e4
LT		 5636 goto repeat;
5637 }
5638 }
5639 }
61df9d91 5640 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 5641 /*
5642 * this also signals 'finished resyncing' to md_stop
5643 */
5644 out:
2ad8b1ef 5645 blk_unplug(mddev->queue);
1da177e4
LT		 5646
5647 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5648
5649 /* tell personality that we are finished */
57afd89f 5650 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 5651
5652 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
ccfcc3c1 5653 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 5654 mddev->curr_resync > 2) {
5655 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5656 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5657 if (mddev->curr_resync >= mddev->recovery_cp) {
5658 printk(KERN_INFO
61df9d91
N		 5659 "md: checkpointing %s of %s.\n",
5660 desc, mdname(mddev));
5fd6c1dc
N		 5661 mddev->recovery_cp = mddev->curr_resync;
5662 }
5663 } else
5664 mddev->recovery_cp = MaxSector;
5665 } else {
5666 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5667 mddev->curr_resync = MaxSector;
5668 ITERATE_RDEV(mddev,rdev,rtmp)
5669 if (rdev->raid_disk >= 0 &&
5670 !test_bit(Faulty, &rdev->flags) &&
5671 !test_bit(In_sync, &rdev->flags) &&
5672 rdev->recovery_offset < mddev->curr_resync)
5673 rdev->recovery_offset = mddev->curr_resync;
5fd6c1dc 5674 }
1da177e4 5675 }
17571284 5676 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5677
1da177e4
LT		 5678 skip:
5679 mddev->curr_resync = 0;
c6207277
N		 5680 mddev->resync_max = MaxSector;
5681 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
1da177e4
LT		 5682 wake_up(&resync_wait);
5683 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5684 md_wakeup_thread(mddev->thread);
c6207277
N		 5685 return;
5686
5687 interrupted:
5688 /*
5689 * got a signal, exit.
5690 */
5691 printk(KERN_INFO
5692 "md: md_do_sync() got signal ... exiting\n");
5693 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5694 goto out;
5695
1da177e4 5696}
29269553 5697EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 5698
5699
b4c4c7b8
N		 5700static int remove_and_add_spares(mddev_t *mddev)
5701{
5702 mdk_rdev_t *rdev;
5703 struct list_head *rtmp;
5704 int spares = 0;
5705
5706 ITERATE_RDEV(mddev,rdev,rtmp)
5707 if (rdev->raid_disk >= 0 &&
c303da6d 5708 !mddev->external &&
b4c4c7b8
N		 5709 (test_bit(Faulty, &rdev->flags) ||
5710 ! test_bit(In_sync, &rdev->flags)) &&
5711 atomic_read(&rdev->nr_pending)==0) {
5712 if (mddev->pers->hot_remove_disk(
5713 mddev, rdev->raid_disk)==0) {
5714 char nm[20];
5715 sprintf(nm,"rd%d", rdev->raid_disk);
5716 sysfs_remove_link(&mddev->kobj, nm);
5717 rdev->raid_disk = -1;
5718 }
5719 }
5720
5721 if (mddev->degraded) {
5722 ITERATE_RDEV(mddev,rdev,rtmp)
5723 if (rdev->raid_disk < 0
5724 && !test_bit(Faulty, &rdev->flags)) {
5725 rdev->recovery_offset = 0;
5726 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5727 char nm[20];
5728 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 5729 if (sysfs_create_link(&mddev->kobj,
5730 &rdev->kobj, nm))
5731 printk(KERN_WARNING
5732 "md: cannot register "
5733 "%s for %s\n",
5734 nm, mdname(mddev));
b4c4c7b8
N		 5735 spares++;
5736 md_new_event(mddev);
5737 } else
5738 break;
5739 }
5740 }
5741 return spares;
5742}
1da177e4
LT		 5743/*
5744 * This routine is regularly called by all per-raid-array threads to
5745 * deal with generic issues like resync and super-block update.
