projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] md: Add '4' to the list of levels for which bitmaps are supported
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
35#include <linux/module.h>
36#include <linux/config.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
LT		 41#include <linux/sysctl.h>
42#include <linux/devfs_fs_kernel.h>
43#include <linux/buffer_head.h> /* for invalidate_bdev */
44#include <linux/suspend.h>
d7603b7e 45#include <linux/poll.h>
1da177e4
LT		 46
47#include <linux/init.h>
48
32a7627c
N		 49#include <linux/file.h>
50
1da177e4
LT		 51#ifdef CONFIG_KMOD
52#include <linux/kmod.h>
53#endif
54
55#include <asm/unaligned.h>
56
57#define MAJOR_NR MD_MAJOR
58#define MD_DRIVER
59
60/* 63 partitions with the alternate major number (mdp) */
61#define MdpMinorShift 6
62
63#define DEBUG 0
64#define dprintk(x...) ((void)(DEBUG && printk(x)))
65
66
67#ifndef MODULE
68static void autostart_arrays (int part);
69#endif
70
2604b703 71static LIST_HEAD(pers_list);
1da177e4
LT		 72static DEFINE_SPINLOCK(pers_lock);
73
74/*
75 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
76 * is 1000 KB/sec, so the extra system load does not show up that much.
77 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 78 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 79 * subsystem is idle. There is also an 'absolute maximum' reconstruction
80 * speed limit - in case reconstruction slows down your system despite
81 * idle IO detection.
82 *
83 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 84 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 85 */
86
87static int sysctl_speed_limit_min = 1000;
88static int sysctl_speed_limit_max = 200000;
88202a0c
N		 89static inline int speed_min(mddev_t *mddev)
90{
91 return mddev->sync_speed_min ?
92 mddev->sync_speed_min : sysctl_speed_limit_min;
93}
94
95static inline int speed_max(mddev_t *mddev)
96{
97 return mddev->sync_speed_max ?
98 mddev->sync_speed_max : sysctl_speed_limit_max;
99}
1da177e4
LT		 100
101static struct ctl_table_header *raid_table_header;
102
103static ctl_table raid_table[] = {
104 {
105 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
106 .procname = "speed_limit_min",
107 .data = &sysctl_speed_limit_min,
108 .maxlen = sizeof(int),
109 .mode = 0644,
110 .proc_handler = &proc_dointvec,
111 },
112 {
113 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
114 .procname = "speed_limit_max",
115 .data = &sysctl_speed_limit_max,
116 .maxlen = sizeof(int),
117 .mode = 0644,
118 .proc_handler = &proc_dointvec,
119 },
120 { .ctl_name = 0 }
121};
122
123static ctl_table raid_dir_table[] = {
124 {
125 .ctl_name = DEV_RAID,
126 .procname = "raid",
127 .maxlen = 0,
128 .mode = 0555,
129 .child = raid_table,
130 },
131 { .ctl_name = 0 }
132};
133
134static ctl_table raid_root_table[] = {
135 {
136 .ctl_name = CTL_DEV,
137 .procname = "dev",
138 .maxlen = 0,
139 .mode = 0555,
140 .child = raid_dir_table,
141 },
142 { .ctl_name = 0 }
143};
144
145static struct block_device_operations md_fops;
146
f91de92e
N		 147static int start_readonly;
148
d7603b7e
N		 149/*
150 * We have a system wide 'event count' that is incremented
151 * on any 'interesting' event, and readers of /proc/mdstat
152 * can use 'poll' or 'select' to find out when the event
153 * count increases.
154 *
155 * Events are:
156 * start array, stop array, error, add device, remove device,
157 * start build, activate spare
158 */
2989ddbd 159static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 160static atomic_t md_event_count;
07dbd377 161static void md_new_event(mddev_t *mddev)
d7603b7e
N		 162{
163 atomic_inc(&md_event_count);
164 wake_up(&md_event_waiters);
165}
166
1da177e4
LT		 167/*
168 * Enables to iterate over all existing md arrays
169 * all_mddevs_lock protects this list.
170 */
171static LIST_HEAD(all_mddevs);
172static DEFINE_SPINLOCK(all_mddevs_lock);
173
174
175/*
176 * iterates through all used mddevs in the system.
177 * We take care to grab the all_mddevs_lock whenever navigating
178 * the list, and to always hold a refcount when unlocked.
179 * Any code which breaks out of this loop while own
180 * a reference to the current mddev and must mddev_put it.
181 */
182#define ITERATE_MDDEV(mddev,tmp) \
183 \
184 for (({ spin_lock(&all_mddevs_lock); \
185 tmp = all_mddevs.next; \
186 mddev = NULL;}); \
187 ({ if (tmp != &all_mddevs) \
188 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
189 spin_unlock(&all_mddevs_lock); \
190 if (mddev) mddev_put(mddev); \
191 mddev = list_entry(tmp, mddev_t, all_mddevs); \
192 tmp != &all_mddevs;}); \
193 ({ spin_lock(&all_mddevs_lock); \
194 tmp = tmp->next;}) \
195 )
196
197
198static int md_fail_request (request_queue_t *q, struct bio *bio)
199{
200 bio_io_error(bio, bio->bi_size);
201 return 0;
202}
203
204static inline mddev_t *mddev_get(mddev_t *mddev)
205{
206 atomic_inc(&mddev->active);
207 return mddev;
208}
209
210static void mddev_put(mddev_t *mddev)
211{
212 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
213 return;
214 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
215 list_del(&mddev->all_mddevs);
1312f40e
AV		 216 /* that blocks */
217 blk_cleanup_queue(mddev->queue);
218 /* that also blocks */
eae1701f 219 kobject_unregister(&mddev->kobj);
1312f40e 220 /* result blows... */
1da177e4
LT		 221 }
222 spin_unlock(&all_mddevs_lock);
223}
224
225static mddev_t * mddev_find(dev_t unit)
226{
227 mddev_t *mddev, *new = NULL;
228
229 retry:
230 spin_lock(&all_mddevs_lock);
231 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
232 if (mddev->unit == unit) {
233 mddev_get(mddev);
234 spin_unlock(&all_mddevs_lock);
990a8baf 235 kfree(new);
1da177e4
LT		 236 return mddev;
237 }
238
239 if (new) {
240 list_add(&new->all_mddevs, &all_mddevs);
241 spin_unlock(&all_mddevs_lock);
242 return new;
243 }
244 spin_unlock(&all_mddevs_lock);
245
9ffae0cf 246 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 247 if (!new)
248 return NULL;
249
1da177e4
LT		 250 new->unit = unit;
251 if (MAJOR(unit) == MD_MAJOR)
252 new->md_minor = MINOR(unit);
253 else
254 new->md_minor = MINOR(unit) >> MdpMinorShift;
255
256 init_MUTEX(&new->reconfig_sem);
257 INIT_LIST_HEAD(&new->disks);
258 INIT_LIST_HEAD(&new->all_mddevs);
259 init_timer(&new->safemode_timer);
260 atomic_set(&new->active, 1);
06d91a5f 261 spin_lock_init(&new->write_lock);
3d310eb7 262 init_waitqueue_head(&new->sb_wait);
1da177e4
LT		 263
264 new->queue = blk_alloc_queue(GFP_KERNEL);
265 if (!new->queue) {
266 kfree(new);
267 return NULL;
268 }
89e5c8b5 269 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
1da177e4
LT		 270
271 blk_queue_make_request(new->queue, md_fail_request);
272
273 goto retry;
274}
275
276static inline int mddev_lock(mddev_t * mddev)
277{
278 return down_interruptible(&mddev->reconfig_sem);
279}
280
281static inline void mddev_lock_uninterruptible(mddev_t * mddev)
282{
283 down(&mddev->reconfig_sem);
284}
285
286static inline int mddev_trylock(mddev_t * mddev)
287{
288 return down_trylock(&mddev->reconfig_sem);
289}
290
291static inline void mddev_unlock(mddev_t * mddev)
292{
293 up(&mddev->reconfig_sem);
294
005eca5e 295 md_wakeup_thread(mddev->thread);
1da177e4
LT		 296}
297
2989ddbd 298static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4
LT		 299{
300 mdk_rdev_t * rdev;
301 struct list_head *tmp;
302
303 ITERATE_RDEV(mddev,rdev,tmp) {
304 if (rdev->desc_nr == nr)
305 return rdev;
306 }
307 return NULL;
308}
309
310static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
311{
312 struct list_head *tmp;
313 mdk_rdev_t *rdev;
314
315 ITERATE_RDEV(mddev,rdev,tmp) {
316 if (rdev->bdev->bd_dev == dev)
317 return rdev;
318 }
319 return NULL;
320}
321
d9d166c2 322static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 323{
324 struct mdk_personality *pers;
d9d166c2
N		 325 list_for_each_entry(pers, &pers_list, list) {
326 if (level != LEVEL_NONE && pers->level == level)
2604b703 327 return pers;
d9d166c2
N		 328 if (strcmp(pers->name, clevel)==0)
329 return pers;
330 }
2604b703
N		 331 return NULL;
332}
333
77933d72 334static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4
LT		 335{
336 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
337 return MD_NEW_SIZE_BLOCKS(size);
338}
339
340static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
341{
342 sector_t size;
343
344 size = rdev->sb_offset;
345
346 if (chunk_size)
347 size &= ~((sector_t)chunk_size/1024 - 1);
348 return size;
349}
350
351static int alloc_disk_sb(mdk_rdev_t * rdev)
352{
353 if (rdev->sb_page)
354 MD_BUG();
355
356 rdev->sb_page = alloc_page(GFP_KERNEL);
357 if (!rdev->sb_page) {
358 printk(KERN_ALERT "md: out of memory.\n");
359 return -EINVAL;
360 }
361
362 return 0;
363}
364
365static void free_disk_sb(mdk_rdev_t * rdev)
366{
367 if (rdev->sb_page) {
2d1f3b5d 368 put_page(rdev->sb_page);
1da177e4
LT		 369 rdev->sb_loaded = 0;
370 rdev->sb_page = NULL;
371 rdev->sb_offset = 0;
372 rdev->size = 0;
373 }
374}
375
376
7bfa19f2
N		 377static int super_written(struct bio *bio, unsigned int bytes_done, int error)
378{
379 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 380 mddev_t *mddev = rdev->mddev;
7bfa19f2
N		 381 if (bio->bi_size)
382 return 1;
383
384 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
a9701a30 385 md_error(mddev, rdev);
7bfa19f2 386
a9701a30
N		 387 if (atomic_dec_and_test(&mddev->pending_writes))
388 wake_up(&mddev->sb_wait);
f8b58edf 389 bio_put(bio);
7bfa19f2
N		 390 return 0;
391}
392
a9701a30
N		 393static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
394{
395 struct bio *bio2 = bio->bi_private;
396 mdk_rdev_t *rdev = bio2->bi_private;
397 mddev_t *mddev = rdev->mddev;
398 if (bio->bi_size)
399 return 1;
400
401 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
402 error == -EOPNOTSUPP) {
403 unsigned long flags;
404 /* barriers don't appear to be supported :-( */
405 set_bit(BarriersNotsupp, &rdev->flags);
406 mddev->barriers_work = 0;
407 spin_lock_irqsave(&mddev->write_lock, flags);
408 bio2->bi_next = mddev->biolist;
409 mddev->biolist = bio2;
410 spin_unlock_irqrestore(&mddev->write_lock, flags);
411 wake_up(&mddev->sb_wait);
412 bio_put(bio);
413 return 0;
414 }
415 bio_put(bio2);
416 bio->bi_private = rdev;
417 return super_written(bio, bytes_done, error);
418}
419
7bfa19f2
N		 420void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
421 sector_t sector, int size, struct page *page)
422{
423 /* write first size bytes of page to sector of rdev
424 * Increment mddev->pending_writes before returning
425 * and decrement it on completion, waking up sb_wait
426 * if zero is reached.
427 * If an error occurred, call md_error
a9701a30
N		 428 *
429 * As we might need to resubmit the request if BIO_RW_BARRIER
430 * causes ENOTSUPP, we allocate a spare bio...
7bfa19f2
N		 431 */
432 struct bio *bio = bio_alloc(GFP_NOIO, 1);
a9701a30 433 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
7bfa19f2
N		 434
435 bio->bi_bdev = rdev->bdev;
436 bio->bi_sector = sector;
437 bio_add_page(bio, page, size, 0);
438 bio->bi_private = rdev;
439 bio->bi_end_io = super_written;
a9701a30
N		 440 bio->bi_rw = rw;
441
7bfa19f2 442 atomic_inc(&mddev->pending_writes);
a9701a30
N		 443 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
444 struct bio *rbio;
445 rw |= (1<<BIO_RW_BARRIER);
446 rbio = bio_clone(bio, GFP_NOIO);
447 rbio->bi_private = bio;
448 rbio->bi_end_io = super_written_barrier;
449 submit_bio(rw, rbio);
450 } else
451 submit_bio(rw, bio);
452}
453
454void md_super_wait(mddev_t *mddev)
455{
456 /* wait for all superblock writes that were scheduled to complete.
457 * if any had to be retried (due to BARRIER problems), retry them
458 */
459 DEFINE_WAIT(wq);
460 for(;;) {
461 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
462 if (atomic_read(&mddev->pending_writes)==0)
463 break;
464 while (mddev->biolist) {
465 struct bio *bio;
466 spin_lock_irq(&mddev->write_lock);
467 bio = mddev->biolist;
468 mddev->biolist = bio->bi_next ;
469 bio->bi_next = NULL;
470 spin_unlock_irq(&mddev->write_lock);
471 submit_bio(bio->bi_rw, bio);
472 }
473 schedule();
474 }
475 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 476}
477
1da177e4
LT		 478static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
479{
480 if (bio->bi_size)
481 return 1;
482
483 complete((struct completion*)bio->bi_private);
484 return 0;
485}
486
a654b9d8 487int sync_page_io(struct block_device *bdev, sector_t sector, int size,
1da177e4
LT		 488 struct page *page, int rw)
489{
baaa2c51 490 struct bio *bio = bio_alloc(GFP_NOIO, 1);
1da177e4
LT		 491 struct completion event;
492 int ret;
493
494 rw |= (1 << BIO_RW_SYNC);
495
496 bio->bi_bdev = bdev;
497 bio->bi_sector = sector;
498 bio_add_page(bio, page, size, 0);
499 init_completion(&event);
500 bio->bi_private = &event;
501 bio->bi_end_io = bi_complete;
502 submit_bio(rw, bio);
503 wait_for_completion(&event);
504
505 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
506 bio_put(bio);
507 return ret;
508}
a8745db2 509EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 510
0002b271 511static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 512{
513 char b[BDEVNAME_SIZE];
514 if (!rdev->sb_page) {
515 MD_BUG();
516 return -EINVAL;
517 }
518 if (rdev->sb_loaded)
519 return 0;
520
521
0002b271 522 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
1da177e4
LT		 523 goto fail;
524 rdev->sb_loaded = 1;
525 return 0;
526
527fail:
528 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
529 bdevname(rdev->bdev,b));
530 return -EINVAL;
531}
532
533static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
534{
535 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
536 (sb1->set_uuid1 == sb2->set_uuid1) &&
537 (sb1->set_uuid2 == sb2->set_uuid2) &&
538 (sb1->set_uuid3 == sb2->set_uuid3))
539
540 return 1;
541
542 return 0;
543}
544
545
546static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
547{
548 int ret;
549 mdp_super_t *tmp1, *tmp2;
550
551 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
552 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
553
554 if (!tmp1 || !tmp2) {
555 ret = 0;
556 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
557 goto abort;
558 }
559
560 *tmp1 = *sb1;
561 *tmp2 = *sb2;
562
563 /*
564 * nr_disks is not constant
565 */
566 tmp1->nr_disks = 0;
567 tmp2->nr_disks = 0;
568
569 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
570 ret = 0;
571 else
572 ret = 1;
573
574abort:
990a8baf
JJ		 575 kfree(tmp1);
576 kfree(tmp2);
1da177e4
LT		 577 return ret;
578}
579
580static unsigned int calc_sb_csum(mdp_super_t * sb)
581{
582 unsigned int disk_csum, csum;
583
584 disk_csum = sb->sb_csum;
585 sb->sb_csum = 0;
586 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
587 sb->sb_csum = disk_csum;
588 return csum;
589}
590
591
592/*
593 * Handle superblock details.
594 * We want to be able to handle multiple superblock formats
595 * so we have a common interface to them all, and an array of
596 * different handlers.
597 * We rely on user-space to write the initial superblock, and support
598 * reading and updating of superblocks.
599 * Interface methods are:
600 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
601 * loads and validates a superblock on dev.
602 * if refdev != NULL, compare superblocks on both devices
603 * Return:
604 * 0 - dev has a superblock that is compatible with refdev
605 * 1 - dev has a superblock that is compatible and newer than refdev
606 * so dev should be used as the refdev in future
607 * -EINVAL superblock incompatible or invalid
608 * -othererror e.g. -EIO
609 *
610 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
611 * Verify that dev is acceptable into mddev.
612 * The first time, mddev->raid_disks will be 0, and data from
613 * dev should be merged in. Subsequent calls check that dev
614 * is new enough. Return 0 or -EINVAL
615 *
616 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
617 * Update the superblock for rdev with data in mddev
618 * This does not write to disc.
619 *
620 */
621
622struct super_type {
623 char *name;
624 struct module *owner;
625 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
626 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
627 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
628};
629
630/*
631 * load_super for 0.90.0
632 */
633static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
634{
635 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
636 mdp_super_t *sb;
637 int ret;
638 sector_t sb_offset;
639
640 /*
641 * Calculate the position of the superblock,
642 * it's at the end of the disk.
643 *
644 * It also happens to be a multiple of 4Kb.
645 */
646 sb_offset = calc_dev_sboffset(rdev->bdev);
647 rdev->sb_offset = sb_offset;
648
0002b271 649 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 650 if (ret) return ret;
651
652 ret = -EINVAL;
653
654 bdevname(rdev->bdev, b);
655 sb = (mdp_super_t*)page_address(rdev->sb_page);
656
657 if (sb->md_magic != MD_SB_MAGIC) {
658 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
659 b);
660 goto abort;
661 }
662
663 if (sb->major_version != 0 ||
664 sb->minor_version != 90) {
665 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
666 sb->major_version, sb->minor_version,
667 b);
668 goto abort;
669 }
670
671 if (sb->raid_disks <= 0)
672 goto abort;
673
674 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
675 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
676 b);
677 goto abort;
678 }
679
680 rdev->preferred_minor = sb->md_minor;
681 rdev->data_offset = 0;
0002b271 682 rdev->sb_size = MD_SB_BYTES;
1da177e4
LT		 683
684 if (sb->level == LEVEL_MULTIPATH)
685 rdev->desc_nr = -1;
686 else
687 rdev->desc_nr = sb->this_disk.number;
688
689 if (refdev == 0)
690 ret = 1;
691 else {
692 __u64 ev1, ev2;
693 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
694 if (!uuid_equal(refsb, sb)) {
695 printk(KERN_WARNING "md: %s has different UUID to %s\n",
696 b, bdevname(refdev->bdev,b2));
697 goto abort;
698 }
699 if (!sb_equal(refsb, sb)) {
700 printk(KERN_WARNING "md: %s has same UUID"
701 " but different superblock to %s\n",
702 b, bdevname(refdev->bdev, b2));
703 goto abort;
704 }
705 ev1 = md_event(sb);
706 ev2 = md_event(refsb);
707 if (ev1 > ev2)
708 ret = 1;
709 else
710 ret = 0;
711 }
712 rdev->size = calc_dev_size(rdev, sb->chunk_size);
713
2bf071bf
N		 714 if (rdev->size < sb->size && sb->level > 1)
715 /* "this cannot possibly happen" ... */
716 ret = -EINVAL;
717
1da177e4
LT		 718 abort:
719 return ret;
720}
721
722/*
723 * validate_super for 0.90.0
724 */
725static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
726{
727 mdp_disk_t *desc;
728 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
729
41158c7e 730 rdev->raid_disk = -1;
b2d444d7 731 rdev->flags = 0;
1da177e4
LT		 732 if (mddev->raid_disks == 0) {
733 mddev->major_version = 0;
734 mddev->minor_version = sb->minor_version;
735 mddev->patch_version = sb->patch_version;
736 mddev->persistent = ! sb->not_persistent;
737 mddev->chunk_size = sb->chunk_size;
738 mddev->ctime = sb->ctime;
739 mddev->utime = sb->utime;
740 mddev->level = sb->level;
d9d166c2 741 mddev->clevel[0] = 0;
1da177e4
LT		 742 mddev->layout = sb->layout;
743 mddev->raid_disks = sb->raid_disks;
744 mddev->size = sb->size;
745 mddev->events = md_event(sb);
9223214e 746 mddev->bitmap_offset = 0;
36fa3063 747 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
1da177e4
LT		 748
749 if (sb->state & (1<<MD_SB_CLEAN))
750 mddev->recovery_cp = MaxSector;
751 else {
752 if (sb->events_hi == sb->cp_events_hi &&
753 sb->events_lo == sb->cp_events_lo) {
754 mddev->recovery_cp = sb->recovery_cp;
755 } else
756 mddev->recovery_cp = 0;
757 }
758
759 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
760 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
761 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
762 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
763
764 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 765
766 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
767 mddev->bitmap_file == NULL) {
c5a10f62
N		 768 if (mddev->level != 1 && mddev->level != 4
769 && mddev->level != 5 && mddev->level != 6
6cce3b23 770 && mddev->level != 10) {
a654b9d8 771 /* FIXME use a better test */
6cce3b23 772 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
a654b9d8
N		 773 return -EINVAL;
774 }
36fa3063 775 mddev->bitmap_offset = mddev->default_bitmap_offset;
a654b9d8
N		 776 }
777
41158c7e
N		 778 } else if (mddev->pers == NULL) {
779 /* Insist on good event counter while assembling */
780 __u64 ev1 = md_event(sb);
1da177e4
LT		 781 ++ev1;
782 if (ev1 < mddev->events)
783 return -EINVAL;
41158c7e
N		 784 } else if (mddev->bitmap) {
785 /* if adding to array with a bitmap, then we can accept an
786 * older device ... but not too old.
787 */
788 __u64 ev1 = md_event(sb);
789 if (ev1 < mddev->bitmap->events_cleared)
790 return 0;
791 } else /* just a hot-add of a new device, leave raid_disk at -1 */
792 return 0;
793
1da177e4 794 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 795 desc = sb->disks + rdev->desc_nr;
796
797 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 798 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 799 else if (desc->state & (1<<MD_DISK_SYNC) &&
800 desc->raid_disk < mddev->raid_disks) {
b2d444d7 801 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 802 rdev->raid_disk = desc->raid_disk;
803 }
8ddf9efe
N		 804 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
805 set_bit(WriteMostly, &rdev->flags);
41158c7e 806 } else /* MULTIPATH are always insync */
b2d444d7 807 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 808 return 0;
809}
810
811/*
812 * sync_super for 0.90.0
813 */
814static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
815{
816 mdp_super_t *sb;
817 struct list_head *tmp;
818 mdk_rdev_t *rdev2;
819 int next_spare = mddev->raid_disks;
19133a42 820
1da177e4
LT		 821
822 /* make rdev->sb match mddev data..
823 *
824 * 1/ zero out disks
825 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
826 * 3/ any empty disks < next_spare become removed
827 *
828 * disks[0] gets initialised to REMOVED because
829 * we cannot be sure from other fields if it has
830 * been initialised or not.
831 */
832 int i;
833 int active=0, working=0,failed=0,spare=0,nr_disks=0;
834
61181565
N		 835 rdev->sb_size = MD_SB_BYTES;
836
1da177e4
LT		 837 sb = (mdp_super_t*)page_address(rdev->sb_page);
838
839 memset(sb, 0, sizeof(*sb));
840
841 sb->md_magic = MD_SB_MAGIC;
842 sb->major_version = mddev->major_version;
843 sb->minor_version = mddev->minor_version;
844 sb->patch_version = mddev->patch_version;
845 sb->gvalid_words = 0; /* ignored */
846 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
847 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
848 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
849 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
850
851 sb->ctime = mddev->ctime;
852 sb->level = mddev->level;
853 sb->size = mddev->size;
854 sb->raid_disks = mddev->raid_disks;
855 sb->md_minor = mddev->md_minor;
856 sb->not_persistent = !mddev->persistent;
857 sb->utime = mddev->utime;
858 sb->state = 0;
859 sb->events_hi = (mddev->events>>32);
860 sb->events_lo = (u32)mddev->events;
861
862 if (mddev->in_sync)
863 {
864 sb->recovery_cp = mddev->recovery_cp;
865 sb->cp_events_hi = (mddev->events>>32);
866 sb->cp_events_lo = (u32)mddev->events;
867 if (mddev->recovery_cp == MaxSector)
868 sb->state = (1<< MD_SB_CLEAN);
869 } else
870 sb->recovery_cp = 0;
871
872 sb->layout = mddev->layout;
873 sb->chunk_size = mddev->chunk_size;
874
a654b9d8
N		 875 if (mddev->bitmap && mddev->bitmap_file == NULL)
876 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
877
1da177e4
LT		 878 sb->disks[0].state = (1<<MD_DISK_REMOVED);
879 ITERATE_RDEV(mddev,rdev2,tmp) {
880 mdp_disk_t *d;
86e6ffdd 881 int desc_nr;
b2d444d7
N		 882 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
883 && !test_bit(Faulty, &rdev2->flags))
86e6ffdd 884 desc_nr = rdev2->raid_disk;
1da177e4 885 else
86e6ffdd 886 desc_nr = next_spare++;
19133a42 887 rdev2->desc_nr = desc_nr;
1da177e4
LT		 888 d = &sb->disks[rdev2->desc_nr];
889 nr_disks++;
890 d->number = rdev2->desc_nr;
891 d->major = MAJOR(rdev2->bdev->bd_dev);
892 d->minor = MINOR(rdev2->bdev->bd_dev);
b2d444d7
N		 893 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
894 && !test_bit(Faulty, &rdev2->flags))
1da177e4
LT		 895 d->raid_disk = rdev2->raid_disk;
896 else
897 d->raid_disk = rdev2->desc_nr; /* compatibility */
b2d444d7 898 if (test_bit(Faulty, &rdev2->flags)) {
1da177e4
LT		 899 d->state = (1<<MD_DISK_FAULTY);
900 failed++;
b2d444d7 901 } else if (test_bit(In_sync, &rdev2->flags)) {
1da177e4
LT		 902 d->state = (1<<MD_DISK_ACTIVE);
903 d->state |= (1<<MD_DISK_SYNC);
904 active++;
905 working++;
906 } else {
907 d->state = 0;
908 spare++;
909 working++;
910 }
8ddf9efe
N		 911 if (test_bit(WriteMostly, &rdev2->flags))
912 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 913 }
1da177e4
LT		 914 /* now set the "removed" and "faulty" bits on any missing devices */
915 for (i=0 ; i < mddev->raid_disks ; i++) {
916 mdp_disk_t *d = &sb->disks[i];
917 if (d->state == 0 && d->number == 0) {
918 d->number = i;
919 d->raid_disk = i;
920 d->state = (1<<MD_DISK_REMOVED);
921 d->state |= (1<<MD_DISK_FAULTY);
922 failed++;
923 }
924 }
925 sb->nr_disks = nr_disks;
926 sb->active_disks = active;
927 sb->working_disks = working;
928 sb->failed_disks = failed;
929 sb->spare_disks = spare;
930
931 sb->this_disk = sb->disks[rdev->desc_nr];
932 sb->sb_csum = calc_sb_csum(sb);
933}
934
935/*
936 * version 1 superblock
937 */
938
939static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
940{
941 unsigned int disk_csum, csum;
942 unsigned long long newcsum;
943 int size = 256 + le32_to_cpu(sb->max_dev)*2;
944 unsigned int *isuper = (unsigned int*)sb;
945 int i;
946
947 disk_csum = sb->sb_csum;
948 sb->sb_csum = 0;
949 newcsum = 0;
950 for (i=0; size>=4; size -= 4 )
951 newcsum += le32_to_cpu(*isuper++);
952
953 if (size == 2)
954 newcsum += le16_to_cpu(*(unsigned short*) isuper);
955
956 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
957 sb->sb_csum = disk_csum;
958 return cpu_to_le32(csum);
959}
960
961static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
962{
963 struct mdp_superblock_1 *sb;
964 int ret;
965 sector_t sb_offset;
966 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 967 int bmask;
1da177e4
LT		 968
969 /*
970 * Calculate the position of the superblock.
971 * It is always aligned to a 4K boundary and
972 * depeding on minor_version, it can be:
973 * 0: At least 8K, but less than 12K, from end of device
974 * 1: At start of device
975 * 2: 4K from start of device.
976 */
977 switch(minor_version) {
978 case 0:
979 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
980 sb_offset -= 8*2;
39730960 981 sb_offset &= ~(sector_t)(4*2-1);
1da177e4
LT		 982 /* convert from sectors to K */
983 sb_offset /= 2;
984 break;
985 case 1:
986 sb_offset = 0;
987 break;
988 case 2:
989 sb_offset = 4;
990 break;
991 default:
992 return -EINVAL;
993 }
994 rdev->sb_offset = sb_offset;
995
0002b271
N		 996 /* superblock is rarely larger than 1K, but it can be larger,
997 * and it is safe to read 4k, so we do that
998 */
999 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1000 if (ret) return ret;
1001
1002
1003 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1004
1005 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1006 sb->major_version != cpu_to_le32(1) ||
1007 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1008 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
71c0805c 1009 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1010 return -EINVAL;
1011
1012 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1013 printk("md: invalid superblock checksum on %s\n",
1014 bdevname(rdev->bdev,b));
1015 return -EINVAL;
1016 }
1017 if (le64_to_cpu(sb->data_size) < 10) {
1018 printk("md: data_size too small on %s\n",
1019 bdevname(rdev->bdev,b));
1020 return -EINVAL;
1021 }
1022 rdev->preferred_minor = 0xffff;
1023 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1024 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1025
0002b271 1026 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
720a3dc3 1027 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
0002b271
N		 1028 if (rdev->sb_size & bmask)
1029 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1030
1da177e4 1031 if (refdev == 0)
8ed75463 1032 ret = 1;
1da177e4
LT		 1033 else {
1034 __u64 ev1, ev2;
1035 struct mdp_superblock_1 *refsb =
1036 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1037
1038 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1039 sb->level != refsb->level ||
1040 sb->layout != refsb->layout ||
1041 sb->chunksize != refsb->chunksize) {
1042 printk(KERN_WARNING "md: %s has strangely different"
1043 " superblock to %s\n",
1044 bdevname(rdev->bdev,b),
1045 bdevname(refdev->bdev,b2));
1046 return -EINVAL;
1047 }
1048 ev1 = le64_to_cpu(sb->events);
1049 ev2 = le64_to_cpu(refsb->events);
1050
1051 if (ev1 > ev2)
8ed75463
N		 1052 ret = 1;
1053 else
1054 ret = 0;
1da177e4
LT		 1055 }
1056 if (minor_version)
1057 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1058 else
1059 rdev->size = rdev->sb_offset;
1060 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1061 return -EINVAL;
1062 rdev->size = le64_to_cpu(sb->data_size)/2;
1063 if (le32_to_cpu(sb->chunksize))
1064 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
2bf071bf
N		 1065
1066 if (le32_to_cpu(sb->size) > rdev->size*2)
1067 return -EINVAL;
8ed75463 1068 return ret;
1da177e4
LT		 1069}
1070
1071static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1072{
1073 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1074
41158c7e 1075 rdev->raid_disk = -1;
b2d444d7 1076 rdev->flags = 0;
1da177e4
LT		 1077 if (mddev->raid_disks == 0) {
1078 mddev->major_version = 1;
1079 mddev->patch_version = 0;
1080 mddev->persistent = 1;
1081 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1082 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1083 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1084 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1085 mddev->clevel[0] = 0;
1da177e4
LT		 1086 mddev->layout = le32_to_cpu(sb->layout);
1087 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1088 mddev->size = le64_to_cpu(sb->size)/2;
1089 mddev->events = le64_to_cpu(sb->events);
9223214e 1090 mddev->bitmap_offset = 0;
29fc7e3e 1091 mddev->default_bitmap_offset = 1024 >> 9;
1da177e4
LT		 1092
1093 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1094 memcpy(mddev->uuid, sb->set_uuid, 16);
1095
1096 mddev->max_disks = (4096-256)/2;
a654b9d8 1097
71c0805c 1098 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
a654b9d8 1099 mddev->bitmap_file == NULL ) {
6cce3b23
N		 1100 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1101 && mddev->level != 10) {
1102 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
a654b9d8
N		 1103 return -EINVAL;
1104 }
1105 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1106 }
41158c7e
N		 1107 } else if (mddev->pers == NULL) {
1108 /* Insist of good event counter while assembling */
1109 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4
LT		 1110 ++ev1;
1111 if (ev1 < mddev->events)
1112 return -EINVAL;
41158c7e
N		 1113 } else if (mddev->bitmap) {
1114 /* If adding to array with a bitmap, then we can accept an
1115 * older device, but not too old.
1116 */
1117 __u64 ev1 = le64_to_cpu(sb->events);
1118 if (ev1 < mddev->bitmap->events_cleared)
1119 return 0;
1120 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1121 return 0;
1da177e4
LT		 1122
1123 if (mddev->level != LEVEL_MULTIPATH) {
1124 int role;
1125 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1126 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1127 switch(role) {
1128 case 0xffff: /* spare */
1da177e4
LT		 1129 break;
1130 case 0xfffe: /* faulty */
b2d444d7 1131 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1132 break;
1133 default:
b2d444d7 1134 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1135 rdev->raid_disk = role;
1136 break;
1137 }
8ddf9efe
N		 1138 if (sb->devflags & WriteMostly1)
1139 set_bit(WriteMostly, &rdev->flags);
41158c7e 1140 } else /* MULTIPATH are always insync */
b2d444d7 1141 set_bit(In_sync, &rdev->flags);
41158c7e 1142
1da177e4
LT		 1143 return 0;
1144}
1145
1146static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1147{
1148 struct mdp_superblock_1 *sb;
1149 struct list_head *tmp;
1150 mdk_rdev_t *rdev2;
1151 int max_dev, i;
1152 /* make rdev->sb match mddev and rdev data. */
1153
1154 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1155
1156 sb->feature_map = 0;
1157 sb->pad0 = 0;
1158 memset(sb->pad1, 0, sizeof(sb->pad1));
1159 memset(sb->pad2, 0, sizeof(sb->pad2));
1160 memset(sb->pad3, 0, sizeof(sb->pad3));
1161
1162 sb->utime = cpu_to_le64((__u64)mddev->utime);
1163 sb->events = cpu_to_le64(mddev->events);
1164 if (mddev->in_sync)
1165 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1166 else
1167 sb->resync_offset = cpu_to_le64(0);
1168
4dbcdc75
N		 1169 sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1170
f0ca340c 1171 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
29fc7e3e 1172 sb->size = cpu_to_le64(mddev->size<<1);
f0ca340c 1173
a654b9d8
N		 1174 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1175 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1176 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8
N		 1177 }
1178
1da177e4
LT		 1179 max_dev = 0;
1180 ITERATE_RDEV(mddev,rdev2,tmp)
1181 if (rdev2->desc_nr+1 > max_dev)
1182 max_dev = rdev2->desc_nr+1;
1183
1184 sb->max_dev = cpu_to_le32(max_dev);
1185 for (i=0; i<max_dev;i++)
1186 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1187
1188 ITERATE_RDEV(mddev,rdev2,tmp) {
1189 i = rdev2->desc_nr;
b2d444d7 1190 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1191 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1192 else if (test_bit(In_sync, &rdev2->flags))
1da177e4
LT		 1193 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1194 else
1195 sb->dev_roles[i] = cpu_to_le16(0xffff);
1196 }
1197
1198 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1199 sb->sb_csum = calc_sb_1_csum(sb);
1200}
1201
1202
75c96f85 1203static struct super_type super_types[] = {
1da177e4
LT		 1204 [0] = {
1205 .name = "0.90.0",
1206 .owner = THIS_MODULE,
1207 .load_super = super_90_load,
1208 .validate_super = super_90_validate,
1209 .sync_super = super_90_sync,
1210 },
1211 [1] = {
1212 .name = "md-1",
1213 .owner = THIS_MODULE,
1214 .load_super = super_1_load,
1215 .validate_super = super_1_validate,
1216 .sync_super = super_1_sync,
1217 },
1218};
1219
1220static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1221{
1222 struct list_head *tmp;
1223 mdk_rdev_t *rdev;
1224
1225 ITERATE_RDEV(mddev,rdev,tmp)
1226 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1227 return rdev;
1228
1229 return NULL;
1230}
1231
1232static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1233{
1234 struct list_head *tmp;
1235 mdk_rdev_t *rdev;
1236
1237 ITERATE_RDEV(mddev1,rdev,tmp)
1238 if (match_dev_unit(mddev2, rdev))
1239 return 1;
1240
1241 return 0;
1242}
1243
1244static LIST_HEAD(pending_raid_disks);
1245
1246static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1247{
1248 mdk_rdev_t *same_pdev;
1249 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
f637b9f9 1250 struct kobject *ko;
1edf80d3 1251 char *s;
1da177e4
LT		 1252
1253 if (rdev->mddev) {
1254 MD_BUG();
1255 return -EINVAL;
1256 }
2bf071bf
N		 1257 /* make sure rdev->size exceeds mddev->size */
1258 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1259 if (mddev->pers)
1260 /* Cannot change size, so fail */
1261 return -ENOSPC;
1262 else
1263 mddev->size = rdev->size;
1264 }
1da177e4
LT		 1265 same_pdev = match_dev_unit(mddev, rdev);
1266 if (same_pdev)
1267 printk(KERN_WARNING
1268 "%s: WARNING: %s appears to be on the same physical"
1269 " disk as %s. True\n protection against single-disk"
1270 " failure might be compromised.\n",
1271 mdname(mddev), bdevname(rdev->bdev,b),
1272 bdevname(same_pdev->bdev,b2));
1273
1274 /* Verify rdev->desc_nr is unique.
1275 * If it is -1, assign a free number, else
1276 * check number is not in use
1277 */
1278 if (rdev->desc_nr < 0) {
1279 int choice = 0;
1280 if (mddev->pers) choice = mddev->raid_disks;
1281 while (find_rdev_nr(mddev, choice))
1282 choice++;
1283 rdev->desc_nr = choice;
1284 } else {
1285 if (find_rdev_nr(mddev, rdev->desc_nr))
1286 return -EBUSY;
1287 }
19133a42
N		 1288 bdevname(rdev->bdev,b);
1289 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1290 return -ENOMEM;
1edf80d3
NB		 1291 while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1292 *s = '!';
1da177e4
LT		 1293
1294 list_add(&rdev->same_set, &mddev->disks);
1295 rdev->mddev = mddev;
19133a42 1296 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1297
9c791977 1298 rdev->kobj.parent = &mddev->kobj;
86e6ffdd
N		 1299 kobject_add(&rdev->kobj);
1300
f637b9f9
N		 1301 if (rdev->bdev->bd_part)
1302 ko = &rdev->bdev->bd_part->kobj;
1303 else
1304 ko = &rdev->bdev->bd_disk->kobj;
1305 sysfs_create_link(&rdev->kobj, ko, "block");
5463c790 1306 bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1da177e4
LT		 1307 return 0;
1308}
1309
1310static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1311{
1312 char b[BDEVNAME_SIZE];
1313 if (!rdev->mddev) {
1314 MD_BUG();
1315 return;
1316 }
5463c790 1317 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1da177e4
LT		 1318 list_del_init(&rdev->same_set);
1319 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1320 rdev->mddev = NULL;
86e6ffdd
N		 1321 sysfs_remove_link(&rdev->kobj, "block");
1322 kobject_del(&rdev->kobj);
1da177e4
LT		 1323}
1324
1325/*
1326 * prevent the device from being mounted, repartitioned or
1327 * otherwise reused by a RAID array (or any other kernel
1328 * subsystem), by bd_claiming the device.