5746 * Raid personalities that don't have a thread (linear/raid0) do not
5747 * need this as they never do any recovery or update the superblock.
5748 *
5749 * It does not do any resync itself, but rather "forks" off other threads
5750 * to do that as needed.
5751 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5752 * "->recovery" and create a thread at ->sync_thread.
5753 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5754 * and wakeups up this thread which will reap the thread and finish up.
5755 * This thread also removes any faulty devices (with nr_pending == 0).
5756 *
5757 * The overall approach is:
5758 * 1/ if the superblock needs updating, update it.
5759 * 2/ If a recovery thread is running, don't do anything else.
5760 * 3/ If recovery has finished, clean up, possibly marking spares active.
5761 * 4/ If there are any faulty devices, remove them.
5762 * 5/ If array is degraded, try to add spares devices
5763 * 6/ If array has spares or is not in-sync, start a resync thread.
5764 */
5765void md_check_recovery(mddev_t *mddev)
5766{
5767 mdk_rdev_t *rdev;
5768 struct list_head *rtmp;
5769
5770
5f40402d
N		 5771 if (mddev->bitmap)
5772 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 5773
5774 if (mddev->ro)
5775 return;
fca4d848
N		 5776
5777 if (signal_pending(current)) {
5778 if (mddev->pers->sync_request) {
5779 printk(KERN_INFO "md: %s in immediate safe mode\n",
5780 mdname(mddev));
5781 mddev->safemode = 2;
5782 }
5783 flush_signals(current);
5784 }
5785
1da177e4 5786 if ( ! (
e691063a 5787 (mddev->flags && !mddev->external) ||
1da177e4 5788 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 5789 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5790 (mddev->safemode == 1) ||
5791 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5792 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 5793 ))
5794 return;
fca4d848 5795
df5b89b3 5796 if (mddev_trylock(mddev)) {
b4c4c7b8 5797 int spares = 0;
fca4d848 5798
a9701a30 5799 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 5800 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5801 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5802 mddev->in_sync = 1;
e691063a
N		 5803 if (mddev->persistent)
5804 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
fca4d848
N		 5805 }
5806 if (mddev->safemode == 1)
5807 mddev->safemode = 0;
a9701a30 5808 spin_unlock_irq(&mddev->write_lock);
fca4d848 5809
850b2b42
N		 5810 if (mddev->flags)
5811 md_update_sb(mddev, 0);
06d91a5f 5812
06d91a5f 5813
1da177e4
LT		 5814 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5815 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5816 /* resync/recovery still happening */
5817 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5818 goto unlock;
5819 }
5820 if (mddev->sync_thread) {
5821 /* resync has finished, collect result */
5822 md_unregister_thread(mddev->sync_thread);
5823 mddev->sync_thread = NULL;
5824 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5825 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5826 /* success...*/
5827 /* activate any spares */
5828 mddev->pers->spare_active(mddev);
5829 }
850b2b42 5830 md_update_sb(mddev, 1);
41158c7e
N		 5831
5832 /* if array is no-longer degraded, then any saved_raid_disk
5833 * information must be scrapped
5834 */
5835 if (!mddev->degraded)
5836 ITERATE_RDEV(mddev,rdev,rtmp)
5837 rdev->saved_raid_disk = -1;
5838
1da177e4
LT		 5839 mddev->recovery = 0;
5840 /* flag recovery needed just to double check */
5841 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d7603b7e 5842 md_new_event(mddev);
1da177e4
LT		 5843 goto unlock;
5844 }
24dd469d
N		 5845 /* Clear some bits that don't mean anything, but
5846 * might be left set
5847 */
5848 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5849 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5850 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5851 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 5852
5fd6c1dc
N		 5853 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5854 goto unlock;
1da177e4
LT		 5855 /* no recovery is running.
5856 * remove any failed drives, then
5857 * add spares if possible.
5858 * Spare are also removed and re-added, to allow
5859 * the personality to fail the re-add.