1329 */
1330static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1331{
1332 int err = 0;
1333 struct block_device *bdev;
1334 char b[BDEVNAME_SIZE];
1335
1336 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1337 if (IS_ERR(bdev)) {
1338 printk(KERN_ERR "md: could not open %s.\n",
1339 __bdevname(dev, b));
1340 return PTR_ERR(bdev);
1341 }
1342 err = bd_claim(bdev, rdev);
1343 if (err) {
1344 printk(KERN_ERR "md: could not bd_claim %s.\n",
1345 bdevname(bdev, b));
1346 blkdev_put(bdev);
1347 return err;
1348 }
1349 rdev->bdev = bdev;
1350 return err;
1351}
1352
1353static void unlock_rdev(mdk_rdev_t *rdev)
1354{
1355 struct block_device *bdev = rdev->bdev;
1356 rdev->bdev = NULL;
1357 if (!bdev)
1358 MD_BUG();
1359 bd_release(bdev);
1360 blkdev_put(bdev);
1361}
1362
1363void md_autodetect_dev(dev_t dev);
1364
1365static void export_rdev(mdk_rdev_t * rdev)
1366{
1367 char b[BDEVNAME_SIZE];
1368 printk(KERN_INFO "md: export_rdev(%s)\n",
1369 bdevname(rdev->bdev,b));
1370 if (rdev->mddev)
1371 MD_BUG();
1372 free_disk_sb(rdev);
1373 list_del_init(&rdev->same_set);
1374#ifndef MODULE
1375 md_autodetect_dev(rdev->bdev->bd_dev);
1376#endif
1377 unlock_rdev(rdev);
86e6ffdd 1378 kobject_put(&rdev->kobj);
1da177e4
LT		 1379}
1380
1381static void kick_rdev_from_array(mdk_rdev_t * rdev)
1382{
1383 unbind_rdev_from_array(rdev);
1384 export_rdev(rdev);
1385}
1386
1387static void export_array(mddev_t *mddev)
1388{
1389 struct list_head *tmp;
1390 mdk_rdev_t *rdev;
1391
1392 ITERATE_RDEV(mddev,rdev,tmp) {
1393 if (!rdev->mddev) {
1394 MD_BUG();
1395 continue;
1396 }
1397 kick_rdev_from_array(rdev);
1398 }
1399 if (!list_empty(&mddev->disks))
1400 MD_BUG();
1401 mddev->raid_disks = 0;
1402 mddev->major_version = 0;
1403}
1404
1405static void print_desc(mdp_disk_t *desc)
1406{
1407 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1408 desc->major,desc->minor,desc->raid_disk,desc->state);
1409}
1410
1411static void print_sb(mdp_super_t *sb)
1412{
1413 int i;
1414
1415 printk(KERN_INFO
1416 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1417 sb->major_version, sb->minor_version, sb->patch_version,
1418 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1419 sb->ctime);
1420 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1421 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1422 sb->md_minor, sb->layout, sb->chunk_size);
1423 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1424 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1425 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1426 sb->failed_disks, sb->spare_disks,
1427 sb->sb_csum, (unsigned long)sb->events_lo);
1428
1429 printk(KERN_INFO);
1430 for (i = 0; i < MD_SB_DISKS; i++) {
1431 mdp_disk_t *desc;
1432
1433 desc = sb->disks + i;
1434 if (desc->number || desc->major || desc->minor ||
1435 desc->raid_disk || (desc->state && (desc->state != 4))) {
1436 printk(" D %2d: ", i);
1437 print_desc(desc);
1438 }
1439 }
1440 printk(KERN_INFO "md: THIS: ");
1441 print_desc(&sb->this_disk);
1442
1443}
1444
1445static void print_rdev(mdk_rdev_t *rdev)
1446{
1447 char b[BDEVNAME_SIZE];
1448 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1449 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
b2d444d7
N		 1450 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1451 rdev->desc_nr);
1da177e4
LT		 1452 if (rdev->sb_loaded) {
1453 printk(KERN_INFO "md: rdev superblock:\n");
1454 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1455 } else
1456 printk(KERN_INFO "md: no rdev superblock!\n");
1457}
1458
1459void md_print_devices(void)
1460{
1461 struct list_head *tmp, *tmp2;
1462 mdk_rdev_t *rdev;
1463 mddev_t *mddev;
1464 char b[BDEVNAME_SIZE];
1465
1466 printk("\n");
1467 printk("md: **********************************\n");
1468 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1469 printk("md: **********************************\n");
1470 ITERATE_MDDEV(mddev,tmp) {
1da177e4 1471
32a7627c
N		 1472 if (mddev->bitmap)
1473 bitmap_print_sb(mddev->bitmap);
1474 else
1475 printk("%s: ", mdname(mddev));
1da177e4
LT		 1476 ITERATE_RDEV(mddev,rdev,tmp2)
1477 printk("<%s>", bdevname(rdev->bdev,b));
1478 printk("\n");
1479
1480 ITERATE_RDEV(mddev,rdev,tmp2)
1481 print_rdev(rdev);
1482 }
1483 printk("md: **********************************\n");
1484 printk("\n");
1485}
1486
1487
1da177e4
LT		 1488static void sync_sbs(mddev_t * mddev)
1489{
1490 mdk_rdev_t *rdev;
1491 struct list_head *tmp;
1492
1493 ITERATE_RDEV(mddev,rdev,tmp) {
1494 super_types[mddev->major_version].
1495 sync_super(mddev, rdev);
1496 rdev->sb_loaded = 1;
1497 }
1498}
1499
1500static void md_update_sb(mddev_t * mddev)
1501{
7bfa19f2 1502 int err;
1da177e4
LT		 1503 struct list_head *tmp;
1504 mdk_rdev_t *rdev;
06d91a5f 1505 int sync_req;
1da177e4 1506
1da177e4 1507repeat:
a9701a30 1508 spin_lock_irq(&mddev->write_lock);
06d91a5f 1509 sync_req = mddev->in_sync;
1da177e4
LT		 1510 mddev->utime = get_seconds();
1511 mddev->events ++;
1512
1513 if (!mddev->events) {
1514 /*
1515 * oops, this 64-bit counter should never wrap.
1516 * Either we are in around ~1 trillion A.C., assuming
1517 * 1 reboot per second, or we have a bug:
1518 */
1519 MD_BUG();
1520 mddev->events --;
1521 }
7bfa19f2 1522 mddev->sb_dirty = 2;
1da177e4
LT		 1523 sync_sbs(mddev);
1524
1525 /*
1526 * do not write anything to disk if using
1527 * nonpersistent superblocks
1528 */
06d91a5f
N		 1529 if (!mddev->persistent) {
1530 mddev->sb_dirty = 0;
a9701a30 1531 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1532 wake_up(&mddev->sb_wait);
1da177e4 1533 return;
06d91a5f 1534 }
a9701a30 1535 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1536
1537 dprintk(KERN_INFO
1538 "md: updating %s RAID superblock on device (in sync %d)\n",
1539 mdname(mddev),mddev->in_sync);
1540
32a7627c 1541 err = bitmap_update_sb(mddev->bitmap);
1da177e4
LT		 1542 ITERATE_RDEV(mddev,rdev,tmp) {
1543 char b[BDEVNAME_SIZE];
1544 dprintk(KERN_INFO "md: ");
b2d444d7 1545 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1546 dprintk("(skipping faulty ");
1547
1548 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1549 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1550 md_super_write(mddev,rdev,
0002b271 1551 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1552 rdev->sb_page);
1553 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1554 bdevname(rdev->bdev,b),
1555 (unsigned long long)rdev->sb_offset);
1556
1da177e4
LT		 1557 } else
1558 dprintk(")\n");
7bfa19f2 1559 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1560 /* only need to write one superblock... */
1561 break;
1562 }
a9701a30 1563 md_super_wait(mddev);
7bfa19f2
N		 1564 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1565
a9701a30 1566 spin_lock_irq(&mddev->write_lock);
7bfa19f2 1567 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
06d91a5f 1568 /* have to write it out again */
a9701a30 1569 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1570 goto repeat;
1571 }
1572 mddev->sb_dirty = 0;
a9701a30 1573 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1574 wake_up(&mddev->sb_wait);
06d91a5f 1575
1da177e4
LT		 1576}
1577
bce74dac
N		 1578/* words written to sysfs files may, or my not, be \n terminated.
1579 * We want to accept with case. For this we use cmd_match.
1580 */
1581static int cmd_match(const char *cmd, const char *str)
1582{
1583 /* See if cmd, written into a sysfs file, matches
1584 * str. They must either be the same, or cmd can
1585 * have a trailing newline
1586 */
1587 while (*cmd && *str && *cmd == *str) {
1588 cmd++;
1589 str++;
1590 }
1591 if (*cmd == '\n')
1592 cmd++;
1593 if (*str || *cmd)
1594 return 0;
1595 return 1;
1596}
1597
86e6ffdd
N		 1598struct rdev_sysfs_entry {
1599 struct attribute attr;
1600 ssize_t (*show)(mdk_rdev_t *, char *);
1601 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1602};
1603
1604static ssize_t
96de1e66 1605state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1606{
1607 char *sep = "";
1608 int len=0;
1609
b2d444d7 1610 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1611 len+= sprintf(page+len, "%sfaulty",sep);
1612 sep = ",";
1613 }
b2d444d7 1614 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1615 len += sprintf(page+len, "%sin_sync",sep);
1616 sep = ",";
1617 }
b2d444d7
N		 1618 if (!test_bit(Faulty, &rdev->flags) &&
1619 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1620 len += sprintf(page+len, "%sspare", sep);
1621 sep = ",";
1622 }
1623 return len+sprintf(page+len, "\n");
1624}
1625
96de1e66
N		 1626static struct rdev_sysfs_entry
1627rdev_state = __ATTR_RO(state);
86e6ffdd
N		 1628
1629static ssize_t
96de1e66 1630super_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1631{
1632 if (rdev->sb_loaded && rdev->sb_size) {
1633 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1634 return rdev->sb_size;
1635 } else
1636 return 0;
1637}
96de1e66
N		 1638static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1639
4dbcdc75
N		 1640static ssize_t
1641errors_show(mdk_rdev_t *rdev, char *page)
1642{
1643 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1644}
1645
1646static ssize_t
1647errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1648{
1649 char *e;
1650 unsigned long n = simple_strtoul(buf, &e, 10);
1651 if (*buf && (*e == 0 || *e == '\n')) {
1652 atomic_set(&rdev->corrected_errors, n);
1653 return len;
1654 }
1655 return -EINVAL;
1656}
1657static struct rdev_sysfs_entry rdev_errors =
1658__ATTR(errors, 0644, errors_show, errors_store);
1659
014236d2
N		 1660static ssize_t
1661slot_show(mdk_rdev_t *rdev, char *page)
1662{
1663 if (rdev->raid_disk < 0)
1664 return sprintf(page, "none\n");
1665 else
1666 return sprintf(page, "%d\n", rdev->raid_disk);
1667}
1668
1669static ssize_t
1670slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1671{
1672 char *e;
1673 int slot = simple_strtoul(buf, &e, 10);
1674 if (strncmp(buf, "none", 4)==0)
1675 slot = -1;
1676 else if (e==buf || (*e && *e!= '\n'))
1677 return -EINVAL;
1678 if (rdev->mddev->pers)
1679 /* Cannot set slot in active array (yet) */
1680 return -EBUSY;
1681 if (slot >= rdev->mddev->raid_disks)
1682 return -ENOSPC;
1683 rdev->raid_disk = slot;
1684 /* assume it is working */
1685 rdev->flags = 0;
1686 set_bit(In_sync, &rdev->flags);
1687 return len;
1688}
1689
1690
1691static struct rdev_sysfs_entry rdev_slot =
1692__ATTR(slot, 0644, slot_show, slot_store);
1693
93c8cad0
N		 1694static ssize_t
1695offset_show(mdk_rdev_t *rdev, char *page)
1696{
6961ece4 1697 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 1698}
1699
1700static ssize_t
1701offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1702{
1703 char *e;
1704 unsigned long long offset = simple_strtoull(buf, &e, 10);
1705 if (e==buf || (*e && *e != '\n'))
1706 return -EINVAL;
1707 if (rdev->mddev->pers)
1708 return -EBUSY;
1709 rdev->data_offset = offset;
1710 return len;
1711}
1712
1713static struct rdev_sysfs_entry rdev_offset =
1714__ATTR(offset, 0644, offset_show, offset_store);
1715
83303b61
N		 1716static ssize_t
1717rdev_size_show(mdk_rdev_t *rdev, char *page)
1718{
1719 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1720}
1721
1722static ssize_t
1723rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1724{
1725 char *e;
1726 unsigned long long size = simple_strtoull(buf, &e, 10);
1727 if (e==buf || (*e && *e != '\n'))
1728 return -EINVAL;
1729 if (rdev->mddev->pers)
1730 return -EBUSY;
1731 rdev->size = size;
1732 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1733 rdev->mddev->size = size;
1734 return len;
1735}
1736
1737static struct rdev_sysfs_entry rdev_size =
1738__ATTR(size, 0644, rdev_size_show, rdev_size_store);
1739
86e6ffdd
N		 1740static struct attribute *rdev_default_attrs[] = {
1741 &rdev_state.attr,
1742 &rdev_super.attr,
4dbcdc75 1743 &rdev_errors.attr,
014236d2 1744 &rdev_slot.attr,
93c8cad0 1745 &rdev_offset.attr,
83303b61 1746 &rdev_size.attr,
86e6ffdd
N		 1747 NULL,
1748};
1749static ssize_t
1750rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1751{
1752 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1753 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1754
1755 if (!entry->show)
1756 return -EIO;
1757 return entry->show(rdev, page);
1758}
1759
1760static ssize_t
1761rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1762 const char *page, size_t length)
1763{
1764 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1765 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1766
1767 if (!entry->store)
1768 return -EIO;
1769 return entry->store(rdev, page, length);
1770}
1771
1772static void rdev_free(struct kobject *ko)
1773{
1774 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1775 kfree(rdev);
1776}
1777static struct sysfs_ops rdev_sysfs_ops = {
1778 .show = rdev_attr_show,
1779 .store = rdev_attr_store,
1780};
1781static struct kobj_type rdev_ktype = {
1782 .release = rdev_free,
1783 .sysfs_ops = &rdev_sysfs_ops,
1784 .default_attrs = rdev_default_attrs,
1785};
1786
1da177e4
LT		 1787/*
1788 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1789 *
1790 * mark the device faulty if:
1791 *
1792 * - the device is nonexistent (zero size)
1793 * - the device has no valid superblock
1794 *
1795 * a faulty rdev _never_ has rdev->sb set.
1796 */
1797static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1798{
1799 char b[BDEVNAME_SIZE];
1800 int err;
1801 mdk_rdev_t *rdev;
1802 sector_t size;
1803
9ffae0cf 1804 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 1805 if (!rdev) {
1806 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1807 return ERR_PTR(-ENOMEM);
1808 }
1da177e4
LT		 1809
1810 if ((err = alloc_disk_sb(rdev)))
1811 goto abort_free;
1812
1813 err = lock_rdev(rdev, newdev);
1814 if (err)
1815 goto abort_free;
1816
86e6ffdd
N		 1817 rdev->kobj.parent = NULL;
1818 rdev->kobj.ktype = &rdev_ktype;
1819 kobject_init(&rdev->kobj);
1820
1da177e4 1821 rdev->desc_nr = -1;
b2d444d7 1822 rdev->flags = 0;
1da177e4
LT		 1823 rdev->data_offset = 0;
1824 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 1825 atomic_set(&rdev->read_errors, 0);
4dbcdc75 1826 atomic_set(&rdev->corrected_errors, 0);
1da177e4
LT		 1827
1828 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1829 if (!size) {
1830 printk(KERN_WARNING
1831 "md: %s has zero or unknown size, marking faulty!\n",
1832 bdevname(rdev->bdev,b));
1833 err = -EINVAL;
1834 goto abort_free;
1835 }
1836
1837 if (super_format >= 0) {
1838 err = super_types[super_format].
1839 load_super(rdev, NULL, super_minor);
1840 if (err == -EINVAL) {
1841 printk(KERN_WARNING
1842 "md: %s has invalid sb, not importing!\n",
1843 bdevname(rdev->bdev,b));
1844 goto abort_free;
1845 }
1846 if (err < 0) {
1847 printk(KERN_WARNING
1848 "md: could not read %s's sb, not importing!\n",
1849 bdevname(rdev->bdev,b));
1850 goto abort_free;
1851 }
1852 }
1853 INIT_LIST_HEAD(&rdev->same_set);
1854
1855 return rdev;
1856
1857abort_free:
1858 if (rdev->sb_page) {
1859 if (rdev->bdev)
1860 unlock_rdev(rdev);
1861 free_disk_sb(rdev);
1862 }
1863 kfree(rdev);
1864 return ERR_PTR(err);
1865}
1866
1867/*
1868 * Check a full RAID array for plausibility
1869 */
1870
1871
a757e64c 1872static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 1873{
1874 int i;
1875 struct list_head *tmp;
1876 mdk_rdev_t *rdev, *freshest;
1877 char b[BDEVNAME_SIZE];
1878
1879 freshest = NULL;
1880 ITERATE_RDEV(mddev,rdev,tmp)
1881 switch (super_types[mddev->major_version].
1882 load_super(rdev, freshest, mddev->minor_version)) {
1883 case 1:
1884 freshest = rdev;
1885 break;
1886 case 0:
1887 break;
1888 default:
1889 printk( KERN_ERR \
1890 "md: fatal superblock inconsistency in %s"
1891 " -- removing from array\n",
1892 bdevname(rdev->bdev,b));
1893 kick_rdev_from_array(rdev);
1894 }
1895
1896
1897 super_types[mddev->major_version].
1898 validate_super(mddev, freshest);
1899
1900 i = 0;
1901 ITERATE_RDEV(mddev,rdev,tmp) {
1902 if (rdev != freshest)
1903 if (super_types[mddev->major_version].
1904 validate_super(mddev, rdev)) {
1905 printk(KERN_WARNING "md: kicking non-fresh %s"
1906 " from array!\n",
1907 bdevname(rdev->bdev,b));
1908 kick_rdev_from_array(rdev);
1909 continue;
1910 }
1911 if (mddev->level == LEVEL_MULTIPATH) {
1912 rdev->desc_nr = i++;
1913 rdev->raid_disk = rdev->desc_nr;
b2d444d7 1914 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1915 }
1916 }
1917
1918
1919
1920 if (mddev->recovery_cp != MaxSector &&
1921 mddev->level >= 1)
1922 printk(KERN_ERR "md: %s: raid array is not clean"
1923 " -- starting background reconstruction\n",
1924 mdname(mddev));
1925
1da177e4
LT		 1926}
1927
eae1701f 1928static ssize_t
96de1e66 1929level_show(mddev_t *mddev, char *page)
eae1701f 1930{
2604b703 1931 struct mdk_personality *p = mddev->pers;
d9d166c2 1932 if (p)
eae1701f 1933 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 1934 else if (mddev->clevel[0])
1935 return sprintf(page, "%s\n", mddev->clevel);
1936 else if (mddev->level != LEVEL_NONE)
1937 return sprintf(page, "%d\n", mddev->level);
1938 else
1939 return 0;
eae1701f
N		 1940}
1941
d9d166c2
N		 1942static ssize_t
1943level_store(mddev_t *mddev, const char *buf, size_t len)
1944{
1945 int rv = len;
1946 if (mddev->pers)
1947 return -EBUSY;
1948 if (len == 0)
1949 return 0;
1950 if (len >= sizeof(mddev->clevel))
1951 return -ENOSPC;
1952 strncpy(mddev->clevel, buf, len);
1953 if (mddev->clevel[len-1] == '\n')
1954 len--;
1955 mddev->clevel[len] = 0;
1956 mddev->level = LEVEL_NONE;
1957 return rv;
1958}
1959
1960static struct md_sysfs_entry md_level =
1961__ATTR(level, 0644, level_show, level_store);
eae1701f
N		 1962
1963static ssize_t
96de1e66 1964raid_disks_show(mddev_t *mddev, char *page)
eae1701f 1965{
bb636547
N		 1966 if (mddev->raid_disks == 0)
1967 return 0;
eae1701f
N		 1968 return sprintf(page, "%d\n", mddev->raid_disks);
1969}
1970
da943b99
N		 1971static int update_raid_disks(mddev_t *mddev, int raid_disks);
1972
1973static ssize_t
1974raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
1975{
1976 /* can only set raid_disks if array is not yet active */
1977 char *e;
1978 int rv = 0;
1979 unsigned long n = simple_strtoul(buf, &e, 10);
1980
1981 if (!*buf || (*e && *e != '\n'))
1982 return -EINVAL;
1983
1984 if (mddev->pers)
1985 rv = update_raid_disks(mddev, n);
1986 else
1987 mddev->raid_disks = n;
1988 return rv ? rv : len;
1989}
1990static struct md_sysfs_entry md_raid_disks =
1991__ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
eae1701f 1992
3b34380a
N		 1993static ssize_t
1994chunk_size_show(mddev_t *mddev, char *page)
1995{
1996 return sprintf(page, "%d\n", mddev->chunk_size);
1997}
1998
1999static ssize_t
2000chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2001{
2002 /* can only set chunk_size if array is not yet active */
2003 char *e;
2004 unsigned long n = simple_strtoul(buf, &e, 10);
2005
2006 if (mddev->pers)
2007 return -EBUSY;
2008 if (!*buf || (*e && *e != '\n'))
2009 return -EINVAL;
2010
2011 mddev->chunk_size = n;
2012 return len;
2013}
2014static struct md_sysfs_entry md_chunk_size =
2015__ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
2016
6d7ff738
N		 2017static ssize_t
2018null_show(mddev_t *mddev, char *page)
2019{
2020 return -EINVAL;
2021}
2022
2023static ssize_t
2024new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2025{
2026 /* buf must be %d:%d\n? giving major and minor numbers */
2027 /* The new device is added to the array.
2028 * If the array has a persistent superblock, we read the
2029 * superblock to initialise info and check validity.
2030 * Otherwise, only checking done is that in bind_rdev_to_array,
2031 * which mainly checks size.
2032 */
2033 char *e;
2034 int major = simple_strtoul(buf, &e, 10);
2035 int minor;
2036 dev_t dev;
2037 mdk_rdev_t *rdev;
2038 int err;
2039
2040 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2041 return -EINVAL;
2042 minor = simple_strtoul(e+1, &e, 10);
2043 if (*e && *e != '\n')
2044 return -EINVAL;
2045 dev = MKDEV(major, minor);
2046 if (major != MAJOR(dev) ||
2047 minor != MINOR(dev))
2048 return -EOVERFLOW;
2049
2050
2051 if (mddev->persistent) {
2052 rdev = md_import_device(dev, mddev->major_version,
2053 mddev->minor_version);
2054 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2055 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2056 mdk_rdev_t, same_set);
2057 err = super_types[mddev->major_version]
2058 .load_super(rdev, rdev0, mddev->minor_version);
2059 if (err < 0)
2060 goto out;
2061 }
2062 } else
2063 rdev = md_import_device(dev, -1, -1);
2064
2065 if (IS_ERR(rdev))
2066 return PTR_ERR(rdev);
2067 err = bind_rdev_to_array(rdev, mddev);
2068 out:
2069 if (err)
2070 export_rdev(rdev);
2071 return err ? err : len;
2072}
2073
2074static struct md_sysfs_entry md_new_device =
2075__ATTR(new_dev, 0200, null_show, new_dev_store);
3b34380a 2076
a35b0d69
N		 2077static ssize_t
2078size_show(mddev_t *mddev, char *page)
2079{
2080 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2081}
2082
2083static int update_size(mddev_t *mddev, unsigned long size);
2084
2085static ssize_t
2086size_store(mddev_t *mddev, const char *buf, size_t len)
2087{
2088 /* If array is inactive, we can reduce the component size, but
2089 * not increase it (except from 0).
2090 * If array is active, we can try an on-line resize
2091 */
2092 char *e;
2093 int err = 0;
2094 unsigned long long size = simple_strtoull(buf, &e, 10);
2095 if (!*buf || *buf == '\n' ||
2096 (*e && *e != '\n'))
2097 return -EINVAL;
2098
2099 if (mddev->pers) {
2100 err = update_size(mddev, size);
2101 md_update_sb(mddev);
2102 } else {
2103 if (mddev->size == 0 ||
2104 mddev->size > size)
2105 mddev->size = size;
2106 else
2107 err = -ENOSPC;
2108 }
2109 return err ? err : len;
2110}
2111
2112static struct md_sysfs_entry md_size =
2113__ATTR(component_size, 0644, size_show, size_store);
2114
8bb93aac
N		 2115
2116/* Metdata version.
2117 * This is either 'none' for arrays with externally managed metadata,
2118 * or N.M for internally known formats
2119 */
2120static ssize_t
2121metadata_show(mddev_t *mddev, char *page)
2122{
2123 if (mddev->persistent)
2124 return sprintf(page, "%d.%d\n",
2125 mddev->major_version, mddev->minor_version);
2126 else
2127 return sprintf(page, "none\n");
2128}
2129
2130static ssize_t
2131metadata_store(mddev_t *mddev, const char *buf, size_t len)
2132{
2133 int major, minor;
2134 char *e;
2135 if (!list_empty(&mddev->disks))
2136 return -EBUSY;
2137
2138 if (cmd_match(buf, "none")) {
2139 mddev->persistent = 0;
2140 mddev->major_version = 0;
2141 mddev->minor_version = 90;
2142 return len;
2143 }
2144 major = simple_strtoul(buf, &e, 10);
2145 if (e==buf || *e != '.')
2146 return -EINVAL;
2147 buf = e+1;
2148 minor = simple_strtoul(buf, &e, 10);
2149 if (e==buf || *e != '\n')
2150 return -EINVAL;
2151 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2152 super_types[major].name == NULL)
2153 return -ENOENT;
2154 mddev->major_version = major;
2155 mddev->minor_version = minor;
2156 mddev->persistent = 1;
2157 return len;
2158}
2159
2160static struct md_sysfs_entry md_metadata =
2161__ATTR(metadata_version, 0644, metadata_show, metadata_store);
2162
24dd469d 2163static ssize_t
7eec314d 2164action_show(mddev_t *mddev, char *page)
24dd469d 2165{
7eec314d 2166 char *type = "idle";
31399d9e
N		 2167 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2168 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2169 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 2170 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2171 type = "resync";
2172 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2173 type = "check";
2174 else
2175 type = "repair";
2176 } else
2177 type = "recover";
2178 }
2179 return sprintf(page, "%s\n", type);
2180}
2181
2182static ssize_t
7eec314d 2183action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 2184{
7eec314d
N		 2185 if (!mddev->pers || !mddev->pers->sync_request)
2186 return -EINVAL;
2187
bce74dac 2188 if (cmd_match(page, "idle")) {
7eec314d
N		 2189 if (mddev->sync_thread) {
2190 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2191 md_unregister_thread(mddev->sync_thread);
2192 mddev->sync_thread = NULL;
2193 mddev->recovery = 0;
2194 }
03c902e1
N		 2195 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2196 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 2197 return -EBUSY;
03c902e1 2198 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
7eec314d
N		 2199 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2200 else {
bce74dac 2201 if (cmd_match(page, "check"))
7eec314d 2202 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
bce74dac 2203 else if (cmd_match(page, "repair"))
7eec314d
N		 2204 return -EINVAL;
2205 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2206 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 2207 }
03c902e1 2208 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 2209 md_wakeup_thread(mddev->thread);
2210 return len;
2211}
2212
9d88883e 2213static ssize_t
96de1e66 2214mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 2215{
2216 return sprintf(page, "%llu\n",
2217 (unsigned long long) mddev->resync_mismatches);
2218}
2219
96de1e66 2220static struct md_sysfs_entry
7eec314d 2221md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 2222
96de1e66
N		 2223
2224static struct md_sysfs_entry
2225md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 2226
88202a0c
N		 2227static ssize_t
2228sync_min_show(mddev_t *mddev, char *page)
2229{
2230 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2231 mddev->sync_speed_min ? "local": "system");
2232}
2233
2234static ssize_t
2235sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2236{
2237 int min;
2238 char *e;
2239 if (strncmp(buf, "system", 6)==0) {
2240 mddev->sync_speed_min = 0;
2241 return len;
2242 }
2243 min = simple_strtoul(buf, &e, 10);
2244 if (buf == e || (*e && *e != '\n') || min <= 0)
2245 return -EINVAL;
2246 mddev->sync_speed_min = min;
2247 return len;
2248}
2249
2250static struct md_sysfs_entry md_sync_min =
2251__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2252
2253static ssize_t
2254sync_max_show(mddev_t *mddev, char *page)
2255{
2256 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2257 mddev->sync_speed_max ? "local": "system");
2258}
2259
2260static ssize_t
2261sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2262{
2263 int max;
2264 char *e;
2265 if (strncmp(buf, "system", 6)==0) {
2266 mddev->sync_speed_max = 0;
2267 return len;
2268 }
2269 max = simple_strtoul(buf, &e, 10);
2270 if (buf == e || (*e && *e != '\n') || max <= 0)
2271 return -EINVAL;
2272 mddev->sync_speed_max = max;
2273 return len;
2274}
2275
2276static struct md_sysfs_entry md_sync_max =
2277__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2278
2279
2280static ssize_t
2281sync_speed_show(mddev_t *mddev, char *page)
2282{
2283 unsigned long resync, dt, db;
2284 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2285 dt = ((jiffies - mddev->resync_mark) / HZ);
2286 if (!dt) dt++;
2287 db = resync - (mddev->resync_mark_cnt);
2288 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2289}
2290
2291static struct md_sysfs_entry
2292md_sync_speed = __ATTR_RO(sync_speed);
2293
2294static ssize_t
2295sync_completed_show(mddev_t *mddev, char *page)
2296{
2297 unsigned long max_blocks, resync;
2298
2299 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2300 max_blocks = mddev->resync_max_sectors;
2301 else
2302 max_blocks = mddev->size << 1;
2303
2304 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2305 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2306}
2307
2308static struct md_sysfs_entry
2309md_sync_completed = __ATTR_RO(sync_completed);
2310
eae1701f
N		 2311static struct attribute *md_default_attrs[] = {
2312 &md_level.attr,
2313 &md_raid_disks.attr,
3b34380a 2314 &md_chunk_size.attr,
a35b0d69 2315 &md_size.attr,
8bb93aac 2316 &md_metadata.attr,
6d7ff738 2317 &md_new_device.attr,
411036fa
N		 2318 NULL,
2319};
2320
2321static struct attribute *md_redundancy_attrs[] = {
24dd469d 2322 &md_scan_mode.attr,
9d88883e 2323 &md_mismatches.attr,
88202a0c
N		 2324 &md_sync_min.attr,
2325 &md_sync_max.attr,
2326 &md_sync_speed.attr,
2327 &md_sync_completed.attr,
eae1701f
N		 2328 NULL,
2329};
411036fa
N		 2330static struct attribute_group md_redundancy_group = {
2331 .name = NULL,
2332 .attrs = md_redundancy_attrs,
2333};
2334
eae1701f
N		 2335
2336static ssize_t
2337md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2338{
2339 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2340 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2341 ssize_t rv;
eae1701f
N		 2342
2343 if (!entry->show)
2344 return -EIO;
96de1e66
N		 2345 mddev_lock(mddev);
2346 rv = entry->show(mddev, page);
2347 mddev_unlock(mddev);
2348 return rv;
eae1701f
N		 2349}
2350
2351static ssize_t
2352md_attr_store(struct kobject *kobj, struct attribute *attr,
2353 const char *page, size_t length)
2354{
2355 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2356 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2357 ssize_t rv;
eae1701f
N		 2358
2359 if (!entry->store)
2360 return -EIO;
96de1e66
N		 2361 mddev_lock(mddev);
2362 rv = entry->store(mddev, page, length);
2363 mddev_unlock(mddev);
2364 return rv;
eae1701f
N		 2365}
2366
2367static void md_free(struct kobject *ko)
2368{
2369 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2370 kfree(mddev);
2371}
2372
2373static struct sysfs_ops md_sysfs_ops = {
2374 .show = md_attr_show,
2375 .store = md_attr_store,
2376};
2377static struct kobj_type md_ktype = {
2378 .release = md_free,
2379 .sysfs_ops = &md_sysfs_ops,
2380 .default_attrs = md_default_attrs,
2381};
2382
1da177e4
LT		 2383int mdp_major = 0;
2384
2385static struct kobject *md_probe(dev_t dev, int *part, void *data)
2386{
2387 static DECLARE_MUTEX(disks_sem);
2388 mddev_t *mddev = mddev_find(dev);
2389 struct gendisk *disk;
2390 int partitioned = (MAJOR(dev) != MD_MAJOR);
2391 int shift = partitioned ? MdpMinorShift : 0;
2392 int unit = MINOR(dev) >> shift;
2393
2394 if (!mddev)
2395 return NULL;
2396
2397 down(&disks_sem);
2398 if (mddev->gendisk) {
2399 up(&disks_sem);
2400 mddev_put(mddev);
2401 return NULL;
2402 }
2403 disk = alloc_disk(1 << shift);
2404 if (!disk) {
2405 up(&disks_sem);
2406 mddev_put(mddev);
2407 return NULL;
2408 }
2409 disk->major = MAJOR(dev);
2410 disk->first_minor = unit << shift;
2411 if (partitioned) {
2412 sprintf(disk->disk_name, "md_d%d", unit);
2413 sprintf(disk->devfs_name, "md/d%d", unit);
2414 } else {
2415 sprintf(disk->disk_name, "md%d", unit);
2416 sprintf(disk->devfs_name, "md/%d", unit);
2417 }
2418 disk->fops = &md_fops;
2419 disk->private_data = mddev;
2420 disk->queue = mddev->queue;
2421 add_disk(disk);
2422 mddev->gendisk = disk;
2423 up(&disks_sem);
9c791977 2424 mddev->kobj.parent = &disk->kobj;
eae1701f
N		 2425 mddev->kobj.k_name = NULL;
2426 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2427 mddev->kobj.ktype = &md_ktype;
2428 kobject_register(&mddev->kobj);
1da177e4
LT		 2429 return NULL;
2430}
2431
2432void md_wakeup_thread(mdk_thread_t *thread);
2433
2434static void md_safemode_timeout(unsigned long data)
2435{
2436 mddev_t *mddev = (mddev_t *) data;
2437
2438 mddev->safemode = 1;
2439 md_wakeup_thread(mddev->thread);
2440}
2441
6ff8d8ec 2442static int start_dirty_degraded;
1da177e4
LT		 2443
2444static int do_md_run(mddev_t * mddev)
2445{
2604b703 2446 int err;
1da177e4
LT		 2447 int chunk_size;
2448 struct list_head *tmp;
2449 mdk_rdev_t *rdev;
2450 struct gendisk *disk;
2604b703 2451 struct mdk_personality *pers;
1da177e4
LT		 2452 char b[BDEVNAME_SIZE];
2453
a757e64c
N		 2454 if (list_empty(&mddev->disks))
2455 /* cannot run an array with no devices.. */
1da177e4 2456 return -EINVAL;
1da177e4
LT		 2457
2458 if (mddev->pers)
2459 return -EBUSY;
2460
2461 /*
2462 * Analyze all RAID superblock(s)
2463 */
a757e64c
N		 2464 if (!mddev->raid_disks)
2465 analyze_sbs(mddev);
1da177e4
LT		 2466
2467 chunk_size = mddev->chunk_size;
2604b703
N		 2468
2469 if (chunk_size) {
1da177e4
LT		 2470 if (chunk_size > MAX_CHUNK_SIZE) {
2471 printk(KERN_ERR "too big chunk_size: %d > %d\n",
2472 chunk_size, MAX_CHUNK_SIZE);
2473 return -EINVAL;
2474 }
2475 /*
2476 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2477 */
2478 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 2479 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 2480 return -EINVAL;
2481 }
2482 if (chunk_size < PAGE_SIZE) {
2483 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2484 chunk_size, PAGE_SIZE);
2485 return -EINVAL;
2486 }
2487
2488 /* devices must have minimum size of one chunk */
2489 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 2490 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2491 continue;
2492 if (rdev->size < chunk_size / 1024) {
2493 printk(KERN_WARNING
2494 "md: Dev %s smaller than chunk_size:"
2495 " %lluk < %dk\n",
2496 bdevname(rdev->bdev,b),
2497 (unsigned long long)rdev->size,
2498 chunk_size / 1024);
2499 return -EINVAL;
2500 }
2501 }
2502 }
2503
1da177e4 2504#ifdef CONFIG_KMOD
d9d166c2
N		 2505 if (mddev->level != LEVEL_NONE)
2506 request_module("md-level-%d", mddev->level);
2507 else if (mddev->clevel[0])
2508 request_module("md-%s", mddev->clevel);
1da177e4
LT		 2509#endif
2510
2511 /*
2512 * Drop all container device buffers, from now on
2513 * the only valid external interface is through the md
2514 * device.
2515 * Also find largest hardsector size
2516 */
2517 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 2518 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2519 continue;
2520 sync_blockdev(rdev->bdev);
2521 invalidate_bdev(rdev->bdev, 0);
2522 }
2523
2524 md_probe(mddev->unit, NULL, NULL);
2525 disk = mddev->gendisk;
2526 if (!disk)
2527 return -ENOMEM;
2528
2529 spin_lock(&pers_lock);
d9d166c2 2530 pers = find_pers(mddev->level, mddev->clevel);
2604b703 2531 if (!pers || !try_module_get(pers->owner)) {
1da177e4 2532 spin_unlock(&pers_lock);
d9d166c2
N		 2533 if (mddev->level != LEVEL_NONE)
2534 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2535 mddev->level);
2536 else
2537 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2538 mddev->clevel);
1da177e4
LT		 2539 return -EINVAL;
2540 }
2604b703 2541 mddev->pers = pers;
1da177e4 2542 spin_unlock(&pers_lock);
d9d166c2
N		 2543 mddev->level = pers->level;
2544 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 2545
657390d2 2546 mddev->recovery = 0;
1da177e4 2547 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 2548 mddev->barriers_work = 1;
6ff8d8ec 2549 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 2550
f91de92e
N		 2551 if (start_readonly)
2552 mddev->ro = 2; /* read-only, but switch on first write */
2553
b15c2e57
N		 2554 err = mddev->pers->run(mddev);
2555 if (!err && mddev->pers->sync_request) {
2556 err = bitmap_create(mddev);
2557 if (err) {
2558 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2559 mdname(mddev), err);
2560 mddev->pers->stop(mddev);
2561 }
2562 }
1da177e4
LT		 2563 if (err) {
2564 printk(KERN_ERR "md: pers->run() failed ...\n");
2565 module_put(mddev->pers->owner);
2566 mddev->pers = NULL;
32a7627c
N		 2567 bitmap_destroy(mddev);
2568 return err;
1da177e4 2569 }
411036fa
N		 2570 if (mddev->pers->sync_request)
2571 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
fd9d49ca
N		 2572 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2573 mddev->ro = 0;
2574
1da177e4
LT		 2575 atomic_set(&mddev->writes_pending,0);
2576 mddev->safemode = 0;
2577 mddev->safemode_timer.function = md_safemode_timeout;
2578 mddev->safemode_timer.data = (unsigned long) mddev;
2579 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2580 mddev->in_sync = 1;
86e6ffdd
N		 2581
2582 ITERATE_RDEV(mddev,rdev,tmp)
2583 if (rdev->raid_disk >= 0) {
2584 char nm[20];
2585 sprintf(nm, "rd%d", rdev->raid_disk);
2586 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2587 }
1da177e4
LT		 2588
2589 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 2590 md_wakeup_thread(mddev->thread);
1da177e4
LT		 2591
2592 if (mddev->sb_dirty)
2593 md_update_sb(mddev);
2594
2595 set_capacity(disk, mddev->array_size<<1);
2596
2597 /* If we call blk_queue_make_request here, it will
2598 * re-initialise max_sectors etc which may have been
2599 * refined inside -> run. So just set the bits we need to set.