5860 */
1da177e4 5861
b4c4c7b8
N		 5862 if (mddev->reshape_position != MaxSector) {
5863 if (mddev->pers->check_reshape(mddev) != 0)
5864 /* Cannot proceed */
5865 goto unlock;
5866 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5867 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 5868 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5869 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5870 } else if (mddev->recovery_cp < MaxSector) {
5871 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5872 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5873 /* nothing to be done ... */
1da177e4 5874 goto unlock;
24dd469d 5875
1da177e4
LT		 5876 if (mddev->pers->sync_request) {
5877 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 5878 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5879 /* We are adding a device or devices to an array
5880 * which has the bitmap stored on all devices.
5881 * So make sure all bitmap pages get written
5882 */
5883 bitmap_write_all(mddev->bitmap);
5884 }
1da177e4
LT		 5885 mddev->sync_thread = md_register_thread(md_do_sync,
5886 mddev,
5887 "%s_resync");
5888 if (!mddev->sync_thread) {
5889 printk(KERN_ERR "%s: could not start resync"
5890 " thread...\n",
5891 mdname(mddev));
5892 /* leave the spares where they are, it shouldn't hurt */
5893 mddev->recovery = 0;
d7603b7e 5894 } else
1da177e4 5895 md_wakeup_thread(mddev->sync_thread);
d7603b7e 5896 md_new_event(mddev);
1da177e4
LT		 5897 }
5898 unlock:
5899 mddev_unlock(mddev);
5900 }
5901}
5902
75c96f85
AB		 5903static int md_notify_reboot(struct notifier_block *this,
5904 unsigned long code, void *x)
1da177e4
LT		 5905{
5906 struct list_head *tmp;
5907 mddev_t *mddev;
5908
5909 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5910
5911 printk(KERN_INFO "md: stopping all md devices.\n");
5912
5913 ITERATE_MDDEV(mddev,tmp)
c71d4887 5914 if (mddev_trylock(mddev)) {
1da177e4 5915 do_md_stop (mddev, 1);
c71d4887
NB		 5916 mddev_unlock(mddev);
5917 }
1da177e4
LT		 5918 /*
5919 * certain more exotic SCSI devices are known to be
5920 * volatile wrt too early system reboots. While the
5921 * right place to handle this issue is the given
5922 * driver, we do want to have a safe RAID driver ...
5923 */
5924 mdelay(1000*1);
5925 }
5926 return NOTIFY_DONE;
5927}
5928
75c96f85 5929static struct notifier_block md_notifier = {
1da177e4
LT		 5930 .notifier_call = md_notify_reboot,
5931 .next = NULL,
5932 .priority = INT_MAX, /* before any real devices */
5933};
5934
5935static void md_geninit(void)
5936{
5937 struct proc_dir_entry *p;
5938
5939 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5940
5941 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5942 if (p)
5943 p->proc_fops = &md_seq_fops;
5944}
5945
75c96f85 5946static int __init md_init(void)
1da177e4 5947{
1da177e4
LT		 5948 if (register_blkdev(MAJOR_NR, "md"))
5949 return -1;
5950 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5951 unregister_blkdev(MAJOR_NR, "md");
5952 return -1;
5953 }
e8703fe1
N		 5954 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5955 md_probe, NULL, NULL);
5956 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 5957 md_probe, NULL, NULL);
5958
1da177e4 5959 register_reboot_notifier(&md_notifier);
0b4d4147 5960 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 5961
5962 md_geninit();
5963 return (0);
5964}
5965
5966
5967#ifndef MODULE
5968
5969/*
5970 * Searches all registered partitions for autorun RAID arrays
5971 * at boot time.