2600 * Most initialisation happended when we called
2601 * blk_queue_make_request(..., md_fail_request)
2602 * earlier.
2603 */
2604 mddev->queue->queuedata = mddev;
2605 mddev->queue->make_request_fn = mddev->pers->make_request;
2606
2607 mddev->changed = 1;
d7603b7e 2608 md_new_event(mddev);
1da177e4
LT		 2609 return 0;
2610}
2611
2612static int restart_array(mddev_t *mddev)
2613{
2614 struct gendisk *disk = mddev->gendisk;
2615 int err;
2616
2617 /*
2618 * Complain if it has no devices
2619 */
2620 err = -ENXIO;
2621 if (list_empty(&mddev->disks))
2622 goto out;
2623
2624 if (mddev->pers) {
2625 err = -EBUSY;
2626 if (!mddev->ro)
2627 goto out;
2628
2629 mddev->safemode = 0;
2630 mddev->ro = 0;
2631 set_disk_ro(disk, 0);
2632
2633 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2634 mdname(mddev));
2635 /*
2636 * Kick recovery or resync if necessary
2637 */
2638 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2639 md_wakeup_thread(mddev->thread);
2640 err = 0;
2641 } else {
2642 printk(KERN_ERR "md: %s has no personality assigned.\n",
2643 mdname(mddev));
2644 err = -EINVAL;
2645 }
2646
2647out:
2648 return err;
2649}
2650
2651static int do_md_stop(mddev_t * mddev, int ro)
2652{
2653 int err = 0;
2654 struct gendisk *disk = mddev->gendisk;
2655
2656 if (mddev->pers) {
2657 if (atomic_read(&mddev->active)>2) {
2658 printk("md: %s still in use.\n",mdname(mddev));
2659 return -EBUSY;
2660 }
2661
2662 if (mddev->sync_thread) {
2663 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2664 md_unregister_thread(mddev->sync_thread);
2665 mddev->sync_thread = NULL;
2666 }
2667
2668 del_timer_sync(&mddev->safemode_timer);
2669
2670 invalidate_partition(disk, 0);
2671
2672 if (ro) {
2673 err = -ENXIO;
f91de92e 2674 if (mddev->ro==1)
1da177e4
LT		 2675 goto out;
2676 mddev->ro = 1;
2677 } else {
6b8b3e8a 2678 bitmap_flush(mddev);
a9701a30 2679 md_super_wait(mddev);
1da177e4
LT		 2680 if (mddev->ro)
2681 set_disk_ro(disk, 0);
2682 blk_queue_make_request(mddev->queue, md_fail_request);
2683 mddev->pers->stop(mddev);
411036fa
N		 2684 if (mddev->pers->sync_request)
2685 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2686
1da177e4
LT		 2687 module_put(mddev->pers->owner);
2688 mddev->pers = NULL;
2689 if (mddev->ro)
2690 mddev->ro = 0;
2691 }
2692 if (!mddev->in_sync) {
2693 /* mark array as shutdown cleanly */
2694 mddev->in_sync = 1;
2695 md_update_sb(mddev);
2696 }
2697 if (ro)
2698 set_disk_ro(disk, 1);
2699 }
32a7627c 2700
1da177e4
LT		 2701 /*
2702 * Free resources if final stop
2703 */
2704 if (!ro) {
86e6ffdd
N		 2705 mdk_rdev_t *rdev;
2706 struct list_head *tmp;
1da177e4
LT		 2707 struct gendisk *disk;
2708 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2709
978f946b
N		 2710 bitmap_destroy(mddev);
2711 if (mddev->bitmap_file) {
2712 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2713 fput(mddev->bitmap_file);
2714 mddev->bitmap_file = NULL;
2715 }
2716 mddev->bitmap_offset = 0;
2717
86e6ffdd
N		 2718 ITERATE_RDEV(mddev,rdev,tmp)
2719 if (rdev->raid_disk >= 0) {
2720 char nm[20];
2721 sprintf(nm, "rd%d", rdev->raid_disk);
2722 sysfs_remove_link(&mddev->kobj, nm);
2723 }
2724
1da177e4
LT		 2725 export_array(mddev);
2726
2727 mddev->array_size = 0;
2728 disk = mddev->gendisk;
2729 if (disk)
2730 set_capacity(disk, 0);
2731 mddev->changed = 1;
2732 } else
2733 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2734 mdname(mddev));
2735 err = 0;
d7603b7e 2736 md_new_event(mddev);
1da177e4
LT		 2737out:
2738 return err;
2739}
2740
2741static void autorun_array(mddev_t *mddev)
2742{
2743 mdk_rdev_t *rdev;
2744 struct list_head *tmp;
2745 int err;
2746
a757e64c 2747 if (list_empty(&mddev->disks))
1da177e4 2748 return;
1da177e4
LT		 2749
2750 printk(KERN_INFO "md: running: ");
2751
2752 ITERATE_RDEV(mddev,rdev,tmp) {
2753 char b[BDEVNAME_SIZE];
2754 printk("<%s>", bdevname(rdev->bdev,b));
2755 }
2756 printk("\n");
2757
2758 err = do_md_run (mddev);
2759 if (err) {
2760 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2761 do_md_stop (mddev, 0);
2762 }
2763}
2764
2765/*
2766 * lets try to run arrays based on all disks that have arrived
2767 * until now. (those are in pending_raid_disks)
2768 *
2769 * the method: pick the first pending disk, collect all disks with
2770 * the same UUID, remove all from the pending list and put them into
2771 * the 'same_array' list. Then order this list based on superblock
2772 * update time (freshest comes first), kick out 'old' disks and
2773 * compare superblocks. If everything's fine then run it.
2774 *
2775 * If "unit" is allocated, then bump its reference count
2776 */
2777static void autorun_devices(int part)
2778{
2779 struct list_head candidates;
2780 struct list_head *tmp;
2781 mdk_rdev_t *rdev0, *rdev;
2782 mddev_t *mddev;
2783 char b[BDEVNAME_SIZE];
2784
2785 printk(KERN_INFO "md: autorun ...\n");
2786 while (!list_empty(&pending_raid_disks)) {
2787 dev_t dev;
2788 rdev0 = list_entry(pending_raid_disks.next,
2789 mdk_rdev_t, same_set);
2790
2791 printk(KERN_INFO "md: considering %s ...\n",
2792 bdevname(rdev0->bdev,b));
2793 INIT_LIST_HEAD(&candidates);
2794 ITERATE_RDEV_PENDING(rdev,tmp)
2795 if (super_90_load(rdev, rdev0, 0) >= 0) {
2796 printk(KERN_INFO "md: adding %s ...\n",
2797 bdevname(rdev->bdev,b));
2798 list_move(&rdev->same_set, &candidates);
2799 }
2800 /*
2801 * now we have a set of devices, with all of them having
2802 * mostly sane superblocks. It's time to allocate the
2803 * mddev.
2804 */
2805 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2806 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2807 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2808 break;
2809 }
2810 if (part)
2811 dev = MKDEV(mdp_major,
2812 rdev0->preferred_minor << MdpMinorShift);
2813 else
2814 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2815
2816 md_probe(dev, NULL, NULL);
2817 mddev = mddev_find(dev);
2818 if (!mddev) {
2819 printk(KERN_ERR
2820 "md: cannot allocate memory for md drive.\n");
2821 break;
2822 }
2823 if (mddev_lock(mddev))
2824 printk(KERN_WARNING "md: %s locked, cannot run\n",
2825 mdname(mddev));
2826 else if (mddev->raid_disks || mddev->major_version
2827 || !list_empty(&mddev->disks)) {
2828 printk(KERN_WARNING
2829 "md: %s already running, cannot run %s\n",
2830 mdname(mddev), bdevname(rdev0->bdev,b));
2831 mddev_unlock(mddev);
2832 } else {
2833 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2834 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2835 list_del_init(&rdev->same_set);
2836 if (bind_rdev_to_array(rdev, mddev))
2837 export_rdev(rdev);
2838 }
2839 autorun_array(mddev);
2840 mddev_unlock(mddev);
2841 }
2842 /* on success, candidates will be empty, on error
2843 * it won't...
2844 */
2845 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2846 export_rdev(rdev);
2847 mddev_put(mddev);
2848 }
2849 printk(KERN_INFO "md: ... autorun DONE.\n");
2850}
2851
2852/*
2853 * import RAID devices based on one partition
2854 * if possible, the array gets run as well.
2855 */
2856
2857static int autostart_array(dev_t startdev)
2858{
2859 char b[BDEVNAME_SIZE];
2860 int err = -EINVAL, i;
2861 mdp_super_t *sb = NULL;
2862 mdk_rdev_t *start_rdev = NULL, *rdev;
2863
2864 start_rdev = md_import_device(startdev, 0, 0);
2865 if (IS_ERR(start_rdev))
2866 return err;
2867
2868
2869 /* NOTE: this can only work for 0.90.0 superblocks */
2870 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2871 if (sb->major_version != 0 ||
2872 sb->minor_version != 90 ) {
2873 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2874 export_rdev(start_rdev);
2875 return err;
2876 }
2877
b2d444d7 2878 if (test_bit(Faulty, &start_rdev->flags)) {
1da177e4
LT		 2879 printk(KERN_WARNING
2880 "md: can not autostart based on faulty %s!\n",
2881 bdevname(start_rdev->bdev,b));
2882 export_rdev(start_rdev);
2883 return err;
2884 }
2885 list_add(&start_rdev->same_set, &pending_raid_disks);
2886
2887 for (i = 0; i < MD_SB_DISKS; i++) {
2888 mdp_disk_t *desc = sb->disks + i;
2889 dev_t dev = MKDEV(desc->major, desc->minor);
2890
2891 if (!dev)
2892 continue;
2893 if (dev == startdev)
2894 continue;
2895 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2896 continue;
2897 rdev = md_import_device(dev, 0, 0);
2898 if (IS_ERR(rdev))
2899 continue;
2900
2901 list_add(&rdev->same_set, &pending_raid_disks);
2902 }
2903
2904 /*
2905 * possibly return codes
2906 */
2907 autorun_devices(0);
2908 return 0;
2909
2910}
2911
2912
2913static int get_version(void __user * arg)
2914{
2915 mdu_version_t ver;
2916
2917 ver.major = MD_MAJOR_VERSION;
2918 ver.minor = MD_MINOR_VERSION;
2919 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2920
2921 if (copy_to_user(arg, &ver, sizeof(ver)))
2922 return -EFAULT;
2923
2924 return 0;
2925}
2926
2927static int get_array_info(mddev_t * mddev, void __user * arg)
2928{
2929 mdu_array_info_t info;
2930 int nr,working,active,failed,spare;
2931 mdk_rdev_t *rdev;
2932 struct list_head *tmp;
2933
2934 nr=working=active=failed=spare=0;
2935 ITERATE_RDEV(mddev,rdev,tmp) {
2936 nr++;
b2d444d7 2937 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2938 failed++;
2939 else {
2940 working++;
b2d444d7 2941 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 2942 active++;
2943 else
2944 spare++;
2945 }
2946 }
2947
2948 info.major_version = mddev->major_version;
2949 info.minor_version = mddev->minor_version;
2950 info.patch_version = MD_PATCHLEVEL_VERSION;
2951 info.ctime = mddev->ctime;
2952 info.level = mddev->level;
2953 info.size = mddev->size;
284ae7ca
N		 2954 if (info.size != mddev->size) /* overflow */
2955 info.size = -1;
1da177e4
LT		 2956 info.nr_disks = nr;
2957 info.raid_disks = mddev->raid_disks;
2958 info.md_minor = mddev->md_minor;
2959 info.not_persistent= !mddev->persistent;
2960
2961 info.utime = mddev->utime;
2962 info.state = 0;
2963 if (mddev->in_sync)
2964 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 2965 if (mddev->bitmap && mddev->bitmap_offset)
2966 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 2967 info.active_disks = active;
2968 info.working_disks = working;
2969 info.failed_disks = failed;
2970 info.spare_disks = spare;
2971
2972 info.layout = mddev->layout;
2973 info.chunk_size = mddev->chunk_size;
2974
2975 if (copy_to_user(arg, &info, sizeof(info)))
2976 return -EFAULT;
2977
2978 return 0;
2979}
2980
87162a28 2981static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 2982{
2983 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2984 char *ptr, *buf = NULL;
2985 int err = -ENOMEM;
2986
2987 file = kmalloc(sizeof(*file), GFP_KERNEL);
2988 if (!file)
2989 goto out;
2990
2991 /* bitmap disabled, zero the first byte and copy out */
2992 if (!mddev->bitmap || !mddev->bitmap->file) {
2993 file->pathname[0] = '\0';
2994 goto copy_out;
2995 }
2996
2997 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2998 if (!buf)
2999 goto out;
3000
3001 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3002 if (!ptr)
3003 goto out;
3004
3005 strcpy(file->pathname, ptr);
3006
3007copy_out:
3008 err = 0;
3009 if (copy_to_user(arg, file, sizeof(*file)))
3010 err = -EFAULT;
3011out:
3012 kfree(buf);
3013 kfree(file);
3014 return err;
3015}
3016
1da177e4
LT		 3017static int get_disk_info(mddev_t * mddev, void __user * arg)
3018{
3019 mdu_disk_info_t info;
3020 unsigned int nr;
3021 mdk_rdev_t *rdev;
3022
3023 if (copy_from_user(&info, arg, sizeof(info)))
3024 return -EFAULT;
3025
3026 nr = info.number;
3027
3028 rdev = find_rdev_nr(mddev, nr);
3029 if (rdev) {
3030 info.major = MAJOR(rdev->bdev->bd_dev);
3031 info.minor = MINOR(rdev->bdev->bd_dev);
3032 info.raid_disk = rdev->raid_disk;
3033 info.state = 0;
b2d444d7 3034 if (test_bit(Faulty, &rdev->flags))
1da177e4 3035 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 3036 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 3037 info.state |= (1<<MD_DISK_ACTIVE);
3038 info.state |= (1<<MD_DISK_SYNC);
3039 }
8ddf9efe
N		 3040 if (test_bit(WriteMostly, &rdev->flags))
3041 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 3042 } else {
3043 info.major = info.minor = 0;
3044 info.raid_disk = -1;
3045 info.state = (1<<MD_DISK_REMOVED);
3046 }
3047
3048 if (copy_to_user(arg, &info, sizeof(info)))
3049 return -EFAULT;
3050
3051 return 0;
3052}
3053
3054static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3055{
3056 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3057 mdk_rdev_t *rdev;
3058 dev_t dev = MKDEV(info->major,info->minor);
3059
3060 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3061 return -EOVERFLOW;
3062
3063 if (!mddev->raid_disks) {
3064 int err;
3065 /* expecting a device which has a superblock */
3066 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3067 if (IS_ERR(rdev)) {
3068 printk(KERN_WARNING
3069 "md: md_import_device returned %ld\n",
3070 PTR_ERR(rdev));
3071 return PTR_ERR(rdev);
3072 }
3073 if (!list_empty(&mddev->disks)) {
3074 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3075 mdk_rdev_t, same_set);
3076 int err = super_types[mddev->major_version]
3077 .load_super(rdev, rdev0, mddev->minor_version);
3078 if (err < 0) {
3079 printk(KERN_WARNING
3080 "md: %s has different UUID to %s\n",
3081 bdevname(rdev->bdev,b),
3082 bdevname(rdev0->bdev,b2));
3083 export_rdev(rdev);
3084 return -EINVAL;
3085 }
3086 }
3087 err = bind_rdev_to_array(rdev, mddev);
3088 if (err)
3089 export_rdev(rdev);
3090 return err;
3091 }
3092
3093 /*
3094 * add_new_disk can be used once the array is assembled
3095 * to add "hot spares". They must already have a superblock
3096 * written
3097 */
3098 if (mddev->pers) {
3099 int err;
3100 if (!mddev->pers->hot_add_disk) {
3101 printk(KERN_WARNING
3102 "%s: personality does not support diskops!\n",
3103 mdname(mddev));
3104 return -EINVAL;
3105 }
7b1e35f6
N		 3106 if (mddev->persistent)
3107 rdev = md_import_device(dev, mddev->major_version,
3108 mddev->minor_version);
3109 else
3110 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 3111 if (IS_ERR(rdev)) {
3112 printk(KERN_WARNING
3113 "md: md_import_device returned %ld\n",
3114 PTR_ERR(rdev));
3115 return PTR_ERR(rdev);
3116 }
41158c7e
N		 3117 /* set save_raid_disk if appropriate */
3118 if (!mddev->persistent) {
3119 if (info->state & (1<<MD_DISK_SYNC) &&
3120 info->raid_disk < mddev->raid_disks)
3121 rdev->raid_disk = info->raid_disk;
3122 else
3123 rdev->raid_disk = -1;
3124 } else
3125 super_types[mddev->major_version].
3126 validate_super(mddev, rdev);
3127 rdev->saved_raid_disk = rdev->raid_disk;
3128
b2d444d7 3129 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 3130 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3131 set_bit(WriteMostly, &rdev->flags);
3132
1da177e4
LT		 3133 rdev->raid_disk = -1;
3134 err = bind_rdev_to_array(rdev, mddev);
3135 if (err)
3136 export_rdev(rdev);
c361777f
N		 3137
3138 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 3139 md_wakeup_thread(mddev->thread);
1da177e4
LT		 3140 return err;
3141 }
3142
3143 /* otherwise, add_new_disk is only allowed
3144 * for major_version==0 superblocks
3145 */
3146 if (mddev->major_version != 0) {
3147 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3148 mdname(mddev));
3149 return -EINVAL;
3150 }
3151
3152 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3153 int err;
3154 rdev = md_import_device (dev, -1, 0);
3155 if (IS_ERR(rdev)) {
3156 printk(KERN_WARNING
3157 "md: error, md_import_device() returned %ld\n",
3158 PTR_ERR(rdev));
3159 return PTR_ERR(rdev);
3160 }
3161 rdev->desc_nr = info->number;
3162 if (info->raid_disk < mddev->raid_disks)
3163 rdev->raid_disk = info->raid_disk;
3164 else
3165 rdev->raid_disk = -1;
3166
b2d444d7
N		 3167 rdev->flags = 0;
3168
1da177e4 3169 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 3170 if (info->state & (1<<MD_DISK_SYNC))
3171 set_bit(In_sync, &rdev->flags);
1da177e4 3172
8ddf9efe
N		 3173 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3174 set_bit(WriteMostly, &rdev->flags);
3175
1da177e4
LT		 3176 if (!mddev->persistent) {
3177 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3178 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3179 } else
3180 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3181 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3182
2bf071bf
N		 3183 err = bind_rdev_to_array(rdev, mddev);
3184 if (err) {
3185 export_rdev(rdev);
3186 return err;
3187 }
1da177e4
LT		 3188 }
3189
3190 return 0;
3191}
3192
3193static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3194{
3195 char b[BDEVNAME_SIZE];
3196 mdk_rdev_t *rdev;
3197
3198 if (!mddev->pers)
3199 return -ENODEV;
3200
3201 rdev = find_rdev(mddev, dev);
3202 if (!rdev)
3203 return -ENXIO;
3204
3205 if (rdev->raid_disk >= 0)
3206 goto busy;
3207
3208 kick_rdev_from_array(rdev);
3209 md_update_sb(mddev);
d7603b7e 3210 md_new_event(mddev);
1da177e4
LT		 3211
3212 return 0;
3213busy:
3214 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3215 bdevname(rdev->bdev,b), mdname(mddev));
3216 return -EBUSY;
3217}
3218
3219static int hot_add_disk(mddev_t * mddev, dev_t dev)
3220{
3221 char b[BDEVNAME_SIZE];
3222 int err;
3223 unsigned int size;
3224 mdk_rdev_t *rdev;
3225
3226 if (!mddev->pers)
3227 return -ENODEV;
3228
3229 if (mddev->major_version != 0) {
3230 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3231 " version-0 superblocks.\n",
3232 mdname(mddev));
3233 return -EINVAL;
3234 }
3235 if (!mddev->pers->hot_add_disk) {
3236 printk(KERN_WARNING
3237 "%s: personality does not support diskops!\n",
3238 mdname(mddev));
3239 return -EINVAL;
3240 }
3241
3242 rdev = md_import_device (dev, -1, 0);
3243 if (IS_ERR(rdev)) {
3244 printk(KERN_WARNING
3245 "md: error, md_import_device() returned %ld\n",
3246 PTR_ERR(rdev));
3247 return -EINVAL;
3248 }
3249
3250 if (mddev->persistent)
3251 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3252 else
3253 rdev->sb_offset =
3254 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3255
3256 size = calc_dev_size(rdev, mddev->chunk_size);
3257 rdev->size = size;
3258
b2d444d7 3259 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 3260 printk(KERN_WARNING
3261 "md: can not hot-add faulty %s disk to %s!\n",
3262 bdevname(rdev->bdev,b), mdname(mddev));
3263 err = -EINVAL;
3264 goto abort_export;
3265 }
b2d444d7 3266 clear_bit(In_sync, &rdev->flags);
1da177e4 3267 rdev->desc_nr = -1;
2bf071bf
N		 3268 err = bind_rdev_to_array(rdev, mddev);
3269 if (err)
3270 goto abort_export;
1da177e4
LT		 3271
3272 /*
3273 * The rest should better be atomic, we can have disk failures
3274 * noticed in interrupt contexts ...
3275 */
3276
3277 if (rdev->desc_nr == mddev->max_disks) {
3278 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3279 mdname(mddev));
3280 err = -EBUSY;
3281 goto abort_unbind_export;
3282 }
3283
3284 rdev->raid_disk = -1;
3285
3286 md_update_sb(mddev);
3287
3288 /*
3289 * Kick recovery, maybe this spare has to be added to the
3290 * array immediately.
3291 */
3292 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3293 md_wakeup_thread(mddev->thread);
d7603b7e 3294 md_new_event(mddev);
1da177e4
LT		 3295 return 0;
3296
3297abort_unbind_export:
3298 unbind_rdev_from_array(rdev);
3299
3300abort_export:
3301 export_rdev(rdev);
3302 return err;
3303}
3304
32a7627c
N		 3305/* similar to deny_write_access, but accounts for our holding a reference
3306 * to the file ourselves */
3307static int deny_bitmap_write_access(struct file * file)
3308{
3309 struct inode *inode = file->f_mapping->host;
3310
3311 spin_lock(&inode->i_lock);
3312 if (atomic_read(&inode->i_writecount) > 1) {
3313 spin_unlock(&inode->i_lock);
3314 return -ETXTBSY;
3315 }
3316 atomic_set(&inode->i_writecount, -1);
3317 spin_unlock(&inode->i_lock);
3318
3319 return 0;
3320}
3321
3322static int set_bitmap_file(mddev_t *mddev, int fd)
3323{
3324 int err;
3325
36fa3063
N		 3326 if (mddev->pers) {
3327 if (!mddev->pers->quiesce)
3328 return -EBUSY;
3329 if (mddev->recovery || mddev->sync_thread)
3330 return -EBUSY;
3331 /* we should be able to change the bitmap.. */
3332 }
32a7627c 3333
32a7627c 3334
36fa3063
N		 3335 if (fd >= 0) {
3336 if (mddev->bitmap)
3337 return -EEXIST; /* cannot add when bitmap is present */
3338 mddev->bitmap_file = fget(fd);
32a7627c 3339
36fa3063
N		 3340 if (mddev->bitmap_file == NULL) {
3341 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3342 mdname(mddev));
3343 return -EBADF;
3344 }
3345
3346 err = deny_bitmap_write_access(mddev->bitmap_file);
3347 if (err) {
3348 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3349 mdname(mddev));
3350 fput(mddev->bitmap_file);
3351 mddev->bitmap_file = NULL;
3352 return err;
3353 }
a654b9d8 3354 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 3355 } else if (mddev->bitmap == NULL)
3356 return -ENOENT; /* cannot remove what isn't there */
3357 err = 0;
3358 if (mddev->pers) {
3359 mddev->pers->quiesce(mddev, 1);
3360 if (fd >= 0)
3361 err = bitmap_create(mddev);
3362 if (fd < 0 || err)
3363 bitmap_destroy(mddev);
3364 mddev->pers->quiesce(mddev, 0);
3365 } else if (fd < 0) {
3366 if (mddev->bitmap_file)
3367 fput(mddev->bitmap_file);
3368 mddev->bitmap_file = NULL;
3369 }
3370
32a7627c
N		 3371 return err;
3372}
3373
1da177e4
LT		 3374/*
3375 * set_array_info is used two different ways
3376 * The original usage is when creating a new array.
3377 * In this usage, raid_disks is > 0 and it together with
3378 * level, size, not_persistent,layout,chunksize determine the
3379 * shape of the array.
3380 * This will always create an array with a type-0.90.0 superblock.
3381 * The newer usage is when assembling an array.
3382 * In this case raid_disks will be 0, and the major_version field is
3383 * use to determine which style super-blocks are to be found on the devices.
3384 * The minor and patch _version numbers are also kept incase the
3385 * super_block handler wishes to interpret them.
3386 */
3387static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3388{
3389
3390 if (info->raid_disks == 0) {
3391 /* just setting version number for superblock loading */
3392 if (info->major_version < 0 ||
3393 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3394 super_types[info->major_version].name == NULL) {
3395 /* maybe try to auto-load a module? */
3396 printk(KERN_INFO
3397 "md: superblock version %d not known\n",
3398 info->major_version);
3399 return -EINVAL;
3400 }
3401 mddev->major_version = info->major_version;
3402 mddev->minor_version = info->minor_version;
3403 mddev->patch_version = info->patch_version;
3404 return 0;
3405 }
3406 mddev->major_version = MD_MAJOR_VERSION;
3407 mddev->minor_version = MD_MINOR_VERSION;
3408 mddev->patch_version = MD_PATCHLEVEL_VERSION;
3409 mddev->ctime = get_seconds();
3410
3411 mddev->level = info->level;
17115e03 3412 mddev->clevel[0] = 0;
1da177e4
LT		 3413 mddev->size = info->size;
3414 mddev->raid_disks = info->raid_disks;
3415 /* don't set md_minor, it is determined by which /dev/md* was
3416 * openned
3417 */
3418 if (info->state & (1<<MD_SB_CLEAN))
3419 mddev->recovery_cp = MaxSector;
3420 else
3421 mddev->recovery_cp = 0;
3422 mddev->persistent = ! info->not_persistent;
3423
3424 mddev->layout = info->layout;
3425 mddev->chunk_size = info->chunk_size;
3426
3427 mddev->max_disks = MD_SB_DISKS;
3428
3429 mddev->sb_dirty = 1;
3430
b2a2703c
N		 3431 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3432 mddev->bitmap_offset = 0;
3433
1da177e4
LT		 3434 /*
3435 * Generate a 128 bit UUID
3436 */
3437 get_random_bytes(mddev->uuid, 16);
3438
3439 return 0;
3440}
3441
a35b0d69
N		 3442static int update_size(mddev_t *mddev, unsigned long size)
3443{
3444 mdk_rdev_t * rdev;
3445 int rv;
3446 struct list_head *tmp;
3447
3448 if (mddev->pers->resize == NULL)
3449 return -EINVAL;
3450 /* The "size" is the amount of each device that is used.
3451 * This can only make sense for arrays with redundancy.
3452 * linear and raid0 always use whatever space is available
3453 * We can only consider changing the size if no resync
3454 * or reconstruction is happening, and if the new size
3455 * is acceptable. It must fit before the sb_offset or,
3456 * if that is <data_offset, it must fit before the
3457 * size of each device.
3458 * If size is zero, we find the largest size that fits.
3459 */
3460 if (mddev->sync_thread)
3461 return -EBUSY;
3462 ITERATE_RDEV(mddev,rdev,tmp) {
3463 sector_t avail;
3464 int fit = (size == 0);
3465 if (rdev->sb_offset > rdev->data_offset)
3466 avail = (rdev->sb_offset*2) - rdev->data_offset;
3467 else
3468 avail = get_capacity(rdev->bdev->bd_disk)
3469 - rdev->data_offset;
3470 if (fit && (size == 0 || size > avail/2))
3471 size = avail/2;
3472 if (avail < ((sector_t)size << 1))
3473 return -ENOSPC;
3474 }
3475 rv = mddev->pers->resize(mddev, (sector_t)size *2);
3476 if (!rv) {
3477 struct block_device *bdev;
3478
3479 bdev = bdget_disk(mddev->gendisk, 0);
3480 if (bdev) {
1b1dcc1b 3481 mutex_lock(&bdev->bd_inode->i_mutex);
6d89332b 3482 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
1b1dcc1b 3483 mutex_unlock(&bdev->bd_inode->i_mutex);
a35b0d69
N		 3484 bdput(bdev);
3485 }
3486 }
3487 return rv;
3488}
3489
da943b99
N		 3490static int update_raid_disks(mddev_t *mddev, int raid_disks)
3491{
3492 int rv;
3493 /* change the number of raid disks */
3494 if (mddev->pers->reshape == NULL)
3495 return -EINVAL;
3496 if (raid_disks <= 0 ||
3497 raid_disks >= mddev->max_disks)
3498 return -EINVAL;
3499 if (mddev->sync_thread)
3500 return -EBUSY;
3501 rv = mddev->pers->reshape(mddev, raid_disks);
da943b99
N		 3502 return rv;
3503}
3504
3505
1da177e4
LT		 3506/*
3507 * update_array_info is used to change the configuration of an
3508 * on-line array.
3509 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3510 * fields in the info are checked against the array.
3511 * Any differences that cannot be handled will cause an error.
3512 * Normally, only one change can be managed at a time.
3513 */
3514static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3515{
3516 int rv = 0;
3517 int cnt = 0;
36fa3063
N		 3518 int state = 0;
3519
3520 /* calculate expected state,ignoring low bits */
3521 if (mddev->bitmap && mddev->bitmap_offset)
3522 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 3523
3524 if (mddev->major_version != info->major_version ||
3525 mddev->minor_version != info->minor_version ||
3526/* mddev->patch_version != info->patch_version || */
3527 mddev->ctime != info->ctime ||
3528 mddev->level != info->level ||
3529/* mddev->layout != info->layout || */
3530 !mddev->persistent != info->not_persistent||
36fa3063
N		 3531 mddev->chunk_size != info->chunk_size ||
3532 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3533 ((state^info->state) & 0xfffffe00)
3534 )
1da177e4
LT		 3535 return -EINVAL;
3536 /* Check there is only one change */
284ae7ca 3537 if (info->size >= 0 && mddev->size != info->size) cnt++;
1da177e4
LT		 3538 if (mddev->raid_disks != info->raid_disks) cnt++;
3539 if (mddev->layout != info->layout) cnt++;
36fa3063 3540 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 3541 if (cnt == 0) return 0;
3542 if (cnt > 1) return -EINVAL;
3543
3544 if (mddev->layout != info->layout) {
3545 /* Change layout
3546 * we don't need to do anything at the md level, the
3547 * personality will take care of it all.
3548 */
3549 if (mddev->pers->reconfig == NULL)
3550 return -EINVAL;
3551 else
3552 return mddev->pers->reconfig(mddev, info->layout, -1);
3553 }
284ae7ca 3554 if (info->size >= 0 && mddev->size != info->size)
a35b0d69
N		 3555 rv = update_size(mddev, info->size);
3556
da943b99
N		 3557 if (mddev->raid_disks != info->raid_disks)
3558 rv = update_raid_disks(mddev, info->raid_disks);
3559
36fa3063
N		 3560 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3561 if (mddev->pers->quiesce == NULL)
3562 return -EINVAL;
3563 if (mddev->recovery || mddev->sync_thread)
3564 return -EBUSY;
3565 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3566 /* add the bitmap */
3567 if (mddev->bitmap)
3568 return -EEXIST;
3569 if (mddev->default_bitmap_offset == 0)
3570 return -EINVAL;
3571 mddev->bitmap_offset = mddev->default_bitmap_offset;
3572 mddev->pers->quiesce(mddev, 1);
3573 rv = bitmap_create(mddev);
3574 if (rv)
3575 bitmap_destroy(mddev);
3576 mddev->pers->quiesce(mddev, 0);
3577 } else {
3578 /* remove the bitmap */
3579 if (!mddev->bitmap)
3580 return -ENOENT;
3581 if (mddev->bitmap->file)
3582 return -EINVAL;
3583 mddev->pers->quiesce(mddev, 1);
3584 bitmap_destroy(mddev);
3585 mddev->pers->quiesce(mddev, 0);
3586 mddev->bitmap_offset = 0;
3587 }
3588 }
1da177e4
LT		 3589 md_update_sb(mddev);
3590 return rv;
3591}
3592
3593static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3594{
3595 mdk_rdev_t *rdev;
3596
3597 if (mddev->pers == NULL)
3598 return -ENODEV;
3599
3600 rdev = find_rdev(mddev, dev);
3601 if (!rdev)
3602 return -ENODEV;
3603
3604 md_error(mddev, rdev);
3605 return 0;
3606}
3607
a885c8c4
CH		 3608static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
3609{
3610 mddev_t *mddev = bdev->bd_disk->private_data;
3611
3612 geo->heads = 2;
3613 geo->sectors = 4;
3614 geo->cylinders = get_capacity(mddev->gendisk) / 8;
3615 return 0;
3616}
3617
1da177e4
LT		 3618static int md_ioctl(struct inode *inode, struct file *file,
3619 unsigned int cmd, unsigned long arg)
3620{
3621 int err = 0;
3622 void __user *argp = (void __user *)arg;
1da177e4
LT		 3623 mddev_t *mddev = NULL;
3624
3625 if (!capable(CAP_SYS_ADMIN))
3626 return -EACCES;
3627
3628 /*
3629 * Commands dealing with the RAID driver but not any
3630 * particular array:
3631 */
3632 switch (cmd)
3633 {
3634 case RAID_VERSION:
3635 err = get_version(argp);
3636 goto done;
3637
3638 case PRINT_RAID_DEBUG:
3639 err = 0;
3640 md_print_devices();
3641 goto done;
3642
3643#ifndef MODULE
3644 case RAID_AUTORUN:
3645 err = 0;
3646 autostart_arrays(arg);
3647 goto done;
3648#endif
3649 default:;
3650 }
3651
3652 /*
3653 * Commands creating/starting a new array:
3654 */
3655
3656 mddev = inode->i_bdev->bd_disk->private_data;
3657
3658 if (!mddev) {
3659 BUG();
3660 goto abort;
3661 }
3662
3663
3664 if (cmd == START_ARRAY) {
3665 /* START_ARRAY doesn't need to lock the array as autostart_array
3666 * does the locking, and it could even be a different array
3667 */
3668 static int cnt = 3;
3669 if (cnt > 0 ) {
3670 printk(KERN_WARNING
3671 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
e8a00334 3672 "This will not be supported beyond July 2006\n",
1da177e4
LT		 3673 current->comm, current->pid);
3674 cnt--;
3675 }
3676 err = autostart_array(new_decode_dev(arg));
3677 if (err) {
3678 printk(KERN_WARNING "md: autostart failed!\n");
3679 goto abort;
3680 }
3681 goto done;
3682 }
3683
3684 err = mddev_lock(mddev);
3685 if (err) {
3686 printk(KERN_INFO
3687 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3688 err, cmd);
3689 goto abort;
3690 }
3691
3692 switch (cmd)
3693 {
3694 case SET_ARRAY_INFO:
3695 {
3696 mdu_array_info_t info;
3697 if (!arg)
3698 memset(&info, 0, sizeof(info));
3699 else if (copy_from_user(&info, argp, sizeof(info))) {
3700 err = -EFAULT;
3701 goto abort_unlock;
3702 }
3703 if (mddev->pers) {
3704 err = update_array_info(mddev, &info);
3705 if (err) {
3706 printk(KERN_WARNING "md: couldn't update"
3707 " array info. %d\n", err);
3708 goto abort_unlock;
3709 }
3710 goto done_unlock;
3711 }
3712 if (!list_empty(&mddev->disks)) {
3713 printk(KERN_WARNING
3714 "md: array %s already has disks!\n",
3715 mdname(mddev));
3716 err = -EBUSY;
3717 goto abort_unlock;
3718 }
3719 if (mddev->raid_disks) {
3720 printk(KERN_WARNING
3721 "md: array %s already initialised!\n",
3722 mdname(mddev));
3723 err = -EBUSY;
3724 goto abort_unlock;
3725 }
3726 err = set_array_info(mddev, &info);
3727 if (err) {
3728 printk(KERN_WARNING "md: couldn't set"
3729 " array info. %d\n", err);
3730 goto abort_unlock;
3731 }
3732 }
3733 goto done_unlock;
3734
3735 default:;
3736 }
3737
3738 /*
3739 * Commands querying/configuring an existing array:
3740 */
32a7627c
N		 3741 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3742 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3743 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3744 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
1da177e4
LT		 3745 err = -ENODEV;
3746 goto abort_unlock;
3747 }
3748
3749 /*
3750 * Commands even a read-only array can execute:
3751 */
3752 switch (cmd)
3753 {
3754 case GET_ARRAY_INFO:
3755 err = get_array_info(mddev, argp);
3756 goto done_unlock;
3757
32a7627c 3758 case GET_BITMAP_FILE:
87162a28 3759 err = get_bitmap_file(mddev, argp);
32a7627c
N		 3760 goto done_unlock;
3761
1da177e4
LT		 3762 case GET_DISK_INFO:
3763 err = get_disk_info(mddev, argp);
3764 goto done_unlock;
3765
3766 case RESTART_ARRAY_RW:
3767 err = restart_array(mddev);
3768 goto done_unlock;
3769
3770 case STOP_ARRAY:
3771 err = do_md_stop (mddev, 0);
3772 goto done_unlock;
3773
3774 case STOP_ARRAY_RO:
3775 err = do_md_stop (mddev, 1);
3776 goto done_unlock;
3777
3778 /*
3779 * We have a problem here : there is no easy way to give a CHS
3780 * virtual geometry. We currently pretend that we have a 2 heads
3781 * 4 sectors (with a BIG number of cylinders...). This drives
3782 * dosfs just mad... ;-)
3783 */
1da177e4
LT		 3784 }
3785
3786 /*
3787 * The remaining ioctls are changing the state of the
f91de92e
N		 3788 * superblock, so we do not allow them on read-only arrays.