5972 */
4d936ec1
ME		 5973
5974static LIST_HEAD(all_detected_devices);
5975struct detected_devices_node {
5976 struct list_head list;
5977 dev_t dev;
5978};
1da177e4
LT		 5979
5980void md_autodetect_dev(dev_t dev)
5981{
4d936ec1
ME		 5982 struct detected_devices_node *node_detected_dev;
5983
5984 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
5985 if (node_detected_dev) {
5986 node_detected_dev->dev = dev;
5987 list_add_tail(&node_detected_dev->list, &all_detected_devices);
5988 } else {
5989 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
5990 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
5991 }
1da177e4
LT		 5992}
5993
5994
5995static void autostart_arrays(int part)
5996{
5997 mdk_rdev_t *rdev;
4d936ec1
ME		 5998 struct detected_devices_node *node_detected_dev;
5999 dev_t dev;
6000 int i_scanned, i_passed;
1da177e4 6001
4d936ec1
ME		 6002 i_scanned = 0;
6003 i_passed = 0;
1da177e4 6004
4d936ec1 6005 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 6006
4d936ec1
ME		 6007 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6008 i_scanned++;
6009 node_detected_dev = list_entry(all_detected_devices.next,
6010 struct detected_devices_node, list);
6011 list_del(&node_detected_dev->list);
6012 dev = node_detected_dev->dev;
6013 kfree(node_detected_dev);
df968c4e 6014 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 6015 if (IS_ERR(rdev))
6016 continue;
6017
b2d444d7 6018 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6019 MD_BUG();
6020 continue;
6021 }
6022 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 6023 i_passed++;
1da177e4 6024 }
4d936ec1
ME		 6025
6026 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6027 i_scanned, i_passed);
1da177e4
LT		 6028
6029 autorun_devices(part);
6030}
6031
fdee8ae4 6032#endif /* !MODULE */
1da177e4
LT		 6033
6034static __exit void md_exit(void)
6035{
6036 mddev_t *mddev;
6037 struct list_head *tmp;
8ab5e4c1 6038
e8703fe1
N		 6039 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6040 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4
LT		 6041
6042 unregister_blkdev(MAJOR_NR,"md");
6043 unregister_blkdev(mdp_major, "mdp");
6044 unregister_reboot_notifier(&md_notifier);
6045 unregister_sysctl_table(raid_table_header);
6046 remove_proc_entry("mdstat", NULL);
6047 ITERATE_MDDEV(mddev,tmp) {
6048 struct gendisk *disk = mddev->gendisk;
6049 if (!disk)
6050 continue;
6051 export_array(mddev);
6052 del_gendisk(disk);
6053 put_disk(disk);
6054 mddev->gendisk = NULL;
6055 mddev_put(mddev);
6056 }
6057}
6058
685784aa 6059subsys_initcall(md_init);
1da177e4
LT		 6060module_exit(md_exit)
6061
f91de92e
N		 6062static int get_ro(char *buffer, struct kernel_param *kp)
6063{
6064 return sprintf(buffer, "%d", start_readonly);
6065}
6066static int set_ro(const char *val, struct kernel_param *kp)
6067{
6068 char *e;
6069 int num = simple_strtoul(val, &e, 10);
6070 if (*val && (*e == '\0' || *e == '\n')) {
6071 start_readonly = num;
4dbcdc75 6072 return 0;
f91de92e
N		 6073 }
6074 return -EINVAL;
6075}
6076
80ca3a44
N		 6077module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6078module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 6079
f91de92e 6080
1da177e4
LT		 6081EXPORT_SYMBOL(register_md_personality);
6082EXPORT_SYMBOL(unregister_md_personality);
6083EXPORT_SYMBOL(md_error);
6084EXPORT_SYMBOL(md_done_sync);
6085EXPORT_SYMBOL(md_write_start);
6086EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 6087EXPORT_SYMBOL(md_register_thread);
6088EXPORT_SYMBOL(md_unregister_thread);
6089EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 6090EXPORT_SYMBOL(md_check_recovery);
6091MODULE_LICENSE("GPL");
aa1595e9 6092MODULE_ALIAS("md");
72008652 6093MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)