3789 * However non-MD ioctls (e.g. get-size) will still come through
3790 * here and hit the 'default' below, so only disallow
3791 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 3792 */
f91de92e
N		 3793 if (_IOC_TYPE(cmd) == MD_MAJOR &&
3794 mddev->ro && mddev->pers) {
3795 if (mddev->ro == 2) {
3796 mddev->ro = 0;
3797 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3798 md_wakeup_thread(mddev->thread);
3799
3800 } else {
3801 err = -EROFS;
3802 goto abort_unlock;
3803 }
1da177e4
LT		 3804 }
3805
3806 switch (cmd)
3807 {
3808 case ADD_NEW_DISK:
3809 {
3810 mdu_disk_info_t info;
3811 if (copy_from_user(&info, argp, sizeof(info)))
3812 err = -EFAULT;
3813 else
3814 err = add_new_disk(mddev, &info);
3815 goto done_unlock;
3816 }
3817
3818 case HOT_REMOVE_DISK:
3819 err = hot_remove_disk(mddev, new_decode_dev(arg));
3820 goto done_unlock;
3821
3822 case HOT_ADD_DISK:
3823 err = hot_add_disk(mddev, new_decode_dev(arg));
3824 goto done_unlock;
3825
3826 case SET_DISK_FAULTY:
3827 err = set_disk_faulty(mddev, new_decode_dev(arg));
3828 goto done_unlock;
3829
3830 case RUN_ARRAY:
3831 err = do_md_run (mddev);
3832 goto done_unlock;
3833
32a7627c
N		 3834 case SET_BITMAP_FILE:
3835 err = set_bitmap_file(mddev, (int)arg);
3836 goto done_unlock;
3837
1da177e4
LT		 3838 default:
3839 if (_IOC_TYPE(cmd) == MD_MAJOR)
3840 printk(KERN_WARNING "md: %s(pid %d) used"
3841 " obsolete MD ioctl, upgrade your"
3842 " software to use new ictls.\n",
3843 current->comm, current->pid);
3844 err = -EINVAL;
3845 goto abort_unlock;
3846 }
3847
3848done_unlock:
3849abort_unlock:
3850 mddev_unlock(mddev);
3851
3852 return err;
3853done:
3854 if (err)
3855 MD_BUG();
3856abort:
3857 return err;
3858}
3859
3860static int md_open(struct inode *inode, struct file *file)
3861{
3862 /*
3863 * Succeed if we can lock the mddev, which confirms that
3864 * it isn't being stopped right now.
3865 */
3866 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3867 int err;
3868
3869 if ((err = mddev_lock(mddev)))
3870 goto out;
3871
3872 err = 0;
3873 mddev_get(mddev);
3874 mddev_unlock(mddev);
3875
3876 check_disk_change(inode->i_bdev);
3877 out:
3878 return err;
3879}
3880
3881static int md_release(struct inode *inode, struct file * file)
3882{
3883 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3884
3885 if (!mddev)
3886 BUG();
3887 mddev_put(mddev);
3888
3889 return 0;
3890}
3891
3892static int md_media_changed(struct gendisk *disk)
3893{
3894 mddev_t *mddev = disk->private_data;
3895
3896 return mddev->changed;
3897}
3898
3899static int md_revalidate(struct gendisk *disk)
3900{
3901 mddev_t *mddev = disk->private_data;
3902
3903 mddev->changed = 0;
3904 return 0;
3905}
3906static struct block_device_operations md_fops =
3907{
3908 .owner = THIS_MODULE,
3909 .open = md_open,
3910 .release = md_release,
3911 .ioctl = md_ioctl,
a885c8c4 3912 .getgeo = md_getgeo,
1da177e4
LT		 3913 .media_changed = md_media_changed,
3914 .revalidate_disk= md_revalidate,
3915};
3916
75c96f85 3917static int md_thread(void * arg)
1da177e4
LT		 3918{
3919 mdk_thread_t *thread = arg;
3920
1da177e4
LT		 3921 /*
3922 * md_thread is a 'system-thread', it's priority should be very
3923 * high. We avoid resource deadlocks individually in each
3924 * raid personality. (RAID5 does preallocation) We also use RR and
3925 * the very same RT priority as kswapd, thus we will never get
3926 * into a priority inversion deadlock.
3927 *
3928 * we definitely have to have equal or higher priority than
3929 * bdflush, otherwise bdflush will deadlock if there are too
3930 * many dirty RAID5 blocks.
3931 */
1da177e4 3932
6985c43f 3933 allow_signal(SIGKILL);
a6fb0934 3934 while (!kthread_should_stop()) {
1da177e4 3935
93588e22
N		 3936 /* We need to wait INTERRUPTIBLE so that
3937 * we don't add to the load-average.
3938 * That means we need to be sure no signals are
3939 * pending
3940 */
3941 if (signal_pending(current))
3942 flush_signals(current);
3943
3944 wait_event_interruptible_timeout
3945 (thread->wqueue,
3946 test_bit(THREAD_WAKEUP, &thread->flags)
3947 || kthread_should_stop(),
3948 thread->timeout);
3e1d1d28 3949 try_to_freeze();
1da177e4
LT		 3950
3951 clear_bit(THREAD_WAKEUP, &thread->flags);
3952
787453c2 3953 thread->run(thread->mddev);
1da177e4 3954 }
a6fb0934 3955
1da177e4
LT		 3956 return 0;
3957}
3958
3959void md_wakeup_thread(mdk_thread_t *thread)
3960{
3961 if (thread) {
3962 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3963 set_bit(THREAD_WAKEUP, &thread->flags);
3964 wake_up(&thread->wqueue);
3965 }
3966}
3967
3968mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3969 const char *name)
3970{
3971 mdk_thread_t *thread;
1da177e4 3972
9ffae0cf 3973 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 3974 if (!thread)
3975 return NULL;
3976
1da177e4
LT		 3977 init_waitqueue_head(&thread->wqueue);
3978
1da177e4
LT		 3979 thread->run = run;
3980 thread->mddev = mddev;
32a7627c 3981 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 3982 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 3983 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 3984 kfree(thread);
3985 return NULL;
3986 }
1da177e4
LT		 3987 return thread;
3988}
3989
1da177e4
LT		 3990void md_unregister_thread(mdk_thread_t *thread)
3991{
d28446fe 3992 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
a6fb0934
N		 3993
3994 kthread_stop(thread->tsk);
1da177e4
LT		 3995 kfree(thread);
3996}
3997
3998void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3999{
4000 if (!mddev) {
4001 MD_BUG();
4002 return;
4003 }
4004
b2d444d7 4005 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 4006 return;
32a7627c 4007/*
1da177e4
LT		 4008 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4009 mdname(mddev),
4010 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4011 __builtin_return_address(0),__builtin_return_address(1),
4012 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 4013*/
1da177e4
LT		 4014 if (!mddev->pers->error_handler)
4015 return;
4016 mddev->pers->error_handler(mddev,rdev);
4017 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4018 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4019 md_wakeup_thread(mddev->thread);
d7603b7e 4020 md_new_event(mddev);
1da177e4
LT		 4021}
4022
4023/* seq_file implementation /proc/mdstat */
4024
4025static void status_unused(struct seq_file *seq)
4026{
4027 int i = 0;
4028 mdk_rdev_t *rdev;
4029 struct list_head *tmp;
4030
4031 seq_printf(seq, "unused devices: ");
4032
4033 ITERATE_RDEV_PENDING(rdev,tmp) {
4034 char b[BDEVNAME_SIZE];
4035 i++;
4036 seq_printf(seq, "%s ",
4037 bdevname(rdev->bdev,b));
4038 }
4039 if (!i)
4040 seq_printf(seq, "<none>");
4041
4042 seq_printf(seq, "\n");
4043}
4044
4045
4046static void status_resync(struct seq_file *seq, mddev_t * mddev)
4047{
4048 unsigned long max_blocks, resync, res, dt, db, rt;
4049
4050 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4051
4052 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4053 max_blocks = mddev->resync_max_sectors >> 1;
4054 else
4055 max_blocks = mddev->size;
4056
4057 /*
4058 * Should not happen.
4059 */
4060 if (!max_blocks) {
4061 MD_BUG();
4062 return;
4063 }
4064 res = (resync/1024)*1000/(max_blocks/1024 + 1);
4065 {
4066 int i, x = res/50, y = 20-x;
4067 seq_printf(seq, "[");
4068 for (i = 0; i < x; i++)
4069 seq_printf(seq, "=");
4070 seq_printf(seq, ">");
4071 for (i = 0; i < y; i++)
4072 seq_printf(seq, ".");
4073 seq_printf(seq, "] ");
4074 }
4075 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
4076 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4077 "resync" : "recovery"),
4078 res/10, res % 10, resync, max_blocks);
4079
4080 /*
4081 * We do not want to overflow, so the order of operands and
4082 * the * 100 / 100 trick are important. We do a +1 to be
4083 * safe against division by zero. We only estimate anyway.
4084 *
4085 * dt: time from mark until now
4086 * db: blocks written from mark until now
4087 * rt: remaining time
4088 */
4089 dt = ((jiffies - mddev->resync_mark) / HZ);
4090 if (!dt) dt++;
4091 db = resync - (mddev->resync_mark_cnt/2);
4092 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
4093
4094 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4095
4096 seq_printf(seq, " speed=%ldK/sec", db/dt);
4097}
4098
4099static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4100{
4101 struct list_head *tmp;
4102 loff_t l = *pos;
4103 mddev_t *mddev;
4104
4105 if (l >= 0x10000)
4106 return NULL;
4107 if (!l--)
4108 /* header */
4109 return (void*)1;
4110
4111 spin_lock(&all_mddevs_lock);
4112 list_for_each(tmp,&all_mddevs)
4113 if (!l--) {
4114 mddev = list_entry(tmp, mddev_t, all_mddevs);
4115 mddev_get(mddev);
4116 spin_unlock(&all_mddevs_lock);
4117 return mddev;
4118 }
4119 spin_unlock(&all_mddevs_lock);
4120 if (!l--)
4121 return (void*)2;/* tail */
4122 return NULL;
4123}
4124
4125static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4126{
4127 struct list_head *tmp;
4128 mddev_t *next_mddev, *mddev = v;
4129
4130 ++*pos;
4131 if (v == (void*)2)
4132 return NULL;
4133
4134 spin_lock(&all_mddevs_lock);
4135 if (v == (void*)1)
4136 tmp = all_mddevs.next;
4137 else
4138 tmp = mddev->all_mddevs.next;
4139 if (tmp != &all_mddevs)
4140 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4141 else {
4142 next_mddev = (void*)2;
4143 *pos = 0x10000;
4144 }
4145 spin_unlock(&all_mddevs_lock);
4146
4147 if (v != (void*)1)
4148 mddev_put(mddev);
4149 return next_mddev;
4150
4151}
4152
4153static void md_seq_stop(struct seq_file *seq, void *v)
4154{
4155 mddev_t *mddev = v;
4156
4157 if (mddev && v != (void*)1 && v != (void*)2)
4158 mddev_put(mddev);
4159}
4160
d7603b7e
N		 4161struct mdstat_info {
4162 int event;
4163};
4164
1da177e4
LT		 4165static int md_seq_show(struct seq_file *seq, void *v)
4166{
4167 mddev_t *mddev = v;
4168 sector_t size;
4169 struct list_head *tmp2;
4170 mdk_rdev_t *rdev;
d7603b7e 4171 struct mdstat_info *mi = seq->private;
32a7627c 4172 struct bitmap *bitmap;
1da177e4
LT		 4173
4174 if (v == (void*)1) {
2604b703 4175 struct mdk_personality *pers;
1da177e4
LT		 4176 seq_printf(seq, "Personalities : ");
4177 spin_lock(&pers_lock);
2604b703
N		 4178 list_for_each_entry(pers, &pers_list, list)
4179 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 4180
4181 spin_unlock(&pers_lock);
4182 seq_printf(seq, "\n");
d7603b7e 4183 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 4184 return 0;
4185 }
4186 if (v == (void*)2) {
4187 status_unused(seq);
4188 return 0;
4189 }
4190
4191 if (mddev_lock(mddev)!=0)
4192 return -EINTR;
4193 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4194 seq_printf(seq, "%s : %sactive", mdname(mddev),
4195 mddev->pers ? "" : "in");
4196 if (mddev->pers) {
f91de92e 4197 if (mddev->ro==1)
1da177e4 4198 seq_printf(seq, " (read-only)");
f91de92e
N		 4199 if (mddev->ro==2)
4200 seq_printf(seq, "(auto-read-only)");
1da177e4
LT		 4201 seq_printf(seq, " %s", mddev->pers->name);
4202 }
4203
4204 size = 0;
4205 ITERATE_RDEV(mddev,rdev,tmp2) {
4206 char b[BDEVNAME_SIZE];
4207 seq_printf(seq, " %s[%d]",
4208 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 4209 if (test_bit(WriteMostly, &rdev->flags))
4210 seq_printf(seq, "(W)");
b2d444d7 4211 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4212 seq_printf(seq, "(F)");
4213 continue;
b325a32e
N		 4214 } else if (rdev->raid_disk < 0)
4215 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 4216 size += rdev->size;
4217 }
4218
4219 if (!list_empty(&mddev->disks)) {
4220 if (mddev->pers)
4221 seq_printf(seq, "\n %llu blocks",
4222 (unsigned long long)mddev->array_size);
4223 else
4224 seq_printf(seq, "\n %llu blocks",
4225 (unsigned long long)size);
4226 }
1cd6bf19
N		 4227 if (mddev->persistent) {
4228 if (mddev->major_version != 0 ||
4229 mddev->minor_version != 90) {
4230 seq_printf(seq," super %d.%d",
4231 mddev->major_version,
4232 mddev->minor_version);
4233 }
4234 } else
4235 seq_printf(seq, " super non-persistent");
1da177e4
LT		 4236
4237 if (mddev->pers) {
4238 mddev->pers->status (seq, mddev);
4239 seq_printf(seq, "\n ");
8e1b39d6
N		 4240 if (mddev->pers->sync_request) {
4241 if (mddev->curr_resync > 2) {
4242 status_resync (seq, mddev);
4243 seq_printf(seq, "\n ");
4244 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4245 seq_printf(seq, "\tresync=DELAYED\n ");
4246 else if (mddev->recovery_cp < MaxSector)
4247 seq_printf(seq, "\tresync=PENDING\n ");
4248 }
32a7627c
N		 4249 } else
4250 seq_printf(seq, "\n ");
4251
4252 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 4253 unsigned long chunk_kb;
4254 unsigned long flags;
32a7627c
N		 4255 spin_lock_irqsave(&bitmap->lock, flags);
4256 chunk_kb = bitmap->chunksize >> 10;
4257 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4258 "%lu%s chunk",
4259 bitmap->pages - bitmap->missing_pages,
4260 bitmap->pages,
4261 (bitmap->pages - bitmap->missing_pages)
4262 << (PAGE_SHIFT - 10),
4263 chunk_kb ? chunk_kb : bitmap->chunksize,
4264 chunk_kb ? "KB" : "B");
78d742d8
N		 4265 if (bitmap->file) {
4266 seq_printf(seq, ", file: ");
4267 seq_path(seq, bitmap->file->f_vfsmnt,
4268 bitmap->file->f_dentry," \t\n");
32a7627c 4269 }
78d742d8 4270
32a7627c
N		 4271 seq_printf(seq, "\n");
4272 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 4273 }
4274
4275 seq_printf(seq, "\n");
4276 }
4277 mddev_unlock(mddev);
4278
4279 return 0;
4280}
4281
4282static struct seq_operations md_seq_ops = {
4283 .start = md_seq_start,
4284 .next = md_seq_next,
4285 .stop = md_seq_stop,
4286 .show = md_seq_show,
4287};
4288
4289static int md_seq_open(struct inode *inode, struct file *file)
4290{
4291 int error;
d7603b7e
N		 4292 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4293 if (mi == NULL)
4294 return -ENOMEM;
1da177e4
LT		 4295
4296 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 4297 if (error)
4298 kfree(mi);
4299 else {
4300 struct seq_file *p = file->private_data;
4301 p->private = mi;
4302 mi->event = atomic_read(&md_event_count);
4303 }
1da177e4
LT		 4304 return error;
4305}
4306
d7603b7e
N		 4307static int md_seq_release(struct inode *inode, struct file *file)
4308{
4309 struct seq_file *m = file->private_data;
4310 struct mdstat_info *mi = m->private;
4311 m->private = NULL;
4312 kfree(mi);
4313 return seq_release(inode, file);
4314}
4315
4316static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4317{
4318 struct seq_file *m = filp->private_data;
4319 struct mdstat_info *mi = m->private;
4320 int mask;
4321
4322 poll_wait(filp, &md_event_waiters, wait);
4323
4324 /* always allow read */
4325 mask = POLLIN | POLLRDNORM;
4326
4327 if (mi->event != atomic_read(&md_event_count))
4328 mask |= POLLERR | POLLPRI;
4329 return mask;
4330}
4331
1da177e4
LT		 4332static struct file_operations md_seq_fops = {
4333 .open = md_seq_open,
4334 .read = seq_read,
4335 .llseek = seq_lseek,
d7603b7e
N		 4336 .release = md_seq_release,
4337 .poll = mdstat_poll,
1da177e4
LT		 4338};
4339
2604b703 4340int register_md_personality(struct mdk_personality *p)
1da177e4 4341{
1da177e4 4342 spin_lock(&pers_lock);
2604b703
N		 4343 list_add_tail(&p->list, &pers_list);
4344 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 4345 spin_unlock(&pers_lock);
4346 return 0;
4347}
4348
2604b703 4349int unregister_md_personality(struct mdk_personality *p)
1da177e4 4350{
2604b703 4351 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 4352 spin_lock(&pers_lock);
2604b703 4353 list_del_init(&p->list);
1da177e4
LT		 4354 spin_unlock(&pers_lock);
4355 return 0;
4356}
4357
4358static int is_mddev_idle(mddev_t *mddev)
4359{
4360 mdk_rdev_t * rdev;
4361 struct list_head *tmp;
4362 int idle;
4363 unsigned long curr_events;
4364
4365 idle = 1;
4366 ITERATE_RDEV(mddev,rdev,tmp) {
4367 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 4368 curr_events = disk_stat_read(disk, sectors[0]) +
4369 disk_stat_read(disk, sectors[1]) -
1da177e4 4370 atomic_read(&disk->sync_io);
c0e48521
N		 4371 /* The difference between curr_events and last_events
4372 * will be affected by any new non-sync IO (making
4373 * curr_events bigger) and any difference in the amount of
4374 * in-flight syncio (making current_events bigger or smaller)
4375 * The amount in-flight is currently limited to
4376 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4377 * which is at most 4096 sectors.
4378 * These numbers are fairly fragile and should be made
4379 * more robust, probably by enforcing the
4380 * 'window size' that md_do_sync sort-of uses.
4381 *
1da177e4
LT		 4382 * Note: the following is an unsigned comparison.
4383 */
c0e48521 4384 if ((curr_events - rdev->last_events + 4096) > 8192) {
1da177e4
LT		 4385 rdev->last_events = curr_events;
4386 idle = 0;
4387 }
4388 }
4389 return idle;
4390}
4391
4392void md_done_sync(mddev_t *mddev, int blocks, int ok)
4393{
4394 /* another "blocks" (512byte) blocks have been synced */
4395 atomic_sub(blocks, &mddev->recovery_active);
4396 wake_up(&mddev->recovery_wait);
4397 if (!ok) {
4398 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4399 md_wakeup_thread(mddev->thread);
4400 // stop recovery, signal do_sync ....
4401 }
4402}
4403
4404
06d91a5f
N		 4405/* md_write_start(mddev, bi)
4406 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 4407 * in superblock) before writing, schedule a superblock update
4408 * and wait for it to complete.
06d91a5f 4409 */
3d310eb7 4410void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 4411{
06d91a5f 4412 if (bio_data_dir(bi) != WRITE)
3d310eb7 4413 return;
06d91a5f 4414
f91de92e
N		 4415 BUG_ON(mddev->ro == 1);
4416 if (mddev->ro == 2) {
4417 /* need to switch to read/write */
4418 mddev->ro = 0;
4419 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4420 md_wakeup_thread(mddev->thread);
4421 }
06d91a5f 4422 atomic_inc(&mddev->writes_pending);
06d91a5f 4423 if (mddev->in_sync) {
a9701a30 4424 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 4425 if (mddev->in_sync) {
4426 mddev->in_sync = 0;
4427 mddev->sb_dirty = 1;
4428 md_wakeup_thread(mddev->thread);
4429 }
a9701a30 4430 spin_unlock_irq(&mddev->write_lock);
06d91a5f 4431 }
3d310eb7 4432 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
1da177e4
LT		 4433}
4434
4435void md_write_end(mddev_t *mddev)
4436{
4437 if (atomic_dec_and_test(&mddev->writes_pending)) {
4438 if (mddev->safemode == 2)
4439 md_wakeup_thread(mddev->thread);
4440 else
4441 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4442 }
4443}
4444
75c96f85 4445static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 4446
4447#define SYNC_MARKS 10
4448#define SYNC_MARK_STEP (3*HZ)
4449static void md_do_sync(mddev_t *mddev)
4450{
4451 mddev_t *mddev2;
4452 unsigned int currspeed = 0,
4453 window;
57afd89f 4454 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 4455 unsigned long mark[SYNC_MARKS];
4456 sector_t mark_cnt[SYNC_MARKS];
4457 int last_mark,m;
4458 struct list_head *tmp;
4459 sector_t last_check;
57afd89f 4460 int skipped = 0;
1da177e4
LT		 4461
4462 /* just incase thread restarts... */
4463 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4464 return;
4465
4466 /* we overload curr_resync somewhat here.
4467 * 0 == not engaged in resync at all
4468 * 2 == checking that there is no conflict with another sync
4469 * 1 == like 2, but have yielded to allow conflicting resync to
4470 * commense
4471 * other == active in resync - this many blocks
4472 *
4473 * Before starting a resync we must have set curr_resync to
4474 * 2, and then checked that every "conflicting" array has curr_resync
4475 * less than ours. When we find one that is the same or higher
4476 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4477 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4478 * This will mean we have to start checking from the beginning again.
4479 *
4480 */
4481
4482 do {
4483 mddev->curr_resync = 2;
4484
4485 try_again:
787453c2 4486 if (kthread_should_stop()) {
6985c43f 4487 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 4488 goto skip;
4489 }
4490 ITERATE_MDDEV(mddev2,tmp) {
1da177e4
LT		 4491 if (mddev2 == mddev)
4492 continue;
4493 if (mddev2->curr_resync &&
4494 match_mddev_units(mddev,mddev2)) {
4495 DEFINE_WAIT(wq);
4496 if (mddev < mddev2 && mddev->curr_resync == 2) {
4497 /* arbitrarily yield */
4498 mddev->curr_resync = 1;
4499 wake_up(&resync_wait);
4500 }
4501 if (mddev > mddev2 && mddev->curr_resync == 1)
4502 /* no need to wait here, we can wait the next
4503 * time 'round when curr_resync == 2
4504 */
4505 continue;
787453c2
N		 4506 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4507 if (!kthread_should_stop() &&
8712e553 4508 mddev2->curr_resync >= mddev->curr_resync) {
1da177e4
LT		 4509 printk(KERN_INFO "md: delaying resync of %s"
4510 " until %s has finished resync (they"
4511 " share one or more physical units)\n",
4512 mdname(mddev), mdname(mddev2));
4513 mddev_put(mddev2);
4514 schedule();
4515 finish_wait(&resync_wait, &wq);
4516 goto try_again;
4517 }
4518 finish_wait(&resync_wait, &wq);
4519 }
4520 }
4521 } while (mddev->curr_resync < 2);
4522
9d88883e 4523 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 4524 /* resync follows the size requested by the personality,
57afd89f 4525 * which defaults to physical size, but can be virtual size
1da177e4
LT		 4526 */
4527 max_sectors = mddev->resync_max_sectors;
9d88883e
N		 4528 mddev->resync_mismatches = 0;
4529 } else
1da177e4
LT		 4530 /* recovery follows the physical size of devices */
4531 max_sectors = mddev->size << 1;
4532
4533 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4534 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
88202a0c 4535 " %d KB/sec/disc.\n", speed_min(mddev));
338cec32 4536 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
1da177e4 4537 "(but not more than %d KB/sec) for reconstruction.\n",
88202a0c 4538 speed_max(mddev));
1da177e4
LT		 4539
4540 is_mddev_idle(mddev); /* this also initializes IO event counters */
32a7627c 4541 /* we don't use the checkpoint if there's a bitmap */
24dd469d
N		 4542 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4543 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1da177e4
LT		 4544 j = mddev->recovery_cp;
4545 else
4546 j = 0;
57afd89f 4547 io_sectors = 0;
1da177e4
LT		 4548 for (m = 0; m < SYNC_MARKS; m++) {
4549 mark[m] = jiffies;
57afd89f 4550 mark_cnt[m] = io_sectors;
1da177e4
LT		 4551 }
4552 last_mark = 0;
4553 mddev->resync_mark = mark[last_mark];
4554 mddev->resync_mark_cnt = mark_cnt[last_mark];
4555
4556 /*
4557 * Tune reconstruction:
4558 */
4559 window = 32*(PAGE_SIZE/512);
4560 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4561 window/2,(unsigned long long) max_sectors/2);
4562
4563 atomic_set(&mddev->recovery_active, 0);
4564 init_waitqueue_head(&mddev->recovery_wait);
4565 last_check = 0;
4566
4567 if (j>2) {
4568 printk(KERN_INFO
4569 "md: resuming recovery of %s from checkpoint.\n",
4570 mdname(mddev));
4571 mddev->curr_resync = j;
4572 }
4573
4574 while (j < max_sectors) {
57afd89f 4575 sector_t sectors;
1da177e4 4576
57afd89f
N		 4577 skipped = 0;
4578 sectors = mddev->pers->sync_request(mddev, j, &skipped,
88202a0c 4579 currspeed < speed_min(mddev));
57afd89f 4580 if (sectors == 0) {
1da177e4
LT		 4581 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4582 goto out;
4583 }
57afd89f
N		 4584
4585 if (!skipped) { /* actual IO requested */
4586 io_sectors += sectors;
4587 atomic_add(sectors, &mddev->recovery_active);
4588 }
4589
1da177e4
LT		 4590 j += sectors;
4591 if (j>1) mddev->curr_resync = j;
d7603b7e
N		 4592 if (last_check == 0)
4593 /* this is the earliers that rebuilt will be
4594 * visible in /proc/mdstat
4595 */
4596 md_new_event(mddev);
57afd89f
N		 4597
4598 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 4599 continue;
4600
57afd89f 4601 last_check = io_sectors;
1da177e4
LT		 4602
4603 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4604 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4605 break;
4606
4607 repeat:
4608 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4609 /* step marks */
4610 int next = (last_mark+1) % SYNC_MARKS;
4611
4612 mddev->resync_mark = mark[next];
4613 mddev->resync_mark_cnt = mark_cnt[next];
4614 mark[next] = jiffies;
57afd89f 4615 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 4616 last_mark = next;
4617 }
4618
4619
787453c2 4620 if (kthread_should_stop()) {
1da177e4
LT		 4621 /*
4622 * got a signal, exit.
4623 */
4624 printk(KERN_INFO
4625 "md: md_do_sync() got signal ... exiting\n");
1da177e4
LT		 4626 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4627 goto out;
4628 }
4629
4630 /*
4631 * this loop exits only if either when we are slower than
4632 * the 'hard' speed limit, or the system was IO-idle for
4633 * a jiffy.
4634 * the system might be non-idle CPU-wise, but we only care
4635 * about not overloading the IO subsystem. (things like an
4636 * e2fsck being done on the RAID array should execute fast)
4637 */
4638 mddev->queue->unplug_fn(mddev->queue);
4639 cond_resched();
4640
57afd89f
N		 4641 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4642 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 4643
88202a0c
N		 4644 if (currspeed > speed_min(mddev)) {
4645 if ((currspeed > speed_max(mddev)) ||
1da177e4 4646 !is_mddev_idle(mddev)) {
c0e48521 4647 msleep(500);
1da177e4
LT		 4648 goto repeat;
4649 }
4650 }
4651 }
4652 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4653 /*
4654 * this also signals 'finished resyncing' to md_stop
4655 */
4656 out:
4657 mddev->queue->unplug_fn(mddev->queue);
4658
4659 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4660
4661 /* tell personality that we are finished */
57afd89f 4662 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 4663
4664 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4665 mddev->curr_resync > 2 &&
4666 mddev->curr_resync >= mddev->recovery_cp) {
4667 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4668 printk(KERN_INFO
4669 "md: checkpointing recovery of %s.\n",
4670 mdname(mddev));
4671 mddev->recovery_cp = mddev->curr_resync;
4672 } else
4673 mddev->recovery_cp = MaxSector;
4674 }
4675
1da177e4
LT		 4676 skip:
4677 mddev->curr_resync = 0;
4678 wake_up(&resync_wait);
4679 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4680 md_wakeup_thread(mddev->thread);
4681}
4682
4683
4684/*
4685 * This routine is regularly called by all per-raid-array threads to
4686 * deal with generic issues like resync and super-block update.
4687 * Raid personalities that don't have a thread (linear/raid0) do not
4688 * need this as they never do any recovery or update the superblock.
4689 *
4690 * It does not do any resync itself, but rather "forks" off other threads
4691 * to do that as needed.
4692 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4693 * "->recovery" and create a thread at ->sync_thread.
4694 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4695 * and wakeups up this thread which will reap the thread and finish up.
4696 * This thread also removes any faulty devices (with nr_pending == 0).
4697 *
4698 * The overall approach is:
4699 * 1/ if the superblock needs updating, update it.
4700 * 2/ If a recovery thread is running, don't do anything else.
4701 * 3/ If recovery has finished, clean up, possibly marking spares active.
4702 * 4/ If there are any faulty devices, remove them.
4703 * 5/ If array is degraded, try to add spares devices
4704 * 6/ If array has spares or is not in-sync, start a resync thread.
4705 */
4706void md_check_recovery(mddev_t *mddev)
4707{
4708 mdk_rdev_t *rdev;
4709 struct list_head *rtmp;
4710
4711
5f40402d
N		 4712 if (mddev->bitmap)
4713 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 4714
4715 if (mddev->ro)
4716 return;
fca4d848
N		 4717
4718 if (signal_pending(current)) {
4719 if (mddev->pers->sync_request) {
4720 printk(KERN_INFO "md: %s in immediate safe mode\n",
4721 mdname(mddev));
4722 mddev->safemode = 2;
4723 }
4724 flush_signals(current);
4725 }
4726
1da177e4
LT		 4727 if ( ! (
4728 mddev->sb_dirty ||
4729 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 4730 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4731 (mddev->safemode == 1) ||
4732 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4733 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 4734 ))
4735 return;
fca4d848 4736
1da177e4
LT		 4737 if (mddev_trylock(mddev)==0) {
4738 int spares =0;
fca4d848 4739
a9701a30 4740 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 4741 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4742 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4743 mddev->in_sync = 1;
4744 mddev->sb_dirty = 1;
4745 }
4746 if (mddev->safemode == 1)
4747 mddev->safemode = 0;
a9701a30 4748 spin_unlock_irq(&mddev->write_lock);
fca4d848 4749
1da177e4
LT		 4750 if (mddev->sb_dirty)
4751 md_update_sb(mddev);
06d91a5f 4752
06d91a5f 4753
1da177e4
LT		 4754 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4755 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4756 /* resync/recovery still happening */
4757 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4758 goto unlock;
4759 }
4760 if (mddev->sync_thread) {
4761 /* resync has finished, collect result */
4762 md_unregister_thread(mddev->sync_thread);
4763 mddev->sync_thread = NULL;
4764 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4765 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4766 /* success...*/
4767 /* activate any spares */
4768 mddev->pers->spare_active(mddev);
4769 }
4770 md_update_sb(mddev);
41158c7e
N		 4771
4772 /* if array is no-longer degraded, then any saved_raid_disk
4773 * information must be scrapped
4774 */
4775 if (!mddev->degraded)
4776 ITERATE_RDEV(mddev,rdev,rtmp)
4777 rdev->saved_raid_disk = -1;
4778
1da177e4
LT		 4779 mddev->recovery = 0;
4780 /* flag recovery needed just to double check */
4781 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d7603b7e 4782 md_new_event(mddev);
1da177e4
LT		 4783 goto unlock;
4784 }
24dd469d
N		 4785 /* Clear some bits that don't mean anything, but
4786 * might be left set
4787 */
4788 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4789 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4790 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4791 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4
LT		 4792
4793 /* no recovery is running.
4794 * remove any failed drives, then
4795 * add spares if possible.
4796 * Spare are also removed and re-added, to allow
4797 * the personality to fail the re-add.
4798 */
4799 ITERATE_RDEV(mddev,rdev,rtmp)
4800 if (rdev->raid_disk >= 0 &&
b2d444d7 4801 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
1da177e4 4802 atomic_read(&rdev->nr_pending)==0) {
86e6ffdd
N		 4803 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4804 char nm[20];
4805 sprintf(nm,"rd%d", rdev->raid_disk);
4806 sysfs_remove_link(&mddev->kobj, nm);
1da177e4 4807 rdev->raid_disk = -1;
86e6ffdd 4808 }
1da177e4
LT		 4809 }
4810
4811 if (mddev->degraded) {
4812 ITERATE_RDEV(mddev,rdev,rtmp)
4813 if (rdev->raid_disk < 0
b2d444d7 4814 && !test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 4815 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4816 char nm[20];
4817 sprintf(nm, "rd%d", rdev->raid_disk);
4818 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
1da177e4 4819 spares++;
d7603b7e 4820 md_new_event(mddev);
86e6ffdd 4821 } else
1da177e4
LT		 4822 break;
4823 }
4824 }
4825
24dd469d
N		 4826 if (spares) {
4827 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4828 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4829 } else if (mddev->recovery_cp < MaxSector) {
4830 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4831 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4832 /* nothing to be done ... */
1da177e4 4833 goto unlock;
24dd469d 4834
1da177e4
LT		 4835 if (mddev->pers->sync_request) {
4836 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 4837 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4838 /* We are adding a device or devices to an array
4839 * which has the bitmap stored on all devices.
4840 * So make sure all bitmap pages get written
4841 */
4842 bitmap_write_all(mddev->bitmap);
4843 }
1da177e4
LT		 4844 mddev->sync_thread = md_register_thread(md_do_sync,
4845 mddev,
4846 "%s_resync");
4847 if (!mddev->sync_thread) {
4848 printk(KERN_ERR "%s: could not start resync"
4849 " thread...\n",
4850 mdname(mddev));
4851 /* leave the spares where they are, it shouldn't hurt */
4852 mddev->recovery = 0;
d7603b7e 4853 } else
1da177e4 4854 md_wakeup_thread(mddev->sync_thread);
d7603b7e 4855 md_new_event(mddev);
1da177e4
LT		 4856 }
4857 unlock:
4858 mddev_unlock(mddev);
4859 }
4860}
4861
75c96f85
AB		 4862static int md_notify_reboot(struct notifier_block *this,
4863 unsigned long code, void *x)
1da177e4
LT		 4864{
4865 struct list_head *tmp;
4866 mddev_t *mddev;
4867
4868 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4869
4870 printk(KERN_INFO "md: stopping all md devices.\n");
4871
4872 ITERATE_MDDEV(mddev,tmp)
4873 if (mddev_trylock(mddev)==0)
4874 do_md_stop (mddev, 1);
4875 /*
4876 * certain more exotic SCSI devices are known to be
4877 * volatile wrt too early system reboots. While the
4878 * right place to handle this issue is the given
4879 * driver, we do want to have a safe RAID driver ...
4880 */
4881 mdelay(1000*1);
4882 }
4883 return NOTIFY_DONE;
4884}
4885
75c96f85 4886static struct notifier_block md_notifier = {
1da177e4
LT		 4887 .notifier_call = md_notify_reboot,
4888 .next = NULL,
4889 .priority = INT_MAX, /* before any real devices */
4890};
4891
4892static void md_geninit(void)
4893{
4894 struct proc_dir_entry *p;
4895
4896 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4897
4898 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4899 if (p)
4900 p->proc_fops = &md_seq_fops;
4901}
4902
75c96f85 4903static int __init md_init(void)
1da177e4
LT		 4904{
4905 int minor;
4906
4907 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4908 " MD_SB_DISKS=%d\n",
4909 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4910 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
bd926c63 4911 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
32a7627c 4912 BITMAP_MINOR);
1da177e4
LT		 4913
4914 if (register_blkdev(MAJOR_NR, "md"))
4915 return -1;
4916 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4917 unregister_blkdev(MAJOR_NR, "md");
4918 return -1;
4919 }
4920 devfs_mk_dir("md");
4921 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4922 md_probe, NULL, NULL);
4923 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4924 md_probe, NULL, NULL);
4925
4926 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4927 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4928 S_IFBLK|S_IRUSR|S_IWUSR,
4929 "md/%d", minor);
4930
4931 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4932 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4933 S_IFBLK|S_IRUSR|S_IWUSR,
4934 "md/mdp%d", minor);
4935
4936
4937 register_reboot_notifier(&md_notifier);
4938 raid_table_header = register_sysctl_table(raid_root_table, 1);
4939
4940 md_geninit();
4941 return (0);
4942}
4943
4944
4945#ifndef MODULE
4946
4947/*
4948 * Searches all registered partitions for autorun RAID arrays
4949 * at boot time.
4950 */
4951static dev_t detected_devices[128];
4952static int dev_cnt;
4953
4954void md_autodetect_dev(dev_t dev)
4955{
4956 if (dev_cnt >= 0 && dev_cnt < 127)
4957 detected_devices[dev_cnt++] = dev;
4958}
4959
4960
4961static void autostart_arrays(int part)
4962{
4963 mdk_rdev_t *rdev;
4964 int i;
4965
4966 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4967
4968 for (i = 0; i < dev_cnt; i++) {
4969 dev_t dev = detected_devices[i];
4970
4971 rdev = md_import_device(dev,0, 0);
4972 if (IS_ERR(rdev))
4973 continue;
4974
b2d444d7 4975 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4976 MD_BUG();
4977 continue;
4978 }
4979 list_add(&rdev->same_set, &pending_raid_disks);
4980 }
4981 dev_cnt = 0;
4982
4983 autorun_devices(part);
4984}
4985
4986#endif
4987
4988static __exit void md_exit(void)
4989{
4990 mddev_t *mddev;
4991 struct list_head *tmp;
4992 int i;
4993 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4994 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4995 for (i=0; i < MAX_MD_DEVS; i++)
4996 devfs_remove("md/%d", i);
4997 for (i=0; i < MAX_MD_DEVS; i++)
4998 devfs_remove("md/d%d", i);
4999
5000 devfs_remove("md");
5001
5002 unregister_blkdev(MAJOR_NR,"md");
5003 unregister_blkdev(mdp_major, "mdp");
5004 unregister_reboot_notifier(&md_notifier);
5005 unregister_sysctl_table(raid_table_header);
5006 remove_proc_entry("mdstat", NULL);
5007 ITERATE_MDDEV(mddev,tmp) {
5008 struct gendisk *disk = mddev->gendisk;
5009 if (!disk)
5010 continue;
5011 export_array(mddev);
5012 del_gendisk(disk);
5013 put_disk(disk);
5014 mddev->gendisk = NULL;
5015 mddev_put(mddev);
5016 }
5017}
5018
5019module_init(md_init)
5020module_exit(md_exit)
5021
f91de92e
N		 5022static int get_ro(char *buffer, struct kernel_param *kp)
5023{
5024 return sprintf(buffer, "%d", start_readonly);
5025}
5026static int set_ro(const char *val, struct kernel_param *kp)
5027{
5028 char *e;
5029 int num = simple_strtoul(val, &e, 10);
5030 if (*val && (*e == '\0' || *e == '\n')) {
5031 start_readonly = num;
4dbcdc75 5032 return 0;
f91de92e
N		 5033 }
5034 return -EINVAL;
5035}
5036
5037module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
6ff8d8ec
N		 5038module_param(start_dirty_degraded, int, 0644);
5039
f91de92e 5040
1da177e4
LT		 5041EXPORT_SYMBOL(register_md_personality);
5042EXPORT_SYMBOL(unregister_md_personality);
5043EXPORT_SYMBOL(md_error);
5044EXPORT_SYMBOL(md_done_sync);
5045EXPORT_SYMBOL(md_write_start);
5046EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 5047EXPORT_SYMBOL(md_register_thread);
5048EXPORT_SYMBOL(md_unregister_thread);
5049EXPORT_SYMBOL(md_wakeup_thread);
5050EXPORT_SYMBOL(md_print_devices);
5051EXPORT_SYMBOL(md_check_recovery);
5052MODULE_LICENSE("GPL");
aa1595e9 5053MODULE_ALIAS("md");
72008652 5054MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)